]>
Commit | Line | Data |
---|---|---|
2e405ad8 JB |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | ||
784352fe | 3 | #include "misc.h" |
2e405ad8 JB |
4 | #include "ctree.h" |
5 | #include "block-group.h" | |
3eeb3226 | 6 | #include "space-info.h" |
9f21246d JB |
7 | #include "disk-io.h" |
8 | #include "free-space-cache.h" | |
9 | #include "free-space-tree.h" | |
e3e0520b JB |
10 | #include "disk-io.h" |
11 | #include "volumes.h" | |
12 | #include "transaction.h" | |
13 | #include "ref-verify.h" | |
4358d963 JB |
14 | #include "sysfs.h" |
15 | #include "tree-log.h" | |
77745c05 | 16 | #include "delalloc-space.h" |
2e405ad8 | 17 | |
878d7b67 JB |
18 | /* |
19 | * Return target flags in extended format or 0 if restripe for this chunk_type | |
20 | * is not in progress | |
21 | * | |
22 | * Should be called with balance_lock held | |
23 | */ | |
e11c0406 | 24 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) |
878d7b67 JB |
25 | { |
26 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
27 | u64 target = 0; | |
28 | ||
29 | if (!bctl) | |
30 | return 0; | |
31 | ||
32 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
33 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
34 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
35 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
36 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
37 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
38 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
39 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
40 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
41 | } | |
42 | ||
43 | return target; | |
44 | } | |
45 | ||
46 | /* | |
47 | * @flags: available profiles in extended format (see ctree.h) | |
48 | * | |
49 | * Return reduced profile in chunk format. If profile changing is in progress | |
50 | * (either running or paused) picks the target profile (if it's already | |
51 | * available), otherwise falls back to plain reducing. | |
52 | */ | |
53 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) | |
54 | { | |
55 | u64 num_devices = fs_info->fs_devices->rw_devices; | |
56 | u64 target; | |
57 | u64 raid_type; | |
58 | u64 allowed = 0; | |
59 | ||
60 | /* | |
61 | * See if restripe for this chunk_type is in progress, if so try to | |
62 | * reduce to the target profile | |
63 | */ | |
64 | spin_lock(&fs_info->balance_lock); | |
e11c0406 | 65 | target = get_restripe_target(fs_info, flags); |
878d7b67 JB |
66 | if (target) { |
67 | /* Pick target profile only if it's already available */ | |
68 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
69 | spin_unlock(&fs_info->balance_lock); | |
70 | return extended_to_chunk(target); | |
71 | } | |
72 | } | |
73 | spin_unlock(&fs_info->balance_lock); | |
74 | ||
75 | /* First, mask out the RAID levels which aren't possible */ | |
76 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { | |
77 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
78 | allowed |= btrfs_raid_array[raid_type].bg_flag; | |
79 | } | |
80 | allowed &= flags; | |
81 | ||
82 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
83 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
84 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
85 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
86 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
87 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
88 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
89 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
90 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
91 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
92 | ||
93 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
94 | ||
95 | return extended_to_chunk(flags | allowed); | |
96 | } | |
97 | ||
98 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) | |
99 | { | |
100 | unsigned seq; | |
101 | u64 flags; | |
102 | ||
103 | do { | |
104 | flags = orig_flags; | |
105 | seq = read_seqbegin(&fs_info->profiles_lock); | |
106 | ||
107 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
108 | flags |= fs_info->avail_data_alloc_bits; | |
109 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
110 | flags |= fs_info->avail_system_alloc_bits; | |
111 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
112 | flags |= fs_info->avail_metadata_alloc_bits; | |
113 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
114 | ||
115 | return btrfs_reduce_alloc_profile(fs_info, flags); | |
116 | } | |
117 | ||
118 | u64 btrfs_get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) | |
119 | { | |
120 | return get_alloc_profile(fs_info, orig_flags); | |
121 | } | |
122 | ||
3cad1284 JB |
123 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
124 | { | |
125 | atomic_inc(&cache->count); | |
126 | } | |
127 | ||
128 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
129 | { | |
130 | if (atomic_dec_and_test(&cache->count)) { | |
131 | WARN_ON(cache->pinned > 0); | |
132 | WARN_ON(cache->reserved > 0); | |
133 | ||
134 | /* | |
135 | * If not empty, someone is still holding mutex of | |
136 | * full_stripe_lock, which can only be released by caller. | |
137 | * And it will definitely cause use-after-free when caller | |
138 | * tries to release full stripe lock. | |
139 | * | |
140 | * No better way to resolve, but only to warn. | |
141 | */ | |
142 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
143 | kfree(cache->free_space_ctl); | |
144 | kfree(cache); | |
145 | } | |
146 | } | |
147 | ||
4358d963 JB |
148 | /* |
149 | * This adds the block group to the fs_info rb tree for the block group cache | |
150 | */ | |
151 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, | |
152 | struct btrfs_block_group_cache *block_group) | |
153 | { | |
154 | struct rb_node **p; | |
155 | struct rb_node *parent = NULL; | |
156 | struct btrfs_block_group_cache *cache; | |
157 | ||
158 | spin_lock(&info->block_group_cache_lock); | |
159 | p = &info->block_group_cache_tree.rb_node; | |
160 | ||
161 | while (*p) { | |
162 | parent = *p; | |
163 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
164 | cache_node); | |
165 | if (block_group->key.objectid < cache->key.objectid) { | |
166 | p = &(*p)->rb_left; | |
167 | } else if (block_group->key.objectid > cache->key.objectid) { | |
168 | p = &(*p)->rb_right; | |
169 | } else { | |
170 | spin_unlock(&info->block_group_cache_lock); | |
171 | return -EEXIST; | |
172 | } | |
173 | } | |
174 | ||
175 | rb_link_node(&block_group->cache_node, parent, p); | |
176 | rb_insert_color(&block_group->cache_node, | |
177 | &info->block_group_cache_tree); | |
178 | ||
179 | if (info->first_logical_byte > block_group->key.objectid) | |
180 | info->first_logical_byte = block_group->key.objectid; | |
181 | ||
182 | spin_unlock(&info->block_group_cache_lock); | |
183 | ||
184 | return 0; | |
185 | } | |
186 | ||
2e405ad8 JB |
187 | /* |
188 | * This will return the block group at or after bytenr if contains is 0, else | |
189 | * it will return the block group that contains the bytenr | |
190 | */ | |
191 | static struct btrfs_block_group_cache *block_group_cache_tree_search( | |
192 | struct btrfs_fs_info *info, u64 bytenr, int contains) | |
193 | { | |
194 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
195 | struct rb_node *n; | |
196 | u64 end, start; | |
197 | ||
198 | spin_lock(&info->block_group_cache_lock); | |
199 | n = info->block_group_cache_tree.rb_node; | |
200 | ||
201 | while (n) { | |
202 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
203 | cache_node); | |
204 | end = cache->key.objectid + cache->key.offset - 1; | |
205 | start = cache->key.objectid; | |
206 | ||
207 | if (bytenr < start) { | |
208 | if (!contains && (!ret || start < ret->key.objectid)) | |
209 | ret = cache; | |
210 | n = n->rb_left; | |
211 | } else if (bytenr > start) { | |
212 | if (contains && bytenr <= end) { | |
213 | ret = cache; | |
214 | break; | |
215 | } | |
216 | n = n->rb_right; | |
217 | } else { | |
218 | ret = cache; | |
219 | break; | |
220 | } | |
221 | } | |
222 | if (ret) { | |
223 | btrfs_get_block_group(ret); | |
224 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) | |
225 | info->first_logical_byte = ret->key.objectid; | |
226 | } | |
227 | spin_unlock(&info->block_group_cache_lock); | |
228 | ||
229 | return ret; | |
230 | } | |
231 | ||
232 | /* | |
233 | * Return the block group that starts at or after bytenr | |
234 | */ | |
235 | struct btrfs_block_group_cache *btrfs_lookup_first_block_group( | |
236 | struct btrfs_fs_info *info, u64 bytenr) | |
237 | { | |
238 | return block_group_cache_tree_search(info, bytenr, 0); | |
239 | } | |
240 | ||
241 | /* | |
242 | * Return the block group that contains the given bytenr | |
243 | */ | |
244 | struct btrfs_block_group_cache *btrfs_lookup_block_group( | |
245 | struct btrfs_fs_info *info, u64 bytenr) | |
246 | { | |
247 | return block_group_cache_tree_search(info, bytenr, 1); | |
248 | } | |
249 | ||
250 | struct btrfs_block_group_cache *btrfs_next_block_group( | |
251 | struct btrfs_block_group_cache *cache) | |
252 | { | |
253 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
254 | struct rb_node *node; | |
255 | ||
256 | spin_lock(&fs_info->block_group_cache_lock); | |
257 | ||
258 | /* If our block group was removed, we need a full search. */ | |
259 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
260 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
261 | ||
262 | spin_unlock(&fs_info->block_group_cache_lock); | |
263 | btrfs_put_block_group(cache); | |
264 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; | |
265 | } | |
266 | node = rb_next(&cache->cache_node); | |
267 | btrfs_put_block_group(cache); | |
268 | if (node) { | |
269 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
270 | cache_node); | |
271 | btrfs_get_block_group(cache); | |
272 | } else | |
273 | cache = NULL; | |
274 | spin_unlock(&fs_info->block_group_cache_lock); | |
275 | return cache; | |
276 | } | |
3eeb3226 JB |
277 | |
278 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
279 | { | |
280 | struct btrfs_block_group_cache *bg; | |
281 | bool ret = true; | |
282 | ||
283 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
284 | if (!bg) | |
285 | return false; | |
286 | ||
287 | spin_lock(&bg->lock); | |
288 | if (bg->ro) | |
289 | ret = false; | |
290 | else | |
291 | atomic_inc(&bg->nocow_writers); | |
292 | spin_unlock(&bg->lock); | |
293 | ||
294 | /* No put on block group, done by btrfs_dec_nocow_writers */ | |
295 | if (!ret) | |
296 | btrfs_put_block_group(bg); | |
297 | ||
298 | return ret; | |
299 | } | |
300 | ||
301 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
302 | { | |
303 | struct btrfs_block_group_cache *bg; | |
304 | ||
305 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
306 | ASSERT(bg); | |
307 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
308 | wake_up_var(&bg->nocow_writers); | |
309 | /* | |
310 | * Once for our lookup and once for the lookup done by a previous call | |
311 | * to btrfs_inc_nocow_writers() | |
312 | */ | |
313 | btrfs_put_block_group(bg); | |
314 | btrfs_put_block_group(bg); | |
315 | } | |
316 | ||
317 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) | |
318 | { | |
319 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); | |
320 | } | |
321 | ||
322 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
323 | const u64 start) | |
324 | { | |
325 | struct btrfs_block_group_cache *bg; | |
326 | ||
327 | bg = btrfs_lookup_block_group(fs_info, start); | |
328 | ASSERT(bg); | |
329 | if (atomic_dec_and_test(&bg->reservations)) | |
330 | wake_up_var(&bg->reservations); | |
331 | btrfs_put_block_group(bg); | |
332 | } | |
333 | ||
334 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) | |
335 | { | |
336 | struct btrfs_space_info *space_info = bg->space_info; | |
337 | ||
338 | ASSERT(bg->ro); | |
339 | ||
340 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
341 | return; | |
342 | ||
343 | /* | |
344 | * Our block group is read only but before we set it to read only, | |
345 | * some task might have had allocated an extent from it already, but it | |
346 | * has not yet created a respective ordered extent (and added it to a | |
347 | * root's list of ordered extents). | |
348 | * Therefore wait for any task currently allocating extents, since the | |
349 | * block group's reservations counter is incremented while a read lock | |
350 | * on the groups' semaphore is held and decremented after releasing | |
351 | * the read access on that semaphore and creating the ordered extent. | |
352 | */ | |
353 | down_write(&space_info->groups_sem); | |
354 | up_write(&space_info->groups_sem); | |
355 | ||
356 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); | |
357 | } | |
9f21246d JB |
358 | |
359 | struct btrfs_caching_control *btrfs_get_caching_control( | |
360 | struct btrfs_block_group_cache *cache) | |
361 | { | |
362 | struct btrfs_caching_control *ctl; | |
363 | ||
364 | spin_lock(&cache->lock); | |
365 | if (!cache->caching_ctl) { | |
366 | spin_unlock(&cache->lock); | |
367 | return NULL; | |
368 | } | |
369 | ||
370 | ctl = cache->caching_ctl; | |
371 | refcount_inc(&ctl->count); | |
372 | spin_unlock(&cache->lock); | |
373 | return ctl; | |
374 | } | |
375 | ||
376 | void btrfs_put_caching_control(struct btrfs_caching_control *ctl) | |
377 | { | |
378 | if (refcount_dec_and_test(&ctl->count)) | |
379 | kfree(ctl); | |
380 | } | |
381 | ||
382 | /* | |
383 | * When we wait for progress in the block group caching, its because our | |
384 | * allocation attempt failed at least once. So, we must sleep and let some | |
385 | * progress happen before we try again. | |
386 | * | |
387 | * This function will sleep at least once waiting for new free space to show | |
388 | * up, and then it will check the block group free space numbers for our min | |
389 | * num_bytes. Another option is to have it go ahead and look in the rbtree for | |
390 | * a free extent of a given size, but this is a good start. | |
391 | * | |
392 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
393 | * any of the information in this block group. | |
394 | */ | |
395 | void btrfs_wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
396 | u64 num_bytes) | |
397 | { | |
398 | struct btrfs_caching_control *caching_ctl; | |
399 | ||
400 | caching_ctl = btrfs_get_caching_control(cache); | |
401 | if (!caching_ctl) | |
402 | return; | |
403 | ||
404 | wait_event(caching_ctl->wait, btrfs_block_group_cache_done(cache) || | |
405 | (cache->free_space_ctl->free_space >= num_bytes)); | |
406 | ||
407 | btrfs_put_caching_control(caching_ctl); | |
408 | } | |
409 | ||
410 | int btrfs_wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
411 | { | |
412 | struct btrfs_caching_control *caching_ctl; | |
413 | int ret = 0; | |
414 | ||
415 | caching_ctl = btrfs_get_caching_control(cache); | |
416 | if (!caching_ctl) | |
417 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; | |
418 | ||
419 | wait_event(caching_ctl->wait, btrfs_block_group_cache_done(cache)); | |
420 | if (cache->cached == BTRFS_CACHE_ERROR) | |
421 | ret = -EIO; | |
422 | btrfs_put_caching_control(caching_ctl); | |
423 | return ret; | |
424 | } | |
425 | ||
426 | #ifdef CONFIG_BTRFS_DEBUG | |
e11c0406 | 427 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
9f21246d JB |
428 | { |
429 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
430 | u64 start = block_group->key.objectid; | |
431 | u64 len = block_group->key.offset; | |
432 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
433 | fs_info->nodesize : fs_info->sectorsize; | |
434 | u64 step = chunk << 1; | |
435 | ||
436 | while (len > chunk) { | |
437 | btrfs_remove_free_space(block_group, start, chunk); | |
438 | start += step; | |
439 | if (len < step) | |
440 | len = 0; | |
441 | else | |
442 | len -= step; | |
443 | } | |
444 | } | |
445 | #endif | |
446 | ||
447 | /* | |
448 | * This is only called by btrfs_cache_block_group, since we could have freed | |
449 | * extents we need to check the pinned_extents for any extents that can't be | |
450 | * used yet since their free space will be released as soon as the transaction | |
451 | * commits. | |
452 | */ | |
453 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, | |
454 | u64 start, u64 end) | |
455 | { | |
456 | struct btrfs_fs_info *info = block_group->fs_info; | |
457 | u64 extent_start, extent_end, size, total_added = 0; | |
458 | int ret; | |
459 | ||
460 | while (start < end) { | |
461 | ret = find_first_extent_bit(info->pinned_extents, start, | |
462 | &extent_start, &extent_end, | |
463 | EXTENT_DIRTY | EXTENT_UPTODATE, | |
464 | NULL); | |
465 | if (ret) | |
466 | break; | |
467 | ||
468 | if (extent_start <= start) { | |
469 | start = extent_end + 1; | |
470 | } else if (extent_start > start && extent_start < end) { | |
471 | size = extent_start - start; | |
472 | total_added += size; | |
473 | ret = btrfs_add_free_space(block_group, start, | |
474 | size); | |
475 | BUG_ON(ret); /* -ENOMEM or logic error */ | |
476 | start = extent_end + 1; | |
477 | } else { | |
478 | break; | |
479 | } | |
480 | } | |
481 | ||
482 | if (start < end) { | |
483 | size = end - start; | |
484 | total_added += size; | |
485 | ret = btrfs_add_free_space(block_group, start, size); | |
486 | BUG_ON(ret); /* -ENOMEM or logic error */ | |
487 | } | |
488 | ||
489 | return total_added; | |
490 | } | |
491 | ||
492 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) | |
493 | { | |
494 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; | |
495 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
496 | struct btrfs_root *extent_root = fs_info->extent_root; | |
497 | struct btrfs_path *path; | |
498 | struct extent_buffer *leaf; | |
499 | struct btrfs_key key; | |
500 | u64 total_found = 0; | |
501 | u64 last = 0; | |
502 | u32 nritems; | |
503 | int ret; | |
504 | bool wakeup = true; | |
505 | ||
506 | path = btrfs_alloc_path(); | |
507 | if (!path) | |
508 | return -ENOMEM; | |
509 | ||
510 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); | |
511 | ||
512 | #ifdef CONFIG_BTRFS_DEBUG | |
513 | /* | |
514 | * If we're fragmenting we don't want to make anybody think we can | |
515 | * allocate from this block group until we've had a chance to fragment | |
516 | * the free space. | |
517 | */ | |
518 | if (btrfs_should_fragment_free_space(block_group)) | |
519 | wakeup = false; | |
520 | #endif | |
521 | /* | |
522 | * We don't want to deadlock with somebody trying to allocate a new | |
523 | * extent for the extent root while also trying to search the extent | |
524 | * root to add free space. So we skip locking and search the commit | |
525 | * root, since its read-only | |
526 | */ | |
527 | path->skip_locking = 1; | |
528 | path->search_commit_root = 1; | |
529 | path->reada = READA_FORWARD; | |
530 | ||
531 | key.objectid = last; | |
532 | key.offset = 0; | |
533 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
534 | ||
535 | next: | |
536 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
537 | if (ret < 0) | |
538 | goto out; | |
539 | ||
540 | leaf = path->nodes[0]; | |
541 | nritems = btrfs_header_nritems(leaf); | |
542 | ||
543 | while (1) { | |
544 | if (btrfs_fs_closing(fs_info) > 1) { | |
545 | last = (u64)-1; | |
546 | break; | |
547 | } | |
548 | ||
549 | if (path->slots[0] < nritems) { | |
550 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
551 | } else { | |
552 | ret = btrfs_find_next_key(extent_root, path, &key, 0, 0); | |
553 | if (ret) | |
554 | break; | |
555 | ||
556 | if (need_resched() || | |
557 | rwsem_is_contended(&fs_info->commit_root_sem)) { | |
558 | if (wakeup) | |
559 | caching_ctl->progress = last; | |
560 | btrfs_release_path(path); | |
561 | up_read(&fs_info->commit_root_sem); | |
562 | mutex_unlock(&caching_ctl->mutex); | |
563 | cond_resched(); | |
564 | mutex_lock(&caching_ctl->mutex); | |
565 | down_read(&fs_info->commit_root_sem); | |
566 | goto next; | |
567 | } | |
568 | ||
569 | ret = btrfs_next_leaf(extent_root, path); | |
570 | if (ret < 0) | |
571 | goto out; | |
572 | if (ret) | |
573 | break; | |
574 | leaf = path->nodes[0]; | |
575 | nritems = btrfs_header_nritems(leaf); | |
576 | continue; | |
577 | } | |
578 | ||
579 | if (key.objectid < last) { | |
580 | key.objectid = last; | |
581 | key.offset = 0; | |
582 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
583 | ||
584 | if (wakeup) | |
585 | caching_ctl->progress = last; | |
586 | btrfs_release_path(path); | |
587 | goto next; | |
588 | } | |
589 | ||
590 | if (key.objectid < block_group->key.objectid) { | |
591 | path->slots[0]++; | |
592 | continue; | |
593 | } | |
594 | ||
595 | if (key.objectid >= block_group->key.objectid + | |
596 | block_group->key.offset) | |
597 | break; | |
598 | ||
599 | if (key.type == BTRFS_EXTENT_ITEM_KEY || | |
600 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
601 | total_found += add_new_free_space(block_group, last, | |
602 | key.objectid); | |
603 | if (key.type == BTRFS_METADATA_ITEM_KEY) | |
604 | last = key.objectid + | |
605 | fs_info->nodesize; | |
606 | else | |
607 | last = key.objectid + key.offset; | |
608 | ||
609 | if (total_found > CACHING_CTL_WAKE_UP) { | |
610 | total_found = 0; | |
611 | if (wakeup) | |
612 | wake_up(&caching_ctl->wait); | |
613 | } | |
614 | } | |
615 | path->slots[0]++; | |
616 | } | |
617 | ret = 0; | |
618 | ||
619 | total_found += add_new_free_space(block_group, last, | |
620 | block_group->key.objectid + | |
621 | block_group->key.offset); | |
622 | caching_ctl->progress = (u64)-1; | |
623 | ||
624 | out: | |
625 | btrfs_free_path(path); | |
626 | return ret; | |
627 | } | |
628 | ||
629 | static noinline void caching_thread(struct btrfs_work *work) | |
630 | { | |
631 | struct btrfs_block_group_cache *block_group; | |
632 | struct btrfs_fs_info *fs_info; | |
633 | struct btrfs_caching_control *caching_ctl; | |
634 | int ret; | |
635 | ||
636 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
637 | block_group = caching_ctl->block_group; | |
638 | fs_info = block_group->fs_info; | |
639 | ||
640 | mutex_lock(&caching_ctl->mutex); | |
641 | down_read(&fs_info->commit_root_sem); | |
642 | ||
643 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) | |
644 | ret = load_free_space_tree(caching_ctl); | |
645 | else | |
646 | ret = load_extent_tree_free(caching_ctl); | |
647 | ||
648 | spin_lock(&block_group->lock); | |
649 | block_group->caching_ctl = NULL; | |
650 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; | |
651 | spin_unlock(&block_group->lock); | |
652 | ||
653 | #ifdef CONFIG_BTRFS_DEBUG | |
654 | if (btrfs_should_fragment_free_space(block_group)) { | |
655 | u64 bytes_used; | |
656 | ||
657 | spin_lock(&block_group->space_info->lock); | |
658 | spin_lock(&block_group->lock); | |
659 | bytes_used = block_group->key.offset - | |
660 | btrfs_block_group_used(&block_group->item); | |
661 | block_group->space_info->bytes_used += bytes_used >> 1; | |
662 | spin_unlock(&block_group->lock); | |
663 | spin_unlock(&block_group->space_info->lock); | |
e11c0406 | 664 | fragment_free_space(block_group); |
9f21246d JB |
665 | } |
666 | #endif | |
667 | ||
668 | caching_ctl->progress = (u64)-1; | |
669 | ||
670 | up_read(&fs_info->commit_root_sem); | |
671 | btrfs_free_excluded_extents(block_group); | |
672 | mutex_unlock(&caching_ctl->mutex); | |
673 | ||
674 | wake_up(&caching_ctl->wait); | |
675 | ||
676 | btrfs_put_caching_control(caching_ctl); | |
677 | btrfs_put_block_group(block_group); | |
678 | } | |
679 | ||
680 | int btrfs_cache_block_group(struct btrfs_block_group_cache *cache, | |
681 | int load_cache_only) | |
682 | { | |
683 | DEFINE_WAIT(wait); | |
684 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
685 | struct btrfs_caching_control *caching_ctl; | |
686 | int ret = 0; | |
687 | ||
688 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); | |
689 | if (!caching_ctl) | |
690 | return -ENOMEM; | |
691 | ||
692 | INIT_LIST_HEAD(&caching_ctl->list); | |
693 | mutex_init(&caching_ctl->mutex); | |
694 | init_waitqueue_head(&caching_ctl->wait); | |
695 | caching_ctl->block_group = cache; | |
696 | caching_ctl->progress = cache->key.objectid; | |
697 | refcount_set(&caching_ctl->count, 1); | |
a0cac0ec | 698 | btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL); |
9f21246d JB |
699 | |
700 | spin_lock(&cache->lock); | |
701 | /* | |
702 | * This should be a rare occasion, but this could happen I think in the | |
703 | * case where one thread starts to load the space cache info, and then | |
704 | * some other thread starts a transaction commit which tries to do an | |
705 | * allocation while the other thread is still loading the space cache | |
706 | * info. The previous loop should have kept us from choosing this block | |
707 | * group, but if we've moved to the state where we will wait on caching | |
708 | * block groups we need to first check if we're doing a fast load here, | |
709 | * so we can wait for it to finish, otherwise we could end up allocating | |
710 | * from a block group who's cache gets evicted for one reason or | |
711 | * another. | |
712 | */ | |
713 | while (cache->cached == BTRFS_CACHE_FAST) { | |
714 | struct btrfs_caching_control *ctl; | |
715 | ||
716 | ctl = cache->caching_ctl; | |
717 | refcount_inc(&ctl->count); | |
718 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
719 | spin_unlock(&cache->lock); | |
720 | ||
721 | schedule(); | |
722 | ||
723 | finish_wait(&ctl->wait, &wait); | |
724 | btrfs_put_caching_control(ctl); | |
725 | spin_lock(&cache->lock); | |
726 | } | |
727 | ||
728 | if (cache->cached != BTRFS_CACHE_NO) { | |
729 | spin_unlock(&cache->lock); | |
730 | kfree(caching_ctl); | |
731 | return 0; | |
732 | } | |
733 | WARN_ON(cache->caching_ctl); | |
734 | cache->caching_ctl = caching_ctl; | |
735 | cache->cached = BTRFS_CACHE_FAST; | |
736 | spin_unlock(&cache->lock); | |
737 | ||
738 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { | |
739 | mutex_lock(&caching_ctl->mutex); | |
740 | ret = load_free_space_cache(cache); | |
741 | ||
742 | spin_lock(&cache->lock); | |
743 | if (ret == 1) { | |
744 | cache->caching_ctl = NULL; | |
745 | cache->cached = BTRFS_CACHE_FINISHED; | |
746 | cache->last_byte_to_unpin = (u64)-1; | |
747 | caching_ctl->progress = (u64)-1; | |
748 | } else { | |
749 | if (load_cache_only) { | |
750 | cache->caching_ctl = NULL; | |
751 | cache->cached = BTRFS_CACHE_NO; | |
752 | } else { | |
753 | cache->cached = BTRFS_CACHE_STARTED; | |
754 | cache->has_caching_ctl = 1; | |
755 | } | |
756 | } | |
757 | spin_unlock(&cache->lock); | |
758 | #ifdef CONFIG_BTRFS_DEBUG | |
759 | if (ret == 1 && | |
760 | btrfs_should_fragment_free_space(cache)) { | |
761 | u64 bytes_used; | |
762 | ||
763 | spin_lock(&cache->space_info->lock); | |
764 | spin_lock(&cache->lock); | |
765 | bytes_used = cache->key.offset - | |
766 | btrfs_block_group_used(&cache->item); | |
767 | cache->space_info->bytes_used += bytes_used >> 1; | |
768 | spin_unlock(&cache->lock); | |
769 | spin_unlock(&cache->space_info->lock); | |
e11c0406 | 770 | fragment_free_space(cache); |
9f21246d JB |
771 | } |
772 | #endif | |
773 | mutex_unlock(&caching_ctl->mutex); | |
774 | ||
775 | wake_up(&caching_ctl->wait); | |
776 | if (ret == 1) { | |
777 | btrfs_put_caching_control(caching_ctl); | |
778 | btrfs_free_excluded_extents(cache); | |
779 | return 0; | |
780 | } | |
781 | } else { | |
782 | /* | |
783 | * We're either using the free space tree or no caching at all. | |
784 | * Set cached to the appropriate value and wakeup any waiters. | |
785 | */ | |
786 | spin_lock(&cache->lock); | |
787 | if (load_cache_only) { | |
788 | cache->caching_ctl = NULL; | |
789 | cache->cached = BTRFS_CACHE_NO; | |
790 | } else { | |
791 | cache->cached = BTRFS_CACHE_STARTED; | |
792 | cache->has_caching_ctl = 1; | |
793 | } | |
794 | spin_unlock(&cache->lock); | |
795 | wake_up(&caching_ctl->wait); | |
796 | } | |
797 | ||
798 | if (load_cache_only) { | |
799 | btrfs_put_caching_control(caching_ctl); | |
800 | return 0; | |
801 | } | |
802 | ||
803 | down_write(&fs_info->commit_root_sem); | |
804 | refcount_inc(&caching_ctl->count); | |
805 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); | |
806 | up_write(&fs_info->commit_root_sem); | |
807 | ||
808 | btrfs_get_block_group(cache); | |
809 | ||
810 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); | |
811 | ||
812 | return ret; | |
813 | } | |
e3e0520b JB |
814 | |
815 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
816 | { | |
817 | u64 extra_flags = chunk_to_extended(flags) & | |
818 | BTRFS_EXTENDED_PROFILE_MASK; | |
819 | ||
820 | write_seqlock(&fs_info->profiles_lock); | |
821 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
822 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
823 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
824 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
825 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
826 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
827 | write_sequnlock(&fs_info->profiles_lock); | |
828 | } | |
829 | ||
830 | /* | |
831 | * Clear incompat bits for the following feature(s): | |
832 | * | |
833 | * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group | |
834 | * in the whole filesystem | |
835 | */ | |
836 | static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
837 | { | |
838 | if (flags & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
839 | struct list_head *head = &fs_info->space_info; | |
840 | struct btrfs_space_info *sinfo; | |
841 | ||
842 | list_for_each_entry_rcu(sinfo, head, list) { | |
843 | bool found = false; | |
844 | ||
845 | down_read(&sinfo->groups_sem); | |
846 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5])) | |
847 | found = true; | |
848 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6])) | |
849 | found = true; | |
850 | up_read(&sinfo->groups_sem); | |
851 | ||
852 | if (found) | |
853 | return; | |
854 | } | |
855 | btrfs_clear_fs_incompat(fs_info, RAID56); | |
856 | } | |
857 | } | |
858 | ||
859 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, | |
860 | u64 group_start, struct extent_map *em) | |
861 | { | |
862 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
863 | struct btrfs_root *root = fs_info->extent_root; | |
864 | struct btrfs_path *path; | |
865 | struct btrfs_block_group_cache *block_group; | |
866 | struct btrfs_free_cluster *cluster; | |
867 | struct btrfs_root *tree_root = fs_info->tree_root; | |
868 | struct btrfs_key key; | |
869 | struct inode *inode; | |
870 | struct kobject *kobj = NULL; | |
871 | int ret; | |
872 | int index; | |
873 | int factor; | |
874 | struct btrfs_caching_control *caching_ctl = NULL; | |
875 | bool remove_em; | |
876 | bool remove_rsv = false; | |
877 | ||
878 | block_group = btrfs_lookup_block_group(fs_info, group_start); | |
879 | BUG_ON(!block_group); | |
880 | BUG_ON(!block_group->ro); | |
881 | ||
882 | trace_btrfs_remove_block_group(block_group); | |
883 | /* | |
884 | * Free the reserved super bytes from this block group before | |
885 | * remove it. | |
886 | */ | |
887 | btrfs_free_excluded_extents(block_group); | |
888 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, | |
889 | block_group->key.offset); | |
890 | ||
891 | memcpy(&key, &block_group->key, sizeof(key)); | |
892 | index = btrfs_bg_flags_to_raid_index(block_group->flags); | |
893 | factor = btrfs_bg_type_to_factor(block_group->flags); | |
894 | ||
895 | /* make sure this block group isn't part of an allocation cluster */ | |
896 | cluster = &fs_info->data_alloc_cluster; | |
897 | spin_lock(&cluster->refill_lock); | |
898 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
899 | spin_unlock(&cluster->refill_lock); | |
900 | ||
901 | /* | |
902 | * make sure this block group isn't part of a metadata | |
903 | * allocation cluster | |
904 | */ | |
905 | cluster = &fs_info->meta_alloc_cluster; | |
906 | spin_lock(&cluster->refill_lock); | |
907 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
908 | spin_unlock(&cluster->refill_lock); | |
909 | ||
910 | path = btrfs_alloc_path(); | |
911 | if (!path) { | |
912 | ret = -ENOMEM; | |
913 | goto out; | |
914 | } | |
915 | ||
916 | /* | |
917 | * get the inode first so any iput calls done for the io_list | |
918 | * aren't the final iput (no unlinks allowed now) | |
919 | */ | |
920 | inode = lookup_free_space_inode(block_group, path); | |
921 | ||
922 | mutex_lock(&trans->transaction->cache_write_mutex); | |
923 | /* | |
924 | * Make sure our free space cache IO is done before removing the | |
925 | * free space inode | |
926 | */ | |
927 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
928 | if (!list_empty(&block_group->io_list)) { | |
929 | list_del_init(&block_group->io_list); | |
930 | ||
931 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
932 | ||
933 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
934 | btrfs_wait_cache_io(trans, block_group, path); | |
935 | btrfs_put_block_group(block_group); | |
936 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
937 | } | |
938 | ||
939 | if (!list_empty(&block_group->dirty_list)) { | |
940 | list_del_init(&block_group->dirty_list); | |
941 | remove_rsv = true; | |
942 | btrfs_put_block_group(block_group); | |
943 | } | |
944 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
945 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
946 | ||
947 | if (!IS_ERR(inode)) { | |
948 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); | |
949 | if (ret) { | |
950 | btrfs_add_delayed_iput(inode); | |
951 | goto out; | |
952 | } | |
953 | clear_nlink(inode); | |
954 | /* One for the block groups ref */ | |
955 | spin_lock(&block_group->lock); | |
956 | if (block_group->iref) { | |
957 | block_group->iref = 0; | |
958 | block_group->inode = NULL; | |
959 | spin_unlock(&block_group->lock); | |
960 | iput(inode); | |
961 | } else { | |
962 | spin_unlock(&block_group->lock); | |
963 | } | |
964 | /* One for our lookup ref */ | |
965 | btrfs_add_delayed_iput(inode); | |
966 | } | |
967 | ||
968 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
969 | key.offset = block_group->key.objectid; | |
970 | key.type = 0; | |
971 | ||
972 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
973 | if (ret < 0) | |
974 | goto out; | |
975 | if (ret > 0) | |
976 | btrfs_release_path(path); | |
977 | if (ret == 0) { | |
978 | ret = btrfs_del_item(trans, tree_root, path); | |
979 | if (ret) | |
980 | goto out; | |
981 | btrfs_release_path(path); | |
982 | } | |
983 | ||
984 | spin_lock(&fs_info->block_group_cache_lock); | |
985 | rb_erase(&block_group->cache_node, | |
986 | &fs_info->block_group_cache_tree); | |
987 | RB_CLEAR_NODE(&block_group->cache_node); | |
988 | ||
989 | if (fs_info->first_logical_byte == block_group->key.objectid) | |
990 | fs_info->first_logical_byte = (u64)-1; | |
991 | spin_unlock(&fs_info->block_group_cache_lock); | |
992 | ||
993 | down_write(&block_group->space_info->groups_sem); | |
994 | /* | |
995 | * we must use list_del_init so people can check to see if they | |
996 | * are still on the list after taking the semaphore | |
997 | */ | |
998 | list_del_init(&block_group->list); | |
999 | if (list_empty(&block_group->space_info->block_groups[index])) { | |
1000 | kobj = block_group->space_info->block_group_kobjs[index]; | |
1001 | block_group->space_info->block_group_kobjs[index] = NULL; | |
1002 | clear_avail_alloc_bits(fs_info, block_group->flags); | |
1003 | } | |
1004 | up_write(&block_group->space_info->groups_sem); | |
1005 | clear_incompat_bg_bits(fs_info, block_group->flags); | |
1006 | if (kobj) { | |
1007 | kobject_del(kobj); | |
1008 | kobject_put(kobj); | |
1009 | } | |
1010 | ||
1011 | if (block_group->has_caching_ctl) | |
1012 | caching_ctl = btrfs_get_caching_control(block_group); | |
1013 | if (block_group->cached == BTRFS_CACHE_STARTED) | |
1014 | btrfs_wait_block_group_cache_done(block_group); | |
1015 | if (block_group->has_caching_ctl) { | |
1016 | down_write(&fs_info->commit_root_sem); | |
1017 | if (!caching_ctl) { | |
1018 | struct btrfs_caching_control *ctl; | |
1019 | ||
1020 | list_for_each_entry(ctl, | |
1021 | &fs_info->caching_block_groups, list) | |
1022 | if (ctl->block_group == block_group) { | |
1023 | caching_ctl = ctl; | |
1024 | refcount_inc(&caching_ctl->count); | |
1025 | break; | |
1026 | } | |
1027 | } | |
1028 | if (caching_ctl) | |
1029 | list_del_init(&caching_ctl->list); | |
1030 | up_write(&fs_info->commit_root_sem); | |
1031 | if (caching_ctl) { | |
1032 | /* Once for the caching bgs list and once for us. */ | |
1033 | btrfs_put_caching_control(caching_ctl); | |
1034 | btrfs_put_caching_control(caching_ctl); | |
1035 | } | |
1036 | } | |
1037 | ||
1038 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
1039 | WARN_ON(!list_empty(&block_group->dirty_list)); | |
1040 | WARN_ON(!list_empty(&block_group->io_list)); | |
1041 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
1042 | ||
1043 | btrfs_remove_free_space_cache(block_group); | |
1044 | ||
1045 | spin_lock(&block_group->space_info->lock); | |
1046 | list_del_init(&block_group->ro_list); | |
1047 | ||
1048 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { | |
1049 | WARN_ON(block_group->space_info->total_bytes | |
1050 | < block_group->key.offset); | |
1051 | WARN_ON(block_group->space_info->bytes_readonly | |
1052 | < block_group->key.offset); | |
1053 | WARN_ON(block_group->space_info->disk_total | |
1054 | < block_group->key.offset * factor); | |
1055 | } | |
1056 | block_group->space_info->total_bytes -= block_group->key.offset; | |
1057 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
1058 | block_group->space_info->disk_total -= block_group->key.offset * factor; | |
1059 | ||
1060 | spin_unlock(&block_group->space_info->lock); | |
1061 | ||
1062 | memcpy(&key, &block_group->key, sizeof(key)); | |
1063 | ||
1064 | mutex_lock(&fs_info->chunk_mutex); | |
1065 | spin_lock(&block_group->lock); | |
1066 | block_group->removed = 1; | |
1067 | /* | |
1068 | * At this point trimming can't start on this block group, because we | |
1069 | * removed the block group from the tree fs_info->block_group_cache_tree | |
1070 | * so no one can't find it anymore and even if someone already got this | |
1071 | * block group before we removed it from the rbtree, they have already | |
1072 | * incremented block_group->trimming - if they didn't, they won't find | |
1073 | * any free space entries because we already removed them all when we | |
1074 | * called btrfs_remove_free_space_cache(). | |
1075 | * | |
1076 | * And we must not remove the extent map from the fs_info->mapping_tree | |
1077 | * to prevent the same logical address range and physical device space | |
1078 | * ranges from being reused for a new block group. This is because our | |
1079 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
1080 | * completely transactionless, so while it is trimming a range the | |
1081 | * currently running transaction might finish and a new one start, | |
1082 | * allowing for new block groups to be created that can reuse the same | |
1083 | * physical device locations unless we take this special care. | |
1084 | * | |
1085 | * There may also be an implicit trim operation if the file system | |
1086 | * is mounted with -odiscard. The same protections must remain | |
1087 | * in place until the extents have been discarded completely when | |
1088 | * the transaction commit has completed. | |
1089 | */ | |
1090 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
1091 | spin_unlock(&block_group->lock); | |
1092 | ||
1093 | mutex_unlock(&fs_info->chunk_mutex); | |
1094 | ||
1095 | ret = remove_block_group_free_space(trans, block_group); | |
1096 | if (ret) | |
1097 | goto out; | |
1098 | ||
1099 | btrfs_put_block_group(block_group); | |
1100 | btrfs_put_block_group(block_group); | |
1101 | ||
1102 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1103 | if (ret > 0) | |
1104 | ret = -EIO; | |
1105 | if (ret < 0) | |
1106 | goto out; | |
1107 | ||
1108 | ret = btrfs_del_item(trans, root, path); | |
1109 | if (ret) | |
1110 | goto out; | |
1111 | ||
1112 | if (remove_em) { | |
1113 | struct extent_map_tree *em_tree; | |
1114 | ||
1115 | em_tree = &fs_info->mapping_tree; | |
1116 | write_lock(&em_tree->lock); | |
1117 | remove_extent_mapping(em_tree, em); | |
1118 | write_unlock(&em_tree->lock); | |
1119 | /* once for the tree */ | |
1120 | free_extent_map(em); | |
1121 | } | |
1122 | out: | |
1123 | if (remove_rsv) | |
1124 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1125 | btrfs_free_path(path); | |
1126 | return ret; | |
1127 | } | |
1128 | ||
1129 | struct btrfs_trans_handle *btrfs_start_trans_remove_block_group( | |
1130 | struct btrfs_fs_info *fs_info, const u64 chunk_offset) | |
1131 | { | |
1132 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; | |
1133 | struct extent_map *em; | |
1134 | struct map_lookup *map; | |
1135 | unsigned int num_items; | |
1136 | ||
1137 | read_lock(&em_tree->lock); | |
1138 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
1139 | read_unlock(&em_tree->lock); | |
1140 | ASSERT(em && em->start == chunk_offset); | |
1141 | ||
1142 | /* | |
1143 | * We need to reserve 3 + N units from the metadata space info in order | |
1144 | * to remove a block group (done at btrfs_remove_chunk() and at | |
1145 | * btrfs_remove_block_group()), which are used for: | |
1146 | * | |
1147 | * 1 unit for adding the free space inode's orphan (located in the tree | |
1148 | * of tree roots). | |
1149 | * 1 unit for deleting the block group item (located in the extent | |
1150 | * tree). | |
1151 | * 1 unit for deleting the free space item (located in tree of tree | |
1152 | * roots). | |
1153 | * N units for deleting N device extent items corresponding to each | |
1154 | * stripe (located in the device tree). | |
1155 | * | |
1156 | * In order to remove a block group we also need to reserve units in the | |
1157 | * system space info in order to update the chunk tree (update one or | |
1158 | * more device items and remove one chunk item), but this is done at | |
1159 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
1160 | */ | |
1161 | map = em->map_lookup; | |
1162 | num_items = 3 + map->num_stripes; | |
1163 | free_extent_map(em); | |
1164 | ||
1165 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, | |
1166 | num_items, 1); | |
1167 | } | |
1168 | ||
26ce2095 JB |
1169 | /* |
1170 | * Mark block group @cache read-only, so later write won't happen to block | |
1171 | * group @cache. | |
1172 | * | |
1173 | * If @force is not set, this function will only mark the block group readonly | |
1174 | * if we have enough free space (1M) in other metadata/system block groups. | |
1175 | * If @force is not set, this function will mark the block group readonly | |
1176 | * without checking free space. | |
1177 | * | |
1178 | * NOTE: This function doesn't care if other block groups can contain all the | |
1179 | * data in this block group. That check should be done by relocation routine, | |
1180 | * not this function. | |
1181 | */ | |
e11c0406 | 1182 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
26ce2095 JB |
1183 | { |
1184 | struct btrfs_space_info *sinfo = cache->space_info; | |
1185 | u64 num_bytes; | |
1186 | u64 sinfo_used; | |
1187 | u64 min_allocable_bytes; | |
1188 | int ret = -ENOSPC; | |
1189 | ||
1190 | /* | |
1191 | * We need some metadata space and system metadata space for | |
1192 | * allocating chunks in some corner cases until we force to set | |
1193 | * it to be readonly. | |
1194 | */ | |
1195 | if ((sinfo->flags & | |
1196 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
1197 | !force) | |
1198 | min_allocable_bytes = SZ_1M; | |
1199 | else | |
1200 | min_allocable_bytes = 0; | |
1201 | ||
1202 | spin_lock(&sinfo->lock); | |
1203 | spin_lock(&cache->lock); | |
1204 | ||
1205 | if (cache->ro) { | |
1206 | cache->ro++; | |
1207 | ret = 0; | |
1208 | goto out; | |
1209 | } | |
1210 | ||
1211 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
1212 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
1213 | sinfo_used = btrfs_space_info_used(sinfo, true); | |
1214 | ||
1215 | /* | |
1216 | * sinfo_used + num_bytes should always <= sinfo->total_bytes. | |
1217 | * | |
1218 | * Here we make sure if we mark this bg RO, we still have enough | |
1219 | * free space as buffer (if min_allocable_bytes is not 0). | |
1220 | */ | |
1221 | if (sinfo_used + num_bytes + min_allocable_bytes <= | |
1222 | sinfo->total_bytes) { | |
1223 | sinfo->bytes_readonly += num_bytes; | |
1224 | cache->ro++; | |
1225 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); | |
1226 | ret = 0; | |
1227 | } | |
1228 | out: | |
1229 | spin_unlock(&cache->lock); | |
1230 | spin_unlock(&sinfo->lock); | |
1231 | if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { | |
1232 | btrfs_info(cache->fs_info, | |
1233 | "unable to make block group %llu ro", | |
1234 | cache->key.objectid); | |
1235 | btrfs_info(cache->fs_info, | |
1236 | "sinfo_used=%llu bg_num_bytes=%llu min_allocable=%llu", | |
1237 | sinfo_used, num_bytes, min_allocable_bytes); | |
1238 | btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0); | |
1239 | } | |
1240 | return ret; | |
1241 | } | |
1242 | ||
e3e0520b JB |
1243 | /* |
1244 | * Process the unused_bgs list and remove any that don't have any allocated | |
1245 | * space inside of them. | |
1246 | */ | |
1247 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
1248 | { | |
1249 | struct btrfs_block_group_cache *block_group; | |
1250 | struct btrfs_space_info *space_info; | |
1251 | struct btrfs_trans_handle *trans; | |
1252 | int ret = 0; | |
1253 | ||
1254 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) | |
1255 | return; | |
1256 | ||
1257 | spin_lock(&fs_info->unused_bgs_lock); | |
1258 | while (!list_empty(&fs_info->unused_bgs)) { | |
1259 | u64 start, end; | |
1260 | int trimming; | |
1261 | ||
1262 | block_group = list_first_entry(&fs_info->unused_bgs, | |
1263 | struct btrfs_block_group_cache, | |
1264 | bg_list); | |
1265 | list_del_init(&block_group->bg_list); | |
1266 | ||
1267 | space_info = block_group->space_info; | |
1268 | ||
1269 | if (ret || btrfs_mixed_space_info(space_info)) { | |
1270 | btrfs_put_block_group(block_group); | |
1271 | continue; | |
1272 | } | |
1273 | spin_unlock(&fs_info->unused_bgs_lock); | |
1274 | ||
1275 | mutex_lock(&fs_info->delete_unused_bgs_mutex); | |
1276 | ||
1277 | /* Don't want to race with allocators so take the groups_sem */ | |
1278 | down_write(&space_info->groups_sem); | |
1279 | spin_lock(&block_group->lock); | |
1280 | if (block_group->reserved || block_group->pinned || | |
1281 | btrfs_block_group_used(&block_group->item) || | |
1282 | block_group->ro || | |
1283 | list_is_singular(&block_group->list)) { | |
1284 | /* | |
1285 | * We want to bail if we made new allocations or have | |
1286 | * outstanding allocations in this block group. We do | |
1287 | * the ro check in case balance is currently acting on | |
1288 | * this block group. | |
1289 | */ | |
1290 | trace_btrfs_skip_unused_block_group(block_group); | |
1291 | spin_unlock(&block_group->lock); | |
1292 | up_write(&space_info->groups_sem); | |
1293 | goto next; | |
1294 | } | |
1295 | spin_unlock(&block_group->lock); | |
1296 | ||
1297 | /* We don't want to force the issue, only flip if it's ok. */ | |
e11c0406 | 1298 | ret = inc_block_group_ro(block_group, 0); |
e3e0520b JB |
1299 | up_write(&space_info->groups_sem); |
1300 | if (ret < 0) { | |
1301 | ret = 0; | |
1302 | goto next; | |
1303 | } | |
1304 | ||
1305 | /* | |
1306 | * Want to do this before we do anything else so we can recover | |
1307 | * properly if we fail to join the transaction. | |
1308 | */ | |
1309 | trans = btrfs_start_trans_remove_block_group(fs_info, | |
1310 | block_group->key.objectid); | |
1311 | if (IS_ERR(trans)) { | |
1312 | btrfs_dec_block_group_ro(block_group); | |
1313 | ret = PTR_ERR(trans); | |
1314 | goto next; | |
1315 | } | |
1316 | ||
1317 | /* | |
1318 | * We could have pending pinned extents for this block group, | |
1319 | * just delete them, we don't care about them anymore. | |
1320 | */ | |
1321 | start = block_group->key.objectid; | |
1322 | end = start + block_group->key.offset - 1; | |
1323 | /* | |
1324 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
1325 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
1326 | * another task might be running finish_extent_commit() for the | |
1327 | * previous transaction N - 1, and have seen a range belonging | |
1328 | * to the block group in freed_extents[] before we were able to | |
1329 | * clear the whole block group range from freed_extents[]. This | |
1330 | * means that task can lookup for the block group after we | |
1331 | * unpinned it from freed_extents[] and removed it, leading to | |
1332 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
1333 | */ | |
1334 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
1335 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, | |
1336 | EXTENT_DIRTY); | |
1337 | if (ret) { | |
1338 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
1339 | btrfs_dec_block_group_ro(block_group); | |
1340 | goto end_trans; | |
1341 | } | |
1342 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
1343 | EXTENT_DIRTY); | |
1344 | if (ret) { | |
1345 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
1346 | btrfs_dec_block_group_ro(block_group); | |
1347 | goto end_trans; | |
1348 | } | |
1349 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
1350 | ||
1351 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
1352 | spin_lock(&space_info->lock); | |
1353 | spin_lock(&block_group->lock); | |
1354 | ||
1355 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, | |
1356 | -block_group->pinned); | |
1357 | space_info->bytes_readonly += block_group->pinned; | |
1358 | percpu_counter_add_batch(&space_info->total_bytes_pinned, | |
1359 | -block_group->pinned, | |
1360 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
1361 | block_group->pinned = 0; | |
1362 | ||
1363 | spin_unlock(&block_group->lock); | |
1364 | spin_unlock(&space_info->lock); | |
1365 | ||
1366 | /* DISCARD can flip during remount */ | |
1367 | trimming = btrfs_test_opt(fs_info, DISCARD); | |
1368 | ||
1369 | /* Implicit trim during transaction commit. */ | |
1370 | if (trimming) | |
1371 | btrfs_get_block_group_trimming(block_group); | |
1372 | ||
1373 | /* | |
1374 | * Btrfs_remove_chunk will abort the transaction if things go | |
1375 | * horribly wrong. | |
1376 | */ | |
1377 | ret = btrfs_remove_chunk(trans, block_group->key.objectid); | |
1378 | ||
1379 | if (ret) { | |
1380 | if (trimming) | |
1381 | btrfs_put_block_group_trimming(block_group); | |
1382 | goto end_trans; | |
1383 | } | |
1384 | ||
1385 | /* | |
1386 | * If we're not mounted with -odiscard, we can just forget | |
1387 | * about this block group. Otherwise we'll need to wait | |
1388 | * until transaction commit to do the actual discard. | |
1389 | */ | |
1390 | if (trimming) { | |
1391 | spin_lock(&fs_info->unused_bgs_lock); | |
1392 | /* | |
1393 | * A concurrent scrub might have added us to the list | |
1394 | * fs_info->unused_bgs, so use a list_move operation | |
1395 | * to add the block group to the deleted_bgs list. | |
1396 | */ | |
1397 | list_move(&block_group->bg_list, | |
1398 | &trans->transaction->deleted_bgs); | |
1399 | spin_unlock(&fs_info->unused_bgs_lock); | |
1400 | btrfs_get_block_group(block_group); | |
1401 | } | |
1402 | end_trans: | |
1403 | btrfs_end_transaction(trans); | |
1404 | next: | |
1405 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
1406 | btrfs_put_block_group(block_group); | |
1407 | spin_lock(&fs_info->unused_bgs_lock); | |
1408 | } | |
1409 | spin_unlock(&fs_info->unused_bgs_lock); | |
1410 | } | |
1411 | ||
1412 | void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg) | |
1413 | { | |
1414 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
1415 | ||
1416 | spin_lock(&fs_info->unused_bgs_lock); | |
1417 | if (list_empty(&bg->bg_list)) { | |
1418 | btrfs_get_block_group(bg); | |
1419 | trace_btrfs_add_unused_block_group(bg); | |
1420 | list_add_tail(&bg->bg_list, &fs_info->unused_bgs); | |
1421 | } | |
1422 | spin_unlock(&fs_info->unused_bgs_lock); | |
1423 | } | |
4358d963 JB |
1424 | |
1425 | static int find_first_block_group(struct btrfs_fs_info *fs_info, | |
1426 | struct btrfs_path *path, | |
1427 | struct btrfs_key *key) | |
1428 | { | |
1429 | struct btrfs_root *root = fs_info->extent_root; | |
1430 | int ret = 0; | |
1431 | struct btrfs_key found_key; | |
1432 | struct extent_buffer *leaf; | |
1433 | struct btrfs_block_group_item bg; | |
1434 | u64 flags; | |
1435 | int slot; | |
1436 | ||
1437 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
1438 | if (ret < 0) | |
1439 | goto out; | |
1440 | ||
1441 | while (1) { | |
1442 | slot = path->slots[0]; | |
1443 | leaf = path->nodes[0]; | |
1444 | if (slot >= btrfs_header_nritems(leaf)) { | |
1445 | ret = btrfs_next_leaf(root, path); | |
1446 | if (ret == 0) | |
1447 | continue; | |
1448 | if (ret < 0) | |
1449 | goto out; | |
1450 | break; | |
1451 | } | |
1452 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
1453 | ||
1454 | if (found_key.objectid >= key->objectid && | |
1455 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { | |
1456 | struct extent_map_tree *em_tree; | |
1457 | struct extent_map *em; | |
1458 | ||
1459 | em_tree = &root->fs_info->mapping_tree; | |
1460 | read_lock(&em_tree->lock); | |
1461 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
1462 | found_key.offset); | |
1463 | read_unlock(&em_tree->lock); | |
1464 | if (!em) { | |
1465 | btrfs_err(fs_info, | |
1466 | "logical %llu len %llu found bg but no related chunk", | |
1467 | found_key.objectid, found_key.offset); | |
1468 | ret = -ENOENT; | |
1469 | } else if (em->start != found_key.objectid || | |
1470 | em->len != found_key.offset) { | |
1471 | btrfs_err(fs_info, | |
1472 | "block group %llu len %llu mismatch with chunk %llu len %llu", | |
1473 | found_key.objectid, found_key.offset, | |
1474 | em->start, em->len); | |
1475 | ret = -EUCLEAN; | |
1476 | } else { | |
1477 | read_extent_buffer(leaf, &bg, | |
1478 | btrfs_item_ptr_offset(leaf, slot), | |
1479 | sizeof(bg)); | |
1480 | flags = btrfs_block_group_flags(&bg) & | |
1481 | BTRFS_BLOCK_GROUP_TYPE_MASK; | |
1482 | ||
1483 | if (flags != (em->map_lookup->type & | |
1484 | BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
1485 | btrfs_err(fs_info, | |
1486 | "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", | |
1487 | found_key.objectid, | |
1488 | found_key.offset, flags, | |
1489 | (BTRFS_BLOCK_GROUP_TYPE_MASK & | |
1490 | em->map_lookup->type)); | |
1491 | ret = -EUCLEAN; | |
1492 | } else { | |
1493 | ret = 0; | |
1494 | } | |
1495 | } | |
1496 | free_extent_map(em); | |
1497 | goto out; | |
1498 | } | |
1499 | path->slots[0]++; | |
1500 | } | |
1501 | out: | |
1502 | return ret; | |
1503 | } | |
1504 | ||
1505 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
1506 | { | |
1507 | u64 extra_flags = chunk_to_extended(flags) & | |
1508 | BTRFS_EXTENDED_PROFILE_MASK; | |
1509 | ||
1510 | write_seqlock(&fs_info->profiles_lock); | |
1511 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
1512 | fs_info->avail_data_alloc_bits |= extra_flags; | |
1513 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
1514 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
1515 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
1516 | fs_info->avail_system_alloc_bits |= extra_flags; | |
1517 | write_sequnlock(&fs_info->profiles_lock); | |
1518 | } | |
1519 | ||
1520 | static int exclude_super_stripes(struct btrfs_block_group_cache *cache) | |
1521 | { | |
1522 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
1523 | u64 bytenr; | |
1524 | u64 *logical; | |
1525 | int stripe_len; | |
1526 | int i, nr, ret; | |
1527 | ||
1528 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { | |
1529 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
1530 | cache->bytes_super += stripe_len; | |
1531 | ret = btrfs_add_excluded_extent(fs_info, cache->key.objectid, | |
1532 | stripe_len); | |
1533 | if (ret) | |
1534 | return ret; | |
1535 | } | |
1536 | ||
1537 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
1538 | bytenr = btrfs_sb_offset(i); | |
1539 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, | |
1540 | bytenr, &logical, &nr, &stripe_len); | |
1541 | if (ret) | |
1542 | return ret; | |
1543 | ||
1544 | while (nr--) { | |
1545 | u64 start, len; | |
1546 | ||
1547 | if (logical[nr] > cache->key.objectid + | |
1548 | cache->key.offset) | |
1549 | continue; | |
1550 | ||
1551 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
1552 | continue; | |
1553 | ||
1554 | start = logical[nr]; | |
1555 | if (start < cache->key.objectid) { | |
1556 | start = cache->key.objectid; | |
1557 | len = (logical[nr] + stripe_len) - start; | |
1558 | } else { | |
1559 | len = min_t(u64, stripe_len, | |
1560 | cache->key.objectid + | |
1561 | cache->key.offset - start); | |
1562 | } | |
1563 | ||
1564 | cache->bytes_super += len; | |
1565 | ret = btrfs_add_excluded_extent(fs_info, start, len); | |
1566 | if (ret) { | |
1567 | kfree(logical); | |
1568 | return ret; | |
1569 | } | |
1570 | } | |
1571 | ||
1572 | kfree(logical); | |
1573 | } | |
1574 | return 0; | |
1575 | } | |
1576 | ||
1577 | static void link_block_group(struct btrfs_block_group_cache *cache) | |
1578 | { | |
1579 | struct btrfs_space_info *space_info = cache->space_info; | |
1580 | int index = btrfs_bg_flags_to_raid_index(cache->flags); | |
1581 | bool first = false; | |
1582 | ||
1583 | down_write(&space_info->groups_sem); | |
1584 | if (list_empty(&space_info->block_groups[index])) | |
1585 | first = true; | |
1586 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
1587 | up_write(&space_info->groups_sem); | |
1588 | ||
1589 | if (first) | |
1590 | btrfs_sysfs_add_block_group_type(cache); | |
1591 | } | |
1592 | ||
1593 | static struct btrfs_block_group_cache *btrfs_create_block_group_cache( | |
1594 | struct btrfs_fs_info *fs_info, u64 start, u64 size) | |
1595 | { | |
1596 | struct btrfs_block_group_cache *cache; | |
1597 | ||
1598 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
1599 | if (!cache) | |
1600 | return NULL; | |
1601 | ||
1602 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
1603 | GFP_NOFS); | |
1604 | if (!cache->free_space_ctl) { | |
1605 | kfree(cache); | |
1606 | return NULL; | |
1607 | } | |
1608 | ||
1609 | cache->key.objectid = start; | |
1610 | cache->key.offset = size; | |
1611 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
1612 | ||
1613 | cache->fs_info = fs_info; | |
1614 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); | |
1615 | set_free_space_tree_thresholds(cache); | |
1616 | ||
1617 | atomic_set(&cache->count, 1); | |
1618 | spin_lock_init(&cache->lock); | |
1619 | init_rwsem(&cache->data_rwsem); | |
1620 | INIT_LIST_HEAD(&cache->list); | |
1621 | INIT_LIST_HEAD(&cache->cluster_list); | |
1622 | INIT_LIST_HEAD(&cache->bg_list); | |
1623 | INIT_LIST_HEAD(&cache->ro_list); | |
1624 | INIT_LIST_HEAD(&cache->dirty_list); | |
1625 | INIT_LIST_HEAD(&cache->io_list); | |
1626 | btrfs_init_free_space_ctl(cache); | |
1627 | atomic_set(&cache->trimming, 0); | |
1628 | mutex_init(&cache->free_space_lock); | |
1629 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); | |
1630 | ||
1631 | return cache; | |
1632 | } | |
1633 | ||
1634 | /* | |
1635 | * Iterate all chunks and verify that each of them has the corresponding block | |
1636 | * group | |
1637 | */ | |
1638 | static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) | |
1639 | { | |
1640 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; | |
1641 | struct extent_map *em; | |
1642 | struct btrfs_block_group_cache *bg; | |
1643 | u64 start = 0; | |
1644 | int ret = 0; | |
1645 | ||
1646 | while (1) { | |
1647 | read_lock(&map_tree->lock); | |
1648 | /* | |
1649 | * lookup_extent_mapping will return the first extent map | |
1650 | * intersecting the range, so setting @len to 1 is enough to | |
1651 | * get the first chunk. | |
1652 | */ | |
1653 | em = lookup_extent_mapping(map_tree, start, 1); | |
1654 | read_unlock(&map_tree->lock); | |
1655 | if (!em) | |
1656 | break; | |
1657 | ||
1658 | bg = btrfs_lookup_block_group(fs_info, em->start); | |
1659 | if (!bg) { | |
1660 | btrfs_err(fs_info, | |
1661 | "chunk start=%llu len=%llu doesn't have corresponding block group", | |
1662 | em->start, em->len); | |
1663 | ret = -EUCLEAN; | |
1664 | free_extent_map(em); | |
1665 | break; | |
1666 | } | |
1667 | if (bg->key.objectid != em->start || | |
1668 | bg->key.offset != em->len || | |
1669 | (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != | |
1670 | (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
1671 | btrfs_err(fs_info, | |
1672 | "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", | |
1673 | em->start, em->len, | |
1674 | em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, | |
1675 | bg->key.objectid, bg->key.offset, | |
1676 | bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); | |
1677 | ret = -EUCLEAN; | |
1678 | free_extent_map(em); | |
1679 | btrfs_put_block_group(bg); | |
1680 | break; | |
1681 | } | |
1682 | start = em->start + em->len; | |
1683 | free_extent_map(em); | |
1684 | btrfs_put_block_group(bg); | |
1685 | } | |
1686 | return ret; | |
1687 | } | |
1688 | ||
1689 | int btrfs_read_block_groups(struct btrfs_fs_info *info) | |
1690 | { | |
1691 | struct btrfs_path *path; | |
1692 | int ret; | |
1693 | struct btrfs_block_group_cache *cache; | |
1694 | struct btrfs_space_info *space_info; | |
1695 | struct btrfs_key key; | |
1696 | struct btrfs_key found_key; | |
1697 | struct extent_buffer *leaf; | |
1698 | int need_clear = 0; | |
1699 | u64 cache_gen; | |
1700 | u64 feature; | |
1701 | int mixed; | |
1702 | ||
1703 | feature = btrfs_super_incompat_flags(info->super_copy); | |
1704 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
1705 | ||
1706 | key.objectid = 0; | |
1707 | key.offset = 0; | |
1708 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
1709 | path = btrfs_alloc_path(); | |
1710 | if (!path) | |
1711 | return -ENOMEM; | |
1712 | path->reada = READA_FORWARD; | |
1713 | ||
1714 | cache_gen = btrfs_super_cache_generation(info->super_copy); | |
1715 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
1716 | btrfs_super_generation(info->super_copy) != cache_gen) | |
1717 | need_clear = 1; | |
1718 | if (btrfs_test_opt(info, CLEAR_CACHE)) | |
1719 | need_clear = 1; | |
1720 | ||
1721 | while (1) { | |
1722 | ret = find_first_block_group(info, path, &key); | |
1723 | if (ret > 0) | |
1724 | break; | |
1725 | if (ret != 0) | |
1726 | goto error; | |
1727 | ||
1728 | leaf = path->nodes[0]; | |
1729 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1730 | ||
1731 | cache = btrfs_create_block_group_cache(info, found_key.objectid, | |
1732 | found_key.offset); | |
1733 | if (!cache) { | |
1734 | ret = -ENOMEM; | |
1735 | goto error; | |
1736 | } | |
1737 | ||
1738 | if (need_clear) { | |
1739 | /* | |
1740 | * When we mount with old space cache, we need to | |
1741 | * set BTRFS_DC_CLEAR and set dirty flag. | |
1742 | * | |
1743 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
1744 | * truncate the old free space cache inode and | |
1745 | * setup a new one. | |
1746 | * b) Setting 'dirty flag' makes sure that we flush | |
1747 | * the new space cache info onto disk. | |
1748 | */ | |
1749 | if (btrfs_test_opt(info, SPACE_CACHE)) | |
1750 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
1751 | } | |
1752 | ||
1753 | read_extent_buffer(leaf, &cache->item, | |
1754 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
1755 | sizeof(cache->item)); | |
1756 | cache->flags = btrfs_block_group_flags(&cache->item); | |
1757 | if (!mixed && | |
1758 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
1759 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
1760 | btrfs_err(info, | |
1761 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
1762 | cache->key.objectid); | |
4b654acd | 1763 | btrfs_put_block_group(cache); |
4358d963 JB |
1764 | ret = -EINVAL; |
1765 | goto error; | |
1766 | } | |
1767 | ||
1768 | key.objectid = found_key.objectid + found_key.offset; | |
1769 | btrfs_release_path(path); | |
1770 | ||
1771 | /* | |
1772 | * We need to exclude the super stripes now so that the space | |
1773 | * info has super bytes accounted for, otherwise we'll think | |
1774 | * we have more space than we actually do. | |
1775 | */ | |
1776 | ret = exclude_super_stripes(cache); | |
1777 | if (ret) { | |
1778 | /* | |
1779 | * We may have excluded something, so call this just in | |
1780 | * case. | |
1781 | */ | |
1782 | btrfs_free_excluded_extents(cache); | |
1783 | btrfs_put_block_group(cache); | |
1784 | goto error; | |
1785 | } | |
1786 | ||
1787 | /* | |
1788 | * Check for two cases, either we are full, and therefore | |
1789 | * don't need to bother with the caching work since we won't | |
1790 | * find any space, or we are empty, and we can just add all | |
1791 | * the space in and be done with it. This saves us _a_lot_ of | |
1792 | * time, particularly in the full case. | |
1793 | */ | |
1794 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
1795 | cache->last_byte_to_unpin = (u64)-1; | |
1796 | cache->cached = BTRFS_CACHE_FINISHED; | |
1797 | btrfs_free_excluded_extents(cache); | |
1798 | } else if (btrfs_block_group_used(&cache->item) == 0) { | |
1799 | cache->last_byte_to_unpin = (u64)-1; | |
1800 | cache->cached = BTRFS_CACHE_FINISHED; | |
1801 | add_new_free_space(cache, found_key.objectid, | |
1802 | found_key.objectid + | |
1803 | found_key.offset); | |
1804 | btrfs_free_excluded_extents(cache); | |
1805 | } | |
1806 | ||
1807 | ret = btrfs_add_block_group_cache(info, cache); | |
1808 | if (ret) { | |
1809 | btrfs_remove_free_space_cache(cache); | |
1810 | btrfs_put_block_group(cache); | |
1811 | goto error; | |
1812 | } | |
1813 | ||
1814 | trace_btrfs_add_block_group(info, cache, 0); | |
1815 | btrfs_update_space_info(info, cache->flags, found_key.offset, | |
1816 | btrfs_block_group_used(&cache->item), | |
1817 | cache->bytes_super, &space_info); | |
1818 | ||
1819 | cache->space_info = space_info; | |
1820 | ||
1821 | link_block_group(cache); | |
1822 | ||
1823 | set_avail_alloc_bits(info, cache->flags); | |
1824 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { | |
e11c0406 | 1825 | inc_block_group_ro(cache, 1); |
4358d963 JB |
1826 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
1827 | ASSERT(list_empty(&cache->bg_list)); | |
1828 | btrfs_mark_bg_unused(cache); | |
1829 | } | |
1830 | } | |
1831 | ||
1832 | list_for_each_entry_rcu(space_info, &info->space_info, list) { | |
1833 | if (!(btrfs_get_alloc_profile(info, space_info->flags) & | |
1834 | (BTRFS_BLOCK_GROUP_RAID10 | | |
1835 | BTRFS_BLOCK_GROUP_RAID1_MASK | | |
1836 | BTRFS_BLOCK_GROUP_RAID56_MASK | | |
1837 | BTRFS_BLOCK_GROUP_DUP))) | |
1838 | continue; | |
1839 | /* | |
1840 | * Avoid allocating from un-mirrored block group if there are | |
1841 | * mirrored block groups. | |
1842 | */ | |
1843 | list_for_each_entry(cache, | |
1844 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
1845 | list) | |
e11c0406 | 1846 | inc_block_group_ro(cache, 1); |
4358d963 JB |
1847 | list_for_each_entry(cache, |
1848 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
1849 | list) | |
e11c0406 | 1850 | inc_block_group_ro(cache, 1); |
4358d963 JB |
1851 | } |
1852 | ||
1853 | btrfs_init_global_block_rsv(info); | |
1854 | ret = check_chunk_block_group_mappings(info); | |
1855 | error: | |
1856 | btrfs_free_path(path); | |
1857 | return ret; | |
1858 | } | |
1859 | ||
1860 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) | |
1861 | { | |
1862 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1863 | struct btrfs_block_group_cache *block_group; | |
1864 | struct btrfs_root *extent_root = fs_info->extent_root; | |
1865 | struct btrfs_block_group_item item; | |
1866 | struct btrfs_key key; | |
1867 | int ret = 0; | |
1868 | ||
1869 | if (!trans->can_flush_pending_bgs) | |
1870 | return; | |
1871 | ||
1872 | while (!list_empty(&trans->new_bgs)) { | |
1873 | block_group = list_first_entry(&trans->new_bgs, | |
1874 | struct btrfs_block_group_cache, | |
1875 | bg_list); | |
1876 | if (ret) | |
1877 | goto next; | |
1878 | ||
1879 | spin_lock(&block_group->lock); | |
1880 | memcpy(&item, &block_group->item, sizeof(item)); | |
1881 | memcpy(&key, &block_group->key, sizeof(key)); | |
1882 | spin_unlock(&block_group->lock); | |
1883 | ||
1884 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
1885 | sizeof(item)); | |
1886 | if (ret) | |
1887 | btrfs_abort_transaction(trans, ret); | |
1888 | ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset); | |
1889 | if (ret) | |
1890 | btrfs_abort_transaction(trans, ret); | |
1891 | add_block_group_free_space(trans, block_group); | |
1892 | /* Already aborted the transaction if it failed. */ | |
1893 | next: | |
1894 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1895 | list_del_init(&block_group->bg_list); | |
1896 | } | |
1897 | btrfs_trans_release_chunk_metadata(trans); | |
1898 | } | |
1899 | ||
1900 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used, | |
1901 | u64 type, u64 chunk_offset, u64 size) | |
1902 | { | |
1903 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1904 | struct btrfs_block_group_cache *cache; | |
1905 | int ret; | |
1906 | ||
1907 | btrfs_set_log_full_commit(trans); | |
1908 | ||
1909 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); | |
1910 | if (!cache) | |
1911 | return -ENOMEM; | |
1912 | ||
1913 | btrfs_set_block_group_used(&cache->item, bytes_used); | |
1914 | btrfs_set_block_group_chunk_objectid(&cache->item, | |
1915 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
1916 | btrfs_set_block_group_flags(&cache->item, type); | |
1917 | ||
1918 | cache->flags = type; | |
1919 | cache->last_byte_to_unpin = (u64)-1; | |
1920 | cache->cached = BTRFS_CACHE_FINISHED; | |
1921 | cache->needs_free_space = 1; | |
1922 | ret = exclude_super_stripes(cache); | |
1923 | if (ret) { | |
1924 | /* We may have excluded something, so call this just in case */ | |
1925 | btrfs_free_excluded_extents(cache); | |
1926 | btrfs_put_block_group(cache); | |
1927 | return ret; | |
1928 | } | |
1929 | ||
1930 | add_new_free_space(cache, chunk_offset, chunk_offset + size); | |
1931 | ||
1932 | btrfs_free_excluded_extents(cache); | |
1933 | ||
1934 | #ifdef CONFIG_BTRFS_DEBUG | |
1935 | if (btrfs_should_fragment_free_space(cache)) { | |
1936 | u64 new_bytes_used = size - bytes_used; | |
1937 | ||
1938 | bytes_used += new_bytes_used >> 1; | |
e11c0406 | 1939 | fragment_free_space(cache); |
4358d963 JB |
1940 | } |
1941 | #endif | |
1942 | /* | |
1943 | * Ensure the corresponding space_info object is created and | |
1944 | * assigned to our block group. We want our bg to be added to the rbtree | |
1945 | * with its ->space_info set. | |
1946 | */ | |
1947 | cache->space_info = btrfs_find_space_info(fs_info, cache->flags); | |
1948 | ASSERT(cache->space_info); | |
1949 | ||
1950 | ret = btrfs_add_block_group_cache(fs_info, cache); | |
1951 | if (ret) { | |
1952 | btrfs_remove_free_space_cache(cache); | |
1953 | btrfs_put_block_group(cache); | |
1954 | return ret; | |
1955 | } | |
1956 | ||
1957 | /* | |
1958 | * Now that our block group has its ->space_info set and is inserted in | |
1959 | * the rbtree, update the space info's counters. | |
1960 | */ | |
1961 | trace_btrfs_add_block_group(fs_info, cache, 1); | |
1962 | btrfs_update_space_info(fs_info, cache->flags, size, bytes_used, | |
1963 | cache->bytes_super, &cache->space_info); | |
1964 | btrfs_update_global_block_rsv(fs_info); | |
1965 | ||
1966 | link_block_group(cache); | |
1967 | ||
1968 | list_add_tail(&cache->bg_list, &trans->new_bgs); | |
1969 | trans->delayed_ref_updates++; | |
1970 | btrfs_update_delayed_refs_rsv(trans); | |
1971 | ||
1972 | set_avail_alloc_bits(fs_info, type); | |
1973 | return 0; | |
1974 | } | |
26ce2095 JB |
1975 | |
1976 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) | |
1977 | { | |
1978 | u64 num_devices; | |
1979 | u64 stripped; | |
1980 | ||
1981 | /* | |
1982 | * if restripe for this chunk_type is on pick target profile and | |
1983 | * return, otherwise do the usual balance | |
1984 | */ | |
e11c0406 | 1985 | stripped = get_restripe_target(fs_info, flags); |
26ce2095 JB |
1986 | if (stripped) |
1987 | return extended_to_chunk(stripped); | |
1988 | ||
1989 | num_devices = fs_info->fs_devices->rw_devices; | |
1990 | ||
1991 | stripped = BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID56_MASK | | |
1992 | BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10; | |
1993 | ||
1994 | if (num_devices == 1) { | |
1995 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
1996 | stripped = flags & ~stripped; | |
1997 | ||
1998 | /* turn raid0 into single device chunks */ | |
1999 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
2000 | return stripped; | |
2001 | ||
2002 | /* turn mirroring into duplication */ | |
2003 | if (flags & (BTRFS_BLOCK_GROUP_RAID1_MASK | | |
2004 | BTRFS_BLOCK_GROUP_RAID10)) | |
2005 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
2006 | } else { | |
2007 | /* they already had raid on here, just return */ | |
2008 | if (flags & stripped) | |
2009 | return flags; | |
2010 | ||
2011 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
2012 | stripped = flags & ~stripped; | |
2013 | ||
2014 | /* switch duplicated blocks with raid1 */ | |
2015 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
2016 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
2017 | ||
2018 | /* this is drive concat, leave it alone */ | |
2019 | } | |
2020 | ||
2021 | return flags; | |
2022 | } | |
2023 | ||
2024 | int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache) | |
2025 | ||
2026 | { | |
2027 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
2028 | struct btrfs_trans_handle *trans; | |
2029 | u64 alloc_flags; | |
2030 | int ret; | |
2031 | ||
2032 | again: | |
2033 | trans = btrfs_join_transaction(fs_info->extent_root); | |
2034 | if (IS_ERR(trans)) | |
2035 | return PTR_ERR(trans); | |
2036 | ||
2037 | /* | |
2038 | * we're not allowed to set block groups readonly after the dirty | |
2039 | * block groups cache has started writing. If it already started, | |
2040 | * back off and let this transaction commit | |
2041 | */ | |
2042 | mutex_lock(&fs_info->ro_block_group_mutex); | |
2043 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { | |
2044 | u64 transid = trans->transid; | |
2045 | ||
2046 | mutex_unlock(&fs_info->ro_block_group_mutex); | |
2047 | btrfs_end_transaction(trans); | |
2048 | ||
2049 | ret = btrfs_wait_for_commit(fs_info, transid); | |
2050 | if (ret) | |
2051 | return ret; | |
2052 | goto again; | |
2053 | } | |
2054 | ||
2055 | /* | |
2056 | * if we are changing raid levels, try to allocate a corresponding | |
2057 | * block group with the new raid level. | |
2058 | */ | |
2059 | alloc_flags = update_block_group_flags(fs_info, cache->flags); | |
2060 | if (alloc_flags != cache->flags) { | |
2061 | ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); | |
2062 | /* | |
2063 | * ENOSPC is allowed here, we may have enough space | |
2064 | * already allocated at the new raid level to | |
2065 | * carry on | |
2066 | */ | |
2067 | if (ret == -ENOSPC) | |
2068 | ret = 0; | |
2069 | if (ret < 0) | |
2070 | goto out; | |
2071 | } | |
2072 | ||
e11c0406 | 2073 | ret = inc_block_group_ro(cache, 0); |
26ce2095 JB |
2074 | if (!ret) |
2075 | goto out; | |
2076 | alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags); | |
2077 | ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); | |
2078 | if (ret < 0) | |
2079 | goto out; | |
e11c0406 | 2080 | ret = inc_block_group_ro(cache, 0); |
26ce2095 JB |
2081 | out: |
2082 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { | |
2083 | alloc_flags = update_block_group_flags(fs_info, cache->flags); | |
2084 | mutex_lock(&fs_info->chunk_mutex); | |
2085 | check_system_chunk(trans, alloc_flags); | |
2086 | mutex_unlock(&fs_info->chunk_mutex); | |
2087 | } | |
2088 | mutex_unlock(&fs_info->ro_block_group_mutex); | |
2089 | ||
2090 | btrfs_end_transaction(trans); | |
2091 | return ret; | |
2092 | } | |
2093 | ||
2094 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) | |
2095 | { | |
2096 | struct btrfs_space_info *sinfo = cache->space_info; | |
2097 | u64 num_bytes; | |
2098 | ||
2099 | BUG_ON(!cache->ro); | |
2100 | ||
2101 | spin_lock(&sinfo->lock); | |
2102 | spin_lock(&cache->lock); | |
2103 | if (!--cache->ro) { | |
2104 | num_bytes = cache->key.offset - cache->reserved - | |
2105 | cache->pinned - cache->bytes_super - | |
2106 | btrfs_block_group_used(&cache->item); | |
2107 | sinfo->bytes_readonly -= num_bytes; | |
2108 | list_del_init(&cache->ro_list); | |
2109 | } | |
2110 | spin_unlock(&cache->lock); | |
2111 | spin_unlock(&sinfo->lock); | |
2112 | } | |
77745c05 JB |
2113 | |
2114 | static int write_one_cache_group(struct btrfs_trans_handle *trans, | |
2115 | struct btrfs_path *path, | |
2116 | struct btrfs_block_group_cache *cache) | |
2117 | { | |
2118 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2119 | int ret; | |
2120 | struct btrfs_root *extent_root = fs_info->extent_root; | |
2121 | unsigned long bi; | |
2122 | struct extent_buffer *leaf; | |
2123 | ||
2124 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); | |
2125 | if (ret) { | |
2126 | if (ret > 0) | |
2127 | ret = -ENOENT; | |
2128 | goto fail; | |
2129 | } | |
2130 | ||
2131 | leaf = path->nodes[0]; | |
2132 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2133 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2134 | btrfs_mark_buffer_dirty(leaf); | |
2135 | fail: | |
2136 | btrfs_release_path(path); | |
2137 | return ret; | |
2138 | ||
2139 | } | |
2140 | ||
2141 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, | |
2142 | struct btrfs_trans_handle *trans, | |
2143 | struct btrfs_path *path) | |
2144 | { | |
2145 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
2146 | struct btrfs_root *root = fs_info->tree_root; | |
2147 | struct inode *inode = NULL; | |
2148 | struct extent_changeset *data_reserved = NULL; | |
2149 | u64 alloc_hint = 0; | |
2150 | int dcs = BTRFS_DC_ERROR; | |
2151 | u64 num_pages = 0; | |
2152 | int retries = 0; | |
2153 | int ret = 0; | |
2154 | ||
2155 | /* | |
2156 | * If this block group is smaller than 100 megs don't bother caching the | |
2157 | * block group. | |
2158 | */ | |
2159 | if (block_group->key.offset < (100 * SZ_1M)) { | |
2160 | spin_lock(&block_group->lock); | |
2161 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2162 | spin_unlock(&block_group->lock); | |
2163 | return 0; | |
2164 | } | |
2165 | ||
2166 | if (trans->aborted) | |
2167 | return 0; | |
2168 | again: | |
2169 | inode = lookup_free_space_inode(block_group, path); | |
2170 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2171 | ret = PTR_ERR(inode); | |
2172 | btrfs_release_path(path); | |
2173 | goto out; | |
2174 | } | |
2175 | ||
2176 | if (IS_ERR(inode)) { | |
2177 | BUG_ON(retries); | |
2178 | retries++; | |
2179 | ||
2180 | if (block_group->ro) | |
2181 | goto out_free; | |
2182 | ||
2183 | ret = create_free_space_inode(trans, block_group, path); | |
2184 | if (ret) | |
2185 | goto out_free; | |
2186 | goto again; | |
2187 | } | |
2188 | ||
2189 | /* | |
2190 | * We want to set the generation to 0, that way if anything goes wrong | |
2191 | * from here on out we know not to trust this cache when we load up next | |
2192 | * time. | |
2193 | */ | |
2194 | BTRFS_I(inode)->generation = 0; | |
2195 | ret = btrfs_update_inode(trans, root, inode); | |
2196 | if (ret) { | |
2197 | /* | |
2198 | * So theoretically we could recover from this, simply set the | |
2199 | * super cache generation to 0 so we know to invalidate the | |
2200 | * cache, but then we'd have to keep track of the block groups | |
2201 | * that fail this way so we know we _have_ to reset this cache | |
2202 | * before the next commit or risk reading stale cache. So to | |
2203 | * limit our exposure to horrible edge cases lets just abort the | |
2204 | * transaction, this only happens in really bad situations | |
2205 | * anyway. | |
2206 | */ | |
2207 | btrfs_abort_transaction(trans, ret); | |
2208 | goto out_put; | |
2209 | } | |
2210 | WARN_ON(ret); | |
2211 | ||
2212 | /* We've already setup this transaction, go ahead and exit */ | |
2213 | if (block_group->cache_generation == trans->transid && | |
2214 | i_size_read(inode)) { | |
2215 | dcs = BTRFS_DC_SETUP; | |
2216 | goto out_put; | |
2217 | } | |
2218 | ||
2219 | if (i_size_read(inode) > 0) { | |
2220 | ret = btrfs_check_trunc_cache_free_space(fs_info, | |
2221 | &fs_info->global_block_rsv); | |
2222 | if (ret) | |
2223 | goto out_put; | |
2224 | ||
2225 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); | |
2226 | if (ret) | |
2227 | goto out_put; | |
2228 | } | |
2229 | ||
2230 | spin_lock(&block_group->lock); | |
2231 | if (block_group->cached != BTRFS_CACHE_FINISHED || | |
2232 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { | |
2233 | /* | |
2234 | * don't bother trying to write stuff out _if_ | |
2235 | * a) we're not cached, | |
2236 | * b) we're with nospace_cache mount option, | |
2237 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
2238 | */ | |
2239 | dcs = BTRFS_DC_WRITTEN; | |
2240 | spin_unlock(&block_group->lock); | |
2241 | goto out_put; | |
2242 | } | |
2243 | spin_unlock(&block_group->lock); | |
2244 | ||
2245 | /* | |
2246 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
2247 | * skip doing the setup, we've already cleared the cache so we're safe. | |
2248 | */ | |
2249 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
2250 | ret = -ENOSPC; | |
2251 | goto out_put; | |
2252 | } | |
2253 | ||
2254 | /* | |
2255 | * Try to preallocate enough space based on how big the block group is. | |
2256 | * Keep in mind this has to include any pinned space which could end up | |
2257 | * taking up quite a bit since it's not folded into the other space | |
2258 | * cache. | |
2259 | */ | |
2260 | num_pages = div_u64(block_group->key.offset, SZ_256M); | |
2261 | if (!num_pages) | |
2262 | num_pages = 1; | |
2263 | ||
2264 | num_pages *= 16; | |
2265 | num_pages *= PAGE_SIZE; | |
2266 | ||
2267 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); | |
2268 | if (ret) | |
2269 | goto out_put; | |
2270 | ||
2271 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
2272 | num_pages, num_pages, | |
2273 | &alloc_hint); | |
2274 | /* | |
2275 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
2276 | * of metadata or split extents when writing the cache out, which means | |
2277 | * we can enospc if we are heavily fragmented in addition to just normal | |
2278 | * out of space conditions. So if we hit this just skip setting up any | |
2279 | * other block groups for this transaction, maybe we'll unpin enough | |
2280 | * space the next time around. | |
2281 | */ | |
2282 | if (!ret) | |
2283 | dcs = BTRFS_DC_SETUP; | |
2284 | else if (ret == -ENOSPC) | |
2285 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
2286 | ||
2287 | out_put: | |
2288 | iput(inode); | |
2289 | out_free: | |
2290 | btrfs_release_path(path); | |
2291 | out: | |
2292 | spin_lock(&block_group->lock); | |
2293 | if (!ret && dcs == BTRFS_DC_SETUP) | |
2294 | block_group->cache_generation = trans->transid; | |
2295 | block_group->disk_cache_state = dcs; | |
2296 | spin_unlock(&block_group->lock); | |
2297 | ||
2298 | extent_changeset_free(data_reserved); | |
2299 | return ret; | |
2300 | } | |
2301 | ||
2302 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) | |
2303 | { | |
2304 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2305 | struct btrfs_block_group_cache *cache, *tmp; | |
2306 | struct btrfs_transaction *cur_trans = trans->transaction; | |
2307 | struct btrfs_path *path; | |
2308 | ||
2309 | if (list_empty(&cur_trans->dirty_bgs) || | |
2310 | !btrfs_test_opt(fs_info, SPACE_CACHE)) | |
2311 | return 0; | |
2312 | ||
2313 | path = btrfs_alloc_path(); | |
2314 | if (!path) | |
2315 | return -ENOMEM; | |
2316 | ||
2317 | /* Could add new block groups, use _safe just in case */ | |
2318 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
2319 | dirty_list) { | |
2320 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
2321 | cache_save_setup(cache, trans, path); | |
2322 | } | |
2323 | ||
2324 | btrfs_free_path(path); | |
2325 | return 0; | |
2326 | } | |
2327 | ||
2328 | /* | |
2329 | * Transaction commit does final block group cache writeback during a critical | |
2330 | * section where nothing is allowed to change the FS. This is required in | |
2331 | * order for the cache to actually match the block group, but can introduce a | |
2332 | * lot of latency into the commit. | |
2333 | * | |
2334 | * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO. | |
2335 | * There's a chance we'll have to redo some of it if the block group changes | |
2336 | * again during the commit, but it greatly reduces the commit latency by | |
2337 | * getting rid of the easy block groups while we're still allowing others to | |
2338 | * join the commit. | |
2339 | */ | |
2340 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) | |
2341 | { | |
2342 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2343 | struct btrfs_block_group_cache *cache; | |
2344 | struct btrfs_transaction *cur_trans = trans->transaction; | |
2345 | int ret = 0; | |
2346 | int should_put; | |
2347 | struct btrfs_path *path = NULL; | |
2348 | LIST_HEAD(dirty); | |
2349 | struct list_head *io = &cur_trans->io_bgs; | |
2350 | int num_started = 0; | |
2351 | int loops = 0; | |
2352 | ||
2353 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2354 | if (list_empty(&cur_trans->dirty_bgs)) { | |
2355 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2356 | return 0; | |
2357 | } | |
2358 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
2359 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2360 | ||
2361 | again: | |
2362 | /* Make sure all the block groups on our dirty list actually exist */ | |
2363 | btrfs_create_pending_block_groups(trans); | |
2364 | ||
2365 | if (!path) { | |
2366 | path = btrfs_alloc_path(); | |
2367 | if (!path) | |
2368 | return -ENOMEM; | |
2369 | } | |
2370 | ||
2371 | /* | |
2372 | * cache_write_mutex is here only to save us from balance or automatic | |
2373 | * removal of empty block groups deleting this block group while we are | |
2374 | * writing out the cache | |
2375 | */ | |
2376 | mutex_lock(&trans->transaction->cache_write_mutex); | |
2377 | while (!list_empty(&dirty)) { | |
2378 | bool drop_reserve = true; | |
2379 | ||
2380 | cache = list_first_entry(&dirty, | |
2381 | struct btrfs_block_group_cache, | |
2382 | dirty_list); | |
2383 | /* | |
2384 | * This can happen if something re-dirties a block group that | |
2385 | * is already under IO. Just wait for it to finish and then do | |
2386 | * it all again | |
2387 | */ | |
2388 | if (!list_empty(&cache->io_list)) { | |
2389 | list_del_init(&cache->io_list); | |
2390 | btrfs_wait_cache_io(trans, cache, path); | |
2391 | btrfs_put_block_group(cache); | |
2392 | } | |
2393 | ||
2394 | ||
2395 | /* | |
2396 | * btrfs_wait_cache_io uses the cache->dirty_list to decide if | |
2397 | * it should update the cache_state. Don't delete until after | |
2398 | * we wait. | |
2399 | * | |
2400 | * Since we're not running in the commit critical section | |
2401 | * we need the dirty_bgs_lock to protect from update_block_group | |
2402 | */ | |
2403 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2404 | list_del_init(&cache->dirty_list); | |
2405 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2406 | ||
2407 | should_put = 1; | |
2408 | ||
2409 | cache_save_setup(cache, trans, path); | |
2410 | ||
2411 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
2412 | cache->io_ctl.inode = NULL; | |
2413 | ret = btrfs_write_out_cache(trans, cache, path); | |
2414 | if (ret == 0 && cache->io_ctl.inode) { | |
2415 | num_started++; | |
2416 | should_put = 0; | |
2417 | ||
2418 | /* | |
2419 | * The cache_write_mutex is protecting the | |
2420 | * io_list, also refer to the definition of | |
2421 | * btrfs_transaction::io_bgs for more details | |
2422 | */ | |
2423 | list_add_tail(&cache->io_list, io); | |
2424 | } else { | |
2425 | /* | |
2426 | * If we failed to write the cache, the | |
2427 | * generation will be bad and life goes on | |
2428 | */ | |
2429 | ret = 0; | |
2430 | } | |
2431 | } | |
2432 | if (!ret) { | |
2433 | ret = write_one_cache_group(trans, path, cache); | |
2434 | /* | |
2435 | * Our block group might still be attached to the list | |
2436 | * of new block groups in the transaction handle of some | |
2437 | * other task (struct btrfs_trans_handle->new_bgs). This | |
2438 | * means its block group item isn't yet in the extent | |
2439 | * tree. If this happens ignore the error, as we will | |
2440 | * try again later in the critical section of the | |
2441 | * transaction commit. | |
2442 | */ | |
2443 | if (ret == -ENOENT) { | |
2444 | ret = 0; | |
2445 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2446 | if (list_empty(&cache->dirty_list)) { | |
2447 | list_add_tail(&cache->dirty_list, | |
2448 | &cur_trans->dirty_bgs); | |
2449 | btrfs_get_block_group(cache); | |
2450 | drop_reserve = false; | |
2451 | } | |
2452 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2453 | } else if (ret) { | |
2454 | btrfs_abort_transaction(trans, ret); | |
2455 | } | |
2456 | } | |
2457 | ||
2458 | /* If it's not on the io list, we need to put the block group */ | |
2459 | if (should_put) | |
2460 | btrfs_put_block_group(cache); | |
2461 | if (drop_reserve) | |
2462 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
2463 | ||
2464 | if (ret) | |
2465 | break; | |
2466 | ||
2467 | /* | |
2468 | * Avoid blocking other tasks for too long. It might even save | |
2469 | * us from writing caches for block groups that are going to be | |
2470 | * removed. | |
2471 | */ | |
2472 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
2473 | mutex_lock(&trans->transaction->cache_write_mutex); | |
2474 | } | |
2475 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
2476 | ||
2477 | /* | |
2478 | * Go through delayed refs for all the stuff we've just kicked off | |
2479 | * and then loop back (just once) | |
2480 | */ | |
2481 | ret = btrfs_run_delayed_refs(trans, 0); | |
2482 | if (!ret && loops == 0) { | |
2483 | loops++; | |
2484 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2485 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
2486 | /* | |
2487 | * dirty_bgs_lock protects us from concurrent block group | |
2488 | * deletes too (not just cache_write_mutex). | |
2489 | */ | |
2490 | if (!list_empty(&dirty)) { | |
2491 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2492 | goto again; | |
2493 | } | |
2494 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2495 | } else if (ret < 0) { | |
2496 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); | |
2497 | } | |
2498 | ||
2499 | btrfs_free_path(path); | |
2500 | return ret; | |
2501 | } | |
2502 | ||
2503 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) | |
2504 | { | |
2505 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2506 | struct btrfs_block_group_cache *cache; | |
2507 | struct btrfs_transaction *cur_trans = trans->transaction; | |
2508 | int ret = 0; | |
2509 | int should_put; | |
2510 | struct btrfs_path *path; | |
2511 | struct list_head *io = &cur_trans->io_bgs; | |
2512 | int num_started = 0; | |
2513 | ||
2514 | path = btrfs_alloc_path(); | |
2515 | if (!path) | |
2516 | return -ENOMEM; | |
2517 | ||
2518 | /* | |
2519 | * Even though we are in the critical section of the transaction commit, | |
2520 | * we can still have concurrent tasks adding elements to this | |
2521 | * transaction's list of dirty block groups. These tasks correspond to | |
2522 | * endio free space workers started when writeback finishes for a | |
2523 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
2524 | * allocate new block groups as a result of COWing nodes of the root | |
2525 | * tree when updating the free space inode. The writeback for the space | |
2526 | * caches is triggered by an earlier call to | |
2527 | * btrfs_start_dirty_block_groups() and iterations of the following | |
2528 | * loop. | |
2529 | * Also we want to do the cache_save_setup first and then run the | |
2530 | * delayed refs to make sure we have the best chance at doing this all | |
2531 | * in one shot. | |
2532 | */ | |
2533 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2534 | while (!list_empty(&cur_trans->dirty_bgs)) { | |
2535 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
2536 | struct btrfs_block_group_cache, | |
2537 | dirty_list); | |
2538 | ||
2539 | /* | |
2540 | * This can happen if cache_save_setup re-dirties a block group | |
2541 | * that is already under IO. Just wait for it to finish and | |
2542 | * then do it all again | |
2543 | */ | |
2544 | if (!list_empty(&cache->io_list)) { | |
2545 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2546 | list_del_init(&cache->io_list); | |
2547 | btrfs_wait_cache_io(trans, cache, path); | |
2548 | btrfs_put_block_group(cache); | |
2549 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2550 | } | |
2551 | ||
2552 | /* | |
2553 | * Don't remove from the dirty list until after we've waited on | |
2554 | * any pending IO | |
2555 | */ | |
2556 | list_del_init(&cache->dirty_list); | |
2557 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2558 | should_put = 1; | |
2559 | ||
2560 | cache_save_setup(cache, trans, path); | |
2561 | ||
2562 | if (!ret) | |
2563 | ret = btrfs_run_delayed_refs(trans, | |
2564 | (unsigned long) -1); | |
2565 | ||
2566 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
2567 | cache->io_ctl.inode = NULL; | |
2568 | ret = btrfs_write_out_cache(trans, cache, path); | |
2569 | if (ret == 0 && cache->io_ctl.inode) { | |
2570 | num_started++; | |
2571 | should_put = 0; | |
2572 | list_add_tail(&cache->io_list, io); | |
2573 | } else { | |
2574 | /* | |
2575 | * If we failed to write the cache, the | |
2576 | * generation will be bad and life goes on | |
2577 | */ | |
2578 | ret = 0; | |
2579 | } | |
2580 | } | |
2581 | if (!ret) { | |
2582 | ret = write_one_cache_group(trans, path, cache); | |
2583 | /* | |
2584 | * One of the free space endio workers might have | |
2585 | * created a new block group while updating a free space | |
2586 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
2587 | * and hasn't released its transaction handle yet, in | |
2588 | * which case the new block group is still attached to | |
2589 | * its transaction handle and its creation has not | |
2590 | * finished yet (no block group item in the extent tree | |
2591 | * yet, etc). If this is the case, wait for all free | |
2592 | * space endio workers to finish and retry. This is a | |
2593 | * a very rare case so no need for a more efficient and | |
2594 | * complex approach. | |
2595 | */ | |
2596 | if (ret == -ENOENT) { | |
2597 | wait_event(cur_trans->writer_wait, | |
2598 | atomic_read(&cur_trans->num_writers) == 1); | |
2599 | ret = write_one_cache_group(trans, path, cache); | |
2600 | } | |
2601 | if (ret) | |
2602 | btrfs_abort_transaction(trans, ret); | |
2603 | } | |
2604 | ||
2605 | /* If its not on the io list, we need to put the block group */ | |
2606 | if (should_put) | |
2607 | btrfs_put_block_group(cache); | |
2608 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
2609 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2610 | } | |
2611 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2612 | ||
2613 | /* | |
2614 | * Refer to the definition of io_bgs member for details why it's safe | |
2615 | * to use it without any locking | |
2616 | */ | |
2617 | while (!list_empty(io)) { | |
2618 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
2619 | io_list); | |
2620 | list_del_init(&cache->io_list); | |
2621 | btrfs_wait_cache_io(trans, cache, path); | |
2622 | btrfs_put_block_group(cache); | |
2623 | } | |
2624 | ||
2625 | btrfs_free_path(path); | |
2626 | return ret; | |
2627 | } | |
606d1bf1 JB |
2628 | |
2629 | int btrfs_update_block_group(struct btrfs_trans_handle *trans, | |
2630 | u64 bytenr, u64 num_bytes, int alloc) | |
2631 | { | |
2632 | struct btrfs_fs_info *info = trans->fs_info; | |
2633 | struct btrfs_block_group_cache *cache = NULL; | |
2634 | u64 total = num_bytes; | |
2635 | u64 old_val; | |
2636 | u64 byte_in_group; | |
2637 | int factor; | |
2638 | int ret = 0; | |
2639 | ||
2640 | /* Block accounting for super block */ | |
2641 | spin_lock(&info->delalloc_root_lock); | |
2642 | old_val = btrfs_super_bytes_used(info->super_copy); | |
2643 | if (alloc) | |
2644 | old_val += num_bytes; | |
2645 | else | |
2646 | old_val -= num_bytes; | |
2647 | btrfs_set_super_bytes_used(info->super_copy, old_val); | |
2648 | spin_unlock(&info->delalloc_root_lock); | |
2649 | ||
2650 | while (total) { | |
2651 | cache = btrfs_lookup_block_group(info, bytenr); | |
2652 | if (!cache) { | |
2653 | ret = -ENOENT; | |
2654 | break; | |
2655 | } | |
2656 | factor = btrfs_bg_type_to_factor(cache->flags); | |
2657 | ||
2658 | /* | |
2659 | * If this block group has free space cache written out, we | |
2660 | * need to make sure to load it if we are removing space. This | |
2661 | * is because we need the unpinning stage to actually add the | |
2662 | * space back to the block group, otherwise we will leak space. | |
2663 | */ | |
a60adce8 | 2664 | if (!alloc && !btrfs_block_group_cache_done(cache)) |
606d1bf1 JB |
2665 | btrfs_cache_block_group(cache, 1); |
2666 | ||
2667 | byte_in_group = bytenr - cache->key.objectid; | |
2668 | WARN_ON(byte_in_group > cache->key.offset); | |
2669 | ||
2670 | spin_lock(&cache->space_info->lock); | |
2671 | spin_lock(&cache->lock); | |
2672 | ||
2673 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
2674 | cache->disk_cache_state < BTRFS_DC_CLEAR) | |
2675 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
2676 | ||
2677 | old_val = btrfs_block_group_used(&cache->item); | |
2678 | num_bytes = min(total, cache->key.offset - byte_in_group); | |
2679 | if (alloc) { | |
2680 | old_val += num_bytes; | |
2681 | btrfs_set_block_group_used(&cache->item, old_val); | |
2682 | cache->reserved -= num_bytes; | |
2683 | cache->space_info->bytes_reserved -= num_bytes; | |
2684 | cache->space_info->bytes_used += num_bytes; | |
2685 | cache->space_info->disk_used += num_bytes * factor; | |
2686 | spin_unlock(&cache->lock); | |
2687 | spin_unlock(&cache->space_info->lock); | |
2688 | } else { | |
2689 | old_val -= num_bytes; | |
2690 | btrfs_set_block_group_used(&cache->item, old_val); | |
2691 | cache->pinned += num_bytes; | |
2692 | btrfs_space_info_update_bytes_pinned(info, | |
2693 | cache->space_info, num_bytes); | |
2694 | cache->space_info->bytes_used -= num_bytes; | |
2695 | cache->space_info->disk_used -= num_bytes * factor; | |
2696 | spin_unlock(&cache->lock); | |
2697 | spin_unlock(&cache->space_info->lock); | |
2698 | ||
606d1bf1 JB |
2699 | percpu_counter_add_batch( |
2700 | &cache->space_info->total_bytes_pinned, | |
2701 | num_bytes, | |
2702 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
2703 | set_extent_dirty(info->pinned_extents, | |
2704 | bytenr, bytenr + num_bytes - 1, | |
2705 | GFP_NOFS | __GFP_NOFAIL); | |
2706 | } | |
2707 | ||
2708 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
2709 | if (list_empty(&cache->dirty_list)) { | |
2710 | list_add_tail(&cache->dirty_list, | |
2711 | &trans->transaction->dirty_bgs); | |
2712 | trans->delayed_ref_updates++; | |
2713 | btrfs_get_block_group(cache); | |
2714 | } | |
2715 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
2716 | ||
2717 | /* | |
2718 | * No longer have used bytes in this block group, queue it for | |
2719 | * deletion. We do this after adding the block group to the | |
2720 | * dirty list to avoid races between cleaner kthread and space | |
2721 | * cache writeout. | |
2722 | */ | |
2723 | if (!alloc && old_val == 0) | |
2724 | btrfs_mark_bg_unused(cache); | |
2725 | ||
2726 | btrfs_put_block_group(cache); | |
2727 | total -= num_bytes; | |
2728 | bytenr += num_bytes; | |
2729 | } | |
2730 | ||
2731 | /* Modified block groups are accounted for in the delayed_refs_rsv. */ | |
2732 | btrfs_update_delayed_refs_rsv(trans); | |
2733 | return ret; | |
2734 | } | |
2735 | ||
2736 | /** | |
2737 | * btrfs_add_reserved_bytes - update the block_group and space info counters | |
2738 | * @cache: The cache we are manipulating | |
2739 | * @ram_bytes: The number of bytes of file content, and will be same to | |
2740 | * @num_bytes except for the compress path. | |
2741 | * @num_bytes: The number of bytes in question | |
2742 | * @delalloc: The blocks are allocated for the delalloc write | |
2743 | * | |
2744 | * This is called by the allocator when it reserves space. If this is a | |
2745 | * reservation and the block group has become read only we cannot make the | |
2746 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
2747 | */ | |
2748 | int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, | |
2749 | u64 ram_bytes, u64 num_bytes, int delalloc) | |
2750 | { | |
2751 | struct btrfs_space_info *space_info = cache->space_info; | |
2752 | int ret = 0; | |
2753 | ||
2754 | spin_lock(&space_info->lock); | |
2755 | spin_lock(&cache->lock); | |
2756 | if (cache->ro) { | |
2757 | ret = -EAGAIN; | |
2758 | } else { | |
2759 | cache->reserved += num_bytes; | |
2760 | space_info->bytes_reserved += num_bytes; | |
a43c3835 JB |
2761 | trace_btrfs_space_reservation(cache->fs_info, "space_info", |
2762 | space_info->flags, num_bytes, 1); | |
606d1bf1 JB |
2763 | btrfs_space_info_update_bytes_may_use(cache->fs_info, |
2764 | space_info, -ram_bytes); | |
2765 | if (delalloc) | |
2766 | cache->delalloc_bytes += num_bytes; | |
2767 | } | |
2768 | spin_unlock(&cache->lock); | |
2769 | spin_unlock(&space_info->lock); | |
2770 | return ret; | |
2771 | } | |
2772 | ||
2773 | /** | |
2774 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
2775 | * @cache: The cache we are manipulating | |
2776 | * @num_bytes: The number of bytes in question | |
2777 | * @delalloc: The blocks are allocated for the delalloc write | |
2778 | * | |
2779 | * This is called by somebody who is freeing space that was never actually used | |
2780 | * on disk. For example if you reserve some space for a new leaf in transaction | |
2781 | * A and before transaction A commits you free that leaf, you call this with | |
2782 | * reserve set to 0 in order to clear the reservation. | |
2783 | */ | |
2784 | void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, | |
2785 | u64 num_bytes, int delalloc) | |
2786 | { | |
2787 | struct btrfs_space_info *space_info = cache->space_info; | |
2788 | ||
2789 | spin_lock(&space_info->lock); | |
2790 | spin_lock(&cache->lock); | |
2791 | if (cache->ro) | |
2792 | space_info->bytes_readonly += num_bytes; | |
2793 | cache->reserved -= num_bytes; | |
2794 | space_info->bytes_reserved -= num_bytes; | |
2795 | space_info->max_extent_size = 0; | |
2796 | ||
2797 | if (delalloc) | |
2798 | cache->delalloc_bytes -= num_bytes; | |
2799 | spin_unlock(&cache->lock); | |
2800 | spin_unlock(&space_info->lock); | |
2801 | } | |
07730d87 JB |
2802 | |
2803 | static void force_metadata_allocation(struct btrfs_fs_info *info) | |
2804 | { | |
2805 | struct list_head *head = &info->space_info; | |
2806 | struct btrfs_space_info *found; | |
2807 | ||
2808 | rcu_read_lock(); | |
2809 | list_for_each_entry_rcu(found, head, list) { | |
2810 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
2811 | found->force_alloc = CHUNK_ALLOC_FORCE; | |
2812 | } | |
2813 | rcu_read_unlock(); | |
2814 | } | |
2815 | ||
2816 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, | |
2817 | struct btrfs_space_info *sinfo, int force) | |
2818 | { | |
2819 | u64 bytes_used = btrfs_space_info_used(sinfo, false); | |
2820 | u64 thresh; | |
2821 | ||
2822 | if (force == CHUNK_ALLOC_FORCE) | |
2823 | return 1; | |
2824 | ||
2825 | /* | |
2826 | * in limited mode, we want to have some free space up to | |
2827 | * about 1% of the FS size. | |
2828 | */ | |
2829 | if (force == CHUNK_ALLOC_LIMITED) { | |
2830 | thresh = btrfs_super_total_bytes(fs_info->super_copy); | |
2831 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); | |
2832 | ||
2833 | if (sinfo->total_bytes - bytes_used < thresh) | |
2834 | return 1; | |
2835 | } | |
2836 | ||
2837 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) | |
2838 | return 0; | |
2839 | return 1; | |
2840 | } | |
2841 | ||
2842 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) | |
2843 | { | |
2844 | u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type); | |
2845 | ||
2846 | return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); | |
2847 | } | |
2848 | ||
2849 | /* | |
2850 | * If force is CHUNK_ALLOC_FORCE: | |
2851 | * - return 1 if it successfully allocates a chunk, | |
2852 | * - return errors including -ENOSPC otherwise. | |
2853 | * If force is NOT CHUNK_ALLOC_FORCE: | |
2854 | * - return 0 if it doesn't need to allocate a new chunk, | |
2855 | * - return 1 if it successfully allocates a chunk, | |
2856 | * - return errors including -ENOSPC otherwise. | |
2857 | */ | |
2858 | int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, | |
2859 | enum btrfs_chunk_alloc_enum force) | |
2860 | { | |
2861 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2862 | struct btrfs_space_info *space_info; | |
2863 | bool wait_for_alloc = false; | |
2864 | bool should_alloc = false; | |
2865 | int ret = 0; | |
2866 | ||
2867 | /* Don't re-enter if we're already allocating a chunk */ | |
2868 | if (trans->allocating_chunk) | |
2869 | return -ENOSPC; | |
2870 | ||
2871 | space_info = btrfs_find_space_info(fs_info, flags); | |
2872 | ASSERT(space_info); | |
2873 | ||
2874 | do { | |
2875 | spin_lock(&space_info->lock); | |
2876 | if (force < space_info->force_alloc) | |
2877 | force = space_info->force_alloc; | |
2878 | should_alloc = should_alloc_chunk(fs_info, space_info, force); | |
2879 | if (space_info->full) { | |
2880 | /* No more free physical space */ | |
2881 | if (should_alloc) | |
2882 | ret = -ENOSPC; | |
2883 | else | |
2884 | ret = 0; | |
2885 | spin_unlock(&space_info->lock); | |
2886 | return ret; | |
2887 | } else if (!should_alloc) { | |
2888 | spin_unlock(&space_info->lock); | |
2889 | return 0; | |
2890 | } else if (space_info->chunk_alloc) { | |
2891 | /* | |
2892 | * Someone is already allocating, so we need to block | |
2893 | * until this someone is finished and then loop to | |
2894 | * recheck if we should continue with our allocation | |
2895 | * attempt. | |
2896 | */ | |
2897 | wait_for_alloc = true; | |
2898 | spin_unlock(&space_info->lock); | |
2899 | mutex_lock(&fs_info->chunk_mutex); | |
2900 | mutex_unlock(&fs_info->chunk_mutex); | |
2901 | } else { | |
2902 | /* Proceed with allocation */ | |
2903 | space_info->chunk_alloc = 1; | |
2904 | wait_for_alloc = false; | |
2905 | spin_unlock(&space_info->lock); | |
2906 | } | |
2907 | ||
2908 | cond_resched(); | |
2909 | } while (wait_for_alloc); | |
2910 | ||
2911 | mutex_lock(&fs_info->chunk_mutex); | |
2912 | trans->allocating_chunk = true; | |
2913 | ||
2914 | /* | |
2915 | * If we have mixed data/metadata chunks we want to make sure we keep | |
2916 | * allocating mixed chunks instead of individual chunks. | |
2917 | */ | |
2918 | if (btrfs_mixed_space_info(space_info)) | |
2919 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
2920 | ||
2921 | /* | |
2922 | * if we're doing a data chunk, go ahead and make sure that | |
2923 | * we keep a reasonable number of metadata chunks allocated in the | |
2924 | * FS as well. | |
2925 | */ | |
2926 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { | |
2927 | fs_info->data_chunk_allocations++; | |
2928 | if (!(fs_info->data_chunk_allocations % | |
2929 | fs_info->metadata_ratio)) | |
2930 | force_metadata_allocation(fs_info); | |
2931 | } | |
2932 | ||
2933 | /* | |
2934 | * Check if we have enough space in SYSTEM chunk because we may need | |
2935 | * to update devices. | |
2936 | */ | |
2937 | check_system_chunk(trans, flags); | |
2938 | ||
2939 | ret = btrfs_alloc_chunk(trans, flags); | |
2940 | trans->allocating_chunk = false; | |
2941 | ||
2942 | spin_lock(&space_info->lock); | |
2943 | if (ret < 0) { | |
2944 | if (ret == -ENOSPC) | |
2945 | space_info->full = 1; | |
2946 | else | |
2947 | goto out; | |
2948 | } else { | |
2949 | ret = 1; | |
2950 | space_info->max_extent_size = 0; | |
2951 | } | |
2952 | ||
2953 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
2954 | out: | |
2955 | space_info->chunk_alloc = 0; | |
2956 | spin_unlock(&space_info->lock); | |
2957 | mutex_unlock(&fs_info->chunk_mutex); | |
2958 | /* | |
2959 | * When we allocate a new chunk we reserve space in the chunk block | |
2960 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
2961 | * add new nodes/leafs to it if we end up needing to do it when | |
2962 | * inserting the chunk item and updating device items as part of the | |
2963 | * second phase of chunk allocation, performed by | |
2964 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
2965 | * large number of new block groups to create in our transaction | |
2966 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
2967 | * in extreme cases - like having a single transaction create many new | |
2968 | * block groups when starting to write out the free space caches of all | |
2969 | * the block groups that were made dirty during the lifetime of the | |
2970 | * transaction. | |
2971 | */ | |
2972 | if (trans->chunk_bytes_reserved >= (u64)SZ_2M) | |
2973 | btrfs_create_pending_block_groups(trans); | |
2974 | ||
2975 | return ret; | |
2976 | } | |
2977 | ||
2978 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) | |
2979 | { | |
2980 | u64 num_dev; | |
2981 | ||
2982 | num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max; | |
2983 | if (!num_dev) | |
2984 | num_dev = fs_info->fs_devices->rw_devices; | |
2985 | ||
2986 | return num_dev; | |
2987 | } | |
2988 | ||
2989 | /* | |
a9143bd3 | 2990 | * Reserve space in the system space for allocating or removing a chunk |
07730d87 JB |
2991 | */ |
2992 | void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) | |
2993 | { | |
2994 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2995 | struct btrfs_space_info *info; | |
2996 | u64 left; | |
2997 | u64 thresh; | |
2998 | int ret = 0; | |
2999 | u64 num_devs; | |
3000 | ||
3001 | /* | |
3002 | * Needed because we can end up allocating a system chunk and for an | |
3003 | * atomic and race free space reservation in the chunk block reserve. | |
3004 | */ | |
3005 | lockdep_assert_held(&fs_info->chunk_mutex); | |
3006 | ||
3007 | info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3008 | spin_lock(&info->lock); | |
3009 | left = info->total_bytes - btrfs_space_info_used(info, true); | |
3010 | spin_unlock(&info->lock); | |
3011 | ||
3012 | num_devs = get_profile_num_devs(fs_info, type); | |
3013 | ||
3014 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
2bd36e7b JB |
3015 | thresh = btrfs_calc_metadata_size(fs_info, num_devs) + |
3016 | btrfs_calc_insert_metadata_size(fs_info, 1); | |
07730d87 JB |
3017 | |
3018 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { | |
3019 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
3020 | left, thresh, type); | |
3021 | btrfs_dump_space_info(fs_info, info, 0, 0); | |
3022 | } | |
3023 | ||
3024 | if (left < thresh) { | |
3025 | u64 flags = btrfs_system_alloc_profile(fs_info); | |
3026 | ||
3027 | /* | |
3028 | * Ignore failure to create system chunk. We might end up not | |
3029 | * needing it, as we might not need to COW all nodes/leafs from | |
3030 | * the paths we visit in the chunk tree (they were already COWed | |
3031 | * or created in the current transaction for example). | |
3032 | */ | |
3033 | ret = btrfs_alloc_chunk(trans, flags); | |
3034 | } | |
3035 | ||
3036 | if (!ret) { | |
3037 | ret = btrfs_block_rsv_add(fs_info->chunk_root, | |
3038 | &fs_info->chunk_block_rsv, | |
3039 | thresh, BTRFS_RESERVE_NO_FLUSH); | |
3040 | if (!ret) | |
3041 | trans->chunk_bytes_reserved += thresh; | |
3042 | } | |
3043 | } | |
3044 | ||
3e43c279 JB |
3045 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
3046 | { | |
3047 | struct btrfs_block_group_cache *block_group; | |
3048 | u64 last = 0; | |
3049 | ||
3050 | while (1) { | |
3051 | struct inode *inode; | |
3052 | ||
3053 | block_group = btrfs_lookup_first_block_group(info, last); | |
3054 | while (block_group) { | |
3055 | btrfs_wait_block_group_cache_done(block_group); | |
3056 | spin_lock(&block_group->lock); | |
3057 | if (block_group->iref) | |
3058 | break; | |
3059 | spin_unlock(&block_group->lock); | |
3060 | block_group = btrfs_next_block_group(block_group); | |
3061 | } | |
3062 | if (!block_group) { | |
3063 | if (last == 0) | |
3064 | break; | |
3065 | last = 0; | |
3066 | continue; | |
3067 | } | |
3068 | ||
3069 | inode = block_group->inode; | |
3070 | block_group->iref = 0; | |
3071 | block_group->inode = NULL; | |
3072 | spin_unlock(&block_group->lock); | |
3073 | ASSERT(block_group->io_ctl.inode == NULL); | |
3074 | iput(inode); | |
3075 | last = block_group->key.objectid + block_group->key.offset; | |
3076 | btrfs_put_block_group(block_group); | |
3077 | } | |
3078 | } | |
3079 | ||
3080 | /* | |
3081 | * Must be called only after stopping all workers, since we could have block | |
3082 | * group caching kthreads running, and therefore they could race with us if we | |
3083 | * freed the block groups before stopping them. | |
3084 | */ | |
3085 | int btrfs_free_block_groups(struct btrfs_fs_info *info) | |
3086 | { | |
3087 | struct btrfs_block_group_cache *block_group; | |
3088 | struct btrfs_space_info *space_info; | |
3089 | struct btrfs_caching_control *caching_ctl; | |
3090 | struct rb_node *n; | |
3091 | ||
3092 | down_write(&info->commit_root_sem); | |
3093 | while (!list_empty(&info->caching_block_groups)) { | |
3094 | caching_ctl = list_entry(info->caching_block_groups.next, | |
3095 | struct btrfs_caching_control, list); | |
3096 | list_del(&caching_ctl->list); | |
3097 | btrfs_put_caching_control(caching_ctl); | |
3098 | } | |
3099 | up_write(&info->commit_root_sem); | |
3100 | ||
3101 | spin_lock(&info->unused_bgs_lock); | |
3102 | while (!list_empty(&info->unused_bgs)) { | |
3103 | block_group = list_first_entry(&info->unused_bgs, | |
3104 | struct btrfs_block_group_cache, | |
3105 | bg_list); | |
3106 | list_del_init(&block_group->bg_list); | |
3107 | btrfs_put_block_group(block_group); | |
3108 | } | |
3109 | spin_unlock(&info->unused_bgs_lock); | |
3110 | ||
3111 | spin_lock(&info->block_group_cache_lock); | |
3112 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
3113 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
3114 | cache_node); | |
3115 | rb_erase(&block_group->cache_node, | |
3116 | &info->block_group_cache_tree); | |
3117 | RB_CLEAR_NODE(&block_group->cache_node); | |
3118 | spin_unlock(&info->block_group_cache_lock); | |
3119 | ||
3120 | down_write(&block_group->space_info->groups_sem); | |
3121 | list_del(&block_group->list); | |
3122 | up_write(&block_group->space_info->groups_sem); | |
3123 | ||
3124 | /* | |
3125 | * We haven't cached this block group, which means we could | |
3126 | * possibly have excluded extents on this block group. | |
3127 | */ | |
3128 | if (block_group->cached == BTRFS_CACHE_NO || | |
3129 | block_group->cached == BTRFS_CACHE_ERROR) | |
3130 | btrfs_free_excluded_extents(block_group); | |
3131 | ||
3132 | btrfs_remove_free_space_cache(block_group); | |
3133 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); | |
3134 | ASSERT(list_empty(&block_group->dirty_list)); | |
3135 | ASSERT(list_empty(&block_group->io_list)); | |
3136 | ASSERT(list_empty(&block_group->bg_list)); | |
3137 | ASSERT(atomic_read(&block_group->count) == 1); | |
3138 | btrfs_put_block_group(block_group); | |
3139 | ||
3140 | spin_lock(&info->block_group_cache_lock); | |
3141 | } | |
3142 | spin_unlock(&info->block_group_cache_lock); | |
3143 | ||
3144 | /* | |
3145 | * Now that all the block groups are freed, go through and free all the | |
3146 | * space_info structs. This is only called during the final stages of | |
3147 | * unmount, and so we know nobody is using them. We call | |
3148 | * synchronize_rcu() once before we start, just to be on the safe side. | |
3149 | */ | |
3150 | synchronize_rcu(); | |
3151 | ||
3152 | btrfs_release_global_block_rsv(info); | |
3153 | ||
3154 | while (!list_empty(&info->space_info)) { | |
3155 | space_info = list_entry(info->space_info.next, | |
3156 | struct btrfs_space_info, | |
3157 | list); | |
3158 | ||
3159 | /* | |
3160 | * Do not hide this behind enospc_debug, this is actually | |
3161 | * important and indicates a real bug if this happens. | |
3162 | */ | |
3163 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
3164 | space_info->bytes_reserved > 0 || | |
3165 | space_info->bytes_may_use > 0)) | |
3166 | btrfs_dump_space_info(info, space_info, 0, 0); | |
3167 | list_del(&space_info->list); | |
3168 | btrfs_sysfs_remove_space_info(space_info); | |
3169 | } | |
3170 | return 0; | |
3171 | } |