]>
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 "volumes.h" |
11 | #include "transaction.h" | |
12 | #include "ref-verify.h" | |
4358d963 JB |
13 | #include "sysfs.h" |
14 | #include "tree-log.h" | |
77745c05 | 15 | #include "delalloc-space.h" |
b0643e59 | 16 | #include "discard.h" |
96a14336 | 17 | #include "raid56.h" |
08e11a3d | 18 | #include "zoned.h" |
2e405ad8 | 19 | |
878d7b67 JB |
20 | /* |
21 | * Return target flags in extended format or 0 if restripe for this chunk_type | |
22 | * is not in progress | |
23 | * | |
24 | * Should be called with balance_lock held | |
25 | */ | |
e11c0406 | 26 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) |
878d7b67 JB |
27 | { |
28 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
29 | u64 target = 0; | |
30 | ||
31 | if (!bctl) | |
32 | return 0; | |
33 | ||
34 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
35 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
36 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
37 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
38 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
39 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
40 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
41 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
42 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
43 | } | |
44 | ||
45 | return target; | |
46 | } | |
47 | ||
48 | /* | |
49 | * @flags: available profiles in extended format (see ctree.h) | |
50 | * | |
51 | * Return reduced profile in chunk format. If profile changing is in progress | |
52 | * (either running or paused) picks the target profile (if it's already | |
53 | * available), otherwise falls back to plain reducing. | |
54 | */ | |
55 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) | |
56 | { | |
57 | u64 num_devices = fs_info->fs_devices->rw_devices; | |
58 | u64 target; | |
59 | u64 raid_type; | |
60 | u64 allowed = 0; | |
61 | ||
62 | /* | |
63 | * See if restripe for this chunk_type is in progress, if so try to | |
64 | * reduce to the target profile | |
65 | */ | |
66 | spin_lock(&fs_info->balance_lock); | |
e11c0406 | 67 | target = get_restripe_target(fs_info, flags); |
878d7b67 | 68 | if (target) { |
162e0a16 JB |
69 | spin_unlock(&fs_info->balance_lock); |
70 | return extended_to_chunk(target); | |
878d7b67 JB |
71 | } |
72 | spin_unlock(&fs_info->balance_lock); | |
73 | ||
74 | /* First, mask out the RAID levels which aren't possible */ | |
75 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { | |
76 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
77 | allowed |= btrfs_raid_array[raid_type].bg_flag; | |
78 | } | |
79 | allowed &= flags; | |
80 | ||
81 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
82 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
83 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
84 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
85 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
86 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
87 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
88 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
89 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
90 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
91 | ||
92 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
93 | ||
94 | return extended_to_chunk(flags | allowed); | |
95 | } | |
96 | ||
ef0a82da | 97 | u64 btrfs_get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
878d7b67 JB |
98 | { |
99 | unsigned seq; | |
100 | u64 flags; | |
101 | ||
102 | do { | |
103 | flags = orig_flags; | |
104 | seq = read_seqbegin(&fs_info->profiles_lock); | |
105 | ||
106 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
107 | flags |= fs_info->avail_data_alloc_bits; | |
108 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
109 | flags |= fs_info->avail_system_alloc_bits; | |
110 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
111 | flags |= fs_info->avail_metadata_alloc_bits; | |
112 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
113 | ||
114 | return btrfs_reduce_alloc_profile(fs_info, flags); | |
115 | } | |
116 | ||
32da5386 | 117 | void btrfs_get_block_group(struct btrfs_block_group *cache) |
3cad1284 | 118 | { |
48aaeebe | 119 | refcount_inc(&cache->refs); |
3cad1284 JB |
120 | } |
121 | ||
32da5386 | 122 | void btrfs_put_block_group(struct btrfs_block_group *cache) |
3cad1284 | 123 | { |
48aaeebe | 124 | if (refcount_dec_and_test(&cache->refs)) { |
3cad1284 JB |
125 | WARN_ON(cache->pinned > 0); |
126 | WARN_ON(cache->reserved > 0); | |
127 | ||
b0643e59 DZ |
128 | /* |
129 | * A block_group shouldn't be on the discard_list anymore. | |
130 | * Remove the block_group from the discard_list to prevent us | |
131 | * from causing a panic due to NULL pointer dereference. | |
132 | */ | |
133 | if (WARN_ON(!list_empty(&cache->discard_list))) | |
134 | btrfs_discard_cancel_work(&cache->fs_info->discard_ctl, | |
135 | cache); | |
136 | ||
3cad1284 JB |
137 | /* |
138 | * If not empty, someone is still holding mutex of | |
139 | * full_stripe_lock, which can only be released by caller. | |
140 | * And it will definitely cause use-after-free when caller | |
141 | * tries to release full stripe lock. | |
142 | * | |
143 | * No better way to resolve, but only to warn. | |
144 | */ | |
145 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
146 | kfree(cache->free_space_ctl); | |
147 | kfree(cache); | |
148 | } | |
149 | } | |
150 | ||
4358d963 JB |
151 | /* |
152 | * This adds the block group to the fs_info rb tree for the block group cache | |
153 | */ | |
154 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, | |
32da5386 | 155 | struct btrfs_block_group *block_group) |
4358d963 JB |
156 | { |
157 | struct rb_node **p; | |
158 | struct rb_node *parent = NULL; | |
32da5386 | 159 | struct btrfs_block_group *cache; |
4358d963 | 160 | |
9afc6649 QW |
161 | ASSERT(block_group->length != 0); |
162 | ||
4358d963 JB |
163 | spin_lock(&info->block_group_cache_lock); |
164 | p = &info->block_group_cache_tree.rb_node; | |
165 | ||
166 | while (*p) { | |
167 | parent = *p; | |
32da5386 | 168 | cache = rb_entry(parent, struct btrfs_block_group, cache_node); |
b3470b5d | 169 | if (block_group->start < cache->start) { |
4358d963 | 170 | p = &(*p)->rb_left; |
b3470b5d | 171 | } else if (block_group->start > cache->start) { |
4358d963 JB |
172 | p = &(*p)->rb_right; |
173 | } else { | |
174 | spin_unlock(&info->block_group_cache_lock); | |
175 | return -EEXIST; | |
176 | } | |
177 | } | |
178 | ||
179 | rb_link_node(&block_group->cache_node, parent, p); | |
180 | rb_insert_color(&block_group->cache_node, | |
181 | &info->block_group_cache_tree); | |
182 | ||
b3470b5d DS |
183 | if (info->first_logical_byte > block_group->start) |
184 | info->first_logical_byte = block_group->start; | |
4358d963 JB |
185 | |
186 | spin_unlock(&info->block_group_cache_lock); | |
187 | ||
188 | return 0; | |
189 | } | |
190 | ||
2e405ad8 JB |
191 | /* |
192 | * This will return the block group at or after bytenr if contains is 0, else | |
193 | * it will return the block group that contains the bytenr | |
194 | */ | |
32da5386 | 195 | static struct btrfs_block_group *block_group_cache_tree_search( |
2e405ad8 JB |
196 | struct btrfs_fs_info *info, u64 bytenr, int contains) |
197 | { | |
32da5386 | 198 | struct btrfs_block_group *cache, *ret = NULL; |
2e405ad8 JB |
199 | struct rb_node *n; |
200 | u64 end, start; | |
201 | ||
202 | spin_lock(&info->block_group_cache_lock); | |
203 | n = info->block_group_cache_tree.rb_node; | |
204 | ||
205 | while (n) { | |
32da5386 | 206 | cache = rb_entry(n, struct btrfs_block_group, cache_node); |
b3470b5d DS |
207 | end = cache->start + cache->length - 1; |
208 | start = cache->start; | |
2e405ad8 JB |
209 | |
210 | if (bytenr < start) { | |
b3470b5d | 211 | if (!contains && (!ret || start < ret->start)) |
2e405ad8 JB |
212 | ret = cache; |
213 | n = n->rb_left; | |
214 | } else if (bytenr > start) { | |
215 | if (contains && bytenr <= end) { | |
216 | ret = cache; | |
217 | break; | |
218 | } | |
219 | n = n->rb_right; | |
220 | } else { | |
221 | ret = cache; | |
222 | break; | |
223 | } | |
224 | } | |
225 | if (ret) { | |
226 | btrfs_get_block_group(ret); | |
b3470b5d DS |
227 | if (bytenr == 0 && info->first_logical_byte > ret->start) |
228 | info->first_logical_byte = ret->start; | |
2e405ad8 JB |
229 | } |
230 | spin_unlock(&info->block_group_cache_lock); | |
231 | ||
232 | return ret; | |
233 | } | |
234 | ||
235 | /* | |
236 | * Return the block group that starts at or after bytenr | |
237 | */ | |
32da5386 | 238 | struct btrfs_block_group *btrfs_lookup_first_block_group( |
2e405ad8 JB |
239 | struct btrfs_fs_info *info, u64 bytenr) |
240 | { | |
241 | return block_group_cache_tree_search(info, bytenr, 0); | |
242 | } | |
243 | ||
244 | /* | |
245 | * Return the block group that contains the given bytenr | |
246 | */ | |
32da5386 | 247 | struct btrfs_block_group *btrfs_lookup_block_group( |
2e405ad8 JB |
248 | struct btrfs_fs_info *info, u64 bytenr) |
249 | { | |
250 | return block_group_cache_tree_search(info, bytenr, 1); | |
251 | } | |
252 | ||
32da5386 DS |
253 | struct btrfs_block_group *btrfs_next_block_group( |
254 | struct btrfs_block_group *cache) | |
2e405ad8 JB |
255 | { |
256 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
257 | struct rb_node *node; | |
258 | ||
259 | spin_lock(&fs_info->block_group_cache_lock); | |
260 | ||
261 | /* If our block group was removed, we need a full search. */ | |
262 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
b3470b5d | 263 | const u64 next_bytenr = cache->start + cache->length; |
2e405ad8 JB |
264 | |
265 | spin_unlock(&fs_info->block_group_cache_lock); | |
266 | btrfs_put_block_group(cache); | |
267 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; | |
268 | } | |
269 | node = rb_next(&cache->cache_node); | |
270 | btrfs_put_block_group(cache); | |
271 | if (node) { | |
32da5386 | 272 | cache = rb_entry(node, struct btrfs_block_group, cache_node); |
2e405ad8 JB |
273 | btrfs_get_block_group(cache); |
274 | } else | |
275 | cache = NULL; | |
276 | spin_unlock(&fs_info->block_group_cache_lock); | |
277 | return cache; | |
278 | } | |
3eeb3226 JB |
279 | |
280 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
281 | { | |
32da5386 | 282 | struct btrfs_block_group *bg; |
3eeb3226 JB |
283 | bool ret = true; |
284 | ||
285 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
286 | if (!bg) | |
287 | return false; | |
288 | ||
289 | spin_lock(&bg->lock); | |
290 | if (bg->ro) | |
291 | ret = false; | |
292 | else | |
293 | atomic_inc(&bg->nocow_writers); | |
294 | spin_unlock(&bg->lock); | |
295 | ||
296 | /* No put on block group, done by btrfs_dec_nocow_writers */ | |
297 | if (!ret) | |
298 | btrfs_put_block_group(bg); | |
299 | ||
300 | return ret; | |
301 | } | |
302 | ||
303 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
304 | { | |
32da5386 | 305 | struct btrfs_block_group *bg; |
3eeb3226 JB |
306 | |
307 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
308 | ASSERT(bg); | |
309 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
310 | wake_up_var(&bg->nocow_writers); | |
311 | /* | |
312 | * Once for our lookup and once for the lookup done by a previous call | |
313 | * to btrfs_inc_nocow_writers() | |
314 | */ | |
315 | btrfs_put_block_group(bg); | |
316 | btrfs_put_block_group(bg); | |
317 | } | |
318 | ||
32da5386 | 319 | void btrfs_wait_nocow_writers(struct btrfs_block_group *bg) |
3eeb3226 JB |
320 | { |
321 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); | |
322 | } | |
323 | ||
324 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
325 | const u64 start) | |
326 | { | |
32da5386 | 327 | struct btrfs_block_group *bg; |
3eeb3226 JB |
328 | |
329 | bg = btrfs_lookup_block_group(fs_info, start); | |
330 | ASSERT(bg); | |
331 | if (atomic_dec_and_test(&bg->reservations)) | |
332 | wake_up_var(&bg->reservations); | |
333 | btrfs_put_block_group(bg); | |
334 | } | |
335 | ||
32da5386 | 336 | void btrfs_wait_block_group_reservations(struct btrfs_block_group *bg) |
3eeb3226 JB |
337 | { |
338 | struct btrfs_space_info *space_info = bg->space_info; | |
339 | ||
340 | ASSERT(bg->ro); | |
341 | ||
342 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
343 | return; | |
344 | ||
345 | /* | |
346 | * Our block group is read only but before we set it to read only, | |
347 | * some task might have had allocated an extent from it already, but it | |
348 | * has not yet created a respective ordered extent (and added it to a | |
349 | * root's list of ordered extents). | |
350 | * Therefore wait for any task currently allocating extents, since the | |
351 | * block group's reservations counter is incremented while a read lock | |
352 | * on the groups' semaphore is held and decremented after releasing | |
353 | * the read access on that semaphore and creating the ordered extent. | |
354 | */ | |
355 | down_write(&space_info->groups_sem); | |
356 | up_write(&space_info->groups_sem); | |
357 | ||
358 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); | |
359 | } | |
9f21246d JB |
360 | |
361 | struct btrfs_caching_control *btrfs_get_caching_control( | |
32da5386 | 362 | struct btrfs_block_group *cache) |
9f21246d JB |
363 | { |
364 | struct btrfs_caching_control *ctl; | |
365 | ||
366 | spin_lock(&cache->lock); | |
367 | if (!cache->caching_ctl) { | |
368 | spin_unlock(&cache->lock); | |
369 | return NULL; | |
370 | } | |
371 | ||
372 | ctl = cache->caching_ctl; | |
373 | refcount_inc(&ctl->count); | |
374 | spin_unlock(&cache->lock); | |
375 | return ctl; | |
376 | } | |
377 | ||
378 | void btrfs_put_caching_control(struct btrfs_caching_control *ctl) | |
379 | { | |
380 | if (refcount_dec_and_test(&ctl->count)) | |
381 | kfree(ctl); | |
382 | } | |
383 | ||
384 | /* | |
385 | * When we wait for progress in the block group caching, its because our | |
386 | * allocation attempt failed at least once. So, we must sleep and let some | |
387 | * progress happen before we try again. | |
388 | * | |
389 | * This function will sleep at least once waiting for new free space to show | |
390 | * up, and then it will check the block group free space numbers for our min | |
391 | * num_bytes. Another option is to have it go ahead and look in the rbtree for | |
392 | * a free extent of a given size, but this is a good start. | |
393 | * | |
394 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
395 | * any of the information in this block group. | |
396 | */ | |
32da5386 | 397 | void btrfs_wait_block_group_cache_progress(struct btrfs_block_group *cache, |
9f21246d JB |
398 | u64 num_bytes) |
399 | { | |
400 | struct btrfs_caching_control *caching_ctl; | |
401 | ||
402 | caching_ctl = btrfs_get_caching_control(cache); | |
403 | if (!caching_ctl) | |
404 | return; | |
405 | ||
32da5386 | 406 | wait_event(caching_ctl->wait, btrfs_block_group_done(cache) || |
9f21246d JB |
407 | (cache->free_space_ctl->free_space >= num_bytes)); |
408 | ||
409 | btrfs_put_caching_control(caching_ctl); | |
410 | } | |
411 | ||
32da5386 | 412 | int btrfs_wait_block_group_cache_done(struct btrfs_block_group *cache) |
9f21246d JB |
413 | { |
414 | struct btrfs_caching_control *caching_ctl; | |
415 | int ret = 0; | |
416 | ||
417 | caching_ctl = btrfs_get_caching_control(cache); | |
418 | if (!caching_ctl) | |
419 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; | |
420 | ||
32da5386 | 421 | wait_event(caching_ctl->wait, btrfs_block_group_done(cache)); |
9f21246d JB |
422 | if (cache->cached == BTRFS_CACHE_ERROR) |
423 | ret = -EIO; | |
424 | btrfs_put_caching_control(caching_ctl); | |
425 | return ret; | |
426 | } | |
427 | ||
e747853c JB |
428 | static bool space_cache_v1_done(struct btrfs_block_group *cache) |
429 | { | |
430 | bool ret; | |
431 | ||
432 | spin_lock(&cache->lock); | |
433 | ret = cache->cached != BTRFS_CACHE_FAST; | |
434 | spin_unlock(&cache->lock); | |
435 | ||
436 | return ret; | |
437 | } | |
438 | ||
439 | void btrfs_wait_space_cache_v1_finished(struct btrfs_block_group *cache, | |
440 | struct btrfs_caching_control *caching_ctl) | |
441 | { | |
442 | wait_event(caching_ctl->wait, space_cache_v1_done(cache)); | |
443 | } | |
444 | ||
9f21246d | 445 | #ifdef CONFIG_BTRFS_DEBUG |
32da5386 | 446 | static void fragment_free_space(struct btrfs_block_group *block_group) |
9f21246d JB |
447 | { |
448 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
b3470b5d DS |
449 | u64 start = block_group->start; |
450 | u64 len = block_group->length; | |
9f21246d JB |
451 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? |
452 | fs_info->nodesize : fs_info->sectorsize; | |
453 | u64 step = chunk << 1; | |
454 | ||
455 | while (len > chunk) { | |
456 | btrfs_remove_free_space(block_group, start, chunk); | |
457 | start += step; | |
458 | if (len < step) | |
459 | len = 0; | |
460 | else | |
461 | len -= step; | |
462 | } | |
463 | } | |
464 | #endif | |
465 | ||
466 | /* | |
467 | * This is only called by btrfs_cache_block_group, since we could have freed | |
468 | * extents we need to check the pinned_extents for any extents that can't be | |
469 | * used yet since their free space will be released as soon as the transaction | |
470 | * commits. | |
471 | */ | |
32da5386 | 472 | u64 add_new_free_space(struct btrfs_block_group *block_group, u64 start, u64 end) |
9f21246d JB |
473 | { |
474 | struct btrfs_fs_info *info = block_group->fs_info; | |
475 | u64 extent_start, extent_end, size, total_added = 0; | |
476 | int ret; | |
477 | ||
478 | while (start < end) { | |
fe119a6e | 479 | ret = find_first_extent_bit(&info->excluded_extents, start, |
9f21246d JB |
480 | &extent_start, &extent_end, |
481 | EXTENT_DIRTY | EXTENT_UPTODATE, | |
482 | NULL); | |
483 | if (ret) | |
484 | break; | |
485 | ||
486 | if (extent_start <= start) { | |
487 | start = extent_end + 1; | |
488 | } else if (extent_start > start && extent_start < end) { | |
489 | size = extent_start - start; | |
490 | total_added += size; | |
b0643e59 DZ |
491 | ret = btrfs_add_free_space_async_trimmed(block_group, |
492 | start, size); | |
9f21246d JB |
493 | BUG_ON(ret); /* -ENOMEM or logic error */ |
494 | start = extent_end + 1; | |
495 | } else { | |
496 | break; | |
497 | } | |
498 | } | |
499 | ||
500 | if (start < end) { | |
501 | size = end - start; | |
502 | total_added += size; | |
b0643e59 DZ |
503 | ret = btrfs_add_free_space_async_trimmed(block_group, start, |
504 | size); | |
9f21246d JB |
505 | BUG_ON(ret); /* -ENOMEM or logic error */ |
506 | } | |
507 | ||
508 | return total_added; | |
509 | } | |
510 | ||
511 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) | |
512 | { | |
32da5386 | 513 | struct btrfs_block_group *block_group = caching_ctl->block_group; |
9f21246d JB |
514 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
515 | struct btrfs_root *extent_root = fs_info->extent_root; | |
516 | struct btrfs_path *path; | |
517 | struct extent_buffer *leaf; | |
518 | struct btrfs_key key; | |
519 | u64 total_found = 0; | |
520 | u64 last = 0; | |
521 | u32 nritems; | |
522 | int ret; | |
523 | bool wakeup = true; | |
524 | ||
525 | path = btrfs_alloc_path(); | |
526 | if (!path) | |
527 | return -ENOMEM; | |
528 | ||
b3470b5d | 529 | last = max_t(u64, block_group->start, BTRFS_SUPER_INFO_OFFSET); |
9f21246d JB |
530 | |
531 | #ifdef CONFIG_BTRFS_DEBUG | |
532 | /* | |
533 | * If we're fragmenting we don't want to make anybody think we can | |
534 | * allocate from this block group until we've had a chance to fragment | |
535 | * the free space. | |
536 | */ | |
537 | if (btrfs_should_fragment_free_space(block_group)) | |
538 | wakeup = false; | |
539 | #endif | |
540 | /* | |
541 | * We don't want to deadlock with somebody trying to allocate a new | |
542 | * extent for the extent root while also trying to search the extent | |
543 | * root to add free space. So we skip locking and search the commit | |
544 | * root, since its read-only | |
545 | */ | |
546 | path->skip_locking = 1; | |
547 | path->search_commit_root = 1; | |
548 | path->reada = READA_FORWARD; | |
549 | ||
550 | key.objectid = last; | |
551 | key.offset = 0; | |
552 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
553 | ||
554 | next: | |
555 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
556 | if (ret < 0) | |
557 | goto out; | |
558 | ||
559 | leaf = path->nodes[0]; | |
560 | nritems = btrfs_header_nritems(leaf); | |
561 | ||
562 | while (1) { | |
563 | if (btrfs_fs_closing(fs_info) > 1) { | |
564 | last = (u64)-1; | |
565 | break; | |
566 | } | |
567 | ||
568 | if (path->slots[0] < nritems) { | |
569 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
570 | } else { | |
571 | ret = btrfs_find_next_key(extent_root, path, &key, 0, 0); | |
572 | if (ret) | |
573 | break; | |
574 | ||
575 | if (need_resched() || | |
576 | rwsem_is_contended(&fs_info->commit_root_sem)) { | |
577 | if (wakeup) | |
578 | caching_ctl->progress = last; | |
579 | btrfs_release_path(path); | |
580 | up_read(&fs_info->commit_root_sem); | |
581 | mutex_unlock(&caching_ctl->mutex); | |
582 | cond_resched(); | |
583 | mutex_lock(&caching_ctl->mutex); | |
584 | down_read(&fs_info->commit_root_sem); | |
585 | goto next; | |
586 | } | |
587 | ||
588 | ret = btrfs_next_leaf(extent_root, path); | |
589 | if (ret < 0) | |
590 | goto out; | |
591 | if (ret) | |
592 | break; | |
593 | leaf = path->nodes[0]; | |
594 | nritems = btrfs_header_nritems(leaf); | |
595 | continue; | |
596 | } | |
597 | ||
598 | if (key.objectid < last) { | |
599 | key.objectid = last; | |
600 | key.offset = 0; | |
601 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
602 | ||
603 | if (wakeup) | |
604 | caching_ctl->progress = last; | |
605 | btrfs_release_path(path); | |
606 | goto next; | |
607 | } | |
608 | ||
b3470b5d | 609 | if (key.objectid < block_group->start) { |
9f21246d JB |
610 | path->slots[0]++; |
611 | continue; | |
612 | } | |
613 | ||
b3470b5d | 614 | if (key.objectid >= block_group->start + block_group->length) |
9f21246d JB |
615 | break; |
616 | ||
617 | if (key.type == BTRFS_EXTENT_ITEM_KEY || | |
618 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
619 | total_found += add_new_free_space(block_group, last, | |
620 | key.objectid); | |
621 | if (key.type == BTRFS_METADATA_ITEM_KEY) | |
622 | last = key.objectid + | |
623 | fs_info->nodesize; | |
624 | else | |
625 | last = key.objectid + key.offset; | |
626 | ||
627 | if (total_found > CACHING_CTL_WAKE_UP) { | |
628 | total_found = 0; | |
629 | if (wakeup) | |
630 | wake_up(&caching_ctl->wait); | |
631 | } | |
632 | } | |
633 | path->slots[0]++; | |
634 | } | |
635 | ret = 0; | |
636 | ||
637 | total_found += add_new_free_space(block_group, last, | |
b3470b5d | 638 | block_group->start + block_group->length); |
9f21246d JB |
639 | caching_ctl->progress = (u64)-1; |
640 | ||
641 | out: | |
642 | btrfs_free_path(path); | |
643 | return ret; | |
644 | } | |
645 | ||
646 | static noinline void caching_thread(struct btrfs_work *work) | |
647 | { | |
32da5386 | 648 | struct btrfs_block_group *block_group; |
9f21246d JB |
649 | struct btrfs_fs_info *fs_info; |
650 | struct btrfs_caching_control *caching_ctl; | |
651 | int ret; | |
652 | ||
653 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
654 | block_group = caching_ctl->block_group; | |
655 | fs_info = block_group->fs_info; | |
656 | ||
657 | mutex_lock(&caching_ctl->mutex); | |
658 | down_read(&fs_info->commit_root_sem); | |
659 | ||
e747853c JB |
660 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
661 | ret = load_free_space_cache(block_group); | |
662 | if (ret == 1) { | |
663 | ret = 0; | |
664 | goto done; | |
665 | } | |
666 | ||
667 | /* | |
668 | * We failed to load the space cache, set ourselves to | |
669 | * CACHE_STARTED and carry on. | |
670 | */ | |
671 | spin_lock(&block_group->lock); | |
672 | block_group->cached = BTRFS_CACHE_STARTED; | |
673 | spin_unlock(&block_group->lock); | |
674 | wake_up(&caching_ctl->wait); | |
675 | } | |
676 | ||
2f96e402 JB |
677 | /* |
678 | * If we are in the transaction that populated the free space tree we | |
679 | * can't actually cache from the free space tree as our commit root and | |
680 | * real root are the same, so we could change the contents of the blocks | |
681 | * while caching. Instead do the slow caching in this case, and after | |
682 | * the transaction has committed we will be safe. | |
683 | */ | |
684 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && | |
685 | !(test_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags))) | |
9f21246d JB |
686 | ret = load_free_space_tree(caching_ctl); |
687 | else | |
688 | ret = load_extent_tree_free(caching_ctl); | |
e747853c | 689 | done: |
9f21246d JB |
690 | spin_lock(&block_group->lock); |
691 | block_group->caching_ctl = NULL; | |
692 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; | |
693 | spin_unlock(&block_group->lock); | |
694 | ||
695 | #ifdef CONFIG_BTRFS_DEBUG | |
696 | if (btrfs_should_fragment_free_space(block_group)) { | |
697 | u64 bytes_used; | |
698 | ||
699 | spin_lock(&block_group->space_info->lock); | |
700 | spin_lock(&block_group->lock); | |
b3470b5d | 701 | bytes_used = block_group->length - block_group->used; |
9f21246d JB |
702 | block_group->space_info->bytes_used += bytes_used >> 1; |
703 | spin_unlock(&block_group->lock); | |
704 | spin_unlock(&block_group->space_info->lock); | |
e11c0406 | 705 | fragment_free_space(block_group); |
9f21246d JB |
706 | } |
707 | #endif | |
708 | ||
709 | caching_ctl->progress = (u64)-1; | |
710 | ||
711 | up_read(&fs_info->commit_root_sem); | |
712 | btrfs_free_excluded_extents(block_group); | |
713 | mutex_unlock(&caching_ctl->mutex); | |
714 | ||
715 | wake_up(&caching_ctl->wait); | |
716 | ||
717 | btrfs_put_caching_control(caching_ctl); | |
718 | btrfs_put_block_group(block_group); | |
719 | } | |
720 | ||
32da5386 | 721 | int btrfs_cache_block_group(struct btrfs_block_group *cache, int load_cache_only) |
9f21246d JB |
722 | { |
723 | DEFINE_WAIT(wait); | |
724 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
e747853c | 725 | struct btrfs_caching_control *caching_ctl = NULL; |
9f21246d JB |
726 | int ret = 0; |
727 | ||
2eda5708 NA |
728 | /* Allocator for zoned filesystems does not use the cache at all */ |
729 | if (btrfs_is_zoned(fs_info)) | |
730 | return 0; | |
731 | ||
9f21246d JB |
732 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
733 | if (!caching_ctl) | |
734 | return -ENOMEM; | |
735 | ||
736 | INIT_LIST_HEAD(&caching_ctl->list); | |
737 | mutex_init(&caching_ctl->mutex); | |
738 | init_waitqueue_head(&caching_ctl->wait); | |
739 | caching_ctl->block_group = cache; | |
b3470b5d | 740 | caching_ctl->progress = cache->start; |
e747853c | 741 | refcount_set(&caching_ctl->count, 2); |
a0cac0ec | 742 | btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL); |
9f21246d JB |
743 | |
744 | spin_lock(&cache->lock); | |
9f21246d | 745 | if (cache->cached != BTRFS_CACHE_NO) { |
9f21246d | 746 | kfree(caching_ctl); |
e747853c JB |
747 | |
748 | caching_ctl = cache->caching_ctl; | |
749 | if (caching_ctl) | |
750 | refcount_inc(&caching_ctl->count); | |
751 | spin_unlock(&cache->lock); | |
752 | goto out; | |
9f21246d JB |
753 | } |
754 | WARN_ON(cache->caching_ctl); | |
755 | cache->caching_ctl = caching_ctl; | |
e747853c JB |
756 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
757 | cache->cached = BTRFS_CACHE_FAST; | |
758 | else | |
759 | cache->cached = BTRFS_CACHE_STARTED; | |
760 | cache->has_caching_ctl = 1; | |
9f21246d JB |
761 | spin_unlock(&cache->lock); |
762 | ||
bbb86a37 | 763 | spin_lock(&fs_info->block_group_cache_lock); |
9f21246d JB |
764 | refcount_inc(&caching_ctl->count); |
765 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); | |
bbb86a37 | 766 | spin_unlock(&fs_info->block_group_cache_lock); |
9f21246d JB |
767 | |
768 | btrfs_get_block_group(cache); | |
769 | ||
770 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); | |
e747853c JB |
771 | out: |
772 | if (load_cache_only && caching_ctl) | |
773 | btrfs_wait_space_cache_v1_finished(cache, caching_ctl); | |
774 | if (caching_ctl) | |
775 | btrfs_put_caching_control(caching_ctl); | |
9f21246d JB |
776 | |
777 | return ret; | |
778 | } | |
e3e0520b JB |
779 | |
780 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
781 | { | |
782 | u64 extra_flags = chunk_to_extended(flags) & | |
783 | BTRFS_EXTENDED_PROFILE_MASK; | |
784 | ||
785 | write_seqlock(&fs_info->profiles_lock); | |
786 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
787 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
788 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
789 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
790 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
791 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
792 | write_sequnlock(&fs_info->profiles_lock); | |
793 | } | |
794 | ||
795 | /* | |
796 | * Clear incompat bits for the following feature(s): | |
797 | * | |
798 | * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group | |
799 | * in the whole filesystem | |
9c907446 DS |
800 | * |
801 | * - RAID1C34 - same as above for RAID1C3 and RAID1C4 block groups | |
e3e0520b JB |
802 | */ |
803 | static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
804 | { | |
9c907446 DS |
805 | bool found_raid56 = false; |
806 | bool found_raid1c34 = false; | |
807 | ||
808 | if ((flags & BTRFS_BLOCK_GROUP_RAID56_MASK) || | |
809 | (flags & BTRFS_BLOCK_GROUP_RAID1C3) || | |
810 | (flags & BTRFS_BLOCK_GROUP_RAID1C4)) { | |
e3e0520b JB |
811 | struct list_head *head = &fs_info->space_info; |
812 | struct btrfs_space_info *sinfo; | |
813 | ||
814 | list_for_each_entry_rcu(sinfo, head, list) { | |
e3e0520b JB |
815 | down_read(&sinfo->groups_sem); |
816 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5])) | |
9c907446 | 817 | found_raid56 = true; |
e3e0520b | 818 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6])) |
9c907446 DS |
819 | found_raid56 = true; |
820 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C3])) | |
821 | found_raid1c34 = true; | |
822 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C4])) | |
823 | found_raid1c34 = true; | |
e3e0520b | 824 | up_read(&sinfo->groups_sem); |
e3e0520b | 825 | } |
d8e6fd5c | 826 | if (!found_raid56) |
9c907446 | 827 | btrfs_clear_fs_incompat(fs_info, RAID56); |
d8e6fd5c | 828 | if (!found_raid1c34) |
9c907446 | 829 | btrfs_clear_fs_incompat(fs_info, RAID1C34); |
e3e0520b JB |
830 | } |
831 | } | |
832 | ||
7357623a QW |
833 | static int remove_block_group_item(struct btrfs_trans_handle *trans, |
834 | struct btrfs_path *path, | |
835 | struct btrfs_block_group *block_group) | |
836 | { | |
837 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
838 | struct btrfs_root *root; | |
839 | struct btrfs_key key; | |
840 | int ret; | |
841 | ||
842 | root = fs_info->extent_root; | |
843 | key.objectid = block_group->start; | |
844 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
845 | key.offset = block_group->length; | |
846 | ||
847 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
848 | if (ret > 0) | |
849 | ret = -ENOENT; | |
850 | if (ret < 0) | |
851 | return ret; | |
852 | ||
853 | ret = btrfs_del_item(trans, root, path); | |
854 | return ret; | |
855 | } | |
856 | ||
e3e0520b JB |
857 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
858 | u64 group_start, struct extent_map *em) | |
859 | { | |
860 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
e3e0520b | 861 | struct btrfs_path *path; |
32da5386 | 862 | struct btrfs_block_group *block_group; |
e3e0520b | 863 | struct btrfs_free_cluster *cluster; |
e3e0520b JB |
864 | struct inode *inode; |
865 | struct kobject *kobj = NULL; | |
866 | int ret; | |
867 | int index; | |
868 | int factor; | |
869 | struct btrfs_caching_control *caching_ctl = NULL; | |
870 | bool remove_em; | |
871 | bool remove_rsv = false; | |
872 | ||
873 | block_group = btrfs_lookup_block_group(fs_info, group_start); | |
874 | BUG_ON(!block_group); | |
875 | BUG_ON(!block_group->ro); | |
876 | ||
877 | trace_btrfs_remove_block_group(block_group); | |
878 | /* | |
879 | * Free the reserved super bytes from this block group before | |
880 | * remove it. | |
881 | */ | |
882 | btrfs_free_excluded_extents(block_group); | |
b3470b5d DS |
883 | btrfs_free_ref_tree_range(fs_info, block_group->start, |
884 | block_group->length); | |
e3e0520b | 885 | |
e3e0520b JB |
886 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
887 | factor = btrfs_bg_type_to_factor(block_group->flags); | |
888 | ||
889 | /* make sure this block group isn't part of an allocation cluster */ | |
890 | cluster = &fs_info->data_alloc_cluster; | |
891 | spin_lock(&cluster->refill_lock); | |
892 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
893 | spin_unlock(&cluster->refill_lock); | |
894 | ||
895 | /* | |
896 | * make sure this block group isn't part of a metadata | |
897 | * allocation cluster | |
898 | */ | |
899 | cluster = &fs_info->meta_alloc_cluster; | |
900 | spin_lock(&cluster->refill_lock); | |
901 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
902 | spin_unlock(&cluster->refill_lock); | |
903 | ||
40ab3be1 NA |
904 | btrfs_clear_treelog_bg(block_group); |
905 | ||
e3e0520b JB |
906 | path = btrfs_alloc_path(); |
907 | if (!path) { | |
908 | ret = -ENOMEM; | |
9fecd132 | 909 | goto out; |
e3e0520b JB |
910 | } |
911 | ||
912 | /* | |
913 | * get the inode first so any iput calls done for the io_list | |
914 | * aren't the final iput (no unlinks allowed now) | |
915 | */ | |
916 | inode = lookup_free_space_inode(block_group, path); | |
917 | ||
918 | mutex_lock(&trans->transaction->cache_write_mutex); | |
919 | /* | |
920 | * Make sure our free space cache IO is done before removing the | |
921 | * free space inode | |
922 | */ | |
923 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
924 | if (!list_empty(&block_group->io_list)) { | |
925 | list_del_init(&block_group->io_list); | |
926 | ||
927 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
928 | ||
929 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
930 | btrfs_wait_cache_io(trans, block_group, path); | |
931 | btrfs_put_block_group(block_group); | |
932 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
933 | } | |
934 | ||
935 | if (!list_empty(&block_group->dirty_list)) { | |
936 | list_del_init(&block_group->dirty_list); | |
937 | remove_rsv = true; | |
938 | btrfs_put_block_group(block_group); | |
939 | } | |
940 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
941 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
942 | ||
36b216c8 BB |
943 | ret = btrfs_remove_free_space_inode(trans, inode, block_group); |
944 | if (ret) | |
9fecd132 | 945 | goto out; |
e3e0520b JB |
946 | |
947 | spin_lock(&fs_info->block_group_cache_lock); | |
948 | rb_erase(&block_group->cache_node, | |
949 | &fs_info->block_group_cache_tree); | |
950 | RB_CLEAR_NODE(&block_group->cache_node); | |
951 | ||
9fecd132 FM |
952 | /* Once for the block groups rbtree */ |
953 | btrfs_put_block_group(block_group); | |
954 | ||
b3470b5d | 955 | if (fs_info->first_logical_byte == block_group->start) |
e3e0520b JB |
956 | fs_info->first_logical_byte = (u64)-1; |
957 | spin_unlock(&fs_info->block_group_cache_lock); | |
958 | ||
959 | down_write(&block_group->space_info->groups_sem); | |
960 | /* | |
961 | * we must use list_del_init so people can check to see if they | |
962 | * are still on the list after taking the semaphore | |
963 | */ | |
964 | list_del_init(&block_group->list); | |
965 | if (list_empty(&block_group->space_info->block_groups[index])) { | |
966 | kobj = block_group->space_info->block_group_kobjs[index]; | |
967 | block_group->space_info->block_group_kobjs[index] = NULL; | |
968 | clear_avail_alloc_bits(fs_info, block_group->flags); | |
969 | } | |
970 | up_write(&block_group->space_info->groups_sem); | |
971 | clear_incompat_bg_bits(fs_info, block_group->flags); | |
972 | if (kobj) { | |
973 | kobject_del(kobj); | |
974 | kobject_put(kobj); | |
975 | } | |
976 | ||
977 | if (block_group->has_caching_ctl) | |
978 | caching_ctl = btrfs_get_caching_control(block_group); | |
979 | if (block_group->cached == BTRFS_CACHE_STARTED) | |
980 | btrfs_wait_block_group_cache_done(block_group); | |
981 | if (block_group->has_caching_ctl) { | |
bbb86a37 | 982 | spin_lock(&fs_info->block_group_cache_lock); |
e3e0520b JB |
983 | if (!caching_ctl) { |
984 | struct btrfs_caching_control *ctl; | |
985 | ||
986 | list_for_each_entry(ctl, | |
987 | &fs_info->caching_block_groups, list) | |
988 | if (ctl->block_group == block_group) { | |
989 | caching_ctl = ctl; | |
990 | refcount_inc(&caching_ctl->count); | |
991 | break; | |
992 | } | |
993 | } | |
994 | if (caching_ctl) | |
995 | list_del_init(&caching_ctl->list); | |
bbb86a37 | 996 | spin_unlock(&fs_info->block_group_cache_lock); |
e3e0520b JB |
997 | if (caching_ctl) { |
998 | /* Once for the caching bgs list and once for us. */ | |
999 | btrfs_put_caching_control(caching_ctl); | |
1000 | btrfs_put_caching_control(caching_ctl); | |
1001 | } | |
1002 | } | |
1003 | ||
1004 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
1005 | WARN_ON(!list_empty(&block_group->dirty_list)); | |
1006 | WARN_ON(!list_empty(&block_group->io_list)); | |
1007 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
1008 | ||
1009 | btrfs_remove_free_space_cache(block_group); | |
1010 | ||
1011 | spin_lock(&block_group->space_info->lock); | |
1012 | list_del_init(&block_group->ro_list); | |
1013 | ||
1014 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { | |
1015 | WARN_ON(block_group->space_info->total_bytes | |
b3470b5d | 1016 | < block_group->length); |
e3e0520b | 1017 | WARN_ON(block_group->space_info->bytes_readonly |
169e0da9 NA |
1018 | < block_group->length - block_group->zone_unusable); |
1019 | WARN_ON(block_group->space_info->bytes_zone_unusable | |
1020 | < block_group->zone_unusable); | |
e3e0520b | 1021 | WARN_ON(block_group->space_info->disk_total |
b3470b5d | 1022 | < block_group->length * factor); |
e3e0520b | 1023 | } |
b3470b5d | 1024 | block_group->space_info->total_bytes -= block_group->length; |
169e0da9 NA |
1025 | block_group->space_info->bytes_readonly -= |
1026 | (block_group->length - block_group->zone_unusable); | |
1027 | block_group->space_info->bytes_zone_unusable -= | |
1028 | block_group->zone_unusable; | |
b3470b5d | 1029 | block_group->space_info->disk_total -= block_group->length * factor; |
e3e0520b JB |
1030 | |
1031 | spin_unlock(&block_group->space_info->lock); | |
1032 | ||
ffcb9d44 FM |
1033 | /* |
1034 | * Remove the free space for the block group from the free space tree | |
1035 | * and the block group's item from the extent tree before marking the | |
1036 | * block group as removed. This is to prevent races with tasks that | |
1037 | * freeze and unfreeze a block group, this task and another task | |
1038 | * allocating a new block group - the unfreeze task ends up removing | |
1039 | * the block group's extent map before the task calling this function | |
1040 | * deletes the block group item from the extent tree, allowing for | |
1041 | * another task to attempt to create another block group with the same | |
1042 | * item key (and failing with -EEXIST and a transaction abort). | |
1043 | */ | |
1044 | ret = remove_block_group_free_space(trans, block_group); | |
1045 | if (ret) | |
1046 | goto out; | |
1047 | ||
1048 | ret = remove_block_group_item(trans, path, block_group); | |
1049 | if (ret < 0) | |
1050 | goto out; | |
1051 | ||
e3e0520b JB |
1052 | spin_lock(&block_group->lock); |
1053 | block_group->removed = 1; | |
1054 | /* | |
6b7304af FM |
1055 | * At this point trimming or scrub can't start on this block group, |
1056 | * because we removed the block group from the rbtree | |
1057 | * fs_info->block_group_cache_tree so no one can't find it anymore and | |
1058 | * even if someone already got this block group before we removed it | |
1059 | * from the rbtree, they have already incremented block_group->frozen - | |
1060 | * if they didn't, for the trimming case they won't find any free space | |
1061 | * entries because we already removed them all when we called | |
1062 | * btrfs_remove_free_space_cache(). | |
e3e0520b JB |
1063 | * |
1064 | * And we must not remove the extent map from the fs_info->mapping_tree | |
1065 | * to prevent the same logical address range and physical device space | |
6b7304af FM |
1066 | * ranges from being reused for a new block group. This is needed to |
1067 | * avoid races with trimming and scrub. | |
1068 | * | |
1069 | * An fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
e3e0520b JB |
1070 | * completely transactionless, so while it is trimming a range the |
1071 | * currently running transaction might finish and a new one start, | |
1072 | * allowing for new block groups to be created that can reuse the same | |
1073 | * physical device locations unless we take this special care. | |
1074 | * | |
1075 | * There may also be an implicit trim operation if the file system | |
1076 | * is mounted with -odiscard. The same protections must remain | |
1077 | * in place until the extents have been discarded completely when | |
1078 | * the transaction commit has completed. | |
1079 | */ | |
6b7304af | 1080 | remove_em = (atomic_read(&block_group->frozen) == 0); |
e3e0520b JB |
1081 | spin_unlock(&block_group->lock); |
1082 | ||
e3e0520b JB |
1083 | if (remove_em) { |
1084 | struct extent_map_tree *em_tree; | |
1085 | ||
1086 | em_tree = &fs_info->mapping_tree; | |
1087 | write_lock(&em_tree->lock); | |
1088 | remove_extent_mapping(em_tree, em); | |
1089 | write_unlock(&em_tree->lock); | |
1090 | /* once for the tree */ | |
1091 | free_extent_map(em); | |
1092 | } | |
f6033c5e | 1093 | |
9fecd132 | 1094 | out: |
f6033c5e XY |
1095 | /* Once for the lookup reference */ |
1096 | btrfs_put_block_group(block_group); | |
e3e0520b JB |
1097 | if (remove_rsv) |
1098 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1099 | btrfs_free_path(path); | |
1100 | return ret; | |
1101 | } | |
1102 | ||
1103 | struct btrfs_trans_handle *btrfs_start_trans_remove_block_group( | |
1104 | struct btrfs_fs_info *fs_info, const u64 chunk_offset) | |
1105 | { | |
1106 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; | |
1107 | struct extent_map *em; | |
1108 | struct map_lookup *map; | |
1109 | unsigned int num_items; | |
1110 | ||
1111 | read_lock(&em_tree->lock); | |
1112 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
1113 | read_unlock(&em_tree->lock); | |
1114 | ASSERT(em && em->start == chunk_offset); | |
1115 | ||
1116 | /* | |
1117 | * We need to reserve 3 + N units from the metadata space info in order | |
1118 | * to remove a block group (done at btrfs_remove_chunk() and at | |
1119 | * btrfs_remove_block_group()), which are used for: | |
1120 | * | |
1121 | * 1 unit for adding the free space inode's orphan (located in the tree | |
1122 | * of tree roots). | |
1123 | * 1 unit for deleting the block group item (located in the extent | |
1124 | * tree). | |
1125 | * 1 unit for deleting the free space item (located in tree of tree | |
1126 | * roots). | |
1127 | * N units for deleting N device extent items corresponding to each | |
1128 | * stripe (located in the device tree). | |
1129 | * | |
1130 | * In order to remove a block group we also need to reserve units in the | |
1131 | * system space info in order to update the chunk tree (update one or | |
1132 | * more device items and remove one chunk item), but this is done at | |
1133 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
1134 | */ | |
1135 | map = em->map_lookup; | |
1136 | num_items = 3 + map->num_stripes; | |
1137 | free_extent_map(em); | |
1138 | ||
1139 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, | |
7f9fe614 | 1140 | num_items); |
e3e0520b JB |
1141 | } |
1142 | ||
26ce2095 JB |
1143 | /* |
1144 | * Mark block group @cache read-only, so later write won't happen to block | |
1145 | * group @cache. | |
1146 | * | |
1147 | * If @force is not set, this function will only mark the block group readonly | |
1148 | * if we have enough free space (1M) in other metadata/system block groups. | |
1149 | * If @force is not set, this function will mark the block group readonly | |
1150 | * without checking free space. | |
1151 | * | |
1152 | * NOTE: This function doesn't care if other block groups can contain all the | |
1153 | * data in this block group. That check should be done by relocation routine, | |
1154 | * not this function. | |
1155 | */ | |
32da5386 | 1156 | static int inc_block_group_ro(struct btrfs_block_group *cache, int force) |
26ce2095 JB |
1157 | { |
1158 | struct btrfs_space_info *sinfo = cache->space_info; | |
1159 | u64 num_bytes; | |
26ce2095 JB |
1160 | int ret = -ENOSPC; |
1161 | ||
26ce2095 JB |
1162 | spin_lock(&sinfo->lock); |
1163 | spin_lock(&cache->lock); | |
1164 | ||
195a49ea FM |
1165 | if (cache->swap_extents) { |
1166 | ret = -ETXTBSY; | |
1167 | goto out; | |
1168 | } | |
1169 | ||
26ce2095 JB |
1170 | if (cache->ro) { |
1171 | cache->ro++; | |
1172 | ret = 0; | |
1173 | goto out; | |
1174 | } | |
1175 | ||
b3470b5d | 1176 | num_bytes = cache->length - cache->reserved - cache->pinned - |
169e0da9 | 1177 | cache->bytes_super - cache->zone_unusable - cache->used; |
26ce2095 JB |
1178 | |
1179 | /* | |
a30a3d20 JB |
1180 | * Data never overcommits, even in mixed mode, so do just the straight |
1181 | * check of left over space in how much we have allocated. | |
26ce2095 | 1182 | */ |
a30a3d20 JB |
1183 | if (force) { |
1184 | ret = 0; | |
1185 | } else if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) { | |
1186 | u64 sinfo_used = btrfs_space_info_used(sinfo, true); | |
1187 | ||
1188 | /* | |
1189 | * Here we make sure if we mark this bg RO, we still have enough | |
1190 | * free space as buffer. | |
1191 | */ | |
1192 | if (sinfo_used + num_bytes <= sinfo->total_bytes) | |
1193 | ret = 0; | |
1194 | } else { | |
1195 | /* | |
1196 | * We overcommit metadata, so we need to do the | |
1197 | * btrfs_can_overcommit check here, and we need to pass in | |
1198 | * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of | |
1199 | * leeway to allow us to mark this block group as read only. | |
1200 | */ | |
1201 | if (btrfs_can_overcommit(cache->fs_info, sinfo, num_bytes, | |
1202 | BTRFS_RESERVE_NO_FLUSH)) | |
1203 | ret = 0; | |
1204 | } | |
1205 | ||
1206 | if (!ret) { | |
26ce2095 | 1207 | sinfo->bytes_readonly += num_bytes; |
169e0da9 NA |
1208 | if (btrfs_is_zoned(cache->fs_info)) { |
1209 | /* Migrate zone_unusable bytes to readonly */ | |
1210 | sinfo->bytes_readonly += cache->zone_unusable; | |
1211 | sinfo->bytes_zone_unusable -= cache->zone_unusable; | |
1212 | cache->zone_unusable = 0; | |
1213 | } | |
26ce2095 JB |
1214 | cache->ro++; |
1215 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); | |
26ce2095 JB |
1216 | } |
1217 | out: | |
1218 | spin_unlock(&cache->lock); | |
1219 | spin_unlock(&sinfo->lock); | |
1220 | if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { | |
1221 | btrfs_info(cache->fs_info, | |
b3470b5d | 1222 | "unable to make block group %llu ro", cache->start); |
26ce2095 JB |
1223 | btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0); |
1224 | } | |
1225 | return ret; | |
1226 | } | |
1227 | ||
fe119a6e NB |
1228 | static bool clean_pinned_extents(struct btrfs_trans_handle *trans, |
1229 | struct btrfs_block_group *bg) | |
45bb5d6a NB |
1230 | { |
1231 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
fe119a6e | 1232 | struct btrfs_transaction *prev_trans = NULL; |
45bb5d6a NB |
1233 | const u64 start = bg->start; |
1234 | const u64 end = start + bg->length - 1; | |
1235 | int ret; | |
1236 | ||
fe119a6e NB |
1237 | spin_lock(&fs_info->trans_lock); |
1238 | if (trans->transaction->list.prev != &fs_info->trans_list) { | |
1239 | prev_trans = list_last_entry(&trans->transaction->list, | |
1240 | struct btrfs_transaction, list); | |
1241 | refcount_inc(&prev_trans->use_count); | |
1242 | } | |
1243 | spin_unlock(&fs_info->trans_lock); | |
1244 | ||
45bb5d6a NB |
1245 | /* |
1246 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
1247 | * btrfs_finish_extent_commit(). If we are at transaction N, another | |
1248 | * task might be running finish_extent_commit() for the previous | |
1249 | * transaction N - 1, and have seen a range belonging to the block | |
fe119a6e NB |
1250 | * group in pinned_extents before we were able to clear the whole block |
1251 | * group range from pinned_extents. This means that task can lookup for | |
1252 | * the block group after we unpinned it from pinned_extents and removed | |
1253 | * it, leading to a BUG_ON() at unpin_extent_range(). | |
45bb5d6a NB |
1254 | */ |
1255 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
fe119a6e NB |
1256 | if (prev_trans) { |
1257 | ret = clear_extent_bits(&prev_trans->pinned_extents, start, end, | |
1258 | EXTENT_DIRTY); | |
1259 | if (ret) | |
534cf531 | 1260 | goto out; |
fe119a6e | 1261 | } |
45bb5d6a | 1262 | |
fe119a6e | 1263 | ret = clear_extent_bits(&trans->transaction->pinned_extents, start, end, |
45bb5d6a | 1264 | EXTENT_DIRTY); |
534cf531 | 1265 | out: |
45bb5d6a | 1266 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
5150bf19 FM |
1267 | if (prev_trans) |
1268 | btrfs_put_transaction(prev_trans); | |
45bb5d6a | 1269 | |
534cf531 | 1270 | return ret == 0; |
45bb5d6a NB |
1271 | } |
1272 | ||
e3e0520b JB |
1273 | /* |
1274 | * Process the unused_bgs list and remove any that don't have any allocated | |
1275 | * space inside of them. | |
1276 | */ | |
1277 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
1278 | { | |
32da5386 | 1279 | struct btrfs_block_group *block_group; |
e3e0520b JB |
1280 | struct btrfs_space_info *space_info; |
1281 | struct btrfs_trans_handle *trans; | |
6e80d4f8 | 1282 | const bool async_trim_enabled = btrfs_test_opt(fs_info, DISCARD_ASYNC); |
e3e0520b JB |
1283 | int ret = 0; |
1284 | ||
1285 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) | |
1286 | return; | |
1287 | ||
ddfd08cb JB |
1288 | /* |
1289 | * Long running balances can keep us blocked here for eternity, so | |
1290 | * simply skip deletion if we're unable to get the mutex. | |
1291 | */ | |
f3372065 | 1292 | if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) |
ddfd08cb JB |
1293 | return; |
1294 | ||
e3e0520b JB |
1295 | spin_lock(&fs_info->unused_bgs_lock); |
1296 | while (!list_empty(&fs_info->unused_bgs)) { | |
e3e0520b JB |
1297 | int trimming; |
1298 | ||
1299 | block_group = list_first_entry(&fs_info->unused_bgs, | |
32da5386 | 1300 | struct btrfs_block_group, |
e3e0520b JB |
1301 | bg_list); |
1302 | list_del_init(&block_group->bg_list); | |
1303 | ||
1304 | space_info = block_group->space_info; | |
1305 | ||
1306 | if (ret || btrfs_mixed_space_info(space_info)) { | |
1307 | btrfs_put_block_group(block_group); | |
1308 | continue; | |
1309 | } | |
1310 | spin_unlock(&fs_info->unused_bgs_lock); | |
1311 | ||
b0643e59 DZ |
1312 | btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); |
1313 | ||
e3e0520b JB |
1314 | /* Don't want to race with allocators so take the groups_sem */ |
1315 | down_write(&space_info->groups_sem); | |
6e80d4f8 DZ |
1316 | |
1317 | /* | |
1318 | * Async discard moves the final block group discard to be prior | |
1319 | * to the unused_bgs code path. Therefore, if it's not fully | |
1320 | * trimmed, punt it back to the async discard lists. | |
1321 | */ | |
1322 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC) && | |
1323 | !btrfs_is_free_space_trimmed(block_group)) { | |
1324 | trace_btrfs_skip_unused_block_group(block_group); | |
1325 | up_write(&space_info->groups_sem); | |
1326 | /* Requeue if we failed because of async discard */ | |
1327 | btrfs_discard_queue_work(&fs_info->discard_ctl, | |
1328 | block_group); | |
1329 | goto next; | |
1330 | } | |
1331 | ||
e3e0520b JB |
1332 | spin_lock(&block_group->lock); |
1333 | if (block_group->reserved || block_group->pinned || | |
bf38be65 | 1334 | block_group->used || block_group->ro || |
e3e0520b JB |
1335 | list_is_singular(&block_group->list)) { |
1336 | /* | |
1337 | * We want to bail if we made new allocations or have | |
1338 | * outstanding allocations in this block group. We do | |
1339 | * the ro check in case balance is currently acting on | |
1340 | * this block group. | |
1341 | */ | |
1342 | trace_btrfs_skip_unused_block_group(block_group); | |
1343 | spin_unlock(&block_group->lock); | |
1344 | up_write(&space_info->groups_sem); | |
1345 | goto next; | |
1346 | } | |
1347 | spin_unlock(&block_group->lock); | |
1348 | ||
1349 | /* We don't want to force the issue, only flip if it's ok. */ | |
e11c0406 | 1350 | ret = inc_block_group_ro(block_group, 0); |
e3e0520b JB |
1351 | up_write(&space_info->groups_sem); |
1352 | if (ret < 0) { | |
1353 | ret = 0; | |
1354 | goto next; | |
1355 | } | |
1356 | ||
1357 | /* | |
1358 | * Want to do this before we do anything else so we can recover | |
1359 | * properly if we fail to join the transaction. | |
1360 | */ | |
1361 | trans = btrfs_start_trans_remove_block_group(fs_info, | |
b3470b5d | 1362 | block_group->start); |
e3e0520b JB |
1363 | if (IS_ERR(trans)) { |
1364 | btrfs_dec_block_group_ro(block_group); | |
1365 | ret = PTR_ERR(trans); | |
1366 | goto next; | |
1367 | } | |
1368 | ||
1369 | /* | |
1370 | * We could have pending pinned extents for this block group, | |
1371 | * just delete them, we don't care about them anymore. | |
1372 | */ | |
534cf531 FM |
1373 | if (!clean_pinned_extents(trans, block_group)) { |
1374 | btrfs_dec_block_group_ro(block_group); | |
e3e0520b | 1375 | goto end_trans; |
534cf531 | 1376 | } |
e3e0520b | 1377 | |
b0643e59 DZ |
1378 | /* |
1379 | * At this point, the block_group is read only and should fail | |
1380 | * new allocations. However, btrfs_finish_extent_commit() can | |
1381 | * cause this block_group to be placed back on the discard | |
1382 | * lists because now the block_group isn't fully discarded. | |
1383 | * Bail here and try again later after discarding everything. | |
1384 | */ | |
1385 | spin_lock(&fs_info->discard_ctl.lock); | |
1386 | if (!list_empty(&block_group->discard_list)) { | |
1387 | spin_unlock(&fs_info->discard_ctl.lock); | |
1388 | btrfs_dec_block_group_ro(block_group); | |
1389 | btrfs_discard_queue_work(&fs_info->discard_ctl, | |
1390 | block_group); | |
1391 | goto end_trans; | |
1392 | } | |
1393 | spin_unlock(&fs_info->discard_ctl.lock); | |
1394 | ||
e3e0520b JB |
1395 | /* Reset pinned so btrfs_put_block_group doesn't complain */ |
1396 | spin_lock(&space_info->lock); | |
1397 | spin_lock(&block_group->lock); | |
1398 | ||
1399 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, | |
1400 | -block_group->pinned); | |
1401 | space_info->bytes_readonly += block_group->pinned; | |
e3e0520b JB |
1402 | block_group->pinned = 0; |
1403 | ||
1404 | spin_unlock(&block_group->lock); | |
1405 | spin_unlock(&space_info->lock); | |
1406 | ||
6e80d4f8 DZ |
1407 | /* |
1408 | * The normal path here is an unused block group is passed here, | |
1409 | * then trimming is handled in the transaction commit path. | |
1410 | * Async discard interposes before this to do the trimming | |
1411 | * before coming down the unused block group path as trimming | |
1412 | * will no longer be done later in the transaction commit path. | |
1413 | */ | |
1414 | if (!async_trim_enabled && btrfs_test_opt(fs_info, DISCARD_ASYNC)) | |
1415 | goto flip_async; | |
1416 | ||
dcba6e48 NA |
1417 | /* |
1418 | * DISCARD can flip during remount. On zoned filesystems, we | |
1419 | * need to reset sequential-required zones. | |
1420 | */ | |
1421 | trimming = btrfs_test_opt(fs_info, DISCARD_SYNC) || | |
1422 | btrfs_is_zoned(fs_info); | |
e3e0520b JB |
1423 | |
1424 | /* Implicit trim during transaction commit. */ | |
1425 | if (trimming) | |
6b7304af | 1426 | btrfs_freeze_block_group(block_group); |
e3e0520b JB |
1427 | |
1428 | /* | |
1429 | * Btrfs_remove_chunk will abort the transaction if things go | |
1430 | * horribly wrong. | |
1431 | */ | |
b3470b5d | 1432 | ret = btrfs_remove_chunk(trans, block_group->start); |
e3e0520b JB |
1433 | |
1434 | if (ret) { | |
1435 | if (trimming) | |
6b7304af | 1436 | btrfs_unfreeze_block_group(block_group); |
e3e0520b JB |
1437 | goto end_trans; |
1438 | } | |
1439 | ||
1440 | /* | |
1441 | * If we're not mounted with -odiscard, we can just forget | |
1442 | * about this block group. Otherwise we'll need to wait | |
1443 | * until transaction commit to do the actual discard. | |
1444 | */ | |
1445 | if (trimming) { | |
1446 | spin_lock(&fs_info->unused_bgs_lock); | |
1447 | /* | |
1448 | * A concurrent scrub might have added us to the list | |
1449 | * fs_info->unused_bgs, so use a list_move operation | |
1450 | * to add the block group to the deleted_bgs list. | |
1451 | */ | |
1452 | list_move(&block_group->bg_list, | |
1453 | &trans->transaction->deleted_bgs); | |
1454 | spin_unlock(&fs_info->unused_bgs_lock); | |
1455 | btrfs_get_block_group(block_group); | |
1456 | } | |
1457 | end_trans: | |
1458 | btrfs_end_transaction(trans); | |
1459 | next: | |
e3e0520b JB |
1460 | btrfs_put_block_group(block_group); |
1461 | spin_lock(&fs_info->unused_bgs_lock); | |
1462 | } | |
1463 | spin_unlock(&fs_info->unused_bgs_lock); | |
f3372065 | 1464 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
6e80d4f8 DZ |
1465 | return; |
1466 | ||
1467 | flip_async: | |
1468 | btrfs_end_transaction(trans); | |
f3372065 | 1469 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
6e80d4f8 DZ |
1470 | btrfs_put_block_group(block_group); |
1471 | btrfs_discard_punt_unused_bgs_list(fs_info); | |
e3e0520b JB |
1472 | } |
1473 | ||
32da5386 | 1474 | void btrfs_mark_bg_unused(struct btrfs_block_group *bg) |
e3e0520b JB |
1475 | { |
1476 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
1477 | ||
1478 | spin_lock(&fs_info->unused_bgs_lock); | |
1479 | if (list_empty(&bg->bg_list)) { | |
1480 | btrfs_get_block_group(bg); | |
1481 | trace_btrfs_add_unused_block_group(bg); | |
1482 | list_add_tail(&bg->bg_list, &fs_info->unused_bgs); | |
1483 | } | |
1484 | spin_unlock(&fs_info->unused_bgs_lock); | |
1485 | } | |
4358d963 | 1486 | |
18bb8bbf JT |
1487 | void btrfs_reclaim_bgs_work(struct work_struct *work) |
1488 | { | |
1489 | struct btrfs_fs_info *fs_info = | |
1490 | container_of(work, struct btrfs_fs_info, reclaim_bgs_work); | |
1491 | struct btrfs_block_group *bg; | |
1492 | struct btrfs_space_info *space_info; | |
1cea5cf0 | 1493 | LIST_HEAD(again_list); |
18bb8bbf JT |
1494 | |
1495 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) | |
1496 | return; | |
1497 | ||
1498 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) | |
1499 | return; | |
1500 | ||
9cc0b837 JT |
1501 | /* |
1502 | * Long running balances can keep us blocked here for eternity, so | |
1503 | * simply skip reclaim if we're unable to get the mutex. | |
1504 | */ | |
1505 | if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) { | |
1506 | btrfs_exclop_finish(fs_info); | |
1507 | return; | |
1508 | } | |
1509 | ||
18bb8bbf JT |
1510 | spin_lock(&fs_info->unused_bgs_lock); |
1511 | while (!list_empty(&fs_info->reclaim_bgs)) { | |
5f93e776 | 1512 | u64 zone_unusable; |
1cea5cf0 FM |
1513 | int ret = 0; |
1514 | ||
18bb8bbf JT |
1515 | bg = list_first_entry(&fs_info->reclaim_bgs, |
1516 | struct btrfs_block_group, | |
1517 | bg_list); | |
1518 | list_del_init(&bg->bg_list); | |
1519 | ||
1520 | space_info = bg->space_info; | |
1521 | spin_unlock(&fs_info->unused_bgs_lock); | |
1522 | ||
1523 | /* Don't race with allocators so take the groups_sem */ | |
1524 | down_write(&space_info->groups_sem); | |
1525 | ||
1526 | spin_lock(&bg->lock); | |
1527 | if (bg->reserved || bg->pinned || bg->ro) { | |
1528 | /* | |
1529 | * We want to bail if we made new allocations or have | |
1530 | * outstanding allocations in this block group. We do | |
1531 | * the ro check in case balance is currently acting on | |
1532 | * this block group. | |
1533 | */ | |
1534 | spin_unlock(&bg->lock); | |
1535 | up_write(&space_info->groups_sem); | |
1536 | goto next; | |
1537 | } | |
1538 | spin_unlock(&bg->lock); | |
1539 | ||
1540 | /* Get out fast, in case we're unmounting the filesystem */ | |
1541 | if (btrfs_fs_closing(fs_info)) { | |
1542 | up_write(&space_info->groups_sem); | |
1543 | goto next; | |
1544 | } | |
1545 | ||
5f93e776 JT |
1546 | /* |
1547 | * Cache the zone_unusable value before turning the block group | |
1548 | * to read only. As soon as the blog group is read only it's | |
1549 | * zone_unusable value gets moved to the block group's read-only | |
1550 | * bytes and isn't available for calculations anymore. | |
1551 | */ | |
1552 | zone_unusable = bg->zone_unusable; | |
18bb8bbf JT |
1553 | ret = inc_block_group_ro(bg, 0); |
1554 | up_write(&space_info->groups_sem); | |
1555 | if (ret < 0) | |
1556 | goto next; | |
1557 | ||
5f93e776 JT |
1558 | btrfs_info(fs_info, |
1559 | "reclaiming chunk %llu with %llu%% used %llu%% unusable", | |
1560 | bg->start, div_u64(bg->used * 100, bg->length), | |
1561 | div64_u64(zone_unusable * 100, bg->length)); | |
18bb8bbf JT |
1562 | trace_btrfs_reclaim_block_group(bg); |
1563 | ret = btrfs_relocate_chunk(fs_info, bg->start); | |
ba86dd9f | 1564 | if (ret && ret != -EAGAIN) |
18bb8bbf JT |
1565 | btrfs_err(fs_info, "error relocating chunk %llu", |
1566 | bg->start); | |
1567 | ||
1568 | next: | |
18bb8bbf | 1569 | spin_lock(&fs_info->unused_bgs_lock); |
1cea5cf0 FM |
1570 | if (ret == -EAGAIN && list_empty(&bg->bg_list)) |
1571 | list_add_tail(&bg->bg_list, &again_list); | |
1572 | else | |
1573 | btrfs_put_block_group(bg); | |
18bb8bbf | 1574 | } |
1cea5cf0 | 1575 | list_splice_tail(&again_list, &fs_info->reclaim_bgs); |
18bb8bbf JT |
1576 | spin_unlock(&fs_info->unused_bgs_lock); |
1577 | mutex_unlock(&fs_info->reclaim_bgs_lock); | |
1578 | btrfs_exclop_finish(fs_info); | |
1579 | } | |
1580 | ||
1581 | void btrfs_reclaim_bgs(struct btrfs_fs_info *fs_info) | |
1582 | { | |
1583 | spin_lock(&fs_info->unused_bgs_lock); | |
1584 | if (!list_empty(&fs_info->reclaim_bgs)) | |
1585 | queue_work(system_unbound_wq, &fs_info->reclaim_bgs_work); | |
1586 | spin_unlock(&fs_info->unused_bgs_lock); | |
1587 | } | |
1588 | ||
1589 | void btrfs_mark_bg_to_reclaim(struct btrfs_block_group *bg) | |
1590 | { | |
1591 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
1592 | ||
1593 | spin_lock(&fs_info->unused_bgs_lock); | |
1594 | if (list_empty(&bg->bg_list)) { | |
1595 | btrfs_get_block_group(bg); | |
1596 | trace_btrfs_add_reclaim_block_group(bg); | |
1597 | list_add_tail(&bg->bg_list, &fs_info->reclaim_bgs); | |
1598 | } | |
1599 | spin_unlock(&fs_info->unused_bgs_lock); | |
1600 | } | |
1601 | ||
e3ba67a1 JT |
1602 | static int read_bg_from_eb(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
1603 | struct btrfs_path *path) | |
1604 | { | |
1605 | struct extent_map_tree *em_tree; | |
1606 | struct extent_map *em; | |
1607 | struct btrfs_block_group_item bg; | |
1608 | struct extent_buffer *leaf; | |
1609 | int slot; | |
1610 | u64 flags; | |
1611 | int ret = 0; | |
1612 | ||
1613 | slot = path->slots[0]; | |
1614 | leaf = path->nodes[0]; | |
1615 | ||
1616 | em_tree = &fs_info->mapping_tree; | |
1617 | read_lock(&em_tree->lock); | |
1618 | em = lookup_extent_mapping(em_tree, key->objectid, key->offset); | |
1619 | read_unlock(&em_tree->lock); | |
1620 | if (!em) { | |
1621 | btrfs_err(fs_info, | |
1622 | "logical %llu len %llu found bg but no related chunk", | |
1623 | key->objectid, key->offset); | |
1624 | return -ENOENT; | |
1625 | } | |
1626 | ||
1627 | if (em->start != key->objectid || em->len != key->offset) { | |
1628 | btrfs_err(fs_info, | |
1629 | "block group %llu len %llu mismatch with chunk %llu len %llu", | |
1630 | key->objectid, key->offset, em->start, em->len); | |
1631 | ret = -EUCLEAN; | |
1632 | goto out_free_em; | |
1633 | } | |
1634 | ||
1635 | read_extent_buffer(leaf, &bg, btrfs_item_ptr_offset(leaf, slot), | |
1636 | sizeof(bg)); | |
1637 | flags = btrfs_stack_block_group_flags(&bg) & | |
1638 | BTRFS_BLOCK_GROUP_TYPE_MASK; | |
1639 | ||
1640 | if (flags != (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
1641 | btrfs_err(fs_info, | |
1642 | "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", | |
1643 | key->objectid, key->offset, flags, | |
1644 | (BTRFS_BLOCK_GROUP_TYPE_MASK & em->map_lookup->type)); | |
1645 | ret = -EUCLEAN; | |
1646 | } | |
1647 | ||
1648 | out_free_em: | |
1649 | free_extent_map(em); | |
1650 | return ret; | |
1651 | } | |
1652 | ||
4358d963 JB |
1653 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
1654 | struct btrfs_path *path, | |
1655 | struct btrfs_key *key) | |
1656 | { | |
1657 | struct btrfs_root *root = fs_info->extent_root; | |
e3ba67a1 | 1658 | int ret; |
4358d963 JB |
1659 | struct btrfs_key found_key; |
1660 | struct extent_buffer *leaf; | |
4358d963 JB |
1661 | int slot; |
1662 | ||
1663 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
1664 | if (ret < 0) | |
e3ba67a1 | 1665 | return ret; |
4358d963 JB |
1666 | |
1667 | while (1) { | |
1668 | slot = path->slots[0]; | |
1669 | leaf = path->nodes[0]; | |
1670 | if (slot >= btrfs_header_nritems(leaf)) { | |
1671 | ret = btrfs_next_leaf(root, path); | |
1672 | if (ret == 0) | |
1673 | continue; | |
1674 | if (ret < 0) | |
1675 | goto out; | |
1676 | break; | |
1677 | } | |
1678 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
1679 | ||
1680 | if (found_key.objectid >= key->objectid && | |
1681 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { | |
e3ba67a1 JT |
1682 | ret = read_bg_from_eb(fs_info, &found_key, path); |
1683 | break; | |
4358d963 | 1684 | } |
e3ba67a1 | 1685 | |
4358d963 JB |
1686 | path->slots[0]++; |
1687 | } | |
1688 | out: | |
1689 | return ret; | |
1690 | } | |
1691 | ||
1692 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
1693 | { | |
1694 | u64 extra_flags = chunk_to_extended(flags) & | |
1695 | BTRFS_EXTENDED_PROFILE_MASK; | |
1696 | ||
1697 | write_seqlock(&fs_info->profiles_lock); | |
1698 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
1699 | fs_info->avail_data_alloc_bits |= extra_flags; | |
1700 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
1701 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
1702 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
1703 | fs_info->avail_system_alloc_bits |= extra_flags; | |
1704 | write_sequnlock(&fs_info->profiles_lock); | |
1705 | } | |
1706 | ||
96a14336 | 1707 | /** |
9ee9b979 NB |
1708 | * Map a physical disk address to a list of logical addresses |
1709 | * | |
1710 | * @fs_info: the filesystem | |
96a14336 | 1711 | * @chunk_start: logical address of block group |
138082f3 | 1712 | * @bdev: physical device to resolve, can be NULL to indicate any device |
96a14336 NB |
1713 | * @physical: physical address to map to logical addresses |
1714 | * @logical: return array of logical addresses which map to @physical | |
1715 | * @naddrs: length of @logical | |
1716 | * @stripe_len: size of IO stripe for the given block group | |
1717 | * | |
1718 | * Maps a particular @physical disk address to a list of @logical addresses. | |
1719 | * Used primarily to exclude those portions of a block group that contain super | |
1720 | * block copies. | |
1721 | */ | |
96a14336 | 1722 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, |
138082f3 NA |
1723 | struct block_device *bdev, u64 physical, u64 **logical, |
1724 | int *naddrs, int *stripe_len) | |
96a14336 NB |
1725 | { |
1726 | struct extent_map *em; | |
1727 | struct map_lookup *map; | |
1728 | u64 *buf; | |
1729 | u64 bytenr; | |
1776ad17 NB |
1730 | u64 data_stripe_length; |
1731 | u64 io_stripe_size; | |
1732 | int i, nr = 0; | |
1733 | int ret = 0; | |
96a14336 NB |
1734 | |
1735 | em = btrfs_get_chunk_map(fs_info, chunk_start, 1); | |
1736 | if (IS_ERR(em)) | |
1737 | return -EIO; | |
1738 | ||
1739 | map = em->map_lookup; | |
9e22b925 | 1740 | data_stripe_length = em->orig_block_len; |
1776ad17 | 1741 | io_stripe_size = map->stripe_len; |
138082f3 | 1742 | chunk_start = em->start; |
96a14336 | 1743 | |
9e22b925 NB |
1744 | /* For RAID5/6 adjust to a full IO stripe length */ |
1745 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
1776ad17 | 1746 | io_stripe_size = map->stripe_len * nr_data_stripes(map); |
96a14336 NB |
1747 | |
1748 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); | |
1776ad17 NB |
1749 | if (!buf) { |
1750 | ret = -ENOMEM; | |
1751 | goto out; | |
1752 | } | |
96a14336 NB |
1753 | |
1754 | for (i = 0; i < map->num_stripes; i++) { | |
1776ad17 NB |
1755 | bool already_inserted = false; |
1756 | u64 stripe_nr; | |
138082f3 | 1757 | u64 offset; |
1776ad17 NB |
1758 | int j; |
1759 | ||
1760 | if (!in_range(physical, map->stripes[i].physical, | |
1761 | data_stripe_length)) | |
96a14336 NB |
1762 | continue; |
1763 | ||
138082f3 NA |
1764 | if (bdev && map->stripes[i].dev->bdev != bdev) |
1765 | continue; | |
1766 | ||
96a14336 | 1767 | stripe_nr = physical - map->stripes[i].physical; |
138082f3 | 1768 | stripe_nr = div64_u64_rem(stripe_nr, map->stripe_len, &offset); |
96a14336 NB |
1769 | |
1770 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
1771 | stripe_nr = stripe_nr * map->num_stripes + i; | |
1772 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); | |
1773 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
1774 | stripe_nr = stripe_nr * map->num_stripes + i; | |
1775 | } | |
1776 | /* | |
1777 | * The remaining case would be for RAID56, multiply by | |
1778 | * nr_data_stripes(). Alternatively, just use rmap_len below | |
1779 | * instead of map->stripe_len | |
1780 | */ | |
1781 | ||
138082f3 | 1782 | bytenr = chunk_start + stripe_nr * io_stripe_size + offset; |
1776ad17 NB |
1783 | |
1784 | /* Ensure we don't add duplicate addresses */ | |
96a14336 | 1785 | for (j = 0; j < nr; j++) { |
1776ad17 NB |
1786 | if (buf[j] == bytenr) { |
1787 | already_inserted = true; | |
96a14336 | 1788 | break; |
1776ad17 | 1789 | } |
96a14336 | 1790 | } |
1776ad17 NB |
1791 | |
1792 | if (!already_inserted) | |
96a14336 | 1793 | buf[nr++] = bytenr; |
96a14336 NB |
1794 | } |
1795 | ||
1796 | *logical = buf; | |
1797 | *naddrs = nr; | |
1776ad17 NB |
1798 | *stripe_len = io_stripe_size; |
1799 | out: | |
96a14336 | 1800 | free_extent_map(em); |
1776ad17 | 1801 | return ret; |
96a14336 NB |
1802 | } |
1803 | ||
32da5386 | 1804 | static int exclude_super_stripes(struct btrfs_block_group *cache) |
4358d963 JB |
1805 | { |
1806 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
12659251 | 1807 | const bool zoned = btrfs_is_zoned(fs_info); |
4358d963 JB |
1808 | u64 bytenr; |
1809 | u64 *logical; | |
1810 | int stripe_len; | |
1811 | int i, nr, ret; | |
1812 | ||
b3470b5d DS |
1813 | if (cache->start < BTRFS_SUPER_INFO_OFFSET) { |
1814 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->start; | |
4358d963 | 1815 | cache->bytes_super += stripe_len; |
b3470b5d | 1816 | ret = btrfs_add_excluded_extent(fs_info, cache->start, |
4358d963 JB |
1817 | stripe_len); |
1818 | if (ret) | |
1819 | return ret; | |
1820 | } | |
1821 | ||
1822 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
1823 | bytenr = btrfs_sb_offset(i); | |
138082f3 | 1824 | ret = btrfs_rmap_block(fs_info, cache->start, NULL, |
4358d963 JB |
1825 | bytenr, &logical, &nr, &stripe_len); |
1826 | if (ret) | |
1827 | return ret; | |
1828 | ||
12659251 NA |
1829 | /* Shouldn't have super stripes in sequential zones */ |
1830 | if (zoned && nr) { | |
1831 | btrfs_err(fs_info, | |
1832 | "zoned: block group %llu must not contain super block", | |
1833 | cache->start); | |
1834 | return -EUCLEAN; | |
1835 | } | |
1836 | ||
4358d963 | 1837 | while (nr--) { |
96f9b0f2 NB |
1838 | u64 len = min_t(u64, stripe_len, |
1839 | cache->start + cache->length - logical[nr]); | |
4358d963 JB |
1840 | |
1841 | cache->bytes_super += len; | |
96f9b0f2 NB |
1842 | ret = btrfs_add_excluded_extent(fs_info, logical[nr], |
1843 | len); | |
4358d963 JB |
1844 | if (ret) { |
1845 | kfree(logical); | |
1846 | return ret; | |
1847 | } | |
1848 | } | |
1849 | ||
1850 | kfree(logical); | |
1851 | } | |
1852 | return 0; | |
1853 | } | |
1854 | ||
32da5386 | 1855 | static void link_block_group(struct btrfs_block_group *cache) |
4358d963 JB |
1856 | { |
1857 | struct btrfs_space_info *space_info = cache->space_info; | |
1858 | int index = btrfs_bg_flags_to_raid_index(cache->flags); | |
4358d963 JB |
1859 | |
1860 | down_write(&space_info->groups_sem); | |
4358d963 JB |
1861 | list_add_tail(&cache->list, &space_info->block_groups[index]); |
1862 | up_write(&space_info->groups_sem); | |
4358d963 JB |
1863 | } |
1864 | ||
32da5386 | 1865 | static struct btrfs_block_group *btrfs_create_block_group_cache( |
9afc6649 | 1866 | struct btrfs_fs_info *fs_info, u64 start) |
4358d963 | 1867 | { |
32da5386 | 1868 | struct btrfs_block_group *cache; |
4358d963 JB |
1869 | |
1870 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
1871 | if (!cache) | |
1872 | return NULL; | |
1873 | ||
1874 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
1875 | GFP_NOFS); | |
1876 | if (!cache->free_space_ctl) { | |
1877 | kfree(cache); | |
1878 | return NULL; | |
1879 | } | |
1880 | ||
b3470b5d | 1881 | cache->start = start; |
4358d963 JB |
1882 | |
1883 | cache->fs_info = fs_info; | |
1884 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); | |
4358d963 | 1885 | |
6e80d4f8 DZ |
1886 | cache->discard_index = BTRFS_DISCARD_INDEX_UNUSED; |
1887 | ||
48aaeebe | 1888 | refcount_set(&cache->refs, 1); |
4358d963 JB |
1889 | spin_lock_init(&cache->lock); |
1890 | init_rwsem(&cache->data_rwsem); | |
1891 | INIT_LIST_HEAD(&cache->list); | |
1892 | INIT_LIST_HEAD(&cache->cluster_list); | |
1893 | INIT_LIST_HEAD(&cache->bg_list); | |
1894 | INIT_LIST_HEAD(&cache->ro_list); | |
b0643e59 | 1895 | INIT_LIST_HEAD(&cache->discard_list); |
4358d963 JB |
1896 | INIT_LIST_HEAD(&cache->dirty_list); |
1897 | INIT_LIST_HEAD(&cache->io_list); | |
cd79909b | 1898 | btrfs_init_free_space_ctl(cache, cache->free_space_ctl); |
6b7304af | 1899 | atomic_set(&cache->frozen, 0); |
4358d963 JB |
1900 | mutex_init(&cache->free_space_lock); |
1901 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); | |
1902 | ||
1903 | return cache; | |
1904 | } | |
1905 | ||
1906 | /* | |
1907 | * Iterate all chunks and verify that each of them has the corresponding block | |
1908 | * group | |
1909 | */ | |
1910 | static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) | |
1911 | { | |
1912 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; | |
1913 | struct extent_map *em; | |
32da5386 | 1914 | struct btrfs_block_group *bg; |
4358d963 JB |
1915 | u64 start = 0; |
1916 | int ret = 0; | |
1917 | ||
1918 | while (1) { | |
1919 | read_lock(&map_tree->lock); | |
1920 | /* | |
1921 | * lookup_extent_mapping will return the first extent map | |
1922 | * intersecting the range, so setting @len to 1 is enough to | |
1923 | * get the first chunk. | |
1924 | */ | |
1925 | em = lookup_extent_mapping(map_tree, start, 1); | |
1926 | read_unlock(&map_tree->lock); | |
1927 | if (!em) | |
1928 | break; | |
1929 | ||
1930 | bg = btrfs_lookup_block_group(fs_info, em->start); | |
1931 | if (!bg) { | |
1932 | btrfs_err(fs_info, | |
1933 | "chunk start=%llu len=%llu doesn't have corresponding block group", | |
1934 | em->start, em->len); | |
1935 | ret = -EUCLEAN; | |
1936 | free_extent_map(em); | |
1937 | break; | |
1938 | } | |
b3470b5d | 1939 | if (bg->start != em->start || bg->length != em->len || |
4358d963 JB |
1940 | (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != |
1941 | (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
1942 | btrfs_err(fs_info, | |
1943 | "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", | |
1944 | em->start, em->len, | |
1945 | em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, | |
b3470b5d | 1946 | bg->start, bg->length, |
4358d963 JB |
1947 | bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); |
1948 | ret = -EUCLEAN; | |
1949 | free_extent_map(em); | |
1950 | btrfs_put_block_group(bg); | |
1951 | break; | |
1952 | } | |
1953 | start = em->start + em->len; | |
1954 | free_extent_map(em); | |
1955 | btrfs_put_block_group(bg); | |
1956 | } | |
1957 | return ret; | |
1958 | } | |
1959 | ||
ffb9e0f0 | 1960 | static int read_one_block_group(struct btrfs_fs_info *info, |
4afd2fe8 | 1961 | struct btrfs_block_group_item *bgi, |
d49a2ddb | 1962 | const struct btrfs_key *key, |
ffb9e0f0 QW |
1963 | int need_clear) |
1964 | { | |
32da5386 | 1965 | struct btrfs_block_group *cache; |
ffb9e0f0 | 1966 | struct btrfs_space_info *space_info; |
ffb9e0f0 | 1967 | const bool mixed = btrfs_fs_incompat(info, MIXED_GROUPS); |
ffb9e0f0 QW |
1968 | int ret; |
1969 | ||
d49a2ddb | 1970 | ASSERT(key->type == BTRFS_BLOCK_GROUP_ITEM_KEY); |
ffb9e0f0 | 1971 | |
9afc6649 | 1972 | cache = btrfs_create_block_group_cache(info, key->objectid); |
ffb9e0f0 QW |
1973 | if (!cache) |
1974 | return -ENOMEM; | |
1975 | ||
4afd2fe8 JT |
1976 | cache->length = key->offset; |
1977 | cache->used = btrfs_stack_block_group_used(bgi); | |
1978 | cache->flags = btrfs_stack_block_group_flags(bgi); | |
9afc6649 | 1979 | |
e3e39c72 MPS |
1980 | set_free_space_tree_thresholds(cache); |
1981 | ||
ffb9e0f0 QW |
1982 | if (need_clear) { |
1983 | /* | |
1984 | * When we mount with old space cache, we need to | |
1985 | * set BTRFS_DC_CLEAR and set dirty flag. | |
1986 | * | |
1987 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
1988 | * truncate the old free space cache inode and | |
1989 | * setup a new one. | |
1990 | * b) Setting 'dirty flag' makes sure that we flush | |
1991 | * the new space cache info onto disk. | |
1992 | */ | |
1993 | if (btrfs_test_opt(info, SPACE_CACHE)) | |
1994 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
1995 | } | |
ffb9e0f0 QW |
1996 | if (!mixed && ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && |
1997 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
1998 | btrfs_err(info, | |
1999 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
2000 | cache->start); | |
2001 | ret = -EINVAL; | |
2002 | goto error; | |
2003 | } | |
2004 | ||
a94794d5 | 2005 | ret = btrfs_load_block_group_zone_info(cache, false); |
08e11a3d NA |
2006 | if (ret) { |
2007 | btrfs_err(info, "zoned: failed to load zone info of bg %llu", | |
2008 | cache->start); | |
2009 | goto error; | |
2010 | } | |
2011 | ||
ffb9e0f0 QW |
2012 | /* |
2013 | * We need to exclude the super stripes now so that the space info has | |
2014 | * super bytes accounted for, otherwise we'll think we have more space | |
2015 | * than we actually do. | |
2016 | */ | |
2017 | ret = exclude_super_stripes(cache); | |
2018 | if (ret) { | |
2019 | /* We may have excluded something, so call this just in case. */ | |
2020 | btrfs_free_excluded_extents(cache); | |
2021 | goto error; | |
2022 | } | |
2023 | ||
2024 | /* | |
169e0da9 NA |
2025 | * For zoned filesystem, space after the allocation offset is the only |
2026 | * free space for a block group. So, we don't need any caching work. | |
2027 | * btrfs_calc_zone_unusable() will set the amount of free space and | |
2028 | * zone_unusable space. | |
2029 | * | |
2030 | * For regular filesystem, check for two cases, either we are full, and | |
2031 | * therefore don't need to bother with the caching work since we won't | |
2032 | * find any space, or we are empty, and we can just add all the space | |
2033 | * in and be done with it. This saves us _a_lot_ of time, particularly | |
2034 | * in the full case. | |
ffb9e0f0 | 2035 | */ |
169e0da9 NA |
2036 | if (btrfs_is_zoned(info)) { |
2037 | btrfs_calc_zone_unusable(cache); | |
2038 | } else if (cache->length == cache->used) { | |
ffb9e0f0 QW |
2039 | cache->last_byte_to_unpin = (u64)-1; |
2040 | cache->cached = BTRFS_CACHE_FINISHED; | |
2041 | btrfs_free_excluded_extents(cache); | |
2042 | } else if (cache->used == 0) { | |
2043 | cache->last_byte_to_unpin = (u64)-1; | |
2044 | cache->cached = BTRFS_CACHE_FINISHED; | |
9afc6649 QW |
2045 | add_new_free_space(cache, cache->start, |
2046 | cache->start + cache->length); | |
ffb9e0f0 QW |
2047 | btrfs_free_excluded_extents(cache); |
2048 | } | |
2049 | ||
2050 | ret = btrfs_add_block_group_cache(info, cache); | |
2051 | if (ret) { | |
2052 | btrfs_remove_free_space_cache(cache); | |
2053 | goto error; | |
2054 | } | |
2055 | trace_btrfs_add_block_group(info, cache, 0); | |
9afc6649 | 2056 | btrfs_update_space_info(info, cache->flags, cache->length, |
169e0da9 NA |
2057 | cache->used, cache->bytes_super, |
2058 | cache->zone_unusable, &space_info); | |
ffb9e0f0 QW |
2059 | |
2060 | cache->space_info = space_info; | |
2061 | ||
2062 | link_block_group(cache); | |
2063 | ||
2064 | set_avail_alloc_bits(info, cache->flags); | |
2065 | if (btrfs_chunk_readonly(info, cache->start)) { | |
2066 | inc_block_group_ro(cache, 1); | |
2067 | } else if (cache->used == 0) { | |
2068 | ASSERT(list_empty(&cache->bg_list)); | |
6e80d4f8 DZ |
2069 | if (btrfs_test_opt(info, DISCARD_ASYNC)) |
2070 | btrfs_discard_queue_work(&info->discard_ctl, cache); | |
2071 | else | |
2072 | btrfs_mark_bg_unused(cache); | |
ffb9e0f0 QW |
2073 | } |
2074 | return 0; | |
2075 | error: | |
2076 | btrfs_put_block_group(cache); | |
2077 | return ret; | |
2078 | } | |
2079 | ||
42437a63 JB |
2080 | static int fill_dummy_bgs(struct btrfs_fs_info *fs_info) |
2081 | { | |
2082 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; | |
2083 | struct btrfs_space_info *space_info; | |
2084 | struct rb_node *node; | |
2085 | int ret = 0; | |
2086 | ||
2087 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { | |
2088 | struct extent_map *em; | |
2089 | struct map_lookup *map; | |
2090 | struct btrfs_block_group *bg; | |
2091 | ||
2092 | em = rb_entry(node, struct extent_map, rb_node); | |
2093 | map = em->map_lookup; | |
2094 | bg = btrfs_create_block_group_cache(fs_info, em->start); | |
2095 | if (!bg) { | |
2096 | ret = -ENOMEM; | |
2097 | break; | |
2098 | } | |
2099 | ||
2100 | /* Fill dummy cache as FULL */ | |
2101 | bg->length = em->len; | |
2102 | bg->flags = map->type; | |
2103 | bg->last_byte_to_unpin = (u64)-1; | |
2104 | bg->cached = BTRFS_CACHE_FINISHED; | |
2105 | bg->used = em->len; | |
2106 | bg->flags = map->type; | |
2107 | ret = btrfs_add_block_group_cache(fs_info, bg); | |
2b29726c QW |
2108 | /* |
2109 | * We may have some valid block group cache added already, in | |
2110 | * that case we skip to the next one. | |
2111 | */ | |
2112 | if (ret == -EEXIST) { | |
2113 | ret = 0; | |
2114 | btrfs_put_block_group(bg); | |
2115 | continue; | |
2116 | } | |
2117 | ||
42437a63 JB |
2118 | if (ret) { |
2119 | btrfs_remove_free_space_cache(bg); | |
2120 | btrfs_put_block_group(bg); | |
2121 | break; | |
2122 | } | |
2b29726c | 2123 | |
42437a63 | 2124 | btrfs_update_space_info(fs_info, bg->flags, em->len, em->len, |
169e0da9 | 2125 | 0, 0, &space_info); |
42437a63 JB |
2126 | bg->space_info = space_info; |
2127 | link_block_group(bg); | |
2128 | ||
2129 | set_avail_alloc_bits(fs_info, bg->flags); | |
2130 | } | |
2131 | if (!ret) | |
2132 | btrfs_init_global_block_rsv(fs_info); | |
2133 | return ret; | |
2134 | } | |
2135 | ||
4358d963 JB |
2136 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
2137 | { | |
2138 | struct btrfs_path *path; | |
2139 | int ret; | |
32da5386 | 2140 | struct btrfs_block_group *cache; |
4358d963 JB |
2141 | struct btrfs_space_info *space_info; |
2142 | struct btrfs_key key; | |
4358d963 JB |
2143 | int need_clear = 0; |
2144 | u64 cache_gen; | |
4358d963 | 2145 | |
42437a63 JB |
2146 | if (!info->extent_root) |
2147 | return fill_dummy_bgs(info); | |
2148 | ||
4358d963 JB |
2149 | key.objectid = 0; |
2150 | key.offset = 0; | |
2151 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
2152 | path = btrfs_alloc_path(); | |
2153 | if (!path) | |
2154 | return -ENOMEM; | |
4358d963 JB |
2155 | |
2156 | cache_gen = btrfs_super_cache_generation(info->super_copy); | |
2157 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
2158 | btrfs_super_generation(info->super_copy) != cache_gen) | |
2159 | need_clear = 1; | |
2160 | if (btrfs_test_opt(info, CLEAR_CACHE)) | |
2161 | need_clear = 1; | |
2162 | ||
2163 | while (1) { | |
4afd2fe8 JT |
2164 | struct btrfs_block_group_item bgi; |
2165 | struct extent_buffer *leaf; | |
2166 | int slot; | |
2167 | ||
4358d963 JB |
2168 | ret = find_first_block_group(info, path, &key); |
2169 | if (ret > 0) | |
2170 | break; | |
2171 | if (ret != 0) | |
2172 | goto error; | |
2173 | ||
4afd2fe8 JT |
2174 | leaf = path->nodes[0]; |
2175 | slot = path->slots[0]; | |
2176 | ||
2177 | read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot), | |
2178 | sizeof(bgi)); | |
2179 | ||
2180 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2181 | btrfs_release_path(path); | |
2182 | ret = read_one_block_group(info, &bgi, &key, need_clear); | |
ffb9e0f0 | 2183 | if (ret < 0) |
4358d963 | 2184 | goto error; |
ffb9e0f0 QW |
2185 | key.objectid += key.offset; |
2186 | key.offset = 0; | |
4358d963 | 2187 | } |
7837fa88 | 2188 | btrfs_release_path(path); |
4358d963 | 2189 | |
72804905 | 2190 | list_for_each_entry(space_info, &info->space_info, list) { |
49ea112d JB |
2191 | int i; |
2192 | ||
2193 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { | |
2194 | if (list_empty(&space_info->block_groups[i])) | |
2195 | continue; | |
2196 | cache = list_first_entry(&space_info->block_groups[i], | |
2197 | struct btrfs_block_group, | |
2198 | list); | |
2199 | btrfs_sysfs_add_block_group_type(cache); | |
2200 | } | |
2201 | ||
4358d963 JB |
2202 | if (!(btrfs_get_alloc_profile(info, space_info->flags) & |
2203 | (BTRFS_BLOCK_GROUP_RAID10 | | |
2204 | BTRFS_BLOCK_GROUP_RAID1_MASK | | |
2205 | BTRFS_BLOCK_GROUP_RAID56_MASK | | |
2206 | BTRFS_BLOCK_GROUP_DUP))) | |
2207 | continue; | |
2208 | /* | |
2209 | * Avoid allocating from un-mirrored block group if there are | |
2210 | * mirrored block groups. | |
2211 | */ | |
2212 | list_for_each_entry(cache, | |
2213 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
2214 | list) | |
e11c0406 | 2215 | inc_block_group_ro(cache, 1); |
4358d963 JB |
2216 | list_for_each_entry(cache, |
2217 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
2218 | list) | |
e11c0406 | 2219 | inc_block_group_ro(cache, 1); |
4358d963 JB |
2220 | } |
2221 | ||
2222 | btrfs_init_global_block_rsv(info); | |
2223 | ret = check_chunk_block_group_mappings(info); | |
2224 | error: | |
2225 | btrfs_free_path(path); | |
2b29726c QW |
2226 | /* |
2227 | * We've hit some error while reading the extent tree, and have | |
2228 | * rescue=ibadroots mount option. | |
2229 | * Try to fill the tree using dummy block groups so that the user can | |
2230 | * continue to mount and grab their data. | |
2231 | */ | |
2232 | if (ret && btrfs_test_opt(info, IGNOREBADROOTS)) | |
2233 | ret = fill_dummy_bgs(info); | |
4358d963 JB |
2234 | return ret; |
2235 | } | |
2236 | ||
79bd3712 FM |
2237 | /* |
2238 | * This function, insert_block_group_item(), belongs to the phase 2 of chunk | |
2239 | * allocation. | |
2240 | * | |
2241 | * See the comment at btrfs_chunk_alloc() for details about the chunk allocation | |
2242 | * phases. | |
2243 | */ | |
97f4728a QW |
2244 | static int insert_block_group_item(struct btrfs_trans_handle *trans, |
2245 | struct btrfs_block_group *block_group) | |
2246 | { | |
2247 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2248 | struct btrfs_block_group_item bgi; | |
2249 | struct btrfs_root *root; | |
2250 | struct btrfs_key key; | |
2251 | ||
2252 | spin_lock(&block_group->lock); | |
2253 | btrfs_set_stack_block_group_used(&bgi, block_group->used); | |
2254 | btrfs_set_stack_block_group_chunk_objectid(&bgi, | |
2255 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
2256 | btrfs_set_stack_block_group_flags(&bgi, block_group->flags); | |
2257 | key.objectid = block_group->start; | |
2258 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
2259 | key.offset = block_group->length; | |
2260 | spin_unlock(&block_group->lock); | |
2261 | ||
2262 | root = fs_info->extent_root; | |
2263 | return btrfs_insert_item(trans, root, &key, &bgi, sizeof(bgi)); | |
2264 | } | |
2265 | ||
2eadb9e7 NB |
2266 | static int insert_dev_extent(struct btrfs_trans_handle *trans, |
2267 | struct btrfs_device *device, u64 chunk_offset, | |
2268 | u64 start, u64 num_bytes) | |
2269 | { | |
2270 | struct btrfs_fs_info *fs_info = device->fs_info; | |
2271 | struct btrfs_root *root = fs_info->dev_root; | |
2272 | struct btrfs_path *path; | |
2273 | struct btrfs_dev_extent *extent; | |
2274 | struct extent_buffer *leaf; | |
2275 | struct btrfs_key key; | |
2276 | int ret; | |
2277 | ||
2278 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); | |
2279 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); | |
2280 | path = btrfs_alloc_path(); | |
2281 | if (!path) | |
2282 | return -ENOMEM; | |
2283 | ||
2284 | key.objectid = device->devid; | |
2285 | key.type = BTRFS_DEV_EXTENT_KEY; | |
2286 | key.offset = start; | |
2287 | ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*extent)); | |
2288 | if (ret) | |
2289 | goto out; | |
2290 | ||
2291 | leaf = path->nodes[0]; | |
2292 | extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent); | |
2293 | btrfs_set_dev_extent_chunk_tree(leaf, extent, BTRFS_CHUNK_TREE_OBJECTID); | |
2294 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, | |
2295 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
2296 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
2297 | ||
2298 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); | |
2299 | btrfs_mark_buffer_dirty(leaf); | |
2300 | out: | |
2301 | btrfs_free_path(path); | |
2302 | return ret; | |
2303 | } | |
2304 | ||
2305 | /* | |
2306 | * This function belongs to phase 2. | |
2307 | * | |
2308 | * See the comment at btrfs_chunk_alloc() for details about the chunk allocation | |
2309 | * phases. | |
2310 | */ | |
2311 | static int insert_dev_extents(struct btrfs_trans_handle *trans, | |
2312 | u64 chunk_offset, u64 chunk_size) | |
2313 | { | |
2314 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2315 | struct btrfs_device *device; | |
2316 | struct extent_map *em; | |
2317 | struct map_lookup *map; | |
2318 | u64 dev_offset; | |
2319 | u64 stripe_size; | |
2320 | int i; | |
2321 | int ret = 0; | |
2322 | ||
2323 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); | |
2324 | if (IS_ERR(em)) | |
2325 | return PTR_ERR(em); | |
2326 | ||
2327 | map = em->map_lookup; | |
2328 | stripe_size = em->orig_block_len; | |
2329 | ||
2330 | /* | |
2331 | * Take the device list mutex to prevent races with the final phase of | |
2332 | * a device replace operation that replaces the device object associated | |
2333 | * with the map's stripes, because the device object's id can change | |
2334 | * at any time during that final phase of the device replace operation | |
2335 | * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the | |
2336 | * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, | |
2337 | * resulting in persisting a device extent item with such ID. | |
2338 | */ | |
2339 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
2340 | for (i = 0; i < map->num_stripes; i++) { | |
2341 | device = map->stripes[i].dev; | |
2342 | dev_offset = map->stripes[i].physical; | |
2343 | ||
2344 | ret = insert_dev_extent(trans, device, chunk_offset, dev_offset, | |
2345 | stripe_size); | |
2346 | if (ret) | |
2347 | break; | |
2348 | } | |
2349 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
2350 | ||
2351 | free_extent_map(em); | |
2352 | return ret; | |
2353 | } | |
2354 | ||
79bd3712 FM |
2355 | /* |
2356 | * This function, btrfs_create_pending_block_groups(), belongs to the phase 2 of | |
2357 | * chunk allocation. | |
2358 | * | |
2359 | * See the comment at btrfs_chunk_alloc() for details about the chunk allocation | |
2360 | * phases. | |
2361 | */ | |
4358d963 JB |
2362 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
2363 | { | |
2364 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2365 | struct btrfs_block_group *block_group; |
4358d963 JB |
2366 | int ret = 0; |
2367 | ||
4358d963 | 2368 | while (!list_empty(&trans->new_bgs)) { |
49ea112d JB |
2369 | int index; |
2370 | ||
4358d963 | 2371 | block_group = list_first_entry(&trans->new_bgs, |
32da5386 | 2372 | struct btrfs_block_group, |
4358d963 JB |
2373 | bg_list); |
2374 | if (ret) | |
2375 | goto next; | |
2376 | ||
49ea112d JB |
2377 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
2378 | ||
97f4728a | 2379 | ret = insert_block_group_item(trans, block_group); |
4358d963 JB |
2380 | if (ret) |
2381 | btrfs_abort_transaction(trans, ret); | |
79bd3712 FM |
2382 | if (!block_group->chunk_item_inserted) { |
2383 | mutex_lock(&fs_info->chunk_mutex); | |
2384 | ret = btrfs_chunk_alloc_add_chunk_item(trans, block_group); | |
2385 | mutex_unlock(&fs_info->chunk_mutex); | |
2386 | if (ret) | |
2387 | btrfs_abort_transaction(trans, ret); | |
2388 | } | |
2eadb9e7 NB |
2389 | ret = insert_dev_extents(trans, block_group->start, |
2390 | block_group->length); | |
4358d963 JB |
2391 | if (ret) |
2392 | btrfs_abort_transaction(trans, ret); | |
2393 | add_block_group_free_space(trans, block_group); | |
49ea112d JB |
2394 | |
2395 | /* | |
2396 | * If we restriped during balance, we may have added a new raid | |
2397 | * type, so now add the sysfs entries when it is safe to do so. | |
2398 | * We don't have to worry about locking here as it's handled in | |
2399 | * btrfs_sysfs_add_block_group_type. | |
2400 | */ | |
2401 | if (block_group->space_info->block_group_kobjs[index] == NULL) | |
2402 | btrfs_sysfs_add_block_group_type(block_group); | |
2403 | ||
4358d963 JB |
2404 | /* Already aborted the transaction if it failed. */ |
2405 | next: | |
2406 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
2407 | list_del_init(&block_group->bg_list); | |
2408 | } | |
2409 | btrfs_trans_release_chunk_metadata(trans); | |
2410 | } | |
2411 | ||
79bd3712 FM |
2412 | struct btrfs_block_group *btrfs_make_block_group(struct btrfs_trans_handle *trans, |
2413 | u64 bytes_used, u64 type, | |
2414 | u64 chunk_offset, u64 size) | |
4358d963 JB |
2415 | { |
2416 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2417 | struct btrfs_block_group *cache; |
4358d963 JB |
2418 | int ret; |
2419 | ||
2420 | btrfs_set_log_full_commit(trans); | |
2421 | ||
9afc6649 | 2422 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset); |
4358d963 | 2423 | if (!cache) |
79bd3712 | 2424 | return ERR_PTR(-ENOMEM); |
4358d963 | 2425 | |
9afc6649 | 2426 | cache->length = size; |
e3e39c72 | 2427 | set_free_space_tree_thresholds(cache); |
bf38be65 | 2428 | cache->used = bytes_used; |
4358d963 JB |
2429 | cache->flags = type; |
2430 | cache->last_byte_to_unpin = (u64)-1; | |
2431 | cache->cached = BTRFS_CACHE_FINISHED; | |
997e3e2e BB |
2432 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
2433 | cache->needs_free_space = 1; | |
08e11a3d | 2434 | |
a94794d5 | 2435 | ret = btrfs_load_block_group_zone_info(cache, true); |
08e11a3d NA |
2436 | if (ret) { |
2437 | btrfs_put_block_group(cache); | |
79bd3712 | 2438 | return ERR_PTR(ret); |
08e11a3d NA |
2439 | } |
2440 | ||
4358d963 JB |
2441 | ret = exclude_super_stripes(cache); |
2442 | if (ret) { | |
2443 | /* We may have excluded something, so call this just in case */ | |
2444 | btrfs_free_excluded_extents(cache); | |
2445 | btrfs_put_block_group(cache); | |
79bd3712 | 2446 | return ERR_PTR(ret); |
4358d963 JB |
2447 | } |
2448 | ||
2449 | add_new_free_space(cache, chunk_offset, chunk_offset + size); | |
2450 | ||
2451 | btrfs_free_excluded_extents(cache); | |
2452 | ||
2453 | #ifdef CONFIG_BTRFS_DEBUG | |
2454 | if (btrfs_should_fragment_free_space(cache)) { | |
2455 | u64 new_bytes_used = size - bytes_used; | |
2456 | ||
2457 | bytes_used += new_bytes_used >> 1; | |
e11c0406 | 2458 | fragment_free_space(cache); |
4358d963 JB |
2459 | } |
2460 | #endif | |
2461 | /* | |
2462 | * Ensure the corresponding space_info object is created and | |
2463 | * assigned to our block group. We want our bg to be added to the rbtree | |
2464 | * with its ->space_info set. | |
2465 | */ | |
2466 | cache->space_info = btrfs_find_space_info(fs_info, cache->flags); | |
2467 | ASSERT(cache->space_info); | |
2468 | ||
2469 | ret = btrfs_add_block_group_cache(fs_info, cache); | |
2470 | if (ret) { | |
2471 | btrfs_remove_free_space_cache(cache); | |
2472 | btrfs_put_block_group(cache); | |
79bd3712 | 2473 | return ERR_PTR(ret); |
4358d963 JB |
2474 | } |
2475 | ||
2476 | /* | |
2477 | * Now that our block group has its ->space_info set and is inserted in | |
2478 | * the rbtree, update the space info's counters. | |
2479 | */ | |
2480 | trace_btrfs_add_block_group(fs_info, cache, 1); | |
2481 | btrfs_update_space_info(fs_info, cache->flags, size, bytes_used, | |
169e0da9 | 2482 | cache->bytes_super, 0, &cache->space_info); |
4358d963 JB |
2483 | btrfs_update_global_block_rsv(fs_info); |
2484 | ||
2485 | link_block_group(cache); | |
2486 | ||
2487 | list_add_tail(&cache->bg_list, &trans->new_bgs); | |
2488 | trans->delayed_ref_updates++; | |
2489 | btrfs_update_delayed_refs_rsv(trans); | |
2490 | ||
2491 | set_avail_alloc_bits(fs_info, type); | |
79bd3712 | 2492 | return cache; |
4358d963 | 2493 | } |
26ce2095 | 2494 | |
b12de528 QW |
2495 | /* |
2496 | * Mark one block group RO, can be called several times for the same block | |
2497 | * group. | |
2498 | * | |
2499 | * @cache: the destination block group | |
2500 | * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to | |
2501 | * ensure we still have some free space after marking this | |
2502 | * block group RO. | |
2503 | */ | |
2504 | int btrfs_inc_block_group_ro(struct btrfs_block_group *cache, | |
2505 | bool do_chunk_alloc) | |
26ce2095 JB |
2506 | { |
2507 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
2508 | struct btrfs_trans_handle *trans; | |
2509 | u64 alloc_flags; | |
2510 | int ret; | |
b6e9f16c | 2511 | bool dirty_bg_running; |
26ce2095 | 2512 | |
b6e9f16c NB |
2513 | do { |
2514 | trans = btrfs_join_transaction(fs_info->extent_root); | |
2515 | if (IS_ERR(trans)) | |
2516 | return PTR_ERR(trans); | |
26ce2095 | 2517 | |
b6e9f16c | 2518 | dirty_bg_running = false; |
26ce2095 | 2519 | |
b6e9f16c NB |
2520 | /* |
2521 | * We're not allowed to set block groups readonly after the dirty | |
2522 | * block group cache has started writing. If it already started, | |
2523 | * back off and let this transaction commit. | |
2524 | */ | |
2525 | mutex_lock(&fs_info->ro_block_group_mutex); | |
2526 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { | |
2527 | u64 transid = trans->transid; | |
26ce2095 | 2528 | |
b6e9f16c NB |
2529 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2530 | btrfs_end_transaction(trans); | |
2531 | ||
2532 | ret = btrfs_wait_for_commit(fs_info, transid); | |
2533 | if (ret) | |
2534 | return ret; | |
2535 | dirty_bg_running = true; | |
2536 | } | |
2537 | } while (dirty_bg_running); | |
26ce2095 | 2538 | |
b12de528 | 2539 | if (do_chunk_alloc) { |
26ce2095 | 2540 | /* |
b12de528 QW |
2541 | * If we are changing raid levels, try to allocate a |
2542 | * corresponding block group with the new raid level. | |
26ce2095 | 2543 | */ |
349e120e | 2544 | alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); |
b12de528 QW |
2545 | if (alloc_flags != cache->flags) { |
2546 | ret = btrfs_chunk_alloc(trans, alloc_flags, | |
2547 | CHUNK_ALLOC_FORCE); | |
2548 | /* | |
2549 | * ENOSPC is allowed here, we may have enough space | |
2550 | * already allocated at the new raid level to carry on | |
2551 | */ | |
2552 | if (ret == -ENOSPC) | |
2553 | ret = 0; | |
2554 | if (ret < 0) | |
2555 | goto out; | |
2556 | } | |
26ce2095 JB |
2557 | } |
2558 | ||
a7a63acc | 2559 | ret = inc_block_group_ro(cache, 0); |
195a49ea | 2560 | if (!do_chunk_alloc || ret == -ETXTBSY) |
b12de528 | 2561 | goto unlock_out; |
26ce2095 JB |
2562 | if (!ret) |
2563 | goto out; | |
2564 | alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags); | |
2565 | ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); | |
2566 | if (ret < 0) | |
2567 | goto out; | |
e11c0406 | 2568 | ret = inc_block_group_ro(cache, 0); |
195a49ea FM |
2569 | if (ret == -ETXTBSY) |
2570 | goto unlock_out; | |
26ce2095 JB |
2571 | out: |
2572 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { | |
349e120e | 2573 | alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); |
26ce2095 JB |
2574 | mutex_lock(&fs_info->chunk_mutex); |
2575 | check_system_chunk(trans, alloc_flags); | |
2576 | mutex_unlock(&fs_info->chunk_mutex); | |
2577 | } | |
b12de528 | 2578 | unlock_out: |
26ce2095 JB |
2579 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2580 | ||
2581 | btrfs_end_transaction(trans); | |
2582 | return ret; | |
2583 | } | |
2584 | ||
32da5386 | 2585 | void btrfs_dec_block_group_ro(struct btrfs_block_group *cache) |
26ce2095 JB |
2586 | { |
2587 | struct btrfs_space_info *sinfo = cache->space_info; | |
2588 | u64 num_bytes; | |
2589 | ||
2590 | BUG_ON(!cache->ro); | |
2591 | ||
2592 | spin_lock(&sinfo->lock); | |
2593 | spin_lock(&cache->lock); | |
2594 | if (!--cache->ro) { | |
169e0da9 NA |
2595 | if (btrfs_is_zoned(cache->fs_info)) { |
2596 | /* Migrate zone_unusable bytes back */ | |
2597 | cache->zone_unusable = cache->alloc_offset - cache->used; | |
2598 | sinfo->bytes_zone_unusable += cache->zone_unusable; | |
2599 | sinfo->bytes_readonly -= cache->zone_unusable; | |
2600 | } | |
f9f28e5b NA |
2601 | num_bytes = cache->length - cache->reserved - |
2602 | cache->pinned - cache->bytes_super - | |
2603 | cache->zone_unusable - cache->used; | |
2604 | sinfo->bytes_readonly -= num_bytes; | |
26ce2095 JB |
2605 | list_del_init(&cache->ro_list); |
2606 | } | |
2607 | spin_unlock(&cache->lock); | |
2608 | spin_unlock(&sinfo->lock); | |
2609 | } | |
77745c05 | 2610 | |
3be4d8ef QW |
2611 | static int update_block_group_item(struct btrfs_trans_handle *trans, |
2612 | struct btrfs_path *path, | |
2613 | struct btrfs_block_group *cache) | |
77745c05 JB |
2614 | { |
2615 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2616 | int ret; | |
3be4d8ef | 2617 | struct btrfs_root *root = fs_info->extent_root; |
77745c05 JB |
2618 | unsigned long bi; |
2619 | struct extent_buffer *leaf; | |
bf38be65 | 2620 | struct btrfs_block_group_item bgi; |
b3470b5d DS |
2621 | struct btrfs_key key; |
2622 | ||
2623 | key.objectid = cache->start; | |
2624 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
2625 | key.offset = cache->length; | |
77745c05 | 2626 | |
3be4d8ef | 2627 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
77745c05 JB |
2628 | if (ret) { |
2629 | if (ret > 0) | |
2630 | ret = -ENOENT; | |
2631 | goto fail; | |
2632 | } | |
2633 | ||
2634 | leaf = path->nodes[0]; | |
2635 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
de0dc456 DS |
2636 | btrfs_set_stack_block_group_used(&bgi, cache->used); |
2637 | btrfs_set_stack_block_group_chunk_objectid(&bgi, | |
3d976388 | 2638 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); |
de0dc456 | 2639 | btrfs_set_stack_block_group_flags(&bgi, cache->flags); |
bf38be65 | 2640 | write_extent_buffer(leaf, &bgi, bi, sizeof(bgi)); |
77745c05 JB |
2641 | btrfs_mark_buffer_dirty(leaf); |
2642 | fail: | |
2643 | btrfs_release_path(path); | |
2644 | return ret; | |
2645 | ||
2646 | } | |
2647 | ||
32da5386 | 2648 | static int cache_save_setup(struct btrfs_block_group *block_group, |
77745c05 JB |
2649 | struct btrfs_trans_handle *trans, |
2650 | struct btrfs_path *path) | |
2651 | { | |
2652 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
2653 | struct btrfs_root *root = fs_info->tree_root; | |
2654 | struct inode *inode = NULL; | |
2655 | struct extent_changeset *data_reserved = NULL; | |
2656 | u64 alloc_hint = 0; | |
2657 | int dcs = BTRFS_DC_ERROR; | |
0044ae11 | 2658 | u64 cache_size = 0; |
77745c05 JB |
2659 | int retries = 0; |
2660 | int ret = 0; | |
2661 | ||
af456a2c BB |
2662 | if (!btrfs_test_opt(fs_info, SPACE_CACHE)) |
2663 | return 0; | |
2664 | ||
77745c05 JB |
2665 | /* |
2666 | * If this block group is smaller than 100 megs don't bother caching the | |
2667 | * block group. | |
2668 | */ | |
b3470b5d | 2669 | if (block_group->length < (100 * SZ_1M)) { |
77745c05 JB |
2670 | spin_lock(&block_group->lock); |
2671 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2672 | spin_unlock(&block_group->lock); | |
2673 | return 0; | |
2674 | } | |
2675 | ||
bf31f87f | 2676 | if (TRANS_ABORTED(trans)) |
77745c05 JB |
2677 | return 0; |
2678 | again: | |
2679 | inode = lookup_free_space_inode(block_group, path); | |
2680 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2681 | ret = PTR_ERR(inode); | |
2682 | btrfs_release_path(path); | |
2683 | goto out; | |
2684 | } | |
2685 | ||
2686 | if (IS_ERR(inode)) { | |
2687 | BUG_ON(retries); | |
2688 | retries++; | |
2689 | ||
2690 | if (block_group->ro) | |
2691 | goto out_free; | |
2692 | ||
2693 | ret = create_free_space_inode(trans, block_group, path); | |
2694 | if (ret) | |
2695 | goto out_free; | |
2696 | goto again; | |
2697 | } | |
2698 | ||
2699 | /* | |
2700 | * We want to set the generation to 0, that way if anything goes wrong | |
2701 | * from here on out we know not to trust this cache when we load up next | |
2702 | * time. | |
2703 | */ | |
2704 | BTRFS_I(inode)->generation = 0; | |
9a56fcd1 | 2705 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
77745c05 JB |
2706 | if (ret) { |
2707 | /* | |
2708 | * So theoretically we could recover from this, simply set the | |
2709 | * super cache generation to 0 so we know to invalidate the | |
2710 | * cache, but then we'd have to keep track of the block groups | |
2711 | * that fail this way so we know we _have_ to reset this cache | |
2712 | * before the next commit or risk reading stale cache. So to | |
2713 | * limit our exposure to horrible edge cases lets just abort the | |
2714 | * transaction, this only happens in really bad situations | |
2715 | * anyway. | |
2716 | */ | |
2717 | btrfs_abort_transaction(trans, ret); | |
2718 | goto out_put; | |
2719 | } | |
2720 | WARN_ON(ret); | |
2721 | ||
2722 | /* We've already setup this transaction, go ahead and exit */ | |
2723 | if (block_group->cache_generation == trans->transid && | |
2724 | i_size_read(inode)) { | |
2725 | dcs = BTRFS_DC_SETUP; | |
2726 | goto out_put; | |
2727 | } | |
2728 | ||
2729 | if (i_size_read(inode) > 0) { | |
2730 | ret = btrfs_check_trunc_cache_free_space(fs_info, | |
2731 | &fs_info->global_block_rsv); | |
2732 | if (ret) | |
2733 | goto out_put; | |
2734 | ||
2735 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); | |
2736 | if (ret) | |
2737 | goto out_put; | |
2738 | } | |
2739 | ||
2740 | spin_lock(&block_group->lock); | |
2741 | if (block_group->cached != BTRFS_CACHE_FINISHED || | |
2742 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { | |
2743 | /* | |
2744 | * don't bother trying to write stuff out _if_ | |
2745 | * a) we're not cached, | |
2746 | * b) we're with nospace_cache mount option, | |
2747 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
2748 | */ | |
2749 | dcs = BTRFS_DC_WRITTEN; | |
2750 | spin_unlock(&block_group->lock); | |
2751 | goto out_put; | |
2752 | } | |
2753 | spin_unlock(&block_group->lock); | |
2754 | ||
2755 | /* | |
2756 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
2757 | * skip doing the setup, we've already cleared the cache so we're safe. | |
2758 | */ | |
2759 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
2760 | ret = -ENOSPC; | |
2761 | goto out_put; | |
2762 | } | |
2763 | ||
2764 | /* | |
2765 | * Try to preallocate enough space based on how big the block group is. | |
2766 | * Keep in mind this has to include any pinned space which could end up | |
2767 | * taking up quite a bit since it's not folded into the other space | |
2768 | * cache. | |
2769 | */ | |
0044ae11 QW |
2770 | cache_size = div_u64(block_group->length, SZ_256M); |
2771 | if (!cache_size) | |
2772 | cache_size = 1; | |
77745c05 | 2773 | |
0044ae11 QW |
2774 | cache_size *= 16; |
2775 | cache_size *= fs_info->sectorsize; | |
77745c05 | 2776 | |
36ea6f3e | 2777 | ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved, 0, |
0044ae11 | 2778 | cache_size); |
77745c05 JB |
2779 | if (ret) |
2780 | goto out_put; | |
2781 | ||
0044ae11 QW |
2782 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, cache_size, |
2783 | cache_size, cache_size, | |
77745c05 JB |
2784 | &alloc_hint); |
2785 | /* | |
2786 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
2787 | * of metadata or split extents when writing the cache out, which means | |
2788 | * we can enospc if we are heavily fragmented in addition to just normal | |
2789 | * out of space conditions. So if we hit this just skip setting up any | |
2790 | * other block groups for this transaction, maybe we'll unpin enough | |
2791 | * space the next time around. | |
2792 | */ | |
2793 | if (!ret) | |
2794 | dcs = BTRFS_DC_SETUP; | |
2795 | else if (ret == -ENOSPC) | |
2796 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
2797 | ||
2798 | out_put: | |
2799 | iput(inode); | |
2800 | out_free: | |
2801 | btrfs_release_path(path); | |
2802 | out: | |
2803 | spin_lock(&block_group->lock); | |
2804 | if (!ret && dcs == BTRFS_DC_SETUP) | |
2805 | block_group->cache_generation = trans->transid; | |
2806 | block_group->disk_cache_state = dcs; | |
2807 | spin_unlock(&block_group->lock); | |
2808 | ||
2809 | extent_changeset_free(data_reserved); | |
2810 | return ret; | |
2811 | } | |
2812 | ||
2813 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) | |
2814 | { | |
2815 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2816 | struct btrfs_block_group *cache, *tmp; |
77745c05 JB |
2817 | struct btrfs_transaction *cur_trans = trans->transaction; |
2818 | struct btrfs_path *path; | |
2819 | ||
2820 | if (list_empty(&cur_trans->dirty_bgs) || | |
2821 | !btrfs_test_opt(fs_info, SPACE_CACHE)) | |
2822 | return 0; | |
2823 | ||
2824 | path = btrfs_alloc_path(); | |
2825 | if (!path) | |
2826 | return -ENOMEM; | |
2827 | ||
2828 | /* Could add new block groups, use _safe just in case */ | |
2829 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
2830 | dirty_list) { | |
2831 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
2832 | cache_save_setup(cache, trans, path); | |
2833 | } | |
2834 | ||
2835 | btrfs_free_path(path); | |
2836 | return 0; | |
2837 | } | |
2838 | ||
2839 | /* | |
2840 | * Transaction commit does final block group cache writeback during a critical | |
2841 | * section where nothing is allowed to change the FS. This is required in | |
2842 | * order for the cache to actually match the block group, but can introduce a | |
2843 | * lot of latency into the commit. | |
2844 | * | |
2845 | * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO. | |
2846 | * There's a chance we'll have to redo some of it if the block group changes | |
2847 | * again during the commit, but it greatly reduces the commit latency by | |
2848 | * getting rid of the easy block groups while we're still allowing others to | |
2849 | * join the commit. | |
2850 | */ | |
2851 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) | |
2852 | { | |
2853 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2854 | struct btrfs_block_group *cache; |
77745c05 JB |
2855 | struct btrfs_transaction *cur_trans = trans->transaction; |
2856 | int ret = 0; | |
2857 | int should_put; | |
2858 | struct btrfs_path *path = NULL; | |
2859 | LIST_HEAD(dirty); | |
2860 | struct list_head *io = &cur_trans->io_bgs; | |
2861 | int num_started = 0; | |
2862 | int loops = 0; | |
2863 | ||
2864 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2865 | if (list_empty(&cur_trans->dirty_bgs)) { | |
2866 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2867 | return 0; | |
2868 | } | |
2869 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
2870 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2871 | ||
2872 | again: | |
2873 | /* Make sure all the block groups on our dirty list actually exist */ | |
2874 | btrfs_create_pending_block_groups(trans); | |
2875 | ||
2876 | if (!path) { | |
2877 | path = btrfs_alloc_path(); | |
938fcbfb JB |
2878 | if (!path) { |
2879 | ret = -ENOMEM; | |
2880 | goto out; | |
2881 | } | |
77745c05 JB |
2882 | } |
2883 | ||
2884 | /* | |
2885 | * cache_write_mutex is here only to save us from balance or automatic | |
2886 | * removal of empty block groups deleting this block group while we are | |
2887 | * writing out the cache | |
2888 | */ | |
2889 | mutex_lock(&trans->transaction->cache_write_mutex); | |
2890 | while (!list_empty(&dirty)) { | |
2891 | bool drop_reserve = true; | |
2892 | ||
32da5386 | 2893 | cache = list_first_entry(&dirty, struct btrfs_block_group, |
77745c05 JB |
2894 | dirty_list); |
2895 | /* | |
2896 | * This can happen if something re-dirties a block group that | |
2897 | * is already under IO. Just wait for it to finish and then do | |
2898 | * it all again | |
2899 | */ | |
2900 | if (!list_empty(&cache->io_list)) { | |
2901 | list_del_init(&cache->io_list); | |
2902 | btrfs_wait_cache_io(trans, cache, path); | |
2903 | btrfs_put_block_group(cache); | |
2904 | } | |
2905 | ||
2906 | ||
2907 | /* | |
2908 | * btrfs_wait_cache_io uses the cache->dirty_list to decide if | |
2909 | * it should update the cache_state. Don't delete until after | |
2910 | * we wait. | |
2911 | * | |
2912 | * Since we're not running in the commit critical section | |
2913 | * we need the dirty_bgs_lock to protect from update_block_group | |
2914 | */ | |
2915 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2916 | list_del_init(&cache->dirty_list); | |
2917 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2918 | ||
2919 | should_put = 1; | |
2920 | ||
2921 | cache_save_setup(cache, trans, path); | |
2922 | ||
2923 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
2924 | cache->io_ctl.inode = NULL; | |
2925 | ret = btrfs_write_out_cache(trans, cache, path); | |
2926 | if (ret == 0 && cache->io_ctl.inode) { | |
2927 | num_started++; | |
2928 | should_put = 0; | |
2929 | ||
2930 | /* | |
2931 | * The cache_write_mutex is protecting the | |
2932 | * io_list, also refer to the definition of | |
2933 | * btrfs_transaction::io_bgs for more details | |
2934 | */ | |
2935 | list_add_tail(&cache->io_list, io); | |
2936 | } else { | |
2937 | /* | |
2938 | * If we failed to write the cache, the | |
2939 | * generation will be bad and life goes on | |
2940 | */ | |
2941 | ret = 0; | |
2942 | } | |
2943 | } | |
2944 | if (!ret) { | |
3be4d8ef | 2945 | ret = update_block_group_item(trans, path, cache); |
77745c05 JB |
2946 | /* |
2947 | * Our block group might still be attached to the list | |
2948 | * of new block groups in the transaction handle of some | |
2949 | * other task (struct btrfs_trans_handle->new_bgs). This | |
2950 | * means its block group item isn't yet in the extent | |
2951 | * tree. If this happens ignore the error, as we will | |
2952 | * try again later in the critical section of the | |
2953 | * transaction commit. | |
2954 | */ | |
2955 | if (ret == -ENOENT) { | |
2956 | ret = 0; | |
2957 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2958 | if (list_empty(&cache->dirty_list)) { | |
2959 | list_add_tail(&cache->dirty_list, | |
2960 | &cur_trans->dirty_bgs); | |
2961 | btrfs_get_block_group(cache); | |
2962 | drop_reserve = false; | |
2963 | } | |
2964 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2965 | } else if (ret) { | |
2966 | btrfs_abort_transaction(trans, ret); | |
2967 | } | |
2968 | } | |
2969 | ||
2970 | /* If it's not on the io list, we need to put the block group */ | |
2971 | if (should_put) | |
2972 | btrfs_put_block_group(cache); | |
2973 | if (drop_reserve) | |
2974 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
77745c05 JB |
2975 | /* |
2976 | * Avoid blocking other tasks for too long. It might even save | |
2977 | * us from writing caches for block groups that are going to be | |
2978 | * removed. | |
2979 | */ | |
2980 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
938fcbfb JB |
2981 | if (ret) |
2982 | goto out; | |
77745c05 JB |
2983 | mutex_lock(&trans->transaction->cache_write_mutex); |
2984 | } | |
2985 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
2986 | ||
2987 | /* | |
2988 | * Go through delayed refs for all the stuff we've just kicked off | |
2989 | * and then loop back (just once) | |
2990 | */ | |
34d1eb0e JB |
2991 | if (!ret) |
2992 | ret = btrfs_run_delayed_refs(trans, 0); | |
77745c05 JB |
2993 | if (!ret && loops == 0) { |
2994 | loops++; | |
2995 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2996 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
2997 | /* | |
2998 | * dirty_bgs_lock protects us from concurrent block group | |
2999 | * deletes too (not just cache_write_mutex). | |
3000 | */ | |
3001 | if (!list_empty(&dirty)) { | |
3002 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3003 | goto again; | |
3004 | } | |
3005 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
938fcbfb JB |
3006 | } |
3007 | out: | |
3008 | if (ret < 0) { | |
3009 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3010 | list_splice_init(&dirty, &cur_trans->dirty_bgs); | |
3011 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
77745c05 JB |
3012 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
3013 | } | |
3014 | ||
3015 | btrfs_free_path(path); | |
3016 | return ret; | |
3017 | } | |
3018 | ||
3019 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) | |
3020 | { | |
3021 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 3022 | struct btrfs_block_group *cache; |
77745c05 JB |
3023 | struct btrfs_transaction *cur_trans = trans->transaction; |
3024 | int ret = 0; | |
3025 | int should_put; | |
3026 | struct btrfs_path *path; | |
3027 | struct list_head *io = &cur_trans->io_bgs; | |
3028 | int num_started = 0; | |
3029 | ||
3030 | path = btrfs_alloc_path(); | |
3031 | if (!path) | |
3032 | return -ENOMEM; | |
3033 | ||
3034 | /* | |
3035 | * Even though we are in the critical section of the transaction commit, | |
3036 | * we can still have concurrent tasks adding elements to this | |
3037 | * transaction's list of dirty block groups. These tasks correspond to | |
3038 | * endio free space workers started when writeback finishes for a | |
3039 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3040 | * allocate new block groups as a result of COWing nodes of the root | |
3041 | * tree when updating the free space inode. The writeback for the space | |
3042 | * caches is triggered by an earlier call to | |
3043 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3044 | * loop. | |
3045 | * Also we want to do the cache_save_setup first and then run the | |
3046 | * delayed refs to make sure we have the best chance at doing this all | |
3047 | * in one shot. | |
3048 | */ | |
3049 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3050 | while (!list_empty(&cur_trans->dirty_bgs)) { | |
3051 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
32da5386 | 3052 | struct btrfs_block_group, |
77745c05 JB |
3053 | dirty_list); |
3054 | ||
3055 | /* | |
3056 | * This can happen if cache_save_setup re-dirties a block group | |
3057 | * that is already under IO. Just wait for it to finish and | |
3058 | * then do it all again | |
3059 | */ | |
3060 | if (!list_empty(&cache->io_list)) { | |
3061 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3062 | list_del_init(&cache->io_list); | |
3063 | btrfs_wait_cache_io(trans, cache, path); | |
3064 | btrfs_put_block_group(cache); | |
3065 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3066 | } | |
3067 | ||
3068 | /* | |
3069 | * Don't remove from the dirty list until after we've waited on | |
3070 | * any pending IO | |
3071 | */ | |
3072 | list_del_init(&cache->dirty_list); | |
3073 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3074 | should_put = 1; | |
3075 | ||
3076 | cache_save_setup(cache, trans, path); | |
3077 | ||
3078 | if (!ret) | |
3079 | ret = btrfs_run_delayed_refs(trans, | |
3080 | (unsigned long) -1); | |
3081 | ||
3082 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3083 | cache->io_ctl.inode = NULL; | |
3084 | ret = btrfs_write_out_cache(trans, cache, path); | |
3085 | if (ret == 0 && cache->io_ctl.inode) { | |
3086 | num_started++; | |
3087 | should_put = 0; | |
3088 | list_add_tail(&cache->io_list, io); | |
3089 | } else { | |
3090 | /* | |
3091 | * If we failed to write the cache, the | |
3092 | * generation will be bad and life goes on | |
3093 | */ | |
3094 | ret = 0; | |
3095 | } | |
3096 | } | |
3097 | if (!ret) { | |
3be4d8ef | 3098 | ret = update_block_group_item(trans, path, cache); |
77745c05 JB |
3099 | /* |
3100 | * One of the free space endio workers might have | |
3101 | * created a new block group while updating a free space | |
3102 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3103 | * and hasn't released its transaction handle yet, in | |
3104 | * which case the new block group is still attached to | |
3105 | * its transaction handle and its creation has not | |
3106 | * finished yet (no block group item in the extent tree | |
3107 | * yet, etc). If this is the case, wait for all free | |
3108 | * space endio workers to finish and retry. This is a | |
260db43c | 3109 | * very rare case so no need for a more efficient and |
77745c05 JB |
3110 | * complex approach. |
3111 | */ | |
3112 | if (ret == -ENOENT) { | |
3113 | wait_event(cur_trans->writer_wait, | |
3114 | atomic_read(&cur_trans->num_writers) == 1); | |
3be4d8ef | 3115 | ret = update_block_group_item(trans, path, cache); |
77745c05 JB |
3116 | } |
3117 | if (ret) | |
3118 | btrfs_abort_transaction(trans, ret); | |
3119 | } | |
3120 | ||
3121 | /* If its not on the io list, we need to put the block group */ | |
3122 | if (should_put) | |
3123 | btrfs_put_block_group(cache); | |
3124 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
3125 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3126 | } | |
3127 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3128 | ||
3129 | /* | |
3130 | * Refer to the definition of io_bgs member for details why it's safe | |
3131 | * to use it without any locking | |
3132 | */ | |
3133 | while (!list_empty(io)) { | |
32da5386 | 3134 | cache = list_first_entry(io, struct btrfs_block_group, |
77745c05 JB |
3135 | io_list); |
3136 | list_del_init(&cache->io_list); | |
3137 | btrfs_wait_cache_io(trans, cache, path); | |
3138 | btrfs_put_block_group(cache); | |
3139 | } | |
3140 | ||
3141 | btrfs_free_path(path); | |
3142 | return ret; | |
3143 | } | |
606d1bf1 JB |
3144 | |
3145 | int btrfs_update_block_group(struct btrfs_trans_handle *trans, | |
3146 | u64 bytenr, u64 num_bytes, int alloc) | |
3147 | { | |
3148 | struct btrfs_fs_info *info = trans->fs_info; | |
32da5386 | 3149 | struct btrfs_block_group *cache = NULL; |
606d1bf1 JB |
3150 | u64 total = num_bytes; |
3151 | u64 old_val; | |
3152 | u64 byte_in_group; | |
3153 | int factor; | |
3154 | int ret = 0; | |
3155 | ||
3156 | /* Block accounting for super block */ | |
3157 | spin_lock(&info->delalloc_root_lock); | |
3158 | old_val = btrfs_super_bytes_used(info->super_copy); | |
3159 | if (alloc) | |
3160 | old_val += num_bytes; | |
3161 | else | |
3162 | old_val -= num_bytes; | |
3163 | btrfs_set_super_bytes_used(info->super_copy, old_val); | |
3164 | spin_unlock(&info->delalloc_root_lock); | |
3165 | ||
3166 | while (total) { | |
3167 | cache = btrfs_lookup_block_group(info, bytenr); | |
3168 | if (!cache) { | |
3169 | ret = -ENOENT; | |
3170 | break; | |
3171 | } | |
3172 | factor = btrfs_bg_type_to_factor(cache->flags); | |
3173 | ||
3174 | /* | |
3175 | * If this block group has free space cache written out, we | |
3176 | * need to make sure to load it if we are removing space. This | |
3177 | * is because we need the unpinning stage to actually add the | |
3178 | * space back to the block group, otherwise we will leak space. | |
3179 | */ | |
32da5386 | 3180 | if (!alloc && !btrfs_block_group_done(cache)) |
606d1bf1 JB |
3181 | btrfs_cache_block_group(cache, 1); |
3182 | ||
b3470b5d DS |
3183 | byte_in_group = bytenr - cache->start; |
3184 | WARN_ON(byte_in_group > cache->length); | |
606d1bf1 JB |
3185 | |
3186 | spin_lock(&cache->space_info->lock); | |
3187 | spin_lock(&cache->lock); | |
3188 | ||
3189 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
3190 | cache->disk_cache_state < BTRFS_DC_CLEAR) | |
3191 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
3192 | ||
bf38be65 | 3193 | old_val = cache->used; |
b3470b5d | 3194 | num_bytes = min(total, cache->length - byte_in_group); |
606d1bf1 JB |
3195 | if (alloc) { |
3196 | old_val += num_bytes; | |
bf38be65 | 3197 | cache->used = old_val; |
606d1bf1 JB |
3198 | cache->reserved -= num_bytes; |
3199 | cache->space_info->bytes_reserved -= num_bytes; | |
3200 | cache->space_info->bytes_used += num_bytes; | |
3201 | cache->space_info->disk_used += num_bytes * factor; | |
3202 | spin_unlock(&cache->lock); | |
3203 | spin_unlock(&cache->space_info->lock); | |
3204 | } else { | |
3205 | old_val -= num_bytes; | |
bf38be65 | 3206 | cache->used = old_val; |
606d1bf1 JB |
3207 | cache->pinned += num_bytes; |
3208 | btrfs_space_info_update_bytes_pinned(info, | |
3209 | cache->space_info, num_bytes); | |
3210 | cache->space_info->bytes_used -= num_bytes; | |
3211 | cache->space_info->disk_used -= num_bytes * factor; | |
3212 | spin_unlock(&cache->lock); | |
3213 | spin_unlock(&cache->space_info->lock); | |
3214 | ||
fe119a6e | 3215 | set_extent_dirty(&trans->transaction->pinned_extents, |
606d1bf1 JB |
3216 | bytenr, bytenr + num_bytes - 1, |
3217 | GFP_NOFS | __GFP_NOFAIL); | |
3218 | } | |
3219 | ||
3220 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
3221 | if (list_empty(&cache->dirty_list)) { | |
3222 | list_add_tail(&cache->dirty_list, | |
3223 | &trans->transaction->dirty_bgs); | |
3224 | trans->delayed_ref_updates++; | |
3225 | btrfs_get_block_group(cache); | |
3226 | } | |
3227 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
3228 | ||
3229 | /* | |
3230 | * No longer have used bytes in this block group, queue it for | |
3231 | * deletion. We do this after adding the block group to the | |
3232 | * dirty list to avoid races between cleaner kthread and space | |
3233 | * cache writeout. | |
3234 | */ | |
6e80d4f8 DZ |
3235 | if (!alloc && old_val == 0) { |
3236 | if (!btrfs_test_opt(info, DISCARD_ASYNC)) | |
3237 | btrfs_mark_bg_unused(cache); | |
3238 | } | |
606d1bf1 JB |
3239 | |
3240 | btrfs_put_block_group(cache); | |
3241 | total -= num_bytes; | |
3242 | bytenr += num_bytes; | |
3243 | } | |
3244 | ||
3245 | /* Modified block groups are accounted for in the delayed_refs_rsv. */ | |
3246 | btrfs_update_delayed_refs_rsv(trans); | |
3247 | return ret; | |
3248 | } | |
3249 | ||
3250 | /** | |
3251 | * btrfs_add_reserved_bytes - update the block_group and space info counters | |
3252 | * @cache: The cache we are manipulating | |
3253 | * @ram_bytes: The number of bytes of file content, and will be same to | |
3254 | * @num_bytes except for the compress path. | |
3255 | * @num_bytes: The number of bytes in question | |
3256 | * @delalloc: The blocks are allocated for the delalloc write | |
3257 | * | |
3258 | * This is called by the allocator when it reserves space. If this is a | |
3259 | * reservation and the block group has become read only we cannot make the | |
3260 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
3261 | */ | |
32da5386 | 3262 | int btrfs_add_reserved_bytes(struct btrfs_block_group *cache, |
606d1bf1 JB |
3263 | u64 ram_bytes, u64 num_bytes, int delalloc) |
3264 | { | |
3265 | struct btrfs_space_info *space_info = cache->space_info; | |
3266 | int ret = 0; | |
3267 | ||
3268 | spin_lock(&space_info->lock); | |
3269 | spin_lock(&cache->lock); | |
3270 | if (cache->ro) { | |
3271 | ret = -EAGAIN; | |
3272 | } else { | |
3273 | cache->reserved += num_bytes; | |
3274 | space_info->bytes_reserved += num_bytes; | |
a43c3835 JB |
3275 | trace_btrfs_space_reservation(cache->fs_info, "space_info", |
3276 | space_info->flags, num_bytes, 1); | |
606d1bf1 JB |
3277 | btrfs_space_info_update_bytes_may_use(cache->fs_info, |
3278 | space_info, -ram_bytes); | |
3279 | if (delalloc) | |
3280 | cache->delalloc_bytes += num_bytes; | |
99ffb43e JB |
3281 | |
3282 | /* | |
3283 | * Compression can use less space than we reserved, so wake | |
3284 | * tickets if that happens | |
3285 | */ | |
3286 | if (num_bytes < ram_bytes) | |
3287 | btrfs_try_granting_tickets(cache->fs_info, space_info); | |
606d1bf1 JB |
3288 | } |
3289 | spin_unlock(&cache->lock); | |
3290 | spin_unlock(&space_info->lock); | |
3291 | return ret; | |
3292 | } | |
3293 | ||
3294 | /** | |
3295 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
3296 | * @cache: The cache we are manipulating | |
3297 | * @num_bytes: The number of bytes in question | |
3298 | * @delalloc: The blocks are allocated for the delalloc write | |
3299 | * | |
3300 | * This is called by somebody who is freeing space that was never actually used | |
3301 | * on disk. For example if you reserve some space for a new leaf in transaction | |
3302 | * A and before transaction A commits you free that leaf, you call this with | |
3303 | * reserve set to 0 in order to clear the reservation. | |
3304 | */ | |
32da5386 | 3305 | void btrfs_free_reserved_bytes(struct btrfs_block_group *cache, |
606d1bf1 JB |
3306 | u64 num_bytes, int delalloc) |
3307 | { | |
3308 | struct btrfs_space_info *space_info = cache->space_info; | |
3309 | ||
3310 | spin_lock(&space_info->lock); | |
3311 | spin_lock(&cache->lock); | |
3312 | if (cache->ro) | |
3313 | space_info->bytes_readonly += num_bytes; | |
3314 | cache->reserved -= num_bytes; | |
3315 | space_info->bytes_reserved -= num_bytes; | |
3316 | space_info->max_extent_size = 0; | |
3317 | ||
3318 | if (delalloc) | |
3319 | cache->delalloc_bytes -= num_bytes; | |
3320 | spin_unlock(&cache->lock); | |
3308234a JB |
3321 | |
3322 | btrfs_try_granting_tickets(cache->fs_info, space_info); | |
606d1bf1 JB |
3323 | spin_unlock(&space_info->lock); |
3324 | } | |
07730d87 JB |
3325 | |
3326 | static void force_metadata_allocation(struct btrfs_fs_info *info) | |
3327 | { | |
3328 | struct list_head *head = &info->space_info; | |
3329 | struct btrfs_space_info *found; | |
3330 | ||
72804905 | 3331 | list_for_each_entry(found, head, list) { |
07730d87 JB |
3332 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) |
3333 | found->force_alloc = CHUNK_ALLOC_FORCE; | |
3334 | } | |
07730d87 JB |
3335 | } |
3336 | ||
3337 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, | |
3338 | struct btrfs_space_info *sinfo, int force) | |
3339 | { | |
3340 | u64 bytes_used = btrfs_space_info_used(sinfo, false); | |
3341 | u64 thresh; | |
3342 | ||
3343 | if (force == CHUNK_ALLOC_FORCE) | |
3344 | return 1; | |
3345 | ||
3346 | /* | |
3347 | * in limited mode, we want to have some free space up to | |
3348 | * about 1% of the FS size. | |
3349 | */ | |
3350 | if (force == CHUNK_ALLOC_LIMITED) { | |
3351 | thresh = btrfs_super_total_bytes(fs_info->super_copy); | |
3352 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); | |
3353 | ||
3354 | if (sinfo->total_bytes - bytes_used < thresh) | |
3355 | return 1; | |
3356 | } | |
3357 | ||
3358 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) | |
3359 | return 0; | |
3360 | return 1; | |
3361 | } | |
3362 | ||
3363 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) | |
3364 | { | |
3365 | u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type); | |
3366 | ||
3367 | return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); | |
3368 | } | |
3369 | ||
79bd3712 FM |
3370 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags) |
3371 | { | |
3372 | struct btrfs_block_group *bg; | |
3373 | int ret; | |
3374 | ||
3375 | /* | |
3376 | * Check if we have enough space in the system space info because we | |
3377 | * will need to update device items in the chunk btree and insert a new | |
3378 | * chunk item in the chunk btree as well. This will allocate a new | |
3379 | * system block group if needed. | |
3380 | */ | |
3381 | check_system_chunk(trans, flags); | |
3382 | ||
3383 | bg = btrfs_alloc_chunk(trans, flags); | |
3384 | if (IS_ERR(bg)) { | |
3385 | ret = PTR_ERR(bg); | |
3386 | goto out; | |
3387 | } | |
3388 | ||
3389 | /* | |
3390 | * If this is a system chunk allocation then stop right here and do not | |
3391 | * add the chunk item to the chunk btree. This is to prevent a deadlock | |
3392 | * because this system chunk allocation can be triggered while COWing | |
3393 | * some extent buffer of the chunk btree and while holding a lock on a | |
3394 | * parent extent buffer, in which case attempting to insert the chunk | |
3395 | * item (or update the device item) would result in a deadlock on that | |
3396 | * parent extent buffer. In this case defer the chunk btree updates to | |
3397 | * the second phase of chunk allocation and keep our reservation until | |
3398 | * the second phase completes. | |
3399 | * | |
3400 | * This is a rare case and can only be triggered by the very few cases | |
3401 | * we have where we need to touch the chunk btree outside chunk allocation | |
3402 | * and chunk removal. These cases are basically adding a device, removing | |
3403 | * a device or resizing a device. | |
3404 | */ | |
3405 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3406 | return 0; | |
3407 | ||
3408 | ret = btrfs_chunk_alloc_add_chunk_item(trans, bg); | |
3409 | /* | |
3410 | * Normally we are not expected to fail with -ENOSPC here, since we have | |
3411 | * previously reserved space in the system space_info and allocated one | |
3412 | * new system chunk if necessary. However there are two exceptions: | |
3413 | * | |
3414 | * 1) We may have enough free space in the system space_info but all the | |
3415 | * existing system block groups have a profile which can not be used | |
3416 | * for extent allocation. | |
3417 | * | |
3418 | * This happens when mounting in degraded mode. For example we have a | |
3419 | * RAID1 filesystem with 2 devices, lose one device and mount the fs | |
3420 | * using the other device in degraded mode. If we then allocate a chunk, | |
3421 | * we may have enough free space in the existing system space_info, but | |
3422 | * none of the block groups can be used for extent allocation since they | |
3423 | * have a RAID1 profile, and because we are in degraded mode with a | |
3424 | * single device, we are forced to allocate a new system chunk with a | |
3425 | * SINGLE profile. Making check_system_chunk() iterate over all system | |
3426 | * block groups and check if they have a usable profile and enough space | |
3427 | * can be slow on very large filesystems, so we tolerate the -ENOSPC and | |
3428 | * try again after forcing allocation of a new system chunk. Like this | |
3429 | * we avoid paying the cost of that search in normal circumstances, when | |
3430 | * we were not mounted in degraded mode; | |
3431 | * | |
3432 | * 2) We had enough free space info the system space_info, and one suitable | |
3433 | * block group to allocate from when we called check_system_chunk() | |
3434 | * above. However right after we called it, the only system block group | |
3435 | * with enough free space got turned into RO mode by a running scrub, | |
3436 | * and in this case we have to allocate a new one and retry. We only | |
3437 | * need do this allocate and retry once, since we have a transaction | |
3438 | * handle and scrub uses the commit root to search for block groups. | |
3439 | */ | |
3440 | if (ret == -ENOSPC) { | |
3441 | const u64 sys_flags = btrfs_system_alloc_profile(trans->fs_info); | |
3442 | struct btrfs_block_group *sys_bg; | |
3443 | ||
3444 | sys_bg = btrfs_alloc_chunk(trans, sys_flags); | |
3445 | if (IS_ERR(sys_bg)) { | |
3446 | ret = PTR_ERR(sys_bg); | |
3447 | btrfs_abort_transaction(trans, ret); | |
3448 | goto out; | |
3449 | } | |
3450 | ||
3451 | ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); | |
3452 | if (ret) { | |
3453 | btrfs_abort_transaction(trans, ret); | |
3454 | goto out; | |
3455 | } | |
3456 | ||
3457 | ret = btrfs_chunk_alloc_add_chunk_item(trans, bg); | |
3458 | if (ret) { | |
3459 | btrfs_abort_transaction(trans, ret); | |
3460 | goto out; | |
3461 | } | |
3462 | } else if (ret) { | |
3463 | btrfs_abort_transaction(trans, ret); | |
3464 | goto out; | |
3465 | } | |
3466 | out: | |
3467 | btrfs_trans_release_chunk_metadata(trans); | |
3468 | ||
3469 | return ret; | |
3470 | } | |
3471 | ||
07730d87 | 3472 | /* |
79bd3712 FM |
3473 | * Chunk allocation is done in 2 phases: |
3474 | * | |
3475 | * 1) Phase 1 - through btrfs_chunk_alloc() we allocate device extents for | |
3476 | * the chunk, the chunk mapping, create its block group and add the items | |
3477 | * that belong in the chunk btree to it - more specifically, we need to | |
3478 | * update device items in the chunk btree and add a new chunk item to it. | |
3479 | * | |
3480 | * 2) Phase 2 - through btrfs_create_pending_block_groups(), we add the block | |
3481 | * group item to the extent btree and the device extent items to the devices | |
3482 | * btree. | |
3483 | * | |
3484 | * This is done to prevent deadlocks. For example when COWing a node from the | |
3485 | * extent btree we are holding a write lock on the node's parent and if we | |
3486 | * trigger chunk allocation and attempted to insert the new block group item | |
3487 | * in the extent btree right way, we could deadlock because the path for the | |
3488 | * insertion can include that parent node. At first glance it seems impossible | |
3489 | * to trigger chunk allocation after starting a transaction since tasks should | |
3490 | * reserve enough transaction units (metadata space), however while that is true | |
3491 | * most of the time, chunk allocation may still be triggered for several reasons: | |
3492 | * | |
3493 | * 1) When reserving metadata, we check if there is enough free space in the | |
3494 | * metadata space_info and therefore don't trigger allocation of a new chunk. | |
3495 | * However later when the task actually tries to COW an extent buffer from | |
3496 | * the extent btree or from the device btree for example, it is forced to | |
3497 | * allocate a new block group (chunk) because the only one that had enough | |
3498 | * free space was just turned to RO mode by a running scrub for example (or | |
3499 | * device replace, block group reclaim thread, etc), so we can not use it | |
3500 | * for allocating an extent and end up being forced to allocate a new one; | |
3501 | * | |
3502 | * 2) Because we only check that the metadata space_info has enough free bytes, | |
3503 | * we end up not allocating a new metadata chunk in that case. However if | |
3504 | * the filesystem was mounted in degraded mode, none of the existing block | |
3505 | * groups might be suitable for extent allocation due to their incompatible | |
3506 | * profile (for e.g. mounting a 2 devices filesystem, where all block groups | |
3507 | * use a RAID1 profile, in degraded mode using a single device). In this case | |
3508 | * when the task attempts to COW some extent buffer of the extent btree for | |
3509 | * example, it will trigger allocation of a new metadata block group with a | |
3510 | * suitable profile (SINGLE profile in the example of the degraded mount of | |
3511 | * the RAID1 filesystem); | |
3512 | * | |
3513 | * 3) The task has reserved enough transaction units / metadata space, but when | |
3514 | * it attempts to COW an extent buffer from the extent or device btree for | |
3515 | * example, it does not find any free extent in any metadata block group, | |
3516 | * therefore forced to try to allocate a new metadata block group. | |
3517 | * This is because some other task allocated all available extents in the | |
3518 | * meanwhile - this typically happens with tasks that don't reserve space | |
3519 | * properly, either intentionally or as a bug. One example where this is | |
3520 | * done intentionally is fsync, as it does not reserve any transaction units | |
3521 | * and ends up allocating a variable number of metadata extents for log | |
3522 | * tree extent buffers. | |
3523 | * | |
3524 | * We also need this 2 phases setup when adding a device to a filesystem with | |
3525 | * a seed device - we must create new metadata and system chunks without adding | |
3526 | * any of the block group items to the chunk, extent and device btrees. If we | |
3527 | * did not do it this way, we would get ENOSPC when attempting to update those | |
3528 | * btrees, since all the chunks from the seed device are read-only. | |
3529 | * | |
3530 | * Phase 1 does the updates and insertions to the chunk btree because if we had | |
3531 | * it done in phase 2 and have a thundering herd of tasks allocating chunks in | |
3532 | * parallel, we risk having too many system chunks allocated by many tasks if | |
3533 | * many tasks reach phase 1 without the previous ones completing phase 2. In the | |
3534 | * extreme case this leads to exhaustion of the system chunk array in the | |
3535 | * superblock. This is easier to trigger if using a btree node/leaf size of 64K | |
3536 | * and with RAID filesystems (so we have more device items in the chunk btree). | |
3537 | * This has happened before and commit eafa4fd0ad0607 ("btrfs: fix exhaustion of | |
3538 | * the system chunk array due to concurrent allocations") provides more details. | |
3539 | * | |
3540 | * For allocation of system chunks, we defer the updates and insertions into the | |
3541 | * chunk btree to phase 2. This is to prevent deadlocks on extent buffers because | |
3542 | * if the chunk allocation is triggered while COWing an extent buffer of the | |
3543 | * chunk btree, we are holding a lock on the parent of that extent buffer and | |
3544 | * doing the chunk btree updates and insertions can require locking that parent. | |
3545 | * This is for the very few and rare cases where we update the chunk btree that | |
3546 | * are not chunk allocation or chunk removal: adding a device, removing a device | |
3547 | * or resizing a device. | |
3548 | * | |
3549 | * The reservation of system space, done through check_system_chunk(), as well | |
3550 | * as all the updates and insertions into the chunk btree must be done while | |
3551 | * holding fs_info->chunk_mutex. This is important to guarantee that while COWing | |
3552 | * an extent buffer from the chunks btree we never trigger allocation of a new | |
3553 | * system chunk, which would result in a deadlock (trying to lock twice an | |
3554 | * extent buffer of the chunk btree, first time before triggering the chunk | |
3555 | * allocation and the second time during chunk allocation while attempting to | |
3556 | * update the chunks btree). The system chunk array is also updated while holding | |
3557 | * that mutex. The same logic applies to removing chunks - we must reserve system | |
3558 | * space, update the chunk btree and the system chunk array in the superblock | |
3559 | * while holding fs_info->chunk_mutex. | |
3560 | * | |
3561 | * This function, btrfs_chunk_alloc(), belongs to phase 1. | |
3562 | * | |
3563 | * If @force is CHUNK_ALLOC_FORCE: | |
07730d87 JB |
3564 | * - return 1 if it successfully allocates a chunk, |
3565 | * - return errors including -ENOSPC otherwise. | |
79bd3712 | 3566 | * If @force is NOT CHUNK_ALLOC_FORCE: |
07730d87 JB |
3567 | * - return 0 if it doesn't need to allocate a new chunk, |
3568 | * - return 1 if it successfully allocates a chunk, | |
3569 | * - return errors including -ENOSPC otherwise. | |
3570 | */ | |
3571 | int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, | |
3572 | enum btrfs_chunk_alloc_enum force) | |
3573 | { | |
3574 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3575 | struct btrfs_space_info *space_info; | |
3576 | bool wait_for_alloc = false; | |
3577 | bool should_alloc = false; | |
3578 | int ret = 0; | |
3579 | ||
3580 | /* Don't re-enter if we're already allocating a chunk */ | |
3581 | if (trans->allocating_chunk) | |
3582 | return -ENOSPC; | |
79bd3712 FM |
3583 | /* |
3584 | * If we are removing a chunk, don't re-enter or we would deadlock. | |
3585 | * System space reservation and system chunk allocation is done by the | |
3586 | * chunk remove operation (btrfs_remove_chunk()). | |
3587 | */ | |
3588 | if (trans->removing_chunk) | |
3589 | return -ENOSPC; | |
07730d87 JB |
3590 | |
3591 | space_info = btrfs_find_space_info(fs_info, flags); | |
3592 | ASSERT(space_info); | |
3593 | ||
3594 | do { | |
3595 | spin_lock(&space_info->lock); | |
3596 | if (force < space_info->force_alloc) | |
3597 | force = space_info->force_alloc; | |
3598 | should_alloc = should_alloc_chunk(fs_info, space_info, force); | |
3599 | if (space_info->full) { | |
3600 | /* No more free physical space */ | |
3601 | if (should_alloc) | |
3602 | ret = -ENOSPC; | |
3603 | else | |
3604 | ret = 0; | |
3605 | spin_unlock(&space_info->lock); | |
3606 | return ret; | |
3607 | } else if (!should_alloc) { | |
3608 | spin_unlock(&space_info->lock); | |
3609 | return 0; | |
3610 | } else if (space_info->chunk_alloc) { | |
3611 | /* | |
3612 | * Someone is already allocating, so we need to block | |
3613 | * until this someone is finished and then loop to | |
3614 | * recheck if we should continue with our allocation | |
3615 | * attempt. | |
3616 | */ | |
3617 | wait_for_alloc = true; | |
3618 | spin_unlock(&space_info->lock); | |
3619 | mutex_lock(&fs_info->chunk_mutex); | |
3620 | mutex_unlock(&fs_info->chunk_mutex); | |
3621 | } else { | |
3622 | /* Proceed with allocation */ | |
3623 | space_info->chunk_alloc = 1; | |
3624 | wait_for_alloc = false; | |
3625 | spin_unlock(&space_info->lock); | |
3626 | } | |
3627 | ||
3628 | cond_resched(); | |
3629 | } while (wait_for_alloc); | |
3630 | ||
3631 | mutex_lock(&fs_info->chunk_mutex); | |
3632 | trans->allocating_chunk = true; | |
3633 | ||
3634 | /* | |
3635 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3636 | * allocating mixed chunks instead of individual chunks. | |
3637 | */ | |
3638 | if (btrfs_mixed_space_info(space_info)) | |
3639 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3640 | ||
3641 | /* | |
3642 | * if we're doing a data chunk, go ahead and make sure that | |
3643 | * we keep a reasonable number of metadata chunks allocated in the | |
3644 | * FS as well. | |
3645 | */ | |
3646 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { | |
3647 | fs_info->data_chunk_allocations++; | |
3648 | if (!(fs_info->data_chunk_allocations % | |
3649 | fs_info->metadata_ratio)) | |
3650 | force_metadata_allocation(fs_info); | |
3651 | } | |
3652 | ||
79bd3712 | 3653 | ret = do_chunk_alloc(trans, flags); |
07730d87 JB |
3654 | trans->allocating_chunk = false; |
3655 | ||
3656 | spin_lock(&space_info->lock); | |
3657 | if (ret < 0) { | |
3658 | if (ret == -ENOSPC) | |
3659 | space_info->full = 1; | |
3660 | else | |
3661 | goto out; | |
3662 | } else { | |
3663 | ret = 1; | |
3664 | space_info->max_extent_size = 0; | |
3665 | } | |
3666 | ||
3667 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
3668 | out: | |
3669 | space_info->chunk_alloc = 0; | |
3670 | spin_unlock(&space_info->lock); | |
3671 | mutex_unlock(&fs_info->chunk_mutex); | |
07730d87 JB |
3672 | |
3673 | return ret; | |
3674 | } | |
3675 | ||
3676 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) | |
3677 | { | |
3678 | u64 num_dev; | |
3679 | ||
3680 | num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max; | |
3681 | if (!num_dev) | |
3682 | num_dev = fs_info->fs_devices->rw_devices; | |
3683 | ||
3684 | return num_dev; | |
3685 | } | |
3686 | ||
3687 | /* | |
a9143bd3 | 3688 | * Reserve space in the system space for allocating or removing a chunk |
07730d87 JB |
3689 | */ |
3690 | void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) | |
3691 | { | |
3692 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3693 | struct btrfs_space_info *info; | |
3694 | u64 left; | |
3695 | u64 thresh; | |
3696 | int ret = 0; | |
3697 | u64 num_devs; | |
3698 | ||
3699 | /* | |
3700 | * Needed because we can end up allocating a system chunk and for an | |
3701 | * atomic and race free space reservation in the chunk block reserve. | |
3702 | */ | |
3703 | lockdep_assert_held(&fs_info->chunk_mutex); | |
3704 | ||
3705 | info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3706 | spin_lock(&info->lock); | |
3707 | left = info->total_bytes - btrfs_space_info_used(info, true); | |
3708 | spin_unlock(&info->lock); | |
3709 | ||
3710 | num_devs = get_profile_num_devs(fs_info, type); | |
3711 | ||
3712 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
2bd36e7b JB |
3713 | thresh = btrfs_calc_metadata_size(fs_info, num_devs) + |
3714 | btrfs_calc_insert_metadata_size(fs_info, 1); | |
07730d87 JB |
3715 | |
3716 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { | |
3717 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
3718 | left, thresh, type); | |
3719 | btrfs_dump_space_info(fs_info, info, 0, 0); | |
3720 | } | |
3721 | ||
3722 | if (left < thresh) { | |
3723 | u64 flags = btrfs_system_alloc_profile(fs_info); | |
79bd3712 | 3724 | struct btrfs_block_group *bg; |
07730d87 JB |
3725 | |
3726 | /* | |
3727 | * Ignore failure to create system chunk. We might end up not | |
3728 | * needing it, as we might not need to COW all nodes/leafs from | |
3729 | * the paths we visit in the chunk tree (they were already COWed | |
3730 | * or created in the current transaction for example). | |
79bd3712 FM |
3731 | * |
3732 | * Also, if our caller is allocating a system chunk, do not | |
3733 | * attempt to insert the chunk item in the chunk btree, as we | |
3734 | * could deadlock on an extent buffer since our caller may be | |
3735 | * COWing an extent buffer from the chunk btree. | |
07730d87 | 3736 | */ |
79bd3712 FM |
3737 | bg = btrfs_alloc_chunk(trans, flags); |
3738 | if (IS_ERR(bg)) { | |
3739 | ret = PTR_ERR(bg); | |
3740 | } else if (!(type & BTRFS_BLOCK_GROUP_SYSTEM)) { | |
3741 | /* | |
3742 | * If we fail to add the chunk item here, we end up | |
3743 | * trying again at phase 2 of chunk allocation, at | |
3744 | * btrfs_create_pending_block_groups(). So ignore | |
3745 | * any error here. | |
3746 | */ | |
3747 | btrfs_chunk_alloc_add_chunk_item(trans, bg); | |
3748 | } | |
07730d87 JB |
3749 | } |
3750 | ||
3751 | if (!ret) { | |
3752 | ret = btrfs_block_rsv_add(fs_info->chunk_root, | |
3753 | &fs_info->chunk_block_rsv, | |
3754 | thresh, BTRFS_RESERVE_NO_FLUSH); | |
1cb3db1c | 3755 | if (!ret) |
07730d87 JB |
3756 | trans->chunk_bytes_reserved += thresh; |
3757 | } | |
3758 | } | |
3759 | ||
3e43c279 JB |
3760 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
3761 | { | |
32da5386 | 3762 | struct btrfs_block_group *block_group; |
3e43c279 JB |
3763 | u64 last = 0; |
3764 | ||
3765 | while (1) { | |
3766 | struct inode *inode; | |
3767 | ||
3768 | block_group = btrfs_lookup_first_block_group(info, last); | |
3769 | while (block_group) { | |
3770 | btrfs_wait_block_group_cache_done(block_group); | |
3771 | spin_lock(&block_group->lock); | |
3772 | if (block_group->iref) | |
3773 | break; | |
3774 | spin_unlock(&block_group->lock); | |
3775 | block_group = btrfs_next_block_group(block_group); | |
3776 | } | |
3777 | if (!block_group) { | |
3778 | if (last == 0) | |
3779 | break; | |
3780 | last = 0; | |
3781 | continue; | |
3782 | } | |
3783 | ||
3784 | inode = block_group->inode; | |
3785 | block_group->iref = 0; | |
3786 | block_group->inode = NULL; | |
3787 | spin_unlock(&block_group->lock); | |
3788 | ASSERT(block_group->io_ctl.inode == NULL); | |
3789 | iput(inode); | |
b3470b5d | 3790 | last = block_group->start + block_group->length; |
3e43c279 JB |
3791 | btrfs_put_block_group(block_group); |
3792 | } | |
3793 | } | |
3794 | ||
3795 | /* | |
3796 | * Must be called only after stopping all workers, since we could have block | |
3797 | * group caching kthreads running, and therefore they could race with us if we | |
3798 | * freed the block groups before stopping them. | |
3799 | */ | |
3800 | int btrfs_free_block_groups(struct btrfs_fs_info *info) | |
3801 | { | |
32da5386 | 3802 | struct btrfs_block_group *block_group; |
3e43c279 JB |
3803 | struct btrfs_space_info *space_info; |
3804 | struct btrfs_caching_control *caching_ctl; | |
3805 | struct rb_node *n; | |
3806 | ||
bbb86a37 | 3807 | spin_lock(&info->block_group_cache_lock); |
3e43c279 JB |
3808 | while (!list_empty(&info->caching_block_groups)) { |
3809 | caching_ctl = list_entry(info->caching_block_groups.next, | |
3810 | struct btrfs_caching_control, list); | |
3811 | list_del(&caching_ctl->list); | |
3812 | btrfs_put_caching_control(caching_ctl); | |
3813 | } | |
bbb86a37 | 3814 | spin_unlock(&info->block_group_cache_lock); |
3e43c279 JB |
3815 | |
3816 | spin_lock(&info->unused_bgs_lock); | |
3817 | while (!list_empty(&info->unused_bgs)) { | |
3818 | block_group = list_first_entry(&info->unused_bgs, | |
32da5386 | 3819 | struct btrfs_block_group, |
3e43c279 JB |
3820 | bg_list); |
3821 | list_del_init(&block_group->bg_list); | |
3822 | btrfs_put_block_group(block_group); | |
3823 | } | |
3824 | spin_unlock(&info->unused_bgs_lock); | |
3825 | ||
18bb8bbf JT |
3826 | spin_lock(&info->unused_bgs_lock); |
3827 | while (!list_empty(&info->reclaim_bgs)) { | |
3828 | block_group = list_first_entry(&info->reclaim_bgs, | |
3829 | struct btrfs_block_group, | |
3830 | bg_list); | |
3831 | list_del_init(&block_group->bg_list); | |
3832 | btrfs_put_block_group(block_group); | |
3833 | } | |
3834 | spin_unlock(&info->unused_bgs_lock); | |
3835 | ||
3e43c279 JB |
3836 | spin_lock(&info->block_group_cache_lock); |
3837 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
32da5386 | 3838 | block_group = rb_entry(n, struct btrfs_block_group, |
3e43c279 JB |
3839 | cache_node); |
3840 | rb_erase(&block_group->cache_node, | |
3841 | &info->block_group_cache_tree); | |
3842 | RB_CLEAR_NODE(&block_group->cache_node); | |
3843 | spin_unlock(&info->block_group_cache_lock); | |
3844 | ||
3845 | down_write(&block_group->space_info->groups_sem); | |
3846 | list_del(&block_group->list); | |
3847 | up_write(&block_group->space_info->groups_sem); | |
3848 | ||
3849 | /* | |
3850 | * We haven't cached this block group, which means we could | |
3851 | * possibly have excluded extents on this block group. | |
3852 | */ | |
3853 | if (block_group->cached == BTRFS_CACHE_NO || | |
3854 | block_group->cached == BTRFS_CACHE_ERROR) | |
3855 | btrfs_free_excluded_extents(block_group); | |
3856 | ||
3857 | btrfs_remove_free_space_cache(block_group); | |
3858 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); | |
3859 | ASSERT(list_empty(&block_group->dirty_list)); | |
3860 | ASSERT(list_empty(&block_group->io_list)); | |
3861 | ASSERT(list_empty(&block_group->bg_list)); | |
48aaeebe | 3862 | ASSERT(refcount_read(&block_group->refs) == 1); |
195a49ea | 3863 | ASSERT(block_group->swap_extents == 0); |
3e43c279 JB |
3864 | btrfs_put_block_group(block_group); |
3865 | ||
3866 | spin_lock(&info->block_group_cache_lock); | |
3867 | } | |
3868 | spin_unlock(&info->block_group_cache_lock); | |
3869 | ||
3e43c279 JB |
3870 | btrfs_release_global_block_rsv(info); |
3871 | ||
3872 | while (!list_empty(&info->space_info)) { | |
3873 | space_info = list_entry(info->space_info.next, | |
3874 | struct btrfs_space_info, | |
3875 | list); | |
3876 | ||
3877 | /* | |
3878 | * Do not hide this behind enospc_debug, this is actually | |
3879 | * important and indicates a real bug if this happens. | |
3880 | */ | |
3881 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
3882 | space_info->bytes_reserved > 0 || | |
3883 | space_info->bytes_may_use > 0)) | |
3884 | btrfs_dump_space_info(info, space_info, 0, 0); | |
d611add4 | 3885 | WARN_ON(space_info->reclaim_size > 0); |
3e43c279 JB |
3886 | list_del(&space_info->list); |
3887 | btrfs_sysfs_remove_space_info(space_info); | |
3888 | } | |
3889 | return 0; | |
3890 | } | |
684b752b FM |
3891 | |
3892 | void btrfs_freeze_block_group(struct btrfs_block_group *cache) | |
3893 | { | |
3894 | atomic_inc(&cache->frozen); | |
3895 | } | |
3896 | ||
3897 | void btrfs_unfreeze_block_group(struct btrfs_block_group *block_group) | |
3898 | { | |
3899 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
3900 | struct extent_map_tree *em_tree; | |
3901 | struct extent_map *em; | |
3902 | bool cleanup; | |
3903 | ||
3904 | spin_lock(&block_group->lock); | |
3905 | cleanup = (atomic_dec_and_test(&block_group->frozen) && | |
3906 | block_group->removed); | |
3907 | spin_unlock(&block_group->lock); | |
3908 | ||
3909 | if (cleanup) { | |
684b752b FM |
3910 | em_tree = &fs_info->mapping_tree; |
3911 | write_lock(&em_tree->lock); | |
3912 | em = lookup_extent_mapping(em_tree, block_group->start, | |
3913 | 1); | |
3914 | BUG_ON(!em); /* logic error, can't happen */ | |
3915 | remove_extent_mapping(em_tree, em); | |
3916 | write_unlock(&em_tree->lock); | |
684b752b FM |
3917 | |
3918 | /* once for us and once for the tree */ | |
3919 | free_extent_map(em); | |
3920 | free_extent_map(em); | |
3921 | ||
3922 | /* | |
3923 | * We may have left one free space entry and other possible | |
3924 | * tasks trimming this block group have left 1 entry each one. | |
3925 | * Free them if any. | |
3926 | */ | |
3927 | __btrfs_remove_free_space_cache(block_group->free_space_ctl); | |
3928 | } | |
3929 | } | |
195a49ea FM |
3930 | |
3931 | bool btrfs_inc_block_group_swap_extents(struct btrfs_block_group *bg) | |
3932 | { | |
3933 | bool ret = true; | |
3934 | ||
3935 | spin_lock(&bg->lock); | |
3936 | if (bg->ro) | |
3937 | ret = false; | |
3938 | else | |
3939 | bg->swap_extents++; | |
3940 | spin_unlock(&bg->lock); | |
3941 | ||
3942 | return ret; | |
3943 | } | |
3944 | ||
3945 | void btrfs_dec_block_group_swap_extents(struct btrfs_block_group *bg, int amount) | |
3946 | { | |
3947 | spin_lock(&bg->lock); | |
3948 | ASSERT(!bg->ro); | |
3949 | ASSERT(bg->swap_extents >= amount); | |
3950 | bg->swap_extents -= amount; | |
3951 | spin_unlock(&bg->lock); | |
3952 | } |