]>
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" |
2e405ad8 | 18 | |
878d7b67 JB |
19 | /* |
20 | * Return target flags in extended format or 0 if restripe for this chunk_type | |
21 | * is not in progress | |
22 | * | |
23 | * Should be called with balance_lock held | |
24 | */ | |
e11c0406 | 25 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) |
878d7b67 JB |
26 | { |
27 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
28 | u64 target = 0; | |
29 | ||
30 | if (!bctl) | |
31 | return 0; | |
32 | ||
33 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
34 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
35 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
36 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
37 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
38 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
39 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
40 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
41 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
42 | } | |
43 | ||
44 | return target; | |
45 | } | |
46 | ||
47 | /* | |
48 | * @flags: available profiles in extended format (see ctree.h) | |
49 | * | |
50 | * Return reduced profile in chunk format. If profile changing is in progress | |
51 | * (either running or paused) picks the target profile (if it's already | |
52 | * available), otherwise falls back to plain reducing. | |
53 | */ | |
54 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) | |
55 | { | |
56 | u64 num_devices = fs_info->fs_devices->rw_devices; | |
57 | u64 target; | |
58 | u64 raid_type; | |
59 | u64 allowed = 0; | |
60 | ||
61 | /* | |
62 | * See if restripe for this chunk_type is in progress, if so try to | |
63 | * reduce to the target profile | |
64 | */ | |
65 | spin_lock(&fs_info->balance_lock); | |
e11c0406 | 66 | target = get_restripe_target(fs_info, flags); |
878d7b67 | 67 | if (target) { |
162e0a16 JB |
68 | spin_unlock(&fs_info->balance_lock); |
69 | return extended_to_chunk(target); | |
878d7b67 JB |
70 | } |
71 | spin_unlock(&fs_info->balance_lock); | |
72 | ||
73 | /* First, mask out the RAID levels which aren't possible */ | |
74 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { | |
75 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
76 | allowed |= btrfs_raid_array[raid_type].bg_flag; | |
77 | } | |
78 | allowed &= flags; | |
79 | ||
80 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
81 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
82 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
83 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
84 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
85 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
86 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
87 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
88 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
89 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
90 | ||
91 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
92 | ||
93 | return extended_to_chunk(flags | allowed); | |
94 | } | |
95 | ||
ef0a82da | 96 | u64 btrfs_get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
878d7b67 JB |
97 | { |
98 | unsigned seq; | |
99 | u64 flags; | |
100 | ||
101 | do { | |
102 | flags = orig_flags; | |
103 | seq = read_seqbegin(&fs_info->profiles_lock); | |
104 | ||
105 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
106 | flags |= fs_info->avail_data_alloc_bits; | |
107 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
108 | flags |= fs_info->avail_system_alloc_bits; | |
109 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
110 | flags |= fs_info->avail_metadata_alloc_bits; | |
111 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
112 | ||
113 | return btrfs_reduce_alloc_profile(fs_info, flags); | |
114 | } | |
115 | ||
32da5386 | 116 | void btrfs_get_block_group(struct btrfs_block_group *cache) |
3cad1284 | 117 | { |
48aaeebe | 118 | refcount_inc(&cache->refs); |
3cad1284 JB |
119 | } |
120 | ||
32da5386 | 121 | void btrfs_put_block_group(struct btrfs_block_group *cache) |
3cad1284 | 122 | { |
48aaeebe | 123 | if (refcount_dec_and_test(&cache->refs)) { |
3cad1284 JB |
124 | WARN_ON(cache->pinned > 0); |
125 | WARN_ON(cache->reserved > 0); | |
126 | ||
b0643e59 DZ |
127 | /* |
128 | * A block_group shouldn't be on the discard_list anymore. | |
129 | * Remove the block_group from the discard_list to prevent us | |
130 | * from causing a panic due to NULL pointer dereference. | |
131 | */ | |
132 | if (WARN_ON(!list_empty(&cache->discard_list))) | |
133 | btrfs_discard_cancel_work(&cache->fs_info->discard_ctl, | |
134 | cache); | |
135 | ||
3cad1284 JB |
136 | /* |
137 | * If not empty, someone is still holding mutex of | |
138 | * full_stripe_lock, which can only be released by caller. | |
139 | * And it will definitely cause use-after-free when caller | |
140 | * tries to release full stripe lock. | |
141 | * | |
142 | * No better way to resolve, but only to warn. | |
143 | */ | |
144 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
145 | kfree(cache->free_space_ctl); | |
146 | kfree(cache); | |
147 | } | |
148 | } | |
149 | ||
4358d963 JB |
150 | /* |
151 | * This adds the block group to the fs_info rb tree for the block group cache | |
152 | */ | |
153 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, | |
32da5386 | 154 | struct btrfs_block_group *block_group) |
4358d963 JB |
155 | { |
156 | struct rb_node **p; | |
157 | struct rb_node *parent = NULL; | |
32da5386 | 158 | struct btrfs_block_group *cache; |
4358d963 | 159 | |
9afc6649 QW |
160 | ASSERT(block_group->length != 0); |
161 | ||
4358d963 JB |
162 | spin_lock(&info->block_group_cache_lock); |
163 | p = &info->block_group_cache_tree.rb_node; | |
164 | ||
165 | while (*p) { | |
166 | parent = *p; | |
32da5386 | 167 | cache = rb_entry(parent, struct btrfs_block_group, cache_node); |
b3470b5d | 168 | if (block_group->start < cache->start) { |
4358d963 | 169 | p = &(*p)->rb_left; |
b3470b5d | 170 | } else if (block_group->start > cache->start) { |
4358d963 JB |
171 | p = &(*p)->rb_right; |
172 | } else { | |
173 | spin_unlock(&info->block_group_cache_lock); | |
174 | return -EEXIST; | |
175 | } | |
176 | } | |
177 | ||
178 | rb_link_node(&block_group->cache_node, parent, p); | |
179 | rb_insert_color(&block_group->cache_node, | |
180 | &info->block_group_cache_tree); | |
181 | ||
b3470b5d DS |
182 | if (info->first_logical_byte > block_group->start) |
183 | info->first_logical_byte = block_group->start; | |
4358d963 JB |
184 | |
185 | spin_unlock(&info->block_group_cache_lock); | |
186 | ||
187 | return 0; | |
188 | } | |
189 | ||
2e405ad8 JB |
190 | /* |
191 | * This will return the block group at or after bytenr if contains is 0, else | |
192 | * it will return the block group that contains the bytenr | |
193 | */ | |
32da5386 | 194 | static struct btrfs_block_group *block_group_cache_tree_search( |
2e405ad8 JB |
195 | struct btrfs_fs_info *info, u64 bytenr, int contains) |
196 | { | |
32da5386 | 197 | struct btrfs_block_group *cache, *ret = NULL; |
2e405ad8 JB |
198 | struct rb_node *n; |
199 | u64 end, start; | |
200 | ||
201 | spin_lock(&info->block_group_cache_lock); | |
202 | n = info->block_group_cache_tree.rb_node; | |
203 | ||
204 | while (n) { | |
32da5386 | 205 | cache = rb_entry(n, struct btrfs_block_group, cache_node); |
b3470b5d DS |
206 | end = cache->start + cache->length - 1; |
207 | start = cache->start; | |
2e405ad8 JB |
208 | |
209 | if (bytenr < start) { | |
b3470b5d | 210 | if (!contains && (!ret || start < ret->start)) |
2e405ad8 JB |
211 | ret = cache; |
212 | n = n->rb_left; | |
213 | } else if (bytenr > start) { | |
214 | if (contains && bytenr <= end) { | |
215 | ret = cache; | |
216 | break; | |
217 | } | |
218 | n = n->rb_right; | |
219 | } else { | |
220 | ret = cache; | |
221 | break; | |
222 | } | |
223 | } | |
224 | if (ret) { | |
225 | btrfs_get_block_group(ret); | |
b3470b5d DS |
226 | if (bytenr == 0 && info->first_logical_byte > ret->start) |
227 | info->first_logical_byte = ret->start; | |
2e405ad8 JB |
228 | } |
229 | spin_unlock(&info->block_group_cache_lock); | |
230 | ||
231 | return ret; | |
232 | } | |
233 | ||
234 | /* | |
235 | * Return the block group that starts at or after bytenr | |
236 | */ | |
32da5386 | 237 | struct btrfs_block_group *btrfs_lookup_first_block_group( |
2e405ad8 JB |
238 | struct btrfs_fs_info *info, u64 bytenr) |
239 | { | |
240 | return block_group_cache_tree_search(info, bytenr, 0); | |
241 | } | |
242 | ||
243 | /* | |
244 | * Return the block group that contains the given bytenr | |
245 | */ | |
32da5386 | 246 | struct btrfs_block_group *btrfs_lookup_block_group( |
2e405ad8 JB |
247 | struct btrfs_fs_info *info, u64 bytenr) |
248 | { | |
249 | return block_group_cache_tree_search(info, bytenr, 1); | |
250 | } | |
251 | ||
32da5386 DS |
252 | struct btrfs_block_group *btrfs_next_block_group( |
253 | struct btrfs_block_group *cache) | |
2e405ad8 JB |
254 | { |
255 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
256 | struct rb_node *node; | |
257 | ||
258 | spin_lock(&fs_info->block_group_cache_lock); | |
259 | ||
260 | /* If our block group was removed, we need a full search. */ | |
261 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
b3470b5d | 262 | const u64 next_bytenr = cache->start + cache->length; |
2e405ad8 JB |
263 | |
264 | spin_unlock(&fs_info->block_group_cache_lock); | |
265 | btrfs_put_block_group(cache); | |
266 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; | |
267 | } | |
268 | node = rb_next(&cache->cache_node); | |
269 | btrfs_put_block_group(cache); | |
270 | if (node) { | |
32da5386 | 271 | cache = rb_entry(node, struct btrfs_block_group, cache_node); |
2e405ad8 JB |
272 | btrfs_get_block_group(cache); |
273 | } else | |
274 | cache = NULL; | |
275 | spin_unlock(&fs_info->block_group_cache_lock); | |
276 | return cache; | |
277 | } | |
3eeb3226 JB |
278 | |
279 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
280 | { | |
32da5386 | 281 | struct btrfs_block_group *bg; |
3eeb3226 JB |
282 | bool ret = true; |
283 | ||
284 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
285 | if (!bg) | |
286 | return false; | |
287 | ||
288 | spin_lock(&bg->lock); | |
289 | if (bg->ro) | |
290 | ret = false; | |
291 | else | |
292 | atomic_inc(&bg->nocow_writers); | |
293 | spin_unlock(&bg->lock); | |
294 | ||
295 | /* No put on block group, done by btrfs_dec_nocow_writers */ | |
296 | if (!ret) | |
297 | btrfs_put_block_group(bg); | |
298 | ||
299 | return ret; | |
300 | } | |
301 | ||
302 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
303 | { | |
32da5386 | 304 | struct btrfs_block_group *bg; |
3eeb3226 JB |
305 | |
306 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
307 | ASSERT(bg); | |
308 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
309 | wake_up_var(&bg->nocow_writers); | |
310 | /* | |
311 | * Once for our lookup and once for the lookup done by a previous call | |
312 | * to btrfs_inc_nocow_writers() | |
313 | */ | |
314 | btrfs_put_block_group(bg); | |
315 | btrfs_put_block_group(bg); | |
316 | } | |
317 | ||
32da5386 | 318 | void btrfs_wait_nocow_writers(struct btrfs_block_group *bg) |
3eeb3226 JB |
319 | { |
320 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); | |
321 | } | |
322 | ||
323 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
324 | const u64 start) | |
325 | { | |
32da5386 | 326 | struct btrfs_block_group *bg; |
3eeb3226 JB |
327 | |
328 | bg = btrfs_lookup_block_group(fs_info, start); | |
329 | ASSERT(bg); | |
330 | if (atomic_dec_and_test(&bg->reservations)) | |
331 | wake_up_var(&bg->reservations); | |
332 | btrfs_put_block_group(bg); | |
333 | } | |
334 | ||
32da5386 | 335 | void btrfs_wait_block_group_reservations(struct btrfs_block_group *bg) |
3eeb3226 JB |
336 | { |
337 | struct btrfs_space_info *space_info = bg->space_info; | |
338 | ||
339 | ASSERT(bg->ro); | |
340 | ||
341 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
342 | return; | |
343 | ||
344 | /* | |
345 | * Our block group is read only but before we set it to read only, | |
346 | * some task might have had allocated an extent from it already, but it | |
347 | * has not yet created a respective ordered extent (and added it to a | |
348 | * root's list of ordered extents). | |
349 | * Therefore wait for any task currently allocating extents, since the | |
350 | * block group's reservations counter is incremented while a read lock | |
351 | * on the groups' semaphore is held and decremented after releasing | |
352 | * the read access on that semaphore and creating the ordered extent. | |
353 | */ | |
354 | down_write(&space_info->groups_sem); | |
355 | up_write(&space_info->groups_sem); | |
356 | ||
357 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); | |
358 | } | |
9f21246d JB |
359 | |
360 | struct btrfs_caching_control *btrfs_get_caching_control( | |
32da5386 | 361 | struct btrfs_block_group *cache) |
9f21246d JB |
362 | { |
363 | struct btrfs_caching_control *ctl; | |
364 | ||
365 | spin_lock(&cache->lock); | |
366 | if (!cache->caching_ctl) { | |
367 | spin_unlock(&cache->lock); | |
368 | return NULL; | |
369 | } | |
370 | ||
371 | ctl = cache->caching_ctl; | |
372 | refcount_inc(&ctl->count); | |
373 | spin_unlock(&cache->lock); | |
374 | return ctl; | |
375 | } | |
376 | ||
377 | void btrfs_put_caching_control(struct btrfs_caching_control *ctl) | |
378 | { | |
379 | if (refcount_dec_and_test(&ctl->count)) | |
380 | kfree(ctl); | |
381 | } | |
382 | ||
383 | /* | |
384 | * When we wait for progress in the block group caching, its because our | |
385 | * allocation attempt failed at least once. So, we must sleep and let some | |
386 | * progress happen before we try again. | |
387 | * | |
388 | * This function will sleep at least once waiting for new free space to show | |
389 | * up, and then it will check the block group free space numbers for our min | |
390 | * num_bytes. Another option is to have it go ahead and look in the rbtree for | |
391 | * a free extent of a given size, but this is a good start. | |
392 | * | |
393 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
394 | * any of the information in this block group. | |
395 | */ | |
32da5386 | 396 | void btrfs_wait_block_group_cache_progress(struct btrfs_block_group *cache, |
9f21246d JB |
397 | u64 num_bytes) |
398 | { | |
399 | struct btrfs_caching_control *caching_ctl; | |
400 | ||
401 | caching_ctl = btrfs_get_caching_control(cache); | |
402 | if (!caching_ctl) | |
403 | return; | |
404 | ||
32da5386 | 405 | wait_event(caching_ctl->wait, btrfs_block_group_done(cache) || |
9f21246d JB |
406 | (cache->free_space_ctl->free_space >= num_bytes)); |
407 | ||
408 | btrfs_put_caching_control(caching_ctl); | |
409 | } | |
410 | ||
32da5386 | 411 | int btrfs_wait_block_group_cache_done(struct btrfs_block_group *cache) |
9f21246d JB |
412 | { |
413 | struct btrfs_caching_control *caching_ctl; | |
414 | int ret = 0; | |
415 | ||
416 | caching_ctl = btrfs_get_caching_control(cache); | |
417 | if (!caching_ctl) | |
418 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; | |
419 | ||
32da5386 | 420 | wait_event(caching_ctl->wait, btrfs_block_group_done(cache)); |
9f21246d JB |
421 | if (cache->cached == BTRFS_CACHE_ERROR) |
422 | ret = -EIO; | |
423 | btrfs_put_caching_control(caching_ctl); | |
424 | return ret; | |
425 | } | |
426 | ||
e747853c JB |
427 | static bool space_cache_v1_done(struct btrfs_block_group *cache) |
428 | { | |
429 | bool ret; | |
430 | ||
431 | spin_lock(&cache->lock); | |
432 | ret = cache->cached != BTRFS_CACHE_FAST; | |
433 | spin_unlock(&cache->lock); | |
434 | ||
435 | return ret; | |
436 | } | |
437 | ||
438 | void btrfs_wait_space_cache_v1_finished(struct btrfs_block_group *cache, | |
439 | struct btrfs_caching_control *caching_ctl) | |
440 | { | |
441 | wait_event(caching_ctl->wait, space_cache_v1_done(cache)); | |
442 | } | |
443 | ||
9f21246d | 444 | #ifdef CONFIG_BTRFS_DEBUG |
32da5386 | 445 | static void fragment_free_space(struct btrfs_block_group *block_group) |
9f21246d JB |
446 | { |
447 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
b3470b5d DS |
448 | u64 start = block_group->start; |
449 | u64 len = block_group->length; | |
9f21246d JB |
450 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? |
451 | fs_info->nodesize : fs_info->sectorsize; | |
452 | u64 step = chunk << 1; | |
453 | ||
454 | while (len > chunk) { | |
455 | btrfs_remove_free_space(block_group, start, chunk); | |
456 | start += step; | |
457 | if (len < step) | |
458 | len = 0; | |
459 | else | |
460 | len -= step; | |
461 | } | |
462 | } | |
463 | #endif | |
464 | ||
465 | /* | |
466 | * This is only called by btrfs_cache_block_group, since we could have freed | |
467 | * extents we need to check the pinned_extents for any extents that can't be | |
468 | * used yet since their free space will be released as soon as the transaction | |
469 | * commits. | |
470 | */ | |
32da5386 | 471 | u64 add_new_free_space(struct btrfs_block_group *block_group, u64 start, u64 end) |
9f21246d JB |
472 | { |
473 | struct btrfs_fs_info *info = block_group->fs_info; | |
474 | u64 extent_start, extent_end, size, total_added = 0; | |
475 | int ret; | |
476 | ||
477 | while (start < end) { | |
fe119a6e | 478 | ret = find_first_extent_bit(&info->excluded_extents, start, |
9f21246d JB |
479 | &extent_start, &extent_end, |
480 | EXTENT_DIRTY | EXTENT_UPTODATE, | |
481 | NULL); | |
482 | if (ret) | |
483 | break; | |
484 | ||
485 | if (extent_start <= start) { | |
486 | start = extent_end + 1; | |
487 | } else if (extent_start > start && extent_start < end) { | |
488 | size = extent_start - start; | |
489 | total_added += size; | |
b0643e59 DZ |
490 | ret = btrfs_add_free_space_async_trimmed(block_group, |
491 | start, size); | |
9f21246d JB |
492 | BUG_ON(ret); /* -ENOMEM or logic error */ |
493 | start = extent_end + 1; | |
494 | } else { | |
495 | break; | |
496 | } | |
497 | } | |
498 | ||
499 | if (start < end) { | |
500 | size = end - start; | |
501 | total_added += size; | |
b0643e59 DZ |
502 | ret = btrfs_add_free_space_async_trimmed(block_group, start, |
503 | size); | |
9f21246d JB |
504 | BUG_ON(ret); /* -ENOMEM or logic error */ |
505 | } | |
506 | ||
507 | return total_added; | |
508 | } | |
509 | ||
510 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) | |
511 | { | |
32da5386 | 512 | struct btrfs_block_group *block_group = caching_ctl->block_group; |
9f21246d JB |
513 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
514 | struct btrfs_root *extent_root = fs_info->extent_root; | |
515 | struct btrfs_path *path; | |
516 | struct extent_buffer *leaf; | |
517 | struct btrfs_key key; | |
518 | u64 total_found = 0; | |
519 | u64 last = 0; | |
520 | u32 nritems; | |
521 | int ret; | |
522 | bool wakeup = true; | |
523 | ||
524 | path = btrfs_alloc_path(); | |
525 | if (!path) | |
526 | return -ENOMEM; | |
527 | ||
b3470b5d | 528 | last = max_t(u64, block_group->start, BTRFS_SUPER_INFO_OFFSET); |
9f21246d JB |
529 | |
530 | #ifdef CONFIG_BTRFS_DEBUG | |
531 | /* | |
532 | * If we're fragmenting we don't want to make anybody think we can | |
533 | * allocate from this block group until we've had a chance to fragment | |
534 | * the free space. | |
535 | */ | |
536 | if (btrfs_should_fragment_free_space(block_group)) | |
537 | wakeup = false; | |
538 | #endif | |
539 | /* | |
540 | * We don't want to deadlock with somebody trying to allocate a new | |
541 | * extent for the extent root while also trying to search the extent | |
542 | * root to add free space. So we skip locking and search the commit | |
543 | * root, since its read-only | |
544 | */ | |
545 | path->skip_locking = 1; | |
546 | path->search_commit_root = 1; | |
547 | path->reada = READA_FORWARD; | |
548 | ||
549 | key.objectid = last; | |
550 | key.offset = 0; | |
551 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
552 | ||
553 | next: | |
554 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
555 | if (ret < 0) | |
556 | goto out; | |
557 | ||
558 | leaf = path->nodes[0]; | |
559 | nritems = btrfs_header_nritems(leaf); | |
560 | ||
561 | while (1) { | |
562 | if (btrfs_fs_closing(fs_info) > 1) { | |
563 | last = (u64)-1; | |
564 | break; | |
565 | } | |
566 | ||
567 | if (path->slots[0] < nritems) { | |
568 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
569 | } else { | |
570 | ret = btrfs_find_next_key(extent_root, path, &key, 0, 0); | |
571 | if (ret) | |
572 | break; | |
573 | ||
574 | if (need_resched() || | |
575 | rwsem_is_contended(&fs_info->commit_root_sem)) { | |
576 | if (wakeup) | |
577 | caching_ctl->progress = last; | |
578 | btrfs_release_path(path); | |
579 | up_read(&fs_info->commit_root_sem); | |
580 | mutex_unlock(&caching_ctl->mutex); | |
581 | cond_resched(); | |
582 | mutex_lock(&caching_ctl->mutex); | |
583 | down_read(&fs_info->commit_root_sem); | |
584 | goto next; | |
585 | } | |
586 | ||
587 | ret = btrfs_next_leaf(extent_root, path); | |
588 | if (ret < 0) | |
589 | goto out; | |
590 | if (ret) | |
591 | break; | |
592 | leaf = path->nodes[0]; | |
593 | nritems = btrfs_header_nritems(leaf); | |
594 | continue; | |
595 | } | |
596 | ||
597 | if (key.objectid < last) { | |
598 | key.objectid = last; | |
599 | key.offset = 0; | |
600 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
601 | ||
602 | if (wakeup) | |
603 | caching_ctl->progress = last; | |
604 | btrfs_release_path(path); | |
605 | goto next; | |
606 | } | |
607 | ||
b3470b5d | 608 | if (key.objectid < block_group->start) { |
9f21246d JB |
609 | path->slots[0]++; |
610 | continue; | |
611 | } | |
612 | ||
b3470b5d | 613 | if (key.objectid >= block_group->start + block_group->length) |
9f21246d JB |
614 | break; |
615 | ||
616 | if (key.type == BTRFS_EXTENT_ITEM_KEY || | |
617 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
618 | total_found += add_new_free_space(block_group, last, | |
619 | key.objectid); | |
620 | if (key.type == BTRFS_METADATA_ITEM_KEY) | |
621 | last = key.objectid + | |
622 | fs_info->nodesize; | |
623 | else | |
624 | last = key.objectid + key.offset; | |
625 | ||
626 | if (total_found > CACHING_CTL_WAKE_UP) { | |
627 | total_found = 0; | |
628 | if (wakeup) | |
629 | wake_up(&caching_ctl->wait); | |
630 | } | |
631 | } | |
632 | path->slots[0]++; | |
633 | } | |
634 | ret = 0; | |
635 | ||
636 | total_found += add_new_free_space(block_group, last, | |
b3470b5d | 637 | block_group->start + block_group->length); |
9f21246d JB |
638 | caching_ctl->progress = (u64)-1; |
639 | ||
640 | out: | |
641 | btrfs_free_path(path); | |
642 | return ret; | |
643 | } | |
644 | ||
645 | static noinline void caching_thread(struct btrfs_work *work) | |
646 | { | |
32da5386 | 647 | struct btrfs_block_group *block_group; |
9f21246d JB |
648 | struct btrfs_fs_info *fs_info; |
649 | struct btrfs_caching_control *caching_ctl; | |
650 | int ret; | |
651 | ||
652 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
653 | block_group = caching_ctl->block_group; | |
654 | fs_info = block_group->fs_info; | |
655 | ||
656 | mutex_lock(&caching_ctl->mutex); | |
657 | down_read(&fs_info->commit_root_sem); | |
658 | ||
e747853c JB |
659 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
660 | ret = load_free_space_cache(block_group); | |
661 | if (ret == 1) { | |
662 | ret = 0; | |
663 | goto done; | |
664 | } | |
665 | ||
666 | /* | |
667 | * We failed to load the space cache, set ourselves to | |
668 | * CACHE_STARTED and carry on. | |
669 | */ | |
670 | spin_lock(&block_group->lock); | |
671 | block_group->cached = BTRFS_CACHE_STARTED; | |
672 | spin_unlock(&block_group->lock); | |
673 | wake_up(&caching_ctl->wait); | |
674 | } | |
675 | ||
9f21246d JB |
676 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
677 | ret = load_free_space_tree(caching_ctl); | |
678 | else | |
679 | ret = load_extent_tree_free(caching_ctl); | |
e747853c | 680 | done: |
9f21246d JB |
681 | spin_lock(&block_group->lock); |
682 | block_group->caching_ctl = NULL; | |
683 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; | |
684 | spin_unlock(&block_group->lock); | |
685 | ||
686 | #ifdef CONFIG_BTRFS_DEBUG | |
687 | if (btrfs_should_fragment_free_space(block_group)) { | |
688 | u64 bytes_used; | |
689 | ||
690 | spin_lock(&block_group->space_info->lock); | |
691 | spin_lock(&block_group->lock); | |
b3470b5d | 692 | bytes_used = block_group->length - block_group->used; |
9f21246d JB |
693 | block_group->space_info->bytes_used += bytes_used >> 1; |
694 | spin_unlock(&block_group->lock); | |
695 | spin_unlock(&block_group->space_info->lock); | |
e11c0406 | 696 | fragment_free_space(block_group); |
9f21246d JB |
697 | } |
698 | #endif | |
699 | ||
700 | caching_ctl->progress = (u64)-1; | |
701 | ||
702 | up_read(&fs_info->commit_root_sem); | |
703 | btrfs_free_excluded_extents(block_group); | |
704 | mutex_unlock(&caching_ctl->mutex); | |
705 | ||
706 | wake_up(&caching_ctl->wait); | |
707 | ||
708 | btrfs_put_caching_control(caching_ctl); | |
709 | btrfs_put_block_group(block_group); | |
710 | } | |
711 | ||
32da5386 | 712 | int btrfs_cache_block_group(struct btrfs_block_group *cache, int load_cache_only) |
9f21246d JB |
713 | { |
714 | DEFINE_WAIT(wait); | |
715 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
e747853c | 716 | struct btrfs_caching_control *caching_ctl = NULL; |
9f21246d JB |
717 | int ret = 0; |
718 | ||
719 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); | |
720 | if (!caching_ctl) | |
721 | return -ENOMEM; | |
722 | ||
723 | INIT_LIST_HEAD(&caching_ctl->list); | |
724 | mutex_init(&caching_ctl->mutex); | |
725 | init_waitqueue_head(&caching_ctl->wait); | |
726 | caching_ctl->block_group = cache; | |
b3470b5d | 727 | caching_ctl->progress = cache->start; |
e747853c | 728 | refcount_set(&caching_ctl->count, 2); |
a0cac0ec | 729 | btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL); |
9f21246d JB |
730 | |
731 | spin_lock(&cache->lock); | |
9f21246d | 732 | if (cache->cached != BTRFS_CACHE_NO) { |
9f21246d | 733 | kfree(caching_ctl); |
e747853c JB |
734 | |
735 | caching_ctl = cache->caching_ctl; | |
736 | if (caching_ctl) | |
737 | refcount_inc(&caching_ctl->count); | |
738 | spin_unlock(&cache->lock); | |
739 | goto out; | |
9f21246d JB |
740 | } |
741 | WARN_ON(cache->caching_ctl); | |
742 | cache->caching_ctl = caching_ctl; | |
e747853c JB |
743 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
744 | cache->cached = BTRFS_CACHE_FAST; | |
745 | else | |
746 | cache->cached = BTRFS_CACHE_STARTED; | |
747 | cache->has_caching_ctl = 1; | |
9f21246d JB |
748 | spin_unlock(&cache->lock); |
749 | ||
bbb86a37 | 750 | spin_lock(&fs_info->block_group_cache_lock); |
9f21246d JB |
751 | refcount_inc(&caching_ctl->count); |
752 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); | |
bbb86a37 | 753 | spin_unlock(&fs_info->block_group_cache_lock); |
9f21246d JB |
754 | |
755 | btrfs_get_block_group(cache); | |
756 | ||
757 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); | |
e747853c JB |
758 | out: |
759 | if (load_cache_only && caching_ctl) | |
760 | btrfs_wait_space_cache_v1_finished(cache, caching_ctl); | |
761 | if (caching_ctl) | |
762 | btrfs_put_caching_control(caching_ctl); | |
9f21246d JB |
763 | |
764 | return ret; | |
765 | } | |
e3e0520b JB |
766 | |
767 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
768 | { | |
769 | u64 extra_flags = chunk_to_extended(flags) & | |
770 | BTRFS_EXTENDED_PROFILE_MASK; | |
771 | ||
772 | write_seqlock(&fs_info->profiles_lock); | |
773 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
774 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
775 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
776 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
777 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
778 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
779 | write_sequnlock(&fs_info->profiles_lock); | |
780 | } | |
781 | ||
782 | /* | |
783 | * Clear incompat bits for the following feature(s): | |
784 | * | |
785 | * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group | |
786 | * in the whole filesystem | |
9c907446 DS |
787 | * |
788 | * - RAID1C34 - same as above for RAID1C3 and RAID1C4 block groups | |
e3e0520b JB |
789 | */ |
790 | static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
791 | { | |
9c907446 DS |
792 | bool found_raid56 = false; |
793 | bool found_raid1c34 = false; | |
794 | ||
795 | if ((flags & BTRFS_BLOCK_GROUP_RAID56_MASK) || | |
796 | (flags & BTRFS_BLOCK_GROUP_RAID1C3) || | |
797 | (flags & BTRFS_BLOCK_GROUP_RAID1C4)) { | |
e3e0520b JB |
798 | struct list_head *head = &fs_info->space_info; |
799 | struct btrfs_space_info *sinfo; | |
800 | ||
801 | list_for_each_entry_rcu(sinfo, head, list) { | |
e3e0520b JB |
802 | down_read(&sinfo->groups_sem); |
803 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5])) | |
9c907446 | 804 | found_raid56 = true; |
e3e0520b | 805 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6])) |
9c907446 DS |
806 | found_raid56 = true; |
807 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C3])) | |
808 | found_raid1c34 = true; | |
809 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C4])) | |
810 | found_raid1c34 = true; | |
e3e0520b | 811 | up_read(&sinfo->groups_sem); |
e3e0520b | 812 | } |
d8e6fd5c | 813 | if (!found_raid56) |
9c907446 | 814 | btrfs_clear_fs_incompat(fs_info, RAID56); |
d8e6fd5c | 815 | if (!found_raid1c34) |
9c907446 | 816 | btrfs_clear_fs_incompat(fs_info, RAID1C34); |
e3e0520b JB |
817 | } |
818 | } | |
819 | ||
7357623a QW |
820 | static int remove_block_group_item(struct btrfs_trans_handle *trans, |
821 | struct btrfs_path *path, | |
822 | struct btrfs_block_group *block_group) | |
823 | { | |
824 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
825 | struct btrfs_root *root; | |
826 | struct btrfs_key key; | |
827 | int ret; | |
828 | ||
829 | root = fs_info->extent_root; | |
830 | key.objectid = block_group->start; | |
831 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
832 | key.offset = block_group->length; | |
833 | ||
834 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
835 | if (ret > 0) | |
836 | ret = -ENOENT; | |
837 | if (ret < 0) | |
838 | return ret; | |
839 | ||
840 | ret = btrfs_del_item(trans, root, path); | |
841 | return ret; | |
842 | } | |
843 | ||
e3e0520b JB |
844 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
845 | u64 group_start, struct extent_map *em) | |
846 | { | |
847 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
e3e0520b | 848 | struct btrfs_path *path; |
32da5386 | 849 | struct btrfs_block_group *block_group; |
e3e0520b JB |
850 | struct btrfs_free_cluster *cluster; |
851 | struct btrfs_root *tree_root = fs_info->tree_root; | |
852 | struct btrfs_key key; | |
853 | struct inode *inode; | |
854 | struct kobject *kobj = NULL; | |
855 | int ret; | |
856 | int index; | |
857 | int factor; | |
858 | struct btrfs_caching_control *caching_ctl = NULL; | |
859 | bool remove_em; | |
860 | bool remove_rsv = false; | |
861 | ||
862 | block_group = btrfs_lookup_block_group(fs_info, group_start); | |
863 | BUG_ON(!block_group); | |
864 | BUG_ON(!block_group->ro); | |
865 | ||
866 | trace_btrfs_remove_block_group(block_group); | |
867 | /* | |
868 | * Free the reserved super bytes from this block group before | |
869 | * remove it. | |
870 | */ | |
871 | btrfs_free_excluded_extents(block_group); | |
b3470b5d DS |
872 | btrfs_free_ref_tree_range(fs_info, block_group->start, |
873 | block_group->length); | |
e3e0520b | 874 | |
e3e0520b JB |
875 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
876 | factor = btrfs_bg_type_to_factor(block_group->flags); | |
877 | ||
878 | /* make sure this block group isn't part of an allocation cluster */ | |
879 | cluster = &fs_info->data_alloc_cluster; | |
880 | spin_lock(&cluster->refill_lock); | |
881 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
882 | spin_unlock(&cluster->refill_lock); | |
883 | ||
884 | /* | |
885 | * make sure this block group isn't part of a metadata | |
886 | * allocation cluster | |
887 | */ | |
888 | cluster = &fs_info->meta_alloc_cluster; | |
889 | spin_lock(&cluster->refill_lock); | |
890 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
891 | spin_unlock(&cluster->refill_lock); | |
892 | ||
893 | path = btrfs_alloc_path(); | |
894 | if (!path) { | |
895 | ret = -ENOMEM; | |
9fecd132 | 896 | goto out; |
e3e0520b JB |
897 | } |
898 | ||
899 | /* | |
900 | * get the inode first so any iput calls done for the io_list | |
901 | * aren't the final iput (no unlinks allowed now) | |
902 | */ | |
903 | inode = lookup_free_space_inode(block_group, path); | |
904 | ||
905 | mutex_lock(&trans->transaction->cache_write_mutex); | |
906 | /* | |
907 | * Make sure our free space cache IO is done before removing the | |
908 | * free space inode | |
909 | */ | |
910 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
911 | if (!list_empty(&block_group->io_list)) { | |
912 | list_del_init(&block_group->io_list); | |
913 | ||
914 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
915 | ||
916 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
917 | btrfs_wait_cache_io(trans, block_group, path); | |
918 | btrfs_put_block_group(block_group); | |
919 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
920 | } | |
921 | ||
922 | if (!list_empty(&block_group->dirty_list)) { | |
923 | list_del_init(&block_group->dirty_list); | |
924 | remove_rsv = true; | |
925 | btrfs_put_block_group(block_group); | |
926 | } | |
927 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
928 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
929 | ||
930 | if (!IS_ERR(inode)) { | |
931 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); | |
932 | if (ret) { | |
933 | btrfs_add_delayed_iput(inode); | |
9fecd132 | 934 | goto out; |
e3e0520b JB |
935 | } |
936 | clear_nlink(inode); | |
937 | /* One for the block groups ref */ | |
938 | spin_lock(&block_group->lock); | |
939 | if (block_group->iref) { | |
940 | block_group->iref = 0; | |
941 | block_group->inode = NULL; | |
942 | spin_unlock(&block_group->lock); | |
943 | iput(inode); | |
944 | } else { | |
945 | spin_unlock(&block_group->lock); | |
946 | } | |
947 | /* One for our lookup ref */ | |
948 | btrfs_add_delayed_iput(inode); | |
949 | } | |
950 | ||
951 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
e3e0520b | 952 | key.type = 0; |
b3470b5d | 953 | key.offset = block_group->start; |
e3e0520b JB |
954 | |
955 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
956 | if (ret < 0) | |
9fecd132 | 957 | goto out; |
e3e0520b JB |
958 | if (ret > 0) |
959 | btrfs_release_path(path); | |
960 | if (ret == 0) { | |
961 | ret = btrfs_del_item(trans, tree_root, path); | |
962 | if (ret) | |
9fecd132 | 963 | goto out; |
e3e0520b JB |
964 | btrfs_release_path(path); |
965 | } | |
966 | ||
967 | spin_lock(&fs_info->block_group_cache_lock); | |
968 | rb_erase(&block_group->cache_node, | |
969 | &fs_info->block_group_cache_tree); | |
970 | RB_CLEAR_NODE(&block_group->cache_node); | |
971 | ||
9fecd132 FM |
972 | /* Once for the block groups rbtree */ |
973 | btrfs_put_block_group(block_group); | |
974 | ||
b3470b5d | 975 | if (fs_info->first_logical_byte == block_group->start) |
e3e0520b JB |
976 | fs_info->first_logical_byte = (u64)-1; |
977 | spin_unlock(&fs_info->block_group_cache_lock); | |
978 | ||
979 | down_write(&block_group->space_info->groups_sem); | |
980 | /* | |
981 | * we must use list_del_init so people can check to see if they | |
982 | * are still on the list after taking the semaphore | |
983 | */ | |
984 | list_del_init(&block_group->list); | |
985 | if (list_empty(&block_group->space_info->block_groups[index])) { | |
986 | kobj = block_group->space_info->block_group_kobjs[index]; | |
987 | block_group->space_info->block_group_kobjs[index] = NULL; | |
988 | clear_avail_alloc_bits(fs_info, block_group->flags); | |
989 | } | |
990 | up_write(&block_group->space_info->groups_sem); | |
991 | clear_incompat_bg_bits(fs_info, block_group->flags); | |
992 | if (kobj) { | |
993 | kobject_del(kobj); | |
994 | kobject_put(kobj); | |
995 | } | |
996 | ||
997 | if (block_group->has_caching_ctl) | |
998 | caching_ctl = btrfs_get_caching_control(block_group); | |
999 | if (block_group->cached == BTRFS_CACHE_STARTED) | |
1000 | btrfs_wait_block_group_cache_done(block_group); | |
1001 | if (block_group->has_caching_ctl) { | |
bbb86a37 | 1002 | spin_lock(&fs_info->block_group_cache_lock); |
e3e0520b JB |
1003 | if (!caching_ctl) { |
1004 | struct btrfs_caching_control *ctl; | |
1005 | ||
1006 | list_for_each_entry(ctl, | |
1007 | &fs_info->caching_block_groups, list) | |
1008 | if (ctl->block_group == block_group) { | |
1009 | caching_ctl = ctl; | |
1010 | refcount_inc(&caching_ctl->count); | |
1011 | break; | |
1012 | } | |
1013 | } | |
1014 | if (caching_ctl) | |
1015 | list_del_init(&caching_ctl->list); | |
bbb86a37 | 1016 | spin_unlock(&fs_info->block_group_cache_lock); |
e3e0520b JB |
1017 | if (caching_ctl) { |
1018 | /* Once for the caching bgs list and once for us. */ | |
1019 | btrfs_put_caching_control(caching_ctl); | |
1020 | btrfs_put_caching_control(caching_ctl); | |
1021 | } | |
1022 | } | |
1023 | ||
1024 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
1025 | WARN_ON(!list_empty(&block_group->dirty_list)); | |
1026 | WARN_ON(!list_empty(&block_group->io_list)); | |
1027 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
1028 | ||
1029 | btrfs_remove_free_space_cache(block_group); | |
1030 | ||
1031 | spin_lock(&block_group->space_info->lock); | |
1032 | list_del_init(&block_group->ro_list); | |
1033 | ||
1034 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { | |
1035 | WARN_ON(block_group->space_info->total_bytes | |
b3470b5d | 1036 | < block_group->length); |
e3e0520b | 1037 | WARN_ON(block_group->space_info->bytes_readonly |
b3470b5d | 1038 | < block_group->length); |
e3e0520b | 1039 | WARN_ON(block_group->space_info->disk_total |
b3470b5d | 1040 | < block_group->length * factor); |
e3e0520b | 1041 | } |
b3470b5d DS |
1042 | block_group->space_info->total_bytes -= block_group->length; |
1043 | block_group->space_info->bytes_readonly -= block_group->length; | |
1044 | block_group->space_info->disk_total -= block_group->length * factor; | |
e3e0520b JB |
1045 | |
1046 | spin_unlock(&block_group->space_info->lock); | |
1047 | ||
ffcb9d44 FM |
1048 | /* |
1049 | * Remove the free space for the block group from the free space tree | |
1050 | * and the block group's item from the extent tree before marking the | |
1051 | * block group as removed. This is to prevent races with tasks that | |
1052 | * freeze and unfreeze a block group, this task and another task | |
1053 | * allocating a new block group - the unfreeze task ends up removing | |
1054 | * the block group's extent map before the task calling this function | |
1055 | * deletes the block group item from the extent tree, allowing for | |
1056 | * another task to attempt to create another block group with the same | |
1057 | * item key (and failing with -EEXIST and a transaction abort). | |
1058 | */ | |
1059 | ret = remove_block_group_free_space(trans, block_group); | |
1060 | if (ret) | |
1061 | goto out; | |
1062 | ||
1063 | ret = remove_block_group_item(trans, path, block_group); | |
1064 | if (ret < 0) | |
1065 | goto out; | |
1066 | ||
e3e0520b JB |
1067 | spin_lock(&block_group->lock); |
1068 | block_group->removed = 1; | |
1069 | /* | |
6b7304af FM |
1070 | * At this point trimming or scrub can't start on this block group, |
1071 | * because we removed the block group from the rbtree | |
1072 | * fs_info->block_group_cache_tree so no one can't find it anymore and | |
1073 | * even if someone already got this block group before we removed it | |
1074 | * from the rbtree, they have already incremented block_group->frozen - | |
1075 | * if they didn't, for the trimming case they won't find any free space | |
1076 | * entries because we already removed them all when we called | |
1077 | * btrfs_remove_free_space_cache(). | |
e3e0520b JB |
1078 | * |
1079 | * And we must not remove the extent map from the fs_info->mapping_tree | |
1080 | * to prevent the same logical address range and physical device space | |
6b7304af FM |
1081 | * ranges from being reused for a new block group. This is needed to |
1082 | * avoid races with trimming and scrub. | |
1083 | * | |
1084 | * An fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
e3e0520b JB |
1085 | * completely transactionless, so while it is trimming a range the |
1086 | * currently running transaction might finish and a new one start, | |
1087 | * allowing for new block groups to be created that can reuse the same | |
1088 | * physical device locations unless we take this special care. | |
1089 | * | |
1090 | * There may also be an implicit trim operation if the file system | |
1091 | * is mounted with -odiscard. The same protections must remain | |
1092 | * in place until the extents have been discarded completely when | |
1093 | * the transaction commit has completed. | |
1094 | */ | |
6b7304af | 1095 | remove_em = (atomic_read(&block_group->frozen) == 0); |
e3e0520b JB |
1096 | spin_unlock(&block_group->lock); |
1097 | ||
e3e0520b JB |
1098 | if (remove_em) { |
1099 | struct extent_map_tree *em_tree; | |
1100 | ||
1101 | em_tree = &fs_info->mapping_tree; | |
1102 | write_lock(&em_tree->lock); | |
1103 | remove_extent_mapping(em_tree, em); | |
1104 | write_unlock(&em_tree->lock); | |
1105 | /* once for the tree */ | |
1106 | free_extent_map(em); | |
1107 | } | |
f6033c5e | 1108 | |
9fecd132 | 1109 | out: |
f6033c5e XY |
1110 | /* Once for the lookup reference */ |
1111 | btrfs_put_block_group(block_group); | |
e3e0520b JB |
1112 | if (remove_rsv) |
1113 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1114 | btrfs_free_path(path); | |
1115 | return ret; | |
1116 | } | |
1117 | ||
1118 | struct btrfs_trans_handle *btrfs_start_trans_remove_block_group( | |
1119 | struct btrfs_fs_info *fs_info, const u64 chunk_offset) | |
1120 | { | |
1121 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; | |
1122 | struct extent_map *em; | |
1123 | struct map_lookup *map; | |
1124 | unsigned int num_items; | |
1125 | ||
1126 | read_lock(&em_tree->lock); | |
1127 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
1128 | read_unlock(&em_tree->lock); | |
1129 | ASSERT(em && em->start == chunk_offset); | |
1130 | ||
1131 | /* | |
1132 | * We need to reserve 3 + N units from the metadata space info in order | |
1133 | * to remove a block group (done at btrfs_remove_chunk() and at | |
1134 | * btrfs_remove_block_group()), which are used for: | |
1135 | * | |
1136 | * 1 unit for adding the free space inode's orphan (located in the tree | |
1137 | * of tree roots). | |
1138 | * 1 unit for deleting the block group item (located in the extent | |
1139 | * tree). | |
1140 | * 1 unit for deleting the free space item (located in tree of tree | |
1141 | * roots). | |
1142 | * N units for deleting N device extent items corresponding to each | |
1143 | * stripe (located in the device tree). | |
1144 | * | |
1145 | * In order to remove a block group we also need to reserve units in the | |
1146 | * system space info in order to update the chunk tree (update one or | |
1147 | * more device items and remove one chunk item), but this is done at | |
1148 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
1149 | */ | |
1150 | map = em->map_lookup; | |
1151 | num_items = 3 + map->num_stripes; | |
1152 | free_extent_map(em); | |
1153 | ||
1154 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, | |
7f9fe614 | 1155 | num_items); |
e3e0520b JB |
1156 | } |
1157 | ||
26ce2095 JB |
1158 | /* |
1159 | * Mark block group @cache read-only, so later write won't happen to block | |
1160 | * group @cache. | |
1161 | * | |
1162 | * If @force is not set, this function will only mark the block group readonly | |
1163 | * if we have enough free space (1M) in other metadata/system block groups. | |
1164 | * If @force is not set, this function will mark the block group readonly | |
1165 | * without checking free space. | |
1166 | * | |
1167 | * NOTE: This function doesn't care if other block groups can contain all the | |
1168 | * data in this block group. That check should be done by relocation routine, | |
1169 | * not this function. | |
1170 | */ | |
32da5386 | 1171 | static int inc_block_group_ro(struct btrfs_block_group *cache, int force) |
26ce2095 JB |
1172 | { |
1173 | struct btrfs_space_info *sinfo = cache->space_info; | |
1174 | u64 num_bytes; | |
26ce2095 JB |
1175 | int ret = -ENOSPC; |
1176 | ||
26ce2095 JB |
1177 | spin_lock(&sinfo->lock); |
1178 | spin_lock(&cache->lock); | |
1179 | ||
1180 | if (cache->ro) { | |
1181 | cache->ro++; | |
1182 | ret = 0; | |
1183 | goto out; | |
1184 | } | |
1185 | ||
b3470b5d | 1186 | num_bytes = cache->length - cache->reserved - cache->pinned - |
bf38be65 | 1187 | cache->bytes_super - cache->used; |
26ce2095 JB |
1188 | |
1189 | /* | |
a30a3d20 JB |
1190 | * Data never overcommits, even in mixed mode, so do just the straight |
1191 | * check of left over space in how much we have allocated. | |
26ce2095 | 1192 | */ |
a30a3d20 JB |
1193 | if (force) { |
1194 | ret = 0; | |
1195 | } else if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) { | |
1196 | u64 sinfo_used = btrfs_space_info_used(sinfo, true); | |
1197 | ||
1198 | /* | |
1199 | * Here we make sure if we mark this bg RO, we still have enough | |
1200 | * free space as buffer. | |
1201 | */ | |
1202 | if (sinfo_used + num_bytes <= sinfo->total_bytes) | |
1203 | ret = 0; | |
1204 | } else { | |
1205 | /* | |
1206 | * We overcommit metadata, so we need to do the | |
1207 | * btrfs_can_overcommit check here, and we need to pass in | |
1208 | * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of | |
1209 | * leeway to allow us to mark this block group as read only. | |
1210 | */ | |
1211 | if (btrfs_can_overcommit(cache->fs_info, sinfo, num_bytes, | |
1212 | BTRFS_RESERVE_NO_FLUSH)) | |
1213 | ret = 0; | |
1214 | } | |
1215 | ||
1216 | if (!ret) { | |
26ce2095 JB |
1217 | sinfo->bytes_readonly += num_bytes; |
1218 | cache->ro++; | |
1219 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); | |
26ce2095 JB |
1220 | } |
1221 | out: | |
1222 | spin_unlock(&cache->lock); | |
1223 | spin_unlock(&sinfo->lock); | |
1224 | if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { | |
1225 | btrfs_info(cache->fs_info, | |
b3470b5d | 1226 | "unable to make block group %llu ro", cache->start); |
26ce2095 JB |
1227 | btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0); |
1228 | } | |
1229 | return ret; | |
1230 | } | |
1231 | ||
fe119a6e NB |
1232 | static bool clean_pinned_extents(struct btrfs_trans_handle *trans, |
1233 | struct btrfs_block_group *bg) | |
45bb5d6a NB |
1234 | { |
1235 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
fe119a6e | 1236 | struct btrfs_transaction *prev_trans = NULL; |
45bb5d6a NB |
1237 | const u64 start = bg->start; |
1238 | const u64 end = start + bg->length - 1; | |
1239 | int ret; | |
1240 | ||
fe119a6e NB |
1241 | spin_lock(&fs_info->trans_lock); |
1242 | if (trans->transaction->list.prev != &fs_info->trans_list) { | |
1243 | prev_trans = list_last_entry(&trans->transaction->list, | |
1244 | struct btrfs_transaction, list); | |
1245 | refcount_inc(&prev_trans->use_count); | |
1246 | } | |
1247 | spin_unlock(&fs_info->trans_lock); | |
1248 | ||
45bb5d6a NB |
1249 | /* |
1250 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
1251 | * btrfs_finish_extent_commit(). If we are at transaction N, another | |
1252 | * task might be running finish_extent_commit() for the previous | |
1253 | * transaction N - 1, and have seen a range belonging to the block | |
fe119a6e NB |
1254 | * group in pinned_extents before we were able to clear the whole block |
1255 | * group range from pinned_extents. This means that task can lookup for | |
1256 | * the block group after we unpinned it from pinned_extents and removed | |
1257 | * it, leading to a BUG_ON() at unpin_extent_range(). | |
45bb5d6a NB |
1258 | */ |
1259 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
fe119a6e NB |
1260 | if (prev_trans) { |
1261 | ret = clear_extent_bits(&prev_trans->pinned_extents, start, end, | |
1262 | EXTENT_DIRTY); | |
1263 | if (ret) | |
534cf531 | 1264 | goto out; |
fe119a6e | 1265 | } |
45bb5d6a | 1266 | |
fe119a6e | 1267 | ret = clear_extent_bits(&trans->transaction->pinned_extents, start, end, |
45bb5d6a | 1268 | EXTENT_DIRTY); |
534cf531 | 1269 | out: |
45bb5d6a | 1270 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
5150bf19 FM |
1271 | if (prev_trans) |
1272 | btrfs_put_transaction(prev_trans); | |
45bb5d6a | 1273 | |
534cf531 | 1274 | return ret == 0; |
45bb5d6a NB |
1275 | } |
1276 | ||
e3e0520b JB |
1277 | /* |
1278 | * Process the unused_bgs list and remove any that don't have any allocated | |
1279 | * space inside of them. | |
1280 | */ | |
1281 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
1282 | { | |
32da5386 | 1283 | struct btrfs_block_group *block_group; |
e3e0520b JB |
1284 | struct btrfs_space_info *space_info; |
1285 | struct btrfs_trans_handle *trans; | |
6e80d4f8 | 1286 | const bool async_trim_enabled = btrfs_test_opt(fs_info, DISCARD_ASYNC); |
e3e0520b JB |
1287 | int ret = 0; |
1288 | ||
1289 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) | |
1290 | return; | |
1291 | ||
1292 | spin_lock(&fs_info->unused_bgs_lock); | |
1293 | while (!list_empty(&fs_info->unused_bgs)) { | |
e3e0520b JB |
1294 | int trimming; |
1295 | ||
1296 | block_group = list_first_entry(&fs_info->unused_bgs, | |
32da5386 | 1297 | struct btrfs_block_group, |
e3e0520b JB |
1298 | bg_list); |
1299 | list_del_init(&block_group->bg_list); | |
1300 | ||
1301 | space_info = block_group->space_info; | |
1302 | ||
1303 | if (ret || btrfs_mixed_space_info(space_info)) { | |
1304 | btrfs_put_block_group(block_group); | |
1305 | continue; | |
1306 | } | |
1307 | spin_unlock(&fs_info->unused_bgs_lock); | |
1308 | ||
b0643e59 DZ |
1309 | btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); |
1310 | ||
e3e0520b JB |
1311 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
1312 | ||
1313 | /* Don't want to race with allocators so take the groups_sem */ | |
1314 | down_write(&space_info->groups_sem); | |
6e80d4f8 DZ |
1315 | |
1316 | /* | |
1317 | * Async discard moves the final block group discard to be prior | |
1318 | * to the unused_bgs code path. Therefore, if it's not fully | |
1319 | * trimmed, punt it back to the async discard lists. | |
1320 | */ | |
1321 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC) && | |
1322 | !btrfs_is_free_space_trimmed(block_group)) { | |
1323 | trace_btrfs_skip_unused_block_group(block_group); | |
1324 | up_write(&space_info->groups_sem); | |
1325 | /* Requeue if we failed because of async discard */ | |
1326 | btrfs_discard_queue_work(&fs_info->discard_ctl, | |
1327 | block_group); | |
1328 | goto next; | |
1329 | } | |
1330 | ||
e3e0520b JB |
1331 | spin_lock(&block_group->lock); |
1332 | if (block_group->reserved || block_group->pinned || | |
bf38be65 | 1333 | block_group->used || block_group->ro || |
e3e0520b JB |
1334 | list_is_singular(&block_group->list)) { |
1335 | /* | |
1336 | * We want to bail if we made new allocations or have | |
1337 | * outstanding allocations in this block group. We do | |
1338 | * the ro check in case balance is currently acting on | |
1339 | * this block group. | |
1340 | */ | |
1341 | trace_btrfs_skip_unused_block_group(block_group); | |
1342 | spin_unlock(&block_group->lock); | |
1343 | up_write(&space_info->groups_sem); | |
1344 | goto next; | |
1345 | } | |
1346 | spin_unlock(&block_group->lock); | |
1347 | ||
1348 | /* We don't want to force the issue, only flip if it's ok. */ | |
e11c0406 | 1349 | ret = inc_block_group_ro(block_group, 0); |
e3e0520b JB |
1350 | up_write(&space_info->groups_sem); |
1351 | if (ret < 0) { | |
1352 | ret = 0; | |
1353 | goto next; | |
1354 | } | |
1355 | ||
1356 | /* | |
1357 | * Want to do this before we do anything else so we can recover | |
1358 | * properly if we fail to join the transaction. | |
1359 | */ | |
1360 | trans = btrfs_start_trans_remove_block_group(fs_info, | |
b3470b5d | 1361 | block_group->start); |
e3e0520b JB |
1362 | if (IS_ERR(trans)) { |
1363 | btrfs_dec_block_group_ro(block_group); | |
1364 | ret = PTR_ERR(trans); | |
1365 | goto next; | |
1366 | } | |
1367 | ||
1368 | /* | |
1369 | * We could have pending pinned extents for this block group, | |
1370 | * just delete them, we don't care about them anymore. | |
1371 | */ | |
534cf531 FM |
1372 | if (!clean_pinned_extents(trans, block_group)) { |
1373 | btrfs_dec_block_group_ro(block_group); | |
e3e0520b | 1374 | goto end_trans; |
534cf531 | 1375 | } |
e3e0520b | 1376 | |
b0643e59 DZ |
1377 | /* |
1378 | * At this point, the block_group is read only and should fail | |
1379 | * new allocations. However, btrfs_finish_extent_commit() can | |
1380 | * cause this block_group to be placed back on the discard | |
1381 | * lists because now the block_group isn't fully discarded. | |
1382 | * Bail here and try again later after discarding everything. | |
1383 | */ | |
1384 | spin_lock(&fs_info->discard_ctl.lock); | |
1385 | if (!list_empty(&block_group->discard_list)) { | |
1386 | spin_unlock(&fs_info->discard_ctl.lock); | |
1387 | btrfs_dec_block_group_ro(block_group); | |
1388 | btrfs_discard_queue_work(&fs_info->discard_ctl, | |
1389 | block_group); | |
1390 | goto end_trans; | |
1391 | } | |
1392 | spin_unlock(&fs_info->discard_ctl.lock); | |
1393 | ||
e3e0520b JB |
1394 | /* Reset pinned so btrfs_put_block_group doesn't complain */ |
1395 | spin_lock(&space_info->lock); | |
1396 | spin_lock(&block_group->lock); | |
1397 | ||
1398 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, | |
1399 | -block_group->pinned); | |
1400 | space_info->bytes_readonly += block_group->pinned; | |
1401 | percpu_counter_add_batch(&space_info->total_bytes_pinned, | |
1402 | -block_group->pinned, | |
1403 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
1404 | block_group->pinned = 0; | |
1405 | ||
1406 | spin_unlock(&block_group->lock); | |
1407 | spin_unlock(&space_info->lock); | |
1408 | ||
6e80d4f8 DZ |
1409 | /* |
1410 | * The normal path here is an unused block group is passed here, | |
1411 | * then trimming is handled in the transaction commit path. | |
1412 | * Async discard interposes before this to do the trimming | |
1413 | * before coming down the unused block group path as trimming | |
1414 | * will no longer be done later in the transaction commit path. | |
1415 | */ | |
1416 | if (!async_trim_enabled && btrfs_test_opt(fs_info, DISCARD_ASYNC)) | |
1417 | goto flip_async; | |
1418 | ||
e3e0520b | 1419 | /* DISCARD can flip during remount */ |
46b27f50 | 1420 | trimming = btrfs_test_opt(fs_info, DISCARD_SYNC); |
e3e0520b JB |
1421 | |
1422 | /* Implicit trim during transaction commit. */ | |
1423 | if (trimming) | |
6b7304af | 1424 | btrfs_freeze_block_group(block_group); |
e3e0520b JB |
1425 | |
1426 | /* | |
1427 | * Btrfs_remove_chunk will abort the transaction if things go | |
1428 | * horribly wrong. | |
1429 | */ | |
b3470b5d | 1430 | ret = btrfs_remove_chunk(trans, block_group->start); |
e3e0520b JB |
1431 | |
1432 | if (ret) { | |
1433 | if (trimming) | |
6b7304af | 1434 | btrfs_unfreeze_block_group(block_group); |
e3e0520b JB |
1435 | goto end_trans; |
1436 | } | |
1437 | ||
1438 | /* | |
1439 | * If we're not mounted with -odiscard, we can just forget | |
1440 | * about this block group. Otherwise we'll need to wait | |
1441 | * until transaction commit to do the actual discard. | |
1442 | */ | |
1443 | if (trimming) { | |
1444 | spin_lock(&fs_info->unused_bgs_lock); | |
1445 | /* | |
1446 | * A concurrent scrub might have added us to the list | |
1447 | * fs_info->unused_bgs, so use a list_move operation | |
1448 | * to add the block group to the deleted_bgs list. | |
1449 | */ | |
1450 | list_move(&block_group->bg_list, | |
1451 | &trans->transaction->deleted_bgs); | |
1452 | spin_unlock(&fs_info->unused_bgs_lock); | |
1453 | btrfs_get_block_group(block_group); | |
1454 | } | |
1455 | end_trans: | |
1456 | btrfs_end_transaction(trans); | |
1457 | next: | |
1458 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
1459 | btrfs_put_block_group(block_group); | |
1460 | spin_lock(&fs_info->unused_bgs_lock); | |
1461 | } | |
1462 | spin_unlock(&fs_info->unused_bgs_lock); | |
6e80d4f8 DZ |
1463 | return; |
1464 | ||
1465 | flip_async: | |
1466 | btrfs_end_transaction(trans); | |
1467 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
1468 | btrfs_put_block_group(block_group); | |
1469 | btrfs_discard_punt_unused_bgs_list(fs_info); | |
e3e0520b JB |
1470 | } |
1471 | ||
32da5386 | 1472 | void btrfs_mark_bg_unused(struct btrfs_block_group *bg) |
e3e0520b JB |
1473 | { |
1474 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
1475 | ||
1476 | spin_lock(&fs_info->unused_bgs_lock); | |
1477 | if (list_empty(&bg->bg_list)) { | |
1478 | btrfs_get_block_group(bg); | |
1479 | trace_btrfs_add_unused_block_group(bg); | |
1480 | list_add_tail(&bg->bg_list, &fs_info->unused_bgs); | |
1481 | } | |
1482 | spin_unlock(&fs_info->unused_bgs_lock); | |
1483 | } | |
4358d963 | 1484 | |
e3ba67a1 JT |
1485 | static int read_bg_from_eb(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
1486 | struct btrfs_path *path) | |
1487 | { | |
1488 | struct extent_map_tree *em_tree; | |
1489 | struct extent_map *em; | |
1490 | struct btrfs_block_group_item bg; | |
1491 | struct extent_buffer *leaf; | |
1492 | int slot; | |
1493 | u64 flags; | |
1494 | int ret = 0; | |
1495 | ||
1496 | slot = path->slots[0]; | |
1497 | leaf = path->nodes[0]; | |
1498 | ||
1499 | em_tree = &fs_info->mapping_tree; | |
1500 | read_lock(&em_tree->lock); | |
1501 | em = lookup_extent_mapping(em_tree, key->objectid, key->offset); | |
1502 | read_unlock(&em_tree->lock); | |
1503 | if (!em) { | |
1504 | btrfs_err(fs_info, | |
1505 | "logical %llu len %llu found bg but no related chunk", | |
1506 | key->objectid, key->offset); | |
1507 | return -ENOENT; | |
1508 | } | |
1509 | ||
1510 | if (em->start != key->objectid || em->len != key->offset) { | |
1511 | btrfs_err(fs_info, | |
1512 | "block group %llu len %llu mismatch with chunk %llu len %llu", | |
1513 | key->objectid, key->offset, em->start, em->len); | |
1514 | ret = -EUCLEAN; | |
1515 | goto out_free_em; | |
1516 | } | |
1517 | ||
1518 | read_extent_buffer(leaf, &bg, btrfs_item_ptr_offset(leaf, slot), | |
1519 | sizeof(bg)); | |
1520 | flags = btrfs_stack_block_group_flags(&bg) & | |
1521 | BTRFS_BLOCK_GROUP_TYPE_MASK; | |
1522 | ||
1523 | if (flags != (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
1524 | btrfs_err(fs_info, | |
1525 | "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", | |
1526 | key->objectid, key->offset, flags, | |
1527 | (BTRFS_BLOCK_GROUP_TYPE_MASK & em->map_lookup->type)); | |
1528 | ret = -EUCLEAN; | |
1529 | } | |
1530 | ||
1531 | out_free_em: | |
1532 | free_extent_map(em); | |
1533 | return ret; | |
1534 | } | |
1535 | ||
4358d963 JB |
1536 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
1537 | struct btrfs_path *path, | |
1538 | struct btrfs_key *key) | |
1539 | { | |
1540 | struct btrfs_root *root = fs_info->extent_root; | |
e3ba67a1 | 1541 | int ret; |
4358d963 JB |
1542 | struct btrfs_key found_key; |
1543 | struct extent_buffer *leaf; | |
4358d963 JB |
1544 | int slot; |
1545 | ||
1546 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
1547 | if (ret < 0) | |
e3ba67a1 | 1548 | return ret; |
4358d963 JB |
1549 | |
1550 | while (1) { | |
1551 | slot = path->slots[0]; | |
1552 | leaf = path->nodes[0]; | |
1553 | if (slot >= btrfs_header_nritems(leaf)) { | |
1554 | ret = btrfs_next_leaf(root, path); | |
1555 | if (ret == 0) | |
1556 | continue; | |
1557 | if (ret < 0) | |
1558 | goto out; | |
1559 | break; | |
1560 | } | |
1561 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
1562 | ||
1563 | if (found_key.objectid >= key->objectid && | |
1564 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { | |
e3ba67a1 JT |
1565 | ret = read_bg_from_eb(fs_info, &found_key, path); |
1566 | break; | |
4358d963 | 1567 | } |
e3ba67a1 | 1568 | |
4358d963 JB |
1569 | path->slots[0]++; |
1570 | } | |
1571 | out: | |
1572 | return ret; | |
1573 | } | |
1574 | ||
1575 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
1576 | { | |
1577 | u64 extra_flags = chunk_to_extended(flags) & | |
1578 | BTRFS_EXTENDED_PROFILE_MASK; | |
1579 | ||
1580 | write_seqlock(&fs_info->profiles_lock); | |
1581 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
1582 | fs_info->avail_data_alloc_bits |= extra_flags; | |
1583 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
1584 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
1585 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
1586 | fs_info->avail_system_alloc_bits |= extra_flags; | |
1587 | write_sequnlock(&fs_info->profiles_lock); | |
1588 | } | |
1589 | ||
96a14336 NB |
1590 | /** |
1591 | * btrfs_rmap_block - Map a physical disk address to a list of logical addresses | |
1592 | * @chunk_start: logical address of block group | |
1593 | * @physical: physical address to map to logical addresses | |
1594 | * @logical: return array of logical addresses which map to @physical | |
1595 | * @naddrs: length of @logical | |
1596 | * @stripe_len: size of IO stripe for the given block group | |
1597 | * | |
1598 | * Maps a particular @physical disk address to a list of @logical addresses. | |
1599 | * Used primarily to exclude those portions of a block group that contain super | |
1600 | * block copies. | |
1601 | */ | |
1602 | EXPORT_FOR_TESTS | |
1603 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, | |
1604 | u64 physical, u64 **logical, int *naddrs, int *stripe_len) | |
1605 | { | |
1606 | struct extent_map *em; | |
1607 | struct map_lookup *map; | |
1608 | u64 *buf; | |
1609 | u64 bytenr; | |
1776ad17 NB |
1610 | u64 data_stripe_length; |
1611 | u64 io_stripe_size; | |
1612 | int i, nr = 0; | |
1613 | int ret = 0; | |
96a14336 NB |
1614 | |
1615 | em = btrfs_get_chunk_map(fs_info, chunk_start, 1); | |
1616 | if (IS_ERR(em)) | |
1617 | return -EIO; | |
1618 | ||
1619 | map = em->map_lookup; | |
9e22b925 | 1620 | data_stripe_length = em->orig_block_len; |
1776ad17 | 1621 | io_stripe_size = map->stripe_len; |
96a14336 | 1622 | |
9e22b925 NB |
1623 | /* For RAID5/6 adjust to a full IO stripe length */ |
1624 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
1776ad17 | 1625 | io_stripe_size = map->stripe_len * nr_data_stripes(map); |
96a14336 NB |
1626 | |
1627 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); | |
1776ad17 NB |
1628 | if (!buf) { |
1629 | ret = -ENOMEM; | |
1630 | goto out; | |
1631 | } | |
96a14336 NB |
1632 | |
1633 | for (i = 0; i < map->num_stripes; i++) { | |
1776ad17 NB |
1634 | bool already_inserted = false; |
1635 | u64 stripe_nr; | |
1636 | int j; | |
1637 | ||
1638 | if (!in_range(physical, map->stripes[i].physical, | |
1639 | data_stripe_length)) | |
96a14336 NB |
1640 | continue; |
1641 | ||
1642 | stripe_nr = physical - map->stripes[i].physical; | |
1643 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); | |
1644 | ||
1645 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
1646 | stripe_nr = stripe_nr * map->num_stripes + i; | |
1647 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); | |
1648 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
1649 | stripe_nr = stripe_nr * map->num_stripes + i; | |
1650 | } | |
1651 | /* | |
1652 | * The remaining case would be for RAID56, multiply by | |
1653 | * nr_data_stripes(). Alternatively, just use rmap_len below | |
1654 | * instead of map->stripe_len | |
1655 | */ | |
1656 | ||
1776ad17 NB |
1657 | bytenr = chunk_start + stripe_nr * io_stripe_size; |
1658 | ||
1659 | /* Ensure we don't add duplicate addresses */ | |
96a14336 | 1660 | for (j = 0; j < nr; j++) { |
1776ad17 NB |
1661 | if (buf[j] == bytenr) { |
1662 | already_inserted = true; | |
96a14336 | 1663 | break; |
1776ad17 | 1664 | } |
96a14336 | 1665 | } |
1776ad17 NB |
1666 | |
1667 | if (!already_inserted) | |
96a14336 | 1668 | buf[nr++] = bytenr; |
96a14336 NB |
1669 | } |
1670 | ||
1671 | *logical = buf; | |
1672 | *naddrs = nr; | |
1776ad17 NB |
1673 | *stripe_len = io_stripe_size; |
1674 | out: | |
96a14336 | 1675 | free_extent_map(em); |
1776ad17 | 1676 | return ret; |
96a14336 NB |
1677 | } |
1678 | ||
32da5386 | 1679 | static int exclude_super_stripes(struct btrfs_block_group *cache) |
4358d963 JB |
1680 | { |
1681 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
12659251 | 1682 | const bool zoned = btrfs_is_zoned(fs_info); |
4358d963 JB |
1683 | u64 bytenr; |
1684 | u64 *logical; | |
1685 | int stripe_len; | |
1686 | int i, nr, ret; | |
1687 | ||
b3470b5d DS |
1688 | if (cache->start < BTRFS_SUPER_INFO_OFFSET) { |
1689 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->start; | |
4358d963 | 1690 | cache->bytes_super += stripe_len; |
b3470b5d | 1691 | ret = btrfs_add_excluded_extent(fs_info, cache->start, |
4358d963 JB |
1692 | stripe_len); |
1693 | if (ret) | |
1694 | return ret; | |
1695 | } | |
1696 | ||
1697 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
1698 | bytenr = btrfs_sb_offset(i); | |
b3470b5d | 1699 | ret = btrfs_rmap_block(fs_info, cache->start, |
4358d963 JB |
1700 | bytenr, &logical, &nr, &stripe_len); |
1701 | if (ret) | |
1702 | return ret; | |
1703 | ||
12659251 NA |
1704 | /* Shouldn't have super stripes in sequential zones */ |
1705 | if (zoned && nr) { | |
1706 | btrfs_err(fs_info, | |
1707 | "zoned: block group %llu must not contain super block", | |
1708 | cache->start); | |
1709 | return -EUCLEAN; | |
1710 | } | |
1711 | ||
4358d963 | 1712 | while (nr--) { |
96f9b0f2 NB |
1713 | u64 len = min_t(u64, stripe_len, |
1714 | cache->start + cache->length - logical[nr]); | |
4358d963 JB |
1715 | |
1716 | cache->bytes_super += len; | |
96f9b0f2 NB |
1717 | ret = btrfs_add_excluded_extent(fs_info, logical[nr], |
1718 | len); | |
4358d963 JB |
1719 | if (ret) { |
1720 | kfree(logical); | |
1721 | return ret; | |
1722 | } | |
1723 | } | |
1724 | ||
1725 | kfree(logical); | |
1726 | } | |
1727 | return 0; | |
1728 | } | |
1729 | ||
32da5386 | 1730 | static void link_block_group(struct btrfs_block_group *cache) |
4358d963 JB |
1731 | { |
1732 | struct btrfs_space_info *space_info = cache->space_info; | |
1733 | int index = btrfs_bg_flags_to_raid_index(cache->flags); | |
4358d963 JB |
1734 | |
1735 | down_write(&space_info->groups_sem); | |
4358d963 JB |
1736 | list_add_tail(&cache->list, &space_info->block_groups[index]); |
1737 | up_write(&space_info->groups_sem); | |
4358d963 JB |
1738 | } |
1739 | ||
32da5386 | 1740 | static struct btrfs_block_group *btrfs_create_block_group_cache( |
9afc6649 | 1741 | struct btrfs_fs_info *fs_info, u64 start) |
4358d963 | 1742 | { |
32da5386 | 1743 | struct btrfs_block_group *cache; |
4358d963 JB |
1744 | |
1745 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
1746 | if (!cache) | |
1747 | return NULL; | |
1748 | ||
1749 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
1750 | GFP_NOFS); | |
1751 | if (!cache->free_space_ctl) { | |
1752 | kfree(cache); | |
1753 | return NULL; | |
1754 | } | |
1755 | ||
b3470b5d | 1756 | cache->start = start; |
4358d963 JB |
1757 | |
1758 | cache->fs_info = fs_info; | |
1759 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); | |
4358d963 | 1760 | |
6e80d4f8 DZ |
1761 | cache->discard_index = BTRFS_DISCARD_INDEX_UNUSED; |
1762 | ||
48aaeebe | 1763 | refcount_set(&cache->refs, 1); |
4358d963 JB |
1764 | spin_lock_init(&cache->lock); |
1765 | init_rwsem(&cache->data_rwsem); | |
1766 | INIT_LIST_HEAD(&cache->list); | |
1767 | INIT_LIST_HEAD(&cache->cluster_list); | |
1768 | INIT_LIST_HEAD(&cache->bg_list); | |
1769 | INIT_LIST_HEAD(&cache->ro_list); | |
b0643e59 | 1770 | INIT_LIST_HEAD(&cache->discard_list); |
4358d963 JB |
1771 | INIT_LIST_HEAD(&cache->dirty_list); |
1772 | INIT_LIST_HEAD(&cache->io_list); | |
cd79909b | 1773 | btrfs_init_free_space_ctl(cache, cache->free_space_ctl); |
6b7304af | 1774 | atomic_set(&cache->frozen, 0); |
4358d963 JB |
1775 | mutex_init(&cache->free_space_lock); |
1776 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); | |
1777 | ||
1778 | return cache; | |
1779 | } | |
1780 | ||
1781 | /* | |
1782 | * Iterate all chunks and verify that each of them has the corresponding block | |
1783 | * group | |
1784 | */ | |
1785 | static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) | |
1786 | { | |
1787 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; | |
1788 | struct extent_map *em; | |
32da5386 | 1789 | struct btrfs_block_group *bg; |
4358d963 JB |
1790 | u64 start = 0; |
1791 | int ret = 0; | |
1792 | ||
1793 | while (1) { | |
1794 | read_lock(&map_tree->lock); | |
1795 | /* | |
1796 | * lookup_extent_mapping will return the first extent map | |
1797 | * intersecting the range, so setting @len to 1 is enough to | |
1798 | * get the first chunk. | |
1799 | */ | |
1800 | em = lookup_extent_mapping(map_tree, start, 1); | |
1801 | read_unlock(&map_tree->lock); | |
1802 | if (!em) | |
1803 | break; | |
1804 | ||
1805 | bg = btrfs_lookup_block_group(fs_info, em->start); | |
1806 | if (!bg) { | |
1807 | btrfs_err(fs_info, | |
1808 | "chunk start=%llu len=%llu doesn't have corresponding block group", | |
1809 | em->start, em->len); | |
1810 | ret = -EUCLEAN; | |
1811 | free_extent_map(em); | |
1812 | break; | |
1813 | } | |
b3470b5d | 1814 | if (bg->start != em->start || bg->length != em->len || |
4358d963 JB |
1815 | (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != |
1816 | (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
1817 | btrfs_err(fs_info, | |
1818 | "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", | |
1819 | em->start, em->len, | |
1820 | em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, | |
b3470b5d | 1821 | bg->start, bg->length, |
4358d963 JB |
1822 | bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); |
1823 | ret = -EUCLEAN; | |
1824 | free_extent_map(em); | |
1825 | btrfs_put_block_group(bg); | |
1826 | break; | |
1827 | } | |
1828 | start = em->start + em->len; | |
1829 | free_extent_map(em); | |
1830 | btrfs_put_block_group(bg); | |
1831 | } | |
1832 | return ret; | |
1833 | } | |
1834 | ||
4c448ce8 | 1835 | static void read_block_group_item(struct btrfs_block_group *cache, |
9afc6649 QW |
1836 | struct btrfs_path *path, |
1837 | const struct btrfs_key *key) | |
1838 | { | |
1839 | struct extent_buffer *leaf = path->nodes[0]; | |
1840 | struct btrfs_block_group_item bgi; | |
1841 | int slot = path->slots[0]; | |
1842 | ||
1843 | cache->length = key->offset; | |
1844 | ||
1845 | read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot), | |
1846 | sizeof(bgi)); | |
1847 | cache->used = btrfs_stack_block_group_used(&bgi); | |
1848 | cache->flags = btrfs_stack_block_group_flags(&bgi); | |
9afc6649 QW |
1849 | } |
1850 | ||
ffb9e0f0 QW |
1851 | static int read_one_block_group(struct btrfs_fs_info *info, |
1852 | struct btrfs_path *path, | |
d49a2ddb | 1853 | const struct btrfs_key *key, |
ffb9e0f0 QW |
1854 | int need_clear) |
1855 | { | |
32da5386 | 1856 | struct btrfs_block_group *cache; |
ffb9e0f0 | 1857 | struct btrfs_space_info *space_info; |
ffb9e0f0 | 1858 | const bool mixed = btrfs_fs_incompat(info, MIXED_GROUPS); |
ffb9e0f0 QW |
1859 | int ret; |
1860 | ||
d49a2ddb | 1861 | ASSERT(key->type == BTRFS_BLOCK_GROUP_ITEM_KEY); |
ffb9e0f0 | 1862 | |
9afc6649 | 1863 | cache = btrfs_create_block_group_cache(info, key->objectid); |
ffb9e0f0 QW |
1864 | if (!cache) |
1865 | return -ENOMEM; | |
1866 | ||
4c448ce8 | 1867 | read_block_group_item(cache, path, key); |
9afc6649 | 1868 | |
e3e39c72 MPS |
1869 | set_free_space_tree_thresholds(cache); |
1870 | ||
ffb9e0f0 QW |
1871 | if (need_clear) { |
1872 | /* | |
1873 | * When we mount with old space cache, we need to | |
1874 | * set BTRFS_DC_CLEAR and set dirty flag. | |
1875 | * | |
1876 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
1877 | * truncate the old free space cache inode and | |
1878 | * setup a new one. | |
1879 | * b) Setting 'dirty flag' makes sure that we flush | |
1880 | * the new space cache info onto disk. | |
1881 | */ | |
1882 | if (btrfs_test_opt(info, SPACE_CACHE)) | |
1883 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
1884 | } | |
ffb9e0f0 QW |
1885 | if (!mixed && ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && |
1886 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
1887 | btrfs_err(info, | |
1888 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
1889 | cache->start); | |
1890 | ret = -EINVAL; | |
1891 | goto error; | |
1892 | } | |
1893 | ||
1894 | /* | |
1895 | * We need to exclude the super stripes now so that the space info has | |
1896 | * super bytes accounted for, otherwise we'll think we have more space | |
1897 | * than we actually do. | |
1898 | */ | |
1899 | ret = exclude_super_stripes(cache); | |
1900 | if (ret) { | |
1901 | /* We may have excluded something, so call this just in case. */ | |
1902 | btrfs_free_excluded_extents(cache); | |
1903 | goto error; | |
1904 | } | |
1905 | ||
1906 | /* | |
1907 | * Check for two cases, either we are full, and therefore don't need | |
1908 | * to bother with the caching work since we won't find any space, or we | |
1909 | * are empty, and we can just add all the space in and be done with it. | |
1910 | * This saves us _a_lot_ of time, particularly in the full case. | |
1911 | */ | |
9afc6649 | 1912 | if (cache->length == cache->used) { |
ffb9e0f0 QW |
1913 | cache->last_byte_to_unpin = (u64)-1; |
1914 | cache->cached = BTRFS_CACHE_FINISHED; | |
1915 | btrfs_free_excluded_extents(cache); | |
1916 | } else if (cache->used == 0) { | |
1917 | cache->last_byte_to_unpin = (u64)-1; | |
1918 | cache->cached = BTRFS_CACHE_FINISHED; | |
9afc6649 QW |
1919 | add_new_free_space(cache, cache->start, |
1920 | cache->start + cache->length); | |
ffb9e0f0 QW |
1921 | btrfs_free_excluded_extents(cache); |
1922 | } | |
1923 | ||
1924 | ret = btrfs_add_block_group_cache(info, cache); | |
1925 | if (ret) { | |
1926 | btrfs_remove_free_space_cache(cache); | |
1927 | goto error; | |
1928 | } | |
1929 | trace_btrfs_add_block_group(info, cache, 0); | |
9afc6649 | 1930 | btrfs_update_space_info(info, cache->flags, cache->length, |
ffb9e0f0 QW |
1931 | cache->used, cache->bytes_super, &space_info); |
1932 | ||
1933 | cache->space_info = space_info; | |
1934 | ||
1935 | link_block_group(cache); | |
1936 | ||
1937 | set_avail_alloc_bits(info, cache->flags); | |
1938 | if (btrfs_chunk_readonly(info, cache->start)) { | |
1939 | inc_block_group_ro(cache, 1); | |
1940 | } else if (cache->used == 0) { | |
1941 | ASSERT(list_empty(&cache->bg_list)); | |
6e80d4f8 DZ |
1942 | if (btrfs_test_opt(info, DISCARD_ASYNC)) |
1943 | btrfs_discard_queue_work(&info->discard_ctl, cache); | |
1944 | else | |
1945 | btrfs_mark_bg_unused(cache); | |
ffb9e0f0 QW |
1946 | } |
1947 | return 0; | |
1948 | error: | |
1949 | btrfs_put_block_group(cache); | |
1950 | return ret; | |
1951 | } | |
1952 | ||
42437a63 JB |
1953 | static int fill_dummy_bgs(struct btrfs_fs_info *fs_info) |
1954 | { | |
1955 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; | |
1956 | struct btrfs_space_info *space_info; | |
1957 | struct rb_node *node; | |
1958 | int ret = 0; | |
1959 | ||
1960 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { | |
1961 | struct extent_map *em; | |
1962 | struct map_lookup *map; | |
1963 | struct btrfs_block_group *bg; | |
1964 | ||
1965 | em = rb_entry(node, struct extent_map, rb_node); | |
1966 | map = em->map_lookup; | |
1967 | bg = btrfs_create_block_group_cache(fs_info, em->start); | |
1968 | if (!bg) { | |
1969 | ret = -ENOMEM; | |
1970 | break; | |
1971 | } | |
1972 | ||
1973 | /* Fill dummy cache as FULL */ | |
1974 | bg->length = em->len; | |
1975 | bg->flags = map->type; | |
1976 | bg->last_byte_to_unpin = (u64)-1; | |
1977 | bg->cached = BTRFS_CACHE_FINISHED; | |
1978 | bg->used = em->len; | |
1979 | bg->flags = map->type; | |
1980 | ret = btrfs_add_block_group_cache(fs_info, bg); | |
1981 | if (ret) { | |
1982 | btrfs_remove_free_space_cache(bg); | |
1983 | btrfs_put_block_group(bg); | |
1984 | break; | |
1985 | } | |
1986 | btrfs_update_space_info(fs_info, bg->flags, em->len, em->len, | |
1987 | 0, &space_info); | |
1988 | bg->space_info = space_info; | |
1989 | link_block_group(bg); | |
1990 | ||
1991 | set_avail_alloc_bits(fs_info, bg->flags); | |
1992 | } | |
1993 | if (!ret) | |
1994 | btrfs_init_global_block_rsv(fs_info); | |
1995 | return ret; | |
1996 | } | |
1997 | ||
4358d963 JB |
1998 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
1999 | { | |
2000 | struct btrfs_path *path; | |
2001 | int ret; | |
32da5386 | 2002 | struct btrfs_block_group *cache; |
4358d963 JB |
2003 | struct btrfs_space_info *space_info; |
2004 | struct btrfs_key key; | |
4358d963 JB |
2005 | int need_clear = 0; |
2006 | u64 cache_gen; | |
4358d963 | 2007 | |
42437a63 JB |
2008 | if (!info->extent_root) |
2009 | return fill_dummy_bgs(info); | |
2010 | ||
4358d963 JB |
2011 | key.objectid = 0; |
2012 | key.offset = 0; | |
2013 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
2014 | path = btrfs_alloc_path(); | |
2015 | if (!path) | |
2016 | return -ENOMEM; | |
4358d963 JB |
2017 | |
2018 | cache_gen = btrfs_super_cache_generation(info->super_copy); | |
2019 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
2020 | btrfs_super_generation(info->super_copy) != cache_gen) | |
2021 | need_clear = 1; | |
2022 | if (btrfs_test_opt(info, CLEAR_CACHE)) | |
2023 | need_clear = 1; | |
2024 | ||
2025 | while (1) { | |
2026 | ret = find_first_block_group(info, path, &key); | |
2027 | if (ret > 0) | |
2028 | break; | |
2029 | if (ret != 0) | |
2030 | goto error; | |
2031 | ||
ffb9e0f0 | 2032 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); |
d49a2ddb | 2033 | ret = read_one_block_group(info, path, &key, need_clear); |
ffb9e0f0 | 2034 | if (ret < 0) |
4358d963 | 2035 | goto error; |
ffb9e0f0 QW |
2036 | key.objectid += key.offset; |
2037 | key.offset = 0; | |
4358d963 | 2038 | btrfs_release_path(path); |
4358d963 | 2039 | } |
7837fa88 | 2040 | btrfs_release_path(path); |
4358d963 | 2041 | |
72804905 | 2042 | list_for_each_entry(space_info, &info->space_info, list) { |
49ea112d JB |
2043 | int i; |
2044 | ||
2045 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { | |
2046 | if (list_empty(&space_info->block_groups[i])) | |
2047 | continue; | |
2048 | cache = list_first_entry(&space_info->block_groups[i], | |
2049 | struct btrfs_block_group, | |
2050 | list); | |
2051 | btrfs_sysfs_add_block_group_type(cache); | |
2052 | } | |
2053 | ||
4358d963 JB |
2054 | if (!(btrfs_get_alloc_profile(info, space_info->flags) & |
2055 | (BTRFS_BLOCK_GROUP_RAID10 | | |
2056 | BTRFS_BLOCK_GROUP_RAID1_MASK | | |
2057 | BTRFS_BLOCK_GROUP_RAID56_MASK | | |
2058 | BTRFS_BLOCK_GROUP_DUP))) | |
2059 | continue; | |
2060 | /* | |
2061 | * Avoid allocating from un-mirrored block group if there are | |
2062 | * mirrored block groups. | |
2063 | */ | |
2064 | list_for_each_entry(cache, | |
2065 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
2066 | list) | |
e11c0406 | 2067 | inc_block_group_ro(cache, 1); |
4358d963 JB |
2068 | list_for_each_entry(cache, |
2069 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
2070 | list) | |
e11c0406 | 2071 | inc_block_group_ro(cache, 1); |
4358d963 JB |
2072 | } |
2073 | ||
2074 | btrfs_init_global_block_rsv(info); | |
2075 | ret = check_chunk_block_group_mappings(info); | |
2076 | error: | |
2077 | btrfs_free_path(path); | |
2078 | return ret; | |
2079 | } | |
2080 | ||
97f4728a QW |
2081 | static int insert_block_group_item(struct btrfs_trans_handle *trans, |
2082 | struct btrfs_block_group *block_group) | |
2083 | { | |
2084 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2085 | struct btrfs_block_group_item bgi; | |
2086 | struct btrfs_root *root; | |
2087 | struct btrfs_key key; | |
2088 | ||
2089 | spin_lock(&block_group->lock); | |
2090 | btrfs_set_stack_block_group_used(&bgi, block_group->used); | |
2091 | btrfs_set_stack_block_group_chunk_objectid(&bgi, | |
2092 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
2093 | btrfs_set_stack_block_group_flags(&bgi, block_group->flags); | |
2094 | key.objectid = block_group->start; | |
2095 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
2096 | key.offset = block_group->length; | |
2097 | spin_unlock(&block_group->lock); | |
2098 | ||
2099 | root = fs_info->extent_root; | |
2100 | return btrfs_insert_item(trans, root, &key, &bgi, sizeof(bgi)); | |
2101 | } | |
2102 | ||
4358d963 JB |
2103 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
2104 | { | |
2105 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2106 | struct btrfs_block_group *block_group; |
4358d963 JB |
2107 | int ret = 0; |
2108 | ||
2109 | if (!trans->can_flush_pending_bgs) | |
2110 | return; | |
2111 | ||
2112 | while (!list_empty(&trans->new_bgs)) { | |
49ea112d JB |
2113 | int index; |
2114 | ||
4358d963 | 2115 | block_group = list_first_entry(&trans->new_bgs, |
32da5386 | 2116 | struct btrfs_block_group, |
4358d963 JB |
2117 | bg_list); |
2118 | if (ret) | |
2119 | goto next; | |
2120 | ||
49ea112d JB |
2121 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
2122 | ||
97f4728a | 2123 | ret = insert_block_group_item(trans, block_group); |
4358d963 JB |
2124 | if (ret) |
2125 | btrfs_abort_transaction(trans, ret); | |
97f4728a QW |
2126 | ret = btrfs_finish_chunk_alloc(trans, block_group->start, |
2127 | block_group->length); | |
4358d963 JB |
2128 | if (ret) |
2129 | btrfs_abort_transaction(trans, ret); | |
2130 | add_block_group_free_space(trans, block_group); | |
49ea112d JB |
2131 | |
2132 | /* | |
2133 | * If we restriped during balance, we may have added a new raid | |
2134 | * type, so now add the sysfs entries when it is safe to do so. | |
2135 | * We don't have to worry about locking here as it's handled in | |
2136 | * btrfs_sysfs_add_block_group_type. | |
2137 | */ | |
2138 | if (block_group->space_info->block_group_kobjs[index] == NULL) | |
2139 | btrfs_sysfs_add_block_group_type(block_group); | |
2140 | ||
4358d963 JB |
2141 | /* Already aborted the transaction if it failed. */ |
2142 | next: | |
2143 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
2144 | list_del_init(&block_group->bg_list); | |
2145 | } | |
2146 | btrfs_trans_release_chunk_metadata(trans); | |
2147 | } | |
2148 | ||
2149 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used, | |
2150 | u64 type, u64 chunk_offset, u64 size) | |
2151 | { | |
2152 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2153 | struct btrfs_block_group *cache; |
4358d963 JB |
2154 | int ret; |
2155 | ||
2156 | btrfs_set_log_full_commit(trans); | |
2157 | ||
9afc6649 | 2158 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset); |
4358d963 JB |
2159 | if (!cache) |
2160 | return -ENOMEM; | |
2161 | ||
9afc6649 | 2162 | cache->length = size; |
e3e39c72 | 2163 | set_free_space_tree_thresholds(cache); |
bf38be65 | 2164 | cache->used = bytes_used; |
4358d963 JB |
2165 | cache->flags = type; |
2166 | cache->last_byte_to_unpin = (u64)-1; | |
2167 | cache->cached = BTRFS_CACHE_FINISHED; | |
2168 | cache->needs_free_space = 1; | |
2169 | ret = exclude_super_stripes(cache); | |
2170 | if (ret) { | |
2171 | /* We may have excluded something, so call this just in case */ | |
2172 | btrfs_free_excluded_extents(cache); | |
2173 | btrfs_put_block_group(cache); | |
2174 | return ret; | |
2175 | } | |
2176 | ||
2177 | add_new_free_space(cache, chunk_offset, chunk_offset + size); | |
2178 | ||
2179 | btrfs_free_excluded_extents(cache); | |
2180 | ||
2181 | #ifdef CONFIG_BTRFS_DEBUG | |
2182 | if (btrfs_should_fragment_free_space(cache)) { | |
2183 | u64 new_bytes_used = size - bytes_used; | |
2184 | ||
2185 | bytes_used += new_bytes_used >> 1; | |
e11c0406 | 2186 | fragment_free_space(cache); |
4358d963 JB |
2187 | } |
2188 | #endif | |
2189 | /* | |
2190 | * Ensure the corresponding space_info object is created and | |
2191 | * assigned to our block group. We want our bg to be added to the rbtree | |
2192 | * with its ->space_info set. | |
2193 | */ | |
2194 | cache->space_info = btrfs_find_space_info(fs_info, cache->flags); | |
2195 | ASSERT(cache->space_info); | |
2196 | ||
2197 | ret = btrfs_add_block_group_cache(fs_info, cache); | |
2198 | if (ret) { | |
2199 | btrfs_remove_free_space_cache(cache); | |
2200 | btrfs_put_block_group(cache); | |
2201 | return ret; | |
2202 | } | |
2203 | ||
2204 | /* | |
2205 | * Now that our block group has its ->space_info set and is inserted in | |
2206 | * the rbtree, update the space info's counters. | |
2207 | */ | |
2208 | trace_btrfs_add_block_group(fs_info, cache, 1); | |
2209 | btrfs_update_space_info(fs_info, cache->flags, size, bytes_used, | |
2210 | cache->bytes_super, &cache->space_info); | |
2211 | btrfs_update_global_block_rsv(fs_info); | |
2212 | ||
2213 | link_block_group(cache); | |
2214 | ||
2215 | list_add_tail(&cache->bg_list, &trans->new_bgs); | |
2216 | trans->delayed_ref_updates++; | |
2217 | btrfs_update_delayed_refs_rsv(trans); | |
2218 | ||
2219 | set_avail_alloc_bits(fs_info, type); | |
2220 | return 0; | |
2221 | } | |
26ce2095 | 2222 | |
b12de528 QW |
2223 | /* |
2224 | * Mark one block group RO, can be called several times for the same block | |
2225 | * group. | |
2226 | * | |
2227 | * @cache: the destination block group | |
2228 | * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to | |
2229 | * ensure we still have some free space after marking this | |
2230 | * block group RO. | |
2231 | */ | |
2232 | int btrfs_inc_block_group_ro(struct btrfs_block_group *cache, | |
2233 | bool do_chunk_alloc) | |
26ce2095 JB |
2234 | { |
2235 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
2236 | struct btrfs_trans_handle *trans; | |
2237 | u64 alloc_flags; | |
2238 | int ret; | |
2239 | ||
2240 | again: | |
2241 | trans = btrfs_join_transaction(fs_info->extent_root); | |
2242 | if (IS_ERR(trans)) | |
2243 | return PTR_ERR(trans); | |
2244 | ||
2245 | /* | |
2246 | * we're not allowed to set block groups readonly after the dirty | |
2247 | * block groups cache has started writing. If it already started, | |
2248 | * back off and let this transaction commit | |
2249 | */ | |
2250 | mutex_lock(&fs_info->ro_block_group_mutex); | |
2251 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { | |
2252 | u64 transid = trans->transid; | |
2253 | ||
2254 | mutex_unlock(&fs_info->ro_block_group_mutex); | |
2255 | btrfs_end_transaction(trans); | |
2256 | ||
2257 | ret = btrfs_wait_for_commit(fs_info, transid); | |
2258 | if (ret) | |
2259 | return ret; | |
2260 | goto again; | |
2261 | } | |
2262 | ||
b12de528 | 2263 | if (do_chunk_alloc) { |
26ce2095 | 2264 | /* |
b12de528 QW |
2265 | * If we are changing raid levels, try to allocate a |
2266 | * corresponding block group with the new raid level. | |
26ce2095 | 2267 | */ |
349e120e | 2268 | alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); |
b12de528 QW |
2269 | if (alloc_flags != cache->flags) { |
2270 | ret = btrfs_chunk_alloc(trans, alloc_flags, | |
2271 | CHUNK_ALLOC_FORCE); | |
2272 | /* | |
2273 | * ENOSPC is allowed here, we may have enough space | |
2274 | * already allocated at the new raid level to carry on | |
2275 | */ | |
2276 | if (ret == -ENOSPC) | |
2277 | ret = 0; | |
2278 | if (ret < 0) | |
2279 | goto out; | |
2280 | } | |
26ce2095 JB |
2281 | } |
2282 | ||
a7a63acc | 2283 | ret = inc_block_group_ro(cache, 0); |
b12de528 QW |
2284 | if (!do_chunk_alloc) |
2285 | goto unlock_out; | |
26ce2095 JB |
2286 | if (!ret) |
2287 | goto out; | |
2288 | alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags); | |
2289 | ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); | |
2290 | if (ret < 0) | |
2291 | goto out; | |
e11c0406 | 2292 | ret = inc_block_group_ro(cache, 0); |
26ce2095 JB |
2293 | out: |
2294 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { | |
349e120e | 2295 | alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); |
26ce2095 JB |
2296 | mutex_lock(&fs_info->chunk_mutex); |
2297 | check_system_chunk(trans, alloc_flags); | |
2298 | mutex_unlock(&fs_info->chunk_mutex); | |
2299 | } | |
b12de528 | 2300 | unlock_out: |
26ce2095 JB |
2301 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2302 | ||
2303 | btrfs_end_transaction(trans); | |
2304 | return ret; | |
2305 | } | |
2306 | ||
32da5386 | 2307 | void btrfs_dec_block_group_ro(struct btrfs_block_group *cache) |
26ce2095 JB |
2308 | { |
2309 | struct btrfs_space_info *sinfo = cache->space_info; | |
2310 | u64 num_bytes; | |
2311 | ||
2312 | BUG_ON(!cache->ro); | |
2313 | ||
2314 | spin_lock(&sinfo->lock); | |
2315 | spin_lock(&cache->lock); | |
2316 | if (!--cache->ro) { | |
b3470b5d | 2317 | num_bytes = cache->length - cache->reserved - |
bf38be65 | 2318 | cache->pinned - cache->bytes_super - cache->used; |
26ce2095 JB |
2319 | sinfo->bytes_readonly -= num_bytes; |
2320 | list_del_init(&cache->ro_list); | |
2321 | } | |
2322 | spin_unlock(&cache->lock); | |
2323 | spin_unlock(&sinfo->lock); | |
2324 | } | |
77745c05 | 2325 | |
3be4d8ef QW |
2326 | static int update_block_group_item(struct btrfs_trans_handle *trans, |
2327 | struct btrfs_path *path, | |
2328 | struct btrfs_block_group *cache) | |
77745c05 JB |
2329 | { |
2330 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2331 | int ret; | |
3be4d8ef | 2332 | struct btrfs_root *root = fs_info->extent_root; |
77745c05 JB |
2333 | unsigned long bi; |
2334 | struct extent_buffer *leaf; | |
bf38be65 | 2335 | struct btrfs_block_group_item bgi; |
b3470b5d DS |
2336 | struct btrfs_key key; |
2337 | ||
2338 | key.objectid = cache->start; | |
2339 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
2340 | key.offset = cache->length; | |
77745c05 | 2341 | |
3be4d8ef | 2342 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
77745c05 JB |
2343 | if (ret) { |
2344 | if (ret > 0) | |
2345 | ret = -ENOENT; | |
2346 | goto fail; | |
2347 | } | |
2348 | ||
2349 | leaf = path->nodes[0]; | |
2350 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
de0dc456 DS |
2351 | btrfs_set_stack_block_group_used(&bgi, cache->used); |
2352 | btrfs_set_stack_block_group_chunk_objectid(&bgi, | |
3d976388 | 2353 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); |
de0dc456 | 2354 | btrfs_set_stack_block_group_flags(&bgi, cache->flags); |
bf38be65 | 2355 | write_extent_buffer(leaf, &bgi, bi, sizeof(bgi)); |
77745c05 JB |
2356 | btrfs_mark_buffer_dirty(leaf); |
2357 | fail: | |
2358 | btrfs_release_path(path); | |
2359 | return ret; | |
2360 | ||
2361 | } | |
2362 | ||
32da5386 | 2363 | static int cache_save_setup(struct btrfs_block_group *block_group, |
77745c05 JB |
2364 | struct btrfs_trans_handle *trans, |
2365 | struct btrfs_path *path) | |
2366 | { | |
2367 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
2368 | struct btrfs_root *root = fs_info->tree_root; | |
2369 | struct inode *inode = NULL; | |
2370 | struct extent_changeset *data_reserved = NULL; | |
2371 | u64 alloc_hint = 0; | |
2372 | int dcs = BTRFS_DC_ERROR; | |
2373 | u64 num_pages = 0; | |
2374 | int retries = 0; | |
2375 | int ret = 0; | |
2376 | ||
2377 | /* | |
2378 | * If this block group is smaller than 100 megs don't bother caching the | |
2379 | * block group. | |
2380 | */ | |
b3470b5d | 2381 | if (block_group->length < (100 * SZ_1M)) { |
77745c05 JB |
2382 | spin_lock(&block_group->lock); |
2383 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2384 | spin_unlock(&block_group->lock); | |
2385 | return 0; | |
2386 | } | |
2387 | ||
bf31f87f | 2388 | if (TRANS_ABORTED(trans)) |
77745c05 JB |
2389 | return 0; |
2390 | again: | |
2391 | inode = lookup_free_space_inode(block_group, path); | |
2392 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2393 | ret = PTR_ERR(inode); | |
2394 | btrfs_release_path(path); | |
2395 | goto out; | |
2396 | } | |
2397 | ||
2398 | if (IS_ERR(inode)) { | |
2399 | BUG_ON(retries); | |
2400 | retries++; | |
2401 | ||
2402 | if (block_group->ro) | |
2403 | goto out_free; | |
2404 | ||
2405 | ret = create_free_space_inode(trans, block_group, path); | |
2406 | if (ret) | |
2407 | goto out_free; | |
2408 | goto again; | |
2409 | } | |
2410 | ||
2411 | /* | |
2412 | * We want to set the generation to 0, that way if anything goes wrong | |
2413 | * from here on out we know not to trust this cache when we load up next | |
2414 | * time. | |
2415 | */ | |
2416 | BTRFS_I(inode)->generation = 0; | |
9a56fcd1 | 2417 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
77745c05 JB |
2418 | if (ret) { |
2419 | /* | |
2420 | * So theoretically we could recover from this, simply set the | |
2421 | * super cache generation to 0 so we know to invalidate the | |
2422 | * cache, but then we'd have to keep track of the block groups | |
2423 | * that fail this way so we know we _have_ to reset this cache | |
2424 | * before the next commit or risk reading stale cache. So to | |
2425 | * limit our exposure to horrible edge cases lets just abort the | |
2426 | * transaction, this only happens in really bad situations | |
2427 | * anyway. | |
2428 | */ | |
2429 | btrfs_abort_transaction(trans, ret); | |
2430 | goto out_put; | |
2431 | } | |
2432 | WARN_ON(ret); | |
2433 | ||
2434 | /* We've already setup this transaction, go ahead and exit */ | |
2435 | if (block_group->cache_generation == trans->transid && | |
2436 | i_size_read(inode)) { | |
2437 | dcs = BTRFS_DC_SETUP; | |
2438 | goto out_put; | |
2439 | } | |
2440 | ||
2441 | if (i_size_read(inode) > 0) { | |
2442 | ret = btrfs_check_trunc_cache_free_space(fs_info, | |
2443 | &fs_info->global_block_rsv); | |
2444 | if (ret) | |
2445 | goto out_put; | |
2446 | ||
2447 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); | |
2448 | if (ret) | |
2449 | goto out_put; | |
2450 | } | |
2451 | ||
2452 | spin_lock(&block_group->lock); | |
2453 | if (block_group->cached != BTRFS_CACHE_FINISHED || | |
2454 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { | |
2455 | /* | |
2456 | * don't bother trying to write stuff out _if_ | |
2457 | * a) we're not cached, | |
2458 | * b) we're with nospace_cache mount option, | |
2459 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
2460 | */ | |
2461 | dcs = BTRFS_DC_WRITTEN; | |
2462 | spin_unlock(&block_group->lock); | |
2463 | goto out_put; | |
2464 | } | |
2465 | spin_unlock(&block_group->lock); | |
2466 | ||
2467 | /* | |
2468 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
2469 | * skip doing the setup, we've already cleared the cache so we're safe. | |
2470 | */ | |
2471 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
2472 | ret = -ENOSPC; | |
2473 | goto out_put; | |
2474 | } | |
2475 | ||
2476 | /* | |
2477 | * Try to preallocate enough space based on how big the block group is. | |
2478 | * Keep in mind this has to include any pinned space which could end up | |
2479 | * taking up quite a bit since it's not folded into the other space | |
2480 | * cache. | |
2481 | */ | |
b3470b5d | 2482 | num_pages = div_u64(block_group->length, SZ_256M); |
77745c05 JB |
2483 | if (!num_pages) |
2484 | num_pages = 1; | |
2485 | ||
2486 | num_pages *= 16; | |
2487 | num_pages *= PAGE_SIZE; | |
2488 | ||
36ea6f3e NB |
2489 | ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved, 0, |
2490 | num_pages); | |
77745c05 JB |
2491 | if (ret) |
2492 | goto out_put; | |
2493 | ||
2494 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
2495 | num_pages, num_pages, | |
2496 | &alloc_hint); | |
2497 | /* | |
2498 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
2499 | * of metadata or split extents when writing the cache out, which means | |
2500 | * we can enospc if we are heavily fragmented in addition to just normal | |
2501 | * out of space conditions. So if we hit this just skip setting up any | |
2502 | * other block groups for this transaction, maybe we'll unpin enough | |
2503 | * space the next time around. | |
2504 | */ | |
2505 | if (!ret) | |
2506 | dcs = BTRFS_DC_SETUP; | |
2507 | else if (ret == -ENOSPC) | |
2508 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
2509 | ||
2510 | out_put: | |
2511 | iput(inode); | |
2512 | out_free: | |
2513 | btrfs_release_path(path); | |
2514 | out: | |
2515 | spin_lock(&block_group->lock); | |
2516 | if (!ret && dcs == BTRFS_DC_SETUP) | |
2517 | block_group->cache_generation = trans->transid; | |
2518 | block_group->disk_cache_state = dcs; | |
2519 | spin_unlock(&block_group->lock); | |
2520 | ||
2521 | extent_changeset_free(data_reserved); | |
2522 | return ret; | |
2523 | } | |
2524 | ||
2525 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) | |
2526 | { | |
2527 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2528 | struct btrfs_block_group *cache, *tmp; |
77745c05 JB |
2529 | struct btrfs_transaction *cur_trans = trans->transaction; |
2530 | struct btrfs_path *path; | |
2531 | ||
2532 | if (list_empty(&cur_trans->dirty_bgs) || | |
2533 | !btrfs_test_opt(fs_info, SPACE_CACHE)) | |
2534 | return 0; | |
2535 | ||
2536 | path = btrfs_alloc_path(); | |
2537 | if (!path) | |
2538 | return -ENOMEM; | |
2539 | ||
2540 | /* Could add new block groups, use _safe just in case */ | |
2541 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
2542 | dirty_list) { | |
2543 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
2544 | cache_save_setup(cache, trans, path); | |
2545 | } | |
2546 | ||
2547 | btrfs_free_path(path); | |
2548 | return 0; | |
2549 | } | |
2550 | ||
2551 | /* | |
2552 | * Transaction commit does final block group cache writeback during a critical | |
2553 | * section where nothing is allowed to change the FS. This is required in | |
2554 | * order for the cache to actually match the block group, but can introduce a | |
2555 | * lot of latency into the commit. | |
2556 | * | |
2557 | * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO. | |
2558 | * There's a chance we'll have to redo some of it if the block group changes | |
2559 | * again during the commit, but it greatly reduces the commit latency by | |
2560 | * getting rid of the easy block groups while we're still allowing others to | |
2561 | * join the commit. | |
2562 | */ | |
2563 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) | |
2564 | { | |
2565 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2566 | struct btrfs_block_group *cache; |
77745c05 JB |
2567 | struct btrfs_transaction *cur_trans = trans->transaction; |
2568 | int ret = 0; | |
2569 | int should_put; | |
2570 | struct btrfs_path *path = NULL; | |
2571 | LIST_HEAD(dirty); | |
2572 | struct list_head *io = &cur_trans->io_bgs; | |
2573 | int num_started = 0; | |
2574 | int loops = 0; | |
2575 | ||
2576 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2577 | if (list_empty(&cur_trans->dirty_bgs)) { | |
2578 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2579 | return 0; | |
2580 | } | |
2581 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
2582 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2583 | ||
2584 | again: | |
2585 | /* Make sure all the block groups on our dirty list actually exist */ | |
2586 | btrfs_create_pending_block_groups(trans); | |
2587 | ||
2588 | if (!path) { | |
2589 | path = btrfs_alloc_path(); | |
2590 | if (!path) | |
2591 | return -ENOMEM; | |
2592 | } | |
2593 | ||
2594 | /* | |
2595 | * cache_write_mutex is here only to save us from balance or automatic | |
2596 | * removal of empty block groups deleting this block group while we are | |
2597 | * writing out the cache | |
2598 | */ | |
2599 | mutex_lock(&trans->transaction->cache_write_mutex); | |
2600 | while (!list_empty(&dirty)) { | |
2601 | bool drop_reserve = true; | |
2602 | ||
32da5386 | 2603 | cache = list_first_entry(&dirty, struct btrfs_block_group, |
77745c05 JB |
2604 | dirty_list); |
2605 | /* | |
2606 | * This can happen if something re-dirties a block group that | |
2607 | * is already under IO. Just wait for it to finish and then do | |
2608 | * it all again | |
2609 | */ | |
2610 | if (!list_empty(&cache->io_list)) { | |
2611 | list_del_init(&cache->io_list); | |
2612 | btrfs_wait_cache_io(trans, cache, path); | |
2613 | btrfs_put_block_group(cache); | |
2614 | } | |
2615 | ||
2616 | ||
2617 | /* | |
2618 | * btrfs_wait_cache_io uses the cache->dirty_list to decide if | |
2619 | * it should update the cache_state. Don't delete until after | |
2620 | * we wait. | |
2621 | * | |
2622 | * Since we're not running in the commit critical section | |
2623 | * we need the dirty_bgs_lock to protect from update_block_group | |
2624 | */ | |
2625 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2626 | list_del_init(&cache->dirty_list); | |
2627 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2628 | ||
2629 | should_put = 1; | |
2630 | ||
2631 | cache_save_setup(cache, trans, path); | |
2632 | ||
2633 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
2634 | cache->io_ctl.inode = NULL; | |
2635 | ret = btrfs_write_out_cache(trans, cache, path); | |
2636 | if (ret == 0 && cache->io_ctl.inode) { | |
2637 | num_started++; | |
2638 | should_put = 0; | |
2639 | ||
2640 | /* | |
2641 | * The cache_write_mutex is protecting the | |
2642 | * io_list, also refer to the definition of | |
2643 | * btrfs_transaction::io_bgs for more details | |
2644 | */ | |
2645 | list_add_tail(&cache->io_list, io); | |
2646 | } else { | |
2647 | /* | |
2648 | * If we failed to write the cache, the | |
2649 | * generation will be bad and life goes on | |
2650 | */ | |
2651 | ret = 0; | |
2652 | } | |
2653 | } | |
2654 | if (!ret) { | |
3be4d8ef | 2655 | ret = update_block_group_item(trans, path, cache); |
77745c05 JB |
2656 | /* |
2657 | * Our block group might still be attached to the list | |
2658 | * of new block groups in the transaction handle of some | |
2659 | * other task (struct btrfs_trans_handle->new_bgs). This | |
2660 | * means its block group item isn't yet in the extent | |
2661 | * tree. If this happens ignore the error, as we will | |
2662 | * try again later in the critical section of the | |
2663 | * transaction commit. | |
2664 | */ | |
2665 | if (ret == -ENOENT) { | |
2666 | ret = 0; | |
2667 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2668 | if (list_empty(&cache->dirty_list)) { | |
2669 | list_add_tail(&cache->dirty_list, | |
2670 | &cur_trans->dirty_bgs); | |
2671 | btrfs_get_block_group(cache); | |
2672 | drop_reserve = false; | |
2673 | } | |
2674 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2675 | } else if (ret) { | |
2676 | btrfs_abort_transaction(trans, ret); | |
2677 | } | |
2678 | } | |
2679 | ||
2680 | /* If it's not on the io list, we need to put the block group */ | |
2681 | if (should_put) | |
2682 | btrfs_put_block_group(cache); | |
2683 | if (drop_reserve) | |
2684 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
2685 | ||
2686 | if (ret) | |
2687 | break; | |
2688 | ||
2689 | /* | |
2690 | * Avoid blocking other tasks for too long. It might even save | |
2691 | * us from writing caches for block groups that are going to be | |
2692 | * removed. | |
2693 | */ | |
2694 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
2695 | mutex_lock(&trans->transaction->cache_write_mutex); | |
2696 | } | |
2697 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
2698 | ||
2699 | /* | |
2700 | * Go through delayed refs for all the stuff we've just kicked off | |
2701 | * and then loop back (just once) | |
2702 | */ | |
2703 | ret = btrfs_run_delayed_refs(trans, 0); | |
2704 | if (!ret && loops == 0) { | |
2705 | loops++; | |
2706 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2707 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
2708 | /* | |
2709 | * dirty_bgs_lock protects us from concurrent block group | |
2710 | * deletes too (not just cache_write_mutex). | |
2711 | */ | |
2712 | if (!list_empty(&dirty)) { | |
2713 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2714 | goto again; | |
2715 | } | |
2716 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2717 | } else if (ret < 0) { | |
2718 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); | |
2719 | } | |
2720 | ||
2721 | btrfs_free_path(path); | |
2722 | return ret; | |
2723 | } | |
2724 | ||
2725 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) | |
2726 | { | |
2727 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2728 | struct btrfs_block_group *cache; |
77745c05 JB |
2729 | struct btrfs_transaction *cur_trans = trans->transaction; |
2730 | int ret = 0; | |
2731 | int should_put; | |
2732 | struct btrfs_path *path; | |
2733 | struct list_head *io = &cur_trans->io_bgs; | |
2734 | int num_started = 0; | |
2735 | ||
2736 | path = btrfs_alloc_path(); | |
2737 | if (!path) | |
2738 | return -ENOMEM; | |
2739 | ||
2740 | /* | |
2741 | * Even though we are in the critical section of the transaction commit, | |
2742 | * we can still have concurrent tasks adding elements to this | |
2743 | * transaction's list of dirty block groups. These tasks correspond to | |
2744 | * endio free space workers started when writeback finishes for a | |
2745 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
2746 | * allocate new block groups as a result of COWing nodes of the root | |
2747 | * tree when updating the free space inode. The writeback for the space | |
2748 | * caches is triggered by an earlier call to | |
2749 | * btrfs_start_dirty_block_groups() and iterations of the following | |
2750 | * loop. | |
2751 | * Also we want to do the cache_save_setup first and then run the | |
2752 | * delayed refs to make sure we have the best chance at doing this all | |
2753 | * in one shot. | |
2754 | */ | |
2755 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2756 | while (!list_empty(&cur_trans->dirty_bgs)) { | |
2757 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
32da5386 | 2758 | struct btrfs_block_group, |
77745c05 JB |
2759 | dirty_list); |
2760 | ||
2761 | /* | |
2762 | * This can happen if cache_save_setup re-dirties a block group | |
2763 | * that is already under IO. Just wait for it to finish and | |
2764 | * then do it all again | |
2765 | */ | |
2766 | if (!list_empty(&cache->io_list)) { | |
2767 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2768 | list_del_init(&cache->io_list); | |
2769 | btrfs_wait_cache_io(trans, cache, path); | |
2770 | btrfs_put_block_group(cache); | |
2771 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2772 | } | |
2773 | ||
2774 | /* | |
2775 | * Don't remove from the dirty list until after we've waited on | |
2776 | * any pending IO | |
2777 | */ | |
2778 | list_del_init(&cache->dirty_list); | |
2779 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2780 | should_put = 1; | |
2781 | ||
2782 | cache_save_setup(cache, trans, path); | |
2783 | ||
2784 | if (!ret) | |
2785 | ret = btrfs_run_delayed_refs(trans, | |
2786 | (unsigned long) -1); | |
2787 | ||
2788 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
2789 | cache->io_ctl.inode = NULL; | |
2790 | ret = btrfs_write_out_cache(trans, cache, path); | |
2791 | if (ret == 0 && cache->io_ctl.inode) { | |
2792 | num_started++; | |
2793 | should_put = 0; | |
2794 | list_add_tail(&cache->io_list, io); | |
2795 | } else { | |
2796 | /* | |
2797 | * If we failed to write the cache, the | |
2798 | * generation will be bad and life goes on | |
2799 | */ | |
2800 | ret = 0; | |
2801 | } | |
2802 | } | |
2803 | if (!ret) { | |
3be4d8ef | 2804 | ret = update_block_group_item(trans, path, cache); |
77745c05 JB |
2805 | /* |
2806 | * One of the free space endio workers might have | |
2807 | * created a new block group while updating a free space | |
2808 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
2809 | * and hasn't released its transaction handle yet, in | |
2810 | * which case the new block group is still attached to | |
2811 | * its transaction handle and its creation has not | |
2812 | * finished yet (no block group item in the extent tree | |
2813 | * yet, etc). If this is the case, wait for all free | |
2814 | * space endio workers to finish and retry. This is a | |
260db43c | 2815 | * very rare case so no need for a more efficient and |
77745c05 JB |
2816 | * complex approach. |
2817 | */ | |
2818 | if (ret == -ENOENT) { | |
2819 | wait_event(cur_trans->writer_wait, | |
2820 | atomic_read(&cur_trans->num_writers) == 1); | |
3be4d8ef | 2821 | ret = update_block_group_item(trans, path, cache); |
77745c05 JB |
2822 | } |
2823 | if (ret) | |
2824 | btrfs_abort_transaction(trans, ret); | |
2825 | } | |
2826 | ||
2827 | /* If its not on the io list, we need to put the block group */ | |
2828 | if (should_put) | |
2829 | btrfs_put_block_group(cache); | |
2830 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
2831 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2832 | } | |
2833 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2834 | ||
2835 | /* | |
2836 | * Refer to the definition of io_bgs member for details why it's safe | |
2837 | * to use it without any locking | |
2838 | */ | |
2839 | while (!list_empty(io)) { | |
32da5386 | 2840 | cache = list_first_entry(io, struct btrfs_block_group, |
77745c05 JB |
2841 | io_list); |
2842 | list_del_init(&cache->io_list); | |
2843 | btrfs_wait_cache_io(trans, cache, path); | |
2844 | btrfs_put_block_group(cache); | |
2845 | } | |
2846 | ||
2847 | btrfs_free_path(path); | |
2848 | return ret; | |
2849 | } | |
606d1bf1 JB |
2850 | |
2851 | int btrfs_update_block_group(struct btrfs_trans_handle *trans, | |
2852 | u64 bytenr, u64 num_bytes, int alloc) | |
2853 | { | |
2854 | struct btrfs_fs_info *info = trans->fs_info; | |
32da5386 | 2855 | struct btrfs_block_group *cache = NULL; |
606d1bf1 JB |
2856 | u64 total = num_bytes; |
2857 | u64 old_val; | |
2858 | u64 byte_in_group; | |
2859 | int factor; | |
2860 | int ret = 0; | |
2861 | ||
2862 | /* Block accounting for super block */ | |
2863 | spin_lock(&info->delalloc_root_lock); | |
2864 | old_val = btrfs_super_bytes_used(info->super_copy); | |
2865 | if (alloc) | |
2866 | old_val += num_bytes; | |
2867 | else | |
2868 | old_val -= num_bytes; | |
2869 | btrfs_set_super_bytes_used(info->super_copy, old_val); | |
2870 | spin_unlock(&info->delalloc_root_lock); | |
2871 | ||
2872 | while (total) { | |
2873 | cache = btrfs_lookup_block_group(info, bytenr); | |
2874 | if (!cache) { | |
2875 | ret = -ENOENT; | |
2876 | break; | |
2877 | } | |
2878 | factor = btrfs_bg_type_to_factor(cache->flags); | |
2879 | ||
2880 | /* | |
2881 | * If this block group has free space cache written out, we | |
2882 | * need to make sure to load it if we are removing space. This | |
2883 | * is because we need the unpinning stage to actually add the | |
2884 | * space back to the block group, otherwise we will leak space. | |
2885 | */ | |
32da5386 | 2886 | if (!alloc && !btrfs_block_group_done(cache)) |
606d1bf1 JB |
2887 | btrfs_cache_block_group(cache, 1); |
2888 | ||
b3470b5d DS |
2889 | byte_in_group = bytenr - cache->start; |
2890 | WARN_ON(byte_in_group > cache->length); | |
606d1bf1 JB |
2891 | |
2892 | spin_lock(&cache->space_info->lock); | |
2893 | spin_lock(&cache->lock); | |
2894 | ||
2895 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
2896 | cache->disk_cache_state < BTRFS_DC_CLEAR) | |
2897 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
2898 | ||
bf38be65 | 2899 | old_val = cache->used; |
b3470b5d | 2900 | num_bytes = min(total, cache->length - byte_in_group); |
606d1bf1 JB |
2901 | if (alloc) { |
2902 | old_val += num_bytes; | |
bf38be65 | 2903 | cache->used = old_val; |
606d1bf1 JB |
2904 | cache->reserved -= num_bytes; |
2905 | cache->space_info->bytes_reserved -= num_bytes; | |
2906 | cache->space_info->bytes_used += num_bytes; | |
2907 | cache->space_info->disk_used += num_bytes * factor; | |
2908 | spin_unlock(&cache->lock); | |
2909 | spin_unlock(&cache->space_info->lock); | |
2910 | } else { | |
2911 | old_val -= num_bytes; | |
bf38be65 | 2912 | cache->used = old_val; |
606d1bf1 JB |
2913 | cache->pinned += num_bytes; |
2914 | btrfs_space_info_update_bytes_pinned(info, | |
2915 | cache->space_info, num_bytes); | |
2916 | cache->space_info->bytes_used -= num_bytes; | |
2917 | cache->space_info->disk_used -= num_bytes * factor; | |
2918 | spin_unlock(&cache->lock); | |
2919 | spin_unlock(&cache->space_info->lock); | |
2920 | ||
606d1bf1 JB |
2921 | percpu_counter_add_batch( |
2922 | &cache->space_info->total_bytes_pinned, | |
2923 | num_bytes, | |
2924 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
fe119a6e | 2925 | set_extent_dirty(&trans->transaction->pinned_extents, |
606d1bf1 JB |
2926 | bytenr, bytenr + num_bytes - 1, |
2927 | GFP_NOFS | __GFP_NOFAIL); | |
2928 | } | |
2929 | ||
2930 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
2931 | if (list_empty(&cache->dirty_list)) { | |
2932 | list_add_tail(&cache->dirty_list, | |
2933 | &trans->transaction->dirty_bgs); | |
2934 | trans->delayed_ref_updates++; | |
2935 | btrfs_get_block_group(cache); | |
2936 | } | |
2937 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
2938 | ||
2939 | /* | |
2940 | * No longer have used bytes in this block group, queue it for | |
2941 | * deletion. We do this after adding the block group to the | |
2942 | * dirty list to avoid races between cleaner kthread and space | |
2943 | * cache writeout. | |
2944 | */ | |
6e80d4f8 DZ |
2945 | if (!alloc && old_val == 0) { |
2946 | if (!btrfs_test_opt(info, DISCARD_ASYNC)) | |
2947 | btrfs_mark_bg_unused(cache); | |
2948 | } | |
606d1bf1 JB |
2949 | |
2950 | btrfs_put_block_group(cache); | |
2951 | total -= num_bytes; | |
2952 | bytenr += num_bytes; | |
2953 | } | |
2954 | ||
2955 | /* Modified block groups are accounted for in the delayed_refs_rsv. */ | |
2956 | btrfs_update_delayed_refs_rsv(trans); | |
2957 | return ret; | |
2958 | } | |
2959 | ||
2960 | /** | |
2961 | * btrfs_add_reserved_bytes - update the block_group and space info counters | |
2962 | * @cache: The cache we are manipulating | |
2963 | * @ram_bytes: The number of bytes of file content, and will be same to | |
2964 | * @num_bytes except for the compress path. | |
2965 | * @num_bytes: The number of bytes in question | |
2966 | * @delalloc: The blocks are allocated for the delalloc write | |
2967 | * | |
2968 | * This is called by the allocator when it reserves space. If this is a | |
2969 | * reservation and the block group has become read only we cannot make the | |
2970 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
2971 | */ | |
32da5386 | 2972 | int btrfs_add_reserved_bytes(struct btrfs_block_group *cache, |
606d1bf1 JB |
2973 | u64 ram_bytes, u64 num_bytes, int delalloc) |
2974 | { | |
2975 | struct btrfs_space_info *space_info = cache->space_info; | |
2976 | int ret = 0; | |
2977 | ||
2978 | spin_lock(&space_info->lock); | |
2979 | spin_lock(&cache->lock); | |
2980 | if (cache->ro) { | |
2981 | ret = -EAGAIN; | |
2982 | } else { | |
2983 | cache->reserved += num_bytes; | |
2984 | space_info->bytes_reserved += num_bytes; | |
a43c3835 JB |
2985 | trace_btrfs_space_reservation(cache->fs_info, "space_info", |
2986 | space_info->flags, num_bytes, 1); | |
606d1bf1 JB |
2987 | btrfs_space_info_update_bytes_may_use(cache->fs_info, |
2988 | space_info, -ram_bytes); | |
2989 | if (delalloc) | |
2990 | cache->delalloc_bytes += num_bytes; | |
99ffb43e JB |
2991 | |
2992 | /* | |
2993 | * Compression can use less space than we reserved, so wake | |
2994 | * tickets if that happens | |
2995 | */ | |
2996 | if (num_bytes < ram_bytes) | |
2997 | btrfs_try_granting_tickets(cache->fs_info, space_info); | |
606d1bf1 JB |
2998 | } |
2999 | spin_unlock(&cache->lock); | |
3000 | spin_unlock(&space_info->lock); | |
3001 | return ret; | |
3002 | } | |
3003 | ||
3004 | /** | |
3005 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
3006 | * @cache: The cache we are manipulating | |
3007 | * @num_bytes: The number of bytes in question | |
3008 | * @delalloc: The blocks are allocated for the delalloc write | |
3009 | * | |
3010 | * This is called by somebody who is freeing space that was never actually used | |
3011 | * on disk. For example if you reserve some space for a new leaf in transaction | |
3012 | * A and before transaction A commits you free that leaf, you call this with | |
3013 | * reserve set to 0 in order to clear the reservation. | |
3014 | */ | |
32da5386 | 3015 | void btrfs_free_reserved_bytes(struct btrfs_block_group *cache, |
606d1bf1 JB |
3016 | u64 num_bytes, int delalloc) |
3017 | { | |
3018 | struct btrfs_space_info *space_info = cache->space_info; | |
3019 | ||
3020 | spin_lock(&space_info->lock); | |
3021 | spin_lock(&cache->lock); | |
3022 | if (cache->ro) | |
3023 | space_info->bytes_readonly += num_bytes; | |
3024 | cache->reserved -= num_bytes; | |
3025 | space_info->bytes_reserved -= num_bytes; | |
3026 | space_info->max_extent_size = 0; | |
3027 | ||
3028 | if (delalloc) | |
3029 | cache->delalloc_bytes -= num_bytes; | |
3030 | spin_unlock(&cache->lock); | |
3308234a JB |
3031 | |
3032 | btrfs_try_granting_tickets(cache->fs_info, space_info); | |
606d1bf1 JB |
3033 | spin_unlock(&space_info->lock); |
3034 | } | |
07730d87 JB |
3035 | |
3036 | static void force_metadata_allocation(struct btrfs_fs_info *info) | |
3037 | { | |
3038 | struct list_head *head = &info->space_info; | |
3039 | struct btrfs_space_info *found; | |
3040 | ||
72804905 | 3041 | list_for_each_entry(found, head, list) { |
07730d87 JB |
3042 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) |
3043 | found->force_alloc = CHUNK_ALLOC_FORCE; | |
3044 | } | |
07730d87 JB |
3045 | } |
3046 | ||
3047 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, | |
3048 | struct btrfs_space_info *sinfo, int force) | |
3049 | { | |
3050 | u64 bytes_used = btrfs_space_info_used(sinfo, false); | |
3051 | u64 thresh; | |
3052 | ||
3053 | if (force == CHUNK_ALLOC_FORCE) | |
3054 | return 1; | |
3055 | ||
3056 | /* | |
3057 | * in limited mode, we want to have some free space up to | |
3058 | * about 1% of the FS size. | |
3059 | */ | |
3060 | if (force == CHUNK_ALLOC_LIMITED) { | |
3061 | thresh = btrfs_super_total_bytes(fs_info->super_copy); | |
3062 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); | |
3063 | ||
3064 | if (sinfo->total_bytes - bytes_used < thresh) | |
3065 | return 1; | |
3066 | } | |
3067 | ||
3068 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) | |
3069 | return 0; | |
3070 | return 1; | |
3071 | } | |
3072 | ||
3073 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) | |
3074 | { | |
3075 | u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type); | |
3076 | ||
3077 | return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); | |
3078 | } | |
3079 | ||
3080 | /* | |
3081 | * If force is CHUNK_ALLOC_FORCE: | |
3082 | * - return 1 if it successfully allocates a chunk, | |
3083 | * - return errors including -ENOSPC otherwise. | |
3084 | * If force is NOT CHUNK_ALLOC_FORCE: | |
3085 | * - return 0 if it doesn't need to allocate a new chunk, | |
3086 | * - return 1 if it successfully allocates a chunk, | |
3087 | * - return errors including -ENOSPC otherwise. | |
3088 | */ | |
3089 | int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, | |
3090 | enum btrfs_chunk_alloc_enum force) | |
3091 | { | |
3092 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3093 | struct btrfs_space_info *space_info; | |
3094 | bool wait_for_alloc = false; | |
3095 | bool should_alloc = false; | |
3096 | int ret = 0; | |
3097 | ||
3098 | /* Don't re-enter if we're already allocating a chunk */ | |
3099 | if (trans->allocating_chunk) | |
3100 | return -ENOSPC; | |
3101 | ||
3102 | space_info = btrfs_find_space_info(fs_info, flags); | |
3103 | ASSERT(space_info); | |
3104 | ||
3105 | do { | |
3106 | spin_lock(&space_info->lock); | |
3107 | if (force < space_info->force_alloc) | |
3108 | force = space_info->force_alloc; | |
3109 | should_alloc = should_alloc_chunk(fs_info, space_info, force); | |
3110 | if (space_info->full) { | |
3111 | /* No more free physical space */ | |
3112 | if (should_alloc) | |
3113 | ret = -ENOSPC; | |
3114 | else | |
3115 | ret = 0; | |
3116 | spin_unlock(&space_info->lock); | |
3117 | return ret; | |
3118 | } else if (!should_alloc) { | |
3119 | spin_unlock(&space_info->lock); | |
3120 | return 0; | |
3121 | } else if (space_info->chunk_alloc) { | |
3122 | /* | |
3123 | * Someone is already allocating, so we need to block | |
3124 | * until this someone is finished and then loop to | |
3125 | * recheck if we should continue with our allocation | |
3126 | * attempt. | |
3127 | */ | |
3128 | wait_for_alloc = true; | |
3129 | spin_unlock(&space_info->lock); | |
3130 | mutex_lock(&fs_info->chunk_mutex); | |
3131 | mutex_unlock(&fs_info->chunk_mutex); | |
3132 | } else { | |
3133 | /* Proceed with allocation */ | |
3134 | space_info->chunk_alloc = 1; | |
3135 | wait_for_alloc = false; | |
3136 | spin_unlock(&space_info->lock); | |
3137 | } | |
3138 | ||
3139 | cond_resched(); | |
3140 | } while (wait_for_alloc); | |
3141 | ||
3142 | mutex_lock(&fs_info->chunk_mutex); | |
3143 | trans->allocating_chunk = true; | |
3144 | ||
3145 | /* | |
3146 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3147 | * allocating mixed chunks instead of individual chunks. | |
3148 | */ | |
3149 | if (btrfs_mixed_space_info(space_info)) | |
3150 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3151 | ||
3152 | /* | |
3153 | * if we're doing a data chunk, go ahead and make sure that | |
3154 | * we keep a reasonable number of metadata chunks allocated in the | |
3155 | * FS as well. | |
3156 | */ | |
3157 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { | |
3158 | fs_info->data_chunk_allocations++; | |
3159 | if (!(fs_info->data_chunk_allocations % | |
3160 | fs_info->metadata_ratio)) | |
3161 | force_metadata_allocation(fs_info); | |
3162 | } | |
3163 | ||
3164 | /* | |
3165 | * Check if we have enough space in SYSTEM chunk because we may need | |
3166 | * to update devices. | |
3167 | */ | |
3168 | check_system_chunk(trans, flags); | |
3169 | ||
3170 | ret = btrfs_alloc_chunk(trans, flags); | |
3171 | trans->allocating_chunk = false; | |
3172 | ||
3173 | spin_lock(&space_info->lock); | |
3174 | if (ret < 0) { | |
3175 | if (ret == -ENOSPC) | |
3176 | space_info->full = 1; | |
3177 | else | |
3178 | goto out; | |
3179 | } else { | |
3180 | ret = 1; | |
3181 | space_info->max_extent_size = 0; | |
3182 | } | |
3183 | ||
3184 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
3185 | out: | |
3186 | space_info->chunk_alloc = 0; | |
3187 | spin_unlock(&space_info->lock); | |
3188 | mutex_unlock(&fs_info->chunk_mutex); | |
3189 | /* | |
3190 | * When we allocate a new chunk we reserve space in the chunk block | |
3191 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
3192 | * add new nodes/leafs to it if we end up needing to do it when | |
3193 | * inserting the chunk item and updating device items as part of the | |
3194 | * second phase of chunk allocation, performed by | |
3195 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
3196 | * large number of new block groups to create in our transaction | |
3197 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
3198 | * in extreme cases - like having a single transaction create many new | |
3199 | * block groups when starting to write out the free space caches of all | |
3200 | * the block groups that were made dirty during the lifetime of the | |
3201 | * transaction. | |
3202 | */ | |
3203 | if (trans->chunk_bytes_reserved >= (u64)SZ_2M) | |
3204 | btrfs_create_pending_block_groups(trans); | |
3205 | ||
3206 | return ret; | |
3207 | } | |
3208 | ||
3209 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) | |
3210 | { | |
3211 | u64 num_dev; | |
3212 | ||
3213 | num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max; | |
3214 | if (!num_dev) | |
3215 | num_dev = fs_info->fs_devices->rw_devices; | |
3216 | ||
3217 | return num_dev; | |
3218 | } | |
3219 | ||
3220 | /* | |
a9143bd3 | 3221 | * Reserve space in the system space for allocating or removing a chunk |
07730d87 JB |
3222 | */ |
3223 | void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) | |
3224 | { | |
3225 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3226 | struct btrfs_space_info *info; | |
3227 | u64 left; | |
3228 | u64 thresh; | |
3229 | int ret = 0; | |
3230 | u64 num_devs; | |
3231 | ||
3232 | /* | |
3233 | * Needed because we can end up allocating a system chunk and for an | |
3234 | * atomic and race free space reservation in the chunk block reserve. | |
3235 | */ | |
3236 | lockdep_assert_held(&fs_info->chunk_mutex); | |
3237 | ||
3238 | info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3239 | spin_lock(&info->lock); | |
3240 | left = info->total_bytes - btrfs_space_info_used(info, true); | |
3241 | spin_unlock(&info->lock); | |
3242 | ||
3243 | num_devs = get_profile_num_devs(fs_info, type); | |
3244 | ||
3245 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
2bd36e7b JB |
3246 | thresh = btrfs_calc_metadata_size(fs_info, num_devs) + |
3247 | btrfs_calc_insert_metadata_size(fs_info, 1); | |
07730d87 JB |
3248 | |
3249 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { | |
3250 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
3251 | left, thresh, type); | |
3252 | btrfs_dump_space_info(fs_info, info, 0, 0); | |
3253 | } | |
3254 | ||
3255 | if (left < thresh) { | |
3256 | u64 flags = btrfs_system_alloc_profile(fs_info); | |
3257 | ||
3258 | /* | |
3259 | * Ignore failure to create system chunk. We might end up not | |
3260 | * needing it, as we might not need to COW all nodes/leafs from | |
3261 | * the paths we visit in the chunk tree (they were already COWed | |
3262 | * or created in the current transaction for example). | |
3263 | */ | |
3264 | ret = btrfs_alloc_chunk(trans, flags); | |
3265 | } | |
3266 | ||
3267 | if (!ret) { | |
3268 | ret = btrfs_block_rsv_add(fs_info->chunk_root, | |
3269 | &fs_info->chunk_block_rsv, | |
3270 | thresh, BTRFS_RESERVE_NO_FLUSH); | |
3271 | if (!ret) | |
3272 | trans->chunk_bytes_reserved += thresh; | |
3273 | } | |
3274 | } | |
3275 | ||
3e43c279 JB |
3276 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
3277 | { | |
32da5386 | 3278 | struct btrfs_block_group *block_group; |
3e43c279 JB |
3279 | u64 last = 0; |
3280 | ||
3281 | while (1) { | |
3282 | struct inode *inode; | |
3283 | ||
3284 | block_group = btrfs_lookup_first_block_group(info, last); | |
3285 | while (block_group) { | |
3286 | btrfs_wait_block_group_cache_done(block_group); | |
3287 | spin_lock(&block_group->lock); | |
3288 | if (block_group->iref) | |
3289 | break; | |
3290 | spin_unlock(&block_group->lock); | |
3291 | block_group = btrfs_next_block_group(block_group); | |
3292 | } | |
3293 | if (!block_group) { | |
3294 | if (last == 0) | |
3295 | break; | |
3296 | last = 0; | |
3297 | continue; | |
3298 | } | |
3299 | ||
3300 | inode = block_group->inode; | |
3301 | block_group->iref = 0; | |
3302 | block_group->inode = NULL; | |
3303 | spin_unlock(&block_group->lock); | |
3304 | ASSERT(block_group->io_ctl.inode == NULL); | |
3305 | iput(inode); | |
b3470b5d | 3306 | last = block_group->start + block_group->length; |
3e43c279 JB |
3307 | btrfs_put_block_group(block_group); |
3308 | } | |
3309 | } | |
3310 | ||
3311 | /* | |
3312 | * Must be called only after stopping all workers, since we could have block | |
3313 | * group caching kthreads running, and therefore they could race with us if we | |
3314 | * freed the block groups before stopping them. | |
3315 | */ | |
3316 | int btrfs_free_block_groups(struct btrfs_fs_info *info) | |
3317 | { | |
32da5386 | 3318 | struct btrfs_block_group *block_group; |
3e43c279 JB |
3319 | struct btrfs_space_info *space_info; |
3320 | struct btrfs_caching_control *caching_ctl; | |
3321 | struct rb_node *n; | |
3322 | ||
bbb86a37 | 3323 | spin_lock(&info->block_group_cache_lock); |
3e43c279 JB |
3324 | while (!list_empty(&info->caching_block_groups)) { |
3325 | caching_ctl = list_entry(info->caching_block_groups.next, | |
3326 | struct btrfs_caching_control, list); | |
3327 | list_del(&caching_ctl->list); | |
3328 | btrfs_put_caching_control(caching_ctl); | |
3329 | } | |
bbb86a37 | 3330 | spin_unlock(&info->block_group_cache_lock); |
3e43c279 JB |
3331 | |
3332 | spin_lock(&info->unused_bgs_lock); | |
3333 | while (!list_empty(&info->unused_bgs)) { | |
3334 | block_group = list_first_entry(&info->unused_bgs, | |
32da5386 | 3335 | struct btrfs_block_group, |
3e43c279 JB |
3336 | bg_list); |
3337 | list_del_init(&block_group->bg_list); | |
3338 | btrfs_put_block_group(block_group); | |
3339 | } | |
3340 | spin_unlock(&info->unused_bgs_lock); | |
3341 | ||
3342 | spin_lock(&info->block_group_cache_lock); | |
3343 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
32da5386 | 3344 | block_group = rb_entry(n, struct btrfs_block_group, |
3e43c279 JB |
3345 | cache_node); |
3346 | rb_erase(&block_group->cache_node, | |
3347 | &info->block_group_cache_tree); | |
3348 | RB_CLEAR_NODE(&block_group->cache_node); | |
3349 | spin_unlock(&info->block_group_cache_lock); | |
3350 | ||
3351 | down_write(&block_group->space_info->groups_sem); | |
3352 | list_del(&block_group->list); | |
3353 | up_write(&block_group->space_info->groups_sem); | |
3354 | ||
3355 | /* | |
3356 | * We haven't cached this block group, which means we could | |
3357 | * possibly have excluded extents on this block group. | |
3358 | */ | |
3359 | if (block_group->cached == BTRFS_CACHE_NO || | |
3360 | block_group->cached == BTRFS_CACHE_ERROR) | |
3361 | btrfs_free_excluded_extents(block_group); | |
3362 | ||
3363 | btrfs_remove_free_space_cache(block_group); | |
3364 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); | |
3365 | ASSERT(list_empty(&block_group->dirty_list)); | |
3366 | ASSERT(list_empty(&block_group->io_list)); | |
3367 | ASSERT(list_empty(&block_group->bg_list)); | |
48aaeebe | 3368 | ASSERT(refcount_read(&block_group->refs) == 1); |
3e43c279 JB |
3369 | btrfs_put_block_group(block_group); |
3370 | ||
3371 | spin_lock(&info->block_group_cache_lock); | |
3372 | } | |
3373 | spin_unlock(&info->block_group_cache_lock); | |
3374 | ||
3e43c279 JB |
3375 | btrfs_release_global_block_rsv(info); |
3376 | ||
3377 | while (!list_empty(&info->space_info)) { | |
3378 | space_info = list_entry(info->space_info.next, | |
3379 | struct btrfs_space_info, | |
3380 | list); | |
3381 | ||
3382 | /* | |
3383 | * Do not hide this behind enospc_debug, this is actually | |
3384 | * important and indicates a real bug if this happens. | |
3385 | */ | |
3386 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
3387 | space_info->bytes_reserved > 0 || | |
3388 | space_info->bytes_may_use > 0)) | |
3389 | btrfs_dump_space_info(info, space_info, 0, 0); | |
d611add4 | 3390 | WARN_ON(space_info->reclaim_size > 0); |
3e43c279 JB |
3391 | list_del(&space_info->list); |
3392 | btrfs_sysfs_remove_space_info(space_info); | |
3393 | } | |
3394 | return 0; | |
3395 | } | |
684b752b FM |
3396 | |
3397 | void btrfs_freeze_block_group(struct btrfs_block_group *cache) | |
3398 | { | |
3399 | atomic_inc(&cache->frozen); | |
3400 | } | |
3401 | ||
3402 | void btrfs_unfreeze_block_group(struct btrfs_block_group *block_group) | |
3403 | { | |
3404 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
3405 | struct extent_map_tree *em_tree; | |
3406 | struct extent_map *em; | |
3407 | bool cleanup; | |
3408 | ||
3409 | spin_lock(&block_group->lock); | |
3410 | cleanup = (atomic_dec_and_test(&block_group->frozen) && | |
3411 | block_group->removed); | |
3412 | spin_unlock(&block_group->lock); | |
3413 | ||
3414 | if (cleanup) { | |
684b752b FM |
3415 | em_tree = &fs_info->mapping_tree; |
3416 | write_lock(&em_tree->lock); | |
3417 | em = lookup_extent_mapping(em_tree, block_group->start, | |
3418 | 1); | |
3419 | BUG_ON(!em); /* logic error, can't happen */ | |
3420 | remove_extent_mapping(em_tree, em); | |
3421 | write_unlock(&em_tree->lock); | |
684b752b FM |
3422 | |
3423 | /* once for us and once for the tree */ | |
3424 | free_extent_map(em); | |
3425 | free_extent_map(em); | |
3426 | ||
3427 | /* | |
3428 | * We may have left one free space entry and other possible | |
3429 | * tasks trimming this block group have left 1 entry each one. | |
3430 | * Free them if any. | |
3431 | */ | |
3432 | __btrfs_remove_free_space_cache(block_group->free_space_ctl); | |
3433 | } | |
3434 | } |