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 | ||
427 | #ifdef CONFIG_BTRFS_DEBUG | |
32da5386 | 428 | static void fragment_free_space(struct btrfs_block_group *block_group) |
9f21246d JB |
429 | { |
430 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
b3470b5d DS |
431 | u64 start = block_group->start; |
432 | u64 len = block_group->length; | |
9f21246d JB |
433 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? |
434 | fs_info->nodesize : fs_info->sectorsize; | |
435 | u64 step = chunk << 1; | |
436 | ||
437 | while (len > chunk) { | |
438 | btrfs_remove_free_space(block_group, start, chunk); | |
439 | start += step; | |
440 | if (len < step) | |
441 | len = 0; | |
442 | else | |
443 | len -= step; | |
444 | } | |
445 | } | |
446 | #endif | |
447 | ||
448 | /* | |
449 | * This is only called by btrfs_cache_block_group, since we could have freed | |
450 | * extents we need to check the pinned_extents for any extents that can't be | |
451 | * used yet since their free space will be released as soon as the transaction | |
452 | * commits. | |
453 | */ | |
32da5386 | 454 | u64 add_new_free_space(struct btrfs_block_group *block_group, u64 start, u64 end) |
9f21246d JB |
455 | { |
456 | struct btrfs_fs_info *info = block_group->fs_info; | |
457 | u64 extent_start, extent_end, size, total_added = 0; | |
458 | int ret; | |
459 | ||
460 | while (start < end) { | |
fe119a6e | 461 | ret = find_first_extent_bit(&info->excluded_extents, start, |
9f21246d JB |
462 | &extent_start, &extent_end, |
463 | EXTENT_DIRTY | EXTENT_UPTODATE, | |
464 | NULL); | |
465 | if (ret) | |
466 | break; | |
467 | ||
468 | if (extent_start <= start) { | |
469 | start = extent_end + 1; | |
470 | } else if (extent_start > start && extent_start < end) { | |
471 | size = extent_start - start; | |
472 | total_added += size; | |
b0643e59 DZ |
473 | ret = btrfs_add_free_space_async_trimmed(block_group, |
474 | start, size); | |
9f21246d JB |
475 | BUG_ON(ret); /* -ENOMEM or logic error */ |
476 | start = extent_end + 1; | |
477 | } else { | |
478 | break; | |
479 | } | |
480 | } | |
481 | ||
482 | if (start < end) { | |
483 | size = end - start; | |
484 | total_added += size; | |
b0643e59 DZ |
485 | ret = btrfs_add_free_space_async_trimmed(block_group, start, |
486 | size); | |
9f21246d JB |
487 | BUG_ON(ret); /* -ENOMEM or logic error */ |
488 | } | |
489 | ||
490 | return total_added; | |
491 | } | |
492 | ||
493 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) | |
494 | { | |
32da5386 | 495 | struct btrfs_block_group *block_group = caching_ctl->block_group; |
9f21246d JB |
496 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
497 | struct btrfs_root *extent_root = fs_info->extent_root; | |
498 | struct btrfs_path *path; | |
499 | struct extent_buffer *leaf; | |
500 | struct btrfs_key key; | |
501 | u64 total_found = 0; | |
502 | u64 last = 0; | |
503 | u32 nritems; | |
504 | int ret; | |
505 | bool wakeup = true; | |
506 | ||
507 | path = btrfs_alloc_path(); | |
508 | if (!path) | |
509 | return -ENOMEM; | |
510 | ||
b3470b5d | 511 | last = max_t(u64, block_group->start, BTRFS_SUPER_INFO_OFFSET); |
9f21246d JB |
512 | |
513 | #ifdef CONFIG_BTRFS_DEBUG | |
514 | /* | |
515 | * If we're fragmenting we don't want to make anybody think we can | |
516 | * allocate from this block group until we've had a chance to fragment | |
517 | * the free space. | |
518 | */ | |
519 | if (btrfs_should_fragment_free_space(block_group)) | |
520 | wakeup = false; | |
521 | #endif | |
522 | /* | |
523 | * We don't want to deadlock with somebody trying to allocate a new | |
524 | * extent for the extent root while also trying to search the extent | |
525 | * root to add free space. So we skip locking and search the commit | |
526 | * root, since its read-only | |
527 | */ | |
528 | path->skip_locking = 1; | |
529 | path->search_commit_root = 1; | |
530 | path->reada = READA_FORWARD; | |
531 | ||
532 | key.objectid = last; | |
533 | key.offset = 0; | |
534 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
535 | ||
536 | next: | |
537 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
538 | if (ret < 0) | |
539 | goto out; | |
540 | ||
541 | leaf = path->nodes[0]; | |
542 | nritems = btrfs_header_nritems(leaf); | |
543 | ||
544 | while (1) { | |
545 | if (btrfs_fs_closing(fs_info) > 1) { | |
546 | last = (u64)-1; | |
547 | break; | |
548 | } | |
549 | ||
550 | if (path->slots[0] < nritems) { | |
551 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
552 | } else { | |
553 | ret = btrfs_find_next_key(extent_root, path, &key, 0, 0); | |
554 | if (ret) | |
555 | break; | |
556 | ||
557 | if (need_resched() || | |
558 | rwsem_is_contended(&fs_info->commit_root_sem)) { | |
559 | if (wakeup) | |
560 | caching_ctl->progress = last; | |
561 | btrfs_release_path(path); | |
562 | up_read(&fs_info->commit_root_sem); | |
563 | mutex_unlock(&caching_ctl->mutex); | |
564 | cond_resched(); | |
565 | mutex_lock(&caching_ctl->mutex); | |
566 | down_read(&fs_info->commit_root_sem); | |
567 | goto next; | |
568 | } | |
569 | ||
570 | ret = btrfs_next_leaf(extent_root, path); | |
571 | if (ret < 0) | |
572 | goto out; | |
573 | if (ret) | |
574 | break; | |
575 | leaf = path->nodes[0]; | |
576 | nritems = btrfs_header_nritems(leaf); | |
577 | continue; | |
578 | } | |
579 | ||
580 | if (key.objectid < last) { | |
581 | key.objectid = last; | |
582 | key.offset = 0; | |
583 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
584 | ||
585 | if (wakeup) | |
586 | caching_ctl->progress = last; | |
587 | btrfs_release_path(path); | |
588 | goto next; | |
589 | } | |
590 | ||
b3470b5d | 591 | if (key.objectid < block_group->start) { |
9f21246d JB |
592 | path->slots[0]++; |
593 | continue; | |
594 | } | |
595 | ||
b3470b5d | 596 | if (key.objectid >= block_group->start + block_group->length) |
9f21246d JB |
597 | break; |
598 | ||
599 | if (key.type == BTRFS_EXTENT_ITEM_KEY || | |
600 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
601 | total_found += add_new_free_space(block_group, last, | |
602 | key.objectid); | |
603 | if (key.type == BTRFS_METADATA_ITEM_KEY) | |
604 | last = key.objectid + | |
605 | fs_info->nodesize; | |
606 | else | |
607 | last = key.objectid + key.offset; | |
608 | ||
609 | if (total_found > CACHING_CTL_WAKE_UP) { | |
610 | total_found = 0; | |
611 | if (wakeup) | |
612 | wake_up(&caching_ctl->wait); | |
613 | } | |
614 | } | |
615 | path->slots[0]++; | |
616 | } | |
617 | ret = 0; | |
618 | ||
619 | total_found += add_new_free_space(block_group, last, | |
b3470b5d | 620 | block_group->start + block_group->length); |
9f21246d JB |
621 | caching_ctl->progress = (u64)-1; |
622 | ||
623 | out: | |
624 | btrfs_free_path(path); | |
625 | return ret; | |
626 | } | |
627 | ||
628 | static noinline void caching_thread(struct btrfs_work *work) | |
629 | { | |
32da5386 | 630 | struct btrfs_block_group *block_group; |
9f21246d JB |
631 | struct btrfs_fs_info *fs_info; |
632 | struct btrfs_caching_control *caching_ctl; | |
633 | int ret; | |
634 | ||
635 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
636 | block_group = caching_ctl->block_group; | |
637 | fs_info = block_group->fs_info; | |
638 | ||
639 | mutex_lock(&caching_ctl->mutex); | |
640 | down_read(&fs_info->commit_root_sem); | |
641 | ||
642 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) | |
643 | ret = load_free_space_tree(caching_ctl); | |
644 | else | |
645 | ret = load_extent_tree_free(caching_ctl); | |
646 | ||
647 | spin_lock(&block_group->lock); | |
648 | block_group->caching_ctl = NULL; | |
649 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; | |
650 | spin_unlock(&block_group->lock); | |
651 | ||
652 | #ifdef CONFIG_BTRFS_DEBUG | |
653 | if (btrfs_should_fragment_free_space(block_group)) { | |
654 | u64 bytes_used; | |
655 | ||
656 | spin_lock(&block_group->space_info->lock); | |
657 | spin_lock(&block_group->lock); | |
b3470b5d | 658 | bytes_used = block_group->length - block_group->used; |
9f21246d JB |
659 | block_group->space_info->bytes_used += bytes_used >> 1; |
660 | spin_unlock(&block_group->lock); | |
661 | spin_unlock(&block_group->space_info->lock); | |
e11c0406 | 662 | fragment_free_space(block_group); |
9f21246d JB |
663 | } |
664 | #endif | |
665 | ||
666 | caching_ctl->progress = (u64)-1; | |
667 | ||
668 | up_read(&fs_info->commit_root_sem); | |
669 | btrfs_free_excluded_extents(block_group); | |
670 | mutex_unlock(&caching_ctl->mutex); | |
671 | ||
672 | wake_up(&caching_ctl->wait); | |
673 | ||
674 | btrfs_put_caching_control(caching_ctl); | |
675 | btrfs_put_block_group(block_group); | |
676 | } | |
677 | ||
32da5386 | 678 | int btrfs_cache_block_group(struct btrfs_block_group *cache, int load_cache_only) |
9f21246d JB |
679 | { |
680 | DEFINE_WAIT(wait); | |
681 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
682 | struct btrfs_caching_control *caching_ctl; | |
683 | int ret = 0; | |
684 | ||
685 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); | |
686 | if (!caching_ctl) | |
687 | return -ENOMEM; | |
688 | ||
689 | INIT_LIST_HEAD(&caching_ctl->list); | |
690 | mutex_init(&caching_ctl->mutex); | |
691 | init_waitqueue_head(&caching_ctl->wait); | |
692 | caching_ctl->block_group = cache; | |
b3470b5d | 693 | caching_ctl->progress = cache->start; |
9f21246d | 694 | refcount_set(&caching_ctl->count, 1); |
a0cac0ec | 695 | btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL); |
9f21246d JB |
696 | |
697 | spin_lock(&cache->lock); | |
9f21246d JB |
698 | if (cache->cached != BTRFS_CACHE_NO) { |
699 | spin_unlock(&cache->lock); | |
700 | kfree(caching_ctl); | |
701 | return 0; | |
702 | } | |
703 | WARN_ON(cache->caching_ctl); | |
704 | cache->caching_ctl = caching_ctl; | |
705 | cache->cached = BTRFS_CACHE_FAST; | |
706 | spin_unlock(&cache->lock); | |
707 | ||
708 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { | |
709 | mutex_lock(&caching_ctl->mutex); | |
710 | ret = load_free_space_cache(cache); | |
711 | ||
712 | spin_lock(&cache->lock); | |
713 | if (ret == 1) { | |
714 | cache->caching_ctl = NULL; | |
715 | cache->cached = BTRFS_CACHE_FINISHED; | |
716 | cache->last_byte_to_unpin = (u64)-1; | |
717 | caching_ctl->progress = (u64)-1; | |
718 | } else { | |
719 | if (load_cache_only) { | |
720 | cache->caching_ctl = NULL; | |
721 | cache->cached = BTRFS_CACHE_NO; | |
722 | } else { | |
723 | cache->cached = BTRFS_CACHE_STARTED; | |
724 | cache->has_caching_ctl = 1; | |
725 | } | |
726 | } | |
727 | spin_unlock(&cache->lock); | |
728 | #ifdef CONFIG_BTRFS_DEBUG | |
729 | if (ret == 1 && | |
730 | btrfs_should_fragment_free_space(cache)) { | |
731 | u64 bytes_used; | |
732 | ||
733 | spin_lock(&cache->space_info->lock); | |
734 | spin_lock(&cache->lock); | |
b3470b5d | 735 | bytes_used = cache->length - cache->used; |
9f21246d JB |
736 | cache->space_info->bytes_used += bytes_used >> 1; |
737 | spin_unlock(&cache->lock); | |
738 | spin_unlock(&cache->space_info->lock); | |
e11c0406 | 739 | fragment_free_space(cache); |
9f21246d JB |
740 | } |
741 | #endif | |
742 | mutex_unlock(&caching_ctl->mutex); | |
743 | ||
744 | wake_up(&caching_ctl->wait); | |
745 | if (ret == 1) { | |
746 | btrfs_put_caching_control(caching_ctl); | |
747 | btrfs_free_excluded_extents(cache); | |
748 | return 0; | |
749 | } | |
750 | } else { | |
751 | /* | |
752 | * We're either using the free space tree or no caching at all. | |
753 | * Set cached to the appropriate value and wakeup any waiters. | |
754 | */ | |
755 | spin_lock(&cache->lock); | |
756 | if (load_cache_only) { | |
757 | cache->caching_ctl = NULL; | |
758 | cache->cached = BTRFS_CACHE_NO; | |
759 | } else { | |
760 | cache->cached = BTRFS_CACHE_STARTED; | |
761 | cache->has_caching_ctl = 1; | |
762 | } | |
763 | spin_unlock(&cache->lock); | |
764 | wake_up(&caching_ctl->wait); | |
765 | } | |
766 | ||
767 | if (load_cache_only) { | |
768 | btrfs_put_caching_control(caching_ctl); | |
769 | return 0; | |
770 | } | |
771 | ||
772 | down_write(&fs_info->commit_root_sem); | |
773 | refcount_inc(&caching_ctl->count); | |
774 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); | |
775 | up_write(&fs_info->commit_root_sem); | |
776 | ||
777 | btrfs_get_block_group(cache); | |
778 | ||
779 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); | |
780 | ||
781 | return ret; | |
782 | } | |
e3e0520b JB |
783 | |
784 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
785 | { | |
786 | u64 extra_flags = chunk_to_extended(flags) & | |
787 | BTRFS_EXTENDED_PROFILE_MASK; | |
788 | ||
789 | write_seqlock(&fs_info->profiles_lock); | |
790 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
791 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
792 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
793 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
794 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
795 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
796 | write_sequnlock(&fs_info->profiles_lock); | |
797 | } | |
798 | ||
799 | /* | |
800 | * Clear incompat bits for the following feature(s): | |
801 | * | |
802 | * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group | |
803 | * in the whole filesystem | |
9c907446 DS |
804 | * |
805 | * - RAID1C34 - same as above for RAID1C3 and RAID1C4 block groups | |
e3e0520b JB |
806 | */ |
807 | static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
808 | { | |
9c907446 DS |
809 | bool found_raid56 = false; |
810 | bool found_raid1c34 = false; | |
811 | ||
812 | if ((flags & BTRFS_BLOCK_GROUP_RAID56_MASK) || | |
813 | (flags & BTRFS_BLOCK_GROUP_RAID1C3) || | |
814 | (flags & BTRFS_BLOCK_GROUP_RAID1C4)) { | |
e3e0520b JB |
815 | struct list_head *head = &fs_info->space_info; |
816 | struct btrfs_space_info *sinfo; | |
817 | ||
818 | list_for_each_entry_rcu(sinfo, head, list) { | |
e3e0520b JB |
819 | down_read(&sinfo->groups_sem); |
820 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5])) | |
9c907446 | 821 | found_raid56 = true; |
e3e0520b | 822 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6])) |
9c907446 DS |
823 | found_raid56 = true; |
824 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C3])) | |
825 | found_raid1c34 = true; | |
826 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C4])) | |
827 | found_raid1c34 = true; | |
e3e0520b | 828 | up_read(&sinfo->groups_sem); |
e3e0520b | 829 | } |
d8e6fd5c | 830 | if (!found_raid56) |
9c907446 | 831 | btrfs_clear_fs_incompat(fs_info, RAID56); |
d8e6fd5c | 832 | if (!found_raid1c34) |
9c907446 | 833 | btrfs_clear_fs_incompat(fs_info, RAID1C34); |
e3e0520b JB |
834 | } |
835 | } | |
836 | ||
7357623a QW |
837 | static int remove_block_group_item(struct btrfs_trans_handle *trans, |
838 | struct btrfs_path *path, | |
839 | struct btrfs_block_group *block_group) | |
840 | { | |
841 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
842 | struct btrfs_root *root; | |
843 | struct btrfs_key key; | |
844 | int ret; | |
845 | ||
846 | root = fs_info->extent_root; | |
847 | key.objectid = block_group->start; | |
848 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
849 | key.offset = block_group->length; | |
850 | ||
851 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
852 | if (ret > 0) | |
853 | ret = -ENOENT; | |
854 | if (ret < 0) | |
855 | return ret; | |
856 | ||
857 | ret = btrfs_del_item(trans, root, path); | |
858 | return ret; | |
859 | } | |
860 | ||
e3e0520b JB |
861 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
862 | u64 group_start, struct extent_map *em) | |
863 | { | |
864 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
e3e0520b | 865 | struct btrfs_path *path; |
32da5386 | 866 | struct btrfs_block_group *block_group; |
e3e0520b JB |
867 | struct btrfs_free_cluster *cluster; |
868 | struct btrfs_root *tree_root = fs_info->tree_root; | |
869 | struct btrfs_key key; | |
870 | struct inode *inode; | |
871 | struct kobject *kobj = NULL; | |
872 | int ret; | |
873 | int index; | |
874 | int factor; | |
875 | struct btrfs_caching_control *caching_ctl = NULL; | |
876 | bool remove_em; | |
877 | bool remove_rsv = false; | |
878 | ||
879 | block_group = btrfs_lookup_block_group(fs_info, group_start); | |
880 | BUG_ON(!block_group); | |
881 | BUG_ON(!block_group->ro); | |
882 | ||
883 | trace_btrfs_remove_block_group(block_group); | |
884 | /* | |
885 | * Free the reserved super bytes from this block group before | |
886 | * remove it. | |
887 | */ | |
888 | btrfs_free_excluded_extents(block_group); | |
b3470b5d DS |
889 | btrfs_free_ref_tree_range(fs_info, block_group->start, |
890 | block_group->length); | |
e3e0520b | 891 | |
e3e0520b JB |
892 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
893 | factor = btrfs_bg_type_to_factor(block_group->flags); | |
894 | ||
895 | /* make sure this block group isn't part of an allocation cluster */ | |
896 | cluster = &fs_info->data_alloc_cluster; | |
897 | spin_lock(&cluster->refill_lock); | |
898 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
899 | spin_unlock(&cluster->refill_lock); | |
900 | ||
901 | /* | |
902 | * make sure this block group isn't part of a metadata | |
903 | * allocation cluster | |
904 | */ | |
905 | cluster = &fs_info->meta_alloc_cluster; | |
906 | spin_lock(&cluster->refill_lock); | |
907 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
908 | spin_unlock(&cluster->refill_lock); | |
909 | ||
910 | path = btrfs_alloc_path(); | |
911 | if (!path) { | |
912 | ret = -ENOMEM; | |
9fecd132 | 913 | goto out; |
e3e0520b JB |
914 | } |
915 | ||
916 | /* | |
917 | * get the inode first so any iput calls done for the io_list | |
918 | * aren't the final iput (no unlinks allowed now) | |
919 | */ | |
920 | inode = lookup_free_space_inode(block_group, path); | |
921 | ||
922 | mutex_lock(&trans->transaction->cache_write_mutex); | |
923 | /* | |
924 | * Make sure our free space cache IO is done before removing the | |
925 | * free space inode | |
926 | */ | |
927 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
928 | if (!list_empty(&block_group->io_list)) { | |
929 | list_del_init(&block_group->io_list); | |
930 | ||
931 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
932 | ||
933 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
934 | btrfs_wait_cache_io(trans, block_group, path); | |
935 | btrfs_put_block_group(block_group); | |
936 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
937 | } | |
938 | ||
939 | if (!list_empty(&block_group->dirty_list)) { | |
940 | list_del_init(&block_group->dirty_list); | |
941 | remove_rsv = true; | |
942 | btrfs_put_block_group(block_group); | |
943 | } | |
944 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
945 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
946 | ||
947 | if (!IS_ERR(inode)) { | |
948 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); | |
949 | if (ret) { | |
950 | btrfs_add_delayed_iput(inode); | |
9fecd132 | 951 | goto out; |
e3e0520b JB |
952 | } |
953 | clear_nlink(inode); | |
954 | /* One for the block groups ref */ | |
955 | spin_lock(&block_group->lock); | |
956 | if (block_group->iref) { | |
957 | block_group->iref = 0; | |
958 | block_group->inode = NULL; | |
959 | spin_unlock(&block_group->lock); | |
960 | iput(inode); | |
961 | } else { | |
962 | spin_unlock(&block_group->lock); | |
963 | } | |
964 | /* One for our lookup ref */ | |
965 | btrfs_add_delayed_iput(inode); | |
966 | } | |
967 | ||
968 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
e3e0520b | 969 | key.type = 0; |
b3470b5d | 970 | key.offset = block_group->start; |
e3e0520b JB |
971 | |
972 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
973 | if (ret < 0) | |
9fecd132 | 974 | goto out; |
e3e0520b JB |
975 | if (ret > 0) |
976 | btrfs_release_path(path); | |
977 | if (ret == 0) { | |
978 | ret = btrfs_del_item(trans, tree_root, path); | |
979 | if (ret) | |
9fecd132 | 980 | goto out; |
e3e0520b JB |
981 | btrfs_release_path(path); |
982 | } | |
983 | ||
984 | spin_lock(&fs_info->block_group_cache_lock); | |
985 | rb_erase(&block_group->cache_node, | |
986 | &fs_info->block_group_cache_tree); | |
987 | RB_CLEAR_NODE(&block_group->cache_node); | |
988 | ||
9fecd132 FM |
989 | /* Once for the block groups rbtree */ |
990 | btrfs_put_block_group(block_group); | |
991 | ||
b3470b5d | 992 | if (fs_info->first_logical_byte == block_group->start) |
e3e0520b JB |
993 | fs_info->first_logical_byte = (u64)-1; |
994 | spin_unlock(&fs_info->block_group_cache_lock); | |
995 | ||
996 | down_write(&block_group->space_info->groups_sem); | |
997 | /* | |
998 | * we must use list_del_init so people can check to see if they | |
999 | * are still on the list after taking the semaphore | |
1000 | */ | |
1001 | list_del_init(&block_group->list); | |
1002 | if (list_empty(&block_group->space_info->block_groups[index])) { | |
1003 | kobj = block_group->space_info->block_group_kobjs[index]; | |
1004 | block_group->space_info->block_group_kobjs[index] = NULL; | |
1005 | clear_avail_alloc_bits(fs_info, block_group->flags); | |
1006 | } | |
1007 | up_write(&block_group->space_info->groups_sem); | |
1008 | clear_incompat_bg_bits(fs_info, block_group->flags); | |
1009 | if (kobj) { | |
1010 | kobject_del(kobj); | |
1011 | kobject_put(kobj); | |
1012 | } | |
1013 | ||
1014 | if (block_group->has_caching_ctl) | |
1015 | caching_ctl = btrfs_get_caching_control(block_group); | |
1016 | if (block_group->cached == BTRFS_CACHE_STARTED) | |
1017 | btrfs_wait_block_group_cache_done(block_group); | |
1018 | if (block_group->has_caching_ctl) { | |
1019 | down_write(&fs_info->commit_root_sem); | |
1020 | if (!caching_ctl) { | |
1021 | struct btrfs_caching_control *ctl; | |
1022 | ||
1023 | list_for_each_entry(ctl, | |
1024 | &fs_info->caching_block_groups, list) | |
1025 | if (ctl->block_group == block_group) { | |
1026 | caching_ctl = ctl; | |
1027 | refcount_inc(&caching_ctl->count); | |
1028 | break; | |
1029 | } | |
1030 | } | |
1031 | if (caching_ctl) | |
1032 | list_del_init(&caching_ctl->list); | |
1033 | up_write(&fs_info->commit_root_sem); | |
1034 | if (caching_ctl) { | |
1035 | /* Once for the caching bgs list and once for us. */ | |
1036 | btrfs_put_caching_control(caching_ctl); | |
1037 | btrfs_put_caching_control(caching_ctl); | |
1038 | } | |
1039 | } | |
1040 | ||
1041 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
1042 | WARN_ON(!list_empty(&block_group->dirty_list)); | |
1043 | WARN_ON(!list_empty(&block_group->io_list)); | |
1044 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
1045 | ||
1046 | btrfs_remove_free_space_cache(block_group); | |
1047 | ||
1048 | spin_lock(&block_group->space_info->lock); | |
1049 | list_del_init(&block_group->ro_list); | |
1050 | ||
1051 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { | |
1052 | WARN_ON(block_group->space_info->total_bytes | |
b3470b5d | 1053 | < block_group->length); |
e3e0520b | 1054 | WARN_ON(block_group->space_info->bytes_readonly |
b3470b5d | 1055 | < block_group->length); |
e3e0520b | 1056 | WARN_ON(block_group->space_info->disk_total |
b3470b5d | 1057 | < block_group->length * factor); |
e3e0520b | 1058 | } |
b3470b5d DS |
1059 | block_group->space_info->total_bytes -= block_group->length; |
1060 | block_group->space_info->bytes_readonly -= block_group->length; | |
1061 | block_group->space_info->disk_total -= block_group->length * factor; | |
e3e0520b JB |
1062 | |
1063 | spin_unlock(&block_group->space_info->lock); | |
1064 | ||
ffcb9d44 FM |
1065 | /* |
1066 | * Remove the free space for the block group from the free space tree | |
1067 | * and the block group's item from the extent tree before marking the | |
1068 | * block group as removed. This is to prevent races with tasks that | |
1069 | * freeze and unfreeze a block group, this task and another task | |
1070 | * allocating a new block group - the unfreeze task ends up removing | |
1071 | * the block group's extent map before the task calling this function | |
1072 | * deletes the block group item from the extent tree, allowing for | |
1073 | * another task to attempt to create another block group with the same | |
1074 | * item key (and failing with -EEXIST and a transaction abort). | |
1075 | */ | |
1076 | ret = remove_block_group_free_space(trans, block_group); | |
1077 | if (ret) | |
1078 | goto out; | |
1079 | ||
1080 | ret = remove_block_group_item(trans, path, block_group); | |
1081 | if (ret < 0) | |
1082 | goto out; | |
1083 | ||
e3e0520b JB |
1084 | spin_lock(&block_group->lock); |
1085 | block_group->removed = 1; | |
1086 | /* | |
6b7304af FM |
1087 | * At this point trimming or scrub can't start on this block group, |
1088 | * because we removed the block group from the rbtree | |
1089 | * fs_info->block_group_cache_tree so no one can't find it anymore and | |
1090 | * even if someone already got this block group before we removed it | |
1091 | * from the rbtree, they have already incremented block_group->frozen - | |
1092 | * if they didn't, for the trimming case they won't find any free space | |
1093 | * entries because we already removed them all when we called | |
1094 | * btrfs_remove_free_space_cache(). | |
e3e0520b JB |
1095 | * |
1096 | * And we must not remove the extent map from the fs_info->mapping_tree | |
1097 | * to prevent the same logical address range and physical device space | |
6b7304af FM |
1098 | * ranges from being reused for a new block group. This is needed to |
1099 | * avoid races with trimming and scrub. | |
1100 | * | |
1101 | * An fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
e3e0520b JB |
1102 | * completely transactionless, so while it is trimming a range the |
1103 | * currently running transaction might finish and a new one start, | |
1104 | * allowing for new block groups to be created that can reuse the same | |
1105 | * physical device locations unless we take this special care. | |
1106 | * | |
1107 | * There may also be an implicit trim operation if the file system | |
1108 | * is mounted with -odiscard. The same protections must remain | |
1109 | * in place until the extents have been discarded completely when | |
1110 | * the transaction commit has completed. | |
1111 | */ | |
6b7304af | 1112 | remove_em = (atomic_read(&block_group->frozen) == 0); |
e3e0520b JB |
1113 | spin_unlock(&block_group->lock); |
1114 | ||
e3e0520b JB |
1115 | if (remove_em) { |
1116 | struct extent_map_tree *em_tree; | |
1117 | ||
1118 | em_tree = &fs_info->mapping_tree; | |
1119 | write_lock(&em_tree->lock); | |
1120 | remove_extent_mapping(em_tree, em); | |
1121 | write_unlock(&em_tree->lock); | |
1122 | /* once for the tree */ | |
1123 | free_extent_map(em); | |
1124 | } | |
f6033c5e | 1125 | |
9fecd132 | 1126 | out: |
f6033c5e XY |
1127 | /* Once for the lookup reference */ |
1128 | btrfs_put_block_group(block_group); | |
e3e0520b JB |
1129 | if (remove_rsv) |
1130 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1131 | btrfs_free_path(path); | |
1132 | return ret; | |
1133 | } | |
1134 | ||
1135 | struct btrfs_trans_handle *btrfs_start_trans_remove_block_group( | |
1136 | struct btrfs_fs_info *fs_info, const u64 chunk_offset) | |
1137 | { | |
1138 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; | |
1139 | struct extent_map *em; | |
1140 | struct map_lookup *map; | |
1141 | unsigned int num_items; | |
1142 | ||
1143 | read_lock(&em_tree->lock); | |
1144 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
1145 | read_unlock(&em_tree->lock); | |
1146 | ASSERT(em && em->start == chunk_offset); | |
1147 | ||
1148 | /* | |
1149 | * We need to reserve 3 + N units from the metadata space info in order | |
1150 | * to remove a block group (done at btrfs_remove_chunk() and at | |
1151 | * btrfs_remove_block_group()), which are used for: | |
1152 | * | |
1153 | * 1 unit for adding the free space inode's orphan (located in the tree | |
1154 | * of tree roots). | |
1155 | * 1 unit for deleting the block group item (located in the extent | |
1156 | * tree). | |
1157 | * 1 unit for deleting the free space item (located in tree of tree | |
1158 | * roots). | |
1159 | * N units for deleting N device extent items corresponding to each | |
1160 | * stripe (located in the device tree). | |
1161 | * | |
1162 | * In order to remove a block group we also need to reserve units in the | |
1163 | * system space info in order to update the chunk tree (update one or | |
1164 | * more device items and remove one chunk item), but this is done at | |
1165 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
1166 | */ | |
1167 | map = em->map_lookup; | |
1168 | num_items = 3 + map->num_stripes; | |
1169 | free_extent_map(em); | |
1170 | ||
1171 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, | |
7f9fe614 | 1172 | num_items); |
e3e0520b JB |
1173 | } |
1174 | ||
26ce2095 JB |
1175 | /* |
1176 | * Mark block group @cache read-only, so later write won't happen to block | |
1177 | * group @cache. | |
1178 | * | |
1179 | * If @force is not set, this function will only mark the block group readonly | |
1180 | * if we have enough free space (1M) in other metadata/system block groups. | |
1181 | * If @force is not set, this function will mark the block group readonly | |
1182 | * without checking free space. | |
1183 | * | |
1184 | * NOTE: This function doesn't care if other block groups can contain all the | |
1185 | * data in this block group. That check should be done by relocation routine, | |
1186 | * not this function. | |
1187 | */ | |
32da5386 | 1188 | static int inc_block_group_ro(struct btrfs_block_group *cache, int force) |
26ce2095 JB |
1189 | { |
1190 | struct btrfs_space_info *sinfo = cache->space_info; | |
1191 | u64 num_bytes; | |
26ce2095 JB |
1192 | int ret = -ENOSPC; |
1193 | ||
26ce2095 JB |
1194 | spin_lock(&sinfo->lock); |
1195 | spin_lock(&cache->lock); | |
1196 | ||
1197 | if (cache->ro) { | |
1198 | cache->ro++; | |
1199 | ret = 0; | |
1200 | goto out; | |
1201 | } | |
1202 | ||
b3470b5d | 1203 | num_bytes = cache->length - cache->reserved - cache->pinned - |
bf38be65 | 1204 | cache->bytes_super - cache->used; |
26ce2095 JB |
1205 | |
1206 | /* | |
a30a3d20 JB |
1207 | * Data never overcommits, even in mixed mode, so do just the straight |
1208 | * check of left over space in how much we have allocated. | |
26ce2095 | 1209 | */ |
a30a3d20 JB |
1210 | if (force) { |
1211 | ret = 0; | |
1212 | } else if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) { | |
1213 | u64 sinfo_used = btrfs_space_info_used(sinfo, true); | |
1214 | ||
1215 | /* | |
1216 | * Here we make sure if we mark this bg RO, we still have enough | |
1217 | * free space as buffer. | |
1218 | */ | |
1219 | if (sinfo_used + num_bytes <= sinfo->total_bytes) | |
1220 | ret = 0; | |
1221 | } else { | |
1222 | /* | |
1223 | * We overcommit metadata, so we need to do the | |
1224 | * btrfs_can_overcommit check here, and we need to pass in | |
1225 | * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of | |
1226 | * leeway to allow us to mark this block group as read only. | |
1227 | */ | |
1228 | if (btrfs_can_overcommit(cache->fs_info, sinfo, num_bytes, | |
1229 | BTRFS_RESERVE_NO_FLUSH)) | |
1230 | ret = 0; | |
1231 | } | |
1232 | ||
1233 | if (!ret) { | |
26ce2095 JB |
1234 | sinfo->bytes_readonly += num_bytes; |
1235 | cache->ro++; | |
1236 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); | |
26ce2095 JB |
1237 | } |
1238 | out: | |
1239 | spin_unlock(&cache->lock); | |
1240 | spin_unlock(&sinfo->lock); | |
1241 | if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { | |
1242 | btrfs_info(cache->fs_info, | |
b3470b5d | 1243 | "unable to make block group %llu ro", cache->start); |
26ce2095 JB |
1244 | btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0); |
1245 | } | |
1246 | return ret; | |
1247 | } | |
1248 | ||
fe119a6e NB |
1249 | static bool clean_pinned_extents(struct btrfs_trans_handle *trans, |
1250 | struct btrfs_block_group *bg) | |
45bb5d6a NB |
1251 | { |
1252 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
fe119a6e | 1253 | struct btrfs_transaction *prev_trans = NULL; |
45bb5d6a NB |
1254 | const u64 start = bg->start; |
1255 | const u64 end = start + bg->length - 1; | |
1256 | int ret; | |
1257 | ||
fe119a6e NB |
1258 | spin_lock(&fs_info->trans_lock); |
1259 | if (trans->transaction->list.prev != &fs_info->trans_list) { | |
1260 | prev_trans = list_last_entry(&trans->transaction->list, | |
1261 | struct btrfs_transaction, list); | |
1262 | refcount_inc(&prev_trans->use_count); | |
1263 | } | |
1264 | spin_unlock(&fs_info->trans_lock); | |
1265 | ||
45bb5d6a NB |
1266 | /* |
1267 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
1268 | * btrfs_finish_extent_commit(). If we are at transaction N, another | |
1269 | * task might be running finish_extent_commit() for the previous | |
1270 | * transaction N - 1, and have seen a range belonging to the block | |
fe119a6e NB |
1271 | * group in pinned_extents before we were able to clear the whole block |
1272 | * group range from pinned_extents. This means that task can lookup for | |
1273 | * the block group after we unpinned it from pinned_extents and removed | |
1274 | * it, leading to a BUG_ON() at unpin_extent_range(). | |
45bb5d6a NB |
1275 | */ |
1276 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
fe119a6e NB |
1277 | if (prev_trans) { |
1278 | ret = clear_extent_bits(&prev_trans->pinned_extents, start, end, | |
1279 | EXTENT_DIRTY); | |
1280 | if (ret) | |
534cf531 | 1281 | goto out; |
fe119a6e | 1282 | } |
45bb5d6a | 1283 | |
fe119a6e | 1284 | ret = clear_extent_bits(&trans->transaction->pinned_extents, start, end, |
45bb5d6a | 1285 | EXTENT_DIRTY); |
534cf531 | 1286 | out: |
45bb5d6a | 1287 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
5150bf19 FM |
1288 | if (prev_trans) |
1289 | btrfs_put_transaction(prev_trans); | |
45bb5d6a | 1290 | |
534cf531 | 1291 | return ret == 0; |
45bb5d6a NB |
1292 | } |
1293 | ||
e3e0520b JB |
1294 | /* |
1295 | * Process the unused_bgs list and remove any that don't have any allocated | |
1296 | * space inside of them. | |
1297 | */ | |
1298 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
1299 | { | |
32da5386 | 1300 | struct btrfs_block_group *block_group; |
e3e0520b JB |
1301 | struct btrfs_space_info *space_info; |
1302 | struct btrfs_trans_handle *trans; | |
6e80d4f8 | 1303 | const bool async_trim_enabled = btrfs_test_opt(fs_info, DISCARD_ASYNC); |
e3e0520b JB |
1304 | int ret = 0; |
1305 | ||
1306 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) | |
1307 | return; | |
1308 | ||
1309 | spin_lock(&fs_info->unused_bgs_lock); | |
1310 | while (!list_empty(&fs_info->unused_bgs)) { | |
e3e0520b JB |
1311 | int trimming; |
1312 | ||
1313 | block_group = list_first_entry(&fs_info->unused_bgs, | |
32da5386 | 1314 | struct btrfs_block_group, |
e3e0520b JB |
1315 | bg_list); |
1316 | list_del_init(&block_group->bg_list); | |
1317 | ||
1318 | space_info = block_group->space_info; | |
1319 | ||
1320 | if (ret || btrfs_mixed_space_info(space_info)) { | |
1321 | btrfs_put_block_group(block_group); | |
1322 | continue; | |
1323 | } | |
1324 | spin_unlock(&fs_info->unused_bgs_lock); | |
1325 | ||
b0643e59 DZ |
1326 | btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); |
1327 | ||
e3e0520b JB |
1328 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
1329 | ||
1330 | /* Don't want to race with allocators so take the groups_sem */ | |
1331 | down_write(&space_info->groups_sem); | |
6e80d4f8 DZ |
1332 | |
1333 | /* | |
1334 | * Async discard moves the final block group discard to be prior | |
1335 | * to the unused_bgs code path. Therefore, if it's not fully | |
1336 | * trimmed, punt it back to the async discard lists. | |
1337 | */ | |
1338 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC) && | |
1339 | !btrfs_is_free_space_trimmed(block_group)) { | |
1340 | trace_btrfs_skip_unused_block_group(block_group); | |
1341 | up_write(&space_info->groups_sem); | |
1342 | /* Requeue if we failed because of async discard */ | |
1343 | btrfs_discard_queue_work(&fs_info->discard_ctl, | |
1344 | block_group); | |
1345 | goto next; | |
1346 | } | |
1347 | ||
e3e0520b JB |
1348 | spin_lock(&block_group->lock); |
1349 | if (block_group->reserved || block_group->pinned || | |
bf38be65 | 1350 | block_group->used || block_group->ro || |
e3e0520b JB |
1351 | list_is_singular(&block_group->list)) { |
1352 | /* | |
1353 | * We want to bail if we made new allocations or have | |
1354 | * outstanding allocations in this block group. We do | |
1355 | * the ro check in case balance is currently acting on | |
1356 | * this block group. | |
1357 | */ | |
1358 | trace_btrfs_skip_unused_block_group(block_group); | |
1359 | spin_unlock(&block_group->lock); | |
1360 | up_write(&space_info->groups_sem); | |
1361 | goto next; | |
1362 | } | |
1363 | spin_unlock(&block_group->lock); | |
1364 | ||
1365 | /* We don't want to force the issue, only flip if it's ok. */ | |
e11c0406 | 1366 | ret = inc_block_group_ro(block_group, 0); |
e3e0520b JB |
1367 | up_write(&space_info->groups_sem); |
1368 | if (ret < 0) { | |
1369 | ret = 0; | |
1370 | goto next; | |
1371 | } | |
1372 | ||
1373 | /* | |
1374 | * Want to do this before we do anything else so we can recover | |
1375 | * properly if we fail to join the transaction. | |
1376 | */ | |
1377 | trans = btrfs_start_trans_remove_block_group(fs_info, | |
b3470b5d | 1378 | block_group->start); |
e3e0520b JB |
1379 | if (IS_ERR(trans)) { |
1380 | btrfs_dec_block_group_ro(block_group); | |
1381 | ret = PTR_ERR(trans); | |
1382 | goto next; | |
1383 | } | |
1384 | ||
1385 | /* | |
1386 | * We could have pending pinned extents for this block group, | |
1387 | * just delete them, we don't care about them anymore. | |
1388 | */ | |
534cf531 FM |
1389 | if (!clean_pinned_extents(trans, block_group)) { |
1390 | btrfs_dec_block_group_ro(block_group); | |
e3e0520b | 1391 | goto end_trans; |
534cf531 | 1392 | } |
e3e0520b | 1393 | |
b0643e59 DZ |
1394 | /* |
1395 | * At this point, the block_group is read only and should fail | |
1396 | * new allocations. However, btrfs_finish_extent_commit() can | |
1397 | * cause this block_group to be placed back on the discard | |
1398 | * lists because now the block_group isn't fully discarded. | |
1399 | * Bail here and try again later after discarding everything. | |
1400 | */ | |
1401 | spin_lock(&fs_info->discard_ctl.lock); | |
1402 | if (!list_empty(&block_group->discard_list)) { | |
1403 | spin_unlock(&fs_info->discard_ctl.lock); | |
1404 | btrfs_dec_block_group_ro(block_group); | |
1405 | btrfs_discard_queue_work(&fs_info->discard_ctl, | |
1406 | block_group); | |
1407 | goto end_trans; | |
1408 | } | |
1409 | spin_unlock(&fs_info->discard_ctl.lock); | |
1410 | ||
e3e0520b JB |
1411 | /* Reset pinned so btrfs_put_block_group doesn't complain */ |
1412 | spin_lock(&space_info->lock); | |
1413 | spin_lock(&block_group->lock); | |
1414 | ||
1415 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, | |
1416 | -block_group->pinned); | |
1417 | space_info->bytes_readonly += block_group->pinned; | |
1418 | percpu_counter_add_batch(&space_info->total_bytes_pinned, | |
1419 | -block_group->pinned, | |
1420 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
1421 | block_group->pinned = 0; | |
1422 | ||
1423 | spin_unlock(&block_group->lock); | |
1424 | spin_unlock(&space_info->lock); | |
1425 | ||
6e80d4f8 DZ |
1426 | /* |
1427 | * The normal path here is an unused block group is passed here, | |
1428 | * then trimming is handled in the transaction commit path. | |
1429 | * Async discard interposes before this to do the trimming | |
1430 | * before coming down the unused block group path as trimming | |
1431 | * will no longer be done later in the transaction commit path. | |
1432 | */ | |
1433 | if (!async_trim_enabled && btrfs_test_opt(fs_info, DISCARD_ASYNC)) | |
1434 | goto flip_async; | |
1435 | ||
e3e0520b | 1436 | /* DISCARD can flip during remount */ |
46b27f50 | 1437 | trimming = btrfs_test_opt(fs_info, DISCARD_SYNC); |
e3e0520b JB |
1438 | |
1439 | /* Implicit trim during transaction commit. */ | |
1440 | if (trimming) | |
6b7304af | 1441 | btrfs_freeze_block_group(block_group); |
e3e0520b JB |
1442 | |
1443 | /* | |
1444 | * Btrfs_remove_chunk will abort the transaction if things go | |
1445 | * horribly wrong. | |
1446 | */ | |
b3470b5d | 1447 | ret = btrfs_remove_chunk(trans, block_group->start); |
e3e0520b JB |
1448 | |
1449 | if (ret) { | |
1450 | if (trimming) | |
6b7304af | 1451 | btrfs_unfreeze_block_group(block_group); |
e3e0520b JB |
1452 | goto end_trans; |
1453 | } | |
1454 | ||
1455 | /* | |
1456 | * If we're not mounted with -odiscard, we can just forget | |
1457 | * about this block group. Otherwise we'll need to wait | |
1458 | * until transaction commit to do the actual discard. | |
1459 | */ | |
1460 | if (trimming) { | |
1461 | spin_lock(&fs_info->unused_bgs_lock); | |
1462 | /* | |
1463 | * A concurrent scrub might have added us to the list | |
1464 | * fs_info->unused_bgs, so use a list_move operation | |
1465 | * to add the block group to the deleted_bgs list. | |
1466 | */ | |
1467 | list_move(&block_group->bg_list, | |
1468 | &trans->transaction->deleted_bgs); | |
1469 | spin_unlock(&fs_info->unused_bgs_lock); | |
1470 | btrfs_get_block_group(block_group); | |
1471 | } | |
1472 | end_trans: | |
1473 | btrfs_end_transaction(trans); | |
1474 | next: | |
1475 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
1476 | btrfs_put_block_group(block_group); | |
1477 | spin_lock(&fs_info->unused_bgs_lock); | |
1478 | } | |
1479 | spin_unlock(&fs_info->unused_bgs_lock); | |
6e80d4f8 DZ |
1480 | return; |
1481 | ||
1482 | flip_async: | |
1483 | btrfs_end_transaction(trans); | |
1484 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
1485 | btrfs_put_block_group(block_group); | |
1486 | btrfs_discard_punt_unused_bgs_list(fs_info); | |
e3e0520b JB |
1487 | } |
1488 | ||
32da5386 | 1489 | void btrfs_mark_bg_unused(struct btrfs_block_group *bg) |
e3e0520b JB |
1490 | { |
1491 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
1492 | ||
1493 | spin_lock(&fs_info->unused_bgs_lock); | |
1494 | if (list_empty(&bg->bg_list)) { | |
1495 | btrfs_get_block_group(bg); | |
1496 | trace_btrfs_add_unused_block_group(bg); | |
1497 | list_add_tail(&bg->bg_list, &fs_info->unused_bgs); | |
1498 | } | |
1499 | spin_unlock(&fs_info->unused_bgs_lock); | |
1500 | } | |
4358d963 | 1501 | |
e3ba67a1 JT |
1502 | static int read_bg_from_eb(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
1503 | struct btrfs_path *path) | |
1504 | { | |
1505 | struct extent_map_tree *em_tree; | |
1506 | struct extent_map *em; | |
1507 | struct btrfs_block_group_item bg; | |
1508 | struct extent_buffer *leaf; | |
1509 | int slot; | |
1510 | u64 flags; | |
1511 | int ret = 0; | |
1512 | ||
1513 | slot = path->slots[0]; | |
1514 | leaf = path->nodes[0]; | |
1515 | ||
1516 | em_tree = &fs_info->mapping_tree; | |
1517 | read_lock(&em_tree->lock); | |
1518 | em = lookup_extent_mapping(em_tree, key->objectid, key->offset); | |
1519 | read_unlock(&em_tree->lock); | |
1520 | if (!em) { | |
1521 | btrfs_err(fs_info, | |
1522 | "logical %llu len %llu found bg but no related chunk", | |
1523 | key->objectid, key->offset); | |
1524 | return -ENOENT; | |
1525 | } | |
1526 | ||
1527 | if (em->start != key->objectid || em->len != key->offset) { | |
1528 | btrfs_err(fs_info, | |
1529 | "block group %llu len %llu mismatch with chunk %llu len %llu", | |
1530 | key->objectid, key->offset, em->start, em->len); | |
1531 | ret = -EUCLEAN; | |
1532 | goto out_free_em; | |
1533 | } | |
1534 | ||
1535 | read_extent_buffer(leaf, &bg, btrfs_item_ptr_offset(leaf, slot), | |
1536 | sizeof(bg)); | |
1537 | flags = btrfs_stack_block_group_flags(&bg) & | |
1538 | BTRFS_BLOCK_GROUP_TYPE_MASK; | |
1539 | ||
1540 | if (flags != (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
1541 | btrfs_err(fs_info, | |
1542 | "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", | |
1543 | key->objectid, key->offset, flags, | |
1544 | (BTRFS_BLOCK_GROUP_TYPE_MASK & em->map_lookup->type)); | |
1545 | ret = -EUCLEAN; | |
1546 | } | |
1547 | ||
1548 | out_free_em: | |
1549 | free_extent_map(em); | |
1550 | return ret; | |
1551 | } | |
1552 | ||
4358d963 JB |
1553 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
1554 | struct btrfs_path *path, | |
1555 | struct btrfs_key *key) | |
1556 | { | |
1557 | struct btrfs_root *root = fs_info->extent_root; | |
e3ba67a1 | 1558 | int ret; |
4358d963 JB |
1559 | struct btrfs_key found_key; |
1560 | struct extent_buffer *leaf; | |
4358d963 JB |
1561 | int slot; |
1562 | ||
1563 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
1564 | if (ret < 0) | |
e3ba67a1 | 1565 | return ret; |
4358d963 JB |
1566 | |
1567 | while (1) { | |
1568 | slot = path->slots[0]; | |
1569 | leaf = path->nodes[0]; | |
1570 | if (slot >= btrfs_header_nritems(leaf)) { | |
1571 | ret = btrfs_next_leaf(root, path); | |
1572 | if (ret == 0) | |
1573 | continue; | |
1574 | if (ret < 0) | |
1575 | goto out; | |
1576 | break; | |
1577 | } | |
1578 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
1579 | ||
1580 | if (found_key.objectid >= key->objectid && | |
1581 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { | |
e3ba67a1 JT |
1582 | ret = read_bg_from_eb(fs_info, &found_key, path); |
1583 | break; | |
4358d963 | 1584 | } |
e3ba67a1 | 1585 | |
4358d963 JB |
1586 | path->slots[0]++; |
1587 | } | |
1588 | out: | |
1589 | return ret; | |
1590 | } | |
1591 | ||
1592 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
1593 | { | |
1594 | u64 extra_flags = chunk_to_extended(flags) & | |
1595 | BTRFS_EXTENDED_PROFILE_MASK; | |
1596 | ||
1597 | write_seqlock(&fs_info->profiles_lock); | |
1598 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
1599 | fs_info->avail_data_alloc_bits |= extra_flags; | |
1600 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
1601 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
1602 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
1603 | fs_info->avail_system_alloc_bits |= extra_flags; | |
1604 | write_sequnlock(&fs_info->profiles_lock); | |
1605 | } | |
1606 | ||
96a14336 NB |
1607 | /** |
1608 | * btrfs_rmap_block - Map a physical disk address to a list of logical addresses | |
1609 | * @chunk_start: logical address of block group | |
1610 | * @physical: physical address to map to logical addresses | |
1611 | * @logical: return array of logical addresses which map to @physical | |
1612 | * @naddrs: length of @logical | |
1613 | * @stripe_len: size of IO stripe for the given block group | |
1614 | * | |
1615 | * Maps a particular @physical disk address to a list of @logical addresses. | |
1616 | * Used primarily to exclude those portions of a block group that contain super | |
1617 | * block copies. | |
1618 | */ | |
1619 | EXPORT_FOR_TESTS | |
1620 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, | |
1621 | u64 physical, u64 **logical, int *naddrs, int *stripe_len) | |
1622 | { | |
1623 | struct extent_map *em; | |
1624 | struct map_lookup *map; | |
1625 | u64 *buf; | |
1626 | u64 bytenr; | |
1776ad17 NB |
1627 | u64 data_stripe_length; |
1628 | u64 io_stripe_size; | |
1629 | int i, nr = 0; | |
1630 | int ret = 0; | |
96a14336 NB |
1631 | |
1632 | em = btrfs_get_chunk_map(fs_info, chunk_start, 1); | |
1633 | if (IS_ERR(em)) | |
1634 | return -EIO; | |
1635 | ||
1636 | map = em->map_lookup; | |
9e22b925 | 1637 | data_stripe_length = em->orig_block_len; |
1776ad17 | 1638 | io_stripe_size = map->stripe_len; |
96a14336 | 1639 | |
9e22b925 NB |
1640 | /* For RAID5/6 adjust to a full IO stripe length */ |
1641 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
1776ad17 | 1642 | io_stripe_size = map->stripe_len * nr_data_stripes(map); |
96a14336 NB |
1643 | |
1644 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); | |
1776ad17 NB |
1645 | if (!buf) { |
1646 | ret = -ENOMEM; | |
1647 | goto out; | |
1648 | } | |
96a14336 NB |
1649 | |
1650 | for (i = 0; i < map->num_stripes; i++) { | |
1776ad17 NB |
1651 | bool already_inserted = false; |
1652 | u64 stripe_nr; | |
1653 | int j; | |
1654 | ||
1655 | if (!in_range(physical, map->stripes[i].physical, | |
1656 | data_stripe_length)) | |
96a14336 NB |
1657 | continue; |
1658 | ||
1659 | stripe_nr = physical - map->stripes[i].physical; | |
1660 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); | |
1661 | ||
1662 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
1663 | stripe_nr = stripe_nr * map->num_stripes + i; | |
1664 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); | |
1665 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
1666 | stripe_nr = stripe_nr * map->num_stripes + i; | |
1667 | } | |
1668 | /* | |
1669 | * The remaining case would be for RAID56, multiply by | |
1670 | * nr_data_stripes(). Alternatively, just use rmap_len below | |
1671 | * instead of map->stripe_len | |
1672 | */ | |
1673 | ||
1776ad17 NB |
1674 | bytenr = chunk_start + stripe_nr * io_stripe_size; |
1675 | ||
1676 | /* Ensure we don't add duplicate addresses */ | |
96a14336 | 1677 | for (j = 0; j < nr; j++) { |
1776ad17 NB |
1678 | if (buf[j] == bytenr) { |
1679 | already_inserted = true; | |
96a14336 | 1680 | break; |
1776ad17 | 1681 | } |
96a14336 | 1682 | } |
1776ad17 NB |
1683 | |
1684 | if (!already_inserted) | |
96a14336 | 1685 | buf[nr++] = bytenr; |
96a14336 NB |
1686 | } |
1687 | ||
1688 | *logical = buf; | |
1689 | *naddrs = nr; | |
1776ad17 NB |
1690 | *stripe_len = io_stripe_size; |
1691 | out: | |
96a14336 | 1692 | free_extent_map(em); |
1776ad17 | 1693 | return ret; |
96a14336 NB |
1694 | } |
1695 | ||
32da5386 | 1696 | static int exclude_super_stripes(struct btrfs_block_group *cache) |
4358d963 JB |
1697 | { |
1698 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
1699 | u64 bytenr; | |
1700 | u64 *logical; | |
1701 | int stripe_len; | |
1702 | int i, nr, ret; | |
1703 | ||
b3470b5d DS |
1704 | if (cache->start < BTRFS_SUPER_INFO_OFFSET) { |
1705 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->start; | |
4358d963 | 1706 | cache->bytes_super += stripe_len; |
b3470b5d | 1707 | ret = btrfs_add_excluded_extent(fs_info, cache->start, |
4358d963 JB |
1708 | stripe_len); |
1709 | if (ret) | |
1710 | return ret; | |
1711 | } | |
1712 | ||
1713 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
1714 | bytenr = btrfs_sb_offset(i); | |
b3470b5d | 1715 | ret = btrfs_rmap_block(fs_info, cache->start, |
4358d963 JB |
1716 | bytenr, &logical, &nr, &stripe_len); |
1717 | if (ret) | |
1718 | return ret; | |
1719 | ||
1720 | while (nr--) { | |
96f9b0f2 NB |
1721 | u64 len = min_t(u64, stripe_len, |
1722 | cache->start + cache->length - logical[nr]); | |
4358d963 JB |
1723 | |
1724 | cache->bytes_super += len; | |
96f9b0f2 NB |
1725 | ret = btrfs_add_excluded_extent(fs_info, logical[nr], |
1726 | len); | |
4358d963 JB |
1727 | if (ret) { |
1728 | kfree(logical); | |
1729 | return ret; | |
1730 | } | |
1731 | } | |
1732 | ||
1733 | kfree(logical); | |
1734 | } | |
1735 | return 0; | |
1736 | } | |
1737 | ||
32da5386 | 1738 | static void link_block_group(struct btrfs_block_group *cache) |
4358d963 JB |
1739 | { |
1740 | struct btrfs_space_info *space_info = cache->space_info; | |
1741 | int index = btrfs_bg_flags_to_raid_index(cache->flags); | |
4358d963 JB |
1742 | |
1743 | down_write(&space_info->groups_sem); | |
4358d963 JB |
1744 | list_add_tail(&cache->list, &space_info->block_groups[index]); |
1745 | up_write(&space_info->groups_sem); | |
4358d963 JB |
1746 | } |
1747 | ||
32da5386 | 1748 | static struct btrfs_block_group *btrfs_create_block_group_cache( |
9afc6649 | 1749 | struct btrfs_fs_info *fs_info, u64 start) |
4358d963 | 1750 | { |
32da5386 | 1751 | struct btrfs_block_group *cache; |
4358d963 JB |
1752 | |
1753 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
1754 | if (!cache) | |
1755 | return NULL; | |
1756 | ||
1757 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
1758 | GFP_NOFS); | |
1759 | if (!cache->free_space_ctl) { | |
1760 | kfree(cache); | |
1761 | return NULL; | |
1762 | } | |
1763 | ||
b3470b5d | 1764 | cache->start = start; |
4358d963 JB |
1765 | |
1766 | cache->fs_info = fs_info; | |
1767 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); | |
4358d963 | 1768 | |
6e80d4f8 DZ |
1769 | cache->discard_index = BTRFS_DISCARD_INDEX_UNUSED; |
1770 | ||
48aaeebe | 1771 | refcount_set(&cache->refs, 1); |
4358d963 JB |
1772 | spin_lock_init(&cache->lock); |
1773 | init_rwsem(&cache->data_rwsem); | |
1774 | INIT_LIST_HEAD(&cache->list); | |
1775 | INIT_LIST_HEAD(&cache->cluster_list); | |
1776 | INIT_LIST_HEAD(&cache->bg_list); | |
1777 | INIT_LIST_HEAD(&cache->ro_list); | |
b0643e59 | 1778 | INIT_LIST_HEAD(&cache->discard_list); |
4358d963 JB |
1779 | INIT_LIST_HEAD(&cache->dirty_list); |
1780 | INIT_LIST_HEAD(&cache->io_list); | |
cd79909b | 1781 | btrfs_init_free_space_ctl(cache, cache->free_space_ctl); |
6b7304af | 1782 | atomic_set(&cache->frozen, 0); |
4358d963 JB |
1783 | mutex_init(&cache->free_space_lock); |
1784 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); | |
1785 | ||
1786 | return cache; | |
1787 | } | |
1788 | ||
1789 | /* | |
1790 | * Iterate all chunks and verify that each of them has the corresponding block | |
1791 | * group | |
1792 | */ | |
1793 | static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) | |
1794 | { | |
1795 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; | |
1796 | struct extent_map *em; | |
32da5386 | 1797 | struct btrfs_block_group *bg; |
4358d963 JB |
1798 | u64 start = 0; |
1799 | int ret = 0; | |
1800 | ||
1801 | while (1) { | |
1802 | read_lock(&map_tree->lock); | |
1803 | /* | |
1804 | * lookup_extent_mapping will return the first extent map | |
1805 | * intersecting the range, so setting @len to 1 is enough to | |
1806 | * get the first chunk. | |
1807 | */ | |
1808 | em = lookup_extent_mapping(map_tree, start, 1); | |
1809 | read_unlock(&map_tree->lock); | |
1810 | if (!em) | |
1811 | break; | |
1812 | ||
1813 | bg = btrfs_lookup_block_group(fs_info, em->start); | |
1814 | if (!bg) { | |
1815 | btrfs_err(fs_info, | |
1816 | "chunk start=%llu len=%llu doesn't have corresponding block group", | |
1817 | em->start, em->len); | |
1818 | ret = -EUCLEAN; | |
1819 | free_extent_map(em); | |
1820 | break; | |
1821 | } | |
b3470b5d | 1822 | if (bg->start != em->start || bg->length != em->len || |
4358d963 JB |
1823 | (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != |
1824 | (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
1825 | btrfs_err(fs_info, | |
1826 | "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", | |
1827 | em->start, em->len, | |
1828 | em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, | |
b3470b5d | 1829 | bg->start, bg->length, |
4358d963 JB |
1830 | bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); |
1831 | ret = -EUCLEAN; | |
1832 | free_extent_map(em); | |
1833 | btrfs_put_block_group(bg); | |
1834 | break; | |
1835 | } | |
1836 | start = em->start + em->len; | |
1837 | free_extent_map(em); | |
1838 | btrfs_put_block_group(bg); | |
1839 | } | |
1840 | return ret; | |
1841 | } | |
1842 | ||
4c448ce8 | 1843 | static void read_block_group_item(struct btrfs_block_group *cache, |
9afc6649 QW |
1844 | struct btrfs_path *path, |
1845 | const struct btrfs_key *key) | |
1846 | { | |
1847 | struct extent_buffer *leaf = path->nodes[0]; | |
1848 | struct btrfs_block_group_item bgi; | |
1849 | int slot = path->slots[0]; | |
1850 | ||
1851 | cache->length = key->offset; | |
1852 | ||
1853 | read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot), | |
1854 | sizeof(bgi)); | |
1855 | cache->used = btrfs_stack_block_group_used(&bgi); | |
1856 | cache->flags = btrfs_stack_block_group_flags(&bgi); | |
9afc6649 QW |
1857 | } |
1858 | ||
ffb9e0f0 QW |
1859 | static int read_one_block_group(struct btrfs_fs_info *info, |
1860 | struct btrfs_path *path, | |
d49a2ddb | 1861 | const struct btrfs_key *key, |
ffb9e0f0 QW |
1862 | int need_clear) |
1863 | { | |
32da5386 | 1864 | struct btrfs_block_group *cache; |
ffb9e0f0 | 1865 | struct btrfs_space_info *space_info; |
ffb9e0f0 | 1866 | const bool mixed = btrfs_fs_incompat(info, MIXED_GROUPS); |
ffb9e0f0 QW |
1867 | int ret; |
1868 | ||
d49a2ddb | 1869 | ASSERT(key->type == BTRFS_BLOCK_GROUP_ITEM_KEY); |
ffb9e0f0 | 1870 | |
9afc6649 | 1871 | cache = btrfs_create_block_group_cache(info, key->objectid); |
ffb9e0f0 QW |
1872 | if (!cache) |
1873 | return -ENOMEM; | |
1874 | ||
4c448ce8 | 1875 | read_block_group_item(cache, path, key); |
9afc6649 | 1876 | |
e3e39c72 MPS |
1877 | set_free_space_tree_thresholds(cache); |
1878 | ||
ffb9e0f0 QW |
1879 | if (need_clear) { |
1880 | /* | |
1881 | * When we mount with old space cache, we need to | |
1882 | * set BTRFS_DC_CLEAR and set dirty flag. | |
1883 | * | |
1884 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
1885 | * truncate the old free space cache inode and | |
1886 | * setup a new one. | |
1887 | * b) Setting 'dirty flag' makes sure that we flush | |
1888 | * the new space cache info onto disk. | |
1889 | */ | |
1890 | if (btrfs_test_opt(info, SPACE_CACHE)) | |
1891 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
1892 | } | |
ffb9e0f0 QW |
1893 | if (!mixed && ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && |
1894 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
1895 | btrfs_err(info, | |
1896 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
1897 | cache->start); | |
1898 | ret = -EINVAL; | |
1899 | goto error; | |
1900 | } | |
1901 | ||
1902 | /* | |
1903 | * We need to exclude the super stripes now so that the space info has | |
1904 | * super bytes accounted for, otherwise we'll think we have more space | |
1905 | * than we actually do. | |
1906 | */ | |
1907 | ret = exclude_super_stripes(cache); | |
1908 | if (ret) { | |
1909 | /* We may have excluded something, so call this just in case. */ | |
1910 | btrfs_free_excluded_extents(cache); | |
1911 | goto error; | |
1912 | } | |
1913 | ||
1914 | /* | |
1915 | * Check for two cases, either we are full, and therefore don't need | |
1916 | * to bother with the caching work since we won't find any space, or we | |
1917 | * are empty, and we can just add all the space in and be done with it. | |
1918 | * This saves us _a_lot_ of time, particularly in the full case. | |
1919 | */ | |
9afc6649 | 1920 | if (cache->length == cache->used) { |
ffb9e0f0 QW |
1921 | cache->last_byte_to_unpin = (u64)-1; |
1922 | cache->cached = BTRFS_CACHE_FINISHED; | |
1923 | btrfs_free_excluded_extents(cache); | |
1924 | } else if (cache->used == 0) { | |
1925 | cache->last_byte_to_unpin = (u64)-1; | |
1926 | cache->cached = BTRFS_CACHE_FINISHED; | |
9afc6649 QW |
1927 | add_new_free_space(cache, cache->start, |
1928 | cache->start + cache->length); | |
ffb9e0f0 QW |
1929 | btrfs_free_excluded_extents(cache); |
1930 | } | |
1931 | ||
1932 | ret = btrfs_add_block_group_cache(info, cache); | |
1933 | if (ret) { | |
1934 | btrfs_remove_free_space_cache(cache); | |
1935 | goto error; | |
1936 | } | |
1937 | trace_btrfs_add_block_group(info, cache, 0); | |
9afc6649 | 1938 | btrfs_update_space_info(info, cache->flags, cache->length, |
ffb9e0f0 QW |
1939 | cache->used, cache->bytes_super, &space_info); |
1940 | ||
1941 | cache->space_info = space_info; | |
1942 | ||
1943 | link_block_group(cache); | |
1944 | ||
1945 | set_avail_alloc_bits(info, cache->flags); | |
1946 | if (btrfs_chunk_readonly(info, cache->start)) { | |
1947 | inc_block_group_ro(cache, 1); | |
1948 | } else if (cache->used == 0) { | |
1949 | ASSERT(list_empty(&cache->bg_list)); | |
6e80d4f8 DZ |
1950 | if (btrfs_test_opt(info, DISCARD_ASYNC)) |
1951 | btrfs_discard_queue_work(&info->discard_ctl, cache); | |
1952 | else | |
1953 | btrfs_mark_bg_unused(cache); | |
ffb9e0f0 QW |
1954 | } |
1955 | return 0; | |
1956 | error: | |
1957 | btrfs_put_block_group(cache); | |
1958 | return ret; | |
1959 | } | |
1960 | ||
42437a63 JB |
1961 | static int fill_dummy_bgs(struct btrfs_fs_info *fs_info) |
1962 | { | |
1963 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; | |
1964 | struct btrfs_space_info *space_info; | |
1965 | struct rb_node *node; | |
1966 | int ret = 0; | |
1967 | ||
1968 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { | |
1969 | struct extent_map *em; | |
1970 | struct map_lookup *map; | |
1971 | struct btrfs_block_group *bg; | |
1972 | ||
1973 | em = rb_entry(node, struct extent_map, rb_node); | |
1974 | map = em->map_lookup; | |
1975 | bg = btrfs_create_block_group_cache(fs_info, em->start); | |
1976 | if (!bg) { | |
1977 | ret = -ENOMEM; | |
1978 | break; | |
1979 | } | |
1980 | ||
1981 | /* Fill dummy cache as FULL */ | |
1982 | bg->length = em->len; | |
1983 | bg->flags = map->type; | |
1984 | bg->last_byte_to_unpin = (u64)-1; | |
1985 | bg->cached = BTRFS_CACHE_FINISHED; | |
1986 | bg->used = em->len; | |
1987 | bg->flags = map->type; | |
1988 | ret = btrfs_add_block_group_cache(fs_info, bg); | |
1989 | if (ret) { | |
1990 | btrfs_remove_free_space_cache(bg); | |
1991 | btrfs_put_block_group(bg); | |
1992 | break; | |
1993 | } | |
1994 | btrfs_update_space_info(fs_info, bg->flags, em->len, em->len, | |
1995 | 0, &space_info); | |
1996 | bg->space_info = space_info; | |
1997 | link_block_group(bg); | |
1998 | ||
1999 | set_avail_alloc_bits(fs_info, bg->flags); | |
2000 | } | |
2001 | if (!ret) | |
2002 | btrfs_init_global_block_rsv(fs_info); | |
2003 | return ret; | |
2004 | } | |
2005 | ||
4358d963 JB |
2006 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
2007 | { | |
2008 | struct btrfs_path *path; | |
2009 | int ret; | |
32da5386 | 2010 | struct btrfs_block_group *cache; |
4358d963 JB |
2011 | struct btrfs_space_info *space_info; |
2012 | struct btrfs_key key; | |
4358d963 JB |
2013 | int need_clear = 0; |
2014 | u64 cache_gen; | |
4358d963 | 2015 | |
42437a63 JB |
2016 | if (!info->extent_root) |
2017 | return fill_dummy_bgs(info); | |
2018 | ||
4358d963 JB |
2019 | key.objectid = 0; |
2020 | key.offset = 0; | |
2021 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
2022 | path = btrfs_alloc_path(); | |
2023 | if (!path) | |
2024 | return -ENOMEM; | |
4358d963 JB |
2025 | |
2026 | cache_gen = btrfs_super_cache_generation(info->super_copy); | |
2027 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
2028 | btrfs_super_generation(info->super_copy) != cache_gen) | |
2029 | need_clear = 1; | |
2030 | if (btrfs_test_opt(info, CLEAR_CACHE)) | |
2031 | need_clear = 1; | |
2032 | ||
2033 | while (1) { | |
2034 | ret = find_first_block_group(info, path, &key); | |
2035 | if (ret > 0) | |
2036 | break; | |
2037 | if (ret != 0) | |
2038 | goto error; | |
2039 | ||
ffb9e0f0 | 2040 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); |
d49a2ddb | 2041 | ret = read_one_block_group(info, path, &key, need_clear); |
ffb9e0f0 | 2042 | if (ret < 0) |
4358d963 | 2043 | goto error; |
ffb9e0f0 QW |
2044 | key.objectid += key.offset; |
2045 | key.offset = 0; | |
4358d963 | 2046 | btrfs_release_path(path); |
4358d963 | 2047 | } |
7837fa88 | 2048 | btrfs_release_path(path); |
4358d963 | 2049 | |
72804905 | 2050 | list_for_each_entry(space_info, &info->space_info, list) { |
49ea112d JB |
2051 | int i; |
2052 | ||
2053 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { | |
2054 | if (list_empty(&space_info->block_groups[i])) | |
2055 | continue; | |
2056 | cache = list_first_entry(&space_info->block_groups[i], | |
2057 | struct btrfs_block_group, | |
2058 | list); | |
2059 | btrfs_sysfs_add_block_group_type(cache); | |
2060 | } | |
2061 | ||
4358d963 JB |
2062 | if (!(btrfs_get_alloc_profile(info, space_info->flags) & |
2063 | (BTRFS_BLOCK_GROUP_RAID10 | | |
2064 | BTRFS_BLOCK_GROUP_RAID1_MASK | | |
2065 | BTRFS_BLOCK_GROUP_RAID56_MASK | | |
2066 | BTRFS_BLOCK_GROUP_DUP))) | |
2067 | continue; | |
2068 | /* | |
2069 | * Avoid allocating from un-mirrored block group if there are | |
2070 | * mirrored block groups. | |
2071 | */ | |
2072 | list_for_each_entry(cache, | |
2073 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
2074 | list) | |
e11c0406 | 2075 | inc_block_group_ro(cache, 1); |
4358d963 JB |
2076 | list_for_each_entry(cache, |
2077 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
2078 | list) | |
e11c0406 | 2079 | inc_block_group_ro(cache, 1); |
4358d963 JB |
2080 | } |
2081 | ||
2082 | btrfs_init_global_block_rsv(info); | |
2083 | ret = check_chunk_block_group_mappings(info); | |
2084 | error: | |
2085 | btrfs_free_path(path); | |
2086 | return ret; | |
2087 | } | |
2088 | ||
97f4728a QW |
2089 | static int insert_block_group_item(struct btrfs_trans_handle *trans, |
2090 | struct btrfs_block_group *block_group) | |
2091 | { | |
2092 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2093 | struct btrfs_block_group_item bgi; | |
2094 | struct btrfs_root *root; | |
2095 | struct btrfs_key key; | |
2096 | ||
2097 | spin_lock(&block_group->lock); | |
2098 | btrfs_set_stack_block_group_used(&bgi, block_group->used); | |
2099 | btrfs_set_stack_block_group_chunk_objectid(&bgi, | |
2100 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
2101 | btrfs_set_stack_block_group_flags(&bgi, block_group->flags); | |
2102 | key.objectid = block_group->start; | |
2103 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
2104 | key.offset = block_group->length; | |
2105 | spin_unlock(&block_group->lock); | |
2106 | ||
2107 | root = fs_info->extent_root; | |
2108 | return btrfs_insert_item(trans, root, &key, &bgi, sizeof(bgi)); | |
2109 | } | |
2110 | ||
4358d963 JB |
2111 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
2112 | { | |
2113 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2114 | struct btrfs_block_group *block_group; |
4358d963 JB |
2115 | int ret = 0; |
2116 | ||
2117 | if (!trans->can_flush_pending_bgs) | |
2118 | return; | |
2119 | ||
2120 | while (!list_empty(&trans->new_bgs)) { | |
49ea112d JB |
2121 | int index; |
2122 | ||
4358d963 | 2123 | block_group = list_first_entry(&trans->new_bgs, |
32da5386 | 2124 | struct btrfs_block_group, |
4358d963 JB |
2125 | bg_list); |
2126 | if (ret) | |
2127 | goto next; | |
2128 | ||
49ea112d JB |
2129 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
2130 | ||
97f4728a | 2131 | ret = insert_block_group_item(trans, block_group); |
4358d963 JB |
2132 | if (ret) |
2133 | btrfs_abort_transaction(trans, ret); | |
97f4728a QW |
2134 | ret = btrfs_finish_chunk_alloc(trans, block_group->start, |
2135 | block_group->length); | |
4358d963 JB |
2136 | if (ret) |
2137 | btrfs_abort_transaction(trans, ret); | |
2138 | add_block_group_free_space(trans, block_group); | |
49ea112d JB |
2139 | |
2140 | /* | |
2141 | * If we restriped during balance, we may have added a new raid | |
2142 | * type, so now add the sysfs entries when it is safe to do so. | |
2143 | * We don't have to worry about locking here as it's handled in | |
2144 | * btrfs_sysfs_add_block_group_type. | |
2145 | */ | |
2146 | if (block_group->space_info->block_group_kobjs[index] == NULL) | |
2147 | btrfs_sysfs_add_block_group_type(block_group); | |
2148 | ||
4358d963 JB |
2149 | /* Already aborted the transaction if it failed. */ |
2150 | next: | |
2151 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
2152 | list_del_init(&block_group->bg_list); | |
2153 | } | |
2154 | btrfs_trans_release_chunk_metadata(trans); | |
2155 | } | |
2156 | ||
2157 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used, | |
2158 | u64 type, u64 chunk_offset, u64 size) | |
2159 | { | |
2160 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2161 | struct btrfs_block_group *cache; |
4358d963 JB |
2162 | int ret; |
2163 | ||
2164 | btrfs_set_log_full_commit(trans); | |
2165 | ||
9afc6649 | 2166 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset); |
4358d963 JB |
2167 | if (!cache) |
2168 | return -ENOMEM; | |
2169 | ||
9afc6649 | 2170 | cache->length = size; |
e3e39c72 | 2171 | set_free_space_tree_thresholds(cache); |
bf38be65 | 2172 | cache->used = bytes_used; |
4358d963 JB |
2173 | cache->flags = type; |
2174 | cache->last_byte_to_unpin = (u64)-1; | |
2175 | cache->cached = BTRFS_CACHE_FINISHED; | |
2176 | cache->needs_free_space = 1; | |
2177 | ret = exclude_super_stripes(cache); | |
2178 | if (ret) { | |
2179 | /* We may have excluded something, so call this just in case */ | |
2180 | btrfs_free_excluded_extents(cache); | |
2181 | btrfs_put_block_group(cache); | |
2182 | return ret; | |
2183 | } | |
2184 | ||
2185 | add_new_free_space(cache, chunk_offset, chunk_offset + size); | |
2186 | ||
2187 | btrfs_free_excluded_extents(cache); | |
2188 | ||
2189 | #ifdef CONFIG_BTRFS_DEBUG | |
2190 | if (btrfs_should_fragment_free_space(cache)) { | |
2191 | u64 new_bytes_used = size - bytes_used; | |
2192 | ||
2193 | bytes_used += new_bytes_used >> 1; | |
e11c0406 | 2194 | fragment_free_space(cache); |
4358d963 JB |
2195 | } |
2196 | #endif | |
2197 | /* | |
2198 | * Ensure the corresponding space_info object is created and | |
2199 | * assigned to our block group. We want our bg to be added to the rbtree | |
2200 | * with its ->space_info set. | |
2201 | */ | |
2202 | cache->space_info = btrfs_find_space_info(fs_info, cache->flags); | |
2203 | ASSERT(cache->space_info); | |
2204 | ||
2205 | ret = btrfs_add_block_group_cache(fs_info, cache); | |
2206 | if (ret) { | |
2207 | btrfs_remove_free_space_cache(cache); | |
2208 | btrfs_put_block_group(cache); | |
2209 | return ret; | |
2210 | } | |
2211 | ||
2212 | /* | |
2213 | * Now that our block group has its ->space_info set and is inserted in | |
2214 | * the rbtree, update the space info's counters. | |
2215 | */ | |
2216 | trace_btrfs_add_block_group(fs_info, cache, 1); | |
2217 | btrfs_update_space_info(fs_info, cache->flags, size, bytes_used, | |
2218 | cache->bytes_super, &cache->space_info); | |
2219 | btrfs_update_global_block_rsv(fs_info); | |
2220 | ||
2221 | link_block_group(cache); | |
2222 | ||
2223 | list_add_tail(&cache->bg_list, &trans->new_bgs); | |
2224 | trans->delayed_ref_updates++; | |
2225 | btrfs_update_delayed_refs_rsv(trans); | |
2226 | ||
2227 | set_avail_alloc_bits(fs_info, type); | |
2228 | return 0; | |
2229 | } | |
26ce2095 | 2230 | |
b12de528 QW |
2231 | /* |
2232 | * Mark one block group RO, can be called several times for the same block | |
2233 | * group. | |
2234 | * | |
2235 | * @cache: the destination block group | |
2236 | * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to | |
2237 | * ensure we still have some free space after marking this | |
2238 | * block group RO. | |
2239 | */ | |
2240 | int btrfs_inc_block_group_ro(struct btrfs_block_group *cache, | |
2241 | bool do_chunk_alloc) | |
26ce2095 JB |
2242 | { |
2243 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
2244 | struct btrfs_trans_handle *trans; | |
2245 | u64 alloc_flags; | |
2246 | int ret; | |
2247 | ||
2248 | again: | |
2249 | trans = btrfs_join_transaction(fs_info->extent_root); | |
2250 | if (IS_ERR(trans)) | |
2251 | return PTR_ERR(trans); | |
2252 | ||
2253 | /* | |
2254 | * we're not allowed to set block groups readonly after the dirty | |
2255 | * block groups cache has started writing. If it already started, | |
2256 | * back off and let this transaction commit | |
2257 | */ | |
2258 | mutex_lock(&fs_info->ro_block_group_mutex); | |
2259 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { | |
2260 | u64 transid = trans->transid; | |
2261 | ||
2262 | mutex_unlock(&fs_info->ro_block_group_mutex); | |
2263 | btrfs_end_transaction(trans); | |
2264 | ||
2265 | ret = btrfs_wait_for_commit(fs_info, transid); | |
2266 | if (ret) | |
2267 | return ret; | |
2268 | goto again; | |
2269 | } | |
2270 | ||
b12de528 | 2271 | if (do_chunk_alloc) { |
26ce2095 | 2272 | /* |
b12de528 QW |
2273 | * If we are changing raid levels, try to allocate a |
2274 | * corresponding block group with the new raid level. | |
26ce2095 | 2275 | */ |
349e120e | 2276 | alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); |
b12de528 QW |
2277 | if (alloc_flags != cache->flags) { |
2278 | ret = btrfs_chunk_alloc(trans, alloc_flags, | |
2279 | CHUNK_ALLOC_FORCE); | |
2280 | /* | |
2281 | * ENOSPC is allowed here, we may have enough space | |
2282 | * already allocated at the new raid level to carry on | |
2283 | */ | |
2284 | if (ret == -ENOSPC) | |
2285 | ret = 0; | |
2286 | if (ret < 0) | |
2287 | goto out; | |
2288 | } | |
26ce2095 JB |
2289 | } |
2290 | ||
a7a63acc | 2291 | ret = inc_block_group_ro(cache, 0); |
b12de528 QW |
2292 | if (!do_chunk_alloc) |
2293 | goto unlock_out; | |
26ce2095 JB |
2294 | if (!ret) |
2295 | goto out; | |
2296 | alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags); | |
2297 | ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); | |
2298 | if (ret < 0) | |
2299 | goto out; | |
e11c0406 | 2300 | ret = inc_block_group_ro(cache, 0); |
26ce2095 JB |
2301 | out: |
2302 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { | |
349e120e | 2303 | alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); |
26ce2095 JB |
2304 | mutex_lock(&fs_info->chunk_mutex); |
2305 | check_system_chunk(trans, alloc_flags); | |
2306 | mutex_unlock(&fs_info->chunk_mutex); | |
2307 | } | |
b12de528 | 2308 | unlock_out: |
26ce2095 JB |
2309 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2310 | ||
2311 | btrfs_end_transaction(trans); | |
2312 | return ret; | |
2313 | } | |
2314 | ||
32da5386 | 2315 | void btrfs_dec_block_group_ro(struct btrfs_block_group *cache) |
26ce2095 JB |
2316 | { |
2317 | struct btrfs_space_info *sinfo = cache->space_info; | |
2318 | u64 num_bytes; | |
2319 | ||
2320 | BUG_ON(!cache->ro); | |
2321 | ||
2322 | spin_lock(&sinfo->lock); | |
2323 | spin_lock(&cache->lock); | |
2324 | if (!--cache->ro) { | |
b3470b5d | 2325 | num_bytes = cache->length - cache->reserved - |
bf38be65 | 2326 | cache->pinned - cache->bytes_super - cache->used; |
26ce2095 JB |
2327 | sinfo->bytes_readonly -= num_bytes; |
2328 | list_del_init(&cache->ro_list); | |
2329 | } | |
2330 | spin_unlock(&cache->lock); | |
2331 | spin_unlock(&sinfo->lock); | |
2332 | } | |
77745c05 | 2333 | |
3be4d8ef QW |
2334 | static int update_block_group_item(struct btrfs_trans_handle *trans, |
2335 | struct btrfs_path *path, | |
2336 | struct btrfs_block_group *cache) | |
77745c05 JB |
2337 | { |
2338 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2339 | int ret; | |
3be4d8ef | 2340 | struct btrfs_root *root = fs_info->extent_root; |
77745c05 JB |
2341 | unsigned long bi; |
2342 | struct extent_buffer *leaf; | |
bf38be65 | 2343 | struct btrfs_block_group_item bgi; |
b3470b5d DS |
2344 | struct btrfs_key key; |
2345 | ||
2346 | key.objectid = cache->start; | |
2347 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
2348 | key.offset = cache->length; | |
77745c05 | 2349 | |
3be4d8ef | 2350 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
77745c05 JB |
2351 | if (ret) { |
2352 | if (ret > 0) | |
2353 | ret = -ENOENT; | |
2354 | goto fail; | |
2355 | } | |
2356 | ||
2357 | leaf = path->nodes[0]; | |
2358 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
de0dc456 DS |
2359 | btrfs_set_stack_block_group_used(&bgi, cache->used); |
2360 | btrfs_set_stack_block_group_chunk_objectid(&bgi, | |
3d976388 | 2361 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); |
de0dc456 | 2362 | btrfs_set_stack_block_group_flags(&bgi, cache->flags); |
bf38be65 | 2363 | write_extent_buffer(leaf, &bgi, bi, sizeof(bgi)); |
77745c05 JB |
2364 | btrfs_mark_buffer_dirty(leaf); |
2365 | fail: | |
2366 | btrfs_release_path(path); | |
2367 | return ret; | |
2368 | ||
2369 | } | |
2370 | ||
32da5386 | 2371 | static int cache_save_setup(struct btrfs_block_group *block_group, |
77745c05 JB |
2372 | struct btrfs_trans_handle *trans, |
2373 | struct btrfs_path *path) | |
2374 | { | |
2375 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
2376 | struct btrfs_root *root = fs_info->tree_root; | |
2377 | struct inode *inode = NULL; | |
2378 | struct extent_changeset *data_reserved = NULL; | |
2379 | u64 alloc_hint = 0; | |
2380 | int dcs = BTRFS_DC_ERROR; | |
2381 | u64 num_pages = 0; | |
2382 | int retries = 0; | |
2383 | int ret = 0; | |
2384 | ||
2385 | /* | |
2386 | * If this block group is smaller than 100 megs don't bother caching the | |
2387 | * block group. | |
2388 | */ | |
b3470b5d | 2389 | if (block_group->length < (100 * SZ_1M)) { |
77745c05 JB |
2390 | spin_lock(&block_group->lock); |
2391 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2392 | spin_unlock(&block_group->lock); | |
2393 | return 0; | |
2394 | } | |
2395 | ||
bf31f87f | 2396 | if (TRANS_ABORTED(trans)) |
77745c05 JB |
2397 | return 0; |
2398 | again: | |
2399 | inode = lookup_free_space_inode(block_group, path); | |
2400 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2401 | ret = PTR_ERR(inode); | |
2402 | btrfs_release_path(path); | |
2403 | goto out; | |
2404 | } | |
2405 | ||
2406 | if (IS_ERR(inode)) { | |
2407 | BUG_ON(retries); | |
2408 | retries++; | |
2409 | ||
2410 | if (block_group->ro) | |
2411 | goto out_free; | |
2412 | ||
2413 | ret = create_free_space_inode(trans, block_group, path); | |
2414 | if (ret) | |
2415 | goto out_free; | |
2416 | goto again; | |
2417 | } | |
2418 | ||
2419 | /* | |
2420 | * We want to set the generation to 0, that way if anything goes wrong | |
2421 | * from here on out we know not to trust this cache when we load up next | |
2422 | * time. | |
2423 | */ | |
2424 | BTRFS_I(inode)->generation = 0; | |
2425 | ret = btrfs_update_inode(trans, root, inode); | |
2426 | if (ret) { | |
2427 | /* | |
2428 | * So theoretically we could recover from this, simply set the | |
2429 | * super cache generation to 0 so we know to invalidate the | |
2430 | * cache, but then we'd have to keep track of the block groups | |
2431 | * that fail this way so we know we _have_ to reset this cache | |
2432 | * before the next commit or risk reading stale cache. So to | |
2433 | * limit our exposure to horrible edge cases lets just abort the | |
2434 | * transaction, this only happens in really bad situations | |
2435 | * anyway. | |
2436 | */ | |
2437 | btrfs_abort_transaction(trans, ret); | |
2438 | goto out_put; | |
2439 | } | |
2440 | WARN_ON(ret); | |
2441 | ||
2442 | /* We've already setup this transaction, go ahead and exit */ | |
2443 | if (block_group->cache_generation == trans->transid && | |
2444 | i_size_read(inode)) { | |
2445 | dcs = BTRFS_DC_SETUP; | |
2446 | goto out_put; | |
2447 | } | |
2448 | ||
2449 | if (i_size_read(inode) > 0) { | |
2450 | ret = btrfs_check_trunc_cache_free_space(fs_info, | |
2451 | &fs_info->global_block_rsv); | |
2452 | if (ret) | |
2453 | goto out_put; | |
2454 | ||
2455 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); | |
2456 | if (ret) | |
2457 | goto out_put; | |
2458 | } | |
2459 | ||
2460 | spin_lock(&block_group->lock); | |
2461 | if (block_group->cached != BTRFS_CACHE_FINISHED || | |
2462 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { | |
2463 | /* | |
2464 | * don't bother trying to write stuff out _if_ | |
2465 | * a) we're not cached, | |
2466 | * b) we're with nospace_cache mount option, | |
2467 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
2468 | */ | |
2469 | dcs = BTRFS_DC_WRITTEN; | |
2470 | spin_unlock(&block_group->lock); | |
2471 | goto out_put; | |
2472 | } | |
2473 | spin_unlock(&block_group->lock); | |
2474 | ||
2475 | /* | |
2476 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
2477 | * skip doing the setup, we've already cleared the cache so we're safe. | |
2478 | */ | |
2479 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
2480 | ret = -ENOSPC; | |
2481 | goto out_put; | |
2482 | } | |
2483 | ||
2484 | /* | |
2485 | * Try to preallocate enough space based on how big the block group is. | |
2486 | * Keep in mind this has to include any pinned space which could end up | |
2487 | * taking up quite a bit since it's not folded into the other space | |
2488 | * cache. | |
2489 | */ | |
b3470b5d | 2490 | num_pages = div_u64(block_group->length, SZ_256M); |
77745c05 JB |
2491 | if (!num_pages) |
2492 | num_pages = 1; | |
2493 | ||
2494 | num_pages *= 16; | |
2495 | num_pages *= PAGE_SIZE; | |
2496 | ||
36ea6f3e NB |
2497 | ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved, 0, |
2498 | num_pages); | |
77745c05 JB |
2499 | if (ret) |
2500 | goto out_put; | |
2501 | ||
2502 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
2503 | num_pages, num_pages, | |
2504 | &alloc_hint); | |
2505 | /* | |
2506 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
2507 | * of metadata or split extents when writing the cache out, which means | |
2508 | * we can enospc if we are heavily fragmented in addition to just normal | |
2509 | * out of space conditions. So if we hit this just skip setting up any | |
2510 | * other block groups for this transaction, maybe we'll unpin enough | |
2511 | * space the next time around. | |
2512 | */ | |
2513 | if (!ret) | |
2514 | dcs = BTRFS_DC_SETUP; | |
2515 | else if (ret == -ENOSPC) | |
2516 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
2517 | ||
2518 | out_put: | |
2519 | iput(inode); | |
2520 | out_free: | |
2521 | btrfs_release_path(path); | |
2522 | out: | |
2523 | spin_lock(&block_group->lock); | |
2524 | if (!ret && dcs == BTRFS_DC_SETUP) | |
2525 | block_group->cache_generation = trans->transid; | |
2526 | block_group->disk_cache_state = dcs; | |
2527 | spin_unlock(&block_group->lock); | |
2528 | ||
2529 | extent_changeset_free(data_reserved); | |
2530 | return ret; | |
2531 | } | |
2532 | ||
2533 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) | |
2534 | { | |
2535 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2536 | struct btrfs_block_group *cache, *tmp; |
77745c05 JB |
2537 | struct btrfs_transaction *cur_trans = trans->transaction; |
2538 | struct btrfs_path *path; | |
2539 | ||
2540 | if (list_empty(&cur_trans->dirty_bgs) || | |
2541 | !btrfs_test_opt(fs_info, SPACE_CACHE)) | |
2542 | return 0; | |
2543 | ||
2544 | path = btrfs_alloc_path(); | |
2545 | if (!path) | |
2546 | return -ENOMEM; | |
2547 | ||
2548 | /* Could add new block groups, use _safe just in case */ | |
2549 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
2550 | dirty_list) { | |
2551 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
2552 | cache_save_setup(cache, trans, path); | |
2553 | } | |
2554 | ||
2555 | btrfs_free_path(path); | |
2556 | return 0; | |
2557 | } | |
2558 | ||
2559 | /* | |
2560 | * Transaction commit does final block group cache writeback during a critical | |
2561 | * section where nothing is allowed to change the FS. This is required in | |
2562 | * order for the cache to actually match the block group, but can introduce a | |
2563 | * lot of latency into the commit. | |
2564 | * | |
2565 | * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO. | |
2566 | * There's a chance we'll have to redo some of it if the block group changes | |
2567 | * again during the commit, but it greatly reduces the commit latency by | |
2568 | * getting rid of the easy block groups while we're still allowing others to | |
2569 | * join the commit. | |
2570 | */ | |
2571 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) | |
2572 | { | |
2573 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2574 | struct btrfs_block_group *cache; |
77745c05 JB |
2575 | struct btrfs_transaction *cur_trans = trans->transaction; |
2576 | int ret = 0; | |
2577 | int should_put; | |
2578 | struct btrfs_path *path = NULL; | |
2579 | LIST_HEAD(dirty); | |
2580 | struct list_head *io = &cur_trans->io_bgs; | |
2581 | int num_started = 0; | |
2582 | int loops = 0; | |
2583 | ||
2584 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2585 | if (list_empty(&cur_trans->dirty_bgs)) { | |
2586 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2587 | return 0; | |
2588 | } | |
2589 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
2590 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2591 | ||
2592 | again: | |
2593 | /* Make sure all the block groups on our dirty list actually exist */ | |
2594 | btrfs_create_pending_block_groups(trans); | |
2595 | ||
2596 | if (!path) { | |
2597 | path = btrfs_alloc_path(); | |
2598 | if (!path) | |
2599 | return -ENOMEM; | |
2600 | } | |
2601 | ||
2602 | /* | |
2603 | * cache_write_mutex is here only to save us from balance or automatic | |
2604 | * removal of empty block groups deleting this block group while we are | |
2605 | * writing out the cache | |
2606 | */ | |
2607 | mutex_lock(&trans->transaction->cache_write_mutex); | |
2608 | while (!list_empty(&dirty)) { | |
2609 | bool drop_reserve = true; | |
2610 | ||
32da5386 | 2611 | cache = list_first_entry(&dirty, struct btrfs_block_group, |
77745c05 JB |
2612 | dirty_list); |
2613 | /* | |
2614 | * This can happen if something re-dirties a block group that | |
2615 | * is already under IO. Just wait for it to finish and then do | |
2616 | * it all again | |
2617 | */ | |
2618 | if (!list_empty(&cache->io_list)) { | |
2619 | list_del_init(&cache->io_list); | |
2620 | btrfs_wait_cache_io(trans, cache, path); | |
2621 | btrfs_put_block_group(cache); | |
2622 | } | |
2623 | ||
2624 | ||
2625 | /* | |
2626 | * btrfs_wait_cache_io uses the cache->dirty_list to decide if | |
2627 | * it should update the cache_state. Don't delete until after | |
2628 | * we wait. | |
2629 | * | |
2630 | * Since we're not running in the commit critical section | |
2631 | * we need the dirty_bgs_lock to protect from update_block_group | |
2632 | */ | |
2633 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2634 | list_del_init(&cache->dirty_list); | |
2635 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2636 | ||
2637 | should_put = 1; | |
2638 | ||
2639 | cache_save_setup(cache, trans, path); | |
2640 | ||
2641 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
2642 | cache->io_ctl.inode = NULL; | |
2643 | ret = btrfs_write_out_cache(trans, cache, path); | |
2644 | if (ret == 0 && cache->io_ctl.inode) { | |
2645 | num_started++; | |
2646 | should_put = 0; | |
2647 | ||
2648 | /* | |
2649 | * The cache_write_mutex is protecting the | |
2650 | * io_list, also refer to the definition of | |
2651 | * btrfs_transaction::io_bgs for more details | |
2652 | */ | |
2653 | list_add_tail(&cache->io_list, io); | |
2654 | } else { | |
2655 | /* | |
2656 | * If we failed to write the cache, the | |
2657 | * generation will be bad and life goes on | |
2658 | */ | |
2659 | ret = 0; | |
2660 | } | |
2661 | } | |
2662 | if (!ret) { | |
3be4d8ef | 2663 | ret = update_block_group_item(trans, path, cache); |
77745c05 JB |
2664 | /* |
2665 | * Our block group might still be attached to the list | |
2666 | * of new block groups in the transaction handle of some | |
2667 | * other task (struct btrfs_trans_handle->new_bgs). This | |
2668 | * means its block group item isn't yet in the extent | |
2669 | * tree. If this happens ignore the error, as we will | |
2670 | * try again later in the critical section of the | |
2671 | * transaction commit. | |
2672 | */ | |
2673 | if (ret == -ENOENT) { | |
2674 | ret = 0; | |
2675 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2676 | if (list_empty(&cache->dirty_list)) { | |
2677 | list_add_tail(&cache->dirty_list, | |
2678 | &cur_trans->dirty_bgs); | |
2679 | btrfs_get_block_group(cache); | |
2680 | drop_reserve = false; | |
2681 | } | |
2682 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2683 | } else if (ret) { | |
2684 | btrfs_abort_transaction(trans, ret); | |
2685 | } | |
2686 | } | |
2687 | ||
2688 | /* If it's not on the io list, we need to put the block group */ | |
2689 | if (should_put) | |
2690 | btrfs_put_block_group(cache); | |
2691 | if (drop_reserve) | |
2692 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
2693 | ||
2694 | if (ret) | |
2695 | break; | |
2696 | ||
2697 | /* | |
2698 | * Avoid blocking other tasks for too long. It might even save | |
2699 | * us from writing caches for block groups that are going to be | |
2700 | * removed. | |
2701 | */ | |
2702 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
2703 | mutex_lock(&trans->transaction->cache_write_mutex); | |
2704 | } | |
2705 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
2706 | ||
2707 | /* | |
2708 | * Go through delayed refs for all the stuff we've just kicked off | |
2709 | * and then loop back (just once) | |
2710 | */ | |
2711 | ret = btrfs_run_delayed_refs(trans, 0); | |
2712 | if (!ret && loops == 0) { | |
2713 | loops++; | |
2714 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2715 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
2716 | /* | |
2717 | * dirty_bgs_lock protects us from concurrent block group | |
2718 | * deletes too (not just cache_write_mutex). | |
2719 | */ | |
2720 | if (!list_empty(&dirty)) { | |
2721 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2722 | goto again; | |
2723 | } | |
2724 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2725 | } else if (ret < 0) { | |
2726 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); | |
2727 | } | |
2728 | ||
2729 | btrfs_free_path(path); | |
2730 | return ret; | |
2731 | } | |
2732 | ||
2733 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) | |
2734 | { | |
2735 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2736 | struct btrfs_block_group *cache; |
77745c05 JB |
2737 | struct btrfs_transaction *cur_trans = trans->transaction; |
2738 | int ret = 0; | |
2739 | int should_put; | |
2740 | struct btrfs_path *path; | |
2741 | struct list_head *io = &cur_trans->io_bgs; | |
2742 | int num_started = 0; | |
2743 | ||
2744 | path = btrfs_alloc_path(); | |
2745 | if (!path) | |
2746 | return -ENOMEM; | |
2747 | ||
2748 | /* | |
2749 | * Even though we are in the critical section of the transaction commit, | |
2750 | * we can still have concurrent tasks adding elements to this | |
2751 | * transaction's list of dirty block groups. These tasks correspond to | |
2752 | * endio free space workers started when writeback finishes for a | |
2753 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
2754 | * allocate new block groups as a result of COWing nodes of the root | |
2755 | * tree when updating the free space inode. The writeback for the space | |
2756 | * caches is triggered by an earlier call to | |
2757 | * btrfs_start_dirty_block_groups() and iterations of the following | |
2758 | * loop. | |
2759 | * Also we want to do the cache_save_setup first and then run the | |
2760 | * delayed refs to make sure we have the best chance at doing this all | |
2761 | * in one shot. | |
2762 | */ | |
2763 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2764 | while (!list_empty(&cur_trans->dirty_bgs)) { | |
2765 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
32da5386 | 2766 | struct btrfs_block_group, |
77745c05 JB |
2767 | dirty_list); |
2768 | ||
2769 | /* | |
2770 | * This can happen if cache_save_setup re-dirties a block group | |
2771 | * that is already under IO. Just wait for it to finish and | |
2772 | * then do it all again | |
2773 | */ | |
2774 | if (!list_empty(&cache->io_list)) { | |
2775 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2776 | list_del_init(&cache->io_list); | |
2777 | btrfs_wait_cache_io(trans, cache, path); | |
2778 | btrfs_put_block_group(cache); | |
2779 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2780 | } | |
2781 | ||
2782 | /* | |
2783 | * Don't remove from the dirty list until after we've waited on | |
2784 | * any pending IO | |
2785 | */ | |
2786 | list_del_init(&cache->dirty_list); | |
2787 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2788 | should_put = 1; | |
2789 | ||
2790 | cache_save_setup(cache, trans, path); | |
2791 | ||
2792 | if (!ret) | |
2793 | ret = btrfs_run_delayed_refs(trans, | |
2794 | (unsigned long) -1); | |
2795 | ||
2796 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
2797 | cache->io_ctl.inode = NULL; | |
2798 | ret = btrfs_write_out_cache(trans, cache, path); | |
2799 | if (ret == 0 && cache->io_ctl.inode) { | |
2800 | num_started++; | |
2801 | should_put = 0; | |
2802 | list_add_tail(&cache->io_list, io); | |
2803 | } else { | |
2804 | /* | |
2805 | * If we failed to write the cache, the | |
2806 | * generation will be bad and life goes on | |
2807 | */ | |
2808 | ret = 0; | |
2809 | } | |
2810 | } | |
2811 | if (!ret) { | |
3be4d8ef | 2812 | ret = update_block_group_item(trans, path, cache); |
77745c05 JB |
2813 | /* |
2814 | * One of the free space endio workers might have | |
2815 | * created a new block group while updating a free space | |
2816 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
2817 | * and hasn't released its transaction handle yet, in | |
2818 | * which case the new block group is still attached to | |
2819 | * its transaction handle and its creation has not | |
2820 | * finished yet (no block group item in the extent tree | |
2821 | * yet, etc). If this is the case, wait for all free | |
2822 | * space endio workers to finish and retry. This is a | |
260db43c | 2823 | * very rare case so no need for a more efficient and |
77745c05 JB |
2824 | * complex approach. |
2825 | */ | |
2826 | if (ret == -ENOENT) { | |
2827 | wait_event(cur_trans->writer_wait, | |
2828 | atomic_read(&cur_trans->num_writers) == 1); | |
3be4d8ef | 2829 | ret = update_block_group_item(trans, path, cache); |
77745c05 JB |
2830 | } |
2831 | if (ret) | |
2832 | btrfs_abort_transaction(trans, ret); | |
2833 | } | |
2834 | ||
2835 | /* If its not on the io list, we need to put the block group */ | |
2836 | if (should_put) | |
2837 | btrfs_put_block_group(cache); | |
2838 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
2839 | spin_lock(&cur_trans->dirty_bgs_lock); | |
2840 | } | |
2841 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
2842 | ||
2843 | /* | |
2844 | * Refer to the definition of io_bgs member for details why it's safe | |
2845 | * to use it without any locking | |
2846 | */ | |
2847 | while (!list_empty(io)) { | |
32da5386 | 2848 | cache = list_first_entry(io, struct btrfs_block_group, |
77745c05 JB |
2849 | io_list); |
2850 | list_del_init(&cache->io_list); | |
2851 | btrfs_wait_cache_io(trans, cache, path); | |
2852 | btrfs_put_block_group(cache); | |
2853 | } | |
2854 | ||
2855 | btrfs_free_path(path); | |
2856 | return ret; | |
2857 | } | |
606d1bf1 JB |
2858 | |
2859 | int btrfs_update_block_group(struct btrfs_trans_handle *trans, | |
2860 | u64 bytenr, u64 num_bytes, int alloc) | |
2861 | { | |
2862 | struct btrfs_fs_info *info = trans->fs_info; | |
32da5386 | 2863 | struct btrfs_block_group *cache = NULL; |
606d1bf1 JB |
2864 | u64 total = num_bytes; |
2865 | u64 old_val; | |
2866 | u64 byte_in_group; | |
2867 | int factor; | |
2868 | int ret = 0; | |
2869 | ||
2870 | /* Block accounting for super block */ | |
2871 | spin_lock(&info->delalloc_root_lock); | |
2872 | old_val = btrfs_super_bytes_used(info->super_copy); | |
2873 | if (alloc) | |
2874 | old_val += num_bytes; | |
2875 | else | |
2876 | old_val -= num_bytes; | |
2877 | btrfs_set_super_bytes_used(info->super_copy, old_val); | |
2878 | spin_unlock(&info->delalloc_root_lock); | |
2879 | ||
2880 | while (total) { | |
2881 | cache = btrfs_lookup_block_group(info, bytenr); | |
2882 | if (!cache) { | |
2883 | ret = -ENOENT; | |
2884 | break; | |
2885 | } | |
2886 | factor = btrfs_bg_type_to_factor(cache->flags); | |
2887 | ||
2888 | /* | |
2889 | * If this block group has free space cache written out, we | |
2890 | * need to make sure to load it if we are removing space. This | |
2891 | * is because we need the unpinning stage to actually add the | |
2892 | * space back to the block group, otherwise we will leak space. | |
2893 | */ | |
32da5386 | 2894 | if (!alloc && !btrfs_block_group_done(cache)) |
606d1bf1 JB |
2895 | btrfs_cache_block_group(cache, 1); |
2896 | ||
b3470b5d DS |
2897 | byte_in_group = bytenr - cache->start; |
2898 | WARN_ON(byte_in_group > cache->length); | |
606d1bf1 JB |
2899 | |
2900 | spin_lock(&cache->space_info->lock); | |
2901 | spin_lock(&cache->lock); | |
2902 | ||
2903 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
2904 | cache->disk_cache_state < BTRFS_DC_CLEAR) | |
2905 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
2906 | ||
bf38be65 | 2907 | old_val = cache->used; |
b3470b5d | 2908 | num_bytes = min(total, cache->length - byte_in_group); |
606d1bf1 JB |
2909 | if (alloc) { |
2910 | old_val += num_bytes; | |
bf38be65 | 2911 | cache->used = old_val; |
606d1bf1 JB |
2912 | cache->reserved -= num_bytes; |
2913 | cache->space_info->bytes_reserved -= num_bytes; | |
2914 | cache->space_info->bytes_used += num_bytes; | |
2915 | cache->space_info->disk_used += num_bytes * factor; | |
2916 | spin_unlock(&cache->lock); | |
2917 | spin_unlock(&cache->space_info->lock); | |
2918 | } else { | |
2919 | old_val -= num_bytes; | |
bf38be65 | 2920 | cache->used = old_val; |
606d1bf1 JB |
2921 | cache->pinned += num_bytes; |
2922 | btrfs_space_info_update_bytes_pinned(info, | |
2923 | cache->space_info, num_bytes); | |
2924 | cache->space_info->bytes_used -= num_bytes; | |
2925 | cache->space_info->disk_used -= num_bytes * factor; | |
2926 | spin_unlock(&cache->lock); | |
2927 | spin_unlock(&cache->space_info->lock); | |
2928 | ||
606d1bf1 JB |
2929 | percpu_counter_add_batch( |
2930 | &cache->space_info->total_bytes_pinned, | |
2931 | num_bytes, | |
2932 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
fe119a6e | 2933 | set_extent_dirty(&trans->transaction->pinned_extents, |
606d1bf1 JB |
2934 | bytenr, bytenr + num_bytes - 1, |
2935 | GFP_NOFS | __GFP_NOFAIL); | |
2936 | } | |
2937 | ||
2938 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
2939 | if (list_empty(&cache->dirty_list)) { | |
2940 | list_add_tail(&cache->dirty_list, | |
2941 | &trans->transaction->dirty_bgs); | |
2942 | trans->delayed_ref_updates++; | |
2943 | btrfs_get_block_group(cache); | |
2944 | } | |
2945 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
2946 | ||
2947 | /* | |
2948 | * No longer have used bytes in this block group, queue it for | |
2949 | * deletion. We do this after adding the block group to the | |
2950 | * dirty list to avoid races between cleaner kthread and space | |
2951 | * cache writeout. | |
2952 | */ | |
6e80d4f8 DZ |
2953 | if (!alloc && old_val == 0) { |
2954 | if (!btrfs_test_opt(info, DISCARD_ASYNC)) | |
2955 | btrfs_mark_bg_unused(cache); | |
2956 | } | |
606d1bf1 JB |
2957 | |
2958 | btrfs_put_block_group(cache); | |
2959 | total -= num_bytes; | |
2960 | bytenr += num_bytes; | |
2961 | } | |
2962 | ||
2963 | /* Modified block groups are accounted for in the delayed_refs_rsv. */ | |
2964 | btrfs_update_delayed_refs_rsv(trans); | |
2965 | return ret; | |
2966 | } | |
2967 | ||
2968 | /** | |
2969 | * btrfs_add_reserved_bytes - update the block_group and space info counters | |
2970 | * @cache: The cache we are manipulating | |
2971 | * @ram_bytes: The number of bytes of file content, and will be same to | |
2972 | * @num_bytes except for the compress path. | |
2973 | * @num_bytes: The number of bytes in question | |
2974 | * @delalloc: The blocks are allocated for the delalloc write | |
2975 | * | |
2976 | * This is called by the allocator when it reserves space. If this is a | |
2977 | * reservation and the block group has become read only we cannot make the | |
2978 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
2979 | */ | |
32da5386 | 2980 | int btrfs_add_reserved_bytes(struct btrfs_block_group *cache, |
606d1bf1 JB |
2981 | u64 ram_bytes, u64 num_bytes, int delalloc) |
2982 | { | |
2983 | struct btrfs_space_info *space_info = cache->space_info; | |
2984 | int ret = 0; | |
2985 | ||
2986 | spin_lock(&space_info->lock); | |
2987 | spin_lock(&cache->lock); | |
2988 | if (cache->ro) { | |
2989 | ret = -EAGAIN; | |
2990 | } else { | |
2991 | cache->reserved += num_bytes; | |
2992 | space_info->bytes_reserved += num_bytes; | |
a43c3835 JB |
2993 | trace_btrfs_space_reservation(cache->fs_info, "space_info", |
2994 | space_info->flags, num_bytes, 1); | |
606d1bf1 JB |
2995 | btrfs_space_info_update_bytes_may_use(cache->fs_info, |
2996 | space_info, -ram_bytes); | |
2997 | if (delalloc) | |
2998 | cache->delalloc_bytes += num_bytes; | |
99ffb43e JB |
2999 | |
3000 | /* | |
3001 | * Compression can use less space than we reserved, so wake | |
3002 | * tickets if that happens | |
3003 | */ | |
3004 | if (num_bytes < ram_bytes) | |
3005 | btrfs_try_granting_tickets(cache->fs_info, space_info); | |
606d1bf1 JB |
3006 | } |
3007 | spin_unlock(&cache->lock); | |
3008 | spin_unlock(&space_info->lock); | |
3009 | return ret; | |
3010 | } | |
3011 | ||
3012 | /** | |
3013 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
3014 | * @cache: The cache we are manipulating | |
3015 | * @num_bytes: The number of bytes in question | |
3016 | * @delalloc: The blocks are allocated for the delalloc write | |
3017 | * | |
3018 | * This is called by somebody who is freeing space that was never actually used | |
3019 | * on disk. For example if you reserve some space for a new leaf in transaction | |
3020 | * A and before transaction A commits you free that leaf, you call this with | |
3021 | * reserve set to 0 in order to clear the reservation. | |
3022 | */ | |
32da5386 | 3023 | void btrfs_free_reserved_bytes(struct btrfs_block_group *cache, |
606d1bf1 JB |
3024 | u64 num_bytes, int delalloc) |
3025 | { | |
3026 | struct btrfs_space_info *space_info = cache->space_info; | |
3027 | ||
3028 | spin_lock(&space_info->lock); | |
3029 | spin_lock(&cache->lock); | |
3030 | if (cache->ro) | |
3031 | space_info->bytes_readonly += num_bytes; | |
3032 | cache->reserved -= num_bytes; | |
3033 | space_info->bytes_reserved -= num_bytes; | |
3034 | space_info->max_extent_size = 0; | |
3035 | ||
3036 | if (delalloc) | |
3037 | cache->delalloc_bytes -= num_bytes; | |
3038 | spin_unlock(&cache->lock); | |
3308234a JB |
3039 | |
3040 | btrfs_try_granting_tickets(cache->fs_info, space_info); | |
606d1bf1 JB |
3041 | spin_unlock(&space_info->lock); |
3042 | } | |
07730d87 JB |
3043 | |
3044 | static void force_metadata_allocation(struct btrfs_fs_info *info) | |
3045 | { | |
3046 | struct list_head *head = &info->space_info; | |
3047 | struct btrfs_space_info *found; | |
3048 | ||
72804905 | 3049 | list_for_each_entry(found, head, list) { |
07730d87 JB |
3050 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) |
3051 | found->force_alloc = CHUNK_ALLOC_FORCE; | |
3052 | } | |
07730d87 JB |
3053 | } |
3054 | ||
3055 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, | |
3056 | struct btrfs_space_info *sinfo, int force) | |
3057 | { | |
3058 | u64 bytes_used = btrfs_space_info_used(sinfo, false); | |
3059 | u64 thresh; | |
3060 | ||
3061 | if (force == CHUNK_ALLOC_FORCE) | |
3062 | return 1; | |
3063 | ||
3064 | /* | |
3065 | * in limited mode, we want to have some free space up to | |
3066 | * about 1% of the FS size. | |
3067 | */ | |
3068 | if (force == CHUNK_ALLOC_LIMITED) { | |
3069 | thresh = btrfs_super_total_bytes(fs_info->super_copy); | |
3070 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); | |
3071 | ||
3072 | if (sinfo->total_bytes - bytes_used < thresh) | |
3073 | return 1; | |
3074 | } | |
3075 | ||
3076 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) | |
3077 | return 0; | |
3078 | return 1; | |
3079 | } | |
3080 | ||
3081 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) | |
3082 | { | |
3083 | u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type); | |
3084 | ||
3085 | return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); | |
3086 | } | |
3087 | ||
3088 | /* | |
3089 | * If force is CHUNK_ALLOC_FORCE: | |
3090 | * - return 1 if it successfully allocates a chunk, | |
3091 | * - return errors including -ENOSPC otherwise. | |
3092 | * If force is NOT CHUNK_ALLOC_FORCE: | |
3093 | * - return 0 if it doesn't need to allocate a new chunk, | |
3094 | * - return 1 if it successfully allocates a chunk, | |
3095 | * - return errors including -ENOSPC otherwise. | |
3096 | */ | |
3097 | int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, | |
3098 | enum btrfs_chunk_alloc_enum force) | |
3099 | { | |
3100 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3101 | struct btrfs_space_info *space_info; | |
3102 | bool wait_for_alloc = false; | |
3103 | bool should_alloc = false; | |
3104 | int ret = 0; | |
3105 | ||
3106 | /* Don't re-enter if we're already allocating a chunk */ | |
3107 | if (trans->allocating_chunk) | |
3108 | return -ENOSPC; | |
3109 | ||
3110 | space_info = btrfs_find_space_info(fs_info, flags); | |
3111 | ASSERT(space_info); | |
3112 | ||
3113 | do { | |
3114 | spin_lock(&space_info->lock); | |
3115 | if (force < space_info->force_alloc) | |
3116 | force = space_info->force_alloc; | |
3117 | should_alloc = should_alloc_chunk(fs_info, space_info, force); | |
3118 | if (space_info->full) { | |
3119 | /* No more free physical space */ | |
3120 | if (should_alloc) | |
3121 | ret = -ENOSPC; | |
3122 | else | |
3123 | ret = 0; | |
3124 | spin_unlock(&space_info->lock); | |
3125 | return ret; | |
3126 | } else if (!should_alloc) { | |
3127 | spin_unlock(&space_info->lock); | |
3128 | return 0; | |
3129 | } else if (space_info->chunk_alloc) { | |
3130 | /* | |
3131 | * Someone is already allocating, so we need to block | |
3132 | * until this someone is finished and then loop to | |
3133 | * recheck if we should continue with our allocation | |
3134 | * attempt. | |
3135 | */ | |
3136 | wait_for_alloc = true; | |
3137 | spin_unlock(&space_info->lock); | |
3138 | mutex_lock(&fs_info->chunk_mutex); | |
3139 | mutex_unlock(&fs_info->chunk_mutex); | |
3140 | } else { | |
3141 | /* Proceed with allocation */ | |
3142 | space_info->chunk_alloc = 1; | |
3143 | wait_for_alloc = false; | |
3144 | spin_unlock(&space_info->lock); | |
3145 | } | |
3146 | ||
3147 | cond_resched(); | |
3148 | } while (wait_for_alloc); | |
3149 | ||
3150 | mutex_lock(&fs_info->chunk_mutex); | |
3151 | trans->allocating_chunk = true; | |
3152 | ||
3153 | /* | |
3154 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3155 | * allocating mixed chunks instead of individual chunks. | |
3156 | */ | |
3157 | if (btrfs_mixed_space_info(space_info)) | |
3158 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3159 | ||
3160 | /* | |
3161 | * if we're doing a data chunk, go ahead and make sure that | |
3162 | * we keep a reasonable number of metadata chunks allocated in the | |
3163 | * FS as well. | |
3164 | */ | |
3165 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { | |
3166 | fs_info->data_chunk_allocations++; | |
3167 | if (!(fs_info->data_chunk_allocations % | |
3168 | fs_info->metadata_ratio)) | |
3169 | force_metadata_allocation(fs_info); | |
3170 | } | |
3171 | ||
3172 | /* | |
3173 | * Check if we have enough space in SYSTEM chunk because we may need | |
3174 | * to update devices. | |
3175 | */ | |
3176 | check_system_chunk(trans, flags); | |
3177 | ||
3178 | ret = btrfs_alloc_chunk(trans, flags); | |
3179 | trans->allocating_chunk = false; | |
3180 | ||
3181 | spin_lock(&space_info->lock); | |
3182 | if (ret < 0) { | |
3183 | if (ret == -ENOSPC) | |
3184 | space_info->full = 1; | |
3185 | else | |
3186 | goto out; | |
3187 | } else { | |
3188 | ret = 1; | |
3189 | space_info->max_extent_size = 0; | |
3190 | } | |
3191 | ||
3192 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
3193 | out: | |
3194 | space_info->chunk_alloc = 0; | |
3195 | spin_unlock(&space_info->lock); | |
3196 | mutex_unlock(&fs_info->chunk_mutex); | |
3197 | /* | |
3198 | * When we allocate a new chunk we reserve space in the chunk block | |
3199 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
3200 | * add new nodes/leafs to it if we end up needing to do it when | |
3201 | * inserting the chunk item and updating device items as part of the | |
3202 | * second phase of chunk allocation, performed by | |
3203 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
3204 | * large number of new block groups to create in our transaction | |
3205 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
3206 | * in extreme cases - like having a single transaction create many new | |
3207 | * block groups when starting to write out the free space caches of all | |
3208 | * the block groups that were made dirty during the lifetime of the | |
3209 | * transaction. | |
3210 | */ | |
3211 | if (trans->chunk_bytes_reserved >= (u64)SZ_2M) | |
3212 | btrfs_create_pending_block_groups(trans); | |
3213 | ||
3214 | return ret; | |
3215 | } | |
3216 | ||
3217 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) | |
3218 | { | |
3219 | u64 num_dev; | |
3220 | ||
3221 | num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max; | |
3222 | if (!num_dev) | |
3223 | num_dev = fs_info->fs_devices->rw_devices; | |
3224 | ||
3225 | return num_dev; | |
3226 | } | |
3227 | ||
3228 | /* | |
a9143bd3 | 3229 | * Reserve space in the system space for allocating or removing a chunk |
07730d87 JB |
3230 | */ |
3231 | void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) | |
3232 | { | |
3233 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3234 | struct btrfs_space_info *info; | |
3235 | u64 left; | |
3236 | u64 thresh; | |
3237 | int ret = 0; | |
3238 | u64 num_devs; | |
3239 | ||
3240 | /* | |
3241 | * Needed because we can end up allocating a system chunk and for an | |
3242 | * atomic and race free space reservation in the chunk block reserve. | |
3243 | */ | |
3244 | lockdep_assert_held(&fs_info->chunk_mutex); | |
3245 | ||
3246 | info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3247 | spin_lock(&info->lock); | |
3248 | left = info->total_bytes - btrfs_space_info_used(info, true); | |
3249 | spin_unlock(&info->lock); | |
3250 | ||
3251 | num_devs = get_profile_num_devs(fs_info, type); | |
3252 | ||
3253 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
2bd36e7b JB |
3254 | thresh = btrfs_calc_metadata_size(fs_info, num_devs) + |
3255 | btrfs_calc_insert_metadata_size(fs_info, 1); | |
07730d87 JB |
3256 | |
3257 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { | |
3258 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
3259 | left, thresh, type); | |
3260 | btrfs_dump_space_info(fs_info, info, 0, 0); | |
3261 | } | |
3262 | ||
3263 | if (left < thresh) { | |
3264 | u64 flags = btrfs_system_alloc_profile(fs_info); | |
3265 | ||
3266 | /* | |
3267 | * Ignore failure to create system chunk. We might end up not | |
3268 | * needing it, as we might not need to COW all nodes/leafs from | |
3269 | * the paths we visit in the chunk tree (they were already COWed | |
3270 | * or created in the current transaction for example). | |
3271 | */ | |
3272 | ret = btrfs_alloc_chunk(trans, flags); | |
3273 | } | |
3274 | ||
3275 | if (!ret) { | |
3276 | ret = btrfs_block_rsv_add(fs_info->chunk_root, | |
3277 | &fs_info->chunk_block_rsv, | |
3278 | thresh, BTRFS_RESERVE_NO_FLUSH); | |
3279 | if (!ret) | |
3280 | trans->chunk_bytes_reserved += thresh; | |
3281 | } | |
3282 | } | |
3283 | ||
3e43c279 JB |
3284 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
3285 | { | |
32da5386 | 3286 | struct btrfs_block_group *block_group; |
3e43c279 JB |
3287 | u64 last = 0; |
3288 | ||
3289 | while (1) { | |
3290 | struct inode *inode; | |
3291 | ||
3292 | block_group = btrfs_lookup_first_block_group(info, last); | |
3293 | while (block_group) { | |
3294 | btrfs_wait_block_group_cache_done(block_group); | |
3295 | spin_lock(&block_group->lock); | |
3296 | if (block_group->iref) | |
3297 | break; | |
3298 | spin_unlock(&block_group->lock); | |
3299 | block_group = btrfs_next_block_group(block_group); | |
3300 | } | |
3301 | if (!block_group) { | |
3302 | if (last == 0) | |
3303 | break; | |
3304 | last = 0; | |
3305 | continue; | |
3306 | } | |
3307 | ||
3308 | inode = block_group->inode; | |
3309 | block_group->iref = 0; | |
3310 | block_group->inode = NULL; | |
3311 | spin_unlock(&block_group->lock); | |
3312 | ASSERT(block_group->io_ctl.inode == NULL); | |
3313 | iput(inode); | |
b3470b5d | 3314 | last = block_group->start + block_group->length; |
3e43c279 JB |
3315 | btrfs_put_block_group(block_group); |
3316 | } | |
3317 | } | |
3318 | ||
3319 | /* | |
3320 | * Must be called only after stopping all workers, since we could have block | |
3321 | * group caching kthreads running, and therefore they could race with us if we | |
3322 | * freed the block groups before stopping them. | |
3323 | */ | |
3324 | int btrfs_free_block_groups(struct btrfs_fs_info *info) | |
3325 | { | |
32da5386 | 3326 | struct btrfs_block_group *block_group; |
3e43c279 JB |
3327 | struct btrfs_space_info *space_info; |
3328 | struct btrfs_caching_control *caching_ctl; | |
3329 | struct rb_node *n; | |
3330 | ||
3331 | down_write(&info->commit_root_sem); | |
3332 | while (!list_empty(&info->caching_block_groups)) { | |
3333 | caching_ctl = list_entry(info->caching_block_groups.next, | |
3334 | struct btrfs_caching_control, list); | |
3335 | list_del(&caching_ctl->list); | |
3336 | btrfs_put_caching_control(caching_ctl); | |
3337 | } | |
3338 | up_write(&info->commit_root_sem); | |
3339 | ||
3340 | spin_lock(&info->unused_bgs_lock); | |
3341 | while (!list_empty(&info->unused_bgs)) { | |
3342 | block_group = list_first_entry(&info->unused_bgs, | |
32da5386 | 3343 | struct btrfs_block_group, |
3e43c279 JB |
3344 | bg_list); |
3345 | list_del_init(&block_group->bg_list); | |
3346 | btrfs_put_block_group(block_group); | |
3347 | } | |
3348 | spin_unlock(&info->unused_bgs_lock); | |
3349 | ||
3350 | spin_lock(&info->block_group_cache_lock); | |
3351 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
32da5386 | 3352 | block_group = rb_entry(n, struct btrfs_block_group, |
3e43c279 JB |
3353 | cache_node); |
3354 | rb_erase(&block_group->cache_node, | |
3355 | &info->block_group_cache_tree); | |
3356 | RB_CLEAR_NODE(&block_group->cache_node); | |
3357 | spin_unlock(&info->block_group_cache_lock); | |
3358 | ||
3359 | down_write(&block_group->space_info->groups_sem); | |
3360 | list_del(&block_group->list); | |
3361 | up_write(&block_group->space_info->groups_sem); | |
3362 | ||
3363 | /* | |
3364 | * We haven't cached this block group, which means we could | |
3365 | * possibly have excluded extents on this block group. | |
3366 | */ | |
3367 | if (block_group->cached == BTRFS_CACHE_NO || | |
3368 | block_group->cached == BTRFS_CACHE_ERROR) | |
3369 | btrfs_free_excluded_extents(block_group); | |
3370 | ||
3371 | btrfs_remove_free_space_cache(block_group); | |
3372 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); | |
3373 | ASSERT(list_empty(&block_group->dirty_list)); | |
3374 | ASSERT(list_empty(&block_group->io_list)); | |
3375 | ASSERT(list_empty(&block_group->bg_list)); | |
48aaeebe | 3376 | ASSERT(refcount_read(&block_group->refs) == 1); |
3e43c279 JB |
3377 | btrfs_put_block_group(block_group); |
3378 | ||
3379 | spin_lock(&info->block_group_cache_lock); | |
3380 | } | |
3381 | spin_unlock(&info->block_group_cache_lock); | |
3382 | ||
3e43c279 JB |
3383 | btrfs_release_global_block_rsv(info); |
3384 | ||
3385 | while (!list_empty(&info->space_info)) { | |
3386 | space_info = list_entry(info->space_info.next, | |
3387 | struct btrfs_space_info, | |
3388 | list); | |
3389 | ||
3390 | /* | |
3391 | * Do not hide this behind enospc_debug, this is actually | |
3392 | * important and indicates a real bug if this happens. | |
3393 | */ | |
3394 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
3395 | space_info->bytes_reserved > 0 || | |
3396 | space_info->bytes_may_use > 0)) | |
3397 | btrfs_dump_space_info(info, space_info, 0, 0); | |
d611add4 | 3398 | WARN_ON(space_info->reclaim_size > 0); |
3e43c279 JB |
3399 | list_del(&space_info->list); |
3400 | btrfs_sysfs_remove_space_info(space_info); | |
3401 | } | |
3402 | return 0; | |
3403 | } | |
684b752b FM |
3404 | |
3405 | void btrfs_freeze_block_group(struct btrfs_block_group *cache) | |
3406 | { | |
3407 | atomic_inc(&cache->frozen); | |
3408 | } | |
3409 | ||
3410 | void btrfs_unfreeze_block_group(struct btrfs_block_group *block_group) | |
3411 | { | |
3412 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
3413 | struct extent_map_tree *em_tree; | |
3414 | struct extent_map *em; | |
3415 | bool cleanup; | |
3416 | ||
3417 | spin_lock(&block_group->lock); | |
3418 | cleanup = (atomic_dec_and_test(&block_group->frozen) && | |
3419 | block_group->removed); | |
3420 | spin_unlock(&block_group->lock); | |
3421 | ||
3422 | if (cleanup) { | |
684b752b FM |
3423 | em_tree = &fs_info->mapping_tree; |
3424 | write_lock(&em_tree->lock); | |
3425 | em = lookup_extent_mapping(em_tree, block_group->start, | |
3426 | 1); | |
3427 | BUG_ON(!em); /* logic error, can't happen */ | |
3428 | remove_extent_mapping(em_tree, em); | |
3429 | write_unlock(&em_tree->lock); | |
684b752b FM |
3430 | |
3431 | /* once for us and once for the tree */ | |
3432 | free_extent_map(em); | |
3433 | free_extent_map(em); | |
3434 | ||
3435 | /* | |
3436 | * We may have left one free space entry and other possible | |
3437 | * tasks trimming this block group have left 1 entry each one. | |
3438 | * Free them if any. | |
3439 | */ | |
3440 | __btrfs_remove_free_space_cache(block_group->free_space_ctl); | |
3441 | } | |
3442 | } |