Commit | Line | Data |
---|---|---|
2e405ad8 JB |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | ||
2ca0ec77 | 3 | #include <linux/list_sort.h> |
784352fe | 4 | #include "misc.h" |
2e405ad8 JB |
5 | #include "ctree.h" |
6 | #include "block-group.h" | |
3eeb3226 | 7 | #include "space-info.h" |
9f21246d JB |
8 | #include "disk-io.h" |
9 | #include "free-space-cache.h" | |
10 | #include "free-space-tree.h" | |
e3e0520b JB |
11 | #include "volumes.h" |
12 | #include "transaction.h" | |
13 | #include "ref-verify.h" | |
4358d963 JB |
14 | #include "sysfs.h" |
15 | #include "tree-log.h" | |
77745c05 | 16 | #include "delalloc-space.h" |
b0643e59 | 17 | #include "discard.h" |
96a14336 | 18 | #include "raid56.h" |
08e11a3d | 19 | #include "zoned.h" |
c7f13d42 | 20 | #include "fs.h" |
07e81dc9 | 21 | #include "accessors.h" |
a0231804 | 22 | #include "extent-tree.h" |
2e405ad8 | 23 | |
06d61cb1 JB |
24 | #ifdef CONFIG_BTRFS_DEBUG |
25 | int btrfs_should_fragment_free_space(struct btrfs_block_group *block_group) | |
26 | { | |
27 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
28 | ||
29 | return (btrfs_test_opt(fs_info, FRAGMENT_METADATA) && | |
30 | block_group->flags & BTRFS_BLOCK_GROUP_METADATA) || | |
31 | (btrfs_test_opt(fs_info, FRAGMENT_DATA) && | |
32 | block_group->flags & BTRFS_BLOCK_GROUP_DATA); | |
33 | } | |
34 | #endif | |
35 | ||
878d7b67 JB |
36 | /* |
37 | * Return target flags in extended format or 0 if restripe for this chunk_type | |
38 | * is not in progress | |
39 | * | |
40 | * Should be called with balance_lock held | |
41 | */ | |
e11c0406 | 42 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) |
878d7b67 JB |
43 | { |
44 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
45 | u64 target = 0; | |
46 | ||
47 | if (!bctl) | |
48 | return 0; | |
49 | ||
50 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
51 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
52 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
53 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
54 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
55 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
56 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
57 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
58 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
59 | } | |
60 | ||
61 | return target; | |
62 | } | |
63 | ||
64 | /* | |
65 | * @flags: available profiles in extended format (see ctree.h) | |
66 | * | |
67 | * Return reduced profile in chunk format. If profile changing is in progress | |
68 | * (either running or paused) picks the target profile (if it's already | |
69 | * available), otherwise falls back to plain reducing. | |
70 | */ | |
71 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) | |
72 | { | |
73 | u64 num_devices = fs_info->fs_devices->rw_devices; | |
74 | u64 target; | |
75 | u64 raid_type; | |
76 | u64 allowed = 0; | |
77 | ||
78 | /* | |
79 | * See if restripe for this chunk_type is in progress, if so try to | |
80 | * reduce to the target profile | |
81 | */ | |
82 | spin_lock(&fs_info->balance_lock); | |
e11c0406 | 83 | target = get_restripe_target(fs_info, flags); |
878d7b67 | 84 | if (target) { |
162e0a16 JB |
85 | spin_unlock(&fs_info->balance_lock); |
86 | return extended_to_chunk(target); | |
878d7b67 JB |
87 | } |
88 | spin_unlock(&fs_info->balance_lock); | |
89 | ||
90 | /* First, mask out the RAID levels which aren't possible */ | |
91 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { | |
92 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
93 | allowed |= btrfs_raid_array[raid_type].bg_flag; | |
94 | } | |
95 | allowed &= flags; | |
96 | ||
97 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
98 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
99 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
100 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
101 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
102 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
103 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
104 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
105 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
106 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
107 | ||
108 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
109 | ||
110 | return extended_to_chunk(flags | allowed); | |
111 | } | |
112 | ||
ef0a82da | 113 | u64 btrfs_get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
878d7b67 JB |
114 | { |
115 | unsigned seq; | |
116 | u64 flags; | |
117 | ||
118 | do { | |
119 | flags = orig_flags; | |
120 | seq = read_seqbegin(&fs_info->profiles_lock); | |
121 | ||
122 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
123 | flags |= fs_info->avail_data_alloc_bits; | |
124 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
125 | flags |= fs_info->avail_system_alloc_bits; | |
126 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
127 | flags |= fs_info->avail_metadata_alloc_bits; | |
128 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
129 | ||
130 | return btrfs_reduce_alloc_profile(fs_info, flags); | |
131 | } | |
132 | ||
32da5386 | 133 | void btrfs_get_block_group(struct btrfs_block_group *cache) |
3cad1284 | 134 | { |
48aaeebe | 135 | refcount_inc(&cache->refs); |
3cad1284 JB |
136 | } |
137 | ||
32da5386 | 138 | void btrfs_put_block_group(struct btrfs_block_group *cache) |
3cad1284 | 139 | { |
48aaeebe | 140 | if (refcount_dec_and_test(&cache->refs)) { |
3cad1284 | 141 | WARN_ON(cache->pinned > 0); |
40cdc509 FM |
142 | /* |
143 | * If there was a failure to cleanup a log tree, very likely due | |
144 | * to an IO failure on a writeback attempt of one or more of its | |
145 | * extent buffers, we could not do proper (and cheap) unaccounting | |
146 | * of their reserved space, so don't warn on reserved > 0 in that | |
147 | * case. | |
148 | */ | |
149 | if (!(cache->flags & BTRFS_BLOCK_GROUP_METADATA) || | |
150 | !BTRFS_FS_LOG_CLEANUP_ERROR(cache->fs_info)) | |
151 | WARN_ON(cache->reserved > 0); | |
3cad1284 | 152 | |
b0643e59 DZ |
153 | /* |
154 | * A block_group shouldn't be on the discard_list anymore. | |
155 | * Remove the block_group from the discard_list to prevent us | |
156 | * from causing a panic due to NULL pointer dereference. | |
157 | */ | |
158 | if (WARN_ON(!list_empty(&cache->discard_list))) | |
159 | btrfs_discard_cancel_work(&cache->fs_info->discard_ctl, | |
160 | cache); | |
161 | ||
3cad1284 JB |
162 | /* |
163 | * If not empty, someone is still holding mutex of | |
164 | * full_stripe_lock, which can only be released by caller. | |
165 | * And it will definitely cause use-after-free when caller | |
166 | * tries to release full stripe lock. | |
167 | * | |
168 | * No better way to resolve, but only to warn. | |
169 | */ | |
170 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
171 | kfree(cache->free_space_ctl); | |
dafc340d | 172 | kfree(cache->physical_map); |
3cad1284 JB |
173 | kfree(cache); |
174 | } | |
175 | } | |
176 | ||
4358d963 JB |
177 | /* |
178 | * This adds the block group to the fs_info rb tree for the block group cache | |
179 | */ | |
180 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, | |
32da5386 | 181 | struct btrfs_block_group *block_group) |
4358d963 JB |
182 | { |
183 | struct rb_node **p; | |
184 | struct rb_node *parent = NULL; | |
32da5386 | 185 | struct btrfs_block_group *cache; |
08dddb29 | 186 | bool leftmost = true; |
4358d963 | 187 | |
9afc6649 QW |
188 | ASSERT(block_group->length != 0); |
189 | ||
16b0c258 | 190 | write_lock(&info->block_group_cache_lock); |
08dddb29 | 191 | p = &info->block_group_cache_tree.rb_root.rb_node; |
4358d963 JB |
192 | |
193 | while (*p) { | |
194 | parent = *p; | |
32da5386 | 195 | cache = rb_entry(parent, struct btrfs_block_group, cache_node); |
b3470b5d | 196 | if (block_group->start < cache->start) { |
4358d963 | 197 | p = &(*p)->rb_left; |
b3470b5d | 198 | } else if (block_group->start > cache->start) { |
4358d963 | 199 | p = &(*p)->rb_right; |
08dddb29 | 200 | leftmost = false; |
4358d963 | 201 | } else { |
16b0c258 | 202 | write_unlock(&info->block_group_cache_lock); |
4358d963 JB |
203 | return -EEXIST; |
204 | } | |
205 | } | |
206 | ||
207 | rb_link_node(&block_group->cache_node, parent, p); | |
08dddb29 FM |
208 | rb_insert_color_cached(&block_group->cache_node, |
209 | &info->block_group_cache_tree, leftmost); | |
4358d963 | 210 | |
16b0c258 | 211 | write_unlock(&info->block_group_cache_lock); |
4358d963 JB |
212 | |
213 | return 0; | |
214 | } | |
215 | ||
2e405ad8 JB |
216 | /* |
217 | * This will return the block group at or after bytenr if contains is 0, else | |
218 | * it will return the block group that contains the bytenr | |
219 | */ | |
32da5386 | 220 | static struct btrfs_block_group *block_group_cache_tree_search( |
2e405ad8 JB |
221 | struct btrfs_fs_info *info, u64 bytenr, int contains) |
222 | { | |
32da5386 | 223 | struct btrfs_block_group *cache, *ret = NULL; |
2e405ad8 JB |
224 | struct rb_node *n; |
225 | u64 end, start; | |
226 | ||
16b0c258 | 227 | read_lock(&info->block_group_cache_lock); |
08dddb29 | 228 | n = info->block_group_cache_tree.rb_root.rb_node; |
2e405ad8 JB |
229 | |
230 | while (n) { | |
32da5386 | 231 | cache = rb_entry(n, struct btrfs_block_group, cache_node); |
b3470b5d DS |
232 | end = cache->start + cache->length - 1; |
233 | start = cache->start; | |
2e405ad8 JB |
234 | |
235 | if (bytenr < start) { | |
b3470b5d | 236 | if (!contains && (!ret || start < ret->start)) |
2e405ad8 JB |
237 | ret = cache; |
238 | n = n->rb_left; | |
239 | } else if (bytenr > start) { | |
240 | if (contains && bytenr <= end) { | |
241 | ret = cache; | |
242 | break; | |
243 | } | |
244 | n = n->rb_right; | |
245 | } else { | |
246 | ret = cache; | |
247 | break; | |
248 | } | |
249 | } | |
08dddb29 | 250 | if (ret) |
2e405ad8 | 251 | btrfs_get_block_group(ret); |
16b0c258 | 252 | read_unlock(&info->block_group_cache_lock); |
2e405ad8 JB |
253 | |
254 | return ret; | |
255 | } | |
256 | ||
257 | /* | |
258 | * Return the block group that starts at or after bytenr | |
259 | */ | |
32da5386 | 260 | struct btrfs_block_group *btrfs_lookup_first_block_group( |
2e405ad8 JB |
261 | struct btrfs_fs_info *info, u64 bytenr) |
262 | { | |
263 | return block_group_cache_tree_search(info, bytenr, 0); | |
264 | } | |
265 | ||
266 | /* | |
267 | * Return the block group that contains the given bytenr | |
268 | */ | |
32da5386 | 269 | struct btrfs_block_group *btrfs_lookup_block_group( |
2e405ad8 JB |
270 | struct btrfs_fs_info *info, u64 bytenr) |
271 | { | |
272 | return block_group_cache_tree_search(info, bytenr, 1); | |
273 | } | |
274 | ||
32da5386 DS |
275 | struct btrfs_block_group *btrfs_next_block_group( |
276 | struct btrfs_block_group *cache) | |
2e405ad8 JB |
277 | { |
278 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
279 | struct rb_node *node; | |
280 | ||
16b0c258 | 281 | read_lock(&fs_info->block_group_cache_lock); |
2e405ad8 JB |
282 | |
283 | /* If our block group was removed, we need a full search. */ | |
284 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
b3470b5d | 285 | const u64 next_bytenr = cache->start + cache->length; |
2e405ad8 | 286 | |
16b0c258 | 287 | read_unlock(&fs_info->block_group_cache_lock); |
2e405ad8 | 288 | btrfs_put_block_group(cache); |
8b01f931 | 289 | return btrfs_lookup_first_block_group(fs_info, next_bytenr); |
2e405ad8 JB |
290 | } |
291 | node = rb_next(&cache->cache_node); | |
292 | btrfs_put_block_group(cache); | |
293 | if (node) { | |
32da5386 | 294 | cache = rb_entry(node, struct btrfs_block_group, cache_node); |
2e405ad8 JB |
295 | btrfs_get_block_group(cache); |
296 | } else | |
297 | cache = NULL; | |
16b0c258 | 298 | read_unlock(&fs_info->block_group_cache_lock); |
2e405ad8 JB |
299 | return cache; |
300 | } | |
3eeb3226 | 301 | |
2306e83e FM |
302 | /** |
303 | * Check if we can do a NOCOW write for a given extent. | |
304 | * | |
305 | * @fs_info: The filesystem information object. | |
306 | * @bytenr: Logical start address of the extent. | |
307 | * | |
308 | * Check if we can do a NOCOW write for the given extent, and increments the | |
309 | * number of NOCOW writers in the block group that contains the extent, as long | |
310 | * as the block group exists and it's currently not in read-only mode. | |
311 | * | |
312 | * Returns: A non-NULL block group pointer if we can do a NOCOW write, the caller | |
313 | * is responsible for calling btrfs_dec_nocow_writers() later. | |
314 | * | |
315 | * Or NULL if we can not do a NOCOW write | |
316 | */ | |
317 | struct btrfs_block_group *btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, | |
318 | u64 bytenr) | |
3eeb3226 | 319 | { |
32da5386 | 320 | struct btrfs_block_group *bg; |
2306e83e | 321 | bool can_nocow = true; |
3eeb3226 JB |
322 | |
323 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
324 | if (!bg) | |
2306e83e | 325 | return NULL; |
3eeb3226 JB |
326 | |
327 | spin_lock(&bg->lock); | |
328 | if (bg->ro) | |
2306e83e | 329 | can_nocow = false; |
3eeb3226 JB |
330 | else |
331 | atomic_inc(&bg->nocow_writers); | |
332 | spin_unlock(&bg->lock); | |
333 | ||
2306e83e | 334 | if (!can_nocow) { |
3eeb3226 | 335 | btrfs_put_block_group(bg); |
2306e83e FM |
336 | return NULL; |
337 | } | |
3eeb3226 | 338 | |
2306e83e FM |
339 | /* No put on block group, done by btrfs_dec_nocow_writers(). */ |
340 | return bg; | |
3eeb3226 JB |
341 | } |
342 | ||
2306e83e FM |
343 | /** |
344 | * Decrement the number of NOCOW writers in a block group. | |
345 | * | |
346 | * @bg: The block group. | |
347 | * | |
348 | * This is meant to be called after a previous call to btrfs_inc_nocow_writers(), | |
349 | * and on the block group returned by that call. Typically this is called after | |
350 | * creating an ordered extent for a NOCOW write, to prevent races with scrub and | |
351 | * relocation. | |
352 | * | |
353 | * After this call, the caller should not use the block group anymore. It it wants | |
354 | * to use it, then it should get a reference on it before calling this function. | |
355 | */ | |
356 | void btrfs_dec_nocow_writers(struct btrfs_block_group *bg) | |
3eeb3226 | 357 | { |
3eeb3226 JB |
358 | if (atomic_dec_and_test(&bg->nocow_writers)) |
359 | wake_up_var(&bg->nocow_writers); | |
2306e83e FM |
360 | |
361 | /* For the lookup done by a previous call to btrfs_inc_nocow_writers(). */ | |
3eeb3226 JB |
362 | btrfs_put_block_group(bg); |
363 | } | |
364 | ||
32da5386 | 365 | void btrfs_wait_nocow_writers(struct btrfs_block_group *bg) |
3eeb3226 JB |
366 | { |
367 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); | |
368 | } | |
369 | ||
370 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
371 | const u64 start) | |
372 | { | |
32da5386 | 373 | struct btrfs_block_group *bg; |
3eeb3226 JB |
374 | |
375 | bg = btrfs_lookup_block_group(fs_info, start); | |
376 | ASSERT(bg); | |
377 | if (atomic_dec_and_test(&bg->reservations)) | |
378 | wake_up_var(&bg->reservations); | |
379 | btrfs_put_block_group(bg); | |
380 | } | |
381 | ||
32da5386 | 382 | void btrfs_wait_block_group_reservations(struct btrfs_block_group *bg) |
3eeb3226 JB |
383 | { |
384 | struct btrfs_space_info *space_info = bg->space_info; | |
385 | ||
386 | ASSERT(bg->ro); | |
387 | ||
388 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
389 | return; | |
390 | ||
391 | /* | |
392 | * Our block group is read only but before we set it to read only, | |
393 | * some task might have had allocated an extent from it already, but it | |
394 | * has not yet created a respective ordered extent (and added it to a | |
395 | * root's list of ordered extents). | |
396 | * Therefore wait for any task currently allocating extents, since the | |
397 | * block group's reservations counter is incremented while a read lock | |
398 | * on the groups' semaphore is held and decremented after releasing | |
399 | * the read access on that semaphore and creating the ordered extent. | |
400 | */ | |
401 | down_write(&space_info->groups_sem); | |
402 | up_write(&space_info->groups_sem); | |
403 | ||
404 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); | |
405 | } | |
9f21246d JB |
406 | |
407 | struct btrfs_caching_control *btrfs_get_caching_control( | |
32da5386 | 408 | struct btrfs_block_group *cache) |
9f21246d JB |
409 | { |
410 | struct btrfs_caching_control *ctl; | |
411 | ||
412 | spin_lock(&cache->lock); | |
413 | if (!cache->caching_ctl) { | |
414 | spin_unlock(&cache->lock); | |
415 | return NULL; | |
416 | } | |
417 | ||
418 | ctl = cache->caching_ctl; | |
419 | refcount_inc(&ctl->count); | |
420 | spin_unlock(&cache->lock); | |
421 | return ctl; | |
422 | } | |
423 | ||
424 | void btrfs_put_caching_control(struct btrfs_caching_control *ctl) | |
425 | { | |
426 | if (refcount_dec_and_test(&ctl->count)) | |
427 | kfree(ctl); | |
428 | } | |
429 | ||
430 | /* | |
431 | * When we wait for progress in the block group caching, its because our | |
432 | * allocation attempt failed at least once. So, we must sleep and let some | |
433 | * progress happen before we try again. | |
434 | * | |
435 | * This function will sleep at least once waiting for new free space to show | |
436 | * up, and then it will check the block group free space numbers for our min | |
437 | * num_bytes. Another option is to have it go ahead and look in the rbtree for | |
438 | * a free extent of a given size, but this is a good start. | |
439 | * | |
440 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
441 | * any of the information in this block group. | |
442 | */ | |
32da5386 | 443 | void btrfs_wait_block_group_cache_progress(struct btrfs_block_group *cache, |
9f21246d JB |
444 | u64 num_bytes) |
445 | { | |
446 | struct btrfs_caching_control *caching_ctl; | |
447 | ||
448 | caching_ctl = btrfs_get_caching_control(cache); | |
449 | if (!caching_ctl) | |
450 | return; | |
451 | ||
32da5386 | 452 | wait_event(caching_ctl->wait, btrfs_block_group_done(cache) || |
9f21246d JB |
453 | (cache->free_space_ctl->free_space >= num_bytes)); |
454 | ||
455 | btrfs_put_caching_control(caching_ctl); | |
456 | } | |
457 | ||
ced8ecf0 OS |
458 | static int btrfs_caching_ctl_wait_done(struct btrfs_block_group *cache, |
459 | struct btrfs_caching_control *caching_ctl) | |
460 | { | |
461 | wait_event(caching_ctl->wait, btrfs_block_group_done(cache)); | |
462 | return cache->cached == BTRFS_CACHE_ERROR ? -EIO : 0; | |
463 | } | |
464 | ||
465 | static int btrfs_wait_block_group_cache_done(struct btrfs_block_group *cache) | |
9f21246d JB |
466 | { |
467 | struct btrfs_caching_control *caching_ctl; | |
ced8ecf0 | 468 | int ret; |
9f21246d JB |
469 | |
470 | caching_ctl = btrfs_get_caching_control(cache); | |
471 | if (!caching_ctl) | |
472 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; | |
ced8ecf0 | 473 | ret = btrfs_caching_ctl_wait_done(cache, caching_ctl); |
9f21246d JB |
474 | btrfs_put_caching_control(caching_ctl); |
475 | return ret; | |
476 | } | |
477 | ||
478 | #ifdef CONFIG_BTRFS_DEBUG | |
32da5386 | 479 | static void fragment_free_space(struct btrfs_block_group *block_group) |
9f21246d JB |
480 | { |
481 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
b3470b5d DS |
482 | u64 start = block_group->start; |
483 | u64 len = block_group->length; | |
9f21246d JB |
484 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? |
485 | fs_info->nodesize : fs_info->sectorsize; | |
486 | u64 step = chunk << 1; | |
487 | ||
488 | while (len > chunk) { | |
489 | btrfs_remove_free_space(block_group, start, chunk); | |
490 | start += step; | |
491 | if (len < step) | |
492 | len = 0; | |
493 | else | |
494 | len -= step; | |
495 | } | |
496 | } | |
497 | #endif | |
498 | ||
499 | /* | |
500 | * This is only called by btrfs_cache_block_group, since we could have freed | |
501 | * extents we need to check the pinned_extents for any extents that can't be | |
502 | * used yet since their free space will be released as soon as the transaction | |
503 | * commits. | |
504 | */ | |
32da5386 | 505 | u64 add_new_free_space(struct btrfs_block_group *block_group, u64 start, u64 end) |
9f21246d JB |
506 | { |
507 | struct btrfs_fs_info *info = block_group->fs_info; | |
508 | u64 extent_start, extent_end, size, total_added = 0; | |
509 | int ret; | |
510 | ||
511 | while (start < end) { | |
fe119a6e | 512 | ret = find_first_extent_bit(&info->excluded_extents, start, |
9f21246d JB |
513 | &extent_start, &extent_end, |
514 | EXTENT_DIRTY | EXTENT_UPTODATE, | |
515 | NULL); | |
516 | if (ret) | |
517 | break; | |
518 | ||
519 | if (extent_start <= start) { | |
520 | start = extent_end + 1; | |
521 | } else if (extent_start > start && extent_start < end) { | |
522 | size = extent_start - start; | |
523 | total_added += size; | |
b0643e59 DZ |
524 | ret = btrfs_add_free_space_async_trimmed(block_group, |
525 | start, size); | |
9f21246d JB |
526 | BUG_ON(ret); /* -ENOMEM or logic error */ |
527 | start = extent_end + 1; | |
528 | } else { | |
529 | break; | |
530 | } | |
531 | } | |
532 | ||
533 | if (start < end) { | |
534 | size = end - start; | |
535 | total_added += size; | |
b0643e59 DZ |
536 | ret = btrfs_add_free_space_async_trimmed(block_group, start, |
537 | size); | |
9f21246d JB |
538 | BUG_ON(ret); /* -ENOMEM or logic error */ |
539 | } | |
540 | ||
541 | return total_added; | |
542 | } | |
543 | ||
544 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) | |
545 | { | |
32da5386 | 546 | struct btrfs_block_group *block_group = caching_ctl->block_group; |
9f21246d | 547 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
29cbcf40 | 548 | struct btrfs_root *extent_root; |
9f21246d JB |
549 | struct btrfs_path *path; |
550 | struct extent_buffer *leaf; | |
551 | struct btrfs_key key; | |
552 | u64 total_found = 0; | |
553 | u64 last = 0; | |
554 | u32 nritems; | |
555 | int ret; | |
556 | bool wakeup = true; | |
557 | ||
558 | path = btrfs_alloc_path(); | |
559 | if (!path) | |
560 | return -ENOMEM; | |
561 | ||
b3470b5d | 562 | last = max_t(u64, block_group->start, BTRFS_SUPER_INFO_OFFSET); |
29cbcf40 | 563 | extent_root = btrfs_extent_root(fs_info, last); |
9f21246d JB |
564 | |
565 | #ifdef CONFIG_BTRFS_DEBUG | |
566 | /* | |
567 | * If we're fragmenting we don't want to make anybody think we can | |
568 | * allocate from this block group until we've had a chance to fragment | |
569 | * the free space. | |
570 | */ | |
571 | if (btrfs_should_fragment_free_space(block_group)) | |
572 | wakeup = false; | |
573 | #endif | |
574 | /* | |
575 | * We don't want to deadlock with somebody trying to allocate a new | |
576 | * extent for the extent root while also trying to search the extent | |
577 | * root to add free space. So we skip locking and search the commit | |
578 | * root, since its read-only | |
579 | */ | |
580 | path->skip_locking = 1; | |
581 | path->search_commit_root = 1; | |
582 | path->reada = READA_FORWARD; | |
583 | ||
584 | key.objectid = last; | |
585 | key.offset = 0; | |
586 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
587 | ||
588 | next: | |
589 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
590 | if (ret < 0) | |
591 | goto out; | |
592 | ||
593 | leaf = path->nodes[0]; | |
594 | nritems = btrfs_header_nritems(leaf); | |
595 | ||
596 | while (1) { | |
597 | if (btrfs_fs_closing(fs_info) > 1) { | |
598 | last = (u64)-1; | |
599 | break; | |
600 | } | |
601 | ||
602 | if (path->slots[0] < nritems) { | |
603 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
604 | } else { | |
605 | ret = btrfs_find_next_key(extent_root, path, &key, 0, 0); | |
606 | if (ret) | |
607 | break; | |
608 | ||
609 | if (need_resched() || | |
610 | rwsem_is_contended(&fs_info->commit_root_sem)) { | |
9f21246d JB |
611 | btrfs_release_path(path); |
612 | up_read(&fs_info->commit_root_sem); | |
613 | mutex_unlock(&caching_ctl->mutex); | |
614 | cond_resched(); | |
615 | mutex_lock(&caching_ctl->mutex); | |
616 | down_read(&fs_info->commit_root_sem); | |
617 | goto next; | |
618 | } | |
619 | ||
620 | ret = btrfs_next_leaf(extent_root, path); | |
621 | if (ret < 0) | |
622 | goto out; | |
623 | if (ret) | |
624 | break; | |
625 | leaf = path->nodes[0]; | |
626 | nritems = btrfs_header_nritems(leaf); | |
627 | continue; | |
628 | } | |
629 | ||
630 | if (key.objectid < last) { | |
631 | key.objectid = last; | |
632 | key.offset = 0; | |
633 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
9f21246d JB |
634 | btrfs_release_path(path); |
635 | goto next; | |
636 | } | |
637 | ||
b3470b5d | 638 | if (key.objectid < block_group->start) { |
9f21246d JB |
639 | path->slots[0]++; |
640 | continue; | |
641 | } | |
642 | ||
b3470b5d | 643 | if (key.objectid >= block_group->start + block_group->length) |
9f21246d JB |
644 | break; |
645 | ||
646 | if (key.type == BTRFS_EXTENT_ITEM_KEY || | |
647 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
648 | total_found += add_new_free_space(block_group, last, | |
649 | key.objectid); | |
650 | if (key.type == BTRFS_METADATA_ITEM_KEY) | |
651 | last = key.objectid + | |
652 | fs_info->nodesize; | |
653 | else | |
654 | last = key.objectid + key.offset; | |
655 | ||
656 | if (total_found > CACHING_CTL_WAKE_UP) { | |
657 | total_found = 0; | |
658 | if (wakeup) | |
659 | wake_up(&caching_ctl->wait); | |
660 | } | |
661 | } | |
662 | path->slots[0]++; | |
663 | } | |
664 | ret = 0; | |
665 | ||
666 | total_found += add_new_free_space(block_group, last, | |
b3470b5d | 667 | block_group->start + block_group->length); |
9f21246d JB |
668 | |
669 | out: | |
670 | btrfs_free_path(path); | |
671 | return ret; | |
672 | } | |
673 | ||
674 | static noinline void caching_thread(struct btrfs_work *work) | |
675 | { | |
32da5386 | 676 | struct btrfs_block_group *block_group; |
9f21246d JB |
677 | struct btrfs_fs_info *fs_info; |
678 | struct btrfs_caching_control *caching_ctl; | |
679 | int ret; | |
680 | ||
681 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
682 | block_group = caching_ctl->block_group; | |
683 | fs_info = block_group->fs_info; | |
684 | ||
685 | mutex_lock(&caching_ctl->mutex); | |
686 | down_read(&fs_info->commit_root_sem); | |
687 | ||
e747853c JB |
688 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
689 | ret = load_free_space_cache(block_group); | |
690 | if (ret == 1) { | |
691 | ret = 0; | |
692 | goto done; | |
693 | } | |
694 | ||
695 | /* | |
696 | * We failed to load the space cache, set ourselves to | |
697 | * CACHE_STARTED and carry on. | |
698 | */ | |
699 | spin_lock(&block_group->lock); | |
700 | block_group->cached = BTRFS_CACHE_STARTED; | |
701 | spin_unlock(&block_group->lock); | |
702 | wake_up(&caching_ctl->wait); | |
703 | } | |
704 | ||
2f96e402 JB |
705 | /* |
706 | * If we are in the transaction that populated the free space tree we | |
707 | * can't actually cache from the free space tree as our commit root and | |
708 | * real root are the same, so we could change the contents of the blocks | |
709 | * while caching. Instead do the slow caching in this case, and after | |
710 | * the transaction has committed we will be safe. | |
711 | */ | |
712 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && | |
713 | !(test_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags))) | |
9f21246d JB |
714 | ret = load_free_space_tree(caching_ctl); |
715 | else | |
716 | ret = load_extent_tree_free(caching_ctl); | |
e747853c | 717 | done: |
9f21246d JB |
718 | spin_lock(&block_group->lock); |
719 | block_group->caching_ctl = NULL; | |
720 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; | |
721 | spin_unlock(&block_group->lock); | |
722 | ||
723 | #ifdef CONFIG_BTRFS_DEBUG | |
724 | if (btrfs_should_fragment_free_space(block_group)) { | |
725 | u64 bytes_used; | |
726 | ||
727 | spin_lock(&block_group->space_info->lock); | |
728 | spin_lock(&block_group->lock); | |
b3470b5d | 729 | bytes_used = block_group->length - block_group->used; |
9f21246d JB |
730 | block_group->space_info->bytes_used += bytes_used >> 1; |
731 | spin_unlock(&block_group->lock); | |
732 | spin_unlock(&block_group->space_info->lock); | |
e11c0406 | 733 | fragment_free_space(block_group); |
9f21246d JB |
734 | } |
735 | #endif | |
736 | ||
9f21246d JB |
737 | up_read(&fs_info->commit_root_sem); |
738 | btrfs_free_excluded_extents(block_group); | |
739 | mutex_unlock(&caching_ctl->mutex); | |
740 | ||
741 | wake_up(&caching_ctl->wait); | |
742 | ||
743 | btrfs_put_caching_control(caching_ctl); | |
744 | btrfs_put_block_group(block_group); | |
745 | } | |
746 | ||
ced8ecf0 | 747 | int btrfs_cache_block_group(struct btrfs_block_group *cache, bool wait) |
9f21246d | 748 | { |
9f21246d | 749 | struct btrfs_fs_info *fs_info = cache->fs_info; |
e747853c | 750 | struct btrfs_caching_control *caching_ctl = NULL; |
9f21246d JB |
751 | int ret = 0; |
752 | ||
2eda5708 NA |
753 | /* Allocator for zoned filesystems does not use the cache at all */ |
754 | if (btrfs_is_zoned(fs_info)) | |
755 | return 0; | |
756 | ||
9f21246d JB |
757 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
758 | if (!caching_ctl) | |
759 | return -ENOMEM; | |
760 | ||
761 | INIT_LIST_HEAD(&caching_ctl->list); | |
762 | mutex_init(&caching_ctl->mutex); | |
763 | init_waitqueue_head(&caching_ctl->wait); | |
764 | caching_ctl->block_group = cache; | |
e747853c | 765 | refcount_set(&caching_ctl->count, 2); |
a0cac0ec | 766 | btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL); |
9f21246d JB |
767 | |
768 | spin_lock(&cache->lock); | |
9f21246d | 769 | if (cache->cached != BTRFS_CACHE_NO) { |
9f21246d | 770 | kfree(caching_ctl); |
e747853c JB |
771 | |
772 | caching_ctl = cache->caching_ctl; | |
773 | if (caching_ctl) | |
774 | refcount_inc(&caching_ctl->count); | |
775 | spin_unlock(&cache->lock); | |
776 | goto out; | |
9f21246d JB |
777 | } |
778 | WARN_ON(cache->caching_ctl); | |
779 | cache->caching_ctl = caching_ctl; | |
ced8ecf0 | 780 | cache->cached = BTRFS_CACHE_STARTED; |
9f21246d JB |
781 | spin_unlock(&cache->lock); |
782 | ||
16b0c258 | 783 | write_lock(&fs_info->block_group_cache_lock); |
9f21246d JB |
784 | refcount_inc(&caching_ctl->count); |
785 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); | |
16b0c258 | 786 | write_unlock(&fs_info->block_group_cache_lock); |
9f21246d JB |
787 | |
788 | btrfs_get_block_group(cache); | |
789 | ||
790 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); | |
e747853c | 791 | out: |
ced8ecf0 OS |
792 | if (wait && caching_ctl) |
793 | ret = btrfs_caching_ctl_wait_done(cache, caching_ctl); | |
e747853c JB |
794 | if (caching_ctl) |
795 | btrfs_put_caching_control(caching_ctl); | |
9f21246d JB |
796 | |
797 | return ret; | |
798 | } | |
e3e0520b JB |
799 | |
800 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
801 | { | |
802 | u64 extra_flags = chunk_to_extended(flags) & | |
803 | BTRFS_EXTENDED_PROFILE_MASK; | |
804 | ||
805 | write_seqlock(&fs_info->profiles_lock); | |
806 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
807 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
808 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
809 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
810 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
811 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
812 | write_sequnlock(&fs_info->profiles_lock); | |
813 | } | |
814 | ||
815 | /* | |
816 | * Clear incompat bits for the following feature(s): | |
817 | * | |
818 | * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group | |
819 | * in the whole filesystem | |
9c907446 DS |
820 | * |
821 | * - RAID1C34 - same as above for RAID1C3 and RAID1C4 block groups | |
e3e0520b JB |
822 | */ |
823 | static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
824 | { | |
9c907446 DS |
825 | bool found_raid56 = false; |
826 | bool found_raid1c34 = false; | |
827 | ||
828 | if ((flags & BTRFS_BLOCK_GROUP_RAID56_MASK) || | |
829 | (flags & BTRFS_BLOCK_GROUP_RAID1C3) || | |
830 | (flags & BTRFS_BLOCK_GROUP_RAID1C4)) { | |
e3e0520b JB |
831 | struct list_head *head = &fs_info->space_info; |
832 | struct btrfs_space_info *sinfo; | |
833 | ||
834 | list_for_each_entry_rcu(sinfo, head, list) { | |
e3e0520b JB |
835 | down_read(&sinfo->groups_sem); |
836 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5])) | |
9c907446 | 837 | found_raid56 = true; |
e3e0520b | 838 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6])) |
9c907446 DS |
839 | found_raid56 = true; |
840 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C3])) | |
841 | found_raid1c34 = true; | |
842 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C4])) | |
843 | found_raid1c34 = true; | |
e3e0520b | 844 | up_read(&sinfo->groups_sem); |
e3e0520b | 845 | } |
d8e6fd5c | 846 | if (!found_raid56) |
9c907446 | 847 | btrfs_clear_fs_incompat(fs_info, RAID56); |
d8e6fd5c | 848 | if (!found_raid1c34) |
9c907446 | 849 | btrfs_clear_fs_incompat(fs_info, RAID1C34); |
e3e0520b JB |
850 | } |
851 | } | |
852 | ||
7357623a QW |
853 | static int remove_block_group_item(struct btrfs_trans_handle *trans, |
854 | struct btrfs_path *path, | |
855 | struct btrfs_block_group *block_group) | |
856 | { | |
857 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
858 | struct btrfs_root *root; | |
859 | struct btrfs_key key; | |
860 | int ret; | |
861 | ||
dfe8aec4 | 862 | root = btrfs_block_group_root(fs_info); |
7357623a QW |
863 | key.objectid = block_group->start; |
864 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
865 | key.offset = block_group->length; | |
866 | ||
867 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
868 | if (ret > 0) | |
869 | ret = -ENOENT; | |
870 | if (ret < 0) | |
871 | return ret; | |
872 | ||
873 | ret = btrfs_del_item(trans, root, path); | |
874 | return ret; | |
875 | } | |
876 | ||
e3e0520b JB |
877 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
878 | u64 group_start, struct extent_map *em) | |
879 | { | |
880 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
e3e0520b | 881 | struct btrfs_path *path; |
32da5386 | 882 | struct btrfs_block_group *block_group; |
e3e0520b | 883 | struct btrfs_free_cluster *cluster; |
e3e0520b JB |
884 | struct inode *inode; |
885 | struct kobject *kobj = NULL; | |
886 | int ret; | |
887 | int index; | |
888 | int factor; | |
889 | struct btrfs_caching_control *caching_ctl = NULL; | |
890 | bool remove_em; | |
891 | bool remove_rsv = false; | |
892 | ||
893 | block_group = btrfs_lookup_block_group(fs_info, group_start); | |
894 | BUG_ON(!block_group); | |
895 | BUG_ON(!block_group->ro); | |
896 | ||
897 | trace_btrfs_remove_block_group(block_group); | |
898 | /* | |
899 | * Free the reserved super bytes from this block group before | |
900 | * remove it. | |
901 | */ | |
902 | btrfs_free_excluded_extents(block_group); | |
b3470b5d DS |
903 | btrfs_free_ref_tree_range(fs_info, block_group->start, |
904 | block_group->length); | |
e3e0520b | 905 | |
e3e0520b JB |
906 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
907 | factor = btrfs_bg_type_to_factor(block_group->flags); | |
908 | ||
909 | /* make sure this block group isn't part of an allocation cluster */ | |
910 | cluster = &fs_info->data_alloc_cluster; | |
911 | spin_lock(&cluster->refill_lock); | |
912 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
913 | spin_unlock(&cluster->refill_lock); | |
914 | ||
915 | /* | |
916 | * make sure this block group isn't part of a metadata | |
917 | * allocation cluster | |
918 | */ | |
919 | cluster = &fs_info->meta_alloc_cluster; | |
920 | spin_lock(&cluster->refill_lock); | |
921 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
922 | spin_unlock(&cluster->refill_lock); | |
923 | ||
40ab3be1 | 924 | btrfs_clear_treelog_bg(block_group); |
c2707a25 | 925 | btrfs_clear_data_reloc_bg(block_group); |
40ab3be1 | 926 | |
e3e0520b JB |
927 | path = btrfs_alloc_path(); |
928 | if (!path) { | |
929 | ret = -ENOMEM; | |
9fecd132 | 930 | goto out; |
e3e0520b JB |
931 | } |
932 | ||
933 | /* | |
934 | * get the inode first so any iput calls done for the io_list | |
935 | * aren't the final iput (no unlinks allowed now) | |
936 | */ | |
937 | inode = lookup_free_space_inode(block_group, path); | |
938 | ||
939 | mutex_lock(&trans->transaction->cache_write_mutex); | |
940 | /* | |
941 | * Make sure our free space cache IO is done before removing the | |
942 | * free space inode | |
943 | */ | |
944 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
945 | if (!list_empty(&block_group->io_list)) { | |
946 | list_del_init(&block_group->io_list); | |
947 | ||
948 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
949 | ||
950 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
951 | btrfs_wait_cache_io(trans, block_group, path); | |
952 | btrfs_put_block_group(block_group); | |
953 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
954 | } | |
955 | ||
956 | if (!list_empty(&block_group->dirty_list)) { | |
957 | list_del_init(&block_group->dirty_list); | |
958 | remove_rsv = true; | |
959 | btrfs_put_block_group(block_group); | |
960 | } | |
961 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
962 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
963 | ||
36b216c8 BB |
964 | ret = btrfs_remove_free_space_inode(trans, inode, block_group); |
965 | if (ret) | |
9fecd132 | 966 | goto out; |
e3e0520b | 967 | |
16b0c258 | 968 | write_lock(&fs_info->block_group_cache_lock); |
08dddb29 FM |
969 | rb_erase_cached(&block_group->cache_node, |
970 | &fs_info->block_group_cache_tree); | |
e3e0520b JB |
971 | RB_CLEAR_NODE(&block_group->cache_node); |
972 | ||
9fecd132 FM |
973 | /* Once for the block groups rbtree */ |
974 | btrfs_put_block_group(block_group); | |
975 | ||
16b0c258 | 976 | write_unlock(&fs_info->block_group_cache_lock); |
e3e0520b JB |
977 | |
978 | down_write(&block_group->space_info->groups_sem); | |
979 | /* | |
980 | * we must use list_del_init so people can check to see if they | |
981 | * are still on the list after taking the semaphore | |
982 | */ | |
983 | list_del_init(&block_group->list); | |
984 | if (list_empty(&block_group->space_info->block_groups[index])) { | |
985 | kobj = block_group->space_info->block_group_kobjs[index]; | |
986 | block_group->space_info->block_group_kobjs[index] = NULL; | |
987 | clear_avail_alloc_bits(fs_info, block_group->flags); | |
988 | } | |
989 | up_write(&block_group->space_info->groups_sem); | |
990 | clear_incompat_bg_bits(fs_info, block_group->flags); | |
991 | if (kobj) { | |
992 | kobject_del(kobj); | |
993 | kobject_put(kobj); | |
994 | } | |
995 | ||
e3e0520b JB |
996 | if (block_group->cached == BTRFS_CACHE_STARTED) |
997 | btrfs_wait_block_group_cache_done(block_group); | |
7b9c293b JB |
998 | |
999 | write_lock(&fs_info->block_group_cache_lock); | |
1000 | caching_ctl = btrfs_get_caching_control(block_group); | |
1001 | if (!caching_ctl) { | |
1002 | struct btrfs_caching_control *ctl; | |
1003 | ||
1004 | list_for_each_entry(ctl, &fs_info->caching_block_groups, list) { | |
1005 | if (ctl->block_group == block_group) { | |
1006 | caching_ctl = ctl; | |
1007 | refcount_inc(&caching_ctl->count); | |
1008 | break; | |
1009 | } | |
e3e0520b JB |
1010 | } |
1011 | } | |
7b9c293b JB |
1012 | if (caching_ctl) |
1013 | list_del_init(&caching_ctl->list); | |
1014 | write_unlock(&fs_info->block_group_cache_lock); | |
1015 | ||
1016 | if (caching_ctl) { | |
1017 | /* Once for the caching bgs list and once for us. */ | |
1018 | btrfs_put_caching_control(caching_ctl); | |
1019 | btrfs_put_caching_control(caching_ctl); | |
1020 | } | |
e3e0520b JB |
1021 | |
1022 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
1023 | WARN_ON(!list_empty(&block_group->dirty_list)); | |
1024 | WARN_ON(!list_empty(&block_group->io_list)); | |
1025 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
1026 | ||
1027 | btrfs_remove_free_space_cache(block_group); | |
1028 | ||
1029 | spin_lock(&block_group->space_info->lock); | |
1030 | list_del_init(&block_group->ro_list); | |
1031 | ||
1032 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { | |
1033 | WARN_ON(block_group->space_info->total_bytes | |
b3470b5d | 1034 | < block_group->length); |
e3e0520b | 1035 | WARN_ON(block_group->space_info->bytes_readonly |
169e0da9 NA |
1036 | < block_group->length - block_group->zone_unusable); |
1037 | WARN_ON(block_group->space_info->bytes_zone_unusable | |
1038 | < block_group->zone_unusable); | |
e3e0520b | 1039 | WARN_ON(block_group->space_info->disk_total |
b3470b5d | 1040 | < block_group->length * factor); |
3349b57f JB |
1041 | WARN_ON(test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, |
1042 | &block_group->runtime_flags) && | |
6a921de5 NA |
1043 | block_group->space_info->active_total_bytes |
1044 | < block_group->length); | |
e3e0520b | 1045 | } |
b3470b5d | 1046 | block_group->space_info->total_bytes -= block_group->length; |
3349b57f | 1047 | if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags)) |
6a921de5 | 1048 | block_group->space_info->active_total_bytes -= block_group->length; |
169e0da9 NA |
1049 | block_group->space_info->bytes_readonly -= |
1050 | (block_group->length - block_group->zone_unusable); | |
1051 | block_group->space_info->bytes_zone_unusable -= | |
1052 | block_group->zone_unusable; | |
b3470b5d | 1053 | block_group->space_info->disk_total -= block_group->length * factor; |
e3e0520b JB |
1054 | |
1055 | spin_unlock(&block_group->space_info->lock); | |
1056 | ||
ffcb9d44 FM |
1057 | /* |
1058 | * Remove the free space for the block group from the free space tree | |
1059 | * and the block group's item from the extent tree before marking the | |
1060 | * block group as removed. This is to prevent races with tasks that | |
1061 | * freeze and unfreeze a block group, this task and another task | |
1062 | * allocating a new block group - the unfreeze task ends up removing | |
1063 | * the block group's extent map before the task calling this function | |
1064 | * deletes the block group item from the extent tree, allowing for | |
1065 | * another task to attempt to create another block group with the same | |
1066 | * item key (and failing with -EEXIST and a transaction abort). | |
1067 | */ | |
1068 | ret = remove_block_group_free_space(trans, block_group); | |
1069 | if (ret) | |
1070 | goto out; | |
1071 | ||
1072 | ret = remove_block_group_item(trans, path, block_group); | |
1073 | if (ret < 0) | |
1074 | goto out; | |
1075 | ||
e3e0520b | 1076 | spin_lock(&block_group->lock); |
3349b57f JB |
1077 | set_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags); |
1078 | ||
e3e0520b | 1079 | /* |
6b7304af FM |
1080 | * At this point trimming or scrub can't start on this block group, |
1081 | * because we removed the block group from the rbtree | |
1082 | * fs_info->block_group_cache_tree so no one can't find it anymore and | |
1083 | * even if someone already got this block group before we removed it | |
1084 | * from the rbtree, they have already incremented block_group->frozen - | |
1085 | * if they didn't, for the trimming case they won't find any free space | |
1086 | * entries because we already removed them all when we called | |
1087 | * btrfs_remove_free_space_cache(). | |
e3e0520b JB |
1088 | * |
1089 | * And we must not remove the extent map from the fs_info->mapping_tree | |
1090 | * to prevent the same logical address range and physical device space | |
6b7304af FM |
1091 | * ranges from being reused for a new block group. This is needed to |
1092 | * avoid races with trimming and scrub. | |
1093 | * | |
1094 | * An fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
e3e0520b JB |
1095 | * completely transactionless, so while it is trimming a range the |
1096 | * currently running transaction might finish and a new one start, | |
1097 | * allowing for new block groups to be created that can reuse the same | |
1098 | * physical device locations unless we take this special care. | |
1099 | * | |
1100 | * There may also be an implicit trim operation if the file system | |
1101 | * is mounted with -odiscard. The same protections must remain | |
1102 | * in place until the extents have been discarded completely when | |
1103 | * the transaction commit has completed. | |
1104 | */ | |
6b7304af | 1105 | remove_em = (atomic_read(&block_group->frozen) == 0); |
e3e0520b JB |
1106 | spin_unlock(&block_group->lock); |
1107 | ||
e3e0520b JB |
1108 | if (remove_em) { |
1109 | struct extent_map_tree *em_tree; | |
1110 | ||
1111 | em_tree = &fs_info->mapping_tree; | |
1112 | write_lock(&em_tree->lock); | |
1113 | remove_extent_mapping(em_tree, em); | |
1114 | write_unlock(&em_tree->lock); | |
1115 | /* once for the tree */ | |
1116 | free_extent_map(em); | |
1117 | } | |
f6033c5e | 1118 | |
9fecd132 | 1119 | out: |
f6033c5e XY |
1120 | /* Once for the lookup reference */ |
1121 | btrfs_put_block_group(block_group); | |
e3e0520b JB |
1122 | if (remove_rsv) |
1123 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1124 | btrfs_free_path(path); | |
1125 | return ret; | |
1126 | } | |
1127 | ||
1128 | struct btrfs_trans_handle *btrfs_start_trans_remove_block_group( | |
1129 | struct btrfs_fs_info *fs_info, const u64 chunk_offset) | |
1130 | { | |
dfe8aec4 | 1131 | struct btrfs_root *root = btrfs_block_group_root(fs_info); |
e3e0520b JB |
1132 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
1133 | struct extent_map *em; | |
1134 | struct map_lookup *map; | |
1135 | unsigned int num_items; | |
1136 | ||
1137 | read_lock(&em_tree->lock); | |
1138 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
1139 | read_unlock(&em_tree->lock); | |
1140 | ASSERT(em && em->start == chunk_offset); | |
1141 | ||
1142 | /* | |
1143 | * We need to reserve 3 + N units from the metadata space info in order | |
1144 | * to remove a block group (done at btrfs_remove_chunk() and at | |
1145 | * btrfs_remove_block_group()), which are used for: | |
1146 | * | |
1147 | * 1 unit for adding the free space inode's orphan (located in the tree | |
1148 | * of tree roots). | |
1149 | * 1 unit for deleting the block group item (located in the extent | |
1150 | * tree). | |
1151 | * 1 unit for deleting the free space item (located in tree of tree | |
1152 | * roots). | |
1153 | * N units for deleting N device extent items corresponding to each | |
1154 | * stripe (located in the device tree). | |
1155 | * | |
1156 | * In order to remove a block group we also need to reserve units in the | |
1157 | * system space info in order to update the chunk tree (update one or | |
1158 | * more device items and remove one chunk item), but this is done at | |
1159 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
1160 | */ | |
1161 | map = em->map_lookup; | |
1162 | num_items = 3 + map->num_stripes; | |
1163 | free_extent_map(em); | |
1164 | ||
dfe8aec4 | 1165 | return btrfs_start_transaction_fallback_global_rsv(root, num_items); |
e3e0520b JB |
1166 | } |
1167 | ||
26ce2095 JB |
1168 | /* |
1169 | * Mark block group @cache read-only, so later write won't happen to block | |
1170 | * group @cache. | |
1171 | * | |
1172 | * If @force is not set, this function will only mark the block group readonly | |
1173 | * if we have enough free space (1M) in other metadata/system block groups. | |
1174 | * If @force is not set, this function will mark the block group readonly | |
1175 | * without checking free space. | |
1176 | * | |
1177 | * NOTE: This function doesn't care if other block groups can contain all the | |
1178 | * data in this block group. That check should be done by relocation routine, | |
1179 | * not this function. | |
1180 | */ | |
32da5386 | 1181 | static int inc_block_group_ro(struct btrfs_block_group *cache, int force) |
26ce2095 JB |
1182 | { |
1183 | struct btrfs_space_info *sinfo = cache->space_info; | |
1184 | u64 num_bytes; | |
26ce2095 JB |
1185 | int ret = -ENOSPC; |
1186 | ||
26ce2095 JB |
1187 | spin_lock(&sinfo->lock); |
1188 | spin_lock(&cache->lock); | |
1189 | ||
195a49ea FM |
1190 | if (cache->swap_extents) { |
1191 | ret = -ETXTBSY; | |
1192 | goto out; | |
1193 | } | |
1194 | ||
26ce2095 JB |
1195 | if (cache->ro) { |
1196 | cache->ro++; | |
1197 | ret = 0; | |
1198 | goto out; | |
1199 | } | |
1200 | ||
b3470b5d | 1201 | num_bytes = cache->length - cache->reserved - cache->pinned - |
169e0da9 | 1202 | cache->bytes_super - cache->zone_unusable - cache->used; |
26ce2095 JB |
1203 | |
1204 | /* | |
a30a3d20 JB |
1205 | * Data never overcommits, even in mixed mode, so do just the straight |
1206 | * check of left over space in how much we have allocated. | |
26ce2095 | 1207 | */ |
a30a3d20 JB |
1208 | if (force) { |
1209 | ret = 0; | |
1210 | } else if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) { | |
1211 | u64 sinfo_used = btrfs_space_info_used(sinfo, true); | |
1212 | ||
1213 | /* | |
1214 | * Here we make sure if we mark this bg RO, we still have enough | |
1215 | * free space as buffer. | |
1216 | */ | |
1217 | if (sinfo_used + num_bytes <= sinfo->total_bytes) | |
1218 | ret = 0; | |
1219 | } else { | |
1220 | /* | |
1221 | * We overcommit metadata, so we need to do the | |
1222 | * btrfs_can_overcommit check here, and we need to pass in | |
1223 | * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of | |
1224 | * leeway to allow us to mark this block group as read only. | |
1225 | */ | |
1226 | if (btrfs_can_overcommit(cache->fs_info, sinfo, num_bytes, | |
1227 | BTRFS_RESERVE_NO_FLUSH)) | |
1228 | ret = 0; | |
1229 | } | |
1230 | ||
1231 | if (!ret) { | |
26ce2095 | 1232 | sinfo->bytes_readonly += num_bytes; |
169e0da9 NA |
1233 | if (btrfs_is_zoned(cache->fs_info)) { |
1234 | /* Migrate zone_unusable bytes to readonly */ | |
1235 | sinfo->bytes_readonly += cache->zone_unusable; | |
1236 | sinfo->bytes_zone_unusable -= cache->zone_unusable; | |
1237 | cache->zone_unusable = 0; | |
1238 | } | |
26ce2095 JB |
1239 | cache->ro++; |
1240 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); | |
26ce2095 JB |
1241 | } |
1242 | out: | |
1243 | spin_unlock(&cache->lock); | |
1244 | spin_unlock(&sinfo->lock); | |
1245 | if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { | |
1246 | btrfs_info(cache->fs_info, | |
b3470b5d | 1247 | "unable to make block group %llu ro", cache->start); |
26ce2095 JB |
1248 | btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0); |
1249 | } | |
1250 | return ret; | |
1251 | } | |
1252 | ||
fe119a6e NB |
1253 | static bool clean_pinned_extents(struct btrfs_trans_handle *trans, |
1254 | struct btrfs_block_group *bg) | |
45bb5d6a NB |
1255 | { |
1256 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
fe119a6e | 1257 | struct btrfs_transaction *prev_trans = NULL; |
45bb5d6a NB |
1258 | const u64 start = bg->start; |
1259 | const u64 end = start + bg->length - 1; | |
1260 | int ret; | |
1261 | ||
fe119a6e NB |
1262 | spin_lock(&fs_info->trans_lock); |
1263 | if (trans->transaction->list.prev != &fs_info->trans_list) { | |
1264 | prev_trans = list_last_entry(&trans->transaction->list, | |
1265 | struct btrfs_transaction, list); | |
1266 | refcount_inc(&prev_trans->use_count); | |
1267 | } | |
1268 | spin_unlock(&fs_info->trans_lock); | |
1269 | ||
45bb5d6a NB |
1270 | /* |
1271 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
1272 | * btrfs_finish_extent_commit(). If we are at transaction N, another | |
1273 | * task might be running finish_extent_commit() for the previous | |
1274 | * transaction N - 1, and have seen a range belonging to the block | |
fe119a6e NB |
1275 | * group in pinned_extents before we were able to clear the whole block |
1276 | * group range from pinned_extents. This means that task can lookup for | |
1277 | * the block group after we unpinned it from pinned_extents and removed | |
1278 | * it, leading to a BUG_ON() at unpin_extent_range(). | |
45bb5d6a NB |
1279 | */ |
1280 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
fe119a6e NB |
1281 | if (prev_trans) { |
1282 | ret = clear_extent_bits(&prev_trans->pinned_extents, start, end, | |
1283 | EXTENT_DIRTY); | |
1284 | if (ret) | |
534cf531 | 1285 | goto out; |
fe119a6e | 1286 | } |
45bb5d6a | 1287 | |
fe119a6e | 1288 | ret = clear_extent_bits(&trans->transaction->pinned_extents, start, end, |
45bb5d6a | 1289 | EXTENT_DIRTY); |
534cf531 | 1290 | out: |
45bb5d6a | 1291 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
5150bf19 FM |
1292 | if (prev_trans) |
1293 | btrfs_put_transaction(prev_trans); | |
45bb5d6a | 1294 | |
534cf531 | 1295 | return ret == 0; |
45bb5d6a NB |
1296 | } |
1297 | ||
e3e0520b JB |
1298 | /* |
1299 | * Process the unused_bgs list and remove any that don't have any allocated | |
1300 | * space inside of them. | |
1301 | */ | |
1302 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
1303 | { | |
32da5386 | 1304 | struct btrfs_block_group *block_group; |
e3e0520b JB |
1305 | struct btrfs_space_info *space_info; |
1306 | struct btrfs_trans_handle *trans; | |
6e80d4f8 | 1307 | const bool async_trim_enabled = btrfs_test_opt(fs_info, DISCARD_ASYNC); |
e3e0520b JB |
1308 | int ret = 0; |
1309 | ||
1310 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) | |
1311 | return; | |
1312 | ||
2f12741f JB |
1313 | if (btrfs_fs_closing(fs_info)) |
1314 | return; | |
1315 | ||
ddfd08cb JB |
1316 | /* |
1317 | * Long running balances can keep us blocked here for eternity, so | |
1318 | * simply skip deletion if we're unable to get the mutex. | |
1319 | */ | |
f3372065 | 1320 | if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) |
ddfd08cb JB |
1321 | return; |
1322 | ||
e3e0520b JB |
1323 | spin_lock(&fs_info->unused_bgs_lock); |
1324 | while (!list_empty(&fs_info->unused_bgs)) { | |
e3e0520b JB |
1325 | int trimming; |
1326 | ||
1327 | block_group = list_first_entry(&fs_info->unused_bgs, | |
32da5386 | 1328 | struct btrfs_block_group, |
e3e0520b JB |
1329 | bg_list); |
1330 | list_del_init(&block_group->bg_list); | |
1331 | ||
1332 | space_info = block_group->space_info; | |
1333 | ||
1334 | if (ret || btrfs_mixed_space_info(space_info)) { | |
1335 | btrfs_put_block_group(block_group); | |
1336 | continue; | |
1337 | } | |
1338 | spin_unlock(&fs_info->unused_bgs_lock); | |
1339 | ||
b0643e59 DZ |
1340 | btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); |
1341 | ||
e3e0520b JB |
1342 | /* Don't want to race with allocators so take the groups_sem */ |
1343 | down_write(&space_info->groups_sem); | |
6e80d4f8 DZ |
1344 | |
1345 | /* | |
1346 | * Async discard moves the final block group discard to be prior | |
1347 | * to the unused_bgs code path. Therefore, if it's not fully | |
1348 | * trimmed, punt it back to the async discard lists. | |
1349 | */ | |
1350 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC) && | |
1351 | !btrfs_is_free_space_trimmed(block_group)) { | |
1352 | trace_btrfs_skip_unused_block_group(block_group); | |
1353 | up_write(&space_info->groups_sem); | |
1354 | /* Requeue if we failed because of async discard */ | |
1355 | btrfs_discard_queue_work(&fs_info->discard_ctl, | |
1356 | block_group); | |
1357 | goto next; | |
1358 | } | |
1359 | ||
e3e0520b JB |
1360 | spin_lock(&block_group->lock); |
1361 | if (block_group->reserved || block_group->pinned || | |
bf38be65 | 1362 | block_group->used || block_group->ro || |
e3e0520b JB |
1363 | list_is_singular(&block_group->list)) { |
1364 | /* | |
1365 | * We want to bail if we made new allocations or have | |
1366 | * outstanding allocations in this block group. We do | |
1367 | * the ro check in case balance is currently acting on | |
1368 | * this block group. | |
1369 | */ | |
1370 | trace_btrfs_skip_unused_block_group(block_group); | |
1371 | spin_unlock(&block_group->lock); | |
1372 | up_write(&space_info->groups_sem); | |
1373 | goto next; | |
1374 | } | |
1375 | spin_unlock(&block_group->lock); | |
1376 | ||
1377 | /* We don't want to force the issue, only flip if it's ok. */ | |
e11c0406 | 1378 | ret = inc_block_group_ro(block_group, 0); |
e3e0520b JB |
1379 | up_write(&space_info->groups_sem); |
1380 | if (ret < 0) { | |
1381 | ret = 0; | |
1382 | goto next; | |
1383 | } | |
1384 | ||
74e91b12 NA |
1385 | ret = btrfs_zone_finish(block_group); |
1386 | if (ret < 0) { | |
1387 | btrfs_dec_block_group_ro(block_group); | |
1388 | if (ret == -EAGAIN) | |
1389 | ret = 0; | |
1390 | goto next; | |
1391 | } | |
1392 | ||
e3e0520b JB |
1393 | /* |
1394 | * Want to do this before we do anything else so we can recover | |
1395 | * properly if we fail to join the transaction. | |
1396 | */ | |
1397 | trans = btrfs_start_trans_remove_block_group(fs_info, | |
b3470b5d | 1398 | block_group->start); |
e3e0520b JB |
1399 | if (IS_ERR(trans)) { |
1400 | btrfs_dec_block_group_ro(block_group); | |
1401 | ret = PTR_ERR(trans); | |
1402 | goto next; | |
1403 | } | |
1404 | ||
1405 | /* | |
1406 | * We could have pending pinned extents for this block group, | |
1407 | * just delete them, we don't care about them anymore. | |
1408 | */ | |
534cf531 FM |
1409 | if (!clean_pinned_extents(trans, block_group)) { |
1410 | btrfs_dec_block_group_ro(block_group); | |
e3e0520b | 1411 | goto end_trans; |
534cf531 | 1412 | } |
e3e0520b | 1413 | |
b0643e59 DZ |
1414 | /* |
1415 | * At this point, the block_group is read only and should fail | |
1416 | * new allocations. However, btrfs_finish_extent_commit() can | |
1417 | * cause this block_group to be placed back on the discard | |
1418 | * lists because now the block_group isn't fully discarded. | |
1419 | * Bail here and try again later after discarding everything. | |
1420 | */ | |
1421 | spin_lock(&fs_info->discard_ctl.lock); | |
1422 | if (!list_empty(&block_group->discard_list)) { | |
1423 | spin_unlock(&fs_info->discard_ctl.lock); | |
1424 | btrfs_dec_block_group_ro(block_group); | |
1425 | btrfs_discard_queue_work(&fs_info->discard_ctl, | |
1426 | block_group); | |
1427 | goto end_trans; | |
1428 | } | |
1429 | spin_unlock(&fs_info->discard_ctl.lock); | |
1430 | ||
e3e0520b JB |
1431 | /* Reset pinned so btrfs_put_block_group doesn't complain */ |
1432 | spin_lock(&space_info->lock); | |
1433 | spin_lock(&block_group->lock); | |
1434 | ||
1435 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, | |
1436 | -block_group->pinned); | |
1437 | space_info->bytes_readonly += block_group->pinned; | |
e3e0520b JB |
1438 | block_group->pinned = 0; |
1439 | ||
1440 | spin_unlock(&block_group->lock); | |
1441 | spin_unlock(&space_info->lock); | |
1442 | ||
6e80d4f8 DZ |
1443 | /* |
1444 | * The normal path here is an unused block group is passed here, | |
1445 | * then trimming is handled in the transaction commit path. | |
1446 | * Async discard interposes before this to do the trimming | |
1447 | * before coming down the unused block group path as trimming | |
1448 | * will no longer be done later in the transaction commit path. | |
1449 | */ | |
1450 | if (!async_trim_enabled && btrfs_test_opt(fs_info, DISCARD_ASYNC)) | |
1451 | goto flip_async; | |
1452 | ||
dcba6e48 NA |
1453 | /* |
1454 | * DISCARD can flip during remount. On zoned filesystems, we | |
1455 | * need to reset sequential-required zones. | |
1456 | */ | |
1457 | trimming = btrfs_test_opt(fs_info, DISCARD_SYNC) || | |
1458 | btrfs_is_zoned(fs_info); | |
e3e0520b JB |
1459 | |
1460 | /* Implicit trim during transaction commit. */ | |
1461 | if (trimming) | |
6b7304af | 1462 | btrfs_freeze_block_group(block_group); |
e3e0520b JB |
1463 | |
1464 | /* | |
1465 | * Btrfs_remove_chunk will abort the transaction if things go | |
1466 | * horribly wrong. | |
1467 | */ | |
b3470b5d | 1468 | ret = btrfs_remove_chunk(trans, block_group->start); |
e3e0520b JB |
1469 | |
1470 | if (ret) { | |
1471 | if (trimming) | |
6b7304af | 1472 | btrfs_unfreeze_block_group(block_group); |
e3e0520b JB |
1473 | goto end_trans; |
1474 | } | |
1475 | ||
1476 | /* | |
1477 | * If we're not mounted with -odiscard, we can just forget | |
1478 | * about this block group. Otherwise we'll need to wait | |
1479 | * until transaction commit to do the actual discard. | |
1480 | */ | |
1481 | if (trimming) { | |
1482 | spin_lock(&fs_info->unused_bgs_lock); | |
1483 | /* | |
1484 | * A concurrent scrub might have added us to the list | |
1485 | * fs_info->unused_bgs, so use a list_move operation | |
1486 | * to add the block group to the deleted_bgs list. | |
1487 | */ | |
1488 | list_move(&block_group->bg_list, | |
1489 | &trans->transaction->deleted_bgs); | |
1490 | spin_unlock(&fs_info->unused_bgs_lock); | |
1491 | btrfs_get_block_group(block_group); | |
1492 | } | |
1493 | end_trans: | |
1494 | btrfs_end_transaction(trans); | |
1495 | next: | |
e3e0520b JB |
1496 | btrfs_put_block_group(block_group); |
1497 | spin_lock(&fs_info->unused_bgs_lock); | |
1498 | } | |
1499 | spin_unlock(&fs_info->unused_bgs_lock); | |
f3372065 | 1500 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
6e80d4f8 DZ |
1501 | return; |
1502 | ||
1503 | flip_async: | |
1504 | btrfs_end_transaction(trans); | |
f3372065 | 1505 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
6e80d4f8 DZ |
1506 | btrfs_put_block_group(block_group); |
1507 | btrfs_discard_punt_unused_bgs_list(fs_info); | |
e3e0520b JB |
1508 | } |
1509 | ||
32da5386 | 1510 | void btrfs_mark_bg_unused(struct btrfs_block_group *bg) |
e3e0520b JB |
1511 | { |
1512 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
1513 | ||
1514 | spin_lock(&fs_info->unused_bgs_lock); | |
1515 | if (list_empty(&bg->bg_list)) { | |
1516 | btrfs_get_block_group(bg); | |
1517 | trace_btrfs_add_unused_block_group(bg); | |
1518 | list_add_tail(&bg->bg_list, &fs_info->unused_bgs); | |
1519 | } | |
1520 | spin_unlock(&fs_info->unused_bgs_lock); | |
1521 | } | |
4358d963 | 1522 | |
2ca0ec77 JT |
1523 | /* |
1524 | * We want block groups with a low number of used bytes to be in the beginning | |
1525 | * of the list, so they will get reclaimed first. | |
1526 | */ | |
1527 | static int reclaim_bgs_cmp(void *unused, const struct list_head *a, | |
1528 | const struct list_head *b) | |
1529 | { | |
1530 | const struct btrfs_block_group *bg1, *bg2; | |
1531 | ||
1532 | bg1 = list_entry(a, struct btrfs_block_group, bg_list); | |
1533 | bg2 = list_entry(b, struct btrfs_block_group, bg_list); | |
1534 | ||
1535 | return bg1->used > bg2->used; | |
1536 | } | |
1537 | ||
3687fcb0 JT |
1538 | static inline bool btrfs_should_reclaim(struct btrfs_fs_info *fs_info) |
1539 | { | |
1540 | if (btrfs_is_zoned(fs_info)) | |
1541 | return btrfs_zoned_should_reclaim(fs_info); | |
1542 | return true; | |
1543 | } | |
1544 | ||
81531225 BB |
1545 | static bool should_reclaim_block_group(struct btrfs_block_group *bg, u64 bytes_freed) |
1546 | { | |
1547 | const struct btrfs_space_info *space_info = bg->space_info; | |
1548 | const int reclaim_thresh = READ_ONCE(space_info->bg_reclaim_threshold); | |
1549 | const u64 new_val = bg->used; | |
1550 | const u64 old_val = new_val + bytes_freed; | |
1551 | u64 thresh; | |
1552 | ||
1553 | if (reclaim_thresh == 0) | |
1554 | return false; | |
1555 | ||
1556 | thresh = div_factor_fine(bg->length, reclaim_thresh); | |
1557 | ||
1558 | /* | |
1559 | * If we were below the threshold before don't reclaim, we are likely a | |
1560 | * brand new block group and we don't want to relocate new block groups. | |
1561 | */ | |
1562 | if (old_val < thresh) | |
1563 | return false; | |
1564 | if (new_val >= thresh) | |
1565 | return false; | |
1566 | return true; | |
1567 | } | |
1568 | ||
18bb8bbf JT |
1569 | void btrfs_reclaim_bgs_work(struct work_struct *work) |
1570 | { | |
1571 | struct btrfs_fs_info *fs_info = | |
1572 | container_of(work, struct btrfs_fs_info, reclaim_bgs_work); | |
1573 | struct btrfs_block_group *bg; | |
1574 | struct btrfs_space_info *space_info; | |
18bb8bbf JT |
1575 | |
1576 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) | |
1577 | return; | |
1578 | ||
2f12741f JB |
1579 | if (btrfs_fs_closing(fs_info)) |
1580 | return; | |
1581 | ||
3687fcb0 JT |
1582 | if (!btrfs_should_reclaim(fs_info)) |
1583 | return; | |
1584 | ||
ca5e4ea0 NA |
1585 | sb_start_write(fs_info->sb); |
1586 | ||
1587 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { | |
1588 | sb_end_write(fs_info->sb); | |
18bb8bbf | 1589 | return; |
ca5e4ea0 | 1590 | } |
18bb8bbf | 1591 | |
9cc0b837 JT |
1592 | /* |
1593 | * Long running balances can keep us blocked here for eternity, so | |
1594 | * simply skip reclaim if we're unable to get the mutex. | |
1595 | */ | |
1596 | if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) { | |
1597 | btrfs_exclop_finish(fs_info); | |
ca5e4ea0 | 1598 | sb_end_write(fs_info->sb); |
9cc0b837 JT |
1599 | return; |
1600 | } | |
1601 | ||
18bb8bbf | 1602 | spin_lock(&fs_info->unused_bgs_lock); |
2ca0ec77 JT |
1603 | /* |
1604 | * Sort happens under lock because we can't simply splice it and sort. | |
1605 | * The block groups might still be in use and reachable via bg_list, | |
1606 | * and their presence in the reclaim_bgs list must be preserved. | |
1607 | */ | |
1608 | list_sort(NULL, &fs_info->reclaim_bgs, reclaim_bgs_cmp); | |
18bb8bbf | 1609 | while (!list_empty(&fs_info->reclaim_bgs)) { |
5f93e776 | 1610 | u64 zone_unusable; |
1cea5cf0 FM |
1611 | int ret = 0; |
1612 | ||
18bb8bbf JT |
1613 | bg = list_first_entry(&fs_info->reclaim_bgs, |
1614 | struct btrfs_block_group, | |
1615 | bg_list); | |
1616 | list_del_init(&bg->bg_list); | |
1617 | ||
1618 | space_info = bg->space_info; | |
1619 | spin_unlock(&fs_info->unused_bgs_lock); | |
1620 | ||
1621 | /* Don't race with allocators so take the groups_sem */ | |
1622 | down_write(&space_info->groups_sem); | |
1623 | ||
1624 | spin_lock(&bg->lock); | |
1625 | if (bg->reserved || bg->pinned || bg->ro) { | |
1626 | /* | |
1627 | * We want to bail if we made new allocations or have | |
1628 | * outstanding allocations in this block group. We do | |
1629 | * the ro check in case balance is currently acting on | |
1630 | * this block group. | |
1631 | */ | |
1632 | spin_unlock(&bg->lock); | |
1633 | up_write(&space_info->groups_sem); | |
1634 | goto next; | |
1635 | } | |
cc4804bf BB |
1636 | if (bg->used == 0) { |
1637 | /* | |
1638 | * It is possible that we trigger relocation on a block | |
1639 | * group as its extents are deleted and it first goes | |
1640 | * below the threshold, then shortly after goes empty. | |
1641 | * | |
1642 | * In this case, relocating it does delete it, but has | |
1643 | * some overhead in relocation specific metadata, looking | |
1644 | * for the non-existent extents and running some extra | |
1645 | * transactions, which we can avoid by using one of the | |
1646 | * other mechanisms for dealing with empty block groups. | |
1647 | */ | |
1648 | if (!btrfs_test_opt(fs_info, DISCARD_ASYNC)) | |
1649 | btrfs_mark_bg_unused(bg); | |
1650 | spin_unlock(&bg->lock); | |
1651 | up_write(&space_info->groups_sem); | |
1652 | goto next; | |
81531225 BB |
1653 | |
1654 | } | |
1655 | /* | |
1656 | * The block group might no longer meet the reclaim condition by | |
1657 | * the time we get around to reclaiming it, so to avoid | |
1658 | * reclaiming overly full block_groups, skip reclaiming them. | |
1659 | * | |
1660 | * Since the decision making process also depends on the amount | |
1661 | * being freed, pass in a fake giant value to skip that extra | |
1662 | * check, which is more meaningful when adding to the list in | |
1663 | * the first place. | |
1664 | */ | |
1665 | if (!should_reclaim_block_group(bg, bg->length)) { | |
1666 | spin_unlock(&bg->lock); | |
1667 | up_write(&space_info->groups_sem); | |
1668 | goto next; | |
cc4804bf | 1669 | } |
18bb8bbf JT |
1670 | spin_unlock(&bg->lock); |
1671 | ||
1672 | /* Get out fast, in case we're unmounting the filesystem */ | |
1673 | if (btrfs_fs_closing(fs_info)) { | |
1674 | up_write(&space_info->groups_sem); | |
1675 | goto next; | |
1676 | } | |
1677 | ||
5f93e776 JT |
1678 | /* |
1679 | * Cache the zone_unusable value before turning the block group | |
1680 | * to read only. As soon as the blog group is read only it's | |
1681 | * zone_unusable value gets moved to the block group's read-only | |
1682 | * bytes and isn't available for calculations anymore. | |
1683 | */ | |
1684 | zone_unusable = bg->zone_unusable; | |
18bb8bbf JT |
1685 | ret = inc_block_group_ro(bg, 0); |
1686 | up_write(&space_info->groups_sem); | |
1687 | if (ret < 0) | |
1688 | goto next; | |
1689 | ||
5f93e776 JT |
1690 | btrfs_info(fs_info, |
1691 | "reclaiming chunk %llu with %llu%% used %llu%% unusable", | |
1692 | bg->start, div_u64(bg->used * 100, bg->length), | |
1693 | div64_u64(zone_unusable * 100, bg->length)); | |
18bb8bbf JT |
1694 | trace_btrfs_reclaim_block_group(bg); |
1695 | ret = btrfs_relocate_chunk(fs_info, bg->start); | |
74944c87 JB |
1696 | if (ret) { |
1697 | btrfs_dec_block_group_ro(bg); | |
18bb8bbf JT |
1698 | btrfs_err(fs_info, "error relocating chunk %llu", |
1699 | bg->start); | |
74944c87 | 1700 | } |
18bb8bbf JT |
1701 | |
1702 | next: | |
d96b3424 | 1703 | btrfs_put_block_group(bg); |
18bb8bbf JT |
1704 | spin_lock(&fs_info->unused_bgs_lock); |
1705 | } | |
1706 | spin_unlock(&fs_info->unused_bgs_lock); | |
1707 | mutex_unlock(&fs_info->reclaim_bgs_lock); | |
1708 | btrfs_exclop_finish(fs_info); | |
ca5e4ea0 | 1709 | sb_end_write(fs_info->sb); |
18bb8bbf JT |
1710 | } |
1711 | ||
1712 | void btrfs_reclaim_bgs(struct btrfs_fs_info *fs_info) | |
1713 | { | |
1714 | spin_lock(&fs_info->unused_bgs_lock); | |
1715 | if (!list_empty(&fs_info->reclaim_bgs)) | |
1716 | queue_work(system_unbound_wq, &fs_info->reclaim_bgs_work); | |
1717 | spin_unlock(&fs_info->unused_bgs_lock); | |
1718 | } | |
1719 | ||
1720 | void btrfs_mark_bg_to_reclaim(struct btrfs_block_group *bg) | |
1721 | { | |
1722 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
1723 | ||
1724 | spin_lock(&fs_info->unused_bgs_lock); | |
1725 | if (list_empty(&bg->bg_list)) { | |
1726 | btrfs_get_block_group(bg); | |
1727 | trace_btrfs_add_reclaim_block_group(bg); | |
1728 | list_add_tail(&bg->bg_list, &fs_info->reclaim_bgs); | |
1729 | } | |
1730 | spin_unlock(&fs_info->unused_bgs_lock); | |
1731 | } | |
1732 | ||
e3ba67a1 JT |
1733 | static int read_bg_from_eb(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
1734 | struct btrfs_path *path) | |
1735 | { | |
1736 | struct extent_map_tree *em_tree; | |
1737 | struct extent_map *em; | |
1738 | struct btrfs_block_group_item bg; | |
1739 | struct extent_buffer *leaf; | |
1740 | int slot; | |
1741 | u64 flags; | |
1742 | int ret = 0; | |
1743 | ||
1744 | slot = path->slots[0]; | |
1745 | leaf = path->nodes[0]; | |
1746 | ||
1747 | em_tree = &fs_info->mapping_tree; | |
1748 | read_lock(&em_tree->lock); | |
1749 | em = lookup_extent_mapping(em_tree, key->objectid, key->offset); | |
1750 | read_unlock(&em_tree->lock); | |
1751 | if (!em) { | |
1752 | btrfs_err(fs_info, | |
1753 | "logical %llu len %llu found bg but no related chunk", | |
1754 | key->objectid, key->offset); | |
1755 | return -ENOENT; | |
1756 | } | |
1757 | ||
1758 | if (em->start != key->objectid || em->len != key->offset) { | |
1759 | btrfs_err(fs_info, | |
1760 | "block group %llu len %llu mismatch with chunk %llu len %llu", | |
1761 | key->objectid, key->offset, em->start, em->len); | |
1762 | ret = -EUCLEAN; | |
1763 | goto out_free_em; | |
1764 | } | |
1765 | ||
1766 | read_extent_buffer(leaf, &bg, btrfs_item_ptr_offset(leaf, slot), | |
1767 | sizeof(bg)); | |
1768 | flags = btrfs_stack_block_group_flags(&bg) & | |
1769 | BTRFS_BLOCK_GROUP_TYPE_MASK; | |
1770 | ||
1771 | if (flags != (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
1772 | btrfs_err(fs_info, | |
1773 | "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", | |
1774 | key->objectid, key->offset, flags, | |
1775 | (BTRFS_BLOCK_GROUP_TYPE_MASK & em->map_lookup->type)); | |
1776 | ret = -EUCLEAN; | |
1777 | } | |
1778 | ||
1779 | out_free_em: | |
1780 | free_extent_map(em); | |
1781 | return ret; | |
1782 | } | |
1783 | ||
4358d963 JB |
1784 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
1785 | struct btrfs_path *path, | |
1786 | struct btrfs_key *key) | |
1787 | { | |
dfe8aec4 | 1788 | struct btrfs_root *root = btrfs_block_group_root(fs_info); |
e3ba67a1 | 1789 | int ret; |
4358d963 | 1790 | struct btrfs_key found_key; |
4358d963 | 1791 | |
36dfbbe2 | 1792 | btrfs_for_each_slot(root, key, &found_key, path, ret) { |
4358d963 JB |
1793 | if (found_key.objectid >= key->objectid && |
1794 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { | |
36dfbbe2 | 1795 | return read_bg_from_eb(fs_info, &found_key, path); |
4358d963 | 1796 | } |
4358d963 | 1797 | } |
4358d963 JB |
1798 | return ret; |
1799 | } | |
1800 | ||
1801 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
1802 | { | |
1803 | u64 extra_flags = chunk_to_extended(flags) & | |
1804 | BTRFS_EXTENDED_PROFILE_MASK; | |
1805 | ||
1806 | write_seqlock(&fs_info->profiles_lock); | |
1807 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
1808 | fs_info->avail_data_alloc_bits |= extra_flags; | |
1809 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
1810 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
1811 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
1812 | fs_info->avail_system_alloc_bits |= extra_flags; | |
1813 | write_sequnlock(&fs_info->profiles_lock); | |
1814 | } | |
1815 | ||
96a14336 | 1816 | /** |
9ee9b979 NB |
1817 | * Map a physical disk address to a list of logical addresses |
1818 | * | |
1819 | * @fs_info: the filesystem | |
96a14336 | 1820 | * @chunk_start: logical address of block group |
138082f3 | 1821 | * @bdev: physical device to resolve, can be NULL to indicate any device |
96a14336 NB |
1822 | * @physical: physical address to map to logical addresses |
1823 | * @logical: return array of logical addresses which map to @physical | |
1824 | * @naddrs: length of @logical | |
1825 | * @stripe_len: size of IO stripe for the given block group | |
1826 | * | |
1827 | * Maps a particular @physical disk address to a list of @logical addresses. | |
1828 | * Used primarily to exclude those portions of a block group that contain super | |
1829 | * block copies. | |
1830 | */ | |
96a14336 | 1831 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, |
138082f3 NA |
1832 | struct block_device *bdev, u64 physical, u64 **logical, |
1833 | int *naddrs, int *stripe_len) | |
96a14336 NB |
1834 | { |
1835 | struct extent_map *em; | |
1836 | struct map_lookup *map; | |
1837 | u64 *buf; | |
1838 | u64 bytenr; | |
1776ad17 NB |
1839 | u64 data_stripe_length; |
1840 | u64 io_stripe_size; | |
1841 | int i, nr = 0; | |
1842 | int ret = 0; | |
96a14336 NB |
1843 | |
1844 | em = btrfs_get_chunk_map(fs_info, chunk_start, 1); | |
1845 | if (IS_ERR(em)) | |
1846 | return -EIO; | |
1847 | ||
1848 | map = em->map_lookup; | |
9e22b925 | 1849 | data_stripe_length = em->orig_block_len; |
1776ad17 | 1850 | io_stripe_size = map->stripe_len; |
138082f3 | 1851 | chunk_start = em->start; |
96a14336 | 1852 | |
9e22b925 NB |
1853 | /* For RAID5/6 adjust to a full IO stripe length */ |
1854 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
1776ad17 | 1855 | io_stripe_size = map->stripe_len * nr_data_stripes(map); |
96a14336 NB |
1856 | |
1857 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); | |
1776ad17 NB |
1858 | if (!buf) { |
1859 | ret = -ENOMEM; | |
1860 | goto out; | |
1861 | } | |
96a14336 NB |
1862 | |
1863 | for (i = 0; i < map->num_stripes; i++) { | |
1776ad17 NB |
1864 | bool already_inserted = false; |
1865 | u64 stripe_nr; | |
138082f3 | 1866 | u64 offset; |
1776ad17 NB |
1867 | int j; |
1868 | ||
1869 | if (!in_range(physical, map->stripes[i].physical, | |
1870 | data_stripe_length)) | |
96a14336 NB |
1871 | continue; |
1872 | ||
138082f3 NA |
1873 | if (bdev && map->stripes[i].dev->bdev != bdev) |
1874 | continue; | |
1875 | ||
96a14336 | 1876 | stripe_nr = physical - map->stripes[i].physical; |
138082f3 | 1877 | stripe_nr = div64_u64_rem(stripe_nr, map->stripe_len, &offset); |
96a14336 | 1878 | |
ac067734 DS |
1879 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
1880 | BTRFS_BLOCK_GROUP_RAID10)) { | |
96a14336 NB |
1881 | stripe_nr = stripe_nr * map->num_stripes + i; |
1882 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); | |
96a14336 NB |
1883 | } |
1884 | /* | |
1885 | * The remaining case would be for RAID56, multiply by | |
1886 | * nr_data_stripes(). Alternatively, just use rmap_len below | |
1887 | * instead of map->stripe_len | |
1888 | */ | |
1889 | ||
138082f3 | 1890 | bytenr = chunk_start + stripe_nr * io_stripe_size + offset; |
1776ad17 NB |
1891 | |
1892 | /* Ensure we don't add duplicate addresses */ | |
96a14336 | 1893 | for (j = 0; j < nr; j++) { |
1776ad17 NB |
1894 | if (buf[j] == bytenr) { |
1895 | already_inserted = true; | |
96a14336 | 1896 | break; |
1776ad17 | 1897 | } |
96a14336 | 1898 | } |
1776ad17 NB |
1899 | |
1900 | if (!already_inserted) | |
96a14336 | 1901 | buf[nr++] = bytenr; |
96a14336 NB |
1902 | } |
1903 | ||
1904 | *logical = buf; | |
1905 | *naddrs = nr; | |
1776ad17 NB |
1906 | *stripe_len = io_stripe_size; |
1907 | out: | |
96a14336 | 1908 | free_extent_map(em); |
1776ad17 | 1909 | return ret; |
96a14336 NB |
1910 | } |
1911 | ||
32da5386 | 1912 | static int exclude_super_stripes(struct btrfs_block_group *cache) |
4358d963 JB |
1913 | { |
1914 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
12659251 | 1915 | const bool zoned = btrfs_is_zoned(fs_info); |
4358d963 JB |
1916 | u64 bytenr; |
1917 | u64 *logical; | |
1918 | int stripe_len; | |
1919 | int i, nr, ret; | |
1920 | ||
b3470b5d DS |
1921 | if (cache->start < BTRFS_SUPER_INFO_OFFSET) { |
1922 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->start; | |
4358d963 | 1923 | cache->bytes_super += stripe_len; |
b3470b5d | 1924 | ret = btrfs_add_excluded_extent(fs_info, cache->start, |
4358d963 JB |
1925 | stripe_len); |
1926 | if (ret) | |
1927 | return ret; | |
1928 | } | |
1929 | ||
1930 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
1931 | bytenr = btrfs_sb_offset(i); | |
138082f3 | 1932 | ret = btrfs_rmap_block(fs_info, cache->start, NULL, |
4358d963 JB |
1933 | bytenr, &logical, &nr, &stripe_len); |
1934 | if (ret) | |
1935 | return ret; | |
1936 | ||
12659251 NA |
1937 | /* Shouldn't have super stripes in sequential zones */ |
1938 | if (zoned && nr) { | |
1939 | btrfs_err(fs_info, | |
1940 | "zoned: block group %llu must not contain super block", | |
1941 | cache->start); | |
1942 | return -EUCLEAN; | |
1943 | } | |
1944 | ||
4358d963 | 1945 | while (nr--) { |
96f9b0f2 NB |
1946 | u64 len = min_t(u64, stripe_len, |
1947 | cache->start + cache->length - logical[nr]); | |
4358d963 JB |
1948 | |
1949 | cache->bytes_super += len; | |
96f9b0f2 NB |
1950 | ret = btrfs_add_excluded_extent(fs_info, logical[nr], |
1951 | len); | |
4358d963 JB |
1952 | if (ret) { |
1953 | kfree(logical); | |
1954 | return ret; | |
1955 | } | |
1956 | } | |
1957 | ||
1958 | kfree(logical); | |
1959 | } | |
1960 | return 0; | |
1961 | } | |
1962 | ||
32da5386 | 1963 | static struct btrfs_block_group *btrfs_create_block_group_cache( |
9afc6649 | 1964 | struct btrfs_fs_info *fs_info, u64 start) |
4358d963 | 1965 | { |
32da5386 | 1966 | struct btrfs_block_group *cache; |
4358d963 JB |
1967 | |
1968 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
1969 | if (!cache) | |
1970 | return NULL; | |
1971 | ||
1972 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
1973 | GFP_NOFS); | |
1974 | if (!cache->free_space_ctl) { | |
1975 | kfree(cache); | |
1976 | return NULL; | |
1977 | } | |
1978 | ||
b3470b5d | 1979 | cache->start = start; |
4358d963 JB |
1980 | |
1981 | cache->fs_info = fs_info; | |
1982 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); | |
4358d963 | 1983 | |
6e80d4f8 DZ |
1984 | cache->discard_index = BTRFS_DISCARD_INDEX_UNUSED; |
1985 | ||
48aaeebe | 1986 | refcount_set(&cache->refs, 1); |
4358d963 JB |
1987 | spin_lock_init(&cache->lock); |
1988 | init_rwsem(&cache->data_rwsem); | |
1989 | INIT_LIST_HEAD(&cache->list); | |
1990 | INIT_LIST_HEAD(&cache->cluster_list); | |
1991 | INIT_LIST_HEAD(&cache->bg_list); | |
1992 | INIT_LIST_HEAD(&cache->ro_list); | |
b0643e59 | 1993 | INIT_LIST_HEAD(&cache->discard_list); |
4358d963 JB |
1994 | INIT_LIST_HEAD(&cache->dirty_list); |
1995 | INIT_LIST_HEAD(&cache->io_list); | |
afba2bc0 | 1996 | INIT_LIST_HEAD(&cache->active_bg_list); |
cd79909b | 1997 | btrfs_init_free_space_ctl(cache, cache->free_space_ctl); |
6b7304af | 1998 | atomic_set(&cache->frozen, 0); |
4358d963 | 1999 | mutex_init(&cache->free_space_lock); |
c29abab4 JB |
2000 | cache->full_stripe_locks_root.root = RB_ROOT; |
2001 | mutex_init(&cache->full_stripe_locks_root.lock); | |
4358d963 JB |
2002 | |
2003 | return cache; | |
2004 | } | |
2005 | ||
2006 | /* | |
2007 | * Iterate all chunks and verify that each of them has the corresponding block | |
2008 | * group | |
2009 | */ | |
2010 | static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) | |
2011 | { | |
2012 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; | |
2013 | struct extent_map *em; | |
32da5386 | 2014 | struct btrfs_block_group *bg; |
4358d963 JB |
2015 | u64 start = 0; |
2016 | int ret = 0; | |
2017 | ||
2018 | while (1) { | |
2019 | read_lock(&map_tree->lock); | |
2020 | /* | |
2021 | * lookup_extent_mapping will return the first extent map | |
2022 | * intersecting the range, so setting @len to 1 is enough to | |
2023 | * get the first chunk. | |
2024 | */ | |
2025 | em = lookup_extent_mapping(map_tree, start, 1); | |
2026 | read_unlock(&map_tree->lock); | |
2027 | if (!em) | |
2028 | break; | |
2029 | ||
2030 | bg = btrfs_lookup_block_group(fs_info, em->start); | |
2031 | if (!bg) { | |
2032 | btrfs_err(fs_info, | |
2033 | "chunk start=%llu len=%llu doesn't have corresponding block group", | |
2034 | em->start, em->len); | |
2035 | ret = -EUCLEAN; | |
2036 | free_extent_map(em); | |
2037 | break; | |
2038 | } | |
b3470b5d | 2039 | if (bg->start != em->start || bg->length != em->len || |
4358d963 JB |
2040 | (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != |
2041 | (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
2042 | btrfs_err(fs_info, | |
2043 | "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", | |
2044 | em->start, em->len, | |
2045 | em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, | |
b3470b5d | 2046 | bg->start, bg->length, |
4358d963 JB |
2047 | bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); |
2048 | ret = -EUCLEAN; | |
2049 | free_extent_map(em); | |
2050 | btrfs_put_block_group(bg); | |
2051 | break; | |
2052 | } | |
2053 | start = em->start + em->len; | |
2054 | free_extent_map(em); | |
2055 | btrfs_put_block_group(bg); | |
2056 | } | |
2057 | return ret; | |
2058 | } | |
2059 | ||
ffb9e0f0 | 2060 | static int read_one_block_group(struct btrfs_fs_info *info, |
4afd2fe8 | 2061 | struct btrfs_block_group_item *bgi, |
d49a2ddb | 2062 | const struct btrfs_key *key, |
ffb9e0f0 QW |
2063 | int need_clear) |
2064 | { | |
32da5386 | 2065 | struct btrfs_block_group *cache; |
ffb9e0f0 | 2066 | const bool mixed = btrfs_fs_incompat(info, MIXED_GROUPS); |
ffb9e0f0 QW |
2067 | int ret; |
2068 | ||
d49a2ddb | 2069 | ASSERT(key->type == BTRFS_BLOCK_GROUP_ITEM_KEY); |
ffb9e0f0 | 2070 | |
9afc6649 | 2071 | cache = btrfs_create_block_group_cache(info, key->objectid); |
ffb9e0f0 QW |
2072 | if (!cache) |
2073 | return -ENOMEM; | |
2074 | ||
4afd2fe8 JT |
2075 | cache->length = key->offset; |
2076 | cache->used = btrfs_stack_block_group_used(bgi); | |
7248e0ce | 2077 | cache->commit_used = cache->used; |
4afd2fe8 | 2078 | cache->flags = btrfs_stack_block_group_flags(bgi); |
f7238e50 | 2079 | cache->global_root_id = btrfs_stack_block_group_chunk_objectid(bgi); |
9afc6649 | 2080 | |
e3e39c72 MPS |
2081 | set_free_space_tree_thresholds(cache); |
2082 | ||
ffb9e0f0 QW |
2083 | if (need_clear) { |
2084 | /* | |
2085 | * When we mount with old space cache, we need to | |
2086 | * set BTRFS_DC_CLEAR and set dirty flag. | |
2087 | * | |
2088 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
2089 | * truncate the old free space cache inode and | |
2090 | * setup a new one. | |
2091 | * b) Setting 'dirty flag' makes sure that we flush | |
2092 | * the new space cache info onto disk. | |
2093 | */ | |
2094 | if (btrfs_test_opt(info, SPACE_CACHE)) | |
2095 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
2096 | } | |
ffb9e0f0 QW |
2097 | if (!mixed && ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && |
2098 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
2099 | btrfs_err(info, | |
2100 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
2101 | cache->start); | |
2102 | ret = -EINVAL; | |
2103 | goto error; | |
2104 | } | |
2105 | ||
a94794d5 | 2106 | ret = btrfs_load_block_group_zone_info(cache, false); |
08e11a3d NA |
2107 | if (ret) { |
2108 | btrfs_err(info, "zoned: failed to load zone info of bg %llu", | |
2109 | cache->start); | |
2110 | goto error; | |
2111 | } | |
2112 | ||
ffb9e0f0 QW |
2113 | /* |
2114 | * We need to exclude the super stripes now so that the space info has | |
2115 | * super bytes accounted for, otherwise we'll think we have more space | |
2116 | * than we actually do. | |
2117 | */ | |
2118 | ret = exclude_super_stripes(cache); | |
2119 | if (ret) { | |
2120 | /* We may have excluded something, so call this just in case. */ | |
2121 | btrfs_free_excluded_extents(cache); | |
2122 | goto error; | |
2123 | } | |
2124 | ||
2125 | /* | |
169e0da9 NA |
2126 | * For zoned filesystem, space after the allocation offset is the only |
2127 | * free space for a block group. So, we don't need any caching work. | |
2128 | * btrfs_calc_zone_unusable() will set the amount of free space and | |
2129 | * zone_unusable space. | |
2130 | * | |
2131 | * For regular filesystem, check for two cases, either we are full, and | |
2132 | * therefore don't need to bother with the caching work since we won't | |
2133 | * find any space, or we are empty, and we can just add all the space | |
2134 | * in and be done with it. This saves us _a_lot_ of time, particularly | |
2135 | * in the full case. | |
ffb9e0f0 | 2136 | */ |
169e0da9 NA |
2137 | if (btrfs_is_zoned(info)) { |
2138 | btrfs_calc_zone_unusable(cache); | |
c46c4247 NA |
2139 | /* Should not have any excluded extents. Just in case, though. */ |
2140 | btrfs_free_excluded_extents(cache); | |
169e0da9 | 2141 | } else if (cache->length == cache->used) { |
ffb9e0f0 QW |
2142 | cache->cached = BTRFS_CACHE_FINISHED; |
2143 | btrfs_free_excluded_extents(cache); | |
2144 | } else if (cache->used == 0) { | |
ffb9e0f0 | 2145 | cache->cached = BTRFS_CACHE_FINISHED; |
9afc6649 QW |
2146 | add_new_free_space(cache, cache->start, |
2147 | cache->start + cache->length); | |
ffb9e0f0 QW |
2148 | btrfs_free_excluded_extents(cache); |
2149 | } | |
2150 | ||
2151 | ret = btrfs_add_block_group_cache(info, cache); | |
2152 | if (ret) { | |
2153 | btrfs_remove_free_space_cache(cache); | |
2154 | goto error; | |
2155 | } | |
2156 | trace_btrfs_add_block_group(info, cache, 0); | |
723de71d | 2157 | btrfs_add_bg_to_space_info(info, cache); |
ffb9e0f0 QW |
2158 | |
2159 | set_avail_alloc_bits(info, cache->flags); | |
a09f23c3 AJ |
2160 | if (btrfs_chunk_writeable(info, cache->start)) { |
2161 | if (cache->used == 0) { | |
2162 | ASSERT(list_empty(&cache->bg_list)); | |
2163 | if (btrfs_test_opt(info, DISCARD_ASYNC)) | |
2164 | btrfs_discard_queue_work(&info->discard_ctl, cache); | |
2165 | else | |
2166 | btrfs_mark_bg_unused(cache); | |
2167 | } | |
2168 | } else { | |
ffb9e0f0 | 2169 | inc_block_group_ro(cache, 1); |
ffb9e0f0 | 2170 | } |
a09f23c3 | 2171 | |
ffb9e0f0 QW |
2172 | return 0; |
2173 | error: | |
2174 | btrfs_put_block_group(cache); | |
2175 | return ret; | |
2176 | } | |
2177 | ||
42437a63 JB |
2178 | static int fill_dummy_bgs(struct btrfs_fs_info *fs_info) |
2179 | { | |
2180 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; | |
42437a63 JB |
2181 | struct rb_node *node; |
2182 | int ret = 0; | |
2183 | ||
2184 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { | |
2185 | struct extent_map *em; | |
2186 | struct map_lookup *map; | |
2187 | struct btrfs_block_group *bg; | |
2188 | ||
2189 | em = rb_entry(node, struct extent_map, rb_node); | |
2190 | map = em->map_lookup; | |
2191 | bg = btrfs_create_block_group_cache(fs_info, em->start); | |
2192 | if (!bg) { | |
2193 | ret = -ENOMEM; | |
2194 | break; | |
2195 | } | |
2196 | ||
2197 | /* Fill dummy cache as FULL */ | |
2198 | bg->length = em->len; | |
2199 | bg->flags = map->type; | |
42437a63 JB |
2200 | bg->cached = BTRFS_CACHE_FINISHED; |
2201 | bg->used = em->len; | |
2202 | bg->flags = map->type; | |
2203 | ret = btrfs_add_block_group_cache(fs_info, bg); | |
2b29726c QW |
2204 | /* |
2205 | * We may have some valid block group cache added already, in | |
2206 | * that case we skip to the next one. | |
2207 | */ | |
2208 | if (ret == -EEXIST) { | |
2209 | ret = 0; | |
2210 | btrfs_put_block_group(bg); | |
2211 | continue; | |
2212 | } | |
2213 | ||
42437a63 JB |
2214 | if (ret) { |
2215 | btrfs_remove_free_space_cache(bg); | |
2216 | btrfs_put_block_group(bg); | |
2217 | break; | |
2218 | } | |
2b29726c | 2219 | |
723de71d | 2220 | btrfs_add_bg_to_space_info(fs_info, bg); |
42437a63 JB |
2221 | |
2222 | set_avail_alloc_bits(fs_info, bg->flags); | |
2223 | } | |
2224 | if (!ret) | |
2225 | btrfs_init_global_block_rsv(fs_info); | |
2226 | return ret; | |
2227 | } | |
2228 | ||
4358d963 JB |
2229 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
2230 | { | |
dfe8aec4 | 2231 | struct btrfs_root *root = btrfs_block_group_root(info); |
4358d963 JB |
2232 | struct btrfs_path *path; |
2233 | int ret; | |
32da5386 | 2234 | struct btrfs_block_group *cache; |
4358d963 JB |
2235 | struct btrfs_space_info *space_info; |
2236 | struct btrfs_key key; | |
4358d963 JB |
2237 | int need_clear = 0; |
2238 | u64 cache_gen; | |
4358d963 | 2239 | |
81d5d614 QW |
2240 | /* |
2241 | * Either no extent root (with ibadroots rescue option) or we have | |
2242 | * unsupported RO options. The fs can never be mounted read-write, so no | |
2243 | * need to waste time searching block group items. | |
2244 | * | |
2245 | * This also allows new extent tree related changes to be RO compat, | |
2246 | * no need for a full incompat flag. | |
2247 | */ | |
2248 | if (!root || (btrfs_super_compat_ro_flags(info->super_copy) & | |
2249 | ~BTRFS_FEATURE_COMPAT_RO_SUPP)) | |
42437a63 JB |
2250 | return fill_dummy_bgs(info); |
2251 | ||
4358d963 JB |
2252 | key.objectid = 0; |
2253 | key.offset = 0; | |
2254 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
2255 | path = btrfs_alloc_path(); | |
2256 | if (!path) | |
2257 | return -ENOMEM; | |
4358d963 JB |
2258 | |
2259 | cache_gen = btrfs_super_cache_generation(info->super_copy); | |
2260 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
2261 | btrfs_super_generation(info->super_copy) != cache_gen) | |
2262 | need_clear = 1; | |
2263 | if (btrfs_test_opt(info, CLEAR_CACHE)) | |
2264 | need_clear = 1; | |
2265 | ||
2266 | while (1) { | |
4afd2fe8 JT |
2267 | struct btrfs_block_group_item bgi; |
2268 | struct extent_buffer *leaf; | |
2269 | int slot; | |
2270 | ||
4358d963 JB |
2271 | ret = find_first_block_group(info, path, &key); |
2272 | if (ret > 0) | |
2273 | break; | |
2274 | if (ret != 0) | |
2275 | goto error; | |
2276 | ||
4afd2fe8 JT |
2277 | leaf = path->nodes[0]; |
2278 | slot = path->slots[0]; | |
2279 | ||
2280 | read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot), | |
2281 | sizeof(bgi)); | |
2282 | ||
2283 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2284 | btrfs_release_path(path); | |
2285 | ret = read_one_block_group(info, &bgi, &key, need_clear); | |
ffb9e0f0 | 2286 | if (ret < 0) |
4358d963 | 2287 | goto error; |
ffb9e0f0 QW |
2288 | key.objectid += key.offset; |
2289 | key.offset = 0; | |
4358d963 | 2290 | } |
7837fa88 | 2291 | btrfs_release_path(path); |
4358d963 | 2292 | |
72804905 | 2293 | list_for_each_entry(space_info, &info->space_info, list) { |
49ea112d JB |
2294 | int i; |
2295 | ||
2296 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { | |
2297 | if (list_empty(&space_info->block_groups[i])) | |
2298 | continue; | |
2299 | cache = list_first_entry(&space_info->block_groups[i], | |
2300 | struct btrfs_block_group, | |
2301 | list); | |
2302 | btrfs_sysfs_add_block_group_type(cache); | |
2303 | } | |
2304 | ||
4358d963 JB |
2305 | if (!(btrfs_get_alloc_profile(info, space_info->flags) & |
2306 | (BTRFS_BLOCK_GROUP_RAID10 | | |
2307 | BTRFS_BLOCK_GROUP_RAID1_MASK | | |
2308 | BTRFS_BLOCK_GROUP_RAID56_MASK | | |
2309 | BTRFS_BLOCK_GROUP_DUP))) | |
2310 | continue; | |
2311 | /* | |
2312 | * Avoid allocating from un-mirrored block group if there are | |
2313 | * mirrored block groups. | |
2314 | */ | |
2315 | list_for_each_entry(cache, | |
2316 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
2317 | list) | |
e11c0406 | 2318 | inc_block_group_ro(cache, 1); |
4358d963 JB |
2319 | list_for_each_entry(cache, |
2320 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
2321 | list) | |
e11c0406 | 2322 | inc_block_group_ro(cache, 1); |
4358d963 JB |
2323 | } |
2324 | ||
2325 | btrfs_init_global_block_rsv(info); | |
2326 | ret = check_chunk_block_group_mappings(info); | |
2327 | error: | |
2328 | btrfs_free_path(path); | |
2b29726c QW |
2329 | /* |
2330 | * We've hit some error while reading the extent tree, and have | |
2331 | * rescue=ibadroots mount option. | |
2332 | * Try to fill the tree using dummy block groups so that the user can | |
2333 | * continue to mount and grab their data. | |
2334 | */ | |
2335 | if (ret && btrfs_test_opt(info, IGNOREBADROOTS)) | |
2336 | ret = fill_dummy_bgs(info); | |
4358d963 JB |
2337 | return ret; |
2338 | } | |
2339 | ||
79bd3712 FM |
2340 | /* |
2341 | * This function, insert_block_group_item(), belongs to the phase 2 of chunk | |
2342 | * allocation. | |
2343 | * | |
2344 | * See the comment at btrfs_chunk_alloc() for details about the chunk allocation | |
2345 | * phases. | |
2346 | */ | |
97f4728a QW |
2347 | static int insert_block_group_item(struct btrfs_trans_handle *trans, |
2348 | struct btrfs_block_group *block_group) | |
2349 | { | |
2350 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2351 | struct btrfs_block_group_item bgi; | |
dfe8aec4 | 2352 | struct btrfs_root *root = btrfs_block_group_root(fs_info); |
97f4728a QW |
2353 | struct btrfs_key key; |
2354 | ||
2355 | spin_lock(&block_group->lock); | |
2356 | btrfs_set_stack_block_group_used(&bgi, block_group->used); | |
2357 | btrfs_set_stack_block_group_chunk_objectid(&bgi, | |
f7238e50 | 2358 | block_group->global_root_id); |
97f4728a QW |
2359 | btrfs_set_stack_block_group_flags(&bgi, block_group->flags); |
2360 | key.objectid = block_group->start; | |
2361 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
2362 | key.offset = block_group->length; | |
2363 | spin_unlock(&block_group->lock); | |
2364 | ||
97f4728a QW |
2365 | return btrfs_insert_item(trans, root, &key, &bgi, sizeof(bgi)); |
2366 | } | |
2367 | ||
2eadb9e7 NB |
2368 | static int insert_dev_extent(struct btrfs_trans_handle *trans, |
2369 | struct btrfs_device *device, u64 chunk_offset, | |
2370 | u64 start, u64 num_bytes) | |
2371 | { | |
2372 | struct btrfs_fs_info *fs_info = device->fs_info; | |
2373 | struct btrfs_root *root = fs_info->dev_root; | |
2374 | struct btrfs_path *path; | |
2375 | struct btrfs_dev_extent *extent; | |
2376 | struct extent_buffer *leaf; | |
2377 | struct btrfs_key key; | |
2378 | int ret; | |
2379 | ||
2380 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); | |
2381 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); | |
2382 | path = btrfs_alloc_path(); | |
2383 | if (!path) | |
2384 | return -ENOMEM; | |
2385 | ||
2386 | key.objectid = device->devid; | |
2387 | key.type = BTRFS_DEV_EXTENT_KEY; | |
2388 | key.offset = start; | |
2389 | ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*extent)); | |
2390 | if (ret) | |
2391 | goto out; | |
2392 | ||
2393 | leaf = path->nodes[0]; | |
2394 | extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent); | |
2395 | btrfs_set_dev_extent_chunk_tree(leaf, extent, BTRFS_CHUNK_TREE_OBJECTID); | |
2396 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, | |
2397 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
2398 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
2399 | ||
2400 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); | |
2401 | btrfs_mark_buffer_dirty(leaf); | |
2402 | out: | |
2403 | btrfs_free_path(path); | |
2404 | return ret; | |
2405 | } | |
2406 | ||
2407 | /* | |
2408 | * This function belongs to phase 2. | |
2409 | * | |
2410 | * See the comment at btrfs_chunk_alloc() for details about the chunk allocation | |
2411 | * phases. | |
2412 | */ | |
2413 | static int insert_dev_extents(struct btrfs_trans_handle *trans, | |
2414 | u64 chunk_offset, u64 chunk_size) | |
2415 | { | |
2416 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2417 | struct btrfs_device *device; | |
2418 | struct extent_map *em; | |
2419 | struct map_lookup *map; | |
2420 | u64 dev_offset; | |
2421 | u64 stripe_size; | |
2422 | int i; | |
2423 | int ret = 0; | |
2424 | ||
2425 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); | |
2426 | if (IS_ERR(em)) | |
2427 | return PTR_ERR(em); | |
2428 | ||
2429 | map = em->map_lookup; | |
2430 | stripe_size = em->orig_block_len; | |
2431 | ||
2432 | /* | |
2433 | * Take the device list mutex to prevent races with the final phase of | |
2434 | * a device replace operation that replaces the device object associated | |
2435 | * with the map's stripes, because the device object's id can change | |
2436 | * at any time during that final phase of the device replace operation | |
2437 | * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the | |
2438 | * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, | |
2439 | * resulting in persisting a device extent item with such ID. | |
2440 | */ | |
2441 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
2442 | for (i = 0; i < map->num_stripes; i++) { | |
2443 | device = map->stripes[i].dev; | |
2444 | dev_offset = map->stripes[i].physical; | |
2445 | ||
2446 | ret = insert_dev_extent(trans, device, chunk_offset, dev_offset, | |
2447 | stripe_size); | |
2448 | if (ret) | |
2449 | break; | |
2450 | } | |
2451 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
2452 | ||
2453 | free_extent_map(em); | |
2454 | return ret; | |
2455 | } | |
2456 | ||
79bd3712 FM |
2457 | /* |
2458 | * This function, btrfs_create_pending_block_groups(), belongs to the phase 2 of | |
2459 | * chunk allocation. | |
2460 | * | |
2461 | * See the comment at btrfs_chunk_alloc() for details about the chunk allocation | |
2462 | * phases. | |
2463 | */ | |
4358d963 JB |
2464 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
2465 | { | |
2466 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2467 | struct btrfs_block_group *block_group; |
4358d963 JB |
2468 | int ret = 0; |
2469 | ||
4358d963 | 2470 | while (!list_empty(&trans->new_bgs)) { |
49ea112d JB |
2471 | int index; |
2472 | ||
4358d963 | 2473 | block_group = list_first_entry(&trans->new_bgs, |
32da5386 | 2474 | struct btrfs_block_group, |
4358d963 JB |
2475 | bg_list); |
2476 | if (ret) | |
2477 | goto next; | |
2478 | ||
49ea112d JB |
2479 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
2480 | ||
97f4728a | 2481 | ret = insert_block_group_item(trans, block_group); |
4358d963 JB |
2482 | if (ret) |
2483 | btrfs_abort_transaction(trans, ret); | |
3349b57f JB |
2484 | if (!test_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, |
2485 | &block_group->runtime_flags)) { | |
79bd3712 FM |
2486 | mutex_lock(&fs_info->chunk_mutex); |
2487 | ret = btrfs_chunk_alloc_add_chunk_item(trans, block_group); | |
2488 | mutex_unlock(&fs_info->chunk_mutex); | |
2489 | if (ret) | |
2490 | btrfs_abort_transaction(trans, ret); | |
2491 | } | |
2eadb9e7 NB |
2492 | ret = insert_dev_extents(trans, block_group->start, |
2493 | block_group->length); | |
4358d963 JB |
2494 | if (ret) |
2495 | btrfs_abort_transaction(trans, ret); | |
2496 | add_block_group_free_space(trans, block_group); | |
49ea112d JB |
2497 | |
2498 | /* | |
2499 | * If we restriped during balance, we may have added a new raid | |
2500 | * type, so now add the sysfs entries when it is safe to do so. | |
2501 | * We don't have to worry about locking here as it's handled in | |
2502 | * btrfs_sysfs_add_block_group_type. | |
2503 | */ | |
2504 | if (block_group->space_info->block_group_kobjs[index] == NULL) | |
2505 | btrfs_sysfs_add_block_group_type(block_group); | |
2506 | ||
4358d963 JB |
2507 | /* Already aborted the transaction if it failed. */ |
2508 | next: | |
2509 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
2510 | list_del_init(&block_group->bg_list); | |
2511 | } | |
2512 | btrfs_trans_release_chunk_metadata(trans); | |
2513 | } | |
2514 | ||
f7238e50 JB |
2515 | /* |
2516 | * For extent tree v2 we use the block_group_item->chunk_offset to point at our | |
2517 | * global root id. For v1 it's always set to BTRFS_FIRST_CHUNK_TREE_OBJECTID. | |
2518 | */ | |
2519 | static u64 calculate_global_root_id(struct btrfs_fs_info *fs_info, u64 offset) | |
2520 | { | |
2521 | u64 div = SZ_1G; | |
2522 | u64 index; | |
2523 | ||
2524 | if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) | |
2525 | return BTRFS_FIRST_CHUNK_TREE_OBJECTID; | |
2526 | ||
2527 | /* If we have a smaller fs index based on 128MiB. */ | |
2528 | if (btrfs_super_total_bytes(fs_info->super_copy) <= (SZ_1G * 10ULL)) | |
2529 | div = SZ_128M; | |
2530 | ||
2531 | offset = div64_u64(offset, div); | |
2532 | div64_u64_rem(offset, fs_info->nr_global_roots, &index); | |
2533 | return index; | |
2534 | } | |
2535 | ||
79bd3712 FM |
2536 | struct btrfs_block_group *btrfs_make_block_group(struct btrfs_trans_handle *trans, |
2537 | u64 bytes_used, u64 type, | |
2538 | u64 chunk_offset, u64 size) | |
4358d963 JB |
2539 | { |
2540 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2541 | struct btrfs_block_group *cache; |
4358d963 JB |
2542 | int ret; |
2543 | ||
2544 | btrfs_set_log_full_commit(trans); | |
2545 | ||
9afc6649 | 2546 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset); |
4358d963 | 2547 | if (!cache) |
79bd3712 | 2548 | return ERR_PTR(-ENOMEM); |
4358d963 | 2549 | |
9afc6649 | 2550 | cache->length = size; |
e3e39c72 | 2551 | set_free_space_tree_thresholds(cache); |
bf38be65 | 2552 | cache->used = bytes_used; |
4358d963 | 2553 | cache->flags = type; |
4358d963 | 2554 | cache->cached = BTRFS_CACHE_FINISHED; |
f7238e50 JB |
2555 | cache->global_root_id = calculate_global_root_id(fs_info, cache->start); |
2556 | ||
997e3e2e BB |
2557 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
2558 | cache->needs_free_space = 1; | |
08e11a3d | 2559 | |
a94794d5 | 2560 | ret = btrfs_load_block_group_zone_info(cache, true); |
08e11a3d NA |
2561 | if (ret) { |
2562 | btrfs_put_block_group(cache); | |
79bd3712 | 2563 | return ERR_PTR(ret); |
08e11a3d NA |
2564 | } |
2565 | ||
4358d963 JB |
2566 | ret = exclude_super_stripes(cache); |
2567 | if (ret) { | |
2568 | /* We may have excluded something, so call this just in case */ | |
2569 | btrfs_free_excluded_extents(cache); | |
2570 | btrfs_put_block_group(cache); | |
79bd3712 | 2571 | return ERR_PTR(ret); |
4358d963 JB |
2572 | } |
2573 | ||
2574 | add_new_free_space(cache, chunk_offset, chunk_offset + size); | |
2575 | ||
2576 | btrfs_free_excluded_extents(cache); | |
2577 | ||
4358d963 JB |
2578 | /* |
2579 | * Ensure the corresponding space_info object is created and | |
2580 | * assigned to our block group. We want our bg to be added to the rbtree | |
2581 | * with its ->space_info set. | |
2582 | */ | |
2583 | cache->space_info = btrfs_find_space_info(fs_info, cache->flags); | |
2584 | ASSERT(cache->space_info); | |
2585 | ||
2586 | ret = btrfs_add_block_group_cache(fs_info, cache); | |
2587 | if (ret) { | |
2588 | btrfs_remove_free_space_cache(cache); | |
2589 | btrfs_put_block_group(cache); | |
79bd3712 | 2590 | return ERR_PTR(ret); |
4358d963 JB |
2591 | } |
2592 | ||
2593 | /* | |
2594 | * Now that our block group has its ->space_info set and is inserted in | |
2595 | * the rbtree, update the space info's counters. | |
2596 | */ | |
2597 | trace_btrfs_add_block_group(fs_info, cache, 1); | |
723de71d | 2598 | btrfs_add_bg_to_space_info(fs_info, cache); |
4358d963 JB |
2599 | btrfs_update_global_block_rsv(fs_info); |
2600 | ||
9d4b0a12 JB |
2601 | #ifdef CONFIG_BTRFS_DEBUG |
2602 | if (btrfs_should_fragment_free_space(cache)) { | |
2603 | u64 new_bytes_used = size - bytes_used; | |
2604 | ||
2605 | cache->space_info->bytes_used += new_bytes_used >> 1; | |
2606 | fragment_free_space(cache); | |
2607 | } | |
2608 | #endif | |
4358d963 JB |
2609 | |
2610 | list_add_tail(&cache->bg_list, &trans->new_bgs); | |
2611 | trans->delayed_ref_updates++; | |
2612 | btrfs_update_delayed_refs_rsv(trans); | |
2613 | ||
2614 | set_avail_alloc_bits(fs_info, type); | |
79bd3712 | 2615 | return cache; |
4358d963 | 2616 | } |
26ce2095 | 2617 | |
b12de528 QW |
2618 | /* |
2619 | * Mark one block group RO, can be called several times for the same block | |
2620 | * group. | |
2621 | * | |
2622 | * @cache: the destination block group | |
2623 | * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to | |
2624 | * ensure we still have some free space after marking this | |
2625 | * block group RO. | |
2626 | */ | |
2627 | int btrfs_inc_block_group_ro(struct btrfs_block_group *cache, | |
2628 | bool do_chunk_alloc) | |
26ce2095 JB |
2629 | { |
2630 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
2631 | struct btrfs_trans_handle *trans; | |
dfe8aec4 | 2632 | struct btrfs_root *root = btrfs_block_group_root(fs_info); |
26ce2095 JB |
2633 | u64 alloc_flags; |
2634 | int ret; | |
b6e9f16c | 2635 | bool dirty_bg_running; |
26ce2095 | 2636 | |
2d192fc4 QW |
2637 | /* |
2638 | * This can only happen when we are doing read-only scrub on read-only | |
2639 | * mount. | |
2640 | * In that case we should not start a new transaction on read-only fs. | |
2641 | * Thus here we skip all chunk allocations. | |
2642 | */ | |
2643 | if (sb_rdonly(fs_info->sb)) { | |
2644 | mutex_lock(&fs_info->ro_block_group_mutex); | |
2645 | ret = inc_block_group_ro(cache, 0); | |
2646 | mutex_unlock(&fs_info->ro_block_group_mutex); | |
2647 | return ret; | |
2648 | } | |
2649 | ||
b6e9f16c | 2650 | do { |
dfe8aec4 | 2651 | trans = btrfs_join_transaction(root); |
b6e9f16c NB |
2652 | if (IS_ERR(trans)) |
2653 | return PTR_ERR(trans); | |
26ce2095 | 2654 | |
b6e9f16c | 2655 | dirty_bg_running = false; |
26ce2095 | 2656 | |
b6e9f16c NB |
2657 | /* |
2658 | * We're not allowed to set block groups readonly after the dirty | |
2659 | * block group cache has started writing. If it already started, | |
2660 | * back off and let this transaction commit. | |
2661 | */ | |
2662 | mutex_lock(&fs_info->ro_block_group_mutex); | |
2663 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { | |
2664 | u64 transid = trans->transid; | |
26ce2095 | 2665 | |
b6e9f16c NB |
2666 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2667 | btrfs_end_transaction(trans); | |
2668 | ||
2669 | ret = btrfs_wait_for_commit(fs_info, transid); | |
2670 | if (ret) | |
2671 | return ret; | |
2672 | dirty_bg_running = true; | |
2673 | } | |
2674 | } while (dirty_bg_running); | |
26ce2095 | 2675 | |
b12de528 | 2676 | if (do_chunk_alloc) { |
26ce2095 | 2677 | /* |
b12de528 QW |
2678 | * If we are changing raid levels, try to allocate a |
2679 | * corresponding block group with the new raid level. | |
26ce2095 | 2680 | */ |
349e120e | 2681 | alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); |
b12de528 QW |
2682 | if (alloc_flags != cache->flags) { |
2683 | ret = btrfs_chunk_alloc(trans, alloc_flags, | |
2684 | CHUNK_ALLOC_FORCE); | |
2685 | /* | |
2686 | * ENOSPC is allowed here, we may have enough space | |
2687 | * already allocated at the new raid level to carry on | |
2688 | */ | |
2689 | if (ret == -ENOSPC) | |
2690 | ret = 0; | |
2691 | if (ret < 0) | |
2692 | goto out; | |
2693 | } | |
26ce2095 JB |
2694 | } |
2695 | ||
a7a63acc | 2696 | ret = inc_block_group_ro(cache, 0); |
195a49ea | 2697 | if (!do_chunk_alloc || ret == -ETXTBSY) |
b12de528 | 2698 | goto unlock_out; |
26ce2095 JB |
2699 | if (!ret) |
2700 | goto out; | |
2701 | alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags); | |
2702 | ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); | |
2703 | if (ret < 0) | |
2704 | goto out; | |
b6a98021 NA |
2705 | /* |
2706 | * We have allocated a new chunk. We also need to activate that chunk to | |
2707 | * grant metadata tickets for zoned filesystem. | |
2708 | */ | |
2709 | ret = btrfs_zoned_activate_one_bg(fs_info, cache->space_info, true); | |
2710 | if (ret < 0) | |
2711 | goto out; | |
2712 | ||
e11c0406 | 2713 | ret = inc_block_group_ro(cache, 0); |
195a49ea FM |
2714 | if (ret == -ETXTBSY) |
2715 | goto unlock_out; | |
26ce2095 JB |
2716 | out: |
2717 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { | |
349e120e | 2718 | alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); |
26ce2095 JB |
2719 | mutex_lock(&fs_info->chunk_mutex); |
2720 | check_system_chunk(trans, alloc_flags); | |
2721 | mutex_unlock(&fs_info->chunk_mutex); | |
2722 | } | |
b12de528 | 2723 | unlock_out: |
26ce2095 JB |
2724 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2725 | ||
2726 | btrfs_end_transaction(trans); | |
2727 | return ret; | |
2728 | } | |
2729 | ||
32da5386 | 2730 | void btrfs_dec_block_group_ro(struct btrfs_block_group *cache) |
26ce2095 JB |
2731 | { |
2732 | struct btrfs_space_info *sinfo = cache->space_info; | |
2733 | u64 num_bytes; | |
2734 | ||
2735 | BUG_ON(!cache->ro); | |
2736 | ||
2737 | spin_lock(&sinfo->lock); | |
2738 | spin_lock(&cache->lock); | |
2739 | if (!--cache->ro) { | |
169e0da9 NA |
2740 | if (btrfs_is_zoned(cache->fs_info)) { |
2741 | /* Migrate zone_unusable bytes back */ | |
98173255 NA |
2742 | cache->zone_unusable = |
2743 | (cache->alloc_offset - cache->used) + | |
2744 | (cache->length - cache->zone_capacity); | |
169e0da9 NA |
2745 | sinfo->bytes_zone_unusable += cache->zone_unusable; |
2746 | sinfo->bytes_readonly -= cache->zone_unusable; | |
2747 | } | |
f9f28e5b NA |
2748 | num_bytes = cache->length - cache->reserved - |
2749 | cache->pinned - cache->bytes_super - | |
2750 | cache->zone_unusable - cache->used; | |
2751 | sinfo->bytes_readonly -= num_bytes; | |
26ce2095 JB |
2752 | list_del_init(&cache->ro_list); |
2753 | } | |
2754 | spin_unlock(&cache->lock); | |
2755 | spin_unlock(&sinfo->lock); | |
2756 | } | |
77745c05 | 2757 | |
3be4d8ef QW |
2758 | static int update_block_group_item(struct btrfs_trans_handle *trans, |
2759 | struct btrfs_path *path, | |
2760 | struct btrfs_block_group *cache) | |
77745c05 JB |
2761 | { |
2762 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2763 | int ret; | |
dfe8aec4 | 2764 | struct btrfs_root *root = btrfs_block_group_root(fs_info); |
77745c05 JB |
2765 | unsigned long bi; |
2766 | struct extent_buffer *leaf; | |
bf38be65 | 2767 | struct btrfs_block_group_item bgi; |
b3470b5d | 2768 | struct btrfs_key key; |
7248e0ce QW |
2769 | u64 old_commit_used; |
2770 | u64 used; | |
2771 | ||
2772 | /* | |
2773 | * Block group items update can be triggered out of commit transaction | |
2774 | * critical section, thus we need a consistent view of used bytes. | |
2775 | * We cannot use cache->used directly outside of the spin lock, as it | |
2776 | * may be changed. | |
2777 | */ | |
2778 | spin_lock(&cache->lock); | |
2779 | old_commit_used = cache->commit_used; | |
2780 | used = cache->used; | |
2781 | /* No change in used bytes, can safely skip it. */ | |
2782 | if (cache->commit_used == used) { | |
2783 | spin_unlock(&cache->lock); | |
2784 | return 0; | |
2785 | } | |
2786 | cache->commit_used = used; | |
2787 | spin_unlock(&cache->lock); | |
b3470b5d DS |
2788 | |
2789 | key.objectid = cache->start; | |
2790 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
2791 | key.offset = cache->length; | |
77745c05 | 2792 | |
3be4d8ef | 2793 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
77745c05 JB |
2794 | if (ret) { |
2795 | if (ret > 0) | |
2796 | ret = -ENOENT; | |
2797 | goto fail; | |
2798 | } | |
2799 | ||
2800 | leaf = path->nodes[0]; | |
2801 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
7248e0ce | 2802 | btrfs_set_stack_block_group_used(&bgi, used); |
de0dc456 | 2803 | btrfs_set_stack_block_group_chunk_objectid(&bgi, |
f7238e50 | 2804 | cache->global_root_id); |
de0dc456 | 2805 | btrfs_set_stack_block_group_flags(&bgi, cache->flags); |
bf38be65 | 2806 | write_extent_buffer(leaf, &bgi, bi, sizeof(bgi)); |
77745c05 JB |
2807 | btrfs_mark_buffer_dirty(leaf); |
2808 | fail: | |
2809 | btrfs_release_path(path); | |
7248e0ce QW |
2810 | /* We didn't update the block group item, need to revert @commit_used. */ |
2811 | if (ret < 0) { | |
2812 | spin_lock(&cache->lock); | |
2813 | cache->commit_used = old_commit_used; | |
2814 | spin_unlock(&cache->lock); | |
2815 | } | |
77745c05 JB |
2816 | return ret; |
2817 | ||
2818 | } | |
2819 | ||
32da5386 | 2820 | static int cache_save_setup(struct btrfs_block_group *block_group, |
77745c05 JB |
2821 | struct btrfs_trans_handle *trans, |
2822 | struct btrfs_path *path) | |
2823 | { | |
2824 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
2825 | struct btrfs_root *root = fs_info->tree_root; | |
2826 | struct inode *inode = NULL; | |
2827 | struct extent_changeset *data_reserved = NULL; | |
2828 | u64 alloc_hint = 0; | |
2829 | int dcs = BTRFS_DC_ERROR; | |
0044ae11 | 2830 | u64 cache_size = 0; |
77745c05 JB |
2831 | int retries = 0; |
2832 | int ret = 0; | |
2833 | ||
af456a2c BB |
2834 | if (!btrfs_test_opt(fs_info, SPACE_CACHE)) |
2835 | return 0; | |
2836 | ||
77745c05 JB |
2837 | /* |
2838 | * If this block group is smaller than 100 megs don't bother caching the | |
2839 | * block group. | |
2840 | */ | |
b3470b5d | 2841 | if (block_group->length < (100 * SZ_1M)) { |
77745c05 JB |
2842 | spin_lock(&block_group->lock); |
2843 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2844 | spin_unlock(&block_group->lock); | |
2845 | return 0; | |
2846 | } | |
2847 | ||
bf31f87f | 2848 | if (TRANS_ABORTED(trans)) |
77745c05 JB |
2849 | return 0; |
2850 | again: | |
2851 | inode = lookup_free_space_inode(block_group, path); | |
2852 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2853 | ret = PTR_ERR(inode); | |
2854 | btrfs_release_path(path); | |
2855 | goto out; | |
2856 | } | |
2857 | ||
2858 | if (IS_ERR(inode)) { | |
2859 | BUG_ON(retries); | |
2860 | retries++; | |
2861 | ||
2862 | if (block_group->ro) | |
2863 | goto out_free; | |
2864 | ||
2865 | ret = create_free_space_inode(trans, block_group, path); | |
2866 | if (ret) | |
2867 | goto out_free; | |
2868 | goto again; | |
2869 | } | |
2870 | ||
2871 | /* | |
2872 | * We want to set the generation to 0, that way if anything goes wrong | |
2873 | * from here on out we know not to trust this cache when we load up next | |
2874 | * time. | |
2875 | */ | |
2876 | BTRFS_I(inode)->generation = 0; | |
9a56fcd1 | 2877 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
77745c05 JB |
2878 | if (ret) { |
2879 | /* | |
2880 | * So theoretically we could recover from this, simply set the | |
2881 | * super cache generation to 0 so we know to invalidate the | |
2882 | * cache, but then we'd have to keep track of the block groups | |
2883 | * that fail this way so we know we _have_ to reset this cache | |
2884 | * before the next commit or risk reading stale cache. So to | |
2885 | * limit our exposure to horrible edge cases lets just abort the | |
2886 | * transaction, this only happens in really bad situations | |
2887 | * anyway. | |
2888 | */ | |
2889 | btrfs_abort_transaction(trans, ret); | |
2890 | goto out_put; | |
2891 | } | |
2892 | WARN_ON(ret); | |
2893 | ||
2894 | /* We've already setup this transaction, go ahead and exit */ | |
2895 | if (block_group->cache_generation == trans->transid && | |
2896 | i_size_read(inode)) { | |
2897 | dcs = BTRFS_DC_SETUP; | |
2898 | goto out_put; | |
2899 | } | |
2900 | ||
2901 | if (i_size_read(inode) > 0) { | |
2902 | ret = btrfs_check_trunc_cache_free_space(fs_info, | |
2903 | &fs_info->global_block_rsv); | |
2904 | if (ret) | |
2905 | goto out_put; | |
2906 | ||
2907 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); | |
2908 | if (ret) | |
2909 | goto out_put; | |
2910 | } | |
2911 | ||
2912 | spin_lock(&block_group->lock); | |
2913 | if (block_group->cached != BTRFS_CACHE_FINISHED || | |
2914 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { | |
2915 | /* | |
2916 | * don't bother trying to write stuff out _if_ | |
2917 | * a) we're not cached, | |
2918 | * b) we're with nospace_cache mount option, | |
2919 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
2920 | */ | |
2921 | dcs = BTRFS_DC_WRITTEN; | |
2922 | spin_unlock(&block_group->lock); | |
2923 | goto out_put; | |
2924 | } | |
2925 | spin_unlock(&block_group->lock); | |
2926 | ||
2927 | /* | |
2928 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
2929 | * skip doing the setup, we've already cleared the cache so we're safe. | |
2930 | */ | |
2931 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
2932 | ret = -ENOSPC; | |
2933 | goto out_put; | |
2934 | } | |
2935 | ||
2936 | /* | |
2937 | * Try to preallocate enough space based on how big the block group is. | |
2938 | * Keep in mind this has to include any pinned space which could end up | |
2939 | * taking up quite a bit since it's not folded into the other space | |
2940 | * cache. | |
2941 | */ | |
0044ae11 QW |
2942 | cache_size = div_u64(block_group->length, SZ_256M); |
2943 | if (!cache_size) | |
2944 | cache_size = 1; | |
77745c05 | 2945 | |
0044ae11 QW |
2946 | cache_size *= 16; |
2947 | cache_size *= fs_info->sectorsize; | |
77745c05 | 2948 | |
36ea6f3e | 2949 | ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved, 0, |
1daedb1d | 2950 | cache_size, false); |
77745c05 JB |
2951 | if (ret) |
2952 | goto out_put; | |
2953 | ||
0044ae11 QW |
2954 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, cache_size, |
2955 | cache_size, cache_size, | |
77745c05 JB |
2956 | &alloc_hint); |
2957 | /* | |
2958 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
2959 | * of metadata or split extents when writing the cache out, which means | |
2960 | * we can enospc if we are heavily fragmented in addition to just normal | |
2961 | * out of space conditions. So if we hit this just skip setting up any | |
2962 | * other block groups for this transaction, maybe we'll unpin enough | |
2963 | * space the next time around. | |
2964 | */ | |
2965 | if (!ret) | |
2966 | dcs = BTRFS_DC_SETUP; | |
2967 | else if (ret == -ENOSPC) | |
2968 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
2969 | ||
2970 | out_put: | |
2971 | iput(inode); | |
2972 | out_free: | |
2973 | btrfs_release_path(path); | |
2974 | out: | |
2975 | spin_lock(&block_group->lock); | |
2976 | if (!ret && dcs == BTRFS_DC_SETUP) | |
2977 | block_group->cache_generation = trans->transid; | |
2978 | block_group->disk_cache_state = dcs; | |
2979 | spin_unlock(&block_group->lock); | |
2980 | ||
2981 | extent_changeset_free(data_reserved); | |
2982 | return ret; | |
2983 | } | |
2984 | ||
2985 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) | |
2986 | { | |
2987 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 2988 | struct btrfs_block_group *cache, *tmp; |
77745c05 JB |
2989 | struct btrfs_transaction *cur_trans = trans->transaction; |
2990 | struct btrfs_path *path; | |
2991 | ||
2992 | if (list_empty(&cur_trans->dirty_bgs) || | |
2993 | !btrfs_test_opt(fs_info, SPACE_CACHE)) | |
2994 | return 0; | |
2995 | ||
2996 | path = btrfs_alloc_path(); | |
2997 | if (!path) | |
2998 | return -ENOMEM; | |
2999 | ||
3000 | /* Could add new block groups, use _safe just in case */ | |
3001 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3002 | dirty_list) { | |
3003 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3004 | cache_save_setup(cache, trans, path); | |
3005 | } | |
3006 | ||
3007 | btrfs_free_path(path); | |
3008 | return 0; | |
3009 | } | |
3010 | ||
3011 | /* | |
3012 | * Transaction commit does final block group cache writeback during a critical | |
3013 | * section where nothing is allowed to change the FS. This is required in | |
3014 | * order for the cache to actually match the block group, but can introduce a | |
3015 | * lot of latency into the commit. | |
3016 | * | |
3017 | * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO. | |
3018 | * There's a chance we'll have to redo some of it if the block group changes | |
3019 | * again during the commit, but it greatly reduces the commit latency by | |
3020 | * getting rid of the easy block groups while we're still allowing others to | |
3021 | * join the commit. | |
3022 | */ | |
3023 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) | |
3024 | { | |
3025 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 3026 | struct btrfs_block_group *cache; |
77745c05 JB |
3027 | struct btrfs_transaction *cur_trans = trans->transaction; |
3028 | int ret = 0; | |
3029 | int should_put; | |
3030 | struct btrfs_path *path = NULL; | |
3031 | LIST_HEAD(dirty); | |
3032 | struct list_head *io = &cur_trans->io_bgs; | |
77745c05 JB |
3033 | int loops = 0; |
3034 | ||
3035 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3036 | if (list_empty(&cur_trans->dirty_bgs)) { | |
3037 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3038 | return 0; | |
3039 | } | |
3040 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
3041 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3042 | ||
3043 | again: | |
3044 | /* Make sure all the block groups on our dirty list actually exist */ | |
3045 | btrfs_create_pending_block_groups(trans); | |
3046 | ||
3047 | if (!path) { | |
3048 | path = btrfs_alloc_path(); | |
938fcbfb JB |
3049 | if (!path) { |
3050 | ret = -ENOMEM; | |
3051 | goto out; | |
3052 | } | |
77745c05 JB |
3053 | } |
3054 | ||
3055 | /* | |
3056 | * cache_write_mutex is here only to save us from balance or automatic | |
3057 | * removal of empty block groups deleting this block group while we are | |
3058 | * writing out the cache | |
3059 | */ | |
3060 | mutex_lock(&trans->transaction->cache_write_mutex); | |
3061 | while (!list_empty(&dirty)) { | |
3062 | bool drop_reserve = true; | |
3063 | ||
32da5386 | 3064 | cache = list_first_entry(&dirty, struct btrfs_block_group, |
77745c05 JB |
3065 | dirty_list); |
3066 | /* | |
3067 | * This can happen if something re-dirties a block group that | |
3068 | * is already under IO. Just wait for it to finish and then do | |
3069 | * it all again | |
3070 | */ | |
3071 | if (!list_empty(&cache->io_list)) { | |
3072 | list_del_init(&cache->io_list); | |
3073 | btrfs_wait_cache_io(trans, cache, path); | |
3074 | btrfs_put_block_group(cache); | |
3075 | } | |
3076 | ||
3077 | ||
3078 | /* | |
3079 | * btrfs_wait_cache_io uses the cache->dirty_list to decide if | |
3080 | * it should update the cache_state. Don't delete until after | |
3081 | * we wait. | |
3082 | * | |
3083 | * Since we're not running in the commit critical section | |
3084 | * we need the dirty_bgs_lock to protect from update_block_group | |
3085 | */ | |
3086 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3087 | list_del_init(&cache->dirty_list); | |
3088 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3089 | ||
3090 | should_put = 1; | |
3091 | ||
3092 | cache_save_setup(cache, trans, path); | |
3093 | ||
3094 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3095 | cache->io_ctl.inode = NULL; | |
3096 | ret = btrfs_write_out_cache(trans, cache, path); | |
3097 | if (ret == 0 && cache->io_ctl.inode) { | |
77745c05 JB |
3098 | should_put = 0; |
3099 | ||
3100 | /* | |
3101 | * The cache_write_mutex is protecting the | |
3102 | * io_list, also refer to the definition of | |
3103 | * btrfs_transaction::io_bgs for more details | |
3104 | */ | |
3105 | list_add_tail(&cache->io_list, io); | |
3106 | } else { | |
3107 | /* | |
3108 | * If we failed to write the cache, the | |
3109 | * generation will be bad and life goes on | |
3110 | */ | |
3111 | ret = 0; | |
3112 | } | |
3113 | } | |
3114 | if (!ret) { | |
3be4d8ef | 3115 | ret = update_block_group_item(trans, path, cache); |
77745c05 JB |
3116 | /* |
3117 | * Our block group might still be attached to the list | |
3118 | * of new block groups in the transaction handle of some | |
3119 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3120 | * means its block group item isn't yet in the extent | |
3121 | * tree. If this happens ignore the error, as we will | |
3122 | * try again later in the critical section of the | |
3123 | * transaction commit. | |
3124 | */ | |
3125 | if (ret == -ENOENT) { | |
3126 | ret = 0; | |
3127 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3128 | if (list_empty(&cache->dirty_list)) { | |
3129 | list_add_tail(&cache->dirty_list, | |
3130 | &cur_trans->dirty_bgs); | |
3131 | btrfs_get_block_group(cache); | |
3132 | drop_reserve = false; | |
3133 | } | |
3134 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3135 | } else if (ret) { | |
3136 | btrfs_abort_transaction(trans, ret); | |
3137 | } | |
3138 | } | |
3139 | ||
3140 | /* If it's not on the io list, we need to put the block group */ | |
3141 | if (should_put) | |
3142 | btrfs_put_block_group(cache); | |
3143 | if (drop_reserve) | |
3144 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
77745c05 JB |
3145 | /* |
3146 | * Avoid blocking other tasks for too long. It might even save | |
3147 | * us from writing caches for block groups that are going to be | |
3148 | * removed. | |
3149 | */ | |
3150 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
938fcbfb JB |
3151 | if (ret) |
3152 | goto out; | |
77745c05 JB |
3153 | mutex_lock(&trans->transaction->cache_write_mutex); |
3154 | } | |
3155 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3156 | ||
3157 | /* | |
3158 | * Go through delayed refs for all the stuff we've just kicked off | |
3159 | * and then loop back (just once) | |
3160 | */ | |
34d1eb0e JB |
3161 | if (!ret) |
3162 | ret = btrfs_run_delayed_refs(trans, 0); | |
77745c05 JB |
3163 | if (!ret && loops == 0) { |
3164 | loops++; | |
3165 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3166 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
3167 | /* | |
3168 | * dirty_bgs_lock protects us from concurrent block group | |
3169 | * deletes too (not just cache_write_mutex). | |
3170 | */ | |
3171 | if (!list_empty(&dirty)) { | |
3172 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3173 | goto again; | |
3174 | } | |
3175 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
938fcbfb JB |
3176 | } |
3177 | out: | |
3178 | if (ret < 0) { | |
3179 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3180 | list_splice_init(&dirty, &cur_trans->dirty_bgs); | |
3181 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
77745c05 JB |
3182 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
3183 | } | |
3184 | ||
3185 | btrfs_free_path(path); | |
3186 | return ret; | |
3187 | } | |
3188 | ||
3189 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) | |
3190 | { | |
3191 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
32da5386 | 3192 | struct btrfs_block_group *cache; |
77745c05 JB |
3193 | struct btrfs_transaction *cur_trans = trans->transaction; |
3194 | int ret = 0; | |
3195 | int should_put; | |
3196 | struct btrfs_path *path; | |
3197 | struct list_head *io = &cur_trans->io_bgs; | |
77745c05 JB |
3198 | |
3199 | path = btrfs_alloc_path(); | |
3200 | if (!path) | |
3201 | return -ENOMEM; | |
3202 | ||
3203 | /* | |
3204 | * Even though we are in the critical section of the transaction commit, | |
3205 | * we can still have concurrent tasks adding elements to this | |
3206 | * transaction's list of dirty block groups. These tasks correspond to | |
3207 | * endio free space workers started when writeback finishes for a | |
3208 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3209 | * allocate new block groups as a result of COWing nodes of the root | |
3210 | * tree when updating the free space inode. The writeback for the space | |
3211 | * caches is triggered by an earlier call to | |
3212 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3213 | * loop. | |
3214 | * Also we want to do the cache_save_setup first and then run the | |
3215 | * delayed refs to make sure we have the best chance at doing this all | |
3216 | * in one shot. | |
3217 | */ | |
3218 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3219 | while (!list_empty(&cur_trans->dirty_bgs)) { | |
3220 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
32da5386 | 3221 | struct btrfs_block_group, |
77745c05 JB |
3222 | dirty_list); |
3223 | ||
3224 | /* | |
3225 | * This can happen if cache_save_setup re-dirties a block group | |
3226 | * that is already under IO. Just wait for it to finish and | |
3227 | * then do it all again | |
3228 | */ | |
3229 | if (!list_empty(&cache->io_list)) { | |
3230 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3231 | list_del_init(&cache->io_list); | |
3232 | btrfs_wait_cache_io(trans, cache, path); | |
3233 | btrfs_put_block_group(cache); | |
3234 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3235 | } | |
3236 | ||
3237 | /* | |
3238 | * Don't remove from the dirty list until after we've waited on | |
3239 | * any pending IO | |
3240 | */ | |
3241 | list_del_init(&cache->dirty_list); | |
3242 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3243 | should_put = 1; | |
3244 | ||
3245 | cache_save_setup(cache, trans, path); | |
3246 | ||
3247 | if (!ret) | |
3248 | ret = btrfs_run_delayed_refs(trans, | |
3249 | (unsigned long) -1); | |
3250 | ||
3251 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3252 | cache->io_ctl.inode = NULL; | |
3253 | ret = btrfs_write_out_cache(trans, cache, path); | |
3254 | if (ret == 0 && cache->io_ctl.inode) { | |
77745c05 JB |
3255 | should_put = 0; |
3256 | list_add_tail(&cache->io_list, io); | |
3257 | } else { | |
3258 | /* | |
3259 | * If we failed to write the cache, the | |
3260 | * generation will be bad and life goes on | |
3261 | */ | |
3262 | ret = 0; | |
3263 | } | |
3264 | } | |
3265 | if (!ret) { | |
3be4d8ef | 3266 | ret = update_block_group_item(trans, path, cache); |
77745c05 JB |
3267 | /* |
3268 | * One of the free space endio workers might have | |
3269 | * created a new block group while updating a free space | |
3270 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3271 | * and hasn't released its transaction handle yet, in | |
3272 | * which case the new block group is still attached to | |
3273 | * its transaction handle and its creation has not | |
3274 | * finished yet (no block group item in the extent tree | |
3275 | * yet, etc). If this is the case, wait for all free | |
3276 | * space endio workers to finish and retry. This is a | |
260db43c | 3277 | * very rare case so no need for a more efficient and |
77745c05 JB |
3278 | * complex approach. |
3279 | */ | |
3280 | if (ret == -ENOENT) { | |
3281 | wait_event(cur_trans->writer_wait, | |
3282 | atomic_read(&cur_trans->num_writers) == 1); | |
3be4d8ef | 3283 | ret = update_block_group_item(trans, path, cache); |
77745c05 JB |
3284 | } |
3285 | if (ret) | |
3286 | btrfs_abort_transaction(trans, ret); | |
3287 | } | |
3288 | ||
3289 | /* If its not on the io list, we need to put the block group */ | |
3290 | if (should_put) | |
3291 | btrfs_put_block_group(cache); | |
3292 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
3293 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3294 | } | |
3295 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3296 | ||
3297 | /* | |
3298 | * Refer to the definition of io_bgs member for details why it's safe | |
3299 | * to use it without any locking | |
3300 | */ | |
3301 | while (!list_empty(io)) { | |
32da5386 | 3302 | cache = list_first_entry(io, struct btrfs_block_group, |
77745c05 JB |
3303 | io_list); |
3304 | list_del_init(&cache->io_list); | |
3305 | btrfs_wait_cache_io(trans, cache, path); | |
3306 | btrfs_put_block_group(cache); | |
3307 | } | |
3308 | ||
3309 | btrfs_free_path(path); | |
3310 | return ret; | |
3311 | } | |
606d1bf1 JB |
3312 | |
3313 | int btrfs_update_block_group(struct btrfs_trans_handle *trans, | |
11b66fa6 | 3314 | u64 bytenr, u64 num_bytes, bool alloc) |
606d1bf1 JB |
3315 | { |
3316 | struct btrfs_fs_info *info = trans->fs_info; | |
32da5386 | 3317 | struct btrfs_block_group *cache = NULL; |
606d1bf1 JB |
3318 | u64 total = num_bytes; |
3319 | u64 old_val; | |
3320 | u64 byte_in_group; | |
3321 | int factor; | |
3322 | int ret = 0; | |
3323 | ||
3324 | /* Block accounting for super block */ | |
3325 | spin_lock(&info->delalloc_root_lock); | |
3326 | old_val = btrfs_super_bytes_used(info->super_copy); | |
3327 | if (alloc) | |
3328 | old_val += num_bytes; | |
3329 | else | |
3330 | old_val -= num_bytes; | |
3331 | btrfs_set_super_bytes_used(info->super_copy, old_val); | |
3332 | spin_unlock(&info->delalloc_root_lock); | |
3333 | ||
3334 | while (total) { | |
ac2f1e63 JB |
3335 | bool reclaim; |
3336 | ||
606d1bf1 JB |
3337 | cache = btrfs_lookup_block_group(info, bytenr); |
3338 | if (!cache) { | |
3339 | ret = -ENOENT; | |
3340 | break; | |
3341 | } | |
3342 | factor = btrfs_bg_type_to_factor(cache->flags); | |
3343 | ||
3344 | /* | |
3345 | * If this block group has free space cache written out, we | |
3346 | * need to make sure to load it if we are removing space. This | |
3347 | * is because we need the unpinning stage to actually add the | |
3348 | * space back to the block group, otherwise we will leak space. | |
3349 | */ | |
32da5386 | 3350 | if (!alloc && !btrfs_block_group_done(cache)) |
ced8ecf0 | 3351 | btrfs_cache_block_group(cache, true); |
606d1bf1 | 3352 | |
b3470b5d DS |
3353 | byte_in_group = bytenr - cache->start; |
3354 | WARN_ON(byte_in_group > cache->length); | |
606d1bf1 JB |
3355 | |
3356 | spin_lock(&cache->space_info->lock); | |
3357 | spin_lock(&cache->lock); | |
3358 | ||
3359 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
3360 | cache->disk_cache_state < BTRFS_DC_CLEAR) | |
3361 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
3362 | ||
bf38be65 | 3363 | old_val = cache->used; |
b3470b5d | 3364 | num_bytes = min(total, cache->length - byte_in_group); |
606d1bf1 JB |
3365 | if (alloc) { |
3366 | old_val += num_bytes; | |
bf38be65 | 3367 | cache->used = old_val; |
606d1bf1 JB |
3368 | cache->reserved -= num_bytes; |
3369 | cache->space_info->bytes_reserved -= num_bytes; | |
3370 | cache->space_info->bytes_used += num_bytes; | |
3371 | cache->space_info->disk_used += num_bytes * factor; | |
3372 | spin_unlock(&cache->lock); | |
3373 | spin_unlock(&cache->space_info->lock); | |
3374 | } else { | |
3375 | old_val -= num_bytes; | |
bf38be65 | 3376 | cache->used = old_val; |
606d1bf1 JB |
3377 | cache->pinned += num_bytes; |
3378 | btrfs_space_info_update_bytes_pinned(info, | |
3379 | cache->space_info, num_bytes); | |
3380 | cache->space_info->bytes_used -= num_bytes; | |
3381 | cache->space_info->disk_used -= num_bytes * factor; | |
ac2f1e63 JB |
3382 | |
3383 | reclaim = should_reclaim_block_group(cache, num_bytes); | |
606d1bf1 JB |
3384 | spin_unlock(&cache->lock); |
3385 | spin_unlock(&cache->space_info->lock); | |
3386 | ||
fe119a6e | 3387 | set_extent_dirty(&trans->transaction->pinned_extents, |
606d1bf1 JB |
3388 | bytenr, bytenr + num_bytes - 1, |
3389 | GFP_NOFS | __GFP_NOFAIL); | |
3390 | } | |
3391 | ||
3392 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
3393 | if (list_empty(&cache->dirty_list)) { | |
3394 | list_add_tail(&cache->dirty_list, | |
3395 | &trans->transaction->dirty_bgs); | |
3396 | trans->delayed_ref_updates++; | |
3397 | btrfs_get_block_group(cache); | |
3398 | } | |
3399 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
3400 | ||
3401 | /* | |
3402 | * No longer have used bytes in this block group, queue it for | |
3403 | * deletion. We do this after adding the block group to the | |
3404 | * dirty list to avoid races between cleaner kthread and space | |
3405 | * cache writeout. | |
3406 | */ | |
6e80d4f8 DZ |
3407 | if (!alloc && old_val == 0) { |
3408 | if (!btrfs_test_opt(info, DISCARD_ASYNC)) | |
3409 | btrfs_mark_bg_unused(cache); | |
ac2f1e63 JB |
3410 | } else if (!alloc && reclaim) { |
3411 | btrfs_mark_bg_to_reclaim(cache); | |
6e80d4f8 | 3412 | } |
606d1bf1 JB |
3413 | |
3414 | btrfs_put_block_group(cache); | |
3415 | total -= num_bytes; | |
3416 | bytenr += num_bytes; | |
3417 | } | |
3418 | ||
3419 | /* Modified block groups are accounted for in the delayed_refs_rsv. */ | |
3420 | btrfs_update_delayed_refs_rsv(trans); | |
3421 | return ret; | |
3422 | } | |
3423 | ||
3424 | /** | |
3425 | * btrfs_add_reserved_bytes - update the block_group and space info counters | |
3426 | * @cache: The cache we are manipulating | |
3427 | * @ram_bytes: The number of bytes of file content, and will be same to | |
3428 | * @num_bytes except for the compress path. | |
3429 | * @num_bytes: The number of bytes in question | |
3430 | * @delalloc: The blocks are allocated for the delalloc write | |
3431 | * | |
3432 | * This is called by the allocator when it reserves space. If this is a | |
3433 | * reservation and the block group has become read only we cannot make the | |
3434 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
3435 | */ | |
32da5386 | 3436 | int btrfs_add_reserved_bytes(struct btrfs_block_group *cache, |
606d1bf1 JB |
3437 | u64 ram_bytes, u64 num_bytes, int delalloc) |
3438 | { | |
3439 | struct btrfs_space_info *space_info = cache->space_info; | |
3440 | int ret = 0; | |
3441 | ||
3442 | spin_lock(&space_info->lock); | |
3443 | spin_lock(&cache->lock); | |
3444 | if (cache->ro) { | |
3445 | ret = -EAGAIN; | |
3446 | } else { | |
3447 | cache->reserved += num_bytes; | |
3448 | space_info->bytes_reserved += num_bytes; | |
a43c3835 JB |
3449 | trace_btrfs_space_reservation(cache->fs_info, "space_info", |
3450 | space_info->flags, num_bytes, 1); | |
606d1bf1 JB |
3451 | btrfs_space_info_update_bytes_may_use(cache->fs_info, |
3452 | space_info, -ram_bytes); | |
3453 | if (delalloc) | |
3454 | cache->delalloc_bytes += num_bytes; | |
99ffb43e JB |
3455 | |
3456 | /* | |
3457 | * Compression can use less space than we reserved, so wake | |
3458 | * tickets if that happens | |
3459 | */ | |
3460 | if (num_bytes < ram_bytes) | |
3461 | btrfs_try_granting_tickets(cache->fs_info, space_info); | |
606d1bf1 JB |
3462 | } |
3463 | spin_unlock(&cache->lock); | |
3464 | spin_unlock(&space_info->lock); | |
3465 | return ret; | |
3466 | } | |
3467 | ||
3468 | /** | |
3469 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
3470 | * @cache: The cache we are manipulating | |
3471 | * @num_bytes: The number of bytes in question | |
3472 | * @delalloc: The blocks are allocated for the delalloc write | |
3473 | * | |
3474 | * This is called by somebody who is freeing space that was never actually used | |
3475 | * on disk. For example if you reserve some space for a new leaf in transaction | |
3476 | * A and before transaction A commits you free that leaf, you call this with | |
3477 | * reserve set to 0 in order to clear the reservation. | |
3478 | */ | |
32da5386 | 3479 | void btrfs_free_reserved_bytes(struct btrfs_block_group *cache, |
606d1bf1 JB |
3480 | u64 num_bytes, int delalloc) |
3481 | { | |
3482 | struct btrfs_space_info *space_info = cache->space_info; | |
3483 | ||
3484 | spin_lock(&space_info->lock); | |
3485 | spin_lock(&cache->lock); | |
3486 | if (cache->ro) | |
3487 | space_info->bytes_readonly += num_bytes; | |
3488 | cache->reserved -= num_bytes; | |
3489 | space_info->bytes_reserved -= num_bytes; | |
3490 | space_info->max_extent_size = 0; | |
3491 | ||
3492 | if (delalloc) | |
3493 | cache->delalloc_bytes -= num_bytes; | |
3494 | spin_unlock(&cache->lock); | |
3308234a JB |
3495 | |
3496 | btrfs_try_granting_tickets(cache->fs_info, space_info); | |
606d1bf1 JB |
3497 | spin_unlock(&space_info->lock); |
3498 | } | |
07730d87 JB |
3499 | |
3500 | static void force_metadata_allocation(struct btrfs_fs_info *info) | |
3501 | { | |
3502 | struct list_head *head = &info->space_info; | |
3503 | struct btrfs_space_info *found; | |
3504 | ||
72804905 | 3505 | list_for_each_entry(found, head, list) { |
07730d87 JB |
3506 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) |
3507 | found->force_alloc = CHUNK_ALLOC_FORCE; | |
3508 | } | |
07730d87 JB |
3509 | } |
3510 | ||
3511 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, | |
3512 | struct btrfs_space_info *sinfo, int force) | |
3513 | { | |
3514 | u64 bytes_used = btrfs_space_info_used(sinfo, false); | |
3515 | u64 thresh; | |
3516 | ||
3517 | if (force == CHUNK_ALLOC_FORCE) | |
3518 | return 1; | |
3519 | ||
3520 | /* | |
3521 | * in limited mode, we want to have some free space up to | |
3522 | * about 1% of the FS size. | |
3523 | */ | |
3524 | if (force == CHUNK_ALLOC_LIMITED) { | |
3525 | thresh = btrfs_super_total_bytes(fs_info->super_copy); | |
3526 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); | |
3527 | ||
3528 | if (sinfo->total_bytes - bytes_used < thresh) | |
3529 | return 1; | |
3530 | } | |
3531 | ||
3532 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) | |
3533 | return 0; | |
3534 | return 1; | |
3535 | } | |
3536 | ||
3537 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) | |
3538 | { | |
3539 | u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type); | |
3540 | ||
3541 | return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); | |
3542 | } | |
3543 | ||
820c363b | 3544 | static struct btrfs_block_group *do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags) |
79bd3712 FM |
3545 | { |
3546 | struct btrfs_block_group *bg; | |
3547 | int ret; | |
3548 | ||
3549 | /* | |
3550 | * Check if we have enough space in the system space info because we | |
3551 | * will need to update device items in the chunk btree and insert a new | |
3552 | * chunk item in the chunk btree as well. This will allocate a new | |
3553 | * system block group if needed. | |
3554 | */ | |
3555 | check_system_chunk(trans, flags); | |
3556 | ||
f6f39f7a | 3557 | bg = btrfs_create_chunk(trans, flags); |
79bd3712 FM |
3558 | if (IS_ERR(bg)) { |
3559 | ret = PTR_ERR(bg); | |
3560 | goto out; | |
3561 | } | |
3562 | ||
79bd3712 FM |
3563 | ret = btrfs_chunk_alloc_add_chunk_item(trans, bg); |
3564 | /* | |
3565 | * Normally we are not expected to fail with -ENOSPC here, since we have | |
3566 | * previously reserved space in the system space_info and allocated one | |
ecd84d54 | 3567 | * new system chunk if necessary. However there are three exceptions: |
79bd3712 FM |
3568 | * |
3569 | * 1) We may have enough free space in the system space_info but all the | |
3570 | * existing system block groups have a profile which can not be used | |
3571 | * for extent allocation. | |
3572 | * | |
3573 | * This happens when mounting in degraded mode. For example we have a | |
3574 | * RAID1 filesystem with 2 devices, lose one device and mount the fs | |
3575 | * using the other device in degraded mode. If we then allocate a chunk, | |
3576 | * we may have enough free space in the existing system space_info, but | |
3577 | * none of the block groups can be used for extent allocation since they | |
3578 | * have a RAID1 profile, and because we are in degraded mode with a | |
3579 | * single device, we are forced to allocate a new system chunk with a | |
3580 | * SINGLE profile. Making check_system_chunk() iterate over all system | |
3581 | * block groups and check if they have a usable profile and enough space | |
3582 | * can be slow on very large filesystems, so we tolerate the -ENOSPC and | |
3583 | * try again after forcing allocation of a new system chunk. Like this | |
3584 | * we avoid paying the cost of that search in normal circumstances, when | |
3585 | * we were not mounted in degraded mode; | |
3586 | * | |
3587 | * 2) We had enough free space info the system space_info, and one suitable | |
3588 | * block group to allocate from when we called check_system_chunk() | |
3589 | * above. However right after we called it, the only system block group | |
3590 | * with enough free space got turned into RO mode by a running scrub, | |
3591 | * and in this case we have to allocate a new one and retry. We only | |
3592 | * need do this allocate and retry once, since we have a transaction | |
ecd84d54 FM |
3593 | * handle and scrub uses the commit root to search for block groups; |
3594 | * | |
3595 | * 3) We had one system block group with enough free space when we called | |
3596 | * check_system_chunk(), but after that, right before we tried to | |
3597 | * allocate the last extent buffer we needed, a discard operation came | |
3598 | * in and it temporarily removed the last free space entry from the | |
3599 | * block group (discard removes a free space entry, discards it, and | |
3600 | * then adds back the entry to the block group cache). | |
79bd3712 FM |
3601 | */ |
3602 | if (ret == -ENOSPC) { | |
3603 | const u64 sys_flags = btrfs_system_alloc_profile(trans->fs_info); | |
3604 | struct btrfs_block_group *sys_bg; | |
3605 | ||
f6f39f7a | 3606 | sys_bg = btrfs_create_chunk(trans, sys_flags); |
79bd3712 FM |
3607 | if (IS_ERR(sys_bg)) { |
3608 | ret = PTR_ERR(sys_bg); | |
3609 | btrfs_abort_transaction(trans, ret); | |
3610 | goto out; | |
3611 | } | |
3612 | ||
3613 | ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); | |
3614 | if (ret) { | |
3615 | btrfs_abort_transaction(trans, ret); | |
3616 | goto out; | |
3617 | } | |
3618 | ||
3619 | ret = btrfs_chunk_alloc_add_chunk_item(trans, bg); | |
3620 | if (ret) { | |
3621 | btrfs_abort_transaction(trans, ret); | |
3622 | goto out; | |
3623 | } | |
3624 | } else if (ret) { | |
3625 | btrfs_abort_transaction(trans, ret); | |
3626 | goto out; | |
3627 | } | |
3628 | out: | |
3629 | btrfs_trans_release_chunk_metadata(trans); | |
3630 | ||
820c363b NA |
3631 | if (ret) |
3632 | return ERR_PTR(ret); | |
3633 | ||
3634 | btrfs_get_block_group(bg); | |
3635 | return bg; | |
79bd3712 FM |
3636 | } |
3637 | ||
07730d87 | 3638 | /* |
79bd3712 FM |
3639 | * Chunk allocation is done in 2 phases: |
3640 | * | |
3641 | * 1) Phase 1 - through btrfs_chunk_alloc() we allocate device extents for | |
3642 | * the chunk, the chunk mapping, create its block group and add the items | |
3643 | * that belong in the chunk btree to it - more specifically, we need to | |
3644 | * update device items in the chunk btree and add a new chunk item to it. | |
3645 | * | |
3646 | * 2) Phase 2 - through btrfs_create_pending_block_groups(), we add the block | |
3647 | * group item to the extent btree and the device extent items to the devices | |
3648 | * btree. | |
3649 | * | |
3650 | * This is done to prevent deadlocks. For example when COWing a node from the | |
3651 | * extent btree we are holding a write lock on the node's parent and if we | |
3652 | * trigger chunk allocation and attempted to insert the new block group item | |
3653 | * in the extent btree right way, we could deadlock because the path for the | |
3654 | * insertion can include that parent node. At first glance it seems impossible | |
3655 | * to trigger chunk allocation after starting a transaction since tasks should | |
3656 | * reserve enough transaction units (metadata space), however while that is true | |
3657 | * most of the time, chunk allocation may still be triggered for several reasons: | |
3658 | * | |
3659 | * 1) When reserving metadata, we check if there is enough free space in the | |
3660 | * metadata space_info and therefore don't trigger allocation of a new chunk. | |
3661 | * However later when the task actually tries to COW an extent buffer from | |
3662 | * the extent btree or from the device btree for example, it is forced to | |
3663 | * allocate a new block group (chunk) because the only one that had enough | |
3664 | * free space was just turned to RO mode by a running scrub for example (or | |
3665 | * device replace, block group reclaim thread, etc), so we can not use it | |
3666 | * for allocating an extent and end up being forced to allocate a new one; | |
3667 | * | |
3668 | * 2) Because we only check that the metadata space_info has enough free bytes, | |
3669 | * we end up not allocating a new metadata chunk in that case. However if | |
3670 | * the filesystem was mounted in degraded mode, none of the existing block | |
3671 | * groups might be suitable for extent allocation due to their incompatible | |
3672 | * profile (for e.g. mounting a 2 devices filesystem, where all block groups | |
3673 | * use a RAID1 profile, in degraded mode using a single device). In this case | |
3674 | * when the task attempts to COW some extent buffer of the extent btree for | |
3675 | * example, it will trigger allocation of a new metadata block group with a | |
3676 | * suitable profile (SINGLE profile in the example of the degraded mount of | |
3677 | * the RAID1 filesystem); | |
3678 | * | |
3679 | * 3) The task has reserved enough transaction units / metadata space, but when | |
3680 | * it attempts to COW an extent buffer from the extent or device btree for | |
3681 | * example, it does not find any free extent in any metadata block group, | |
3682 | * therefore forced to try to allocate a new metadata block group. | |
3683 | * This is because some other task allocated all available extents in the | |
3684 | * meanwhile - this typically happens with tasks that don't reserve space | |
3685 | * properly, either intentionally or as a bug. One example where this is | |
3686 | * done intentionally is fsync, as it does not reserve any transaction units | |
3687 | * and ends up allocating a variable number of metadata extents for log | |
ecd84d54 FM |
3688 | * tree extent buffers; |
3689 | * | |
3690 | * 4) The task has reserved enough transaction units / metadata space, but right | |
3691 | * before it tries to allocate the last extent buffer it needs, a discard | |
3692 | * operation comes in and, temporarily, removes the last free space entry from | |
3693 | * the only metadata block group that had free space (discard starts by | |
3694 | * removing a free space entry from a block group, then does the discard | |
3695 | * operation and, once it's done, it adds back the free space entry to the | |
3696 | * block group). | |
79bd3712 FM |
3697 | * |
3698 | * We also need this 2 phases setup when adding a device to a filesystem with | |
3699 | * a seed device - we must create new metadata and system chunks without adding | |
3700 | * any of the block group items to the chunk, extent and device btrees. If we | |
3701 | * did not do it this way, we would get ENOSPC when attempting to update those | |
3702 | * btrees, since all the chunks from the seed device are read-only. | |
3703 | * | |
3704 | * Phase 1 does the updates and insertions to the chunk btree because if we had | |
3705 | * it done in phase 2 and have a thundering herd of tasks allocating chunks in | |
3706 | * parallel, we risk having too many system chunks allocated by many tasks if | |
3707 | * many tasks reach phase 1 without the previous ones completing phase 2. In the | |
3708 | * extreme case this leads to exhaustion of the system chunk array in the | |
3709 | * superblock. This is easier to trigger if using a btree node/leaf size of 64K | |
3710 | * and with RAID filesystems (so we have more device items in the chunk btree). | |
3711 | * This has happened before and commit eafa4fd0ad0607 ("btrfs: fix exhaustion of | |
3712 | * the system chunk array due to concurrent allocations") provides more details. | |
3713 | * | |
2bb2e00e FM |
3714 | * Allocation of system chunks does not happen through this function. A task that |
3715 | * needs to update the chunk btree (the only btree that uses system chunks), must | |
3716 | * preallocate chunk space by calling either check_system_chunk() or | |
3717 | * btrfs_reserve_chunk_metadata() - the former is used when allocating a data or | |
3718 | * metadata chunk or when removing a chunk, while the later is used before doing | |
3719 | * a modification to the chunk btree - use cases for the later are adding, | |
3720 | * removing and resizing a device as well as relocation of a system chunk. | |
3721 | * See the comment below for more details. | |
79bd3712 FM |
3722 | * |
3723 | * The reservation of system space, done through check_system_chunk(), as well | |
3724 | * as all the updates and insertions into the chunk btree must be done while | |
3725 | * holding fs_info->chunk_mutex. This is important to guarantee that while COWing | |
3726 | * an extent buffer from the chunks btree we never trigger allocation of a new | |
3727 | * system chunk, which would result in a deadlock (trying to lock twice an | |
3728 | * extent buffer of the chunk btree, first time before triggering the chunk | |
3729 | * allocation and the second time during chunk allocation while attempting to | |
3730 | * update the chunks btree). The system chunk array is also updated while holding | |
3731 | * that mutex. The same logic applies to removing chunks - we must reserve system | |
3732 | * space, update the chunk btree and the system chunk array in the superblock | |
3733 | * while holding fs_info->chunk_mutex. | |
3734 | * | |
3735 | * This function, btrfs_chunk_alloc(), belongs to phase 1. | |
3736 | * | |
3737 | * If @force is CHUNK_ALLOC_FORCE: | |
07730d87 JB |
3738 | * - return 1 if it successfully allocates a chunk, |
3739 | * - return errors including -ENOSPC otherwise. | |
79bd3712 | 3740 | * If @force is NOT CHUNK_ALLOC_FORCE: |
07730d87 JB |
3741 | * - return 0 if it doesn't need to allocate a new chunk, |
3742 | * - return 1 if it successfully allocates a chunk, | |
3743 | * - return errors including -ENOSPC otherwise. | |
3744 | */ | |
3745 | int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, | |
3746 | enum btrfs_chunk_alloc_enum force) | |
3747 | { | |
3748 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3749 | struct btrfs_space_info *space_info; | |
820c363b | 3750 | struct btrfs_block_group *ret_bg; |
07730d87 JB |
3751 | bool wait_for_alloc = false; |
3752 | bool should_alloc = false; | |
760e69c4 | 3753 | bool from_extent_allocation = false; |
07730d87 JB |
3754 | int ret = 0; |
3755 | ||
760e69c4 NA |
3756 | if (force == CHUNK_ALLOC_FORCE_FOR_EXTENT) { |
3757 | from_extent_allocation = true; | |
3758 | force = CHUNK_ALLOC_FORCE; | |
3759 | } | |
3760 | ||
07730d87 JB |
3761 | /* Don't re-enter if we're already allocating a chunk */ |
3762 | if (trans->allocating_chunk) | |
3763 | return -ENOSPC; | |
79bd3712 | 3764 | /* |
2bb2e00e FM |
3765 | * Allocation of system chunks can not happen through this path, as we |
3766 | * could end up in a deadlock if we are allocating a data or metadata | |
3767 | * chunk and there is another task modifying the chunk btree. | |
3768 | * | |
3769 | * This is because while we are holding the chunk mutex, we will attempt | |
3770 | * to add the new chunk item to the chunk btree or update an existing | |
3771 | * device item in the chunk btree, while the other task that is modifying | |
3772 | * the chunk btree is attempting to COW an extent buffer while holding a | |
3773 | * lock on it and on its parent - if the COW operation triggers a system | |
3774 | * chunk allocation, then we can deadlock because we are holding the | |
3775 | * chunk mutex and we may need to access that extent buffer or its parent | |
3776 | * in order to add the chunk item or update a device item. | |
3777 | * | |
3778 | * Tasks that want to modify the chunk tree should reserve system space | |
3779 | * before updating the chunk btree, by calling either | |
3780 | * btrfs_reserve_chunk_metadata() or check_system_chunk(). | |
3781 | * It's possible that after a task reserves the space, it still ends up | |
3782 | * here - this happens in the cases described above at do_chunk_alloc(). | |
3783 | * The task will have to either retry or fail. | |
79bd3712 | 3784 | */ |
2bb2e00e | 3785 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
79bd3712 | 3786 | return -ENOSPC; |
07730d87 JB |
3787 | |
3788 | space_info = btrfs_find_space_info(fs_info, flags); | |
3789 | ASSERT(space_info); | |
3790 | ||
3791 | do { | |
3792 | spin_lock(&space_info->lock); | |
3793 | if (force < space_info->force_alloc) | |
3794 | force = space_info->force_alloc; | |
3795 | should_alloc = should_alloc_chunk(fs_info, space_info, force); | |
3796 | if (space_info->full) { | |
3797 | /* No more free physical space */ | |
3798 | if (should_alloc) | |
3799 | ret = -ENOSPC; | |
3800 | else | |
3801 | ret = 0; | |
3802 | spin_unlock(&space_info->lock); | |
3803 | return ret; | |
3804 | } else if (!should_alloc) { | |
3805 | spin_unlock(&space_info->lock); | |
3806 | return 0; | |
3807 | } else if (space_info->chunk_alloc) { | |
3808 | /* | |
3809 | * Someone is already allocating, so we need to block | |
3810 | * until this someone is finished and then loop to | |
3811 | * recheck if we should continue with our allocation | |
3812 | * attempt. | |
3813 | */ | |
3814 | wait_for_alloc = true; | |
1314ca78 | 3815 | force = CHUNK_ALLOC_NO_FORCE; |
07730d87 JB |
3816 | spin_unlock(&space_info->lock); |
3817 | mutex_lock(&fs_info->chunk_mutex); | |
3818 | mutex_unlock(&fs_info->chunk_mutex); | |
3819 | } else { | |
3820 | /* Proceed with allocation */ | |
3821 | space_info->chunk_alloc = 1; | |
3822 | wait_for_alloc = false; | |
3823 | spin_unlock(&space_info->lock); | |
3824 | } | |
3825 | ||
3826 | cond_resched(); | |
3827 | } while (wait_for_alloc); | |
3828 | ||
3829 | mutex_lock(&fs_info->chunk_mutex); | |
3830 | trans->allocating_chunk = true; | |
3831 | ||
3832 | /* | |
3833 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3834 | * allocating mixed chunks instead of individual chunks. | |
3835 | */ | |
3836 | if (btrfs_mixed_space_info(space_info)) | |
3837 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3838 | ||
3839 | /* | |
3840 | * if we're doing a data chunk, go ahead and make sure that | |
3841 | * we keep a reasonable number of metadata chunks allocated in the | |
3842 | * FS as well. | |
3843 | */ | |
3844 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { | |
3845 | fs_info->data_chunk_allocations++; | |
3846 | if (!(fs_info->data_chunk_allocations % | |
3847 | fs_info->metadata_ratio)) | |
3848 | force_metadata_allocation(fs_info); | |
3849 | } | |
3850 | ||
820c363b | 3851 | ret_bg = do_chunk_alloc(trans, flags); |
07730d87 JB |
3852 | trans->allocating_chunk = false; |
3853 | ||
760e69c4 | 3854 | if (IS_ERR(ret_bg)) { |
820c363b | 3855 | ret = PTR_ERR(ret_bg); |
760e69c4 NA |
3856 | } else if (from_extent_allocation) { |
3857 | /* | |
3858 | * New block group is likely to be used soon. Try to activate | |
3859 | * it now. Failure is OK for now. | |
3860 | */ | |
3861 | btrfs_zone_activate(ret_bg); | |
3862 | } | |
3863 | ||
3864 | if (!ret) | |
820c363b NA |
3865 | btrfs_put_block_group(ret_bg); |
3866 | ||
07730d87 JB |
3867 | spin_lock(&space_info->lock); |
3868 | if (ret < 0) { | |
3869 | if (ret == -ENOSPC) | |
3870 | space_info->full = 1; | |
3871 | else | |
3872 | goto out; | |
3873 | } else { | |
3874 | ret = 1; | |
3875 | space_info->max_extent_size = 0; | |
3876 | } | |
3877 | ||
3878 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
3879 | out: | |
3880 | space_info->chunk_alloc = 0; | |
3881 | spin_unlock(&space_info->lock); | |
3882 | mutex_unlock(&fs_info->chunk_mutex); | |
07730d87 JB |
3883 | |
3884 | return ret; | |
3885 | } | |
3886 | ||
3887 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) | |
3888 | { | |
3889 | u64 num_dev; | |
3890 | ||
3891 | num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max; | |
3892 | if (!num_dev) | |
3893 | num_dev = fs_info->fs_devices->rw_devices; | |
3894 | ||
3895 | return num_dev; | |
3896 | } | |
3897 | ||
2bb2e00e FM |
3898 | static void reserve_chunk_space(struct btrfs_trans_handle *trans, |
3899 | u64 bytes, | |
3900 | u64 type) | |
07730d87 JB |
3901 | { |
3902 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3903 | struct btrfs_space_info *info; | |
3904 | u64 left; | |
07730d87 | 3905 | int ret = 0; |
07730d87 JB |
3906 | |
3907 | /* | |
3908 | * Needed because we can end up allocating a system chunk and for an | |
3909 | * atomic and race free space reservation in the chunk block reserve. | |
3910 | */ | |
3911 | lockdep_assert_held(&fs_info->chunk_mutex); | |
3912 | ||
3913 | info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3914 | spin_lock(&info->lock); | |
3915 | left = info->total_bytes - btrfs_space_info_used(info, true); | |
3916 | spin_unlock(&info->lock); | |
3917 | ||
2bb2e00e | 3918 | if (left < bytes && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
07730d87 | 3919 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", |
2bb2e00e | 3920 | left, bytes, type); |
07730d87 JB |
3921 | btrfs_dump_space_info(fs_info, info, 0, 0); |
3922 | } | |
3923 | ||
2bb2e00e | 3924 | if (left < bytes) { |
07730d87 | 3925 | u64 flags = btrfs_system_alloc_profile(fs_info); |
79bd3712 | 3926 | struct btrfs_block_group *bg; |
07730d87 JB |
3927 | |
3928 | /* | |
3929 | * Ignore failure to create system chunk. We might end up not | |
3930 | * needing it, as we might not need to COW all nodes/leafs from | |
3931 | * the paths we visit in the chunk tree (they were already COWed | |
3932 | * or created in the current transaction for example). | |
3933 | */ | |
f6f39f7a | 3934 | bg = btrfs_create_chunk(trans, flags); |
79bd3712 FM |
3935 | if (IS_ERR(bg)) { |
3936 | ret = PTR_ERR(bg); | |
2bb2e00e | 3937 | } else { |
b6a98021 NA |
3938 | /* |
3939 | * We have a new chunk. We also need to activate it for | |
3940 | * zoned filesystem. | |
3941 | */ | |
3942 | ret = btrfs_zoned_activate_one_bg(fs_info, info, true); | |
3943 | if (ret < 0) | |
3944 | return; | |
3945 | ||
79bd3712 FM |
3946 | /* |
3947 | * If we fail to add the chunk item here, we end up | |
3948 | * trying again at phase 2 of chunk allocation, at | |
3949 | * btrfs_create_pending_block_groups(). So ignore | |
2bb2e00e FM |
3950 | * any error here. An ENOSPC here could happen, due to |
3951 | * the cases described at do_chunk_alloc() - the system | |
3952 | * block group we just created was just turned into RO | |
3953 | * mode by a scrub for example, or a running discard | |
3954 | * temporarily removed its free space entries, etc. | |
79bd3712 FM |
3955 | */ |
3956 | btrfs_chunk_alloc_add_chunk_item(trans, bg); | |
3957 | } | |
07730d87 JB |
3958 | } |
3959 | ||
3960 | if (!ret) { | |
9270501c | 3961 | ret = btrfs_block_rsv_add(fs_info, |
07730d87 | 3962 | &fs_info->chunk_block_rsv, |
2bb2e00e | 3963 | bytes, BTRFS_RESERVE_NO_FLUSH); |
1cb3db1c | 3964 | if (!ret) |
2bb2e00e | 3965 | trans->chunk_bytes_reserved += bytes; |
07730d87 JB |
3966 | } |
3967 | } | |
3968 | ||
2bb2e00e FM |
3969 | /* |
3970 | * Reserve space in the system space for allocating or removing a chunk. | |
3971 | * The caller must be holding fs_info->chunk_mutex. | |
3972 | */ | |
3973 | void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) | |
3974 | { | |
3975 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3976 | const u64 num_devs = get_profile_num_devs(fs_info, type); | |
3977 | u64 bytes; | |
3978 | ||
3979 | /* num_devs device items to update and 1 chunk item to add or remove. */ | |
3980 | bytes = btrfs_calc_metadata_size(fs_info, num_devs) + | |
3981 | btrfs_calc_insert_metadata_size(fs_info, 1); | |
3982 | ||
3983 | reserve_chunk_space(trans, bytes, type); | |
3984 | } | |
3985 | ||
3986 | /* | |
3987 | * Reserve space in the system space, if needed, for doing a modification to the | |
3988 | * chunk btree. | |
3989 | * | |
3990 | * @trans: A transaction handle. | |
3991 | * @is_item_insertion: Indicate if the modification is for inserting a new item | |
3992 | * in the chunk btree or if it's for the deletion or update | |
3993 | * of an existing item. | |
3994 | * | |
3995 | * This is used in a context where we need to update the chunk btree outside | |
3996 | * block group allocation and removal, to avoid a deadlock with a concurrent | |
3997 | * task that is allocating a metadata or data block group and therefore needs to | |
3998 | * update the chunk btree while holding the chunk mutex. After the update to the | |
3999 | * chunk btree is done, btrfs_trans_release_chunk_metadata() should be called. | |
4000 | * | |
4001 | */ | |
4002 | void btrfs_reserve_chunk_metadata(struct btrfs_trans_handle *trans, | |
4003 | bool is_item_insertion) | |
4004 | { | |
4005 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
4006 | u64 bytes; | |
4007 | ||
4008 | if (is_item_insertion) | |
4009 | bytes = btrfs_calc_insert_metadata_size(fs_info, 1); | |
4010 | else | |
4011 | bytes = btrfs_calc_metadata_size(fs_info, 1); | |
4012 | ||
4013 | mutex_lock(&fs_info->chunk_mutex); | |
4014 | reserve_chunk_space(trans, bytes, BTRFS_BLOCK_GROUP_SYSTEM); | |
4015 | mutex_unlock(&fs_info->chunk_mutex); | |
4016 | } | |
4017 | ||
3e43c279 JB |
4018 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
4019 | { | |
32da5386 | 4020 | struct btrfs_block_group *block_group; |
3e43c279 | 4021 | |
50c31eaa JB |
4022 | block_group = btrfs_lookup_first_block_group(info, 0); |
4023 | while (block_group) { | |
4024 | btrfs_wait_block_group_cache_done(block_group); | |
4025 | spin_lock(&block_group->lock); | |
4026 | if (test_and_clear_bit(BLOCK_GROUP_FLAG_IREF, | |
4027 | &block_group->runtime_flags)) { | |
4028 | struct inode *inode = block_group->inode; | |
4029 | ||
4030 | block_group->inode = NULL; | |
3e43c279 | 4031 | spin_unlock(&block_group->lock); |
3e43c279 | 4032 | |
50c31eaa JB |
4033 | ASSERT(block_group->io_ctl.inode == NULL); |
4034 | iput(inode); | |
4035 | } else { | |
4036 | spin_unlock(&block_group->lock); | |
4037 | } | |
4038 | block_group = btrfs_next_block_group(block_group); | |
3e43c279 JB |
4039 | } |
4040 | } | |
4041 | ||
4042 | /* | |
4043 | * Must be called only after stopping all workers, since we could have block | |
4044 | * group caching kthreads running, and therefore they could race with us if we | |
4045 | * freed the block groups before stopping them. | |
4046 | */ | |
4047 | int btrfs_free_block_groups(struct btrfs_fs_info *info) | |
4048 | { | |
32da5386 | 4049 | struct btrfs_block_group *block_group; |
3e43c279 JB |
4050 | struct btrfs_space_info *space_info; |
4051 | struct btrfs_caching_control *caching_ctl; | |
4052 | struct rb_node *n; | |
4053 | ||
16b0c258 | 4054 | write_lock(&info->block_group_cache_lock); |
3e43c279 JB |
4055 | while (!list_empty(&info->caching_block_groups)) { |
4056 | caching_ctl = list_entry(info->caching_block_groups.next, | |
4057 | struct btrfs_caching_control, list); | |
4058 | list_del(&caching_ctl->list); | |
4059 | btrfs_put_caching_control(caching_ctl); | |
4060 | } | |
16b0c258 | 4061 | write_unlock(&info->block_group_cache_lock); |
3e43c279 JB |
4062 | |
4063 | spin_lock(&info->unused_bgs_lock); | |
4064 | while (!list_empty(&info->unused_bgs)) { | |
4065 | block_group = list_first_entry(&info->unused_bgs, | |
32da5386 | 4066 | struct btrfs_block_group, |
3e43c279 JB |
4067 | bg_list); |
4068 | list_del_init(&block_group->bg_list); | |
4069 | btrfs_put_block_group(block_group); | |
4070 | } | |
3e43c279 | 4071 | |
18bb8bbf JT |
4072 | while (!list_empty(&info->reclaim_bgs)) { |
4073 | block_group = list_first_entry(&info->reclaim_bgs, | |
4074 | struct btrfs_block_group, | |
4075 | bg_list); | |
4076 | list_del_init(&block_group->bg_list); | |
4077 | btrfs_put_block_group(block_group); | |
4078 | } | |
4079 | spin_unlock(&info->unused_bgs_lock); | |
4080 | ||
afba2bc0 NA |
4081 | spin_lock(&info->zone_active_bgs_lock); |
4082 | while (!list_empty(&info->zone_active_bgs)) { | |
4083 | block_group = list_first_entry(&info->zone_active_bgs, | |
4084 | struct btrfs_block_group, | |
4085 | active_bg_list); | |
4086 | list_del_init(&block_group->active_bg_list); | |
4087 | btrfs_put_block_group(block_group); | |
4088 | } | |
4089 | spin_unlock(&info->zone_active_bgs_lock); | |
4090 | ||
16b0c258 | 4091 | write_lock(&info->block_group_cache_lock); |
08dddb29 | 4092 | while ((n = rb_last(&info->block_group_cache_tree.rb_root)) != NULL) { |
32da5386 | 4093 | block_group = rb_entry(n, struct btrfs_block_group, |
3e43c279 | 4094 | cache_node); |
08dddb29 FM |
4095 | rb_erase_cached(&block_group->cache_node, |
4096 | &info->block_group_cache_tree); | |
3e43c279 | 4097 | RB_CLEAR_NODE(&block_group->cache_node); |
16b0c258 | 4098 | write_unlock(&info->block_group_cache_lock); |
3e43c279 JB |
4099 | |
4100 | down_write(&block_group->space_info->groups_sem); | |
4101 | list_del(&block_group->list); | |
4102 | up_write(&block_group->space_info->groups_sem); | |
4103 | ||
4104 | /* | |
4105 | * We haven't cached this block group, which means we could | |
4106 | * possibly have excluded extents on this block group. | |
4107 | */ | |
4108 | if (block_group->cached == BTRFS_CACHE_NO || | |
4109 | block_group->cached == BTRFS_CACHE_ERROR) | |
4110 | btrfs_free_excluded_extents(block_group); | |
4111 | ||
4112 | btrfs_remove_free_space_cache(block_group); | |
4113 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); | |
4114 | ASSERT(list_empty(&block_group->dirty_list)); | |
4115 | ASSERT(list_empty(&block_group->io_list)); | |
4116 | ASSERT(list_empty(&block_group->bg_list)); | |
48aaeebe | 4117 | ASSERT(refcount_read(&block_group->refs) == 1); |
195a49ea | 4118 | ASSERT(block_group->swap_extents == 0); |
3e43c279 JB |
4119 | btrfs_put_block_group(block_group); |
4120 | ||
16b0c258 | 4121 | write_lock(&info->block_group_cache_lock); |
3e43c279 | 4122 | } |
16b0c258 | 4123 | write_unlock(&info->block_group_cache_lock); |
3e43c279 | 4124 | |
3e43c279 JB |
4125 | btrfs_release_global_block_rsv(info); |
4126 | ||
4127 | while (!list_empty(&info->space_info)) { | |
4128 | space_info = list_entry(info->space_info.next, | |
4129 | struct btrfs_space_info, | |
4130 | list); | |
4131 | ||
4132 | /* | |
4133 | * Do not hide this behind enospc_debug, this is actually | |
4134 | * important and indicates a real bug if this happens. | |
4135 | */ | |
4136 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
3e43c279 JB |
4137 | space_info->bytes_may_use > 0)) |
4138 | btrfs_dump_space_info(info, space_info, 0, 0); | |
40cdc509 FM |
4139 | |
4140 | /* | |
4141 | * If there was a failure to cleanup a log tree, very likely due | |
4142 | * to an IO failure on a writeback attempt of one or more of its | |
4143 | * extent buffers, we could not do proper (and cheap) unaccounting | |
4144 | * of their reserved space, so don't warn on bytes_reserved > 0 in | |
4145 | * that case. | |
4146 | */ | |
4147 | if (!(space_info->flags & BTRFS_BLOCK_GROUP_METADATA) || | |
4148 | !BTRFS_FS_LOG_CLEANUP_ERROR(info)) { | |
4149 | if (WARN_ON(space_info->bytes_reserved > 0)) | |
4150 | btrfs_dump_space_info(info, space_info, 0, 0); | |
4151 | } | |
4152 | ||
d611add4 | 4153 | WARN_ON(space_info->reclaim_size > 0); |
3e43c279 JB |
4154 | list_del(&space_info->list); |
4155 | btrfs_sysfs_remove_space_info(space_info); | |
4156 | } | |
4157 | return 0; | |
4158 | } | |
684b752b FM |
4159 | |
4160 | void btrfs_freeze_block_group(struct btrfs_block_group *cache) | |
4161 | { | |
4162 | atomic_inc(&cache->frozen); | |
4163 | } | |
4164 | ||
4165 | void btrfs_unfreeze_block_group(struct btrfs_block_group *block_group) | |
4166 | { | |
4167 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
4168 | struct extent_map_tree *em_tree; | |
4169 | struct extent_map *em; | |
4170 | bool cleanup; | |
4171 | ||
4172 | spin_lock(&block_group->lock); | |
4173 | cleanup = (atomic_dec_and_test(&block_group->frozen) && | |
3349b57f | 4174 | test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)); |
684b752b FM |
4175 | spin_unlock(&block_group->lock); |
4176 | ||
4177 | if (cleanup) { | |
684b752b FM |
4178 | em_tree = &fs_info->mapping_tree; |
4179 | write_lock(&em_tree->lock); | |
4180 | em = lookup_extent_mapping(em_tree, block_group->start, | |
4181 | 1); | |
4182 | BUG_ON(!em); /* logic error, can't happen */ | |
4183 | remove_extent_mapping(em_tree, em); | |
4184 | write_unlock(&em_tree->lock); | |
684b752b FM |
4185 | |
4186 | /* once for us and once for the tree */ | |
4187 | free_extent_map(em); | |
4188 | free_extent_map(em); | |
4189 | ||
4190 | /* | |
4191 | * We may have left one free space entry and other possible | |
4192 | * tasks trimming this block group have left 1 entry each one. | |
4193 | * Free them if any. | |
4194 | */ | |
fc80f7ac | 4195 | btrfs_remove_free_space_cache(block_group); |
684b752b FM |
4196 | } |
4197 | } | |
195a49ea FM |
4198 | |
4199 | bool btrfs_inc_block_group_swap_extents(struct btrfs_block_group *bg) | |
4200 | { | |
4201 | bool ret = true; | |
4202 | ||
4203 | spin_lock(&bg->lock); | |
4204 | if (bg->ro) | |
4205 | ret = false; | |
4206 | else | |
4207 | bg->swap_extents++; | |
4208 | spin_unlock(&bg->lock); | |
4209 | ||
4210 | return ret; | |
4211 | } | |
4212 | ||
4213 | void btrfs_dec_block_group_swap_extents(struct btrfs_block_group *bg, int amount) | |
4214 | { | |
4215 | spin_lock(&bg->lock); | |
4216 | ASSERT(!bg->ro); | |
4217 | ASSERT(bg->swap_extents >= amount); | |
4218 | bg->swap_extents -= amount; | |
4219 | spin_unlock(&bg->lock); | |
4220 | } |