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