]>
Commit | Line | Data |
---|---|---|
280c2908 JB |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | ||
784352fe | 3 | #include "misc.h" |
280c2908 JB |
4 | #include "ctree.h" |
5 | #include "space-info.h" | |
6 | #include "sysfs.h" | |
7 | #include "volumes.h" | |
5da6afeb | 8 | #include "free-space-cache.h" |
0d9764f6 JB |
9 | #include "ordered-data.h" |
10 | #include "transaction.h" | |
aac0023c | 11 | #include "block-group.h" |
280c2908 | 12 | |
4b8b0528 JB |
13 | /* |
14 | * HOW DOES SPACE RESERVATION WORK | |
15 | * | |
16 | * If you want to know about delalloc specifically, there is a separate comment | |
17 | * for that with the delalloc code. This comment is about how the whole system | |
18 | * works generally. | |
19 | * | |
20 | * BASIC CONCEPTS | |
21 | * | |
22 | * 1) space_info. This is the ultimate arbiter of how much space we can use. | |
23 | * There's a description of the bytes_ fields with the struct declaration, | |
24 | * refer to that for specifics on each field. Suffice it to say that for | |
25 | * reservations we care about total_bytes - SUM(space_info->bytes_) when | |
26 | * determining if there is space to make an allocation. There is a space_info | |
27 | * for METADATA, SYSTEM, and DATA areas. | |
28 | * | |
29 | * 2) block_rsv's. These are basically buckets for every different type of | |
30 | * metadata reservation we have. You can see the comment in the block_rsv | |
31 | * code on the rules for each type, but generally block_rsv->reserved is how | |
32 | * much space is accounted for in space_info->bytes_may_use. | |
33 | * | |
34 | * 3) btrfs_calc*_size. These are the worst case calculations we used based | |
35 | * on the number of items we will want to modify. We have one for changing | |
36 | * items, and one for inserting new items. Generally we use these helpers to | |
37 | * determine the size of the block reserves, and then use the actual bytes | |
38 | * values to adjust the space_info counters. | |
39 | * | |
40 | * MAKING RESERVATIONS, THE NORMAL CASE | |
41 | * | |
42 | * We call into either btrfs_reserve_data_bytes() or | |
43 | * btrfs_reserve_metadata_bytes(), depending on which we're looking for, with | |
44 | * num_bytes we want to reserve. | |
45 | * | |
46 | * ->reserve | |
47 | * space_info->bytes_may_reserve += num_bytes | |
48 | * | |
49 | * ->extent allocation | |
50 | * Call btrfs_add_reserved_bytes() which does | |
51 | * space_info->bytes_may_reserve -= num_bytes | |
52 | * space_info->bytes_reserved += extent_bytes | |
53 | * | |
54 | * ->insert reference | |
55 | * Call btrfs_update_block_group() which does | |
56 | * space_info->bytes_reserved -= extent_bytes | |
57 | * space_info->bytes_used += extent_bytes | |
58 | * | |
59 | * MAKING RESERVATIONS, FLUSHING NORMALLY (non-priority) | |
60 | * | |
61 | * Assume we are unable to simply make the reservation because we do not have | |
62 | * enough space | |
63 | * | |
64 | * -> __reserve_bytes | |
65 | * create a reserve_ticket with ->bytes set to our reservation, add it to | |
66 | * the tail of space_info->tickets, kick async flush thread | |
67 | * | |
68 | * ->handle_reserve_ticket | |
69 | * wait on ticket->wait for ->bytes to be reduced to 0, or ->error to be set | |
70 | * on the ticket. | |
71 | * | |
72 | * -> btrfs_async_reclaim_metadata_space/btrfs_async_reclaim_data_space | |
73 | * Flushes various things attempting to free up space. | |
74 | * | |
75 | * -> btrfs_try_granting_tickets() | |
76 | * This is called by anything that either subtracts space from | |
77 | * space_info->bytes_may_use, ->bytes_pinned, etc, or adds to the | |
78 | * space_info->total_bytes. This loops through the ->priority_tickets and | |
79 | * then the ->tickets list checking to see if the reservation can be | |
80 | * completed. If it can the space is added to space_info->bytes_may_use and | |
81 | * the ticket is woken up. | |
82 | * | |
83 | * -> ticket wakeup | |
84 | * Check if ->bytes == 0, if it does we got our reservation and we can carry | |
85 | * on, if not return the appropriate error (ENOSPC, but can be EINTR if we | |
86 | * were interrupted.) | |
87 | * | |
88 | * MAKING RESERVATIONS, FLUSHING HIGH PRIORITY | |
89 | * | |
90 | * Same as the above, except we add ourselves to the | |
91 | * space_info->priority_tickets, and we do not use ticket->wait, we simply | |
92 | * call flush_space() ourselves for the states that are safe for us to call | |
93 | * without deadlocking and hope for the best. | |
94 | * | |
95 | * THE FLUSHING STATES | |
96 | * | |
97 | * Generally speaking we will have two cases for each state, a "nice" state | |
98 | * and a "ALL THE THINGS" state. In btrfs we delay a lot of work in order to | |
99 | * reduce the locking over head on the various trees, and even to keep from | |
100 | * doing any work at all in the case of delayed refs. Each of these delayed | |
101 | * things however hold reservations, and so letting them run allows us to | |
102 | * reclaim space so we can make new reservations. | |
103 | * | |
104 | * FLUSH_DELAYED_ITEMS | |
105 | * Every inode has a delayed item to update the inode. Take a simple write | |
106 | * for example, we would update the inode item at write time to update the | |
107 | * mtime, and then again at finish_ordered_io() time in order to update the | |
108 | * isize or bytes. We keep these delayed items to coalesce these operations | |
109 | * into a single operation done on demand. These are an easy way to reclaim | |
110 | * metadata space. | |
111 | * | |
112 | * FLUSH_DELALLOC | |
113 | * Look at the delalloc comment to get an idea of how much space is reserved | |
114 | * for delayed allocation. We can reclaim some of this space simply by | |
115 | * running delalloc, but usually we need to wait for ordered extents to | |
116 | * reclaim the bulk of this space. | |
117 | * | |
118 | * FLUSH_DELAYED_REFS | |
119 | * We have a block reserve for the outstanding delayed refs space, and every | |
120 | * delayed ref operation holds a reservation. Running these is a quick way | |
121 | * to reclaim space, but we want to hold this until the end because COW can | |
122 | * churn a lot and we can avoid making some extent tree modifications if we | |
123 | * are able to delay for as long as possible. | |
124 | * | |
125 | * ALLOC_CHUNK | |
126 | * We will skip this the first time through space reservation, because of | |
127 | * overcommit and we don't want to have a lot of useless metadata space when | |
128 | * our worst case reservations will likely never come true. | |
129 | * | |
130 | * RUN_DELAYED_IPUTS | |
131 | * If we're freeing inodes we're likely freeing checksums, file extent | |
132 | * items, and extent tree items. Loads of space could be freed up by these | |
133 | * operations, however they won't be usable until the transaction commits. | |
134 | * | |
135 | * COMMIT_TRANS | |
136 | * may_commit_transaction() is the ultimate arbiter on whether we commit the | |
137 | * transaction or not. In order to avoid constantly churning we do all the | |
138 | * above flushing first and then commit the transaction as the last resort. | |
139 | * However we need to take into account things like pinned space that would | |
140 | * be freed, plus any delayed work we may not have gotten rid of in the case | |
141 | * of metadata. | |
142 | * | |
143 | * OVERCOMMIT | |
144 | * | |
145 | * Because we hold so many reservations for metadata we will allow you to | |
146 | * reserve more space than is currently free in the currently allocate | |
147 | * metadata space. This only happens with metadata, data does not allow | |
148 | * overcommitting. | |
149 | * | |
150 | * You can see the current logic for when we allow overcommit in | |
151 | * btrfs_can_overcommit(), but it only applies to unallocated space. If there | |
152 | * is no unallocated space to be had, all reservations are kept within the | |
153 | * free space in the allocated metadata chunks. | |
154 | * | |
155 | * Because of overcommitting, you generally want to use the | |
156 | * btrfs_can_overcommit() logic for metadata allocations, as it does the right | |
157 | * thing with or without extra unallocated space. | |
158 | */ | |
159 | ||
e1f60a65 | 160 | u64 __pure btrfs_space_info_used(struct btrfs_space_info *s_info, |
280c2908 JB |
161 | bool may_use_included) |
162 | { | |
163 | ASSERT(s_info); | |
164 | return s_info->bytes_used + s_info->bytes_reserved + | |
165 | s_info->bytes_pinned + s_info->bytes_readonly + | |
166 | (may_use_included ? s_info->bytes_may_use : 0); | |
167 | } | |
168 | ||
169 | /* | |
170 | * after adding space to the filesystem, we need to clear the full flags | |
171 | * on all the space infos. | |
172 | */ | |
173 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
174 | { | |
175 | struct list_head *head = &info->space_info; | |
176 | struct btrfs_space_info *found; | |
177 | ||
72804905 | 178 | list_for_each_entry(found, head, list) |
280c2908 | 179 | found->full = 0; |
280c2908 JB |
180 | } |
181 | ||
280c2908 JB |
182 | static int create_space_info(struct btrfs_fs_info *info, u64 flags) |
183 | { | |
184 | ||
185 | struct btrfs_space_info *space_info; | |
186 | int i; | |
187 | int ret; | |
188 | ||
189 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); | |
190 | if (!space_info) | |
191 | return -ENOMEM; | |
192 | ||
193 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, | |
194 | GFP_KERNEL); | |
195 | if (ret) { | |
196 | kfree(space_info); | |
197 | return ret; | |
198 | } | |
199 | ||
200 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
201 | INIT_LIST_HEAD(&space_info->block_groups[i]); | |
202 | init_rwsem(&space_info->groups_sem); | |
203 | spin_lock_init(&space_info->lock); | |
204 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
205 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
280c2908 JB |
206 | INIT_LIST_HEAD(&space_info->ro_bgs); |
207 | INIT_LIST_HEAD(&space_info->tickets); | |
208 | INIT_LIST_HEAD(&space_info->priority_tickets); | |
209 | ||
b882327a DS |
210 | ret = btrfs_sysfs_add_space_info_type(info, space_info); |
211 | if (ret) | |
280c2908 | 212 | return ret; |
280c2908 | 213 | |
72804905 | 214 | list_add(&space_info->list, &info->space_info); |
280c2908 JB |
215 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
216 | info->data_sinfo = space_info; | |
217 | ||
218 | return ret; | |
219 | } | |
220 | ||
221 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) | |
222 | { | |
223 | struct btrfs_super_block *disk_super; | |
224 | u64 features; | |
225 | u64 flags; | |
226 | int mixed = 0; | |
227 | int ret; | |
228 | ||
229 | disk_super = fs_info->super_copy; | |
230 | if (!btrfs_super_root(disk_super)) | |
231 | return -EINVAL; | |
232 | ||
233 | features = btrfs_super_incompat_flags(disk_super); | |
234 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
235 | mixed = 1; | |
236 | ||
237 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
238 | ret = create_space_info(fs_info, flags); | |
239 | if (ret) | |
240 | goto out; | |
241 | ||
242 | if (mixed) { | |
243 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
244 | ret = create_space_info(fs_info, flags); | |
245 | } else { | |
246 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
247 | ret = create_space_info(fs_info, flags); | |
248 | if (ret) | |
249 | goto out; | |
250 | ||
251 | flags = BTRFS_BLOCK_GROUP_DATA; | |
252 | ret = create_space_info(fs_info, flags); | |
253 | } | |
254 | out: | |
255 | return ret; | |
256 | } | |
257 | ||
258 | void btrfs_update_space_info(struct btrfs_fs_info *info, u64 flags, | |
259 | u64 total_bytes, u64 bytes_used, | |
260 | u64 bytes_readonly, | |
261 | struct btrfs_space_info **space_info) | |
262 | { | |
263 | struct btrfs_space_info *found; | |
264 | int factor; | |
265 | ||
266 | factor = btrfs_bg_type_to_factor(flags); | |
267 | ||
268 | found = btrfs_find_space_info(info, flags); | |
269 | ASSERT(found); | |
270 | spin_lock(&found->lock); | |
271 | found->total_bytes += total_bytes; | |
272 | found->disk_total += total_bytes * factor; | |
273 | found->bytes_used += bytes_used; | |
274 | found->disk_used += bytes_used * factor; | |
275 | found->bytes_readonly += bytes_readonly; | |
276 | if (total_bytes > 0) | |
277 | found->full = 0; | |
18fa2284 | 278 | btrfs_try_granting_tickets(info, found); |
280c2908 JB |
279 | spin_unlock(&found->lock); |
280 | *space_info = found; | |
281 | } | |
282 | ||
283 | struct btrfs_space_info *btrfs_find_space_info(struct btrfs_fs_info *info, | |
284 | u64 flags) | |
285 | { | |
286 | struct list_head *head = &info->space_info; | |
287 | struct btrfs_space_info *found; | |
288 | ||
289 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; | |
290 | ||
72804905 JB |
291 | list_for_each_entry(found, head, list) { |
292 | if (found->flags & flags) | |
280c2908 | 293 | return found; |
280c2908 | 294 | } |
280c2908 JB |
295 | return NULL; |
296 | } | |
41783ef2 | 297 | |
fa121a26 JB |
298 | static u64 calc_available_free_space(struct btrfs_fs_info *fs_info, |
299 | struct btrfs_space_info *space_info, | |
300 | enum btrfs_reserve_flush_enum flush) | |
41783ef2 | 301 | { |
41783ef2 | 302 | u64 profile; |
41783ef2 | 303 | u64 avail; |
41783ef2 JB |
304 | int factor; |
305 | ||
9f246926 | 306 | if (space_info->flags & BTRFS_BLOCK_GROUP_SYSTEM) |
41783ef2 JB |
307 | profile = btrfs_system_alloc_profile(fs_info); |
308 | else | |
309 | profile = btrfs_metadata_alloc_profile(fs_info); | |
310 | ||
41783ef2 JB |
311 | avail = atomic64_read(&fs_info->free_chunk_space); |
312 | ||
313 | /* | |
314 | * If we have dup, raid1 or raid10 then only half of the free | |
315 | * space is actually usable. For raid56, the space info used | |
316 | * doesn't include the parity drive, so we don't have to | |
317 | * change the math | |
318 | */ | |
319 | factor = btrfs_bg_type_to_factor(profile); | |
320 | avail = div_u64(avail, factor); | |
321 | ||
322 | /* | |
323 | * If we aren't flushing all things, let us overcommit up to | |
324 | * 1/2th of the space. If we can flush, don't let us overcommit | |
325 | * too much, let it overcommit up to 1/8 of the space. | |
326 | */ | |
327 | if (flush == BTRFS_RESERVE_FLUSH_ALL) | |
328 | avail >>= 3; | |
329 | else | |
330 | avail >>= 1; | |
fa121a26 JB |
331 | return avail; |
332 | } | |
333 | ||
334 | int btrfs_can_overcommit(struct btrfs_fs_info *fs_info, | |
335 | struct btrfs_space_info *space_info, u64 bytes, | |
336 | enum btrfs_reserve_flush_enum flush) | |
337 | { | |
338 | u64 avail; | |
339 | u64 used; | |
340 | ||
341 | /* Don't overcommit when in mixed mode */ | |
342 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
343 | return 0; | |
344 | ||
345 | used = btrfs_space_info_used(space_info, true); | |
346 | avail = calc_available_free_space(fs_info, space_info, flush); | |
41783ef2 JB |
347 | |
348 | if (used + bytes < space_info->total_bytes + avail) | |
349 | return 1; | |
350 | return 0; | |
351 | } | |
b338b013 | 352 | |
d611add4 FM |
353 | static void remove_ticket(struct btrfs_space_info *space_info, |
354 | struct reserve_ticket *ticket) | |
355 | { | |
356 | if (!list_empty(&ticket->list)) { | |
357 | list_del_init(&ticket->list); | |
358 | ASSERT(space_info->reclaim_size >= ticket->bytes); | |
359 | space_info->reclaim_size -= ticket->bytes; | |
360 | } | |
361 | } | |
362 | ||
b338b013 JB |
363 | /* |
364 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
365 | * basically when we're returning space from block_rsv's. | |
366 | */ | |
18fa2284 JB |
367 | void btrfs_try_granting_tickets(struct btrfs_fs_info *fs_info, |
368 | struct btrfs_space_info *space_info) | |
b338b013 | 369 | { |
b338b013 | 370 | struct list_head *head; |
b338b013 | 371 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; |
b338b013 | 372 | |
18fa2284 | 373 | lockdep_assert_held(&space_info->lock); |
b338b013 | 374 | |
18fa2284 | 375 | head = &space_info->priority_tickets; |
b338b013 | 376 | again: |
91182645 JB |
377 | while (!list_empty(head)) { |
378 | struct reserve_ticket *ticket; | |
379 | u64 used = btrfs_space_info_used(space_info, true); | |
380 | ||
381 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
382 | ||
383 | /* Check and see if our ticket can be satisified now. */ | |
384 | if ((used + ticket->bytes <= space_info->total_bytes) || | |
a30a3d20 JB |
385 | btrfs_can_overcommit(fs_info, space_info, ticket->bytes, |
386 | flush)) { | |
91182645 JB |
387 | btrfs_space_info_update_bytes_may_use(fs_info, |
388 | space_info, | |
389 | ticket->bytes); | |
d611add4 | 390 | remove_ticket(space_info, ticket); |
b338b013 JB |
391 | ticket->bytes = 0; |
392 | space_info->tickets_id++; | |
393 | wake_up(&ticket->wait); | |
394 | } else { | |
91182645 | 395 | break; |
b338b013 JB |
396 | } |
397 | } | |
398 | ||
91182645 | 399 | if (head == &space_info->priority_tickets) { |
b338b013 JB |
400 | head = &space_info->tickets; |
401 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
402 | goto again; | |
403 | } | |
b338b013 | 404 | } |
5da6afeb JB |
405 | |
406 | #define DUMP_BLOCK_RSV(fs_info, rsv_name) \ | |
407 | do { \ | |
408 | struct btrfs_block_rsv *__rsv = &(fs_info)->rsv_name; \ | |
409 | spin_lock(&__rsv->lock); \ | |
410 | btrfs_info(fs_info, #rsv_name ": size %llu reserved %llu", \ | |
411 | __rsv->size, __rsv->reserved); \ | |
412 | spin_unlock(&__rsv->lock); \ | |
413 | } while (0) | |
414 | ||
84fe47a4 JB |
415 | static void __btrfs_dump_space_info(struct btrfs_fs_info *fs_info, |
416 | struct btrfs_space_info *info) | |
5da6afeb | 417 | { |
84fe47a4 | 418 | lockdep_assert_held(&info->lock); |
5da6afeb | 419 | |
5da6afeb JB |
420 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
421 | info->flags, | |
422 | info->total_bytes - btrfs_space_info_used(info, true), | |
423 | info->full ? "" : "not "); | |
424 | btrfs_info(fs_info, | |
425 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
426 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
427 | info->bytes_reserved, info->bytes_may_use, | |
428 | info->bytes_readonly); | |
5da6afeb JB |
429 | |
430 | DUMP_BLOCK_RSV(fs_info, global_block_rsv); | |
431 | DUMP_BLOCK_RSV(fs_info, trans_block_rsv); | |
432 | DUMP_BLOCK_RSV(fs_info, chunk_block_rsv); | |
433 | DUMP_BLOCK_RSV(fs_info, delayed_block_rsv); | |
434 | DUMP_BLOCK_RSV(fs_info, delayed_refs_rsv); | |
435 | ||
84fe47a4 JB |
436 | } |
437 | ||
438 | void btrfs_dump_space_info(struct btrfs_fs_info *fs_info, | |
439 | struct btrfs_space_info *info, u64 bytes, | |
440 | int dump_block_groups) | |
441 | { | |
32da5386 | 442 | struct btrfs_block_group *cache; |
84fe47a4 JB |
443 | int index = 0; |
444 | ||
445 | spin_lock(&info->lock); | |
446 | __btrfs_dump_space_info(fs_info, info); | |
447 | spin_unlock(&info->lock); | |
448 | ||
5da6afeb JB |
449 | if (!dump_block_groups) |
450 | return; | |
451 | ||
452 | down_read(&info->groups_sem); | |
453 | again: | |
454 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
455 | spin_lock(&cache->lock); | |
456 | btrfs_info(fs_info, | |
457 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
b3470b5d | 458 | cache->start, cache->length, cache->used, cache->pinned, |
5da6afeb | 459 | cache->reserved, cache->ro ? "[readonly]" : ""); |
5da6afeb | 460 | spin_unlock(&cache->lock); |
ab0db043 | 461 | btrfs_dump_free_space(cache, bytes); |
5da6afeb JB |
462 | } |
463 | if (++index < BTRFS_NR_RAID_TYPES) | |
464 | goto again; | |
465 | up_read(&info->groups_sem); | |
466 | } | |
0d9764f6 | 467 | |
0d9764f6 JB |
468 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
469 | u64 to_reclaim) | |
470 | { | |
471 | u64 bytes; | |
472 | u64 nr; | |
473 | ||
2bd36e7b | 474 | bytes = btrfs_calc_insert_metadata_size(fs_info, 1); |
0d9764f6 JB |
475 | nr = div64_u64(to_reclaim, bytes); |
476 | if (!nr) | |
477 | nr = 1; | |
478 | return nr; | |
479 | } | |
480 | ||
481 | #define EXTENT_SIZE_PER_ITEM SZ_256K | |
482 | ||
483 | /* | |
484 | * shrink metadata reservation for delalloc | |
485 | */ | |
920a9958 JB |
486 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, |
487 | struct btrfs_space_info *space_info, | |
488 | u64 to_reclaim, bool wait_ordered) | |
0d9764f6 | 489 | { |
0d9764f6 JB |
490 | struct btrfs_trans_handle *trans; |
491 | u64 delalloc_bytes; | |
492 | u64 dio_bytes; | |
0d9764f6 JB |
493 | u64 items; |
494 | long time_left; | |
0d9764f6 JB |
495 | int loops; |
496 | ||
497 | /* Calc the number of the pages we need flush for space reservation */ | |
d7f81fac JB |
498 | if (to_reclaim == U64_MAX) { |
499 | items = U64_MAX; | |
500 | } else { | |
501 | /* | |
502 | * to_reclaim is set to however much metadata we need to | |
503 | * reclaim, but reclaiming that much data doesn't really track | |
504 | * exactly, so increase the amount to reclaim by 2x in order to | |
505 | * make sure we're flushing enough delalloc to hopefully reclaim | |
506 | * some metadata reservations. | |
507 | */ | |
508 | items = calc_reclaim_items_nr(fs_info, to_reclaim) * 2; | |
509 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; | |
510 | } | |
0d9764f6 JB |
511 | |
512 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
0d9764f6 JB |
513 | |
514 | delalloc_bytes = percpu_counter_sum_positive( | |
515 | &fs_info->delalloc_bytes); | |
516 | dio_bytes = percpu_counter_sum_positive(&fs_info->dio_bytes); | |
517 | if (delalloc_bytes == 0 && dio_bytes == 0) { | |
518 | if (trans) | |
519 | return; | |
520 | if (wait_ordered) | |
521 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); | |
522 | return; | |
523 | } | |
524 | ||
525 | /* | |
526 | * If we are doing more ordered than delalloc we need to just wait on | |
527 | * ordered extents, otherwise we'll waste time trying to flush delalloc | |
528 | * that likely won't give us the space back we need. | |
529 | */ | |
530 | if (dio_bytes > delalloc_bytes) | |
531 | wait_ordered = true; | |
532 | ||
533 | loops = 0; | |
534 | while ((delalloc_bytes || dio_bytes) && loops < 3) { | |
e076ab2a JB |
535 | u64 nr_pages = min(delalloc_bytes, to_reclaim) >> PAGE_SHIFT; |
536 | ||
537 | btrfs_start_delalloc_roots(fs_info, nr_pages, true); | |
0d9764f6 | 538 | |
0d9764f6 JB |
539 | loops++; |
540 | if (wait_ordered && !trans) { | |
541 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); | |
542 | } else { | |
543 | time_left = schedule_timeout_killable(1); | |
544 | if (time_left) | |
545 | break; | |
546 | } | |
448b966b JB |
547 | |
548 | spin_lock(&space_info->lock); | |
549 | if (list_empty(&space_info->tickets) && | |
550 | list_empty(&space_info->priority_tickets)) { | |
551 | spin_unlock(&space_info->lock); | |
552 | break; | |
553 | } | |
554 | spin_unlock(&space_info->lock); | |
555 | ||
0d9764f6 JB |
556 | delalloc_bytes = percpu_counter_sum_positive( |
557 | &fs_info->delalloc_bytes); | |
558 | dio_bytes = percpu_counter_sum_positive(&fs_info->dio_bytes); | |
559 | } | |
560 | } | |
561 | ||
562 | /** | |
563 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
564 | * @root - the root we're allocating for | |
565 | * @bytes - the number of bytes we want to reserve | |
566 | * @force - force the commit | |
567 | * | |
568 | * This will check to make sure that committing the transaction will actually | |
569 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
570 | * will return -ENOSPC. | |
571 | */ | |
572 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, | |
bb86bd3d | 573 | struct btrfs_space_info *space_info) |
0d9764f6 JB |
574 | { |
575 | struct reserve_ticket *ticket = NULL; | |
576 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; | |
577 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; | |
bb4f58a7 | 578 | struct btrfs_block_rsv *trans_rsv = &fs_info->trans_block_rsv; |
0d9764f6 | 579 | struct btrfs_trans_handle *trans; |
0d9764f6 | 580 | u64 reclaim_bytes = 0; |
bb86bd3d | 581 | u64 bytes_needed = 0; |
00c0135e | 582 | u64 cur_free_bytes = 0; |
0d9764f6 JB |
583 | |
584 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
585 | if (trans) | |
586 | return -EAGAIN; | |
587 | ||
588 | spin_lock(&space_info->lock); | |
00c0135e JB |
589 | cur_free_bytes = btrfs_space_info_used(space_info, true); |
590 | if (cur_free_bytes < space_info->total_bytes) | |
591 | cur_free_bytes = space_info->total_bytes - cur_free_bytes; | |
592 | else | |
593 | cur_free_bytes = 0; | |
594 | ||
0d9764f6 JB |
595 | if (!list_empty(&space_info->priority_tickets)) |
596 | ticket = list_first_entry(&space_info->priority_tickets, | |
597 | struct reserve_ticket, list); | |
598 | else if (!list_empty(&space_info->tickets)) | |
599 | ticket = list_first_entry(&space_info->tickets, | |
600 | struct reserve_ticket, list); | |
a1ed0a82 JB |
601 | if (ticket) |
602 | bytes_needed = ticket->bytes; | |
00c0135e JB |
603 | |
604 | if (bytes_needed > cur_free_bytes) | |
605 | bytes_needed -= cur_free_bytes; | |
606 | else | |
607 | bytes_needed = 0; | |
0d9764f6 JB |
608 | spin_unlock(&space_info->lock); |
609 | ||
610 | if (!bytes_needed) | |
611 | return 0; | |
612 | ||
613 | trans = btrfs_join_transaction(fs_info->extent_root); | |
614 | if (IS_ERR(trans)) | |
615 | return PTR_ERR(trans); | |
616 | ||
617 | /* | |
618 | * See if there is enough pinned space to make this reservation, or if | |
619 | * we have block groups that are going to be freed, allowing us to | |
620 | * possibly do a chunk allocation the next loop through. | |
621 | */ | |
bb86bd3d | 622 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags) || |
0d9764f6 JB |
623 | __percpu_counter_compare(&space_info->total_bytes_pinned, |
624 | bytes_needed, | |
625 | BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0) | |
626 | goto commit; | |
627 | ||
628 | /* | |
a1ed0a82 JB |
629 | * See if there is some space in the delayed insertion reserve for this |
630 | * reservation. If the space_info's don't match (like for DATA or | |
631 | * SYSTEM) then just go enospc, reclaiming this space won't recover any | |
632 | * space to satisfy those reservations. | |
0d9764f6 JB |
633 | */ |
634 | if (space_info != delayed_rsv->space_info) | |
635 | goto enospc; | |
636 | ||
637 | spin_lock(&delayed_rsv->lock); | |
638 | reclaim_bytes += delayed_rsv->reserved; | |
639 | spin_unlock(&delayed_rsv->lock); | |
640 | ||
641 | spin_lock(&delayed_refs_rsv->lock); | |
642 | reclaim_bytes += delayed_refs_rsv->reserved; | |
643 | spin_unlock(&delayed_refs_rsv->lock); | |
bb4f58a7 JB |
644 | |
645 | spin_lock(&trans_rsv->lock); | |
646 | reclaim_bytes += trans_rsv->reserved; | |
647 | spin_unlock(&trans_rsv->lock); | |
648 | ||
0d9764f6 JB |
649 | if (reclaim_bytes >= bytes_needed) |
650 | goto commit; | |
651 | bytes_needed -= reclaim_bytes; | |
652 | ||
653 | if (__percpu_counter_compare(&space_info->total_bytes_pinned, | |
654 | bytes_needed, | |
655 | BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) | |
656 | goto enospc; | |
657 | ||
658 | commit: | |
659 | return btrfs_commit_transaction(trans); | |
660 | enospc: | |
661 | btrfs_end_transaction(trans); | |
662 | return -ENOSPC; | |
663 | } | |
664 | ||
665 | /* | |
666 | * Try to flush some data based on policy set by @state. This is only advisory | |
667 | * and may fail for various reasons. The caller is supposed to examine the | |
668 | * state of @space_info to detect the outcome. | |
669 | */ | |
670 | static void flush_space(struct btrfs_fs_info *fs_info, | |
671 | struct btrfs_space_info *space_info, u64 num_bytes, | |
672 | int state) | |
673 | { | |
674 | struct btrfs_root *root = fs_info->extent_root; | |
675 | struct btrfs_trans_handle *trans; | |
676 | int nr; | |
677 | int ret = 0; | |
678 | ||
679 | switch (state) { | |
680 | case FLUSH_DELAYED_ITEMS_NR: | |
681 | case FLUSH_DELAYED_ITEMS: | |
682 | if (state == FLUSH_DELAYED_ITEMS_NR) | |
683 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; | |
684 | else | |
685 | nr = -1; | |
686 | ||
687 | trans = btrfs_join_transaction(root); | |
688 | if (IS_ERR(trans)) { | |
689 | ret = PTR_ERR(trans); | |
690 | break; | |
691 | } | |
692 | ret = btrfs_run_delayed_items_nr(trans, nr); | |
693 | btrfs_end_transaction(trans); | |
694 | break; | |
695 | case FLUSH_DELALLOC: | |
696 | case FLUSH_DELALLOC_WAIT: | |
920a9958 | 697 | shrink_delalloc(fs_info, space_info, num_bytes, |
0d9764f6 JB |
698 | state == FLUSH_DELALLOC_WAIT); |
699 | break; | |
700 | case FLUSH_DELAYED_REFS_NR: | |
701 | case FLUSH_DELAYED_REFS: | |
702 | trans = btrfs_join_transaction(root); | |
703 | if (IS_ERR(trans)) { | |
704 | ret = PTR_ERR(trans); | |
705 | break; | |
706 | } | |
707 | if (state == FLUSH_DELAYED_REFS_NR) | |
708 | nr = calc_reclaim_items_nr(fs_info, num_bytes); | |
709 | else | |
710 | nr = 0; | |
711 | btrfs_run_delayed_refs(trans, nr); | |
712 | btrfs_end_transaction(trans); | |
713 | break; | |
714 | case ALLOC_CHUNK: | |
715 | case ALLOC_CHUNK_FORCE: | |
716 | trans = btrfs_join_transaction(root); | |
717 | if (IS_ERR(trans)) { | |
718 | ret = PTR_ERR(trans); | |
719 | break; | |
720 | } | |
721 | ret = btrfs_chunk_alloc(trans, | |
c6c45303 | 722 | btrfs_get_alloc_profile(fs_info, space_info->flags), |
0d9764f6 JB |
723 | (state == ALLOC_CHUNK) ? CHUNK_ALLOC_NO_FORCE : |
724 | CHUNK_ALLOC_FORCE); | |
725 | btrfs_end_transaction(trans); | |
726 | if (ret > 0 || ret == -ENOSPC) | |
727 | ret = 0; | |
728 | break; | |
844245b4 | 729 | case RUN_DELAYED_IPUTS: |
0d9764f6 JB |
730 | /* |
731 | * If we have pending delayed iputs then we could free up a | |
732 | * bunch of pinned space, so make sure we run the iputs before | |
733 | * we do our pinned bytes check below. | |
734 | */ | |
735 | btrfs_run_delayed_iputs(fs_info); | |
736 | btrfs_wait_on_delayed_iputs(fs_info); | |
844245b4 JB |
737 | break; |
738 | case COMMIT_TRANS: | |
bb86bd3d | 739 | ret = may_commit_transaction(fs_info, space_info); |
0d9764f6 JB |
740 | break; |
741 | default: | |
742 | ret = -ENOSPC; | |
743 | break; | |
744 | } | |
745 | ||
746 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, | |
747 | ret); | |
748 | return; | |
749 | } | |
750 | ||
751 | static inline u64 | |
752 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, | |
9f246926 | 753 | struct btrfs_space_info *space_info) |
0d9764f6 | 754 | { |
0d9764f6 | 755 | u64 used; |
fa121a26 | 756 | u64 avail; |
0d9764f6 | 757 | u64 expected; |
db161806 | 758 | u64 to_reclaim = space_info->reclaim_size; |
0d9764f6 | 759 | |
db161806 | 760 | lockdep_assert_held(&space_info->lock); |
fa121a26 JB |
761 | |
762 | avail = calc_available_free_space(fs_info, space_info, | |
763 | BTRFS_RESERVE_FLUSH_ALL); | |
764 | used = btrfs_space_info_used(space_info, true); | |
765 | ||
766 | /* | |
767 | * We may be flushing because suddenly we have less space than we had | |
768 | * before, and now we're well over-committed based on our current free | |
769 | * space. If that's the case add in our overage so we make sure to put | |
770 | * appropriate pressure on the flushing state machine. | |
771 | */ | |
772 | if (space_info->total_bytes + avail < used) | |
773 | to_reclaim += used - (space_info->total_bytes + avail); | |
774 | ||
0d9764f6 JB |
775 | if (to_reclaim) |
776 | return to_reclaim; | |
777 | ||
778 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); | |
a30a3d20 JB |
779 | if (btrfs_can_overcommit(fs_info, space_info, to_reclaim, |
780 | BTRFS_RESERVE_FLUSH_ALL)) | |
0d9764f6 JB |
781 | return 0; |
782 | ||
783 | used = btrfs_space_info_used(space_info, true); | |
784 | ||
a30a3d20 JB |
785 | if (btrfs_can_overcommit(fs_info, space_info, SZ_1M, |
786 | BTRFS_RESERVE_FLUSH_ALL)) | |
0d9764f6 JB |
787 | expected = div_factor_fine(space_info->total_bytes, 95); |
788 | else | |
789 | expected = div_factor_fine(space_info->total_bytes, 90); | |
790 | ||
791 | if (used > expected) | |
792 | to_reclaim = used - expected; | |
793 | else | |
794 | to_reclaim = 0; | |
795 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
796 | space_info->bytes_reserved); | |
797 | return to_reclaim; | |
798 | } | |
799 | ||
800 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, | |
801 | struct btrfs_space_info *space_info, | |
9f246926 | 802 | u64 used) |
0d9764f6 JB |
803 | { |
804 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); | |
805 | ||
806 | /* If we're just plain full then async reclaim just slows us down. */ | |
807 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) | |
808 | return 0; | |
809 | ||
9f246926 | 810 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info)) |
0d9764f6 JB |
811 | return 0; |
812 | ||
813 | return (used >= thresh && !btrfs_fs_closing(fs_info) && | |
814 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
815 | } | |
816 | ||
7f9fe614 JB |
817 | static bool steal_from_global_rsv(struct btrfs_fs_info *fs_info, |
818 | struct btrfs_space_info *space_info, | |
819 | struct reserve_ticket *ticket) | |
820 | { | |
821 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
822 | u64 min_bytes; | |
823 | ||
824 | if (global_rsv->space_info != space_info) | |
825 | return false; | |
826 | ||
827 | spin_lock(&global_rsv->lock); | |
e6549c2a | 828 | min_bytes = div_factor(global_rsv->size, 1); |
7f9fe614 JB |
829 | if (global_rsv->reserved < min_bytes + ticket->bytes) { |
830 | spin_unlock(&global_rsv->lock); | |
831 | return false; | |
832 | } | |
833 | global_rsv->reserved -= ticket->bytes; | |
6d548b9e | 834 | remove_ticket(space_info, ticket); |
7f9fe614 | 835 | ticket->bytes = 0; |
7f9fe614 JB |
836 | wake_up(&ticket->wait); |
837 | space_info->tickets_id++; | |
838 | if (global_rsv->reserved < global_rsv->size) | |
839 | global_rsv->full = 0; | |
840 | spin_unlock(&global_rsv->lock); | |
841 | ||
842 | return true; | |
843 | } | |
844 | ||
2341ccd1 JB |
845 | /* |
846 | * maybe_fail_all_tickets - we've exhausted our flushing, start failing tickets | |
847 | * @fs_info - fs_info for this fs | |
848 | * @space_info - the space info we were flushing | |
849 | * | |
850 | * We call this when we've exhausted our flushing ability and haven't made | |
851 | * progress in satisfying tickets. The reservation code handles tickets in | |
852 | * order, so if there is a large ticket first and then smaller ones we could | |
853 | * very well satisfy the smaller tickets. This will attempt to wake up any | |
854 | * tickets in the list to catch this case. | |
855 | * | |
856 | * This function returns true if it was able to make progress by clearing out | |
857 | * other tickets, or if it stumbles across a ticket that was smaller than the | |
858 | * first ticket. | |
859 | */ | |
860 | static bool maybe_fail_all_tickets(struct btrfs_fs_info *fs_info, | |
861 | struct btrfs_space_info *space_info) | |
0d9764f6 JB |
862 | { |
863 | struct reserve_ticket *ticket; | |
2341ccd1 JB |
864 | u64 tickets_id = space_info->tickets_id; |
865 | u64 first_ticket_bytes = 0; | |
866 | ||
84fe47a4 JB |
867 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
868 | btrfs_info(fs_info, "cannot satisfy tickets, dumping space info"); | |
869 | __btrfs_dump_space_info(fs_info, space_info); | |
870 | } | |
871 | ||
2341ccd1 JB |
872 | while (!list_empty(&space_info->tickets) && |
873 | tickets_id == space_info->tickets_id) { | |
874 | ticket = list_first_entry(&space_info->tickets, | |
875 | struct reserve_ticket, list); | |
876 | ||
7f9fe614 JB |
877 | if (ticket->steal && |
878 | steal_from_global_rsv(fs_info, space_info, ticket)) | |
879 | return true; | |
880 | ||
2341ccd1 JB |
881 | /* |
882 | * may_commit_transaction will avoid committing the transaction | |
883 | * if it doesn't feel like the space reclaimed by the commit | |
884 | * would result in the ticket succeeding. However if we have a | |
885 | * smaller ticket in the queue it may be small enough to be | |
886 | * satisified by committing the transaction, so if any | |
887 | * subsequent ticket is smaller than the first ticket go ahead | |
888 | * and send us back for another loop through the enospc flushing | |
889 | * code. | |
890 | */ | |
891 | if (first_ticket_bytes == 0) | |
892 | first_ticket_bytes = ticket->bytes; | |
893 | else if (first_ticket_bytes > ticket->bytes) | |
894 | return true; | |
0d9764f6 | 895 | |
84fe47a4 JB |
896 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) |
897 | btrfs_info(fs_info, "failing ticket with %llu bytes", | |
898 | ticket->bytes); | |
899 | ||
d611add4 | 900 | remove_ticket(space_info, ticket); |
0d9764f6 JB |
901 | ticket->error = -ENOSPC; |
902 | wake_up(&ticket->wait); | |
2341ccd1 JB |
903 | |
904 | /* | |
905 | * We're just throwing tickets away, so more flushing may not | |
906 | * trip over btrfs_try_granting_tickets, so we need to call it | |
907 | * here to see if we can make progress with the next ticket in | |
908 | * the list. | |
909 | */ | |
910 | btrfs_try_granting_tickets(fs_info, space_info); | |
0d9764f6 | 911 | } |
2341ccd1 | 912 | return (tickets_id != space_info->tickets_id); |
0d9764f6 JB |
913 | } |
914 | ||
915 | /* | |
916 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
917 | * will loop and continuously try to flush as long as we are making progress. | |
918 | * We count progress as clearing off tickets each time we have to loop. | |
919 | */ | |
920 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) | |
921 | { | |
922 | struct btrfs_fs_info *fs_info; | |
923 | struct btrfs_space_info *space_info; | |
924 | u64 to_reclaim; | |
925 | int flush_state; | |
926 | int commit_cycles = 0; | |
927 | u64 last_tickets_id; | |
928 | ||
929 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
930 | space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
931 | ||
932 | spin_lock(&space_info->lock); | |
9f246926 | 933 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info); |
0d9764f6 JB |
934 | if (!to_reclaim) { |
935 | space_info->flush = 0; | |
936 | spin_unlock(&space_info->lock); | |
937 | return; | |
938 | } | |
939 | last_tickets_id = space_info->tickets_id; | |
940 | spin_unlock(&space_info->lock); | |
941 | ||
942 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
943 | do { | |
944 | flush_space(fs_info, space_info, to_reclaim, flush_state); | |
945 | spin_lock(&space_info->lock); | |
946 | if (list_empty(&space_info->tickets)) { | |
947 | space_info->flush = 0; | |
948 | spin_unlock(&space_info->lock); | |
949 | return; | |
950 | } | |
951 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, | |
9f246926 | 952 | space_info); |
0d9764f6 JB |
953 | if (last_tickets_id == space_info->tickets_id) { |
954 | flush_state++; | |
955 | } else { | |
956 | last_tickets_id = space_info->tickets_id; | |
957 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
958 | if (commit_cycles) | |
959 | commit_cycles--; | |
960 | } | |
961 | ||
962 | /* | |
963 | * We don't want to force a chunk allocation until we've tried | |
964 | * pretty hard to reclaim space. Think of the case where we | |
965 | * freed up a bunch of space and so have a lot of pinned space | |
966 | * to reclaim. We would rather use that than possibly create a | |
967 | * underutilized metadata chunk. So if this is our first run | |
968 | * through the flushing state machine skip ALLOC_CHUNK_FORCE and | |
969 | * commit the transaction. If nothing has changed the next go | |
970 | * around then we can force a chunk allocation. | |
971 | */ | |
972 | if (flush_state == ALLOC_CHUNK_FORCE && !commit_cycles) | |
973 | flush_state++; | |
974 | ||
975 | if (flush_state > COMMIT_TRANS) { | |
976 | commit_cycles++; | |
977 | if (commit_cycles > 2) { | |
2341ccd1 | 978 | if (maybe_fail_all_tickets(fs_info, space_info)) { |
0d9764f6 JB |
979 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
980 | commit_cycles--; | |
981 | } else { | |
982 | space_info->flush = 0; | |
983 | } | |
984 | } else { | |
985 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
986 | } | |
987 | } | |
988 | spin_unlock(&space_info->lock); | |
989 | } while (flush_state <= COMMIT_TRANS); | |
990 | } | |
991 | ||
1a7a92c8 JB |
992 | /* |
993 | * FLUSH_DELALLOC_WAIT: | |
994 | * Space is freed from flushing delalloc in one of two ways. | |
995 | * | |
996 | * 1) compression is on and we allocate less space than we reserved | |
997 | * 2) we are overwriting existing space | |
998 | * | |
999 | * For #1 that extra space is reclaimed as soon as the delalloc pages are | |
1000 | * COWed, by way of btrfs_add_reserved_bytes() which adds the actual extent | |
1001 | * length to ->bytes_reserved, and subtracts the reserved space from | |
1002 | * ->bytes_may_use. | |
1003 | * | |
1004 | * For #2 this is trickier. Once the ordered extent runs we will drop the | |
1005 | * extent in the range we are overwriting, which creates a delayed ref for | |
1006 | * that freed extent. This however is not reclaimed until the transaction | |
1007 | * commits, thus the next stages. | |
1008 | * | |
1009 | * RUN_DELAYED_IPUTS | |
1010 | * If we are freeing inodes, we want to make sure all delayed iputs have | |
1011 | * completed, because they could have been on an inode with i_nlink == 0, and | |
1012 | * thus have been truncated and freed up space. But again this space is not | |
1013 | * immediately re-usable, it comes in the form of a delayed ref, which must be | |
1014 | * run and then the transaction must be committed. | |
1015 | * | |
1016 | * FLUSH_DELAYED_REFS | |
1017 | * The above two cases generate delayed refs that will affect | |
1018 | * ->total_bytes_pinned. However this counter can be inconsistent with | |
1019 | * reality if there are outstanding delayed refs. This is because we adjust | |
1020 | * the counter based solely on the current set of delayed refs and disregard | |
1021 | * any on-disk state which might include more refs. So for example, if we | |
1022 | * have an extent with 2 references, but we only drop 1, we'll see that there | |
1023 | * is a negative delayed ref count for the extent and assume that the space | |
1024 | * will be freed, and thus increase ->total_bytes_pinned. | |
1025 | * | |
1026 | * Running the delayed refs gives us the actual real view of what will be | |
1027 | * freed at the transaction commit time. This stage will not actually free | |
1028 | * space for us, it just makes sure that may_commit_transaction() has all of | |
1029 | * the information it needs to make the right decision. | |
1030 | * | |
1031 | * COMMIT_TRANS | |
1032 | * This is where we reclaim all of the pinned space generated by the previous | |
1033 | * two stages. We will not commit the transaction if we don't think we're | |
1034 | * likely to satisfy our request, which means if our current free space + | |
1035 | * total_bytes_pinned < reservation we will not commit. This is why the | |
1036 | * previous states are actually important, to make sure we know for sure | |
1037 | * whether committing the transaction will allow us to make progress. | |
c4923027 JB |
1038 | * |
1039 | * ALLOC_CHUNK_FORCE | |
1040 | * For data we start with alloc chunk force, however we could have been full | |
1041 | * before, and then the transaction commit could have freed new block groups, | |
1042 | * so if we now have space to allocate do the force chunk allocation. | |
1a7a92c8 | 1043 | */ |
57056740 JB |
1044 | static const enum btrfs_flush_state data_flush_states[] = { |
1045 | FLUSH_DELALLOC_WAIT, | |
1046 | RUN_DELAYED_IPUTS, | |
1047 | FLUSH_DELAYED_REFS, | |
1048 | COMMIT_TRANS, | |
c4923027 | 1049 | ALLOC_CHUNK_FORCE, |
57056740 JB |
1050 | }; |
1051 | ||
1052 | static void btrfs_async_reclaim_data_space(struct work_struct *work) | |
0d9764f6 | 1053 | { |
57056740 JB |
1054 | struct btrfs_fs_info *fs_info; |
1055 | struct btrfs_space_info *space_info; | |
1056 | u64 last_tickets_id; | |
1057 | int flush_state = 0; | |
1058 | ||
1059 | fs_info = container_of(work, struct btrfs_fs_info, async_data_reclaim_work); | |
1060 | space_info = fs_info->data_sinfo; | |
1061 | ||
1062 | spin_lock(&space_info->lock); | |
1063 | if (list_empty(&space_info->tickets)) { | |
1064 | space_info->flush = 0; | |
1065 | spin_unlock(&space_info->lock); | |
1066 | return; | |
1067 | } | |
1068 | last_tickets_id = space_info->tickets_id; | |
1069 | spin_unlock(&space_info->lock); | |
1070 | ||
1071 | while (!space_info->full) { | |
1072 | flush_space(fs_info, space_info, U64_MAX, ALLOC_CHUNK_FORCE); | |
1073 | spin_lock(&space_info->lock); | |
1074 | if (list_empty(&space_info->tickets)) { | |
1075 | space_info->flush = 0; | |
1076 | spin_unlock(&space_info->lock); | |
1077 | return; | |
1078 | } | |
1079 | last_tickets_id = space_info->tickets_id; | |
1080 | spin_unlock(&space_info->lock); | |
1081 | } | |
1082 | ||
1083 | while (flush_state < ARRAY_SIZE(data_flush_states)) { | |
1084 | flush_space(fs_info, space_info, U64_MAX, | |
1085 | data_flush_states[flush_state]); | |
1086 | spin_lock(&space_info->lock); | |
1087 | if (list_empty(&space_info->tickets)) { | |
1088 | space_info->flush = 0; | |
1089 | spin_unlock(&space_info->lock); | |
1090 | return; | |
1091 | } | |
1092 | ||
1093 | if (last_tickets_id == space_info->tickets_id) { | |
1094 | flush_state++; | |
1095 | } else { | |
1096 | last_tickets_id = space_info->tickets_id; | |
1097 | flush_state = 0; | |
1098 | } | |
1099 | ||
1100 | if (flush_state >= ARRAY_SIZE(data_flush_states)) { | |
1101 | if (space_info->full) { | |
1102 | if (maybe_fail_all_tickets(fs_info, space_info)) | |
1103 | flush_state = 0; | |
1104 | else | |
1105 | space_info->flush = 0; | |
1106 | } else { | |
1107 | flush_state = 0; | |
1108 | } | |
1109 | } | |
1110 | spin_unlock(&space_info->lock); | |
1111 | } | |
1112 | } | |
1113 | ||
1114 | void btrfs_init_async_reclaim_work(struct btrfs_fs_info *fs_info) | |
1115 | { | |
1116 | INIT_WORK(&fs_info->async_reclaim_work, btrfs_async_reclaim_metadata_space); | |
1117 | INIT_WORK(&fs_info->async_data_reclaim_work, btrfs_async_reclaim_data_space); | |
0d9764f6 JB |
1118 | } |
1119 | ||
1120 | static const enum btrfs_flush_state priority_flush_states[] = { | |
1121 | FLUSH_DELAYED_ITEMS_NR, | |
1122 | FLUSH_DELAYED_ITEMS, | |
1123 | ALLOC_CHUNK, | |
1124 | }; | |
1125 | ||
d3984c90 JB |
1126 | static const enum btrfs_flush_state evict_flush_states[] = { |
1127 | FLUSH_DELAYED_ITEMS_NR, | |
1128 | FLUSH_DELAYED_ITEMS, | |
1129 | FLUSH_DELAYED_REFS_NR, | |
1130 | FLUSH_DELAYED_REFS, | |
1131 | FLUSH_DELALLOC, | |
1132 | FLUSH_DELALLOC_WAIT, | |
1133 | ALLOC_CHUNK, | |
1134 | COMMIT_TRANS, | |
1135 | }; | |
1136 | ||
0d9764f6 | 1137 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, |
9ce2f423 JB |
1138 | struct btrfs_space_info *space_info, |
1139 | struct reserve_ticket *ticket, | |
1140 | const enum btrfs_flush_state *states, | |
1141 | int states_nr) | |
0d9764f6 JB |
1142 | { |
1143 | u64 to_reclaim; | |
1144 | int flush_state; | |
1145 | ||
1146 | spin_lock(&space_info->lock); | |
9f246926 | 1147 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info); |
0d9764f6 JB |
1148 | if (!to_reclaim) { |
1149 | spin_unlock(&space_info->lock); | |
1150 | return; | |
1151 | } | |
1152 | spin_unlock(&space_info->lock); | |
1153 | ||
1154 | flush_state = 0; | |
1155 | do { | |
9ce2f423 | 1156 | flush_space(fs_info, space_info, to_reclaim, states[flush_state]); |
0d9764f6 JB |
1157 | flush_state++; |
1158 | spin_lock(&space_info->lock); | |
1159 | if (ticket->bytes == 0) { | |
1160 | spin_unlock(&space_info->lock); | |
1161 | return; | |
1162 | } | |
1163 | spin_unlock(&space_info->lock); | |
9ce2f423 | 1164 | } while (flush_state < states_nr); |
0d9764f6 JB |
1165 | } |
1166 | ||
1004f686 JB |
1167 | static void priority_reclaim_data_space(struct btrfs_fs_info *fs_info, |
1168 | struct btrfs_space_info *space_info, | |
57056740 | 1169 | struct reserve_ticket *ticket) |
1004f686 | 1170 | { |
1004f686 JB |
1171 | while (!space_info->full) { |
1172 | flush_space(fs_info, space_info, U64_MAX, ALLOC_CHUNK_FORCE); | |
1173 | spin_lock(&space_info->lock); | |
1174 | if (ticket->bytes == 0) { | |
1175 | spin_unlock(&space_info->lock); | |
1176 | return; | |
1177 | } | |
1178 | spin_unlock(&space_info->lock); | |
1179 | } | |
1004f686 JB |
1180 | } |
1181 | ||
374bf9c5 JB |
1182 | static void wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
1183 | struct btrfs_space_info *space_info, | |
1184 | struct reserve_ticket *ticket) | |
0d9764f6 JB |
1185 | |
1186 | { | |
1187 | DEFINE_WAIT(wait); | |
0d9764f6 JB |
1188 | int ret = 0; |
1189 | ||
1190 | spin_lock(&space_info->lock); | |
1191 | while (ticket->bytes > 0 && ticket->error == 0) { | |
1192 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
1193 | if (ret) { | |
0cab7acc FM |
1194 | /* |
1195 | * Delete us from the list. After we unlock the space | |
1196 | * info, we don't want the async reclaim job to reserve | |
1197 | * space for this ticket. If that would happen, then the | |
1198 | * ticket's task would not known that space was reserved | |
1199 | * despite getting an error, resulting in a space leak | |
1200 | * (bytes_may_use counter of our space_info). | |
1201 | */ | |
d611add4 | 1202 | remove_ticket(space_info, ticket); |
374bf9c5 | 1203 | ticket->error = -EINTR; |
0d9764f6 JB |
1204 | break; |
1205 | } | |
1206 | spin_unlock(&space_info->lock); | |
1207 | ||
1208 | schedule(); | |
1209 | ||
1210 | finish_wait(&ticket->wait, &wait); | |
1211 | spin_lock(&space_info->lock); | |
1212 | } | |
0d9764f6 | 1213 | spin_unlock(&space_info->lock); |
0d9764f6 JB |
1214 | } |
1215 | ||
03235279 JB |
1216 | /** |
1217 | * handle_reserve_ticket - do the appropriate flushing and waiting for a ticket | |
1218 | * @fs_info - the fs | |
1219 | * @space_info - the space_info for the reservation | |
1220 | * @ticket - the ticket for the reservation | |
1221 | * @flush - how much we can flush | |
1222 | * | |
1223 | * This does the work of figuring out how to flush for the ticket, waiting for | |
1224 | * the reservation, and returning the appropriate error if there is one. | |
1225 | */ | |
1226 | static int handle_reserve_ticket(struct btrfs_fs_info *fs_info, | |
1227 | struct btrfs_space_info *space_info, | |
1228 | struct reserve_ticket *ticket, | |
1229 | enum btrfs_reserve_flush_enum flush) | |
1230 | { | |
03235279 JB |
1231 | int ret; |
1232 | ||
d3984c90 | 1233 | switch (flush) { |
57056740 | 1234 | case BTRFS_RESERVE_FLUSH_DATA: |
d3984c90 | 1235 | case BTRFS_RESERVE_FLUSH_ALL: |
7f9fe614 | 1236 | case BTRFS_RESERVE_FLUSH_ALL_STEAL: |
03235279 | 1237 | wait_reserve_ticket(fs_info, space_info, ticket); |
d3984c90 JB |
1238 | break; |
1239 | case BTRFS_RESERVE_FLUSH_LIMIT: | |
9ce2f423 JB |
1240 | priority_reclaim_metadata_space(fs_info, space_info, ticket, |
1241 | priority_flush_states, | |
1242 | ARRAY_SIZE(priority_flush_states)); | |
d3984c90 JB |
1243 | break; |
1244 | case BTRFS_RESERVE_FLUSH_EVICT: | |
1245 | priority_reclaim_metadata_space(fs_info, space_info, ticket, | |
1246 | evict_flush_states, | |
1247 | ARRAY_SIZE(evict_flush_states)); | |
1248 | break; | |
1004f686 | 1249 | case BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE: |
57056740 | 1250 | priority_reclaim_data_space(fs_info, space_info, ticket); |
1004f686 | 1251 | break; |
d3984c90 JB |
1252 | default: |
1253 | ASSERT(0); | |
1254 | break; | |
1255 | } | |
03235279 JB |
1256 | |
1257 | spin_lock(&space_info->lock); | |
1258 | ret = ticket->error; | |
1259 | if (ticket->bytes || ticket->error) { | |
0cab7acc | 1260 | /* |
42a72cb7 JB |
1261 | * We were a priority ticket, so we need to delete ourselves |
1262 | * from the list. Because we could have other priority tickets | |
1263 | * behind us that require less space, run | |
1264 | * btrfs_try_granting_tickets() to see if their reservations can | |
1265 | * now be made. | |
0cab7acc | 1266 | */ |
42a72cb7 JB |
1267 | if (!list_empty(&ticket->list)) { |
1268 | remove_ticket(space_info, ticket); | |
1269 | btrfs_try_granting_tickets(fs_info, space_info); | |
1270 | } | |
1271 | ||
03235279 JB |
1272 | if (!ret) |
1273 | ret = -ENOSPC; | |
1274 | } | |
1275 | spin_unlock(&space_info->lock); | |
03235279 | 1276 | ASSERT(list_empty(&ticket->list)); |
0cab7acc FM |
1277 | /* |
1278 | * Check that we can't have an error set if the reservation succeeded, | |
1279 | * as that would confuse tasks and lead them to error out without | |
1280 | * releasing reserved space (if an error happens the expectation is that | |
1281 | * space wasn't reserved at all). | |
1282 | */ | |
1283 | ASSERT(!(ticket->bytes == 0 && ticket->error)); | |
03235279 JB |
1284 | return ret; |
1285 | } | |
1286 | ||
666daa9f JB |
1287 | /* |
1288 | * This returns true if this flush state will go through the ordinary flushing | |
1289 | * code. | |
1290 | */ | |
1291 | static inline bool is_normal_flushing(enum btrfs_reserve_flush_enum flush) | |
1292 | { | |
1293 | return (flush == BTRFS_RESERVE_FLUSH_ALL) || | |
1294 | (flush == BTRFS_RESERVE_FLUSH_ALL_STEAL); | |
1295 | } | |
1296 | ||
0d9764f6 JB |
1297 | /** |
1298 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
1299 | * @root - the root we're allocating for | |
1300 | * @space_info - the space info we want to allocate from | |
1301 | * @orig_bytes - the number of bytes we want | |
1302 | * @flush - whether or not we can flush to make our reservation | |
1303 | * | |
1304 | * This will reserve orig_bytes number of bytes from the space info associated | |
1305 | * with the block_rsv. If there is not enough space it will make an attempt to | |
1306 | * flush out space to make room. It will do this by flushing delalloc if | |
1307 | * possible or committing the transaction. If flush is 0 then no attempts to | |
1308 | * regain reservations will be made and this will fail if there is not enough | |
1309 | * space already. | |
1310 | */ | |
f3bda421 JB |
1311 | static int __reserve_bytes(struct btrfs_fs_info *fs_info, |
1312 | struct btrfs_space_info *space_info, u64 orig_bytes, | |
1313 | enum btrfs_reserve_flush_enum flush) | |
0d9764f6 | 1314 | { |
57056740 | 1315 | struct work_struct *async_work; |
0d9764f6 JB |
1316 | struct reserve_ticket ticket; |
1317 | u64 used; | |
0d9764f6 | 1318 | int ret = 0; |
ef1317a1 | 1319 | bool pending_tickets; |
0d9764f6 JB |
1320 | |
1321 | ASSERT(orig_bytes); | |
1322 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); | |
1323 | ||
57056740 JB |
1324 | if (flush == BTRFS_RESERVE_FLUSH_DATA) |
1325 | async_work = &fs_info->async_data_reclaim_work; | |
1326 | else | |
1327 | async_work = &fs_info->async_reclaim_work; | |
1328 | ||
0d9764f6 JB |
1329 | spin_lock(&space_info->lock); |
1330 | ret = -ENOSPC; | |
1331 | used = btrfs_space_info_used(space_info, true); | |
666daa9f JB |
1332 | |
1333 | /* | |
1334 | * We don't want NO_FLUSH allocations to jump everybody, they can | |
1335 | * generally handle ENOSPC in a different way, so treat them the same as | |
1336 | * normal flushers when it comes to skipping pending tickets. | |
1337 | */ | |
1338 | if (is_normal_flushing(flush) || (flush == BTRFS_RESERVE_NO_FLUSH)) | |
1339 | pending_tickets = !list_empty(&space_info->tickets) || | |
1340 | !list_empty(&space_info->priority_tickets); | |
1341 | else | |
1342 | pending_tickets = !list_empty(&space_info->priority_tickets); | |
0d9764f6 JB |
1343 | |
1344 | /* | |
9b4851bc GR |
1345 | * Carry on if we have enough space (short-circuit) OR call |
1346 | * can_overcommit() to ensure we can overcommit to continue. | |
0d9764f6 | 1347 | */ |
ef1317a1 JB |
1348 | if (!pending_tickets && |
1349 | ((used + orig_bytes <= space_info->total_bytes) || | |
a30a3d20 | 1350 | btrfs_can_overcommit(fs_info, space_info, orig_bytes, flush))) { |
0d9764f6 JB |
1351 | btrfs_space_info_update_bytes_may_use(fs_info, space_info, |
1352 | orig_bytes); | |
0d9764f6 JB |
1353 | ret = 0; |
1354 | } | |
1355 | ||
1356 | /* | |
1357 | * If we couldn't make a reservation then setup our reservation ticket | |
1358 | * and kick the async worker if it's not already running. | |
1359 | * | |
1360 | * If we are a priority flusher then we just need to add our ticket to | |
1361 | * the list and we will do our own flushing further down. | |
1362 | */ | |
1363 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { | |
0d9764f6 JB |
1364 | ticket.bytes = orig_bytes; |
1365 | ticket.error = 0; | |
db161806 | 1366 | space_info->reclaim_size += ticket.bytes; |
0d9764f6 | 1367 | init_waitqueue_head(&ticket.wait); |
7f9fe614 JB |
1368 | ticket.steal = (flush == BTRFS_RESERVE_FLUSH_ALL_STEAL); |
1369 | if (flush == BTRFS_RESERVE_FLUSH_ALL || | |
57056740 JB |
1370 | flush == BTRFS_RESERVE_FLUSH_ALL_STEAL || |
1371 | flush == BTRFS_RESERVE_FLUSH_DATA) { | |
0d9764f6 JB |
1372 | list_add_tail(&ticket.list, &space_info->tickets); |
1373 | if (!space_info->flush) { | |
1374 | space_info->flush = 1; | |
1375 | trace_btrfs_trigger_flush(fs_info, | |
1376 | space_info->flags, | |
1377 | orig_bytes, flush, | |
1378 | "enospc"); | |
57056740 | 1379 | queue_work(system_unbound_wq, async_work); |
0d9764f6 JB |
1380 | } |
1381 | } else { | |
1382 | list_add_tail(&ticket.list, | |
1383 | &space_info->priority_tickets); | |
1384 | } | |
1385 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
1386 | used += orig_bytes; | |
1387 | /* | |
1388 | * We will do the space reservation dance during log replay, | |
1389 | * which means we won't have fs_info->fs_root set, so don't do | |
1390 | * the async reclaim as we will panic. | |
1391 | */ | |
1392 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && | |
9f246926 | 1393 | need_do_async_reclaim(fs_info, space_info, used) && |
0d9764f6 JB |
1394 | !work_busy(&fs_info->async_reclaim_work)) { |
1395 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
1396 | orig_bytes, flush, "preempt"); | |
1397 | queue_work(system_unbound_wq, | |
1398 | &fs_info->async_reclaim_work); | |
1399 | } | |
1400 | } | |
1401 | spin_unlock(&space_info->lock); | |
1402 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) | |
1403 | return ret; | |
1404 | ||
03235279 | 1405 | return handle_reserve_ticket(fs_info, space_info, &ticket, flush); |
0d9764f6 JB |
1406 | } |
1407 | ||
1408 | /** | |
1409 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
1410 | * @root - the root we're allocating for | |
1411 | * @block_rsv - the block_rsv we're allocating for | |
1412 | * @orig_bytes - the number of bytes we want | |
1413 | * @flush - whether or not we can flush to make our reservation | |
1414 | * | |
1415 | * This will reserve orig_bytes number of bytes from the space info associated | |
1416 | * with the block_rsv. If there is not enough space it will make an attempt to | |
1417 | * flush out space to make room. It will do this by flushing delalloc if | |
1418 | * possible or committing the transaction. If flush is 0 then no attempts to | |
1419 | * regain reservations will be made and this will fail if there is not enough | |
1420 | * space already. | |
1421 | */ | |
1422 | int btrfs_reserve_metadata_bytes(struct btrfs_root *root, | |
1423 | struct btrfs_block_rsv *block_rsv, | |
1424 | u64 orig_bytes, | |
1425 | enum btrfs_reserve_flush_enum flush) | |
1426 | { | |
1427 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1428 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
1429 | int ret; | |
0d9764f6 | 1430 | |
f3bda421 | 1431 | ret = __reserve_bytes(fs_info, block_rsv->space_info, orig_bytes, flush); |
0d9764f6 JB |
1432 | if (ret == -ENOSPC && |
1433 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
1434 | if (block_rsv != global_rsv && | |
1435 | !btrfs_block_rsv_use_bytes(global_rsv, orig_bytes)) | |
1436 | ret = 0; | |
1437 | } | |
1438 | if (ret == -ENOSPC) { | |
1439 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", | |
1440 | block_rsv->space_info->flags, | |
1441 | orig_bytes, 1); | |
1442 | ||
1443 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) | |
1444 | btrfs_dump_space_info(fs_info, block_rsv->space_info, | |
1445 | orig_bytes, 0); | |
1446 | } | |
1447 | return ret; | |
1448 | } | |
8698fc4e JB |
1449 | |
1450 | /** | |
1451 | * btrfs_reserve_data_bytes - try to reserve data bytes for an allocation | |
1452 | * @fs_info - the filesystem | |
1453 | * @bytes - the number of bytes we need | |
1454 | * @flush - how we are allowed to flush | |
1455 | * | |
1456 | * This will reserve bytes from the data space info. If there is not enough | |
1457 | * space then we will attempt to flush space as specified by flush. | |
1458 | */ | |
1459 | int btrfs_reserve_data_bytes(struct btrfs_fs_info *fs_info, u64 bytes, | |
1460 | enum btrfs_reserve_flush_enum flush) | |
1461 | { | |
1462 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; | |
f3bda421 | 1463 | int ret; |
8698fc4e | 1464 | |
f3bda421 JB |
1465 | ASSERT(flush == BTRFS_RESERVE_FLUSH_DATA || |
1466 | flush == BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE); | |
8698fc4e JB |
1467 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_DATA); |
1468 | ||
f3bda421 JB |
1469 | ret = __reserve_bytes(fs_info, data_sinfo, bytes, flush); |
1470 | if (ret == -ENOSPC) { | |
1471 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", | |
8698fc4e | 1472 | data_sinfo->flags, bytes, 1); |
f3bda421 JB |
1473 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) |
1474 | btrfs_dump_space_info(fs_info, data_sinfo, bytes, 0); | |
1475 | } | |
8698fc4e JB |
1476 | return ret; |
1477 | } |