1 // SPDX-License-Identifier: GPL-2.0
2
3 #include "misc.h"
4 #include "ctree.h"
5 #include "block-rsv.h"
6 #include "space-info.h"
7 #include "transaction.h"
8 #include "block-group.h"
9 #include "fs.h"
10 #include "accessors.h"
11
12 /*
13 * HOW DO BLOCK RESERVES WORK
14 *
15 * Think of block_rsv's as buckets for logically grouped metadata
16 * reservations. Each block_rsv has a ->size and a ->reserved. ->size is
17 * how large we want our block rsv to be, ->reserved is how much space is
18 * currently reserved for this block reserve.
19 *
20 * ->failfast exists for the truncate case, and is described below.
21 *
22 * NORMAL OPERATION
23 *
24 * -> Reserve
25 * Entrance: btrfs_block_rsv_add, btrfs_block_rsv_refill
26 *
27 * We call into btrfs_reserve_metadata_bytes() with our bytes, which is
28 * accounted for in space_info->bytes_may_use, and then add the bytes to
29 * ->reserved, and ->size in the case of btrfs_block_rsv_add.
30 *
31 * ->size is an over-estimation of how much we may use for a particular
32 * operation.
33 *
34 * -> Use
35 * Entrance: btrfs_use_block_rsv
36 *
37 * When we do a btrfs_alloc_tree_block() we call into btrfs_use_block_rsv()
38 * to determine the appropriate block_rsv to use, and then verify that
39 * ->reserved has enough space for our tree block allocation. Once
40 * successful we subtract fs_info->nodesize from ->reserved.
41 *
42 * -> Finish
43 * Entrance: btrfs_block_rsv_release
44 *
45 * We are finished with our operation, subtract our individual reservation
46 * from ->size, and then subtract ->size from ->reserved and free up the
47 * excess if there is any.
48 *
49 * There is some logic here to refill the delayed refs rsv or the global rsv
50 * as needed, otherwise the excess is subtracted from
51 * space_info->bytes_may_use.
52 *
53 * TYPES OF BLOCK RESERVES
54 *
55 * BLOCK_RSV_TRANS, BLOCK_RSV_DELOPS, BLOCK_RSV_CHUNK
56 * These behave normally, as described above, just within the confines of the
57 * lifetime of their particular operation (transaction for the whole trans
58 * handle lifetime, for example).
59 *
60 * BLOCK_RSV_GLOBAL
61 * It is impossible to properly account for all the space that may be required
62 * to make our extent tree updates. This block reserve acts as an overflow
63 * buffer in case our delayed refs reserve does not reserve enough space to
64 * update the extent tree.
65 *
66 * We can steal from this in some cases as well, notably on evict() or
67 * truncate() in order to help users recover from ENOSPC conditions.
68 *
69 * BLOCK_RSV_DELALLOC
70 * The individual item sizes are determined by the per-inode size
71 * calculations, which are described with the delalloc code. This is pretty
72 * straightforward, it's just the calculation of ->size encodes a lot of
73 * different items, and thus it gets used when updating inodes, inserting file
74 * extents, and inserting checksums.
75 *
76 * BLOCK_RSV_DELREFS
77 * We keep a running tally of how many delayed refs we have on the system.
78 * We assume each one of these delayed refs are going to use a full
79 * reservation. We use the transaction items and pre-reserve space for every
80 * operation, and use this reservation to refill any gap between ->size and
81 * ->reserved that may exist.
82 *
83 * From there it's straightforward, removing a delayed ref means we remove its
84 * count from ->size and free up reservations as necessary. Since this is
85 * the most dynamic block reserve in the system, we will try to refill this
86 * block reserve first with any excess returned by any other block reserve.
87 *
88 * BLOCK_RSV_EMPTY
89 * This is the fallback block reserve to make us try to reserve space if we
90 * don't have a specific bucket for this allocation. It is mostly used for
91 * updating the device tree and such, since that is a separate pool we're
92 * content to just reserve space from the space_info on demand.
93 *
94 * BLOCK_RSV_TEMP
95 * This is used by things like truncate and iput. We will temporarily
96 * allocate a block reserve, set it to some size, and then truncate bytes
97 * until we have no space left. With ->failfast set we'll simply return
98 * ENOSPC from btrfs_use_block_rsv() to signal that we need to unwind and try
99 * to make a new reservation. This is because these operations are
100 * unbounded, so we want to do as much work as we can, and then back off and
101 * re-reserve.
102 */
103
104 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
105 struct btrfs_block_rsv *block_rsv,
106 struct btrfs_block_rsv *dest, u64 num_bytes,
107 u64 *qgroup_to_release_ret)
108 {
109 struct btrfs_space_info *space_info = block_rsv->space_info;
110 u64 qgroup_to_release = 0;
111 u64 ret;
112
113 spin_lock(&block_rsv->lock);
114 if (num_bytes == (u64)-1) {
115 num_bytes = block_rsv->size;
116 qgroup_to_release = block_rsv->qgroup_rsv_size;
117 }
118 block_rsv->size -= num_bytes;
119 if (block_rsv->reserved >= block_rsv->size) {
120 num_bytes = block_rsv->reserved - block_rsv->size;
121 block_rsv->reserved = block_rsv->size;
122 block_rsv->full = true;
123 } else {
124 num_bytes = 0;
125 }
126 if (qgroup_to_release_ret &&
127 block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
128 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
129 block_rsv->qgroup_rsv_size;
130 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
131 } else {
132 qgroup_to_release = 0;
133 }
134 spin_unlock(&block_rsv->lock);
135
136 ret = num_bytes;
137 if (num_bytes > 0) {
138 if (dest) {
139 spin_lock(&dest->lock);
140 if (!dest->full) {
141 u64 bytes_to_add;
142
143 bytes_to_add = dest->size - dest->reserved;
144 bytes_to_add = min(num_bytes, bytes_to_add);
145 dest->reserved += bytes_to_add;
146 if (dest->reserved >= dest->size)
147 dest->full = true;
148 num_bytes -= bytes_to_add;
149 }
150 spin_unlock(&dest->lock);
151 }
152 if (num_bytes)
153 btrfs_space_info_free_bytes_may_use(fs_info,
154 space_info,
155 num_bytes);
156 }
157 if (qgroup_to_release_ret)
158 *qgroup_to_release_ret = qgroup_to_release;
159 return ret;
160 }
161
162 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
163 struct btrfs_block_rsv *dst, u64 num_bytes,
164 bool update_size)
165 {
166 int ret;
167
168 ret = btrfs_block_rsv_use_bytes(src, num_bytes);
169 if (ret)
170 return ret;
171
172 btrfs_block_rsv_add_bytes(dst, num_bytes, update_size);
173 return 0;
174 }
175
176 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, enum btrfs_rsv_type type)
177 {
178 memset(rsv, 0, sizeof(*rsv));
179 spin_lock_init(&rsv->lock);
180 rsv->type = type;
181 }
182
183 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
184 struct btrfs_block_rsv *rsv,
185 enum btrfs_rsv_type type)
186 {
187 btrfs_init_block_rsv(rsv, type);
188 rsv->space_info = btrfs_find_space_info(fs_info,
189 BTRFS_BLOCK_GROUP_METADATA);
190 }
191
192 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
193 enum btrfs_rsv_type type)
194 {
195 struct btrfs_block_rsv *block_rsv;
196
197 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
198 if (!block_rsv)
199 return NULL;
200
201 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
202 return block_rsv;
203 }
204
205 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
206 struct btrfs_block_rsv *rsv)
207 {
208 if (!rsv)
209 return;
210 btrfs_block_rsv_release(fs_info, rsv, (u64)-1, NULL);
211 kfree(rsv);
212 }
213
214 int btrfs_block_rsv_add(struct btrfs_fs_info *fs_info,
215 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
216 enum btrfs_reserve_flush_enum flush)
217 {
218 int ret;
219
220 if (num_bytes == 0)
221 return 0;
222
223 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv->space_info,
224 num_bytes, flush);
225 if (!ret)
226 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, true);
227
228 return ret;
229 }
230
231 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_percent)
232 {
233 u64 num_bytes = 0;
234 int ret = -ENOSPC;
235
236 spin_lock(&block_rsv->lock);
237 num_bytes = mult_perc(block_rsv->size, min_percent);
238 if (block_rsv->reserved >= num_bytes)
239 ret = 0;
240 spin_unlock(&block_rsv->lock);
241
242 return ret;
243 }
244
245 int btrfs_block_rsv_refill(struct btrfs_fs_info *fs_info,
246 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
247 enum btrfs_reserve_flush_enum flush)
248 {
249 int ret = -ENOSPC;
250
251 if (!block_rsv)
252 return 0;
253
254 spin_lock(&block_rsv->lock);
255 if (block_rsv->reserved >= num_bytes)
256 ret = 0;
257 else
258 num_bytes -= block_rsv->reserved;
259 spin_unlock(&block_rsv->lock);
260
261 if (!ret)
262 return 0;
263
264 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv->space_info,
265 num_bytes, flush);
266 if (!ret) {
267 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, false);
268 return 0;
269 }
270
271 return ret;
272 }
273
274 u64 btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
275 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
276 u64 *qgroup_to_release)
277 {
278 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
279 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
280 struct btrfs_block_rsv *target = NULL;
281
282 /*
283 * If we are a delayed block reserve then push to the global rsv,
284 * otherwise dump into the global delayed reserve if it is not full.
285 */
286 if (block_rsv->type == BTRFS_BLOCK_RSV_DELOPS)
287 target = global_rsv;
288 else if (block_rsv != global_rsv && !btrfs_block_rsv_full(delayed_rsv))
289 target = delayed_rsv;
290
291 if (target && block_rsv->space_info != target->space_info)
292 target = NULL;
293
294 return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
295 qgroup_to_release);
296 }
297
298 int btrfs_block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, u64 num_bytes)
299 {
300 int ret = -ENOSPC;
301
302 spin_lock(&block_rsv->lock);
303 if (block_rsv->reserved >= num_bytes) {
304 block_rsv->reserved -= num_bytes;
305 if (block_rsv->reserved < block_rsv->size)
306 block_rsv->full = false;
307 ret = 0;
308 }
309 spin_unlock(&block_rsv->lock);
310 return ret;
311 }
312
313 void btrfs_block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
314 u64 num_bytes, bool update_size)
315 {
316 spin_lock(&block_rsv->lock);
317 block_rsv->reserved += num_bytes;
318 if (update_size)
319 block_rsv->size += num_bytes;
320 else if (block_rsv->reserved >= block_rsv->size)
321 block_rsv->full = true;
322 spin_unlock(&block_rsv->lock);
323 }
324
325 void btrfs_update_global_block_rsv(struct btrfs_fs_info *fs_info)
326 {
327 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
328 struct btrfs_space_info *sinfo = block_rsv->space_info;
329 struct btrfs_root *root, *tmp;
330 u64 num_bytes = btrfs_root_used(&fs_info->tree_root->root_item);
331 unsigned int min_items = 1;
332
333 /*
334 * The global block rsv is based on the size of the extent tree, the
335 * checksum tree and the root tree. If the fs is empty we want to set
336 * it to a minimal amount for safety.
337 *
338 * We also are going to need to modify the minimum of the tree root and
339 * any global roots we could touch.
340 */
341 read_lock(&fs_info->global_root_lock);
342 rbtree_postorder_for_each_entry_safe(root, tmp, &fs_info->global_root_tree,
343 rb_node) {
344 if (btrfs_root_id(root) == BTRFS_EXTENT_TREE_OBJECTID ||
345 btrfs_root_id(root) == BTRFS_CSUM_TREE_OBJECTID ||
346 btrfs_root_id(root) == BTRFS_FREE_SPACE_TREE_OBJECTID) {
347 num_bytes += btrfs_root_used(&root->root_item);
348 min_items++;
349 }
350 }
351 read_unlock(&fs_info->global_root_lock);
352
353 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
354 num_bytes += btrfs_root_used(&fs_info->block_group_root->root_item);
355 min_items++;
356 }
357
358 if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
359 num_bytes += btrfs_root_used(&fs_info->stripe_root->root_item);
360 min_items++;
361 }
362
363 /*
364 * But we also want to reserve enough space so we can do the fallback
365 * global reserve for an unlink, which is an additional
366 * BTRFS_UNLINK_METADATA_UNITS items.
367 *
368 * But we also need space for the delayed ref updates from the unlink,
369 * so add BTRFS_UNLINK_METADATA_UNITS units for delayed refs, one for
370 * each unlink metadata item.
371 */
372 min_items += BTRFS_UNLINK_METADATA_UNITS;
373
374 num_bytes = max_t(u64, num_bytes,
375 btrfs_calc_insert_metadata_size(fs_info, min_items) +
376 btrfs_calc_delayed_ref_bytes(fs_info,
377 BTRFS_UNLINK_METADATA_UNITS));
378
379 spin_lock(&sinfo->lock);
380 spin_lock(&block_rsv->lock);
381
382 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
383
384 if (block_rsv->reserved < block_rsv->size) {
385 num_bytes = block_rsv->size - block_rsv->reserved;
386 btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
387 num_bytes);
388 block_rsv->reserved = block_rsv->size;
389 } else if (block_rsv->reserved > block_rsv->size) {
390 num_bytes = block_rsv->reserved - block_rsv->size;
391 btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
392 -num_bytes);
393 block_rsv->reserved = block_rsv->size;
394 btrfs_try_granting_tickets(fs_info, sinfo);
395 }
396
397 block_rsv->full = (block_rsv->reserved == block_rsv->size);
398
399 if (block_rsv->size >= sinfo->total_bytes)
400 sinfo->force_alloc = CHUNK_ALLOC_FORCE;
401 spin_unlock(&block_rsv->lock);
402 spin_unlock(&sinfo->lock);
403 }
404
405 void btrfs_init_root_block_rsv(struct btrfs_root *root)
406 {
407 struct btrfs_fs_info *fs_info = root->fs_info;
408
409 switch (btrfs_root_id(root)) {
410 case BTRFS_CSUM_TREE_OBJECTID:
411 case BTRFS_EXTENT_TREE_OBJECTID:
412 case BTRFS_FREE_SPACE_TREE_OBJECTID:
413 case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
414 case BTRFS_RAID_STRIPE_TREE_OBJECTID:
415 root->block_rsv = &fs_info->delayed_refs_rsv;
416 break;
417 case BTRFS_ROOT_TREE_OBJECTID:
418 case BTRFS_DEV_TREE_OBJECTID:
419 case BTRFS_QUOTA_TREE_OBJECTID:
420 root->block_rsv = &fs_info->global_block_rsv;
421 break;
422 case BTRFS_CHUNK_TREE_OBJECTID:
423 root->block_rsv = &fs_info->chunk_block_rsv;
424 break;
425 default:
426 root->block_rsv = NULL;
427 break;
428 }
429 }
430
431 void btrfs_init_global_block_rsv(struct btrfs_fs_info *fs_info)
432 {
433 struct btrfs_space_info *space_info;
434
435 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
436 fs_info->chunk_block_rsv.space_info = space_info;
437
438 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
439 fs_info->global_block_rsv.space_info = space_info;
440 fs_info->trans_block_rsv.space_info = space_info;
441 fs_info->empty_block_rsv.space_info = space_info;
442 fs_info->delayed_block_rsv.space_info = space_info;
443 fs_info->delayed_refs_rsv.space_info = space_info;
444
445 btrfs_update_global_block_rsv(fs_info);
446 }
447
448 void btrfs_release_global_block_rsv(struct btrfs_fs_info *fs_info)
449 {
450 btrfs_block_rsv_release(fs_info, &fs_info->global_block_rsv, (u64)-1,
451 NULL);
452 WARN_ON(fs_info->trans_block_rsv.size > 0);
453 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
454 WARN_ON(fs_info->chunk_block_rsv.size > 0);
455 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
456 WARN_ON(fs_info->delayed_block_rsv.size > 0);
457 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
458 WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
459 WARN_ON(fs_info->delayed_refs_rsv.size > 0);
460 }
461
462 static struct btrfs_block_rsv *get_block_rsv(
463 const struct btrfs_trans_handle *trans,
464 const struct btrfs_root *root)
465 {
466 struct btrfs_fs_info *fs_info = root->fs_info;
467 struct btrfs_block_rsv *block_rsv = NULL;
468
469 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
470 (root == fs_info->uuid_root) ||
471 (trans->adding_csums && btrfs_root_id(root) == BTRFS_CSUM_TREE_OBJECTID))
472 block_rsv = trans->block_rsv;
473
474 if (!block_rsv)
475 block_rsv = root->block_rsv;
476
477 if (!block_rsv)
478 block_rsv = &fs_info->empty_block_rsv;
479
480 return block_rsv;
481 }
482
483 struct btrfs_block_rsv *btrfs_use_block_rsv(struct btrfs_trans_handle *trans,
484 struct btrfs_root *root,
485 u32 blocksize)
486 {
487 struct btrfs_fs_info *fs_info = root->fs_info;
488 struct btrfs_block_rsv *block_rsv;
489 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
490 int ret;
491 bool global_updated = false;
492
493 block_rsv = get_block_rsv(trans, root);
494
495 if (unlikely(btrfs_block_rsv_size(block_rsv) == 0))
496 goto try_reserve;
497 again:
498 ret = btrfs_block_rsv_use_bytes(block_rsv, blocksize);
499 if (!ret)
500 return block_rsv;
501
502 if (block_rsv->failfast)
503 return ERR_PTR(ret);
504
505 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
506 global_updated = true;
507 btrfs_update_global_block_rsv(fs_info);
508 goto again;
509 }
510
511 /*
512 * The global reserve still exists to save us from ourselves, so don't
513 * warn_on if we are short on our delayed refs reserve.
514 */
515 if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
516 btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
517 static DEFINE_RATELIMIT_STATE(_rs,
518 DEFAULT_RATELIMIT_INTERVAL * 10,
519 /*DEFAULT_RATELIMIT_BURST*/ 1);
520 if (__ratelimit(&_rs))
521 WARN(1, KERN_DEBUG
522 "BTRFS: block rsv %d returned %d\n",
523 block_rsv->type, ret);
524 }
525 try_reserve:
526 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv->space_info,
527 blocksize, BTRFS_RESERVE_NO_FLUSH);
528 if (!ret)
529 return block_rsv;
530 /*
531 * If we couldn't reserve metadata bytes try and use some from
532 * the global reserve if its space type is the same as the global
533 * reservation.
534 */
535 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
536 block_rsv->space_info == global_rsv->space_info) {
537 ret = btrfs_block_rsv_use_bytes(global_rsv, blocksize);
538 if (!ret)
539 return global_rsv;
540 }
541
542 /*
543 * All hope is lost, but of course our reservations are overly
544 * pessimistic, so instead of possibly having an ENOSPC abort here, try
545 * one last time to force a reservation if there's enough actual space
546 * on disk to make the reservation.
547 */
548 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv->space_info, blocksize,
549 BTRFS_RESERVE_FLUSH_EMERGENCY);
550 if (!ret)
551 return block_rsv;
552
553 return ERR_PTR(ret);
554 }
555
556 int btrfs_check_trunc_cache_free_space(struct btrfs_fs_info *fs_info,
557 struct btrfs_block_rsv *rsv)
558 {
559 u64 needed_bytes;
560 int ret;
561
562 /* 1 for slack space, 1 for updating the inode */
563 needed_bytes = btrfs_calc_insert_metadata_size(fs_info, 1) +
564 btrfs_calc_metadata_size(fs_info, 1);
565
566 spin_lock(&rsv->lock);
567 if (rsv->reserved < needed_bytes)
568 ret = -ENOSPC;
569 else
570 ret = 0;
571 spin_unlock(&rsv->lock);
572 return ret;
573 }
574
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.