1 // SPDX-License-Identifier: GPL-2.0
6 #include "space-info.h"
7 #include "transaction.h"
10 * HOW DO BLOCK RESERVES WORK
12 * Think of block_rsv's as buckets for logically grouped metadata
13 * reservations. Each block_rsv has a ->size and a ->reserved. ->size is
14 * how large we want our block rsv to be, ->reserved is how much space is
15 * currently reserved for this block reserve.
17 * ->failfast exists for the truncate case, and is described below.
22 * Entrance: btrfs_block_rsv_add, btrfs_block_rsv_refill
24 * We call into btrfs_reserve_metadata_bytes() with our bytes, which is
25 * accounted for in space_info->bytes_may_use, and then add the bytes to
26 * ->reserved, and ->size in the case of btrfs_block_rsv_add.
28 * ->size is an over-estimation of how much we may use for a particular
32 * Entrance: btrfs_use_block_rsv
34 * When we do a btrfs_alloc_tree_block() we call into btrfs_use_block_rsv()
35 * to determine the appropriate block_rsv to use, and then verify that
36 * ->reserved has enough space for our tree block allocation. Once
37 * successful we subtract fs_info->nodesize from ->reserved.
40 * Entrance: btrfs_block_rsv_release
42 * We are finished with our operation, subtract our individual reservation
43 * from ->size, and then subtract ->size from ->reserved and free up the
44 * excess if there is any.
46 * There is some logic here to refill the delayed refs rsv or the global rsv
47 * as needed, otherwise the excess is subtracted from
48 * space_info->bytes_may_use.
50 * TYPES OF BLOCK RESERVES
52 * BLOCK_RSV_TRANS, BLOCK_RSV_DELOPS, BLOCK_RSV_CHUNK
53 * These behave normally, as described above, just within the confines of the
54 * lifetime of their particular operation (transaction for the whole trans
55 * handle lifetime, for example).
58 * It is impossible to properly account for all the space that may be required
59 * to make our extent tree updates. This block reserve acts as an overflow
60 * buffer in case our delayed refs reserve does not reserve enough space to
61 * update the extent tree.
63 * We can steal from this in some cases as well, notably on evict() or
64 * truncate() in order to help users recover from ENOSPC conditions.
67 * The individual item sizes are determined by the per-inode size
68 * calculations, which are described with the delalloc code. This is pretty
69 * straightforward, it's just the calculation of ->size encodes a lot of
70 * different items, and thus it gets used when updating inodes, inserting file
71 * extents, and inserting checksums.
74 * We keep a running tally of how many delayed refs we have on the system.
75 * We assume each one of these delayed refs are going to use a full
76 * reservation. We use the transaction items and pre-reserve space for every
77 * operation, and use this reservation to refill any gap between ->size and
78 * ->reserved that may exist.
80 * From there it's straightforward, removing a delayed ref means we remove its
81 * count from ->size and free up reservations as necessary. Since this is
82 * the most dynamic block reserve in the system, we will try to refill this
83 * block reserve first with any excess returned by any other block reserve.
86 * This is the fallback block reserve to make us try to reserve space if we
87 * don't have a specific bucket for this allocation. It is mostly used for
88 * updating the device tree and such, since that is a separate pool we're
89 * content to just reserve space from the space_info on demand.
92 * This is used by things like truncate and iput. We will temporarily
93 * allocate a block reserve, set it to some size, and then truncate bytes
94 * until we have no space left. With ->failfast set we'll simply return
95 * ENOSPC from btrfs_use_block_rsv() to signal that we need to unwind and try
96 * to make a new reservation. This is because these operations are
97 * unbounded, so we want to do as much work as we can, and then back off and
101 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
102 struct btrfs_block_rsv *block_rsv,
103 struct btrfs_block_rsv *dest, u64 num_bytes,
104 u64 *qgroup_to_release_ret)
106 struct btrfs_space_info *space_info = block_rsv->space_info;
107 u64 qgroup_to_release = 0;
110 spin_lock(&block_rsv->lock);
111 if (num_bytes == (u64)-1) {
112 num_bytes = block_rsv->size;
113 qgroup_to_release = block_rsv->qgroup_rsv_size;
115 block_rsv->size -= num_bytes;
116 if (block_rsv->reserved >= block_rsv->size) {
117 num_bytes = block_rsv->reserved - block_rsv->size;
118 block_rsv->reserved = block_rsv->size;
123 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
124 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
125 block_rsv->qgroup_rsv_size;
126 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
128 qgroup_to_release = 0;
130 spin_unlock(&block_rsv->lock);
135 spin_lock(&dest->lock);
139 bytes_to_add = dest->size - dest->reserved;
140 bytes_to_add = min(num_bytes, bytes_to_add);
141 dest->reserved += bytes_to_add;
142 if (dest->reserved >= dest->size)
144 num_bytes -= bytes_to_add;
146 spin_unlock(&dest->lock);
149 btrfs_space_info_free_bytes_may_use(fs_info,
153 if (qgroup_to_release_ret)
154 *qgroup_to_release_ret = qgroup_to_release;
158 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
159 struct btrfs_block_rsv *dst, u64 num_bytes,
164 ret = btrfs_block_rsv_use_bytes(src, num_bytes);
168 btrfs_block_rsv_add_bytes(dst, num_bytes, update_size);
172 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
174 memset(rsv, 0, sizeof(*rsv));
175 spin_lock_init(&rsv->lock);
179 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
180 struct btrfs_block_rsv *rsv,
183 btrfs_init_block_rsv(rsv, type);
184 rsv->space_info = btrfs_find_space_info(fs_info,
185 BTRFS_BLOCK_GROUP_METADATA);
188 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
191 struct btrfs_block_rsv *block_rsv;
193 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
197 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
201 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
202 struct btrfs_block_rsv *rsv)
206 btrfs_block_rsv_release(fs_info, rsv, (u64)-1, NULL);
210 int btrfs_block_rsv_add(struct btrfs_root *root,
211 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
212 enum btrfs_reserve_flush_enum flush)
219 ret = btrfs_reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
221 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, true);
226 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
234 spin_lock(&block_rsv->lock);
235 num_bytes = div_factor(block_rsv->size, min_factor);
236 if (block_rsv->reserved >= num_bytes)
238 spin_unlock(&block_rsv->lock);
243 int btrfs_block_rsv_refill(struct btrfs_root *root,
244 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
245 enum btrfs_reserve_flush_enum flush)
253 spin_lock(&block_rsv->lock);
254 num_bytes = min_reserved;
255 if (block_rsv->reserved >= num_bytes)
258 num_bytes -= block_rsv->reserved;
259 spin_unlock(&block_rsv->lock);
264 ret = btrfs_reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
266 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, false);
273 u64 btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
274 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
275 u64 *qgroup_to_release)
277 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
278 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
279 struct btrfs_block_rsv *target = NULL;
282 * If we are the delayed_rsv then push to the global rsv, otherwise dump
283 * into the delayed rsv if it is not full.
285 if (block_rsv == delayed_rsv)
287 else if (block_rsv != global_rsv && !delayed_rsv->full)
288 target = delayed_rsv;
290 if (target && block_rsv->space_info != target->space_info)
293 return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
297 int btrfs_block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, u64 num_bytes)
301 spin_lock(&block_rsv->lock);
302 if (block_rsv->reserved >= num_bytes) {
303 block_rsv->reserved -= num_bytes;
304 if (block_rsv->reserved < block_rsv->size)
308 spin_unlock(&block_rsv->lock);
312 void btrfs_block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
313 u64 num_bytes, bool update_size)
315 spin_lock(&block_rsv->lock);
316 block_rsv->reserved += num_bytes;
318 block_rsv->size += num_bytes;
319 else if (block_rsv->reserved >= block_rsv->size)
321 spin_unlock(&block_rsv->lock);
324 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
325 struct btrfs_block_rsv *dest, u64 num_bytes,
328 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
331 if (global_rsv->space_info != dest->space_info)
334 spin_lock(&global_rsv->lock);
335 min_bytes = div_factor(global_rsv->size, min_factor);
336 if (global_rsv->reserved < min_bytes + num_bytes) {
337 spin_unlock(&global_rsv->lock);
340 global_rsv->reserved -= num_bytes;
341 if (global_rsv->reserved < global_rsv->size)
342 global_rsv->full = 0;
343 spin_unlock(&global_rsv->lock);
345 btrfs_block_rsv_add_bytes(dest, num_bytes, true);
349 void btrfs_update_global_block_rsv(struct btrfs_fs_info *fs_info)
351 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
352 struct btrfs_space_info *sinfo = block_rsv->space_info;
357 * The global block rsv is based on the size of the extent tree, the
358 * checksum tree and the root tree. If the fs is empty we want to set
359 * it to a minimal amount for safety.
361 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
362 btrfs_root_used(&fs_info->csum_root->root_item) +
363 btrfs_root_used(&fs_info->tree_root->root_item);
366 * We at a minimum are going to modify the csum root, the tree root, and
372 * But we also want to reserve enough space so we can do the fallback
373 * global reserve for an unlink, which is an additional 5 items (see the
374 * comment in __unlink_start_trans for what we're modifying.)
376 * But we also need space for the delayed ref updates from the unlink,
377 * so its 10, 5 for the actual operation, and 5 for the delayed ref
382 num_bytes = max_t(u64, num_bytes,
383 btrfs_calc_insert_metadata_size(fs_info, min_items));
385 spin_lock(&sinfo->lock);
386 spin_lock(&block_rsv->lock);
388 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
390 if (block_rsv->reserved < block_rsv->size) {
391 num_bytes = block_rsv->size - block_rsv->reserved;
392 btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
394 block_rsv->reserved = block_rsv->size;
395 } else if (block_rsv->reserved > block_rsv->size) {
396 num_bytes = block_rsv->reserved - block_rsv->size;
397 btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
399 block_rsv->reserved = block_rsv->size;
400 btrfs_try_granting_tickets(fs_info, sinfo);
403 if (block_rsv->reserved == block_rsv->size)
408 spin_unlock(&block_rsv->lock);
409 spin_unlock(&sinfo->lock);
412 void btrfs_init_global_block_rsv(struct btrfs_fs_info *fs_info)
414 struct btrfs_space_info *space_info;
416 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
417 fs_info->chunk_block_rsv.space_info = space_info;
419 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
420 fs_info->global_block_rsv.space_info = space_info;
421 fs_info->trans_block_rsv.space_info = space_info;
422 fs_info->empty_block_rsv.space_info = space_info;
423 fs_info->delayed_block_rsv.space_info = space_info;
424 fs_info->delayed_refs_rsv.space_info = space_info;
426 fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv;
427 fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv;
428 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
429 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
430 if (fs_info->quota_root)
431 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
432 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
434 btrfs_update_global_block_rsv(fs_info);
437 void btrfs_release_global_block_rsv(struct btrfs_fs_info *fs_info)
439 btrfs_block_rsv_release(fs_info, &fs_info->global_block_rsv, (u64)-1,
441 WARN_ON(fs_info->trans_block_rsv.size > 0);
442 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
443 WARN_ON(fs_info->chunk_block_rsv.size > 0);
444 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
445 WARN_ON(fs_info->delayed_block_rsv.size > 0);
446 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
447 WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
448 WARN_ON(fs_info->delayed_refs_rsv.size > 0);
451 static struct btrfs_block_rsv *get_block_rsv(
452 const struct btrfs_trans_handle *trans,
453 const struct btrfs_root *root)
455 struct btrfs_fs_info *fs_info = root->fs_info;
456 struct btrfs_block_rsv *block_rsv = NULL;
458 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
459 (root == fs_info->csum_root && trans->adding_csums) ||
460 (root == fs_info->uuid_root))
461 block_rsv = trans->block_rsv;
464 block_rsv = root->block_rsv;
467 block_rsv = &fs_info->empty_block_rsv;
472 struct btrfs_block_rsv *btrfs_use_block_rsv(struct btrfs_trans_handle *trans,
473 struct btrfs_root *root,
476 struct btrfs_fs_info *fs_info = root->fs_info;
477 struct btrfs_block_rsv *block_rsv;
478 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
480 bool global_updated = false;
482 block_rsv = get_block_rsv(trans, root);
484 if (unlikely(block_rsv->size == 0))
487 ret = btrfs_block_rsv_use_bytes(block_rsv, blocksize);
491 if (block_rsv->failfast)
494 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
495 global_updated = true;
496 btrfs_update_global_block_rsv(fs_info);
501 * The global reserve still exists to save us from ourselves, so don't
502 * warn_on if we are short on our delayed refs reserve.
504 if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
505 btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
506 static DEFINE_RATELIMIT_STATE(_rs,
507 DEFAULT_RATELIMIT_INTERVAL * 10,
508 /*DEFAULT_RATELIMIT_BURST*/ 1);
509 if (__ratelimit(&_rs))
511 "BTRFS: block rsv returned %d\n", ret);
514 ret = btrfs_reserve_metadata_bytes(root, block_rsv, blocksize,
515 BTRFS_RESERVE_NO_FLUSH);
519 * If we couldn't reserve metadata bytes try and use some from
520 * the global reserve if its space type is the same as the global
523 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
524 block_rsv->space_info == global_rsv->space_info) {
525 ret = btrfs_block_rsv_use_bytes(global_rsv, blocksize);