1 // SPDX-License-Identifier: GPL-2.0
6 #include "space-info.h"
7 #include "transaction.h"
8 #include "block-group.h"
11 #include "accessors.h"
14 * HOW DO BLOCK RESERVES WORK
16 * Think of block_rsv's as buckets for logically grouped metadata
17 * reservations. Each block_rsv has a ->size and a ->reserved. ->size is
18 * how large we want our block rsv to be, ->reserved is how much space is
19 * currently reserved for this block reserve.
21 * ->failfast exists for the truncate case, and is described below.
26 * Entrance: btrfs_block_rsv_add, btrfs_block_rsv_refill
28 * We call into btrfs_reserve_metadata_bytes() with our bytes, which is
29 * accounted for in space_info->bytes_may_use, and then add the bytes to
30 * ->reserved, and ->size in the case of btrfs_block_rsv_add.
32 * ->size is an over-estimation of how much we may use for a particular
36 * Entrance: btrfs_use_block_rsv
38 * When we do a btrfs_alloc_tree_block() we call into btrfs_use_block_rsv()
39 * to determine the appropriate block_rsv to use, and then verify that
40 * ->reserved has enough space for our tree block allocation. Once
41 * successful we subtract fs_info->nodesize from ->reserved.
44 * Entrance: btrfs_block_rsv_release
46 * We are finished with our operation, subtract our individual reservation
47 * from ->size, and then subtract ->size from ->reserved and free up the
48 * excess if there is any.
50 * There is some logic here to refill the delayed refs rsv or the global rsv
51 * as needed, otherwise the excess is subtracted from
52 * space_info->bytes_may_use.
54 * TYPES OF BLOCK RESERVES
56 * BLOCK_RSV_TRANS, BLOCK_RSV_DELOPS, BLOCK_RSV_CHUNK
57 * These behave normally, as described above, just within the confines of the
58 * lifetime of their particular operation (transaction for the whole trans
59 * handle lifetime, for example).
62 * It is impossible to properly account for all the space that may be required
63 * to make our extent tree updates. This block reserve acts as an overflow
64 * buffer in case our delayed refs reserve does not reserve enough space to
65 * update the extent tree.
67 * We can steal from this in some cases as well, notably on evict() or
68 * truncate() in order to help users recover from ENOSPC conditions.
71 * The individual item sizes are determined by the per-inode size
72 * calculations, which are described with the delalloc code. This is pretty
73 * straightforward, it's just the calculation of ->size encodes a lot of
74 * different items, and thus it gets used when updating inodes, inserting file
75 * extents, and inserting checksums.
78 * We keep a running tally of how many delayed refs we have on the system.
79 * We assume each one of these delayed refs are going to use a full
80 * reservation. We use the transaction items and pre-reserve space for every
81 * operation, and use this reservation to refill any gap between ->size and
82 * ->reserved that may exist.
84 * From there it's straightforward, removing a delayed ref means we remove its
85 * count from ->size and free up reservations as necessary. Since this is
86 * the most dynamic block reserve in the system, we will try to refill this
87 * block reserve first with any excess returned by any other block reserve.
90 * This is the fallback block reserve to make us try to reserve space if we
91 * don't have a specific bucket for this allocation. It is mostly used for
92 * updating the device tree and such, since that is a separate pool we're
93 * content to just reserve space from the space_info on demand.
96 * This is used by things like truncate and iput. We will temporarily
97 * allocate a block reserve, set it to some size, and then truncate bytes
98 * until we have no space left. With ->failfast set we'll simply return
99 * ENOSPC from btrfs_use_block_rsv() to signal that we need to unwind and try
100 * to make a new reservation. This is because these operations are
101 * unbounded, so we want to do as much work as we can, and then back off and
105 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
106 struct btrfs_block_rsv *block_rsv,
107 struct btrfs_block_rsv *dest, u64 num_bytes,
108 u64 *qgroup_to_release_ret)
110 struct btrfs_space_info *space_info = block_rsv->space_info;
111 u64 qgroup_to_release = 0;
114 spin_lock(&block_rsv->lock);
115 if (num_bytes == (u64)-1) {
116 num_bytes = block_rsv->size;
117 qgroup_to_release = block_rsv->qgroup_rsv_size;
119 block_rsv->size -= num_bytes;
120 if (block_rsv->reserved >= block_rsv->size) {
121 num_bytes = block_rsv->reserved - block_rsv->size;
122 block_rsv->reserved = block_rsv->size;
123 block_rsv->full = true;
127 if (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;
132 qgroup_to_release = 0;
134 spin_unlock(&block_rsv->lock);
139 spin_lock(&dest->lock);
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)
148 num_bytes -= bytes_to_add;
150 spin_unlock(&dest->lock);
153 btrfs_space_info_free_bytes_may_use(fs_info,
157 if (qgroup_to_release_ret)
158 *qgroup_to_release_ret = qgroup_to_release;
162 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
163 struct btrfs_block_rsv *dst, u64 num_bytes,
168 ret = btrfs_block_rsv_use_bytes(src, num_bytes);
172 btrfs_block_rsv_add_bytes(dst, num_bytes, update_size);
176 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, enum btrfs_rsv_type type)
178 memset(rsv, 0, sizeof(*rsv));
179 spin_lock_init(&rsv->lock);
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)
187 btrfs_init_block_rsv(rsv, type);
188 rsv->space_info = btrfs_find_space_info(fs_info,
189 BTRFS_BLOCK_GROUP_METADATA);
192 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
193 enum btrfs_rsv_type type)
195 struct btrfs_block_rsv *block_rsv;
197 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
201 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
205 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
206 struct btrfs_block_rsv *rsv)
210 btrfs_block_rsv_release(fs_info, rsv, (u64)-1, NULL);
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)
223 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv, num_bytes, flush);
225 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, true);
230 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_percent)
238 spin_lock(&block_rsv->lock);
239 num_bytes = mult_perc(block_rsv->size, min_percent);
240 if (block_rsv->reserved >= num_bytes)
242 spin_unlock(&block_rsv->lock);
247 int btrfs_block_rsv_refill(struct btrfs_fs_info *fs_info,
248 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
249 enum btrfs_reserve_flush_enum flush)
257 spin_lock(&block_rsv->lock);
258 num_bytes = min_reserved;
259 if (block_rsv->reserved >= num_bytes)
262 num_bytes -= block_rsv->reserved;
263 spin_unlock(&block_rsv->lock);
268 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv, num_bytes, flush);
270 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, false);
277 u64 btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
278 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
279 u64 *qgroup_to_release)
281 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
282 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
283 struct btrfs_block_rsv *target = NULL;
286 * If we are the delayed_rsv then push to the global rsv, otherwise dump
287 * into the delayed rsv if it is not full.
289 if (block_rsv == delayed_rsv)
291 else if (block_rsv != global_rsv && !btrfs_block_rsv_full(delayed_rsv))
292 target = delayed_rsv;
294 if (target && block_rsv->space_info != target->space_info)
297 return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
301 int btrfs_block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, u64 num_bytes)
305 spin_lock(&block_rsv->lock);
306 if (block_rsv->reserved >= num_bytes) {
307 block_rsv->reserved -= num_bytes;
308 if (block_rsv->reserved < block_rsv->size)
309 block_rsv->full = false;
312 spin_unlock(&block_rsv->lock);
316 void btrfs_block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
317 u64 num_bytes, bool update_size)
319 spin_lock(&block_rsv->lock);
320 block_rsv->reserved += num_bytes;
322 block_rsv->size += num_bytes;
323 else if (block_rsv->reserved >= block_rsv->size)
324 block_rsv->full = true;
325 spin_unlock(&block_rsv->lock);
328 void btrfs_update_global_block_rsv(struct btrfs_fs_info *fs_info)
330 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
331 struct btrfs_space_info *sinfo = block_rsv->space_info;
332 struct btrfs_root *root, *tmp;
333 u64 num_bytes = btrfs_root_used(&fs_info->tree_root->root_item);
334 unsigned int min_items = 1;
337 * The global block rsv is based on the size of the extent tree, the
338 * checksum tree and the root tree. If the fs is empty we want to set
339 * it to a minimal amount for safety.
341 * We also are going to need to modify the minimum of the tree root and
342 * any global roots we could touch.
344 read_lock(&fs_info->global_root_lock);
345 rbtree_postorder_for_each_entry_safe(root, tmp, &fs_info->global_root_tree,
347 if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
348 root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID ||
349 root->root_key.objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) {
350 num_bytes += btrfs_root_used(&root->root_item);
354 read_unlock(&fs_info->global_root_lock);
357 * But we also want to reserve enough space so we can do the fallback
358 * global reserve for an unlink, which is an additional 5 items (see the
359 * comment in __unlink_start_trans for what we're modifying.)
361 * But we also need space for the delayed ref updates from the unlink,
362 * so its 10, 5 for the actual operation, and 5 for the delayed ref
367 num_bytes = max_t(u64, num_bytes,
368 btrfs_calc_insert_metadata_size(fs_info, min_items));
370 spin_lock(&sinfo->lock);
371 spin_lock(&block_rsv->lock);
373 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
375 if (block_rsv->reserved < block_rsv->size) {
376 num_bytes = block_rsv->size - block_rsv->reserved;
377 btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
379 block_rsv->reserved = block_rsv->size;
380 } else if (block_rsv->reserved > block_rsv->size) {
381 num_bytes = block_rsv->reserved - block_rsv->size;
382 btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
384 block_rsv->reserved = block_rsv->size;
385 btrfs_try_granting_tickets(fs_info, sinfo);
388 block_rsv->full = (block_rsv->reserved == block_rsv->size);
390 if (block_rsv->size >= sinfo->total_bytes)
391 sinfo->force_alloc = CHUNK_ALLOC_FORCE;
392 spin_unlock(&block_rsv->lock);
393 spin_unlock(&sinfo->lock);
396 void btrfs_init_root_block_rsv(struct btrfs_root *root)
398 struct btrfs_fs_info *fs_info = root->fs_info;
400 switch (root->root_key.objectid) {
401 case BTRFS_CSUM_TREE_OBJECTID:
402 case BTRFS_EXTENT_TREE_OBJECTID:
403 case BTRFS_FREE_SPACE_TREE_OBJECTID:
404 case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
405 root->block_rsv = &fs_info->delayed_refs_rsv;
407 case BTRFS_ROOT_TREE_OBJECTID:
408 case BTRFS_DEV_TREE_OBJECTID:
409 case BTRFS_QUOTA_TREE_OBJECTID:
410 root->block_rsv = &fs_info->global_block_rsv;
412 case BTRFS_CHUNK_TREE_OBJECTID:
413 root->block_rsv = &fs_info->chunk_block_rsv;
416 root->block_rsv = NULL;
421 void btrfs_init_global_block_rsv(struct btrfs_fs_info *fs_info)
423 struct btrfs_space_info *space_info;
425 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
426 fs_info->chunk_block_rsv.space_info = space_info;
428 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
429 fs_info->global_block_rsv.space_info = space_info;
430 fs_info->trans_block_rsv.space_info = space_info;
431 fs_info->empty_block_rsv.space_info = space_info;
432 fs_info->delayed_block_rsv.space_info = space_info;
433 fs_info->delayed_refs_rsv.space_info = space_info;
435 btrfs_update_global_block_rsv(fs_info);
438 void btrfs_release_global_block_rsv(struct btrfs_fs_info *fs_info)
440 btrfs_block_rsv_release(fs_info, &fs_info->global_block_rsv, (u64)-1,
442 WARN_ON(fs_info->trans_block_rsv.size > 0);
443 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
444 WARN_ON(fs_info->chunk_block_rsv.size > 0);
445 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
446 WARN_ON(fs_info->delayed_block_rsv.size > 0);
447 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
448 WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
449 WARN_ON(fs_info->delayed_refs_rsv.size > 0);
452 static struct btrfs_block_rsv *get_block_rsv(
453 const struct btrfs_trans_handle *trans,
454 const struct btrfs_root *root)
456 struct btrfs_fs_info *fs_info = root->fs_info;
457 struct btrfs_block_rsv *block_rsv = NULL;
459 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
460 (root == fs_info->uuid_root) ||
461 (trans->adding_csums &&
462 root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID))
463 block_rsv = trans->block_rsv;
466 block_rsv = root->block_rsv;
469 block_rsv = &fs_info->empty_block_rsv;
474 struct btrfs_block_rsv *btrfs_use_block_rsv(struct btrfs_trans_handle *trans,
475 struct btrfs_root *root,
478 struct btrfs_fs_info *fs_info = root->fs_info;
479 struct btrfs_block_rsv *block_rsv;
480 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
482 bool global_updated = false;
484 block_rsv = get_block_rsv(trans, root);
486 if (unlikely(block_rsv->size == 0))
489 ret = btrfs_block_rsv_use_bytes(block_rsv, blocksize);
493 if (block_rsv->failfast)
496 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
497 global_updated = true;
498 btrfs_update_global_block_rsv(fs_info);
503 * The global reserve still exists to save us from ourselves, so don't
504 * warn_on if we are short on our delayed refs reserve.
506 if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
507 btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
508 static DEFINE_RATELIMIT_STATE(_rs,
509 DEFAULT_RATELIMIT_INTERVAL * 10,
510 /*DEFAULT_RATELIMIT_BURST*/ 1);
511 if (__ratelimit(&_rs))
513 "BTRFS: block rsv %d returned %d\n",
514 block_rsv->type, ret);
517 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv, blocksize,
518 BTRFS_RESERVE_NO_FLUSH);
522 * If we couldn't reserve metadata bytes try and use some from
523 * the global reserve if its space type is the same as the global
526 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
527 block_rsv->space_info == global_rsv->space_info) {
528 ret = btrfs_block_rsv_use_bytes(global_rsv, blocksize);
534 * All hope is lost, but of course our reservations are overly
535 * pessimistic, so instead of possibly having an ENOSPC abort here, try
536 * one last time to force a reservation if there's enough actual space
537 * on disk to make the reservation.
539 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv, blocksize,
540 BTRFS_RESERVE_FLUSH_EMERGENCY);