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 (qgroup_to_release_ret &&
128 block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
129 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
130 block_rsv->qgroup_rsv_size;
131 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
133 qgroup_to_release = 0;
135 spin_unlock(&block_rsv->lock);
140 spin_lock(&dest->lock);
144 bytes_to_add = dest->size - dest->reserved;
145 bytes_to_add = min(num_bytes, bytes_to_add);
146 dest->reserved += bytes_to_add;
147 if (dest->reserved >= dest->size)
149 num_bytes -= bytes_to_add;
151 spin_unlock(&dest->lock);
154 btrfs_space_info_free_bytes_may_use(fs_info,
158 if (qgroup_to_release_ret)
159 *qgroup_to_release_ret = qgroup_to_release;
163 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
164 struct btrfs_block_rsv *dst, u64 num_bytes,
169 ret = btrfs_block_rsv_use_bytes(src, num_bytes);
173 btrfs_block_rsv_add_bytes(dst, num_bytes, update_size);
177 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, enum btrfs_rsv_type type)
179 memset(rsv, 0, sizeof(*rsv));
180 spin_lock_init(&rsv->lock);
184 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
185 struct btrfs_block_rsv *rsv,
186 enum btrfs_rsv_type type)
188 btrfs_init_block_rsv(rsv, type);
189 rsv->space_info = btrfs_find_space_info(fs_info,
190 BTRFS_BLOCK_GROUP_METADATA);
193 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
194 enum btrfs_rsv_type type)
196 struct btrfs_block_rsv *block_rsv;
198 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
202 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
206 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
207 struct btrfs_block_rsv *rsv)
211 btrfs_block_rsv_release(fs_info, rsv, (u64)-1, NULL);
215 int btrfs_block_rsv_add(struct btrfs_fs_info *fs_info,
216 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
217 enum btrfs_reserve_flush_enum flush)
224 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv, num_bytes, flush);
226 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, true);
231 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_percent)
236 spin_lock(&block_rsv->lock);
237 num_bytes = mult_perc(block_rsv->size, min_percent);
238 if (block_rsv->reserved >= num_bytes)
240 spin_unlock(&block_rsv->lock);
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)
254 spin_lock(&block_rsv->lock);
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(fs_info, 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 && !btrfs_block_rsv_full(delayed_rsv))
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)
305 block_rsv->full = false;
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)
320 block_rsv->full = true;
321 spin_unlock(&block_rsv->lock);
324 void btrfs_update_global_block_rsv(struct btrfs_fs_info *fs_info)
326 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
327 struct btrfs_space_info *sinfo = block_rsv->space_info;
328 struct btrfs_root *root, *tmp;
329 u64 num_bytes = btrfs_root_used(&fs_info->tree_root->root_item);
330 unsigned int min_items = 1;
333 * The global block rsv is based on the size of the extent tree, the
334 * checksum tree and the root tree. If the fs is empty we want to set
335 * it to a minimal amount for safety.
337 * We also are going to need to modify the minimum of the tree root and
338 * any global roots we could touch.
340 read_lock(&fs_info->global_root_lock);
341 rbtree_postorder_for_each_entry_safe(root, tmp, &fs_info->global_root_tree,
343 if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
344 root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID ||
345 root->root_key.objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) {
346 num_bytes += btrfs_root_used(&root->root_item);
350 read_unlock(&fs_info->global_root_lock);
352 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
353 num_bytes += btrfs_root_used(&fs_info->block_group_root->root_item);
358 * But we also want to reserve enough space so we can do the fallback
359 * global reserve for an unlink, which is an additional
360 * BTRFS_UNLINK_METADATA_UNITS items.
362 * But we also need space for the delayed ref updates from the unlink,
363 * so add BTRFS_UNLINK_METADATA_UNITS units for delayed refs, one for
364 * each unlink metadata item.
366 min_items += BTRFS_UNLINK_METADATA_UNITS;
368 num_bytes = max_t(u64, num_bytes,
369 btrfs_calc_insert_metadata_size(fs_info, min_items) +
370 btrfs_calc_delayed_ref_bytes(fs_info,
371 BTRFS_UNLINK_METADATA_UNITS));
373 spin_lock(&sinfo->lock);
374 spin_lock(&block_rsv->lock);
376 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
378 if (block_rsv->reserved < block_rsv->size) {
379 num_bytes = block_rsv->size - block_rsv->reserved;
380 btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
382 block_rsv->reserved = block_rsv->size;
383 } else if (block_rsv->reserved > block_rsv->size) {
384 num_bytes = block_rsv->reserved - block_rsv->size;
385 btrfs_space_info_update_bytes_may_use(fs_info, sinfo,
387 block_rsv->reserved = block_rsv->size;
388 btrfs_try_granting_tickets(fs_info, sinfo);
391 block_rsv->full = (block_rsv->reserved == block_rsv->size);
393 if (block_rsv->size >= sinfo->total_bytes)
394 sinfo->force_alloc = CHUNK_ALLOC_FORCE;
395 spin_unlock(&block_rsv->lock);
396 spin_unlock(&sinfo->lock);
399 void btrfs_init_root_block_rsv(struct btrfs_root *root)
401 struct btrfs_fs_info *fs_info = root->fs_info;
403 switch (root->root_key.objectid) {
404 case BTRFS_CSUM_TREE_OBJECTID:
405 case BTRFS_EXTENT_TREE_OBJECTID:
406 case BTRFS_FREE_SPACE_TREE_OBJECTID:
407 case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
408 root->block_rsv = &fs_info->delayed_refs_rsv;
410 case BTRFS_ROOT_TREE_OBJECTID:
411 case BTRFS_DEV_TREE_OBJECTID:
412 case BTRFS_QUOTA_TREE_OBJECTID:
413 root->block_rsv = &fs_info->global_block_rsv;
415 case BTRFS_CHUNK_TREE_OBJECTID:
416 root->block_rsv = &fs_info->chunk_block_rsv;
419 root->block_rsv = NULL;
424 void btrfs_init_global_block_rsv(struct btrfs_fs_info *fs_info)
426 struct btrfs_space_info *space_info;
428 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
429 fs_info->chunk_block_rsv.space_info = space_info;
431 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
432 fs_info->global_block_rsv.space_info = space_info;
433 fs_info->trans_block_rsv.space_info = space_info;
434 fs_info->empty_block_rsv.space_info = space_info;
435 fs_info->delayed_block_rsv.space_info = space_info;
436 fs_info->delayed_refs_rsv.space_info = space_info;
438 btrfs_update_global_block_rsv(fs_info);
441 void btrfs_release_global_block_rsv(struct btrfs_fs_info *fs_info)
443 btrfs_block_rsv_release(fs_info, &fs_info->global_block_rsv, (u64)-1,
445 WARN_ON(fs_info->trans_block_rsv.size > 0);
446 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
447 WARN_ON(fs_info->chunk_block_rsv.size > 0);
448 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
449 WARN_ON(fs_info->delayed_block_rsv.size > 0);
450 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
451 WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
452 WARN_ON(fs_info->delayed_refs_rsv.size > 0);
455 static struct btrfs_block_rsv *get_block_rsv(
456 const struct btrfs_trans_handle *trans,
457 const struct btrfs_root *root)
459 struct btrfs_fs_info *fs_info = root->fs_info;
460 struct btrfs_block_rsv *block_rsv = NULL;
462 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
463 (root == fs_info->uuid_root) ||
464 (trans->adding_csums &&
465 root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID))
466 block_rsv = trans->block_rsv;
469 block_rsv = root->block_rsv;
472 block_rsv = &fs_info->empty_block_rsv;
477 struct btrfs_block_rsv *btrfs_use_block_rsv(struct btrfs_trans_handle *trans,
478 struct btrfs_root *root,
481 struct btrfs_fs_info *fs_info = root->fs_info;
482 struct btrfs_block_rsv *block_rsv;
483 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
485 bool global_updated = false;
487 block_rsv = get_block_rsv(trans, root);
489 if (unlikely(block_rsv->size == 0))
492 ret = btrfs_block_rsv_use_bytes(block_rsv, blocksize);
496 if (block_rsv->failfast)
499 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
500 global_updated = true;
501 btrfs_update_global_block_rsv(fs_info);
506 * The global reserve still exists to save us from ourselves, so don't
507 * warn_on if we are short on our delayed refs reserve.
509 if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
510 btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
511 static DEFINE_RATELIMIT_STATE(_rs,
512 DEFAULT_RATELIMIT_INTERVAL * 10,
513 /*DEFAULT_RATELIMIT_BURST*/ 1);
514 if (__ratelimit(&_rs))
516 "BTRFS: block rsv %d returned %d\n",
517 block_rsv->type, ret);
520 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv, blocksize,
521 BTRFS_RESERVE_NO_FLUSH);
525 * If we couldn't reserve metadata bytes try and use some from
526 * the global reserve if its space type is the same as the global
529 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
530 block_rsv->space_info == global_rsv->space_info) {
531 ret = btrfs_block_rsv_use_bytes(global_rsv, blocksize);
537 * All hope is lost, but of course our reservations are overly
538 * pessimistic, so instead of possibly having an ENOSPC abort here, try
539 * one last time to force a reservation if there's enough actual space
540 * on disk to make the reservation.
542 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv, blocksize,
543 BTRFS_RESERVE_FLUSH_EMERGENCY);
550 int btrfs_check_trunc_cache_free_space(struct btrfs_fs_info *fs_info,
551 struct btrfs_block_rsv *rsv)
556 /* 1 for slack space, 1 for updating the inode */
557 needed_bytes = btrfs_calc_insert_metadata_size(fs_info, 1) +
558 btrfs_calc_metadata_size(fs_info, 1);
560 spin_lock(&rsv->lock);
561 if (rsv->reserved < needed_bytes)
565 spin_unlock(&rsv->lock);