1 // SPDX-License-Identifier: GPL-2.0
6 #include "space-info.h"
7 #include "transaction.h"
8 #include "block-group.h"
10 #include "accessors.h"
13 * HOW DO BLOCK RESERVES WORK
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.
20 * ->failfast exists for the truncate case, and is described below.
25 * Entrance: btrfs_block_rsv_add, btrfs_block_rsv_refill
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.
31 * ->size is an over-estimation of how much we may use for a particular
35 * Entrance: btrfs_use_block_rsv
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.
43 * Entrance: btrfs_block_rsv_release
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.
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.
53 * TYPES OF BLOCK RESERVES
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).
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.
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.
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.
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.
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.
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.
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
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)
109 struct btrfs_space_info *space_info = block_rsv->space_info;
110 u64 qgroup_to_release = 0;
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;
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;
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;
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(space_info, num_bytes);
155 if (qgroup_to_release_ret)
156 *qgroup_to_release_ret = qgroup_to_release;
160 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
161 struct btrfs_block_rsv *dst, u64 num_bytes,
166 ret = btrfs_block_rsv_use_bytes(src, num_bytes);
170 btrfs_block_rsv_add_bytes(dst, num_bytes, update_size);
174 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, enum btrfs_rsv_type type)
176 memset(rsv, 0, sizeof(*rsv));
177 spin_lock_init(&rsv->lock);
181 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
182 struct btrfs_block_rsv *rsv,
183 enum btrfs_rsv_type type)
185 btrfs_init_block_rsv(rsv, type);
186 rsv->space_info = btrfs_find_space_info(fs_info,
187 BTRFS_BLOCK_GROUP_METADATA);
190 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
191 enum btrfs_rsv_type type)
193 struct btrfs_block_rsv *block_rsv;
195 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
199 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
203 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
204 struct btrfs_block_rsv *rsv)
208 btrfs_block_rsv_release(fs_info, rsv, (u64)-1, NULL);
212 int btrfs_block_rsv_add(struct btrfs_fs_info *fs_info,
213 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
214 enum btrfs_reserve_flush_enum flush)
221 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv->space_info,
224 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, true);
229 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_percent)
234 spin_lock(&block_rsv->lock);
235 num_bytes = mult_perc(block_rsv->size, min_percent);
236 if (block_rsv->reserved >= num_bytes)
238 spin_unlock(&block_rsv->lock);
243 int btrfs_block_rsv_refill(struct btrfs_fs_info *fs_info,
244 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
245 enum btrfs_reserve_flush_enum flush)
252 spin_lock(&block_rsv->lock);
253 if (block_rsv->reserved >= num_bytes)
256 num_bytes -= block_rsv->reserved;
257 spin_unlock(&block_rsv->lock);
262 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv->space_info,
265 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, false);
272 u64 btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
273 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
274 u64 *qgroup_to_release)
276 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
277 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
278 struct btrfs_block_rsv *target = NULL;
281 * If we are a delayed block reserve then push to the global rsv,
282 * otherwise dump into the global delayed reserve if it is not full.
284 if (block_rsv->type == BTRFS_BLOCK_RSV_DELOPS)
286 else if (block_rsv != global_rsv && !btrfs_block_rsv_full(delayed_rsv))
287 target = delayed_rsv;
289 if (target && block_rsv->space_info != target->space_info)
292 return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
296 int btrfs_block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, u64 num_bytes)
300 spin_lock(&block_rsv->lock);
301 if (block_rsv->reserved >= num_bytes) {
302 block_rsv->reserved -= num_bytes;
303 if (block_rsv->reserved < block_rsv->size)
304 block_rsv->full = false;
307 spin_unlock(&block_rsv->lock);
311 void btrfs_block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
312 u64 num_bytes, bool update_size)
314 spin_lock(&block_rsv->lock);
315 block_rsv->reserved += num_bytes;
317 block_rsv->size += num_bytes;
318 else if (block_rsv->reserved >= block_rsv->size)
319 block_rsv->full = true;
320 spin_unlock(&block_rsv->lock);
323 void btrfs_update_global_block_rsv(struct btrfs_fs_info *fs_info)
325 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
326 struct btrfs_space_info *sinfo = block_rsv->space_info;
327 struct btrfs_root *root, *tmp;
328 u64 num_bytes = btrfs_root_used(&fs_info->tree_root->root_item);
329 unsigned int min_items = 1;
332 * The global block rsv is based on the size of the extent tree, the
333 * checksum tree and the root tree. If the fs is empty we want to set
334 * it to a minimal amount for safety.
336 * We also are going to need to modify the minimum of the tree root and
337 * any global roots we could touch.
339 read_lock(&fs_info->global_root_lock);
340 rbtree_postorder_for_each_entry_safe(root, tmp, &fs_info->global_root_tree,
342 if (btrfs_root_id(root) == BTRFS_EXTENT_TREE_OBJECTID ||
343 btrfs_root_id(root) == BTRFS_CSUM_TREE_OBJECTID ||
344 btrfs_root_id(root) == BTRFS_FREE_SPACE_TREE_OBJECTID) {
345 num_bytes += btrfs_root_used(&root->root_item);
349 read_unlock(&fs_info->global_root_lock);
351 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
352 num_bytes += btrfs_root_used(&fs_info->block_group_root->root_item);
356 if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
357 num_bytes += btrfs_root_used(&fs_info->stripe_root->root_item);
362 * But we also want to reserve enough space so we can do the fallback
363 * global reserve for an unlink, which is an additional
364 * BTRFS_UNLINK_METADATA_UNITS items.
366 * But we also need space for the delayed ref updates from the unlink,
367 * so add BTRFS_UNLINK_METADATA_UNITS units for delayed refs, one for
368 * each unlink metadata item.
370 min_items += BTRFS_UNLINK_METADATA_UNITS;
372 num_bytes = max_t(u64, num_bytes,
373 btrfs_calc_insert_metadata_size(fs_info, min_items) +
374 btrfs_calc_delayed_ref_bytes(fs_info,
375 BTRFS_UNLINK_METADATA_UNITS));
377 spin_lock(&sinfo->lock);
378 spin_lock(&block_rsv->lock);
380 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
382 if (block_rsv->reserved < block_rsv->size) {
383 num_bytes = block_rsv->size - block_rsv->reserved;
384 btrfs_space_info_update_bytes_may_use(sinfo, num_bytes);
385 block_rsv->reserved = block_rsv->size;
386 } else if (block_rsv->reserved > block_rsv->size) {
387 num_bytes = block_rsv->reserved - block_rsv->size;
388 btrfs_space_info_update_bytes_may_use(sinfo, -num_bytes);
389 block_rsv->reserved = block_rsv->size;
390 btrfs_try_granting_tickets(fs_info, sinfo);
393 block_rsv->full = (block_rsv->reserved == block_rsv->size);
395 if (block_rsv->size >= sinfo->total_bytes)
396 sinfo->force_alloc = CHUNK_ALLOC_FORCE;
397 spin_unlock(&block_rsv->lock);
398 spin_unlock(&sinfo->lock);
401 void btrfs_init_root_block_rsv(struct btrfs_root *root)
403 struct btrfs_fs_info *fs_info = root->fs_info;
405 switch (btrfs_root_id(root)) {
406 case BTRFS_CSUM_TREE_OBJECTID:
407 case BTRFS_EXTENT_TREE_OBJECTID:
408 case BTRFS_FREE_SPACE_TREE_OBJECTID:
409 case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
410 case BTRFS_RAID_STRIPE_TREE_OBJECTID:
411 root->block_rsv = &fs_info->delayed_refs_rsv;
413 case BTRFS_ROOT_TREE_OBJECTID:
414 case BTRFS_DEV_TREE_OBJECTID:
415 case BTRFS_QUOTA_TREE_OBJECTID:
416 root->block_rsv = &fs_info->global_block_rsv;
418 case BTRFS_CHUNK_TREE_OBJECTID:
419 root->block_rsv = &fs_info->chunk_block_rsv;
422 root->block_rsv = NULL;
427 void btrfs_init_global_block_rsv(struct btrfs_fs_info *fs_info)
429 struct btrfs_space_info *space_info;
431 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
432 fs_info->chunk_block_rsv.space_info = space_info;
434 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
435 fs_info->global_block_rsv.space_info = space_info;
436 fs_info->trans_block_rsv.space_info = space_info;
437 fs_info->empty_block_rsv.space_info = space_info;
438 fs_info->delayed_block_rsv.space_info = space_info;
439 fs_info->delayed_refs_rsv.space_info = space_info;
441 btrfs_update_global_block_rsv(fs_info);
444 void btrfs_release_global_block_rsv(struct btrfs_fs_info *fs_info)
446 btrfs_block_rsv_release(fs_info, &fs_info->global_block_rsv, (u64)-1,
448 WARN_ON(fs_info->trans_block_rsv.size > 0);
449 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
450 WARN_ON(fs_info->chunk_block_rsv.size > 0);
451 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
452 WARN_ON(fs_info->delayed_block_rsv.size > 0);
453 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
454 WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
455 WARN_ON(fs_info->delayed_refs_rsv.size > 0);
458 static struct btrfs_block_rsv *get_block_rsv(
459 const struct btrfs_trans_handle *trans,
460 const struct btrfs_root *root)
462 struct btrfs_fs_info *fs_info = root->fs_info;
463 struct btrfs_block_rsv *block_rsv = NULL;
465 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
466 (root == fs_info->uuid_root) ||
467 (trans->adding_csums && btrfs_root_id(root) == BTRFS_CSUM_TREE_OBJECTID))
468 block_rsv = trans->block_rsv;
471 block_rsv = root->block_rsv;
474 block_rsv = &fs_info->empty_block_rsv;
479 struct btrfs_block_rsv *btrfs_use_block_rsv(struct btrfs_trans_handle *trans,
480 struct btrfs_root *root,
483 struct btrfs_fs_info *fs_info = root->fs_info;
484 struct btrfs_block_rsv *block_rsv;
485 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
487 bool global_updated = false;
489 block_rsv = get_block_rsv(trans, root);
491 if (unlikely(btrfs_block_rsv_size(block_rsv) == 0))
494 ret = btrfs_block_rsv_use_bytes(block_rsv, blocksize);
498 if (block_rsv->failfast)
501 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
502 global_updated = true;
503 btrfs_update_global_block_rsv(fs_info);
508 * The global reserve still exists to save us from ourselves, so don't
509 * warn_on if we are short on our delayed refs reserve.
511 if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
512 btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
513 static DEFINE_RATELIMIT_STATE(_rs,
514 DEFAULT_RATELIMIT_INTERVAL * 10,
515 /*DEFAULT_RATELIMIT_BURST*/ 1);
516 if (__ratelimit(&_rs))
518 "BTRFS: block rsv %d returned %d\n",
519 block_rsv->type, ret);
522 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv->space_info,
523 blocksize, BTRFS_RESERVE_NO_FLUSH);
527 * If we couldn't reserve metadata bytes try and use some from
528 * the global reserve if its space type is the same as the global
531 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
532 block_rsv->space_info == global_rsv->space_info) {
533 ret = btrfs_block_rsv_use_bytes(global_rsv, blocksize);
539 * All hope is lost, but of course our reservations are overly
540 * pessimistic, so instead of possibly having an ENOSPC abort here, try
541 * one last time to force a reservation if there's enough actual space
542 * on disk to make the reservation.
544 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv->space_info, blocksize,
545 BTRFS_RESERVE_FLUSH_EMERGENCY);
552 int btrfs_check_trunc_cache_free_space(const struct btrfs_fs_info *fs_info,
553 struct btrfs_block_rsv *rsv)
558 /* 1 for slack space, 1 for updating the inode */
559 needed_bytes = btrfs_calc_insert_metadata_size(fs_info, 1) +
560 btrfs_calc_metadata_size(fs_info, 1);
562 spin_lock(&rsv->lock);
563 if (rsv->reserved < needed_bytes)
567 spin_unlock(&rsv->lock);