1 // SPDX-License-Identifier: GPL-2.0
5 #include "space-info.h"
8 #include "free-space-cache.h"
9 #include "ordered-data.h"
10 #include "transaction.h"
11 #include "block-group.h"
14 * HOW DOES SPACE RESERVATION WORK
16 * If you want to know about delalloc specifically, there is a separate comment
17 * for that with the delalloc code. This comment is about how the whole system
22 * 1) space_info. This is the ultimate arbiter of how much space we can use.
23 * There's a description of the bytes_ fields with the struct declaration,
24 * refer to that for specifics on each field. Suffice it to say that for
25 * reservations we care about total_bytes - SUM(space_info->bytes_) when
26 * determining if there is space to make an allocation. There is a space_info
27 * for METADATA, SYSTEM, and DATA areas.
29 * 2) block_rsv's. These are basically buckets for every different type of
30 * metadata reservation we have. You can see the comment in the block_rsv
31 * code on the rules for each type, but generally block_rsv->reserved is how
32 * much space is accounted for in space_info->bytes_may_use.
34 * 3) btrfs_calc*_size. These are the worst case calculations we used based
35 * on the number of items we will want to modify. We have one for changing
36 * items, and one for inserting new items. Generally we use these helpers to
37 * determine the size of the block reserves, and then use the actual bytes
38 * values to adjust the space_info counters.
40 * MAKING RESERVATIONS, THE NORMAL CASE
42 * We call into either btrfs_reserve_data_bytes() or
43 * btrfs_reserve_metadata_bytes(), depending on which we're looking for, with
44 * num_bytes we want to reserve.
47 * space_info->bytes_may_reserve += num_bytes
50 * Call btrfs_add_reserved_bytes() which does
51 * space_info->bytes_may_reserve -= num_bytes
52 * space_info->bytes_reserved += extent_bytes
55 * Call btrfs_update_block_group() which does
56 * space_info->bytes_reserved -= extent_bytes
57 * space_info->bytes_used += extent_bytes
59 * MAKING RESERVATIONS, FLUSHING NORMALLY (non-priority)
61 * Assume we are unable to simply make the reservation because we do not have
65 * create a reserve_ticket with ->bytes set to our reservation, add it to
66 * the tail of space_info->tickets, kick async flush thread
68 * ->handle_reserve_ticket
69 * wait on ticket->wait for ->bytes to be reduced to 0, or ->error to be set
72 * -> btrfs_async_reclaim_metadata_space/btrfs_async_reclaim_data_space
73 * Flushes various things attempting to free up space.
75 * -> btrfs_try_granting_tickets()
76 * This is called by anything that either subtracts space from
77 * space_info->bytes_may_use, ->bytes_pinned, etc, or adds to the
78 * space_info->total_bytes. This loops through the ->priority_tickets and
79 * then the ->tickets list checking to see if the reservation can be
80 * completed. If it can the space is added to space_info->bytes_may_use and
81 * the ticket is woken up.
84 * Check if ->bytes == 0, if it does we got our reservation and we can carry
85 * on, if not return the appropriate error (ENOSPC, but can be EINTR if we
88 * MAKING RESERVATIONS, FLUSHING HIGH PRIORITY
90 * Same as the above, except we add ourselves to the
91 * space_info->priority_tickets, and we do not use ticket->wait, we simply
92 * call flush_space() ourselves for the states that are safe for us to call
93 * without deadlocking and hope for the best.
97 * Generally speaking we will have two cases for each state, a "nice" state
98 * and a "ALL THE THINGS" state. In btrfs we delay a lot of work in order to
99 * reduce the locking over head on the various trees, and even to keep from
100 * doing any work at all in the case of delayed refs. Each of these delayed
101 * things however hold reservations, and so letting them run allows us to
102 * reclaim space so we can make new reservations.
104 * FLUSH_DELAYED_ITEMS
105 * Every inode has a delayed item to update the inode. Take a simple write
106 * for example, we would update the inode item at write time to update the
107 * mtime, and then again at finish_ordered_io() time in order to update the
108 * isize or bytes. We keep these delayed items to coalesce these operations
109 * into a single operation done on demand. These are an easy way to reclaim
113 * Look at the delalloc comment to get an idea of how much space is reserved
114 * for delayed allocation. We can reclaim some of this space simply by
115 * running delalloc, but usually we need to wait for ordered extents to
116 * reclaim the bulk of this space.
119 * We have a block reserve for the outstanding delayed refs space, and every
120 * delayed ref operation holds a reservation. Running these is a quick way
121 * to reclaim space, but we want to hold this until the end because COW can
122 * churn a lot and we can avoid making some extent tree modifications if we
123 * are able to delay for as long as possible.
126 * We will skip this the first time through space reservation, because of
127 * overcommit and we don't want to have a lot of useless metadata space when
128 * our worst case reservations will likely never come true.
131 * If we're freeing inodes we're likely freeing checksums, file extent
132 * items, and extent tree items. Loads of space could be freed up by these
133 * operations, however they won't be usable until the transaction commits.
136 * This will commit the transaction. Historically we had a lot of logic
137 * surrounding whether or not we'd commit the transaction, but this waits born
138 * out of a pre-tickets era where we could end up committing the transaction
139 * thousands of times in a row without making progress. Now thanks to our
140 * ticketing system we know if we're not making progress and can error
141 * everybody out after a few commits rather than burning the disk hoping for
142 * a different answer.
146 * Because we hold so many reservations for metadata we will allow you to
147 * reserve more space than is currently free in the currently allocate
148 * metadata space. This only happens with metadata, data does not allow
151 * You can see the current logic for when we allow overcommit in
152 * btrfs_can_overcommit(), but it only applies to unallocated space. If there
153 * is no unallocated space to be had, all reservations are kept within the
154 * free space in the allocated metadata chunks.
156 * Because of overcommitting, you generally want to use the
157 * btrfs_can_overcommit() logic for metadata allocations, as it does the right
158 * thing with or without extra unallocated space.
161 u64 __pure btrfs_space_info_used(struct btrfs_space_info *s_info,
162 bool may_use_included)
165 return s_info->bytes_used + s_info->bytes_reserved +
166 s_info->bytes_pinned + s_info->bytes_readonly +
167 s_info->bytes_zone_unusable +
168 (may_use_included ? s_info->bytes_may_use : 0);
172 * after adding space to the filesystem, we need to clear the full flags
173 * on all the space infos.
175 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
177 struct list_head *head = &info->space_info;
178 struct btrfs_space_info *found;
180 list_for_each_entry(found, head, list)
185 * Block groups with more than this value (percents) of unusable space will be
186 * scheduled for background reclaim.
188 #define BTRFS_DEFAULT_ZONED_RECLAIM_THRESH (75)
190 static int create_space_info(struct btrfs_fs_info *info, u64 flags)
193 struct btrfs_space_info *space_info;
197 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
201 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
202 INIT_LIST_HEAD(&space_info->block_groups[i]);
203 init_rwsem(&space_info->groups_sem);
204 spin_lock_init(&space_info->lock);
205 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
206 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
207 INIT_LIST_HEAD(&space_info->ro_bgs);
208 INIT_LIST_HEAD(&space_info->tickets);
209 INIT_LIST_HEAD(&space_info->priority_tickets);
210 space_info->clamp = 1;
212 if (btrfs_is_zoned(info))
213 space_info->bg_reclaim_threshold = BTRFS_DEFAULT_ZONED_RECLAIM_THRESH;
215 ret = btrfs_sysfs_add_space_info_type(info, space_info);
219 list_add(&space_info->list, &info->space_info);
220 if (flags & BTRFS_BLOCK_GROUP_DATA)
221 info->data_sinfo = space_info;
226 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
228 struct btrfs_super_block *disk_super;
234 disk_super = fs_info->super_copy;
235 if (!btrfs_super_root(disk_super))
238 features = btrfs_super_incompat_flags(disk_super);
239 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
242 flags = BTRFS_BLOCK_GROUP_SYSTEM;
243 ret = create_space_info(fs_info, flags);
248 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
249 ret = create_space_info(fs_info, flags);
251 flags = BTRFS_BLOCK_GROUP_METADATA;
252 ret = create_space_info(fs_info, flags);
256 flags = BTRFS_BLOCK_GROUP_DATA;
257 ret = create_space_info(fs_info, flags);
263 void btrfs_update_space_info(struct btrfs_fs_info *info, u64 flags,
264 u64 total_bytes, u64 bytes_used,
265 u64 bytes_readonly, u64 bytes_zone_unusable,
266 struct btrfs_space_info **space_info)
268 struct btrfs_space_info *found;
271 factor = btrfs_bg_type_to_factor(flags);
273 found = btrfs_find_space_info(info, flags);
275 spin_lock(&found->lock);
276 found->total_bytes += total_bytes;
277 found->disk_total += total_bytes * factor;
278 found->bytes_used += bytes_used;
279 found->disk_used += bytes_used * factor;
280 found->bytes_readonly += bytes_readonly;
281 found->bytes_zone_unusable += bytes_zone_unusable;
284 btrfs_try_granting_tickets(info, found);
285 spin_unlock(&found->lock);
289 struct btrfs_space_info *btrfs_find_space_info(struct btrfs_fs_info *info,
292 struct list_head *head = &info->space_info;
293 struct btrfs_space_info *found;
295 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
297 list_for_each_entry(found, head, list) {
298 if (found->flags & flags)
304 static u64 calc_available_free_space(struct btrfs_fs_info *fs_info,
305 struct btrfs_space_info *space_info,
306 enum btrfs_reserve_flush_enum flush)
312 if (space_info->flags & BTRFS_BLOCK_GROUP_SYSTEM)
313 profile = btrfs_system_alloc_profile(fs_info);
315 profile = btrfs_metadata_alloc_profile(fs_info);
317 avail = atomic64_read(&fs_info->free_chunk_space);
320 * If we have dup, raid1 or raid10 then only half of the free
321 * space is actually usable. For raid56, the space info used
322 * doesn't include the parity drive, so we don't have to
325 factor = btrfs_bg_type_to_factor(profile);
326 avail = div_u64(avail, factor);
329 * If we aren't flushing all things, let us overcommit up to
330 * 1/2th of the space. If we can flush, don't let us overcommit
331 * too much, let it overcommit up to 1/8 of the space.
333 if (flush == BTRFS_RESERVE_FLUSH_ALL)
340 int btrfs_can_overcommit(struct btrfs_fs_info *fs_info,
341 struct btrfs_space_info *space_info, u64 bytes,
342 enum btrfs_reserve_flush_enum flush)
347 /* Don't overcommit when in mixed mode */
348 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
351 used = btrfs_space_info_used(space_info, true);
352 avail = calc_available_free_space(fs_info, space_info, flush);
354 if (used + bytes < space_info->total_bytes + avail)
359 static void remove_ticket(struct btrfs_space_info *space_info,
360 struct reserve_ticket *ticket)
362 if (!list_empty(&ticket->list)) {
363 list_del_init(&ticket->list);
364 ASSERT(space_info->reclaim_size >= ticket->bytes);
365 space_info->reclaim_size -= ticket->bytes;
370 * This is for space we already have accounted in space_info->bytes_may_use, so
371 * basically when we're returning space from block_rsv's.
373 void btrfs_try_granting_tickets(struct btrfs_fs_info *fs_info,
374 struct btrfs_space_info *space_info)
376 struct list_head *head;
377 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
379 lockdep_assert_held(&space_info->lock);
381 head = &space_info->priority_tickets;
383 while (!list_empty(head)) {
384 struct reserve_ticket *ticket;
385 u64 used = btrfs_space_info_used(space_info, true);
387 ticket = list_first_entry(head, struct reserve_ticket, list);
389 /* Check and see if our ticket can be satisfied now. */
390 if ((used + ticket->bytes <= space_info->total_bytes) ||
391 btrfs_can_overcommit(fs_info, space_info, ticket->bytes,
393 btrfs_space_info_update_bytes_may_use(fs_info,
396 remove_ticket(space_info, ticket);
398 space_info->tickets_id++;
399 wake_up(&ticket->wait);
405 if (head == &space_info->priority_tickets) {
406 head = &space_info->tickets;
407 flush = BTRFS_RESERVE_FLUSH_ALL;
412 #define DUMP_BLOCK_RSV(fs_info, rsv_name) \
414 struct btrfs_block_rsv *__rsv = &(fs_info)->rsv_name; \
415 spin_lock(&__rsv->lock); \
416 btrfs_info(fs_info, #rsv_name ": size %llu reserved %llu", \
417 __rsv->size, __rsv->reserved); \
418 spin_unlock(&__rsv->lock); \
421 static void __btrfs_dump_space_info(struct btrfs_fs_info *fs_info,
422 struct btrfs_space_info *info)
424 lockdep_assert_held(&info->lock);
426 /* The free space could be negative in case of overcommit */
427 btrfs_info(fs_info, "space_info %llu has %lld free, is %sfull",
429 (s64)(info->total_bytes - btrfs_space_info_used(info, true)),
430 info->full ? "" : "not ");
432 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu zone_unusable=%llu",
433 info->total_bytes, info->bytes_used, info->bytes_pinned,
434 info->bytes_reserved, info->bytes_may_use,
435 info->bytes_readonly, info->bytes_zone_unusable);
437 DUMP_BLOCK_RSV(fs_info, global_block_rsv);
438 DUMP_BLOCK_RSV(fs_info, trans_block_rsv);
439 DUMP_BLOCK_RSV(fs_info, chunk_block_rsv);
440 DUMP_BLOCK_RSV(fs_info, delayed_block_rsv);
441 DUMP_BLOCK_RSV(fs_info, delayed_refs_rsv);
445 void btrfs_dump_space_info(struct btrfs_fs_info *fs_info,
446 struct btrfs_space_info *info, u64 bytes,
447 int dump_block_groups)
449 struct btrfs_block_group *cache;
452 spin_lock(&info->lock);
453 __btrfs_dump_space_info(fs_info, info);
454 spin_unlock(&info->lock);
456 if (!dump_block_groups)
459 down_read(&info->groups_sem);
461 list_for_each_entry(cache, &info->block_groups[index], list) {
462 spin_lock(&cache->lock);
464 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %llu zone_unusable %s",
465 cache->start, cache->length, cache->used, cache->pinned,
466 cache->reserved, cache->zone_unusable,
467 cache->ro ? "[readonly]" : "");
468 spin_unlock(&cache->lock);
469 btrfs_dump_free_space(cache, bytes);
471 if (++index < BTRFS_NR_RAID_TYPES)
473 up_read(&info->groups_sem);
476 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
482 bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
483 nr = div64_u64(to_reclaim, bytes);
489 #define EXTENT_SIZE_PER_ITEM SZ_256K
492 * shrink metadata reservation for delalloc
494 static void shrink_delalloc(struct btrfs_fs_info *fs_info,
495 struct btrfs_space_info *space_info,
496 u64 to_reclaim, bool wait_ordered,
499 struct btrfs_trans_handle *trans;
506 delalloc_bytes = percpu_counter_sum_positive(&fs_info->delalloc_bytes);
507 ordered_bytes = percpu_counter_sum_positive(&fs_info->ordered_bytes);
508 if (delalloc_bytes == 0 && ordered_bytes == 0)
511 /* Calc the number of the pages we need flush for space reservation */
512 if (to_reclaim == U64_MAX) {
516 * to_reclaim is set to however much metadata we need to
517 * reclaim, but reclaiming that much data doesn't really track
518 * exactly. What we really want to do is reclaim full inode's
519 * worth of reservations, however that's not available to us
520 * here. We will take a fraction of the delalloc bytes for our
521 * flushing loops and hope for the best. Delalloc will expand
522 * the amount we write to cover an entire dirty extent, which
523 * will reclaim the metadata reservation for that range. If
524 * it's not enough subsequent flush stages will be more
527 to_reclaim = max(to_reclaim, delalloc_bytes >> 3);
528 items = calc_reclaim_items_nr(fs_info, to_reclaim) * 2;
531 trans = current->journal_info;
534 * If we are doing more ordered than delalloc we need to just wait on
535 * ordered extents, otherwise we'll waste time trying to flush delalloc
536 * that likely won't give us the space back we need.
538 if (ordered_bytes > delalloc_bytes && !for_preempt)
542 while ((delalloc_bytes || ordered_bytes) && loops < 3) {
543 u64 temp = min(delalloc_bytes, to_reclaim) >> PAGE_SHIFT;
544 long nr_pages = min_t(u64, temp, LONG_MAX);
547 btrfs_start_delalloc_roots(fs_info, nr_pages, true);
550 * We need to make sure any outstanding async pages are now
551 * processed before we continue. This is because things like
552 * sync_inode() try to be smart and skip writing if the inode is
553 * marked clean. We don't use filemap_fwrite for flushing
554 * because we want to control how many pages we write out at a
555 * time, thus this is the only safe way to make sure we've
556 * waited for outstanding compressed workers to have started
557 * their jobs and thus have ordered extents set up properly.
559 * This exists because we do not want to wait for each
560 * individual inode to finish its async work, we simply want to
561 * start the IO on everybody, and then come back here and wait
562 * for all of the async work to catch up. Once we're done with
563 * that we know we'll have ordered extents for everything and we
564 * can decide if we wait for that or not.
566 * If we choose to replace this in the future, make absolutely
567 * sure that the proper waiting is being done in the async case,
568 * as there have been bugs in that area before.
570 async_pages = atomic_read(&fs_info->async_delalloc_pages);
575 * We don't want to wait forever, if we wrote less pages in this
576 * loop than we have outstanding, only wait for that number of
577 * pages, otherwise we can wait for all async pages to finish
580 if (async_pages > nr_pages)
581 async_pages -= nr_pages;
584 wait_event(fs_info->async_submit_wait,
585 atomic_read(&fs_info->async_delalloc_pages) <=
589 if (wait_ordered && !trans) {
590 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
592 time_left = schedule_timeout_killable(1);
598 * If we are for preemption we just want a one-shot of delalloc
599 * flushing so we can stop flushing if we decide we don't need
605 spin_lock(&space_info->lock);
606 if (list_empty(&space_info->tickets) &&
607 list_empty(&space_info->priority_tickets)) {
608 spin_unlock(&space_info->lock);
611 spin_unlock(&space_info->lock);
613 delalloc_bytes = percpu_counter_sum_positive(
614 &fs_info->delalloc_bytes);
615 ordered_bytes = percpu_counter_sum_positive(
616 &fs_info->ordered_bytes);
621 * Try to flush some data based on policy set by @state. This is only advisory
622 * and may fail for various reasons. The caller is supposed to examine the
623 * state of @space_info to detect the outcome.
625 static void flush_space(struct btrfs_fs_info *fs_info,
626 struct btrfs_space_info *space_info, u64 num_bytes,
627 enum btrfs_flush_state state, bool for_preempt)
629 struct btrfs_root *root = fs_info->tree_root;
630 struct btrfs_trans_handle *trans;
635 case FLUSH_DELAYED_ITEMS_NR:
636 case FLUSH_DELAYED_ITEMS:
637 if (state == FLUSH_DELAYED_ITEMS_NR)
638 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
642 trans = btrfs_join_transaction(root);
644 ret = PTR_ERR(trans);
647 ret = btrfs_run_delayed_items_nr(trans, nr);
648 btrfs_end_transaction(trans);
651 case FLUSH_DELALLOC_WAIT:
652 case FLUSH_DELALLOC_FULL:
653 if (state == FLUSH_DELALLOC_FULL)
655 shrink_delalloc(fs_info, space_info, num_bytes,
656 state != FLUSH_DELALLOC, for_preempt);
658 case FLUSH_DELAYED_REFS_NR:
659 case FLUSH_DELAYED_REFS:
660 trans = btrfs_join_transaction(root);
662 ret = PTR_ERR(trans);
665 if (state == FLUSH_DELAYED_REFS_NR)
666 nr = calc_reclaim_items_nr(fs_info, num_bytes);
669 btrfs_run_delayed_refs(trans, nr);
670 btrfs_end_transaction(trans);
673 case ALLOC_CHUNK_FORCE:
674 trans = btrfs_join_transaction(root);
676 ret = PTR_ERR(trans);
679 ret = btrfs_chunk_alloc(trans,
680 btrfs_get_alloc_profile(fs_info, space_info->flags),
681 (state == ALLOC_CHUNK) ? CHUNK_ALLOC_NO_FORCE :
683 btrfs_end_transaction(trans);
684 if (ret > 0 || ret == -ENOSPC)
687 case RUN_DELAYED_IPUTS:
689 * If we have pending delayed iputs then we could free up a
690 * bunch of pinned space, so make sure we run the iputs before
691 * we do our pinned bytes check below.
693 btrfs_run_delayed_iputs(fs_info);
694 btrfs_wait_on_delayed_iputs(fs_info);
697 ASSERT(current->journal_info == NULL);
698 trans = btrfs_join_transaction(root);
700 ret = PTR_ERR(trans);
703 ret = btrfs_commit_transaction(trans);
710 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
716 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
717 struct btrfs_space_info *space_info)
721 u64 to_reclaim = space_info->reclaim_size;
723 lockdep_assert_held(&space_info->lock);
725 avail = calc_available_free_space(fs_info, space_info,
726 BTRFS_RESERVE_FLUSH_ALL);
727 used = btrfs_space_info_used(space_info, true);
730 * We may be flushing because suddenly we have less space than we had
731 * before, and now we're well over-committed based on our current free
732 * space. If that's the case add in our overage so we make sure to put
733 * appropriate pressure on the flushing state machine.
735 if (space_info->total_bytes + avail < used)
736 to_reclaim += used - (space_info->total_bytes + avail);
741 static bool need_preemptive_reclaim(struct btrfs_fs_info *fs_info,
742 struct btrfs_space_info *space_info)
744 u64 global_rsv_size = fs_info->global_block_rsv.reserved;
745 u64 ordered, delalloc;
746 u64 thresh = div_factor_fine(space_info->total_bytes, 90);
749 lockdep_assert_held(&space_info->lock);
751 /* If we're just plain full then async reclaim just slows us down. */
752 if ((space_info->bytes_used + space_info->bytes_reserved +
753 global_rsv_size) >= thresh)
756 used = space_info->bytes_may_use + space_info->bytes_pinned;
758 /* The total flushable belongs to the global rsv, don't flush. */
759 if (global_rsv_size >= used)
763 * 128MiB is 1/4 of the maximum global rsv size. If we have less than
764 * that devoted to other reservations then there's no sense in flushing,
765 * we don't have a lot of things that need flushing.
767 if (used - global_rsv_size <= SZ_128M)
771 * We have tickets queued, bail so we don't compete with the async
774 if (space_info->reclaim_size)
778 * If we have over half of the free space occupied by reservations or
779 * pinned then we want to start flushing.
781 * We do not do the traditional thing here, which is to say
783 * if (used >= ((total_bytes + avail) / 2))
786 * because this doesn't quite work how we want. If we had more than 50%
787 * of the space_info used by bytes_used and we had 0 available we'd just
788 * constantly run the background flusher. Instead we want it to kick in
789 * if our reclaimable space exceeds our clamped free space.
791 * Our clamping range is 2^1 -> 2^8. Practically speaking that means
794 * Amount of RAM Minimum threshold Maximum threshold
797 * 128GiB 512MiB 64GiB
802 * These are the range our thresholds will fall in, corresponding to how
803 * much delalloc we need for the background flusher to kick in.
806 thresh = calc_available_free_space(fs_info, space_info,
807 BTRFS_RESERVE_FLUSH_ALL);
808 used = space_info->bytes_used + space_info->bytes_reserved +
809 space_info->bytes_readonly + global_rsv_size;
810 if (used < space_info->total_bytes)
811 thresh += space_info->total_bytes - used;
812 thresh >>= space_info->clamp;
814 used = space_info->bytes_pinned;
817 * If we have more ordered bytes than delalloc bytes then we're either
818 * doing a lot of DIO, or we simply don't have a lot of delalloc waiting
819 * around. Preemptive flushing is only useful in that it can free up
820 * space before tickets need to wait for things to finish. In the case
821 * of ordered extents, preemptively waiting on ordered extents gets us
822 * nothing, if our reservations are tied up in ordered extents we'll
823 * simply have to slow down writers by forcing them to wait on ordered
826 * In the case that ordered is larger than delalloc, only include the
827 * block reserves that we would actually be able to directly reclaim
828 * from. In this case if we're heavy on metadata operations this will
829 * clearly be heavy enough to warrant preemptive flushing. In the case
830 * of heavy DIO or ordered reservations, preemptive flushing will just
831 * waste time and cause us to slow down.
833 * We want to make sure we truly are maxed out on ordered however, so
834 * cut ordered in half, and if it's still higher than delalloc then we
835 * can keep flushing. This is to avoid the case where we start
836 * flushing, and now delalloc == ordered and we stop preemptively
837 * flushing when we could still have several gigs of delalloc to flush.
839 ordered = percpu_counter_read_positive(&fs_info->ordered_bytes) >> 1;
840 delalloc = percpu_counter_read_positive(&fs_info->delalloc_bytes);
841 if (ordered >= delalloc)
842 used += fs_info->delayed_refs_rsv.reserved +
843 fs_info->delayed_block_rsv.reserved;
845 used += space_info->bytes_may_use - global_rsv_size;
847 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
848 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
851 static bool steal_from_global_rsv(struct btrfs_fs_info *fs_info,
852 struct btrfs_space_info *space_info,
853 struct reserve_ticket *ticket)
855 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
861 if (global_rsv->space_info != space_info)
864 spin_lock(&global_rsv->lock);
865 min_bytes = div_factor(global_rsv->size, 1);
866 if (global_rsv->reserved < min_bytes + ticket->bytes) {
867 spin_unlock(&global_rsv->lock);
870 global_rsv->reserved -= ticket->bytes;
871 remove_ticket(space_info, ticket);
873 wake_up(&ticket->wait);
874 space_info->tickets_id++;
875 if (global_rsv->reserved < global_rsv->size)
876 global_rsv->full = 0;
877 spin_unlock(&global_rsv->lock);
883 * maybe_fail_all_tickets - we've exhausted our flushing, start failing tickets
884 * @fs_info - fs_info for this fs
885 * @space_info - the space info we were flushing
887 * We call this when we've exhausted our flushing ability and haven't made
888 * progress in satisfying tickets. The reservation code handles tickets in
889 * order, so if there is a large ticket first and then smaller ones we could
890 * very well satisfy the smaller tickets. This will attempt to wake up any
891 * tickets in the list to catch this case.
893 * This function returns true if it was able to make progress by clearing out
894 * other tickets, or if it stumbles across a ticket that was smaller than the
897 static bool maybe_fail_all_tickets(struct btrfs_fs_info *fs_info,
898 struct btrfs_space_info *space_info)
900 struct reserve_ticket *ticket;
901 u64 tickets_id = space_info->tickets_id;
902 const bool aborted = BTRFS_FS_ERROR(fs_info);
904 trace_btrfs_fail_all_tickets(fs_info, space_info);
906 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
907 btrfs_info(fs_info, "cannot satisfy tickets, dumping space info");
908 __btrfs_dump_space_info(fs_info, space_info);
911 while (!list_empty(&space_info->tickets) &&
912 tickets_id == space_info->tickets_id) {
913 ticket = list_first_entry(&space_info->tickets,
914 struct reserve_ticket, list);
916 if (!aborted && steal_from_global_rsv(fs_info, space_info, ticket))
919 if (!aborted && btrfs_test_opt(fs_info, ENOSPC_DEBUG))
920 btrfs_info(fs_info, "failing ticket with %llu bytes",
923 remove_ticket(space_info, ticket);
925 ticket->error = -EIO;
927 ticket->error = -ENOSPC;
928 wake_up(&ticket->wait);
931 * We're just throwing tickets away, so more flushing may not
932 * trip over btrfs_try_granting_tickets, so we need to call it
933 * here to see if we can make progress with the next ticket in
937 btrfs_try_granting_tickets(fs_info, space_info);
939 return (tickets_id != space_info->tickets_id);
943 * This is for normal flushers, we can wait all goddamned day if we want to. We
944 * will loop and continuously try to flush as long as we are making progress.
945 * We count progress as clearing off tickets each time we have to loop.
947 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
949 struct btrfs_fs_info *fs_info;
950 struct btrfs_space_info *space_info;
952 enum btrfs_flush_state flush_state;
953 int commit_cycles = 0;
956 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
957 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
959 spin_lock(&space_info->lock);
960 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info);
962 space_info->flush = 0;
963 spin_unlock(&space_info->lock);
966 last_tickets_id = space_info->tickets_id;
967 spin_unlock(&space_info->lock);
969 flush_state = FLUSH_DELAYED_ITEMS_NR;
971 flush_space(fs_info, space_info, to_reclaim, flush_state, false);
972 spin_lock(&space_info->lock);
973 if (list_empty(&space_info->tickets)) {
974 space_info->flush = 0;
975 spin_unlock(&space_info->lock);
978 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
980 if (last_tickets_id == space_info->tickets_id) {
983 last_tickets_id = space_info->tickets_id;
984 flush_state = FLUSH_DELAYED_ITEMS_NR;
990 * We do not want to empty the system of delalloc unless we're
991 * under heavy pressure, so allow one trip through the flushing
992 * logic before we start doing a FLUSH_DELALLOC_FULL.
994 if (flush_state == FLUSH_DELALLOC_FULL && !commit_cycles)
998 * We don't want to force a chunk allocation until we've tried
999 * pretty hard to reclaim space. Think of the case where we
1000 * freed up a bunch of space and so have a lot of pinned space
1001 * to reclaim. We would rather use that than possibly create a
1002 * underutilized metadata chunk. So if this is our first run
1003 * through the flushing state machine skip ALLOC_CHUNK_FORCE and
1004 * commit the transaction. If nothing has changed the next go
1005 * around then we can force a chunk allocation.
1007 if (flush_state == ALLOC_CHUNK_FORCE && !commit_cycles)
1010 if (flush_state > COMMIT_TRANS) {
1012 if (commit_cycles > 2) {
1013 if (maybe_fail_all_tickets(fs_info, space_info)) {
1014 flush_state = FLUSH_DELAYED_ITEMS_NR;
1017 space_info->flush = 0;
1020 flush_state = FLUSH_DELAYED_ITEMS_NR;
1023 spin_unlock(&space_info->lock);
1024 } while (flush_state <= COMMIT_TRANS);
1028 * This handles pre-flushing of metadata space before we get to the point that
1029 * we need to start blocking threads on tickets. The logic here is different
1030 * from the other flush paths because it doesn't rely on tickets to tell us how
1031 * much we need to flush, instead it attempts to keep us below the 80% full
1032 * watermark of space by flushing whichever reservation pool is currently the
1035 static void btrfs_preempt_reclaim_metadata_space(struct work_struct *work)
1037 struct btrfs_fs_info *fs_info;
1038 struct btrfs_space_info *space_info;
1039 struct btrfs_block_rsv *delayed_block_rsv;
1040 struct btrfs_block_rsv *delayed_refs_rsv;
1041 struct btrfs_block_rsv *global_rsv;
1042 struct btrfs_block_rsv *trans_rsv;
1045 fs_info = container_of(work, struct btrfs_fs_info,
1046 preempt_reclaim_work);
1047 space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
1048 delayed_block_rsv = &fs_info->delayed_block_rsv;
1049 delayed_refs_rsv = &fs_info->delayed_refs_rsv;
1050 global_rsv = &fs_info->global_block_rsv;
1051 trans_rsv = &fs_info->trans_block_rsv;
1053 spin_lock(&space_info->lock);
1054 while (need_preemptive_reclaim(fs_info, space_info)) {
1055 enum btrfs_flush_state flush;
1056 u64 delalloc_size = 0;
1057 u64 to_reclaim, block_rsv_size;
1058 u64 global_rsv_size = global_rsv->reserved;
1063 * We don't have a precise counter for the metadata being
1064 * reserved for delalloc, so we'll approximate it by subtracting
1065 * out the block rsv's space from the bytes_may_use. If that
1066 * amount is higher than the individual reserves, then we can
1067 * assume it's tied up in delalloc reservations.
1069 block_rsv_size = global_rsv_size +
1070 delayed_block_rsv->reserved +
1071 delayed_refs_rsv->reserved +
1072 trans_rsv->reserved;
1073 if (block_rsv_size < space_info->bytes_may_use)
1074 delalloc_size = space_info->bytes_may_use - block_rsv_size;
1077 * We don't want to include the global_rsv in our calculation,
1078 * because that's space we can't touch. Subtract it from the
1079 * block_rsv_size for the next checks.
1081 block_rsv_size -= global_rsv_size;
1084 * We really want to avoid flushing delalloc too much, as it
1085 * could result in poor allocation patterns, so only flush it if
1086 * it's larger than the rest of the pools combined.
1088 if (delalloc_size > block_rsv_size) {
1089 to_reclaim = delalloc_size;
1090 flush = FLUSH_DELALLOC;
1091 } else if (space_info->bytes_pinned >
1092 (delayed_block_rsv->reserved +
1093 delayed_refs_rsv->reserved)) {
1094 to_reclaim = space_info->bytes_pinned;
1095 flush = COMMIT_TRANS;
1096 } else if (delayed_block_rsv->reserved >
1097 delayed_refs_rsv->reserved) {
1098 to_reclaim = delayed_block_rsv->reserved;
1099 flush = FLUSH_DELAYED_ITEMS_NR;
1101 to_reclaim = delayed_refs_rsv->reserved;
1102 flush = FLUSH_DELAYED_REFS_NR;
1105 spin_unlock(&space_info->lock);
1108 * We don't want to reclaim everything, just a portion, so scale
1109 * down the to_reclaim by 1/4. If it takes us down to 0,
1110 * reclaim 1 items worth.
1114 to_reclaim = btrfs_calc_insert_metadata_size(fs_info, 1);
1115 flush_space(fs_info, space_info, to_reclaim, flush, true);
1117 spin_lock(&space_info->lock);
1120 /* We only went through once, back off our clamping. */
1121 if (loops == 1 && !space_info->reclaim_size)
1122 space_info->clamp = max(1, space_info->clamp - 1);
1123 trace_btrfs_done_preemptive_reclaim(fs_info, space_info);
1124 spin_unlock(&space_info->lock);
1128 * FLUSH_DELALLOC_WAIT:
1129 * Space is freed from flushing delalloc in one of two ways.
1131 * 1) compression is on and we allocate less space than we reserved
1132 * 2) we are overwriting existing space
1134 * For #1 that extra space is reclaimed as soon as the delalloc pages are
1135 * COWed, by way of btrfs_add_reserved_bytes() which adds the actual extent
1136 * length to ->bytes_reserved, and subtracts the reserved space from
1139 * For #2 this is trickier. Once the ordered extent runs we will drop the
1140 * extent in the range we are overwriting, which creates a delayed ref for
1141 * that freed extent. This however is not reclaimed until the transaction
1142 * commits, thus the next stages.
1145 * If we are freeing inodes, we want to make sure all delayed iputs have
1146 * completed, because they could have been on an inode with i_nlink == 0, and
1147 * thus have been truncated and freed up space. But again this space is not
1148 * immediately re-usable, it comes in the form of a delayed ref, which must be
1149 * run and then the transaction must be committed.
1152 * This is where we reclaim all of the pinned space generated by running the
1156 * For data we start with alloc chunk force, however we could have been full
1157 * before, and then the transaction commit could have freed new block groups,
1158 * so if we now have space to allocate do the force chunk allocation.
1160 static const enum btrfs_flush_state data_flush_states[] = {
1161 FLUSH_DELALLOC_FULL,
1167 static void btrfs_async_reclaim_data_space(struct work_struct *work)
1169 struct btrfs_fs_info *fs_info;
1170 struct btrfs_space_info *space_info;
1171 u64 last_tickets_id;
1172 enum btrfs_flush_state flush_state = 0;
1174 fs_info = container_of(work, struct btrfs_fs_info, async_data_reclaim_work);
1175 space_info = fs_info->data_sinfo;
1177 spin_lock(&space_info->lock);
1178 if (list_empty(&space_info->tickets)) {
1179 space_info->flush = 0;
1180 spin_unlock(&space_info->lock);
1183 last_tickets_id = space_info->tickets_id;
1184 spin_unlock(&space_info->lock);
1186 while (!space_info->full) {
1187 flush_space(fs_info, space_info, U64_MAX, ALLOC_CHUNK_FORCE, false);
1188 spin_lock(&space_info->lock);
1189 if (list_empty(&space_info->tickets)) {
1190 space_info->flush = 0;
1191 spin_unlock(&space_info->lock);
1195 /* Something happened, fail everything and bail. */
1196 if (BTRFS_FS_ERROR(fs_info))
1198 last_tickets_id = space_info->tickets_id;
1199 spin_unlock(&space_info->lock);
1202 while (flush_state < ARRAY_SIZE(data_flush_states)) {
1203 flush_space(fs_info, space_info, U64_MAX,
1204 data_flush_states[flush_state], false);
1205 spin_lock(&space_info->lock);
1206 if (list_empty(&space_info->tickets)) {
1207 space_info->flush = 0;
1208 spin_unlock(&space_info->lock);
1212 if (last_tickets_id == space_info->tickets_id) {
1215 last_tickets_id = space_info->tickets_id;
1219 if (flush_state >= ARRAY_SIZE(data_flush_states)) {
1220 if (space_info->full) {
1221 if (maybe_fail_all_tickets(fs_info, space_info))
1224 space_info->flush = 0;
1229 /* Something happened, fail everything and bail. */
1230 if (BTRFS_FS_ERROR(fs_info))
1234 spin_unlock(&space_info->lock);
1239 maybe_fail_all_tickets(fs_info, space_info);
1240 space_info->flush = 0;
1241 spin_unlock(&space_info->lock);
1244 void btrfs_init_async_reclaim_work(struct btrfs_fs_info *fs_info)
1246 INIT_WORK(&fs_info->async_reclaim_work, btrfs_async_reclaim_metadata_space);
1247 INIT_WORK(&fs_info->async_data_reclaim_work, btrfs_async_reclaim_data_space);
1248 INIT_WORK(&fs_info->preempt_reclaim_work,
1249 btrfs_preempt_reclaim_metadata_space);
1252 static const enum btrfs_flush_state priority_flush_states[] = {
1253 FLUSH_DELAYED_ITEMS_NR,
1254 FLUSH_DELAYED_ITEMS,
1258 static const enum btrfs_flush_state evict_flush_states[] = {
1259 FLUSH_DELAYED_ITEMS_NR,
1260 FLUSH_DELAYED_ITEMS,
1261 FLUSH_DELAYED_REFS_NR,
1264 FLUSH_DELALLOC_WAIT,
1265 FLUSH_DELALLOC_FULL,
1270 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
1271 struct btrfs_space_info *space_info,
1272 struct reserve_ticket *ticket,
1273 const enum btrfs_flush_state *states,
1277 int flush_state = 0;
1279 spin_lock(&space_info->lock);
1280 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info);
1282 * This is the priority reclaim path, so to_reclaim could be >0 still
1283 * because we may have only satisified the priority tickets and still
1284 * left non priority tickets on the list. We would then have
1285 * to_reclaim but ->bytes == 0.
1287 if (ticket->bytes == 0) {
1288 spin_unlock(&space_info->lock);
1292 while (flush_state < states_nr) {
1293 spin_unlock(&space_info->lock);
1294 flush_space(fs_info, space_info, to_reclaim, states[flush_state],
1297 spin_lock(&space_info->lock);
1298 if (ticket->bytes == 0) {
1299 spin_unlock(&space_info->lock);
1304 /* Attempt to steal from the global rsv if we can. */
1305 if (!steal_from_global_rsv(fs_info, space_info, ticket)) {
1306 ticket->error = -ENOSPC;
1307 remove_ticket(space_info, ticket);
1311 * We must run try_granting_tickets here because we could be a large
1312 * ticket in front of a smaller ticket that can now be satisfied with
1313 * the available space.
1315 btrfs_try_granting_tickets(fs_info, space_info);
1316 spin_unlock(&space_info->lock);
1319 static void priority_reclaim_data_space(struct btrfs_fs_info *fs_info,
1320 struct btrfs_space_info *space_info,
1321 struct reserve_ticket *ticket)
1323 spin_lock(&space_info->lock);
1325 /* We could have been granted before we got here. */
1326 if (ticket->bytes == 0) {
1327 spin_unlock(&space_info->lock);
1331 while (!space_info->full) {
1332 spin_unlock(&space_info->lock);
1333 flush_space(fs_info, space_info, U64_MAX, ALLOC_CHUNK_FORCE, false);
1334 spin_lock(&space_info->lock);
1335 if (ticket->bytes == 0) {
1336 spin_unlock(&space_info->lock);
1341 ticket->error = -ENOSPC;
1342 remove_ticket(space_info, ticket);
1343 btrfs_try_granting_tickets(fs_info, space_info);
1344 spin_unlock(&space_info->lock);
1347 static void wait_reserve_ticket(struct btrfs_fs_info *fs_info,
1348 struct btrfs_space_info *space_info,
1349 struct reserve_ticket *ticket)
1355 spin_lock(&space_info->lock);
1356 while (ticket->bytes > 0 && ticket->error == 0) {
1357 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
1360 * Delete us from the list. After we unlock the space
1361 * info, we don't want the async reclaim job to reserve
1362 * space for this ticket. If that would happen, then the
1363 * ticket's task would not known that space was reserved
1364 * despite getting an error, resulting in a space leak
1365 * (bytes_may_use counter of our space_info).
1367 remove_ticket(space_info, ticket);
1368 ticket->error = -EINTR;
1371 spin_unlock(&space_info->lock);
1375 finish_wait(&ticket->wait, &wait);
1376 spin_lock(&space_info->lock);
1378 spin_unlock(&space_info->lock);
1382 * Do the appropriate flushing and waiting for a ticket
1384 * @fs_info: the filesystem
1385 * @space_info: space info for the reservation
1386 * @ticket: ticket for the reservation
1387 * @start_ns: timestamp when the reservation started
1388 * @orig_bytes: amount of bytes originally reserved
1389 * @flush: how much we can flush
1391 * This does the work of figuring out how to flush for the ticket, waiting for
1392 * the reservation, and returning the appropriate error if there is one.
1394 static int handle_reserve_ticket(struct btrfs_fs_info *fs_info,
1395 struct btrfs_space_info *space_info,
1396 struct reserve_ticket *ticket,
1397 u64 start_ns, u64 orig_bytes,
1398 enum btrfs_reserve_flush_enum flush)
1403 case BTRFS_RESERVE_FLUSH_DATA:
1404 case BTRFS_RESERVE_FLUSH_ALL:
1405 case BTRFS_RESERVE_FLUSH_ALL_STEAL:
1406 wait_reserve_ticket(fs_info, space_info, ticket);
1408 case BTRFS_RESERVE_FLUSH_LIMIT:
1409 priority_reclaim_metadata_space(fs_info, space_info, ticket,
1410 priority_flush_states,
1411 ARRAY_SIZE(priority_flush_states));
1413 case BTRFS_RESERVE_FLUSH_EVICT:
1414 priority_reclaim_metadata_space(fs_info, space_info, ticket,
1416 ARRAY_SIZE(evict_flush_states));
1418 case BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE:
1419 priority_reclaim_data_space(fs_info, space_info, ticket);
1426 ret = ticket->error;
1427 ASSERT(list_empty(&ticket->list));
1429 * Check that we can't have an error set if the reservation succeeded,
1430 * as that would confuse tasks and lead them to error out without
1431 * releasing reserved space (if an error happens the expectation is that
1432 * space wasn't reserved at all).
1434 ASSERT(!(ticket->bytes == 0 && ticket->error));
1435 trace_btrfs_reserve_ticket(fs_info, space_info->flags, orig_bytes,
1436 start_ns, flush, ticket->error);
1441 * This returns true if this flush state will go through the ordinary flushing
1444 static inline bool is_normal_flushing(enum btrfs_reserve_flush_enum flush)
1446 return (flush == BTRFS_RESERVE_FLUSH_ALL) ||
1447 (flush == BTRFS_RESERVE_FLUSH_ALL_STEAL);
1450 static inline void maybe_clamp_preempt(struct btrfs_fs_info *fs_info,
1451 struct btrfs_space_info *space_info)
1453 u64 ordered = percpu_counter_sum_positive(&fs_info->ordered_bytes);
1454 u64 delalloc = percpu_counter_sum_positive(&fs_info->delalloc_bytes);
1457 * If we're heavy on ordered operations then clamping won't help us. We
1458 * need to clamp specifically to keep up with dirty'ing buffered
1459 * writers, because there's not a 1:1 correlation of writing delalloc
1460 * and freeing space, like there is with flushing delayed refs or
1461 * delayed nodes. If we're already more ordered than delalloc then
1462 * we're keeping up, otherwise we aren't and should probably clamp.
1464 if (ordered < delalloc)
1465 space_info->clamp = min(space_info->clamp + 1, 8);
1468 static inline bool can_steal(enum btrfs_reserve_flush_enum flush)
1470 return (flush == BTRFS_RESERVE_FLUSH_ALL_STEAL ||
1471 flush == BTRFS_RESERVE_FLUSH_EVICT);
1475 * Try to reserve bytes from the block_rsv's space
1477 * @fs_info: the filesystem
1478 * @space_info: space info we want to allocate from
1479 * @orig_bytes: number of bytes we want
1480 * @flush: whether or not we can flush to make our reservation
1482 * This will reserve orig_bytes number of bytes from the space info associated
1483 * with the block_rsv. If there is not enough space it will make an attempt to
1484 * flush out space to make room. It will do this by flushing delalloc if
1485 * possible or committing the transaction. If flush is 0 then no attempts to
1486 * regain reservations will be made and this will fail if there is not enough
1489 static int __reserve_bytes(struct btrfs_fs_info *fs_info,
1490 struct btrfs_space_info *space_info, u64 orig_bytes,
1491 enum btrfs_reserve_flush_enum flush)
1493 struct work_struct *async_work;
1494 struct reserve_ticket ticket;
1498 bool pending_tickets;
1501 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
1503 if (flush == BTRFS_RESERVE_FLUSH_DATA)
1504 async_work = &fs_info->async_data_reclaim_work;
1506 async_work = &fs_info->async_reclaim_work;
1508 spin_lock(&space_info->lock);
1510 used = btrfs_space_info_used(space_info, true);
1513 * We don't want NO_FLUSH allocations to jump everybody, they can
1514 * generally handle ENOSPC in a different way, so treat them the same as
1515 * normal flushers when it comes to skipping pending tickets.
1517 if (is_normal_flushing(flush) || (flush == BTRFS_RESERVE_NO_FLUSH))
1518 pending_tickets = !list_empty(&space_info->tickets) ||
1519 !list_empty(&space_info->priority_tickets);
1521 pending_tickets = !list_empty(&space_info->priority_tickets);
1524 * Carry on if we have enough space (short-circuit) OR call
1525 * can_overcommit() to ensure we can overcommit to continue.
1527 if (!pending_tickets &&
1528 ((used + orig_bytes <= space_info->total_bytes) ||
1529 btrfs_can_overcommit(fs_info, space_info, orig_bytes, flush))) {
1530 btrfs_space_info_update_bytes_may_use(fs_info, space_info,
1536 * If we couldn't make a reservation then setup our reservation ticket
1537 * and kick the async worker if it's not already running.
1539 * If we are a priority flusher then we just need to add our ticket to
1540 * the list and we will do our own flushing further down.
1542 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
1543 ticket.bytes = orig_bytes;
1545 space_info->reclaim_size += ticket.bytes;
1546 init_waitqueue_head(&ticket.wait);
1547 ticket.steal = can_steal(flush);
1548 if (trace_btrfs_reserve_ticket_enabled())
1549 start_ns = ktime_get_ns();
1551 if (flush == BTRFS_RESERVE_FLUSH_ALL ||
1552 flush == BTRFS_RESERVE_FLUSH_ALL_STEAL ||
1553 flush == BTRFS_RESERVE_FLUSH_DATA) {
1554 list_add_tail(&ticket.list, &space_info->tickets);
1555 if (!space_info->flush) {
1557 * We were forced to add a reserve ticket, so
1558 * our preemptive flushing is unable to keep
1559 * up. Clamp down on the threshold for the
1560 * preemptive flushing in order to keep up with
1563 maybe_clamp_preempt(fs_info, space_info);
1565 space_info->flush = 1;
1566 trace_btrfs_trigger_flush(fs_info,
1570 queue_work(system_unbound_wq, async_work);
1573 list_add_tail(&ticket.list,
1574 &space_info->priority_tickets);
1576 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
1579 * We will do the space reservation dance during log replay,
1580 * which means we won't have fs_info->fs_root set, so don't do
1581 * the async reclaim as we will panic.
1583 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
1584 !work_busy(&fs_info->preempt_reclaim_work) &&
1585 need_preemptive_reclaim(fs_info, space_info)) {
1586 trace_btrfs_trigger_flush(fs_info, space_info->flags,
1587 orig_bytes, flush, "preempt");
1588 queue_work(system_unbound_wq,
1589 &fs_info->preempt_reclaim_work);
1592 spin_unlock(&space_info->lock);
1593 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
1596 return handle_reserve_ticket(fs_info, space_info, &ticket, start_ns,
1601 * Trye to reserve metadata bytes from the block_rsv's space
1603 * @fs_info: the filesystem
1604 * @block_rsv: block_rsv we're allocating for
1605 * @orig_bytes: number of bytes we want
1606 * @flush: whether or not we can flush to make our reservation
1608 * This will reserve orig_bytes number of bytes from the space info associated
1609 * with the block_rsv. If there is not enough space it will make an attempt to
1610 * flush out space to make room. It will do this by flushing delalloc if
1611 * possible or committing the transaction. If flush is 0 then no attempts to
1612 * regain reservations will be made and this will fail if there is not enough
1615 int btrfs_reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
1616 struct btrfs_block_rsv *block_rsv,
1618 enum btrfs_reserve_flush_enum flush)
1622 ret = __reserve_bytes(fs_info, block_rsv->space_info, orig_bytes, flush);
1623 if (ret == -ENOSPC) {
1624 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
1625 block_rsv->space_info->flags,
1628 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
1629 btrfs_dump_space_info(fs_info, block_rsv->space_info,
1636 * Try to reserve data bytes for an allocation
1638 * @fs_info: the filesystem
1639 * @bytes: number of bytes we need
1640 * @flush: how we are allowed to flush
1642 * This will reserve bytes from the data space info. If there is not enough
1643 * space then we will attempt to flush space as specified by flush.
1645 int btrfs_reserve_data_bytes(struct btrfs_fs_info *fs_info, u64 bytes,
1646 enum btrfs_reserve_flush_enum flush)
1648 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
1651 ASSERT(flush == BTRFS_RESERVE_FLUSH_DATA ||
1652 flush == BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE);
1653 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_DATA);
1655 ret = __reserve_bytes(fs_info, data_sinfo, bytes, flush);
1656 if (ret == -ENOSPC) {
1657 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
1658 data_sinfo->flags, bytes, 1);
1659 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
1660 btrfs_dump_space_info(fs_info, data_sinfo, bytes, 0);