1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright 2012 Google, Inc.
5 * Foreground allocator code: allocate buckets from freelist, and allocate in
6 * sector granularity from writepoints.
8 * bch2_bucket_alloc() allocates a single bucket from a specific device.
10 * bch2_bucket_alloc_set() allocates one or more buckets from different devices
11 * in a given filesystem.
15 #include "alloc_background.h"
16 #include "alloc_foreground.h"
17 #include "btree_iter.h"
18 #include "btree_update.h"
21 #include "buckets_waiting_for_journal.h"
24 #include "disk_groups.h"
30 #include <linux/math64.h>
31 #include <linux/rculist.h>
32 #include <linux/rcupdate.h>
33 #include <trace/events/bcachefs.h>
35 const char * const bch2_alloc_reserves[] = {
43 * Open buckets represent a bucket that's currently being allocated from. They
46 * - They track buckets that have been partially allocated, allowing for
47 * sub-bucket sized allocations - they're used by the sector allocator below
49 * - They provide a reference to the buckets they own that mark and sweep GC
50 * can find, until the new allocation has a pointer to it inserted into the
53 * When allocating some space with the sector allocator, the allocation comes
54 * with a reference to an open bucket - the caller is required to put that
55 * reference _after_ doing the index update that makes its allocation reachable.
58 static void bch2_open_bucket_hash_add(struct bch_fs *c, struct open_bucket *ob)
60 open_bucket_idx_t idx = ob - c->open_buckets;
61 open_bucket_idx_t *slot = open_bucket_hashslot(c, ob->dev, ob->bucket);
67 static void bch2_open_bucket_hash_remove(struct bch_fs *c, struct open_bucket *ob)
69 open_bucket_idx_t idx = ob - c->open_buckets;
70 open_bucket_idx_t *slot = open_bucket_hashslot(c, ob->dev, ob->bucket);
72 while (*slot != idx) {
74 slot = &c->open_buckets[*slot].hash;
81 void __bch2_open_bucket_put(struct bch_fs *c, struct open_bucket *ob)
83 struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
86 bch2_ec_bucket_written(c, ob);
90 percpu_down_read(&c->mark_lock);
96 spin_unlock(&ob->lock);
97 percpu_up_read(&c->mark_lock);
99 spin_lock(&c->freelist_lock);
100 bch2_open_bucket_hash_remove(c, ob);
102 ob->freelist = c->open_buckets_freelist;
103 c->open_buckets_freelist = ob - c->open_buckets;
105 c->open_buckets_nr_free++;
106 ca->nr_open_buckets--;
107 spin_unlock(&c->freelist_lock);
109 closure_wake_up(&c->open_buckets_wait);
112 void bch2_open_bucket_write_error(struct bch_fs *c,
113 struct open_buckets *obs,
116 struct open_bucket *ob;
119 open_bucket_for_each(c, obs, ob, i)
120 if (ob->dev == dev && ob->ec)
121 bch2_ec_bucket_cancel(c, ob);
124 static struct open_bucket *bch2_open_bucket_alloc(struct bch_fs *c)
126 struct open_bucket *ob;
128 BUG_ON(!c->open_buckets_freelist || !c->open_buckets_nr_free);
130 ob = c->open_buckets + c->open_buckets_freelist;
131 c->open_buckets_freelist = ob->freelist;
132 atomic_set(&ob->pin, 1);
135 c->open_buckets_nr_free--;
139 static void open_bucket_free_unused(struct bch_fs *c,
140 struct write_point *wp,
141 struct open_bucket *ob)
143 struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
144 bool may_realloc = wp->data_type == BCH_DATA_user;
146 BUG_ON(ca->open_buckets_partial_nr >
147 ARRAY_SIZE(ca->open_buckets_partial));
149 if (ca->open_buckets_partial_nr <
150 ARRAY_SIZE(ca->open_buckets_partial) &&
152 spin_lock(&c->freelist_lock);
153 ob->on_partial_list = true;
154 ca->open_buckets_partial[ca->open_buckets_partial_nr++] =
155 ob - c->open_buckets;
156 spin_unlock(&c->freelist_lock);
158 closure_wake_up(&c->open_buckets_wait);
159 closure_wake_up(&c->freelist_wait);
161 bch2_open_bucket_put(c, ob);
165 /* _only_ for allocating the journal on a new device: */
166 long bch2_bucket_alloc_new_fs(struct bch_dev *ca)
168 while (ca->new_fs_bucket_idx < ca->mi.nbuckets) {
169 u64 b = ca->new_fs_bucket_idx++;
171 if (!is_superblock_bucket(ca, b) &&
172 (!ca->buckets_nouse || !test_bit(b, ca->buckets_nouse)))
179 static inline unsigned open_buckets_reserved(enum alloc_reserve reserve)
183 case RESERVE_btree_movinggc:
185 case RESERVE_movinggc:
186 return OPEN_BUCKETS_COUNT / 4;
188 return OPEN_BUCKETS_COUNT / 2;
192 static struct open_bucket *__try_alloc_bucket(struct bch_fs *c, struct bch_dev *ca,
194 enum alloc_reserve reserve,
195 struct bch_alloc_v4 *a,
197 u64 *skipped_need_journal_commit,
201 struct open_bucket *ob;
203 if (unlikely(ca->buckets_nouse && test_bit(bucket, ca->buckets_nouse))) {
208 if (bch2_bucket_is_open(c, ca->dev_idx, bucket)) {
213 if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
214 c->journal.flushed_seq_ondisk, ca->dev_idx, bucket)) {
215 (*skipped_need_journal_commit)++;
219 spin_lock(&c->freelist_lock);
221 if (unlikely(c->open_buckets_nr_free <= open_buckets_reserved(reserve))) {
223 closure_wait(&c->open_buckets_wait, cl);
225 if (!c->blocked_allocate_open_bucket)
226 c->blocked_allocate_open_bucket = local_clock();
228 spin_unlock(&c->freelist_lock);
229 return ERR_PTR(-OPEN_BUCKETS_EMPTY);
232 /* Recheck under lock: */
233 if (bch2_bucket_is_open(c, ca->dev_idx, bucket)) {
234 spin_unlock(&c->freelist_lock);
239 ob = bch2_open_bucket_alloc(c);
241 spin_lock(&ob->lock);
244 ob->sectors_free = ca->mi.bucket_size;
245 ob->alloc_reserve = reserve;
246 ob->dev = ca->dev_idx;
249 spin_unlock(&ob->lock);
251 ca->nr_open_buckets++;
252 bch2_open_bucket_hash_add(c, ob);
254 if (c->blocked_allocate_open_bucket) {
255 bch2_time_stats_update(
256 &c->times[BCH_TIME_blocked_allocate_open_bucket],
257 c->blocked_allocate_open_bucket);
258 c->blocked_allocate_open_bucket = 0;
261 if (c->blocked_allocate) {
262 bch2_time_stats_update(
263 &c->times[BCH_TIME_blocked_allocate],
264 c->blocked_allocate);
265 c->blocked_allocate = 0;
268 spin_unlock(&c->freelist_lock);
270 trace_bucket_alloc(ca, bch2_alloc_reserves[reserve]);
274 static struct open_bucket *try_alloc_bucket(struct btree_trans *trans, struct bch_dev *ca,
275 enum alloc_reserve reserve, u64 free_entry,
277 u64 *skipped_need_journal_commit,
281 struct bch_fs *c = trans->c;
282 struct btree_iter iter;
284 struct open_bucket *ob;
285 struct bch_alloc_v4 a;
286 u64 b = free_entry & ~(~0ULL << 56);
287 unsigned genbits = free_entry >> 56;
288 struct printbuf buf = PRINTBUF;
291 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS(ca->dev_idx, b), BTREE_ITER_CACHED);
292 k = bch2_btree_iter_peek_slot(&iter);
299 bch2_alloc_to_v4(k, &a);
301 if (bch2_fs_inconsistent_on(a.data_type != BCH_DATA_free, c,
302 "non free bucket in freespace btree (state %s)\n"
304 " at %llu (genbits %u)",
305 bch2_data_types[a.data_type],
306 (bch2_bkey_val_to_text(&buf, c, k), buf.buf),
307 free_entry, genbits)) {
312 if (bch2_fs_inconsistent_on(genbits != (alloc_freespace_genbits(a) >> 56), c,
313 "bucket in freespace btree with wrong genbits (got %u should be %llu)\n"
315 genbits, alloc_freespace_genbits(a) >> 56,
316 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
321 if (bch2_fs_inconsistent_on(b < ca->mi.first_bucket || b >= ca->mi.nbuckets, c,
322 "freespace btree has bucket outside allowed range (got %llu, valid %u-%llu)",
323 b, ca->mi.first_bucket, ca->mi.nbuckets)) {
328 ob = __try_alloc_bucket(c, ca, b, reserve, &a,
330 skipped_need_journal_commit,
334 bch2_trans_iter_exit(trans, &iter);
339 static struct open_bucket *try_alloc_partial_bucket(struct bch_fs *c, struct bch_dev *ca,
340 enum alloc_reserve reserve)
342 struct open_bucket *ob;
345 spin_lock(&c->freelist_lock);
347 for (i = ca->open_buckets_partial_nr - 1; i >= 0; --i) {
348 ob = c->open_buckets + ca->open_buckets_partial[i];
350 if (reserve <= ob->alloc_reserve) {
351 array_remove_item(ca->open_buckets_partial,
352 ca->open_buckets_partial_nr,
354 ob->on_partial_list = false;
355 ob->alloc_reserve = reserve;
356 spin_unlock(&c->freelist_lock);
361 spin_unlock(&c->freelist_lock);
366 * This path is for before the freespace btree is initialized:
368 * If ca->new_fs_bucket_idx is nonzero, we haven't yet marked superblock &
369 * journal buckets - journal buckets will be < ca->new_fs_bucket_idx
371 static noinline struct open_bucket *
372 bch2_bucket_alloc_trans_early(struct btree_trans *trans,
374 enum alloc_reserve reserve,
378 u64 *skipped_need_journal_commit,
382 struct btree_iter iter;
384 struct open_bucket *ob = NULL;
387 *cur_bucket = max_t(u64, *cur_bucket, ca->mi.first_bucket);
388 *cur_bucket = max_t(u64, *cur_bucket, ca->new_fs_bucket_idx);
390 for_each_btree_key(trans, iter, BTREE_ID_alloc, POS(ca->dev_idx, *cur_bucket),
391 BTREE_ITER_SLOTS, k, ret) {
392 struct bch_alloc_v4 a;
394 if (bkey_cmp(k.k->p, POS(ca->dev_idx, ca->mi.nbuckets)) >= 0)
397 if (ca->new_fs_bucket_idx &&
398 is_superblock_bucket(ca, k.k->p.offset))
401 bch2_alloc_to_v4(k, &a);
403 if (a.data_type != BCH_DATA_free)
408 ob = __try_alloc_bucket(trans->c, ca, k.k->p.offset, reserve, &a,
410 skipped_need_journal_commit,
416 bch2_trans_iter_exit(trans, &iter);
418 *cur_bucket = iter.pos.offset;
420 return ob ?: ERR_PTR(ret ?: -FREELIST_EMPTY);
423 static struct open_bucket *bch2_bucket_alloc_trans(struct btree_trans *trans,
425 enum alloc_reserve reserve,
429 u64 *skipped_need_journal_commit,
433 struct btree_iter iter;
435 struct open_bucket *ob = NULL;
438 if (unlikely(!ca->mi.freespace_initialized))
439 return bch2_bucket_alloc_trans_early(trans, ca, reserve,
443 skipped_need_journal_commit,
447 BUG_ON(ca->new_fs_bucket_idx);
449 for_each_btree_key(trans, iter, BTREE_ID_freespace,
450 POS(ca->dev_idx, *cur_bucket), 0, k, ret) {
451 if (k.k->p.inode != ca->dev_idx)
454 for (*cur_bucket = max(*cur_bucket, bkey_start_offset(k.k));
455 *cur_bucket != k.k->p.offset && !ob;
457 if (btree_trans_too_many_iters(trans)) {
458 ob = ERR_PTR(-EINTR);
464 ob = try_alloc_bucket(trans, ca, reserve,
467 skipped_need_journal_commit,
474 bch2_trans_iter_exit(trans, &iter);
476 return ob ?: ERR_PTR(ret);
480 * bch_bucket_alloc - allocate a single bucket from a specific device
482 * Returns index of bucket on success, 0 on failure
484 struct open_bucket *bch2_bucket_alloc(struct bch_fs *c, struct bch_dev *ca,
485 enum alloc_reserve reserve,
486 bool may_alloc_partial,
489 struct open_bucket *ob = NULL;
490 struct bch_dev_usage usage;
493 u64 buckets_seen = 0;
494 u64 skipped_open = 0;
495 u64 skipped_need_journal_commit = 0;
496 u64 skipped_nouse = 0;
497 bool waiting = false;
500 usage = bch2_dev_usage_read(ca);
501 avail = __dev_buckets_available(ca, usage,reserve);
503 if (usage.d[BCH_DATA_need_discard].buckets > avail)
506 if (usage.d[BCH_DATA_need_gc_gens].buckets > avail)
509 if (should_invalidate_buckets(ca, usage))
510 bch2_do_invalidates(c);
513 if (cl && !waiting) {
514 closure_wait(&c->freelist_wait, cl);
519 if (!c->blocked_allocate)
520 c->blocked_allocate = local_clock();
522 ob = ERR_PTR(-FREELIST_EMPTY);
527 closure_wake_up(&c->freelist_wait);
529 if (may_alloc_partial) {
530 ob = try_alloc_partial_bucket(c, ca, reserve);
535 ret = bch2_trans_do(c, NULL, NULL, 0,
536 PTR_ERR_OR_ZERO(ob = bch2_bucket_alloc_trans(&trans, ca, reserve,
540 &skipped_need_journal_commit,
544 if (skipped_need_journal_commit * 2 > avail)
545 bch2_journal_flush_async(&c->journal, NULL);
548 ob = ERR_PTR(ret ?: -FREELIST_EMPTY);
551 trace_bucket_alloc_fail(ca, bch2_alloc_reserves[reserve], avail,
554 skipped_need_journal_commit,
556 cl == NULL, PTR_ERR(ob));
557 atomic_long_inc(&c->bucket_alloc_fail);
563 static int __dev_stripe_cmp(struct dev_stripe_state *stripe,
564 unsigned l, unsigned r)
566 return ((stripe->next_alloc[l] > stripe->next_alloc[r]) -
567 (stripe->next_alloc[l] < stripe->next_alloc[r]));
570 #define dev_stripe_cmp(l, r) __dev_stripe_cmp(stripe, l, r)
572 struct dev_alloc_list bch2_dev_alloc_list(struct bch_fs *c,
573 struct dev_stripe_state *stripe,
574 struct bch_devs_mask *devs)
576 struct dev_alloc_list ret = { .nr = 0 };
579 for_each_set_bit(i, devs->d, BCH_SB_MEMBERS_MAX)
580 ret.devs[ret.nr++] = i;
582 bubble_sort(ret.devs, ret.nr, dev_stripe_cmp);
586 void bch2_dev_stripe_increment(struct bch_dev *ca,
587 struct dev_stripe_state *stripe)
589 u64 *v = stripe->next_alloc + ca->dev_idx;
590 u64 free_space = dev_buckets_available(ca, RESERVE_none);
591 u64 free_space_inv = free_space
592 ? div64_u64(1ULL << 48, free_space)
596 if (*v + free_space_inv >= *v)
597 *v += free_space_inv;
601 for (v = stripe->next_alloc;
602 v < stripe->next_alloc + ARRAY_SIZE(stripe->next_alloc); v++)
603 *v = *v < scale ? 0 : *v - scale;
606 #define BUCKET_MAY_ALLOC_PARTIAL (1 << 0)
607 #define BUCKET_ALLOC_USE_DURABILITY (1 << 1)
609 static void add_new_bucket(struct bch_fs *c,
610 struct open_buckets *ptrs,
611 struct bch_devs_mask *devs_may_alloc,
612 unsigned *nr_effective,
615 struct open_bucket *ob)
617 unsigned durability =
618 bch_dev_bkey_exists(c, ob->dev)->mi.durability;
620 __clear_bit(ob->dev, devs_may_alloc->d);
621 *nr_effective += (flags & BUCKET_ALLOC_USE_DURABILITY)
623 *have_cache |= !durability;
625 ob_push(c, ptrs, ob);
628 int bch2_bucket_alloc_set(struct bch_fs *c,
629 struct open_buckets *ptrs,
630 struct dev_stripe_state *stripe,
631 struct bch_devs_mask *devs_may_alloc,
632 unsigned nr_replicas,
633 unsigned *nr_effective,
635 enum alloc_reserve reserve,
639 struct dev_alloc_list devs_sorted =
640 bch2_dev_alloc_list(c, stripe, devs_may_alloc);
643 int ret = -INSUFFICIENT_DEVICES;
646 BUG_ON(*nr_effective >= nr_replicas);
648 for (i = 0; i < devs_sorted.nr; i++) {
649 struct open_bucket *ob;
651 dev = devs_sorted.devs[i];
654 ca = rcu_dereference(c->devs[dev]);
656 percpu_ref_get(&ca->ref);
662 if (!ca->mi.durability && *have_cache) {
663 percpu_ref_put(&ca->ref);
667 ob = bch2_bucket_alloc(c, ca, reserve,
668 flags & BUCKET_MAY_ALLOC_PARTIAL, cl);
670 bch2_dev_stripe_increment(ca, stripe);
671 percpu_ref_put(&ca->ref);
681 add_new_bucket(c, ptrs, devs_may_alloc,
682 nr_effective, have_cache, flags, ob);
684 if (*nr_effective >= nr_replicas) {
693 /* Allocate from stripes: */
696 * if we can't allocate a new stripe because there are already too many
697 * partially filled stripes, force allocating from an existing stripe even when
698 * it's to a device we don't want:
701 static int bucket_alloc_from_stripe(struct bch_fs *c,
702 struct open_buckets *ptrs,
703 struct write_point *wp,
704 struct bch_devs_mask *devs_may_alloc,
706 unsigned erasure_code,
707 unsigned nr_replicas,
708 unsigned *nr_effective,
713 struct dev_alloc_list devs_sorted;
714 struct ec_stripe_head *h;
715 struct open_bucket *ob;
725 if (ec_open_bucket(c, ptrs))
728 h = bch2_ec_stripe_head_get(c, target, 0, nr_replicas - 1,
729 wp == &c->copygc_write_point,
736 devs_sorted = bch2_dev_alloc_list(c, &wp->stripe, devs_may_alloc);
738 for (i = 0; i < devs_sorted.nr; i++)
739 for (ec_idx = 0; ec_idx < h->s->nr_data; ec_idx++) {
740 if (!h->s->blocks[ec_idx])
743 ob = c->open_buckets + h->s->blocks[ec_idx];
744 if (ob->dev == devs_sorted.devs[i] &&
745 !test_and_set_bit(ec_idx, h->s->blocks_allocated))
750 ca = bch_dev_bkey_exists(c, ob->dev);
755 add_new_bucket(c, ptrs, devs_may_alloc,
756 nr_effective, have_cache, flags, ob);
757 atomic_inc(&h->s->pin);
759 bch2_ec_stripe_head_put(c, h);
763 /* Sector allocator */
765 static void get_buckets_from_writepoint(struct bch_fs *c,
766 struct open_buckets *ptrs,
767 struct write_point *wp,
768 struct bch_devs_mask *devs_may_alloc,
769 unsigned nr_replicas,
770 unsigned *nr_effective,
775 struct open_buckets ptrs_skip = { .nr = 0 };
776 struct open_bucket *ob;
779 open_bucket_for_each(c, &wp->ptrs, ob, i) {
780 struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
782 if (*nr_effective < nr_replicas &&
783 test_bit(ob->dev, devs_may_alloc->d) &&
784 (ca->mi.durability ||
785 (wp->data_type == BCH_DATA_user && !*have_cache)) &&
786 (ob->ec || !need_ec)) {
787 add_new_bucket(c, ptrs, devs_may_alloc,
788 nr_effective, have_cache,
791 ob_push(c, &ptrs_skip, ob);
794 wp->ptrs = ptrs_skip;
797 static int open_bucket_add_buckets(struct bch_fs *c,
798 struct open_buckets *ptrs,
799 struct write_point *wp,
800 struct bch_devs_list *devs_have,
802 unsigned erasure_code,
803 unsigned nr_replicas,
804 unsigned *nr_effective,
806 enum alloc_reserve reserve,
810 struct bch_devs_mask devs;
811 struct open_bucket *ob;
812 struct closure *cl = NULL;
817 devs = target_rw_devs(c, wp->data_type, target);
820 /* Don't allocate from devices we already have pointers to: */
821 for (i = 0; i < devs_have->nr; i++)
822 __clear_bit(devs_have->devs[i], devs.d);
824 open_bucket_for_each(c, ptrs, ob, i)
825 __clear_bit(ob->dev, devs.d);
828 if (!ec_open_bucket(c, ptrs)) {
829 get_buckets_from_writepoint(c, ptrs, wp, &devs,
830 nr_replicas, nr_effective,
831 have_cache, flags, true);
832 if (*nr_effective >= nr_replicas)
836 if (!ec_open_bucket(c, ptrs)) {
837 ret = bucket_alloc_from_stripe(c, ptrs, wp, &devs,
838 target, erasure_code,
839 nr_replicas, nr_effective,
840 have_cache, flags, _cl);
841 if (ret == -FREELIST_EMPTY ||
842 ret == -OPEN_BUCKETS_EMPTY)
844 if (*nr_effective >= nr_replicas)
849 get_buckets_from_writepoint(c, ptrs, wp, &devs,
850 nr_replicas, nr_effective,
851 have_cache, flags, false);
852 if (*nr_effective >= nr_replicas)
857 * Try nonblocking first, so that if one device is full we'll try from
860 ret = bch2_bucket_alloc_set(c, ptrs, &wp->stripe, &devs,
861 nr_replicas, nr_effective, have_cache,
863 if (ret && ret != -INSUFFICIENT_DEVICES && !cl && _cl) {
871 void bch2_open_buckets_stop_dev(struct bch_fs *c, struct bch_dev *ca,
872 struct open_buckets *obs)
874 struct open_buckets ptrs = { .nr = 0 };
875 struct open_bucket *ob, *ob2;
878 open_bucket_for_each(c, obs, ob, i) {
879 bool drop = !ca || ob->dev == ca->dev_idx;
881 if (!drop && ob->ec) {
882 mutex_lock(&ob->ec->lock);
883 for (j = 0; j < ob->ec->new_stripe.key.v.nr_blocks; j++) {
884 if (!ob->ec->blocks[j])
887 ob2 = c->open_buckets + ob->ec->blocks[j];
888 drop |= ob2->dev == ca->dev_idx;
890 mutex_unlock(&ob->ec->lock);
894 bch2_open_bucket_put(c, ob);
896 ob_push(c, &ptrs, ob);
902 void bch2_writepoint_stop(struct bch_fs *c, struct bch_dev *ca,
903 struct write_point *wp)
905 mutex_lock(&wp->lock);
906 bch2_open_buckets_stop_dev(c, ca, &wp->ptrs);
907 mutex_unlock(&wp->lock);
910 static inline struct hlist_head *writepoint_hash(struct bch_fs *c,
911 unsigned long write_point)
914 hash_long(write_point, ilog2(ARRAY_SIZE(c->write_points_hash)));
916 return &c->write_points_hash[hash];
919 static struct write_point *__writepoint_find(struct hlist_head *head,
920 unsigned long write_point)
922 struct write_point *wp;
925 hlist_for_each_entry_rcu(wp, head, node)
926 if (wp->write_point == write_point)
934 static inline bool too_many_writepoints(struct bch_fs *c, unsigned factor)
936 u64 stranded = c->write_points_nr * c->bucket_size_max;
937 u64 free = bch2_fs_usage_read_short(c).free;
939 return stranded * factor > free;
942 static bool try_increase_writepoints(struct bch_fs *c)
944 struct write_point *wp;
946 if (c->write_points_nr == ARRAY_SIZE(c->write_points) ||
947 too_many_writepoints(c, 32))
950 wp = c->write_points + c->write_points_nr++;
951 hlist_add_head_rcu(&wp->node, writepoint_hash(c, wp->write_point));
955 static bool try_decrease_writepoints(struct bch_fs *c,
958 struct write_point *wp;
960 mutex_lock(&c->write_points_hash_lock);
961 if (c->write_points_nr < old_nr) {
962 mutex_unlock(&c->write_points_hash_lock);
966 if (c->write_points_nr == 1 ||
967 !too_many_writepoints(c, 8)) {
968 mutex_unlock(&c->write_points_hash_lock);
972 wp = c->write_points + --c->write_points_nr;
974 hlist_del_rcu(&wp->node);
975 mutex_unlock(&c->write_points_hash_lock);
977 bch2_writepoint_stop(c, NULL, wp);
981 static struct write_point *writepoint_find(struct bch_fs *c,
982 unsigned long write_point)
984 struct write_point *wp, *oldest;
985 struct hlist_head *head;
987 if (!(write_point & 1UL)) {
988 wp = (struct write_point *) write_point;
989 mutex_lock(&wp->lock);
993 head = writepoint_hash(c, write_point);
995 wp = __writepoint_find(head, write_point);
998 mutex_lock(&wp->lock);
999 if (wp->write_point == write_point)
1001 mutex_unlock(&wp->lock);
1004 restart_find_oldest:
1006 for (wp = c->write_points;
1007 wp < c->write_points + c->write_points_nr; wp++)
1008 if (!oldest || time_before64(wp->last_used, oldest->last_used))
1011 mutex_lock(&oldest->lock);
1012 mutex_lock(&c->write_points_hash_lock);
1013 if (oldest >= c->write_points + c->write_points_nr ||
1014 try_increase_writepoints(c)) {
1015 mutex_unlock(&c->write_points_hash_lock);
1016 mutex_unlock(&oldest->lock);
1017 goto restart_find_oldest;
1020 wp = __writepoint_find(head, write_point);
1021 if (wp && wp != oldest) {
1022 mutex_unlock(&c->write_points_hash_lock);
1023 mutex_unlock(&oldest->lock);
1028 hlist_del_rcu(&wp->node);
1029 wp->write_point = write_point;
1030 hlist_add_head_rcu(&wp->node, head);
1031 mutex_unlock(&c->write_points_hash_lock);
1033 wp->last_used = sched_clock();
1038 * Get us an open_bucket we can allocate from, return with it locked:
1040 struct write_point *bch2_alloc_sectors_start(struct bch_fs *c,
1042 unsigned erasure_code,
1043 struct write_point_specifier write_point,
1044 struct bch_devs_list *devs_have,
1045 unsigned nr_replicas,
1046 unsigned nr_replicas_required,
1047 enum alloc_reserve reserve,
1051 struct write_point *wp;
1052 struct open_bucket *ob;
1053 struct open_buckets ptrs;
1054 unsigned nr_effective, write_points_nr;
1055 unsigned ob_flags = 0;
1060 if (!(flags & BCH_WRITE_ONLY_SPECIFIED_DEVS))
1061 ob_flags |= BUCKET_ALLOC_USE_DURABILITY;
1063 BUG_ON(!nr_replicas || !nr_replicas_required);
1067 write_points_nr = c->write_points_nr;
1070 wp = writepoint_find(c, write_point.v);
1072 if (wp->data_type == BCH_DATA_user)
1073 ob_flags |= BUCKET_MAY_ALLOC_PARTIAL;
1075 /* metadata may not allocate on cache devices: */
1076 if (wp->data_type != BCH_DATA_user)
1079 if (!target || (flags & BCH_WRITE_ONLY_SPECIFIED_DEVS)) {
1080 ret = open_bucket_add_buckets(c, &ptrs, wp, devs_have,
1081 target, erasure_code,
1082 nr_replicas, &nr_effective,
1083 &have_cache, reserve,
1086 ret = open_bucket_add_buckets(c, &ptrs, wp, devs_have,
1087 target, erasure_code,
1088 nr_replicas, &nr_effective,
1089 &have_cache, reserve,
1094 ret = open_bucket_add_buckets(c, &ptrs, wp, devs_have,
1096 nr_replicas, &nr_effective,
1097 &have_cache, reserve,
1101 BUG_ON(!ret && nr_effective < nr_replicas);
1103 if (erasure_code && !ec_open_bucket(c, &ptrs))
1104 pr_debug("failed to get ec bucket: ret %u", ret);
1106 if (ret == -INSUFFICIENT_DEVICES &&
1107 nr_effective >= nr_replicas_required)
1113 /* Free buckets we didn't use: */
1114 open_bucket_for_each(c, &wp->ptrs, ob, i)
1115 open_bucket_free_unused(c, wp, ob);
1119 wp->sectors_free = UINT_MAX;
1121 open_bucket_for_each(c, &wp->ptrs, ob, i)
1122 wp->sectors_free = min(wp->sectors_free, ob->sectors_free);
1124 BUG_ON(!wp->sectors_free || wp->sectors_free == UINT_MAX);
1128 open_bucket_for_each(c, &wp->ptrs, ob, i)
1129 if (ptrs.nr < ARRAY_SIZE(ptrs.v))
1130 ob_push(c, &ptrs, ob);
1132 open_bucket_free_unused(c, wp, ob);
1135 mutex_unlock(&wp->lock);
1137 if (ret == -FREELIST_EMPTY &&
1138 try_decrease_writepoints(c, write_points_nr))
1142 case -OPEN_BUCKETS_EMPTY:
1143 case -FREELIST_EMPTY:
1144 return cl ? ERR_PTR(-EAGAIN) : ERR_PTR(-ENOSPC);
1145 case -INSUFFICIENT_DEVICES:
1146 return ERR_PTR(-EROFS);
1148 return ERR_PTR(ret);
1152 struct bch_extent_ptr bch2_ob_ptr(struct bch_fs *c, struct open_bucket *ob)
1154 struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
1156 return (struct bch_extent_ptr) {
1157 .type = 1 << BCH_EXTENT_ENTRY_ptr,
1160 .offset = bucket_to_sector(ca, ob->bucket) +
1161 ca->mi.bucket_size -
1167 * Append pointers to the space we just allocated to @k, and mark @sectors space
1168 * as allocated out of @ob
1170 void bch2_alloc_sectors_append_ptrs(struct bch_fs *c, struct write_point *wp,
1171 struct bkey_i *k, unsigned sectors,
1175 struct open_bucket *ob;
1178 BUG_ON(sectors > wp->sectors_free);
1179 wp->sectors_free -= sectors;
1181 open_bucket_for_each(c, &wp->ptrs, ob, i) {
1182 struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
1183 struct bch_extent_ptr ptr = bch2_ob_ptr(c, ob);
1185 ptr.cached = cached ||
1186 (!ca->mi.durability &&
1187 wp->data_type == BCH_DATA_user);
1189 bch2_bkey_append_ptr(k, ptr);
1191 BUG_ON(sectors > ob->sectors_free);
1192 ob->sectors_free -= sectors;
1197 * Append pointers to the space we just allocated to @k, and mark @sectors space
1198 * as allocated out of @ob
1200 void bch2_alloc_sectors_done(struct bch_fs *c, struct write_point *wp)
1202 struct open_buckets ptrs = { .nr = 0 }, keep = { .nr = 0 };
1203 struct open_bucket *ob;
1206 open_bucket_for_each(c, &wp->ptrs, ob, i)
1207 ob_push(c, !ob->sectors_free ? &ptrs : &keep, ob);
1210 mutex_unlock(&wp->lock);
1212 bch2_open_buckets_put(c, &ptrs);
1215 static inline void writepoint_init(struct write_point *wp,
1216 enum bch_data_type type)
1218 mutex_init(&wp->lock);
1219 wp->data_type = type;
1222 void bch2_fs_allocator_foreground_init(struct bch_fs *c)
1224 struct open_bucket *ob;
1225 struct write_point *wp;
1227 mutex_init(&c->write_points_hash_lock);
1228 c->write_points_nr = ARRAY_SIZE(c->write_points);
1230 /* open bucket 0 is a sentinal NULL: */
1231 spin_lock_init(&c->open_buckets[0].lock);
1233 for (ob = c->open_buckets + 1;
1234 ob < c->open_buckets + ARRAY_SIZE(c->open_buckets); ob++) {
1235 spin_lock_init(&ob->lock);
1236 c->open_buckets_nr_free++;
1238 ob->freelist = c->open_buckets_freelist;
1239 c->open_buckets_freelist = ob - c->open_buckets;
1242 writepoint_init(&c->btree_write_point, BCH_DATA_btree);
1243 writepoint_init(&c->rebalance_write_point, BCH_DATA_user);
1244 writepoint_init(&c->copygc_write_point, BCH_DATA_user);
1246 for (wp = c->write_points;
1247 wp < c->write_points + c->write_points_nr; wp++) {
1248 writepoint_init(wp, BCH_DATA_user);
1250 wp->last_used = sched_clock();
1251 wp->write_point = (unsigned long) wp;
1252 hlist_add_head_rcu(&wp->node,
1253 writepoint_hash(c, wp->write_point));
1257 void bch2_open_buckets_to_text(struct printbuf *out, struct bch_fs *c)
1259 struct open_bucket *ob;
1261 for (ob = c->open_buckets;
1262 ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
1264 spin_lock(&ob->lock);
1265 if (ob->valid && !ob->on_partial_list) {
1266 pr_buf(out, "%zu ref %u type %s %u:%llu:%u\n",
1267 ob - c->open_buckets,
1268 atomic_read(&ob->pin),
1269 bch2_data_types[ob->data_type],
1270 ob->dev, ob->bucket, ob->gen);
1272 spin_unlock(&ob->lock);