1 // SPDX-License-Identifier: GPL-2.0
3 * Code for manipulating bucket marks for garbage collection.
5 * Copyright 2014 Datera, Inc.
9 #include "alloc_background.h"
10 #include "backpointers.h"
13 #include "btree_update.h"
15 #include "buckets_waiting_for_journal.h"
23 #include "subvolume.h"
26 #include <linux/preempt.h>
28 static inline void fs_usage_data_type_to_base(struct bch_fs_usage *fs_usage,
29 enum bch_data_type data_type,
34 fs_usage->btree += sectors;
38 fs_usage->data += sectors;
41 fs_usage->cached += sectors;
48 void bch2_fs_usage_initialize(struct bch_fs *c)
50 struct bch_fs_usage *usage;
54 percpu_down_write(&c->mark_lock);
55 usage = c->usage_base;
57 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
58 bch2_fs_usage_acc_to_base(c, i);
60 for (i = 0; i < BCH_REPLICAS_MAX; i++)
61 usage->reserved += usage->persistent_reserved[i];
63 for (i = 0; i < c->replicas.nr; i++) {
64 struct bch_replicas_entry *e =
65 cpu_replicas_entry(&c->replicas, i);
67 fs_usage_data_type_to_base(usage, e->data_type, usage->replicas[i]);
70 for_each_member_device(ca, c, i) {
71 struct bch_dev_usage dev = bch2_dev_usage_read(ca);
73 usage->hidden += (dev.d[BCH_DATA_sb].buckets +
74 dev.d[BCH_DATA_journal].buckets) *
78 percpu_up_write(&c->mark_lock);
81 static inline struct bch_dev_usage *dev_usage_ptr(struct bch_dev *ca,
85 BUG_ON(!gc && !journal_seq);
87 return this_cpu_ptr(gc
89 : ca->usage[journal_seq & JOURNAL_BUF_MASK]);
92 void bch2_dev_usage_read_fast(struct bch_dev *ca, struct bch_dev_usage *usage)
94 struct bch_fs *c = ca->fs;
95 unsigned seq, i, u64s = dev_usage_u64s();
98 seq = read_seqcount_begin(&c->usage_lock);
99 memcpy(usage, ca->usage_base, u64s * sizeof(u64));
100 for (i = 0; i < ARRAY_SIZE(ca->usage); i++)
101 acc_u64s_percpu((u64 *) usage, (u64 __percpu *) ca->usage[i], u64s);
102 } while (read_seqcount_retry(&c->usage_lock, seq));
105 u64 bch2_fs_usage_read_one(struct bch_fs *c, u64 *v)
107 ssize_t offset = v - (u64 *) c->usage_base;
111 BUG_ON(offset < 0 || offset >= fs_usage_u64s(c));
112 percpu_rwsem_assert_held(&c->mark_lock);
115 seq = read_seqcount_begin(&c->usage_lock);
118 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
119 ret += percpu_u64_get((u64 __percpu *) c->usage[i] + offset);
120 } while (read_seqcount_retry(&c->usage_lock, seq));
125 struct bch_fs_usage_online *bch2_fs_usage_read(struct bch_fs *c)
127 struct bch_fs_usage_online *ret;
128 unsigned nr_replicas = READ_ONCE(c->replicas.nr);
131 ret = kmalloc(__fs_usage_online_u64s(nr_replicas) * sizeof(u64), GFP_KERNEL);
135 percpu_down_read(&c->mark_lock);
137 if (nr_replicas != c->replicas.nr) {
138 nr_replicas = c->replicas.nr;
139 percpu_up_read(&c->mark_lock);
144 ret->online_reserved = percpu_u64_get(c->online_reserved);
147 seq = read_seqcount_begin(&c->usage_lock);
148 unsafe_memcpy(&ret->u, c->usage_base,
149 __fs_usage_u64s(nr_replicas) * sizeof(u64),
150 "embedded variable length struct");
151 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
152 acc_u64s_percpu((u64 *) &ret->u, (u64 __percpu *) c->usage[i],
153 __fs_usage_u64s(nr_replicas));
154 } while (read_seqcount_retry(&c->usage_lock, seq));
159 void bch2_fs_usage_acc_to_base(struct bch_fs *c, unsigned idx)
162 unsigned i, u64s = fs_usage_u64s(c);
164 BUG_ON(idx >= ARRAY_SIZE(c->usage));
167 write_seqcount_begin(&c->usage_lock);
169 acc_u64s_percpu((u64 *) c->usage_base,
170 (u64 __percpu *) c->usage[idx], u64s);
171 percpu_memset(c->usage[idx], 0, u64s * sizeof(u64));
174 for_each_member_device_rcu(ca, c, i, NULL) {
175 u64s = dev_usage_u64s();
177 acc_u64s_percpu((u64 *) ca->usage_base,
178 (u64 __percpu *) ca->usage[idx], u64s);
179 percpu_memset(ca->usage[idx], 0, u64s * sizeof(u64));
183 write_seqcount_end(&c->usage_lock);
187 void bch2_fs_usage_to_text(struct printbuf *out,
189 struct bch_fs_usage_online *fs_usage)
193 prt_printf(out, "capacity:\t\t\t%llu\n", c->capacity);
195 prt_printf(out, "hidden:\t\t\t\t%llu\n",
197 prt_printf(out, "data:\t\t\t\t%llu\n",
199 prt_printf(out, "cached:\t\t\t\t%llu\n",
201 prt_printf(out, "reserved:\t\t\t%llu\n",
202 fs_usage->u.reserved);
203 prt_printf(out, "nr_inodes:\t\t\t%llu\n",
204 fs_usage->u.nr_inodes);
205 prt_printf(out, "online reserved:\t\t%llu\n",
206 fs_usage->online_reserved);
209 i < ARRAY_SIZE(fs_usage->u.persistent_reserved);
211 prt_printf(out, "%u replicas:\n", i + 1);
212 prt_printf(out, "\treserved:\t\t%llu\n",
213 fs_usage->u.persistent_reserved[i]);
216 for (i = 0; i < c->replicas.nr; i++) {
217 struct bch_replicas_entry *e =
218 cpu_replicas_entry(&c->replicas, i);
220 prt_printf(out, "\t");
221 bch2_replicas_entry_to_text(out, e);
222 prt_printf(out, ":\t%llu\n", fs_usage->u.replicas[i]);
226 static u64 reserve_factor(u64 r)
228 return r + (round_up(r, (1 << RESERVE_FACTOR)) >> RESERVE_FACTOR);
231 u64 bch2_fs_sectors_used(struct bch_fs *c, struct bch_fs_usage_online *fs_usage)
233 return min(fs_usage->u.hidden +
236 reserve_factor(fs_usage->u.reserved +
237 fs_usage->online_reserved),
241 static struct bch_fs_usage_short
242 __bch2_fs_usage_read_short(struct bch_fs *c)
244 struct bch_fs_usage_short ret;
247 ret.capacity = c->capacity -
248 bch2_fs_usage_read_one(c, &c->usage_base->hidden);
250 data = bch2_fs_usage_read_one(c, &c->usage_base->data) +
251 bch2_fs_usage_read_one(c, &c->usage_base->btree);
252 reserved = bch2_fs_usage_read_one(c, &c->usage_base->reserved) +
253 percpu_u64_get(c->online_reserved);
255 ret.used = min(ret.capacity, data + reserve_factor(reserved));
256 ret.free = ret.capacity - ret.used;
258 ret.nr_inodes = bch2_fs_usage_read_one(c, &c->usage_base->nr_inodes);
263 struct bch_fs_usage_short
264 bch2_fs_usage_read_short(struct bch_fs *c)
266 struct bch_fs_usage_short ret;
268 percpu_down_read(&c->mark_lock);
269 ret = __bch2_fs_usage_read_short(c);
270 percpu_up_read(&c->mark_lock);
275 void bch2_dev_usage_init(struct bch_dev *ca)
277 ca->usage_base->d[BCH_DATA_free].buckets = ca->mi.nbuckets - ca->mi.first_bucket;
280 static inline int bucket_sectors_fragmented(struct bch_dev *ca,
281 struct bch_alloc_v4 a)
283 return a.dirty_sectors
284 ? max(0, (int) ca->mi.bucket_size - (int) a.dirty_sectors)
288 static void bch2_dev_usage_update(struct bch_fs *c, struct bch_dev *ca,
289 struct bch_alloc_v4 old,
290 struct bch_alloc_v4 new,
291 u64 journal_seq, bool gc)
293 struct bch_fs_usage *fs_usage;
294 struct bch_dev_usage *u;
297 fs_usage = fs_usage_ptr(c, journal_seq, gc);
299 if (data_type_is_hidden(old.data_type))
300 fs_usage->hidden -= ca->mi.bucket_size;
301 if (data_type_is_hidden(new.data_type))
302 fs_usage->hidden += ca->mi.bucket_size;
304 u = dev_usage_ptr(ca, journal_seq, gc);
306 u->d[old.data_type].buckets--;
307 u->d[new.data_type].buckets++;
309 u->buckets_ec -= (int) !!old.stripe;
310 u->buckets_ec += (int) !!new.stripe;
312 u->d[old.data_type].sectors -= old.dirty_sectors;
313 u->d[new.data_type].sectors += new.dirty_sectors;
315 u->d[BCH_DATA_cached].sectors += new.cached_sectors;
316 u->d[BCH_DATA_cached].sectors -= old.cached_sectors;
318 u->d[old.data_type].fragmented -= bucket_sectors_fragmented(ca, old);
319 u->d[new.data_type].fragmented += bucket_sectors_fragmented(ca, new);
324 static void bch2_dev_usage_update_m(struct bch_fs *c, struct bch_dev *ca,
325 struct bucket old, struct bucket new,
326 u64 journal_seq, bool gc)
328 struct bch_alloc_v4 old_a = {
330 .data_type = old.data_type,
331 .dirty_sectors = old.dirty_sectors,
332 .cached_sectors = old.cached_sectors,
333 .stripe = old.stripe,
335 struct bch_alloc_v4 new_a = {
337 .data_type = new.data_type,
338 .dirty_sectors = new.dirty_sectors,
339 .cached_sectors = new.cached_sectors,
340 .stripe = new.stripe,
343 bch2_dev_usage_update(c, ca, old_a, new_a, journal_seq, gc);
346 static inline int __update_replicas(struct bch_fs *c,
347 struct bch_fs_usage *fs_usage,
348 struct bch_replicas_entry *r,
351 int idx = bch2_replicas_entry_idx(c, r);
356 fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
357 fs_usage->replicas[idx] += sectors;
361 static inline int update_replicas(struct bch_fs *c, struct bkey_s_c k,
362 struct bch_replicas_entry *r, s64 sectors,
363 unsigned journal_seq, bool gc)
365 struct bch_fs_usage *fs_usage;
367 struct printbuf buf = PRINTBUF;
369 percpu_down_read(&c->mark_lock);
371 idx = bch2_replicas_entry_idx(c, r);
373 fsck_err(c, ptr_to_missing_replicas_entry,
374 "no replicas entry\n while marking %s",
375 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
376 percpu_up_read(&c->mark_lock);
377 ret = bch2_mark_replicas(c, r);
378 percpu_down_read(&c->mark_lock);
382 idx = bch2_replicas_entry_idx(c, r);
390 fs_usage = fs_usage_ptr(c, journal_seq, gc);
391 fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
392 fs_usage->replicas[idx] += sectors;
396 percpu_up_read(&c->mark_lock);
401 static inline int update_cached_sectors(struct bch_fs *c,
403 unsigned dev, s64 sectors,
404 unsigned journal_seq, bool gc)
406 struct bch_replicas_padded r;
408 bch2_replicas_entry_cached(&r.e, dev);
410 return update_replicas(c, k, &r.e, sectors, journal_seq, gc);
413 static int __replicas_deltas_realloc(struct btree_trans *trans, unsigned more,
416 struct replicas_delta_list *d = trans->fs_usage_deltas;
417 unsigned new_size = d ? (d->size + more) * 2 : 128;
418 unsigned alloc_size = sizeof(*d) + new_size;
420 WARN_ON_ONCE(alloc_size > REPLICAS_DELTA_LIST_MAX);
422 if (!d || d->used + more > d->size) {
423 d = krealloc(d, alloc_size, gfp|__GFP_ZERO);
426 if (alloc_size > REPLICAS_DELTA_LIST_MAX)
429 d = mempool_alloc(&trans->c->replicas_delta_pool, gfp);
433 memset(d, 0, REPLICAS_DELTA_LIST_MAX);
435 if (trans->fs_usage_deltas)
436 memcpy(d, trans->fs_usage_deltas,
437 trans->fs_usage_deltas->size + sizeof(*d));
439 new_size = REPLICAS_DELTA_LIST_MAX - sizeof(*d);
440 kfree(trans->fs_usage_deltas);
444 trans->fs_usage_deltas = d;
450 int bch2_replicas_deltas_realloc(struct btree_trans *trans, unsigned more)
452 return allocate_dropping_locks_errcode(trans,
453 __replicas_deltas_realloc(trans, more, _gfp));
456 static inline int update_replicas_list(struct btree_trans *trans,
457 struct bch_replicas_entry *r,
460 struct replicas_delta_list *d;
461 struct replicas_delta *n;
468 b = replicas_entry_bytes(r) + 8;
469 ret = bch2_replicas_deltas_realloc(trans, b);
473 d = trans->fs_usage_deltas;
474 n = (void *) d->d + d->used;
476 unsafe_memcpy((void *) n + offsetof(struct replicas_delta, r),
477 r, replicas_entry_bytes(r),
478 "flexible array member embedded in strcuct with padding");
479 bch2_replicas_entry_sort(&n->r);
484 static inline int update_cached_sectors_list(struct btree_trans *trans,
485 unsigned dev, s64 sectors)
487 struct bch_replicas_padded r;
489 bch2_replicas_entry_cached(&r.e, dev);
491 return update_replicas_list(trans, &r.e, sectors);
494 int bch2_mark_alloc(struct btree_trans *trans,
495 enum btree_id btree, unsigned level,
496 struct bkey_s_c old, struct bkey_s_c new,
499 bool gc = flags & BTREE_TRIGGER_GC;
500 u64 journal_seq = trans->journal_res.seq;
501 u64 bucket_journal_seq;
502 struct bch_fs *c = trans->c;
503 struct bch_alloc_v4 old_a_convert, new_a_convert;
504 const struct bch_alloc_v4 *old_a, *new_a;
509 * alloc btree is read in by bch2_alloc_read, not gc:
511 if ((flags & BTREE_TRIGGER_GC) &&
512 !(flags & BTREE_TRIGGER_BUCKET_INVALIDATE))
515 if (bch2_trans_inconsistent_on(!bch2_dev_bucket_exists(c, new.k->p), trans,
516 "alloc key for invalid device or bucket"))
519 ca = bch_dev_bkey_exists(c, new.k->p.inode);
521 old_a = bch2_alloc_to_v4(old, &old_a_convert);
522 new_a = bch2_alloc_to_v4(new, &new_a_convert);
524 bucket_journal_seq = new_a->journal_seq;
526 if ((flags & BTREE_TRIGGER_INSERT) &&
527 data_type_is_empty(old_a->data_type) !=
528 data_type_is_empty(new_a->data_type) &&
529 new.k->type == KEY_TYPE_alloc_v4) {
530 struct bch_alloc_v4 *v = (struct bch_alloc_v4 *) new.v;
532 EBUG_ON(!journal_seq);
535 * If the btree updates referring to a bucket weren't flushed
536 * before the bucket became empty again, then the we don't have
537 * to wait on a journal flush before we can reuse the bucket:
539 v->journal_seq = bucket_journal_seq =
540 data_type_is_empty(new_a->data_type) &&
541 (journal_seq == v->journal_seq ||
542 bch2_journal_noflush_seq(&c->journal, v->journal_seq))
546 if (!data_type_is_empty(old_a->data_type) &&
547 data_type_is_empty(new_a->data_type) &&
548 bucket_journal_seq) {
549 ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
550 c->journal.flushed_seq_ondisk,
551 new.k->p.inode, new.k->p.offset,
554 bch2_fs_fatal_error(c,
555 "error setting bucket_needs_journal_commit: %i", ret);
560 percpu_down_read(&c->mark_lock);
561 if (!gc && new_a->gen != old_a->gen)
562 *bucket_gen(ca, new.k->p.offset) = new_a->gen;
564 bch2_dev_usage_update(c, ca, *old_a, *new_a, journal_seq, gc);
567 struct bucket *g = gc_bucket(ca, new.k->p.offset);
573 g->data_type = new_a->data_type;
574 g->stripe = new_a->stripe;
575 g->stripe_redundancy = new_a->stripe_redundancy;
576 g->dirty_sectors = new_a->dirty_sectors;
577 g->cached_sectors = new_a->cached_sectors;
581 percpu_up_read(&c->mark_lock);
584 * need to know if we're getting called from the invalidate path or
588 if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
589 old_a->cached_sectors) {
590 ret = update_cached_sectors(c, new, ca->dev_idx,
591 -((s64) old_a->cached_sectors),
594 bch2_fs_fatal_error(c, "%s(): no replicas entry while updating cached sectors",
600 if (new_a->data_type == BCH_DATA_free &&
601 (!new_a->journal_seq || new_a->journal_seq < c->journal.flushed_seq_ondisk))
602 closure_wake_up(&c->freelist_wait);
604 if (new_a->data_type == BCH_DATA_need_discard &&
605 (!bucket_journal_seq || bucket_journal_seq < c->journal.flushed_seq_ondisk))
608 if (old_a->data_type != BCH_DATA_cached &&
609 new_a->data_type == BCH_DATA_cached &&
610 should_invalidate_buckets(ca, bch2_dev_usage_read(ca)))
611 bch2_do_invalidates(c);
613 if (new_a->data_type == BCH_DATA_need_gc_gens)
619 int bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca,
620 size_t b, enum bch_data_type data_type,
621 unsigned sectors, struct gc_pos pos,
624 struct bucket old, new, *g;
627 BUG_ON(!(flags & BTREE_TRIGGER_GC));
628 BUG_ON(data_type != BCH_DATA_sb &&
629 data_type != BCH_DATA_journal);
632 * Backup superblock might be past the end of our normal usable space:
634 if (b >= ca->mi.nbuckets)
637 percpu_down_read(&c->mark_lock);
638 g = gc_bucket(ca, b);
643 if (bch2_fs_inconsistent_on(g->data_type &&
644 g->data_type != data_type, c,
645 "different types of data in same bucket: %s, %s",
646 bch2_data_types[g->data_type],
647 bch2_data_types[data_type])) {
652 if (bch2_fs_inconsistent_on((u64) g->dirty_sectors + sectors > ca->mi.bucket_size, c,
653 "bucket %u:%zu gen %u data type %s sector count overflow: %u + %u > bucket size",
654 ca->dev_idx, b, g->gen,
655 bch2_data_types[g->data_type ?: data_type],
656 g->dirty_sectors, sectors)) {
662 g->data_type = data_type;
663 g->dirty_sectors += sectors;
668 bch2_dev_usage_update_m(c, ca, old, new, 0, true);
669 percpu_up_read(&c->mark_lock);
673 static int check_bucket_ref(struct btree_trans *trans,
675 const struct bch_extent_ptr *ptr,
676 s64 sectors, enum bch_data_type ptr_data_type,
677 u8 b_gen, u8 bucket_data_type,
678 u32 dirty_sectors, u32 cached_sectors)
680 struct bch_fs *c = trans->c;
681 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
682 size_t bucket_nr = PTR_BUCKET_NR(ca, ptr);
683 u32 bucket_sectors = !ptr->cached
686 struct printbuf buf = PRINTBUF;
689 if (bucket_data_type == BCH_DATA_cached)
690 bucket_data_type = BCH_DATA_user;
692 if ((bucket_data_type == BCH_DATA_stripe && ptr_data_type == BCH_DATA_user) ||
693 (bucket_data_type == BCH_DATA_user && ptr_data_type == BCH_DATA_stripe))
694 bucket_data_type = ptr_data_type = BCH_DATA_stripe;
696 if (gen_after(ptr->gen, b_gen)) {
697 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
698 BCH_FSCK_ERR_ptr_gen_newer_than_bucket_gen,
699 "bucket %u:%zu gen %u data type %s: ptr gen %u newer than bucket gen\n"
701 ptr->dev, bucket_nr, b_gen,
702 bch2_data_types[bucket_data_type ?: ptr_data_type],
704 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
709 if (gen_cmp(b_gen, ptr->gen) > BUCKET_GC_GEN_MAX) {
710 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
711 BCH_FSCK_ERR_ptr_too_stale,
712 "bucket %u:%zu gen %u data type %s: ptr gen %u too stale\n"
714 ptr->dev, bucket_nr, b_gen,
715 bch2_data_types[bucket_data_type ?: ptr_data_type],
717 (printbuf_reset(&buf),
718 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
723 if (b_gen != ptr->gen && !ptr->cached) {
724 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
725 BCH_FSCK_ERR_stale_dirty_ptr,
726 "bucket %u:%zu gen %u (mem gen %u) data type %s: stale dirty ptr (gen %u)\n"
728 ptr->dev, bucket_nr, b_gen,
729 *bucket_gen(ca, bucket_nr),
730 bch2_data_types[bucket_data_type ?: ptr_data_type],
732 (printbuf_reset(&buf),
733 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
738 if (b_gen != ptr->gen) {
743 if (!data_type_is_empty(bucket_data_type) &&
745 bucket_data_type != ptr_data_type) {
746 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
747 BCH_FSCK_ERR_ptr_bucket_data_type_mismatch,
748 "bucket %u:%zu gen %u different types of data in same bucket: %s, %s\n"
750 ptr->dev, bucket_nr, b_gen,
751 bch2_data_types[bucket_data_type],
752 bch2_data_types[ptr_data_type],
753 (printbuf_reset(&buf),
754 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
759 if ((u64) bucket_sectors + sectors > U32_MAX) {
760 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
761 BCH_FSCK_ERR_bucket_sector_count_overflow,
762 "bucket %u:%zu gen %u data type %s sector count overflow: %u + %lli > U32_MAX\n"
764 ptr->dev, bucket_nr, b_gen,
765 bch2_data_types[bucket_data_type ?: ptr_data_type],
766 bucket_sectors, sectors,
767 (printbuf_reset(&buf),
768 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
776 bch2_dump_trans_updates(trans);
780 static int mark_stripe_bucket(struct btree_trans *trans,
785 struct bch_fs *c = trans->c;
786 u64 journal_seq = trans->journal_res.seq;
787 const struct bch_stripe *s = bkey_s_c_to_stripe(k).v;
788 unsigned nr_data = s->nr_blocks - s->nr_redundant;
789 bool parity = ptr_idx >= nr_data;
790 enum bch_data_type data_type = parity ? BCH_DATA_parity : BCH_DATA_stripe;
791 s64 sectors = parity ? le16_to_cpu(s->sectors) : 0;
792 const struct bch_extent_ptr *ptr = s->ptrs + ptr_idx;
793 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
794 struct bucket old, new, *g;
795 struct printbuf buf = PRINTBUF;
798 BUG_ON(!(flags & BTREE_TRIGGER_GC));
800 /* * XXX doesn't handle deletion */
802 percpu_down_read(&c->mark_lock);
803 g = PTR_GC_BUCKET(ca, ptr);
805 if (g->dirty_sectors ||
806 (g->stripe && g->stripe != k.k->p.offset)) {
807 bch2_fs_inconsistent(c,
808 "bucket %u:%zu gen %u: multiple stripes using same bucket\n%s",
809 ptr->dev, PTR_BUCKET_NR(ca, ptr), g->gen,
810 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
818 ret = check_bucket_ref(trans, k, ptr, sectors, data_type,
819 g->gen, g->data_type,
820 g->dirty_sectors, g->cached_sectors);
824 g->data_type = data_type;
825 g->dirty_sectors += sectors;
827 g->stripe = k.k->p.offset;
828 g->stripe_redundancy = s->nr_redundant;
833 bch2_dev_usage_update_m(c, ca, old, new, journal_seq, true);
834 percpu_up_read(&c->mark_lock);
839 static int __mark_pointer(struct btree_trans *trans,
841 const struct bch_extent_ptr *ptr,
842 s64 sectors, enum bch_data_type ptr_data_type,
843 u8 bucket_gen, u8 *bucket_data_type,
844 u32 *dirty_sectors, u32 *cached_sectors)
846 u32 *dst_sectors = !ptr->cached
849 int ret = check_bucket_ref(trans, k, ptr, sectors, ptr_data_type,
850 bucket_gen, *bucket_data_type,
851 *dirty_sectors, *cached_sectors);
856 *dst_sectors += sectors;
857 *bucket_data_type = *dirty_sectors || *cached_sectors
862 static int bch2_mark_pointer(struct btree_trans *trans,
863 enum btree_id btree_id, unsigned level,
865 struct extent_ptr_decoded p,
869 u64 journal_seq = trans->journal_res.seq;
870 struct bch_fs *c = trans->c;
871 struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
872 struct bucket old, new, *g;
873 enum bch_data_type data_type = bkey_ptr_data_type(btree_id, level, k, p);
877 BUG_ON(!(flags & BTREE_TRIGGER_GC));
879 percpu_down_read(&c->mark_lock);
880 g = PTR_GC_BUCKET(ca, &p.ptr);
884 bucket_data_type = g->data_type;
885 ret = __mark_pointer(trans, k, &p.ptr, sectors,
891 g->data_type = bucket_data_type;
896 bch2_dev_usage_update_m(c, ca, old, new, journal_seq, true);
897 percpu_up_read(&c->mark_lock);
902 static int bch2_mark_stripe_ptr(struct btree_trans *trans,
904 struct bch_extent_stripe_ptr p,
905 enum bch_data_type data_type,
909 struct bch_fs *c = trans->c;
910 struct bch_replicas_padded r;
913 BUG_ON(!(flags & BTREE_TRIGGER_GC));
915 m = genradix_ptr_alloc(&c->gc_stripes, p.idx, GFP_KERNEL);
917 bch_err(c, "error allocating memory for gc_stripes, idx %llu",
919 return -BCH_ERR_ENOMEM_mark_stripe_ptr;
922 mutex_lock(&c->ec_stripes_heap_lock);
924 if (!m || !m->alive) {
925 mutex_unlock(&c->ec_stripes_heap_lock);
926 bch_err_ratelimited(c, "pointer to nonexistent stripe %llu",
928 bch2_inconsistent_error(c);
932 m->block_sectors[p.block] += sectors;
935 mutex_unlock(&c->ec_stripes_heap_lock);
937 r.e.data_type = data_type;
938 update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, true);
943 static int __mark_extent(struct btree_trans *trans,
944 enum btree_id btree_id, unsigned level,
945 struct bkey_s_c k, unsigned flags)
947 u64 journal_seq = trans->journal_res.seq;
948 struct bch_fs *c = trans->c;
949 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
950 const union bch_extent_entry *entry;
951 struct extent_ptr_decoded p;
952 struct bch_replicas_padded r;
953 enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
956 s64 sectors = bkey_is_btree_ptr(k.k)
959 s64 dirty_sectors = 0;
963 BUG_ON(!(flags & BTREE_TRIGGER_GC));
965 r.e.data_type = data_type;
969 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
970 s64 disk_sectors = ptr_disk_sectors(sectors, p);
972 if (flags & BTREE_TRIGGER_OVERWRITE)
973 disk_sectors = -disk_sectors;
975 ret = bch2_mark_pointer(trans, btree_id, level, k, p, disk_sectors, flags);
983 ret = update_cached_sectors(c, k, p.ptr.dev,
984 disk_sectors, journal_seq, true);
986 bch2_fs_fatal_error(c, "%s(): no replicas entry while updating cached sectors",
991 } else if (!p.has_ec) {
992 dirty_sectors += disk_sectors;
993 r.e.devs[r.e.nr_devs++] = p.ptr.dev;
995 ret = bch2_mark_stripe_ptr(trans, k, p.ec, data_type,
996 disk_sectors, flags);
1001 * There may be other dirty pointers in this extent, but
1002 * if so they're not required for mounting if we have an
1003 * erasure coded pointer in this extent:
1005 r.e.nr_required = 0;
1010 ret = update_replicas(c, k, &r.e, dirty_sectors, journal_seq, true);
1012 struct printbuf buf = PRINTBUF;
1014 bch2_bkey_val_to_text(&buf, c, k);
1015 bch2_fs_fatal_error(c, "%s(): no replicas entry for %s", __func__, buf.buf);
1016 printbuf_exit(&buf);
1024 int bch2_mark_extent(struct btree_trans *trans,
1025 enum btree_id btree_id, unsigned level,
1026 struct bkey_s_c old, struct bkey_s_c new,
1029 return mem_trigger_run_overwrite_then_insert(__mark_extent, trans, btree_id, level, old, new, flags);
1032 int bch2_mark_stripe(struct btree_trans *trans,
1033 enum btree_id btree_id, unsigned level,
1034 struct bkey_s_c old, struct bkey_s_c new,
1037 bool gc = flags & BTREE_TRIGGER_GC;
1038 u64 journal_seq = trans->journal_res.seq;
1039 struct bch_fs *c = trans->c;
1040 u64 idx = new.k->p.offset;
1041 const struct bch_stripe *old_s = old.k->type == KEY_TYPE_stripe
1042 ? bkey_s_c_to_stripe(old).v : NULL;
1043 const struct bch_stripe *new_s = new.k->type == KEY_TYPE_stripe
1044 ? bkey_s_c_to_stripe(new).v : NULL;
1048 BUG_ON(gc && old_s);
1051 struct stripe *m = genradix_ptr(&c->stripes, idx);
1054 struct printbuf buf1 = PRINTBUF;
1055 struct printbuf buf2 = PRINTBUF;
1057 bch2_bkey_val_to_text(&buf1, c, old);
1058 bch2_bkey_val_to_text(&buf2, c, new);
1059 bch_err_ratelimited(c, "error marking nonexistent stripe %llu while marking\n"
1061 "new %s", idx, buf1.buf, buf2.buf);
1062 printbuf_exit(&buf2);
1063 printbuf_exit(&buf1);
1064 bch2_inconsistent_error(c);
1069 bch2_stripes_heap_del(c, m, idx);
1071 memset(m, 0, sizeof(*m));
1073 m->sectors = le16_to_cpu(new_s->sectors);
1074 m->algorithm = new_s->algorithm;
1075 m->nr_blocks = new_s->nr_blocks;
1076 m->nr_redundant = new_s->nr_redundant;
1077 m->blocks_nonempty = 0;
1079 for (i = 0; i < new_s->nr_blocks; i++)
1080 m->blocks_nonempty += !!stripe_blockcount_get(new_s, i);
1083 bch2_stripes_heap_insert(c, m, idx);
1085 bch2_stripes_heap_update(c, m, idx);
1088 struct gc_stripe *m =
1089 genradix_ptr_alloc(&c->gc_stripes, idx, GFP_KERNEL);
1092 bch_err(c, "error allocating memory for gc_stripes, idx %llu",
1094 return -BCH_ERR_ENOMEM_mark_stripe;
1097 * This will be wrong when we bring back runtime gc: we should
1098 * be unmarking the old key and then marking the new key
1101 m->sectors = le16_to_cpu(new_s->sectors);
1102 m->nr_blocks = new_s->nr_blocks;
1103 m->nr_redundant = new_s->nr_redundant;
1105 for (i = 0; i < new_s->nr_blocks; i++)
1106 m->ptrs[i] = new_s->ptrs[i];
1108 bch2_bkey_to_replicas(&m->r.e, new);
1111 * gc recalculates this field from stripe ptr
1114 memset(m->block_sectors, 0, sizeof(m->block_sectors));
1116 for (i = 0; i < new_s->nr_blocks; i++) {
1117 ret = mark_stripe_bucket(trans, new, i, flags);
1122 ret = update_replicas(c, new, &m->r.e,
1123 ((s64) m->sectors * m->nr_redundant),
1126 struct printbuf buf = PRINTBUF;
1128 bch2_bkey_val_to_text(&buf, c, new);
1129 bch2_fs_fatal_error(c, "no replicas entry for %s", buf.buf);
1130 printbuf_exit(&buf);
1138 static int __mark_reservation(struct btree_trans *trans,
1139 enum btree_id btree_id, unsigned level,
1140 struct bkey_s_c k, unsigned flags)
1142 struct bch_fs *c = trans->c;
1143 struct bch_fs_usage *fs_usage;
1144 unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
1145 s64 sectors = (s64) k.k->size;
1147 BUG_ON(!(flags & BTREE_TRIGGER_GC));
1149 if (flags & BTREE_TRIGGER_OVERWRITE)
1151 sectors *= replicas;
1153 percpu_down_read(&c->mark_lock);
1156 fs_usage = fs_usage_ptr(c, trans->journal_res.seq, flags & BTREE_TRIGGER_GC);
1157 replicas = clamp_t(unsigned, replicas, 1,
1158 ARRAY_SIZE(fs_usage->persistent_reserved));
1160 fs_usage->reserved += sectors;
1161 fs_usage->persistent_reserved[replicas - 1] += sectors;
1164 percpu_up_read(&c->mark_lock);
1169 int bch2_mark_reservation(struct btree_trans *trans,
1170 enum btree_id btree_id, unsigned level,
1171 struct bkey_s_c old, struct bkey_s_c new,
1174 return mem_trigger_run_overwrite_then_insert(__mark_reservation, trans, btree_id, level, old, new, flags);
1177 static s64 __bch2_mark_reflink_p(struct btree_trans *trans,
1178 struct bkey_s_c_reflink_p p,
1180 u64 *idx, unsigned flags, size_t r_idx)
1182 struct bch_fs *c = trans->c;
1183 struct reflink_gc *r;
1184 int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
1187 struct printbuf buf = PRINTBUF;
1189 if (r_idx >= c->reflink_gc_nr)
1192 r = genradix_ptr(&c->reflink_gc_table, r_idx);
1193 next_idx = min(next_idx, r->offset - r->size);
1194 if (*idx < next_idx)
1197 BUG_ON((s64) r->refcount + add < 0);
1203 if (fsck_err(c, reflink_p_to_missing_reflink_v,
1204 "pointer to missing indirect extent\n"
1206 " missing range %llu-%llu",
1207 (bch2_bkey_val_to_text(&buf, c, p.s_c), buf.buf),
1209 struct bkey_i_error *new;
1211 new = bch2_trans_kmalloc(trans, sizeof(*new));
1212 ret = PTR_ERR_OR_ZERO(new);
1217 new->k.type = KEY_TYPE_error;
1218 new->k.p = bkey_start_pos(p.k);
1219 new->k.p.offset += *idx - start;
1220 bch2_key_resize(&new->k, next_idx - *idx);
1221 ret = bch2_btree_insert_trans(trans, BTREE_ID_extents, &new->k_i,
1222 BTREE_TRIGGER_NORUN);
1228 printbuf_exit(&buf);
1232 static int __mark_reflink_p(struct btree_trans *trans,
1233 enum btree_id btree_id, unsigned level,
1234 struct bkey_s_c k, unsigned flags)
1236 struct bch_fs *c = trans->c;
1237 struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
1238 struct reflink_gc *ref;
1240 u64 idx = le64_to_cpu(p.v->idx), start = idx;
1241 u64 end = le64_to_cpu(p.v->idx) + p.k->size;
1244 BUG_ON(!(flags & BTREE_TRIGGER_GC));
1246 if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_reflink_p_fix) {
1247 idx -= le32_to_cpu(p.v->front_pad);
1248 end += le32_to_cpu(p.v->back_pad);
1252 r = c->reflink_gc_nr;
1254 m = l + (r - l) / 2;
1256 ref = genradix_ptr(&c->reflink_gc_table, m);
1257 if (ref->offset <= idx)
1263 while (idx < end && !ret)
1264 ret = __bch2_mark_reflink_p(trans, p, start, end,
1270 int bch2_mark_reflink_p(struct btree_trans *trans,
1271 enum btree_id btree_id, unsigned level,
1272 struct bkey_s_c old, struct bkey_s_c new,
1275 return mem_trigger_run_overwrite_then_insert(__mark_reflink_p, trans, btree_id, level, old, new, flags);
1278 void bch2_trans_fs_usage_revert(struct btree_trans *trans,
1279 struct replicas_delta_list *deltas)
1281 struct bch_fs *c = trans->c;
1282 struct bch_fs_usage *dst;
1283 struct replicas_delta *d, *top = (void *) deltas->d + deltas->used;
1287 percpu_down_read(&c->mark_lock);
1289 dst = fs_usage_ptr(c, trans->journal_res.seq, false);
1291 /* revert changes: */
1292 for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
1293 switch (d->r.data_type) {
1294 case BCH_DATA_btree:
1296 case BCH_DATA_parity:
1299 BUG_ON(__update_replicas(c, dst, &d->r, -d->delta));
1302 dst->nr_inodes -= deltas->nr_inodes;
1304 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1305 added -= deltas->persistent_reserved[i];
1306 dst->reserved -= deltas->persistent_reserved[i];
1307 dst->persistent_reserved[i] -= deltas->persistent_reserved[i];
1311 trans->disk_res->sectors += added;
1312 this_cpu_add(*c->online_reserved, added);
1316 percpu_up_read(&c->mark_lock);
1319 int bch2_trans_fs_usage_apply(struct btree_trans *trans,
1320 struct replicas_delta_list *deltas)
1322 struct bch_fs *c = trans->c;
1323 static int warned_disk_usage = 0;
1325 u64 disk_res_sectors = trans->disk_res ? trans->disk_res->sectors : 0;
1326 struct replicas_delta *d, *d2;
1327 struct replicas_delta *top = (void *) deltas->d + deltas->used;
1328 struct bch_fs_usage *dst;
1329 s64 added = 0, should_not_have_added;
1332 percpu_down_read(&c->mark_lock);
1334 dst = fs_usage_ptr(c, trans->journal_res.seq, false);
1336 for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
1337 switch (d->r.data_type) {
1338 case BCH_DATA_btree:
1340 case BCH_DATA_parity:
1344 if (__update_replicas(c, dst, &d->r, d->delta))
1348 dst->nr_inodes += deltas->nr_inodes;
1350 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1351 added += deltas->persistent_reserved[i];
1352 dst->reserved += deltas->persistent_reserved[i];
1353 dst->persistent_reserved[i] += deltas->persistent_reserved[i];
1357 * Not allowed to reduce sectors_available except by getting a
1360 should_not_have_added = added - (s64) disk_res_sectors;
1361 if (unlikely(should_not_have_added > 0)) {
1362 u64 old, new, v = atomic64_read(&c->sectors_available);
1366 new = max_t(s64, 0, old - should_not_have_added);
1367 } while ((v = atomic64_cmpxchg(&c->sectors_available,
1370 added -= should_not_have_added;
1375 trans->disk_res->sectors -= added;
1376 this_cpu_sub(*c->online_reserved, added);
1380 percpu_up_read(&c->mark_lock);
1382 if (unlikely(warn) && !xchg(&warned_disk_usage, 1))
1383 bch2_trans_inconsistent(trans,
1384 "disk usage increased %lli more than %llu sectors reserved)",
1385 should_not_have_added, disk_res_sectors);
1388 /* revert changes: */
1389 for (d2 = deltas->d; d2 != d; d2 = replicas_delta_next(d2))
1390 BUG_ON(__update_replicas(c, dst, &d2->r, -d2->delta));
1393 percpu_up_read(&c->mark_lock);
1399 static inline int bch2_trans_mark_pointer(struct btree_trans *trans,
1400 enum btree_id btree_id, unsigned level,
1401 struct bkey_s_c k, struct extent_ptr_decoded p,
1404 bool insert = !(flags & BTREE_TRIGGER_OVERWRITE);
1405 struct btree_iter iter;
1406 struct bkey_i_alloc_v4 *a;
1408 struct bch_backpointer bp;
1412 bch2_extent_ptr_to_bp(trans->c, btree_id, level, k, p, &bucket, &bp);
1413 sectors = bp.bucket_len;
1417 a = bch2_trans_start_alloc_update(trans, &iter, bucket);
1421 ret = __mark_pointer(trans, k, &p.ptr, sectors, bp.data_type,
1422 a->v.gen, &a->v.data_type,
1423 &a->v.dirty_sectors, &a->v.cached_sectors) ?:
1424 bch2_trans_update(trans, &iter, &a->k_i, 0);
1425 bch2_trans_iter_exit(trans, &iter);
1430 if (!p.ptr.cached) {
1431 ret = bch2_bucket_backpointer_mod(trans, bucket, bp, k, insert);
1439 static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
1440 struct extent_ptr_decoded p,
1441 s64 sectors, enum bch_data_type data_type)
1443 struct btree_iter iter;
1444 struct bkey_i_stripe *s;
1445 struct bch_replicas_padded r;
1448 s = bch2_bkey_get_mut_typed(trans, &iter,
1449 BTREE_ID_stripes, POS(0, p.ec.idx),
1450 BTREE_ITER_WITH_UPDATES, stripe);
1451 ret = PTR_ERR_OR_ZERO(s);
1452 if (unlikely(ret)) {
1453 bch2_trans_inconsistent_on(bch2_err_matches(ret, ENOENT), trans,
1454 "pointer to nonexistent stripe %llu",
1459 if (!bch2_ptr_matches_stripe(&s->v, p)) {
1460 bch2_trans_inconsistent(trans,
1461 "stripe pointer doesn't match stripe %llu",
1467 stripe_blockcount_set(&s->v, p.ec.block,
1468 stripe_blockcount_get(&s->v, p.ec.block) +
1471 bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(&s->k_i));
1472 r.e.data_type = data_type;
1473 ret = update_replicas_list(trans, &r.e, sectors);
1475 bch2_trans_iter_exit(trans, &iter);
1479 static int __trans_mark_extent(struct btree_trans *trans,
1480 enum btree_id btree_id, unsigned level,
1481 struct bkey_s_c k, unsigned flags)
1483 struct bch_fs *c = trans->c;
1484 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1485 const union bch_extent_entry *entry;
1486 struct extent_ptr_decoded p;
1487 struct bch_replicas_padded r;
1488 enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
1491 s64 sectors = bkey_is_btree_ptr(k.k)
1494 s64 dirty_sectors = 0;
1498 r.e.data_type = data_type;
1500 r.e.nr_required = 1;
1502 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
1503 s64 disk_sectors = ptr_disk_sectors(sectors, p);
1505 if (flags & BTREE_TRIGGER_OVERWRITE)
1506 disk_sectors = -disk_sectors;
1508 ret = bch2_trans_mark_pointer(trans, btree_id, level, k, p, flags);
1516 ret = update_cached_sectors_list(trans, p.ptr.dev,
1521 } else if (!p.has_ec) {
1522 dirty_sectors += disk_sectors;
1523 r.e.devs[r.e.nr_devs++] = p.ptr.dev;
1525 ret = bch2_trans_mark_stripe_ptr(trans, p,
1526 disk_sectors, data_type);
1530 r.e.nr_required = 0;
1535 ret = update_replicas_list(trans, &r.e, dirty_sectors);
1540 int bch2_trans_mark_extent(struct btree_trans *trans,
1541 enum btree_id btree_id, unsigned level,
1542 struct bkey_s_c old, struct bkey_i *new,
1545 struct bch_fs *c = trans->c;
1546 int mod = (int) bch2_bkey_needs_rebalance(c, bkey_i_to_s_c(new)) -
1547 (int) bch2_bkey_needs_rebalance(c, old);
1550 int ret = bch2_btree_bit_mod(trans, BTREE_ID_rebalance_work, new->k.p, mod > 0);
1555 return trigger_run_overwrite_then_insert(__trans_mark_extent, trans, btree_id, level, old, new, flags);
1558 static int bch2_trans_mark_stripe_bucket(struct btree_trans *trans,
1559 struct bkey_s_c_stripe s,
1560 unsigned idx, bool deleting)
1562 struct bch_fs *c = trans->c;
1563 const struct bch_extent_ptr *ptr = &s.v->ptrs[idx];
1564 struct btree_iter iter;
1565 struct bkey_i_alloc_v4 *a;
1566 enum bch_data_type data_type = idx >= s.v->nr_blocks - s.v->nr_redundant
1567 ? BCH_DATA_parity : 0;
1568 s64 sectors = data_type ? le16_to_cpu(s.v->sectors) : 0;
1574 a = bch2_trans_start_alloc_update(trans, &iter, PTR_BUCKET_POS(c, ptr));
1578 ret = check_bucket_ref(trans, s.s_c, ptr, sectors, data_type,
1579 a->v.gen, a->v.data_type,
1580 a->v.dirty_sectors, a->v.cached_sectors);
1585 if (bch2_trans_inconsistent_on(a->v.stripe ||
1586 a->v.stripe_redundancy, trans,
1587 "bucket %llu:%llu gen %u data type %s dirty_sectors %u: multiple stripes using same bucket (%u, %llu)",
1588 iter.pos.inode, iter.pos.offset, a->v.gen,
1589 bch2_data_types[a->v.data_type],
1591 a->v.stripe, s.k->p.offset)) {
1596 if (bch2_trans_inconsistent_on(data_type && a->v.dirty_sectors, trans,
1597 "bucket %llu:%llu gen %u data type %s dirty_sectors %u: data already in stripe bucket %llu",
1598 iter.pos.inode, iter.pos.offset, a->v.gen,
1599 bch2_data_types[a->v.data_type],
1606 a->v.stripe = s.k->p.offset;
1607 a->v.stripe_redundancy = s.v->nr_redundant;
1608 a->v.data_type = BCH_DATA_stripe;
1610 if (bch2_trans_inconsistent_on(a->v.stripe != s.k->p.offset ||
1611 a->v.stripe_redundancy != s.v->nr_redundant, trans,
1612 "bucket %llu:%llu gen %u: not marked as stripe when deleting stripe %llu (got %u)",
1613 iter.pos.inode, iter.pos.offset, a->v.gen,
1614 s.k->p.offset, a->v.stripe)) {
1620 a->v.stripe_redundancy = 0;
1621 a->v.data_type = alloc_data_type(a->v, BCH_DATA_user);
1624 a->v.dirty_sectors += sectors;
1626 a->v.data_type = !deleting ? data_type : 0;
1628 ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
1632 bch2_trans_iter_exit(trans, &iter);
1636 int bch2_trans_mark_stripe(struct btree_trans *trans,
1637 enum btree_id btree_id, unsigned level,
1638 struct bkey_s_c old, struct bkey_i *new,
1641 const struct bch_stripe *old_s = NULL;
1642 struct bch_stripe *new_s = NULL;
1643 struct bch_replicas_padded r;
1644 unsigned i, nr_blocks;
1647 if (old.k->type == KEY_TYPE_stripe)
1648 old_s = bkey_s_c_to_stripe(old).v;
1649 if (new->k.type == KEY_TYPE_stripe)
1650 new_s = &bkey_i_to_stripe(new)->v;
1653 * If the pointers aren't changing, we don't need to do anything:
1655 if (new_s && old_s &&
1656 new_s->nr_blocks == old_s->nr_blocks &&
1657 new_s->nr_redundant == old_s->nr_redundant &&
1658 !memcmp(old_s->ptrs, new_s->ptrs,
1659 new_s->nr_blocks * sizeof(struct bch_extent_ptr)))
1662 BUG_ON(new_s && old_s &&
1663 (new_s->nr_blocks != old_s->nr_blocks ||
1664 new_s->nr_redundant != old_s->nr_redundant));
1666 nr_blocks = new_s ? new_s->nr_blocks : old_s->nr_blocks;
1669 s64 sectors = le16_to_cpu(new_s->sectors);
1671 bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(new));
1672 ret = update_replicas_list(trans, &r.e, sectors * new_s->nr_redundant);
1678 s64 sectors = -((s64) le16_to_cpu(old_s->sectors));
1680 bch2_bkey_to_replicas(&r.e, old);
1681 ret = update_replicas_list(trans, &r.e, sectors * old_s->nr_redundant);
1686 for (i = 0; i < nr_blocks; i++) {
1687 if (new_s && old_s &&
1688 !memcmp(&new_s->ptrs[i],
1690 sizeof(new_s->ptrs[i])))
1694 ret = bch2_trans_mark_stripe_bucket(trans,
1695 bkey_i_to_s_c_stripe(new), i, false);
1701 ret = bch2_trans_mark_stripe_bucket(trans,
1702 bkey_s_c_to_stripe(old), i, true);
1711 static int __trans_mark_reservation(struct btree_trans *trans,
1712 enum btree_id btree_id, unsigned level,
1713 struct bkey_s_c k, unsigned flags)
1715 unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
1716 s64 sectors = (s64) k.k->size;
1717 struct replicas_delta_list *d;
1720 if (flags & BTREE_TRIGGER_OVERWRITE)
1722 sectors *= replicas;
1724 ret = bch2_replicas_deltas_realloc(trans, 0);
1728 d = trans->fs_usage_deltas;
1729 replicas = clamp_t(unsigned, replicas, 1,
1730 ARRAY_SIZE(d->persistent_reserved));
1732 d->persistent_reserved[replicas - 1] += sectors;
1736 int bch2_trans_mark_reservation(struct btree_trans *trans,
1737 enum btree_id btree_id, unsigned level,
1738 struct bkey_s_c old,
1742 return trigger_run_overwrite_then_insert(__trans_mark_reservation, trans, btree_id, level, old, new, flags);
1745 static int trans_mark_reflink_p_segment(struct btree_trans *trans,
1746 struct bkey_s_c_reflink_p p,
1747 u64 *idx, unsigned flags)
1749 struct bch_fs *c = trans->c;
1750 struct btree_iter iter;
1753 int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
1754 struct printbuf buf = PRINTBUF;
1757 k = bch2_bkey_get_mut_noupdate(trans, &iter,
1758 BTREE_ID_reflink, POS(0, *idx),
1759 BTREE_ITER_WITH_UPDATES);
1760 ret = PTR_ERR_OR_ZERO(k);
1764 refcount = bkey_refcount(k);
1766 bch2_bkey_val_to_text(&buf, c, p.s_c);
1767 bch2_trans_inconsistent(trans,
1768 "nonexistent indirect extent at %llu while marking\n %s",
1774 if (!*refcount && (flags & BTREE_TRIGGER_OVERWRITE)) {
1775 bch2_bkey_val_to_text(&buf, c, p.s_c);
1776 bch2_trans_inconsistent(trans,
1777 "indirect extent refcount underflow at %llu while marking\n %s",
1783 if (flags & BTREE_TRIGGER_INSERT) {
1784 struct bch_reflink_p *v = (struct bch_reflink_p *) p.v;
1787 pad = max_t(s64, le32_to_cpu(v->front_pad),
1788 le64_to_cpu(v->idx) - bkey_start_offset(&k->k));
1789 BUG_ON(pad > U32_MAX);
1790 v->front_pad = cpu_to_le32(pad);
1792 pad = max_t(s64, le32_to_cpu(v->back_pad),
1793 k->k.p.offset - p.k->size - le64_to_cpu(v->idx));
1794 BUG_ON(pad > U32_MAX);
1795 v->back_pad = cpu_to_le32(pad);
1798 le64_add_cpu(refcount, add);
1800 bch2_btree_iter_set_pos_to_extent_start(&iter);
1801 ret = bch2_trans_update(trans, &iter, k, 0);
1805 *idx = k->k.p.offset;
1807 bch2_trans_iter_exit(trans, &iter);
1808 printbuf_exit(&buf);
1812 static int __trans_mark_reflink_p(struct btree_trans *trans,
1813 enum btree_id btree_id, unsigned level,
1814 struct bkey_s_c k, unsigned flags)
1816 struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
1820 idx = le64_to_cpu(p.v->idx) - le32_to_cpu(p.v->front_pad);
1821 end_idx = le64_to_cpu(p.v->idx) + p.k->size +
1822 le32_to_cpu(p.v->back_pad);
1824 while (idx < end_idx && !ret)
1825 ret = trans_mark_reflink_p_segment(trans, p, &idx, flags);
1829 int bch2_trans_mark_reflink_p(struct btree_trans *trans,
1830 enum btree_id btree_id, unsigned level,
1831 struct bkey_s_c old,
1835 if (flags & BTREE_TRIGGER_INSERT) {
1836 struct bch_reflink_p *v = &bkey_i_to_reflink_p(new)->v;
1838 v->front_pad = v->back_pad = 0;
1841 return trigger_run_overwrite_then_insert(__trans_mark_reflink_p, trans, btree_id, level, old, new, flags);
1844 static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
1845 struct bch_dev *ca, size_t b,
1846 enum bch_data_type type,
1849 struct bch_fs *c = trans->c;
1850 struct btree_iter iter;
1851 struct bkey_i_alloc_v4 *a;
1855 * Backup superblock might be past the end of our normal usable space:
1857 if (b >= ca->mi.nbuckets)
1860 a = bch2_trans_start_alloc_update(trans, &iter, POS(ca->dev_idx, b));
1864 if (a->v.data_type && type && a->v.data_type != type) {
1865 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
1866 BCH_FSCK_ERR_bucket_metadata_type_mismatch,
1867 "bucket %llu:%llu gen %u different types of data in same bucket: %s, %s\n"
1869 iter.pos.inode, iter.pos.offset, a->v.gen,
1870 bch2_data_types[a->v.data_type],
1871 bch2_data_types[type],
1872 bch2_data_types[type]);
1877 if (a->v.data_type != type ||
1878 a->v.dirty_sectors != sectors) {
1879 a->v.data_type = type;
1880 a->v.dirty_sectors = sectors;
1881 ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
1884 bch2_trans_iter_exit(trans, &iter);
1888 int bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
1889 struct bch_dev *ca, size_t b,
1890 enum bch_data_type type,
1893 return commit_do(trans, NULL, NULL, 0,
1894 __bch2_trans_mark_metadata_bucket(trans, ca, b, type, sectors));
1897 static int bch2_trans_mark_metadata_sectors(struct btree_trans *trans,
1900 enum bch_data_type type,
1901 u64 *bucket, unsigned *bucket_sectors)
1904 u64 b = sector_to_bucket(ca, start);
1906 min_t(u64, bucket_to_sector(ca, b + 1), end) - start;
1908 if (b != *bucket && *bucket_sectors) {
1909 int ret = bch2_trans_mark_metadata_bucket(trans, ca, *bucket,
1910 type, *bucket_sectors);
1914 *bucket_sectors = 0;
1918 *bucket_sectors += sectors;
1920 } while (start < end);
1925 static int __bch2_trans_mark_dev_sb(struct btree_trans *trans,
1928 struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
1930 unsigned i, bucket_sectors = 0;
1933 for (i = 0; i < layout->nr_superblocks; i++) {
1934 u64 offset = le64_to_cpu(layout->sb_offset[i]);
1936 if (offset == BCH_SB_SECTOR) {
1937 ret = bch2_trans_mark_metadata_sectors(trans, ca,
1939 BCH_DATA_sb, &bucket, &bucket_sectors);
1944 ret = bch2_trans_mark_metadata_sectors(trans, ca, offset,
1945 offset + (1 << layout->sb_max_size_bits),
1946 BCH_DATA_sb, &bucket, &bucket_sectors);
1951 if (bucket_sectors) {
1952 ret = bch2_trans_mark_metadata_bucket(trans, ca,
1953 bucket, BCH_DATA_sb, bucket_sectors);
1958 for (i = 0; i < ca->journal.nr; i++) {
1959 ret = bch2_trans_mark_metadata_bucket(trans, ca,
1960 ca->journal.buckets[i],
1961 BCH_DATA_journal, ca->mi.bucket_size);
1969 int bch2_trans_mark_dev_sb(struct bch_fs *c, struct bch_dev *ca)
1971 int ret = bch2_trans_run(c, __bch2_trans_mark_dev_sb(trans, ca));
1978 int bch2_trans_mark_dev_sbs(struct bch_fs *c)
1983 for_each_online_member(ca, c, i) {
1984 int ret = bch2_trans_mark_dev_sb(c, ca);
1986 percpu_ref_put(&ca->ref);
1994 /* Disk reservations: */
1996 #define SECTORS_CACHE 1024
1998 int __bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
1999 u64 sectors, int flags)
2001 struct bch_fs_pcpu *pcpu;
2003 s64 sectors_available;
2006 percpu_down_read(&c->mark_lock);
2008 pcpu = this_cpu_ptr(c->pcpu);
2010 if (sectors <= pcpu->sectors_available)
2013 v = atomic64_read(&c->sectors_available);
2016 get = min((u64) sectors + SECTORS_CACHE, old);
2018 if (get < sectors) {
2022 } while ((v = atomic64_cmpxchg(&c->sectors_available,
2023 old, old - get)) != old);
2025 pcpu->sectors_available += get;
2028 pcpu->sectors_available -= sectors;
2029 this_cpu_add(*c->online_reserved, sectors);
2030 res->sectors += sectors;
2033 percpu_up_read(&c->mark_lock);
2037 mutex_lock(&c->sectors_available_lock);
2039 percpu_u64_set(&c->pcpu->sectors_available, 0);
2040 sectors_available = avail_factor(__bch2_fs_usage_read_short(c).free);
2042 if (sectors <= sectors_available ||
2043 (flags & BCH_DISK_RESERVATION_NOFAIL)) {
2044 atomic64_set(&c->sectors_available,
2045 max_t(s64, 0, sectors_available - sectors));
2046 this_cpu_add(*c->online_reserved, sectors);
2047 res->sectors += sectors;
2050 atomic64_set(&c->sectors_available, sectors_available);
2051 ret = -BCH_ERR_ENOSPC_disk_reservation;
2054 mutex_unlock(&c->sectors_available_lock);
2055 percpu_up_read(&c->mark_lock);
2060 /* Startup/shutdown: */
2062 static void bucket_gens_free_rcu(struct rcu_head *rcu)
2064 struct bucket_gens *buckets =
2065 container_of(rcu, struct bucket_gens, rcu);
2067 kvpfree(buckets, sizeof(*buckets) + buckets->nbuckets);
2070 int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
2072 struct bucket_gens *bucket_gens = NULL, *old_bucket_gens = NULL;
2073 unsigned long *buckets_nouse = NULL;
2074 bool resize = ca->bucket_gens != NULL;
2077 if (!(bucket_gens = kvpmalloc(sizeof(struct bucket_gens) + nbuckets,
2078 GFP_KERNEL|__GFP_ZERO))) {
2079 ret = -BCH_ERR_ENOMEM_bucket_gens;
2083 if ((c->opts.buckets_nouse &&
2084 !(buckets_nouse = kvpmalloc(BITS_TO_LONGS(nbuckets) *
2085 sizeof(unsigned long),
2086 GFP_KERNEL|__GFP_ZERO)))) {
2087 ret = -BCH_ERR_ENOMEM_buckets_nouse;
2091 bucket_gens->first_bucket = ca->mi.first_bucket;
2092 bucket_gens->nbuckets = nbuckets;
2094 bch2_copygc_stop(c);
2097 down_write(&c->gc_lock);
2098 down_write(&ca->bucket_lock);
2099 percpu_down_write(&c->mark_lock);
2102 old_bucket_gens = rcu_dereference_protected(ca->bucket_gens, 1);
2105 size_t n = min(bucket_gens->nbuckets, old_bucket_gens->nbuckets);
2107 memcpy(bucket_gens->b,
2111 memcpy(buckets_nouse,
2113 BITS_TO_LONGS(n) * sizeof(unsigned long));
2116 rcu_assign_pointer(ca->bucket_gens, bucket_gens);
2117 bucket_gens = old_bucket_gens;
2119 swap(ca->buckets_nouse, buckets_nouse);
2121 nbuckets = ca->mi.nbuckets;
2124 percpu_up_write(&c->mark_lock);
2125 up_write(&ca->bucket_lock);
2126 up_write(&c->gc_lock);
2131 kvpfree(buckets_nouse,
2132 BITS_TO_LONGS(nbuckets) * sizeof(unsigned long));
2134 call_rcu(&bucket_gens->rcu, bucket_gens_free_rcu);
2139 void bch2_dev_buckets_free(struct bch_dev *ca)
2143 kvpfree(ca->buckets_nouse,
2144 BITS_TO_LONGS(ca->mi.nbuckets) * sizeof(unsigned long));
2145 kvpfree(rcu_dereference_protected(ca->bucket_gens, 1),
2146 sizeof(struct bucket_gens) + ca->mi.nbuckets);
2148 for (i = 0; i < ARRAY_SIZE(ca->usage); i++)
2149 free_percpu(ca->usage[i]);
2150 kfree(ca->usage_base);
2153 int bch2_dev_buckets_alloc(struct bch_fs *c, struct bch_dev *ca)
2157 ca->usage_base = kzalloc(sizeof(struct bch_dev_usage), GFP_KERNEL);
2158 if (!ca->usage_base)
2159 return -BCH_ERR_ENOMEM_usage_init;
2161 for (i = 0; i < ARRAY_SIZE(ca->usage); i++) {
2162 ca->usage[i] = alloc_percpu(struct bch_dev_usage);
2164 return -BCH_ERR_ENOMEM_usage_init;
2167 return bch2_dev_buckets_resize(c, ca, ca->mi.nbuckets);