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"
25 #include <linux/preempt.h>
26 #include <trace/events/bcachefs.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 static inline struct bch_fs_usage *fs_usage_ptr(struct bch_fs *c,
106 unsigned journal_seq,
109 percpu_rwsem_assert_held(&c->mark_lock);
110 BUG_ON(!gc && !journal_seq);
112 return this_cpu_ptr(gc
114 : c->usage[journal_seq & JOURNAL_BUF_MASK]);
117 u64 bch2_fs_usage_read_one(struct bch_fs *c, u64 *v)
119 ssize_t offset = v - (u64 *) c->usage_base;
123 BUG_ON(offset < 0 || offset >= fs_usage_u64s(c));
124 percpu_rwsem_assert_held(&c->mark_lock);
127 seq = read_seqcount_begin(&c->usage_lock);
130 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
131 ret += percpu_u64_get((u64 __percpu *) c->usage[i] + offset);
132 } while (read_seqcount_retry(&c->usage_lock, seq));
137 struct bch_fs_usage_online *bch2_fs_usage_read(struct bch_fs *c)
139 struct bch_fs_usage_online *ret;
140 unsigned seq, i, v, u64s = fs_usage_u64s(c) + 1;
142 ret = kmalloc(u64s * sizeof(u64), GFP_NOFS);
146 percpu_down_read(&c->mark_lock);
148 v = fs_usage_u64s(c) + 1;
149 if (unlikely(u64s != v)) {
151 percpu_up_read(&c->mark_lock);
156 ret->online_reserved = percpu_u64_get(c->online_reserved);
159 seq = read_seqcount_begin(&c->usage_lock);
160 unsafe_memcpy(&ret->u, c->usage_base, u64s * sizeof(u64),
161 "embedded variable length struct");
162 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
163 acc_u64s_percpu((u64 *) &ret->u, (u64 __percpu *) c->usage[i], u64s);
164 } while (read_seqcount_retry(&c->usage_lock, seq));
169 void bch2_fs_usage_acc_to_base(struct bch_fs *c, unsigned idx)
172 unsigned i, u64s = fs_usage_u64s(c);
174 BUG_ON(idx >= ARRAY_SIZE(c->usage));
177 write_seqcount_begin(&c->usage_lock);
179 acc_u64s_percpu((u64 *) c->usage_base,
180 (u64 __percpu *) c->usage[idx], u64s);
181 percpu_memset(c->usage[idx], 0, u64s * sizeof(u64));
184 for_each_member_device_rcu(ca, c, i, NULL) {
185 u64s = dev_usage_u64s();
187 acc_u64s_percpu((u64 *) ca->usage_base,
188 (u64 __percpu *) ca->usage[idx], u64s);
189 percpu_memset(ca->usage[idx], 0, u64s * sizeof(u64));
193 write_seqcount_end(&c->usage_lock);
197 void bch2_fs_usage_to_text(struct printbuf *out,
199 struct bch_fs_usage_online *fs_usage)
203 prt_printf(out, "capacity:\t\t\t%llu\n", c->capacity);
205 prt_printf(out, "hidden:\t\t\t\t%llu\n",
207 prt_printf(out, "data:\t\t\t\t%llu\n",
209 prt_printf(out, "cached:\t\t\t\t%llu\n",
211 prt_printf(out, "reserved:\t\t\t%llu\n",
212 fs_usage->u.reserved);
213 prt_printf(out, "nr_inodes:\t\t\t%llu\n",
214 fs_usage->u.nr_inodes);
215 prt_printf(out, "online reserved:\t\t%llu\n",
216 fs_usage->online_reserved);
219 i < ARRAY_SIZE(fs_usage->u.persistent_reserved);
221 prt_printf(out, "%u replicas:\n", i + 1);
222 prt_printf(out, "\treserved:\t\t%llu\n",
223 fs_usage->u.persistent_reserved[i]);
226 for (i = 0; i < c->replicas.nr; i++) {
227 struct bch_replicas_entry *e =
228 cpu_replicas_entry(&c->replicas, i);
230 prt_printf(out, "\t");
231 bch2_replicas_entry_to_text(out, e);
232 prt_printf(out, ":\t%llu\n", fs_usage->u.replicas[i]);
236 static u64 reserve_factor(u64 r)
238 return r + (round_up(r, (1 << RESERVE_FACTOR)) >> RESERVE_FACTOR);
241 u64 bch2_fs_sectors_used(struct bch_fs *c, struct bch_fs_usage_online *fs_usage)
243 return min(fs_usage->u.hidden +
246 reserve_factor(fs_usage->u.reserved +
247 fs_usage->online_reserved),
251 static struct bch_fs_usage_short
252 __bch2_fs_usage_read_short(struct bch_fs *c)
254 struct bch_fs_usage_short ret;
257 ret.capacity = c->capacity -
258 bch2_fs_usage_read_one(c, &c->usage_base->hidden);
260 data = bch2_fs_usage_read_one(c, &c->usage_base->data) +
261 bch2_fs_usage_read_one(c, &c->usage_base->btree);
262 reserved = bch2_fs_usage_read_one(c, &c->usage_base->reserved) +
263 percpu_u64_get(c->online_reserved);
265 ret.used = min(ret.capacity, data + reserve_factor(reserved));
266 ret.free = ret.capacity - ret.used;
268 ret.nr_inodes = bch2_fs_usage_read_one(c, &c->usage_base->nr_inodes);
273 struct bch_fs_usage_short
274 bch2_fs_usage_read_short(struct bch_fs *c)
276 struct bch_fs_usage_short ret;
278 percpu_down_read(&c->mark_lock);
279 ret = __bch2_fs_usage_read_short(c);
280 percpu_up_read(&c->mark_lock);
285 void bch2_dev_usage_init(struct bch_dev *ca)
287 ca->usage_base->d[BCH_DATA_free].buckets = ca->mi.nbuckets - ca->mi.first_bucket;
290 static inline int bucket_sectors_fragmented(struct bch_dev *ca,
291 struct bch_alloc_v4 a)
293 return a.dirty_sectors
294 ? max(0, (int) ca->mi.bucket_size - (int) a.dirty_sectors)
298 static void bch2_dev_usage_update(struct bch_fs *c, struct bch_dev *ca,
299 struct bch_alloc_v4 old,
300 struct bch_alloc_v4 new,
301 u64 journal_seq, bool gc)
303 struct bch_fs_usage *fs_usage;
304 struct bch_dev_usage *u;
307 fs_usage = fs_usage_ptr(c, journal_seq, gc);
309 if (data_type_is_hidden(old.data_type))
310 fs_usage->hidden -= ca->mi.bucket_size;
311 if (data_type_is_hidden(new.data_type))
312 fs_usage->hidden += ca->mi.bucket_size;
314 u = dev_usage_ptr(ca, journal_seq, gc);
316 u->d[old.data_type].buckets--;
317 u->d[new.data_type].buckets++;
319 u->buckets_ec -= (int) !!old.stripe;
320 u->buckets_ec += (int) !!new.stripe;
322 u->d[old.data_type].sectors -= old.dirty_sectors;
323 u->d[new.data_type].sectors += new.dirty_sectors;
325 u->d[BCH_DATA_cached].sectors += new.cached_sectors;
326 u->d[BCH_DATA_cached].sectors -= old.cached_sectors;
328 u->d[old.data_type].fragmented -= bucket_sectors_fragmented(ca, old);
329 u->d[new.data_type].fragmented += bucket_sectors_fragmented(ca, new);
334 static void bch2_dev_usage_update_m(struct bch_fs *c, struct bch_dev *ca,
335 struct bucket old, struct bucket new,
336 u64 journal_seq, bool gc)
338 struct bch_alloc_v4 old_a = {
340 .data_type = old.data_type,
341 .dirty_sectors = old.dirty_sectors,
342 .cached_sectors = old.cached_sectors,
343 .stripe = old.stripe,
345 struct bch_alloc_v4 new_a = {
347 .data_type = new.data_type,
348 .dirty_sectors = new.dirty_sectors,
349 .cached_sectors = new.cached_sectors,
350 .stripe = new.stripe,
353 bch2_dev_usage_update(c, ca, old_a, new_a, journal_seq, gc);
356 static inline int __update_replicas(struct bch_fs *c,
357 struct bch_fs_usage *fs_usage,
358 struct bch_replicas_entry *r,
361 int idx = bch2_replicas_entry_idx(c, r);
366 fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
367 fs_usage->replicas[idx] += sectors;
371 static inline int update_replicas(struct bch_fs *c, struct bkey_s_c k,
372 struct bch_replicas_entry *r, s64 sectors,
373 unsigned journal_seq, bool gc)
375 struct bch_fs_usage __percpu *fs_usage;
377 struct printbuf buf = PRINTBUF;
379 percpu_down_read(&c->mark_lock);
382 idx = bch2_replicas_entry_idx(c, r);
384 fsck_err(c, "no replicas entry\n"
386 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
387 percpu_up_read(&c->mark_lock);
388 ret = bch2_mark_replicas(c, r);
389 percpu_down_read(&c->mark_lock);
393 idx = bch2_replicas_entry_idx(c, r);
401 fs_usage = fs_usage_ptr(c, journal_seq, gc);
402 fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
403 fs_usage->replicas[idx] += sectors;
407 percpu_up_read(&c->mark_lock);
412 static inline int update_cached_sectors(struct bch_fs *c,
414 unsigned dev, s64 sectors,
415 unsigned journal_seq, bool gc)
417 struct bch_replicas_padded r;
419 bch2_replicas_entry_cached(&r.e, dev);
421 return update_replicas(c, k, &r.e, sectors, journal_seq, gc);
424 static struct replicas_delta_list *
425 replicas_deltas_realloc(struct btree_trans *trans, unsigned more)
427 struct replicas_delta_list *d = trans->fs_usage_deltas;
428 unsigned new_size = d ? (d->size + more) * 2 : 128;
429 unsigned alloc_size = sizeof(*d) + new_size;
431 WARN_ON_ONCE(alloc_size > REPLICAS_DELTA_LIST_MAX);
433 if (!d || d->used + more > d->size) {
434 d = krealloc(d, alloc_size, GFP_NOIO|__GFP_ZERO);
436 BUG_ON(!d && alloc_size > REPLICAS_DELTA_LIST_MAX);
439 d = mempool_alloc(&trans->c->replicas_delta_pool, GFP_NOIO);
440 memset(d, 0, REPLICAS_DELTA_LIST_MAX);
442 if (trans->fs_usage_deltas)
443 memcpy(d, trans->fs_usage_deltas,
444 trans->fs_usage_deltas->size + sizeof(*d));
446 new_size = REPLICAS_DELTA_LIST_MAX - sizeof(*d);
447 kfree(trans->fs_usage_deltas);
451 trans->fs_usage_deltas = d;
456 static inline void update_replicas_list(struct btree_trans *trans,
457 struct bch_replicas_entry *r,
460 struct replicas_delta_list *d;
461 struct replicas_delta *n;
467 b = replicas_entry_bytes(r) + 8;
468 d = replicas_deltas_realloc(trans, b);
470 n = (void *) d->d + d->used;
472 memcpy((void *) n + offsetof(struct replicas_delta, r),
473 r, replicas_entry_bytes(r));
474 bch2_replicas_entry_sort(&n->r);
478 static inline void update_cached_sectors_list(struct btree_trans *trans,
479 unsigned dev, s64 sectors)
481 struct bch_replicas_padded r;
483 bch2_replicas_entry_cached(&r.e, dev);
485 update_replicas_list(trans, &r.e, sectors);
488 int bch2_mark_alloc(struct btree_trans *trans,
489 struct bkey_s_c old, struct bkey_s_c new,
492 bool gc = flags & BTREE_TRIGGER_GC;
493 u64 journal_seq = trans->journal_res.seq;
494 u64 bucket_journal_seq;
495 struct bch_fs *c = trans->c;
496 struct bch_alloc_v4 old_a_convert, new_a_convert;
497 const struct bch_alloc_v4 *old_a, *new_a;
502 * alloc btree is read in by bch2_alloc_read, not gc:
504 if ((flags & BTREE_TRIGGER_GC) &&
505 !(flags & BTREE_TRIGGER_BUCKET_INVALIDATE))
508 if (bch2_trans_inconsistent_on(!bch2_dev_bucket_exists(c, new.k->p), trans,
509 "alloc key for invalid device or bucket"))
512 ca = bch_dev_bkey_exists(c, new.k->p.inode);
514 old_a = bch2_alloc_to_v4(old, &old_a_convert);
515 new_a = bch2_alloc_to_v4(new, &new_a_convert);
517 bucket_journal_seq = new_a->journal_seq;
519 if ((flags & BTREE_TRIGGER_INSERT) &&
520 data_type_is_empty(old_a->data_type) !=
521 data_type_is_empty(new_a->data_type) &&
522 new.k->type == KEY_TYPE_alloc_v4) {
523 struct bch_alloc_v4 *v = (struct bch_alloc_v4 *) new.v;
525 EBUG_ON(!journal_seq);
528 * If the btree updates referring to a bucket weren't flushed
529 * before the bucket became empty again, then the we don't have
530 * to wait on a journal flush before we can reuse the bucket:
532 v->journal_seq = bucket_journal_seq =
533 data_type_is_empty(new_a->data_type) &&
534 (journal_seq == v->journal_seq ||
535 bch2_journal_noflush_seq(&c->journal, v->journal_seq))
539 if (!data_type_is_empty(old_a->data_type) &&
540 data_type_is_empty(new_a->data_type) &&
541 bucket_journal_seq) {
542 ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
543 c->journal.flushed_seq_ondisk,
544 new.k->p.inode, new.k->p.offset,
547 bch2_fs_fatal_error(c,
548 "error setting bucket_needs_journal_commit: %i", ret);
553 percpu_down_read(&c->mark_lock);
554 if (!gc && new_a->gen != old_a->gen)
555 *bucket_gen(ca, new.k->p.offset) = new_a->gen;
557 bch2_dev_usage_update(c, ca, *old_a, *new_a, journal_seq, gc);
560 struct bucket *g = gc_bucket(ca, new.k->p.offset);
566 g->data_type = new_a->data_type;
567 g->stripe = new_a->stripe;
568 g->stripe_redundancy = new_a->stripe_redundancy;
569 g->dirty_sectors = new_a->dirty_sectors;
570 g->cached_sectors = new_a->cached_sectors;
574 percpu_up_read(&c->mark_lock);
577 * need to know if we're getting called from the invalidate path or
581 if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
582 old_a->cached_sectors) {
583 ret = update_cached_sectors(c, new, ca->dev_idx,
584 -((s64) old_a->cached_sectors),
587 bch2_fs_fatal_error(c, "%s(): no replicas entry while updating cached sectors",
593 if (new_a->data_type == BCH_DATA_free &&
594 (!new_a->journal_seq || new_a->journal_seq < c->journal.flushed_seq_ondisk))
595 closure_wake_up(&c->freelist_wait);
597 if (new_a->data_type == BCH_DATA_need_discard &&
598 (!bucket_journal_seq || bucket_journal_seq < c->journal.flushed_seq_ondisk))
601 if (old_a->data_type != BCH_DATA_cached &&
602 new_a->data_type == BCH_DATA_cached &&
603 should_invalidate_buckets(ca, bch2_dev_usage_read(ca)))
604 bch2_do_invalidates(c);
606 if (new_a->data_type == BCH_DATA_need_gc_gens)
612 int bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca,
613 size_t b, enum bch_data_type data_type,
614 unsigned sectors, struct gc_pos pos,
617 struct bucket old, new, *g;
620 BUG_ON(!(flags & BTREE_TRIGGER_GC));
621 BUG_ON(data_type != BCH_DATA_sb &&
622 data_type != BCH_DATA_journal);
625 * Backup superblock might be past the end of our normal usable space:
627 if (b >= ca->mi.nbuckets)
630 percpu_down_read(&c->mark_lock);
631 g = gc_bucket(ca, b);
636 if (bch2_fs_inconsistent_on(g->data_type &&
637 g->data_type != data_type, c,
638 "different types of data in same bucket: %s, %s",
639 bch2_data_types[g->data_type],
640 bch2_data_types[data_type])) {
645 if (bch2_fs_inconsistent_on((u64) g->dirty_sectors + sectors > ca->mi.bucket_size, c,
646 "bucket %u:%zu gen %u data type %s sector count overflow: %u + %u > bucket size",
647 ca->dev_idx, b, g->gen,
648 bch2_data_types[g->data_type ?: data_type],
649 g->dirty_sectors, sectors)) {
655 g->data_type = data_type;
656 g->dirty_sectors += sectors;
661 bch2_dev_usage_update_m(c, ca, old, new, 0, true);
662 percpu_up_read(&c->mark_lock);
666 static int check_bucket_ref(struct btree_trans *trans,
668 const struct bch_extent_ptr *ptr,
669 s64 sectors, enum bch_data_type ptr_data_type,
670 u8 b_gen, u8 bucket_data_type,
671 u32 dirty_sectors, u32 cached_sectors)
673 struct bch_fs *c = trans->c;
674 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
675 size_t bucket_nr = PTR_BUCKET_NR(ca, ptr);
676 u16 bucket_sectors = !ptr->cached
679 struct printbuf buf = PRINTBUF;
682 if (bucket_data_type == BCH_DATA_cached)
683 bucket_data_type = BCH_DATA_user;
685 if ((bucket_data_type == BCH_DATA_stripe && ptr_data_type == BCH_DATA_user) ||
686 (bucket_data_type == BCH_DATA_user && ptr_data_type == BCH_DATA_stripe))
687 bucket_data_type = ptr_data_type = BCH_DATA_stripe;
689 if (gen_after(ptr->gen, b_gen)) {
690 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
691 "bucket %u:%zu gen %u data type %s: ptr gen %u newer than bucket gen\n"
693 ptr->dev, bucket_nr, b_gen,
694 bch2_data_types[bucket_data_type ?: ptr_data_type],
696 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
701 if (gen_cmp(b_gen, ptr->gen) > BUCKET_GC_GEN_MAX) {
702 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
703 "bucket %u:%zu gen %u data type %s: ptr gen %u too stale\n"
705 ptr->dev, bucket_nr, b_gen,
706 bch2_data_types[bucket_data_type ?: ptr_data_type],
708 (printbuf_reset(&buf),
709 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
714 if (b_gen != ptr->gen && !ptr->cached) {
715 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
716 "bucket %u:%zu gen %u (mem gen %u) data type %s: stale dirty ptr (gen %u)\n"
718 ptr->dev, bucket_nr, b_gen,
719 *bucket_gen(ca, bucket_nr),
720 bch2_data_types[bucket_data_type ?: ptr_data_type],
722 (printbuf_reset(&buf),
723 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
728 if (b_gen != ptr->gen) {
733 if (!data_type_is_empty(bucket_data_type) &&
735 bucket_data_type != ptr_data_type) {
736 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
737 "bucket %u:%zu gen %u different types of data in same bucket: %s, %s\n"
739 ptr->dev, bucket_nr, b_gen,
740 bch2_data_types[bucket_data_type],
741 bch2_data_types[ptr_data_type],
742 (printbuf_reset(&buf),
743 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
748 if ((unsigned) (bucket_sectors + sectors) > U32_MAX) {
749 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
750 "bucket %u:%zu gen %u data type %s sector count overflow: %u + %lli > U16_MAX\n"
752 ptr->dev, bucket_nr, b_gen,
753 bch2_data_types[bucket_data_type ?: ptr_data_type],
754 bucket_sectors, sectors,
755 (printbuf_reset(&buf),
756 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
764 bch2_dump_trans_updates(trans);
768 static int mark_stripe_bucket(struct btree_trans *trans,
773 struct bch_fs *c = trans->c;
774 u64 journal_seq = trans->journal_res.seq;
775 const struct bch_stripe *s = bkey_s_c_to_stripe(k).v;
776 unsigned nr_data = s->nr_blocks - s->nr_redundant;
777 bool parity = ptr_idx >= nr_data;
778 enum bch_data_type data_type = parity ? BCH_DATA_parity : 0;
779 s64 sectors = parity ? le16_to_cpu(s->sectors) : 0;
780 const struct bch_extent_ptr *ptr = s->ptrs + ptr_idx;
781 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
782 struct bucket old, new, *g;
783 struct printbuf buf = PRINTBUF;
786 BUG_ON(!(flags & BTREE_TRIGGER_GC));
788 /* * XXX doesn't handle deletion */
790 percpu_down_read(&c->mark_lock);
792 g = PTR_GC_BUCKET(ca, ptr);
794 if (g->dirty_sectors ||
795 (g->stripe && g->stripe != k.k->p.offset)) {
796 bch2_fs_inconsistent(c,
797 "bucket %u:%zu gen %u: multiple stripes using same bucket\n%s",
798 ptr->dev, PTR_BUCKET_NR(ca, ptr), g->gen,
799 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
807 ret = check_bucket_ref(trans, k, ptr, sectors, data_type,
808 g->gen, g->data_type,
809 g->dirty_sectors, g->cached_sectors);
814 g->data_type = data_type;
815 g->dirty_sectors += sectors;
817 g->stripe = k.k->p.offset;
818 g->stripe_redundancy = s->nr_redundant;
823 bch2_dev_usage_update_m(c, ca, old, new, journal_seq, true);
824 percpu_up_read(&c->mark_lock);
829 static int __mark_pointer(struct btree_trans *trans,
831 const struct bch_extent_ptr *ptr,
832 s64 sectors, enum bch_data_type ptr_data_type,
833 u8 bucket_gen, u8 *bucket_data_type,
834 u32 *dirty_sectors, u32 *cached_sectors)
836 u32 *dst_sectors = !ptr->cached
839 int ret = check_bucket_ref(trans, k, ptr, sectors, ptr_data_type,
840 bucket_gen, *bucket_data_type,
841 *dirty_sectors, *cached_sectors);
846 *dst_sectors += sectors;
847 *bucket_data_type = *dirty_sectors || *cached_sectors
852 static int bch2_mark_pointer(struct btree_trans *trans,
854 struct extent_ptr_decoded p,
855 s64 sectors, enum bch_data_type data_type,
858 u64 journal_seq = trans->journal_res.seq;
859 struct bch_fs *c = trans->c;
860 struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
861 struct bucket old, new, *g;
865 BUG_ON(!(flags & BTREE_TRIGGER_GC));
867 percpu_down_read(&c->mark_lock);
868 g = PTR_GC_BUCKET(ca, &p.ptr);
872 bucket_data_type = g->data_type;
873 ret = __mark_pointer(trans, k, &p.ptr, sectors,
879 g->data_type = bucket_data_type;
884 bch2_dev_usage_update_m(c, ca, old, new, journal_seq, true);
885 percpu_up_read(&c->mark_lock);
890 static int bch2_mark_stripe_ptr(struct btree_trans *trans,
892 struct bch_extent_stripe_ptr p,
893 enum bch_data_type data_type,
897 struct bch_fs *c = trans->c;
898 struct bch_replicas_padded r;
901 BUG_ON(!(flags & BTREE_TRIGGER_GC));
903 m = genradix_ptr_alloc(&c->gc_stripes, p.idx, GFP_KERNEL);
905 bch_err(c, "error allocating memory for gc_stripes, idx %llu",
910 mutex_lock(&c->ec_stripes_heap_lock);
912 if (!m || !m->alive) {
913 mutex_unlock(&c->ec_stripes_heap_lock);
914 bch_err_ratelimited(c, "pointer to nonexistent stripe %llu",
916 bch2_inconsistent_error(c);
920 m->block_sectors[p.block] += sectors;
923 mutex_unlock(&c->ec_stripes_heap_lock);
925 r.e.data_type = data_type;
926 update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, true);
931 int bch2_mark_extent(struct btree_trans *trans,
932 struct bkey_s_c old, struct bkey_s_c new,
935 u64 journal_seq = trans->journal_res.seq;
936 struct bch_fs *c = trans->c;
937 struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old : new;
938 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
939 const union bch_extent_entry *entry;
940 struct extent_ptr_decoded p;
941 struct bch_replicas_padded r;
942 enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
945 s64 sectors = bkey_is_btree_ptr(k.k)
948 s64 dirty_sectors = 0;
952 BUG_ON(!(flags & BTREE_TRIGGER_GC));
954 r.e.data_type = data_type;
958 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
959 s64 disk_sectors = ptr_disk_sectors(sectors, p);
961 if (flags & BTREE_TRIGGER_OVERWRITE)
962 disk_sectors = -disk_sectors;
964 ret = bch2_mark_pointer(trans, k, p, disk_sectors,
973 ret = update_cached_sectors(c, k, p.ptr.dev,
974 disk_sectors, journal_seq, true);
976 bch2_fs_fatal_error(c, "%s(): no replicas entry while updating cached sectors",
981 } else if (!p.has_ec) {
982 dirty_sectors += disk_sectors;
983 r.e.devs[r.e.nr_devs++] = p.ptr.dev;
985 ret = bch2_mark_stripe_ptr(trans, k, p.ec, data_type,
986 disk_sectors, flags);
991 * There may be other dirty pointers in this extent, but
992 * if so they're not required for mounting if we have an
993 * erasure coded pointer in this extent:
1000 ret = update_replicas(c, k, &r.e, dirty_sectors, journal_seq, true);
1002 struct printbuf buf = PRINTBUF;
1004 bch2_bkey_val_to_text(&buf, c, k);
1005 bch2_fs_fatal_error(c, "%s(): no replicas entry for %s", __func__, buf.buf);
1006 printbuf_exit(&buf);
1014 int bch2_mark_stripe(struct btree_trans *trans,
1015 struct bkey_s_c old, struct bkey_s_c new,
1018 bool gc = flags & BTREE_TRIGGER_GC;
1019 u64 journal_seq = trans->journal_res.seq;
1020 struct bch_fs *c = trans->c;
1021 u64 idx = new.k->p.offset;
1022 const struct bch_stripe *old_s = old.k->type == KEY_TYPE_stripe
1023 ? bkey_s_c_to_stripe(old).v : NULL;
1024 const struct bch_stripe *new_s = new.k->type == KEY_TYPE_stripe
1025 ? bkey_s_c_to_stripe(new).v : NULL;
1029 BUG_ON(gc && old_s);
1032 struct stripe *m = genradix_ptr(&c->stripes, idx);
1035 struct printbuf buf1 = PRINTBUF;
1036 struct printbuf buf2 = PRINTBUF;
1038 bch2_bkey_val_to_text(&buf1, c, old);
1039 bch2_bkey_val_to_text(&buf2, c, new);
1040 bch_err_ratelimited(c, "error marking nonexistent stripe %llu while marking\n"
1042 "new %s", idx, buf1.buf, buf2.buf);
1043 printbuf_exit(&buf2);
1044 printbuf_exit(&buf1);
1045 bch2_inconsistent_error(c);
1050 bch2_stripes_heap_del(c, m, idx);
1052 memset(m, 0, sizeof(*m));
1054 m->sectors = le16_to_cpu(new_s->sectors);
1055 m->algorithm = new_s->algorithm;
1056 m->nr_blocks = new_s->nr_blocks;
1057 m->nr_redundant = new_s->nr_redundant;
1058 m->blocks_nonempty = 0;
1060 for (i = 0; i < new_s->nr_blocks; i++)
1061 m->blocks_nonempty += !!stripe_blockcount_get(new_s, i);
1064 bch2_stripes_heap_insert(c, m, idx);
1066 bch2_stripes_heap_update(c, m, idx);
1069 struct gc_stripe *m =
1070 genradix_ptr_alloc(&c->gc_stripes, idx, GFP_KERNEL);
1073 bch_err(c, "error allocating memory for gc_stripes, idx %llu",
1078 * This will be wrong when we bring back runtime gc: we should
1079 * be unmarking the old key and then marking the new key
1082 m->sectors = le16_to_cpu(new_s->sectors);
1083 m->nr_blocks = new_s->nr_blocks;
1084 m->nr_redundant = new_s->nr_redundant;
1086 for (i = 0; i < new_s->nr_blocks; i++)
1087 m->ptrs[i] = new_s->ptrs[i];
1089 bch2_bkey_to_replicas(&m->r.e, new);
1092 * gc recalculates this field from stripe ptr
1095 memset(m->block_sectors, 0, sizeof(m->block_sectors));
1097 for (i = 0; i < new_s->nr_blocks; i++) {
1098 ret = mark_stripe_bucket(trans, new, i, flags);
1103 ret = update_replicas(c, new, &m->r.e,
1104 ((s64) m->sectors * m->nr_redundant),
1107 struct printbuf buf = PRINTBUF;
1109 bch2_bkey_val_to_text(&buf, c, new);
1110 bch2_fs_fatal_error(c, "no replicas entry for %s", buf.buf);
1111 printbuf_exit(&buf);
1119 int bch2_mark_inode(struct btree_trans *trans,
1120 struct bkey_s_c old, struct bkey_s_c new,
1123 struct bch_fs *c = trans->c;
1124 struct bch_fs_usage __percpu *fs_usage;
1125 u64 journal_seq = trans->journal_res.seq;
1127 if (flags & BTREE_TRIGGER_INSERT) {
1128 struct bch_inode_v3 *v = (struct bch_inode_v3 *) new.v;
1130 BUG_ON(!journal_seq);
1131 BUG_ON(new.k->type != KEY_TYPE_inode_v3);
1133 v->bi_journal_seq = cpu_to_le64(journal_seq);
1136 if (flags & BTREE_TRIGGER_GC) {
1137 percpu_down_read(&c->mark_lock);
1140 fs_usage = fs_usage_ptr(c, journal_seq, flags & BTREE_TRIGGER_GC);
1141 fs_usage->nr_inodes += bkey_is_inode(new.k);
1142 fs_usage->nr_inodes -= bkey_is_inode(old.k);
1145 percpu_up_read(&c->mark_lock);
1150 int bch2_mark_reservation(struct btree_trans *trans,
1151 struct bkey_s_c old, struct bkey_s_c new,
1154 struct bch_fs *c = trans->c;
1155 struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old : new;
1156 struct bch_fs_usage __percpu *fs_usage;
1157 unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
1158 s64 sectors = (s64) k.k->size;
1160 BUG_ON(!(flags & BTREE_TRIGGER_GC));
1162 if (flags & BTREE_TRIGGER_OVERWRITE)
1164 sectors *= replicas;
1166 percpu_down_read(&c->mark_lock);
1169 fs_usage = fs_usage_ptr(c, trans->journal_res.seq, flags & BTREE_TRIGGER_GC);
1170 replicas = clamp_t(unsigned, replicas, 1,
1171 ARRAY_SIZE(fs_usage->persistent_reserved));
1173 fs_usage->reserved += sectors;
1174 fs_usage->persistent_reserved[replicas - 1] += sectors;
1177 percpu_up_read(&c->mark_lock);
1182 static s64 __bch2_mark_reflink_p(struct btree_trans *trans,
1183 struct bkey_s_c_reflink_p p,
1185 u64 *idx, unsigned flags, size_t r_idx)
1187 struct bch_fs *c = trans->c;
1188 struct reflink_gc *r;
1189 int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
1192 struct printbuf buf = PRINTBUF;
1194 if (r_idx >= c->reflink_gc_nr)
1197 r = genradix_ptr(&c->reflink_gc_table, r_idx);
1198 next_idx = min(next_idx, r->offset - r->size);
1199 if (*idx < next_idx)
1202 BUG_ON((s64) r->refcount + add < 0);
1208 if (fsck_err(c, "pointer to missing indirect extent\n"
1210 " missing range %llu-%llu",
1211 (bch2_bkey_val_to_text(&buf, c, p.s_c), buf.buf),
1213 struct bkey_i_error *new;
1215 new = bch2_trans_kmalloc(trans, sizeof(*new));
1216 ret = PTR_ERR_OR_ZERO(new);
1221 new->k.type = KEY_TYPE_error;
1222 new->k.p = bkey_start_pos(p.k);
1223 new->k.p.offset += *idx - start;
1224 bch2_key_resize(&new->k, next_idx - *idx);
1225 ret = __bch2_btree_insert(trans, BTREE_ID_extents, &new->k_i,
1226 BTREE_TRIGGER_NORUN);
1232 printbuf_exit(&buf);
1236 int bch2_mark_reflink_p(struct btree_trans *trans,
1237 struct bkey_s_c old, struct bkey_s_c new,
1240 struct bch_fs *c = trans->c;
1241 struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old : new;
1242 struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
1243 struct reflink_gc *ref;
1245 u64 idx = le64_to_cpu(p.v->idx), start = idx;
1246 u64 end = le64_to_cpu(p.v->idx) + p.k->size;
1249 BUG_ON(!(flags & BTREE_TRIGGER_GC));
1251 if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix) {
1252 idx -= le32_to_cpu(p.v->front_pad);
1253 end += le32_to_cpu(p.v->back_pad);
1257 r = c->reflink_gc_nr;
1259 m = l + (r - l) / 2;
1261 ref = genradix_ptr(&c->reflink_gc_table, m);
1262 if (ref->offset <= idx)
1268 while (idx < end && !ret)
1269 ret = __bch2_mark_reflink_p(trans, p, start, end,
1275 void bch2_trans_fs_usage_revert(struct btree_trans *trans,
1276 struct replicas_delta_list *deltas)
1278 struct bch_fs *c = trans->c;
1279 struct bch_fs_usage *dst;
1280 struct replicas_delta *d, *top = (void *) deltas->d + deltas->used;
1284 percpu_down_read(&c->mark_lock);
1286 dst = fs_usage_ptr(c, trans->journal_res.seq, false);
1288 /* revert changes: */
1289 for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
1290 switch (d->r.data_type) {
1291 case BCH_DATA_btree:
1293 case BCH_DATA_parity:
1296 BUG_ON(__update_replicas(c, dst, &d->r, -d->delta));
1299 dst->nr_inodes -= deltas->nr_inodes;
1301 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1302 added -= deltas->persistent_reserved[i];
1303 dst->reserved -= deltas->persistent_reserved[i];
1304 dst->persistent_reserved[i] -= deltas->persistent_reserved[i];
1308 trans->disk_res->sectors += added;
1309 this_cpu_add(*c->online_reserved, added);
1313 percpu_up_read(&c->mark_lock);
1316 int bch2_trans_fs_usage_apply(struct btree_trans *trans,
1317 struct replicas_delta_list *deltas)
1319 struct bch_fs *c = trans->c;
1320 static int warned_disk_usage = 0;
1322 unsigned disk_res_sectors = trans->disk_res ? trans->disk_res->sectors : 0;
1323 struct replicas_delta *d = deltas->d, *d2;
1324 struct replicas_delta *top = (void *) deltas->d + deltas->used;
1325 struct bch_fs_usage *dst;
1326 s64 added = 0, should_not_have_added;
1329 percpu_down_read(&c->mark_lock);
1331 dst = fs_usage_ptr(c, trans->journal_res.seq, false);
1333 for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
1334 switch (d->r.data_type) {
1335 case BCH_DATA_btree:
1337 case BCH_DATA_parity:
1341 if (__update_replicas(c, dst, &d->r, d->delta))
1345 dst->nr_inodes += deltas->nr_inodes;
1347 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1348 added += deltas->persistent_reserved[i];
1349 dst->reserved += deltas->persistent_reserved[i];
1350 dst->persistent_reserved[i] += deltas->persistent_reserved[i];
1354 * Not allowed to reduce sectors_available except by getting a
1357 should_not_have_added = added - (s64) disk_res_sectors;
1358 if (unlikely(should_not_have_added > 0)) {
1359 u64 old, new, v = atomic64_read(&c->sectors_available);
1363 new = max_t(s64, 0, old - should_not_have_added);
1364 } while ((v = atomic64_cmpxchg(&c->sectors_available,
1367 added -= should_not_have_added;
1372 trans->disk_res->sectors -= added;
1373 this_cpu_sub(*c->online_reserved, added);
1377 percpu_up_read(&c->mark_lock);
1379 if (unlikely(warn) && !xchg(&warned_disk_usage, 1))
1380 bch2_trans_inconsistent(trans,
1381 "disk usage increased %lli more than %u sectors reserved)",
1382 should_not_have_added, disk_res_sectors);
1385 /* revert changes: */
1386 for (d2 = deltas->d; d2 != d; d2 = replicas_delta_next(d2))
1387 BUG_ON(__update_replicas(c, dst, &d2->r, -d2->delta));
1390 percpu_up_read(&c->mark_lock);
1396 static inline int bch2_trans_mark_pointer(struct btree_trans *trans,
1397 enum btree_id btree_id, unsigned level,
1398 struct bkey_s_c k, struct extent_ptr_decoded p,
1401 bool insert = !(flags & BTREE_TRIGGER_OVERWRITE);
1402 struct btree_iter iter;
1403 struct bkey_i_alloc_v4 *a;
1404 struct bpos bucket_pos;
1405 struct bch_backpointer bp;
1409 bch2_extent_ptr_to_bp(trans->c, btree_id, level, k, p, &bucket_pos, &bp);
1410 sectors = bp.bucket_len;
1414 a = bch2_trans_start_alloc_update(trans, &iter, bucket_pos);
1418 ret = __mark_pointer(trans, k, &p.ptr, sectors, bp.data_type,
1419 a->v.gen, &a->v.data_type,
1420 &a->v.dirty_sectors, &a->v.cached_sectors);
1424 if (!p.ptr.cached) {
1425 ret = bch2_bucket_backpointer_mod(trans, a, bp, k, insert);
1430 ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
1432 bch2_trans_iter_exit(trans, &iter);
1436 static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
1437 struct extent_ptr_decoded p,
1438 s64 sectors, enum bch_data_type data_type)
1440 struct btree_iter iter;
1441 struct bkey_i_stripe *s;
1442 struct bch_replicas_padded r;
1445 bch2_trans_iter_init(trans, &iter, BTREE_ID_stripes, POS(0, p.ec.idx),
1447 BTREE_ITER_WITH_UPDATES);
1448 s = bch2_bkey_get_mut_typed(trans, &iter, stripe);
1449 ret = PTR_ERR_OR_ZERO(s);
1450 if (unlikely(ret)) {
1451 bch2_trans_inconsistent_on(ret == -ENOENT, trans,
1452 "pointer to nonexistent stripe %llu",
1457 if (!bch2_ptr_matches_stripe(&s->v, p)) {
1458 bch2_trans_inconsistent(trans,
1459 "stripe pointer doesn't match stripe %llu",
1465 stripe_blockcount_set(&s->v, p.ec.block,
1466 stripe_blockcount_get(&s->v, p.ec.block) +
1469 ret = bch2_trans_update(trans, &iter, &s->k_i, 0);
1473 bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(&s->k_i));
1474 r.e.data_type = data_type;
1475 update_replicas_list(trans, &r.e, sectors);
1477 bch2_trans_iter_exit(trans, &iter);
1481 int bch2_trans_mark_extent(struct btree_trans *trans,
1482 enum btree_id btree_id, unsigned level,
1483 struct bkey_s_c old, struct bkey_i *new,
1486 struct bch_fs *c = trans->c;
1487 struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE
1489 : bkey_i_to_s_c(new);
1490 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1491 const union bch_extent_entry *entry;
1492 struct extent_ptr_decoded p;
1493 struct bch_replicas_padded r;
1494 enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
1497 s64 sectors = bkey_is_btree_ptr(k.k)
1500 s64 dirty_sectors = 0;
1504 r.e.data_type = data_type;
1506 r.e.nr_required = 1;
1508 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
1509 s64 disk_sectors = ptr_disk_sectors(sectors, p);
1511 if (flags & BTREE_TRIGGER_OVERWRITE)
1512 disk_sectors = -disk_sectors;
1514 ret = bch2_trans_mark_pointer(trans, btree_id, level, k, p, flags);
1522 update_cached_sectors_list(trans, p.ptr.dev,
1524 } else if (!p.has_ec) {
1525 dirty_sectors += disk_sectors;
1526 r.e.devs[r.e.nr_devs++] = p.ptr.dev;
1528 ret = bch2_trans_mark_stripe_ptr(trans, p,
1529 disk_sectors, data_type);
1533 r.e.nr_required = 0;
1538 update_replicas_list(trans, &r.e, dirty_sectors);
1543 static int bch2_trans_mark_stripe_bucket(struct btree_trans *trans,
1544 struct bkey_s_c_stripe s,
1545 unsigned idx, bool deleting)
1547 struct bch_fs *c = trans->c;
1548 const struct bch_extent_ptr *ptr = &s.v->ptrs[idx];
1549 struct btree_iter iter;
1550 struct bkey_i_alloc_v4 *a;
1551 enum bch_data_type data_type = idx >= s.v->nr_blocks - s.v->nr_redundant
1552 ? BCH_DATA_parity : 0;
1553 s64 sectors = data_type ? le16_to_cpu(s.v->sectors) : 0;
1559 a = bch2_trans_start_alloc_update(trans, &iter, PTR_BUCKET_POS(c, ptr));
1563 ret = check_bucket_ref(trans, s.s_c, ptr, sectors, data_type,
1564 a->v.gen, a->v.data_type,
1565 a->v.dirty_sectors, a->v.cached_sectors);
1570 if (bch2_trans_inconsistent_on(a->v.stripe ||
1571 a->v.stripe_redundancy, trans,
1572 "bucket %llu:%llu gen %u data type %s dirty_sectors %u: multiple stripes using same bucket (%u, %llu)",
1573 iter.pos.inode, iter.pos.offset, a->v.gen,
1574 bch2_data_types[a->v.data_type],
1576 a->v.stripe, s.k->p.offset)) {
1581 if (bch2_trans_inconsistent_on(data_type && a->v.dirty_sectors, trans,
1582 "bucket %llu:%llu gen %u data type %s dirty_sectors %u: data already in stripe bucket %llu",
1583 iter.pos.inode, iter.pos.offset, a->v.gen,
1584 bch2_data_types[a->v.data_type],
1591 a->v.stripe = s.k->p.offset;
1592 a->v.stripe_redundancy = s.v->nr_redundant;
1594 if (bch2_trans_inconsistent_on(a->v.stripe != s.k->p.offset ||
1595 a->v.stripe_redundancy != s.v->nr_redundant, trans,
1596 "bucket %llu:%llu gen %u: not marked as stripe when deleting stripe %llu (got %u)",
1597 iter.pos.inode, iter.pos.offset, a->v.gen,
1598 s.k->p.offset, a->v.stripe)) {
1604 a->v.stripe_redundancy = 0;
1607 a->v.dirty_sectors += sectors;
1609 a->v.data_type = !deleting ? data_type : 0;
1611 ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
1615 bch2_trans_iter_exit(trans, &iter);
1619 int bch2_trans_mark_stripe(struct btree_trans *trans,
1620 enum btree_id btree_id, unsigned level,
1621 struct bkey_s_c old, struct bkey_i *new,
1624 const struct bch_stripe *old_s = NULL;
1625 struct bch_stripe *new_s = NULL;
1626 struct bch_replicas_padded r;
1627 unsigned i, nr_blocks;
1630 if (old.k->type == KEY_TYPE_stripe)
1631 old_s = bkey_s_c_to_stripe(old).v;
1632 if (new->k.type == KEY_TYPE_stripe)
1633 new_s = &bkey_i_to_stripe(new)->v;
1636 * If the pointers aren't changing, we don't need to do anything:
1638 if (new_s && old_s &&
1639 new_s->nr_blocks == old_s->nr_blocks &&
1640 new_s->nr_redundant == old_s->nr_redundant &&
1641 !memcmp(old_s->ptrs, new_s->ptrs,
1642 new_s->nr_blocks * sizeof(struct bch_extent_ptr)))
1645 BUG_ON(new_s && old_s &&
1646 (new_s->nr_blocks != old_s->nr_blocks ||
1647 new_s->nr_redundant != old_s->nr_redundant));
1649 nr_blocks = new_s ? new_s->nr_blocks : old_s->nr_blocks;
1652 s64 sectors = le16_to_cpu(new_s->sectors);
1654 bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(new));
1655 update_replicas_list(trans, &r.e, sectors * new_s->nr_redundant);
1659 s64 sectors = -((s64) le16_to_cpu(old_s->sectors));
1661 bch2_bkey_to_replicas(&r.e, old);
1662 update_replicas_list(trans, &r.e, sectors * old_s->nr_redundant);
1665 for (i = 0; i < nr_blocks; i++) {
1666 if (new_s && old_s &&
1667 !memcmp(&new_s->ptrs[i],
1669 sizeof(new_s->ptrs[i])))
1673 ret = bch2_trans_mark_stripe_bucket(trans,
1674 bkey_i_to_s_c_stripe(new), i, false);
1680 ret = bch2_trans_mark_stripe_bucket(trans,
1681 bkey_s_c_to_stripe(old), i, true);
1690 int bch2_trans_mark_inode(struct btree_trans *trans,
1691 enum btree_id btree_id, unsigned level,
1692 struct bkey_s_c old,
1696 int nr = bkey_is_inode(&new->k) - bkey_is_inode(old.k);
1699 struct replicas_delta_list *d =
1700 replicas_deltas_realloc(trans, 0);
1707 int bch2_trans_mark_reservation(struct btree_trans *trans,
1708 enum btree_id btree_id, unsigned level,
1709 struct bkey_s_c old,
1713 struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE
1715 : bkey_i_to_s_c(new);
1716 unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
1717 s64 sectors = (s64) k.k->size;
1718 struct replicas_delta_list *d;
1720 if (flags & BTREE_TRIGGER_OVERWRITE)
1722 sectors *= replicas;
1724 d = replicas_deltas_realloc(trans, 0);
1726 replicas = clamp_t(unsigned, replicas, 1,
1727 ARRAY_SIZE(d->persistent_reserved));
1729 d->persistent_reserved[replicas - 1] += sectors;
1733 static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
1734 struct bkey_s_c_reflink_p p,
1735 u64 *idx, unsigned flags)
1737 struct bch_fs *c = trans->c;
1738 struct btree_iter iter;
1741 int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
1742 struct printbuf buf = PRINTBUF;
1745 bch2_trans_iter_init(trans, &iter, BTREE_ID_reflink, POS(0, *idx),
1747 BTREE_ITER_WITH_UPDATES);
1748 k = bch2_bkey_get_mut(trans, &iter);
1749 ret = PTR_ERR_OR_ZERO(k);
1753 refcount = bkey_refcount(k);
1755 bch2_bkey_val_to_text(&buf, c, p.s_c);
1756 bch2_trans_inconsistent(trans,
1757 "nonexistent indirect extent at %llu while marking\n %s",
1763 if (!*refcount && (flags & BTREE_TRIGGER_OVERWRITE)) {
1764 bch2_bkey_val_to_text(&buf, c, p.s_c);
1765 bch2_trans_inconsistent(trans,
1766 "indirect extent refcount underflow at %llu while marking\n %s",
1772 if (flags & BTREE_TRIGGER_INSERT) {
1773 struct bch_reflink_p *v = (struct bch_reflink_p *) p.v;
1776 pad = max_t(s64, le32_to_cpu(v->front_pad),
1777 le64_to_cpu(v->idx) - bkey_start_offset(&k->k));
1778 BUG_ON(pad > U32_MAX);
1779 v->front_pad = cpu_to_le32(pad);
1781 pad = max_t(s64, le32_to_cpu(v->back_pad),
1782 k->k.p.offset - p.k->size - le64_to_cpu(v->idx));
1783 BUG_ON(pad > U32_MAX);
1784 v->back_pad = cpu_to_le32(pad);
1787 le64_add_cpu(refcount, add);
1789 bch2_btree_iter_set_pos_to_extent_start(&iter);
1790 ret = bch2_trans_update(trans, &iter, k, 0);
1794 *idx = k->k.p.offset;
1796 bch2_trans_iter_exit(trans, &iter);
1797 printbuf_exit(&buf);
1801 int bch2_trans_mark_reflink_p(struct btree_trans *trans,
1802 enum btree_id btree_id, unsigned level,
1803 struct bkey_s_c old,
1807 struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE
1809 : bkey_i_to_s_c(new);
1810 struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
1814 if (flags & BTREE_TRIGGER_INSERT) {
1815 struct bch_reflink_p *v = (struct bch_reflink_p *) p.v;
1817 v->front_pad = v->back_pad = 0;
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 = __bch2_trans_mark_reflink_p(trans, p, &idx, flags);
1830 static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
1831 struct bch_dev *ca, size_t b,
1832 enum bch_data_type type,
1835 struct bch_fs *c = trans->c;
1836 struct btree_iter iter;
1837 struct bkey_i_alloc_v4 *a;
1841 * Backup superblock might be past the end of our normal usable space:
1843 if (b >= ca->mi.nbuckets)
1846 a = bch2_trans_start_alloc_update(trans, &iter, POS(ca->dev_idx, b));
1850 if (a->v.data_type && a->v.data_type != type) {
1851 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
1852 "bucket %llu:%llu gen %u different types of data in same bucket: %s, %s\n"
1854 iter.pos.inode, iter.pos.offset, a->v.gen,
1855 bch2_data_types[a->v.data_type],
1856 bch2_data_types[type],
1857 bch2_data_types[type]);
1862 a->v.data_type = type;
1863 a->v.dirty_sectors = sectors;
1865 ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
1869 bch2_trans_iter_exit(trans, &iter);
1873 int bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
1874 struct bch_dev *ca, size_t b,
1875 enum bch_data_type type,
1878 return commit_do(trans, NULL, NULL, 0,
1879 __bch2_trans_mark_metadata_bucket(trans, ca, b, type, sectors));
1882 static int bch2_trans_mark_metadata_sectors(struct btree_trans *trans,
1885 enum bch_data_type type,
1886 u64 *bucket, unsigned *bucket_sectors)
1889 u64 b = sector_to_bucket(ca, start);
1891 min_t(u64, bucket_to_sector(ca, b + 1), end) - start;
1893 if (b != *bucket && *bucket_sectors) {
1894 int ret = bch2_trans_mark_metadata_bucket(trans, ca, *bucket,
1895 type, *bucket_sectors);
1899 *bucket_sectors = 0;
1903 *bucket_sectors += sectors;
1905 } while (start < end);
1910 static int __bch2_trans_mark_dev_sb(struct btree_trans *trans,
1913 struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
1915 unsigned i, bucket_sectors = 0;
1918 for (i = 0; i < layout->nr_superblocks; i++) {
1919 u64 offset = le64_to_cpu(layout->sb_offset[i]);
1921 if (offset == BCH_SB_SECTOR) {
1922 ret = bch2_trans_mark_metadata_sectors(trans, ca,
1924 BCH_DATA_sb, &bucket, &bucket_sectors);
1929 ret = bch2_trans_mark_metadata_sectors(trans, ca, offset,
1930 offset + (1 << layout->sb_max_size_bits),
1931 BCH_DATA_sb, &bucket, &bucket_sectors);
1936 if (bucket_sectors) {
1937 ret = bch2_trans_mark_metadata_bucket(trans, ca,
1938 bucket, BCH_DATA_sb, bucket_sectors);
1943 for (i = 0; i < ca->journal.nr; i++) {
1944 ret = bch2_trans_mark_metadata_bucket(trans, ca,
1945 ca->journal.buckets[i],
1946 BCH_DATA_journal, ca->mi.bucket_size);
1954 int bch2_trans_mark_dev_sb(struct bch_fs *c, struct bch_dev *ca)
1956 return bch2_trans_run(c, __bch2_trans_mark_dev_sb(&trans, ca));
1959 /* Disk reservations: */
1961 #define SECTORS_CACHE 1024
1963 int __bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
1964 u64 sectors, int flags)
1966 struct bch_fs_pcpu *pcpu;
1968 s64 sectors_available;
1971 percpu_down_read(&c->mark_lock);
1973 pcpu = this_cpu_ptr(c->pcpu);
1975 if (sectors <= pcpu->sectors_available)
1978 v = atomic64_read(&c->sectors_available);
1981 get = min((u64) sectors + SECTORS_CACHE, old);
1983 if (get < sectors) {
1987 } while ((v = atomic64_cmpxchg(&c->sectors_available,
1988 old, old - get)) != old);
1990 pcpu->sectors_available += get;
1993 pcpu->sectors_available -= sectors;
1994 this_cpu_add(*c->online_reserved, sectors);
1995 res->sectors += sectors;
1998 percpu_up_read(&c->mark_lock);
2002 mutex_lock(&c->sectors_available_lock);
2004 percpu_u64_set(&c->pcpu->sectors_available, 0);
2005 sectors_available = avail_factor(__bch2_fs_usage_read_short(c).free);
2007 if (sectors <= sectors_available ||
2008 (flags & BCH_DISK_RESERVATION_NOFAIL)) {
2009 atomic64_set(&c->sectors_available,
2010 max_t(s64, 0, sectors_available - sectors));
2011 this_cpu_add(*c->online_reserved, sectors);
2012 res->sectors += sectors;
2015 atomic64_set(&c->sectors_available, sectors_available);
2016 ret = -BCH_ERR_ENOSPC_disk_reservation;
2019 mutex_unlock(&c->sectors_available_lock);
2020 percpu_up_read(&c->mark_lock);
2025 /* Startup/shutdown: */
2027 static void bucket_gens_free_rcu(struct rcu_head *rcu)
2029 struct bucket_gens *buckets =
2030 container_of(rcu, struct bucket_gens, rcu);
2032 kvpfree(buckets, sizeof(*buckets) + buckets->nbuckets);
2035 int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
2037 struct bucket_gens *bucket_gens = NULL, *old_bucket_gens = NULL;
2038 unsigned long *buckets_nouse = NULL;
2039 bool resize = ca->bucket_gens != NULL;
2042 if (!(bucket_gens = kvpmalloc(sizeof(struct bucket_gens) + nbuckets,
2043 GFP_KERNEL|__GFP_ZERO)) ||
2044 (c->opts.buckets_nouse &&
2045 !(buckets_nouse = kvpmalloc(BITS_TO_LONGS(nbuckets) *
2046 sizeof(unsigned long),
2047 GFP_KERNEL|__GFP_ZERO))))
2050 bucket_gens->first_bucket = ca->mi.first_bucket;
2051 bucket_gens->nbuckets = nbuckets;
2053 bch2_copygc_stop(c);
2056 down_write(&c->gc_lock);
2057 down_write(&ca->bucket_lock);
2058 percpu_down_write(&c->mark_lock);
2061 old_bucket_gens = rcu_dereference_protected(ca->bucket_gens, 1);
2064 size_t n = min(bucket_gens->nbuckets, old_bucket_gens->nbuckets);
2066 memcpy(bucket_gens->b,
2070 memcpy(buckets_nouse,
2072 BITS_TO_LONGS(n) * sizeof(unsigned long));
2075 rcu_assign_pointer(ca->bucket_gens, bucket_gens);
2076 bucket_gens = old_bucket_gens;
2078 swap(ca->buckets_nouse, buckets_nouse);
2080 nbuckets = ca->mi.nbuckets;
2083 percpu_up_write(&c->mark_lock);
2084 up_write(&ca->bucket_lock);
2085 up_write(&c->gc_lock);
2090 kvpfree(buckets_nouse,
2091 BITS_TO_LONGS(nbuckets) * sizeof(unsigned long));
2093 call_rcu(&bucket_gens->rcu, bucket_gens_free_rcu);
2098 void bch2_dev_buckets_free(struct bch_dev *ca)
2102 kvpfree(ca->buckets_nouse,
2103 BITS_TO_LONGS(ca->mi.nbuckets) * sizeof(unsigned long));
2104 kvpfree(rcu_dereference_protected(ca->bucket_gens, 1),
2105 sizeof(struct bucket_gens) + ca->mi.nbuckets);
2107 for (i = 0; i < ARRAY_SIZE(ca->usage); i++)
2108 free_percpu(ca->usage[i]);
2109 kfree(ca->usage_base);
2112 int bch2_dev_buckets_alloc(struct bch_fs *c, struct bch_dev *ca)
2116 ca->usage_base = kzalloc(sizeof(struct bch_dev_usage), GFP_KERNEL);
2117 if (!ca->usage_base)
2120 for (i = 0; i < ARRAY_SIZE(ca->usage); i++) {
2121 ca->usage[i] = alloc_percpu(struct bch_dev_usage);
2126 return bch2_dev_buckets_resize(c, ca, ca->mi.nbuckets);