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 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 nr_replicas = READ_ONCE(c->replicas.nr);
143 ret = kmalloc(__fs_usage_online_u64s(nr_replicas) * sizeof(u64), GFP_KERNEL);
147 percpu_down_read(&c->mark_lock);
149 if (nr_replicas != c->replicas.nr) {
150 nr_replicas = c->replicas.nr;
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,
161 __fs_usage_u64s(nr_replicas) * sizeof(u64),
162 "embedded variable length struct");
163 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
164 acc_u64s_percpu((u64 *) &ret->u, (u64 __percpu *) c->usage[i],
165 __fs_usage_u64s(nr_replicas));
166 } while (read_seqcount_retry(&c->usage_lock, seq));
171 void bch2_fs_usage_acc_to_base(struct bch_fs *c, unsigned idx)
174 unsigned i, u64s = fs_usage_u64s(c);
176 BUG_ON(idx >= ARRAY_SIZE(c->usage));
179 write_seqcount_begin(&c->usage_lock);
181 acc_u64s_percpu((u64 *) c->usage_base,
182 (u64 __percpu *) c->usage[idx], u64s);
183 percpu_memset(c->usage[idx], 0, u64s * sizeof(u64));
186 for_each_member_device_rcu(ca, c, i, NULL) {
187 u64s = dev_usage_u64s();
189 acc_u64s_percpu((u64 *) ca->usage_base,
190 (u64 __percpu *) ca->usage[idx], u64s);
191 percpu_memset(ca->usage[idx], 0, u64s * sizeof(u64));
195 write_seqcount_end(&c->usage_lock);
199 void bch2_fs_usage_to_text(struct printbuf *out,
201 struct bch_fs_usage_online *fs_usage)
205 prt_printf(out, "capacity:\t\t\t%llu\n", c->capacity);
207 prt_printf(out, "hidden:\t\t\t\t%llu\n",
209 prt_printf(out, "data:\t\t\t\t%llu\n",
211 prt_printf(out, "cached:\t\t\t\t%llu\n",
213 prt_printf(out, "reserved:\t\t\t%llu\n",
214 fs_usage->u.reserved);
215 prt_printf(out, "nr_inodes:\t\t\t%llu\n",
216 fs_usage->u.nr_inodes);
217 prt_printf(out, "online reserved:\t\t%llu\n",
218 fs_usage->online_reserved);
221 i < ARRAY_SIZE(fs_usage->u.persistent_reserved);
223 prt_printf(out, "%u replicas:\n", i + 1);
224 prt_printf(out, "\treserved:\t\t%llu\n",
225 fs_usage->u.persistent_reserved[i]);
228 for (i = 0; i < c->replicas.nr; i++) {
229 struct bch_replicas_entry *e =
230 cpu_replicas_entry(&c->replicas, i);
232 prt_printf(out, "\t");
233 bch2_replicas_entry_to_text(out, e);
234 prt_printf(out, ":\t%llu\n", fs_usage->u.replicas[i]);
238 static u64 reserve_factor(u64 r)
240 return r + (round_up(r, (1 << RESERVE_FACTOR)) >> RESERVE_FACTOR);
243 u64 bch2_fs_sectors_used(struct bch_fs *c, struct bch_fs_usage_online *fs_usage)
245 return min(fs_usage->u.hidden +
248 reserve_factor(fs_usage->u.reserved +
249 fs_usage->online_reserved),
253 static struct bch_fs_usage_short
254 __bch2_fs_usage_read_short(struct bch_fs *c)
256 struct bch_fs_usage_short ret;
259 ret.capacity = c->capacity -
260 bch2_fs_usage_read_one(c, &c->usage_base->hidden);
262 data = bch2_fs_usage_read_one(c, &c->usage_base->data) +
263 bch2_fs_usage_read_one(c, &c->usage_base->btree);
264 reserved = bch2_fs_usage_read_one(c, &c->usage_base->reserved) +
265 percpu_u64_get(c->online_reserved);
267 ret.used = min(ret.capacity, data + reserve_factor(reserved));
268 ret.free = ret.capacity - ret.used;
270 ret.nr_inodes = bch2_fs_usage_read_one(c, &c->usage_base->nr_inodes);
275 struct bch_fs_usage_short
276 bch2_fs_usage_read_short(struct bch_fs *c)
278 struct bch_fs_usage_short ret;
280 percpu_down_read(&c->mark_lock);
281 ret = __bch2_fs_usage_read_short(c);
282 percpu_up_read(&c->mark_lock);
287 void bch2_dev_usage_init(struct bch_dev *ca)
289 ca->usage_base->d[BCH_DATA_free].buckets = ca->mi.nbuckets - ca->mi.first_bucket;
292 static inline int bucket_sectors_fragmented(struct bch_dev *ca,
293 struct bch_alloc_v4 a)
295 return a.dirty_sectors
296 ? max(0, (int) ca->mi.bucket_size - (int) a.dirty_sectors)
300 static void bch2_dev_usage_update(struct bch_fs *c, struct bch_dev *ca,
301 struct bch_alloc_v4 old,
302 struct bch_alloc_v4 new,
303 u64 journal_seq, bool gc)
305 struct bch_fs_usage *fs_usage;
306 struct bch_dev_usage *u;
309 fs_usage = fs_usage_ptr(c, journal_seq, gc);
311 if (data_type_is_hidden(old.data_type))
312 fs_usage->hidden -= ca->mi.bucket_size;
313 if (data_type_is_hidden(new.data_type))
314 fs_usage->hidden += ca->mi.bucket_size;
316 u = dev_usage_ptr(ca, journal_seq, gc);
318 u->d[old.data_type].buckets--;
319 u->d[new.data_type].buckets++;
321 u->buckets_ec -= (int) !!old.stripe;
322 u->buckets_ec += (int) !!new.stripe;
324 u->d[old.data_type].sectors -= old.dirty_sectors;
325 u->d[new.data_type].sectors += new.dirty_sectors;
327 u->d[BCH_DATA_cached].sectors += new.cached_sectors;
328 u->d[BCH_DATA_cached].sectors -= old.cached_sectors;
330 u->d[old.data_type].fragmented -= bucket_sectors_fragmented(ca, old);
331 u->d[new.data_type].fragmented += bucket_sectors_fragmented(ca, new);
336 static void bch2_dev_usage_update_m(struct bch_fs *c, struct bch_dev *ca,
337 struct bucket old, struct bucket new,
338 u64 journal_seq, bool gc)
340 struct bch_alloc_v4 old_a = {
342 .data_type = old.data_type,
343 .dirty_sectors = old.dirty_sectors,
344 .cached_sectors = old.cached_sectors,
345 .stripe = old.stripe,
347 struct bch_alloc_v4 new_a = {
349 .data_type = new.data_type,
350 .dirty_sectors = new.dirty_sectors,
351 .cached_sectors = new.cached_sectors,
352 .stripe = new.stripe,
355 bch2_dev_usage_update(c, ca, old_a, new_a, journal_seq, gc);
358 static inline int __update_replicas(struct bch_fs *c,
359 struct bch_fs_usage *fs_usage,
360 struct bch_replicas_entry *r,
363 int idx = bch2_replicas_entry_idx(c, r);
368 fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
369 fs_usage->replicas[idx] += sectors;
373 static inline int update_replicas(struct bch_fs *c, struct bkey_s_c k,
374 struct bch_replicas_entry *r, s64 sectors,
375 unsigned journal_seq, bool gc)
377 struct bch_fs_usage *fs_usage;
379 struct printbuf buf = PRINTBUF;
381 percpu_down_read(&c->mark_lock);
384 idx = bch2_replicas_entry_idx(c, r);
386 fsck_err(c, "no replicas entry\n"
388 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
389 percpu_up_read(&c->mark_lock);
390 ret = bch2_mark_replicas(c, r);
391 percpu_down_read(&c->mark_lock);
395 idx = bch2_replicas_entry_idx(c, r);
403 fs_usage = fs_usage_ptr(c, journal_seq, gc);
404 fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
405 fs_usage->replicas[idx] += sectors;
409 percpu_up_read(&c->mark_lock);
414 static inline int update_cached_sectors(struct bch_fs *c,
416 unsigned dev, s64 sectors,
417 unsigned journal_seq, bool gc)
419 struct bch_replicas_padded r;
421 bch2_replicas_entry_cached(&r.e, dev);
423 return update_replicas(c, k, &r.e, sectors, journal_seq, gc);
426 static int __replicas_deltas_realloc(struct btree_trans *trans, unsigned more,
429 struct replicas_delta_list *d = trans->fs_usage_deltas;
430 unsigned new_size = d ? (d->size + more) * 2 : 128;
431 unsigned alloc_size = sizeof(*d) + new_size;
433 WARN_ON_ONCE(alloc_size > REPLICAS_DELTA_LIST_MAX);
435 if (!d || d->used + more > d->size) {
436 d = krealloc(d, alloc_size, gfp|__GFP_ZERO);
439 if (alloc_size > REPLICAS_DELTA_LIST_MAX)
442 d = mempool_alloc(&trans->c->replicas_delta_pool, gfp);
446 memset(d, 0, REPLICAS_DELTA_LIST_MAX);
448 if (trans->fs_usage_deltas)
449 memcpy(d, trans->fs_usage_deltas,
450 trans->fs_usage_deltas->size + sizeof(*d));
452 new_size = REPLICAS_DELTA_LIST_MAX - sizeof(*d);
453 kfree(trans->fs_usage_deltas);
457 trans->fs_usage_deltas = d;
463 static int replicas_deltas_realloc(struct btree_trans *trans, unsigned more)
465 return allocate_dropping_locks_errcode(trans,
466 __replicas_deltas_realloc(trans, more, _gfp));
469 static inline int update_replicas_list(struct btree_trans *trans,
470 struct bch_replicas_entry *r,
473 struct replicas_delta_list *d;
474 struct replicas_delta *n;
481 b = replicas_entry_bytes(r) + 8;
482 ret = replicas_deltas_realloc(trans, b);
486 d = trans->fs_usage_deltas;
487 n = (void *) d->d + d->used;
489 memcpy((void *) n + offsetof(struct replicas_delta, r),
490 r, replicas_entry_bytes(r));
491 bch2_replicas_entry_sort(&n->r);
496 static inline int update_cached_sectors_list(struct btree_trans *trans,
497 unsigned dev, s64 sectors)
499 struct bch_replicas_padded r;
501 bch2_replicas_entry_cached(&r.e, dev);
503 return update_replicas_list(trans, &r.e, sectors);
506 int bch2_mark_alloc(struct btree_trans *trans,
507 enum btree_id btree, unsigned level,
508 struct bkey_s_c old, struct bkey_s_c new,
511 bool gc = flags & BTREE_TRIGGER_GC;
512 u64 journal_seq = trans->journal_res.seq;
513 u64 bucket_journal_seq;
514 struct bch_fs *c = trans->c;
515 struct bch_alloc_v4 old_a_convert, new_a_convert;
516 const struct bch_alloc_v4 *old_a, *new_a;
521 * alloc btree is read in by bch2_alloc_read, not gc:
523 if ((flags & BTREE_TRIGGER_GC) &&
524 !(flags & BTREE_TRIGGER_BUCKET_INVALIDATE))
527 if (bch2_trans_inconsistent_on(!bch2_dev_bucket_exists(c, new.k->p), trans,
528 "alloc key for invalid device or bucket"))
531 ca = bch_dev_bkey_exists(c, new.k->p.inode);
533 old_a = bch2_alloc_to_v4(old, &old_a_convert);
534 new_a = bch2_alloc_to_v4(new, &new_a_convert);
536 bucket_journal_seq = new_a->journal_seq;
538 if ((flags & BTREE_TRIGGER_INSERT) &&
539 data_type_is_empty(old_a->data_type) !=
540 data_type_is_empty(new_a->data_type) &&
541 new.k->type == KEY_TYPE_alloc_v4) {
542 struct bch_alloc_v4 *v = (struct bch_alloc_v4 *) new.v;
544 EBUG_ON(!journal_seq);
547 * If the btree updates referring to a bucket weren't flushed
548 * before the bucket became empty again, then the we don't have
549 * to wait on a journal flush before we can reuse the bucket:
551 v->journal_seq = bucket_journal_seq =
552 data_type_is_empty(new_a->data_type) &&
553 (journal_seq == v->journal_seq ||
554 bch2_journal_noflush_seq(&c->journal, v->journal_seq))
558 if (!data_type_is_empty(old_a->data_type) &&
559 data_type_is_empty(new_a->data_type) &&
560 bucket_journal_seq) {
561 ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
562 c->journal.flushed_seq_ondisk,
563 new.k->p.inode, new.k->p.offset,
566 bch2_fs_fatal_error(c,
567 "error setting bucket_needs_journal_commit: %i", ret);
572 percpu_down_read(&c->mark_lock);
573 if (!gc && new_a->gen != old_a->gen)
574 *bucket_gen(ca, new.k->p.offset) = new_a->gen;
576 bch2_dev_usage_update(c, ca, *old_a, *new_a, journal_seq, gc);
579 struct bucket *g = gc_bucket(ca, new.k->p.offset);
585 g->data_type = new_a->data_type;
586 g->stripe = new_a->stripe;
587 g->stripe_redundancy = new_a->stripe_redundancy;
588 g->dirty_sectors = new_a->dirty_sectors;
589 g->cached_sectors = new_a->cached_sectors;
593 percpu_up_read(&c->mark_lock);
596 * need to know if we're getting called from the invalidate path or
600 if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
601 old_a->cached_sectors) {
602 ret = update_cached_sectors(c, new, ca->dev_idx,
603 -((s64) old_a->cached_sectors),
606 bch2_fs_fatal_error(c, "%s(): no replicas entry while updating cached sectors",
612 if (new_a->data_type == BCH_DATA_free &&
613 (!new_a->journal_seq || new_a->journal_seq < c->journal.flushed_seq_ondisk))
614 closure_wake_up(&c->freelist_wait);
616 if (new_a->data_type == BCH_DATA_need_discard &&
617 (!bucket_journal_seq || bucket_journal_seq < c->journal.flushed_seq_ondisk))
620 if (old_a->data_type != BCH_DATA_cached &&
621 new_a->data_type == BCH_DATA_cached &&
622 should_invalidate_buckets(ca, bch2_dev_usage_read(ca)))
623 bch2_do_invalidates(c);
625 if (new_a->data_type == BCH_DATA_need_gc_gens)
631 int bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca,
632 size_t b, enum bch_data_type data_type,
633 unsigned sectors, struct gc_pos pos,
636 struct bucket old, new, *g;
639 BUG_ON(!(flags & BTREE_TRIGGER_GC));
640 BUG_ON(data_type != BCH_DATA_sb &&
641 data_type != BCH_DATA_journal);
644 * Backup superblock might be past the end of our normal usable space:
646 if (b >= ca->mi.nbuckets)
649 percpu_down_read(&c->mark_lock);
650 g = gc_bucket(ca, b);
655 if (bch2_fs_inconsistent_on(g->data_type &&
656 g->data_type != data_type, c,
657 "different types of data in same bucket: %s, %s",
658 bch2_data_types[g->data_type],
659 bch2_data_types[data_type])) {
664 if (bch2_fs_inconsistent_on((u64) g->dirty_sectors + sectors > ca->mi.bucket_size, c,
665 "bucket %u:%zu gen %u data type %s sector count overflow: %u + %u > bucket size",
666 ca->dev_idx, b, g->gen,
667 bch2_data_types[g->data_type ?: data_type],
668 g->dirty_sectors, sectors)) {
674 g->data_type = data_type;
675 g->dirty_sectors += sectors;
680 bch2_dev_usage_update_m(c, ca, old, new, 0, true);
681 percpu_up_read(&c->mark_lock);
685 static int check_bucket_ref(struct btree_trans *trans,
687 const struct bch_extent_ptr *ptr,
688 s64 sectors, enum bch_data_type ptr_data_type,
689 u8 b_gen, u8 bucket_data_type,
690 u32 dirty_sectors, u32 cached_sectors)
692 struct bch_fs *c = trans->c;
693 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
694 size_t bucket_nr = PTR_BUCKET_NR(ca, ptr);
695 u16 bucket_sectors = !ptr->cached
698 struct printbuf buf = PRINTBUF;
701 if (bucket_data_type == BCH_DATA_cached)
702 bucket_data_type = BCH_DATA_user;
704 if ((bucket_data_type == BCH_DATA_stripe && ptr_data_type == BCH_DATA_user) ||
705 (bucket_data_type == BCH_DATA_user && ptr_data_type == BCH_DATA_stripe))
706 bucket_data_type = ptr_data_type = BCH_DATA_stripe;
708 if (gen_after(ptr->gen, b_gen)) {
709 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
710 "bucket %u:%zu gen %u data type %s: ptr gen %u newer than bucket gen\n"
712 ptr->dev, bucket_nr, b_gen,
713 bch2_data_types[bucket_data_type ?: ptr_data_type],
715 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
720 if (gen_cmp(b_gen, ptr->gen) > BUCKET_GC_GEN_MAX) {
721 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
722 "bucket %u:%zu gen %u data type %s: ptr gen %u too stale\n"
724 ptr->dev, bucket_nr, b_gen,
725 bch2_data_types[bucket_data_type ?: ptr_data_type],
727 (printbuf_reset(&buf),
728 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
733 if (b_gen != ptr->gen && !ptr->cached) {
734 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
735 "bucket %u:%zu gen %u (mem gen %u) data type %s: stale dirty ptr (gen %u)\n"
737 ptr->dev, bucket_nr, b_gen,
738 *bucket_gen(ca, bucket_nr),
739 bch2_data_types[bucket_data_type ?: ptr_data_type],
741 (printbuf_reset(&buf),
742 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
747 if (b_gen != ptr->gen) {
752 if (!data_type_is_empty(bucket_data_type) &&
754 bucket_data_type != ptr_data_type) {
755 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
756 "bucket %u:%zu gen %u different types of data in same bucket: %s, %s\n"
758 ptr->dev, bucket_nr, b_gen,
759 bch2_data_types[bucket_data_type],
760 bch2_data_types[ptr_data_type],
761 (printbuf_reset(&buf),
762 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
767 if ((unsigned) (bucket_sectors + sectors) > U32_MAX) {
768 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
769 "bucket %u:%zu gen %u data type %s sector count overflow: %u + %lli > U16_MAX\n"
771 ptr->dev, bucket_nr, b_gen,
772 bch2_data_types[bucket_data_type ?: ptr_data_type],
773 bucket_sectors, sectors,
774 (printbuf_reset(&buf),
775 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
783 bch2_dump_trans_updates(trans);
787 static int mark_stripe_bucket(struct btree_trans *trans,
792 struct bch_fs *c = trans->c;
793 u64 journal_seq = trans->journal_res.seq;
794 const struct bch_stripe *s = bkey_s_c_to_stripe(k).v;
795 unsigned nr_data = s->nr_blocks - s->nr_redundant;
796 bool parity = ptr_idx >= nr_data;
797 enum bch_data_type data_type = parity ? BCH_DATA_parity : BCH_DATA_stripe;
798 s64 sectors = parity ? le16_to_cpu(s->sectors) : 0;
799 const struct bch_extent_ptr *ptr = s->ptrs + ptr_idx;
800 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
801 struct bucket old, new, *g;
802 struct printbuf buf = PRINTBUF;
805 BUG_ON(!(flags & BTREE_TRIGGER_GC));
807 /* * XXX doesn't handle deletion */
809 percpu_down_read(&c->mark_lock);
811 g = PTR_GC_BUCKET(ca, ptr);
813 if (g->dirty_sectors ||
814 (g->stripe && g->stripe != k.k->p.offset)) {
815 bch2_fs_inconsistent(c,
816 "bucket %u:%zu gen %u: multiple stripes using same bucket\n%s",
817 ptr->dev, PTR_BUCKET_NR(ca, ptr), g->gen,
818 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
826 ret = check_bucket_ref(trans, k, ptr, sectors, data_type,
827 g->gen, g->data_type,
828 g->dirty_sectors, g->cached_sectors);
832 g->data_type = data_type;
833 g->dirty_sectors += sectors;
835 g->stripe = k.k->p.offset;
836 g->stripe_redundancy = s->nr_redundant;
841 bch2_dev_usage_update_m(c, ca, old, new, journal_seq, true);
842 percpu_up_read(&c->mark_lock);
847 static int __mark_pointer(struct btree_trans *trans,
849 const struct bch_extent_ptr *ptr,
850 s64 sectors, enum bch_data_type ptr_data_type,
851 u8 bucket_gen, u8 *bucket_data_type,
852 u32 *dirty_sectors, u32 *cached_sectors)
854 u32 *dst_sectors = !ptr->cached
857 int ret = check_bucket_ref(trans, k, ptr, sectors, ptr_data_type,
858 bucket_gen, *bucket_data_type,
859 *dirty_sectors, *cached_sectors);
864 *dst_sectors += sectors;
865 *bucket_data_type = *dirty_sectors || *cached_sectors
870 static int bch2_mark_pointer(struct btree_trans *trans,
871 enum btree_id btree_id, unsigned level,
873 struct extent_ptr_decoded p,
877 u64 journal_seq = trans->journal_res.seq;
878 struct bch_fs *c = trans->c;
879 struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
880 struct bucket old, new, *g;
881 enum bch_data_type data_type = bkey_ptr_data_type(btree_id, level, k, p);
885 BUG_ON(!(flags & BTREE_TRIGGER_GC));
887 percpu_down_read(&c->mark_lock);
888 g = PTR_GC_BUCKET(ca, &p.ptr);
892 bucket_data_type = g->data_type;
893 ret = __mark_pointer(trans, k, &p.ptr, sectors,
899 g->data_type = bucket_data_type;
904 bch2_dev_usage_update_m(c, ca, old, new, journal_seq, true);
905 percpu_up_read(&c->mark_lock);
910 static int bch2_mark_stripe_ptr(struct btree_trans *trans,
912 struct bch_extent_stripe_ptr p,
913 enum bch_data_type data_type,
917 struct bch_fs *c = trans->c;
918 struct bch_replicas_padded r;
921 BUG_ON(!(flags & BTREE_TRIGGER_GC));
923 m = genradix_ptr_alloc(&c->gc_stripes, p.idx, GFP_KERNEL);
925 bch_err(c, "error allocating memory for gc_stripes, idx %llu",
927 return -BCH_ERR_ENOMEM_mark_stripe_ptr;
930 mutex_lock(&c->ec_stripes_heap_lock);
932 if (!m || !m->alive) {
933 mutex_unlock(&c->ec_stripes_heap_lock);
934 bch_err_ratelimited(c, "pointer to nonexistent stripe %llu",
936 bch2_inconsistent_error(c);
940 m->block_sectors[p.block] += sectors;
943 mutex_unlock(&c->ec_stripes_heap_lock);
945 r.e.data_type = data_type;
946 update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, true);
951 static int __mark_extent(struct btree_trans *trans,
952 enum btree_id btree_id, unsigned level,
953 struct bkey_s_c k, unsigned flags)
955 u64 journal_seq = trans->journal_res.seq;
956 struct bch_fs *c = trans->c;
957 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
958 const union bch_extent_entry *entry;
959 struct extent_ptr_decoded p;
960 struct bch_replicas_padded r;
961 enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
964 s64 sectors = bkey_is_btree_ptr(k.k)
967 s64 dirty_sectors = 0;
971 BUG_ON(!(flags & BTREE_TRIGGER_GC));
973 r.e.data_type = data_type;
977 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
978 s64 disk_sectors = ptr_disk_sectors(sectors, p);
980 if (flags & BTREE_TRIGGER_OVERWRITE)
981 disk_sectors = -disk_sectors;
983 ret = bch2_mark_pointer(trans, btree_id, level, k, p, disk_sectors, flags);
991 ret = update_cached_sectors(c, k, p.ptr.dev,
992 disk_sectors, journal_seq, true);
994 bch2_fs_fatal_error(c, "%s(): no replicas entry while updating cached sectors",
999 } else if (!p.has_ec) {
1000 dirty_sectors += disk_sectors;
1001 r.e.devs[r.e.nr_devs++] = p.ptr.dev;
1003 ret = bch2_mark_stripe_ptr(trans, k, p.ec, data_type,
1004 disk_sectors, flags);
1009 * There may be other dirty pointers in this extent, but
1010 * if so they're not required for mounting if we have an
1011 * erasure coded pointer in this extent:
1013 r.e.nr_required = 0;
1018 ret = update_replicas(c, k, &r.e, dirty_sectors, journal_seq, true);
1020 struct printbuf buf = PRINTBUF;
1022 bch2_bkey_val_to_text(&buf, c, k);
1023 bch2_fs_fatal_error(c, "%s(): no replicas entry for %s", __func__, buf.buf);
1024 printbuf_exit(&buf);
1032 int bch2_mark_extent(struct btree_trans *trans,
1033 enum btree_id btree_id, unsigned level,
1034 struct bkey_s_c old, struct bkey_s_c new,
1037 return mem_trigger_run_insert_then_overwrite(__mark_extent, trans, btree_id, level, old, new, flags);
1040 int bch2_mark_stripe(struct btree_trans *trans,
1041 enum btree_id btree_id, unsigned level,
1042 struct bkey_s_c old, struct bkey_s_c new,
1045 bool gc = flags & BTREE_TRIGGER_GC;
1046 u64 journal_seq = trans->journal_res.seq;
1047 struct bch_fs *c = trans->c;
1048 u64 idx = new.k->p.offset;
1049 const struct bch_stripe *old_s = old.k->type == KEY_TYPE_stripe
1050 ? bkey_s_c_to_stripe(old).v : NULL;
1051 const struct bch_stripe *new_s = new.k->type == KEY_TYPE_stripe
1052 ? bkey_s_c_to_stripe(new).v : NULL;
1056 BUG_ON(gc && old_s);
1059 struct stripe *m = genradix_ptr(&c->stripes, idx);
1062 struct printbuf buf1 = PRINTBUF;
1063 struct printbuf buf2 = PRINTBUF;
1065 bch2_bkey_val_to_text(&buf1, c, old);
1066 bch2_bkey_val_to_text(&buf2, c, new);
1067 bch_err_ratelimited(c, "error marking nonexistent stripe %llu while marking\n"
1069 "new %s", idx, buf1.buf, buf2.buf);
1070 printbuf_exit(&buf2);
1071 printbuf_exit(&buf1);
1072 bch2_inconsistent_error(c);
1077 bch2_stripes_heap_del(c, m, idx);
1079 memset(m, 0, sizeof(*m));
1081 m->sectors = le16_to_cpu(new_s->sectors);
1082 m->algorithm = new_s->algorithm;
1083 m->nr_blocks = new_s->nr_blocks;
1084 m->nr_redundant = new_s->nr_redundant;
1085 m->blocks_nonempty = 0;
1087 for (i = 0; i < new_s->nr_blocks; i++)
1088 m->blocks_nonempty += !!stripe_blockcount_get(new_s, i);
1091 bch2_stripes_heap_insert(c, m, idx);
1093 bch2_stripes_heap_update(c, m, idx);
1096 struct gc_stripe *m =
1097 genradix_ptr_alloc(&c->gc_stripes, idx, GFP_KERNEL);
1100 bch_err(c, "error allocating memory for gc_stripes, idx %llu",
1102 return -BCH_ERR_ENOMEM_mark_stripe;
1105 * This will be wrong when we bring back runtime gc: we should
1106 * be unmarking the old key and then marking the new key
1109 m->sectors = le16_to_cpu(new_s->sectors);
1110 m->nr_blocks = new_s->nr_blocks;
1111 m->nr_redundant = new_s->nr_redundant;
1113 for (i = 0; i < new_s->nr_blocks; i++)
1114 m->ptrs[i] = new_s->ptrs[i];
1116 bch2_bkey_to_replicas(&m->r.e, new);
1119 * gc recalculates this field from stripe ptr
1122 memset(m->block_sectors, 0, sizeof(m->block_sectors));
1124 for (i = 0; i < new_s->nr_blocks; i++) {
1125 ret = mark_stripe_bucket(trans, new, i, flags);
1130 ret = update_replicas(c, new, &m->r.e,
1131 ((s64) m->sectors * m->nr_redundant),
1134 struct printbuf buf = PRINTBUF;
1136 bch2_bkey_val_to_text(&buf, c, new);
1137 bch2_fs_fatal_error(c, "no replicas entry for %s", buf.buf);
1138 printbuf_exit(&buf);
1146 int bch2_mark_inode(struct btree_trans *trans,
1147 enum btree_id btree_id, unsigned level,
1148 struct bkey_s_c old, struct bkey_s_c new,
1151 struct bch_fs *c = trans->c;
1152 struct bch_fs_usage *fs_usage;
1153 u64 journal_seq = trans->journal_res.seq;
1155 if (flags & BTREE_TRIGGER_INSERT) {
1156 struct bch_inode_v3 *v = (struct bch_inode_v3 *) new.v;
1158 BUG_ON(!journal_seq);
1159 BUG_ON(new.k->type != KEY_TYPE_inode_v3);
1161 v->bi_journal_seq = cpu_to_le64(journal_seq);
1164 if (flags & BTREE_TRIGGER_GC) {
1165 percpu_down_read(&c->mark_lock);
1168 fs_usage = fs_usage_ptr(c, journal_seq, flags & BTREE_TRIGGER_GC);
1169 fs_usage->nr_inodes += bkey_is_inode(new.k);
1170 fs_usage->nr_inodes -= bkey_is_inode(old.k);
1173 percpu_up_read(&c->mark_lock);
1178 static int __mark_reservation(struct btree_trans *trans,
1179 enum btree_id btree_id, unsigned level,
1180 struct bkey_s_c k, unsigned flags)
1182 struct bch_fs *c = trans->c;
1183 struct bch_fs_usage *fs_usage;
1184 unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
1185 s64 sectors = (s64) k.k->size;
1187 BUG_ON(!(flags & BTREE_TRIGGER_GC));
1189 if (flags & BTREE_TRIGGER_OVERWRITE)
1191 sectors *= replicas;
1193 percpu_down_read(&c->mark_lock);
1196 fs_usage = fs_usage_ptr(c, trans->journal_res.seq, flags & BTREE_TRIGGER_GC);
1197 replicas = clamp_t(unsigned, replicas, 1,
1198 ARRAY_SIZE(fs_usage->persistent_reserved));
1200 fs_usage->reserved += sectors;
1201 fs_usage->persistent_reserved[replicas - 1] += sectors;
1204 percpu_up_read(&c->mark_lock);
1209 int bch2_mark_reservation(struct btree_trans *trans,
1210 enum btree_id btree_id, unsigned level,
1211 struct bkey_s_c old, struct bkey_s_c new,
1214 return mem_trigger_run_insert_then_overwrite(__mark_reservation, trans, btree_id, level, old, new, flags);
1217 static s64 __bch2_mark_reflink_p(struct btree_trans *trans,
1218 struct bkey_s_c_reflink_p p,
1220 u64 *idx, unsigned flags, size_t r_idx)
1222 struct bch_fs *c = trans->c;
1223 struct reflink_gc *r;
1224 int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
1227 struct printbuf buf = PRINTBUF;
1229 if (r_idx >= c->reflink_gc_nr)
1232 r = genradix_ptr(&c->reflink_gc_table, r_idx);
1233 next_idx = min(next_idx, r->offset - r->size);
1234 if (*idx < next_idx)
1237 BUG_ON((s64) r->refcount + add < 0);
1243 if (fsck_err(c, "pointer to missing indirect extent\n"
1245 " missing range %llu-%llu",
1246 (bch2_bkey_val_to_text(&buf, c, p.s_c), buf.buf),
1248 struct bkey_i_error *new;
1250 new = bch2_trans_kmalloc(trans, sizeof(*new));
1251 ret = PTR_ERR_OR_ZERO(new);
1256 new->k.type = KEY_TYPE_error;
1257 new->k.p = bkey_start_pos(p.k);
1258 new->k.p.offset += *idx - start;
1259 bch2_key_resize(&new->k, next_idx - *idx);
1260 ret = __bch2_btree_insert(trans, BTREE_ID_extents, &new->k_i,
1261 BTREE_TRIGGER_NORUN);
1267 printbuf_exit(&buf);
1271 static int __mark_reflink_p(struct btree_trans *trans,
1272 enum btree_id btree_id, unsigned level,
1273 struct bkey_s_c k, unsigned flags)
1275 struct bch_fs *c = trans->c;
1276 struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
1277 struct reflink_gc *ref;
1279 u64 idx = le64_to_cpu(p.v->idx), start = idx;
1280 u64 end = le64_to_cpu(p.v->idx) + p.k->size;
1283 BUG_ON(!(flags & BTREE_TRIGGER_GC));
1285 if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix) {
1286 idx -= le32_to_cpu(p.v->front_pad);
1287 end += le32_to_cpu(p.v->back_pad);
1291 r = c->reflink_gc_nr;
1293 m = l + (r - l) / 2;
1295 ref = genradix_ptr(&c->reflink_gc_table, m);
1296 if (ref->offset <= idx)
1302 while (idx < end && !ret)
1303 ret = __bch2_mark_reflink_p(trans, p, start, end,
1309 int bch2_mark_reflink_p(struct btree_trans *trans,
1310 enum btree_id btree_id, unsigned level,
1311 struct bkey_s_c old, struct bkey_s_c new,
1314 return mem_trigger_run_insert_then_overwrite(__mark_reflink_p, trans, btree_id, level, old, new, flags);
1317 void bch2_trans_fs_usage_revert(struct btree_trans *trans,
1318 struct replicas_delta_list *deltas)
1320 struct bch_fs *c = trans->c;
1321 struct bch_fs_usage *dst;
1322 struct replicas_delta *d, *top = (void *) deltas->d + deltas->used;
1326 percpu_down_read(&c->mark_lock);
1328 dst = fs_usage_ptr(c, trans->journal_res.seq, false);
1330 /* revert changes: */
1331 for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
1332 switch (d->r.data_type) {
1333 case BCH_DATA_btree:
1335 case BCH_DATA_parity:
1338 BUG_ON(__update_replicas(c, dst, &d->r, -d->delta));
1341 dst->nr_inodes -= deltas->nr_inodes;
1343 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1344 added -= deltas->persistent_reserved[i];
1345 dst->reserved -= deltas->persistent_reserved[i];
1346 dst->persistent_reserved[i] -= deltas->persistent_reserved[i];
1350 trans->disk_res->sectors += added;
1351 this_cpu_add(*c->online_reserved, added);
1355 percpu_up_read(&c->mark_lock);
1358 int bch2_trans_fs_usage_apply(struct btree_trans *trans,
1359 struct replicas_delta_list *deltas)
1361 struct bch_fs *c = trans->c;
1362 static int warned_disk_usage = 0;
1364 unsigned disk_res_sectors = trans->disk_res ? trans->disk_res->sectors : 0;
1365 struct replicas_delta *d = deltas->d, *d2;
1366 struct replicas_delta *top = (void *) deltas->d + deltas->used;
1367 struct bch_fs_usage *dst;
1368 s64 added = 0, should_not_have_added;
1371 percpu_down_read(&c->mark_lock);
1373 dst = fs_usage_ptr(c, trans->journal_res.seq, false);
1375 for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
1376 switch (d->r.data_type) {
1377 case BCH_DATA_btree:
1379 case BCH_DATA_parity:
1383 if (__update_replicas(c, dst, &d->r, d->delta))
1387 dst->nr_inodes += deltas->nr_inodes;
1389 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1390 added += deltas->persistent_reserved[i];
1391 dst->reserved += deltas->persistent_reserved[i];
1392 dst->persistent_reserved[i] += deltas->persistent_reserved[i];
1396 * Not allowed to reduce sectors_available except by getting a
1399 should_not_have_added = added - (s64) disk_res_sectors;
1400 if (unlikely(should_not_have_added > 0)) {
1401 u64 old, new, v = atomic64_read(&c->sectors_available);
1405 new = max_t(s64, 0, old - should_not_have_added);
1406 } while ((v = atomic64_cmpxchg(&c->sectors_available,
1409 added -= should_not_have_added;
1414 trans->disk_res->sectors -= added;
1415 this_cpu_sub(*c->online_reserved, added);
1419 percpu_up_read(&c->mark_lock);
1421 if (unlikely(warn) && !xchg(&warned_disk_usage, 1))
1422 bch2_trans_inconsistent(trans,
1423 "disk usage increased %lli more than %u sectors reserved)",
1424 should_not_have_added, disk_res_sectors);
1427 /* revert changes: */
1428 for (d2 = deltas->d; d2 != d; d2 = replicas_delta_next(d2))
1429 BUG_ON(__update_replicas(c, dst, &d2->r, -d2->delta));
1432 percpu_up_read(&c->mark_lock);
1438 static inline int bch2_trans_mark_pointer(struct btree_trans *trans,
1439 enum btree_id btree_id, unsigned level,
1440 struct bkey_s_c k, struct extent_ptr_decoded p,
1443 bool insert = !(flags & BTREE_TRIGGER_OVERWRITE);
1444 struct btree_iter iter;
1445 struct bkey_i_alloc_v4 *a;
1447 struct bch_backpointer bp;
1451 bch2_extent_ptr_to_bp(trans->c, btree_id, level, k, p, &bucket, &bp);
1452 sectors = bp.bucket_len;
1456 a = bch2_trans_start_alloc_update(trans, &iter, bucket);
1460 ret = __mark_pointer(trans, k, &p.ptr, sectors, bp.data_type,
1461 a->v.gen, &a->v.data_type,
1462 &a->v.dirty_sectors, &a->v.cached_sectors) ?:
1463 bch2_trans_update(trans, &iter, &a->k_i, 0);
1464 bch2_trans_iter_exit(trans, &iter);
1469 if (!p.ptr.cached) {
1470 ret = bch2_bucket_backpointer_mod(trans, bucket, bp, k, insert);
1478 static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
1479 struct extent_ptr_decoded p,
1480 s64 sectors, enum bch_data_type data_type)
1482 struct btree_iter iter;
1483 struct bkey_i_stripe *s;
1484 struct bch_replicas_padded r;
1487 s = bch2_bkey_get_mut_typed(trans, &iter,
1488 BTREE_ID_stripes, POS(0, p.ec.idx),
1489 BTREE_ITER_WITH_UPDATES, stripe);
1490 ret = PTR_ERR_OR_ZERO(s);
1491 if (unlikely(ret)) {
1492 bch2_trans_inconsistent_on(bch2_err_matches(ret, ENOENT), trans,
1493 "pointer to nonexistent stripe %llu",
1498 if (!bch2_ptr_matches_stripe(&s->v, p)) {
1499 bch2_trans_inconsistent(trans,
1500 "stripe pointer doesn't match stripe %llu",
1506 stripe_blockcount_set(&s->v, p.ec.block,
1507 stripe_blockcount_get(&s->v, p.ec.block) +
1510 bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(&s->k_i));
1511 r.e.data_type = data_type;
1512 ret = update_replicas_list(trans, &r.e, sectors);
1514 bch2_trans_iter_exit(trans, &iter);
1518 static int __trans_mark_extent(struct btree_trans *trans,
1519 enum btree_id btree_id, unsigned level,
1520 struct bkey_s_c k, unsigned flags)
1522 struct bch_fs *c = trans->c;
1523 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1524 const union bch_extent_entry *entry;
1525 struct extent_ptr_decoded p;
1526 struct bch_replicas_padded r;
1527 enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
1530 s64 sectors = bkey_is_btree_ptr(k.k)
1533 s64 dirty_sectors = 0;
1537 r.e.data_type = data_type;
1539 r.e.nr_required = 1;
1541 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
1542 s64 disk_sectors = ptr_disk_sectors(sectors, p);
1544 if (flags & BTREE_TRIGGER_OVERWRITE)
1545 disk_sectors = -disk_sectors;
1547 ret = bch2_trans_mark_pointer(trans, btree_id, level, k, p, flags);
1555 ret = update_cached_sectors_list(trans, p.ptr.dev,
1560 } else if (!p.has_ec) {
1561 dirty_sectors += disk_sectors;
1562 r.e.devs[r.e.nr_devs++] = p.ptr.dev;
1564 ret = bch2_trans_mark_stripe_ptr(trans, p,
1565 disk_sectors, data_type);
1569 r.e.nr_required = 0;
1574 ret = update_replicas_list(trans, &r.e, dirty_sectors);
1579 int bch2_trans_mark_extent(struct btree_trans *trans,
1580 enum btree_id btree_id, unsigned level,
1581 struct bkey_s_c old, struct bkey_i *new,
1584 return trigger_run_insert_then_overwrite(__trans_mark_extent, trans, btree_id, level, old, new, flags);
1587 static int bch2_trans_mark_stripe_bucket(struct btree_trans *trans,
1588 struct bkey_s_c_stripe s,
1589 unsigned idx, bool deleting)
1591 struct bch_fs *c = trans->c;
1592 const struct bch_extent_ptr *ptr = &s.v->ptrs[idx];
1593 struct btree_iter iter;
1594 struct bkey_i_alloc_v4 *a;
1595 enum bch_data_type data_type = idx >= s.v->nr_blocks - s.v->nr_redundant
1596 ? BCH_DATA_parity : 0;
1597 s64 sectors = data_type ? le16_to_cpu(s.v->sectors) : 0;
1603 a = bch2_trans_start_alloc_update(trans, &iter, PTR_BUCKET_POS(c, ptr));
1607 ret = check_bucket_ref(trans, s.s_c, ptr, sectors, data_type,
1608 a->v.gen, a->v.data_type,
1609 a->v.dirty_sectors, a->v.cached_sectors);
1614 if (bch2_trans_inconsistent_on(a->v.stripe ||
1615 a->v.stripe_redundancy, trans,
1616 "bucket %llu:%llu gen %u data type %s dirty_sectors %u: multiple stripes using same bucket (%u, %llu)",
1617 iter.pos.inode, iter.pos.offset, a->v.gen,
1618 bch2_data_types[a->v.data_type],
1620 a->v.stripe, s.k->p.offset)) {
1625 if (bch2_trans_inconsistent_on(data_type && a->v.dirty_sectors, trans,
1626 "bucket %llu:%llu gen %u data type %s dirty_sectors %u: data already in stripe bucket %llu",
1627 iter.pos.inode, iter.pos.offset, a->v.gen,
1628 bch2_data_types[a->v.data_type],
1635 a->v.stripe = s.k->p.offset;
1636 a->v.stripe_redundancy = s.v->nr_redundant;
1637 a->v.data_type = BCH_DATA_stripe;
1639 if (bch2_trans_inconsistent_on(a->v.stripe != s.k->p.offset ||
1640 a->v.stripe_redundancy != s.v->nr_redundant, trans,
1641 "bucket %llu:%llu gen %u: not marked as stripe when deleting stripe %llu (got %u)",
1642 iter.pos.inode, iter.pos.offset, a->v.gen,
1643 s.k->p.offset, a->v.stripe)) {
1649 a->v.stripe_redundancy = 0;
1650 a->v.data_type = alloc_data_type(a->v, BCH_DATA_user);
1653 a->v.dirty_sectors += sectors;
1655 a->v.data_type = !deleting ? data_type : 0;
1657 ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
1661 bch2_trans_iter_exit(trans, &iter);
1665 int bch2_trans_mark_stripe(struct btree_trans *trans,
1666 enum btree_id btree_id, unsigned level,
1667 struct bkey_s_c old, struct bkey_i *new,
1670 const struct bch_stripe *old_s = NULL;
1671 struct bch_stripe *new_s = NULL;
1672 struct bch_replicas_padded r;
1673 unsigned i, nr_blocks;
1676 if (old.k->type == KEY_TYPE_stripe)
1677 old_s = bkey_s_c_to_stripe(old).v;
1678 if (new->k.type == KEY_TYPE_stripe)
1679 new_s = &bkey_i_to_stripe(new)->v;
1682 * If the pointers aren't changing, we don't need to do anything:
1684 if (new_s && old_s &&
1685 new_s->nr_blocks == old_s->nr_blocks &&
1686 new_s->nr_redundant == old_s->nr_redundant &&
1687 !memcmp(old_s->ptrs, new_s->ptrs,
1688 new_s->nr_blocks * sizeof(struct bch_extent_ptr)))
1691 BUG_ON(new_s && old_s &&
1692 (new_s->nr_blocks != old_s->nr_blocks ||
1693 new_s->nr_redundant != old_s->nr_redundant));
1695 nr_blocks = new_s ? new_s->nr_blocks : old_s->nr_blocks;
1698 s64 sectors = le16_to_cpu(new_s->sectors);
1700 bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(new));
1701 ret = update_replicas_list(trans, &r.e, sectors * new_s->nr_redundant);
1707 s64 sectors = -((s64) le16_to_cpu(old_s->sectors));
1709 bch2_bkey_to_replicas(&r.e, old);
1710 ret = update_replicas_list(trans, &r.e, sectors * old_s->nr_redundant);
1715 for (i = 0; i < nr_blocks; i++) {
1716 if (new_s && old_s &&
1717 !memcmp(&new_s->ptrs[i],
1719 sizeof(new_s->ptrs[i])))
1723 ret = bch2_trans_mark_stripe_bucket(trans,
1724 bkey_i_to_s_c_stripe(new), i, false);
1730 ret = bch2_trans_mark_stripe_bucket(trans,
1731 bkey_s_c_to_stripe(old), i, true);
1740 int bch2_trans_mark_inode(struct btree_trans *trans,
1741 enum btree_id btree_id, unsigned level,
1742 struct bkey_s_c old,
1746 int nr = bkey_is_inode(&new->k) - bkey_is_inode(old.k);
1749 int ret = replicas_deltas_realloc(trans, 0);
1750 struct replicas_delta_list *d = trans->fs_usage_deltas;
1761 static int __trans_mark_reservation(struct btree_trans *trans,
1762 enum btree_id btree_id, unsigned level,
1763 struct bkey_s_c k, unsigned flags)
1765 unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
1766 s64 sectors = (s64) k.k->size;
1767 struct replicas_delta_list *d;
1770 if (flags & BTREE_TRIGGER_OVERWRITE)
1772 sectors *= replicas;
1774 ret = replicas_deltas_realloc(trans, 0);
1778 d = trans->fs_usage_deltas;
1779 replicas = clamp_t(unsigned, replicas, 1,
1780 ARRAY_SIZE(d->persistent_reserved));
1782 d->persistent_reserved[replicas - 1] += sectors;
1786 int bch2_trans_mark_reservation(struct btree_trans *trans,
1787 enum btree_id btree_id, unsigned level,
1788 struct bkey_s_c old,
1792 return trigger_run_insert_then_overwrite(__trans_mark_reservation, trans, btree_id, level, old, new, flags);
1795 static int trans_mark_reflink_p_segment(struct btree_trans *trans,
1796 struct bkey_s_c_reflink_p p,
1797 u64 *idx, unsigned flags)
1799 struct bch_fs *c = trans->c;
1800 struct btree_iter iter;
1803 int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
1804 struct printbuf buf = PRINTBUF;
1807 k = bch2_bkey_get_mut_noupdate(trans, &iter,
1808 BTREE_ID_reflink, POS(0, *idx),
1809 BTREE_ITER_WITH_UPDATES);
1810 ret = PTR_ERR_OR_ZERO(k);
1814 refcount = bkey_refcount(k);
1816 bch2_bkey_val_to_text(&buf, c, p.s_c);
1817 bch2_trans_inconsistent(trans,
1818 "nonexistent indirect extent at %llu while marking\n %s",
1824 if (!*refcount && (flags & BTREE_TRIGGER_OVERWRITE)) {
1825 bch2_bkey_val_to_text(&buf, c, p.s_c);
1826 bch2_trans_inconsistent(trans,
1827 "indirect extent refcount underflow at %llu while marking\n %s",
1833 if (flags & BTREE_TRIGGER_INSERT) {
1834 struct bch_reflink_p *v = (struct bch_reflink_p *) p.v;
1837 pad = max_t(s64, le32_to_cpu(v->front_pad),
1838 le64_to_cpu(v->idx) - bkey_start_offset(&k->k));
1839 BUG_ON(pad > U32_MAX);
1840 v->front_pad = cpu_to_le32(pad);
1842 pad = max_t(s64, le32_to_cpu(v->back_pad),
1843 k->k.p.offset - p.k->size - le64_to_cpu(v->idx));
1844 BUG_ON(pad > U32_MAX);
1845 v->back_pad = cpu_to_le32(pad);
1848 le64_add_cpu(refcount, add);
1850 bch2_btree_iter_set_pos_to_extent_start(&iter);
1851 ret = bch2_trans_update(trans, &iter, k, 0);
1855 *idx = k->k.p.offset;
1857 bch2_trans_iter_exit(trans, &iter);
1858 printbuf_exit(&buf);
1862 static int __trans_mark_reflink_p(struct btree_trans *trans,
1863 enum btree_id btree_id, unsigned level,
1864 struct bkey_s_c k, unsigned flags)
1866 struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
1870 idx = le64_to_cpu(p.v->idx) - le32_to_cpu(p.v->front_pad);
1871 end_idx = le64_to_cpu(p.v->idx) + p.k->size +
1872 le32_to_cpu(p.v->back_pad);
1874 while (idx < end_idx && !ret)
1875 ret = trans_mark_reflink_p_segment(trans, p, &idx, flags);
1879 int bch2_trans_mark_reflink_p(struct btree_trans *trans,
1880 enum btree_id btree_id, unsigned level,
1881 struct bkey_s_c old,
1885 if (flags & BTREE_TRIGGER_INSERT) {
1886 struct bch_reflink_p *v = &bkey_i_to_reflink_p(new)->v;
1888 v->front_pad = v->back_pad = 0;
1891 return trigger_run_insert_then_overwrite(__trans_mark_reflink_p, trans, btree_id, level, old, new, flags);
1894 static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
1895 struct bch_dev *ca, size_t b,
1896 enum bch_data_type type,
1899 struct bch_fs *c = trans->c;
1900 struct btree_iter iter;
1901 struct bkey_i_alloc_v4 *a;
1905 * Backup superblock might be past the end of our normal usable space:
1907 if (b >= ca->mi.nbuckets)
1910 a = bch2_trans_start_alloc_update(trans, &iter, POS(ca->dev_idx, b));
1914 if (a->v.data_type && type && a->v.data_type != type) {
1915 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
1916 "bucket %llu:%llu gen %u different types of data in same bucket: %s, %s\n"
1918 iter.pos.inode, iter.pos.offset, a->v.gen,
1919 bch2_data_types[a->v.data_type],
1920 bch2_data_types[type],
1921 bch2_data_types[type]);
1926 a->v.data_type = type;
1927 a->v.dirty_sectors = sectors;
1929 ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
1933 bch2_trans_iter_exit(trans, &iter);
1937 int bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
1938 struct bch_dev *ca, size_t b,
1939 enum bch_data_type type,
1942 return commit_do(trans, NULL, NULL, 0,
1943 __bch2_trans_mark_metadata_bucket(trans, ca, b, type, sectors));
1946 static int bch2_trans_mark_metadata_sectors(struct btree_trans *trans,
1949 enum bch_data_type type,
1950 u64 *bucket, unsigned *bucket_sectors)
1953 u64 b = sector_to_bucket(ca, start);
1955 min_t(u64, bucket_to_sector(ca, b + 1), end) - start;
1957 if (b != *bucket && *bucket_sectors) {
1958 int ret = bch2_trans_mark_metadata_bucket(trans, ca, *bucket,
1959 type, *bucket_sectors);
1963 *bucket_sectors = 0;
1967 *bucket_sectors += sectors;
1969 } while (start < end);
1974 static int __bch2_trans_mark_dev_sb(struct btree_trans *trans,
1977 struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
1979 unsigned i, bucket_sectors = 0;
1982 for (i = 0; i < layout->nr_superblocks; i++) {
1983 u64 offset = le64_to_cpu(layout->sb_offset[i]);
1985 if (offset == BCH_SB_SECTOR) {
1986 ret = bch2_trans_mark_metadata_sectors(trans, ca,
1988 BCH_DATA_sb, &bucket, &bucket_sectors);
1993 ret = bch2_trans_mark_metadata_sectors(trans, ca, offset,
1994 offset + (1 << layout->sb_max_size_bits),
1995 BCH_DATA_sb, &bucket, &bucket_sectors);
2000 if (bucket_sectors) {
2001 ret = bch2_trans_mark_metadata_bucket(trans, ca,
2002 bucket, BCH_DATA_sb, bucket_sectors);
2007 for (i = 0; i < ca->journal.nr; i++) {
2008 ret = bch2_trans_mark_metadata_bucket(trans, ca,
2009 ca->journal.buckets[i],
2010 BCH_DATA_journal, ca->mi.bucket_size);
2018 int bch2_trans_mark_dev_sb(struct bch_fs *c, struct bch_dev *ca)
2020 int ret = bch2_trans_run(c, __bch2_trans_mark_dev_sb(&trans, ca));
2026 /* Disk reservations: */
2028 #define SECTORS_CACHE 1024
2030 int __bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
2031 u64 sectors, int flags)
2033 struct bch_fs_pcpu *pcpu;
2035 s64 sectors_available;
2038 percpu_down_read(&c->mark_lock);
2040 pcpu = this_cpu_ptr(c->pcpu);
2042 if (sectors <= pcpu->sectors_available)
2045 v = atomic64_read(&c->sectors_available);
2048 get = min((u64) sectors + SECTORS_CACHE, old);
2050 if (get < sectors) {
2054 } while ((v = atomic64_cmpxchg(&c->sectors_available,
2055 old, old - get)) != old);
2057 pcpu->sectors_available += get;
2060 pcpu->sectors_available -= sectors;
2061 this_cpu_add(*c->online_reserved, sectors);
2062 res->sectors += sectors;
2065 percpu_up_read(&c->mark_lock);
2069 mutex_lock(&c->sectors_available_lock);
2071 percpu_u64_set(&c->pcpu->sectors_available, 0);
2072 sectors_available = avail_factor(__bch2_fs_usage_read_short(c).free);
2074 if (sectors <= sectors_available ||
2075 (flags & BCH_DISK_RESERVATION_NOFAIL)) {
2076 atomic64_set(&c->sectors_available,
2077 max_t(s64, 0, sectors_available - sectors));
2078 this_cpu_add(*c->online_reserved, sectors);
2079 res->sectors += sectors;
2082 atomic64_set(&c->sectors_available, sectors_available);
2083 ret = -BCH_ERR_ENOSPC_disk_reservation;
2086 mutex_unlock(&c->sectors_available_lock);
2087 percpu_up_read(&c->mark_lock);
2092 /* Startup/shutdown: */
2094 static void bucket_gens_free_rcu(struct rcu_head *rcu)
2096 struct bucket_gens *buckets =
2097 container_of(rcu, struct bucket_gens, rcu);
2099 kvpfree(buckets, sizeof(*buckets) + buckets->nbuckets);
2102 int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
2104 struct bucket_gens *bucket_gens = NULL, *old_bucket_gens = NULL;
2105 unsigned long *buckets_nouse = NULL;
2106 bool resize = ca->bucket_gens != NULL;
2109 if (!(bucket_gens = kvpmalloc(sizeof(struct bucket_gens) + nbuckets,
2110 GFP_KERNEL|__GFP_ZERO))) {
2111 ret = -BCH_ERR_ENOMEM_bucket_gens;
2115 if ((c->opts.buckets_nouse &&
2116 !(buckets_nouse = kvpmalloc(BITS_TO_LONGS(nbuckets) *
2117 sizeof(unsigned long),
2118 GFP_KERNEL|__GFP_ZERO)))) {
2119 ret = -BCH_ERR_ENOMEM_buckets_nouse;
2123 bucket_gens->first_bucket = ca->mi.first_bucket;
2124 bucket_gens->nbuckets = nbuckets;
2126 bch2_copygc_stop(c);
2129 down_write(&c->gc_lock);
2130 down_write(&ca->bucket_lock);
2131 percpu_down_write(&c->mark_lock);
2134 old_bucket_gens = rcu_dereference_protected(ca->bucket_gens, 1);
2137 size_t n = min(bucket_gens->nbuckets, old_bucket_gens->nbuckets);
2139 memcpy(bucket_gens->b,
2143 memcpy(buckets_nouse,
2145 BITS_TO_LONGS(n) * sizeof(unsigned long));
2148 rcu_assign_pointer(ca->bucket_gens, bucket_gens);
2149 bucket_gens = old_bucket_gens;
2151 swap(ca->buckets_nouse, buckets_nouse);
2153 nbuckets = ca->mi.nbuckets;
2156 percpu_up_write(&c->mark_lock);
2157 up_write(&ca->bucket_lock);
2158 up_write(&c->gc_lock);
2163 kvpfree(buckets_nouse,
2164 BITS_TO_LONGS(nbuckets) * sizeof(unsigned long));
2166 call_rcu(&bucket_gens->rcu, bucket_gens_free_rcu);
2171 void bch2_dev_buckets_free(struct bch_dev *ca)
2175 kvpfree(ca->buckets_nouse,
2176 BITS_TO_LONGS(ca->mi.nbuckets) * sizeof(unsigned long));
2177 kvpfree(rcu_dereference_protected(ca->bucket_gens, 1),
2178 sizeof(struct bucket_gens) + ca->mi.nbuckets);
2180 for (i = 0; i < ARRAY_SIZE(ca->usage); i++)
2181 free_percpu(ca->usage[i]);
2182 kfree(ca->usage_base);
2185 int bch2_dev_buckets_alloc(struct bch_fs *c, struct bch_dev *ca)
2189 ca->usage_base = kzalloc(sizeof(struct bch_dev_usage), GFP_KERNEL);
2190 if (!ca->usage_base)
2191 return -BCH_ERR_ENOMEM_usage_init;
2193 for (i = 0; i < ARRAY_SIZE(ca->usage); i++) {
2194 ca->usage[i] = alloc_percpu(struct bch_dev_usage);
2196 return -BCH_ERR_ENOMEM_usage_init;
2199 return bch2_dev_buckets_resize(c, ca, ca->mi.nbuckets);