1 // SPDX-License-Identifier: GPL-2.0
3 #include "alloc_background.h"
4 #include "alloc_foreground.h"
5 #include "backpointers.h"
6 #include "btree_cache.h"
8 #include "btree_key_cache.h"
9 #include "btree_update.h"
10 #include "btree_update_interior.h"
12 #include "btree_write_buffer.h"
14 #include "buckets_waiting_for_journal.h"
24 #include <linux/kthread.h>
25 #include <linux/math64.h>
26 #include <linux/random.h>
27 #include <linux/rculist.h>
28 #include <linux/rcupdate.h>
29 #include <linux/sched/task.h>
30 #include <linux/sort.h>
32 /* Persistent alloc info: */
34 static const unsigned BCH_ALLOC_V1_FIELD_BYTES[] = {
35 #define x(name, bits) [BCH_ALLOC_FIELD_V1_##name] = bits / 8,
40 struct bkey_alloc_unpacked {
47 #define x(_name, _bits) u##_bits _name;
52 static inline u64 alloc_field_v1_get(const struct bch_alloc *a,
53 const void **p, unsigned field)
55 unsigned bytes = BCH_ALLOC_V1_FIELD_BYTES[field];
58 if (!(a->fields & (1 << field)))
63 v = *((const u8 *) *p);
82 static void bch2_alloc_unpack_v1(struct bkey_alloc_unpacked *out,
85 const struct bch_alloc *in = bkey_s_c_to_alloc(k).v;
86 const void *d = in->data;
91 #define x(_name, _bits) out->_name = alloc_field_v1_get(in, &d, idx++);
96 static int bch2_alloc_unpack_v2(struct bkey_alloc_unpacked *out,
99 struct bkey_s_c_alloc_v2 a = bkey_s_c_to_alloc_v2(k);
100 const u8 *in = a.v->data;
101 const u8 *end = bkey_val_end(a);
102 unsigned fieldnr = 0;
107 out->oldest_gen = a.v->oldest_gen;
108 out->data_type = a.v->data_type;
110 #define x(_name, _bits) \
111 if (fieldnr < a.v->nr_fields) { \
112 ret = bch2_varint_decode_fast(in, end, &v); \
120 if (v != out->_name) \
124 BCH_ALLOC_FIELDS_V2()
129 static int bch2_alloc_unpack_v3(struct bkey_alloc_unpacked *out,
132 struct bkey_s_c_alloc_v3 a = bkey_s_c_to_alloc_v3(k);
133 const u8 *in = a.v->data;
134 const u8 *end = bkey_val_end(a);
135 unsigned fieldnr = 0;
140 out->oldest_gen = a.v->oldest_gen;
141 out->data_type = a.v->data_type;
142 out->need_discard = BCH_ALLOC_V3_NEED_DISCARD(a.v);
143 out->need_inc_gen = BCH_ALLOC_V3_NEED_INC_GEN(a.v);
144 out->journal_seq = le64_to_cpu(a.v->journal_seq);
146 #define x(_name, _bits) \
147 if (fieldnr < a.v->nr_fields) { \
148 ret = bch2_varint_decode_fast(in, end, &v); \
156 if (v != out->_name) \
160 BCH_ALLOC_FIELDS_V2()
165 static struct bkey_alloc_unpacked bch2_alloc_unpack(struct bkey_s_c k)
167 struct bkey_alloc_unpacked ret = { .gen = 0 };
171 bch2_alloc_unpack_v1(&ret, k);
173 case KEY_TYPE_alloc_v2:
174 bch2_alloc_unpack_v2(&ret, k);
176 case KEY_TYPE_alloc_v3:
177 bch2_alloc_unpack_v3(&ret, k);
184 static unsigned bch_alloc_v1_val_u64s(const struct bch_alloc *a)
186 unsigned i, bytes = offsetof(struct bch_alloc, data);
188 for (i = 0; i < ARRAY_SIZE(BCH_ALLOC_V1_FIELD_BYTES); i++)
189 if (a->fields & (1 << i))
190 bytes += BCH_ALLOC_V1_FIELD_BYTES[i];
192 return DIV_ROUND_UP(bytes, sizeof(u64));
195 int bch2_alloc_v1_invalid(struct bch_fs *c, struct bkey_s_c k,
196 enum bkey_invalid_flags flags,
197 struct printbuf *err)
199 struct bkey_s_c_alloc a = bkey_s_c_to_alloc(k);
202 /* allow for unknown fields */
203 bkey_fsck_err_on(bkey_val_u64s(a.k) < bch_alloc_v1_val_u64s(a.v), c, err,
204 alloc_v1_val_size_bad,
205 "incorrect value size (%zu < %u)",
206 bkey_val_u64s(a.k), bch_alloc_v1_val_u64s(a.v));
211 int bch2_alloc_v2_invalid(struct bch_fs *c, struct bkey_s_c k,
212 enum bkey_invalid_flags flags,
213 struct printbuf *err)
215 struct bkey_alloc_unpacked u;
218 bkey_fsck_err_on(bch2_alloc_unpack_v2(&u, k), c, err,
219 alloc_v2_unpack_error,
225 int bch2_alloc_v3_invalid(struct bch_fs *c, struct bkey_s_c k,
226 enum bkey_invalid_flags flags,
227 struct printbuf *err)
229 struct bkey_alloc_unpacked u;
232 bkey_fsck_err_on(bch2_alloc_unpack_v3(&u, k), c, err,
233 alloc_v2_unpack_error,
239 int bch2_alloc_v4_invalid(struct bch_fs *c, struct bkey_s_c k,
240 enum bkey_invalid_flags flags, struct printbuf *err)
242 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
245 bkey_fsck_err_on(alloc_v4_u64s(a.v) > bkey_val_u64s(k.k), c, err,
246 alloc_v4_val_size_bad,
247 "bad val size (%u > %zu)",
248 alloc_v4_u64s(a.v), bkey_val_u64s(k.k));
250 bkey_fsck_err_on(!BCH_ALLOC_V4_BACKPOINTERS_START(a.v) &&
251 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v), c, err,
252 alloc_v4_backpointers_start_bad,
253 "invalid backpointers_start");
255 bkey_fsck_err_on(alloc_data_type(*a.v, a.v->data_type) != a.v->data_type, c, err,
256 alloc_key_data_type_bad,
257 "invalid data type (got %u should be %u)",
258 a.v->data_type, alloc_data_type(*a.v, a.v->data_type));
260 switch (a.v->data_type) {
262 case BCH_DATA_need_gc_gens:
263 case BCH_DATA_need_discard:
264 bkey_fsck_err_on(bch2_bucket_sectors(*a.v) || a.v->stripe,
265 c, err, alloc_key_empty_but_have_data,
266 "empty data type free but have data");
269 case BCH_DATA_journal:
272 case BCH_DATA_parity:
273 bkey_fsck_err_on(!bch2_bucket_sectors_dirty(*a.v),
274 c, err, alloc_key_dirty_sectors_0,
275 "data_type %s but dirty_sectors==0",
276 bch2_data_type_str(a.v->data_type));
278 case BCH_DATA_cached:
279 bkey_fsck_err_on(!a.v->cached_sectors ||
280 bch2_bucket_sectors_dirty(*a.v) ||
282 c, err, alloc_key_cached_inconsistency,
283 "data type inconsistency");
285 bkey_fsck_err_on(!a.v->io_time[READ] &&
286 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_to_lru_refs,
287 c, err, alloc_key_cached_but_read_time_zero,
288 "cached bucket with read_time == 0");
290 case BCH_DATA_stripe:
297 void bch2_alloc_v4_swab(struct bkey_s k)
299 struct bch_alloc_v4 *a = bkey_s_to_alloc_v4(k).v;
300 struct bch_backpointer *bp, *bps;
302 a->journal_seq = swab64(a->journal_seq);
303 a->flags = swab32(a->flags);
304 a->dirty_sectors = swab32(a->dirty_sectors);
305 a->cached_sectors = swab32(a->cached_sectors);
306 a->io_time[0] = swab64(a->io_time[0]);
307 a->io_time[1] = swab64(a->io_time[1]);
308 a->stripe = swab32(a->stripe);
309 a->nr_external_backpointers = swab32(a->nr_external_backpointers);
310 a->fragmentation_lru = swab64(a->fragmentation_lru);
312 bps = alloc_v4_backpointers(a);
313 for (bp = bps; bp < bps + BCH_ALLOC_V4_NR_BACKPOINTERS(a); bp++) {
314 bp->bucket_offset = swab40(bp->bucket_offset);
315 bp->bucket_len = swab32(bp->bucket_len);
316 bch2_bpos_swab(&bp->pos);
320 void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
322 struct bch_alloc_v4 _a;
323 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a);
326 printbuf_indent_add(out, 2);
328 prt_printf(out, "gen %u oldest_gen %u data_type ", a->gen, a->oldest_gen);
329 bch2_prt_data_type(out, a->data_type);
331 prt_printf(out, "journal_seq %llu", a->journal_seq);
333 prt_printf(out, "need_discard %llu", BCH_ALLOC_V4_NEED_DISCARD(a));
335 prt_printf(out, "need_inc_gen %llu", BCH_ALLOC_V4_NEED_INC_GEN(a));
337 prt_printf(out, "dirty_sectors %u", a->dirty_sectors);
339 prt_printf(out, "cached_sectors %u", a->cached_sectors);
341 prt_printf(out, "stripe %u", a->stripe);
343 prt_printf(out, "stripe_redundancy %u", a->stripe_redundancy);
345 prt_printf(out, "io_time[READ] %llu", a->io_time[READ]);
347 prt_printf(out, "io_time[WRITE] %llu", a->io_time[WRITE]);
349 prt_printf(out, "fragmentation %llu", a->fragmentation_lru);
351 prt_printf(out, "bp_start %llu", BCH_ALLOC_V4_BACKPOINTERS_START(a));
352 printbuf_indent_sub(out, 2);
355 void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out)
357 if (k.k->type == KEY_TYPE_alloc_v4) {
360 *out = *bkey_s_c_to_alloc_v4(k).v;
362 src = alloc_v4_backpointers(out);
363 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
364 dst = alloc_v4_backpointers(out);
367 memset(src, 0, dst - src);
369 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(out, 0);
371 struct bkey_alloc_unpacked u = bch2_alloc_unpack(k);
373 *out = (struct bch_alloc_v4) {
374 .journal_seq = u.journal_seq,
375 .flags = u.need_discard,
377 .oldest_gen = u.oldest_gen,
378 .data_type = u.data_type,
379 .stripe_redundancy = u.stripe_redundancy,
380 .dirty_sectors = u.dirty_sectors,
381 .cached_sectors = u.cached_sectors,
382 .io_time[READ] = u.read_time,
383 .io_time[WRITE] = u.write_time,
387 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
391 static noinline struct bkey_i_alloc_v4 *
392 __bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
394 struct bkey_i_alloc_v4 *ret;
396 ret = bch2_trans_kmalloc(trans, max(bkey_bytes(k.k), sizeof(struct bkey_i_alloc_v4)));
400 if (k.k->type == KEY_TYPE_alloc_v4) {
403 bkey_reassemble(&ret->k_i, k);
405 src = alloc_v4_backpointers(&ret->v);
406 SET_BCH_ALLOC_V4_BACKPOINTERS_START(&ret->v, BCH_ALLOC_V4_U64s);
407 dst = alloc_v4_backpointers(&ret->v);
410 memset(src, 0, dst - src);
412 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(&ret->v, 0);
413 set_alloc_v4_u64s(ret);
415 bkey_alloc_v4_init(&ret->k_i);
417 bch2_alloc_to_v4(k, &ret->v);
422 static inline struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut_inlined(struct btree_trans *trans, struct bkey_s_c k)
424 struct bkey_s_c_alloc_v4 a;
426 if (likely(k.k->type == KEY_TYPE_alloc_v4) &&
427 ((a = bkey_s_c_to_alloc_v4(k), true) &&
428 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v) == 0))
429 return bch2_bkey_make_mut_noupdate_typed(trans, k, alloc_v4);
431 return __bch2_alloc_to_v4_mut(trans, k);
434 struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
436 return bch2_alloc_to_v4_mut_inlined(trans, k);
439 struct bkey_i_alloc_v4 *
440 bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter,
444 struct bkey_i_alloc_v4 *a;
447 k = bch2_bkey_get_iter(trans, iter, BTREE_ID_alloc, pos,
448 BTREE_ITER_WITH_UPDATES|
455 a = bch2_alloc_to_v4_mut_inlined(trans, k);
456 ret = PTR_ERR_OR_ZERO(a);
461 bch2_trans_iter_exit(trans, iter);
465 static struct bpos alloc_gens_pos(struct bpos pos, unsigned *offset)
467 *offset = pos.offset & KEY_TYPE_BUCKET_GENS_MASK;
469 pos.offset >>= KEY_TYPE_BUCKET_GENS_BITS;
473 static struct bpos bucket_gens_pos_to_alloc(struct bpos pos, unsigned offset)
475 pos.offset <<= KEY_TYPE_BUCKET_GENS_BITS;
476 pos.offset += offset;
480 static unsigned alloc_gen(struct bkey_s_c k, unsigned offset)
482 return k.k->type == KEY_TYPE_bucket_gens
483 ? bkey_s_c_to_bucket_gens(k).v->gens[offset]
487 int bch2_bucket_gens_invalid(struct bch_fs *c, struct bkey_s_c k,
488 enum bkey_invalid_flags flags,
489 struct printbuf *err)
493 bkey_fsck_err_on(bkey_val_bytes(k.k) != sizeof(struct bch_bucket_gens), c, err,
494 bucket_gens_val_size_bad,
495 "bad val size (%zu != %zu)",
496 bkey_val_bytes(k.k), sizeof(struct bch_bucket_gens));
501 void bch2_bucket_gens_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
503 struct bkey_s_c_bucket_gens g = bkey_s_c_to_bucket_gens(k);
506 for (i = 0; i < ARRAY_SIZE(g.v->gens); i++) {
509 prt_printf(out, "%u", g.v->gens[i]);
513 int bch2_bucket_gens_init(struct bch_fs *c)
515 struct btree_trans *trans = bch2_trans_get(c);
516 struct bkey_i_bucket_gens g;
517 bool have_bucket_gens_key = false;
520 ret = for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
521 BTREE_ITER_PREFETCH, k, ({
523 * Not a fsck error because this is checked/repaired by
524 * bch2_check_alloc_key() which runs later:
526 if (!bch2_dev_bucket_exists(c, k.k->p))
529 struct bch_alloc_v4 a;
530 u8 gen = bch2_alloc_to_v4(k, &a)->gen;
532 struct bpos pos = alloc_gens_pos(iter.pos, &offset);
534 if (have_bucket_gens_key && bkey_cmp(iter.pos, pos)) {
535 ret = commit_do(trans, NULL, NULL,
536 BCH_TRANS_COMMIT_no_enospc,
537 bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
540 have_bucket_gens_key = false;
543 if (!have_bucket_gens_key) {
544 bkey_bucket_gens_init(&g.k_i);
546 have_bucket_gens_key = true;
549 g.v.gens[offset] = gen;
553 if (have_bucket_gens_key && !ret)
554 ret = commit_do(trans, NULL, NULL,
555 BCH_TRANS_COMMIT_no_enospc,
556 bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
558 bch2_trans_put(trans);
564 int bch2_alloc_read(struct bch_fs *c)
566 struct btree_trans *trans = bch2_trans_get(c);
569 down_read(&c->gc_lock);
571 if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_bucket_gens) {
572 ret = for_each_btree_key(trans, iter, BTREE_ID_bucket_gens, POS_MIN,
573 BTREE_ITER_PREFETCH, k, ({
574 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
575 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
577 if (k.k->type != KEY_TYPE_bucket_gens)
580 const struct bch_bucket_gens *g = bkey_s_c_to_bucket_gens(k).v;
583 * Not a fsck error because this is checked/repaired by
584 * bch2_check_alloc_key() which runs later:
586 if (!bch2_dev_exists2(c, k.k->p.inode))
589 struct bch_dev *ca = bch_dev_bkey_exists(c, k.k->p.inode);
591 for (u64 b = max_t(u64, ca->mi.first_bucket, start);
592 b < min_t(u64, ca->mi.nbuckets, end);
594 *bucket_gen(ca, b) = g->gens[b & KEY_TYPE_BUCKET_GENS_MASK];
598 ret = for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
599 BTREE_ITER_PREFETCH, k, ({
601 * Not a fsck error because this is checked/repaired by
602 * bch2_check_alloc_key() which runs later:
604 if (!bch2_dev_bucket_exists(c, k.k->p))
607 struct bch_dev *ca = bch_dev_bkey_exists(c, k.k->p.inode);
609 struct bch_alloc_v4 a;
610 *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen;
615 bch2_trans_put(trans);
616 up_read(&c->gc_lock);
622 /* Free space/discard btree: */
624 static int bch2_bucket_do_index(struct btree_trans *trans,
625 struct bkey_s_c alloc_k,
626 const struct bch_alloc_v4 *a,
629 struct bch_fs *c = trans->c;
630 struct bch_dev *ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
631 struct btree_iter iter;
635 enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted;
636 enum bch_bkey_type new_type = set ? KEY_TYPE_set : KEY_TYPE_deleted;
637 struct printbuf buf = PRINTBUF;
640 if (a->data_type != BCH_DATA_free &&
641 a->data_type != BCH_DATA_need_discard)
644 k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k));
649 k->k.type = new_type;
651 switch (a->data_type) {
653 btree = BTREE_ID_freespace;
654 k->k.p = alloc_freespace_pos(alloc_k.k->p, *a);
655 bch2_key_resize(&k->k, 1);
657 case BCH_DATA_need_discard:
658 btree = BTREE_ID_need_discard;
659 k->k.p = alloc_k.k->p;
665 old = bch2_bkey_get_iter(trans, &iter, btree,
666 bkey_start_pos(&k->k),
672 if (ca->mi.freespace_initialized &&
673 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info &&
674 bch2_trans_inconsistent_on(old.k->type != old_type, trans,
675 "incorrect key when %s %s:%llu:%llu:0 (got %s should be %s)\n"
677 set ? "setting" : "clearing",
678 bch2_btree_id_str(btree),
681 bch2_bkey_types[old.k->type],
682 bch2_bkey_types[old_type],
683 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
688 ret = bch2_trans_update(trans, &iter, k, 0);
690 bch2_trans_iter_exit(trans, &iter);
695 static noinline int bch2_bucket_gen_update(struct btree_trans *trans,
696 struct bpos bucket, u8 gen)
698 struct btree_iter iter;
700 struct bpos pos = alloc_gens_pos(bucket, &offset);
701 struct bkey_i_bucket_gens *g;
705 g = bch2_trans_kmalloc(trans, sizeof(*g));
706 ret = PTR_ERR_OR_ZERO(g);
710 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_bucket_gens, pos,
712 BTREE_ITER_WITH_UPDATES);
717 if (k.k->type != KEY_TYPE_bucket_gens) {
718 bkey_bucket_gens_init(&g->k_i);
721 bkey_reassemble(&g->k_i, k);
724 g->v.gens[offset] = gen;
726 ret = bch2_trans_update(trans, &iter, &g->k_i, 0);
727 bch2_trans_iter_exit(trans, &iter);
731 int bch2_trigger_alloc(struct btree_trans *trans,
732 enum btree_id btree, unsigned level,
733 struct bkey_s_c old, struct bkey_s new,
736 struct bch_fs *c = trans->c;
739 if (bch2_trans_inconsistent_on(!bch2_dev_bucket_exists(c, new.k->p), trans,
740 "alloc key for invalid device or bucket"))
743 struct bch_dev *ca = bch_dev_bkey_exists(c, new.k->p.inode);
745 struct bch_alloc_v4 old_a_convert;
746 const struct bch_alloc_v4 *old_a = bch2_alloc_to_v4(old, &old_a_convert);
748 if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
749 struct bch_alloc_v4 *new_a = bkey_s_to_alloc_v4(new).v;
751 new_a->data_type = alloc_data_type(*new_a, new_a->data_type);
753 if (bch2_bucket_sectors(*new_a) > bch2_bucket_sectors(*old_a)) {
754 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
755 new_a->io_time[WRITE]= max_t(u64, 1, atomic64_read(&c->io_clock[WRITE].now));
756 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true);
757 SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true);
760 if (data_type_is_empty(new_a->data_type) &&
761 BCH_ALLOC_V4_NEED_INC_GEN(new_a) &&
762 !bch2_bucket_is_open_safe(c, new.k->p.inode, new.k->p.offset)) {
764 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false);
767 if (old_a->data_type != new_a->data_type ||
768 (new_a->data_type == BCH_DATA_free &&
769 alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) {
770 ret = bch2_bucket_do_index(trans, old, old_a, false) ?:
771 bch2_bucket_do_index(trans, new.s_c, new_a, true);
776 if (new_a->data_type == BCH_DATA_cached &&
777 !new_a->io_time[READ])
778 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
780 u64 old_lru = alloc_lru_idx_read(*old_a);
781 u64 new_lru = alloc_lru_idx_read(*new_a);
782 if (old_lru != new_lru) {
783 ret = bch2_lru_change(trans, new.k->p.inode,
784 bucket_to_u64(new.k->p),
790 new_a->fragmentation_lru = alloc_lru_idx_fragmentation(*new_a,
791 bch_dev_bkey_exists(c, new.k->p.inode));
792 if (old_a->fragmentation_lru != new_a->fragmentation_lru) {
793 ret = bch2_lru_change(trans,
794 BCH_LRU_FRAGMENTATION_START,
795 bucket_to_u64(new.k->p),
796 old_a->fragmentation_lru, new_a->fragmentation_lru);
801 if (old_a->gen != new_a->gen) {
802 ret = bch2_bucket_gen_update(trans, new.k->p, new_a->gen);
808 * need to know if we're getting called from the invalidate path or
812 if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
813 old_a->cached_sectors) {
814 ret = bch2_update_cached_sectors_list(trans, new.k->p.inode,
815 -((s64) old_a->cached_sectors));
821 if ((flags & BTREE_TRIGGER_ATOMIC) && (flags & BTREE_TRIGGER_INSERT)) {
822 struct bch_alloc_v4 *new_a = bkey_s_to_alloc_v4(new).v;
823 u64 journal_seq = trans->journal_res.seq;
824 u64 bucket_journal_seq = new_a->journal_seq;
826 if ((flags & BTREE_TRIGGER_INSERT) &&
827 data_type_is_empty(old_a->data_type) !=
828 data_type_is_empty(new_a->data_type) &&
829 new.k->type == KEY_TYPE_alloc_v4) {
830 struct bch_alloc_v4 *v = bkey_s_to_alloc_v4(new).v;
833 * If the btree updates referring to a bucket weren't flushed
834 * before the bucket became empty again, then the we don't have
835 * to wait on a journal flush before we can reuse the bucket:
837 v->journal_seq = bucket_journal_seq =
838 data_type_is_empty(new_a->data_type) &&
839 (journal_seq == v->journal_seq ||
840 bch2_journal_noflush_seq(&c->journal, v->journal_seq))
844 if (!data_type_is_empty(old_a->data_type) &&
845 data_type_is_empty(new_a->data_type) &&
846 bucket_journal_seq) {
847 ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
848 c->journal.flushed_seq_ondisk,
849 new.k->p.inode, new.k->p.offset,
852 bch2_fs_fatal_error(c,
853 "error setting bucket_needs_journal_commit: %i", ret);
858 percpu_down_read(&c->mark_lock);
859 if (new_a->gen != old_a->gen)
860 *bucket_gen(ca, new.k->p.offset) = new_a->gen;
862 bch2_dev_usage_update(c, ca, old_a, new_a, journal_seq, false);
864 if (new_a->data_type == BCH_DATA_free &&
865 (!new_a->journal_seq || new_a->journal_seq < c->journal.flushed_seq_ondisk))
866 closure_wake_up(&c->freelist_wait);
868 if (new_a->data_type == BCH_DATA_need_discard &&
869 (!bucket_journal_seq || bucket_journal_seq < c->journal.flushed_seq_ondisk))
872 if (old_a->data_type != BCH_DATA_cached &&
873 new_a->data_type == BCH_DATA_cached &&
874 should_invalidate_buckets(ca, bch2_dev_usage_read(ca)))
875 bch2_do_invalidates(c);
877 if (new_a->data_type == BCH_DATA_need_gc_gens)
879 percpu_up_read(&c->mark_lock);
882 if ((flags & BTREE_TRIGGER_GC) &&
883 (flags & BTREE_TRIGGER_BUCKET_INVALIDATE)) {
884 struct bch_alloc_v4 new_a_convert;
885 const struct bch_alloc_v4 *new_a = bch2_alloc_to_v4(new.s_c, &new_a_convert);
887 percpu_down_read(&c->mark_lock);
888 struct bucket *g = gc_bucket(ca, new.k->p.offset);
894 g->data_type = new_a->data_type;
895 g->stripe = new_a->stripe;
896 g->stripe_redundancy = new_a->stripe_redundancy;
897 g->dirty_sectors = new_a->dirty_sectors;
898 g->cached_sectors = new_a->cached_sectors;
901 percpu_up_read(&c->mark_lock);
908 * This synthesizes deleted extents for holes, similar to BTREE_ITER_SLOTS for
909 * extents style btrees, but works on non-extents btrees:
911 static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos end, struct bkey *hole)
913 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
921 struct btree_iter iter2;
924 bch2_trans_copy_iter(&iter2, iter);
926 struct btree_path *path = btree_iter_path(iter->trans, iter);
927 if (!bpos_eq(path->l[0].b->key.k.p, SPOS_MAX))
928 end = bkey_min(end, bpos_nosnap_successor(path->l[0].b->key.k.p));
930 end = bkey_min(end, POS(iter->pos.inode, iter->pos.offset + U32_MAX - 1));
933 * btree node min/max is a closed interval, upto takes a half
936 k = bch2_btree_iter_peek_upto(&iter2, end);
938 bch2_trans_iter_exit(iter->trans, &iter2);
940 BUG_ON(next.offset >= iter->pos.offset + U32_MAX);
948 bch2_key_resize(hole, next.offset - iter->pos.offset);
949 return (struct bkey_s_c) { hole, NULL };
953 static bool next_bucket(struct bch_fs *c, struct bpos *bucket)
957 if (bch2_dev_bucket_exists(c, *bucket))
960 if (bch2_dev_exists2(c, bucket->inode)) {
961 ca = bch_dev_bkey_exists(c, bucket->inode);
963 if (bucket->offset < ca->mi.first_bucket) {
964 bucket->offset = ca->mi.first_bucket;
973 ca = __bch2_next_dev_idx(c, bucket->inode, NULL);
975 *bucket = POS(ca->dev_idx, ca->mi.first_bucket);
981 static struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, struct bkey *hole)
983 struct bch_fs *c = iter->trans->c;
986 k = bch2_get_key_or_hole(iter, POS_MAX, hole);
991 struct bpos bucket = bkey_start_pos(k.k);
993 if (!bch2_dev_bucket_exists(c, bucket)) {
994 if (!next_bucket(c, &bucket))
995 return bkey_s_c_null;
997 bch2_btree_iter_set_pos(iter, bucket);
1001 if (!bch2_dev_bucket_exists(c, k.k->p)) {
1002 struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode);
1004 bch2_key_resize(hole, ca->mi.nbuckets - bucket.offset);
1011 static noinline_for_stack
1012 int bch2_check_alloc_key(struct btree_trans *trans,
1013 struct bkey_s_c alloc_k,
1014 struct btree_iter *alloc_iter,
1015 struct btree_iter *discard_iter,
1016 struct btree_iter *freespace_iter,
1017 struct btree_iter *bucket_gens_iter)
1019 struct bch_fs *c = trans->c;
1021 struct bch_alloc_v4 a_convert;
1022 const struct bch_alloc_v4 *a;
1023 unsigned discard_key_type, freespace_key_type;
1024 unsigned gens_offset;
1026 struct printbuf buf = PRINTBUF;
1029 if (fsck_err_on(!bch2_dev_bucket_exists(c, alloc_k.k->p), c,
1030 alloc_key_to_missing_dev_bucket,
1031 "alloc key for invalid device:bucket %llu:%llu",
1032 alloc_k.k->p.inode, alloc_k.k->p.offset))
1033 return bch2_btree_delete_at(trans, alloc_iter, 0);
1035 ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
1036 if (!ca->mi.freespace_initialized)
1039 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1041 discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0;
1042 bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p);
1043 k = bch2_btree_iter_peek_slot(discard_iter);
1048 if (k.k->type != discard_key_type &&
1049 (c->opts.reconstruct_alloc ||
1050 fsck_err(c, need_discard_key_wrong,
1051 "incorrect key in need_discard btree (got %s should be %s)\n"
1053 bch2_bkey_types[k.k->type],
1054 bch2_bkey_types[discard_key_type],
1055 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1056 struct bkey_i *update =
1057 bch2_trans_kmalloc(trans, sizeof(*update));
1059 ret = PTR_ERR_OR_ZERO(update);
1063 bkey_init(&update->k);
1064 update->k.type = discard_key_type;
1065 update->k.p = discard_iter->pos;
1067 ret = bch2_trans_update(trans, discard_iter, update, 0);
1072 freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0;
1073 bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a));
1074 k = bch2_btree_iter_peek_slot(freespace_iter);
1079 if (k.k->type != freespace_key_type &&
1080 (c->opts.reconstruct_alloc ||
1081 fsck_err(c, freespace_key_wrong,
1082 "incorrect key in freespace btree (got %s should be %s)\n"
1084 bch2_bkey_types[k.k->type],
1085 bch2_bkey_types[freespace_key_type],
1086 (printbuf_reset(&buf),
1087 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1088 struct bkey_i *update =
1089 bch2_trans_kmalloc(trans, sizeof(*update));
1091 ret = PTR_ERR_OR_ZERO(update);
1095 bkey_init(&update->k);
1096 update->k.type = freespace_key_type;
1097 update->k.p = freespace_iter->pos;
1098 bch2_key_resize(&update->k, 1);
1100 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1105 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(alloc_k.k->p, &gens_offset));
1106 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1111 if (a->gen != alloc_gen(k, gens_offset) &&
1112 (c->opts.reconstruct_alloc ||
1113 fsck_err(c, bucket_gens_key_wrong,
1114 "incorrect gen in bucket_gens btree (got %u should be %u)\n"
1116 alloc_gen(k, gens_offset), a->gen,
1117 (printbuf_reset(&buf),
1118 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1119 struct bkey_i_bucket_gens *g =
1120 bch2_trans_kmalloc(trans, sizeof(*g));
1122 ret = PTR_ERR_OR_ZERO(g);
1126 if (k.k->type == KEY_TYPE_bucket_gens) {
1127 bkey_reassemble(&g->k_i, k);
1129 bkey_bucket_gens_init(&g->k_i);
1130 g->k.p = alloc_gens_pos(alloc_k.k->p, &gens_offset);
1133 g->v.gens[gens_offset] = a->gen;
1135 ret = bch2_trans_update(trans, bucket_gens_iter, &g->k_i, 0);
1141 printbuf_exit(&buf);
1145 static noinline_for_stack
1146 int bch2_check_alloc_hole_freespace(struct btree_trans *trans,
1149 struct btree_iter *freespace_iter)
1151 struct bch_fs *c = trans->c;
1154 struct printbuf buf = PRINTBUF;
1157 ca = bch_dev_bkey_exists(c, start.inode);
1158 if (!ca->mi.freespace_initialized)
1161 bch2_btree_iter_set_pos(freespace_iter, start);
1163 k = bch2_btree_iter_peek_slot(freespace_iter);
1168 *end = bkey_min(k.k->p, *end);
1170 if (k.k->type != KEY_TYPE_set &&
1171 (c->opts.reconstruct_alloc ||
1172 fsck_err(c, freespace_hole_missing,
1173 "hole in alloc btree missing in freespace btree\n"
1174 " device %llu buckets %llu-%llu",
1175 freespace_iter->pos.inode,
1176 freespace_iter->pos.offset,
1178 struct bkey_i *update =
1179 bch2_trans_kmalloc(trans, sizeof(*update));
1181 ret = PTR_ERR_OR_ZERO(update);
1185 bkey_init(&update->k);
1186 update->k.type = KEY_TYPE_set;
1187 update->k.p = freespace_iter->pos;
1188 bch2_key_resize(&update->k,
1189 min_t(u64, U32_MAX, end->offset -
1190 freespace_iter->pos.offset));
1192 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1198 printbuf_exit(&buf);
1202 static noinline_for_stack
1203 int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans,
1206 struct btree_iter *bucket_gens_iter)
1208 struct bch_fs *c = trans->c;
1210 struct printbuf buf = PRINTBUF;
1211 unsigned i, gens_offset, gens_end_offset;
1214 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(start, &gens_offset));
1216 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1221 if (bkey_cmp(alloc_gens_pos(start, &gens_offset),
1222 alloc_gens_pos(*end, &gens_end_offset)))
1223 gens_end_offset = KEY_TYPE_BUCKET_GENS_NR;
1225 if (k.k->type == KEY_TYPE_bucket_gens) {
1226 struct bkey_i_bucket_gens g;
1227 bool need_update = false;
1229 bkey_reassemble(&g.k_i, k);
1231 for (i = gens_offset; i < gens_end_offset; i++) {
1232 if (fsck_err_on(g.v.gens[i], c,
1233 bucket_gens_hole_wrong,
1234 "hole in alloc btree at %llu:%llu with nonzero gen in bucket_gens btree (%u)",
1235 bucket_gens_pos_to_alloc(k.k->p, i).inode,
1236 bucket_gens_pos_to_alloc(k.k->p, i).offset,
1244 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1246 ret = PTR_ERR_OR_ZERO(u);
1250 memcpy(u, &g, sizeof(g));
1252 ret = bch2_trans_update(trans, bucket_gens_iter, u, 0);
1258 *end = bkey_min(*end, bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0));
1261 printbuf_exit(&buf);
1265 static noinline_for_stack int bch2_check_discard_freespace_key(struct btree_trans *trans,
1266 struct btree_iter *iter)
1268 struct bch_fs *c = trans->c;
1269 struct btree_iter alloc_iter;
1270 struct bkey_s_c alloc_k;
1271 struct bch_alloc_v4 a_convert;
1272 const struct bch_alloc_v4 *a;
1275 enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard
1276 ? BCH_DATA_need_discard
1278 struct printbuf buf = PRINTBUF;
1282 pos.offset &= ~(~0ULL << 56);
1283 genbits = iter->pos.offset & (~0ULL << 56);
1285 alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc, pos, 0);
1286 ret = bkey_err(alloc_k);
1290 if (fsck_err_on(!bch2_dev_bucket_exists(c, pos), c,
1291 need_discard_freespace_key_to_invalid_dev_bucket,
1292 "entry in %s btree for nonexistant dev:bucket %llu:%llu",
1293 bch2_btree_id_str(iter->btree_id), pos.inode, pos.offset))
1296 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1298 if (fsck_err_on(a->data_type != state ||
1299 (state == BCH_DATA_free &&
1300 genbits != alloc_freespace_genbits(*a)), c,
1301 need_discard_freespace_key_bad,
1302 "%s\n incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)",
1303 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
1304 bch2_btree_id_str(iter->btree_id),
1307 a->data_type == state,
1308 genbits >> 56, alloc_freespace_genbits(*a) >> 56))
1312 set_btree_iter_dontneed(&alloc_iter);
1313 bch2_trans_iter_exit(trans, &alloc_iter);
1314 printbuf_exit(&buf);
1317 ret = bch2_btree_delete_extent_at(trans, iter,
1318 iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0) ?:
1319 bch2_trans_commit(trans, NULL, NULL,
1320 BCH_TRANS_COMMIT_no_enospc);
1325 * We've already checked that generation numbers in the bucket_gens btree are
1326 * valid for buckets that exist; this just checks for keys for nonexistent
1329 static noinline_for_stack
1330 int bch2_check_bucket_gens_key(struct btree_trans *trans,
1331 struct btree_iter *iter,
1334 struct bch_fs *c = trans->c;
1335 struct bkey_i_bucket_gens g;
1337 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
1338 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
1340 bool need_update = false, dev_exists;
1341 struct printbuf buf = PRINTBUF;
1344 BUG_ON(k.k->type != KEY_TYPE_bucket_gens);
1345 bkey_reassemble(&g.k_i, k);
1347 /* if no bch_dev, skip out whether we repair or not */
1348 dev_exists = bch2_dev_exists2(c, k.k->p.inode);
1350 if (fsck_err_on(!dev_exists, c,
1351 bucket_gens_to_invalid_dev,
1352 "bucket_gens key for invalid device:\n %s",
1353 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1354 ret = bch2_btree_delete_at(trans, iter, 0);
1359 ca = bch_dev_bkey_exists(c, k.k->p.inode);
1360 if (fsck_err_on(end <= ca->mi.first_bucket ||
1361 start >= ca->mi.nbuckets, c,
1362 bucket_gens_to_invalid_buckets,
1363 "bucket_gens key for invalid buckets:\n %s",
1364 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1365 ret = bch2_btree_delete_at(trans, iter, 0);
1369 for (b = start; b < ca->mi.first_bucket; b++)
1370 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1371 bucket_gens_nonzero_for_invalid_buckets,
1372 "bucket_gens key has nonzero gen for invalid bucket")) {
1373 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1377 for (b = ca->mi.nbuckets; b < end; b++)
1378 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1379 bucket_gens_nonzero_for_invalid_buckets,
1380 "bucket_gens key has nonzero gen for invalid bucket")) {
1381 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1386 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1388 ret = PTR_ERR_OR_ZERO(u);
1392 memcpy(u, &g, sizeof(g));
1393 ret = bch2_trans_update(trans, iter, u, 0);
1397 printbuf_exit(&buf);
1401 int bch2_check_alloc_info(struct bch_fs *c)
1403 struct btree_trans *trans = bch2_trans_get(c);
1404 struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter;
1409 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS_MIN,
1410 BTREE_ITER_PREFETCH);
1411 bch2_trans_iter_init(trans, &discard_iter, BTREE_ID_need_discard, POS_MIN,
1412 BTREE_ITER_PREFETCH);
1413 bch2_trans_iter_init(trans, &freespace_iter, BTREE_ID_freespace, POS_MIN,
1414 BTREE_ITER_PREFETCH);
1415 bch2_trans_iter_init(trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN,
1416 BTREE_ITER_PREFETCH);
1421 bch2_trans_begin(trans);
1423 k = bch2_get_key_or_real_bucket_hole(&iter, &hole);
1432 next = bpos_nosnap_successor(k.k->p);
1434 ret = bch2_check_alloc_key(trans,
1444 ret = bch2_check_alloc_hole_freespace(trans,
1445 bkey_start_pos(k.k),
1448 bch2_check_alloc_hole_bucket_gens(trans,
1449 bkey_start_pos(k.k),
1456 ret = bch2_trans_commit(trans, NULL, NULL,
1457 BCH_TRANS_COMMIT_no_enospc);
1461 bch2_btree_iter_set_pos(&iter, next);
1463 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1468 bch2_trans_iter_exit(trans, &bucket_gens_iter);
1469 bch2_trans_iter_exit(trans, &freespace_iter);
1470 bch2_trans_iter_exit(trans, &discard_iter);
1471 bch2_trans_iter_exit(trans, &iter);
1476 ret = for_each_btree_key(trans, iter,
1477 BTREE_ID_need_discard, POS_MIN,
1478 BTREE_ITER_PREFETCH, k,
1479 bch2_check_discard_freespace_key(trans, &iter));
1483 bch2_trans_iter_init(trans, &iter, BTREE_ID_freespace, POS_MIN,
1484 BTREE_ITER_PREFETCH);
1486 bch2_trans_begin(trans);
1487 k = bch2_btree_iter_peek(&iter);
1491 ret = bkey_err(k) ?:
1492 bch2_check_discard_freespace_key(trans, &iter);
1493 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1498 struct printbuf buf = PRINTBUF;
1499 bch2_bkey_val_to_text(&buf, c, k);
1501 bch_err(c, "while checking %s", buf.buf);
1502 printbuf_exit(&buf);
1506 bch2_btree_iter_set_pos(&iter, bpos_nosnap_successor(iter.pos));
1508 bch2_trans_iter_exit(trans, &iter);
1512 ret = for_each_btree_key_commit(trans, iter,
1513 BTREE_ID_bucket_gens, POS_MIN,
1514 BTREE_ITER_PREFETCH, k,
1515 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1516 bch2_check_bucket_gens_key(trans, &iter, k));
1518 bch2_trans_put(trans);
1523 static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
1524 struct btree_iter *alloc_iter)
1526 struct bch_fs *c = trans->c;
1527 struct btree_iter lru_iter;
1528 struct bch_alloc_v4 a_convert;
1529 const struct bch_alloc_v4 *a;
1530 struct bkey_s_c alloc_k, lru_k;
1531 struct printbuf buf = PRINTBUF;
1534 alloc_k = bch2_btree_iter_peek(alloc_iter);
1538 ret = bkey_err(alloc_k);
1542 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1544 if (a->data_type != BCH_DATA_cached)
1547 if (fsck_err_on(!a->io_time[READ], c,
1548 alloc_key_cached_but_read_time_zero,
1549 "cached bucket with read_time 0\n"
1551 (printbuf_reset(&buf),
1552 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1553 struct bkey_i_alloc_v4 *a_mut =
1554 bch2_alloc_to_v4_mut(trans, alloc_k);
1555 ret = PTR_ERR_OR_ZERO(a_mut);
1559 a_mut->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now);
1560 ret = bch2_trans_update(trans, alloc_iter,
1561 &a_mut->k_i, BTREE_TRIGGER_NORUN);
1568 lru_k = bch2_bkey_get_iter(trans, &lru_iter, BTREE_ID_lru,
1569 lru_pos(alloc_k.k->p.inode,
1570 bucket_to_u64(alloc_k.k->p),
1571 a->io_time[READ]), 0);
1572 ret = bkey_err(lru_k);
1576 if (fsck_err_on(lru_k.k->type != KEY_TYPE_set, c,
1577 alloc_key_to_missing_lru_entry,
1578 "missing lru entry\n"
1580 (printbuf_reset(&buf),
1581 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1582 ret = bch2_lru_set(trans,
1584 bucket_to_u64(alloc_k.k->p),
1591 bch2_trans_iter_exit(trans, &lru_iter);
1592 printbuf_exit(&buf);
1596 int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
1598 int ret = bch2_trans_run(c,
1599 for_each_btree_key_commit(trans, iter, BTREE_ID_alloc,
1600 POS_MIN, BTREE_ITER_PREFETCH, k,
1601 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1602 bch2_check_alloc_to_lru_ref(trans, &iter)));
1607 struct discard_buckets_state {
1610 u64 need_journal_commit;
1613 u64 need_journal_commit_this_dev;
1616 static void discard_buckets_next_dev(struct bch_fs *c, struct discard_buckets_state *s, struct bch_dev *ca)
1621 if (s->ca && s->need_journal_commit_this_dev >
1622 bch2_dev_usage_read(s->ca).d[BCH_DATA_free].buckets)
1623 bch2_journal_flush_async(&c->journal, NULL);
1626 percpu_ref_put(&s->ca->ref);
1628 percpu_ref_get(&ca->ref);
1630 s->need_journal_commit_this_dev = 0;
1633 static int bch2_discard_one_bucket(struct btree_trans *trans,
1634 struct btree_iter *need_discard_iter,
1635 struct bpos *discard_pos_done,
1636 struct discard_buckets_state *s)
1638 struct bch_fs *c = trans->c;
1639 struct bpos pos = need_discard_iter->pos;
1640 struct btree_iter iter = { NULL };
1643 struct bkey_i_alloc_v4 *a;
1644 struct printbuf buf = PRINTBUF;
1647 ca = bch_dev_bkey_exists(c, pos.inode);
1649 if (!percpu_ref_tryget(&ca->io_ref)) {
1650 bch2_btree_iter_set_pos(need_discard_iter, POS(pos.inode + 1, 0));
1654 discard_buckets_next_dev(c, s, ca);
1656 if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) {
1661 if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
1662 c->journal.flushed_seq_ondisk,
1663 pos.inode, pos.offset)) {
1664 s->need_journal_commit++;
1665 s->need_journal_commit_this_dev++;
1669 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc,
1670 need_discard_iter->pos,
1676 a = bch2_alloc_to_v4_mut(trans, k);
1677 ret = PTR_ERR_OR_ZERO(a);
1681 if (BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
1683 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1687 if (a->v.journal_seq > c->journal.flushed_seq_ondisk) {
1688 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1689 bch2_trans_inconsistent(trans,
1690 "clearing need_discard but journal_seq %llu > flushed_seq %llu\n"
1693 c->journal.flushed_seq_ondisk,
1694 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1700 if (a->v.data_type != BCH_DATA_need_discard) {
1701 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1702 bch2_trans_inconsistent(trans,
1703 "bucket incorrectly set in need_discard btree\n"
1705 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1712 if (!bkey_eq(*discard_pos_done, iter.pos) &&
1713 ca->mi.discard && !c->opts.nochanges) {
1715 * This works without any other locks because this is the only
1716 * thread that removes items from the need_discard tree
1718 bch2_trans_unlock_long(trans);
1719 blkdev_issue_discard(ca->disk_sb.bdev,
1720 k.k->p.offset * ca->mi.bucket_size,
1723 *discard_pos_done = iter.pos;
1725 ret = bch2_trans_relock_notrace(trans);
1730 SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
1731 a->v.data_type = alloc_data_type(a->v, a->v.data_type);
1733 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
1734 bch2_trans_commit(trans, NULL, NULL,
1735 BCH_WATERMARK_btree|
1736 BCH_TRANS_COMMIT_no_enospc);
1740 count_event(c, bucket_discard);
1744 bch2_trans_iter_exit(trans, &iter);
1745 percpu_ref_put(&ca->io_ref);
1746 printbuf_exit(&buf);
1750 static void bch2_do_discards_work(struct work_struct *work)
1752 struct bch_fs *c = container_of(work, struct bch_fs, discard_work);
1753 struct discard_buckets_state s = {};
1754 struct bpos discard_pos_done = POS_MAX;
1758 * We're doing the commit in bch2_discard_one_bucket instead of using
1759 * for_each_btree_key_commit() so that we can increment counters after
1760 * successful commit:
1762 ret = bch2_trans_run(c,
1763 for_each_btree_key(trans, iter,
1764 BTREE_ID_need_discard, POS_MIN, 0, k,
1765 bch2_discard_one_bucket(trans, &iter, &discard_pos_done, &s)));
1767 discard_buckets_next_dev(c, &s, NULL);
1769 trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded,
1772 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1775 void bch2_do_discards(struct bch_fs *c)
1777 if (bch2_write_ref_tryget(c, BCH_WRITE_REF_discard) &&
1778 !queue_work(c->write_ref_wq, &c->discard_work))
1779 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1782 static int invalidate_one_bucket(struct btree_trans *trans,
1783 struct btree_iter *lru_iter,
1784 struct bkey_s_c lru_k,
1785 s64 *nr_to_invalidate)
1787 struct bch_fs *c = trans->c;
1788 struct btree_iter alloc_iter = { NULL };
1789 struct bkey_i_alloc_v4 *a = NULL;
1790 struct printbuf buf = PRINTBUF;
1791 struct bpos bucket = u64_to_bucket(lru_k.k->p.offset);
1792 unsigned cached_sectors;
1795 if (*nr_to_invalidate <= 0)
1798 if (!bch2_dev_bucket_exists(c, bucket)) {
1799 prt_str(&buf, "lru entry points to invalid bucket");
1803 if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset))
1806 a = bch2_trans_start_alloc_update(trans, &alloc_iter, bucket);
1807 ret = PTR_ERR_OR_ZERO(a);
1811 /* We expect harmless races here due to the btree write buffer: */
1812 if (lru_pos_time(lru_iter->pos) != alloc_lru_idx_read(a->v))
1815 BUG_ON(a->v.data_type != BCH_DATA_cached);
1817 if (!a->v.cached_sectors)
1818 bch_err(c, "invalidating empty bucket, confused");
1820 cached_sectors = a->v.cached_sectors;
1822 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1825 a->v.dirty_sectors = 0;
1826 a->v.cached_sectors = 0;
1827 a->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now);
1828 a->v.io_time[WRITE] = atomic64_read(&c->io_clock[WRITE].now);
1830 ret = bch2_trans_update(trans, &alloc_iter, &a->k_i,
1831 BTREE_TRIGGER_BUCKET_INVALIDATE) ?:
1832 bch2_trans_commit(trans, NULL, NULL,
1833 BCH_WATERMARK_btree|
1834 BCH_TRANS_COMMIT_no_enospc);
1838 trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors);
1839 --*nr_to_invalidate;
1841 bch2_trans_iter_exit(trans, &alloc_iter);
1842 printbuf_exit(&buf);
1845 prt_str(&buf, "\n lru key: ");
1846 bch2_bkey_val_to_text(&buf, c, lru_k);
1848 prt_str(&buf, "\n lru entry: ");
1849 bch2_lru_pos_to_text(&buf, lru_iter->pos);
1851 prt_str(&buf, "\n alloc key: ");
1853 bch2_bpos_to_text(&buf, bucket);
1855 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i));
1857 bch_err(c, "%s", buf.buf);
1858 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_lrus) {
1859 bch2_inconsistent_error(c);
1866 static void bch2_do_invalidates_work(struct work_struct *work)
1868 struct bch_fs *c = container_of(work, struct bch_fs, invalidate_work);
1869 struct btree_trans *trans = bch2_trans_get(c);
1872 ret = bch2_btree_write_buffer_tryflush(trans);
1876 for_each_member_device(c, ca) {
1877 s64 nr_to_invalidate =
1878 should_invalidate_buckets(ca, bch2_dev_usage_read(ca));
1880 ret = for_each_btree_key_upto(trans, iter, BTREE_ID_lru,
1881 lru_pos(ca->dev_idx, 0, 0),
1882 lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX),
1883 BTREE_ITER_INTENT, k,
1884 invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate));
1887 percpu_ref_put(&ca->ref);
1892 bch2_trans_put(trans);
1893 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1896 void bch2_do_invalidates(struct bch_fs *c)
1898 if (bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate) &&
1899 !queue_work(c->write_ref_wq, &c->invalidate_work))
1900 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1903 int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca,
1904 u64 bucket_start, u64 bucket_end)
1906 struct btree_trans *trans = bch2_trans_get(c);
1907 struct btree_iter iter;
1910 struct bpos end = POS(ca->dev_idx, bucket_end);
1911 struct bch_member *m;
1912 unsigned long last_updated = jiffies;
1915 BUG_ON(bucket_start > bucket_end);
1916 BUG_ON(bucket_end > ca->mi.nbuckets);
1918 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
1919 POS(ca->dev_idx, max_t(u64, ca->mi.first_bucket, bucket_start)),
1920 BTREE_ITER_PREFETCH);
1922 * Scan the alloc btree for every bucket on @ca, and add buckets to the
1923 * freespace/need_discard/need_gc_gens btrees as needed:
1926 if (last_updated + HZ * 10 < jiffies) {
1927 bch_info(ca, "%s: currently at %llu/%llu",
1928 __func__, iter.pos.offset, ca->mi.nbuckets);
1929 last_updated = jiffies;
1932 bch2_trans_begin(trans);
1934 if (bkey_ge(iter.pos, end)) {
1939 k = bch2_get_key_or_hole(&iter, end, &hole);
1946 * We process live keys in the alloc btree one at a
1949 struct bch_alloc_v4 a_convert;
1950 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert);
1952 ret = bch2_bucket_do_index(trans, k, a, true) ?:
1953 bch2_trans_commit(trans, NULL, NULL,
1954 BCH_TRANS_COMMIT_no_enospc);
1958 bch2_btree_iter_advance(&iter);
1960 struct bkey_i *freespace;
1962 freespace = bch2_trans_kmalloc(trans, sizeof(*freespace));
1963 ret = PTR_ERR_OR_ZERO(freespace);
1967 bkey_init(&freespace->k);
1968 freespace->k.type = KEY_TYPE_set;
1969 freespace->k.p = k.k->p;
1970 freespace->k.size = k.k->size;
1972 ret = bch2_btree_insert_trans(trans, BTREE_ID_freespace, freespace, 0) ?:
1973 bch2_trans_commit(trans, NULL, NULL,
1974 BCH_TRANS_COMMIT_no_enospc);
1978 bch2_btree_iter_set_pos(&iter, k.k->p);
1981 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1987 bch2_trans_iter_exit(trans, &iter);
1988 bch2_trans_put(trans);
1991 bch_err_msg(ca, ret, "initializing free space");
1995 mutex_lock(&c->sb_lock);
1996 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1997 SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true);
1998 mutex_unlock(&c->sb_lock);
2003 int bch2_fs_freespace_init(struct bch_fs *c)
2006 bool doing_init = false;
2009 * We can crash during the device add path, so we need to check this on
2013 for_each_member_device(c, ca) {
2014 if (ca->mi.freespace_initialized)
2018 bch_info(c, "initializing freespace");
2022 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
2024 percpu_ref_put(&ca->ref);
2031 mutex_lock(&c->sb_lock);
2032 bch2_write_super(c);
2033 mutex_unlock(&c->sb_lock);
2034 bch_verbose(c, "done initializing freespace");
2040 /* Bucket IO clocks: */
2042 int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
2043 size_t bucket_nr, int rw)
2045 struct bch_fs *c = trans->c;
2046 struct btree_iter iter;
2047 struct bkey_i_alloc_v4 *a;
2051 a = bch2_trans_start_alloc_update(trans, &iter, POS(dev, bucket_nr));
2052 ret = PTR_ERR_OR_ZERO(a);
2056 now = atomic64_read(&c->io_clock[rw].now);
2057 if (a->v.io_time[rw] == now)
2060 a->v.io_time[rw] = now;
2062 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
2063 bch2_trans_commit(trans, NULL, NULL, 0);
2065 bch2_trans_iter_exit(trans, &iter);
2069 /* Startup/shutdown (ro/rw): */
2071 void bch2_recalc_capacity(struct bch_fs *c)
2073 u64 capacity = 0, reserved_sectors = 0, gc_reserve;
2074 unsigned bucket_size_max = 0;
2075 unsigned long ra_pages = 0;
2077 lockdep_assert_held(&c->state_lock);
2079 for_each_online_member(c, ca) {
2080 struct backing_dev_info *bdi = ca->disk_sb.bdev->bd_disk->bdi;
2082 ra_pages += bdi->ra_pages;
2085 bch2_set_ra_pages(c, ra_pages);
2087 for_each_rw_member(c, ca) {
2088 u64 dev_reserve = 0;
2091 * We need to reserve buckets (from the number
2092 * of currently available buckets) against
2093 * foreground writes so that mainly copygc can
2094 * make forward progress.
2096 * We need enough to refill the various reserves
2097 * from scratch - copygc will use its entire
2098 * reserve all at once, then run against when
2099 * its reserve is refilled (from the formerly
2100 * available buckets).
2102 * This reserve is just used when considering if
2103 * allocations for foreground writes must wait -
2104 * not -ENOSPC calculations.
2107 dev_reserve += ca->nr_btree_reserve * 2;
2108 dev_reserve += ca->mi.nbuckets >> 6; /* copygc reserve */
2110 dev_reserve += 1; /* btree write point */
2111 dev_reserve += 1; /* copygc write point */
2112 dev_reserve += 1; /* rebalance write point */
2114 dev_reserve *= ca->mi.bucket_size;
2116 capacity += bucket_to_sector(ca, ca->mi.nbuckets -
2117 ca->mi.first_bucket);
2119 reserved_sectors += dev_reserve * 2;
2121 bucket_size_max = max_t(unsigned, bucket_size_max,
2122 ca->mi.bucket_size);
2125 gc_reserve = c->opts.gc_reserve_bytes
2126 ? c->opts.gc_reserve_bytes >> 9
2127 : div64_u64(capacity * c->opts.gc_reserve_percent, 100);
2129 reserved_sectors = max(gc_reserve, reserved_sectors);
2131 reserved_sectors = min(reserved_sectors, capacity);
2133 c->capacity = capacity - reserved_sectors;
2135 c->bucket_size_max = bucket_size_max;
2137 /* Wake up case someone was waiting for buckets */
2138 closure_wake_up(&c->freelist_wait);
2141 u64 bch2_min_rw_member_capacity(struct bch_fs *c)
2145 for_each_rw_member(c, ca)
2146 ret = min(ret, ca->mi.nbuckets * ca->mi.bucket_size);
2150 static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca)
2152 struct open_bucket *ob;
2155 for (ob = c->open_buckets;
2156 ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
2158 spin_lock(&ob->lock);
2159 if (ob->valid && !ob->on_partial_list &&
2160 ob->dev == ca->dev_idx)
2162 spin_unlock(&ob->lock);
2168 /* device goes ro: */
2169 void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca)
2173 /* First, remove device from allocation groups: */
2175 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2176 clear_bit(ca->dev_idx, c->rw_devs[i].d);
2179 * Capacity is calculated based off of devices in allocation groups:
2181 bch2_recalc_capacity(c);
2183 bch2_open_buckets_stop(c, ca, false);
2186 * Wake up threads that were blocked on allocation, so they can notice
2187 * the device can no longer be removed and the capacity has changed:
2189 closure_wake_up(&c->freelist_wait);
2192 * journal_res_get() can block waiting for free space in the journal -
2193 * it needs to notice there may not be devices to allocate from anymore:
2195 wake_up(&c->journal.wait);
2197 /* Now wait for any in flight writes: */
2199 closure_wait_event(&c->open_buckets_wait,
2200 !bch2_dev_has_open_write_point(c, ca));
2203 /* device goes rw: */
2204 void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca)
2208 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2209 if (ca->mi.data_allowed & (1 << i))
2210 set_bit(ca->dev_idx, c->rw_devs[i].d);
2213 void bch2_fs_allocator_background_init(struct bch_fs *c)
2215 spin_lock_init(&c->freelist_lock);
2216 INIT_WORK(&c->discard_work, bch2_do_discards_work);
2217 INIT_WORK(&c->invalidate_work, bch2_do_invalidates_work);