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(a.v->dirty_sectors ||
265 a.v->cached_sectors ||
267 alloc_key_empty_but_have_data,
268 "empty data type free but have data");
271 case BCH_DATA_journal:
274 case BCH_DATA_parity:
275 bkey_fsck_err_on(!a.v->dirty_sectors, c, err,
276 alloc_key_dirty_sectors_0,
277 "data_type %s but dirty_sectors==0",
278 bch2_data_types[a.v->data_type]);
280 case BCH_DATA_cached:
281 bkey_fsck_err_on(!a.v->cached_sectors ||
282 a.v->dirty_sectors ||
284 alloc_key_cached_inconsistency,
285 "data type inconsistency");
287 bkey_fsck_err_on(!a.v->io_time[READ] &&
288 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_to_lru_refs,
290 alloc_key_cached_but_read_time_zero,
291 "cached bucket with read_time == 0");
293 case BCH_DATA_stripe:
300 static inline u64 swab40(u64 x)
302 return (((x & 0x00000000ffULL) << 32)|
303 ((x & 0x000000ff00ULL) << 16)|
304 ((x & 0x0000ff0000ULL) >> 0)|
305 ((x & 0x00ff000000ULL) >> 16)|
306 ((x & 0xff00000000ULL) >> 32));
309 void bch2_alloc_v4_swab(struct bkey_s k)
311 struct bch_alloc_v4 *a = bkey_s_to_alloc_v4(k).v;
312 struct bch_backpointer *bp, *bps;
314 a->journal_seq = swab64(a->journal_seq);
315 a->flags = swab32(a->flags);
316 a->dirty_sectors = swab32(a->dirty_sectors);
317 a->cached_sectors = swab32(a->cached_sectors);
318 a->io_time[0] = swab64(a->io_time[0]);
319 a->io_time[1] = swab64(a->io_time[1]);
320 a->stripe = swab32(a->stripe);
321 a->nr_external_backpointers = swab32(a->nr_external_backpointers);
323 bps = alloc_v4_backpointers(a);
324 for (bp = bps; bp < bps + BCH_ALLOC_V4_NR_BACKPOINTERS(a); bp++) {
325 bp->bucket_offset = swab40(bp->bucket_offset);
326 bp->bucket_len = swab32(bp->bucket_len);
327 bch2_bpos_swab(&bp->pos);
331 void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
333 struct bch_alloc_v4 _a;
334 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a);
338 printbuf_indent_add(out, 2);
340 prt_printf(out, "gen %u oldest_gen %u data_type %s",
341 a->gen, a->oldest_gen,
342 a->data_type < BCH_DATA_NR
343 ? bch2_data_types[a->data_type]
344 : "(invalid data type)");
346 prt_printf(out, "journal_seq %llu", a->journal_seq);
348 prt_printf(out, "need_discard %llu", BCH_ALLOC_V4_NEED_DISCARD(a));
350 prt_printf(out, "need_inc_gen %llu", BCH_ALLOC_V4_NEED_INC_GEN(a));
352 prt_printf(out, "dirty_sectors %u", a->dirty_sectors);
354 prt_printf(out, "cached_sectors %u", a->cached_sectors);
356 prt_printf(out, "stripe %u", a->stripe);
358 prt_printf(out, "stripe_redundancy %u", a->stripe_redundancy);
360 prt_printf(out, "io_time[READ] %llu", a->io_time[READ]);
362 prt_printf(out, "io_time[WRITE] %llu", a->io_time[WRITE]);
364 prt_printf(out, "fragmentation %llu", a->fragmentation_lru);
366 prt_printf(out, "bp_start %llu", BCH_ALLOC_V4_BACKPOINTERS_START(a));
369 if (BCH_ALLOC_V4_NR_BACKPOINTERS(a)) {
370 struct bkey_s_c_alloc_v4 a_raw = bkey_s_c_to_alloc_v4(k);
371 const struct bch_backpointer *bps = alloc_v4_backpointers_c(a_raw.v);
373 prt_printf(out, "backpointers: %llu", BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v));
374 printbuf_indent_add(out, 2);
376 for (i = 0; i < BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v); i++) {
378 bch2_backpointer_to_text(out, &bps[i]);
381 printbuf_indent_sub(out, 2);
384 printbuf_indent_sub(out, 2);
387 void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out)
389 if (k.k->type == KEY_TYPE_alloc_v4) {
392 *out = *bkey_s_c_to_alloc_v4(k).v;
394 src = alloc_v4_backpointers(out);
395 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
396 dst = alloc_v4_backpointers(out);
399 memset(src, 0, dst - src);
401 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(out, 0);
403 struct bkey_alloc_unpacked u = bch2_alloc_unpack(k);
405 *out = (struct bch_alloc_v4) {
406 .journal_seq = u.journal_seq,
407 .flags = u.need_discard,
409 .oldest_gen = u.oldest_gen,
410 .data_type = u.data_type,
411 .stripe_redundancy = u.stripe_redundancy,
412 .dirty_sectors = u.dirty_sectors,
413 .cached_sectors = u.cached_sectors,
414 .io_time[READ] = u.read_time,
415 .io_time[WRITE] = u.write_time,
419 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
423 static noinline struct bkey_i_alloc_v4 *
424 __bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
426 struct bkey_i_alloc_v4 *ret;
428 ret = bch2_trans_kmalloc(trans, max(bkey_bytes(k.k), sizeof(struct bkey_i_alloc_v4)));
432 if (k.k->type == KEY_TYPE_alloc_v4) {
435 bkey_reassemble(&ret->k_i, k);
437 src = alloc_v4_backpointers(&ret->v);
438 SET_BCH_ALLOC_V4_BACKPOINTERS_START(&ret->v, BCH_ALLOC_V4_U64s);
439 dst = alloc_v4_backpointers(&ret->v);
442 memset(src, 0, dst - src);
444 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(&ret->v, 0);
445 set_alloc_v4_u64s(ret);
447 bkey_alloc_v4_init(&ret->k_i);
449 bch2_alloc_to_v4(k, &ret->v);
454 static inline struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut_inlined(struct btree_trans *trans, struct bkey_s_c k)
456 struct bkey_s_c_alloc_v4 a;
458 if (likely(k.k->type == KEY_TYPE_alloc_v4) &&
459 ((a = bkey_s_c_to_alloc_v4(k), true) &&
460 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v) == 0))
461 return bch2_bkey_make_mut_noupdate_typed(trans, k, alloc_v4);
463 return __bch2_alloc_to_v4_mut(trans, k);
466 struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
468 return bch2_alloc_to_v4_mut_inlined(trans, k);
471 struct bkey_i_alloc_v4 *
472 bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter,
476 struct bkey_i_alloc_v4 *a;
479 k = bch2_bkey_get_iter(trans, iter, BTREE_ID_alloc, pos,
480 BTREE_ITER_WITH_UPDATES|
487 a = bch2_alloc_to_v4_mut_inlined(trans, k);
488 ret = PTR_ERR_OR_ZERO(a);
493 bch2_trans_iter_exit(trans, iter);
497 static struct bpos alloc_gens_pos(struct bpos pos, unsigned *offset)
499 *offset = pos.offset & KEY_TYPE_BUCKET_GENS_MASK;
501 pos.offset >>= KEY_TYPE_BUCKET_GENS_BITS;
505 static struct bpos bucket_gens_pos_to_alloc(struct bpos pos, unsigned offset)
507 pos.offset <<= KEY_TYPE_BUCKET_GENS_BITS;
508 pos.offset += offset;
512 static unsigned alloc_gen(struct bkey_s_c k, unsigned offset)
514 return k.k->type == KEY_TYPE_bucket_gens
515 ? bkey_s_c_to_bucket_gens(k).v->gens[offset]
519 int bch2_bucket_gens_invalid(struct bch_fs *c, struct bkey_s_c k,
520 enum bkey_invalid_flags flags,
521 struct printbuf *err)
525 bkey_fsck_err_on(bkey_val_bytes(k.k) != sizeof(struct bch_bucket_gens), c, err,
526 bucket_gens_val_size_bad,
527 "bad val size (%zu != %zu)",
528 bkey_val_bytes(k.k), sizeof(struct bch_bucket_gens));
533 void bch2_bucket_gens_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
535 struct bkey_s_c_bucket_gens g = bkey_s_c_to_bucket_gens(k);
538 for (i = 0; i < ARRAY_SIZE(g.v->gens); i++) {
541 prt_printf(out, "%u", g.v->gens[i]);
545 int bch2_bucket_gens_init(struct bch_fs *c)
547 struct btree_trans *trans = bch2_trans_get(c);
548 struct btree_iter iter;
550 struct bch_alloc_v4 a;
551 struct bkey_i_bucket_gens g;
552 bool have_bucket_gens_key = false;
558 for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
559 BTREE_ITER_PREFETCH, k, ret) {
561 * Not a fsck error because this is checked/repaired by
562 * bch2_check_alloc_key() which runs later:
564 if (!bch2_dev_bucket_exists(c, k.k->p))
567 gen = bch2_alloc_to_v4(k, &a)->gen;
568 pos = alloc_gens_pos(iter.pos, &offset);
570 if (have_bucket_gens_key && bkey_cmp(iter.pos, pos)) {
571 ret = commit_do(trans, NULL, NULL,
573 BTREE_INSERT_LAZY_RW,
574 bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
577 have_bucket_gens_key = false;
580 if (!have_bucket_gens_key) {
581 bkey_bucket_gens_init(&g.k_i);
583 have_bucket_gens_key = true;
586 g.v.gens[offset] = gen;
588 bch2_trans_iter_exit(trans, &iter);
590 if (have_bucket_gens_key && !ret)
591 ret = commit_do(trans, NULL, NULL,
593 BTREE_INSERT_LAZY_RW,
594 bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
596 bch2_trans_put(trans);
603 int bch2_alloc_read(struct bch_fs *c)
605 struct btree_trans *trans = bch2_trans_get(c);
606 struct btree_iter iter;
611 down_read(&c->gc_lock);
613 if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_bucket_gens) {
614 const struct bch_bucket_gens *g;
617 for_each_btree_key(trans, iter, BTREE_ID_bucket_gens, POS_MIN,
618 BTREE_ITER_PREFETCH, k, ret) {
619 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
620 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
622 if (k.k->type != KEY_TYPE_bucket_gens)
625 g = bkey_s_c_to_bucket_gens(k).v;
628 * Not a fsck error because this is checked/repaired by
629 * bch2_check_alloc_key() which runs later:
631 if (!bch2_dev_exists2(c, k.k->p.inode))
634 ca = bch_dev_bkey_exists(c, k.k->p.inode);
636 for (b = max_t(u64, ca->mi.first_bucket, start);
637 b < min_t(u64, ca->mi.nbuckets, end);
639 *bucket_gen(ca, b) = g->gens[b & KEY_TYPE_BUCKET_GENS_MASK];
641 bch2_trans_iter_exit(trans, &iter);
643 struct bch_alloc_v4 a;
645 for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
646 BTREE_ITER_PREFETCH, k, ret) {
648 * Not a fsck error because this is checked/repaired by
649 * bch2_check_alloc_key() which runs later:
651 if (!bch2_dev_bucket_exists(c, k.k->p))
654 ca = bch_dev_bkey_exists(c, k.k->p.inode);
656 *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen;
658 bch2_trans_iter_exit(trans, &iter);
661 bch2_trans_put(trans);
662 up_read(&c->gc_lock);
670 /* Free space/discard btree: */
672 static int bch2_bucket_do_index(struct btree_trans *trans,
673 struct bkey_s_c alloc_k,
674 const struct bch_alloc_v4 *a,
677 struct bch_fs *c = trans->c;
678 struct bch_dev *ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
679 struct btree_iter iter;
683 enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted;
684 enum bch_bkey_type new_type = set ? KEY_TYPE_set : KEY_TYPE_deleted;
685 struct printbuf buf = PRINTBUF;
688 if (a->data_type != BCH_DATA_free &&
689 a->data_type != BCH_DATA_need_discard)
692 k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k));
697 k->k.type = new_type;
699 switch (a->data_type) {
701 btree = BTREE_ID_freespace;
702 k->k.p = alloc_freespace_pos(alloc_k.k->p, *a);
703 bch2_key_resize(&k->k, 1);
705 case BCH_DATA_need_discard:
706 btree = BTREE_ID_need_discard;
707 k->k.p = alloc_k.k->p;
713 old = bch2_bkey_get_iter(trans, &iter, btree,
714 bkey_start_pos(&k->k),
720 if (ca->mi.freespace_initialized &&
721 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info &&
722 bch2_trans_inconsistent_on(old.k->type != old_type, trans,
723 "incorrect key when %s %s:%llu:%llu:0 (got %s should be %s)\n"
725 set ? "setting" : "clearing",
726 bch2_btree_id_str(btree),
729 bch2_bkey_types[old.k->type],
730 bch2_bkey_types[old_type],
731 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
736 ret = bch2_trans_update(trans, &iter, k, 0);
738 bch2_trans_iter_exit(trans, &iter);
743 static noinline int bch2_bucket_gen_update(struct btree_trans *trans,
744 struct bpos bucket, u8 gen)
746 struct btree_iter iter;
748 struct bpos pos = alloc_gens_pos(bucket, &offset);
749 struct bkey_i_bucket_gens *g;
753 g = bch2_trans_kmalloc(trans, sizeof(*g));
754 ret = PTR_ERR_OR_ZERO(g);
758 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_bucket_gens, pos,
760 BTREE_ITER_WITH_UPDATES);
765 if (k.k->type != KEY_TYPE_bucket_gens) {
766 bkey_bucket_gens_init(&g->k_i);
769 bkey_reassemble(&g->k_i, k);
772 g->v.gens[offset] = gen;
774 ret = bch2_trans_update(trans, &iter, &g->k_i, 0);
775 bch2_trans_iter_exit(trans, &iter);
779 int bch2_trans_mark_alloc(struct btree_trans *trans,
780 enum btree_id btree_id, unsigned level,
781 struct bkey_s_c old, struct bkey_i *new,
784 struct bch_fs *c = trans->c;
785 struct bch_alloc_v4 old_a_convert, *new_a;
786 const struct bch_alloc_v4 *old_a;
787 u64 old_lru, new_lru;
791 * Deletion only happens in the device removal path, with
792 * BTREE_TRIGGER_NORUN:
794 BUG_ON(new->k.type != KEY_TYPE_alloc_v4);
796 old_a = bch2_alloc_to_v4(old, &old_a_convert);
797 new_a = &bkey_i_to_alloc_v4(new)->v;
799 new_a->data_type = alloc_data_type(*new_a, new_a->data_type);
801 if (new_a->dirty_sectors > old_a->dirty_sectors ||
802 new_a->cached_sectors > old_a->cached_sectors) {
803 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
804 new_a->io_time[WRITE]= max_t(u64, 1, atomic64_read(&c->io_clock[WRITE].now));
805 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true);
806 SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true);
809 if (data_type_is_empty(new_a->data_type) &&
810 BCH_ALLOC_V4_NEED_INC_GEN(new_a) &&
811 !bch2_bucket_is_open_safe(c, new->k.p.inode, new->k.p.offset)) {
813 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false);
816 if (old_a->data_type != new_a->data_type ||
817 (new_a->data_type == BCH_DATA_free &&
818 alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) {
819 ret = bch2_bucket_do_index(trans, old, old_a, false) ?:
820 bch2_bucket_do_index(trans, bkey_i_to_s_c(new), new_a, true);
825 if (new_a->data_type == BCH_DATA_cached &&
826 !new_a->io_time[READ])
827 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
829 old_lru = alloc_lru_idx_read(*old_a);
830 new_lru = alloc_lru_idx_read(*new_a);
832 if (old_lru != new_lru) {
833 ret = bch2_lru_change(trans, new->k.p.inode,
834 bucket_to_u64(new->k.p),
840 new_a->fragmentation_lru = alloc_lru_idx_fragmentation(*new_a,
841 bch_dev_bkey_exists(c, new->k.p.inode));
843 if (old_a->fragmentation_lru != new_a->fragmentation_lru) {
844 ret = bch2_lru_change(trans,
845 BCH_LRU_FRAGMENTATION_START,
846 bucket_to_u64(new->k.p),
847 old_a->fragmentation_lru, new_a->fragmentation_lru);
852 if (old_a->gen != new_a->gen) {
853 ret = bch2_bucket_gen_update(trans, new->k.p, new_a->gen);
862 * This synthesizes deleted extents for holes, similar to BTREE_ITER_SLOTS for
863 * extents style btrees, but works on non-extents btrees:
865 static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos end, struct bkey *hole)
867 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
875 struct btree_iter iter2;
878 bch2_trans_copy_iter(&iter2, iter);
880 if (!bpos_eq(iter->path->l[0].b->key.k.p, SPOS_MAX))
881 end = bkey_min(end, bpos_nosnap_successor(iter->path->l[0].b->key.k.p));
883 end = bkey_min(end, POS(iter->pos.inode, iter->pos.offset + U32_MAX - 1));
886 * btree node min/max is a closed interval, upto takes a half
889 k = bch2_btree_iter_peek_upto(&iter2, end);
891 bch2_trans_iter_exit(iter->trans, &iter2);
893 BUG_ON(next.offset >= iter->pos.offset + U32_MAX);
901 bch2_key_resize(hole, next.offset - iter->pos.offset);
902 return (struct bkey_s_c) { hole, NULL };
906 static bool next_bucket(struct bch_fs *c, struct bpos *bucket)
911 if (bch2_dev_bucket_exists(c, *bucket))
914 if (bch2_dev_exists2(c, bucket->inode)) {
915 ca = bch_dev_bkey_exists(c, bucket->inode);
917 if (bucket->offset < ca->mi.first_bucket) {
918 bucket->offset = ca->mi.first_bucket;
927 iter = bucket->inode;
928 ca = __bch2_next_dev(c, &iter, NULL);
930 *bucket = POS(ca->dev_idx, ca->mi.first_bucket);
936 static struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, struct bkey *hole)
938 struct bch_fs *c = iter->trans->c;
941 k = bch2_get_key_or_hole(iter, POS_MAX, hole);
946 struct bpos bucket = bkey_start_pos(k.k);
948 if (!bch2_dev_bucket_exists(c, bucket)) {
949 if (!next_bucket(c, &bucket))
950 return bkey_s_c_null;
952 bch2_btree_iter_set_pos(iter, bucket);
956 if (!bch2_dev_bucket_exists(c, k.k->p)) {
957 struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode);
959 bch2_key_resize(hole, ca->mi.nbuckets - bucket.offset);
966 static noinline_for_stack
967 int bch2_check_alloc_key(struct btree_trans *trans,
968 struct bkey_s_c alloc_k,
969 struct btree_iter *alloc_iter,
970 struct btree_iter *discard_iter,
971 struct btree_iter *freespace_iter,
972 struct btree_iter *bucket_gens_iter)
974 struct bch_fs *c = trans->c;
976 struct bch_alloc_v4 a_convert;
977 const struct bch_alloc_v4 *a;
978 unsigned discard_key_type, freespace_key_type;
979 unsigned gens_offset;
981 struct printbuf buf = PRINTBUF;
984 if (fsck_err_on(!bch2_dev_bucket_exists(c, alloc_k.k->p), c,
985 alloc_key_to_missing_dev_bucket,
986 "alloc key for invalid device:bucket %llu:%llu",
987 alloc_k.k->p.inode, alloc_k.k->p.offset))
988 return bch2_btree_delete_at(trans, alloc_iter, 0);
990 ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
991 if (!ca->mi.freespace_initialized)
994 a = bch2_alloc_to_v4(alloc_k, &a_convert);
996 discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0;
997 bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p);
998 k = bch2_btree_iter_peek_slot(discard_iter);
1003 if (k.k->type != discard_key_type &&
1004 (c->opts.reconstruct_alloc ||
1005 fsck_err(c, need_discard_key_wrong,
1006 "incorrect key in need_discard btree (got %s should be %s)\n"
1008 bch2_bkey_types[k.k->type],
1009 bch2_bkey_types[discard_key_type],
1010 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1011 struct bkey_i *update =
1012 bch2_trans_kmalloc(trans, sizeof(*update));
1014 ret = PTR_ERR_OR_ZERO(update);
1018 bkey_init(&update->k);
1019 update->k.type = discard_key_type;
1020 update->k.p = discard_iter->pos;
1022 ret = bch2_trans_update(trans, discard_iter, update, 0);
1027 freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0;
1028 bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a));
1029 k = bch2_btree_iter_peek_slot(freespace_iter);
1034 if (k.k->type != freespace_key_type &&
1035 (c->opts.reconstruct_alloc ||
1036 fsck_err(c, freespace_key_wrong,
1037 "incorrect key in freespace btree (got %s should be %s)\n"
1039 bch2_bkey_types[k.k->type],
1040 bch2_bkey_types[freespace_key_type],
1041 (printbuf_reset(&buf),
1042 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1043 struct bkey_i *update =
1044 bch2_trans_kmalloc(trans, sizeof(*update));
1046 ret = PTR_ERR_OR_ZERO(update);
1050 bkey_init(&update->k);
1051 update->k.type = freespace_key_type;
1052 update->k.p = freespace_iter->pos;
1053 bch2_key_resize(&update->k, 1);
1055 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1060 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(alloc_k.k->p, &gens_offset));
1061 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1066 if (a->gen != alloc_gen(k, gens_offset) &&
1067 (c->opts.reconstruct_alloc ||
1068 fsck_err(c, bucket_gens_key_wrong,
1069 "incorrect gen in bucket_gens btree (got %u should be %u)\n"
1071 alloc_gen(k, gens_offset), a->gen,
1072 (printbuf_reset(&buf),
1073 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1074 struct bkey_i_bucket_gens *g =
1075 bch2_trans_kmalloc(trans, sizeof(*g));
1077 ret = PTR_ERR_OR_ZERO(g);
1081 if (k.k->type == KEY_TYPE_bucket_gens) {
1082 bkey_reassemble(&g->k_i, k);
1084 bkey_bucket_gens_init(&g->k_i);
1085 g->k.p = alloc_gens_pos(alloc_k.k->p, &gens_offset);
1088 g->v.gens[gens_offset] = a->gen;
1090 ret = bch2_trans_update(trans, bucket_gens_iter, &g->k_i, 0);
1096 printbuf_exit(&buf);
1100 static noinline_for_stack
1101 int bch2_check_alloc_hole_freespace(struct btree_trans *trans,
1104 struct btree_iter *freespace_iter)
1106 struct bch_fs *c = trans->c;
1109 struct printbuf buf = PRINTBUF;
1112 ca = bch_dev_bkey_exists(c, start.inode);
1113 if (!ca->mi.freespace_initialized)
1116 bch2_btree_iter_set_pos(freespace_iter, start);
1118 k = bch2_btree_iter_peek_slot(freespace_iter);
1123 *end = bkey_min(k.k->p, *end);
1125 if (k.k->type != KEY_TYPE_set &&
1126 (c->opts.reconstruct_alloc ||
1127 fsck_err(c, freespace_hole_missing,
1128 "hole in alloc btree missing in freespace btree\n"
1129 " device %llu buckets %llu-%llu",
1130 freespace_iter->pos.inode,
1131 freespace_iter->pos.offset,
1133 struct bkey_i *update =
1134 bch2_trans_kmalloc(trans, sizeof(*update));
1136 ret = PTR_ERR_OR_ZERO(update);
1140 bkey_init(&update->k);
1141 update->k.type = KEY_TYPE_set;
1142 update->k.p = freespace_iter->pos;
1143 bch2_key_resize(&update->k,
1144 min_t(u64, U32_MAX, end->offset -
1145 freespace_iter->pos.offset));
1147 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1153 printbuf_exit(&buf);
1157 static noinline_for_stack
1158 int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans,
1161 struct btree_iter *bucket_gens_iter)
1163 struct bch_fs *c = trans->c;
1165 struct printbuf buf = PRINTBUF;
1166 unsigned i, gens_offset, gens_end_offset;
1169 if (c->sb.version < bcachefs_metadata_version_bucket_gens)
1172 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(start, &gens_offset));
1174 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1179 if (bkey_cmp(alloc_gens_pos(start, &gens_offset),
1180 alloc_gens_pos(*end, &gens_end_offset)))
1181 gens_end_offset = KEY_TYPE_BUCKET_GENS_NR;
1183 if (k.k->type == KEY_TYPE_bucket_gens) {
1184 struct bkey_i_bucket_gens g;
1185 bool need_update = false;
1187 bkey_reassemble(&g.k_i, k);
1189 for (i = gens_offset; i < gens_end_offset; i++) {
1190 if (fsck_err_on(g.v.gens[i], c,
1191 bucket_gens_hole_wrong,
1192 "hole in alloc btree at %llu:%llu with nonzero gen in bucket_gens btree (%u)",
1193 bucket_gens_pos_to_alloc(k.k->p, i).inode,
1194 bucket_gens_pos_to_alloc(k.k->p, i).offset,
1202 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1204 ret = PTR_ERR_OR_ZERO(u);
1208 memcpy(u, &g, sizeof(g));
1210 ret = bch2_trans_update(trans, bucket_gens_iter, u, 0);
1216 *end = bkey_min(*end, bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0));
1219 printbuf_exit(&buf);
1223 static noinline_for_stack int __bch2_check_discard_freespace_key(struct btree_trans *trans,
1224 struct btree_iter *iter)
1226 struct bch_fs *c = trans->c;
1227 struct btree_iter alloc_iter;
1228 struct bkey_s_c alloc_k;
1229 struct bch_alloc_v4 a_convert;
1230 const struct bch_alloc_v4 *a;
1233 enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard
1234 ? BCH_DATA_need_discard
1236 struct printbuf buf = PRINTBUF;
1240 pos.offset &= ~(~0ULL << 56);
1241 genbits = iter->pos.offset & (~0ULL << 56);
1243 alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc, pos, 0);
1244 ret = bkey_err(alloc_k);
1248 if (fsck_err_on(!bch2_dev_bucket_exists(c, pos), c,
1249 need_discard_freespace_key_to_invalid_dev_bucket,
1250 "entry in %s btree for nonexistant dev:bucket %llu:%llu",
1251 bch2_btree_id_str(iter->btree_id), pos.inode, pos.offset))
1254 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1256 if (fsck_err_on(a->data_type != state ||
1257 (state == BCH_DATA_free &&
1258 genbits != alloc_freespace_genbits(*a)), c,
1259 need_discard_freespace_key_bad,
1260 "%s\n incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)",
1261 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
1262 bch2_btree_id_str(iter->btree_id),
1265 a->data_type == state,
1266 genbits >> 56, alloc_freespace_genbits(*a) >> 56))
1270 set_btree_iter_dontneed(&alloc_iter);
1271 bch2_trans_iter_exit(trans, &alloc_iter);
1272 printbuf_exit(&buf);
1275 ret = bch2_btree_delete_extent_at(trans, iter,
1276 iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0) ?:
1277 bch2_trans_commit(trans, NULL, NULL,
1278 BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW);
1282 static int bch2_check_discard_freespace_key(struct btree_trans *trans,
1283 struct btree_iter *iter,
1286 if (!btree_id_is_extents(iter->btree_id)) {
1287 return __bch2_check_discard_freespace_key(trans, iter);
1291 while (!bkey_eq(iter->pos, end) &&
1292 !(ret = btree_trans_too_many_iters(trans) ?:
1293 __bch2_check_discard_freespace_key(trans, iter)))
1294 bch2_btree_iter_set_pos(iter, bpos_nosnap_successor(iter->pos));
1301 * We've already checked that generation numbers in the bucket_gens btree are
1302 * valid for buckets that exist; this just checks for keys for nonexistent
1305 static noinline_for_stack
1306 int bch2_check_bucket_gens_key(struct btree_trans *trans,
1307 struct btree_iter *iter,
1310 struct bch_fs *c = trans->c;
1311 struct bkey_i_bucket_gens g;
1313 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
1314 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
1316 bool need_update = false, dev_exists;
1317 struct printbuf buf = PRINTBUF;
1320 BUG_ON(k.k->type != KEY_TYPE_bucket_gens);
1321 bkey_reassemble(&g.k_i, k);
1323 /* if no bch_dev, skip out whether we repair or not */
1324 dev_exists = bch2_dev_exists2(c, k.k->p.inode);
1326 if (fsck_err_on(!dev_exists, c,
1327 bucket_gens_to_invalid_dev,
1328 "bucket_gens key for invalid device:\n %s",
1329 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1330 ret = bch2_btree_delete_at(trans, iter, 0);
1335 ca = bch_dev_bkey_exists(c, k.k->p.inode);
1336 if (fsck_err_on(end <= ca->mi.first_bucket ||
1337 start >= ca->mi.nbuckets, c,
1338 bucket_gens_to_invalid_buckets,
1339 "bucket_gens key for invalid buckets:\n %s",
1340 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1341 ret = bch2_btree_delete_at(trans, iter, 0);
1345 for (b = start; b < ca->mi.first_bucket; b++)
1346 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1347 bucket_gens_nonzero_for_invalid_buckets,
1348 "bucket_gens key has nonzero gen for invalid bucket")) {
1349 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1353 for (b = ca->mi.nbuckets; b < end; b++)
1354 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1355 bucket_gens_nonzero_for_invalid_buckets,
1356 "bucket_gens key has nonzero gen for invalid bucket")) {
1357 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1362 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1364 ret = PTR_ERR_OR_ZERO(u);
1368 memcpy(u, &g, sizeof(g));
1369 ret = bch2_trans_update(trans, iter, u, 0);
1373 printbuf_exit(&buf);
1377 int bch2_check_alloc_info(struct bch_fs *c)
1379 struct btree_trans *trans = bch2_trans_get(c);
1380 struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter;
1385 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS_MIN,
1386 BTREE_ITER_PREFETCH);
1387 bch2_trans_iter_init(trans, &discard_iter, BTREE_ID_need_discard, POS_MIN,
1388 BTREE_ITER_PREFETCH);
1389 bch2_trans_iter_init(trans, &freespace_iter, BTREE_ID_freespace, POS_MIN,
1390 BTREE_ITER_PREFETCH);
1391 bch2_trans_iter_init(trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN,
1392 BTREE_ITER_PREFETCH);
1397 bch2_trans_begin(trans);
1399 k = bch2_get_key_or_real_bucket_hole(&iter, &hole);
1408 next = bpos_nosnap_successor(k.k->p);
1410 ret = bch2_check_alloc_key(trans,
1420 ret = bch2_check_alloc_hole_freespace(trans,
1421 bkey_start_pos(k.k),
1424 bch2_check_alloc_hole_bucket_gens(trans,
1425 bkey_start_pos(k.k),
1432 ret = bch2_trans_commit(trans, NULL, NULL,
1433 BTREE_INSERT_NOFAIL|
1434 BTREE_INSERT_LAZY_RW);
1438 bch2_btree_iter_set_pos(&iter, next);
1440 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1445 bch2_trans_iter_exit(trans, &bucket_gens_iter);
1446 bch2_trans_iter_exit(trans, &freespace_iter);
1447 bch2_trans_iter_exit(trans, &discard_iter);
1448 bch2_trans_iter_exit(trans, &iter);
1453 ret = for_each_btree_key2(trans, iter,
1454 BTREE_ID_need_discard, POS_MIN,
1455 BTREE_ITER_PREFETCH, k,
1456 bch2_check_discard_freespace_key(trans, &iter, k.k->p)) ?:
1457 for_each_btree_key2(trans, iter,
1458 BTREE_ID_freespace, POS_MIN,
1459 BTREE_ITER_PREFETCH, k,
1460 bch2_check_discard_freespace_key(trans, &iter, k.k->p)) ?:
1461 for_each_btree_key_commit(trans, iter,
1462 BTREE_ID_bucket_gens, POS_MIN,
1463 BTREE_ITER_PREFETCH, k,
1464 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
1465 bch2_check_bucket_gens_key(trans, &iter, k));
1467 bch2_trans_put(trans);
1473 static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
1474 struct btree_iter *alloc_iter)
1476 struct bch_fs *c = trans->c;
1477 struct btree_iter lru_iter;
1478 struct bch_alloc_v4 a_convert;
1479 const struct bch_alloc_v4 *a;
1480 struct bkey_s_c alloc_k, lru_k;
1481 struct printbuf buf = PRINTBUF;
1484 alloc_k = bch2_btree_iter_peek(alloc_iter);
1488 ret = bkey_err(alloc_k);
1492 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1494 if (a->data_type != BCH_DATA_cached)
1497 lru_k = bch2_bkey_get_iter(trans, &lru_iter, BTREE_ID_lru,
1498 lru_pos(alloc_k.k->p.inode,
1499 bucket_to_u64(alloc_k.k->p),
1500 a->io_time[READ]), 0);
1501 ret = bkey_err(lru_k);
1505 if (fsck_err_on(!a->io_time[READ], c,
1506 alloc_key_cached_but_read_time_zero,
1507 "cached bucket with read_time 0\n"
1509 (printbuf_reset(&buf),
1510 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)) ||
1511 fsck_err_on(lru_k.k->type != KEY_TYPE_set, c,
1512 alloc_key_to_missing_lru_entry,
1513 "missing lru entry\n"
1515 (printbuf_reset(&buf),
1516 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1517 u64 read_time = a->io_time[READ] ?:
1518 atomic64_read(&c->io_clock[READ].now);
1520 ret = bch2_lru_set(trans,
1522 bucket_to_u64(alloc_k.k->p),
1527 if (a->io_time[READ] != read_time) {
1528 struct bkey_i_alloc_v4 *a_mut =
1529 bch2_alloc_to_v4_mut(trans, alloc_k);
1530 ret = PTR_ERR_OR_ZERO(a_mut);
1534 a_mut->v.io_time[READ] = read_time;
1535 ret = bch2_trans_update(trans, alloc_iter,
1536 &a_mut->k_i, BTREE_TRIGGER_NORUN);
1543 bch2_trans_iter_exit(trans, &lru_iter);
1544 printbuf_exit(&buf);
1548 int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
1550 struct btree_iter iter;
1554 ret = bch2_trans_run(c,
1555 for_each_btree_key_commit(trans, iter, BTREE_ID_alloc,
1556 POS_MIN, BTREE_ITER_PREFETCH, k,
1557 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
1558 bch2_check_alloc_to_lru_ref(trans, &iter)));
1564 static int bch2_discard_one_bucket(struct btree_trans *trans,
1565 struct btree_iter *need_discard_iter,
1566 struct bpos *discard_pos_done,
1569 u64 *need_journal_commit,
1572 struct bch_fs *c = trans->c;
1573 struct bpos pos = need_discard_iter->pos;
1574 struct btree_iter iter = { NULL };
1577 struct bkey_i_alloc_v4 *a;
1578 struct printbuf buf = PRINTBUF;
1581 ca = bch_dev_bkey_exists(c, pos.inode);
1582 if (!percpu_ref_tryget(&ca->io_ref)) {
1583 bch2_btree_iter_set_pos(need_discard_iter, POS(pos.inode + 1, 0));
1587 if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) {
1592 if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
1593 c->journal.flushed_seq_ondisk,
1594 pos.inode, pos.offset)) {
1595 (*need_journal_commit)++;
1599 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc,
1600 need_discard_iter->pos,
1606 a = bch2_alloc_to_v4_mut(trans, k);
1607 ret = PTR_ERR_OR_ZERO(a);
1611 if (BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
1613 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1617 if (a->v.journal_seq > c->journal.flushed_seq_ondisk) {
1618 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1619 bch2_trans_inconsistent(trans,
1620 "clearing need_discard but journal_seq %llu > flushed_seq %llu\n"
1623 c->journal.flushed_seq_ondisk,
1624 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1630 if (a->v.data_type != BCH_DATA_need_discard) {
1631 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1632 bch2_trans_inconsistent(trans,
1633 "bucket incorrectly set in need_discard btree\n"
1635 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1642 if (!bkey_eq(*discard_pos_done, iter.pos) &&
1643 ca->mi.discard && !c->opts.nochanges) {
1645 * This works without any other locks because this is the only
1646 * thread that removes items from the need_discard tree
1648 bch2_trans_unlock(trans);
1649 blkdev_issue_discard(ca->disk_sb.bdev,
1650 k.k->p.offset * ca->mi.bucket_size,
1653 *discard_pos_done = iter.pos;
1655 ret = bch2_trans_relock_notrace(trans);
1660 SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
1661 a->v.data_type = alloc_data_type(a->v, a->v.data_type);
1663 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
1664 bch2_trans_commit(trans, NULL, NULL,
1665 BCH_WATERMARK_btree|
1666 BTREE_INSERT_NOFAIL);
1670 this_cpu_inc(c->counters[BCH_COUNTER_bucket_discard]);
1674 bch2_trans_iter_exit(trans, &iter);
1675 percpu_ref_put(&ca->io_ref);
1676 printbuf_exit(&buf);
1680 static void bch2_do_discards_work(struct work_struct *work)
1682 struct bch_fs *c = container_of(work, struct bch_fs, discard_work);
1683 struct btree_iter iter;
1685 u64 seen = 0, open = 0, need_journal_commit = 0, discarded = 0;
1686 struct bpos discard_pos_done = POS_MAX;
1690 * We're doing the commit in bch2_discard_one_bucket instead of using
1691 * for_each_btree_key_commit() so that we can increment counters after
1692 * successful commit:
1694 ret = bch2_trans_run(c,
1695 for_each_btree_key2(trans, iter,
1696 BTREE_ID_need_discard, POS_MIN, 0, k,
1697 bch2_discard_one_bucket(trans, &iter, &discard_pos_done,
1700 &need_journal_commit,
1703 if (need_journal_commit * 2 > seen)
1704 bch2_journal_flush_async(&c->journal, NULL);
1706 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1708 trace_discard_buckets(c, seen, open, need_journal_commit, discarded,
1712 void bch2_do_discards(struct bch_fs *c)
1714 if (bch2_write_ref_tryget(c, BCH_WRITE_REF_discard) &&
1715 !queue_work(c->write_ref_wq, &c->discard_work))
1716 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1719 static int invalidate_one_bucket(struct btree_trans *trans,
1720 struct btree_iter *lru_iter,
1721 struct bkey_s_c lru_k,
1722 s64 *nr_to_invalidate)
1724 struct bch_fs *c = trans->c;
1725 struct btree_iter alloc_iter = { NULL };
1726 struct bkey_i_alloc_v4 *a = NULL;
1727 struct printbuf buf = PRINTBUF;
1728 struct bpos bucket = u64_to_bucket(lru_k.k->p.offset);
1729 unsigned cached_sectors;
1732 if (*nr_to_invalidate <= 0)
1735 if (!bch2_dev_bucket_exists(c, bucket)) {
1736 prt_str(&buf, "lru entry points to invalid bucket");
1740 if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset))
1743 a = bch2_trans_start_alloc_update(trans, &alloc_iter, bucket);
1744 ret = PTR_ERR_OR_ZERO(a);
1748 /* We expect harmless races here due to the btree write buffer: */
1749 if (lru_pos_time(lru_iter->pos) != alloc_lru_idx_read(a->v))
1752 BUG_ON(a->v.data_type != BCH_DATA_cached);
1754 if (!a->v.cached_sectors)
1755 bch_err(c, "invalidating empty bucket, confused");
1757 cached_sectors = a->v.cached_sectors;
1759 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1762 a->v.dirty_sectors = 0;
1763 a->v.cached_sectors = 0;
1764 a->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now);
1765 a->v.io_time[WRITE] = atomic64_read(&c->io_clock[WRITE].now);
1767 ret = bch2_trans_update(trans, &alloc_iter, &a->k_i,
1768 BTREE_TRIGGER_BUCKET_INVALIDATE) ?:
1769 bch2_trans_commit(trans, NULL, NULL,
1770 BCH_WATERMARK_btree|
1771 BTREE_INSERT_NOFAIL);
1775 trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors);
1776 --*nr_to_invalidate;
1778 bch2_trans_iter_exit(trans, &alloc_iter);
1779 printbuf_exit(&buf);
1782 prt_str(&buf, "\n lru key: ");
1783 bch2_bkey_val_to_text(&buf, c, lru_k);
1785 prt_str(&buf, "\n lru entry: ");
1786 bch2_lru_pos_to_text(&buf, lru_iter->pos);
1788 prt_str(&buf, "\n alloc key: ");
1790 bch2_bpos_to_text(&buf, bucket);
1792 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i));
1794 bch_err(c, "%s", buf.buf);
1795 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_lrus) {
1796 bch2_inconsistent_error(c);
1803 static void bch2_do_invalidates_work(struct work_struct *work)
1805 struct bch_fs *c = container_of(work, struct bch_fs, invalidate_work);
1807 struct btree_trans *trans = bch2_trans_get(c);
1808 struct btree_iter iter;
1813 ret = bch2_btree_write_buffer_flush(trans);
1817 for_each_member_device(ca, c, i) {
1818 s64 nr_to_invalidate =
1819 should_invalidate_buckets(ca, bch2_dev_usage_read(ca));
1821 ret = for_each_btree_key2_upto(trans, iter, BTREE_ID_lru,
1822 lru_pos(ca->dev_idx, 0, 0),
1823 lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX),
1824 BTREE_ITER_INTENT, k,
1825 invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate));
1828 percpu_ref_put(&ca->ref);
1833 bch2_trans_put(trans);
1834 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1837 void bch2_do_invalidates(struct bch_fs *c)
1839 if (bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate) &&
1840 !queue_work(c->write_ref_wq, &c->invalidate_work))
1841 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1844 int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca,
1845 u64 bucket_start, u64 bucket_end)
1847 struct btree_trans *trans = bch2_trans_get(c);
1848 struct btree_iter iter;
1851 struct bpos end = POS(ca->dev_idx, bucket_end);
1852 struct bch_member *m;
1853 unsigned long last_updated = jiffies;
1856 BUG_ON(bucket_start > bucket_end);
1857 BUG_ON(bucket_end > ca->mi.nbuckets);
1859 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
1860 POS(ca->dev_idx, max_t(u64, ca->mi.first_bucket, bucket_start)),
1861 BTREE_ITER_PREFETCH);
1863 * Scan the alloc btree for every bucket on @ca, and add buckets to the
1864 * freespace/need_discard/need_gc_gens btrees as needed:
1867 if (last_updated + HZ * 10 < jiffies) {
1868 bch_info(ca, "%s: currently at %llu/%llu",
1869 __func__, iter.pos.offset, ca->mi.nbuckets);
1870 last_updated = jiffies;
1873 bch2_trans_begin(trans);
1875 if (bkey_ge(iter.pos, end)) {
1880 k = bch2_get_key_or_hole(&iter, end, &hole);
1887 * We process live keys in the alloc btree one at a
1890 struct bch_alloc_v4 a_convert;
1891 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert);
1893 ret = bch2_bucket_do_index(trans, k, a, true) ?:
1894 bch2_trans_commit(trans, NULL, NULL,
1895 BTREE_INSERT_LAZY_RW|
1896 BTREE_INSERT_NOFAIL);
1900 bch2_btree_iter_advance(&iter);
1902 struct bkey_i *freespace;
1904 freespace = bch2_trans_kmalloc(trans, sizeof(*freespace));
1905 ret = PTR_ERR_OR_ZERO(freespace);
1909 bkey_init(&freespace->k);
1910 freespace->k.type = KEY_TYPE_set;
1911 freespace->k.p = k.k->p;
1912 freespace->k.size = k.k->size;
1914 ret = bch2_btree_insert_trans(trans, BTREE_ID_freespace, freespace, 0) ?:
1915 bch2_trans_commit(trans, NULL, NULL,
1916 BTREE_INSERT_LAZY_RW|
1917 BTREE_INSERT_NOFAIL);
1921 bch2_btree_iter_set_pos(&iter, k.k->p);
1924 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1930 bch2_trans_iter_exit(trans, &iter);
1931 bch2_trans_put(trans);
1934 bch_err_msg(ca, ret, "initializing free space");
1938 mutex_lock(&c->sb_lock);
1939 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1940 SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true);
1941 mutex_unlock(&c->sb_lock);
1946 int bch2_fs_freespace_init(struct bch_fs *c)
1951 bool doing_init = false;
1954 * We can crash during the device add path, so we need to check this on
1958 for_each_member_device(ca, c, i) {
1959 if (ca->mi.freespace_initialized)
1963 bch_info(c, "initializing freespace");
1967 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1969 percpu_ref_put(&ca->ref);
1976 mutex_lock(&c->sb_lock);
1977 bch2_write_super(c);
1978 mutex_unlock(&c->sb_lock);
1979 bch_verbose(c, "done initializing freespace");
1985 /* Bucket IO clocks: */
1987 int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
1988 size_t bucket_nr, int rw)
1990 struct bch_fs *c = trans->c;
1991 struct btree_iter iter;
1992 struct bkey_i_alloc_v4 *a;
1996 a = bch2_trans_start_alloc_update(trans, &iter, POS(dev, bucket_nr));
1997 ret = PTR_ERR_OR_ZERO(a);
2001 now = atomic64_read(&c->io_clock[rw].now);
2002 if (a->v.io_time[rw] == now)
2005 a->v.io_time[rw] = now;
2007 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
2008 bch2_trans_commit(trans, NULL, NULL, 0);
2010 bch2_trans_iter_exit(trans, &iter);
2014 /* Startup/shutdown (ro/rw): */
2016 void bch2_recalc_capacity(struct bch_fs *c)
2019 u64 capacity = 0, reserved_sectors = 0, gc_reserve;
2020 unsigned bucket_size_max = 0;
2021 unsigned long ra_pages = 0;
2024 lockdep_assert_held(&c->state_lock);
2026 for_each_online_member(ca, c, i) {
2027 struct backing_dev_info *bdi = ca->disk_sb.bdev->bd_disk->bdi;
2029 ra_pages += bdi->ra_pages;
2032 bch2_set_ra_pages(c, ra_pages);
2034 for_each_rw_member(ca, c, i) {
2035 u64 dev_reserve = 0;
2038 * We need to reserve buckets (from the number
2039 * of currently available buckets) against
2040 * foreground writes so that mainly copygc can
2041 * make forward progress.
2043 * We need enough to refill the various reserves
2044 * from scratch - copygc will use its entire
2045 * reserve all at once, then run against when
2046 * its reserve is refilled (from the formerly
2047 * available buckets).
2049 * This reserve is just used when considering if
2050 * allocations for foreground writes must wait -
2051 * not -ENOSPC calculations.
2054 dev_reserve += ca->nr_btree_reserve * 2;
2055 dev_reserve += ca->mi.nbuckets >> 6; /* copygc reserve */
2057 dev_reserve += 1; /* btree write point */
2058 dev_reserve += 1; /* copygc write point */
2059 dev_reserve += 1; /* rebalance write point */
2061 dev_reserve *= ca->mi.bucket_size;
2063 capacity += bucket_to_sector(ca, ca->mi.nbuckets -
2064 ca->mi.first_bucket);
2066 reserved_sectors += dev_reserve * 2;
2068 bucket_size_max = max_t(unsigned, bucket_size_max,
2069 ca->mi.bucket_size);
2072 gc_reserve = c->opts.gc_reserve_bytes
2073 ? c->opts.gc_reserve_bytes >> 9
2074 : div64_u64(capacity * c->opts.gc_reserve_percent, 100);
2076 reserved_sectors = max(gc_reserve, reserved_sectors);
2078 reserved_sectors = min(reserved_sectors, capacity);
2080 c->capacity = capacity - reserved_sectors;
2082 c->bucket_size_max = bucket_size_max;
2084 /* Wake up case someone was waiting for buckets */
2085 closure_wake_up(&c->freelist_wait);
2088 static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca)
2090 struct open_bucket *ob;
2093 for (ob = c->open_buckets;
2094 ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
2096 spin_lock(&ob->lock);
2097 if (ob->valid && !ob->on_partial_list &&
2098 ob->dev == ca->dev_idx)
2100 spin_unlock(&ob->lock);
2106 /* device goes ro: */
2107 void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca)
2111 /* First, remove device from allocation groups: */
2113 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2114 clear_bit(ca->dev_idx, c->rw_devs[i].d);
2117 * Capacity is calculated based off of devices in allocation groups:
2119 bch2_recalc_capacity(c);
2121 bch2_open_buckets_stop(c, ca, false);
2124 * Wake up threads that were blocked on allocation, so they can notice
2125 * the device can no longer be removed and the capacity has changed:
2127 closure_wake_up(&c->freelist_wait);
2130 * journal_res_get() can block waiting for free space in the journal -
2131 * it needs to notice there may not be devices to allocate from anymore:
2133 wake_up(&c->journal.wait);
2135 /* Now wait for any in flight writes: */
2137 closure_wait_event(&c->open_buckets_wait,
2138 !bch2_dev_has_open_write_point(c, ca));
2141 /* device goes rw: */
2142 void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca)
2146 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2147 if (ca->mi.data_allowed & (1 << i))
2148 set_bit(ca->dev_idx, c->rw_devs[i].d);
2151 void bch2_fs_allocator_background_init(struct bch_fs *c)
2153 spin_lock_init(&c->freelist_lock);
2154 INIT_WORK(&c->discard_work, bch2_do_discards_work);
2155 INIT_WORK(&c->invalidate_work, bch2_do_invalidates_work);