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 void bch2_alloc_v4_swab(struct bkey_s k)
302 struct bch_alloc_v4 *a = bkey_s_to_alloc_v4(k).v;
303 struct bch_backpointer *bp, *bps;
305 a->journal_seq = swab64(a->journal_seq);
306 a->flags = swab32(a->flags);
307 a->dirty_sectors = swab32(a->dirty_sectors);
308 a->cached_sectors = swab32(a->cached_sectors);
309 a->io_time[0] = swab64(a->io_time[0]);
310 a->io_time[1] = swab64(a->io_time[1]);
311 a->stripe = swab32(a->stripe);
312 a->nr_external_backpointers = swab32(a->nr_external_backpointers);
313 a->fragmentation_lru = swab64(a->fragmentation_lru);
315 bps = alloc_v4_backpointers(a);
316 for (bp = bps; bp < bps + BCH_ALLOC_V4_NR_BACKPOINTERS(a); bp++) {
317 bp->bucket_offset = swab40(bp->bucket_offset);
318 bp->bucket_len = swab32(bp->bucket_len);
319 bch2_bpos_swab(&bp->pos);
323 void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
325 struct bch_alloc_v4 _a;
326 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a);
330 printbuf_indent_add(out, 2);
332 prt_printf(out, "gen %u oldest_gen %u data_type %s",
333 a->gen, a->oldest_gen,
334 a->data_type < BCH_DATA_NR
335 ? bch2_data_types[a->data_type]
336 : "(invalid data type)");
338 prt_printf(out, "journal_seq %llu", a->journal_seq);
340 prt_printf(out, "need_discard %llu", BCH_ALLOC_V4_NEED_DISCARD(a));
342 prt_printf(out, "need_inc_gen %llu", BCH_ALLOC_V4_NEED_INC_GEN(a));
344 prt_printf(out, "dirty_sectors %u", a->dirty_sectors);
346 prt_printf(out, "cached_sectors %u", a->cached_sectors);
348 prt_printf(out, "stripe %u", a->stripe);
350 prt_printf(out, "stripe_redundancy %u", a->stripe_redundancy);
352 prt_printf(out, "io_time[READ] %llu", a->io_time[READ]);
354 prt_printf(out, "io_time[WRITE] %llu", a->io_time[WRITE]);
356 prt_printf(out, "fragmentation %llu", a->fragmentation_lru);
358 prt_printf(out, "bp_start %llu", BCH_ALLOC_V4_BACKPOINTERS_START(a));
361 if (BCH_ALLOC_V4_NR_BACKPOINTERS(a)) {
362 struct bkey_s_c_alloc_v4 a_raw = bkey_s_c_to_alloc_v4(k);
363 const struct bch_backpointer *bps = alloc_v4_backpointers_c(a_raw.v);
365 prt_printf(out, "backpointers: %llu", BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v));
366 printbuf_indent_add(out, 2);
368 for (i = 0; i < BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v); i++) {
370 bch2_backpointer_to_text(out, &bps[i]);
373 printbuf_indent_sub(out, 2);
376 printbuf_indent_sub(out, 2);
379 void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out)
381 if (k.k->type == KEY_TYPE_alloc_v4) {
384 *out = *bkey_s_c_to_alloc_v4(k).v;
386 src = alloc_v4_backpointers(out);
387 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
388 dst = alloc_v4_backpointers(out);
391 memset(src, 0, dst - src);
393 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(out, 0);
395 struct bkey_alloc_unpacked u = bch2_alloc_unpack(k);
397 *out = (struct bch_alloc_v4) {
398 .journal_seq = u.journal_seq,
399 .flags = u.need_discard,
401 .oldest_gen = u.oldest_gen,
402 .data_type = u.data_type,
403 .stripe_redundancy = u.stripe_redundancy,
404 .dirty_sectors = u.dirty_sectors,
405 .cached_sectors = u.cached_sectors,
406 .io_time[READ] = u.read_time,
407 .io_time[WRITE] = u.write_time,
411 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
415 static noinline struct bkey_i_alloc_v4 *
416 __bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
418 struct bkey_i_alloc_v4 *ret;
420 ret = bch2_trans_kmalloc(trans, max(bkey_bytes(k.k), sizeof(struct bkey_i_alloc_v4)));
424 if (k.k->type == KEY_TYPE_alloc_v4) {
427 bkey_reassemble(&ret->k_i, k);
429 src = alloc_v4_backpointers(&ret->v);
430 SET_BCH_ALLOC_V4_BACKPOINTERS_START(&ret->v, BCH_ALLOC_V4_U64s);
431 dst = alloc_v4_backpointers(&ret->v);
434 memset(src, 0, dst - src);
436 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(&ret->v, 0);
437 set_alloc_v4_u64s(ret);
439 bkey_alloc_v4_init(&ret->k_i);
441 bch2_alloc_to_v4(k, &ret->v);
446 static inline struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut_inlined(struct btree_trans *trans, struct bkey_s_c k)
448 struct bkey_s_c_alloc_v4 a;
450 if (likely(k.k->type == KEY_TYPE_alloc_v4) &&
451 ((a = bkey_s_c_to_alloc_v4(k), true) &&
452 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v) == 0))
453 return bch2_bkey_make_mut_noupdate_typed(trans, k, alloc_v4);
455 return __bch2_alloc_to_v4_mut(trans, k);
458 struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
460 return bch2_alloc_to_v4_mut_inlined(trans, k);
463 struct bkey_i_alloc_v4 *
464 bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter,
468 struct bkey_i_alloc_v4 *a;
471 k = bch2_bkey_get_iter(trans, iter, BTREE_ID_alloc, pos,
472 BTREE_ITER_WITH_UPDATES|
479 a = bch2_alloc_to_v4_mut_inlined(trans, k);
480 ret = PTR_ERR_OR_ZERO(a);
485 bch2_trans_iter_exit(trans, iter);
489 static struct bpos alloc_gens_pos(struct bpos pos, unsigned *offset)
491 *offset = pos.offset & KEY_TYPE_BUCKET_GENS_MASK;
493 pos.offset >>= KEY_TYPE_BUCKET_GENS_BITS;
497 static struct bpos bucket_gens_pos_to_alloc(struct bpos pos, unsigned offset)
499 pos.offset <<= KEY_TYPE_BUCKET_GENS_BITS;
500 pos.offset += offset;
504 static unsigned alloc_gen(struct bkey_s_c k, unsigned offset)
506 return k.k->type == KEY_TYPE_bucket_gens
507 ? bkey_s_c_to_bucket_gens(k).v->gens[offset]
511 int bch2_bucket_gens_invalid(struct bch_fs *c, struct bkey_s_c k,
512 enum bkey_invalid_flags flags,
513 struct printbuf *err)
517 bkey_fsck_err_on(bkey_val_bytes(k.k) != sizeof(struct bch_bucket_gens), c, err,
518 bucket_gens_val_size_bad,
519 "bad val size (%zu != %zu)",
520 bkey_val_bytes(k.k), sizeof(struct bch_bucket_gens));
525 void bch2_bucket_gens_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
527 struct bkey_s_c_bucket_gens g = bkey_s_c_to_bucket_gens(k);
530 for (i = 0; i < ARRAY_SIZE(g.v->gens); i++) {
533 prt_printf(out, "%u", g.v->gens[i]);
537 int bch2_bucket_gens_init(struct bch_fs *c)
539 struct btree_trans *trans = bch2_trans_get(c);
540 struct btree_iter iter;
542 struct bch_alloc_v4 a;
543 struct bkey_i_bucket_gens g;
544 bool have_bucket_gens_key = false;
550 for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
551 BTREE_ITER_PREFETCH, k, ret) {
553 * Not a fsck error because this is checked/repaired by
554 * bch2_check_alloc_key() which runs later:
556 if (!bch2_dev_bucket_exists(c, k.k->p))
559 gen = bch2_alloc_to_v4(k, &a)->gen;
560 pos = alloc_gens_pos(iter.pos, &offset);
562 if (have_bucket_gens_key && bkey_cmp(iter.pos, pos)) {
563 ret = commit_do(trans, NULL, NULL,
564 BCH_TRANS_COMMIT_no_enospc|
565 BCH_TRANS_COMMIT_lazy_rw,
566 bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
569 have_bucket_gens_key = false;
572 if (!have_bucket_gens_key) {
573 bkey_bucket_gens_init(&g.k_i);
575 have_bucket_gens_key = true;
578 g.v.gens[offset] = gen;
580 bch2_trans_iter_exit(trans, &iter);
582 if (have_bucket_gens_key && !ret)
583 ret = commit_do(trans, NULL, NULL,
584 BCH_TRANS_COMMIT_no_enospc|
585 BCH_TRANS_COMMIT_lazy_rw,
586 bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
588 bch2_trans_put(trans);
595 int bch2_alloc_read(struct bch_fs *c)
597 struct btree_trans *trans = bch2_trans_get(c);
598 struct btree_iter iter;
603 down_read(&c->gc_lock);
605 if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_bucket_gens) {
606 const struct bch_bucket_gens *g;
609 for_each_btree_key(trans, iter, BTREE_ID_bucket_gens, POS_MIN,
610 BTREE_ITER_PREFETCH, k, ret) {
611 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
612 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
614 if (k.k->type != KEY_TYPE_bucket_gens)
617 g = bkey_s_c_to_bucket_gens(k).v;
620 * Not a fsck error because this is checked/repaired by
621 * bch2_check_alloc_key() which runs later:
623 if (!bch2_dev_exists2(c, k.k->p.inode))
626 ca = bch_dev_bkey_exists(c, k.k->p.inode);
628 for (b = max_t(u64, ca->mi.first_bucket, start);
629 b < min_t(u64, ca->mi.nbuckets, end);
631 *bucket_gen(ca, b) = g->gens[b & KEY_TYPE_BUCKET_GENS_MASK];
633 bch2_trans_iter_exit(trans, &iter);
635 struct bch_alloc_v4 a;
637 for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
638 BTREE_ITER_PREFETCH, k, ret) {
640 * Not a fsck error because this is checked/repaired by
641 * bch2_check_alloc_key() which runs later:
643 if (!bch2_dev_bucket_exists(c, k.k->p))
646 ca = bch_dev_bkey_exists(c, k.k->p.inode);
648 *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen;
650 bch2_trans_iter_exit(trans, &iter);
653 bch2_trans_put(trans);
654 up_read(&c->gc_lock);
662 /* Free space/discard btree: */
664 static int bch2_bucket_do_index(struct btree_trans *trans,
665 struct bkey_s_c alloc_k,
666 const struct bch_alloc_v4 *a,
669 struct bch_fs *c = trans->c;
670 struct bch_dev *ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
671 struct btree_iter iter;
675 enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted;
676 enum bch_bkey_type new_type = set ? KEY_TYPE_set : KEY_TYPE_deleted;
677 struct printbuf buf = PRINTBUF;
680 if (a->data_type != BCH_DATA_free &&
681 a->data_type != BCH_DATA_need_discard)
684 k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k));
689 k->k.type = new_type;
691 switch (a->data_type) {
693 btree = BTREE_ID_freespace;
694 k->k.p = alloc_freespace_pos(alloc_k.k->p, *a);
695 bch2_key_resize(&k->k, 1);
697 case BCH_DATA_need_discard:
698 btree = BTREE_ID_need_discard;
699 k->k.p = alloc_k.k->p;
705 old = bch2_bkey_get_iter(trans, &iter, btree,
706 bkey_start_pos(&k->k),
712 if (ca->mi.freespace_initialized &&
713 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info &&
714 bch2_trans_inconsistent_on(old.k->type != old_type, trans,
715 "incorrect key when %s %s:%llu:%llu:0 (got %s should be %s)\n"
717 set ? "setting" : "clearing",
718 bch2_btree_id_str(btree),
721 bch2_bkey_types[old.k->type],
722 bch2_bkey_types[old_type],
723 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
728 ret = bch2_trans_update(trans, &iter, k, 0);
730 bch2_trans_iter_exit(trans, &iter);
735 static noinline int bch2_bucket_gen_update(struct btree_trans *trans,
736 struct bpos bucket, u8 gen)
738 struct btree_iter iter;
740 struct bpos pos = alloc_gens_pos(bucket, &offset);
741 struct bkey_i_bucket_gens *g;
745 g = bch2_trans_kmalloc(trans, sizeof(*g));
746 ret = PTR_ERR_OR_ZERO(g);
750 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_bucket_gens, pos,
752 BTREE_ITER_WITH_UPDATES);
757 if (k.k->type != KEY_TYPE_bucket_gens) {
758 bkey_bucket_gens_init(&g->k_i);
761 bkey_reassemble(&g->k_i, k);
764 g->v.gens[offset] = gen;
766 ret = bch2_trans_update(trans, &iter, &g->k_i, 0);
767 bch2_trans_iter_exit(trans, &iter);
771 int bch2_trans_mark_alloc(struct btree_trans *trans,
772 enum btree_id btree_id, unsigned level,
773 struct bkey_s_c old, struct bkey_i *new,
776 struct bch_fs *c = trans->c;
777 struct bch_alloc_v4 old_a_convert, *new_a;
778 const struct bch_alloc_v4 *old_a;
779 u64 old_lru, new_lru;
783 * Deletion only happens in the device removal path, with
784 * BTREE_TRIGGER_NORUN:
786 BUG_ON(new->k.type != KEY_TYPE_alloc_v4);
788 old_a = bch2_alloc_to_v4(old, &old_a_convert);
789 new_a = &bkey_i_to_alloc_v4(new)->v;
791 new_a->data_type = alloc_data_type(*new_a, new_a->data_type);
793 if (new_a->dirty_sectors > old_a->dirty_sectors ||
794 new_a->cached_sectors > old_a->cached_sectors) {
795 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
796 new_a->io_time[WRITE]= max_t(u64, 1, atomic64_read(&c->io_clock[WRITE].now));
797 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true);
798 SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true);
801 if (data_type_is_empty(new_a->data_type) &&
802 BCH_ALLOC_V4_NEED_INC_GEN(new_a) &&
803 !bch2_bucket_is_open_safe(c, new->k.p.inode, new->k.p.offset)) {
805 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false);
808 if (old_a->data_type != new_a->data_type ||
809 (new_a->data_type == BCH_DATA_free &&
810 alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) {
811 ret = bch2_bucket_do_index(trans, old, old_a, false) ?:
812 bch2_bucket_do_index(trans, bkey_i_to_s_c(new), new_a, true);
817 if (new_a->data_type == BCH_DATA_cached &&
818 !new_a->io_time[READ])
819 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
821 old_lru = alloc_lru_idx_read(*old_a);
822 new_lru = alloc_lru_idx_read(*new_a);
824 if (old_lru != new_lru) {
825 ret = bch2_lru_change(trans, new->k.p.inode,
826 bucket_to_u64(new->k.p),
832 new_a->fragmentation_lru = alloc_lru_idx_fragmentation(*new_a,
833 bch_dev_bkey_exists(c, new->k.p.inode));
835 if (old_a->fragmentation_lru != new_a->fragmentation_lru) {
836 ret = bch2_lru_change(trans,
837 BCH_LRU_FRAGMENTATION_START,
838 bucket_to_u64(new->k.p),
839 old_a->fragmentation_lru, new_a->fragmentation_lru);
844 if (old_a->gen != new_a->gen) {
845 ret = bch2_bucket_gen_update(trans, new->k.p, new_a->gen);
851 * need to know if we're getting called from the invalidate path or
855 if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
856 old_a->cached_sectors) {
857 ret = bch2_update_cached_sectors_list(trans, new->k.p.inode,
858 -((s64) old_a->cached_sectors));
867 * This synthesizes deleted extents for holes, similar to BTREE_ITER_SLOTS for
868 * extents style btrees, but works on non-extents btrees:
870 static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos end, struct bkey *hole)
872 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
880 struct btree_iter iter2;
883 bch2_trans_copy_iter(&iter2, iter);
885 if (!bpos_eq(iter->path->l[0].b->key.k.p, SPOS_MAX))
886 end = bkey_min(end, bpos_nosnap_successor(iter->path->l[0].b->key.k.p));
888 end = bkey_min(end, POS(iter->pos.inode, iter->pos.offset + U32_MAX - 1));
891 * btree node min/max is a closed interval, upto takes a half
894 k = bch2_btree_iter_peek_upto(&iter2, end);
896 bch2_trans_iter_exit(iter->trans, &iter2);
898 BUG_ON(next.offset >= iter->pos.offset + U32_MAX);
906 bch2_key_resize(hole, next.offset - iter->pos.offset);
907 return (struct bkey_s_c) { hole, NULL };
911 static bool next_bucket(struct bch_fs *c, struct bpos *bucket)
916 if (bch2_dev_bucket_exists(c, *bucket))
919 if (bch2_dev_exists2(c, bucket->inode)) {
920 ca = bch_dev_bkey_exists(c, bucket->inode);
922 if (bucket->offset < ca->mi.first_bucket) {
923 bucket->offset = ca->mi.first_bucket;
932 iter = bucket->inode;
933 ca = __bch2_next_dev(c, &iter, NULL);
935 *bucket = POS(ca->dev_idx, ca->mi.first_bucket);
941 static struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, struct bkey *hole)
943 struct bch_fs *c = iter->trans->c;
946 k = bch2_get_key_or_hole(iter, POS_MAX, hole);
951 struct bpos bucket = bkey_start_pos(k.k);
953 if (!bch2_dev_bucket_exists(c, bucket)) {
954 if (!next_bucket(c, &bucket))
955 return bkey_s_c_null;
957 bch2_btree_iter_set_pos(iter, bucket);
961 if (!bch2_dev_bucket_exists(c, k.k->p)) {
962 struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode);
964 bch2_key_resize(hole, ca->mi.nbuckets - bucket.offset);
971 static noinline_for_stack
972 int bch2_check_alloc_key(struct btree_trans *trans,
973 struct bkey_s_c alloc_k,
974 struct btree_iter *alloc_iter,
975 struct btree_iter *discard_iter,
976 struct btree_iter *freespace_iter,
977 struct btree_iter *bucket_gens_iter)
979 struct bch_fs *c = trans->c;
981 struct bch_alloc_v4 a_convert;
982 const struct bch_alloc_v4 *a;
983 unsigned discard_key_type, freespace_key_type;
984 unsigned gens_offset;
986 struct printbuf buf = PRINTBUF;
989 if (fsck_err_on(!bch2_dev_bucket_exists(c, alloc_k.k->p), c,
990 alloc_key_to_missing_dev_bucket,
991 "alloc key for invalid device:bucket %llu:%llu",
992 alloc_k.k->p.inode, alloc_k.k->p.offset))
993 return bch2_btree_delete_at(trans, alloc_iter, 0);
995 ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
996 if (!ca->mi.freespace_initialized)
999 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1001 discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0;
1002 bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p);
1003 k = bch2_btree_iter_peek_slot(discard_iter);
1008 if (k.k->type != discard_key_type &&
1009 (c->opts.reconstruct_alloc ||
1010 fsck_err(c, need_discard_key_wrong,
1011 "incorrect key in need_discard btree (got %s should be %s)\n"
1013 bch2_bkey_types[k.k->type],
1014 bch2_bkey_types[discard_key_type],
1015 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1016 struct bkey_i *update =
1017 bch2_trans_kmalloc(trans, sizeof(*update));
1019 ret = PTR_ERR_OR_ZERO(update);
1023 bkey_init(&update->k);
1024 update->k.type = discard_key_type;
1025 update->k.p = discard_iter->pos;
1027 ret = bch2_trans_update(trans, discard_iter, update, 0);
1032 freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0;
1033 bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a));
1034 k = bch2_btree_iter_peek_slot(freespace_iter);
1039 if (k.k->type != freespace_key_type &&
1040 (c->opts.reconstruct_alloc ||
1041 fsck_err(c, freespace_key_wrong,
1042 "incorrect key in freespace btree (got %s should be %s)\n"
1044 bch2_bkey_types[k.k->type],
1045 bch2_bkey_types[freespace_key_type],
1046 (printbuf_reset(&buf),
1047 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1048 struct bkey_i *update =
1049 bch2_trans_kmalloc(trans, sizeof(*update));
1051 ret = PTR_ERR_OR_ZERO(update);
1055 bkey_init(&update->k);
1056 update->k.type = freespace_key_type;
1057 update->k.p = freespace_iter->pos;
1058 bch2_key_resize(&update->k, 1);
1060 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1065 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(alloc_k.k->p, &gens_offset));
1066 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1071 if (a->gen != alloc_gen(k, gens_offset) &&
1072 (c->opts.reconstruct_alloc ||
1073 fsck_err(c, bucket_gens_key_wrong,
1074 "incorrect gen in bucket_gens btree (got %u should be %u)\n"
1076 alloc_gen(k, gens_offset), a->gen,
1077 (printbuf_reset(&buf),
1078 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1079 struct bkey_i_bucket_gens *g =
1080 bch2_trans_kmalloc(trans, sizeof(*g));
1082 ret = PTR_ERR_OR_ZERO(g);
1086 if (k.k->type == KEY_TYPE_bucket_gens) {
1087 bkey_reassemble(&g->k_i, k);
1089 bkey_bucket_gens_init(&g->k_i);
1090 g->k.p = alloc_gens_pos(alloc_k.k->p, &gens_offset);
1093 g->v.gens[gens_offset] = a->gen;
1095 ret = bch2_trans_update(trans, bucket_gens_iter, &g->k_i, 0);
1101 printbuf_exit(&buf);
1105 static noinline_for_stack
1106 int bch2_check_alloc_hole_freespace(struct btree_trans *trans,
1109 struct btree_iter *freespace_iter)
1111 struct bch_fs *c = trans->c;
1114 struct printbuf buf = PRINTBUF;
1117 ca = bch_dev_bkey_exists(c, start.inode);
1118 if (!ca->mi.freespace_initialized)
1121 bch2_btree_iter_set_pos(freespace_iter, start);
1123 k = bch2_btree_iter_peek_slot(freespace_iter);
1128 *end = bkey_min(k.k->p, *end);
1130 if (k.k->type != KEY_TYPE_set &&
1131 (c->opts.reconstruct_alloc ||
1132 fsck_err(c, freespace_hole_missing,
1133 "hole in alloc btree missing in freespace btree\n"
1134 " device %llu buckets %llu-%llu",
1135 freespace_iter->pos.inode,
1136 freespace_iter->pos.offset,
1138 struct bkey_i *update =
1139 bch2_trans_kmalloc(trans, sizeof(*update));
1141 ret = PTR_ERR_OR_ZERO(update);
1145 bkey_init(&update->k);
1146 update->k.type = KEY_TYPE_set;
1147 update->k.p = freespace_iter->pos;
1148 bch2_key_resize(&update->k,
1149 min_t(u64, U32_MAX, end->offset -
1150 freespace_iter->pos.offset));
1152 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1158 printbuf_exit(&buf);
1162 static noinline_for_stack
1163 int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans,
1166 struct btree_iter *bucket_gens_iter)
1168 struct bch_fs *c = trans->c;
1170 struct printbuf buf = PRINTBUF;
1171 unsigned i, gens_offset, gens_end_offset;
1174 if (c->sb.version < bcachefs_metadata_version_bucket_gens)
1177 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(start, &gens_offset));
1179 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1184 if (bkey_cmp(alloc_gens_pos(start, &gens_offset),
1185 alloc_gens_pos(*end, &gens_end_offset)))
1186 gens_end_offset = KEY_TYPE_BUCKET_GENS_NR;
1188 if (k.k->type == KEY_TYPE_bucket_gens) {
1189 struct bkey_i_bucket_gens g;
1190 bool need_update = false;
1192 bkey_reassemble(&g.k_i, k);
1194 for (i = gens_offset; i < gens_end_offset; i++) {
1195 if (fsck_err_on(g.v.gens[i], c,
1196 bucket_gens_hole_wrong,
1197 "hole in alloc btree at %llu:%llu with nonzero gen in bucket_gens btree (%u)",
1198 bucket_gens_pos_to_alloc(k.k->p, i).inode,
1199 bucket_gens_pos_to_alloc(k.k->p, i).offset,
1207 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1209 ret = PTR_ERR_OR_ZERO(u);
1213 memcpy(u, &g, sizeof(g));
1215 ret = bch2_trans_update(trans, bucket_gens_iter, u, 0);
1221 *end = bkey_min(*end, bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0));
1224 printbuf_exit(&buf);
1228 static noinline_for_stack int bch2_check_discard_freespace_key(struct btree_trans *trans,
1229 struct btree_iter *iter)
1231 struct bch_fs *c = trans->c;
1232 struct btree_iter alloc_iter;
1233 struct bkey_s_c alloc_k;
1234 struct bch_alloc_v4 a_convert;
1235 const struct bch_alloc_v4 *a;
1238 enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard
1239 ? BCH_DATA_need_discard
1241 struct printbuf buf = PRINTBUF;
1245 pos.offset &= ~(~0ULL << 56);
1246 genbits = iter->pos.offset & (~0ULL << 56);
1248 alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc, pos, 0);
1249 ret = bkey_err(alloc_k);
1253 if (fsck_err_on(!bch2_dev_bucket_exists(c, pos), c,
1254 need_discard_freespace_key_to_invalid_dev_bucket,
1255 "entry in %s btree for nonexistant dev:bucket %llu:%llu",
1256 bch2_btree_id_str(iter->btree_id), pos.inode, pos.offset))
1259 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1261 if (fsck_err_on(a->data_type != state ||
1262 (state == BCH_DATA_free &&
1263 genbits != alloc_freespace_genbits(*a)), c,
1264 need_discard_freespace_key_bad,
1265 "%s\n incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)",
1266 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
1267 bch2_btree_id_str(iter->btree_id),
1270 a->data_type == state,
1271 genbits >> 56, alloc_freespace_genbits(*a) >> 56))
1275 set_btree_iter_dontneed(&alloc_iter);
1276 bch2_trans_iter_exit(trans, &alloc_iter);
1277 printbuf_exit(&buf);
1280 ret = bch2_btree_delete_extent_at(trans, iter,
1281 iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0) ?:
1282 bch2_trans_commit(trans, NULL, NULL,
1283 BCH_TRANS_COMMIT_no_enospc|BCH_TRANS_COMMIT_lazy_rw);
1288 * We've already checked that generation numbers in the bucket_gens btree are
1289 * valid for buckets that exist; this just checks for keys for nonexistent
1292 static noinline_for_stack
1293 int bch2_check_bucket_gens_key(struct btree_trans *trans,
1294 struct btree_iter *iter,
1297 struct bch_fs *c = trans->c;
1298 struct bkey_i_bucket_gens g;
1300 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
1301 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
1303 bool need_update = false, dev_exists;
1304 struct printbuf buf = PRINTBUF;
1307 BUG_ON(k.k->type != KEY_TYPE_bucket_gens);
1308 bkey_reassemble(&g.k_i, k);
1310 /* if no bch_dev, skip out whether we repair or not */
1311 dev_exists = bch2_dev_exists2(c, k.k->p.inode);
1313 if (fsck_err_on(!dev_exists, c,
1314 bucket_gens_to_invalid_dev,
1315 "bucket_gens key for invalid device:\n %s",
1316 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1317 ret = bch2_btree_delete_at(trans, iter, 0);
1322 ca = bch_dev_bkey_exists(c, k.k->p.inode);
1323 if (fsck_err_on(end <= ca->mi.first_bucket ||
1324 start >= ca->mi.nbuckets, c,
1325 bucket_gens_to_invalid_buckets,
1326 "bucket_gens key for invalid buckets:\n %s",
1327 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1328 ret = bch2_btree_delete_at(trans, iter, 0);
1332 for (b = start; b < ca->mi.first_bucket; b++)
1333 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1334 bucket_gens_nonzero_for_invalid_buckets,
1335 "bucket_gens key has nonzero gen for invalid bucket")) {
1336 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1340 for (b = ca->mi.nbuckets; b < end; b++)
1341 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1342 bucket_gens_nonzero_for_invalid_buckets,
1343 "bucket_gens key has nonzero gen for invalid bucket")) {
1344 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1349 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1351 ret = PTR_ERR_OR_ZERO(u);
1355 memcpy(u, &g, sizeof(g));
1356 ret = bch2_trans_update(trans, iter, u, 0);
1360 printbuf_exit(&buf);
1364 int bch2_check_alloc_info(struct bch_fs *c)
1366 struct btree_trans *trans = bch2_trans_get(c);
1367 struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter;
1372 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS_MIN,
1373 BTREE_ITER_PREFETCH);
1374 bch2_trans_iter_init(trans, &discard_iter, BTREE_ID_need_discard, POS_MIN,
1375 BTREE_ITER_PREFETCH);
1376 bch2_trans_iter_init(trans, &freespace_iter, BTREE_ID_freespace, POS_MIN,
1377 BTREE_ITER_PREFETCH);
1378 bch2_trans_iter_init(trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN,
1379 BTREE_ITER_PREFETCH);
1384 bch2_trans_begin(trans);
1386 k = bch2_get_key_or_real_bucket_hole(&iter, &hole);
1395 next = bpos_nosnap_successor(k.k->p);
1397 ret = bch2_check_alloc_key(trans,
1407 ret = bch2_check_alloc_hole_freespace(trans,
1408 bkey_start_pos(k.k),
1411 bch2_check_alloc_hole_bucket_gens(trans,
1412 bkey_start_pos(k.k),
1419 ret = bch2_trans_commit(trans, NULL, NULL,
1420 BCH_TRANS_COMMIT_no_enospc|
1421 BCH_TRANS_COMMIT_lazy_rw);
1425 bch2_btree_iter_set_pos(&iter, next);
1427 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1432 bch2_trans_iter_exit(trans, &bucket_gens_iter);
1433 bch2_trans_iter_exit(trans, &freespace_iter);
1434 bch2_trans_iter_exit(trans, &discard_iter);
1435 bch2_trans_iter_exit(trans, &iter);
1440 ret = for_each_btree_key2(trans, iter,
1441 BTREE_ID_need_discard, POS_MIN,
1442 BTREE_ITER_PREFETCH, k,
1443 bch2_check_discard_freespace_key(trans, &iter));
1447 bch2_trans_iter_init(trans, &iter, BTREE_ID_freespace, POS_MIN,
1448 BTREE_ITER_PREFETCH);
1450 bch2_trans_begin(trans);
1451 k = bch2_btree_iter_peek(&iter);
1455 ret = bkey_err(k) ?:
1456 bch2_check_discard_freespace_key(trans, &iter);
1457 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1462 struct printbuf buf = PRINTBUF;
1463 bch2_bkey_val_to_text(&buf, c, k);
1465 bch_err(c, "while checking %s", buf.buf);
1466 printbuf_exit(&buf);
1470 bch2_btree_iter_set_pos(&iter, bpos_nosnap_successor(iter.pos));
1472 bch2_trans_iter_exit(trans, &iter);
1476 ret = for_each_btree_key_commit(trans, iter,
1477 BTREE_ID_bucket_gens, POS_MIN,
1478 BTREE_ITER_PREFETCH, k,
1479 NULL, NULL, BCH_TRANS_COMMIT_no_enospc|BCH_TRANS_COMMIT_lazy_rw,
1480 bch2_check_bucket_gens_key(trans, &iter, k));
1482 bch2_trans_put(trans);
1488 static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
1489 struct btree_iter *alloc_iter)
1491 struct bch_fs *c = trans->c;
1492 struct btree_iter lru_iter;
1493 struct bch_alloc_v4 a_convert;
1494 const struct bch_alloc_v4 *a;
1495 struct bkey_s_c alloc_k, lru_k;
1496 struct printbuf buf = PRINTBUF;
1499 alloc_k = bch2_btree_iter_peek(alloc_iter);
1503 ret = bkey_err(alloc_k);
1507 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1509 if (a->data_type != BCH_DATA_cached)
1512 lru_k = bch2_bkey_get_iter(trans, &lru_iter, BTREE_ID_lru,
1513 lru_pos(alloc_k.k->p.inode,
1514 bucket_to_u64(alloc_k.k->p),
1515 a->io_time[READ]), 0);
1516 ret = bkey_err(lru_k);
1520 if (fsck_err_on(!a->io_time[READ], c,
1521 alloc_key_cached_but_read_time_zero,
1522 "cached bucket with read_time 0\n"
1524 (printbuf_reset(&buf),
1525 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)) ||
1526 fsck_err_on(lru_k.k->type != KEY_TYPE_set, c,
1527 alloc_key_to_missing_lru_entry,
1528 "missing lru entry\n"
1530 (printbuf_reset(&buf),
1531 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1532 u64 read_time = a->io_time[READ] ?:
1533 atomic64_read(&c->io_clock[READ].now);
1535 ret = bch2_lru_set(trans,
1537 bucket_to_u64(alloc_k.k->p),
1542 if (a->io_time[READ] != read_time) {
1543 struct bkey_i_alloc_v4 *a_mut =
1544 bch2_alloc_to_v4_mut(trans, alloc_k);
1545 ret = PTR_ERR_OR_ZERO(a_mut);
1549 a_mut->v.io_time[READ] = read_time;
1550 ret = bch2_trans_update(trans, alloc_iter,
1551 &a_mut->k_i, BTREE_TRIGGER_NORUN);
1558 bch2_trans_iter_exit(trans, &lru_iter);
1559 printbuf_exit(&buf);
1563 int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
1565 struct btree_iter iter;
1569 ret = bch2_trans_run(c,
1570 for_each_btree_key_commit(trans, iter, BTREE_ID_alloc,
1571 POS_MIN, BTREE_ITER_PREFETCH, k,
1572 NULL, NULL, BCH_TRANS_COMMIT_no_enospc|BCH_TRANS_COMMIT_lazy_rw,
1573 bch2_check_alloc_to_lru_ref(trans, &iter)));
1579 static int bch2_discard_one_bucket(struct btree_trans *trans,
1580 struct btree_iter *need_discard_iter,
1581 struct bpos *discard_pos_done,
1584 u64 *need_journal_commit,
1587 struct bch_fs *c = trans->c;
1588 struct bpos pos = need_discard_iter->pos;
1589 struct btree_iter iter = { NULL };
1592 struct bkey_i_alloc_v4 *a;
1593 struct printbuf buf = PRINTBUF;
1596 ca = bch_dev_bkey_exists(c, pos.inode);
1597 if (!percpu_ref_tryget(&ca->io_ref)) {
1598 bch2_btree_iter_set_pos(need_discard_iter, POS(pos.inode + 1, 0));
1602 if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) {
1607 if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
1608 c->journal.flushed_seq_ondisk,
1609 pos.inode, pos.offset)) {
1610 (*need_journal_commit)++;
1614 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc,
1615 need_discard_iter->pos,
1621 a = bch2_alloc_to_v4_mut(trans, k);
1622 ret = PTR_ERR_OR_ZERO(a);
1626 if (BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
1628 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1632 if (a->v.journal_seq > c->journal.flushed_seq_ondisk) {
1633 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1634 bch2_trans_inconsistent(trans,
1635 "clearing need_discard but journal_seq %llu > flushed_seq %llu\n"
1638 c->journal.flushed_seq_ondisk,
1639 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1645 if (a->v.data_type != BCH_DATA_need_discard) {
1646 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1647 bch2_trans_inconsistent(trans,
1648 "bucket incorrectly set in need_discard btree\n"
1650 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1657 if (!bkey_eq(*discard_pos_done, iter.pos) &&
1658 ca->mi.discard && !c->opts.nochanges) {
1660 * This works without any other locks because this is the only
1661 * thread that removes items from the need_discard tree
1663 bch2_trans_unlock(trans);
1664 blkdev_issue_discard(ca->disk_sb.bdev,
1665 k.k->p.offset * ca->mi.bucket_size,
1668 *discard_pos_done = iter.pos;
1670 ret = bch2_trans_relock_notrace(trans);
1675 SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
1676 a->v.data_type = alloc_data_type(a->v, a->v.data_type);
1678 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
1679 bch2_trans_commit(trans, NULL, NULL,
1680 BCH_WATERMARK_btree|
1681 BCH_TRANS_COMMIT_no_enospc);
1685 this_cpu_inc(c->counters[BCH_COUNTER_bucket_discard]);
1689 bch2_trans_iter_exit(trans, &iter);
1690 percpu_ref_put(&ca->io_ref);
1691 printbuf_exit(&buf);
1695 static void bch2_do_discards_work(struct work_struct *work)
1697 struct bch_fs *c = container_of(work, struct bch_fs, discard_work);
1698 struct btree_iter iter;
1700 u64 seen = 0, open = 0, need_journal_commit = 0, discarded = 0;
1701 struct bpos discard_pos_done = POS_MAX;
1705 * We're doing the commit in bch2_discard_one_bucket instead of using
1706 * for_each_btree_key_commit() so that we can increment counters after
1707 * successful commit:
1709 ret = bch2_trans_run(c,
1710 for_each_btree_key2(trans, iter,
1711 BTREE_ID_need_discard, POS_MIN, 0, k,
1712 bch2_discard_one_bucket(trans, &iter, &discard_pos_done,
1715 &need_journal_commit,
1718 if (need_journal_commit * 2 > seen)
1719 bch2_journal_flush_async(&c->journal, NULL);
1721 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1723 trace_discard_buckets(c, seen, open, need_journal_commit, discarded,
1727 void bch2_do_discards(struct bch_fs *c)
1729 if (bch2_write_ref_tryget(c, BCH_WRITE_REF_discard) &&
1730 !queue_work(c->write_ref_wq, &c->discard_work))
1731 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1734 static int invalidate_one_bucket(struct btree_trans *trans,
1735 struct btree_iter *lru_iter,
1736 struct bkey_s_c lru_k,
1737 s64 *nr_to_invalidate)
1739 struct bch_fs *c = trans->c;
1740 struct btree_iter alloc_iter = { NULL };
1741 struct bkey_i_alloc_v4 *a = NULL;
1742 struct printbuf buf = PRINTBUF;
1743 struct bpos bucket = u64_to_bucket(lru_k.k->p.offset);
1744 unsigned cached_sectors;
1747 if (*nr_to_invalidate <= 0)
1750 if (!bch2_dev_bucket_exists(c, bucket)) {
1751 prt_str(&buf, "lru entry points to invalid bucket");
1755 if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset))
1758 a = bch2_trans_start_alloc_update(trans, &alloc_iter, bucket);
1759 ret = PTR_ERR_OR_ZERO(a);
1763 /* We expect harmless races here due to the btree write buffer: */
1764 if (lru_pos_time(lru_iter->pos) != alloc_lru_idx_read(a->v))
1767 BUG_ON(a->v.data_type != BCH_DATA_cached);
1769 if (!a->v.cached_sectors)
1770 bch_err(c, "invalidating empty bucket, confused");
1772 cached_sectors = a->v.cached_sectors;
1774 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1777 a->v.dirty_sectors = 0;
1778 a->v.cached_sectors = 0;
1779 a->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now);
1780 a->v.io_time[WRITE] = atomic64_read(&c->io_clock[WRITE].now);
1782 ret = bch2_trans_update(trans, &alloc_iter, &a->k_i,
1783 BTREE_TRIGGER_BUCKET_INVALIDATE) ?:
1784 bch2_trans_commit(trans, NULL, NULL,
1785 BCH_WATERMARK_btree|
1786 BCH_TRANS_COMMIT_no_enospc);
1790 trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors);
1791 --*nr_to_invalidate;
1793 bch2_trans_iter_exit(trans, &alloc_iter);
1794 printbuf_exit(&buf);
1797 prt_str(&buf, "\n lru key: ");
1798 bch2_bkey_val_to_text(&buf, c, lru_k);
1800 prt_str(&buf, "\n lru entry: ");
1801 bch2_lru_pos_to_text(&buf, lru_iter->pos);
1803 prt_str(&buf, "\n alloc key: ");
1805 bch2_bpos_to_text(&buf, bucket);
1807 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i));
1809 bch_err(c, "%s", buf.buf);
1810 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_lrus) {
1811 bch2_inconsistent_error(c);
1818 static void bch2_do_invalidates_work(struct work_struct *work)
1820 struct bch_fs *c = container_of(work, struct bch_fs, invalidate_work);
1822 struct btree_trans *trans = bch2_trans_get(c);
1823 struct btree_iter iter;
1828 ret = bch2_btree_write_buffer_tryflush(trans);
1832 for_each_member_device(ca, c, i) {
1833 s64 nr_to_invalidate =
1834 should_invalidate_buckets(ca, bch2_dev_usage_read(ca));
1836 ret = for_each_btree_key2_upto(trans, iter, BTREE_ID_lru,
1837 lru_pos(ca->dev_idx, 0, 0),
1838 lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX),
1839 BTREE_ITER_INTENT, k,
1840 invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate));
1843 percpu_ref_put(&ca->ref);
1848 bch2_trans_put(trans);
1849 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1852 void bch2_do_invalidates(struct bch_fs *c)
1854 if (bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate) &&
1855 !queue_work(c->write_ref_wq, &c->invalidate_work))
1856 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1859 int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca,
1860 u64 bucket_start, u64 bucket_end)
1862 struct btree_trans *trans = bch2_trans_get(c);
1863 struct btree_iter iter;
1866 struct bpos end = POS(ca->dev_idx, bucket_end);
1867 struct bch_member *m;
1868 unsigned long last_updated = jiffies;
1871 BUG_ON(bucket_start > bucket_end);
1872 BUG_ON(bucket_end > ca->mi.nbuckets);
1874 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
1875 POS(ca->dev_idx, max_t(u64, ca->mi.first_bucket, bucket_start)),
1876 BTREE_ITER_PREFETCH);
1878 * Scan the alloc btree for every bucket on @ca, and add buckets to the
1879 * freespace/need_discard/need_gc_gens btrees as needed:
1882 if (last_updated + HZ * 10 < jiffies) {
1883 bch_info(ca, "%s: currently at %llu/%llu",
1884 __func__, iter.pos.offset, ca->mi.nbuckets);
1885 last_updated = jiffies;
1888 bch2_trans_begin(trans);
1890 if (bkey_ge(iter.pos, end)) {
1895 k = bch2_get_key_or_hole(&iter, end, &hole);
1902 * We process live keys in the alloc btree one at a
1905 struct bch_alloc_v4 a_convert;
1906 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert);
1908 ret = bch2_bucket_do_index(trans, k, a, true) ?:
1909 bch2_trans_commit(trans, NULL, NULL,
1910 BCH_TRANS_COMMIT_lazy_rw|
1911 BCH_TRANS_COMMIT_no_enospc);
1915 bch2_btree_iter_advance(&iter);
1917 struct bkey_i *freespace;
1919 freespace = bch2_trans_kmalloc(trans, sizeof(*freespace));
1920 ret = PTR_ERR_OR_ZERO(freespace);
1924 bkey_init(&freespace->k);
1925 freespace->k.type = KEY_TYPE_set;
1926 freespace->k.p = k.k->p;
1927 freespace->k.size = k.k->size;
1929 ret = bch2_btree_insert_trans(trans, BTREE_ID_freespace, freespace, 0) ?:
1930 bch2_trans_commit(trans, NULL, NULL,
1931 BCH_TRANS_COMMIT_lazy_rw|
1932 BCH_TRANS_COMMIT_no_enospc);
1936 bch2_btree_iter_set_pos(&iter, k.k->p);
1939 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1945 bch2_trans_iter_exit(trans, &iter);
1946 bch2_trans_put(trans);
1949 bch_err_msg(ca, ret, "initializing free space");
1953 mutex_lock(&c->sb_lock);
1954 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1955 SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true);
1956 mutex_unlock(&c->sb_lock);
1961 int bch2_fs_freespace_init(struct bch_fs *c)
1966 bool doing_init = false;
1969 * We can crash during the device add path, so we need to check this on
1973 for_each_member_device(ca, c, i) {
1974 if (ca->mi.freespace_initialized)
1978 bch_info(c, "initializing freespace");
1982 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1984 percpu_ref_put(&ca->ref);
1991 mutex_lock(&c->sb_lock);
1992 bch2_write_super(c);
1993 mutex_unlock(&c->sb_lock);
1994 bch_verbose(c, "done initializing freespace");
2000 /* Bucket IO clocks: */
2002 int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
2003 size_t bucket_nr, int rw)
2005 struct bch_fs *c = trans->c;
2006 struct btree_iter iter;
2007 struct bkey_i_alloc_v4 *a;
2011 a = bch2_trans_start_alloc_update(trans, &iter, POS(dev, bucket_nr));
2012 ret = PTR_ERR_OR_ZERO(a);
2016 now = atomic64_read(&c->io_clock[rw].now);
2017 if (a->v.io_time[rw] == now)
2020 a->v.io_time[rw] = now;
2022 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
2023 bch2_trans_commit(trans, NULL, NULL, 0);
2025 bch2_trans_iter_exit(trans, &iter);
2029 /* Startup/shutdown (ro/rw): */
2031 void bch2_recalc_capacity(struct bch_fs *c)
2034 u64 capacity = 0, reserved_sectors = 0, gc_reserve;
2035 unsigned bucket_size_max = 0;
2036 unsigned long ra_pages = 0;
2039 lockdep_assert_held(&c->state_lock);
2041 for_each_online_member(ca, c, i) {
2042 struct backing_dev_info *bdi = ca->disk_sb.bdev->bd_disk->bdi;
2044 ra_pages += bdi->ra_pages;
2047 bch2_set_ra_pages(c, ra_pages);
2049 for_each_rw_member(ca, c, i) {
2050 u64 dev_reserve = 0;
2053 * We need to reserve buckets (from the number
2054 * of currently available buckets) against
2055 * foreground writes so that mainly copygc can
2056 * make forward progress.
2058 * We need enough to refill the various reserves
2059 * from scratch - copygc will use its entire
2060 * reserve all at once, then run against when
2061 * its reserve is refilled (from the formerly
2062 * available buckets).
2064 * This reserve is just used when considering if
2065 * allocations for foreground writes must wait -
2066 * not -ENOSPC calculations.
2069 dev_reserve += ca->nr_btree_reserve * 2;
2070 dev_reserve += ca->mi.nbuckets >> 6; /* copygc reserve */
2072 dev_reserve += 1; /* btree write point */
2073 dev_reserve += 1; /* copygc write point */
2074 dev_reserve += 1; /* rebalance write point */
2076 dev_reserve *= ca->mi.bucket_size;
2078 capacity += bucket_to_sector(ca, ca->mi.nbuckets -
2079 ca->mi.first_bucket);
2081 reserved_sectors += dev_reserve * 2;
2083 bucket_size_max = max_t(unsigned, bucket_size_max,
2084 ca->mi.bucket_size);
2087 gc_reserve = c->opts.gc_reserve_bytes
2088 ? c->opts.gc_reserve_bytes >> 9
2089 : div64_u64(capacity * c->opts.gc_reserve_percent, 100);
2091 reserved_sectors = max(gc_reserve, reserved_sectors);
2093 reserved_sectors = min(reserved_sectors, capacity);
2095 c->capacity = capacity - reserved_sectors;
2097 c->bucket_size_max = bucket_size_max;
2099 /* Wake up case someone was waiting for buckets */
2100 closure_wake_up(&c->freelist_wait);
2103 u64 bch2_min_rw_member_capacity(struct bch_fs *c)
2109 for_each_rw_member(ca, c, i)
2110 ret = min(ret, ca->mi.nbuckets * ca->mi.bucket_size);
2114 static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca)
2116 struct open_bucket *ob;
2119 for (ob = c->open_buckets;
2120 ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
2122 spin_lock(&ob->lock);
2123 if (ob->valid && !ob->on_partial_list &&
2124 ob->dev == ca->dev_idx)
2126 spin_unlock(&ob->lock);
2132 /* device goes ro: */
2133 void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca)
2137 /* First, remove device from allocation groups: */
2139 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2140 clear_bit(ca->dev_idx, c->rw_devs[i].d);
2143 * Capacity is calculated based off of devices in allocation groups:
2145 bch2_recalc_capacity(c);
2147 bch2_open_buckets_stop(c, ca, false);
2150 * Wake up threads that were blocked on allocation, so they can notice
2151 * the device can no longer be removed and the capacity has changed:
2153 closure_wake_up(&c->freelist_wait);
2156 * journal_res_get() can block waiting for free space in the journal -
2157 * it needs to notice there may not be devices to allocate from anymore:
2159 wake_up(&c->journal.wait);
2161 /* Now wait for any in flight writes: */
2163 closure_wait_event(&c->open_buckets_wait,
2164 !bch2_dev_has_open_write_point(c, ca));
2167 /* device goes rw: */
2168 void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca)
2172 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2173 if (ca->mi.data_allowed & (1 << i))
2174 set_bit(ca->dev_idx, c->rw_devs[i].d);
2177 void bch2_fs_allocator_background_init(struct bch_fs *c)
2179 spin_lock_init(&c->freelist_lock);
2180 INIT_WORK(&c->discard_work, bch2_do_discards_work);
2181 INIT_WORK(&c->invalidate_work, bch2_do_invalidates_work);