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 inline void alloc_field_v1_put(struct bkey_i_alloc *a, void **p,
83 unsigned field, u64 v)
85 unsigned bytes = BCH_ALLOC_V1_FIELD_BYTES[field];
90 a->v.fields |= 1 << field;
97 *((__le16 *) *p) = cpu_to_le16(v);
100 *((__le32 *) *p) = cpu_to_le32(v);
103 *((__le64 *) *p) = cpu_to_le64(v);
112 static void bch2_alloc_unpack_v1(struct bkey_alloc_unpacked *out,
115 const struct bch_alloc *in = bkey_s_c_to_alloc(k).v;
116 const void *d = in->data;
121 #define x(_name, _bits) out->_name = alloc_field_v1_get(in, &d, idx++);
122 BCH_ALLOC_FIELDS_V1()
126 static int bch2_alloc_unpack_v2(struct bkey_alloc_unpacked *out,
129 struct bkey_s_c_alloc_v2 a = bkey_s_c_to_alloc_v2(k);
130 const u8 *in = a.v->data;
131 const u8 *end = bkey_val_end(a);
132 unsigned fieldnr = 0;
137 out->oldest_gen = a.v->oldest_gen;
138 out->data_type = a.v->data_type;
140 #define x(_name, _bits) \
141 if (fieldnr < a.v->nr_fields) { \
142 ret = bch2_varint_decode_fast(in, end, &v); \
150 if (v != out->_name) \
154 BCH_ALLOC_FIELDS_V2()
159 static int bch2_alloc_unpack_v3(struct bkey_alloc_unpacked *out,
162 struct bkey_s_c_alloc_v3 a = bkey_s_c_to_alloc_v3(k);
163 const u8 *in = a.v->data;
164 const u8 *end = bkey_val_end(a);
165 unsigned fieldnr = 0;
170 out->oldest_gen = a.v->oldest_gen;
171 out->data_type = a.v->data_type;
172 out->need_discard = BCH_ALLOC_V3_NEED_DISCARD(a.v);
173 out->need_inc_gen = BCH_ALLOC_V3_NEED_INC_GEN(a.v);
174 out->journal_seq = le64_to_cpu(a.v->journal_seq);
176 #define x(_name, _bits) \
177 if (fieldnr < a.v->nr_fields) { \
178 ret = bch2_varint_decode_fast(in, end, &v); \
186 if (v != out->_name) \
190 BCH_ALLOC_FIELDS_V2()
195 static struct bkey_alloc_unpacked bch2_alloc_unpack(struct bkey_s_c k)
197 struct bkey_alloc_unpacked ret = { .gen = 0 };
201 bch2_alloc_unpack_v1(&ret, k);
203 case KEY_TYPE_alloc_v2:
204 bch2_alloc_unpack_v2(&ret, k);
206 case KEY_TYPE_alloc_v3:
207 bch2_alloc_unpack_v3(&ret, k);
214 static unsigned bch_alloc_v1_val_u64s(const struct bch_alloc *a)
216 unsigned i, bytes = offsetof(struct bch_alloc, data);
218 for (i = 0; i < ARRAY_SIZE(BCH_ALLOC_V1_FIELD_BYTES); i++)
219 if (a->fields & (1 << i))
220 bytes += BCH_ALLOC_V1_FIELD_BYTES[i];
222 return DIV_ROUND_UP(bytes, sizeof(u64));
225 int bch2_alloc_v1_invalid(const struct bch_fs *c, struct bkey_s_c k,
226 enum bkey_invalid_flags flags,
227 struct printbuf *err)
229 struct bkey_s_c_alloc a = bkey_s_c_to_alloc(k);
231 /* allow for unknown fields */
232 if (bkey_val_u64s(a.k) < bch_alloc_v1_val_u64s(a.v)) {
233 prt_printf(err, "incorrect value size (%zu < %u)",
234 bkey_val_u64s(a.k), bch_alloc_v1_val_u64s(a.v));
235 return -BCH_ERR_invalid_bkey;
241 int bch2_alloc_v2_invalid(const struct bch_fs *c, struct bkey_s_c k,
242 enum bkey_invalid_flags flags,
243 struct printbuf *err)
245 struct bkey_alloc_unpacked u;
247 if (bch2_alloc_unpack_v2(&u, k)) {
248 prt_printf(err, "unpack error");
249 return -BCH_ERR_invalid_bkey;
255 int bch2_alloc_v3_invalid(const struct bch_fs *c, struct bkey_s_c k,
256 enum bkey_invalid_flags flags,
257 struct printbuf *err)
259 struct bkey_alloc_unpacked u;
261 if (bch2_alloc_unpack_v3(&u, k)) {
262 prt_printf(err, "unpack error");
263 return -BCH_ERR_invalid_bkey;
269 int bch2_alloc_v4_invalid(const struct bch_fs *c, struct bkey_s_c k,
270 unsigned flags, struct printbuf *err)
272 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
273 int rw = flags & WRITE;
275 if (alloc_v4_u64s(a.v) > bkey_val_u64s(k.k)) {
276 prt_printf(err, "bad val size (%u > %lu)",
277 alloc_v4_u64s(a.v), bkey_val_u64s(k.k));
278 return -BCH_ERR_invalid_bkey;
281 if (!BCH_ALLOC_V4_BACKPOINTERS_START(a.v) &&
282 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v)) {
283 prt_printf(err, "invalid backpointers_start");
284 return -BCH_ERR_invalid_bkey;
288 !(flags & BKEY_INVALID_JOURNAL) &&
289 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_btree_backpointers) {
290 unsigned i, bp_len = 0;
292 for (i = 0; i < BCH_ALLOC_V4_NR_BACKPOINTERS(a.v); i++)
293 bp_len += alloc_v4_backpointers_c(a.v)[i].bucket_len;
295 if (bp_len > a.v->dirty_sectors) {
296 prt_printf(err, "too many backpointers");
297 return -BCH_ERR_invalid_bkey;
302 if (alloc_data_type(*a.v, a.v->data_type) != a.v->data_type) {
303 prt_printf(err, "invalid data type (got %u should be %u)",
304 a.v->data_type, alloc_data_type(*a.v, a.v->data_type));
305 return -BCH_ERR_invalid_bkey;
308 switch (a.v->data_type) {
310 case BCH_DATA_need_gc_gens:
311 case BCH_DATA_need_discard:
312 if (a.v->dirty_sectors ||
313 a.v->cached_sectors ||
315 prt_printf(err, "empty data type free but have data");
316 return -BCH_ERR_invalid_bkey;
320 case BCH_DATA_journal:
323 case BCH_DATA_parity:
324 if (!a.v->dirty_sectors) {
325 prt_printf(err, "data_type %s but dirty_sectors==0",
326 bch2_data_types[a.v->data_type]);
327 return -BCH_ERR_invalid_bkey;
330 case BCH_DATA_cached:
331 if (!a.v->cached_sectors ||
332 a.v->dirty_sectors ||
334 prt_printf(err, "data type inconsistency");
335 return -BCH_ERR_invalid_bkey;
338 if (!a.v->io_time[READ] &&
339 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_to_lru_refs) {
340 prt_printf(err, "cached bucket with read_time == 0");
341 return -BCH_ERR_invalid_bkey;
344 case BCH_DATA_stripe:
346 prt_printf(err, "data_type %s but stripe==0",
347 bch2_data_types[a.v->data_type]);
348 return -BCH_ERR_invalid_bkey;
357 static inline u64 swab40(u64 x)
359 return (((x & 0x00000000ffULL) << 32)|
360 ((x & 0x000000ff00ULL) << 16)|
361 ((x & 0x0000ff0000ULL) >> 0)|
362 ((x & 0x00ff000000ULL) >> 16)|
363 ((x & 0xff00000000ULL) >> 32));
366 void bch2_alloc_v4_swab(struct bkey_s k)
368 struct bch_alloc_v4 *a = bkey_s_to_alloc_v4(k).v;
369 struct bch_backpointer *bp, *bps;
371 a->journal_seq = swab64(a->journal_seq);
372 a->flags = swab32(a->flags);
373 a->dirty_sectors = swab32(a->dirty_sectors);
374 a->cached_sectors = swab32(a->cached_sectors);
375 a->io_time[0] = swab64(a->io_time[0]);
376 a->io_time[1] = swab64(a->io_time[1]);
377 a->stripe = swab32(a->stripe);
378 a->nr_external_backpointers = swab32(a->nr_external_backpointers);
380 bps = alloc_v4_backpointers(a);
381 for (bp = bps; bp < bps + BCH_ALLOC_V4_NR_BACKPOINTERS(a); bp++) {
382 bp->bucket_offset = swab40(bp->bucket_offset);
383 bp->bucket_len = swab32(bp->bucket_len);
384 bch2_bpos_swab(&bp->pos);
388 void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
390 struct bch_alloc_v4 _a;
391 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a);
395 printbuf_indent_add(out, 2);
397 prt_printf(out, "gen %u oldest_gen %u data_type %s",
398 a->gen, a->oldest_gen,
399 a->data_type < BCH_DATA_NR
400 ? bch2_data_types[a->data_type]
401 : "(invalid data type)");
403 prt_printf(out, "journal_seq %llu", a->journal_seq);
405 prt_printf(out, "need_discard %llu", BCH_ALLOC_V4_NEED_DISCARD(a));
407 prt_printf(out, "need_inc_gen %llu", BCH_ALLOC_V4_NEED_INC_GEN(a));
409 prt_printf(out, "dirty_sectors %u", a->dirty_sectors);
411 prt_printf(out, "cached_sectors %u", a->cached_sectors);
413 prt_printf(out, "stripe %u", a->stripe);
415 prt_printf(out, "stripe_redundancy %u", a->stripe_redundancy);
417 prt_printf(out, "io_time[READ] %llu", a->io_time[READ]);
419 prt_printf(out, "io_time[WRITE] %llu", a->io_time[WRITE]);
421 prt_printf(out, "fragmentation %llu", a->fragmentation_lru);
423 prt_printf(out, "bp_start %llu", BCH_ALLOC_V4_BACKPOINTERS_START(a));
426 if (BCH_ALLOC_V4_NR_BACKPOINTERS(a)) {
427 struct bkey_s_c_alloc_v4 a_raw = bkey_s_c_to_alloc_v4(k);
428 const struct bch_backpointer *bps = alloc_v4_backpointers_c(a_raw.v);
430 prt_printf(out, "backpointers: %llu", BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v));
431 printbuf_indent_add(out, 2);
433 for (i = 0; i < BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v); i++) {
435 bch2_backpointer_to_text(out, &bps[i]);
438 printbuf_indent_sub(out, 2);
441 printbuf_indent_sub(out, 2);
444 void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out)
446 if (k.k->type == KEY_TYPE_alloc_v4) {
449 *out = *bkey_s_c_to_alloc_v4(k).v;
451 src = alloc_v4_backpointers(out);
452 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
453 dst = alloc_v4_backpointers(out);
456 memset(src, 0, dst - src);
458 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(out, 0);
460 struct bkey_alloc_unpacked u = bch2_alloc_unpack(k);
462 *out = (struct bch_alloc_v4) {
463 .journal_seq = u.journal_seq,
464 .flags = u.need_discard,
466 .oldest_gen = u.oldest_gen,
467 .data_type = u.data_type,
468 .stripe_redundancy = u.stripe_redundancy,
469 .dirty_sectors = u.dirty_sectors,
470 .cached_sectors = u.cached_sectors,
471 .io_time[READ] = u.read_time,
472 .io_time[WRITE] = u.write_time,
476 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
480 static noinline struct bkey_i_alloc_v4 *
481 __bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
483 struct bkey_i_alloc_v4 *ret;
485 ret = bch2_trans_kmalloc(trans, max(bkey_bytes(k.k), sizeof(struct bkey_i_alloc_v4)));
489 if (k.k->type == KEY_TYPE_alloc_v4) {
492 bkey_reassemble(&ret->k_i, k);
494 src = alloc_v4_backpointers(&ret->v);
495 SET_BCH_ALLOC_V4_BACKPOINTERS_START(&ret->v, BCH_ALLOC_V4_U64s);
496 dst = alloc_v4_backpointers(&ret->v);
499 memset(src, 0, dst - src);
501 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(&ret->v, 0);
502 set_alloc_v4_u64s(ret);
504 bkey_alloc_v4_init(&ret->k_i);
506 bch2_alloc_to_v4(k, &ret->v);
511 static inline struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut_inlined(struct btree_trans *trans, struct bkey_s_c k)
513 struct bkey_s_c_alloc_v4 a;
515 if (likely(k.k->type == KEY_TYPE_alloc_v4) &&
516 ((a = bkey_s_c_to_alloc_v4(k), true) &&
517 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v) == 0))
518 return bch2_bkey_make_mut_noupdate_typed(trans, k, alloc_v4);
520 return __bch2_alloc_to_v4_mut(trans, k);
523 struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
525 return bch2_alloc_to_v4_mut_inlined(trans, k);
528 struct bkey_i_alloc_v4 *
529 bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter,
533 struct bkey_i_alloc_v4 *a;
536 k = bch2_bkey_get_iter(trans, iter, BTREE_ID_alloc, pos,
537 BTREE_ITER_WITH_UPDATES|
544 a = bch2_alloc_to_v4_mut_inlined(trans, k);
545 ret = PTR_ERR_OR_ZERO(a);
550 bch2_trans_iter_exit(trans, iter);
554 static struct bpos alloc_gens_pos(struct bpos pos, unsigned *offset)
556 *offset = pos.offset & KEY_TYPE_BUCKET_GENS_MASK;
558 pos.offset >>= KEY_TYPE_BUCKET_GENS_BITS;
562 static struct bpos bucket_gens_pos_to_alloc(struct bpos pos, unsigned offset)
564 pos.offset <<= KEY_TYPE_BUCKET_GENS_BITS;
565 pos.offset += offset;
569 static unsigned alloc_gen(struct bkey_s_c k, unsigned offset)
571 return k.k->type == KEY_TYPE_bucket_gens
572 ? bkey_s_c_to_bucket_gens(k).v->gens[offset]
576 int bch2_bucket_gens_invalid(const struct bch_fs *c, struct bkey_s_c k,
577 enum bkey_invalid_flags flags,
578 struct printbuf *err)
580 if (bkey_val_bytes(k.k) != sizeof(struct bch_bucket_gens)) {
581 prt_printf(err, "bad val size (%lu != %zu)",
582 bkey_val_bytes(k.k), sizeof(struct bch_bucket_gens));
583 return -BCH_ERR_invalid_bkey;
589 void bch2_bucket_gens_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
591 struct bkey_s_c_bucket_gens g = bkey_s_c_to_bucket_gens(k);
594 for (i = 0; i < ARRAY_SIZE(g.v->gens); i++) {
597 prt_printf(out, "%u", g.v->gens[i]);
601 int bch2_bucket_gens_init(struct bch_fs *c)
603 struct btree_trans trans;
604 struct btree_iter iter;
606 struct bch_alloc_v4 a;
607 struct bkey_i_bucket_gens g;
608 bool have_bucket_gens_key = false;
614 bch2_trans_init(&trans, c, 0, 0);
616 for_each_btree_key(&trans, iter, BTREE_ID_alloc, POS_MIN,
617 BTREE_ITER_PREFETCH, k, ret) {
619 * Not a fsck error because this is checked/repaired by
620 * bch2_check_alloc_key() which runs later:
622 if (!bch2_dev_bucket_exists(c, k.k->p))
625 gen = bch2_alloc_to_v4(k, &a)->gen;
626 pos = alloc_gens_pos(iter.pos, &offset);
628 if (have_bucket_gens_key && bkey_cmp(iter.pos, pos)) {
629 ret = commit_do(&trans, NULL, NULL,
631 BTREE_INSERT_LAZY_RW,
632 __bch2_btree_insert(&trans, BTREE_ID_bucket_gens, &g.k_i, 0));
635 have_bucket_gens_key = false;
638 if (!have_bucket_gens_key) {
639 bkey_bucket_gens_init(&g.k_i);
641 have_bucket_gens_key = true;
644 g.v.gens[offset] = gen;
646 bch2_trans_iter_exit(&trans, &iter);
648 if (have_bucket_gens_key && !ret)
649 ret = commit_do(&trans, NULL, NULL,
651 BTREE_INSERT_LAZY_RW,
652 __bch2_btree_insert(&trans, BTREE_ID_bucket_gens, &g.k_i, 0));
654 bch2_trans_exit(&trans);
661 int bch2_alloc_read(struct bch_fs *c)
663 struct btree_trans trans;
664 struct btree_iter iter;
669 down_read(&c->gc_lock);
670 bch2_trans_init(&trans, c, 0, 0);
672 if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_bucket_gens) {
673 const struct bch_bucket_gens *g;
676 for_each_btree_key(&trans, iter, BTREE_ID_bucket_gens, POS_MIN,
677 BTREE_ITER_PREFETCH, k, ret) {
678 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
679 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
681 if (k.k->type != KEY_TYPE_bucket_gens)
684 g = bkey_s_c_to_bucket_gens(k).v;
687 * Not a fsck error because this is checked/repaired by
688 * bch2_check_alloc_key() which runs later:
690 if (!bch2_dev_exists2(c, k.k->p.inode))
693 ca = bch_dev_bkey_exists(c, k.k->p.inode);
695 for (b = max_t(u64, ca->mi.first_bucket, start);
696 b < min_t(u64, ca->mi.nbuckets, end);
698 *bucket_gen(ca, b) = g->gens[b & KEY_TYPE_BUCKET_GENS_MASK];
700 bch2_trans_iter_exit(&trans, &iter);
702 struct bch_alloc_v4 a;
704 for_each_btree_key(&trans, iter, BTREE_ID_alloc, POS_MIN,
705 BTREE_ITER_PREFETCH, k, ret) {
707 * Not a fsck error because this is checked/repaired by
708 * bch2_check_alloc_key() which runs later:
710 if (!bch2_dev_bucket_exists(c, k.k->p))
713 ca = bch_dev_bkey_exists(c, k.k->p.inode);
715 *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen;
717 bch2_trans_iter_exit(&trans, &iter);
720 bch2_trans_exit(&trans);
721 up_read(&c->gc_lock);
729 /* Free space/discard btree: */
731 static int bch2_bucket_do_index(struct btree_trans *trans,
732 struct bkey_s_c alloc_k,
733 const struct bch_alloc_v4 *a,
736 struct bch_fs *c = trans->c;
737 struct bch_dev *ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
738 struct btree_iter iter;
742 enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted;
743 enum bch_bkey_type new_type = set ? KEY_TYPE_set : KEY_TYPE_deleted;
744 struct printbuf buf = PRINTBUF;
747 if (a->data_type != BCH_DATA_free &&
748 a->data_type != BCH_DATA_need_discard)
751 k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k));
756 k->k.type = new_type;
758 switch (a->data_type) {
760 btree = BTREE_ID_freespace;
761 k->k.p = alloc_freespace_pos(alloc_k.k->p, *a);
762 bch2_key_resize(&k->k, 1);
764 case BCH_DATA_need_discard:
765 btree = BTREE_ID_need_discard;
766 k->k.p = alloc_k.k->p;
772 old = bch2_bkey_get_iter(trans, &iter, btree,
773 bkey_start_pos(&k->k),
779 if (ca->mi.freespace_initialized &&
780 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info &&
781 bch2_trans_inconsistent_on(old.k->type != old_type, trans,
782 "incorrect key when %s %s:%llu:%llu:0 (got %s should be %s)\n"
784 set ? "setting" : "clearing",
785 bch2_btree_ids[btree],
788 bch2_bkey_types[old.k->type],
789 bch2_bkey_types[old_type],
790 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
795 ret = bch2_trans_update(trans, &iter, k, 0);
797 bch2_trans_iter_exit(trans, &iter);
802 static noinline int bch2_bucket_gen_update(struct btree_trans *trans,
803 struct bpos bucket, u8 gen)
805 struct btree_iter iter;
807 struct bpos pos = alloc_gens_pos(bucket, &offset);
808 struct bkey_i_bucket_gens *g;
812 g = bch2_trans_kmalloc(trans, sizeof(*g));
813 ret = PTR_ERR_OR_ZERO(g);
817 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_bucket_gens, pos,
819 BTREE_ITER_WITH_UPDATES);
824 if (k.k->type != KEY_TYPE_bucket_gens) {
825 bkey_bucket_gens_init(&g->k_i);
828 bkey_reassemble(&g->k_i, k);
831 g->v.gens[offset] = gen;
833 ret = bch2_trans_update(trans, &iter, &g->k_i, 0);
834 bch2_trans_iter_exit(trans, &iter);
838 int bch2_trans_mark_alloc(struct btree_trans *trans,
839 enum btree_id btree_id, unsigned level,
840 struct bkey_s_c old, struct bkey_i *new,
843 struct bch_fs *c = trans->c;
844 struct bch_alloc_v4 old_a_convert, *new_a;
845 const struct bch_alloc_v4 *old_a;
846 u64 old_lru, new_lru;
850 * Deletion only happens in the device removal path, with
851 * BTREE_TRIGGER_NORUN:
853 BUG_ON(new->k.type != KEY_TYPE_alloc_v4);
855 old_a = bch2_alloc_to_v4(old, &old_a_convert);
856 new_a = &bkey_i_to_alloc_v4(new)->v;
858 new_a->data_type = alloc_data_type(*new_a, new_a->data_type);
860 if (new_a->dirty_sectors > old_a->dirty_sectors ||
861 new_a->cached_sectors > old_a->cached_sectors) {
862 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
863 new_a->io_time[WRITE]= max_t(u64, 1, atomic64_read(&c->io_clock[WRITE].now));
864 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true);
865 SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true);
868 if (data_type_is_empty(new_a->data_type) &&
869 BCH_ALLOC_V4_NEED_INC_GEN(new_a) &&
870 !bch2_bucket_is_open_safe(c, new->k.p.inode, new->k.p.offset)) {
872 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false);
875 if (old_a->data_type != new_a->data_type ||
876 (new_a->data_type == BCH_DATA_free &&
877 alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) {
878 ret = bch2_bucket_do_index(trans, old, old_a, false) ?:
879 bch2_bucket_do_index(trans, bkey_i_to_s_c(new), new_a, true);
884 if (new_a->data_type == BCH_DATA_cached &&
885 !new_a->io_time[READ])
886 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
888 old_lru = alloc_lru_idx_read(*old_a);
889 new_lru = alloc_lru_idx_read(*new_a);
891 if (old_lru != new_lru) {
892 ret = bch2_lru_change(trans, new->k.p.inode,
893 bucket_to_u64(new->k.p),
899 new_a->fragmentation_lru = alloc_lru_idx_fragmentation(*new_a,
900 bch_dev_bkey_exists(c, new->k.p.inode));
902 if (old_a->fragmentation_lru != new_a->fragmentation_lru) {
903 ret = bch2_lru_change(trans,
904 BCH_LRU_FRAGMENTATION_START,
905 bucket_to_u64(new->k.p),
906 old_a->fragmentation_lru, new_a->fragmentation_lru);
911 if (old_a->gen != new_a->gen) {
912 ret = bch2_bucket_gen_update(trans, new->k.p, new_a->gen);
921 * This synthesizes deleted extents for holes, similar to BTREE_ITER_SLOTS for
922 * extents style btrees, but works on non-extents btrees:
924 static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos end, struct bkey *hole)
926 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
934 struct btree_iter iter2;
937 bch2_trans_copy_iter(&iter2, iter);
939 if (!bpos_eq(iter->path->l[0].b->key.k.p, SPOS_MAX))
940 end = bkey_min(end, bpos_nosnap_successor(iter->path->l[0].b->key.k.p));
942 end = bkey_min(end, POS(iter->pos.inode, iter->pos.offset + U32_MAX - 1));
945 * btree node min/max is a closed interval, upto takes a half
948 k = bch2_btree_iter_peek_upto(&iter2, end);
950 bch2_trans_iter_exit(iter->trans, &iter2);
952 BUG_ON(next.offset >= iter->pos.offset + U32_MAX);
960 bch2_key_resize(hole, next.offset - iter->pos.offset);
961 return (struct bkey_s_c) { hole, NULL };
965 static bool next_bucket(struct bch_fs *c, struct bpos *bucket)
970 if (bch2_dev_bucket_exists(c, *bucket))
973 if (bch2_dev_exists2(c, bucket->inode)) {
974 ca = bch_dev_bkey_exists(c, bucket->inode);
976 if (bucket->offset < ca->mi.first_bucket) {
977 bucket->offset = ca->mi.first_bucket;
986 iter = bucket->inode;
987 ca = __bch2_next_dev(c, &iter, NULL);
989 *bucket = POS(ca->dev_idx, ca->mi.first_bucket);
995 static struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, struct bkey *hole)
997 struct bch_fs *c = iter->trans->c;
1000 k = bch2_get_key_or_hole(iter, POS_MAX, hole);
1005 struct bpos bucket = bkey_start_pos(k.k);
1007 if (!bch2_dev_bucket_exists(c, bucket)) {
1008 if (!next_bucket(c, &bucket))
1009 return bkey_s_c_null;
1011 bch2_btree_iter_set_pos(iter, bucket);
1015 if (!bch2_dev_bucket_exists(c, k.k->p)) {
1016 struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode);
1018 bch2_key_resize(hole, ca->mi.nbuckets - bucket.offset);
1025 static noinline_for_stack
1026 int bch2_check_alloc_key(struct btree_trans *trans,
1027 struct bkey_s_c alloc_k,
1028 struct btree_iter *alloc_iter,
1029 struct btree_iter *discard_iter,
1030 struct btree_iter *freespace_iter,
1031 struct btree_iter *bucket_gens_iter)
1033 struct bch_fs *c = trans->c;
1035 struct bch_alloc_v4 a_convert;
1036 const struct bch_alloc_v4 *a;
1037 unsigned discard_key_type, freespace_key_type;
1038 unsigned gens_offset;
1040 struct printbuf buf = PRINTBUF;
1043 if (fsck_err_on(!bch2_dev_bucket_exists(c, alloc_k.k->p), c,
1044 "alloc key for invalid device:bucket %llu:%llu",
1045 alloc_k.k->p.inode, alloc_k.k->p.offset))
1046 return bch2_btree_delete_at(trans, alloc_iter, 0);
1048 ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
1049 if (!ca->mi.freespace_initialized)
1052 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1054 discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0;
1055 bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p);
1056 k = bch2_btree_iter_peek_slot(discard_iter);
1061 if (k.k->type != discard_key_type &&
1062 (c->opts.reconstruct_alloc ||
1063 fsck_err(c, "incorrect key in need_discard btree (got %s should be %s)\n"
1065 bch2_bkey_types[k.k->type],
1066 bch2_bkey_types[discard_key_type],
1067 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1068 struct bkey_i *update =
1069 bch2_trans_kmalloc(trans, sizeof(*update));
1071 ret = PTR_ERR_OR_ZERO(update);
1075 bkey_init(&update->k);
1076 update->k.type = discard_key_type;
1077 update->k.p = discard_iter->pos;
1079 ret = bch2_trans_update(trans, discard_iter, update, 0);
1084 freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0;
1085 bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a));
1086 k = bch2_btree_iter_peek_slot(freespace_iter);
1091 if (k.k->type != freespace_key_type &&
1092 (c->opts.reconstruct_alloc ||
1093 fsck_err(c, "incorrect key in freespace btree (got %s should be %s)\n"
1095 bch2_bkey_types[k.k->type],
1096 bch2_bkey_types[freespace_key_type],
1097 (printbuf_reset(&buf),
1098 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1099 struct bkey_i *update =
1100 bch2_trans_kmalloc(trans, sizeof(*update));
1102 ret = PTR_ERR_OR_ZERO(update);
1106 bkey_init(&update->k);
1107 update->k.type = freespace_key_type;
1108 update->k.p = freespace_iter->pos;
1109 bch2_key_resize(&update->k, 1);
1111 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1116 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(alloc_k.k->p, &gens_offset));
1117 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1122 if (a->gen != alloc_gen(k, gens_offset) &&
1123 (c->opts.reconstruct_alloc ||
1124 fsck_err(c, "incorrect gen in bucket_gens btree (got %u should be %u)\n"
1126 alloc_gen(k, gens_offset), a->gen,
1127 (printbuf_reset(&buf),
1128 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1129 struct bkey_i_bucket_gens *g =
1130 bch2_trans_kmalloc(trans, sizeof(*g));
1132 ret = PTR_ERR_OR_ZERO(g);
1136 if (k.k->type == KEY_TYPE_bucket_gens) {
1137 bkey_reassemble(&g->k_i, k);
1139 bkey_bucket_gens_init(&g->k_i);
1140 g->k.p = alloc_gens_pos(alloc_k.k->p, &gens_offset);
1143 g->v.gens[gens_offset] = a->gen;
1145 ret = bch2_trans_update(trans, bucket_gens_iter, &g->k_i, 0);
1151 printbuf_exit(&buf);
1155 static noinline_for_stack
1156 int bch2_check_alloc_hole_freespace(struct btree_trans *trans,
1159 struct btree_iter *freespace_iter)
1161 struct bch_fs *c = trans->c;
1164 struct printbuf buf = PRINTBUF;
1167 ca = bch_dev_bkey_exists(c, start.inode);
1168 if (!ca->mi.freespace_initialized)
1171 bch2_btree_iter_set_pos(freespace_iter, start);
1173 k = bch2_btree_iter_peek_slot(freespace_iter);
1178 *end = bkey_min(k.k->p, *end);
1180 if (k.k->type != KEY_TYPE_set &&
1181 (c->opts.reconstruct_alloc ||
1182 fsck_err(c, "hole in alloc btree missing in freespace btree\n"
1183 " device %llu buckets %llu-%llu",
1184 freespace_iter->pos.inode,
1185 freespace_iter->pos.offset,
1187 struct bkey_i *update =
1188 bch2_trans_kmalloc(trans, sizeof(*update));
1190 ret = PTR_ERR_OR_ZERO(update);
1194 bkey_init(&update->k);
1195 update->k.type = KEY_TYPE_set;
1196 update->k.p = freespace_iter->pos;
1197 bch2_key_resize(&update->k,
1198 min_t(u64, U32_MAX, end->offset -
1199 freespace_iter->pos.offset));
1201 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1207 printbuf_exit(&buf);
1211 static noinline_for_stack
1212 int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans,
1215 struct btree_iter *bucket_gens_iter)
1217 struct bch_fs *c = trans->c;
1219 struct printbuf buf = PRINTBUF;
1220 unsigned i, gens_offset, gens_end_offset;
1223 if (c->sb.version < bcachefs_metadata_version_bucket_gens)
1226 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(start, &gens_offset));
1228 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1233 if (bkey_cmp(alloc_gens_pos(start, &gens_offset),
1234 alloc_gens_pos(*end, &gens_end_offset)))
1235 gens_end_offset = KEY_TYPE_BUCKET_GENS_NR;
1237 if (k.k->type == KEY_TYPE_bucket_gens) {
1238 struct bkey_i_bucket_gens g;
1239 bool need_update = false;
1241 bkey_reassemble(&g.k_i, k);
1243 for (i = gens_offset; i < gens_end_offset; i++) {
1244 if (fsck_err_on(g.v.gens[i], c,
1245 "hole in alloc btree at %llu:%llu with nonzero gen in bucket_gens btree (%u)",
1246 bucket_gens_pos_to_alloc(k.k->p, i).inode,
1247 bucket_gens_pos_to_alloc(k.k->p, i).offset,
1255 struct bkey_i *k = bch2_trans_kmalloc(trans, sizeof(g));
1257 ret = PTR_ERR_OR_ZERO(k);
1261 memcpy(k, &g, sizeof(g));
1263 ret = bch2_trans_update(trans, bucket_gens_iter, k, 0);
1269 *end = bkey_min(*end, bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0));
1272 printbuf_exit(&buf);
1276 static noinline_for_stack int __bch2_check_discard_freespace_key(struct btree_trans *trans,
1277 struct btree_iter *iter)
1279 struct bch_fs *c = trans->c;
1280 struct btree_iter alloc_iter;
1281 struct bkey_s_c alloc_k;
1282 struct bch_alloc_v4 a_convert;
1283 const struct bch_alloc_v4 *a;
1286 enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard
1287 ? BCH_DATA_need_discard
1289 struct printbuf buf = PRINTBUF;
1293 pos.offset &= ~(~0ULL << 56);
1294 genbits = iter->pos.offset & (~0ULL << 56);
1296 alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc, pos, 0);
1297 ret = bkey_err(alloc_k);
1301 if (fsck_err_on(!bch2_dev_bucket_exists(c, pos), c,
1302 "entry in %s btree for nonexistant dev:bucket %llu:%llu",
1303 bch2_btree_ids[iter->btree_id], pos.inode, pos.offset))
1306 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1308 if (fsck_err_on(a->data_type != state ||
1309 (state == BCH_DATA_free &&
1310 genbits != alloc_freespace_genbits(*a)), c,
1311 "%s\n incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)",
1312 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
1313 bch2_btree_ids[iter->btree_id],
1316 a->data_type == state,
1317 genbits >> 56, alloc_freespace_genbits(*a) >> 56))
1321 set_btree_iter_dontneed(&alloc_iter);
1322 bch2_trans_iter_exit(trans, &alloc_iter);
1323 printbuf_exit(&buf);
1326 ret = bch2_btree_delete_extent_at(trans, iter,
1327 iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0) ?:
1328 bch2_trans_commit(trans, NULL, NULL,
1329 BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW);
1333 static int bch2_check_discard_freespace_key(struct btree_trans *trans,
1334 struct btree_iter *iter,
1337 if (!btree_node_type_is_extents(iter->btree_id)) {
1338 return __bch2_check_discard_freespace_key(trans, iter);
1342 while (!bkey_eq(iter->pos, end) &&
1343 !(ret = btree_trans_too_many_iters(trans) ?:
1344 __bch2_check_discard_freespace_key(trans, iter)))
1345 bch2_btree_iter_set_pos(iter, bpos_nosnap_successor(iter->pos));
1352 * We've already checked that generation numbers in the bucket_gens btree are
1353 * valid for buckets that exist; this just checks for keys for nonexistent
1356 static noinline_for_stack
1357 int bch2_check_bucket_gens_key(struct btree_trans *trans,
1358 struct btree_iter *iter,
1361 struct bch_fs *c = trans->c;
1362 struct bkey_i_bucket_gens g;
1364 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
1365 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
1367 bool need_update = false, dev_exists;
1368 struct printbuf buf = PRINTBUF;
1371 BUG_ON(k.k->type != KEY_TYPE_bucket_gens);
1372 bkey_reassemble(&g.k_i, k);
1374 /* if no bch_dev, skip out whether we repair or not */
1375 dev_exists = bch2_dev_exists2(c, k.k->p.inode);
1377 if (fsck_err_on(!dev_exists, c,
1378 "bucket_gens key for invalid device:\n %s",
1379 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1380 ret = bch2_btree_delete_at(trans, iter, 0);
1385 ca = bch_dev_bkey_exists(c, k.k->p.inode);
1386 if (fsck_err_on(end <= ca->mi.first_bucket ||
1387 start >= ca->mi.nbuckets, c,
1388 "bucket_gens key for invalid buckets:\n %s",
1389 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1390 ret = bch2_btree_delete_at(trans, iter, 0);
1394 for (b = start; b < ca->mi.first_bucket; b++)
1395 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1396 "bucket_gens key has nonzero gen for invalid bucket")) {
1397 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1401 for (b = ca->mi.nbuckets; b < end; b++)
1402 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1403 "bucket_gens key has nonzero gen for invalid bucket")) {
1404 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1411 k = bch2_trans_kmalloc(trans, sizeof(g));
1412 ret = PTR_ERR_OR_ZERO(k);
1416 memcpy(k, &g, sizeof(g));
1417 ret = bch2_trans_update(trans, iter, k, 0);
1421 printbuf_exit(&buf);
1425 int bch2_check_alloc_info(struct bch_fs *c)
1427 struct btree_trans trans;
1428 struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter;
1433 bch2_trans_init(&trans, c, 0, 0);
1435 bch2_trans_iter_init(&trans, &iter, BTREE_ID_alloc, POS_MIN,
1436 BTREE_ITER_PREFETCH);
1437 bch2_trans_iter_init(&trans, &discard_iter, BTREE_ID_need_discard, POS_MIN,
1438 BTREE_ITER_PREFETCH);
1439 bch2_trans_iter_init(&trans, &freespace_iter, BTREE_ID_freespace, POS_MIN,
1440 BTREE_ITER_PREFETCH);
1441 bch2_trans_iter_init(&trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN,
1442 BTREE_ITER_PREFETCH);
1447 bch2_trans_begin(&trans);
1449 k = bch2_get_key_or_real_bucket_hole(&iter, &hole);
1458 next = bpos_nosnap_successor(k.k->p);
1460 ret = bch2_check_alloc_key(&trans,
1470 ret = bch2_check_alloc_hole_freespace(&trans,
1471 bkey_start_pos(k.k),
1474 bch2_check_alloc_hole_bucket_gens(&trans,
1475 bkey_start_pos(k.k),
1482 ret = bch2_trans_commit(&trans, NULL, NULL,
1483 BTREE_INSERT_NOFAIL|
1484 BTREE_INSERT_LAZY_RW);
1488 bch2_btree_iter_set_pos(&iter, next);
1490 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1495 bch2_trans_iter_exit(&trans, &bucket_gens_iter);
1496 bch2_trans_iter_exit(&trans, &freespace_iter);
1497 bch2_trans_iter_exit(&trans, &discard_iter);
1498 bch2_trans_iter_exit(&trans, &iter);
1503 ret = for_each_btree_key2(&trans, iter,
1504 BTREE_ID_need_discard, POS_MIN,
1505 BTREE_ITER_PREFETCH, k,
1506 bch2_check_discard_freespace_key(&trans, &iter, k.k->p)) ?:
1507 for_each_btree_key2(&trans, iter,
1508 BTREE_ID_freespace, POS_MIN,
1509 BTREE_ITER_PREFETCH, k,
1510 bch2_check_discard_freespace_key(&trans, &iter, k.k->p)) ?:
1511 for_each_btree_key_commit(&trans, iter,
1512 BTREE_ID_bucket_gens, POS_MIN,
1513 BTREE_ITER_PREFETCH, k,
1514 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
1515 bch2_check_bucket_gens_key(&trans, &iter, k));
1517 bch2_trans_exit(&trans);
1523 static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
1524 struct btree_iter *alloc_iter)
1526 struct bch_fs *c = trans->c;
1527 struct btree_iter lru_iter;
1528 struct bch_alloc_v4 a_convert;
1529 const struct bch_alloc_v4 *a;
1530 struct bkey_s_c alloc_k, lru_k;
1531 struct printbuf buf = PRINTBUF;
1534 alloc_k = bch2_btree_iter_peek(alloc_iter);
1538 ret = bkey_err(alloc_k);
1542 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1544 if (a->data_type != BCH_DATA_cached)
1547 lru_k = bch2_bkey_get_iter(trans, &lru_iter, BTREE_ID_lru,
1548 lru_pos(alloc_k.k->p.inode,
1549 bucket_to_u64(alloc_k.k->p),
1550 a->io_time[READ]), 0);
1551 ret = bkey_err(lru_k);
1555 if (fsck_err_on(!a->io_time[READ], c,
1556 "cached bucket with read_time 0\n"
1558 (printbuf_reset(&buf),
1559 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)) ||
1560 fsck_err_on(lru_k.k->type != KEY_TYPE_set, c,
1561 "missing lru entry\n"
1563 (printbuf_reset(&buf),
1564 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1565 u64 read_time = a->io_time[READ] ?:
1566 atomic64_read(&c->io_clock[READ].now);
1568 ret = bch2_lru_set(trans,
1570 bucket_to_u64(alloc_k.k->p),
1575 if (a->io_time[READ] != read_time) {
1576 struct bkey_i_alloc_v4 *a_mut =
1577 bch2_alloc_to_v4_mut(trans, alloc_k);
1578 ret = PTR_ERR_OR_ZERO(a_mut);
1582 a_mut->v.io_time[READ] = read_time;
1583 ret = bch2_trans_update(trans, alloc_iter,
1584 &a_mut->k_i, BTREE_TRIGGER_NORUN);
1591 bch2_trans_iter_exit(trans, &lru_iter);
1592 printbuf_exit(&buf);
1596 int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
1598 struct btree_iter iter;
1602 ret = bch2_trans_run(c,
1603 for_each_btree_key_commit(&trans, iter, BTREE_ID_alloc,
1604 POS_MIN, BTREE_ITER_PREFETCH, k,
1605 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
1606 bch2_check_alloc_to_lru_ref(&trans, &iter)));
1612 static int bch2_discard_one_bucket(struct btree_trans *trans,
1613 struct btree_iter *need_discard_iter,
1614 struct bpos *discard_pos_done,
1617 u64 *need_journal_commit,
1620 struct bch_fs *c = trans->c;
1621 struct bpos pos = need_discard_iter->pos;
1622 struct btree_iter iter = { NULL };
1625 struct bkey_i_alloc_v4 *a;
1626 struct printbuf buf = PRINTBUF;
1629 ca = bch_dev_bkey_exists(c, pos.inode);
1630 if (!percpu_ref_tryget(&ca->io_ref)) {
1631 bch2_btree_iter_set_pos(need_discard_iter, POS(pos.inode + 1, 0));
1635 if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) {
1640 if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
1641 c->journal.flushed_seq_ondisk,
1642 pos.inode, pos.offset)) {
1643 (*need_journal_commit)++;
1647 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc,
1648 need_discard_iter->pos,
1654 a = bch2_alloc_to_v4_mut(trans, k);
1655 ret = PTR_ERR_OR_ZERO(a);
1659 if (BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
1661 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1665 if (a->v.journal_seq > c->journal.flushed_seq_ondisk) {
1666 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1667 bch2_trans_inconsistent(trans,
1668 "clearing need_discard but journal_seq %llu > flushed_seq %llu\n"
1671 c->journal.flushed_seq_ondisk,
1672 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1678 if (a->v.data_type != BCH_DATA_need_discard) {
1679 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1680 bch2_trans_inconsistent(trans,
1681 "bucket incorrectly set in need_discard btree\n"
1683 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1690 if (!bkey_eq(*discard_pos_done, iter.pos) &&
1691 ca->mi.discard && !c->opts.nochanges) {
1693 * This works without any other locks because this is the only
1694 * thread that removes items from the need_discard tree
1696 bch2_trans_unlock(trans);
1697 blkdev_issue_discard(ca->disk_sb.bdev,
1698 k.k->p.offset * ca->mi.bucket_size,
1701 *discard_pos_done = iter.pos;
1703 ret = bch2_trans_relock_notrace(trans);
1708 SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
1709 a->v.data_type = alloc_data_type(a->v, a->v.data_type);
1711 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
1712 bch2_trans_commit(trans, NULL, NULL,
1713 BCH_WATERMARK_btree|
1714 BTREE_INSERT_NOFAIL);
1718 this_cpu_inc(c->counters[BCH_COUNTER_bucket_discard]);
1722 bch2_trans_iter_exit(trans, &iter);
1723 percpu_ref_put(&ca->io_ref);
1724 printbuf_exit(&buf);
1728 static void bch2_do_discards_work(struct work_struct *work)
1730 struct bch_fs *c = container_of(work, struct bch_fs, discard_work);
1731 struct btree_trans trans;
1732 struct btree_iter iter;
1734 u64 seen = 0, open = 0, need_journal_commit = 0, discarded = 0;
1735 struct bpos discard_pos_done = POS_MAX;
1738 bch2_trans_init(&trans, c, 0, 0);
1741 * We're doing the commit in bch2_discard_one_bucket instead of using
1742 * for_each_btree_key_commit() so that we can increment counters after
1743 * successful commit:
1745 ret = for_each_btree_key2(&trans, iter,
1746 BTREE_ID_need_discard, POS_MIN, 0, k,
1747 bch2_discard_one_bucket(&trans, &iter, &discard_pos_done,
1750 &need_journal_commit,
1753 bch2_trans_exit(&trans);
1755 if (need_journal_commit * 2 > seen)
1756 bch2_journal_flush_async(&c->journal, NULL);
1758 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1760 trace_discard_buckets(c, seen, open, need_journal_commit, discarded,
1764 void bch2_do_discards(struct bch_fs *c)
1766 if (bch2_write_ref_tryget(c, BCH_WRITE_REF_discard) &&
1767 !queue_work(c->write_ref_wq, &c->discard_work))
1768 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1771 static int invalidate_one_bucket(struct btree_trans *trans,
1772 struct btree_iter *lru_iter,
1773 struct bkey_s_c lru_k,
1774 s64 *nr_to_invalidate)
1776 struct bch_fs *c = trans->c;
1777 struct btree_iter alloc_iter = { NULL };
1778 struct bkey_i_alloc_v4 *a = NULL;
1779 struct printbuf buf = PRINTBUF;
1780 struct bpos bucket = u64_to_bucket(lru_k.k->p.offset);
1781 unsigned cached_sectors;
1784 if (*nr_to_invalidate <= 0)
1787 if (!bch2_dev_bucket_exists(c, bucket)) {
1788 prt_str(&buf, "lru entry points to invalid bucket");
1792 if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset))
1795 a = bch2_trans_start_alloc_update(trans, &alloc_iter, bucket);
1796 ret = PTR_ERR_OR_ZERO(a);
1800 /* We expect harmless races here due to the btree write buffer: */
1801 if (lru_pos_time(lru_iter->pos) != alloc_lru_idx_read(a->v))
1804 BUG_ON(a->v.data_type != BCH_DATA_cached);
1806 if (!a->v.cached_sectors)
1807 bch_err(c, "invalidating empty bucket, confused");
1809 cached_sectors = a->v.cached_sectors;
1811 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1814 a->v.dirty_sectors = 0;
1815 a->v.cached_sectors = 0;
1816 a->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now);
1817 a->v.io_time[WRITE] = atomic64_read(&c->io_clock[WRITE].now);
1819 ret = bch2_trans_update(trans, &alloc_iter, &a->k_i,
1820 BTREE_TRIGGER_BUCKET_INVALIDATE) ?:
1821 bch2_trans_commit(trans, NULL, NULL,
1822 BCH_WATERMARK_btree|
1823 BTREE_INSERT_NOFAIL);
1827 trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors);
1828 --*nr_to_invalidate;
1830 bch2_trans_iter_exit(trans, &alloc_iter);
1831 printbuf_exit(&buf);
1834 prt_str(&buf, "\n lru key: ");
1835 bch2_bkey_val_to_text(&buf, c, lru_k);
1837 prt_str(&buf, "\n lru entry: ");
1838 bch2_lru_pos_to_text(&buf, lru_iter->pos);
1840 prt_str(&buf, "\n alloc key: ");
1842 bch2_bpos_to_text(&buf, bucket);
1844 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i));
1846 bch_err(c, "%s", buf.buf);
1847 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_lrus) {
1848 bch2_inconsistent_error(c);
1855 static void bch2_do_invalidates_work(struct work_struct *work)
1857 struct bch_fs *c = container_of(work, struct bch_fs, invalidate_work);
1859 struct btree_trans trans;
1860 struct btree_iter iter;
1865 bch2_trans_init(&trans, c, 0, 0);
1867 ret = bch2_btree_write_buffer_flush(&trans);
1871 for_each_member_device(ca, c, i) {
1872 s64 nr_to_invalidate =
1873 should_invalidate_buckets(ca, bch2_dev_usage_read(ca));
1875 ret = for_each_btree_key2_upto(&trans, iter, BTREE_ID_lru,
1876 lru_pos(ca->dev_idx, 0, 0),
1877 lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX),
1878 BTREE_ITER_INTENT, k,
1879 invalidate_one_bucket(&trans, &iter, k, &nr_to_invalidate));
1882 percpu_ref_put(&ca->ref);
1887 bch2_trans_exit(&trans);
1888 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1891 void bch2_do_invalidates(struct bch_fs *c)
1893 if (bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate) &&
1894 !queue_work(c->write_ref_wq, &c->invalidate_work))
1895 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1898 static int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca,
1899 unsigned long *last_updated)
1901 struct btree_trans trans;
1902 struct btree_iter iter;
1905 struct bpos end = POS(ca->dev_idx, ca->mi.nbuckets);
1906 struct bch_member *m;
1909 bch2_trans_init(&trans, c, 0, 0);
1911 bch2_trans_iter_init(&trans, &iter, BTREE_ID_alloc,
1912 POS(ca->dev_idx, ca->mi.first_bucket),
1913 BTREE_ITER_PREFETCH);
1915 * Scan the alloc btree for every bucket on @ca, and add buckets to the
1916 * freespace/need_discard/need_gc_gens btrees as needed:
1919 if (*last_updated + HZ * 10 < jiffies) {
1920 bch_info(ca, "%s: currently at %llu/%llu",
1921 __func__, iter.pos.offset, ca->mi.nbuckets);
1922 *last_updated = jiffies;
1925 bch2_trans_begin(&trans);
1927 if (bkey_ge(iter.pos, end)) {
1932 k = bch2_get_key_or_hole(&iter, end, &hole);
1939 * We process live keys in the alloc btree one at a
1942 struct bch_alloc_v4 a_convert;
1943 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert);
1945 ret = bch2_bucket_do_index(&trans, k, a, true) ?:
1946 bch2_trans_commit(&trans, NULL, NULL,
1947 BTREE_INSERT_LAZY_RW|
1948 BTREE_INSERT_NOFAIL);
1952 bch2_btree_iter_advance(&iter);
1954 struct bkey_i *freespace;
1956 freespace = bch2_trans_kmalloc(&trans, sizeof(*freespace));
1957 ret = PTR_ERR_OR_ZERO(freespace);
1961 bkey_init(&freespace->k);
1962 freespace->k.type = KEY_TYPE_set;
1963 freespace->k.p = k.k->p;
1964 freespace->k.size = k.k->size;
1966 ret = __bch2_btree_insert(&trans, BTREE_ID_freespace, freespace, 0) ?:
1967 bch2_trans_commit(&trans, NULL, NULL,
1968 BTREE_INSERT_LAZY_RW|
1969 BTREE_INSERT_NOFAIL);
1973 bch2_btree_iter_set_pos(&iter, k.k->p);
1976 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1982 bch2_trans_iter_exit(&trans, &iter);
1983 bch2_trans_exit(&trans);
1986 bch_err(ca, "error initializing free space: %s", bch2_err_str(ret));
1990 mutex_lock(&c->sb_lock);
1991 m = bch2_sb_get_members(c->disk_sb.sb)->members + ca->dev_idx;
1992 SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true);
1993 mutex_unlock(&c->sb_lock);
1998 int bch2_fs_freespace_init(struct bch_fs *c)
2003 bool doing_init = false;
2004 unsigned long last_updated = jiffies;
2007 * We can crash during the device add path, so we need to check this on
2011 for_each_member_device(ca, c, i) {
2012 if (ca->mi.freespace_initialized)
2016 bch_info(c, "initializing freespace");
2020 ret = bch2_dev_freespace_init(c, ca, &last_updated);
2022 percpu_ref_put(&ca->ref);
2029 mutex_lock(&c->sb_lock);
2030 bch2_write_super(c);
2031 mutex_unlock(&c->sb_lock);
2032 bch_verbose(c, "done initializing freespace");
2038 /* Bucket IO clocks: */
2040 int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
2041 size_t bucket_nr, int rw)
2043 struct bch_fs *c = trans->c;
2044 struct btree_iter iter;
2045 struct bkey_i_alloc_v4 *a;
2049 a = bch2_trans_start_alloc_update(trans, &iter, POS(dev, bucket_nr));
2050 ret = PTR_ERR_OR_ZERO(a);
2054 now = atomic64_read(&c->io_clock[rw].now);
2055 if (a->v.io_time[rw] == now)
2058 a->v.io_time[rw] = now;
2060 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
2061 bch2_trans_commit(trans, NULL, NULL, 0);
2063 bch2_trans_iter_exit(trans, &iter);
2067 /* Startup/shutdown (ro/rw): */
2069 void bch2_recalc_capacity(struct bch_fs *c)
2072 u64 capacity = 0, reserved_sectors = 0, gc_reserve;
2073 unsigned bucket_size_max = 0;
2074 unsigned long ra_pages = 0;
2077 lockdep_assert_held(&c->state_lock);
2079 for_each_online_member(ca, c, i) {
2080 struct backing_dev_info *bdi = ca->disk_sb.bdev->bd_disk->bdi;
2082 ra_pages += bdi->ra_pages;
2085 bch2_set_ra_pages(c, ra_pages);
2087 for_each_rw_member(ca, c, i) {
2088 u64 dev_reserve = 0;
2091 * We need to reserve buckets (from the number
2092 * of currently available buckets) against
2093 * foreground writes so that mainly copygc can
2094 * make forward progress.
2096 * We need enough to refill the various reserves
2097 * from scratch - copygc will use its entire
2098 * reserve all at once, then run against when
2099 * its reserve is refilled (from the formerly
2100 * available buckets).
2102 * This reserve is just used when considering if
2103 * allocations for foreground writes must wait -
2104 * not -ENOSPC calculations.
2107 dev_reserve += ca->nr_btree_reserve * 2;
2108 dev_reserve += ca->mi.nbuckets >> 6; /* copygc reserve */
2110 dev_reserve += 1; /* btree write point */
2111 dev_reserve += 1; /* copygc write point */
2112 dev_reserve += 1; /* rebalance write point */
2114 dev_reserve *= ca->mi.bucket_size;
2116 capacity += bucket_to_sector(ca, ca->mi.nbuckets -
2117 ca->mi.first_bucket);
2119 reserved_sectors += dev_reserve * 2;
2121 bucket_size_max = max_t(unsigned, bucket_size_max,
2122 ca->mi.bucket_size);
2125 gc_reserve = c->opts.gc_reserve_bytes
2126 ? c->opts.gc_reserve_bytes >> 9
2127 : div64_u64(capacity * c->opts.gc_reserve_percent, 100);
2129 reserved_sectors = max(gc_reserve, reserved_sectors);
2131 reserved_sectors = min(reserved_sectors, capacity);
2133 c->capacity = capacity - reserved_sectors;
2135 c->bucket_size_max = bucket_size_max;
2137 /* Wake up case someone was waiting for buckets */
2138 closure_wake_up(&c->freelist_wait);
2141 static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca)
2143 struct open_bucket *ob;
2146 for (ob = c->open_buckets;
2147 ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
2149 spin_lock(&ob->lock);
2150 if (ob->valid && !ob->on_partial_list &&
2151 ob->dev == ca->dev_idx)
2153 spin_unlock(&ob->lock);
2159 /* device goes ro: */
2160 void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca)
2164 /* First, remove device from allocation groups: */
2166 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2167 clear_bit(ca->dev_idx, c->rw_devs[i].d);
2170 * Capacity is calculated based off of devices in allocation groups:
2172 bch2_recalc_capacity(c);
2174 bch2_open_buckets_stop(c, ca, false);
2177 * Wake up threads that were blocked on allocation, so they can notice
2178 * the device can no longer be removed and the capacity has changed:
2180 closure_wake_up(&c->freelist_wait);
2183 * journal_res_get() can block waiting for free space in the journal -
2184 * it needs to notice there may not be devices to allocate from anymore:
2186 wake_up(&c->journal.wait);
2188 /* Now wait for any in flight writes: */
2190 closure_wait_event(&c->open_buckets_wait,
2191 !bch2_dev_has_open_write_point(c, ca));
2194 /* device goes rw: */
2195 void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca)
2199 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2200 if (ca->mi.data_allowed & (1 << i))
2201 set_bit(ca->dev_idx, c->rw_devs[i].d);
2204 void bch2_fs_allocator_background_init(struct bch_fs *c)
2206 spin_lock_init(&c->freelist_lock);
2207 INIT_WORK(&c->discard_work, bch2_do_discards_work);
2208 INIT_WORK(&c->invalidate_work, bch2_do_invalidates_work);