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"
23 #include <linux/kthread.h>
24 #include <linux/math64.h>
25 #include <linux/random.h>
26 #include <linux/rculist.h>
27 #include <linux/rcupdate.h>
28 #include <linux/sched/task.h>
29 #include <linux/sort.h>
30 #include <trace/events/bcachefs.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 unsigned flags, struct printbuf *err)
228 struct bkey_s_c_alloc a = bkey_s_c_to_alloc(k);
230 /* allow for unknown fields */
231 if (bkey_val_u64s(a.k) < bch_alloc_v1_val_u64s(a.v)) {
232 prt_printf(err, "incorrect value size (%zu < %u)",
233 bkey_val_u64s(a.k), bch_alloc_v1_val_u64s(a.v));
234 return -BCH_ERR_invalid_bkey;
240 int bch2_alloc_v2_invalid(const struct bch_fs *c, struct bkey_s_c k,
241 unsigned flags, struct printbuf *err)
243 struct bkey_alloc_unpacked u;
245 if (bch2_alloc_unpack_v2(&u, k)) {
246 prt_printf(err, "unpack error");
247 return -BCH_ERR_invalid_bkey;
253 int bch2_alloc_v3_invalid(const struct bch_fs *c, struct bkey_s_c k,
254 unsigned flags, struct printbuf *err)
256 struct bkey_alloc_unpacked u;
258 if (bch2_alloc_unpack_v3(&u, k)) {
259 prt_printf(err, "unpack error");
260 return -BCH_ERR_invalid_bkey;
266 int bch2_alloc_v4_invalid(const struct bch_fs *c, struct bkey_s_c k,
267 unsigned flags, struct printbuf *err)
269 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
270 int rw = flags & WRITE;
272 if (alloc_v4_u64s(a.v) != bkey_val_u64s(k.k)) {
273 prt_printf(err, "bad val size (%lu != %u)",
274 bkey_val_u64s(k.k), alloc_v4_u64s(a.v));
275 return -BCH_ERR_invalid_bkey;
278 if (!BCH_ALLOC_V4_BACKPOINTERS_START(a.v) &&
279 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v)) {
280 prt_printf(err, "invalid backpointers_start");
281 return -BCH_ERR_invalid_bkey;
285 !(flags & BKEY_INVALID_FROM_JOURNAL) &&
286 test_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags)) {
287 unsigned i, bp_len = 0;
289 for (i = 0; i < BCH_ALLOC_V4_NR_BACKPOINTERS(a.v); i++)
290 bp_len += alloc_v4_backpointers_c(a.v)[i].bucket_len;
292 if (bp_len > a.v->dirty_sectors) {
293 prt_printf(err, "too many backpointers");
294 return -BCH_ERR_invalid_bkey;
299 if (alloc_data_type(*a.v, a.v->data_type) != a.v->data_type) {
300 prt_printf(err, "invalid data type (got %u should be %u)",
301 a.v->data_type, alloc_data_type(*a.v, a.v->data_type));
302 return -BCH_ERR_invalid_bkey;
305 switch (a.v->data_type) {
307 case BCH_DATA_need_gc_gens:
308 case BCH_DATA_need_discard:
309 if (a.v->dirty_sectors ||
310 a.v->cached_sectors ||
312 prt_printf(err, "empty data type free but have data");
313 return -BCH_ERR_invalid_bkey;
317 case BCH_DATA_journal:
320 case BCH_DATA_parity:
321 if (!a.v->dirty_sectors) {
322 prt_printf(err, "data_type %s but dirty_sectors==0",
323 bch2_data_types[a.v->data_type]);
324 return -BCH_ERR_invalid_bkey;
327 case BCH_DATA_cached:
328 if (!a.v->cached_sectors ||
329 a.v->dirty_sectors ||
331 prt_printf(err, "data type inconsistency");
332 return -BCH_ERR_invalid_bkey;
335 if (!a.v->io_time[READ] &&
336 test_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags)) {
337 prt_printf(err, "cached bucket with read_time == 0");
338 return -BCH_ERR_invalid_bkey;
341 case BCH_DATA_stripe:
343 prt_printf(err, "data_type %s but stripe==0",
344 bch2_data_types[a.v->data_type]);
345 return -BCH_ERR_invalid_bkey;
354 static inline u64 swab40(u64 x)
356 return (((x & 0x00000000ffULL) << 32)|
357 ((x & 0x000000ff00ULL) << 16)|
358 ((x & 0x0000ff0000ULL) >> 0)|
359 ((x & 0x00ff000000ULL) >> 16)|
360 ((x & 0xff00000000ULL) >> 32));
363 void bch2_alloc_v4_swab(struct bkey_s k)
365 struct bch_alloc_v4 *a = bkey_s_to_alloc_v4(k).v;
366 struct bch_backpointer *bp, *bps;
368 a->journal_seq = swab64(a->journal_seq);
369 a->flags = swab32(a->flags);
370 a->dirty_sectors = swab32(a->dirty_sectors);
371 a->cached_sectors = swab32(a->cached_sectors);
372 a->io_time[0] = swab64(a->io_time[0]);
373 a->io_time[1] = swab64(a->io_time[1]);
374 a->stripe = swab32(a->stripe);
375 a->nr_external_backpointers = swab32(a->nr_external_backpointers);
377 bps = alloc_v4_backpointers(a);
378 for (bp = bps; bp < bps + BCH_ALLOC_V4_NR_BACKPOINTERS(a); bp++) {
379 bp->bucket_offset = swab40(bp->bucket_offset);
380 bp->bucket_len = swab32(bp->bucket_len);
381 bch2_bpos_swab(&bp->pos);
385 void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
387 struct bch_alloc_v4 _a;
388 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a);
392 printbuf_indent_add(out, 2);
394 prt_printf(out, "gen %u oldest_gen %u data_type %s",
395 a->gen, a->oldest_gen,
396 a->data_type < BCH_DATA_NR
397 ? bch2_data_types[a->data_type]
398 : "(invalid data type)");
400 prt_printf(out, "journal_seq %llu", a->journal_seq);
402 prt_printf(out, "need_discard %llu", BCH_ALLOC_V4_NEED_DISCARD(a));
404 prt_printf(out, "need_inc_gen %llu", BCH_ALLOC_V4_NEED_INC_GEN(a));
406 prt_printf(out, "dirty_sectors %u", a->dirty_sectors);
408 prt_printf(out, "cached_sectors %u", a->cached_sectors);
410 prt_printf(out, "stripe %u", a->stripe);
412 prt_printf(out, "stripe_redundancy %u", a->stripe_redundancy);
414 prt_printf(out, "io_time[READ] %llu", a->io_time[READ]);
416 prt_printf(out, "io_time[WRITE] %llu", a->io_time[WRITE]);
418 prt_printf(out, "fragmentation %llu", a->fragmentation_lru);
420 prt_printf(out, "bp_start %llu", BCH_ALLOC_V4_BACKPOINTERS_START(a));
423 if (BCH_ALLOC_V4_NR_BACKPOINTERS(a)) {
424 struct bkey_s_c_alloc_v4 a_raw = bkey_s_c_to_alloc_v4(k);
425 const struct bch_backpointer *bps = alloc_v4_backpointers_c(a_raw.v);
427 prt_printf(out, "backpointers: %llu", BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v));
428 printbuf_indent_add(out, 2);
430 for (i = 0; i < BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v); i++) {
432 bch2_backpointer_to_text(out, &bps[i]);
435 printbuf_indent_sub(out, 2);
438 printbuf_indent_sub(out, 2);
441 void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out)
443 if (k.k->type == KEY_TYPE_alloc_v4) {
446 *out = *bkey_s_c_to_alloc_v4(k).v;
448 src = alloc_v4_backpointers(out);
449 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
450 dst = alloc_v4_backpointers(out);
453 memset(src, 0, dst - src);
455 struct bkey_alloc_unpacked u = bch2_alloc_unpack(k);
457 *out = (struct bch_alloc_v4) {
458 .journal_seq = u.journal_seq,
459 .flags = u.need_discard,
461 .oldest_gen = u.oldest_gen,
462 .data_type = u.data_type,
463 .stripe_redundancy = u.stripe_redundancy,
464 .dirty_sectors = u.dirty_sectors,
465 .cached_sectors = u.cached_sectors,
466 .io_time[READ] = u.read_time,
467 .io_time[WRITE] = u.write_time,
471 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
475 static noinline struct bkey_i_alloc_v4 *
476 __bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
478 struct bkey_i_alloc_v4 *ret;
479 if (k.k->type == KEY_TYPE_alloc_v4) {
480 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
481 unsigned bytes = sizeof(struct bkey_i_alloc_v4) +
482 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v) *
483 sizeof(struct bch_backpointer);
487 * Reserve space for one more backpointer here:
488 * Not sketchy at doing it this way, nope...
490 ret = bch2_trans_kmalloc(trans, bytes + sizeof(struct bch_backpointer));
494 bkey_reassemble(&ret->k_i, k);
496 src = alloc_v4_backpointers(&ret->v);
497 SET_BCH_ALLOC_V4_BACKPOINTERS_START(&ret->v, BCH_ALLOC_V4_U64s);
498 dst = alloc_v4_backpointers(&ret->v);
500 memmove(dst, src, BCH_ALLOC_V4_NR_BACKPOINTERS(&ret->v) *
501 sizeof(struct bch_backpointer));
503 memset(src, 0, dst - src);
504 set_alloc_v4_u64s(ret);
506 ret = bch2_trans_kmalloc(trans, sizeof(struct bkey_i_alloc_v4) +
507 sizeof(struct bch_backpointer));
511 bkey_alloc_v4_init(&ret->k_i);
513 bch2_alloc_to_v4(k, &ret->v);
518 static inline struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut_inlined(struct btree_trans *trans, struct bkey_s_c k)
520 if (likely(k.k->type == KEY_TYPE_alloc_v4) &&
521 BCH_ALLOC_V4_BACKPOINTERS_START(bkey_s_c_to_alloc_v4(k).v) == BCH_ALLOC_V4_U64s) {
523 * Reserve space for one more backpointer here:
524 * Not sketchy at doing it this way, nope...
526 struct bkey_i_alloc_v4 *ret =
527 bch2_trans_kmalloc_nomemzero(trans, bkey_bytes(k.k) + sizeof(struct bch_backpointer));
529 bkey_reassemble(&ret->k_i, k);
533 return __bch2_alloc_to_v4_mut(trans, k);
536 struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
538 return bch2_alloc_to_v4_mut_inlined(trans, k);
541 struct bkey_i_alloc_v4 *
542 bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter,
546 struct bkey_i_alloc_v4 *a;
549 bch2_trans_iter_init(trans, iter, BTREE_ID_alloc, pos,
550 BTREE_ITER_WITH_UPDATES|
553 k = bch2_btree_iter_peek_slot(iter);
558 a = bch2_alloc_to_v4_mut_inlined(trans, k);
559 ret = PTR_ERR_OR_ZERO(a);
564 bch2_trans_iter_exit(trans, iter);
568 int bch2_alloc_read(struct bch_fs *c)
570 struct btree_trans trans;
571 struct btree_iter iter;
573 struct bch_alloc_v4 a;
577 bch2_trans_init(&trans, c, 0, 0);
579 for_each_btree_key(&trans, iter, BTREE_ID_alloc, POS_MIN,
580 BTREE_ITER_PREFETCH, k, ret) {
582 * Not a fsck error because this is checked/repaired by
583 * bch2_check_alloc_key() which runs later:
585 if (!bch2_dev_bucket_exists(c, k.k->p))
588 ca = bch_dev_bkey_exists(c, k.k->p.inode);
590 *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen;
592 bch2_trans_iter_exit(&trans, &iter);
594 bch2_trans_exit(&trans);
597 bch_err(c, "error reading alloc info: %s", bch2_err_str(ret));
602 static struct bpos alloc_gens_pos(struct bpos pos, unsigned *offset)
604 *offset = pos.offset & KEY_TYPE_BUCKET_GENS_MASK;
606 pos.offset >>= KEY_TYPE_BUCKET_GENS_BITS;
610 static struct bpos bucket_gens_pos_to_alloc(struct bpos pos, unsigned offset)
612 pos.offset <<= KEY_TYPE_BUCKET_GENS_BITS;
613 pos.offset += offset;
617 static unsigned alloc_gen(struct bkey_s_c k, unsigned offset)
619 return k.k->type == KEY_TYPE_bucket_gens
620 ? bkey_s_c_to_bucket_gens(k).v->gens[offset]
624 int bch2_bucket_gens_invalid(const struct bch_fs *c, struct bkey_s_c k,
625 unsigned flags, struct printbuf *err)
627 if (bkey_val_bytes(k.k) != sizeof(struct bch_bucket_gens)) {
628 prt_printf(err, "bad val size (%lu != %zu)",
629 bkey_val_bytes(k.k), sizeof(struct bch_bucket_gens));
630 return -BCH_ERR_invalid_bkey;
636 void bch2_bucket_gens_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
638 struct bkey_s_c_bucket_gens g = bkey_s_c_to_bucket_gens(k);
641 for (i = 0; i < ARRAY_SIZE(g.v->gens); i++) {
644 prt_printf(out, "%u", g.v->gens[i]);
648 int bch2_bucket_gens_init(struct bch_fs *c)
650 struct btree_trans trans;
651 struct btree_iter iter;
653 struct bch_alloc_v4 a;
654 struct bkey_i_bucket_gens g;
655 bool have_bucket_gens_key = false;
661 bch2_trans_init(&trans, c, 0, 0);
663 for_each_btree_key(&trans, iter, BTREE_ID_alloc, POS_MIN,
664 BTREE_ITER_PREFETCH, k, ret) {
666 * Not a fsck error because this is checked/repaired by
667 * bch2_check_alloc_key() which runs later:
669 if (!bch2_dev_bucket_exists(c, k.k->p))
672 gen = bch2_alloc_to_v4(k, &a)->gen;
673 pos = alloc_gens_pos(iter.pos, &offset);
675 if (have_bucket_gens_key && bkey_cmp(iter.pos, pos)) {
676 ret = commit_do(&trans, NULL, NULL,
678 BTREE_INSERT_LAZY_RW,
679 __bch2_btree_insert(&trans, BTREE_ID_bucket_gens, &g.k_i, 0));
682 have_bucket_gens_key = false;
685 if (!have_bucket_gens_key) {
686 bkey_bucket_gens_init(&g.k_i);
688 have_bucket_gens_key = true;
691 g.v.gens[offset] = gen;
693 bch2_trans_iter_exit(&trans, &iter);
695 if (have_bucket_gens_key && !ret)
696 ret = commit_do(&trans, NULL, NULL,
698 BTREE_INSERT_LAZY_RW,
699 __bch2_btree_insert(&trans, BTREE_ID_bucket_gens, &g.k_i, 0));
701 bch2_trans_exit(&trans);
704 bch_err(c, "%s: error %s", __func__, bch2_err_str(ret));
709 int bch2_bucket_gens_read(struct bch_fs *c)
711 struct btree_trans trans;
712 struct btree_iter iter;
714 const struct bch_bucket_gens *g;
719 bch2_trans_init(&trans, c, 0, 0);
721 for_each_btree_key(&trans, iter, BTREE_ID_bucket_gens, POS_MIN,
722 BTREE_ITER_PREFETCH, k, ret) {
723 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
724 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
726 if (k.k->type != KEY_TYPE_bucket_gens)
729 g = bkey_s_c_to_bucket_gens(k).v;
732 * Not a fsck error because this is checked/repaired by
733 * bch2_check_alloc_key() which runs later:
735 if (!bch2_dev_exists2(c, k.k->p.inode))
738 ca = bch_dev_bkey_exists(c, k.k->p.inode);
740 for (b = max_t(u64, ca->mi.first_bucket, start);
741 b < min_t(u64, ca->mi.nbuckets, end);
743 *bucket_gen(ca, b) = g->gens[b & KEY_TYPE_BUCKET_GENS_MASK];
745 bch2_trans_iter_exit(&trans, &iter);
747 bch2_trans_exit(&trans);
750 bch_err(c, "error reading alloc info: %s", bch2_err_str(ret));
755 /* Free space/discard btree: */
757 static int bch2_bucket_do_index(struct btree_trans *trans,
758 struct bkey_s_c alloc_k,
759 const struct bch_alloc_v4 *a,
762 struct bch_fs *c = trans->c;
763 struct bch_dev *ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
764 struct btree_iter iter;
768 enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted;
769 enum bch_bkey_type new_type = set ? KEY_TYPE_set : KEY_TYPE_deleted;
770 struct printbuf buf = PRINTBUF;
773 if (a->data_type != BCH_DATA_free &&
774 a->data_type != BCH_DATA_need_discard)
777 k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k));
782 k->k.type = new_type;
784 switch (a->data_type) {
786 btree = BTREE_ID_freespace;
787 k->k.p = alloc_freespace_pos(alloc_k.k->p, *a);
788 bch2_key_resize(&k->k, 1);
790 case BCH_DATA_need_discard:
791 btree = BTREE_ID_need_discard;
792 k->k.p = alloc_k.k->p;
798 bch2_trans_iter_init(trans, &iter, btree,
799 bkey_start_pos(&k->k),
801 old = bch2_btree_iter_peek_slot(&iter);
806 if (ca->mi.freespace_initialized &&
807 test_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags) &&
808 bch2_trans_inconsistent_on(old.k->type != old_type, trans,
809 "incorrect key when %s %s btree (got %s should be %s)\n"
811 set ? "setting" : "clearing",
812 bch2_btree_ids[btree],
813 bch2_bkey_types[old.k->type],
814 bch2_bkey_types[old_type],
815 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
820 ret = bch2_trans_update(trans, &iter, k, 0);
822 bch2_trans_iter_exit(trans, &iter);
827 static noinline int bch2_bucket_gen_update(struct btree_trans *trans,
828 struct bpos bucket, u8 gen)
830 struct btree_iter iter;
832 struct bpos pos = alloc_gens_pos(bucket, &offset);
833 struct bkey_i_bucket_gens *g;
837 g = bch2_trans_kmalloc(trans, sizeof(*g));
838 ret = PTR_ERR_OR_ZERO(g);
842 bch2_trans_iter_init(trans, &iter, BTREE_ID_bucket_gens, pos,
844 BTREE_ITER_WITH_UPDATES);
845 k = bch2_btree_iter_peek_slot(&iter);
850 if (k.k->type != KEY_TYPE_bucket_gens) {
851 bkey_bucket_gens_init(&g->k_i);
854 bkey_reassemble(&g->k_i, k);
857 g->v.gens[offset] = gen;
859 ret = bch2_trans_update(trans, &iter, &g->k_i, 0);
861 bch2_trans_iter_exit(trans, &iter);
865 int bch2_trans_mark_alloc(struct btree_trans *trans,
866 enum btree_id btree_id, unsigned level,
867 struct bkey_s_c old, struct bkey_i *new,
870 struct bch_fs *c = trans->c;
871 struct bch_alloc_v4 old_a_convert, *new_a;
872 const struct bch_alloc_v4 *old_a;
873 u64 old_lru, new_lru;
877 * Deletion only happens in the device removal path, with
878 * BTREE_TRIGGER_NORUN:
880 BUG_ON(new->k.type != KEY_TYPE_alloc_v4);
882 old_a = bch2_alloc_to_v4(old, &old_a_convert);
883 new_a = &bkey_i_to_alloc_v4(new)->v;
885 new_a->data_type = alloc_data_type(*new_a, new_a->data_type);
887 if (new_a->dirty_sectors > old_a->dirty_sectors ||
888 new_a->cached_sectors > old_a->cached_sectors) {
889 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
890 new_a->io_time[WRITE]= max_t(u64, 1, atomic64_read(&c->io_clock[WRITE].now));
891 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true);
892 SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true);
895 if (data_type_is_empty(new_a->data_type) &&
896 BCH_ALLOC_V4_NEED_INC_GEN(new_a) &&
897 !bch2_bucket_is_open_safe(c, new->k.p.inode, new->k.p.offset)) {
899 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false);
902 if (old_a->data_type != new_a->data_type ||
903 (new_a->data_type == BCH_DATA_free &&
904 alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) {
905 ret = bch2_bucket_do_index(trans, old, old_a, false) ?:
906 bch2_bucket_do_index(trans, bkey_i_to_s_c(new), new_a, true);
911 if (new_a->data_type == BCH_DATA_cached &&
912 !new_a->io_time[READ])
913 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
915 old_lru = alloc_lru_idx_read(*old_a);
916 new_lru = alloc_lru_idx_read(*new_a);
918 if (old_lru != new_lru) {
919 ret = bch2_lru_change(trans, new->k.p.inode,
920 bucket_to_u64(new->k.p),
926 new_a->fragmentation_lru = alloc_lru_idx_fragmentation(*new_a,
927 bch_dev_bkey_exists(c, new->k.p.inode));
929 if (old_a->fragmentation_lru != new_a->fragmentation_lru) {
930 ret = bch2_lru_change(trans,
931 BCH_LRU_FRAGMENTATION_START,
932 bucket_to_u64(new->k.p),
933 old_a->fragmentation_lru, new_a->fragmentation_lru);
938 if (old_a->gen != new_a->gen) {
939 ret = bch2_bucket_gen_update(trans, new->k.p, new_a->gen);
948 * This synthesizes deleted extents for holes, similar to BTREE_ITER_SLOTS for
949 * extents style btrees, but works on non-extents btrees:
951 struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos end, struct bkey *hole)
953 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
961 struct btree_iter iter2;
964 bch2_trans_copy_iter(&iter2, iter);
965 k = bch2_btree_iter_peek_upto(&iter2,
966 bkey_min(bkey_min(end,
967 iter->path->l[0].b->key.k.p),
968 POS(iter->pos.inode, iter->pos.offset + U32_MAX - 1)));
970 bch2_trans_iter_exit(iter->trans, &iter2);
972 BUG_ON(next.offset >= iter->pos.offset + U32_MAX);
980 bch2_key_resize(hole, next.offset - iter->pos.offset);
981 return (struct bkey_s_c) { hole, NULL };
985 static bool next_bucket(struct bch_fs *c, struct bpos *bucket)
990 if (bch2_dev_bucket_exists(c, *bucket))
993 if (bch2_dev_exists2(c, bucket->inode)) {
994 ca = bch_dev_bkey_exists(c, bucket->inode);
996 if (bucket->offset < ca->mi.first_bucket) {
997 bucket->offset = ca->mi.first_bucket;
1006 iter = bucket->inode;
1007 ca = __bch2_next_dev(c, &iter, NULL);
1009 *bucket = POS(ca->dev_idx, ca->mi.first_bucket);
1015 struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, struct bkey *hole)
1017 struct bch_fs *c = iter->trans->c;
1020 k = bch2_get_key_or_hole(iter, POS_MAX, hole);
1025 struct bpos bucket = bkey_start_pos(k.k);
1027 if (!bch2_dev_bucket_exists(c, bucket)) {
1028 if (!next_bucket(c, &bucket))
1029 return bkey_s_c_null;
1031 bch2_btree_iter_set_pos(iter, bucket);
1035 if (!bch2_dev_bucket_exists(c, k.k->p)) {
1036 struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode);
1038 bch2_key_resize(hole, ca->mi.nbuckets - bucket.offset);
1045 static int bch2_check_alloc_key(struct btree_trans *trans,
1046 struct bkey_s_c alloc_k,
1047 struct btree_iter *alloc_iter,
1048 struct btree_iter *discard_iter,
1049 struct btree_iter *freespace_iter,
1050 struct btree_iter *bucket_gens_iter)
1052 struct bch_fs *c = trans->c;
1054 struct bch_alloc_v4 a_convert;
1055 const struct bch_alloc_v4 *a;
1056 unsigned discard_key_type, freespace_key_type;
1057 unsigned gens_offset;
1059 struct printbuf buf = PRINTBUF;
1062 if (fsck_err_on(!bch2_dev_bucket_exists(c, alloc_k.k->p), c,
1063 "alloc key for invalid device:bucket %llu:%llu",
1064 alloc_k.k->p.inode, alloc_k.k->p.offset))
1065 return bch2_btree_delete_at(trans, alloc_iter, 0);
1067 ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
1068 if (!ca->mi.freespace_initialized)
1071 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1073 discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0;
1074 bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p);
1075 k = bch2_btree_iter_peek_slot(discard_iter);
1080 if (k.k->type != discard_key_type &&
1081 (c->opts.reconstruct_alloc ||
1082 fsck_err(c, "incorrect key in need_discard btree (got %s should be %s)\n"
1084 bch2_bkey_types[k.k->type],
1085 bch2_bkey_types[discard_key_type],
1086 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1087 struct bkey_i *update =
1088 bch2_trans_kmalloc(trans, sizeof(*update));
1090 ret = PTR_ERR_OR_ZERO(update);
1094 bkey_init(&update->k);
1095 update->k.type = discard_key_type;
1096 update->k.p = discard_iter->pos;
1098 ret = bch2_trans_update(trans, discard_iter, update, 0);
1103 freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0;
1104 bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a));
1105 k = bch2_btree_iter_peek_slot(freespace_iter);
1110 if (k.k->type != freespace_key_type &&
1111 (c->opts.reconstruct_alloc ||
1112 fsck_err(c, "incorrect key in freespace btree (got %s should be %s)\n"
1114 bch2_bkey_types[k.k->type],
1115 bch2_bkey_types[freespace_key_type],
1116 (printbuf_reset(&buf),
1117 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1118 struct bkey_i *update =
1119 bch2_trans_kmalloc(trans, sizeof(*update));
1121 ret = PTR_ERR_OR_ZERO(update);
1125 bkey_init(&update->k);
1126 update->k.type = freespace_key_type;
1127 update->k.p = freespace_iter->pos;
1128 bch2_key_resize(&update->k, 1);
1130 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1135 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(alloc_k.k->p, &gens_offset));
1136 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1141 if (a->gen != alloc_gen(k, gens_offset) &&
1142 (c->opts.reconstruct_alloc ||
1143 fsck_err(c, "incorrect gen in bucket_gens btree (got %u should be %u)\n"
1145 alloc_gen(k, gens_offset), a->gen,
1146 (printbuf_reset(&buf),
1147 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1148 struct bkey_i_bucket_gens *g =
1149 bch2_trans_kmalloc(trans, sizeof(*g));
1151 ret = PTR_ERR_OR_ZERO(g);
1155 if (k.k->type == KEY_TYPE_bucket_gens) {
1156 bkey_reassemble(&g->k_i, k);
1158 bkey_bucket_gens_init(&g->k_i);
1159 g->k.p = alloc_gens_pos(alloc_k.k->p, &gens_offset);
1162 g->v.gens[gens_offset] = a->gen;
1164 ret = bch2_trans_update(trans, bucket_gens_iter, &g->k_i, 0);
1170 printbuf_exit(&buf);
1174 static int bch2_check_alloc_hole_freespace(struct btree_trans *trans,
1177 struct btree_iter *freespace_iter)
1179 struct bch_fs *c = trans->c;
1182 struct printbuf buf = PRINTBUF;
1185 ca = bch_dev_bkey_exists(c, start.inode);
1186 if (!ca->mi.freespace_initialized)
1189 bch2_btree_iter_set_pos(freespace_iter, start);
1191 k = bch2_btree_iter_peek_slot(freespace_iter);
1196 *end = bkey_min(k.k->p, *end);
1198 if (k.k->type != KEY_TYPE_set &&
1199 (c->opts.reconstruct_alloc ||
1200 fsck_err(c, "hole in alloc btree missing in freespace btree\n"
1201 " device %llu buckets %llu-%llu",
1202 freespace_iter->pos.inode,
1203 freespace_iter->pos.offset,
1205 struct bkey_i *update =
1206 bch2_trans_kmalloc(trans, sizeof(*update));
1208 ret = PTR_ERR_OR_ZERO(update);
1212 bkey_init(&update->k);
1213 update->k.type = KEY_TYPE_set;
1214 update->k.p = freespace_iter->pos;
1215 bch2_key_resize(&update->k,
1216 min_t(u64, U32_MAX, end->offset -
1217 freespace_iter->pos.offset));
1219 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1225 printbuf_exit(&buf);
1229 static int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans,
1232 struct btree_iter *bucket_gens_iter)
1234 struct bch_fs *c = trans->c;
1236 struct printbuf buf = PRINTBUF;
1237 unsigned i, gens_offset, gens_end_offset;
1240 if (c->sb.version < bcachefs_metadata_version_bucket_gens &&
1241 !c->opts.version_upgrade)
1244 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(start, &gens_offset));
1246 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1251 if (bkey_cmp(alloc_gens_pos(start, &gens_offset),
1252 alloc_gens_pos(*end, &gens_end_offset)))
1253 gens_end_offset = KEY_TYPE_BUCKET_GENS_NR;
1255 if (k.k->type == KEY_TYPE_bucket_gens) {
1256 struct bkey_i_bucket_gens g;
1257 bool need_update = false;
1259 bkey_reassemble(&g.k_i, k);
1261 for (i = gens_offset; i < gens_end_offset; i++) {
1262 if (fsck_err_on(g.v.gens[i], c,
1263 "hole in alloc btree at %llu:%llu with nonzero gen in bucket_gens btree (%u)",
1264 bucket_gens_pos_to_alloc(k.k->p, i).inode,
1265 bucket_gens_pos_to_alloc(k.k->p, i).offset,
1273 struct bkey_i *k = bch2_trans_kmalloc(trans, sizeof(g));
1275 ret = PTR_ERR_OR_ZERO(k);
1279 memcpy(k, &g, sizeof(g));
1281 ret = bch2_trans_update(trans, bucket_gens_iter, k, 0);
1287 *end = bkey_min(*end, bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0));
1290 printbuf_exit(&buf);
1294 static int bch2_check_discard_freespace_key(struct btree_trans *trans,
1295 struct btree_iter *iter)
1297 struct bch_fs *c = trans->c;
1298 struct btree_iter alloc_iter;
1299 struct bkey_s_c alloc_k;
1300 struct bch_alloc_v4 a_convert;
1301 const struct bch_alloc_v4 *a;
1304 enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard
1305 ? BCH_DATA_need_discard
1307 struct printbuf buf = PRINTBUF;
1311 pos.offset &= ~(~0ULL << 56);
1312 genbits = iter->pos.offset & (~0ULL << 56);
1314 bch2_trans_iter_init(trans, &alloc_iter, BTREE_ID_alloc, pos, 0);
1316 if (fsck_err_on(!bch2_dev_bucket_exists(c, pos), c,
1317 "entry in %s btree for nonexistant dev:bucket %llu:%llu",
1318 bch2_btree_ids[iter->btree_id], pos.inode, pos.offset))
1321 alloc_k = bch2_btree_iter_peek_slot(&alloc_iter);
1322 ret = bkey_err(alloc_k);
1326 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1328 if (fsck_err_on(a->data_type != state ||
1329 (state == BCH_DATA_free &&
1330 genbits != alloc_freespace_genbits(*a)), c,
1331 "%s\n incorrectly set in %s index (free %u, genbits %llu should be %llu)",
1332 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
1333 bch2_btree_ids[iter->btree_id],
1334 a->data_type == state,
1335 genbits >> 56, alloc_freespace_genbits(*a) >> 56))
1340 bch2_trans_iter_exit(trans, &alloc_iter);
1341 printbuf_exit(&buf);
1344 ret = bch2_btree_delete_extent_at(trans, iter,
1345 iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0);
1350 * We've already checked that generation numbers in the bucket_gens btree are
1351 * valid for buckets that exist; this just checks for keys for nonexistent
1354 static int bch2_check_bucket_gens_key(struct btree_trans *trans,
1355 struct btree_iter *iter,
1358 struct bch_fs *c = trans->c;
1359 struct bkey_i_bucket_gens g;
1361 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
1362 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
1364 bool need_update = false;
1365 struct printbuf buf = PRINTBUF;
1368 BUG_ON(k.k->type != KEY_TYPE_bucket_gens);
1369 bkey_reassemble(&g.k_i, k);
1371 if (fsck_err_on(!bch2_dev_exists2(c, k.k->p.inode), c,
1372 "bucket_gens key for invalid device:\n %s",
1373 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1374 ret = bch2_btree_delete_at(trans, iter, 0);
1378 ca = bch_dev_bkey_exists(c, k.k->p.inode);
1379 if (fsck_err_on(end <= ca->mi.first_bucket ||
1380 start >= ca->mi.nbuckets, c,
1381 "bucket_gens key for invalid buckets:\n %s",
1382 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1383 ret = bch2_btree_delete_at(trans, iter, 0);
1387 for (b = start; b < ca->mi.first_bucket; b++)
1388 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1389 "bucket_gens key has nonzero gen for invalid bucket")) {
1390 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1394 for (b = ca->mi.nbuckets; b < end; 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;
1404 k = bch2_trans_kmalloc(trans, sizeof(g));
1405 ret = PTR_ERR_OR_ZERO(k);
1409 memcpy(k, &g, sizeof(g));
1410 ret = bch2_trans_update(trans, iter, k, 0);
1414 printbuf_exit(&buf);
1418 int bch2_check_alloc_info(struct bch_fs *c)
1420 struct btree_trans trans;
1421 struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter;
1426 bch2_trans_init(&trans, c, 0, 0);
1428 bch2_trans_iter_init(&trans, &iter, BTREE_ID_alloc, POS_MIN,
1429 BTREE_ITER_PREFETCH);
1430 bch2_trans_iter_init(&trans, &discard_iter, BTREE_ID_need_discard, POS_MIN,
1431 BTREE_ITER_PREFETCH);
1432 bch2_trans_iter_init(&trans, &freespace_iter, BTREE_ID_freespace, POS_MIN,
1433 BTREE_ITER_PREFETCH);
1434 bch2_trans_iter_init(&trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN,
1435 BTREE_ITER_PREFETCH);
1440 bch2_trans_begin(&trans);
1442 k = bch2_get_key_or_real_bucket_hole(&iter, &hole);
1451 next = bpos_nosnap_successor(k.k->p);
1453 ret = bch2_check_alloc_key(&trans,
1463 ret = bch2_check_alloc_hole_freespace(&trans,
1464 bkey_start_pos(k.k),
1467 bch2_check_alloc_hole_bucket_gens(&trans,
1468 bkey_start_pos(k.k),
1475 ret = bch2_trans_commit(&trans, NULL, NULL,
1476 BTREE_INSERT_NOFAIL|
1477 BTREE_INSERT_LAZY_RW);
1481 bch2_btree_iter_set_pos(&iter, next);
1483 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1488 bch2_trans_iter_exit(&trans, &bucket_gens_iter);
1489 bch2_trans_iter_exit(&trans, &freespace_iter);
1490 bch2_trans_iter_exit(&trans, &discard_iter);
1491 bch2_trans_iter_exit(&trans, &iter);
1496 ret = for_each_btree_key_commit(&trans, iter,
1497 BTREE_ID_need_discard, POS_MIN,
1498 BTREE_ITER_PREFETCH, k,
1499 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
1500 bch2_check_discard_freespace_key(&trans, &iter)) ?:
1501 for_each_btree_key_commit(&trans, iter,
1502 BTREE_ID_freespace, POS_MIN,
1503 BTREE_ITER_PREFETCH, k,
1504 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
1505 bch2_check_discard_freespace_key(&trans, &iter)) ?:
1506 for_each_btree_key_commit(&trans, iter,
1507 BTREE_ID_bucket_gens, POS_MIN,
1508 BTREE_ITER_PREFETCH, k,
1509 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
1510 bch2_check_bucket_gens_key(&trans, &iter, k));
1512 bch2_trans_exit(&trans);
1513 return ret < 0 ? ret : 0;
1516 static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
1517 struct btree_iter *alloc_iter)
1519 struct bch_fs *c = trans->c;
1520 struct btree_iter lru_iter;
1521 struct bch_alloc_v4 a_convert;
1522 const struct bch_alloc_v4 *a;
1523 struct bkey_s_c alloc_k, k;
1524 struct printbuf buf = PRINTBUF;
1527 alloc_k = bch2_btree_iter_peek(alloc_iter);
1531 ret = bkey_err(alloc_k);
1535 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1537 if (a->data_type != BCH_DATA_cached)
1540 bch2_trans_iter_init(trans, &lru_iter, BTREE_ID_lru,
1541 lru_pos(alloc_k.k->p.inode,
1542 bucket_to_u64(alloc_k.k->p),
1543 a->io_time[READ]), 0);
1544 k = bch2_btree_iter_peek_slot(&lru_iter);
1549 if (fsck_err_on(!a->io_time[READ], c,
1550 "cached bucket with read_time 0\n"
1552 (printbuf_reset(&buf),
1553 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)) ||
1554 fsck_err_on(k.k->type != KEY_TYPE_set, c,
1555 "missing lru entry\n"
1557 (printbuf_reset(&buf),
1558 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1559 u64 read_time = a->io_time[READ] ?:
1560 atomic64_read(&c->io_clock[READ].now);
1562 ret = bch2_lru_set(trans,
1564 bucket_to_u64(alloc_k.k->p),
1569 if (a->io_time[READ] != read_time) {
1570 struct bkey_i_alloc_v4 *a_mut =
1571 bch2_alloc_to_v4_mut(trans, alloc_k);
1572 ret = PTR_ERR_OR_ZERO(a_mut);
1576 a_mut->v.io_time[READ] = read_time;
1577 ret = bch2_trans_update(trans, alloc_iter,
1578 &a_mut->k_i, BTREE_TRIGGER_NORUN);
1585 bch2_trans_iter_exit(trans, &lru_iter);
1586 printbuf_exit(&buf);
1590 int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
1592 struct btree_trans trans;
1593 struct btree_iter iter;
1597 bch2_trans_init(&trans, c, 0, 0);
1599 for_each_btree_key_commit(&trans, iter, BTREE_ID_alloc,
1600 POS_MIN, BTREE_ITER_PREFETCH, k,
1601 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
1602 bch2_check_alloc_to_lru_ref(&trans, &iter));
1604 bch2_trans_exit(&trans);
1605 return ret < 0 ? ret : 0;
1608 static int bch2_discard_one_bucket(struct btree_trans *trans,
1609 struct btree_iter *need_discard_iter,
1610 struct bpos *discard_pos_done,
1613 u64 *need_journal_commit,
1616 struct bch_fs *c = trans->c;
1617 struct bpos pos = need_discard_iter->pos;
1618 struct btree_iter iter = { NULL };
1621 struct bkey_i_alloc_v4 *a;
1622 struct printbuf buf = PRINTBUF;
1625 ca = bch_dev_bkey_exists(c, pos.inode);
1626 if (!percpu_ref_tryget(&ca->io_ref)) {
1627 bch2_btree_iter_set_pos(need_discard_iter, POS(pos.inode + 1, 0));
1631 if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) {
1636 if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
1637 c->journal.flushed_seq_ondisk,
1638 pos.inode, pos.offset)) {
1639 (*need_journal_commit)++;
1643 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
1644 need_discard_iter->pos,
1646 k = bch2_btree_iter_peek_slot(&iter);
1651 a = bch2_alloc_to_v4_mut(trans, k);
1652 ret = PTR_ERR_OR_ZERO(a);
1656 if (BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
1658 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1662 if (a->v.journal_seq > c->journal.flushed_seq_ondisk) {
1663 if (test_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags)) {
1664 bch2_trans_inconsistent(trans,
1665 "clearing need_discard but journal_seq %llu > flushed_seq %llu\n"
1668 c->journal.flushed_seq_ondisk,
1669 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1675 if (a->v.data_type != BCH_DATA_need_discard) {
1676 if (test_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags)) {
1677 bch2_trans_inconsistent(trans,
1678 "bucket incorrectly set in need_discard btree\n"
1680 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1687 if (!bkey_eq(*discard_pos_done, iter.pos) &&
1688 ca->mi.discard && !c->opts.nochanges) {
1690 * This works without any other locks because this is the only
1691 * thread that removes items from the need_discard tree
1693 bch2_trans_unlock(trans);
1694 blkdev_issue_discard(ca->disk_sb.bdev,
1695 k.k->p.offset * ca->mi.bucket_size,
1698 *discard_pos_done = iter.pos;
1700 ret = bch2_trans_relock_notrace(trans);
1705 SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
1706 a->v.data_type = alloc_data_type(a->v, a->v.data_type);
1708 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
1709 bch2_trans_commit(trans, NULL, NULL,
1710 BTREE_INSERT_USE_RESERVE|BTREE_INSERT_NOFAIL);
1714 this_cpu_inc(c->counters[BCH_COUNTER_bucket_discard]);
1718 bch2_trans_iter_exit(trans, &iter);
1719 percpu_ref_put(&ca->io_ref);
1720 printbuf_exit(&buf);
1724 static void bch2_do_discards_work(struct work_struct *work)
1726 struct bch_fs *c = container_of(work, struct bch_fs, discard_work);
1727 struct btree_trans trans;
1728 struct btree_iter iter;
1730 u64 seen = 0, open = 0, need_journal_commit = 0, discarded = 0;
1731 struct bpos discard_pos_done = POS_MAX;
1734 bch2_trans_init(&trans, c, 0, 0);
1737 * We're doing the commit in bch2_discard_one_bucket instead of using
1738 * for_each_btree_key_commit() so that we can increment counters after
1739 * successful commit:
1741 ret = for_each_btree_key2(&trans, iter,
1742 BTREE_ID_need_discard, POS_MIN, 0, k,
1743 bch2_discard_one_bucket(&trans, &iter, &discard_pos_done,
1746 &need_journal_commit,
1749 bch2_trans_exit(&trans);
1751 if (need_journal_commit * 2 > seen)
1752 bch2_journal_flush_async(&c->journal, NULL);
1754 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1756 trace_discard_buckets(c, seen, open, need_journal_commit, discarded,
1760 void bch2_do_discards(struct bch_fs *c)
1762 if (bch2_write_ref_tryget(c, BCH_WRITE_REF_discard) &&
1763 !queue_work(system_long_wq, &c->discard_work))
1764 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1767 static int invalidate_one_bucket(struct btree_trans *trans,
1768 struct btree_iter *lru_iter,
1769 struct bkey_s_c lru_k,
1770 s64 *nr_to_invalidate)
1772 struct bch_fs *c = trans->c;
1773 struct btree_iter alloc_iter = { NULL };
1774 struct bkey_i_alloc_v4 *a = NULL;
1775 struct printbuf buf = PRINTBUF;
1776 struct bpos bucket = u64_to_bucket(lru_k.k->p.offset);
1777 unsigned cached_sectors;
1780 if (*nr_to_invalidate <= 0)
1783 if (!bch2_dev_bucket_exists(c, bucket)) {
1784 prt_str(&buf, "lru entry points to invalid bucket");
1788 if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset))
1791 a = bch2_trans_start_alloc_update(trans, &alloc_iter, bucket);
1792 ret = PTR_ERR_OR_ZERO(a);
1796 /* We expect harmless races here due to the btree write buffer: */
1797 if (lru_pos_time(lru_iter->pos) != alloc_lru_idx_read(a->v))
1800 BUG_ON(a->v.data_type != BCH_DATA_cached);
1802 if (!a->v.cached_sectors)
1803 bch_err(c, "invalidating empty bucket, confused");
1805 cached_sectors = a->v.cached_sectors;
1807 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1810 a->v.dirty_sectors = 0;
1811 a->v.cached_sectors = 0;
1812 a->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now);
1813 a->v.io_time[WRITE] = atomic64_read(&c->io_clock[WRITE].now);
1815 ret = bch2_trans_update(trans, &alloc_iter, &a->k_i,
1816 BTREE_TRIGGER_BUCKET_INVALIDATE) ?:
1817 bch2_trans_commit(trans, NULL, NULL,
1818 BTREE_INSERT_USE_RESERVE|BTREE_INSERT_NOFAIL);
1822 trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors);
1823 --*nr_to_invalidate;
1825 bch2_trans_iter_exit(trans, &alloc_iter);
1826 printbuf_exit(&buf);
1829 prt_str(&buf, "\n lru key: ");
1830 bch2_bkey_val_to_text(&buf, c, lru_k);
1832 prt_str(&buf, "\n lru entry: ");
1833 bch2_lru_pos_to_text(&buf, lru_iter->pos);
1835 prt_str(&buf, "\n alloc key: ");
1837 bch2_bpos_to_text(&buf, bucket);
1839 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i));
1841 bch_err(c, "%s", buf.buf);
1842 if (test_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags)) {
1843 bch2_inconsistent_error(c);
1850 static void bch2_do_invalidates_work(struct work_struct *work)
1852 struct bch_fs *c = container_of(work, struct bch_fs, invalidate_work);
1854 struct btree_trans trans;
1855 struct btree_iter iter;
1860 bch2_trans_init(&trans, c, 0, 0);
1862 ret = bch2_btree_write_buffer_flush(&trans);
1866 for_each_member_device(ca, c, i) {
1867 s64 nr_to_invalidate =
1868 should_invalidate_buckets(ca, bch2_dev_usage_read(ca));
1870 ret = for_each_btree_key2_upto(&trans, iter, BTREE_ID_lru,
1871 lru_pos(ca->dev_idx, 0, 0),
1872 lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX),
1873 BTREE_ITER_INTENT, k,
1874 invalidate_one_bucket(&trans, &iter, k, &nr_to_invalidate));
1877 percpu_ref_put(&ca->ref);
1882 bch2_trans_exit(&trans);
1883 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1886 void bch2_do_invalidates(struct bch_fs *c)
1888 if (bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate) &&
1889 !queue_work(system_long_wq, &c->invalidate_work))
1890 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1893 static int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca)
1895 struct btree_trans trans;
1896 struct btree_iter iter;
1899 struct bpos end = POS(ca->dev_idx, ca->mi.nbuckets);
1900 struct bch_member *m;
1903 bch2_trans_init(&trans, c, 0, 0);
1905 bch2_trans_iter_init(&trans, &iter, BTREE_ID_alloc,
1906 POS(ca->dev_idx, ca->mi.first_bucket),
1907 BTREE_ITER_PREFETCH);
1909 * Scan the alloc btree for every bucket on @ca, and add buckets to the
1910 * freespace/need_discard/need_gc_gens btrees as needed:
1913 bch2_trans_begin(&trans);
1915 if (bkey_ge(iter.pos, end)) {
1920 k = bch2_get_key_or_hole(&iter, end, &hole);
1927 * We process live keys in the alloc btree one at a
1930 struct bch_alloc_v4 a_convert;
1931 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert);
1933 ret = bch2_bucket_do_index(&trans, k, a, true) ?:
1934 bch2_trans_commit(&trans, NULL, NULL,
1935 BTREE_INSERT_LAZY_RW|
1936 BTREE_INSERT_NOFAIL);
1940 bch2_btree_iter_advance(&iter);
1942 struct bkey_i *freespace;
1944 freespace = bch2_trans_kmalloc(&trans, sizeof(*freespace));
1945 ret = PTR_ERR_OR_ZERO(freespace);
1949 bkey_init(&freespace->k);
1950 freespace->k.type = KEY_TYPE_set;
1951 freespace->k.p = k.k->p;
1952 freespace->k.size = k.k->size;
1954 ret = __bch2_btree_insert(&trans, BTREE_ID_freespace, freespace, 0) ?:
1955 bch2_trans_commit(&trans, NULL, NULL,
1956 BTREE_INSERT_LAZY_RW|
1957 BTREE_INSERT_NOFAIL);
1961 bch2_btree_iter_set_pos(&iter, k.k->p);
1964 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1970 bch2_trans_iter_exit(&trans, &iter);
1971 bch2_trans_exit(&trans);
1974 bch_err(ca, "error initializing free space: %s", bch2_err_str(ret));
1978 mutex_lock(&c->sb_lock);
1979 m = bch2_sb_get_members(c->disk_sb.sb)->members + ca->dev_idx;
1980 SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true);
1981 mutex_unlock(&c->sb_lock);
1986 int bch2_fs_freespace_init(struct bch_fs *c)
1991 bool doing_init = false;
1994 * We can crash during the device add path, so we need to check this on
1998 for_each_member_device(ca, c, i) {
1999 if (ca->mi.freespace_initialized)
2003 bch_info(c, "initializing freespace");
2007 ret = bch2_dev_freespace_init(c, ca);
2009 percpu_ref_put(&ca->ref);
2015 mutex_lock(&c->sb_lock);
2016 bch2_write_super(c);
2017 mutex_unlock(&c->sb_lock);
2019 bch_verbose(c, "done initializing freespace");
2025 /* Bucket IO clocks: */
2027 int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
2028 size_t bucket_nr, int rw)
2030 struct bch_fs *c = trans->c;
2031 struct btree_iter iter;
2032 struct bkey_i_alloc_v4 *a;
2036 a = bch2_trans_start_alloc_update(trans, &iter, POS(dev, bucket_nr));
2037 ret = PTR_ERR_OR_ZERO(a);
2041 now = atomic64_read(&c->io_clock[rw].now);
2042 if (a->v.io_time[rw] == now)
2045 a->v.io_time[rw] = now;
2047 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
2048 bch2_trans_commit(trans, NULL, NULL, 0);
2050 bch2_trans_iter_exit(trans, &iter);
2054 /* Startup/shutdown (ro/rw): */
2056 void bch2_recalc_capacity(struct bch_fs *c)
2059 u64 capacity = 0, reserved_sectors = 0, gc_reserve;
2060 unsigned bucket_size_max = 0;
2061 unsigned long ra_pages = 0;
2064 lockdep_assert_held(&c->state_lock);
2066 for_each_online_member(ca, c, i) {
2067 struct backing_dev_info *bdi = ca->disk_sb.bdev->bd_disk->bdi;
2069 ra_pages += bdi->ra_pages;
2072 bch2_set_ra_pages(c, ra_pages);
2074 for_each_rw_member(ca, c, i) {
2075 u64 dev_reserve = 0;
2078 * We need to reserve buckets (from the number
2079 * of currently available buckets) against
2080 * foreground writes so that mainly copygc can
2081 * make forward progress.
2083 * We need enough to refill the various reserves
2084 * from scratch - copygc will use its entire
2085 * reserve all at once, then run against when
2086 * its reserve is refilled (from the formerly
2087 * available buckets).
2089 * This reserve is just used when considering if
2090 * allocations for foreground writes must wait -
2091 * not -ENOSPC calculations.
2094 dev_reserve += ca->nr_btree_reserve * 2;
2095 dev_reserve += ca->mi.nbuckets >> 6; /* copygc reserve */
2097 dev_reserve += 1; /* btree write point */
2098 dev_reserve += 1; /* copygc write point */
2099 dev_reserve += 1; /* rebalance write point */
2101 dev_reserve *= ca->mi.bucket_size;
2103 capacity += bucket_to_sector(ca, ca->mi.nbuckets -
2104 ca->mi.first_bucket);
2106 reserved_sectors += dev_reserve * 2;
2108 bucket_size_max = max_t(unsigned, bucket_size_max,
2109 ca->mi.bucket_size);
2112 gc_reserve = c->opts.gc_reserve_bytes
2113 ? c->opts.gc_reserve_bytes >> 9
2114 : div64_u64(capacity * c->opts.gc_reserve_percent, 100);
2116 reserved_sectors = max(gc_reserve, reserved_sectors);
2118 reserved_sectors = min(reserved_sectors, capacity);
2120 c->capacity = capacity - reserved_sectors;
2122 c->bucket_size_max = bucket_size_max;
2124 /* Wake up case someone was waiting for buckets */
2125 closure_wake_up(&c->freelist_wait);
2128 static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca)
2130 struct open_bucket *ob;
2133 for (ob = c->open_buckets;
2134 ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
2136 spin_lock(&ob->lock);
2137 if (ob->valid && !ob->on_partial_list &&
2138 ob->dev == ca->dev_idx)
2140 spin_unlock(&ob->lock);
2146 /* device goes ro: */
2147 void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca)
2151 /* First, remove device from allocation groups: */
2153 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2154 clear_bit(ca->dev_idx, c->rw_devs[i].d);
2157 * Capacity is calculated based off of devices in allocation groups:
2159 bch2_recalc_capacity(c);
2161 bch2_open_buckets_stop(c, ca, false);
2164 * Wake up threads that were blocked on allocation, so they can notice
2165 * the device can no longer be removed and the capacity has changed:
2167 closure_wake_up(&c->freelist_wait);
2170 * journal_res_get() can block waiting for free space in the journal -
2171 * it needs to notice there may not be devices to allocate from anymore:
2173 wake_up(&c->journal.wait);
2175 /* Now wait for any in flight writes: */
2177 closure_wait_event(&c->open_buckets_wait,
2178 !bch2_dev_has_open_write_point(c, ca));
2181 /* device goes rw: */
2182 void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca)
2186 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2187 if (ca->mi.data_allowed & (1 << i))
2188 set_bit(ca->dev_idx, c->rw_devs[i].d);
2191 void bch2_fs_allocator_background_init(struct bch_fs *c)
2193 spin_lock_init(&c->freelist_lock);
2194 INIT_WORK(&c->discard_work, bch2_do_discards_work);
2195 INIT_WORK(&c->invalidate_work, bch2_do_invalidates_work);