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"
13 #include "buckets_waiting_for_journal.h"
22 #include <linux/kthread.h>
23 #include <linux/math64.h>
24 #include <linux/random.h>
25 #include <linux/rculist.h>
26 #include <linux/rcupdate.h>
27 #include <linux/sched/task.h>
28 #include <linux/sort.h>
29 #include <trace/events/bcachefs.h>
31 /* Persistent alloc info: */
33 static const unsigned BCH_ALLOC_V1_FIELD_BYTES[] = {
34 #define x(name, bits) [BCH_ALLOC_FIELD_V1_##name] = bits / 8,
39 struct bkey_alloc_unpacked {
46 #define x(_name, _bits) u##_bits _name;
51 static inline u64 alloc_field_v1_get(const struct bch_alloc *a,
52 const void **p, unsigned field)
54 unsigned bytes = BCH_ALLOC_V1_FIELD_BYTES[field];
57 if (!(a->fields & (1 << field)))
62 v = *((const u8 *) *p);
81 static inline void alloc_field_v1_put(struct bkey_i_alloc *a, void **p,
82 unsigned field, u64 v)
84 unsigned bytes = BCH_ALLOC_V1_FIELD_BYTES[field];
89 a->v.fields |= 1 << field;
96 *((__le16 *) *p) = cpu_to_le16(v);
99 *((__le32 *) *p) = cpu_to_le32(v);
102 *((__le64 *) *p) = cpu_to_le64(v);
111 static void bch2_alloc_unpack_v1(struct bkey_alloc_unpacked *out,
114 const struct bch_alloc *in = bkey_s_c_to_alloc(k).v;
115 const void *d = in->data;
120 #define x(_name, _bits) out->_name = alloc_field_v1_get(in, &d, idx++);
121 BCH_ALLOC_FIELDS_V1()
125 static int bch2_alloc_unpack_v2(struct bkey_alloc_unpacked *out,
128 struct bkey_s_c_alloc_v2 a = bkey_s_c_to_alloc_v2(k);
129 const u8 *in = a.v->data;
130 const u8 *end = bkey_val_end(a);
131 unsigned fieldnr = 0;
136 out->oldest_gen = a.v->oldest_gen;
137 out->data_type = a.v->data_type;
139 #define x(_name, _bits) \
140 if (fieldnr < a.v->nr_fields) { \
141 ret = bch2_varint_decode_fast(in, end, &v); \
149 if (v != out->_name) \
153 BCH_ALLOC_FIELDS_V2()
158 static int bch2_alloc_unpack_v3(struct bkey_alloc_unpacked *out,
161 struct bkey_s_c_alloc_v3 a = bkey_s_c_to_alloc_v3(k);
162 const u8 *in = a.v->data;
163 const u8 *end = bkey_val_end(a);
164 unsigned fieldnr = 0;
169 out->oldest_gen = a.v->oldest_gen;
170 out->data_type = a.v->data_type;
171 out->need_discard = BCH_ALLOC_V3_NEED_DISCARD(a.v);
172 out->need_inc_gen = BCH_ALLOC_V3_NEED_INC_GEN(a.v);
173 out->journal_seq = le64_to_cpu(a.v->journal_seq);
175 #define x(_name, _bits) \
176 if (fieldnr < a.v->nr_fields) { \
177 ret = bch2_varint_decode_fast(in, end, &v); \
185 if (v != out->_name) \
189 BCH_ALLOC_FIELDS_V2()
194 static struct bkey_alloc_unpacked bch2_alloc_unpack(struct bkey_s_c k)
196 struct bkey_alloc_unpacked ret = { .gen = 0 };
200 bch2_alloc_unpack_v1(&ret, k);
202 case KEY_TYPE_alloc_v2:
203 bch2_alloc_unpack_v2(&ret, k);
205 case KEY_TYPE_alloc_v3:
206 bch2_alloc_unpack_v3(&ret, k);
213 static unsigned bch_alloc_v1_val_u64s(const struct bch_alloc *a)
215 unsigned i, bytes = offsetof(struct bch_alloc, data);
217 for (i = 0; i < ARRAY_SIZE(BCH_ALLOC_V1_FIELD_BYTES); i++)
218 if (a->fields & (1 << i))
219 bytes += BCH_ALLOC_V1_FIELD_BYTES[i];
221 return DIV_ROUND_UP(bytes, sizeof(u64));
224 int bch2_alloc_v1_invalid(const struct bch_fs *c, struct bkey_s_c k,
225 int rw, struct printbuf *err)
227 struct bkey_s_c_alloc a = bkey_s_c_to_alloc(k);
229 /* allow for unknown fields */
230 if (bkey_val_u64s(a.k) < bch_alloc_v1_val_u64s(a.v)) {
231 prt_printf(err, "incorrect value size (%zu < %u)",
232 bkey_val_u64s(a.k), bch_alloc_v1_val_u64s(a.v));
233 return -BCH_ERR_invalid_bkey;
239 int bch2_alloc_v2_invalid(const struct bch_fs *c, struct bkey_s_c k,
240 int rw, struct printbuf *err)
242 struct bkey_alloc_unpacked u;
244 if (bch2_alloc_unpack_v2(&u, k)) {
245 prt_printf(err, "unpack error");
246 return -BCH_ERR_invalid_bkey;
252 int bch2_alloc_v3_invalid(const struct bch_fs *c, struct bkey_s_c k,
253 int rw, struct printbuf *err)
255 struct bkey_alloc_unpacked u;
257 if (bch2_alloc_unpack_v3(&u, k)) {
258 prt_printf(err, "unpack error");
259 return -BCH_ERR_invalid_bkey;
265 int bch2_alloc_v4_invalid(const struct bch_fs *c, struct bkey_s_c k,
266 int rw, struct printbuf *err)
268 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
270 if (alloc_v4_u64s(a.v) != bkey_val_u64s(k.k)) {
271 prt_printf(err, "bad val size (%lu != %u)",
272 bkey_val_u64s(k.k), alloc_v4_u64s(a.v));
273 return -BCH_ERR_invalid_bkey;
276 if (!BCH_ALLOC_V4_BACKPOINTERS_START(a.v) &&
277 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v)) {
278 prt_printf(err, "invalid backpointers_start");
279 return -BCH_ERR_invalid_bkey;
283 * XXX this is wrong, we'll be checking updates that happened from
284 * before BCH_FS_CHECK_BACKPOINTERS_DONE
286 if (rw == WRITE && 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, bch2_data_types[a->data_type]);
397 prt_printf(out, "journal_seq %llu", a->journal_seq);
399 prt_printf(out, "need_discard %llu", BCH_ALLOC_V4_NEED_DISCARD(a));
401 prt_printf(out, "need_inc_gen %llu", BCH_ALLOC_V4_NEED_INC_GEN(a));
403 prt_printf(out, "dirty_sectors %u", a->dirty_sectors);
405 prt_printf(out, "cached_sectors %u", a->cached_sectors);
407 prt_printf(out, "stripe %u", a->stripe);
409 prt_printf(out, "stripe_redundancy %u", a->stripe_redundancy);
411 prt_printf(out, "io_time[READ] %llu", a->io_time[READ]);
413 prt_printf(out, "io_time[WRITE] %llu", a->io_time[WRITE]);
416 if (k.k->type == KEY_TYPE_alloc_v4) {
417 struct bkey_s_c_alloc_v4 a_raw = bkey_s_c_to_alloc_v4(k);
418 const struct bch_backpointer *bps = alloc_v4_backpointers_c(a_raw.v);
420 prt_printf(out, "bp_start %llu", BCH_ALLOC_V4_BACKPOINTERS_START(a_raw.v));
423 prt_printf(out, "backpointers: %llu", BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v));
424 printbuf_indent_add(out, 2);
426 for (i = 0; i < BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v); i++) {
428 bch2_backpointer_to_text(out, &bps[i]);
431 printbuf_indent_sub(out, 2);
434 printbuf_indent_sub(out, 2);
437 void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out)
439 if (k.k->type == KEY_TYPE_alloc_v4) {
442 *out = *bkey_s_c_to_alloc_v4(k).v;
444 src = alloc_v4_backpointers(out);
445 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
446 dst = alloc_v4_backpointers(out);
449 memset(src, 0, dst - src);
451 struct bkey_alloc_unpacked u = bch2_alloc_unpack(k);
453 *out = (struct bch_alloc_v4) {
454 .journal_seq = u.journal_seq,
455 .flags = u.need_discard,
457 .oldest_gen = u.oldest_gen,
458 .data_type = u.data_type,
459 .stripe_redundancy = u.stripe_redundancy,
460 .dirty_sectors = u.dirty_sectors,
461 .cached_sectors = u.cached_sectors,
462 .io_time[READ] = u.read_time,
463 .io_time[WRITE] = u.write_time,
467 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
471 static noinline struct bkey_i_alloc_v4 *
472 __bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
474 struct bkey_i_alloc_v4 *ret;
475 if (k.k->type == KEY_TYPE_alloc_v4) {
476 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
477 unsigned bytes = sizeof(struct bkey_i_alloc_v4) +
478 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v) *
479 sizeof(struct bch_backpointer);
483 * Reserve space for one more backpointer here:
484 * Not sketchy at doing it this way, nope...
486 ret = bch2_trans_kmalloc(trans, bytes + sizeof(struct bch_backpointer));
490 bkey_reassemble(&ret->k_i, k);
492 src = alloc_v4_backpointers(&ret->v);
493 SET_BCH_ALLOC_V4_BACKPOINTERS_START(&ret->v, BCH_ALLOC_V4_U64s);
494 dst = alloc_v4_backpointers(&ret->v);
496 memmove(dst, src, BCH_ALLOC_V4_NR_BACKPOINTERS(&ret->v) *
497 sizeof(struct bch_backpointer));
499 memset(src, 0, dst - src);
500 set_alloc_v4_u64s(ret);
502 ret = bch2_trans_kmalloc(trans, sizeof(struct bkey_i_alloc_v4) +
503 sizeof(struct bch_backpointer));
507 bkey_alloc_v4_init(&ret->k_i);
509 bch2_alloc_to_v4(k, &ret->v);
514 static inline struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut_inlined(struct btree_trans *trans, struct bkey_s_c k)
516 if (likely(k.k->type == KEY_TYPE_alloc_v4) &&
517 BCH_ALLOC_V4_BACKPOINTERS_START(bkey_s_c_to_alloc_v4(k).v) == BCH_ALLOC_V4_U64s) {
519 * Reserve space for one more backpointer here:
520 * Not sketchy at doing it this way, nope...
522 struct bkey_i_alloc_v4 *ret =
523 bch2_trans_kmalloc_nomemzero(trans, bkey_bytes(k.k) + sizeof(struct bch_backpointer));
525 bkey_reassemble(&ret->k_i, k);
529 return __bch2_alloc_to_v4_mut(trans, k);
532 struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
534 return bch2_alloc_to_v4_mut_inlined(trans, k);
537 struct bkey_i_alloc_v4 *
538 bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter,
542 struct bkey_i_alloc_v4 *a;
545 bch2_trans_iter_init(trans, iter, BTREE_ID_alloc, pos,
546 BTREE_ITER_WITH_UPDATES|
549 k = bch2_btree_iter_peek_slot(iter);
554 a = bch2_alloc_to_v4_mut_inlined(trans, k);
555 ret = PTR_ERR_OR_ZERO(a);
560 bch2_trans_iter_exit(trans, iter);
564 int bch2_alloc_read(struct bch_fs *c)
566 struct btree_trans trans;
567 struct btree_iter iter;
569 struct bch_alloc_v4 a;
573 bch2_trans_init(&trans, c, 0, 0);
575 for_each_btree_key(&trans, iter, BTREE_ID_alloc, POS_MIN,
576 BTREE_ITER_PREFETCH, k, ret) {
578 * Not a fsck error because this is checked/repaired by
579 * bch2_check_alloc_key() which runs later:
581 if (!bch2_dev_bucket_exists(c, k.k->p))
584 ca = bch_dev_bkey_exists(c, k.k->p.inode);
586 *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen;
588 bch2_trans_iter_exit(&trans, &iter);
590 bch2_trans_exit(&trans);
593 bch_err(c, "error reading alloc info: %s", bch2_err_str(ret));
598 static struct bpos alloc_gens_pos(struct bpos pos, unsigned *offset)
600 *offset = pos.offset & KEY_TYPE_BUCKET_GENS_MASK;
602 pos.offset >>= KEY_TYPE_BUCKET_GENS_BITS;
606 static struct bpos bucket_gens_pos_to_alloc(struct bpos pos, unsigned offset)
608 pos.offset <<= KEY_TYPE_BUCKET_GENS_BITS;
609 pos.offset += offset;
613 static unsigned alloc_gen(struct bkey_s_c k, unsigned offset)
615 return k.k->type == KEY_TYPE_bucket_gens
616 ? bkey_s_c_to_bucket_gens(k).v->gens[offset]
620 int bch2_bucket_gens_invalid(const struct bch_fs *c, struct bkey_s_c k,
621 int rw, struct printbuf *err)
623 if (bkey_val_bytes(k.k) != sizeof(struct bch_bucket_gens)) {
624 prt_printf(err, "bad val size (%lu != %zu)",
625 bkey_val_bytes(k.k), sizeof(struct bch_bucket_gens));
626 return -BCH_ERR_invalid_bkey;
632 void bch2_bucket_gens_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
634 struct bkey_s_c_bucket_gens g = bkey_s_c_to_bucket_gens(k);
637 for (i = 0; i < ARRAY_SIZE(g.v->gens); i++) {
640 prt_printf(out, "%u", g.v->gens[i]);
644 int bch2_bucket_gens_init(struct bch_fs *c)
646 struct btree_trans trans;
647 struct btree_iter iter;
649 struct bch_alloc_v4 a;
650 struct bkey_i_bucket_gens g;
651 bool have_bucket_gens_key = false;
657 bch2_trans_init(&trans, c, 0, 0);
659 for_each_btree_key(&trans, iter, BTREE_ID_alloc, POS_MIN,
660 BTREE_ITER_PREFETCH, k, ret) {
662 * Not a fsck error because this is checked/repaired by
663 * bch2_check_alloc_key() which runs later:
665 if (!bch2_dev_bucket_exists(c, k.k->p))
668 gen = bch2_alloc_to_v4(k, &a)->gen;
669 pos = alloc_gens_pos(iter.pos, &offset);
671 if (have_bucket_gens_key && bkey_cmp(iter.pos, pos)) {
672 ret = commit_do(&trans, NULL, NULL,
674 BTREE_INSERT_LAZY_RW,
675 __bch2_btree_insert(&trans, BTREE_ID_bucket_gens, &g.k_i));
678 have_bucket_gens_key = false;
681 if (!have_bucket_gens_key) {
682 bkey_bucket_gens_init(&g.k_i);
684 have_bucket_gens_key = true;
687 g.v.gens[offset] = gen;
689 bch2_trans_iter_exit(&trans, &iter);
691 if (have_bucket_gens_key && !ret)
692 ret = commit_do(&trans, NULL, NULL,
694 BTREE_INSERT_LAZY_RW,
695 __bch2_btree_insert(&trans, BTREE_ID_bucket_gens, &g.k_i));
697 bch2_trans_exit(&trans);
700 bch_err(c, "%s: error %s", __func__, bch2_err_str(ret));
705 int bch2_bucket_gens_read(struct bch_fs *c)
707 struct btree_trans trans;
708 struct btree_iter iter;
710 const struct bch_bucket_gens *g;
715 bch2_trans_init(&trans, c, 0, 0);
717 for_each_btree_key(&trans, iter, BTREE_ID_bucket_gens, POS_MIN,
718 BTREE_ITER_PREFETCH, k, ret) {
719 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
720 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
722 if (k.k->type != KEY_TYPE_bucket_gens)
725 g = bkey_s_c_to_bucket_gens(k).v;
728 * Not a fsck error because this is checked/repaired by
729 * bch2_check_alloc_key() which runs later:
731 if (!bch2_dev_exists2(c, k.k->p.inode))
734 ca = bch_dev_bkey_exists(c, k.k->p.inode);
736 for (b = max_t(u64, ca->mi.first_bucket, start);
737 b < min_t(u64, ca->mi.nbuckets, end);
739 *bucket_gen(ca, b) = g->gens[b & KEY_TYPE_BUCKET_GENS_MASK];
741 bch2_trans_iter_exit(&trans, &iter);
743 bch2_trans_exit(&trans);
746 bch_err(c, "error reading alloc info: %s", bch2_err_str(ret));
751 /* Free space/discard btree: */
753 static int bch2_bucket_do_index(struct btree_trans *trans,
754 struct bkey_s_c alloc_k,
755 const struct bch_alloc_v4 *a,
758 struct bch_fs *c = trans->c;
759 struct bch_dev *ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
760 struct btree_iter iter;
764 enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted;
765 enum bch_bkey_type new_type = set ? KEY_TYPE_set : KEY_TYPE_deleted;
766 struct printbuf buf = PRINTBUF;
769 if (a->data_type != BCH_DATA_free &&
770 a->data_type != BCH_DATA_need_discard)
773 k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k));
778 k->k.type = new_type;
780 switch (a->data_type) {
782 btree = BTREE_ID_freespace;
783 k->k.p = alloc_freespace_pos(alloc_k.k->p, *a);
784 bch2_key_resize(&k->k, 1);
786 case BCH_DATA_need_discard:
787 btree = BTREE_ID_need_discard;
788 k->k.p = alloc_k.k->p;
794 bch2_trans_iter_init(trans, &iter, btree,
795 bkey_start_pos(&k->k),
797 old = bch2_btree_iter_peek_slot(&iter);
802 if (ca->mi.freespace_initialized &&
803 bch2_trans_inconsistent_on(old.k->type != old_type, trans,
804 "incorrect key when %s %s btree (got %s should be %s)\n"
806 set ? "setting" : "clearing",
807 bch2_btree_ids[btree],
808 bch2_bkey_types[old.k->type],
809 bch2_bkey_types[old_type],
810 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
815 ret = bch2_trans_update(trans, &iter, k, 0);
817 bch2_trans_iter_exit(trans, &iter);
822 static noinline int bch2_bucket_gen_update(struct btree_trans *trans,
823 struct bpos bucket, u8 gen)
825 struct btree_iter iter;
827 struct bpos pos = alloc_gens_pos(bucket, &offset);
828 struct bkey_i_bucket_gens *g;
832 g = bch2_trans_kmalloc(trans, sizeof(*g));
833 ret = PTR_ERR_OR_ZERO(g);
837 bch2_trans_iter_init(trans, &iter, BTREE_ID_bucket_gens, pos,
839 BTREE_ITER_WITH_UPDATES);
840 k = bch2_btree_iter_peek_slot(&iter);
845 if (k.k->type != KEY_TYPE_bucket_gens) {
846 bkey_bucket_gens_init(&g->k_i);
849 bkey_reassemble(&g->k_i, k);
852 g->v.gens[offset] = gen;
854 ret = bch2_trans_update(trans, &iter, &g->k_i, 0);
856 bch2_trans_iter_exit(trans, &iter);
860 int bch2_trans_mark_alloc(struct btree_trans *trans,
861 enum btree_id btree_id, unsigned level,
862 struct bkey_s_c old, struct bkey_i *new,
865 struct bch_fs *c = trans->c;
866 struct bch_alloc_v4 old_a_convert, *new_a;
867 const struct bch_alloc_v4 *old_a;
868 u64 old_lru, new_lru;
872 * Deletion only happens in the device removal path, with
873 * BTREE_TRIGGER_NORUN:
875 BUG_ON(new->k.type != KEY_TYPE_alloc_v4);
877 old_a = bch2_alloc_to_v4(old, &old_a_convert);
878 new_a = &bkey_i_to_alloc_v4(new)->v;
880 new_a->data_type = alloc_data_type(*new_a, new_a->data_type);
882 if (new_a->dirty_sectors > old_a->dirty_sectors ||
883 new_a->cached_sectors > old_a->cached_sectors) {
884 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
885 new_a->io_time[WRITE]= max_t(u64, 1, atomic64_read(&c->io_clock[WRITE].now));
886 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true);
887 SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true);
890 if (data_type_is_empty(new_a->data_type) &&
891 BCH_ALLOC_V4_NEED_INC_GEN(new_a) &&
892 !bch2_bucket_is_open_safe(c, new->k.p.inode, new->k.p.offset)) {
894 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false);
897 if (old_a->data_type != new_a->data_type ||
898 (new_a->data_type == BCH_DATA_free &&
899 alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) {
900 ret = bch2_bucket_do_index(trans, old, old_a, false) ?:
901 bch2_bucket_do_index(trans, bkey_i_to_s_c(new), new_a, true);
906 if (new_a->data_type == BCH_DATA_cached &&
907 !new_a->io_time[READ])
908 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
910 old_lru = alloc_lru_idx(*old_a);
911 new_lru = alloc_lru_idx(*new_a);
913 if (old_lru != new_lru) {
914 ret = bch2_lru_change(trans, new->k.p.inode, new->k.p.offset,
915 old_lru, &new_lru, old);
919 if (new_a->data_type == BCH_DATA_cached)
920 new_a->io_time[READ] = new_lru;
923 if (old_a->gen != new_a->gen) {
924 ret = bch2_bucket_gen_update(trans, new->k.p, new_a->gen);
933 * This synthesizes deleted extents for holes, similar to BTREE_ITER_SLOTS for
934 * extents style btrees, but works on non-extents btrees:
936 struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos end, struct bkey *hole)
938 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
946 struct btree_iter iter2;
949 bch2_trans_copy_iter(&iter2, iter);
950 k = bch2_btree_iter_peek_upto(&iter2,
951 bkey_min(bkey_min(end,
952 iter->path->l[0].b->key.k.p),
953 POS(iter->pos.inode, iter->pos.offset + U32_MAX - 1)));
955 bch2_trans_iter_exit(iter->trans, &iter2);
957 BUG_ON(next.offset >= iter->pos.offset + U32_MAX);
965 bch2_key_resize(hole, next.offset - iter->pos.offset);
966 return (struct bkey_s_c) { hole, NULL };
970 static bool next_bucket(struct bch_fs *c, struct bpos *bucket)
975 if (bch2_dev_bucket_exists(c, *bucket))
978 if (bch2_dev_exists2(c, bucket->inode)) {
979 ca = bch_dev_bkey_exists(c, bucket->inode);
981 if (bucket->offset < ca->mi.first_bucket) {
982 bucket->offset = ca->mi.first_bucket;
991 iter = bucket->inode;
992 ca = __bch2_next_dev(c, &iter, NULL);
994 bucket->offset = ca->mi.first_bucket;
1000 struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, struct bkey *hole)
1002 struct bch_fs *c = iter->trans->c;
1005 k = bch2_get_key_or_hole(iter, POS_MAX, hole);
1010 struct bpos bucket = bkey_start_pos(k.k);
1012 if (!bch2_dev_bucket_exists(c, bucket)) {
1013 if (!next_bucket(c, &bucket))
1014 return bkey_s_c_null;
1016 bch2_btree_iter_set_pos(iter, bucket);
1020 if (!bch2_dev_bucket_exists(c, k.k->p)) {
1021 struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode);
1023 bch2_key_resize(hole, ca->mi.nbuckets - bucket.offset);
1030 static int bch2_check_alloc_key(struct btree_trans *trans,
1031 struct bkey_s_c alloc_k,
1032 struct btree_iter *alloc_iter,
1033 struct btree_iter *discard_iter,
1034 struct btree_iter *freespace_iter,
1035 struct btree_iter *bucket_gens_iter)
1037 struct bch_fs *c = trans->c;
1039 struct bch_alloc_v4 a_convert;
1040 const struct bch_alloc_v4 *a;
1041 unsigned discard_key_type, freespace_key_type;
1042 unsigned gens_offset;
1044 struct printbuf buf = PRINTBUF;
1047 if (fsck_err_on(!bch2_dev_bucket_exists(c, alloc_k.k->p), c,
1048 "alloc key for invalid device:bucket %llu:%llu",
1049 alloc_k.k->p.inode, alloc_k.k->p.offset))
1050 return bch2_btree_delete_at(trans, alloc_iter, 0);
1052 ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
1053 if (!ca->mi.freespace_initialized)
1056 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1058 discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0;
1059 bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p);
1060 k = bch2_btree_iter_peek_slot(discard_iter);
1065 if (k.k->type != discard_key_type &&
1066 (c->opts.reconstruct_alloc ||
1067 fsck_err(c, "incorrect key in need_discard btree (got %s should be %s)\n"
1069 bch2_bkey_types[k.k->type],
1070 bch2_bkey_types[discard_key_type],
1071 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1072 struct bkey_i *update =
1073 bch2_trans_kmalloc(trans, sizeof(*update));
1075 ret = PTR_ERR_OR_ZERO(update);
1079 bkey_init(&update->k);
1080 update->k.type = discard_key_type;
1081 update->k.p = discard_iter->pos;
1083 ret = bch2_trans_update(trans, discard_iter, update, 0);
1088 freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0;
1089 bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a));
1090 k = bch2_btree_iter_peek_slot(freespace_iter);
1095 if (k.k->type != freespace_key_type &&
1096 (c->opts.reconstruct_alloc ||
1097 fsck_err(c, "incorrect key in freespace btree (got %s should be %s)\n"
1099 bch2_bkey_types[k.k->type],
1100 bch2_bkey_types[freespace_key_type],
1101 (printbuf_reset(&buf),
1102 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1103 struct bkey_i *update =
1104 bch2_trans_kmalloc(trans, sizeof(*update));
1106 ret = PTR_ERR_OR_ZERO(update);
1110 bkey_init(&update->k);
1111 update->k.type = freespace_key_type;
1112 update->k.p = freespace_iter->pos;
1113 bch2_key_resize(&update->k, 1);
1115 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1120 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(alloc_k.k->p, &gens_offset));
1121 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1126 if (a->gen != alloc_gen(k, gens_offset) &&
1127 (c->opts.reconstruct_alloc ||
1128 fsck_err(c, "incorrect gen in bucket_gens btree (got %u should be %u)\n"
1130 alloc_gen(k, gens_offset), a->gen,
1131 (printbuf_reset(&buf),
1132 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1133 struct bkey_i_bucket_gens *g =
1134 bch2_trans_kmalloc(trans, sizeof(*g));
1136 ret = PTR_ERR_OR_ZERO(g);
1140 if (k.k->type == KEY_TYPE_bucket_gens) {
1141 bkey_reassemble(&g->k_i, k);
1143 bkey_bucket_gens_init(&g->k_i);
1144 g->k.p = alloc_gens_pos(alloc_k.k->p, &gens_offset);
1147 g->v.gens[gens_offset] = a->gen;
1149 ret = bch2_trans_update(trans, bucket_gens_iter, &g->k_i, 0);
1155 printbuf_exit(&buf);
1159 static int bch2_check_alloc_hole_freespace(struct btree_trans *trans,
1162 struct btree_iter *freespace_iter)
1164 struct bch_fs *c = trans->c;
1167 struct printbuf buf = PRINTBUF;
1170 ca = bch_dev_bkey_exists(c, start.inode);
1171 if (!ca->mi.freespace_initialized)
1174 bch2_btree_iter_set_pos(freespace_iter, start);
1176 k = bch2_btree_iter_peek_slot(freespace_iter);
1181 *end = bkey_min(k.k->p, *end);
1183 if (k.k->type != KEY_TYPE_set &&
1184 (c->opts.reconstruct_alloc ||
1185 fsck_err(c, "hole in alloc btree missing in freespace btree\n"
1186 " device %llu buckets %llu-%llu",
1187 freespace_iter->pos.inode,
1188 freespace_iter->pos.offset,
1190 struct bkey_i *update =
1191 bch2_trans_kmalloc(trans, sizeof(*update));
1193 ret = PTR_ERR_OR_ZERO(update);
1197 bkey_init(&update->k);
1198 update->k.type = KEY_TYPE_set;
1199 update->k.p = freespace_iter->pos;
1200 bch2_key_resize(&update->k,
1201 min_t(u64, U32_MAX, end->offset -
1202 freespace_iter->pos.offset));
1204 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1210 printbuf_exit(&buf);
1214 static int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans,
1217 struct btree_iter *bucket_gens_iter)
1219 struct bch_fs *c = trans->c;
1221 struct printbuf buf = PRINTBUF;
1222 unsigned i, gens_offset, gens_end_offset;
1225 if (c->sb.version < bcachefs_metadata_version_bucket_gens &&
1226 !c->opts.version_upgrade)
1229 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(start, &gens_offset));
1231 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1236 if (bkey_cmp(alloc_gens_pos(start, &gens_offset),
1237 alloc_gens_pos(*end, &gens_end_offset)))
1238 gens_end_offset = KEY_TYPE_BUCKET_GENS_NR;
1240 if (k.k->type == KEY_TYPE_bucket_gens) {
1241 struct bkey_i_bucket_gens g;
1242 bool need_update = false;
1244 bkey_reassemble(&g.k_i, k);
1246 for (i = gens_offset; i < gens_end_offset; i++) {
1247 if (fsck_err_on(g.v.gens[i], c,
1248 "hole in alloc btree at %llu:%llu with nonzero gen in bucket_gens btree (%u)",
1249 bucket_gens_pos_to_alloc(k.k->p, i).inode,
1250 bucket_gens_pos_to_alloc(k.k->p, i).offset,
1258 struct bkey_i *k = bch2_trans_kmalloc(trans, sizeof(g));
1260 ret = PTR_ERR_OR_ZERO(k);
1264 memcpy(k, &g, sizeof(g));
1266 ret = bch2_trans_update(trans, bucket_gens_iter, k, 0);
1272 *end = bkey_min(*end, bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0));
1275 printbuf_exit(&buf);
1279 static int bch2_check_discard_freespace_key(struct btree_trans *trans,
1280 struct btree_iter *iter)
1282 struct bch_fs *c = trans->c;
1283 struct btree_iter alloc_iter;
1284 struct bkey_s_c alloc_k;
1285 struct bch_alloc_v4 a_convert;
1286 const struct bch_alloc_v4 *a;
1289 enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard
1290 ? BCH_DATA_need_discard
1292 struct printbuf buf = PRINTBUF;
1296 pos.offset &= ~(~0ULL << 56);
1297 genbits = iter->pos.offset & (~0ULL << 56);
1299 bch2_trans_iter_init(trans, &alloc_iter, BTREE_ID_alloc, pos, 0);
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 alloc_k = bch2_btree_iter_peek_slot(&alloc_iter);
1307 ret = bkey_err(alloc_k);
1311 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1313 if (fsck_err_on(a->data_type != state ||
1314 (state == BCH_DATA_free &&
1315 genbits != alloc_freespace_genbits(*a)), c,
1316 "%s\n incorrectly set in %s index (free %u, genbits %llu should be %llu)",
1317 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
1318 bch2_btree_ids[iter->btree_id],
1319 a->data_type == state,
1320 genbits >> 56, alloc_freespace_genbits(*a) >> 56))
1325 bch2_trans_iter_exit(trans, &alloc_iter);
1326 printbuf_exit(&buf);
1329 ret = bch2_btree_delete_extent_at(trans, iter,
1330 iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0);
1335 * We've already checked that generation numbers in the bucket_gens btree are
1336 * valid for buckets that exist; this just checks for keys for nonexistent
1339 static int bch2_check_bucket_gens_key(struct btree_trans *trans,
1340 struct btree_iter *iter,
1343 struct bch_fs *c = trans->c;
1344 struct bkey_i_bucket_gens g;
1346 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
1347 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
1349 bool need_update = false;
1350 struct printbuf buf = PRINTBUF;
1353 BUG_ON(k.k->type != KEY_TYPE_bucket_gens);
1354 bkey_reassemble(&g.k_i, k);
1356 if (fsck_err_on(!bch2_dev_exists2(c, k.k->p.inode), c,
1357 "bucket_gens key for invalid device:\n %s",
1358 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1359 ret = bch2_btree_delete_at(trans, iter, 0);
1363 ca = bch_dev_bkey_exists(c, k.k->p.inode);
1364 if (fsck_err_on(end <= ca->mi.first_bucket ||
1365 start >= ca->mi.nbuckets, c,
1366 "bucket_gens key for invalid buckets:\n %s",
1367 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1368 ret = bch2_btree_delete_at(trans, iter, 0);
1372 for (b = start; b < ca->mi.first_bucket; b++)
1373 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1374 "bucket_gens key has nonzero gen for invalid bucket")) {
1375 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1379 for (b = ca->mi.nbuckets; b < end; b++)
1380 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1381 "bucket_gens key has nonzero gen for invalid bucket")) {
1382 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1389 k = bch2_trans_kmalloc(trans, sizeof(g));
1390 ret = PTR_ERR_OR_ZERO(k);
1394 memcpy(k, &g, sizeof(g));
1395 ret = bch2_trans_update(trans, iter, k, 0);
1399 printbuf_exit(&buf);
1403 int bch2_check_alloc_info(struct bch_fs *c)
1405 struct btree_trans trans;
1406 struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter;
1411 bch2_trans_init(&trans, c, 0, 0);
1413 bch2_trans_iter_init(&trans, &iter, BTREE_ID_alloc, POS_MIN,
1414 BTREE_ITER_PREFETCH);
1415 bch2_trans_iter_init(&trans, &discard_iter, BTREE_ID_need_discard, POS_MIN,
1416 BTREE_ITER_PREFETCH);
1417 bch2_trans_iter_init(&trans, &freespace_iter, BTREE_ID_freespace, POS_MIN,
1418 BTREE_ITER_PREFETCH);
1419 bch2_trans_iter_init(&trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN,
1420 BTREE_ITER_PREFETCH);
1425 bch2_trans_begin(&trans);
1427 k = bch2_get_key_or_real_bucket_hole(&iter, &hole);
1436 next = bpos_nosnap_successor(k.k->p);
1438 ret = bch2_check_alloc_key(&trans,
1448 ret = bch2_check_alloc_hole_freespace(&trans,
1449 bkey_start_pos(k.k),
1452 bch2_check_alloc_hole_bucket_gens(&trans,
1453 bkey_start_pos(k.k),
1460 ret = bch2_trans_commit(&trans, NULL, NULL,
1461 BTREE_INSERT_NOFAIL|
1462 BTREE_INSERT_LAZY_RW);
1466 bch2_btree_iter_set_pos(&iter, next);
1468 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1473 bch2_trans_iter_exit(&trans, &bucket_gens_iter);
1474 bch2_trans_iter_exit(&trans, &freespace_iter);
1475 bch2_trans_iter_exit(&trans, &discard_iter);
1476 bch2_trans_iter_exit(&trans, &iter);
1481 ret = for_each_btree_key_commit(&trans, iter,
1482 BTREE_ID_need_discard, POS_MIN,
1483 BTREE_ITER_PREFETCH, k,
1484 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
1485 bch2_check_discard_freespace_key(&trans, &iter)) ?:
1486 for_each_btree_key_commit(&trans, iter,
1487 BTREE_ID_freespace, POS_MIN,
1488 BTREE_ITER_PREFETCH, k,
1489 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
1490 bch2_check_discard_freespace_key(&trans, &iter)) ?:
1491 for_each_btree_key_commit(&trans, iter,
1492 BTREE_ID_bucket_gens, POS_MIN,
1493 BTREE_ITER_PREFETCH, k,
1494 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
1495 bch2_check_bucket_gens_key(&trans, &iter, k));
1497 bch2_trans_exit(&trans);
1498 return ret < 0 ? ret : 0;
1501 static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
1502 struct btree_iter *alloc_iter)
1504 struct bch_fs *c = trans->c;
1505 struct btree_iter lru_iter;
1506 struct bch_alloc_v4 a_convert;
1507 const struct bch_alloc_v4 *a;
1508 struct bkey_s_c alloc_k, k;
1509 struct printbuf buf = PRINTBUF;
1510 struct printbuf buf2 = PRINTBUF;
1513 alloc_k = bch2_btree_iter_peek(alloc_iter);
1517 ret = bkey_err(alloc_k);
1521 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1523 if (a->data_type != BCH_DATA_cached)
1526 bch2_trans_iter_init(trans, &lru_iter, BTREE_ID_lru,
1527 POS(alloc_k.k->p.inode, a->io_time[READ]), 0);
1529 k = bch2_btree_iter_peek_slot(&lru_iter);
1534 if (fsck_err_on(!a->io_time[READ], c,
1535 "cached bucket with read_time 0\n"
1537 (printbuf_reset(&buf),
1538 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)) ||
1539 fsck_err_on(k.k->type != KEY_TYPE_lru ||
1540 le64_to_cpu(bkey_s_c_to_lru(k).v->idx) != alloc_k.k->p.offset, c,
1541 "incorrect/missing lru entry\n"
1544 (printbuf_reset(&buf),
1545 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
1546 (bch2_bkey_val_to_text(&buf2, c, k), buf2.buf))) {
1547 u64 read_time = a->io_time[READ] ?:
1548 atomic64_read(&c->io_clock[READ].now);
1550 ret = bch2_lru_set(trans,
1552 alloc_k.k->p.offset,
1557 if (a->io_time[READ] != read_time) {
1558 struct bkey_i_alloc_v4 *a_mut =
1559 bch2_alloc_to_v4_mut(trans, alloc_k);
1560 ret = PTR_ERR_OR_ZERO(a_mut);
1564 a_mut->v.io_time[READ] = read_time;
1565 ret = bch2_trans_update(trans, alloc_iter,
1566 &a_mut->k_i, BTREE_TRIGGER_NORUN);
1573 bch2_trans_iter_exit(trans, &lru_iter);
1574 printbuf_exit(&buf2);
1575 printbuf_exit(&buf);
1579 int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
1581 struct btree_trans trans;
1582 struct btree_iter iter;
1586 bch2_trans_init(&trans, c, 0, 0);
1588 for_each_btree_key_commit(&trans, iter, BTREE_ID_alloc,
1589 POS_MIN, BTREE_ITER_PREFETCH, k,
1590 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
1591 bch2_check_alloc_to_lru_ref(&trans, &iter));
1593 bch2_trans_exit(&trans);
1594 return ret < 0 ? ret : 0;
1597 static int bch2_discard_one_bucket(struct btree_trans *trans,
1598 struct btree_iter *need_discard_iter,
1599 struct bpos *discard_pos_done,
1602 u64 *need_journal_commit,
1605 struct bch_fs *c = trans->c;
1606 struct bpos pos = need_discard_iter->pos;
1607 struct btree_iter iter = { NULL };
1610 struct bkey_i_alloc_v4 *a;
1611 struct printbuf buf = PRINTBUF;
1612 bool did_discard = false;
1615 ca = bch_dev_bkey_exists(c, pos.inode);
1616 if (!percpu_ref_tryget(&ca->io_ref)) {
1617 bch2_btree_iter_set_pos(need_discard_iter, POS(pos.inode + 1, 0));
1621 if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) {
1626 if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
1627 c->journal.flushed_seq_ondisk,
1628 pos.inode, pos.offset)) {
1629 (*need_journal_commit)++;
1633 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
1634 need_discard_iter->pos,
1636 k = bch2_btree_iter_peek_slot(&iter);
1641 a = bch2_alloc_to_v4_mut(trans, k);
1642 ret = PTR_ERR_OR_ZERO(a);
1646 if (BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
1648 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1652 if (bch2_trans_inconsistent_on(a->v.journal_seq > c->journal.flushed_seq_ondisk, trans,
1653 "clearing need_discard but journal_seq %llu > flushed_seq %llu\n"
1656 c->journal.flushed_seq_ondisk,
1657 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1662 if (bch2_trans_inconsistent_on(a->v.data_type != BCH_DATA_need_discard, trans,
1663 "bucket incorrectly set in need_discard btree\n"
1665 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1670 if (!bkey_eq(*discard_pos_done, iter.pos) &&
1671 ca->mi.discard && !c->opts.nochanges) {
1673 * This works without any other locks because this is the only
1674 * thread that removes items from the need_discard tree
1676 bch2_trans_unlock(trans);
1677 blkdev_issue_discard(ca->disk_sb.bdev,
1678 k.k->p.offset * ca->mi.bucket_size,
1682 ret = bch2_trans_relock(trans);
1687 *discard_pos_done = iter.pos;
1690 SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
1691 a->v.data_type = alloc_data_type(a->v, a->v.data_type);
1693 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
1694 bch2_trans_commit(trans, NULL, NULL,
1695 BTREE_INSERT_USE_RESERVE|BTREE_INSERT_NOFAIL);
1700 this_cpu_inc(c->counters[BCH_COUNTER_bucket_discard]);
1704 bch2_trans_iter_exit(trans, &iter);
1705 percpu_ref_put(&ca->io_ref);
1706 printbuf_exit(&buf);
1710 static void bch2_do_discards_work(struct work_struct *work)
1712 struct bch_fs *c = container_of(work, struct bch_fs, discard_work);
1713 struct btree_trans trans;
1714 struct btree_iter iter;
1716 u64 seen = 0, open = 0, need_journal_commit = 0, discarded = 0;
1717 struct bpos discard_pos_done = POS_MAX;
1720 bch2_trans_init(&trans, c, 0, 0);
1723 * We're doing the commit in bch2_discard_one_bucket instead of using
1724 * for_each_btree_key_commit() so that we can increment counters after
1725 * successful commit:
1727 ret = for_each_btree_key2(&trans, iter,
1728 BTREE_ID_need_discard, POS_MIN, 0, k,
1729 bch2_discard_one_bucket(&trans, &iter, &discard_pos_done,
1732 &need_journal_commit,
1735 bch2_trans_exit(&trans);
1737 if (need_journal_commit * 2 > seen)
1738 bch2_journal_flush_async(&c->journal, NULL);
1740 percpu_ref_put(&c->writes);
1742 trace_discard_buckets(c, seen, open, need_journal_commit, discarded,
1746 void bch2_do_discards(struct bch_fs *c)
1748 if (percpu_ref_tryget_live(&c->writes) &&
1749 !queue_work(system_long_wq, &c->discard_work))
1750 percpu_ref_put(&c->writes);
1753 static int invalidate_one_bucket(struct btree_trans *trans,
1754 struct btree_iter *lru_iter, struct bkey_s_c k,
1755 unsigned dev_idx, s64 *nr_to_invalidate)
1757 struct bch_fs *c = trans->c;
1758 struct btree_iter alloc_iter = { NULL };
1759 struct bkey_i_alloc_v4 *a;
1761 struct printbuf buf = PRINTBUF;
1762 unsigned cached_sectors;
1765 if (*nr_to_invalidate <= 0 || k.k->p.inode != dev_idx)
1768 if (k.k->type != KEY_TYPE_lru) {
1769 prt_printf(&buf, "non lru key in lru btree:\n ");
1770 bch2_bkey_val_to_text(&buf, c, k);
1772 if (!test_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags)) {
1773 bch_err(c, "%s", buf.buf);
1775 bch2_trans_inconsistent(trans, "%s", buf.buf);
1782 bucket = POS(dev_idx, le64_to_cpu(bkey_s_c_to_lru(k).v->idx));
1784 a = bch2_trans_start_alloc_update(trans, &alloc_iter, bucket);
1785 ret = PTR_ERR_OR_ZERO(a);
1789 if (k.k->p.offset != alloc_lru_idx(a->v)) {
1790 prt_printf(&buf, "alloc key does not point back to lru entry when invalidating bucket:\n ");
1791 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i));
1792 prt_printf(&buf, "\n ");
1793 bch2_bkey_val_to_text(&buf, c, k);
1795 if (!test_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags)) {
1796 bch_err(c, "%s", buf.buf);
1798 bch2_trans_inconsistent(trans, "%s", buf.buf);
1805 if (!a->v.cached_sectors)
1806 bch_err(c, "invalidating empty bucket, confused");
1808 cached_sectors = a->v.cached_sectors;
1810 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1813 a->v.dirty_sectors = 0;
1814 a->v.cached_sectors = 0;
1815 a->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now);
1816 a->v.io_time[WRITE] = atomic64_read(&c->io_clock[WRITE].now);
1818 ret = bch2_trans_update(trans, &alloc_iter, &a->k_i,
1819 BTREE_TRIGGER_BUCKET_INVALIDATE) ?:
1820 bch2_trans_commit(trans, NULL, NULL,
1821 BTREE_INSERT_USE_RESERVE|BTREE_INSERT_NOFAIL);
1825 trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors);
1826 --*nr_to_invalidate;
1828 bch2_trans_iter_exit(trans, &alloc_iter);
1829 printbuf_exit(&buf);
1833 static void bch2_do_invalidates_work(struct work_struct *work)
1835 struct bch_fs *c = container_of(work, struct bch_fs, invalidate_work);
1837 struct btree_trans trans;
1838 struct btree_iter iter;
1843 bch2_trans_init(&trans, c, 0, 0);
1845 for_each_member_device(ca, c, i) {
1846 s64 nr_to_invalidate =
1847 should_invalidate_buckets(ca, bch2_dev_usage_read(ca));
1849 ret = for_each_btree_key2(&trans, iter, BTREE_ID_lru,
1850 POS(ca->dev_idx, 0), BTREE_ITER_INTENT, k,
1851 invalidate_one_bucket(&trans, &iter, k, ca->dev_idx, &nr_to_invalidate));
1854 percpu_ref_put(&ca->ref);
1859 bch2_trans_exit(&trans);
1860 percpu_ref_put(&c->writes);
1863 void bch2_do_invalidates(struct bch_fs *c)
1865 if (percpu_ref_tryget_live(&c->writes) &&
1866 !queue_work(system_long_wq, &c->invalidate_work))
1867 percpu_ref_put(&c->writes);
1870 static int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca)
1872 struct btree_trans trans;
1873 struct btree_iter iter;
1876 struct bpos end = POS(ca->dev_idx, ca->mi.nbuckets);
1877 struct bch_member *m;
1880 bch2_trans_init(&trans, c, 0, 0);
1882 bch2_trans_iter_init(&trans, &iter, BTREE_ID_alloc,
1883 POS(ca->dev_idx, ca->mi.first_bucket),
1884 BTREE_ITER_PREFETCH);
1886 * Scan the alloc btree for every bucket on @ca, and add buckets to the
1887 * freespace/need_discard/need_gc_gens btrees as needed:
1890 bch2_trans_begin(&trans);
1892 if (bkey_ge(iter.pos, end)) {
1897 k = bch2_get_key_or_hole(&iter, end, &hole);
1904 * We process live keys in the alloc btree one at a
1907 struct bch_alloc_v4 a_convert;
1908 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert);
1910 ret = bch2_bucket_do_index(&trans, k, a, true) ?:
1911 bch2_trans_commit(&trans, NULL, NULL,
1912 BTREE_INSERT_LAZY_RW|
1913 BTREE_INSERT_NOFAIL);
1917 bch2_btree_iter_advance(&iter);
1919 struct bkey_i *freespace;
1921 freespace = bch2_trans_kmalloc(&trans, sizeof(*freespace));
1922 ret = PTR_ERR_OR_ZERO(freespace);
1926 bkey_init(&freespace->k);
1927 freespace->k.type = KEY_TYPE_set;
1928 freespace->k.p = k.k->p;
1929 freespace->k.size = k.k->size;
1931 ret = __bch2_btree_insert(&trans, BTREE_ID_freespace, freespace) ?:
1932 bch2_trans_commit(&trans, NULL, NULL,
1933 BTREE_INSERT_LAZY_RW|
1934 BTREE_INSERT_NOFAIL);
1938 bch2_btree_iter_set_pos(&iter, k.k->p);
1941 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1947 bch2_trans_iter_exit(&trans, &iter);
1948 bch2_trans_exit(&trans);
1951 bch_err(ca, "error initializing free space: %s", bch2_err_str(ret));
1955 mutex_lock(&c->sb_lock);
1956 m = bch2_sb_get_members(c->disk_sb.sb)->members + ca->dev_idx;
1957 SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true);
1958 mutex_unlock(&c->sb_lock);
1963 int bch2_fs_freespace_init(struct bch_fs *c)
1968 bool doing_init = false;
1971 * We can crash during the device add path, so we need to check this on
1975 for_each_member_device(ca, c, i) {
1976 if (ca->mi.freespace_initialized)
1980 bch_info(c, "initializing freespace");
1984 ret = bch2_dev_freespace_init(c, ca);
1986 percpu_ref_put(&ca->ref);
1992 mutex_lock(&c->sb_lock);
1993 bch2_write_super(c);
1994 mutex_unlock(&c->sb_lock);
1996 bch_verbose(c, "done initializing freespace");
2002 /* Bucket IO clocks: */
2004 int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
2005 size_t bucket_nr, int rw)
2007 struct bch_fs *c = trans->c;
2008 struct btree_iter iter;
2009 struct bkey_i_alloc_v4 *a;
2013 a = bch2_trans_start_alloc_update(trans, &iter, POS(dev, bucket_nr));
2014 ret = PTR_ERR_OR_ZERO(a);
2018 now = atomic64_read(&c->io_clock[rw].now);
2019 if (a->v.io_time[rw] == now)
2022 a->v.io_time[rw] = now;
2024 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
2025 bch2_trans_commit(trans, NULL, NULL, 0);
2027 bch2_trans_iter_exit(trans, &iter);
2031 /* Startup/shutdown (ro/rw): */
2033 void bch2_recalc_capacity(struct bch_fs *c)
2036 u64 capacity = 0, reserved_sectors = 0, gc_reserve;
2037 unsigned bucket_size_max = 0;
2038 unsigned long ra_pages = 0;
2041 lockdep_assert_held(&c->state_lock);
2043 for_each_online_member(ca, c, i) {
2044 struct backing_dev_info *bdi = ca->disk_sb.bdev->bd_disk->bdi;
2046 ra_pages += bdi->ra_pages;
2049 bch2_set_ra_pages(c, ra_pages);
2051 for_each_rw_member(ca, c, i) {
2052 u64 dev_reserve = 0;
2055 * We need to reserve buckets (from the number
2056 * of currently available buckets) against
2057 * foreground writes so that mainly copygc can
2058 * make forward progress.
2060 * We need enough to refill the various reserves
2061 * from scratch - copygc will use its entire
2062 * reserve all at once, then run against when
2063 * its reserve is refilled (from the formerly
2064 * available buckets).
2066 * This reserve is just used when considering if
2067 * allocations for foreground writes must wait -
2068 * not -ENOSPC calculations.
2071 dev_reserve += ca->nr_btree_reserve * 2;
2072 dev_reserve += ca->mi.nbuckets >> 6; /* copygc reserve */
2074 dev_reserve += 1; /* btree write point */
2075 dev_reserve += 1; /* copygc write point */
2076 dev_reserve += 1; /* rebalance write point */
2078 dev_reserve *= ca->mi.bucket_size;
2080 capacity += bucket_to_sector(ca, ca->mi.nbuckets -
2081 ca->mi.first_bucket);
2083 reserved_sectors += dev_reserve * 2;
2085 bucket_size_max = max_t(unsigned, bucket_size_max,
2086 ca->mi.bucket_size);
2089 gc_reserve = c->opts.gc_reserve_bytes
2090 ? c->opts.gc_reserve_bytes >> 9
2091 : div64_u64(capacity * c->opts.gc_reserve_percent, 100);
2093 reserved_sectors = max(gc_reserve, reserved_sectors);
2095 reserved_sectors = min(reserved_sectors, capacity);
2097 c->capacity = capacity - reserved_sectors;
2099 c->bucket_size_max = bucket_size_max;
2101 /* Wake up case someone was waiting for buckets */
2102 closure_wake_up(&c->freelist_wait);
2105 static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca)
2107 struct open_bucket *ob;
2110 for (ob = c->open_buckets;
2111 ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
2113 spin_lock(&ob->lock);
2114 if (ob->valid && !ob->on_partial_list &&
2115 ob->dev == ca->dev_idx)
2117 spin_unlock(&ob->lock);
2123 /* device goes ro: */
2124 void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca)
2128 /* First, remove device from allocation groups: */
2130 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2131 clear_bit(ca->dev_idx, c->rw_devs[i].d);
2134 * Capacity is calculated based off of devices in allocation groups:
2136 bch2_recalc_capacity(c);
2138 /* Next, close write points that point to this device... */
2139 for (i = 0; i < ARRAY_SIZE(c->write_points); i++)
2140 bch2_writepoint_stop(c, ca, &c->write_points[i]);
2142 bch2_writepoint_stop(c, ca, &c->copygc_write_point);
2143 bch2_writepoint_stop(c, ca, &c->rebalance_write_point);
2144 bch2_writepoint_stop(c, ca, &c->btree_write_point);
2146 mutex_lock(&c->btree_reserve_cache_lock);
2147 while (c->btree_reserve_cache_nr) {
2148 struct btree_alloc *a =
2149 &c->btree_reserve_cache[--c->btree_reserve_cache_nr];
2151 bch2_open_buckets_put(c, &a->ob);
2153 mutex_unlock(&c->btree_reserve_cache_lock);
2156 struct open_bucket *ob;
2158 spin_lock(&c->freelist_lock);
2159 if (!ca->open_buckets_partial_nr) {
2160 spin_unlock(&c->freelist_lock);
2163 ob = c->open_buckets +
2164 ca->open_buckets_partial[--ca->open_buckets_partial_nr];
2165 ob->on_partial_list = false;
2166 spin_unlock(&c->freelist_lock);
2168 bch2_open_bucket_put(c, ob);
2171 bch2_ec_stop_dev(c, ca);
2174 * Wake up threads that were blocked on allocation, so they can notice
2175 * the device can no longer be removed and the capacity has changed:
2177 closure_wake_up(&c->freelist_wait);
2180 * journal_res_get() can block waiting for free space in the journal -
2181 * it needs to notice there may not be devices to allocate from anymore:
2183 wake_up(&c->journal.wait);
2185 /* Now wait for any in flight writes: */
2187 closure_wait_event(&c->open_buckets_wait,
2188 !bch2_dev_has_open_write_point(c, ca));
2191 /* device goes rw: */
2192 void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca)
2196 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2197 if (ca->mi.data_allowed & (1 << i))
2198 set_bit(ca->dev_idx, c->rw_devs[i].d);
2201 void bch2_fs_allocator_background_init(struct bch_fs *c)
2203 spin_lock_init(&c->freelist_lock);
2204 INIT_WORK(&c->discard_work, bch2_do_discards_work);
2205 INIT_WORK(&c->invalidate_work, bch2_do_invalidates_work);