1 // SPDX-License-Identifier: GPL-2.0
3 #include "alloc_background.h"
4 #include "alloc_foreground.h"
5 #include "backpointers.h"
6 #include "btree_cache.h"
8 #include "btree_key_cache.h"
9 #include "btree_update.h"
10 #include "btree_update_interior.h"
12 #include "btree_write_buffer.h"
14 #include "buckets_waiting_for_journal.h"
24 #include <linux/kthread.h>
25 #include <linux/math64.h>
26 #include <linux/random.h>
27 #include <linux/rculist.h>
28 #include <linux/rcupdate.h>
29 #include <linux/sched/task.h>
30 #include <linux/sort.h>
32 /* Persistent alloc info: */
34 static const unsigned BCH_ALLOC_V1_FIELD_BYTES[] = {
35 #define x(name, bits) [BCH_ALLOC_FIELD_V1_##name] = bits / 8,
40 struct bkey_alloc_unpacked {
47 #define x(_name, _bits) u##_bits _name;
52 static inline u64 alloc_field_v1_get(const struct bch_alloc *a,
53 const void **p, unsigned field)
55 unsigned bytes = BCH_ALLOC_V1_FIELD_BYTES[field];
58 if (!(a->fields & (1 << field)))
63 v = *((const u8 *) *p);
82 static void bch2_alloc_unpack_v1(struct bkey_alloc_unpacked *out,
85 const struct bch_alloc *in = bkey_s_c_to_alloc(k).v;
86 const void *d = in->data;
91 #define x(_name, _bits) out->_name = alloc_field_v1_get(in, &d, idx++);
96 static int bch2_alloc_unpack_v2(struct bkey_alloc_unpacked *out,
99 struct bkey_s_c_alloc_v2 a = bkey_s_c_to_alloc_v2(k);
100 const u8 *in = a.v->data;
101 const u8 *end = bkey_val_end(a);
102 unsigned fieldnr = 0;
107 out->oldest_gen = a.v->oldest_gen;
108 out->data_type = a.v->data_type;
110 #define x(_name, _bits) \
111 if (fieldnr < a.v->nr_fields) { \
112 ret = bch2_varint_decode_fast(in, end, &v); \
120 if (v != out->_name) \
124 BCH_ALLOC_FIELDS_V2()
129 static int bch2_alloc_unpack_v3(struct bkey_alloc_unpacked *out,
132 struct bkey_s_c_alloc_v3 a = bkey_s_c_to_alloc_v3(k);
133 const u8 *in = a.v->data;
134 const u8 *end = bkey_val_end(a);
135 unsigned fieldnr = 0;
140 out->oldest_gen = a.v->oldest_gen;
141 out->data_type = a.v->data_type;
142 out->need_discard = BCH_ALLOC_V3_NEED_DISCARD(a.v);
143 out->need_inc_gen = BCH_ALLOC_V3_NEED_INC_GEN(a.v);
144 out->journal_seq = le64_to_cpu(a.v->journal_seq);
146 #define x(_name, _bits) \
147 if (fieldnr < a.v->nr_fields) { \
148 ret = bch2_varint_decode_fast(in, end, &v); \
156 if (v != out->_name) \
160 BCH_ALLOC_FIELDS_V2()
165 static struct bkey_alloc_unpacked bch2_alloc_unpack(struct bkey_s_c k)
167 struct bkey_alloc_unpacked ret = { .gen = 0 };
171 bch2_alloc_unpack_v1(&ret, k);
173 case KEY_TYPE_alloc_v2:
174 bch2_alloc_unpack_v2(&ret, k);
176 case KEY_TYPE_alloc_v3:
177 bch2_alloc_unpack_v3(&ret, k);
184 static unsigned bch_alloc_v1_val_u64s(const struct bch_alloc *a)
186 unsigned i, bytes = offsetof(struct bch_alloc, data);
188 for (i = 0; i < ARRAY_SIZE(BCH_ALLOC_V1_FIELD_BYTES); i++)
189 if (a->fields & (1 << i))
190 bytes += BCH_ALLOC_V1_FIELD_BYTES[i];
192 return DIV_ROUND_UP(bytes, sizeof(u64));
195 int bch2_alloc_v1_invalid(struct bch_fs *c, struct bkey_s_c k,
196 enum bkey_invalid_flags flags,
197 struct printbuf *err)
199 struct bkey_s_c_alloc a = bkey_s_c_to_alloc(k);
202 /* allow for unknown fields */
203 bkey_fsck_err_on(bkey_val_u64s(a.k) < bch_alloc_v1_val_u64s(a.v), c, err,
204 alloc_v1_val_size_bad,
205 "incorrect value size (%zu < %u)",
206 bkey_val_u64s(a.k), bch_alloc_v1_val_u64s(a.v));
211 int bch2_alloc_v2_invalid(struct bch_fs *c, struct bkey_s_c k,
212 enum bkey_invalid_flags flags,
213 struct printbuf *err)
215 struct bkey_alloc_unpacked u;
218 bkey_fsck_err_on(bch2_alloc_unpack_v2(&u, k), c, err,
219 alloc_v2_unpack_error,
225 int bch2_alloc_v3_invalid(struct bch_fs *c, struct bkey_s_c k,
226 enum bkey_invalid_flags flags,
227 struct printbuf *err)
229 struct bkey_alloc_unpacked u;
232 bkey_fsck_err_on(bch2_alloc_unpack_v3(&u, k), c, err,
233 alloc_v2_unpack_error,
239 int bch2_alloc_v4_invalid(struct bch_fs *c, struct bkey_s_c k,
240 enum bkey_invalid_flags flags, struct printbuf *err)
242 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
245 bkey_fsck_err_on(alloc_v4_u64s(a.v) > bkey_val_u64s(k.k), c, err,
246 alloc_v4_val_size_bad,
247 "bad val size (%u > %zu)",
248 alloc_v4_u64s(a.v), bkey_val_u64s(k.k));
250 bkey_fsck_err_on(!BCH_ALLOC_V4_BACKPOINTERS_START(a.v) &&
251 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v), c, err,
252 alloc_v4_backpointers_start_bad,
253 "invalid backpointers_start");
255 bkey_fsck_err_on(alloc_data_type(*a.v, a.v->data_type) != a.v->data_type, c, err,
256 alloc_key_data_type_bad,
257 "invalid data type (got %u should be %u)",
258 a.v->data_type, alloc_data_type(*a.v, a.v->data_type));
260 switch (a.v->data_type) {
262 case BCH_DATA_need_gc_gens:
263 case BCH_DATA_need_discard:
264 bkey_fsck_err_on(bch2_bucket_sectors(*a.v) || a.v->stripe,
265 c, err, alloc_key_empty_but_have_data,
266 "empty data type free but have data");
269 case BCH_DATA_journal:
272 case BCH_DATA_parity:
273 bkey_fsck_err_on(!bch2_bucket_sectors_dirty(*a.v),
274 c, err, alloc_key_dirty_sectors_0,
275 "data_type %s but dirty_sectors==0",
276 bch2_data_types[a.v->data_type]);
278 case BCH_DATA_cached:
279 bkey_fsck_err_on(!a.v->cached_sectors ||
280 bch2_bucket_sectors_dirty(*a.v) ||
282 c, err, alloc_key_cached_inconsistency,
283 "data type inconsistency");
285 bkey_fsck_err_on(!a.v->io_time[READ] &&
286 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_to_lru_refs,
287 c, err, alloc_key_cached_but_read_time_zero,
288 "cached bucket with read_time == 0");
290 case BCH_DATA_stripe:
297 void bch2_alloc_v4_swab(struct bkey_s k)
299 struct bch_alloc_v4 *a = bkey_s_to_alloc_v4(k).v;
300 struct bch_backpointer *bp, *bps;
302 a->journal_seq = swab64(a->journal_seq);
303 a->flags = swab32(a->flags);
304 a->dirty_sectors = swab32(a->dirty_sectors);
305 a->cached_sectors = swab32(a->cached_sectors);
306 a->io_time[0] = swab64(a->io_time[0]);
307 a->io_time[1] = swab64(a->io_time[1]);
308 a->stripe = swab32(a->stripe);
309 a->nr_external_backpointers = swab32(a->nr_external_backpointers);
310 a->fragmentation_lru = swab64(a->fragmentation_lru);
312 bps = alloc_v4_backpointers(a);
313 for (bp = bps; bp < bps + BCH_ALLOC_V4_NR_BACKPOINTERS(a); bp++) {
314 bp->bucket_offset = swab40(bp->bucket_offset);
315 bp->bucket_len = swab32(bp->bucket_len);
316 bch2_bpos_swab(&bp->pos);
320 void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
322 struct bch_alloc_v4 _a;
323 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a);
327 printbuf_indent_add(out, 2);
329 prt_printf(out, "gen %u oldest_gen %u data_type %s",
330 a->gen, a->oldest_gen,
331 a->data_type < BCH_DATA_NR
332 ? bch2_data_types[a->data_type]
333 : "(invalid data type)");
335 prt_printf(out, "journal_seq %llu", a->journal_seq);
337 prt_printf(out, "need_discard %llu", BCH_ALLOC_V4_NEED_DISCARD(a));
339 prt_printf(out, "need_inc_gen %llu", BCH_ALLOC_V4_NEED_INC_GEN(a));
341 prt_printf(out, "dirty_sectors %u", a->dirty_sectors);
343 prt_printf(out, "cached_sectors %u", a->cached_sectors);
345 prt_printf(out, "stripe %u", a->stripe);
347 prt_printf(out, "stripe_redundancy %u", a->stripe_redundancy);
349 prt_printf(out, "io_time[READ] %llu", a->io_time[READ]);
351 prt_printf(out, "io_time[WRITE] %llu", a->io_time[WRITE]);
353 prt_printf(out, "fragmentation %llu", a->fragmentation_lru);
355 prt_printf(out, "bp_start %llu", BCH_ALLOC_V4_BACKPOINTERS_START(a));
358 if (BCH_ALLOC_V4_NR_BACKPOINTERS(a)) {
359 struct bkey_s_c_alloc_v4 a_raw = bkey_s_c_to_alloc_v4(k);
360 const struct bch_backpointer *bps = alloc_v4_backpointers_c(a_raw.v);
362 prt_printf(out, "backpointers: %llu", BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v));
363 printbuf_indent_add(out, 2);
365 for (i = 0; i < BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v); i++) {
367 bch2_backpointer_to_text(out, &bps[i]);
370 printbuf_indent_sub(out, 2);
373 printbuf_indent_sub(out, 2);
376 void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out)
378 if (k.k->type == KEY_TYPE_alloc_v4) {
381 *out = *bkey_s_c_to_alloc_v4(k).v;
383 src = alloc_v4_backpointers(out);
384 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
385 dst = alloc_v4_backpointers(out);
388 memset(src, 0, dst - src);
390 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(out, 0);
392 struct bkey_alloc_unpacked u = bch2_alloc_unpack(k);
394 *out = (struct bch_alloc_v4) {
395 .journal_seq = u.journal_seq,
396 .flags = u.need_discard,
398 .oldest_gen = u.oldest_gen,
399 .data_type = u.data_type,
400 .stripe_redundancy = u.stripe_redundancy,
401 .dirty_sectors = u.dirty_sectors,
402 .cached_sectors = u.cached_sectors,
403 .io_time[READ] = u.read_time,
404 .io_time[WRITE] = u.write_time,
408 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
412 static noinline struct bkey_i_alloc_v4 *
413 __bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
415 struct bkey_i_alloc_v4 *ret;
417 ret = bch2_trans_kmalloc(trans, max(bkey_bytes(k.k), sizeof(struct bkey_i_alloc_v4)));
421 if (k.k->type == KEY_TYPE_alloc_v4) {
424 bkey_reassemble(&ret->k_i, k);
426 src = alloc_v4_backpointers(&ret->v);
427 SET_BCH_ALLOC_V4_BACKPOINTERS_START(&ret->v, BCH_ALLOC_V4_U64s);
428 dst = alloc_v4_backpointers(&ret->v);
431 memset(src, 0, dst - src);
433 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(&ret->v, 0);
434 set_alloc_v4_u64s(ret);
436 bkey_alloc_v4_init(&ret->k_i);
438 bch2_alloc_to_v4(k, &ret->v);
443 static inline struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut_inlined(struct btree_trans *trans, struct bkey_s_c k)
445 struct bkey_s_c_alloc_v4 a;
447 if (likely(k.k->type == KEY_TYPE_alloc_v4) &&
448 ((a = bkey_s_c_to_alloc_v4(k), true) &&
449 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v) == 0))
450 return bch2_bkey_make_mut_noupdate_typed(trans, k, alloc_v4);
452 return __bch2_alloc_to_v4_mut(trans, k);
455 struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
457 return bch2_alloc_to_v4_mut_inlined(trans, k);
460 struct bkey_i_alloc_v4 *
461 bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter,
465 struct bkey_i_alloc_v4 *a;
468 k = bch2_bkey_get_iter(trans, iter, BTREE_ID_alloc, pos,
469 BTREE_ITER_WITH_UPDATES|
476 a = bch2_alloc_to_v4_mut_inlined(trans, k);
477 ret = PTR_ERR_OR_ZERO(a);
482 bch2_trans_iter_exit(trans, iter);
486 static struct bpos alloc_gens_pos(struct bpos pos, unsigned *offset)
488 *offset = pos.offset & KEY_TYPE_BUCKET_GENS_MASK;
490 pos.offset >>= KEY_TYPE_BUCKET_GENS_BITS;
494 static struct bpos bucket_gens_pos_to_alloc(struct bpos pos, unsigned offset)
496 pos.offset <<= KEY_TYPE_BUCKET_GENS_BITS;
497 pos.offset += offset;
501 static unsigned alloc_gen(struct bkey_s_c k, unsigned offset)
503 return k.k->type == KEY_TYPE_bucket_gens
504 ? bkey_s_c_to_bucket_gens(k).v->gens[offset]
508 int bch2_bucket_gens_invalid(struct bch_fs *c, struct bkey_s_c k,
509 enum bkey_invalid_flags flags,
510 struct printbuf *err)
514 bkey_fsck_err_on(bkey_val_bytes(k.k) != sizeof(struct bch_bucket_gens), c, err,
515 bucket_gens_val_size_bad,
516 "bad val size (%zu != %zu)",
517 bkey_val_bytes(k.k), sizeof(struct bch_bucket_gens));
522 void bch2_bucket_gens_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
524 struct bkey_s_c_bucket_gens g = bkey_s_c_to_bucket_gens(k);
527 for (i = 0; i < ARRAY_SIZE(g.v->gens); i++) {
530 prt_printf(out, "%u", g.v->gens[i]);
534 int bch2_bucket_gens_init(struct bch_fs *c)
536 struct btree_trans *trans = bch2_trans_get(c);
537 struct btree_iter iter;
539 struct bch_alloc_v4 a;
540 struct bkey_i_bucket_gens g;
541 bool have_bucket_gens_key = false;
547 for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
548 BTREE_ITER_PREFETCH, k, ret) {
550 * Not a fsck error because this is checked/repaired by
551 * bch2_check_alloc_key() which runs later:
553 if (!bch2_dev_bucket_exists(c, k.k->p))
556 gen = bch2_alloc_to_v4(k, &a)->gen;
557 pos = alloc_gens_pos(iter.pos, &offset);
559 if (have_bucket_gens_key && bkey_cmp(iter.pos, pos)) {
560 ret = commit_do(trans, NULL, NULL,
561 BCH_TRANS_COMMIT_no_enospc,
562 bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
565 have_bucket_gens_key = false;
568 if (!have_bucket_gens_key) {
569 bkey_bucket_gens_init(&g.k_i);
571 have_bucket_gens_key = true;
574 g.v.gens[offset] = gen;
576 bch2_trans_iter_exit(trans, &iter);
578 if (have_bucket_gens_key && !ret)
579 ret = commit_do(trans, NULL, NULL,
580 BCH_TRANS_COMMIT_no_enospc,
581 bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
583 bch2_trans_put(trans);
590 int bch2_alloc_read(struct bch_fs *c)
592 struct btree_trans *trans = bch2_trans_get(c);
593 struct btree_iter iter;
598 down_read(&c->gc_lock);
600 if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_bucket_gens) {
601 const struct bch_bucket_gens *g;
604 for_each_btree_key(trans, iter, BTREE_ID_bucket_gens, POS_MIN,
605 BTREE_ITER_PREFETCH, k, ret) {
606 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
607 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
609 if (k.k->type != KEY_TYPE_bucket_gens)
612 g = bkey_s_c_to_bucket_gens(k).v;
615 * Not a fsck error because this is checked/repaired by
616 * bch2_check_alloc_key() which runs later:
618 if (!bch2_dev_exists2(c, k.k->p.inode))
621 ca = bch_dev_bkey_exists(c, k.k->p.inode);
623 for (b = max_t(u64, ca->mi.first_bucket, start);
624 b < min_t(u64, ca->mi.nbuckets, end);
626 *bucket_gen(ca, b) = g->gens[b & KEY_TYPE_BUCKET_GENS_MASK];
628 bch2_trans_iter_exit(trans, &iter);
630 struct bch_alloc_v4 a;
632 for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
633 BTREE_ITER_PREFETCH, k, ret) {
635 * Not a fsck error because this is checked/repaired by
636 * bch2_check_alloc_key() which runs later:
638 if (!bch2_dev_bucket_exists(c, k.k->p))
641 ca = bch_dev_bkey_exists(c, k.k->p.inode);
643 *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen;
645 bch2_trans_iter_exit(trans, &iter);
648 bch2_trans_put(trans);
649 up_read(&c->gc_lock);
657 /* Free space/discard btree: */
659 static int bch2_bucket_do_index(struct btree_trans *trans,
660 struct bkey_s_c alloc_k,
661 const struct bch_alloc_v4 *a,
664 struct bch_fs *c = trans->c;
665 struct bch_dev *ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
666 struct btree_iter iter;
670 enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted;
671 enum bch_bkey_type new_type = set ? KEY_TYPE_set : KEY_TYPE_deleted;
672 struct printbuf buf = PRINTBUF;
675 if (a->data_type != BCH_DATA_free &&
676 a->data_type != BCH_DATA_need_discard)
679 k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k));
684 k->k.type = new_type;
686 switch (a->data_type) {
688 btree = BTREE_ID_freespace;
689 k->k.p = alloc_freespace_pos(alloc_k.k->p, *a);
690 bch2_key_resize(&k->k, 1);
692 case BCH_DATA_need_discard:
693 btree = BTREE_ID_need_discard;
694 k->k.p = alloc_k.k->p;
700 old = bch2_bkey_get_iter(trans, &iter, btree,
701 bkey_start_pos(&k->k),
707 if (ca->mi.freespace_initialized &&
708 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info &&
709 bch2_trans_inconsistent_on(old.k->type != old_type, trans,
710 "incorrect key when %s %s:%llu:%llu:0 (got %s should be %s)\n"
712 set ? "setting" : "clearing",
713 bch2_btree_id_str(btree),
716 bch2_bkey_types[old.k->type],
717 bch2_bkey_types[old_type],
718 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
723 ret = bch2_trans_update(trans, &iter, k, 0);
725 bch2_trans_iter_exit(trans, &iter);
730 static noinline int bch2_bucket_gen_update(struct btree_trans *trans,
731 struct bpos bucket, u8 gen)
733 struct btree_iter iter;
735 struct bpos pos = alloc_gens_pos(bucket, &offset);
736 struct bkey_i_bucket_gens *g;
740 g = bch2_trans_kmalloc(trans, sizeof(*g));
741 ret = PTR_ERR_OR_ZERO(g);
745 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_bucket_gens, pos,
747 BTREE_ITER_WITH_UPDATES);
752 if (k.k->type != KEY_TYPE_bucket_gens) {
753 bkey_bucket_gens_init(&g->k_i);
756 bkey_reassemble(&g->k_i, k);
759 g->v.gens[offset] = gen;
761 ret = bch2_trans_update(trans, &iter, &g->k_i, 0);
762 bch2_trans_iter_exit(trans, &iter);
766 int bch2_trans_mark_alloc(struct btree_trans *trans,
767 enum btree_id btree_id, unsigned level,
768 struct bkey_s_c old, struct bkey_i *new,
771 struct bch_fs *c = trans->c;
772 struct bch_alloc_v4 old_a_convert, *new_a;
773 const struct bch_alloc_v4 *old_a;
774 u64 old_lru, new_lru;
778 * Deletion only happens in the device removal path, with
779 * BTREE_TRIGGER_NORUN:
781 BUG_ON(new->k.type != KEY_TYPE_alloc_v4);
783 old_a = bch2_alloc_to_v4(old, &old_a_convert);
784 new_a = &bkey_i_to_alloc_v4(new)->v;
786 new_a->data_type = alloc_data_type(*new_a, new_a->data_type);
788 if (bch2_bucket_sectors(*new_a) > bch2_bucket_sectors(*old_a)) {
789 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
790 new_a->io_time[WRITE]= max_t(u64, 1, atomic64_read(&c->io_clock[WRITE].now));
791 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true);
792 SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true);
795 if (data_type_is_empty(new_a->data_type) &&
796 BCH_ALLOC_V4_NEED_INC_GEN(new_a) &&
797 !bch2_bucket_is_open_safe(c, new->k.p.inode, new->k.p.offset)) {
799 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false);
802 if (old_a->data_type != new_a->data_type ||
803 (new_a->data_type == BCH_DATA_free &&
804 alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) {
805 ret = bch2_bucket_do_index(trans, old, old_a, false) ?:
806 bch2_bucket_do_index(trans, bkey_i_to_s_c(new), new_a, true);
811 if (new_a->data_type == BCH_DATA_cached &&
812 !new_a->io_time[READ])
813 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
815 old_lru = alloc_lru_idx_read(*old_a);
816 new_lru = alloc_lru_idx_read(*new_a);
818 if (old_lru != new_lru) {
819 ret = bch2_lru_change(trans, new->k.p.inode,
820 bucket_to_u64(new->k.p),
826 new_a->fragmentation_lru = alloc_lru_idx_fragmentation(*new_a,
827 bch_dev_bkey_exists(c, new->k.p.inode));
829 if (old_a->fragmentation_lru != new_a->fragmentation_lru) {
830 ret = bch2_lru_change(trans,
831 BCH_LRU_FRAGMENTATION_START,
832 bucket_to_u64(new->k.p),
833 old_a->fragmentation_lru, new_a->fragmentation_lru);
838 if (old_a->gen != new_a->gen) {
839 ret = bch2_bucket_gen_update(trans, new->k.p, new_a->gen);
845 * need to know if we're getting called from the invalidate path or
849 if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
850 old_a->cached_sectors) {
851 ret = bch2_update_cached_sectors_list(trans, new->k.p.inode,
852 -((s64) old_a->cached_sectors));
861 * This synthesizes deleted extents for holes, similar to BTREE_ITER_SLOTS for
862 * extents style btrees, but works on non-extents btrees:
864 static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos end, struct bkey *hole)
866 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
874 struct btree_iter iter2;
877 bch2_trans_copy_iter(&iter2, iter);
879 if (!bpos_eq(iter->path->l[0].b->key.k.p, SPOS_MAX))
880 end = bkey_min(end, bpos_nosnap_successor(iter->path->l[0].b->key.k.p));
882 end = bkey_min(end, POS(iter->pos.inode, iter->pos.offset + U32_MAX - 1));
885 * btree node min/max is a closed interval, upto takes a half
888 k = bch2_btree_iter_peek_upto(&iter2, end);
890 bch2_trans_iter_exit(iter->trans, &iter2);
892 BUG_ON(next.offset >= iter->pos.offset + U32_MAX);
900 bch2_key_resize(hole, next.offset - iter->pos.offset);
901 return (struct bkey_s_c) { hole, NULL };
905 static bool next_bucket(struct bch_fs *c, struct bpos *bucket)
910 if (bch2_dev_bucket_exists(c, *bucket))
913 if (bch2_dev_exists2(c, bucket->inode)) {
914 ca = bch_dev_bkey_exists(c, bucket->inode);
916 if (bucket->offset < ca->mi.first_bucket) {
917 bucket->offset = ca->mi.first_bucket;
926 iter = bucket->inode;
927 ca = __bch2_next_dev(c, &iter, NULL);
929 *bucket = POS(ca->dev_idx, ca->mi.first_bucket);
935 static struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, struct bkey *hole)
937 struct bch_fs *c = iter->trans->c;
940 k = bch2_get_key_or_hole(iter, POS_MAX, hole);
945 struct bpos bucket = bkey_start_pos(k.k);
947 if (!bch2_dev_bucket_exists(c, bucket)) {
948 if (!next_bucket(c, &bucket))
949 return bkey_s_c_null;
951 bch2_btree_iter_set_pos(iter, bucket);
955 if (!bch2_dev_bucket_exists(c, k.k->p)) {
956 struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode);
958 bch2_key_resize(hole, ca->mi.nbuckets - bucket.offset);
965 static noinline_for_stack
966 int bch2_check_alloc_key(struct btree_trans *trans,
967 struct bkey_s_c alloc_k,
968 struct btree_iter *alloc_iter,
969 struct btree_iter *discard_iter,
970 struct btree_iter *freespace_iter,
971 struct btree_iter *bucket_gens_iter)
973 struct bch_fs *c = trans->c;
975 struct bch_alloc_v4 a_convert;
976 const struct bch_alloc_v4 *a;
977 unsigned discard_key_type, freespace_key_type;
978 unsigned gens_offset;
980 struct printbuf buf = PRINTBUF;
983 if (fsck_err_on(!bch2_dev_bucket_exists(c, alloc_k.k->p), c,
984 alloc_key_to_missing_dev_bucket,
985 "alloc key for invalid device:bucket %llu:%llu",
986 alloc_k.k->p.inode, alloc_k.k->p.offset))
987 return bch2_btree_delete_at(trans, alloc_iter, 0);
989 ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
990 if (!ca->mi.freespace_initialized)
993 a = bch2_alloc_to_v4(alloc_k, &a_convert);
995 discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0;
996 bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p);
997 k = bch2_btree_iter_peek_slot(discard_iter);
1002 if (k.k->type != discard_key_type &&
1003 (c->opts.reconstruct_alloc ||
1004 fsck_err(c, need_discard_key_wrong,
1005 "incorrect key in need_discard btree (got %s should be %s)\n"
1007 bch2_bkey_types[k.k->type],
1008 bch2_bkey_types[discard_key_type],
1009 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1010 struct bkey_i *update =
1011 bch2_trans_kmalloc(trans, sizeof(*update));
1013 ret = PTR_ERR_OR_ZERO(update);
1017 bkey_init(&update->k);
1018 update->k.type = discard_key_type;
1019 update->k.p = discard_iter->pos;
1021 ret = bch2_trans_update(trans, discard_iter, update, 0);
1026 freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0;
1027 bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a));
1028 k = bch2_btree_iter_peek_slot(freespace_iter);
1033 if (k.k->type != freespace_key_type &&
1034 (c->opts.reconstruct_alloc ||
1035 fsck_err(c, freespace_key_wrong,
1036 "incorrect key in freespace btree (got %s should be %s)\n"
1038 bch2_bkey_types[k.k->type],
1039 bch2_bkey_types[freespace_key_type],
1040 (printbuf_reset(&buf),
1041 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1042 struct bkey_i *update =
1043 bch2_trans_kmalloc(trans, sizeof(*update));
1045 ret = PTR_ERR_OR_ZERO(update);
1049 bkey_init(&update->k);
1050 update->k.type = freespace_key_type;
1051 update->k.p = freespace_iter->pos;
1052 bch2_key_resize(&update->k, 1);
1054 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1059 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(alloc_k.k->p, &gens_offset));
1060 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1065 if (a->gen != alloc_gen(k, gens_offset) &&
1066 (c->opts.reconstruct_alloc ||
1067 fsck_err(c, bucket_gens_key_wrong,
1068 "incorrect gen in bucket_gens btree (got %u should be %u)\n"
1070 alloc_gen(k, gens_offset), a->gen,
1071 (printbuf_reset(&buf),
1072 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1073 struct bkey_i_bucket_gens *g =
1074 bch2_trans_kmalloc(trans, sizeof(*g));
1076 ret = PTR_ERR_OR_ZERO(g);
1080 if (k.k->type == KEY_TYPE_bucket_gens) {
1081 bkey_reassemble(&g->k_i, k);
1083 bkey_bucket_gens_init(&g->k_i);
1084 g->k.p = alloc_gens_pos(alloc_k.k->p, &gens_offset);
1087 g->v.gens[gens_offset] = a->gen;
1089 ret = bch2_trans_update(trans, bucket_gens_iter, &g->k_i, 0);
1095 printbuf_exit(&buf);
1099 static noinline_for_stack
1100 int bch2_check_alloc_hole_freespace(struct btree_trans *trans,
1103 struct btree_iter *freespace_iter)
1105 struct bch_fs *c = trans->c;
1108 struct printbuf buf = PRINTBUF;
1111 ca = bch_dev_bkey_exists(c, start.inode);
1112 if (!ca->mi.freespace_initialized)
1115 bch2_btree_iter_set_pos(freespace_iter, start);
1117 k = bch2_btree_iter_peek_slot(freespace_iter);
1122 *end = bkey_min(k.k->p, *end);
1124 if (k.k->type != KEY_TYPE_set &&
1125 (c->opts.reconstruct_alloc ||
1126 fsck_err(c, freespace_hole_missing,
1127 "hole in alloc btree missing in freespace btree\n"
1128 " device %llu buckets %llu-%llu",
1129 freespace_iter->pos.inode,
1130 freespace_iter->pos.offset,
1132 struct bkey_i *update =
1133 bch2_trans_kmalloc(trans, sizeof(*update));
1135 ret = PTR_ERR_OR_ZERO(update);
1139 bkey_init(&update->k);
1140 update->k.type = KEY_TYPE_set;
1141 update->k.p = freespace_iter->pos;
1142 bch2_key_resize(&update->k,
1143 min_t(u64, U32_MAX, end->offset -
1144 freespace_iter->pos.offset));
1146 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1152 printbuf_exit(&buf);
1156 static noinline_for_stack
1157 int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans,
1160 struct btree_iter *bucket_gens_iter)
1162 struct bch_fs *c = trans->c;
1164 struct printbuf buf = PRINTBUF;
1165 unsigned i, gens_offset, gens_end_offset;
1168 if (c->sb.version < bcachefs_metadata_version_bucket_gens)
1171 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(start, &gens_offset));
1173 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1178 if (bkey_cmp(alloc_gens_pos(start, &gens_offset),
1179 alloc_gens_pos(*end, &gens_end_offset)))
1180 gens_end_offset = KEY_TYPE_BUCKET_GENS_NR;
1182 if (k.k->type == KEY_TYPE_bucket_gens) {
1183 struct bkey_i_bucket_gens g;
1184 bool need_update = false;
1186 bkey_reassemble(&g.k_i, k);
1188 for (i = gens_offset; i < gens_end_offset; i++) {
1189 if (fsck_err_on(g.v.gens[i], c,
1190 bucket_gens_hole_wrong,
1191 "hole in alloc btree at %llu:%llu with nonzero gen in bucket_gens btree (%u)",
1192 bucket_gens_pos_to_alloc(k.k->p, i).inode,
1193 bucket_gens_pos_to_alloc(k.k->p, i).offset,
1201 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1203 ret = PTR_ERR_OR_ZERO(u);
1207 memcpy(u, &g, sizeof(g));
1209 ret = bch2_trans_update(trans, bucket_gens_iter, u, 0);
1215 *end = bkey_min(*end, bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0));
1218 printbuf_exit(&buf);
1222 static noinline_for_stack int bch2_check_discard_freespace_key(struct btree_trans *trans,
1223 struct btree_iter *iter)
1225 struct bch_fs *c = trans->c;
1226 struct btree_iter alloc_iter;
1227 struct bkey_s_c alloc_k;
1228 struct bch_alloc_v4 a_convert;
1229 const struct bch_alloc_v4 *a;
1232 enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard
1233 ? BCH_DATA_need_discard
1235 struct printbuf buf = PRINTBUF;
1239 pos.offset &= ~(~0ULL << 56);
1240 genbits = iter->pos.offset & (~0ULL << 56);
1242 alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc, pos, 0);
1243 ret = bkey_err(alloc_k);
1247 if (fsck_err_on(!bch2_dev_bucket_exists(c, pos), c,
1248 need_discard_freespace_key_to_invalid_dev_bucket,
1249 "entry in %s btree for nonexistant dev:bucket %llu:%llu",
1250 bch2_btree_id_str(iter->btree_id), pos.inode, pos.offset))
1253 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1255 if (fsck_err_on(a->data_type != state ||
1256 (state == BCH_DATA_free &&
1257 genbits != alloc_freespace_genbits(*a)), c,
1258 need_discard_freespace_key_bad,
1259 "%s\n incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)",
1260 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
1261 bch2_btree_id_str(iter->btree_id),
1264 a->data_type == state,
1265 genbits >> 56, alloc_freespace_genbits(*a) >> 56))
1269 set_btree_iter_dontneed(&alloc_iter);
1270 bch2_trans_iter_exit(trans, &alloc_iter);
1271 printbuf_exit(&buf);
1274 ret = bch2_btree_delete_extent_at(trans, iter,
1275 iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0) ?:
1276 bch2_trans_commit(trans, NULL, NULL,
1277 BCH_TRANS_COMMIT_no_enospc);
1282 * We've already checked that generation numbers in the bucket_gens btree are
1283 * valid for buckets that exist; this just checks for keys for nonexistent
1286 static noinline_for_stack
1287 int bch2_check_bucket_gens_key(struct btree_trans *trans,
1288 struct btree_iter *iter,
1291 struct bch_fs *c = trans->c;
1292 struct bkey_i_bucket_gens g;
1294 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
1295 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
1297 bool need_update = false, dev_exists;
1298 struct printbuf buf = PRINTBUF;
1301 BUG_ON(k.k->type != KEY_TYPE_bucket_gens);
1302 bkey_reassemble(&g.k_i, k);
1304 /* if no bch_dev, skip out whether we repair or not */
1305 dev_exists = bch2_dev_exists2(c, k.k->p.inode);
1307 if (fsck_err_on(!dev_exists, c,
1308 bucket_gens_to_invalid_dev,
1309 "bucket_gens key for invalid device:\n %s",
1310 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1311 ret = bch2_btree_delete_at(trans, iter, 0);
1316 ca = bch_dev_bkey_exists(c, k.k->p.inode);
1317 if (fsck_err_on(end <= ca->mi.first_bucket ||
1318 start >= ca->mi.nbuckets, c,
1319 bucket_gens_to_invalid_buckets,
1320 "bucket_gens key for invalid buckets:\n %s",
1321 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1322 ret = bch2_btree_delete_at(trans, iter, 0);
1326 for (b = start; b < ca->mi.first_bucket; b++)
1327 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1328 bucket_gens_nonzero_for_invalid_buckets,
1329 "bucket_gens key has nonzero gen for invalid bucket")) {
1330 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1334 for (b = ca->mi.nbuckets; b < end; b++)
1335 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1336 bucket_gens_nonzero_for_invalid_buckets,
1337 "bucket_gens key has nonzero gen for invalid bucket")) {
1338 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1343 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1345 ret = PTR_ERR_OR_ZERO(u);
1349 memcpy(u, &g, sizeof(g));
1350 ret = bch2_trans_update(trans, iter, u, 0);
1354 printbuf_exit(&buf);
1358 int bch2_check_alloc_info(struct bch_fs *c)
1360 struct btree_trans *trans = bch2_trans_get(c);
1361 struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter;
1366 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS_MIN,
1367 BTREE_ITER_PREFETCH);
1368 bch2_trans_iter_init(trans, &discard_iter, BTREE_ID_need_discard, POS_MIN,
1369 BTREE_ITER_PREFETCH);
1370 bch2_trans_iter_init(trans, &freespace_iter, BTREE_ID_freespace, POS_MIN,
1371 BTREE_ITER_PREFETCH);
1372 bch2_trans_iter_init(trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN,
1373 BTREE_ITER_PREFETCH);
1378 bch2_trans_begin(trans);
1380 k = bch2_get_key_or_real_bucket_hole(&iter, &hole);
1389 next = bpos_nosnap_successor(k.k->p);
1391 ret = bch2_check_alloc_key(trans,
1401 ret = bch2_check_alloc_hole_freespace(trans,
1402 bkey_start_pos(k.k),
1405 bch2_check_alloc_hole_bucket_gens(trans,
1406 bkey_start_pos(k.k),
1413 ret = bch2_trans_commit(trans, NULL, NULL,
1414 BCH_TRANS_COMMIT_no_enospc);
1418 bch2_btree_iter_set_pos(&iter, next);
1420 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1425 bch2_trans_iter_exit(trans, &bucket_gens_iter);
1426 bch2_trans_iter_exit(trans, &freespace_iter);
1427 bch2_trans_iter_exit(trans, &discard_iter);
1428 bch2_trans_iter_exit(trans, &iter);
1433 ret = for_each_btree_key2(trans, iter,
1434 BTREE_ID_need_discard, POS_MIN,
1435 BTREE_ITER_PREFETCH, k,
1436 bch2_check_discard_freespace_key(trans, &iter));
1440 bch2_trans_iter_init(trans, &iter, BTREE_ID_freespace, POS_MIN,
1441 BTREE_ITER_PREFETCH);
1443 bch2_trans_begin(trans);
1444 k = bch2_btree_iter_peek(&iter);
1448 ret = bkey_err(k) ?:
1449 bch2_check_discard_freespace_key(trans, &iter);
1450 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1455 struct printbuf buf = PRINTBUF;
1456 bch2_bkey_val_to_text(&buf, c, k);
1458 bch_err(c, "while checking %s", buf.buf);
1459 printbuf_exit(&buf);
1463 bch2_btree_iter_set_pos(&iter, bpos_nosnap_successor(iter.pos));
1465 bch2_trans_iter_exit(trans, &iter);
1469 ret = for_each_btree_key_commit(trans, iter,
1470 BTREE_ID_bucket_gens, POS_MIN,
1471 BTREE_ITER_PREFETCH, k,
1472 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1473 bch2_check_bucket_gens_key(trans, &iter, k));
1475 bch2_trans_put(trans);
1481 static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
1482 struct btree_iter *alloc_iter)
1484 struct bch_fs *c = trans->c;
1485 struct btree_iter lru_iter;
1486 struct bch_alloc_v4 a_convert;
1487 const struct bch_alloc_v4 *a;
1488 struct bkey_s_c alloc_k, lru_k;
1489 struct printbuf buf = PRINTBUF;
1492 alloc_k = bch2_btree_iter_peek(alloc_iter);
1496 ret = bkey_err(alloc_k);
1500 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1502 if (a->data_type != BCH_DATA_cached)
1505 if (fsck_err_on(!a->io_time[READ], c,
1506 alloc_key_cached_but_read_time_zero,
1507 "cached bucket with read_time 0\n"
1509 (printbuf_reset(&buf),
1510 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1511 struct bkey_i_alloc_v4 *a_mut =
1512 bch2_alloc_to_v4_mut(trans, alloc_k);
1513 ret = PTR_ERR_OR_ZERO(a_mut);
1517 a_mut->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now);
1518 ret = bch2_trans_update(trans, alloc_iter,
1519 &a_mut->k_i, BTREE_TRIGGER_NORUN);
1526 lru_k = bch2_bkey_get_iter(trans, &lru_iter, BTREE_ID_lru,
1527 lru_pos(alloc_k.k->p.inode,
1528 bucket_to_u64(alloc_k.k->p),
1529 a->io_time[READ]), 0);
1530 ret = bkey_err(lru_k);
1534 if (fsck_err_on(lru_k.k->type != KEY_TYPE_set, c,
1535 alloc_key_to_missing_lru_entry,
1536 "missing lru entry\n"
1538 (printbuf_reset(&buf),
1539 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1540 ret = bch2_lru_set(trans,
1542 bucket_to_u64(alloc_k.k->p),
1549 bch2_trans_iter_exit(trans, &lru_iter);
1550 printbuf_exit(&buf);
1554 int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
1556 struct btree_iter iter;
1559 int ret = bch2_trans_run(c,
1560 for_each_btree_key_commit(trans, iter, BTREE_ID_alloc,
1561 POS_MIN, BTREE_ITER_PREFETCH, k,
1562 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1563 bch2_check_alloc_to_lru_ref(trans, &iter)));
1568 static int bch2_discard_one_bucket(struct btree_trans *trans,
1569 struct btree_iter *need_discard_iter,
1570 struct bpos *discard_pos_done,
1573 u64 *need_journal_commit,
1576 struct bch_fs *c = trans->c;
1577 struct bpos pos = need_discard_iter->pos;
1578 struct btree_iter iter = { NULL };
1581 struct bkey_i_alloc_v4 *a;
1582 struct printbuf buf = PRINTBUF;
1585 ca = bch_dev_bkey_exists(c, pos.inode);
1586 if (!percpu_ref_tryget(&ca->io_ref)) {
1587 bch2_btree_iter_set_pos(need_discard_iter, POS(pos.inode + 1, 0));
1591 if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) {
1596 if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
1597 c->journal.flushed_seq_ondisk,
1598 pos.inode, pos.offset)) {
1599 (*need_journal_commit)++;
1603 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc,
1604 need_discard_iter->pos,
1610 a = bch2_alloc_to_v4_mut(trans, k);
1611 ret = PTR_ERR_OR_ZERO(a);
1615 if (BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
1617 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1621 if (a->v.journal_seq > c->journal.flushed_seq_ondisk) {
1622 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1623 bch2_trans_inconsistent(trans,
1624 "clearing need_discard but journal_seq %llu > flushed_seq %llu\n"
1627 c->journal.flushed_seq_ondisk,
1628 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1634 if (a->v.data_type != BCH_DATA_need_discard) {
1635 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1636 bch2_trans_inconsistent(trans,
1637 "bucket incorrectly set in need_discard btree\n"
1639 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1646 if (!bkey_eq(*discard_pos_done, iter.pos) &&
1647 ca->mi.discard && !c->opts.nochanges) {
1649 * This works without any other locks because this is the only
1650 * thread that removes items from the need_discard tree
1652 bch2_trans_unlock(trans);
1653 blkdev_issue_discard(ca->disk_sb.bdev,
1654 k.k->p.offset * ca->mi.bucket_size,
1657 *discard_pos_done = iter.pos;
1659 ret = bch2_trans_relock_notrace(trans);
1664 SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
1665 a->v.data_type = alloc_data_type(a->v, a->v.data_type);
1667 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
1668 bch2_trans_commit(trans, NULL, NULL,
1669 BCH_WATERMARK_btree|
1670 BCH_TRANS_COMMIT_no_enospc);
1674 count_event(c, bucket_discard);
1678 bch2_trans_iter_exit(trans, &iter);
1679 percpu_ref_put(&ca->io_ref);
1680 printbuf_exit(&buf);
1684 static void bch2_do_discards_work(struct work_struct *work)
1686 struct bch_fs *c = container_of(work, struct bch_fs, discard_work);
1687 struct btree_iter iter;
1689 u64 seen = 0, open = 0, need_journal_commit = 0, discarded = 0;
1690 struct bpos discard_pos_done = POS_MAX;
1694 * We're doing the commit in bch2_discard_one_bucket instead of using
1695 * for_each_btree_key_commit() so that we can increment counters after
1696 * successful commit:
1698 ret = bch2_trans_run(c,
1699 for_each_btree_key2(trans, iter,
1700 BTREE_ID_need_discard, POS_MIN, 0, k,
1701 bch2_discard_one_bucket(trans, &iter, &discard_pos_done,
1704 &need_journal_commit,
1707 if (need_journal_commit * 2 > seen)
1708 bch2_journal_flush_async(&c->journal, NULL);
1710 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1712 trace_discard_buckets(c, seen, open, need_journal_commit, discarded,
1716 void bch2_do_discards(struct bch_fs *c)
1718 if (bch2_write_ref_tryget(c, BCH_WRITE_REF_discard) &&
1719 !queue_work(c->write_ref_wq, &c->discard_work))
1720 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1723 static int invalidate_one_bucket(struct btree_trans *trans,
1724 struct btree_iter *lru_iter,
1725 struct bkey_s_c lru_k,
1726 s64 *nr_to_invalidate)
1728 struct bch_fs *c = trans->c;
1729 struct btree_iter alloc_iter = { NULL };
1730 struct bkey_i_alloc_v4 *a = NULL;
1731 struct printbuf buf = PRINTBUF;
1732 struct bpos bucket = u64_to_bucket(lru_k.k->p.offset);
1733 unsigned cached_sectors;
1736 if (*nr_to_invalidate <= 0)
1739 if (!bch2_dev_bucket_exists(c, bucket)) {
1740 prt_str(&buf, "lru entry points to invalid bucket");
1744 if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset))
1747 a = bch2_trans_start_alloc_update(trans, &alloc_iter, bucket);
1748 ret = PTR_ERR_OR_ZERO(a);
1752 /* We expect harmless races here due to the btree write buffer: */
1753 if (lru_pos_time(lru_iter->pos) != alloc_lru_idx_read(a->v))
1756 BUG_ON(a->v.data_type != BCH_DATA_cached);
1758 if (!a->v.cached_sectors)
1759 bch_err(c, "invalidating empty bucket, confused");
1761 cached_sectors = a->v.cached_sectors;
1763 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1766 a->v.dirty_sectors = 0;
1767 a->v.cached_sectors = 0;
1768 a->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now);
1769 a->v.io_time[WRITE] = atomic64_read(&c->io_clock[WRITE].now);
1771 ret = bch2_trans_update(trans, &alloc_iter, &a->k_i,
1772 BTREE_TRIGGER_BUCKET_INVALIDATE) ?:
1773 bch2_trans_commit(trans, NULL, NULL,
1774 BCH_WATERMARK_btree|
1775 BCH_TRANS_COMMIT_no_enospc);
1779 trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors);
1780 --*nr_to_invalidate;
1782 bch2_trans_iter_exit(trans, &alloc_iter);
1783 printbuf_exit(&buf);
1786 prt_str(&buf, "\n lru key: ");
1787 bch2_bkey_val_to_text(&buf, c, lru_k);
1789 prt_str(&buf, "\n lru entry: ");
1790 bch2_lru_pos_to_text(&buf, lru_iter->pos);
1792 prt_str(&buf, "\n alloc key: ");
1794 bch2_bpos_to_text(&buf, bucket);
1796 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i));
1798 bch_err(c, "%s", buf.buf);
1799 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_lrus) {
1800 bch2_inconsistent_error(c);
1807 static void bch2_do_invalidates_work(struct work_struct *work)
1809 struct bch_fs *c = container_of(work, struct bch_fs, invalidate_work);
1811 struct btree_trans *trans = bch2_trans_get(c);
1812 struct btree_iter iter;
1817 ret = bch2_btree_write_buffer_tryflush(trans);
1821 for_each_member_device(ca, c, i) {
1822 s64 nr_to_invalidate =
1823 should_invalidate_buckets(ca, bch2_dev_usage_read(ca));
1825 ret = for_each_btree_key2_upto(trans, iter, BTREE_ID_lru,
1826 lru_pos(ca->dev_idx, 0, 0),
1827 lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX),
1828 BTREE_ITER_INTENT, k,
1829 invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate));
1832 percpu_ref_put(&ca->ref);
1837 bch2_trans_put(trans);
1838 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1841 void bch2_do_invalidates(struct bch_fs *c)
1843 if (bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate) &&
1844 !queue_work(c->write_ref_wq, &c->invalidate_work))
1845 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1848 int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca,
1849 u64 bucket_start, u64 bucket_end)
1851 struct btree_trans *trans = bch2_trans_get(c);
1852 struct btree_iter iter;
1855 struct bpos end = POS(ca->dev_idx, bucket_end);
1856 struct bch_member *m;
1857 unsigned long last_updated = jiffies;
1860 BUG_ON(bucket_start > bucket_end);
1861 BUG_ON(bucket_end > ca->mi.nbuckets);
1863 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
1864 POS(ca->dev_idx, max_t(u64, ca->mi.first_bucket, bucket_start)),
1865 BTREE_ITER_PREFETCH);
1867 * Scan the alloc btree for every bucket on @ca, and add buckets to the
1868 * freespace/need_discard/need_gc_gens btrees as needed:
1871 if (last_updated + HZ * 10 < jiffies) {
1872 bch_info(ca, "%s: currently at %llu/%llu",
1873 __func__, iter.pos.offset, ca->mi.nbuckets);
1874 last_updated = jiffies;
1877 bch2_trans_begin(trans);
1879 if (bkey_ge(iter.pos, end)) {
1884 k = bch2_get_key_or_hole(&iter, end, &hole);
1891 * We process live keys in the alloc btree one at a
1894 struct bch_alloc_v4 a_convert;
1895 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert);
1897 ret = bch2_bucket_do_index(trans, k, a, true) ?:
1898 bch2_trans_commit(trans, NULL, NULL,
1899 BCH_TRANS_COMMIT_no_enospc);
1903 bch2_btree_iter_advance(&iter);
1905 struct bkey_i *freespace;
1907 freespace = bch2_trans_kmalloc(trans, sizeof(*freespace));
1908 ret = PTR_ERR_OR_ZERO(freespace);
1912 bkey_init(&freespace->k);
1913 freespace->k.type = KEY_TYPE_set;
1914 freespace->k.p = k.k->p;
1915 freespace->k.size = k.k->size;
1917 ret = bch2_btree_insert_trans(trans, BTREE_ID_freespace, freespace, 0) ?:
1918 bch2_trans_commit(trans, NULL, NULL,
1919 BCH_TRANS_COMMIT_no_enospc);
1923 bch2_btree_iter_set_pos(&iter, k.k->p);
1926 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1932 bch2_trans_iter_exit(trans, &iter);
1933 bch2_trans_put(trans);
1936 bch_err_msg(ca, ret, "initializing free space");
1940 mutex_lock(&c->sb_lock);
1941 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1942 SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true);
1943 mutex_unlock(&c->sb_lock);
1948 int bch2_fs_freespace_init(struct bch_fs *c)
1953 bool doing_init = false;
1956 * We can crash during the device add path, so we need to check this on
1960 for_each_member_device(ca, c, i) {
1961 if (ca->mi.freespace_initialized)
1965 bch_info(c, "initializing freespace");
1969 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1971 percpu_ref_put(&ca->ref);
1978 mutex_lock(&c->sb_lock);
1979 bch2_write_super(c);
1980 mutex_unlock(&c->sb_lock);
1981 bch_verbose(c, "done initializing freespace");
1987 /* Bucket IO clocks: */
1989 int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
1990 size_t bucket_nr, int rw)
1992 struct bch_fs *c = trans->c;
1993 struct btree_iter iter;
1994 struct bkey_i_alloc_v4 *a;
1998 a = bch2_trans_start_alloc_update(trans, &iter, POS(dev, bucket_nr));
1999 ret = PTR_ERR_OR_ZERO(a);
2003 now = atomic64_read(&c->io_clock[rw].now);
2004 if (a->v.io_time[rw] == now)
2007 a->v.io_time[rw] = now;
2009 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
2010 bch2_trans_commit(trans, NULL, NULL, 0);
2012 bch2_trans_iter_exit(trans, &iter);
2016 /* Startup/shutdown (ro/rw): */
2018 void bch2_recalc_capacity(struct bch_fs *c)
2021 u64 capacity = 0, reserved_sectors = 0, gc_reserve;
2022 unsigned bucket_size_max = 0;
2023 unsigned long ra_pages = 0;
2026 lockdep_assert_held(&c->state_lock);
2028 for_each_online_member(ca, c, i) {
2029 struct backing_dev_info *bdi = ca->disk_sb.bdev->bd_disk->bdi;
2031 ra_pages += bdi->ra_pages;
2034 bch2_set_ra_pages(c, ra_pages);
2036 for_each_rw_member(ca, c, i) {
2037 u64 dev_reserve = 0;
2040 * We need to reserve buckets (from the number
2041 * of currently available buckets) against
2042 * foreground writes so that mainly copygc can
2043 * make forward progress.
2045 * We need enough to refill the various reserves
2046 * from scratch - copygc will use its entire
2047 * reserve all at once, then run against when
2048 * its reserve is refilled (from the formerly
2049 * available buckets).
2051 * This reserve is just used when considering if
2052 * allocations for foreground writes must wait -
2053 * not -ENOSPC calculations.
2056 dev_reserve += ca->nr_btree_reserve * 2;
2057 dev_reserve += ca->mi.nbuckets >> 6; /* copygc reserve */
2059 dev_reserve += 1; /* btree write point */
2060 dev_reserve += 1; /* copygc write point */
2061 dev_reserve += 1; /* rebalance write point */
2063 dev_reserve *= ca->mi.bucket_size;
2065 capacity += bucket_to_sector(ca, ca->mi.nbuckets -
2066 ca->mi.first_bucket);
2068 reserved_sectors += dev_reserve * 2;
2070 bucket_size_max = max_t(unsigned, bucket_size_max,
2071 ca->mi.bucket_size);
2074 gc_reserve = c->opts.gc_reserve_bytes
2075 ? c->opts.gc_reserve_bytes >> 9
2076 : div64_u64(capacity * c->opts.gc_reserve_percent, 100);
2078 reserved_sectors = max(gc_reserve, reserved_sectors);
2080 reserved_sectors = min(reserved_sectors, capacity);
2082 c->capacity = capacity - reserved_sectors;
2084 c->bucket_size_max = bucket_size_max;
2086 /* Wake up case someone was waiting for buckets */
2087 closure_wake_up(&c->freelist_wait);
2090 u64 bch2_min_rw_member_capacity(struct bch_fs *c)
2096 for_each_rw_member(ca, c, i)
2097 ret = min(ret, ca->mi.nbuckets * ca->mi.bucket_size);
2101 static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca)
2103 struct open_bucket *ob;
2106 for (ob = c->open_buckets;
2107 ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
2109 spin_lock(&ob->lock);
2110 if (ob->valid && !ob->on_partial_list &&
2111 ob->dev == ca->dev_idx)
2113 spin_unlock(&ob->lock);
2119 /* device goes ro: */
2120 void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca)
2124 /* First, remove device from allocation groups: */
2126 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2127 clear_bit(ca->dev_idx, c->rw_devs[i].d);
2130 * Capacity is calculated based off of devices in allocation groups:
2132 bch2_recalc_capacity(c);
2134 bch2_open_buckets_stop(c, ca, false);
2137 * Wake up threads that were blocked on allocation, so they can notice
2138 * the device can no longer be removed and the capacity has changed:
2140 closure_wake_up(&c->freelist_wait);
2143 * journal_res_get() can block waiting for free space in the journal -
2144 * it needs to notice there may not be devices to allocate from anymore:
2146 wake_up(&c->journal.wait);
2148 /* Now wait for any in flight writes: */
2150 closure_wait_event(&c->open_buckets_wait,
2151 !bch2_dev_has_open_write_point(c, ca));
2154 /* device goes rw: */
2155 void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca)
2159 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2160 if (ca->mi.data_allowed & (1 << i))
2161 set_bit(ca->dev_idx, c->rw_devs[i].d);
2164 void bch2_fs_allocator_background_init(struct bch_fs *c)
2166 spin_lock_init(&c->freelist_lock);
2167 INIT_WORK(&c->discard_work, bch2_do_discards_work);
2168 INIT_WORK(&c->invalidate_work, bch2_do_invalidates_work);