X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Falloc_background.c;h=c7be6afe89553b154ff8ba1d2899908026c5451a;hb=e773e86495534ede8efc0f3059cf0939c374a925;hp=702d6ec55a9310909a63d320f113cac03bef044c;hpb=5a5a6c25a93e83e0b1e659d7808e7636d0b770ed;p=bcachefs-tools-debian diff --git a/libbcachefs/alloc_background.c b/libbcachefs/alloc_background.c index 702d6ec..c7be6af 100644 --- a/libbcachefs/alloc_background.c +++ b/libbcachefs/alloc_background.c @@ -9,6 +9,7 @@ #include "btree_update.h" #include "btree_update_interior.h" #include "btree_gc.h" +#include "btree_write_buffer.h" #include "buckets.h" #include "buckets_waiting_for_journal.h" #include "clock.h" @@ -17,6 +18,7 @@ #include "error.h" #include "lru.h" #include "recovery.h" +#include "trace.h" #include "varint.h" #include @@ -26,7 +28,6 @@ #include #include #include -#include /* Persistent alloc info: */ @@ -78,36 +79,6 @@ static inline u64 alloc_field_v1_get(const struct bch_alloc *a, return v; } -static inline void alloc_field_v1_put(struct bkey_i_alloc *a, void **p, - unsigned field, u64 v) -{ - unsigned bytes = BCH_ALLOC_V1_FIELD_BYTES[field]; - - if (!v) - return; - - a->v.fields |= 1 << field; - - switch (bytes) { - case 1: - *((u8 *) *p) = v; - break; - case 2: - *((__le16 *) *p) = cpu_to_le16(v); - break; - case 4: - *((__le32 *) *p) = cpu_to_le32(v); - break; - case 8: - *((__le64 *) *p) = cpu_to_le64(v); - break; - default: - BUG(); - } - - *p += bytes; -} - static void bch2_alloc_unpack_v1(struct bkey_alloc_unpacked *out, struct bkey_s_c k) { @@ -221,143 +192,106 @@ static unsigned bch_alloc_v1_val_u64s(const struct bch_alloc *a) return DIV_ROUND_UP(bytes, sizeof(u64)); } -int bch2_alloc_v1_invalid(const struct bch_fs *c, struct bkey_s_c k, - int rw, struct printbuf *err) +int bch2_alloc_v1_invalid(struct bch_fs *c, struct bkey_s_c k, + enum bkey_invalid_flags flags, + struct printbuf *err) { struct bkey_s_c_alloc a = bkey_s_c_to_alloc(k); + int ret = 0; /* allow for unknown fields */ - if (bkey_val_u64s(a.k) < bch_alloc_v1_val_u64s(a.v)) { - prt_printf(err, "incorrect value size (%zu < %u)", - bkey_val_u64s(a.k), bch_alloc_v1_val_u64s(a.v)); - return -BCH_ERR_invalid_bkey; - } - - return 0; + bkey_fsck_err_on(bkey_val_u64s(a.k) < bch_alloc_v1_val_u64s(a.v), c, err, + alloc_v1_val_size_bad, + "incorrect value size (%zu < %u)", + bkey_val_u64s(a.k), bch_alloc_v1_val_u64s(a.v)); +fsck_err: + return ret; } -int bch2_alloc_v2_invalid(const struct bch_fs *c, struct bkey_s_c k, - int rw, struct printbuf *err) +int bch2_alloc_v2_invalid(struct bch_fs *c, struct bkey_s_c k, + enum bkey_invalid_flags flags, + struct printbuf *err) { struct bkey_alloc_unpacked u; + int ret = 0; - if (bch2_alloc_unpack_v2(&u, k)) { - prt_printf(err, "unpack error"); - return -BCH_ERR_invalid_bkey; - } - - return 0; + bkey_fsck_err_on(bch2_alloc_unpack_v2(&u, k), c, err, + alloc_v2_unpack_error, + "unpack error"); +fsck_err: + return ret; } -int bch2_alloc_v3_invalid(const struct bch_fs *c, struct bkey_s_c k, - int rw, struct printbuf *err) +int bch2_alloc_v3_invalid(struct bch_fs *c, struct bkey_s_c k, + enum bkey_invalid_flags flags, + struct printbuf *err) { struct bkey_alloc_unpacked u; + int ret = 0; - if (bch2_alloc_unpack_v3(&u, k)) { - prt_printf(err, "unpack error"); - return -BCH_ERR_invalid_bkey; - } - - return 0; + bkey_fsck_err_on(bch2_alloc_unpack_v3(&u, k), c, err, + alloc_v2_unpack_error, + "unpack error"); +fsck_err: + return ret; } -int bch2_alloc_v4_invalid(const struct bch_fs *c, struct bkey_s_c k, - int rw, struct printbuf *err) +int bch2_alloc_v4_invalid(struct bch_fs *c, struct bkey_s_c k, + enum bkey_invalid_flags flags, struct printbuf *err) { struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k); + int ret = 0; - if (alloc_v4_u64s(a.v) != bkey_val_u64s(k.k)) { - prt_printf(err, "bad val size (%lu != %u)", - bkey_val_u64s(k.k), alloc_v4_u64s(a.v)); - return -BCH_ERR_invalid_bkey; - } - - if (!BCH_ALLOC_V4_BACKPOINTERS_START(a.v) && - BCH_ALLOC_V4_NR_BACKPOINTERS(a.v)) { - prt_printf(err, "invalid backpointers_start"); - return -BCH_ERR_invalid_bkey; - } - - /* - * XXX this is wrong, we'll be checking updates that happened from - * before BCH_FS_CHECK_BACKPOINTERS_DONE - */ - if (rw == WRITE && test_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags)) { - unsigned i, bp_len = 0; + bkey_fsck_err_on(alloc_v4_u64s(a.v) > bkey_val_u64s(k.k), c, err, + alloc_v4_val_size_bad, + "bad val size (%u > %zu)", + alloc_v4_u64s(a.v), bkey_val_u64s(k.k)); - for (i = 0; i < BCH_ALLOC_V4_NR_BACKPOINTERS(a.v); i++) - bp_len += alloc_v4_backpointers_c(a.v)[i].bucket_len; + bkey_fsck_err_on(!BCH_ALLOC_V4_BACKPOINTERS_START(a.v) && + BCH_ALLOC_V4_NR_BACKPOINTERS(a.v), c, err, + alloc_v4_backpointers_start_bad, + "invalid backpointers_start"); - if (bp_len > a.v->dirty_sectors) { - prt_printf(err, "too many backpointers"); - return -BCH_ERR_invalid_bkey; - } - } - - if (rw == WRITE) { - if (alloc_data_type(*a.v, a.v->data_type) != a.v->data_type) { - prt_printf(err, "invalid data type (got %u should be %u)", - a.v->data_type, alloc_data_type(*a.v, a.v->data_type)); - return -BCH_ERR_invalid_bkey; - } - - switch (a.v->data_type) { - case BCH_DATA_free: - case BCH_DATA_need_gc_gens: - case BCH_DATA_need_discard: - if (a.v->dirty_sectors || - a.v->cached_sectors || - a.v->stripe) { - prt_printf(err, "empty data type free but have data"); - return -BCH_ERR_invalid_bkey; - } - break; - case BCH_DATA_sb: - case BCH_DATA_journal: - case BCH_DATA_btree: - case BCH_DATA_user: - case BCH_DATA_parity: - if (!a.v->dirty_sectors) { - prt_printf(err, "data_type %s but dirty_sectors==0", - bch2_data_types[a.v->data_type]); - return -BCH_ERR_invalid_bkey; - } - break; - case BCH_DATA_cached: - if (!a.v->cached_sectors || - a.v->dirty_sectors || - a.v->stripe) { - prt_printf(err, "data type inconsistency"); - return -BCH_ERR_invalid_bkey; - } + bkey_fsck_err_on(alloc_data_type(*a.v, a.v->data_type) != a.v->data_type, c, err, + alloc_key_data_type_bad, + "invalid data type (got %u should be %u)", + a.v->data_type, alloc_data_type(*a.v, a.v->data_type)); - if (!a.v->io_time[READ] && - test_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags)) { - prt_printf(err, "cached bucket with read_time == 0"); - return -BCH_ERR_invalid_bkey; - } - break; - case BCH_DATA_stripe: - if (!a.v->stripe) { - prt_printf(err, "data_type %s but stripe==0", - bch2_data_types[a.v->data_type]); - return -BCH_ERR_invalid_bkey; - } - break; - } + switch (a.v->data_type) { + case BCH_DATA_free: + case BCH_DATA_need_gc_gens: + case BCH_DATA_need_discard: + bkey_fsck_err_on(bch2_bucket_sectors(*a.v) || a.v->stripe, + c, err, alloc_key_empty_but_have_data, + "empty data type free but have data"); + break; + case BCH_DATA_sb: + case BCH_DATA_journal: + case BCH_DATA_btree: + case BCH_DATA_user: + case BCH_DATA_parity: + bkey_fsck_err_on(!bch2_bucket_sectors_dirty(*a.v), + c, err, alloc_key_dirty_sectors_0, + "data_type %s but dirty_sectors==0", + bch2_data_type_str(a.v->data_type)); + break; + case BCH_DATA_cached: + bkey_fsck_err_on(!a.v->cached_sectors || + bch2_bucket_sectors_dirty(*a.v) || + a.v->stripe, + c, err, alloc_key_cached_inconsistency, + "data type inconsistency"); + + bkey_fsck_err_on(!a.v->io_time[READ] && + c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_to_lru_refs, + c, err, alloc_key_cached_but_read_time_zero, + "cached bucket with read_time == 0"); + break; + case BCH_DATA_stripe: + break; } - - return 0; -} - -static inline u64 swab40(u64 x) -{ - return (((x & 0x00000000ffULL) << 32)| - ((x & 0x000000ff00ULL) << 16)| - ((x & 0x0000ff0000ULL) >> 0)| - ((x & 0x00ff000000ULL) >> 16)| - ((x & 0xff00000000ULL) >> 32)); +fsck_err: + return ret; } void bch2_alloc_v4_swab(struct bkey_s k) @@ -373,6 +307,7 @@ void bch2_alloc_v4_swab(struct bkey_s k) a->io_time[1] = swab64(a->io_time[1]); a->stripe = swab32(a->stripe); a->nr_external_backpointers = swab32(a->nr_external_backpointers); + a->fragmentation_lru = swab64(a->fragmentation_lru); bps = alloc_v4_backpointers(a); for (bp = bps; bp < bps + BCH_ALLOC_V4_NR_BACKPOINTERS(a); bp++) { @@ -386,14 +321,12 @@ void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c { struct bch_alloc_v4 _a; const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a); - const struct bch_backpointer *bps; - unsigned i; prt_newline(out); printbuf_indent_add(out, 2); - prt_printf(out, "gen %u oldest_gen %u data_type %s", - a->gen, a->oldest_gen, bch2_data_types[a->data_type]); + prt_printf(out, "gen %u oldest_gen %u data_type ", a->gen, a->oldest_gen); + bch2_prt_data_type(out, a->data_type); prt_newline(out); prt_printf(out, "journal_seq %llu", a->journal_seq); prt_newline(out); @@ -413,33 +346,27 @@ void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c prt_newline(out); prt_printf(out, "io_time[WRITE] %llu", a->io_time[WRITE]); prt_newline(out); - prt_printf(out, "backpointers: %llu", BCH_ALLOC_V4_NR_BACKPOINTERS(a)); - printbuf_indent_add(out, 2); - - bps = alloc_v4_backpointers_c(a); - for (i = 0; i < BCH_ALLOC_V4_NR_BACKPOINTERS(a); i++) { - prt_newline(out); - bch2_backpointer_to_text(out, &bps[i]); - } - - printbuf_indent_sub(out, 4); + prt_printf(out, "fragmentation %llu", a->fragmentation_lru); + prt_newline(out); + prt_printf(out, "bp_start %llu", BCH_ALLOC_V4_BACKPOINTERS_START(a)); + printbuf_indent_sub(out, 2); } void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out) { if (k.k->type == KEY_TYPE_alloc_v4) { - int d; + void *src, *dst; *out = *bkey_s_c_to_alloc_v4(k).v; - d = (int) BCH_ALLOC_V4_U64s - - (int) (BCH_ALLOC_V4_BACKPOINTERS_START(out) ?: BCH_ALLOC_V4_U64s_V0); - if (unlikely(d > 0)) { - memset((u64 *) out + BCH_ALLOC_V4_BACKPOINTERS_START(out), - 0, - d * sizeof(u64)); - SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s); - } + src = alloc_v4_backpointers(out); + SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s); + dst = alloc_v4_backpointers(out); + + if (src < dst) + memset(src, 0, dst - src); + + SET_BCH_ALLOC_V4_NR_BACKPOINTERS(out, 0); } else { struct bkey_alloc_unpacked u = bch2_alloc_unpack(k); @@ -465,20 +392,13 @@ static noinline struct bkey_i_alloc_v4 * __bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k) { struct bkey_i_alloc_v4 *ret; - unsigned bytes = k.k->type == KEY_TYPE_alloc_v4 - ? bkey_bytes(k.k) - : sizeof(struct bkey_i_alloc_v4); - /* - * Reserve space for one more backpointer here: - * Not sketchy at doing it this way, nope... - */ - ret = bch2_trans_kmalloc(trans, bytes + sizeof(struct bch_backpointer)); + ret = bch2_trans_kmalloc(trans, max(bkey_bytes(k.k), sizeof(struct bkey_i_alloc_v4))); if (IS_ERR(ret)) return ret; if (k.k->type == KEY_TYPE_alloc_v4) { - struct bch_backpointer *src, *dst; + void *src, *dst; bkey_reassemble(&ret->k_i, k); @@ -486,9 +406,10 @@ __bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k) SET_BCH_ALLOC_V4_BACKPOINTERS_START(&ret->v, BCH_ALLOC_V4_U64s); dst = alloc_v4_backpointers(&ret->v); - memmove(dst, src, BCH_ALLOC_V4_NR_BACKPOINTERS(&ret->v) * - sizeof(struct bch_backpointer)); - memset(src, 0, dst - src); + if (src < dst) + memset(src, 0, dst - src); + + SET_BCH_ALLOC_V4_NR_BACKPOINTERS(&ret->v, 0); set_alloc_v4_u64s(ret); } else { bkey_alloc_v4_init(&ret->k_i); @@ -500,20 +421,12 @@ __bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k) static inline struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut_inlined(struct btree_trans *trans, struct bkey_s_c k) { + struct bkey_s_c_alloc_v4 a; + if (likely(k.k->type == KEY_TYPE_alloc_v4) && - BCH_ALLOC_V4_BACKPOINTERS_START(bkey_s_c_to_alloc_v4(k).v) == BCH_ALLOC_V4_U64s) { - /* - * Reserve space for one more backpointer here: - * Not sketchy at doing it this way, nope... - */ - struct bkey_i_alloc_v4 *ret = - bch2_trans_kmalloc_nomemzero(trans, bkey_bytes(k.k) + sizeof(struct bch_backpointer)); - if (!IS_ERR(ret)) { - bkey_reassemble(&ret->k_i, k); - memset((void *) ret + bkey_bytes(k.k), 0, sizeof(struct bch_backpointer)); - } - return ret; - } + ((a = bkey_s_c_to_alloc_v4(k), true) && + BCH_ALLOC_V4_NR_BACKPOINTERS(a.v) == 0)) + return bch2_bkey_make_mut_noupdate_typed(trans, k, alloc_v4); return __bch2_alloc_to_v4_mut(trans, k); } @@ -531,14 +444,13 @@ bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter struct bkey_i_alloc_v4 *a; int ret; - bch2_trans_iter_init(trans, iter, BTREE_ID_alloc, pos, + k = bch2_bkey_get_iter(trans, iter, BTREE_ID_alloc, pos, BTREE_ITER_WITH_UPDATES| BTREE_ITER_CACHED| BTREE_ITER_INTENT); - k = bch2_btree_iter_peek_slot(iter); ret = bkey_err(k); if (unlikely(ret)) - goto err; + return ERR_PTR(ret); a = bch2_alloc_to_v4_mut_inlined(trans, k); ret = PTR_ERR_OR_ZERO(a); @@ -550,19 +462,63 @@ err: return ERR_PTR(ret); } -int bch2_alloc_read(struct bch_fs *c) +static struct bpos alloc_gens_pos(struct bpos pos, unsigned *offset) { - struct btree_trans trans; - struct btree_iter iter; - struct bkey_s_c k; - struct bch_alloc_v4 a; - struct bch_dev *ca; - int ret; + *offset = pos.offset & KEY_TYPE_BUCKET_GENS_MASK; - bch2_trans_init(&trans, c, 0, 0); + pos.offset >>= KEY_TYPE_BUCKET_GENS_BITS; + return pos; +} - for_each_btree_key(&trans, iter, BTREE_ID_alloc, POS_MIN, - BTREE_ITER_PREFETCH, k, ret) { +static struct bpos bucket_gens_pos_to_alloc(struct bpos pos, unsigned offset) +{ + pos.offset <<= KEY_TYPE_BUCKET_GENS_BITS; + pos.offset += offset; + return pos; +} + +static unsigned alloc_gen(struct bkey_s_c k, unsigned offset) +{ + return k.k->type == KEY_TYPE_bucket_gens + ? bkey_s_c_to_bucket_gens(k).v->gens[offset] + : 0; +} + +int bch2_bucket_gens_invalid(struct bch_fs *c, struct bkey_s_c k, + enum bkey_invalid_flags flags, + struct printbuf *err) +{ + int ret = 0; + + bkey_fsck_err_on(bkey_val_bytes(k.k) != sizeof(struct bch_bucket_gens), c, err, + bucket_gens_val_size_bad, + "bad val size (%zu != %zu)", + bkey_val_bytes(k.k), sizeof(struct bch_bucket_gens)); +fsck_err: + return ret; +} + +void bch2_bucket_gens_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k) +{ + struct bkey_s_c_bucket_gens g = bkey_s_c_to_bucket_gens(k); + unsigned i; + + for (i = 0; i < ARRAY_SIZE(g.v->gens); i++) { + if (i) + prt_char(out, ' '); + prt_printf(out, "%u", g.v->gens[i]); + } +} + +int bch2_bucket_gens_init(struct bch_fs *c) +{ + struct btree_trans *trans = bch2_trans_get(c); + struct bkey_i_bucket_gens g; + bool have_bucket_gens_key = false; + int ret; + + ret = for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN, + BTREE_ITER_PREFETCH, k, ({ /* * Not a fsck error because this is checked/repaired by * bch2_check_alloc_key() which runs later: @@ -570,17 +526,96 @@ int bch2_alloc_read(struct bch_fs *c) if (!bch2_dev_bucket_exists(c, k.k->p)) continue; - ca = bch_dev_bkey_exists(c, k.k->p.inode); + struct bch_alloc_v4 a; + u8 gen = bch2_alloc_to_v4(k, &a)->gen; + unsigned offset; + struct bpos pos = alloc_gens_pos(iter.pos, &offset); - *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen; - } - bch2_trans_iter_exit(&trans, &iter); + if (have_bucket_gens_key && bkey_cmp(iter.pos, pos)) { + ret = commit_do(trans, NULL, NULL, + BCH_TRANS_COMMIT_no_enospc, + bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0)); + if (ret) + break; + have_bucket_gens_key = false; + } - bch2_trans_exit(&trans); + if (!have_bucket_gens_key) { + bkey_bucket_gens_init(&g.k_i); + g.k.p = pos; + have_bucket_gens_key = true; + } - if (ret) - bch_err(c, "error reading alloc info: %s", bch2_err_str(ret)); + g.v.gens[offset] = gen; + 0; + })); + + if (have_bucket_gens_key && !ret) + ret = commit_do(trans, NULL, NULL, + BCH_TRANS_COMMIT_no_enospc, + bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0)); + bch2_trans_put(trans); + + bch_err_fn(c, ret); + return ret; +} + +int bch2_alloc_read(struct bch_fs *c) +{ + struct btree_trans *trans = bch2_trans_get(c); + int ret; + + down_read(&c->gc_lock); + + if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_bucket_gens) { + ret = for_each_btree_key(trans, iter, BTREE_ID_bucket_gens, POS_MIN, + BTREE_ITER_PREFETCH, k, ({ + u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset; + u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset; + + if (k.k->type != KEY_TYPE_bucket_gens) + continue; + + const struct bch_bucket_gens *g = bkey_s_c_to_bucket_gens(k).v; + + /* + * Not a fsck error because this is checked/repaired by + * bch2_check_alloc_key() which runs later: + */ + if (!bch2_dev_exists2(c, k.k->p.inode)) + continue; + + struct bch_dev *ca = bch_dev_bkey_exists(c, k.k->p.inode); + + for (u64 b = max_t(u64, ca->mi.first_bucket, start); + b < min_t(u64, ca->mi.nbuckets, end); + b++) + *bucket_gen(ca, b) = g->gens[b & KEY_TYPE_BUCKET_GENS_MASK]; + 0; + })); + } else { + ret = for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN, + BTREE_ITER_PREFETCH, k, ({ + /* + * Not a fsck error because this is checked/repaired by + * bch2_check_alloc_key() which runs later: + */ + if (!bch2_dev_bucket_exists(c, k.k->p)) + continue; + + struct bch_dev *ca = bch_dev_bkey_exists(c, k.k->p.inode); + + struct bch_alloc_v4 a; + *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen; + 0; + })); + } + + bch2_trans_put(trans); + up_read(&c->gc_lock); + + bch_err_fn(c, ret); return ret; } @@ -627,20 +662,22 @@ static int bch2_bucket_do_index(struct btree_trans *trans, return 0; } - bch2_trans_iter_init(trans, &iter, btree, + old = bch2_bkey_get_iter(trans, &iter, btree, bkey_start_pos(&k->k), BTREE_ITER_INTENT); - old = bch2_btree_iter_peek_slot(&iter); ret = bkey_err(old); if (ret) - goto err; + return ret; if (ca->mi.freespace_initialized && + c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info && bch2_trans_inconsistent_on(old.k->type != old_type, trans, - "incorrect key when %s %s btree (got %s should be %s)\n" + "incorrect key when %s %s:%llu:%llu:0 (got %s should be %s)\n" " for %s", set ? "setting" : "clearing", - bch2_btree_ids[btree], + bch2_btree_id_str(btree), + iter.pos.inode, + iter.pos.offset, bch2_bkey_types[old.k->type], bch2_bkey_types[old_type], (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) { @@ -655,97 +692,347 @@ err: return ret; } -int bch2_trans_mark_alloc(struct btree_trans *trans, - enum btree_id btree_id, unsigned level, - struct bkey_s_c old, struct bkey_i *new, - unsigned flags) +static noinline int bch2_bucket_gen_update(struct btree_trans *trans, + struct bpos bucket, u8 gen) +{ + struct btree_iter iter; + unsigned offset; + struct bpos pos = alloc_gens_pos(bucket, &offset); + struct bkey_i_bucket_gens *g; + struct bkey_s_c k; + int ret; + + g = bch2_trans_kmalloc(trans, sizeof(*g)); + ret = PTR_ERR_OR_ZERO(g); + if (ret) + return ret; + + k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_bucket_gens, pos, + BTREE_ITER_INTENT| + BTREE_ITER_WITH_UPDATES); + ret = bkey_err(k); + if (ret) + return ret; + + if (k.k->type != KEY_TYPE_bucket_gens) { + bkey_bucket_gens_init(&g->k_i); + g->k.p = iter.pos; + } else { + bkey_reassemble(&g->k_i, k); + } + + g->v.gens[offset] = gen; + + ret = bch2_trans_update(trans, &iter, &g->k_i, 0); + bch2_trans_iter_exit(trans, &iter); + return ret; +} + +int bch2_trigger_alloc(struct btree_trans *trans, + enum btree_id btree, unsigned level, + struct bkey_s_c old, struct bkey_s new, + unsigned flags) { struct bch_fs *c = trans->c; - struct bch_alloc_v4 old_a_convert, *new_a; - const struct bch_alloc_v4 *old_a; - u64 old_lru, new_lru; int ret = 0; - /* - * Deletion only happens in the device removal path, with - * BTREE_TRIGGER_NORUN: - */ - BUG_ON(new->k.type != KEY_TYPE_alloc_v4); + if (bch2_trans_inconsistent_on(!bch2_dev_bucket_exists(c, new.k->p), trans, + "alloc key for invalid device or bucket")) + return -EIO; - old_a = bch2_alloc_to_v4(old, &old_a_convert); - new_a = &bkey_i_to_alloc_v4(new)->v; + struct bch_dev *ca = bch_dev_bkey_exists(c, new.k->p.inode); - new_a->data_type = alloc_data_type(*new_a, new_a->data_type); + struct bch_alloc_v4 old_a_convert; + const struct bch_alloc_v4 *old_a = bch2_alloc_to_v4(old, &old_a_convert); - if (new_a->dirty_sectors > old_a->dirty_sectors || - new_a->cached_sectors > old_a->cached_sectors) { - new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now)); - new_a->io_time[WRITE]= max_t(u64, 1, atomic64_read(&c->io_clock[WRITE].now)); - SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true); - SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true); - } + if (flags & BTREE_TRIGGER_TRANSACTIONAL) { + struct bch_alloc_v4 *new_a = bkey_s_to_alloc_v4(new).v; + + new_a->data_type = alloc_data_type(*new_a, new_a->data_type); + + if (bch2_bucket_sectors(*new_a) > bch2_bucket_sectors(*old_a)) { + new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now)); + new_a->io_time[WRITE]= max_t(u64, 1, atomic64_read(&c->io_clock[WRITE].now)); + SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true); + SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true); + } + + if (data_type_is_empty(new_a->data_type) && + BCH_ALLOC_V4_NEED_INC_GEN(new_a) && + !bch2_bucket_is_open_safe(c, new.k->p.inode, new.k->p.offset)) { + new_a->gen++; + SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false); + } - if (data_type_is_empty(new_a->data_type) && - BCH_ALLOC_V4_NEED_INC_GEN(new_a) && - !bch2_bucket_is_open_safe(c, new->k.p.inode, new->k.p.offset)) { - new_a->gen++; - SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false); + if (old_a->data_type != new_a->data_type || + (new_a->data_type == BCH_DATA_free && + alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) { + ret = bch2_bucket_do_index(trans, old, old_a, false) ?: + bch2_bucket_do_index(trans, new.s_c, new_a, true); + if (ret) + return ret; + } + + if (new_a->data_type == BCH_DATA_cached && + !new_a->io_time[READ]) + new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now)); + + u64 old_lru = alloc_lru_idx_read(*old_a); + u64 new_lru = alloc_lru_idx_read(*new_a); + if (old_lru != new_lru) { + ret = bch2_lru_change(trans, new.k->p.inode, + bucket_to_u64(new.k->p), + old_lru, new_lru); + if (ret) + return ret; + } + + new_a->fragmentation_lru = alloc_lru_idx_fragmentation(*new_a, + bch_dev_bkey_exists(c, new.k->p.inode)); + if (old_a->fragmentation_lru != new_a->fragmentation_lru) { + ret = bch2_lru_change(trans, + BCH_LRU_FRAGMENTATION_START, + bucket_to_u64(new.k->p), + old_a->fragmentation_lru, new_a->fragmentation_lru); + if (ret) + return ret; + } + + if (old_a->gen != new_a->gen) { + ret = bch2_bucket_gen_update(trans, new.k->p, new_a->gen); + if (ret) + return ret; + } + + /* + * need to know if we're getting called from the invalidate path or + * not: + */ + + if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) && + old_a->cached_sectors) { + ret = bch2_update_cached_sectors_list(trans, new.k->p.inode, + -((s64) old_a->cached_sectors)); + if (ret) + return ret; + } } - if (old_a->data_type != new_a->data_type || - (new_a->data_type == BCH_DATA_free && - alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) { - ret = bch2_bucket_do_index(trans, old, old_a, false) ?: - bch2_bucket_do_index(trans, bkey_i_to_s_c(new), new_a, true); - if (ret) - return ret; + if ((flags & BTREE_TRIGGER_ATOMIC) && (flags & BTREE_TRIGGER_INSERT)) { + struct bch_alloc_v4 *new_a = bkey_s_to_alloc_v4(new).v; + u64 journal_seq = trans->journal_res.seq; + u64 bucket_journal_seq = new_a->journal_seq; + + if ((flags & BTREE_TRIGGER_INSERT) && + data_type_is_empty(old_a->data_type) != + data_type_is_empty(new_a->data_type) && + new.k->type == KEY_TYPE_alloc_v4) { + struct bch_alloc_v4 *v = bkey_s_to_alloc_v4(new).v; + + /* + * If the btree updates referring to a bucket weren't flushed + * before the bucket became empty again, then the we don't have + * to wait on a journal flush before we can reuse the bucket: + */ + v->journal_seq = bucket_journal_seq = + data_type_is_empty(new_a->data_type) && + (journal_seq == v->journal_seq || + bch2_journal_noflush_seq(&c->journal, v->journal_seq)) + ? 0 : journal_seq; + } + + if (!data_type_is_empty(old_a->data_type) && + data_type_is_empty(new_a->data_type) && + bucket_journal_seq) { + ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal, + c->journal.flushed_seq_ondisk, + new.k->p.inode, new.k->p.offset, + bucket_journal_seq); + if (ret) { + bch2_fs_fatal_error(c, + "error setting bucket_needs_journal_commit: %i", ret); + return ret; + } + } + + percpu_down_read(&c->mark_lock); + if (new_a->gen != old_a->gen) + *bucket_gen(ca, new.k->p.offset) = new_a->gen; + + bch2_dev_usage_update(c, ca, old_a, new_a, journal_seq, false); + percpu_up_read(&c->mark_lock); + +#define eval_state(_a, expr) ({ const struct bch_alloc_v4 *a = _a; expr; }) +#define statechange(expr) !eval_state(old_a, expr) && eval_state(new_a, expr) +#define bucket_flushed(a) (!a->journal_seq || a->journal_seq <= c->journal.flushed_seq_ondisk) + + if (statechange(a->data_type == BCH_DATA_free && + bucket_flushed(a))) + closure_wake_up(&c->freelist_wait); + + if (statechange(a->data_type == BCH_DATA_need_discard && + bucket_flushed(a)) && + !bch2_bucket_is_open(c, new.k->p.inode, new.k->p.offset)) + bch2_do_discards(c); + + if (statechange(a->data_type == BCH_DATA_cached) && + !bch2_bucket_is_open(c, new.k->p.inode, new.k->p.offset) && + should_invalidate_buckets(ca, bch2_dev_usage_read(ca))) + bch2_do_invalidates(c); + + if (statechange(a->data_type == BCH_DATA_need_gc_gens)) + bch2_do_gc_gens(c); } - if (new_a->data_type == BCH_DATA_cached && - !new_a->io_time[READ]) - new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now)); + if ((flags & BTREE_TRIGGER_GC) && + (flags & BTREE_TRIGGER_BUCKET_INVALIDATE)) { + struct bch_alloc_v4 new_a_convert; + const struct bch_alloc_v4 *new_a = bch2_alloc_to_v4(new.s_c, &new_a_convert); - old_lru = alloc_lru_idx(*old_a); - new_lru = alloc_lru_idx(*new_a); + percpu_down_read(&c->mark_lock); + struct bucket *g = gc_bucket(ca, new.k->p.offset); - if (old_lru != new_lru) { - ret = bch2_lru_change(trans, new->k.p.inode, new->k.p.offset, - old_lru, &new_lru, old); - if (ret) - return ret; + bucket_lock(g); + + g->gen_valid = 1; + g->gen = new_a->gen; + g->data_type = new_a->data_type; + g->stripe = new_a->stripe; + g->stripe_redundancy = new_a->stripe_redundancy; + g->dirty_sectors = new_a->dirty_sectors; + g->cached_sectors = new_a->cached_sectors; - if (new_a->data_type == BCH_DATA_cached) - new_a->io_time[READ] = new_lru; + bucket_unlock(g); + percpu_up_read(&c->mark_lock); } return 0; } -static int bch2_check_alloc_key(struct btree_trans *trans, - struct btree_iter *alloc_iter, - struct btree_iter *discard_iter, - struct btree_iter *freespace_iter) +/* + * This synthesizes deleted extents for holes, similar to BTREE_ITER_SLOTS for + * extents style btrees, but works on non-extents btrees: + */ +static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos end, struct bkey *hole) +{ + struct bkey_s_c k = bch2_btree_iter_peek_slot(iter); + + if (bkey_err(k)) + return k; + + if (k.k->type) { + return k; + } else { + struct btree_iter iter2; + struct bpos next; + + bch2_trans_copy_iter(&iter2, iter); + + struct btree_path *path = btree_iter_path(iter->trans, iter); + if (!bpos_eq(path->l[0].b->key.k.p, SPOS_MAX)) + end = bkey_min(end, bpos_nosnap_successor(path->l[0].b->key.k.p)); + + end = bkey_min(end, POS(iter->pos.inode, iter->pos.offset + U32_MAX - 1)); + + /* + * btree node min/max is a closed interval, upto takes a half + * open interval: + */ + k = bch2_btree_iter_peek_upto(&iter2, end); + next = iter2.pos; + bch2_trans_iter_exit(iter->trans, &iter2); + + BUG_ON(next.offset >= iter->pos.offset + U32_MAX); + + if (bkey_err(k)) + return k; + + bkey_init(hole); + hole->p = iter->pos; + + bch2_key_resize(hole, next.offset - iter->pos.offset); + return (struct bkey_s_c) { hole, NULL }; + } +} + +static bool next_bucket(struct bch_fs *c, struct bpos *bucket) +{ + struct bch_dev *ca; + + if (bch2_dev_bucket_exists(c, *bucket)) + return true; + + if (bch2_dev_exists2(c, bucket->inode)) { + ca = bch_dev_bkey_exists(c, bucket->inode); + + if (bucket->offset < ca->mi.first_bucket) { + bucket->offset = ca->mi.first_bucket; + return true; + } + + bucket->inode++; + bucket->offset = 0; + } + + rcu_read_lock(); + ca = __bch2_next_dev_idx(c, bucket->inode, NULL); + if (ca) + *bucket = POS(ca->dev_idx, ca->mi.first_bucket); + rcu_read_unlock(); + + return ca != NULL; +} + +static struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, struct bkey *hole) +{ + struct bch_fs *c = iter->trans->c; + struct bkey_s_c k; +again: + k = bch2_get_key_or_hole(iter, POS_MAX, hole); + if (bkey_err(k)) + return k; + + if (!k.k->type) { + struct bpos bucket = bkey_start_pos(k.k); + + if (!bch2_dev_bucket_exists(c, bucket)) { + if (!next_bucket(c, &bucket)) + return bkey_s_c_null; + + bch2_btree_iter_set_pos(iter, bucket); + goto again; + } + + if (!bch2_dev_bucket_exists(c, k.k->p)) { + struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode); + + bch2_key_resize(hole, ca->mi.nbuckets - bucket.offset); + } + } + + return k; +} + +static noinline_for_stack +int bch2_check_alloc_key(struct btree_trans *trans, + struct bkey_s_c alloc_k, + struct btree_iter *alloc_iter, + struct btree_iter *discard_iter, + struct btree_iter *freespace_iter, + struct btree_iter *bucket_gens_iter) { struct bch_fs *c = trans->c; struct bch_dev *ca; struct bch_alloc_v4 a_convert; const struct bch_alloc_v4 *a; unsigned discard_key_type, freespace_key_type; - struct bkey_s_c alloc_k, k; + unsigned gens_offset; + struct bkey_s_c k; struct printbuf buf = PRINTBUF; int ret; - alloc_k = bch2_dev_bucket_exists(c, alloc_iter->pos) - ? bch2_btree_iter_peek_slot(alloc_iter) - : bch2_btree_iter_peek(alloc_iter); - if (!alloc_k.k) - return 1; - - ret = bkey_err(alloc_k); - if (ret) - return ret; - if (fsck_err_on(!bch2_dev_bucket_exists(c, alloc_k.k->p), c, + alloc_key_to_missing_dev_bucket, "alloc key for invalid device:bucket %llu:%llu", alloc_k.k->p.inode, alloc_k.k->p.offset)) return bch2_btree_delete_at(trans, alloc_iter, 0); @@ -756,14 +1043,8 @@ static int bch2_check_alloc_key(struct btree_trans *trans, a = bch2_alloc_to_v4(alloc_k, &a_convert); - discard_key_type = a->data_type == BCH_DATA_need_discard - ? KEY_TYPE_set : 0; - freespace_key_type = a->data_type == BCH_DATA_free - ? KEY_TYPE_set : 0; - + discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0; bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p); - bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a)); - k = bch2_btree_iter_peek_slot(discard_iter); ret = bkey_err(k); if (ret) @@ -771,7 +1052,8 @@ static int bch2_check_alloc_key(struct btree_trans *trans, if (k.k->type != discard_key_type && (c->opts.reconstruct_alloc || - fsck_err(c, "incorrect key in need_discard btree (got %s should be %s)\n" + fsck_err(c, need_discard_key_wrong, + "incorrect key in need_discard btree (got %s should be %s)\n" " %s", bch2_bkey_types[k.k->type], bch2_bkey_types[discard_key_type], @@ -783,28 +1065,121 @@ static int bch2_check_alloc_key(struct btree_trans *trans, if (ret) goto err; - bkey_init(&update->k); - update->k.type = discard_key_type; - update->k.p = discard_iter->pos; + bkey_init(&update->k); + update->k.type = discard_key_type; + update->k.p = discard_iter->pos; + + ret = bch2_trans_update(trans, discard_iter, update, 0); + if (ret) + goto err; + } + + freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0; + bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a)); + k = bch2_btree_iter_peek_slot(freespace_iter); + ret = bkey_err(k); + if (ret) + goto err; + + if (k.k->type != freespace_key_type && + (c->opts.reconstruct_alloc || + fsck_err(c, freespace_key_wrong, + "incorrect key in freespace btree (got %s should be %s)\n" + " %s", + bch2_bkey_types[k.k->type], + bch2_bkey_types[freespace_key_type], + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) { + struct bkey_i *update = + bch2_trans_kmalloc(trans, sizeof(*update)); + + ret = PTR_ERR_OR_ZERO(update); + if (ret) + goto err; + + bkey_init(&update->k); + update->k.type = freespace_key_type; + update->k.p = freespace_iter->pos; + bch2_key_resize(&update->k, 1); + + ret = bch2_trans_update(trans, freespace_iter, update, 0); + if (ret) + goto err; + } + + bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(alloc_k.k->p, &gens_offset)); + k = bch2_btree_iter_peek_slot(bucket_gens_iter); + ret = bkey_err(k); + if (ret) + goto err; + + if (a->gen != alloc_gen(k, gens_offset) && + (c->opts.reconstruct_alloc || + fsck_err(c, bucket_gens_key_wrong, + "incorrect gen in bucket_gens btree (got %u should be %u)\n" + " %s", + alloc_gen(k, gens_offset), a->gen, + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) { + struct bkey_i_bucket_gens *g = + bch2_trans_kmalloc(trans, sizeof(*g)); + + ret = PTR_ERR_OR_ZERO(g); + if (ret) + goto err; + + if (k.k->type == KEY_TYPE_bucket_gens) { + bkey_reassemble(&g->k_i, k); + } else { + bkey_bucket_gens_init(&g->k_i); + g->k.p = alloc_gens_pos(alloc_k.k->p, &gens_offset); + } + + g->v.gens[gens_offset] = a->gen; + + ret = bch2_trans_update(trans, bucket_gens_iter, &g->k_i, 0); + if (ret) + goto err; + } +err: +fsck_err: + printbuf_exit(&buf); + return ret; +} + +static noinline_for_stack +int bch2_check_alloc_hole_freespace(struct btree_trans *trans, + struct bpos start, + struct bpos *end, + struct btree_iter *freespace_iter) +{ + struct bch_fs *c = trans->c; + struct bch_dev *ca; + struct bkey_s_c k; + struct printbuf buf = PRINTBUF; + int ret; + + ca = bch_dev_bkey_exists(c, start.inode); + if (!ca->mi.freespace_initialized) + return 0; - ret = bch2_trans_update(trans, discard_iter, update, 0); - if (ret) - goto err; - } + bch2_btree_iter_set_pos(freespace_iter, start); k = bch2_btree_iter_peek_slot(freespace_iter); ret = bkey_err(k); if (ret) goto err; - if (k.k->type != freespace_key_type && + *end = bkey_min(k.k->p, *end); + + if (k.k->type != KEY_TYPE_set && (c->opts.reconstruct_alloc || - fsck_err(c, "incorrect key in freespace btree (got %s should be %s)\n" - " %s", - bch2_bkey_types[k.k->type], - bch2_bkey_types[freespace_key_type], - (printbuf_reset(&buf), - bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) { + fsck_err(c, freespace_hole_missing, + "hole in alloc btree missing in freespace btree\n" + " device %llu buckets %llu-%llu", + freespace_iter->pos.inode, + freespace_iter->pos.offset, + end->offset))) { struct bkey_i *update = bch2_trans_kmalloc(trans, sizeof(*update)); @@ -813,9 +1188,11 @@ static int bch2_check_alloc_key(struct btree_trans *trans, goto err; bkey_init(&update->k); - update->k.type = freespace_key_type; + update->k.type = KEY_TYPE_set; update->k.p = freespace_iter->pos; - bch2_key_resize(&update->k, 1); + bch2_key_resize(&update->k, + min_t(u64, U32_MAX, end->offset - + freespace_iter->pos.offset)); ret = bch2_trans_update(trans, freespace_iter, update, 0); if (ret) @@ -827,8 +1204,71 @@ fsck_err: return ret; } -static int bch2_check_discard_freespace_key(struct btree_trans *trans, - struct btree_iter *iter) +static noinline_for_stack +int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans, + struct bpos start, + struct bpos *end, + struct btree_iter *bucket_gens_iter) +{ + struct bch_fs *c = trans->c; + struct bkey_s_c k; + struct printbuf buf = PRINTBUF; + unsigned i, gens_offset, gens_end_offset; + int ret; + + bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(start, &gens_offset)); + + k = bch2_btree_iter_peek_slot(bucket_gens_iter); + ret = bkey_err(k); + if (ret) + goto err; + + if (bkey_cmp(alloc_gens_pos(start, &gens_offset), + alloc_gens_pos(*end, &gens_end_offset))) + gens_end_offset = KEY_TYPE_BUCKET_GENS_NR; + + if (k.k->type == KEY_TYPE_bucket_gens) { + struct bkey_i_bucket_gens g; + bool need_update = false; + + bkey_reassemble(&g.k_i, k); + + for (i = gens_offset; i < gens_end_offset; i++) { + if (fsck_err_on(g.v.gens[i], c, + bucket_gens_hole_wrong, + "hole in alloc btree at %llu:%llu with nonzero gen in bucket_gens btree (%u)", + bucket_gens_pos_to_alloc(k.k->p, i).inode, + bucket_gens_pos_to_alloc(k.k->p, i).offset, + g.v.gens[i])) { + g.v.gens[i] = 0; + need_update = true; + } + } + + if (need_update) { + struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g)); + + ret = PTR_ERR_OR_ZERO(u); + if (ret) + goto err; + + memcpy(u, &g, sizeof(g)); + + ret = bch2_trans_update(trans, bucket_gens_iter, u, 0); + if (ret) + goto err; + } + } + + *end = bkey_min(*end, bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0)); +err: +fsck_err: + printbuf_exit(&buf); + return ret; +} + +static noinline_for_stack int bch2_check_discard_freespace_key(struct btree_trans *trans, + struct btree_iter *iter) { struct bch_fs *c = trans->c; struct btree_iter alloc_iter; @@ -847,88 +1287,242 @@ static int bch2_check_discard_freespace_key(struct btree_trans *trans, pos.offset &= ~(~0ULL << 56); genbits = iter->pos.offset & (~0ULL << 56); - bch2_trans_iter_init(trans, &alloc_iter, BTREE_ID_alloc, pos, 0); + alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc, pos, 0); + ret = bkey_err(alloc_k); + if (ret) + return ret; if (fsck_err_on(!bch2_dev_bucket_exists(c, pos), c, + need_discard_freespace_key_to_invalid_dev_bucket, "entry in %s btree for nonexistant dev:bucket %llu:%llu", - bch2_btree_ids[iter->btree_id], pos.inode, pos.offset)) + bch2_btree_id_str(iter->btree_id), pos.inode, pos.offset)) goto delete; - alloc_k = bch2_btree_iter_peek_slot(&alloc_iter); - ret = bkey_err(alloc_k); - if (ret) - goto err; - a = bch2_alloc_to_v4(alloc_k, &a_convert); if (fsck_err_on(a->data_type != state || (state == BCH_DATA_free && genbits != alloc_freespace_genbits(*a)), c, - "%s\n incorrectly set in %s index (free %u, genbits %llu should be %llu)", + need_discard_freespace_key_bad, + "%s\n incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)", (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf), - bch2_btree_ids[iter->btree_id], + bch2_btree_id_str(iter->btree_id), + iter->pos.inode, + iter->pos.offset, a->data_type == state, genbits >> 56, alloc_freespace_genbits(*a) >> 56)) goto delete; out: -err: fsck_err: + set_btree_iter_dontneed(&alloc_iter); bch2_trans_iter_exit(trans, &alloc_iter); printbuf_exit(&buf); return ret; delete: - ret = bch2_btree_delete_extent_at(trans, iter, - iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0); + ret = bch2_btree_delete_extent_at(trans, iter, + iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0) ?: + bch2_trans_commit(trans, NULL, NULL, + BCH_TRANS_COMMIT_no_enospc); goto out; } +/* + * We've already checked that generation numbers in the bucket_gens btree are + * valid for buckets that exist; this just checks for keys for nonexistent + * buckets. + */ +static noinline_for_stack +int bch2_check_bucket_gens_key(struct btree_trans *trans, + struct btree_iter *iter, + struct bkey_s_c k) +{ + struct bch_fs *c = trans->c; + struct bkey_i_bucket_gens g; + struct bch_dev *ca; + u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset; + u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset; + u64 b; + bool need_update = false, dev_exists; + struct printbuf buf = PRINTBUF; + int ret = 0; + + BUG_ON(k.k->type != KEY_TYPE_bucket_gens); + bkey_reassemble(&g.k_i, k); + + /* if no bch_dev, skip out whether we repair or not */ + dev_exists = bch2_dev_exists2(c, k.k->p.inode); + if (!dev_exists) { + if (fsck_err_on(!dev_exists, c, + bucket_gens_to_invalid_dev, + "bucket_gens key for invalid device:\n %s", + (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) { + ret = bch2_btree_delete_at(trans, iter, 0); + } + goto out; + } + + ca = bch_dev_bkey_exists(c, k.k->p.inode); + if (fsck_err_on(end <= ca->mi.first_bucket || + start >= ca->mi.nbuckets, c, + bucket_gens_to_invalid_buckets, + "bucket_gens key for invalid buckets:\n %s", + (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) { + ret = bch2_btree_delete_at(trans, iter, 0); + goto out; + } + + for (b = start; b < ca->mi.first_bucket; b++) + if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c, + bucket_gens_nonzero_for_invalid_buckets, + "bucket_gens key has nonzero gen for invalid bucket")) { + g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0; + need_update = true; + } + + for (b = ca->mi.nbuckets; b < end; b++) + if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c, + bucket_gens_nonzero_for_invalid_buckets, + "bucket_gens key has nonzero gen for invalid bucket")) { + g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0; + need_update = true; + } + + if (need_update) { + struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g)); + + ret = PTR_ERR_OR_ZERO(u); + if (ret) + goto out; + + memcpy(u, &g, sizeof(g)); + ret = bch2_trans_update(trans, iter, u, 0); + } +out: +fsck_err: + printbuf_exit(&buf); + return ret; +} + int bch2_check_alloc_info(struct bch_fs *c) { - struct btree_trans trans; - struct btree_iter iter, discard_iter, freespace_iter; + struct btree_trans *trans = bch2_trans_get(c); + struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter; + struct bkey hole; struct bkey_s_c k; int ret = 0; - bch2_trans_init(&trans, c, 0, 0); - - bch2_trans_iter_init(&trans, &iter, BTREE_ID_alloc, POS_MIN, + bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS_MIN, BTREE_ITER_PREFETCH); - bch2_trans_iter_init(&trans, &discard_iter, BTREE_ID_need_discard, POS_MIN, + bch2_trans_iter_init(trans, &discard_iter, BTREE_ID_need_discard, POS_MIN, BTREE_ITER_PREFETCH); - bch2_trans_iter_init(&trans, &freespace_iter, BTREE_ID_freespace, POS_MIN, + bch2_trans_iter_init(trans, &freespace_iter, BTREE_ID_freespace, POS_MIN, BTREE_ITER_PREFETCH); + bch2_trans_iter_init(trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN, + BTREE_ITER_PREFETCH); + while (1) { - ret = commit_do(&trans, NULL, NULL, - BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW, - bch2_check_alloc_key(&trans, &iter, - &discard_iter, - &freespace_iter)); + struct bpos next; + + bch2_trans_begin(trans); + + k = bch2_get_key_or_real_bucket_hole(&iter, &hole); + ret = bkey_err(k); if (ret) + goto bkey_err; + + if (!k.k) break; - bch2_btree_iter_advance(&iter); + if (k.k->type) { + next = bpos_nosnap_successor(k.k->p); + + ret = bch2_check_alloc_key(trans, + k, &iter, + &discard_iter, + &freespace_iter, + &bucket_gens_iter); + if (ret) + goto bkey_err; + } else { + next = k.k->p; + + ret = bch2_check_alloc_hole_freespace(trans, + bkey_start_pos(k.k), + &next, + &freespace_iter) ?: + bch2_check_alloc_hole_bucket_gens(trans, + bkey_start_pos(k.k), + &next, + &bucket_gens_iter); + if (ret) + goto bkey_err; + } + + ret = bch2_trans_commit(trans, NULL, NULL, + BCH_TRANS_COMMIT_no_enospc); + if (ret) + goto bkey_err; + + bch2_btree_iter_set_pos(&iter, next); +bkey_err: + if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) + continue; + if (ret) + break; } - bch2_trans_iter_exit(&trans, &freespace_iter); - bch2_trans_iter_exit(&trans, &discard_iter); - bch2_trans_iter_exit(&trans, &iter); + bch2_trans_iter_exit(trans, &bucket_gens_iter); + bch2_trans_iter_exit(trans, &freespace_iter); + bch2_trans_iter_exit(trans, &discard_iter); + bch2_trans_iter_exit(trans, &iter); if (ret < 0) goto err; - ret = for_each_btree_key_commit(&trans, iter, + ret = for_each_btree_key(trans, iter, BTREE_ID_need_discard, POS_MIN, BTREE_ITER_PREFETCH, k, - NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW, - bch2_check_discard_freespace_key(&trans, &iter)) ?: - for_each_btree_key_commit(&trans, iter, - BTREE_ID_freespace, POS_MIN, + bch2_check_discard_freespace_key(trans, &iter)); + if (ret) + goto err; + + bch2_trans_iter_init(trans, &iter, BTREE_ID_freespace, POS_MIN, + BTREE_ITER_PREFETCH); + while (1) { + bch2_trans_begin(trans); + k = bch2_btree_iter_peek(&iter); + if (!k.k) + break; + + ret = bkey_err(k) ?: + bch2_check_discard_freespace_key(trans, &iter); + if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) { + ret = 0; + continue; + } + if (ret) { + struct printbuf buf = PRINTBUF; + bch2_bkey_val_to_text(&buf, c, k); + + bch_err(c, "while checking %s", buf.buf); + printbuf_exit(&buf); + break; + } + + bch2_btree_iter_set_pos(&iter, bpos_nosnap_successor(iter.pos)); + } + bch2_trans_iter_exit(trans, &iter); + if (ret) + goto err; + + ret = for_each_btree_key_commit(trans, iter, + BTREE_ID_bucket_gens, POS_MIN, BTREE_ITER_PREFETCH, k, - NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW, - bch2_check_discard_freespace_key(&trans, &iter)); + NULL, NULL, BCH_TRANS_COMMIT_no_enospc, + bch2_check_bucket_gens_key(trans, &iter, k)); err: - bch2_trans_exit(&trans); - return ret < 0 ? ret : 0; + bch2_trans_put(trans); + bch_err_fn(c, ret); + return ret; } static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans, @@ -938,9 +1532,8 @@ static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans, struct btree_iter lru_iter; struct bch_alloc_v4 a_convert; const struct bch_alloc_v4 *a; - struct bkey_s_c alloc_k, k; + struct bkey_s_c alloc_k, lru_k; struct printbuf buf = PRINTBUF; - struct printbuf buf2 = PRINTBUF; int ret; alloc_k = bch2_btree_iter_peek(alloc_iter); @@ -956,84 +1549,96 @@ static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans, if (a->data_type != BCH_DATA_cached) return 0; - bch2_trans_iter_init(trans, &lru_iter, BTREE_ID_lru, - POS(alloc_k.k->p.inode, a->io_time[READ]), 0); - - k = bch2_btree_iter_peek_slot(&lru_iter); - ret = bkey_err(k); - if (ret) - goto err; - if (fsck_err_on(!a->io_time[READ], c, + alloc_key_cached_but_read_time_zero, "cached bucket with read_time 0\n" " %s", (printbuf_reset(&buf), - bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)) || - fsck_err_on(k.k->type != KEY_TYPE_lru || - le64_to_cpu(bkey_s_c_to_lru(k).v->idx) != alloc_k.k->p.offset, c, - "incorrect/missing lru entry\n" - " %s\n" + bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) { + struct bkey_i_alloc_v4 *a_mut = + bch2_alloc_to_v4_mut(trans, alloc_k); + ret = PTR_ERR_OR_ZERO(a_mut); + if (ret) + goto err; + + a_mut->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now); + ret = bch2_trans_update(trans, alloc_iter, + &a_mut->k_i, BTREE_TRIGGER_NORUN); + if (ret) + goto err; + + a = &a_mut->v; + } + + lru_k = bch2_bkey_get_iter(trans, &lru_iter, BTREE_ID_lru, + lru_pos(alloc_k.k->p.inode, + bucket_to_u64(alloc_k.k->p), + a->io_time[READ]), 0); + ret = bkey_err(lru_k); + if (ret) + return ret; + + if (fsck_err_on(lru_k.k->type != KEY_TYPE_set, c, + alloc_key_to_missing_lru_entry, + "missing lru entry\n" " %s", (printbuf_reset(&buf), - bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf), - (bch2_bkey_val_to_text(&buf2, c, k), buf2.buf))) { - u64 read_time = a->io_time[READ] ?: - atomic64_read(&c->io_clock[READ].now); - + bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) { ret = bch2_lru_set(trans, alloc_k.k->p.inode, - alloc_k.k->p.offset, - &read_time); + bucket_to_u64(alloc_k.k->p), + a->io_time[READ]); if (ret) goto err; - - if (a->io_time[READ] != read_time) { - struct bkey_i_alloc_v4 *a_mut = - bch2_alloc_to_v4_mut(trans, alloc_k); - ret = PTR_ERR_OR_ZERO(a_mut); - if (ret) - goto err; - - a_mut->v.io_time[READ] = read_time; - ret = bch2_trans_update(trans, alloc_iter, - &a_mut->k_i, BTREE_TRIGGER_NORUN); - if (ret) - goto err; - } } err: fsck_err: bch2_trans_iter_exit(trans, &lru_iter); - printbuf_exit(&buf2); printbuf_exit(&buf); return ret; } int bch2_check_alloc_to_lru_refs(struct bch_fs *c) { - struct btree_trans trans; - struct btree_iter iter; - struct bkey_s_c k; - int ret = 0; + int ret = bch2_trans_run(c, + for_each_btree_key_commit(trans, iter, BTREE_ID_alloc, + POS_MIN, BTREE_ITER_PREFETCH, k, + NULL, NULL, BCH_TRANS_COMMIT_no_enospc, + bch2_check_alloc_to_lru_ref(trans, &iter))); + bch_err_fn(c, ret); + return ret; +} + +struct discard_buckets_state { + u64 seen; + u64 open; + u64 need_journal_commit; + u64 discarded; + struct bch_dev *ca; + u64 need_journal_commit_this_dev; +}; - bch2_trans_init(&trans, c, 0, 0); +static void discard_buckets_next_dev(struct bch_fs *c, struct discard_buckets_state *s, struct bch_dev *ca) +{ + if (s->ca == ca) + return; - for_each_btree_key_commit(&trans, iter, BTREE_ID_alloc, - POS_MIN, BTREE_ITER_PREFETCH, k, - NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW, - bch2_check_alloc_to_lru_ref(&trans, &iter)); + if (s->ca && s->need_journal_commit_this_dev > + bch2_dev_usage_read(s->ca).d[BCH_DATA_free].buckets) + bch2_journal_flush_async(&c->journal, NULL); - bch2_trans_exit(&trans); - return ret < 0 ? ret : 0; + if (s->ca) + percpu_ref_put(&s->ca->ref); + if (ca) + percpu_ref_get(&ca->ref); + s->ca = ca; + s->need_journal_commit_this_dev = 0; } static int bch2_discard_one_bucket(struct btree_trans *trans, struct btree_iter *need_discard_iter, struct bpos *discard_pos_done, - u64 *seen, - u64 *open, - u64 *need_journal_commit, - u64 *discarded) + struct discard_buckets_state *s) { struct bch_fs *c = trans->c; struct bpos pos = need_discard_iter->pos; @@ -1042,31 +1647,33 @@ static int bch2_discard_one_bucket(struct btree_trans *trans, struct bch_dev *ca; struct bkey_i_alloc_v4 *a; struct printbuf buf = PRINTBUF; - bool did_discard = false; int ret = 0; ca = bch_dev_bkey_exists(c, pos.inode); + if (!percpu_ref_tryget(&ca->io_ref)) { bch2_btree_iter_set_pos(need_discard_iter, POS(pos.inode + 1, 0)); return 0; } + discard_buckets_next_dev(c, s, ca); + if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) { - (*open)++; + s->open++; goto out; } if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal, c->journal.flushed_seq_ondisk, pos.inode, pos.offset)) { - (*need_journal_commit)++; + s->need_journal_commit++; + s->need_journal_commit_this_dev++; goto out; } - bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, - need_discard_iter->pos, - BTREE_ITER_CACHED); - k = bch2_btree_iter_peek_slot(&iter); + k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc, + need_discard_iter->pos, + BTREE_ITER_CACHED); ret = bkey_err(k); if (ret) goto out; @@ -1082,21 +1689,28 @@ static int bch2_discard_one_bucket(struct btree_trans *trans, goto write; } - if (bch2_trans_inconsistent_on(a->v.journal_seq > c->journal.flushed_seq_ondisk, trans, - "clearing need_discard but journal_seq %llu > flushed_seq %llu\n" - "%s", - a->v.journal_seq, - c->journal.flushed_seq_ondisk, - (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) { - ret = -EIO; + if (a->v.journal_seq > c->journal.flushed_seq_ondisk) { + if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) { + bch2_trans_inconsistent(trans, + "clearing need_discard but journal_seq %llu > flushed_seq %llu\n" + "%s", + a->v.journal_seq, + c->journal.flushed_seq_ondisk, + (bch2_bkey_val_to_text(&buf, c, k), buf.buf)); + ret = -EIO; + } goto out; } - if (bch2_trans_inconsistent_on(a->v.data_type != BCH_DATA_need_discard, trans, - "bucket incorrectly set in need_discard btree\n" - "%s", - (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) { - ret = -EIO; + if (a->v.data_type != BCH_DATA_need_discard) { + if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) { + bch2_trans_inconsistent(trans, + "bucket incorrectly set in need_discard btree\n" + "%s", + (bch2_bkey_val_to_text(&buf, c, k), buf.buf)); + ret = -EIO; + } + goto out; } @@ -1106,34 +1720,32 @@ static int bch2_discard_one_bucket(struct btree_trans *trans, * This works without any other locks because this is the only * thread that removes items from the need_discard tree */ - bch2_trans_unlock(trans); + bch2_trans_unlock_long(trans); blkdev_issue_discard(ca->disk_sb.bdev, k.k->p.offset * ca->mi.bucket_size, ca->mi.bucket_size, GFP_KERNEL); + *discard_pos_done = iter.pos; - ret = bch2_trans_relock(trans); + ret = bch2_trans_relock_notrace(trans); if (ret) goto out; } - *discard_pos_done = iter.pos; - did_discard = true; - SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false); a->v.data_type = alloc_data_type(a->v, a->v.data_type); write: ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?: bch2_trans_commit(trans, NULL, NULL, - BTREE_INSERT_USE_RESERVE|BTREE_INSERT_NOFAIL); + BCH_WATERMARK_btree| + BCH_TRANS_COMMIT_no_enospc); if (ret) goto out; - if (did_discard) { - this_cpu_inc(c->counters[BCH_COUNTER_bucket_discard]); - (*discarded)++; - } + count_event(c, bucket_discard); + s->discarded++; out: + s->seen++; bch2_trans_iter_exit(trans, &iter); percpu_ref_put(&ca->io_ref); printbuf_exit(&buf); @@ -1143,97 +1755,69 @@ out: static void bch2_do_discards_work(struct work_struct *work) { struct bch_fs *c = container_of(work, struct bch_fs, discard_work); - struct btree_trans trans; - struct btree_iter iter; - struct bkey_s_c k; - u64 seen = 0, open = 0, need_journal_commit = 0, discarded = 0; + struct discard_buckets_state s = {}; struct bpos discard_pos_done = POS_MAX; int ret; - bch2_trans_init(&trans, c, 0, 0); - /* * We're doing the commit in bch2_discard_one_bucket instead of using * for_each_btree_key_commit() so that we can increment counters after * successful commit: */ - ret = for_each_btree_key2(&trans, iter, - BTREE_ID_need_discard, POS_MIN, 0, k, - bch2_discard_one_bucket(&trans, &iter, &discard_pos_done, - &seen, - &open, - &need_journal_commit, - &discarded)); - - bch2_trans_exit(&trans); + ret = bch2_trans_run(c, + for_each_btree_key(trans, iter, + BTREE_ID_need_discard, POS_MIN, 0, k, + bch2_discard_one_bucket(trans, &iter, &discard_pos_done, &s))); - if (need_journal_commit * 2 > seen) - bch2_journal_flush_async(&c->journal, NULL); - - percpu_ref_put(&c->writes); + discard_buckets_next_dev(c, &s, NULL); - trace_discard_buckets(c, seen, open, need_journal_commit, discarded, + trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded, bch2_err_str(ret)); + + bch2_write_ref_put(c, BCH_WRITE_REF_discard); } void bch2_do_discards(struct bch_fs *c) { - if (percpu_ref_tryget_live(&c->writes) && - !queue_work(system_long_wq, &c->discard_work)) - percpu_ref_put(&c->writes); + if (bch2_write_ref_tryget(c, BCH_WRITE_REF_discard) && + !queue_work(c->write_ref_wq, &c->discard_work)) + bch2_write_ref_put(c, BCH_WRITE_REF_discard); } static int invalidate_one_bucket(struct btree_trans *trans, - struct btree_iter *lru_iter, struct bkey_s_c k, - unsigned dev_idx, s64 *nr_to_invalidate) + struct btree_iter *lru_iter, + struct bkey_s_c lru_k, + s64 *nr_to_invalidate) { struct bch_fs *c = trans->c; struct btree_iter alloc_iter = { NULL }; - struct bkey_i_alloc_v4 *a; - struct bpos bucket; + struct bkey_i_alloc_v4 *a = NULL; struct printbuf buf = PRINTBUF; + struct bpos bucket = u64_to_bucket(lru_k.k->p.offset); unsigned cached_sectors; int ret = 0; - if (*nr_to_invalidate <= 0 || k.k->p.inode != dev_idx) + if (*nr_to_invalidate <= 0) return 1; - if (k.k->type != KEY_TYPE_lru) { - prt_printf(&buf, "non lru key in lru btree:\n "); - bch2_bkey_val_to_text(&buf, c, k); - - if (!test_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags)) { - bch_err(c, "%s", buf.buf); - } else { - bch2_trans_inconsistent(trans, "%s", buf.buf); - ret = -EINVAL; - } - - goto out; + if (!bch2_dev_bucket_exists(c, bucket)) { + prt_str(&buf, "lru entry points to invalid bucket"); + goto err; } - bucket = POS(dev_idx, le64_to_cpu(bkey_s_c_to_lru(k).v->idx)); + if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset)) + return 0; a = bch2_trans_start_alloc_update(trans, &alloc_iter, bucket); ret = PTR_ERR_OR_ZERO(a); if (ret) goto out; - if (k.k->p.offset != alloc_lru_idx(a->v)) { - prt_printf(&buf, "alloc key does not point back to lru entry when invalidating bucket:\n "); - bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i)); - prt_printf(&buf, "\n "); - bch2_bkey_val_to_text(&buf, c, k); - - if (!test_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags)) { - bch_err(c, "%s", buf.buf); - } else { - bch2_trans_inconsistent(trans, "%s", buf.buf); - ret = -EINVAL; - } - + /* We expect harmless races here due to the btree write buffer: */ + if (lru_pos_time(lru_iter->pos) != alloc_lru_idx_read(a->v)) goto out; - } + + BUG_ON(a->v.data_type != BCH_DATA_cached); if (!a->v.cached_sectors) bch_err(c, "invalidating empty bucket, confused"); @@ -1251,7 +1835,8 @@ static int invalidate_one_bucket(struct btree_trans *trans, ret = bch2_trans_update(trans, &alloc_iter, &a->k_i, BTREE_TRIGGER_BUCKET_INVALIDATE) ?: bch2_trans_commit(trans, NULL, NULL, - BTREE_INSERT_USE_RESERVE|BTREE_INSERT_NOFAIL); + BCH_WATERMARK_btree| + BCH_TRANS_COMMIT_no_enospc); if (ret) goto out; @@ -1261,71 +1846,102 @@ out: bch2_trans_iter_exit(trans, &alloc_iter); printbuf_exit(&buf); return ret; +err: + prt_str(&buf, "\n lru key: "); + bch2_bkey_val_to_text(&buf, c, lru_k); + + prt_str(&buf, "\n lru entry: "); + bch2_lru_pos_to_text(&buf, lru_iter->pos); + + prt_str(&buf, "\n alloc key: "); + if (!a) + bch2_bpos_to_text(&buf, bucket); + else + bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i)); + + bch_err(c, "%s", buf.buf); + if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_lrus) { + bch2_inconsistent_error(c); + ret = -EINVAL; + } + + goto out; } static void bch2_do_invalidates_work(struct work_struct *work) { struct bch_fs *c = container_of(work, struct bch_fs, invalidate_work); - struct bch_dev *ca; - struct btree_trans trans; - struct btree_iter iter; - struct bkey_s_c k; - unsigned i; + struct btree_trans *trans = bch2_trans_get(c); int ret = 0; - bch2_trans_init(&trans, c, 0, 0); + ret = bch2_btree_write_buffer_tryflush(trans); + if (ret) + goto err; - for_each_member_device(ca, c, i) { + for_each_member_device(c, ca) { s64 nr_to_invalidate = should_invalidate_buckets(ca, bch2_dev_usage_read(ca)); - ret = for_each_btree_key2(&trans, iter, BTREE_ID_lru, - POS(ca->dev_idx, 0), BTREE_ITER_INTENT, k, - invalidate_one_bucket(&trans, &iter, k, ca->dev_idx, &nr_to_invalidate)); + ret = for_each_btree_key_upto(trans, iter, BTREE_ID_lru, + lru_pos(ca->dev_idx, 0, 0), + lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX), + BTREE_ITER_INTENT, k, + invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate)); if (ret < 0) { percpu_ref_put(&ca->ref); break; } } - - bch2_trans_exit(&trans); - percpu_ref_put(&c->writes); +err: + bch2_trans_put(trans); + bch2_write_ref_put(c, BCH_WRITE_REF_invalidate); } void bch2_do_invalidates(struct bch_fs *c) { - if (percpu_ref_tryget_live(&c->writes) && - !queue_work(system_long_wq, &c->invalidate_work)) - percpu_ref_put(&c->writes); + if (bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate) && + !queue_work(c->write_ref_wq, &c->invalidate_work)) + bch2_write_ref_put(c, BCH_WRITE_REF_invalidate); } -static int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca) +int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca, + u64 bucket_start, u64 bucket_end) { - struct btree_trans trans; + struct btree_trans *trans = bch2_trans_get(c); struct btree_iter iter; struct bkey_s_c k; - struct bpos end = POS(ca->dev_idx, ca->mi.nbuckets); + struct bkey hole; + struct bpos end = POS(ca->dev_idx, bucket_end); struct bch_member *m; + unsigned long last_updated = jiffies; int ret; - bch2_trans_init(&trans, c, 0, 0); + BUG_ON(bucket_start > bucket_end); + BUG_ON(bucket_end > ca->mi.nbuckets); - bch2_trans_iter_init(&trans, &iter, BTREE_ID_alloc, - POS(ca->dev_idx, ca->mi.first_bucket), - BTREE_ITER_PREFETCH); + bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, + POS(ca->dev_idx, max_t(u64, ca->mi.first_bucket, bucket_start)), + BTREE_ITER_PREFETCH); /* * Scan the alloc btree for every bucket on @ca, and add buckets to the * freespace/need_discard/need_gc_gens btrees as needed: */ while (1) { - bch2_trans_begin(&trans); - ret = 0; + if (last_updated + HZ * 10 < jiffies) { + bch_info(ca, "%s: currently at %llu/%llu", + __func__, iter.pos.offset, ca->mi.nbuckets); + last_updated = jiffies; + } + + bch2_trans_begin(trans); - if (bkey_ge(iter.pos, end)) + if (bkey_ge(iter.pos, end)) { + ret = 0; break; + } - k = bch2_btree_iter_peek_slot(&iter); + k = bch2_get_key_or_hole(&iter, end, &hole); ret = bkey_err(k); if (ret) goto bkey_err; @@ -1338,63 +1954,33 @@ static int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca) struct bch_alloc_v4 a_convert; const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert); - ret = bch2_bucket_do_index(&trans, k, a, true) ?: - bch2_trans_commit(&trans, NULL, NULL, - BTREE_INSERT_LAZY_RW| - BTREE_INSERT_NOFAIL); + ret = bch2_bucket_do_index(trans, k, a, true) ?: + bch2_trans_commit(trans, NULL, NULL, + BCH_TRANS_COMMIT_no_enospc); if (ret) goto bkey_err; bch2_btree_iter_advance(&iter); } else { - /* - * When there's a hole, process a whole range of keys - * all at once: - * - * This is similar to how extent btree iterators in - * slots mode will synthesize a whole range - a - * KEY_TYPE_deleted extent. - * - * But alloc keys aren't extents (they have zero size), - * so we're open coding it here: - */ - struct btree_iter iter2; struct bkey_i *freespace; - struct bpos next; - - bch2_trans_copy_iter(&iter2, &iter); - k = bch2_btree_iter_peek_upto(&iter2, - bkey_min(bkey_min(end, - iter.path->l[0].b->key.k.p), - POS(iter.pos.inode, iter.pos.offset + U32_MAX - 1))); - next = iter2.pos; - ret = bkey_err(k); - bch2_trans_iter_exit(&trans, &iter2); - - BUG_ON(next.offset >= iter.pos.offset + U32_MAX); - - if (ret) - goto bkey_err; - freespace = bch2_trans_kmalloc(&trans, sizeof(*freespace)); + freespace = bch2_trans_kmalloc(trans, sizeof(*freespace)); ret = PTR_ERR_OR_ZERO(freespace); if (ret) goto bkey_err; bkey_init(&freespace->k); - freespace->k.type = KEY_TYPE_set; - freespace->k.p = iter.pos; + freespace->k.type = KEY_TYPE_set; + freespace->k.p = k.k->p; + freespace->k.size = k.k->size; - bch2_key_resize(&freespace->k, next.offset - iter.pos.offset); - - ret = __bch2_btree_insert(&trans, BTREE_ID_freespace, freespace) ?: - bch2_trans_commit(&trans, NULL, NULL, - BTREE_INSERT_LAZY_RW| - BTREE_INSERT_NOFAIL); + ret = bch2_btree_insert_trans(trans, BTREE_ID_freespace, freespace, 0) ?: + bch2_trans_commit(trans, NULL, NULL, + BCH_TRANS_COMMIT_no_enospc); if (ret) goto bkey_err; - bch2_btree_iter_set_pos(&iter, next); + bch2_btree_iter_set_pos(&iter, k.k->p); } bkey_err: if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) @@ -1403,16 +1989,16 @@ bkey_err: break; } - bch2_trans_iter_exit(&trans, &iter); - bch2_trans_exit(&trans); + bch2_trans_iter_exit(trans, &iter); + bch2_trans_put(trans); if (ret < 0) { - bch_err(ca, "error initializing free space: %s", bch2_err_str(ret)); + bch_err_msg(ca, ret, "initializing free space"); return ret; } mutex_lock(&c->sb_lock); - m = bch2_sb_get_members(c->disk_sb.sb)->members + ca->dev_idx; + m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx); SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true); mutex_unlock(&c->sb_lock); @@ -1421,8 +2007,6 @@ bkey_err: int bch2_fs_freespace_init(struct bch_fs *c) { - struct bch_dev *ca; - unsigned i; int ret = 0; bool doing_init = false; @@ -1431,7 +2015,7 @@ int bch2_fs_freespace_init(struct bch_fs *c) * every mount: */ - for_each_member_device(ca, c, i) { + for_each_member_device(c, ca) { if (ca->mi.freespace_initialized) continue; @@ -1440,9 +2024,10 @@ int bch2_fs_freespace_init(struct bch_fs *c) doing_init = true; } - ret = bch2_dev_freespace_init(c, ca); + ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets); if (ret) { percpu_ref_put(&ca->ref); + bch_err_fn(c, ret); return ret; } } @@ -1451,11 +2036,10 @@ int bch2_fs_freespace_init(struct bch_fs *c) mutex_lock(&c->sb_lock); bch2_write_super(c); mutex_unlock(&c->sb_lock); - bch_verbose(c, "done initializing freespace"); } - return ret; + return 0; } /* Bucket IO clocks: */ @@ -1491,15 +2075,13 @@ out: void bch2_recalc_capacity(struct bch_fs *c) { - struct bch_dev *ca; u64 capacity = 0, reserved_sectors = 0, gc_reserve; unsigned bucket_size_max = 0; unsigned long ra_pages = 0; - unsigned i; lockdep_assert_held(&c->state_lock); - for_each_online_member(ca, c, i) { + for_each_online_member(c, ca) { struct backing_dev_info *bdi = ca->disk_sb.bdev->bd_disk->bdi; ra_pages += bdi->ra_pages; @@ -1507,7 +2089,7 @@ void bch2_recalc_capacity(struct bch_fs *c) bch2_set_ra_pages(c, ra_pages); - for_each_rw_member(ca, c, i) { + for_each_rw_member(c, ca) { u64 dev_reserve = 0; /* @@ -1561,6 +2143,15 @@ void bch2_recalc_capacity(struct bch_fs *c) closure_wake_up(&c->freelist_wait); } +u64 bch2_min_rw_member_capacity(struct bch_fs *c) +{ + u64 ret = U64_MAX; + + for_each_rw_member(c, ca) + ret = min(ret, ca->mi.nbuckets * ca->mi.bucket_size); + return ret; +} + static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca) { struct open_bucket *ob; @@ -1594,40 +2185,7 @@ void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca) */ bch2_recalc_capacity(c); - /* Next, close write points that point to this device... */ - for (i = 0; i < ARRAY_SIZE(c->write_points); i++) - bch2_writepoint_stop(c, ca, &c->write_points[i]); - - bch2_writepoint_stop(c, ca, &c->copygc_write_point); - bch2_writepoint_stop(c, ca, &c->rebalance_write_point); - bch2_writepoint_stop(c, ca, &c->btree_write_point); - - mutex_lock(&c->btree_reserve_cache_lock); - while (c->btree_reserve_cache_nr) { - struct btree_alloc *a = - &c->btree_reserve_cache[--c->btree_reserve_cache_nr]; - - bch2_open_buckets_put(c, &a->ob); - } - mutex_unlock(&c->btree_reserve_cache_lock); - - while (1) { - struct open_bucket *ob; - - spin_lock(&c->freelist_lock); - if (!ca->open_buckets_partial_nr) { - spin_unlock(&c->freelist_lock); - break; - } - ob = c->open_buckets + - ca->open_buckets_partial[--ca->open_buckets_partial_nr]; - ob->on_partial_list = false; - spin_unlock(&c->freelist_lock); - - bch2_open_bucket_put(c, ob); - } - - bch2_ec_stop_dev(c, ca); + bch2_open_buckets_stop(c, ca, false); /* * Wake up threads that were blocked on allocation, so they can notice