X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Falloc_background.c;h=1acdee8226eb1debaa3e2c8070cdd7b2c4bc4809;hb=934a84dfaf719af82dadbbe0e2480baff03c905b;hp=d114ba0949a8aff1823dae597b126bedbbadd4e0;hpb=6ac37db8415c636607d878c16af8346df55668f4;p=bcachefs-tools-debian diff --git a/libbcachefs/alloc_background.c b/libbcachefs/alloc_background.c index d114ba0..1acdee8 100644 --- a/libbcachefs/alloc_background.c +++ b/libbcachefs/alloc_background.c @@ -2,6 +2,7 @@ #include "bcachefs.h" #include "alloc_background.h" #include "alloc_foreground.h" +#include "backpointers.h" #include "btree_cache.h" #include "btree_io.h" #include "btree_key_cache.h" @@ -37,8 +38,6 @@ static const unsigned BCH_ALLOC_V1_FIELD_BYTES[] = { struct bkey_alloc_unpacked { u64 journal_seq; - u64 bucket; - u8 dev; u8 gen; u8 oldest_gen; u8 data_type; @@ -194,11 +193,7 @@ static int bch2_alloc_unpack_v3(struct bkey_alloc_unpacked *out, static struct bkey_alloc_unpacked bch2_alloc_unpack(struct bkey_s_c k) { - struct bkey_alloc_unpacked ret = { - .dev = k.k->p.inode, - .bucket = k.k->p.offset, - .gen = 0, - }; + struct bkey_alloc_unpacked ret = { .gen = 0 }; switch (k.k->type) { case KEY_TYPE_alloc: @@ -215,73 +210,6 @@ static struct bkey_alloc_unpacked bch2_alloc_unpack(struct bkey_s_c k) return ret; } -void bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out) -{ - if (k.k->type == KEY_TYPE_alloc_v4) { - *out = *bkey_s_c_to_alloc_v4(k).v; - } else { - struct bkey_alloc_unpacked u = bch2_alloc_unpack(k); - - *out = (struct bch_alloc_v4) { - .journal_seq = u.journal_seq, - .flags = u.need_discard, - .gen = u.gen, - .oldest_gen = u.oldest_gen, - .data_type = u.data_type, - .stripe_redundancy = u.stripe_redundancy, - .dirty_sectors = u.dirty_sectors, - .cached_sectors = u.cached_sectors, - .io_time[READ] = u.read_time, - .io_time[WRITE] = u.write_time, - .stripe = u.stripe, - }; - } -} - -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; - - if (k.k->type == KEY_TYPE_alloc_v4) { - ret = bch2_trans_kmalloc(trans, bkey_bytes(k.k)); - if (!IS_ERR(ret)) - bkey_reassemble(&ret->k_i, k); - } else { - ret = bch2_trans_kmalloc(trans, sizeof(*ret)); - if (!IS_ERR(ret)) { - bkey_alloc_v4_init(&ret->k_i); - ret->k.p = k.k->p; - bch2_alloc_to_v4(k, &ret->v); - } - } - return ret; -} - -struct bkey_i_alloc_v4 * -bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter, - struct bpos pos) -{ - struct bkey_s_c k; - struct bkey_i_alloc_v4 *a; - int ret; - - bch2_trans_iter_init(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 (ret) { - bch2_trans_iter_exit(trans, iter); - return ERR_PTR(ret); - } - - a = bch2_alloc_to_v4_mut(trans, k); - if (IS_ERR(a)) - bch2_trans_iter_exit(trans, iter); - return a; -} - static unsigned bch_alloc_v1_val_u64s(const struct bch_alloc *a) { unsigned i, bytes = offsetof(struct bch_alloc, data); @@ -302,7 +230,7 @@ int bch2_alloc_v1_invalid(const struct bch_fs *c, struct bkey_s_c k, 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 -EINVAL; + return -BCH_ERR_invalid_bkey; } return 0; @@ -315,7 +243,7 @@ int bch2_alloc_v2_invalid(const struct bch_fs *c, struct bkey_s_c k, if (bch2_alloc_unpack_v2(&u, k)) { prt_printf(err, "unpack error"); - return -EINVAL; + return -BCH_ERR_invalid_bkey; } return 0; @@ -328,7 +256,7 @@ int bch2_alloc_v3_invalid(const struct bch_fs *c, struct bkey_s_c k, if (bch2_alloc_unpack_v3(&u, k)) { prt_printf(err, "unpack error"); - return -EINVAL; + return -BCH_ERR_invalid_bkey; } return 0; @@ -339,17 +267,39 @@ int bch2_alloc_v4_invalid(const struct bch_fs *c, struct bkey_s_c k, { struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k); - if (bkey_val_bytes(k.k) != sizeof(struct bch_alloc_v4)) { - prt_printf(err, "bad val size (%zu != %zu)", - bkey_val_bytes(k.k), sizeof(struct bch_alloc_v4)); - return -EINVAL; + 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; + + for (i = 0; i < BCH_ALLOC_V4_NR_BACKPOINTERS(a.v); i++) + bp_len += alloc_v4_backpointers_c(a.v)[i].bucket_len; + + 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 -EINVAL; + return -BCH_ERR_invalid_bkey; } switch (a.v->data_type) { @@ -360,7 +310,7 @@ int bch2_alloc_v4_invalid(const struct bch_fs *c, struct bkey_s_c k, a.v->cached_sectors || a.v->stripe) { prt_printf(err, "empty data type free but have data"); - return -EINVAL; + return -BCH_ERR_invalid_bkey; } break; case BCH_DATA_sb: @@ -371,7 +321,7 @@ int bch2_alloc_v4_invalid(const struct bch_fs *c, struct bkey_s_c k, if (!a.v->dirty_sectors) { prt_printf(err, "data_type %s but dirty_sectors==0", bch2_data_types[a.v->data_type]); - return -EINVAL; + return -BCH_ERR_invalid_bkey; } break; case BCH_DATA_cached: @@ -379,20 +329,20 @@ int bch2_alloc_v4_invalid(const struct bch_fs *c, struct bkey_s_c k, a.v->dirty_sectors || a.v->stripe) { prt_printf(err, "data type inconsistency"); - return -EINVAL; + return -BCH_ERR_invalid_bkey; } 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 -EINVAL; + 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 -EINVAL; + return -BCH_ERR_invalid_bkey; } break; } @@ -401,9 +351,19 @@ int bch2_alloc_v4_invalid(const struct bch_fs *c, struct bkey_s_c k, return 0; } +static inline u64 swab40(u64 x) +{ + return (((x & 0x00000000ffULL) << 32)| + ((x & 0x000000ff00ULL) << 16)| + ((x & 0x0000ff0000ULL) >> 0)| + ((x & 0x00ff000000ULL) >> 16)| + ((x & 0xff00000000ULL) >> 32)); +} + void bch2_alloc_v4_swab(struct bkey_s k) { struct bch_alloc_v4 *a = bkey_s_to_alloc_v4(k).v; + struct bch_backpointer *bp, *bps; a->journal_seq = swab64(a->journal_seq); a->flags = swab32(a->flags); @@ -413,25 +373,184 @@ 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); + + bps = alloc_v4_backpointers(a); + for (bp = bps; bp < bps + BCH_ALLOC_V4_NR_BACKPOINTERS(a); bp++) { + bp->bucket_offset = swab40(bp->bucket_offset); + bp->bucket_len = swab32(bp->bucket_len); + bch2_bpos_swab(&bp->pos); + } } void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k) { - struct bch_alloc_v4 a; + struct bch_alloc_v4 _a; + const struct bch_alloc_v4 *a = &_a; + const struct bch_backpointer *bps; + unsigned i; - bch2_alloc_to_v4(k, &a); + if (k.k->type == KEY_TYPE_alloc_v4) + a = bkey_s_c_to_alloc_v4(k).v; + else + bch2_alloc_to_v4(k, &_a); + + 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_newline(out); + prt_printf(out, "journal_seq %llu", a->journal_seq); + prt_newline(out); + prt_printf(out, "need_discard %llu", BCH_ALLOC_V4_NEED_DISCARD(a)); + prt_newline(out); + prt_printf(out, "need_inc_gen %llu", BCH_ALLOC_V4_NEED_INC_GEN(a)); + prt_newline(out); + prt_printf(out, "dirty_sectors %u", a->dirty_sectors); + prt_newline(out); + prt_printf(out, "cached_sectors %u", a->cached_sectors); + prt_newline(out); + prt_printf(out, "stripe %u", a->stripe); + prt_newline(out); + prt_printf(out, "stripe_redundancy %u", a->stripe_redundancy); + prt_newline(out); + prt_printf(out, "io_time[READ] %llu", a->io_time[READ]); + 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); +} + +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; + + *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); + } + } else { + struct bkey_alloc_unpacked u = bch2_alloc_unpack(k); + + *out = (struct bch_alloc_v4) { + .journal_seq = u.journal_seq, + .flags = u.need_discard, + .gen = u.gen, + .oldest_gen = u.oldest_gen, + .data_type = u.data_type, + .stripe_redundancy = u.stripe_redundancy, + .dirty_sectors = u.dirty_sectors, + .cached_sectors = u.cached_sectors, + .io_time[READ] = u.read_time, + .io_time[WRITE] = u.write_time, + .stripe = u.stripe, + }; + + SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s); + } +} + +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)); + if (IS_ERR(ret)) + return ret; + + if (k.k->type == KEY_TYPE_alloc_v4) { + struct bch_backpointer *src, *dst; + + bkey_reassemble(&ret->k_i, k); + + src = alloc_v4_backpointers(&ret->v); + 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); + set_alloc_v4_u64s(ret); + } else { + bkey_alloc_v4_init(&ret->k_i); + ret->k.p = k.k->p; + bch2_alloc_to_v4(k, &ret->v); + } + return ret; +} + +static inline struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut_inlined(struct btree_trans *trans, struct bkey_s_c k) +{ + 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(trans, bkey_bytes(k.k) + sizeof(struct bch_backpointer)); + if (!IS_ERR(ret)) + bkey_reassemble(&ret->k_i, k); + return ret; + } + + return __bch2_alloc_to_v4_mut(trans, k); +} + +struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k) +{ + return bch2_alloc_to_v4_mut_inlined(trans, k); +} + +struct bkey_i_alloc_v4 * +bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter, + struct bpos pos) +{ + struct bkey_s_c k; + struct bkey_i_alloc_v4 *a; + int ret; + + bch2_trans_iter_init(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; - prt_printf(out, "gen %u oldest_gen %u data_type %s journal_seq %llu need_discard %llu need_inc_gen %llu", - a.gen, a.oldest_gen, bch2_data_types[a.data_type], - a.journal_seq, - BCH_ALLOC_V4_NEED_DISCARD(&a), - BCH_ALLOC_V4_NEED_INC_GEN(&a)); - prt_printf(out, " dirty_sectors %u", a.dirty_sectors); - prt_printf(out, " cached_sectors %u", a.cached_sectors); - prt_printf(out, " stripe %u", a.stripe); - prt_printf(out, " stripe_redundancy %u", a.stripe_redundancy); - prt_printf(out, " read_time %llu", a.io_time[READ]); - prt_printf(out, " write_time %llu", a.io_time[WRITE]); + a = bch2_alloc_to_v4_mut_inlined(trans, k); + ret = PTR_ERR_OR_ZERO(a); + if (unlikely(ret)) + goto err; + return a; +err: + bch2_trans_iter_exit(trans, iter); + return ERR_PTR(ret); } int bch2_alloc_read(struct bch_fs *c) @@ -464,7 +583,7 @@ int bch2_alloc_read(struct bch_fs *c) bch2_trans_exit(&trans); if (ret) - bch_err(c, "error reading alloc info: %i", ret); + bch_err(c, "error reading alloc info: %s", bch2_err_str(ret)); return ret; } @@ -606,21 +725,23 @@ int bch2_trans_mark_alloc(struct btree_trans *trans, } static int bch2_check_alloc_key(struct btree_trans *trans, - struct btree_iter *alloc_iter) + struct btree_iter *alloc_iter, + struct btree_iter *discard_iter, + struct btree_iter *freespace_iter) { struct bch_fs *c = trans->c; struct bch_dev *ca; - struct btree_iter discard_iter, freespace_iter; struct bch_alloc_v4 a; unsigned discard_key_type, freespace_key_type; struct bkey_s_c alloc_k, k; struct printbuf buf = PRINTBUF; - struct printbuf buf2 = PRINTBUF; int ret; - alloc_k = bch2_btree_iter_peek(alloc_iter); + 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 0; + return 1; ret = bkey_err(alloc_k); if (ret) @@ -642,22 +763,21 @@ static int bch2_check_alloc_key(struct btree_trans *trans, freespace_key_type = a.data_type == BCH_DATA_free ? KEY_TYPE_set : 0; - bch2_trans_iter_init(trans, &discard_iter, BTREE_ID_need_discard, - alloc_k.k->p, 0); - bch2_trans_iter_init(trans, &freespace_iter, BTREE_ID_freespace, - alloc_freespace_pos(alloc_k.k->p, a), 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); + k = bch2_btree_iter_peek_slot(discard_iter); ret = bkey_err(k); if (ret) goto err; - if (fsck_err_on(k.k->type != discard_key_type, c, - "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], - (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) { + 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" + " %s", + bch2_bkey_types[k.k->type], + bch2_bkey_types[discard_key_type], + (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) { struct bkey_i *update = bch2_trans_kmalloc(trans, sizeof(*update)); @@ -667,25 +787,26 @@ static int bch2_check_alloc_key(struct btree_trans *trans, bkey_init(&update->k); update->k.type = discard_key_type; - update->k.p = discard_iter.pos; + update->k.p = discard_iter->pos; - ret = bch2_trans_update(trans, &discard_iter, update, 0); + ret = bch2_trans_update(trans, discard_iter, update, 0); if (ret) goto err; } - k = bch2_btree_iter_peek_slot(&freespace_iter); + k = bch2_btree_iter_peek_slot(freespace_iter); ret = bkey_err(k); if (ret) goto err; - if (fsck_err_on(k.k->type != freespace_key_type, 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))) { + if (k.k->type != freespace_key_type && + (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)))) { struct bkey_i *update = bch2_trans_kmalloc(trans, sizeof(*update)); @@ -695,18 +816,15 @@ static int bch2_check_alloc_key(struct btree_trans *trans, bkey_init(&update->k); update->k.type = freespace_key_type; - update->k.p = freespace_iter.pos; + update->k.p = freespace_iter->pos; bch2_key_resize(&update->k, 1); - ret = bch2_trans_update(trans, &freespace_iter, update, 0); + ret = bch2_trans_update(trans, freespace_iter, update, 0); if (ret) goto err; } err: fsck_err: - bch2_trans_iter_exit(trans, &freespace_iter); - bch2_trans_iter_exit(trans, &discard_iter); - printbuf_exit(&buf2); printbuf_exit(&buf); return ret; } @@ -716,7 +834,7 @@ static int bch2_check_discard_freespace_key(struct btree_trans *trans, { struct bch_fs *c = trans->c; struct btree_iter alloc_iter; - struct bkey_s_c k, freespace_k; + struct bkey_s_c alloc_k; struct bch_alloc_v4 a; u64 genbits; struct bpos pos; @@ -726,14 +844,6 @@ static int bch2_check_discard_freespace_key(struct btree_trans *trans, struct printbuf buf = PRINTBUF; int ret; - freespace_k = bch2_btree_iter_peek(iter); - if (!freespace_k.k) - return 1; - - ret = bkey_err(freespace_k); - if (ret) - return ret; - pos = iter->pos; pos.offset &= ~(~0ULL << 56); genbits = iter->pos.offset & (~0ULL << 56); @@ -745,18 +855,18 @@ static int bch2_check_discard_freespace_key(struct btree_trans *trans, bch2_btree_ids[iter->btree_id], pos.inode, pos.offset)) goto delete; - k = bch2_btree_iter_peek_slot(&alloc_iter); - ret = bkey_err(k); + alloc_k = bch2_btree_iter_peek_slot(&alloc_iter); + ret = bkey_err(alloc_k); if (ret) goto err; - bch2_alloc_to_v4(k, &a); + bch2_alloc_to_v4(alloc_k, &a); 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)", - (bch2_bkey_val_to_text(&buf, c, k), buf.buf), + (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf), bch2_btree_ids[iter->btree_id], a.data_type == state, genbits >> 56, alloc_freespace_genbits(a) >> 56)) @@ -776,50 +886,47 @@ delete: int bch2_check_alloc_info(struct bch_fs *c) { struct btree_trans trans; - struct btree_iter iter; + struct btree_iter iter, discard_iter, freespace_iter; struct bkey_s_c k; int ret = 0; bch2_trans_init(&trans, c, 0, 0); - for_each_btree_key(&trans, iter, BTREE_ID_alloc, POS_MIN, - BTREE_ITER_PREFETCH, k, ret) { - ret = __bch2_trans_do(&trans, NULL, NULL, 0, - bch2_check_alloc_key(&trans, &iter)); - if (ret) - break; - } - bch2_trans_iter_exit(&trans, &iter); - - if (ret) - goto err; - - bch2_trans_iter_init(&trans, &iter, BTREE_ID_need_discard, 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, + BTREE_ITER_PREFETCH); + bch2_trans_iter_init(&trans, &freespace_iter, BTREE_ID_freespace, POS_MIN, BTREE_ITER_PREFETCH); while (1) { - ret = __bch2_trans_do(&trans, NULL, NULL, 0, - bch2_check_discard_freespace_key(&trans, &iter)); + ret = commit_do(&trans, NULL, NULL, + BTREE_INSERT_NOFAIL| + BTREE_INSERT_LAZY_RW, + bch2_check_alloc_key(&trans, &iter, + &discard_iter, + &freespace_iter)); if (ret) break; - bch2_btree_iter_set_pos(&iter, bpos_nosnap_successor(iter.pos)); + bch2_btree_iter_advance(&iter); } + bch2_trans_iter_exit(&trans, &freespace_iter); + bch2_trans_iter_exit(&trans, &discard_iter); bch2_trans_iter_exit(&trans, &iter); - if (ret) + if (ret < 0) goto err; - bch2_trans_iter_init(&trans, &iter, BTREE_ID_freespace, POS_MIN, - BTREE_ITER_PREFETCH); - while (1) { - ret = __bch2_trans_do(&trans, NULL, NULL, 0, - bch2_check_discard_freespace_key(&trans, &iter)); - if (ret) - break; - - bch2_btree_iter_set_pos(&iter, bpos_nosnap_successor(iter.pos)); - } - bch2_trans_iter_exit(&trans, &iter); + ret = for_each_btree_key_commit(&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, + BTREE_ITER_PREFETCH, k, + NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW, + bch2_check_discard_freespace_key(&trans, &iter)); err: bch2_trans_exit(&trans); return ret < 0 ? ret : 0; @@ -913,32 +1020,53 @@ int bch2_check_alloc_to_lru_refs(struct bch_fs *c) bch2_trans_init(&trans, c, 0, 0); - for_each_btree_key(&trans, iter, BTREE_ID_alloc, POS_MIN, - BTREE_ITER_PREFETCH, k, ret) { - ret = __bch2_trans_do(&trans, NULL, NULL, - BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW, - bch2_check_alloc_to_lru_ref(&trans, &iter)); - if (ret) - break; - } - bch2_trans_iter_exit(&trans, &iter); + 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)); bch2_trans_exit(&trans); return ret < 0 ? ret : 0; } -static int bch2_clear_need_discard(struct btree_trans *trans, struct bpos pos, - struct bch_dev *ca, bool *discard_done) +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 bch_fs *c = trans->c; - struct btree_iter iter; + struct bpos pos = need_discard_iter->pos; + struct btree_iter iter = { NULL }; struct bkey_s_c k; + struct bch_dev *ca; struct bkey_i_alloc_v4 *a; struct printbuf buf = PRINTBUF; - int ret; + 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; + } + + if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) { + (*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)++; + goto out; + } - bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, pos, + bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, + need_discard_iter->pos, BTREE_ITER_CACHED); k = bch2_btree_iter_peek_slot(&iter); ret = bkey_err(k); @@ -974,7 +1102,8 @@ static int bch2_clear_need_discard(struct btree_trans *trans, struct bpos pos, goto out; } - if (!*discard_done && ca->mi.discard && !c->opts.nochanges) { + if (!bkey_eq(*discard_pos_done, iter.pos) && + ca->mi.discard && !c->opts.nochanges) { /* * This works without any other locks because this is the only * thread that removes items from the need_discard tree @@ -983,20 +1112,32 @@ static int bch2_clear_need_discard(struct btree_trans *trans, struct bpos pos, blkdev_issue_discard(ca->disk_sb.bdev, k.k->p.offset * ca->mi.bucket_size, ca->mi.bucket_size, - GFP_KERNEL, 0); - *discard_done = true; + GFP_KERNEL); - ret = bch2_trans_relock(trans) ? 0 : -EINTR; + ret = bch2_trans_relock(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); + ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?: + bch2_trans_commit(trans, NULL, NULL, + BTREE_INSERT_USE_RESERVE|BTREE_INSERT_NOFAIL); + if (ret) + goto out; + + if (did_discard) { + this_cpu_inc(c->counters[BCH_COUNTER_bucket_discard]); + (*discarded)++; + } out: bch2_trans_iter_exit(trans, &iter); + percpu_ref_put(&ca->io_ref); printbuf_exit(&buf); return ret; } @@ -1004,60 +1145,27 @@ out: static void bch2_do_discards_work(struct work_struct *work) { struct bch_fs *c = container_of(work, struct bch_fs, discard_work); - struct bch_dev *ca = NULL; 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 bpos discard_pos_done = POS_MAX; int ret; bch2_trans_init(&trans, c, 0, 0); - for_each_btree_key(&trans, iter, BTREE_ID_need_discard, - POS_MIN, 0, k, ret) { - bool discard_done = false; - - if (ca && k.k->p.inode != ca->dev_idx) { - percpu_ref_put(&ca->io_ref); - ca = NULL; - } - - if (!ca) { - ca = bch_dev_bkey_exists(c, k.k->p.inode); - if (!percpu_ref_tryget(&ca->io_ref)) { - ca = NULL; - bch2_btree_iter_set_pos(&iter, POS(k.k->p.inode + 1, 0)); - continue; - } - } - - seen++; - - if (bch2_bucket_is_open_safe(c, k.k->p.inode, k.k->p.offset)) { - open++; - continue; - } - - if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal, - c->journal.flushed_seq_ondisk, - k.k->p.inode, k.k->p.offset)) { - need_journal_commit++; - continue; - } - - ret = __bch2_trans_do(&trans, NULL, NULL, - BTREE_INSERT_USE_RESERVE| - BTREE_INSERT_NOFAIL, - bch2_clear_need_discard(&trans, k.k->p, ca, &discard_done)); - if (ret) - break; - - discarded++; - } - bch2_trans_iter_exit(&trans, &iter); - - if (ca) - percpu_ref_put(&ca->io_ref); + /* + * 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); @@ -1066,36 +1174,31 @@ static void bch2_do_discards_work(struct work_struct *work) percpu_ref_put(&c->writes); - trace_discard_buckets(c, seen, open, need_journal_commit, discarded, ret); + trace_discard_buckets(c, seen, open, need_journal_commit, discarded, + bch2_err_str(ret)); } void bch2_do_discards(struct bch_fs *c) { - if (percpu_ref_tryget(&c->writes) && + if (percpu_ref_tryget_live(&c->writes) && !queue_work(system_long_wq, &c->discard_work)) percpu_ref_put(&c->writes); } -static int invalidate_one_bucket(struct btree_trans *trans, struct bch_dev *ca) +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 bch_fs *c = trans->c; - struct btree_iter lru_iter, alloc_iter = { NULL }; - struct bkey_s_c k; + struct btree_iter alloc_iter = { NULL }; struct bkey_i_alloc_v4 *a; - u64 bucket, idx; + struct bpos bucket; struct printbuf buf = PRINTBUF; - int ret; - - bch2_trans_iter_init(trans, &lru_iter, BTREE_ID_lru, - POS(ca->dev_idx, 0), 0); -next_lru: - k = bch2_btree_iter_peek(&lru_iter); - ret = bkey_err(k); - if (ret) - goto out; + unsigned cached_sectors; + int ret = 0; - if (!k.k || k.k->p.inode != ca->dev_idx) - goto out; + if (*nr_to_invalidate <= 0 || k.k->p.inode != dev_idx) + return 1; if (k.k->type != KEY_TYPE_lru) { prt_printf(&buf, "non lru key in lru btree:\n "); @@ -1103,25 +1206,22 @@ next_lru: if (!test_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags)) { bch_err(c, "%s", buf.buf); - bch2_btree_iter_advance(&lru_iter); - goto next_lru; } else { bch2_trans_inconsistent(trans, "%s", buf.buf); ret = -EINVAL; - goto out; } + + goto out; } - idx = k.k->p.offset; - bucket = le64_to_cpu(bkey_s_c_to_lru(k).v->idx); + bucket = POS(dev_idx, le64_to_cpu(bkey_s_c_to_lru(k).v->idx)); - a = bch2_trans_start_alloc_update(trans, &alloc_iter, - POS(ca->dev_idx, bucket)); + a = bch2_trans_start_alloc_update(trans, &alloc_iter, bucket); ret = PTR_ERR_OR_ZERO(a); if (ret) goto out; - if (idx != alloc_lru_idx(a->v)) { + 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 "); @@ -1129,15 +1229,19 @@ next_lru: if (!test_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags)) { bch_err(c, "%s", buf.buf); - bch2_btree_iter_advance(&lru_iter); - goto next_lru; } else { bch2_trans_inconsistent(trans, "%s", buf.buf); ret = -EINVAL; - goto out; } + + goto out; } + if (!a->v.cached_sectors) + bch_err(c, "invalidating empty bucket, confused"); + + cached_sectors = a->v.cached_sectors; + SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false); a->v.gen++; a->v.data_type = 0; @@ -1146,15 +1250,17 @@ next_lru: a->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now); a->v.io_time[WRITE] = atomic64_read(&c->io_clock[WRITE].now); - ret = bch2_trans_update(trans, &alloc_iter, &a->k_i, - BTREE_TRIGGER_BUCKET_INVALIDATE); + 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); if (ret) goto out; - trace_invalidate_bucket(c, a->k.p.inode, a->k.p.offset); + trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors); + --*nr_to_invalidate; out: bch2_trans_iter_exit(trans, &alloc_iter); - bch2_trans_iter_exit(trans, &lru_iter); printbuf_exit(&buf); return ret; } @@ -1164,6 +1270,8 @@ 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; int ret = 0; @@ -1173,11 +1281,14 @@ static void bch2_do_invalidates_work(struct work_struct *work) s64 nr_to_invalidate = should_invalidate_buckets(ca, bch2_dev_usage_read(ca)); - while (!ret && nr_to_invalidate-- >= 0) - ret = __bch2_trans_do(&trans, NULL, NULL, - BTREE_INSERT_USE_RESERVE| - BTREE_INSERT_NOFAIL, - invalidate_one_bucket(&trans, 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)); + + if (ret < 0) { + percpu_ref_put(&ca->ref); + break; + } } bch2_trans_exit(&trans); @@ -1186,8 +1297,21 @@ static void bch2_do_invalidates_work(struct work_struct *work) void bch2_do_invalidates(struct bch_fs *c) { - if (percpu_ref_tryget(&c->writes)) - queue_work(system_long_wq, &c->invalidate_work); + if (percpu_ref_tryget_live(&c->writes) && + !queue_work(system_long_wq, &c->invalidate_work)) + percpu_ref_put(&c->writes); +} + +static int bucket_freespace_init(struct btree_trans *trans, struct btree_iter *iter, + struct bkey_s_c k, struct bch_dev *ca) +{ + struct bch_alloc_v4 a; + + if (iter->pos.offset >= ca->mi.nbuckets) + return 1; + + bch2_alloc_to_v4(k, &a); + return bch2_bucket_do_index(trans, k, &a, true); } static int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca) @@ -1195,32 +1319,21 @@ static int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca) struct btree_trans trans; struct btree_iter iter; struct bkey_s_c k; - struct bch_alloc_v4 a; struct bch_member *m; int ret; bch2_trans_init(&trans, c, 0, 0); - for_each_btree_key(&trans, iter, BTREE_ID_alloc, - POS(ca->dev_idx, ca->mi.first_bucket), - BTREE_ITER_SLOTS| - BTREE_ITER_PREFETCH, k, ret) { - if (iter.pos.offset >= ca->mi.nbuckets) - break; - - bch2_alloc_to_v4(k, &a); - ret = __bch2_trans_do(&trans, NULL, NULL, - BTREE_INSERT_LAZY_RW, - bch2_bucket_do_index(&trans, k, &a, true)); - if (ret) - break; - } - bch2_trans_iter_exit(&trans, &iter); + ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_alloc, + POS(ca->dev_idx, ca->mi.first_bucket), + BTREE_ITER_SLOTS|BTREE_ITER_PREFETCH, k, + NULL, NULL, BTREE_INSERT_LAZY_RW, + bucket_freespace_init(&trans, &iter, k, ca)); bch2_trans_exit(&trans); - if (ret) { - bch_err(ca, "error initializing free space: %i", ret); + if (ret < 0) { + bch_err(ca, "error initializing free space: %s", bch2_err_str(ret)); return ret; } @@ -1229,7 +1342,7 @@ static int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca) SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true); mutex_unlock(&c->sb_lock); - return ret; + return 0; } int bch2_fs_freespace_init(struct bch_fs *c)