diff options
Diffstat (limited to 'fs/bcachefs/alloc_background.c')
| -rw-r--r-- | fs/bcachefs/alloc_background.c | 1525 |
1 files changed, 872 insertions, 653 deletions
diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c index 4ff56fa4d539..3ea809990ef1 100644 --- a/fs/bcachefs/alloc_background.c +++ b/fs/bcachefs/alloc_background.c @@ -3,6 +3,7 @@ #include "alloc_background.h" #include "alloc_foreground.h" #include "backpointers.h" +#include "bkey_buf.h" #include "btree_cache.h" #include "btree_io.h" #include "btree_key_cache.h" @@ -14,6 +15,7 @@ #include "buckets_waiting_for_journal.h" #include "clock.h" #include "debug.h" +#include "disk_accounting.h" #include "ec.h" #include "error.h" #include "lru.h" @@ -28,8 +30,9 @@ #include <linux/rcupdate.h> #include <linux/sched/task.h> #include <linux/sort.h> +#include <linux/jiffies.h> -static void bch2_discard_one_bucket_fast(struct bch_fs *c, struct bpos bucket); +static void bch2_discard_one_bucket_fast(struct bch_dev *, u64); /* Persistent alloc info: */ @@ -194,99 +197,119 @@ static unsigned bch_alloc_v1_val_u64s(const struct bch_alloc *a) return DIV_ROUND_UP(bytes, sizeof(u64)); } -int bch2_alloc_v1_invalid(struct bch_fs *c, struct bkey_s_c k, - enum bkey_invalid_flags flags, - struct printbuf *err) +int bch2_alloc_v1_validate(struct bch_fs *c, struct bkey_s_c k, + struct bkey_validate_context from) { struct bkey_s_c_alloc a = bkey_s_c_to_alloc(k); int ret = 0; /* allow for unknown fields */ - bkey_fsck_err_on(bkey_val_u64s(a.k) < bch_alloc_v1_val_u64s(a.v), c, err, - alloc_v1_val_size_bad, + bkey_fsck_err_on(bkey_val_u64s(a.k) < bch_alloc_v1_val_u64s(a.v), + c, 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(struct bch_fs *c, struct bkey_s_c k, - enum bkey_invalid_flags flags, - struct printbuf *err) +int bch2_alloc_v2_validate(struct bch_fs *c, struct bkey_s_c k, + struct bkey_validate_context from) { struct bkey_alloc_unpacked u; int ret = 0; - bkey_fsck_err_on(bch2_alloc_unpack_v2(&u, k), c, err, - alloc_v2_unpack_error, + bkey_fsck_err_on(bch2_alloc_unpack_v2(&u, k), + c, alloc_v2_unpack_error, "unpack error"); fsck_err: return ret; } -int bch2_alloc_v3_invalid(struct bch_fs *c, struct bkey_s_c k, - enum bkey_invalid_flags flags, - struct printbuf *err) +int bch2_alloc_v3_validate(struct bch_fs *c, struct bkey_s_c k, + struct bkey_validate_context from) { struct bkey_alloc_unpacked u; int ret = 0; - bkey_fsck_err_on(bch2_alloc_unpack_v3(&u, k), c, err, - alloc_v2_unpack_error, + bkey_fsck_err_on(bch2_alloc_unpack_v3(&u, k), + c, alloc_v2_unpack_error, "unpack error"); fsck_err: return ret; } -int bch2_alloc_v4_invalid(struct bch_fs *c, struct bkey_s_c k, - enum bkey_invalid_flags flags, struct printbuf *err) +int bch2_alloc_v4_validate(struct bch_fs *c, struct bkey_s_c k, + struct bkey_validate_context from) { - struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k); + struct bch_alloc_v4 a; int ret = 0; - bkey_fsck_err_on(alloc_v4_u64s(a.v) > bkey_val_u64s(k.k), c, err, - alloc_v4_val_size_bad, + bkey_val_copy(&a, bkey_s_c_to_alloc_v4(k)); + + bkey_fsck_err_on(alloc_v4_u64s_noerror(&a) > bkey_val_u64s(k.k), + c, alloc_v4_val_size_bad, "bad val size (%u > %zu)", - alloc_v4_u64s(a.v), bkey_val_u64s(k.k)); + alloc_v4_u64s_noerror(&a), bkey_val_u64s(k.k)); - 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, + bkey_fsck_err_on(!BCH_ALLOC_V4_BACKPOINTERS_START(&a) && + BCH_ALLOC_V4_NR_BACKPOINTERS(&a), + c, alloc_v4_backpointers_start_bad, "invalid backpointers_start"); - bkey_fsck_err_on(alloc_data_type(*a.v, a.v->data_type) != a.v->data_type, c, err, - alloc_key_data_type_bad, + bkey_fsck_err_on(alloc_data_type(a, a.data_type) != a.data_type, + c, 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)); + a.data_type, alloc_data_type(a, a.data_type)); + + for (unsigned i = 0; i < 2; i++) + bkey_fsck_err_on(a.io_time[i] > LRU_TIME_MAX, + c, alloc_key_io_time_bad, + "invalid io_time[%s]: %llu, max %llu", + i == READ ? "read" : "write", + a.io_time[i], LRU_TIME_MAX); + + unsigned stripe_sectors = BCH_ALLOC_V4_BACKPOINTERS_START(&a) * sizeof(u64) > + offsetof(struct bch_alloc_v4, stripe_sectors) + ? a.stripe_sectors + : 0; - switch (a.v->data_type) { + switch (a.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"); + bkey_fsck_err_on(stripe_sectors || + a.dirty_sectors || + a.cached_sectors || + a.stripe, + c, alloc_key_empty_but_have_data, + "empty data type free but have data %u.%u.%u %u", + stripe_sectors, + a.dirty_sectors, + a.cached_sectors, + a.stripe); 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, + bkey_fsck_err_on(!a.dirty_sectors && + !stripe_sectors, + c, alloc_key_dirty_sectors_0, "data_type %s but dirty_sectors==0", - bch2_data_type_str(a.v->data_type)); + bch2_data_type_str(a.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, + bkey_fsck_err_on(!a.cached_sectors || + a.dirty_sectors || + stripe_sectors || + a.stripe, + c, alloc_key_cached_inconsistency, "data type inconsistency"); - bkey_fsck_err_on(!a.v->io_time[READ] && + bkey_fsck_err_on(!a.io_time[READ] && c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_to_lru_refs, - c, err, alloc_key_cached_but_read_time_zero, + c, alloc_key_cached_but_read_time_zero, "cached bucket with read_time == 0"); break; case BCH_DATA_stripe: @@ -299,9 +322,9 @@ fsck_err: 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->journal_seq_nonempty = swab64(a->journal_seq_nonempty); + a->journal_seq_empty = swab64(a->journal_seq_empty); a->flags = swab32(a->flags); a->dirty_sectors = swab32(a->dirty_sectors); a->cached_sectors = swab32(a->cached_sectors); @@ -309,20 +332,14 @@ 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++) { - bp->bucket_offset = swab40(bp->bucket_offset); - bp->bucket_len = swab32(bp->bucket_len); - bch2_bpos_swab(&bp->pos); - } + a->stripe_sectors = swab32(a->stripe_sectors); } void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k) { struct bch_alloc_v4 _a; const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a); + struct bch_dev *ca = c ? bch2_dev_bucket_tryget_noerror(c, k.k->p) : NULL; prt_newline(out); printbuf_indent_add(out, 2); @@ -330,28 +347,24 @@ void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c 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); - 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, "fragmentation %llu", a->fragmentation_lru); - prt_newline(out); - prt_printf(out, "bp_start %llu", BCH_ALLOC_V4_BACKPOINTERS_START(a)); + prt_printf(out, "journal_seq_nonempty %llu\n", a->journal_seq_nonempty); + prt_printf(out, "journal_seq_empty %llu\n", a->journal_seq_empty); + prt_printf(out, "need_discard %llu\n", BCH_ALLOC_V4_NEED_DISCARD(a)); + prt_printf(out, "need_inc_gen %llu\n", BCH_ALLOC_V4_NEED_INC_GEN(a)); + prt_printf(out, "dirty_sectors %u\n", a->dirty_sectors); + prt_printf(out, "stripe_sectors %u\n", a->stripe_sectors); + prt_printf(out, "cached_sectors %u\n", a->cached_sectors); + prt_printf(out, "stripe %u\n", a->stripe); + prt_printf(out, "stripe_redundancy %u\n", a->stripe_redundancy); + prt_printf(out, "io_time[READ] %llu\n", a->io_time[READ]); + prt_printf(out, "io_time[WRITE] %llu\n", a->io_time[WRITE]); + + if (ca) + prt_printf(out, "fragmentation %llu\n", alloc_lru_idx_fragmentation(*a, ca)); + prt_printf(out, "bp_start %llu\n", BCH_ALLOC_V4_BACKPOINTERS_START(a)); printbuf_indent_sub(out, 2); + + bch2_dev_put(ca); } void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out) @@ -373,7 +386,7 @@ void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out) struct bkey_alloc_unpacked u = bch2_alloc_unpack(k); *out = (struct bch_alloc_v4) { - .journal_seq = u.journal_seq, + .journal_seq_nonempty = u.journal_seq, .flags = u.need_discard, .gen = u.gen, .oldest_gen = u.oldest_gen, @@ -439,22 +452,18 @@ struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *trans, struct b } struct bkey_i_alloc_v4 * -bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter, - struct bpos pos) +bch2_trans_start_alloc_update_noupdate(struct btree_trans *trans, struct btree_iter *iter, + struct bpos pos) { - struct bkey_s_c k; - struct bkey_i_alloc_v4 *a; - int ret; - - k = bch2_bkey_get_iter(trans, iter, BTREE_ID_alloc, pos, - BTREE_ITER_WITH_UPDATES| - BTREE_ITER_CACHED| - BTREE_ITER_INTENT); - ret = bkey_err(k); + struct bkey_s_c k = bch2_bkey_get_iter(trans, iter, BTREE_ID_alloc, pos, + BTREE_ITER_with_updates| + BTREE_ITER_cached| + BTREE_ITER_intent); + int ret = bkey_err(k); if (unlikely(ret)) return ERR_PTR(ret); - a = bch2_alloc_to_v4_mut_inlined(trans, k); + struct bkey_i_alloc_v4 *a = bch2_alloc_to_v4_mut_inlined(trans, k); ret = PTR_ERR_OR_ZERO(a); if (unlikely(ret)) goto err; @@ -464,6 +473,21 @@ err: return ERR_PTR(ret); } +__flatten +struct bkey_i_alloc_v4 *bch2_trans_start_alloc_update(struct btree_trans *trans, struct bpos pos, + enum btree_iter_update_trigger_flags flags) +{ + struct btree_iter iter; + struct bkey_i_alloc_v4 *a = bch2_trans_start_alloc_update_noupdate(trans, &iter, pos); + int ret = PTR_ERR_OR_ZERO(a); + if (ret) + return ERR_PTR(ret); + + ret = bch2_trans_update(trans, &iter, &a->k_i, flags); + bch2_trans_iter_exit(trans, &iter); + return unlikely(ret) ? ERR_PTR(ret) : a; +} + static struct bpos alloc_gens_pos(struct bpos pos, unsigned *offset) { *offset = pos.offset & KEY_TYPE_BUCKET_GENS_MASK; @@ -486,14 +510,13 @@ static unsigned alloc_gen(struct bkey_s_c k, unsigned 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 bch2_bucket_gens_validate(struct bch_fs *c, struct bkey_s_c k, + struct bkey_validate_context from) { int ret = 0; - bkey_fsck_err_on(bkey_val_bytes(k.k) != sizeof(struct bch_bucket_gens), c, err, - bucket_gens_val_size_bad, + bkey_fsck_err_on(bkey_val_bytes(k.k) != sizeof(struct bch_bucket_gens), + c, bucket_gens_val_size_bad, "bad val size (%zu != %zu)", bkey_val_bytes(k.k), sizeof(struct bch_bucket_gens)); fsck_err: @@ -520,7 +543,7 @@ int bch2_bucket_gens_init(struct bch_fs *c) int ret; ret = for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN, - BTREE_ITER_PREFETCH, k, ({ + BTREE_ITER_prefetch, k, ({ /* * Not a fsck error because this is checked/repaired by * bch2_check_alloc_key() which runs later: @@ -534,7 +557,7 @@ int bch2_bucket_gens_init(struct bch_fs *c) struct bpos pos = alloc_gens_pos(iter.pos, &offset); int ret2 = 0; - if (have_bucket_gens_key && bkey_cmp(iter.pos, pos)) { + if (have_bucket_gens_key && !bkey_eq(g.k.p, pos)) { ret2 = bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0) ?: bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc); if (ret2) @@ -567,29 +590,29 @@ iter_err: int bch2_alloc_read(struct bch_fs *c) { struct btree_trans *trans = bch2_trans_get(c); + struct bch_dev *ca = NULL; 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, ({ + 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; - + ca = bch2_dev_iterate(c, ca, k.k->p.inode); /* * 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)) + if (!ca) { + bch2_btree_iter_set_pos(&iter, POS(k.k->p.inode + 1, 0)); continue; + } - struct bch_dev *ca = bch_dev_bkey_exists(c, k.k->p.inode); + const struct bch_bucket_gens *g = bkey_s_c_to_bucket_gens(k).v; for (u64 b = max_t(u64, ca->mi.first_bucket, start); b < min_t(u64, ca->mi.nbuckets, end); @@ -599,15 +622,26 @@ int bch2_alloc_read(struct bch_fs *c) })); } else { ret = for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN, - BTREE_ITER_PREFETCH, k, ({ + BTREE_ITER_prefetch, k, ({ + ca = bch2_dev_iterate(c, ca, k.k->p.inode); /* * 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)) + if (!ca) { + bch2_btree_iter_set_pos(&iter, POS(k.k->p.inode + 1, 0)); continue; + } - struct bch_dev *ca = bch_dev_bkey_exists(c, k.k->p.inode); + if (k.k->p.offset < ca->mi.first_bucket) { + bch2_btree_iter_set_pos(&iter, POS(k.k->p.inode, ca->mi.first_bucket)); + continue; + } + + if (k.k->p.offset >= ca->mi.nbuckets) { + bch2_btree_iter_set_pos(&iter, POS(k.k->p.inode + 1, 0)); + continue; + } struct bch_alloc_v4 a; *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen; @@ -615,8 +649,8 @@ int bch2_alloc_read(struct bch_fs *c) })); } + bch2_dev_put(ca); bch2_trans_put(trans); - up_read(&c->gc_lock); bch_err_fn(c, ret); return ret; @@ -624,74 +658,80 @@ int bch2_alloc_read(struct bch_fs *c) /* Free space/discard btree: */ +static int __need_discard_or_freespace_err(struct btree_trans *trans, + struct bkey_s_c alloc_k, + bool set, bool discard, bool repair) +{ + struct bch_fs *c = trans->c; + enum bch_fsck_flags flags = FSCK_CAN_IGNORE|(repair ? FSCK_CAN_FIX : 0); + enum bch_sb_error_id err_id = discard + ? BCH_FSCK_ERR_need_discard_key_wrong + : BCH_FSCK_ERR_freespace_key_wrong; + enum btree_id btree = discard ? BTREE_ID_need_discard : BTREE_ID_freespace; + struct printbuf buf = PRINTBUF; + + bch2_bkey_val_to_text(&buf, c, alloc_k); + + int ret = __bch2_fsck_err(NULL, trans, flags, err_id, + "bucket incorrectly %sset in %s btree\n" + " %s", + set ? "" : "un", + bch2_btree_id_str(btree), + buf.buf); + if (ret == -BCH_ERR_fsck_ignore || + ret == -BCH_ERR_fsck_errors_not_fixed) + ret = 0; + + printbuf_exit(&buf); + return ret; +} + +#define need_discard_or_freespace_err(...) \ + fsck_err_wrap(__need_discard_or_freespace_err(__VA_ARGS__)) + +#define need_discard_or_freespace_err_on(cond, ...) \ + (unlikely(cond) ? need_discard_or_freespace_err(__VA_ARGS__) : false) + static int bch2_bucket_do_index(struct btree_trans *trans, + struct bch_dev *ca, struct bkey_s_c alloc_k, const struct bch_alloc_v4 *a, bool set) { - struct bch_fs *c = trans->c; - struct bch_dev *ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode); - struct btree_iter iter; - struct bkey_s_c old; - struct bkey_i *k; enum btree_id btree; - enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted; - enum bch_bkey_type new_type = set ? KEY_TYPE_set : KEY_TYPE_deleted; - struct printbuf buf = PRINTBUF; - int ret; + struct bpos pos; if (a->data_type != BCH_DATA_free && a->data_type != BCH_DATA_need_discard) return 0; - k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k)); - if (IS_ERR(k)) - return PTR_ERR(k); - - bkey_init(&k->k); - k->k.type = new_type; - switch (a->data_type) { case BCH_DATA_free: btree = BTREE_ID_freespace; - k->k.p = alloc_freespace_pos(alloc_k.k->p, *a); - bch2_key_resize(&k->k, 1); + pos = alloc_freespace_pos(alloc_k.k->p, *a); break; case BCH_DATA_need_discard: btree = BTREE_ID_need_discard; - k->k.p = alloc_k.k->p; + pos = alloc_k.k->p; break; default: return 0; } - old = bch2_bkey_get_iter(trans, &iter, btree, - bkey_start_pos(&k->k), - BTREE_ITER_INTENT); - ret = bkey_err(old); + struct btree_iter iter; + struct bkey_s_c old = bch2_bkey_get_iter(trans, &iter, btree, pos, BTREE_ITER_intent); + int ret = bkey_err(old); if (ret) 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:%llu:%llu:0 (got %s should be %s)\n" - " for %s", - set ? "setting" : "clearing", - 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))) { - ret = -EIO; - goto err; - } + need_discard_or_freespace_err_on(ca->mi.freespace_initialized && + !old.k->type != set, + trans, alloc_k, set, + btree == BTREE_ID_need_discard, false); - ret = bch2_trans_update(trans, &iter, k, 0); -err: + ret = bch2_btree_bit_mod_iter(trans, &iter, set); +fsck_err: bch2_trans_iter_exit(trans, &iter); - printbuf_exit(&buf); return ret; } @@ -711,8 +751,8 @@ static noinline int bch2_bucket_gen_update(struct btree_trans *trans, return ret; k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_bucket_gens, pos, - BTREE_ITER_INTENT| - BTREE_ITER_WITH_UPDATES); + BTREE_ITER_intent| + BTREE_ITER_with_updates); ret = bkey_err(k); if (ret) return ret; @@ -731,31 +771,99 @@ static noinline int bch2_bucket_gen_update(struct btree_trans *trans, return ret; } +static inline int bch2_dev_data_type_accounting_mod(struct btree_trans *trans, struct bch_dev *ca, + enum bch_data_type data_type, + s64 delta_buckets, + s64 delta_sectors, + s64 delta_fragmented, unsigned flags) +{ + struct disk_accounting_pos acc = { + .type = BCH_DISK_ACCOUNTING_dev_data_type, + .dev_data_type.dev = ca->dev_idx, + .dev_data_type.data_type = data_type, + }; + s64 d[3] = { delta_buckets, delta_sectors, delta_fragmented }; + + return bch2_disk_accounting_mod(trans, &acc, d, 3, flags & BTREE_TRIGGER_gc); +} + +int bch2_alloc_key_to_dev_counters(struct btree_trans *trans, struct bch_dev *ca, + const struct bch_alloc_v4 *old, + const struct bch_alloc_v4 *new, + unsigned flags) +{ + s64 old_sectors = bch2_bucket_sectors(*old); + s64 new_sectors = bch2_bucket_sectors(*new); + if (old->data_type != new->data_type) { + int ret = bch2_dev_data_type_accounting_mod(trans, ca, new->data_type, + 1, new_sectors, bch2_bucket_sectors_fragmented(ca, *new), flags) ?: + bch2_dev_data_type_accounting_mod(trans, ca, old->data_type, + -1, -old_sectors, -bch2_bucket_sectors_fragmented(ca, *old), flags); + if (ret) + return ret; + } else if (old_sectors != new_sectors) { + int ret = bch2_dev_data_type_accounting_mod(trans, ca, new->data_type, + 0, + new_sectors - old_sectors, + bch2_bucket_sectors_fragmented(ca, *new) - + bch2_bucket_sectors_fragmented(ca, *old), flags); + if (ret) + return ret; + } + + s64 old_unstriped = bch2_bucket_sectors_unstriped(*old); + s64 new_unstriped = bch2_bucket_sectors_unstriped(*new); + if (old_unstriped != new_unstriped) { + int ret = bch2_dev_data_type_accounting_mod(trans, ca, BCH_DATA_unstriped, + !!new_unstriped - !!old_unstriped, + new_unstriped - old_unstriped, + 0, + flags); + if (ret) + return ret; + } + + return 0; +} + 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) + enum btree_iter_update_trigger_flags flags) { struct bch_fs *c = trans->c; + struct printbuf buf = PRINTBUF; int ret = 0; - if (bch2_trans_inconsistent_on(!bch2_dev_bucket_exists(c, new.k->p), trans, - "alloc key for invalid device or bucket")) + struct bch_dev *ca = bch2_dev_bucket_tryget(c, new.k->p); + if (!ca) return -EIO; - struct bch_dev *ca = bch_dev_bkey_exists(c, new.k->p.inode); - struct bch_alloc_v4 old_a_convert; const struct bch_alloc_v4 *old_a = bch2_alloc_to_v4(old, &old_a_convert); - if (flags & BTREE_TRIGGER_TRANSACTIONAL) { - struct bch_alloc_v4 *new_a = bkey_s_to_alloc_v4(new).v; + struct bch_alloc_v4 *new_a; + if (likely(new.k->type == KEY_TYPE_alloc_v4)) { + new_a = bkey_s_to_alloc_v4(new).v; + } else { + BUG_ON(!(flags & (BTREE_TRIGGER_gc|BTREE_TRIGGER_check_repair))); - new_a->data_type = alloc_data_type(*new_a, new_a->data_type); + struct bkey_i_alloc_v4 *new_ka = bch2_alloc_to_v4_mut_inlined(trans, new.s_c); + ret = PTR_ERR_OR_ZERO(new_ka); + if (unlikely(ret)) + goto err; + new_a = &new_ka->v; + } + + if (flags & BTREE_TRIGGER_transactional) { + alloc_data_type_set(new_a, new_a->data_type); + + int is_empty_delta = (int) data_type_is_empty(new_a->data_type) - + (int) data_type_is_empty(old_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)); + if (is_empty_delta < 0) { + new_a->io_time[READ] = bch2_current_io_time(c, READ); + new_a->io_time[WRITE]= bch2_current_io_time(c, WRITE); SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true); SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true); } @@ -765,20 +873,21 @@ int bch2_trigger_alloc(struct btree_trans *trans, !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); + alloc_data_type_set(new_a, new_a->data_type); } 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); + ret = bch2_bucket_do_index(trans, ca, old, old_a, false) ?: + bch2_bucket_do_index(trans, ca, new.s_c, new_a, true); if (ret) - return ret; + goto err; } 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)); + new_a->io_time[READ] = bch2_current_io_time(c, READ); u64 old_lru = alloc_lru_idx_read(*old_a); u64 new_lru = alloc_lru_idx_read(*new_a); @@ -787,133 +896,150 @@ int bch2_trigger_alloc(struct btree_trans *trans, bucket_to_u64(new.k->p), old_lru, new_lru); if (ret) - return ret; + goto err; } - 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) { + old_lru = alloc_lru_idx_fragmentation(*old_a, ca); + new_lru = alloc_lru_idx_fragmentation(*new_a, ca); + if (old_lru != new_lru) { ret = bch2_lru_change(trans, BCH_LRU_FRAGMENTATION_START, bucket_to_u64(new.k->p), - old_a->fragmentation_lru, new_a->fragmentation_lru); + old_lru, new_lru); if (ret) - return ret; + goto err; } if (old_a->gen != new_a->gen) { ret = bch2_bucket_gen_update(trans, new.k->p, new_a->gen); if (ret) - return ret; + goto err; } - /* - * need to know if we're getting called from the invalidate path or - * not: - */ - - if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) && + 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)); + ret = bch2_mod_dev_cached_sectors(trans, ca->dev_idx, + -((s64) old_a->cached_sectors), + flags & BTREE_TRIGGER_gc); if (ret) - return ret; + goto err; } + + ret = bch2_alloc_key_to_dev_counters(trans, ca, old_a, new_a, flags); + if (ret) + goto err; } - 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_atomic) && (flags & BTREE_TRIGGER_insert)) { + u64 transaction_seq = trans->journal_res.seq; + BUG_ON(!transaction_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 (log_fsck_err_on(transaction_seq && new_a->journal_seq_nonempty > transaction_seq, + trans, alloc_key_journal_seq_in_future, + "bucket journal seq in future (currently at %llu)\n%s", + journal_cur_seq(&c->journal), + (bch2_bkey_val_to_text(&buf, c, new.s_c), buf.buf))) + new_a->journal_seq_nonempty = transaction_seq; - /* - * 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; + int is_empty_delta = (int) data_type_is_empty(new_a->data_type) - + (int) data_type_is_empty(old_a->data_type); + + /* + * Record journal sequence number of empty -> nonempty transition: + * Note that there may be multiple empty -> nonempty + * transitions, data in a bucket may be overwritten while we're + * still writing to it - so be careful to only record the first: + * */ + if (is_empty_delta < 0 && + new_a->journal_seq_empty <= c->journal.flushed_seq_ondisk) { + new_a->journal_seq_nonempty = transaction_seq; + new_a->journal_seq_empty = 0; } - 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, - "setting bucket_needs_journal_commit: %s", bch2_err_str(ret)); - return ret; + /* + * Bucket becomes empty: mark it as waiting for a journal flush, + * unless updates since empty -> nonempty transition were never + * flushed - we may need to ask the journal not to flush + * intermediate sequence numbers: + */ + if (is_empty_delta > 0) { + if (new_a->journal_seq_nonempty == transaction_seq || + bch2_journal_noflush_seq(&c->journal, + new_a->journal_seq_nonempty, + transaction_seq)) { + new_a->journal_seq_nonempty = new_a->journal_seq_empty = 0; + } else { + new_a->journal_seq_empty = transaction_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, + transaction_seq); + if (bch2_fs_fatal_err_on(ret, c, + "setting bucket_needs_journal_commit: %s", + bch2_err_str(ret))) + goto err; } } - 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); + if (new_a->gen != old_a->gen) { + rcu_read_lock(); + u8 *gen = bucket_gen(ca, new.k->p.offset); + if (unlikely(!gen)) { + rcu_read_unlock(); + goto invalid_bucket; + } + *gen = new_a->gen; + rcu_read_unlock(); + } #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) +#define bucket_flushed(a) (a->journal_seq_empty <= c->journal.flushed_seq_ondisk) if (statechange(a->data_type == BCH_DATA_free) && bucket_flushed(new_a)) closure_wake_up(&c->freelist_wait); if (statechange(a->data_type == BCH_DATA_need_discard) && - !bch2_bucket_is_open(c, new.k->p.inode, new.k->p.offset) && + !bch2_bucket_is_open_safe(c, new.k->p.inode, new.k->p.offset) && bucket_flushed(new_a)) - bch2_discard_one_bucket_fast(c, new.k->p); + bch2_discard_one_bucket_fast(ca, new.k->p.offset); 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); + bch2_dev_do_invalidates(ca); if (statechange(a->data_type == BCH_DATA_need_gc_gens)) - bch2_do_gc_gens(c); + bch2_gc_gens_async(c); } - 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); - - percpu_down_read(&c->mark_lock); + if ((flags & BTREE_TRIGGER_gc) && (flags & BTREE_TRIGGER_insert)) { + rcu_read_lock(); struct bucket *g = gc_bucket(ca, new.k->p.offset); - - 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; - - bucket_unlock(g); - percpu_up_read(&c->mark_lock); + if (unlikely(!g)) { + rcu_read_unlock(); + goto invalid_bucket; + } + g->gen_valid = 1; + g->gen = new_a->gen; + rcu_read_unlock(); } - - return 0; +err: +fsck_err: + printbuf_exit(&buf); + bch2_dev_put(ca); + return ret; +invalid_bucket: + bch2_fs_inconsistent(c, "reference to invalid bucket\n %s", + (bch2_bkey_val_to_text(&buf, c, new.s_c), buf.buf)); + ret = -EIO; + goto err; } /* - * This synthesizes deleted extents for holes, similar to BTREE_ITER_SLOTS for + * 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) @@ -941,7 +1067,7 @@ static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos * btree node min/max is a closed interval, upto takes a half * open interval: */ - k = bch2_btree_iter_peek_upto(&iter2, end); + k = bch2_btree_iter_peek_max(&iter2, end); next = iter2.pos; bch2_trans_iter_exit(iter->trans, &iter2); @@ -958,35 +1084,34 @@ static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos } } -static bool next_bucket(struct bch_fs *c, struct bpos *bucket) +static bool next_bucket(struct bch_fs *c, struct bch_dev **ca, struct bpos *bucket) { - struct bch_dev *ca; - - if (bch2_dev_bucket_exists(c, *bucket)) - return true; + if (*ca) { + if (bucket->offset < (*ca)->mi.first_bucket) + bucket->offset = (*ca)->mi.first_bucket; - 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; + if (bucket->offset < (*ca)->mi.nbuckets) return true; - } + bch2_dev_put(*ca); + *ca = NULL; 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); + *ca = __bch2_next_dev_idx(c, bucket->inode, NULL); + if (*ca) { + *bucket = POS((*ca)->dev_idx, (*ca)->mi.first_bucket); + bch2_dev_get(*ca); + } rcu_read_unlock(); - return ca != NULL; + return *ca != NULL; } -static struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, struct bkey *hole) +static struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, + struct bch_dev **ca, struct bkey *hole) { struct bch_fs *c = iter->trans->c; struct bkey_s_c k; @@ -995,22 +1120,21 @@ again: if (bkey_err(k)) return k; + *ca = bch2_dev_iterate_noerror(c, *ca, k.k->p.inode); + if (!k.k->type) { - struct bpos bucket = bkey_start_pos(k.k); + struct bpos hole_start = bkey_start_pos(k.k); - if (!bch2_dev_bucket_exists(c, bucket)) { - if (!next_bucket(c, &bucket)) + if (!*ca || !bucket_valid(*ca, hole_start.offset)) { + if (!next_bucket(c, ca, &hole_start)) return bkey_s_c_null; - bch2_btree_iter_set_pos(iter, bucket); + bch2_btree_iter_set_pos(iter, hole_start); 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); - } + if (k.k->p.offset > (*ca)->mi.nbuckets) + bch2_key_resize(hole, (*ca)->mi.nbuckets - hole_start.offset); } return k; @@ -1025,85 +1149,51 @@ int bch2_check_alloc_key(struct btree_trans *trans, 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; unsigned gens_offset; struct bkey_s_c k; struct printbuf buf = PRINTBUF; - int ret; + int ret = 0; - if (fsck_err_on(!bch2_dev_bucket_exists(c, alloc_k.k->p), c, - alloc_key_to_missing_dev_bucket, + struct bch_dev *ca = bch2_dev_bucket_tryget_noerror(c, alloc_k.k->p); + if (fsck_err_on(!ca, + trans, 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); + ret = bch2_btree_delete_at(trans, alloc_iter, 0); + if (!ca) + return ret; - ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode); if (!ca->mi.freespace_initialized) - return 0; + goto out; a = bch2_alloc_to_v4(alloc_k, &a_convert); - 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); 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, 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], - (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 = discard_key_type; - update->k.p = discard_iter->pos; - - ret = bch2_trans_update(trans, discard_iter, update, 0); + bool is_discarded = a->data_type == BCH_DATA_need_discard; + if (need_discard_or_freespace_err_on(!!k.k->type != is_discarded, + trans, alloc_k, !is_discarded, true, true)) { + ret = bch2_btree_bit_mod_iter(trans, discard_iter, is_discarded); 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 (fsck_err_on(k.k->type != freespace_key_type, - 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); + bool is_free = a->data_type == BCH_DATA_free; + if (need_discard_or_freespace_err_on(!!k.k->type != is_free, + trans, alloc_k, !is_free, false, true)) { + ret = bch2_btree_bit_mod_iter(trans, freespace_iter, is_free); if (ret) goto err; } @@ -1115,7 +1205,7 @@ int bch2_check_alloc_key(struct btree_trans *trans, goto err; if (fsck_err_on(a->gen != alloc_gen(k, gens_offset), - c, bucket_gens_key_wrong, + trans, bucket_gens_key_wrong, "incorrect gen in bucket_gens btree (got %u should be %u)\n" " %s", alloc_gen(k, gens_offset), a->gen, @@ -1141,25 +1231,25 @@ int bch2_check_alloc_key(struct btree_trans *trans, if (ret) goto err; } +out: err: fsck_err: + bch2_dev_put(ca); printbuf_exit(&buf); return ret; } static noinline_for_stack int bch2_check_alloc_hole_freespace(struct btree_trans *trans, + struct bch_dev *ca, 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; @@ -1173,7 +1263,7 @@ int bch2_check_alloc_hole_freespace(struct btree_trans *trans, *end = bkey_min(k.k->p, *end); if (fsck_err_on(k.k->type != KEY_TYPE_set, - c, freespace_hole_missing, + trans, freespace_hole_missing, "hole in alloc btree missing in freespace btree\n" " device %llu buckets %llu-%llu", freespace_iter->pos.inode, @@ -1209,7 +1299,6 @@ int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans, 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; @@ -1233,7 +1322,7 @@ int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans, bkey_reassemble(&g.k_i, k); for (i = gens_offset; i < gens_end_offset; i++) { - if (fsck_err_on(g.v.gens[i], c, + if (fsck_err_on(g.v.gens[i], trans, 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, @@ -1266,63 +1355,129 @@ fsck_err: return ret; } -static noinline_for_stack int bch2_check_discard_freespace_key(struct btree_trans *trans, - struct btree_iter *iter) +struct check_discard_freespace_key_async { + struct work_struct work; + struct bch_fs *c; + struct bbpos pos; +}; + +static int bch2_recheck_discard_freespace_key(struct btree_trans *trans, struct bbpos pos) +{ + struct btree_iter iter; + struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, pos.btree, pos.pos, 0); + int ret = bkey_err(k); + if (ret) + return ret; + + u8 gen; + ret = k.k->type != KEY_TYPE_set + ? bch2_check_discard_freespace_key(trans, &iter, &gen, false) + : 0; + bch2_trans_iter_exit(trans, &iter); + return ret; +} + +static void check_discard_freespace_key_work(struct work_struct *work) +{ + struct check_discard_freespace_key_async *w = + container_of(work, struct check_discard_freespace_key_async, work); + + bch2_trans_do(w->c, bch2_recheck_discard_freespace_key(trans, w->pos)); + bch2_write_ref_put(w->c, BCH_WRITE_REF_check_discard_freespace_key); + kfree(w); +} + +int bch2_check_discard_freespace_key(struct btree_trans *trans, struct btree_iter *iter, u8 *gen, + bool async_repair) { struct bch_fs *c = trans->c; - struct btree_iter alloc_iter; - struct bkey_s_c alloc_k; - struct bch_alloc_v4 a_convert; - const struct bch_alloc_v4 *a; - u64 genbits; - struct bpos pos; enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard ? BCH_DATA_need_discard : BCH_DATA_free; struct printbuf buf = PRINTBUF; - int ret; - pos = iter->pos; - pos.offset &= ~(~0ULL << 56); - genbits = iter->pos.offset & (~0ULL << 56); + struct bpos bucket = iter->pos; + bucket.offset &= ~(~0ULL << 56); + u64 genbits = iter->pos.offset & (~0ULL << 56); - alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc, pos, 0); - ret = bkey_err(alloc_k); + struct btree_iter alloc_iter; + struct bkey_s_c alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, + BTREE_ID_alloc, bucket, + async_repair ? BTREE_ITER_cached : 0); + int 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_id_str(iter->btree_id), pos.inode, pos.offset)) - goto delete; + if (!bch2_dev_bucket_exists(c, bucket)) { + if (fsck_err(trans, need_discard_freespace_key_to_invalid_dev_bucket, + "entry in %s btree for nonexistant dev:bucket %llu:%llu", + bch2_btree_id_str(iter->btree_id), bucket.inode, bucket.offset)) + goto delete; + ret = 1; + goto out; + } - a = bch2_alloc_to_v4(alloc_k, &a_convert); + struct bch_alloc_v4 a_convert; + const struct bch_alloc_v4 *a = bch2_alloc_to_v4(alloc_k, &a_convert); + + if (a->data_type != state || + (state == BCH_DATA_free && + genbits != alloc_freespace_genbits(*a))) { + if (fsck_err(trans, 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_id_str(iter->btree_id), + iter->pos.inode, + iter->pos.offset, + a->data_type == state, + genbits >> 56, alloc_freespace_genbits(*a) >> 56)) + goto delete; + ret = 1; + goto out; + } - if (fsck_err_on(a->data_type != state || - (state == BCH_DATA_free && - genbits != alloc_freespace_genbits(*a)), c, - 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_id_str(iter->btree_id), - iter->pos.inode, - iter->pos.offset, - a->data_type == state, - genbits >> 56, alloc_freespace_genbits(*a) >> 56)) - goto delete; + *gen = a->gen; out: fsck_err: - set_btree_iter_dontneed(&alloc_iter); + bch2_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) ?: - bch2_trans_commit(trans, NULL, NULL, - BCH_TRANS_COMMIT_no_enospc); - goto out; + if (!async_repair) { + ret = bch2_btree_bit_mod_iter(trans, iter, false) ?: + bch2_trans_commit(trans, NULL, NULL, + BCH_TRANS_COMMIT_no_enospc) ?: + -BCH_ERR_transaction_restart_commit; + goto out; + } else { + /* + * We can't repair here when called from the allocator path: the + * commit will recurse back into the allocator + */ + struct check_discard_freespace_key_async *w = + kzalloc(sizeof(*w), GFP_KERNEL); + if (!w) + goto out; + + if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_check_discard_freespace_key)) { + kfree(w); + goto out; + } + + INIT_WORK(&w->work, check_discard_freespace_key_work); + w->c = c; + w->pos = BBPOS(iter->btree_id, iter->pos); + queue_work(c->write_ref_wq, &w->work); + goto out; + } +} + +static int bch2_check_discard_freespace_key_fsck(struct btree_trans *trans, struct btree_iter *iter) +{ + u8 gen; + int ret = bch2_check_discard_freespace_key(trans, iter, &gen, false); + return ret < 0 ? ret : 0; } /* @@ -1337,33 +1492,28 @@ int bch2_check_bucket_gens_key(struct btree_trans *trans, { 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; + bool need_update = false; 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))) { + struct bch_dev *ca = bch2_dev_tryget_noerror(c, k.k->p.inode); + if (!ca) { + if (fsck_err(trans, 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, + start >= ca->mi.nbuckets, + trans, 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); @@ -1371,16 +1521,16 @@ int bch2_check_bucket_gens_key(struct btree_trans *trans, } 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, + if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], + trans, 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, + if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], + trans, 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; @@ -1398,6 +1548,7 @@ int bch2_check_bucket_gens_key(struct btree_trans *trans, } out: fsck_err: + bch2_dev_put(ca); printbuf_exit(&buf); return ret; } @@ -1406,25 +1557,26 @@ int bch2_check_alloc_info(struct bch_fs *c) { struct btree_trans *trans = bch2_trans_get(c); struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter; + struct bch_dev *ca = NULL; struct bkey hole; struct bkey_s_c k; int ret = 0; bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS_MIN, - BTREE_ITER_PREFETCH); + BTREE_ITER_prefetch); bch2_trans_iter_init(trans, &discard_iter, BTREE_ID_need_discard, POS_MIN, - BTREE_ITER_PREFETCH); + BTREE_ITER_prefetch); bch2_trans_iter_init(trans, &freespace_iter, BTREE_ID_freespace, POS_MIN, - BTREE_ITER_PREFETCH); + BTREE_ITER_prefetch); bch2_trans_iter_init(trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN, - BTREE_ITER_PREFETCH); + BTREE_ITER_prefetch); while (1) { struct bpos next; bch2_trans_begin(trans); - k = bch2_get_key_or_real_bucket_hole(&iter, &hole); + k = bch2_get_key_or_real_bucket_hole(&iter, &ca, &hole); ret = bkey_err(k); if (ret) goto bkey_err; @@ -1445,7 +1597,7 @@ int bch2_check_alloc_info(struct bch_fs *c) } else { next = k.k->p; - ret = bch2_check_alloc_hole_freespace(trans, + ret = bch2_check_alloc_hole_freespace(trans, ca, bkey_start_pos(k.k), &next, &freespace_iter) ?: @@ -1473,19 +1625,21 @@ bkey_err: bch2_trans_iter_exit(trans, &freespace_iter); bch2_trans_iter_exit(trans, &discard_iter); bch2_trans_iter_exit(trans, &iter); + bch2_dev_put(ca); + ca = NULL; if (ret < 0) goto err; ret = for_each_btree_key(trans, iter, BTREE_ID_need_discard, POS_MIN, - BTREE_ITER_PREFETCH, k, - bch2_check_discard_freespace_key(trans, &iter)); + BTREE_ITER_prefetch, k, + bch2_check_discard_freespace_key_fsck(trans, &iter)); if (ret) goto err; bch2_trans_iter_init(trans, &iter, BTREE_ID_freespace, POS_MIN, - BTREE_ITER_PREFETCH); + BTREE_ITER_prefetch); while (1) { bch2_trans_begin(trans); k = bch2_btree_iter_peek(&iter); @@ -1493,7 +1647,7 @@ bkey_err: break; ret = bkey_err(k) ?: - bch2_check_discard_freespace_key(trans, &iter); + bch2_check_discard_freespace_key_fsck(trans, &iter); if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) { ret = 0; continue; @@ -1515,7 +1669,7 @@ bkey_err: ret = for_each_btree_key_commit(trans, iter, BTREE_ID_bucket_gens, POS_MIN, - BTREE_ITER_PREFETCH, k, + BTREE_ITER_prefetch, k, NULL, NULL, BCH_TRANS_COMMIT_no_enospc, bch2_check_bucket_gens_key(trans, &iter, k)); err: @@ -1525,13 +1679,13 @@ err: } static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans, - struct btree_iter *alloc_iter) + struct btree_iter *alloc_iter, + struct bkey_buf *last_flushed) { struct bch_fs *c = trans->c; - struct btree_iter lru_iter; struct bch_alloc_v4 a_convert; const struct bch_alloc_v4 *a; - struct bkey_s_c alloc_k, lru_k; + struct bkey_s_c alloc_k; struct printbuf buf = PRINTBUF; int ret; @@ -1543,13 +1697,25 @@ static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans, if (ret) return ret; + struct bch_dev *ca = bch2_dev_tryget_noerror(c, alloc_k.k->p.inode); + if (!ca) + return 0; + a = bch2_alloc_to_v4(alloc_k, &a_convert); + u64 lru_idx = alloc_lru_idx_fragmentation(*a, ca); + if (lru_idx) { + ret = bch2_lru_check_set(trans, BCH_LRU_FRAGMENTATION_START, + lru_idx, alloc_k, last_flushed); + if (ret) + goto err; + } + if (a->data_type != BCH_DATA_cached) - return 0; + goto err; - if (fsck_err_on(!a->io_time[READ], c, - alloc_key_cached_but_read_time_zero, + if (fsck_err_on(!a->io_time[READ], + trans, alloc_key_cached_but_read_time_zero, "cached bucket with read_time 0\n" " %s", (printbuf_reset(&buf), @@ -1560,82 +1726,76 @@ static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans, if (ret) goto err; - a_mut->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now); + a_mut->v.io_time[READ] = bch2_current_io_time(c, READ); ret = bch2_trans_update(trans, alloc_iter, - &a_mut->k_i, BTREE_TRIGGER_NORUN); + &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); + ret = bch2_lru_check_set(trans, alloc_k.k->p.inode, a->io_time[READ], + alloc_k, last_flushed); 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))) { - ret = bch2_lru_set(trans, - alloc_k.k->p.inode, - bucket_to_u64(alloc_k.k->p), - a->io_time[READ]); - if (ret) - goto err; - } + goto err; err: fsck_err: - bch2_trans_iter_exit(trans, &lru_iter); + bch2_dev_put(ca); printbuf_exit(&buf); return ret; } int bch2_check_alloc_to_lru_refs(struct bch_fs *c) { + struct bkey_buf last_flushed; + + bch2_bkey_buf_init(&last_flushed); + bkey_init(&last_flushed.k->k); + int ret = bch2_trans_run(c, for_each_btree_key_commit(trans, iter, BTREE_ID_alloc, - POS_MIN, BTREE_ITER_PREFETCH, k, + POS_MIN, BTREE_ITER_prefetch, k, NULL, NULL, BCH_TRANS_COMMIT_no_enospc, - bch2_check_alloc_to_lru_ref(trans, &iter))); + bch2_check_alloc_to_lru_ref(trans, &iter, &last_flushed))); + + bch2_bkey_buf_exit(&last_flushed, c); bch_err_fn(c, ret); return ret; } -static int discard_in_flight_add(struct bch_fs *c, struct bpos bucket) +static int discard_in_flight_add(struct bch_dev *ca, u64 bucket, bool in_progress) { int ret; - mutex_lock(&c->discard_buckets_in_flight_lock); - darray_for_each(c->discard_buckets_in_flight, i) - if (bkey_eq(*i, bucket)) { - ret = -EEXIST; + mutex_lock(&ca->discard_buckets_in_flight_lock); + darray_for_each(ca->discard_buckets_in_flight, i) + if (i->bucket == bucket) { + ret = -BCH_ERR_EEXIST_discard_in_flight_add; goto out; } - ret = darray_push(&c->discard_buckets_in_flight, bucket); + ret = darray_push(&ca->discard_buckets_in_flight, ((struct discard_in_flight) { + .in_progress = in_progress, + .bucket = bucket, + })); out: - mutex_unlock(&c->discard_buckets_in_flight_lock); + mutex_unlock(&ca->discard_buckets_in_flight_lock); return ret; } -static void discard_in_flight_remove(struct bch_fs *c, struct bpos bucket) +static void discard_in_flight_remove(struct bch_dev *ca, u64 bucket) { - mutex_lock(&c->discard_buckets_in_flight_lock); - darray_for_each(c->discard_buckets_in_flight, i) - if (bkey_eq(*i, bucket)) { - darray_remove_item(&c->discard_buckets_in_flight, i); + mutex_lock(&ca->discard_buckets_in_flight_lock); + darray_for_each(ca->discard_buckets_in_flight, i) + if (i->bucket == bucket) { + BUG_ON(!i->in_progress); + darray_remove_item(&ca->discard_buckets_in_flight, i); goto found; } BUG(); found: - mutex_unlock(&c->discard_buckets_in_flight_lock); + mutex_unlock(&ca->discard_buckets_in_flight_lock); } struct discard_buckets_state { @@ -1643,67 +1803,40 @@ struct discard_buckets_state { u64 open; u64 need_journal_commit; u64 discarded; - struct bch_dev *ca; - u64 need_journal_commit_this_dev; }; -static void discard_buckets_next_dev(struct bch_fs *c, struct discard_buckets_state *s, struct bch_dev *ca) -{ - if (s->ca == ca) - return; - - 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); - - 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 bch_dev *ca, struct btree_iter *need_discard_iter, struct bpos *discard_pos_done, - struct discard_buckets_state *s) + struct discard_buckets_state *s, + bool fastpath) { struct bch_fs *c = trans->c; 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; bool discard_locked = 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)) { s->open++; goto out; } - if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal, - c->journal.flushed_seq_ondisk, - pos.inode, pos.offset)) { - s->need_journal_commit++; - s->need_journal_commit_this_dev++; + u64 seq_ready = bch2_bucket_journal_seq_ready(&c->buckets_waiting_for_journal, + pos.inode, pos.offset); + if (seq_ready > c->journal.flushed_seq_ondisk) { + if (seq_ready > c->journal.flushing_seq) + s->need_journal_commit++; goto out; } k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc, need_discard_iter->pos, - BTREE_ITER_CACHED); + BTREE_ITER_cached); ret = bkey_err(k); if (ret) goto out; @@ -1713,88 +1846,73 @@ static int bch2_discard_one_bucket(struct btree_trans *trans, if (ret) goto out; - if (a->v.dirty_sectors) { - if (bch2_trans_inconsistent_on(c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info, - trans, "attempting to discard bucket with dirty data\n%s", - (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) - ret = -EIO; - goto out; - } - if (a->v.data_type != BCH_DATA_need_discard) { - if (data_type_is_empty(a->v.data_type) && - BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) { - a->v.gen++; - SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false); - goto write; + if (need_discard_or_freespace_err(trans, k, true, true, true)) { + ret = bch2_btree_bit_mod_iter(trans, need_discard_iter, false); + if (ret) + goto out; + goto commit; } - if (bch2_trans_inconsistent_on(c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info, - trans, "bucket incorrectly set in need_discard btree\n" - "%s", - (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) - ret = -EIO; - goto out; - } - - if (a->v.journal_seq > c->journal.flushed_seq_ondisk) { - if (bch2_trans_inconsistent_on(c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info, - 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 (discard_in_flight_add(c, SPOS(iter.pos.inode, iter.pos.offset, true))) - goto out; + if (!fastpath) { + if (discard_in_flight_add(ca, iter.pos.offset, true)) + goto out; - discard_locked = true; + discard_locked = true; + } - 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 - */ - 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); + if (!bkey_eq(*discard_pos_done, iter.pos)) { + s->discarded++; *discard_pos_done = iter.pos; - ret = bch2_trans_relock_notrace(trans); - if (ret) - goto out; + if (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 + */ + 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); + ret = bch2_trans_relock_notrace(trans); + if (ret) + goto out; + } } 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, - BCH_WATERMARK_btree| - BCH_TRANS_COMMIT_no_enospc); + alloc_data_type_set(&a->v, a->v.data_type); + + ret = bch2_trans_update(trans, &iter, &a->k_i, 0); + if (ret) + goto out; +commit: + ret = bch2_trans_commit(trans, NULL, NULL, + BCH_WATERMARK_btree| + BCH_TRANS_COMMIT_no_enospc); if (ret) goto out; count_event(c, bucket_discard); - s->discarded++; out: +fsck_err: if (discard_locked) - discard_in_flight_remove(c, iter.pos); - s->seen++; + discard_in_flight_remove(ca, iter.pos.offset); + if (!ret) + s->seen++; bch2_trans_iter_exit(trans, &iter); - percpu_ref_put(&ca->io_ref); printbuf_exit(&buf); return ret; } 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 = container_of(work, struct bch_dev, discard_work); + struct bch_fs *c = ca->fs; struct discard_buckets_state s = {}; struct bpos discard_pos_done = POS_MAX; int ret; @@ -1805,111 +1923,136 @@ static void bch2_do_discards_work(struct work_struct *work) * successful commit: */ 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))); + for_each_btree_key_max(trans, iter, + BTREE_ID_need_discard, + POS(ca->dev_idx, 0), + POS(ca->dev_idx, U64_MAX), 0, k, + bch2_discard_one_bucket(trans, ca, &iter, &discard_pos_done, &s, false))); - discard_buckets_next_dev(c, &s, NULL); + if (s.need_journal_commit > dev_buckets_available(ca, BCH_WATERMARK_normal)) + bch2_journal_flush_async(&c->journal, NULL); trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded, bch2_err_str(ret)); + percpu_ref_put(&ca->io_ref); bch2_write_ref_put(c, BCH_WRITE_REF_discard); } -void bch2_do_discards(struct bch_fs *c) +void bch2_dev_do_discards(struct bch_dev *ca) { - 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); + struct bch_fs *c = ca->fs; + + if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_discard)) + return; + + if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE)) + goto put_write_ref; + + if (queue_work(c->write_ref_wq, &ca->discard_work)) + return; + + percpu_ref_put(&ca->io_ref); +put_write_ref: + bch2_write_ref_put(c, BCH_WRITE_REF_discard); } -static int bch2_clear_bucket_needs_discard(struct btree_trans *trans, struct bpos bucket) +void bch2_do_discards(struct bch_fs *c) { - struct btree_iter iter; - bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, bucket, BTREE_ITER_INTENT); - struct bkey_s_c k = bch2_btree_iter_peek_slot(&iter); - int ret = bkey_err(k); - if (ret) - goto err; + for_each_member_device(c, ca) + bch2_dev_do_discards(ca); +} - struct bkey_i_alloc_v4 *a = bch2_alloc_to_v4_mut(trans, k); - ret = PTR_ERR_OR_ZERO(a); +static int bch2_do_discards_fast_one(struct btree_trans *trans, + struct bch_dev *ca, + u64 bucket, + struct bpos *discard_pos_done, + struct discard_buckets_state *s) +{ + struct btree_iter need_discard_iter; + struct bkey_s_c discard_k = bch2_bkey_get_iter(trans, &need_discard_iter, + BTREE_ID_need_discard, POS(ca->dev_idx, bucket), 0); + int ret = bkey_err(discard_k); if (ret) - goto err; + return ret; - BUG_ON(a->v.dirty_sectors); - SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false); - a->v.data_type = alloc_data_type(a->v, a->v.data_type); + if (log_fsck_err_on(discard_k.k->type != KEY_TYPE_set, + trans, discarding_bucket_not_in_need_discard_btree, + "attempting to discard bucket %u:%llu not in need_discard btree", + ca->dev_idx, bucket)) + goto out; - ret = bch2_trans_update(trans, &iter, &a->k_i, 0); -err: - bch2_trans_iter_exit(trans, &iter); + ret = bch2_discard_one_bucket(trans, ca, &need_discard_iter, discard_pos_done, s, true); +out: +fsck_err: + bch2_trans_iter_exit(trans, &need_discard_iter); return ret; } static void bch2_do_discards_fast_work(struct work_struct *work) { - struct bch_fs *c = container_of(work, struct bch_fs, discard_fast_work); + struct bch_dev *ca = container_of(work, struct bch_dev, discard_fast_work); + struct bch_fs *c = ca->fs; + struct discard_buckets_state s = {}; + struct bpos discard_pos_done = POS_MAX; + struct btree_trans *trans = bch2_trans_get(c); + int ret = 0; while (1) { bool got_bucket = false; - struct bpos bucket; - struct bch_dev *ca; - - mutex_lock(&c->discard_buckets_in_flight_lock); - darray_for_each(c->discard_buckets_in_flight, i) { - if (i->snapshot) - continue; + u64 bucket; - ca = bch_dev_bkey_exists(c, i->inode); - - if (!percpu_ref_tryget(&ca->io_ref)) { - darray_remove_item(&c->discard_buckets_in_flight, i); + mutex_lock(&ca->discard_buckets_in_flight_lock); + darray_for_each(ca->discard_buckets_in_flight, i) { + if (i->in_progress) continue; - } got_bucket = true; - bucket = *i; - i->snapshot = true; + bucket = i->bucket; + i->in_progress = true; break; } - mutex_unlock(&c->discard_buckets_in_flight_lock); + mutex_unlock(&ca->discard_buckets_in_flight_lock); if (!got_bucket) break; - if (ca->mi.discard && !c->opts.nochanges) - blkdev_issue_discard(ca->disk_sb.bdev, - bucket.offset * ca->mi.bucket_size, - ca->mi.bucket_size, - GFP_KERNEL); - - int ret = bch2_trans_do(c, NULL, NULL, - BCH_WATERMARK_btree| - BCH_TRANS_COMMIT_no_enospc, - bch2_clear_bucket_needs_discard(trans, bucket)); + ret = lockrestart_do(trans, + bch2_do_discards_fast_one(trans, ca, bucket, &discard_pos_done, &s)); bch_err_fn(c, ret); - percpu_ref_put(&ca->io_ref); - discard_in_flight_remove(c, bucket); + discard_in_flight_remove(ca, bucket); if (ret) break; } + trace_discard_buckets_fast(c, s.seen, s.open, s.need_journal_commit, s.discarded, bch2_err_str(ret)); + + bch2_trans_put(trans); + percpu_ref_put(&ca->io_ref); bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast); } -static void bch2_discard_one_bucket_fast(struct bch_fs *c, struct bpos bucket) +static void bch2_discard_one_bucket_fast(struct bch_dev *ca, u64 bucket) { - struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode); + struct bch_fs *c = ca->fs; + + if (discard_in_flight_add(ca, bucket, false)) + return; + + if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_discard_fast)) + return; + + if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE)) + goto put_ref; + + if (queue_work(c->write_ref_wq, &ca->discard_fast_work)) + return; - if (!percpu_ref_is_dying(&ca->io_ref) && - !discard_in_flight_add(c, bucket) && - bch2_write_ref_tryget(c, BCH_WRITE_REF_discard_fast) && - !queue_work(c->write_ref_wq, &c->discard_fast_work)) - bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast); + percpu_ref_put(&ca->io_ref); +put_ref: + bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast); } static int invalidate_one_bucket(struct btree_trans *trans, @@ -1918,7 +2061,6 @@ static int invalidate_one_bucket(struct btree_trans *trans, s64 *nr_to_invalidate) { struct bch_fs *c = trans->c; - struct btree_iter alloc_iter = { NULL }; struct bkey_i_alloc_v4 *a = NULL; struct printbuf buf = PRINTBUF; struct bpos bucket = u64_to_bucket(lru_k.k->p.offset); @@ -1929,14 +2071,17 @@ static int invalidate_one_bucket(struct btree_trans *trans, return 1; if (!bch2_dev_bucket_exists(c, bucket)) { - prt_str(&buf, "lru entry points to invalid bucket"); - goto err; + if (fsck_err(trans, lru_entry_to_invalid_bucket, + "lru key points to nonexistent device:bucket %llu:%llu", + bucket.inode, bucket.offset)) + return bch2_btree_bit_mod_buffered(trans, BTREE_ID_lru, lru_iter->pos, false); + goto out; } if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset)) return 0; - a = bch2_trans_start_alloc_update(trans, &alloc_iter, bucket); + a = bch2_trans_start_alloc_update(trans, bucket, BTREE_TRIGGER_bucket_invalidate); ret = PTR_ERR_OR_ZERO(a); if (ret) goto out; @@ -1957,49 +2102,44 @@ static int invalidate_one_bucket(struct btree_trans *trans, a->v.gen++; a->v.data_type = 0; a->v.dirty_sectors = 0; + a->v.stripe_sectors = 0; a->v.cached_sectors = 0; - 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) ?: - bch2_trans_commit(trans, NULL, NULL, - BCH_WATERMARK_btree| - BCH_TRANS_COMMIT_no_enospc); + a->v.io_time[READ] = bch2_current_io_time(c, READ); + a->v.io_time[WRITE] = bch2_current_io_time(c, WRITE); + + ret = bch2_trans_commit(trans, NULL, NULL, + BCH_WATERMARK_btree| + BCH_TRANS_COMMIT_no_enospc); if (ret) goto out; trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors); --*nr_to_invalidate; out: - bch2_trans_iter_exit(trans, &alloc_iter); +fsck_err: 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; +} + +static struct bkey_s_c next_lru_key(struct btree_trans *trans, struct btree_iter *iter, + struct bch_dev *ca, bool *wrapped) +{ + struct bkey_s_c k; +again: + k = bch2_btree_iter_peek_max(iter, lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX)); + if (!k.k && !*wrapped) { + bch2_btree_iter_set_pos(iter, lru_pos(ca->dev_idx, 0, 0)); + *wrapped = true; + goto again; } - goto out; + return k; } 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 = container_of(work, struct bch_dev, invalidate_work); + struct bch_fs *c = ca->fs; struct btree_trans *trans = bch2_trans_get(c); int ret = 0; @@ -2007,31 +2147,64 @@ static void bch2_do_invalidates_work(struct work_struct *work) if (ret) goto err; - for_each_member_device(c, ca) { - s64 nr_to_invalidate = - should_invalidate_buckets(ca, bch2_dev_usage_read(ca)); + s64 nr_to_invalidate = + should_invalidate_buckets(ca, bch2_dev_usage_read(ca)); + struct btree_iter iter; + bool wrapped = false; + + bch2_trans_iter_init(trans, &iter, BTREE_ID_lru, + lru_pos(ca->dev_idx, 0, + ((bch2_current_io_time(c, READ) + U32_MAX) & + LRU_TIME_MAX)), 0); - 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)); + while (true) { + bch2_trans_begin(trans); - if (ret < 0) { - percpu_ref_put(&ca->ref); + struct bkey_s_c k = next_lru_key(trans, &iter, ca, &wrapped); + ret = bkey_err(k); + if (ret) + goto restart_err; + if (!k.k) break; - } + + ret = invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate); +restart_err: + if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) + continue; + if (ret) + break; + + bch2_btree_iter_advance(&iter); } + bch2_trans_iter_exit(trans, &iter); err: bch2_trans_put(trans); + percpu_ref_put(&ca->io_ref); + bch2_write_ref_put(c, BCH_WRITE_REF_invalidate); +} + +void bch2_dev_do_invalidates(struct bch_dev *ca) +{ + struct bch_fs *c = ca->fs; + + if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate)) + return; + + if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE)) + goto put_ref; + + if (queue_work(c->write_ref_wq, &ca->invalidate_work)) + return; + + percpu_ref_put(&ca->io_ref); +put_ref: bch2_write_ref_put(c, BCH_WRITE_REF_invalidate); } void bch2_do_invalidates(struct bch_fs *c) { - 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); + for_each_member_device(c, ca) + bch2_dev_do_invalidates(ca); } int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca, @@ -2051,13 +2224,13 @@ int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca, 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); + 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) { - if (last_updated + HZ * 10 < jiffies) { + if (time_after(jiffies, last_updated + HZ * 10)) { bch_info(ca, "%s: currently at %llu/%llu", __func__, iter.pos.offset, ca->mi.nbuckets); last_updated = jiffies; @@ -2083,7 +2256,7 @@ 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) ?: + ret = bch2_bucket_do_index(trans, ca, k, a, true) ?: bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc); if (ret) @@ -2155,7 +2328,7 @@ int bch2_fs_freespace_init(struct bch_fs *c) ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets); if (ret) { - percpu_ref_put(&ca->ref); + bch2_dev_put(ca); bch_err_fn(c, ret); return ret; } @@ -2171,23 +2344,51 @@ int bch2_fs_freespace_init(struct bch_fs *c) return 0; } +/* device removal */ + +int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca) +{ + struct bpos start = POS(ca->dev_idx, 0); + struct bpos end = POS(ca->dev_idx, U64_MAX); + int ret; + + /* + * We clear the LRU and need_discard btrees first so that we don't race + * with bch2_do_invalidates() and bch2_do_discards() + */ + ret = bch2_dev_remove_stripes(c, ca->dev_idx) ?: + bch2_btree_delete_range(c, BTREE_ID_lru, start, end, + BTREE_TRIGGER_norun, NULL) ?: + bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end, + BTREE_TRIGGER_norun, NULL) ?: + bch2_btree_delete_range(c, BTREE_ID_freespace, start, end, + BTREE_TRIGGER_norun, NULL) ?: + bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end, + BTREE_TRIGGER_norun, NULL) ?: + bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end, + BTREE_TRIGGER_norun, NULL) ?: + bch2_btree_delete_range(c, BTREE_ID_alloc, start, end, + BTREE_TRIGGER_norun, NULL) ?: + bch2_dev_usage_remove(c, ca->dev_idx); + bch_err_msg(ca, ret, "removing dev alloc info"); + return ret; +} + /* Bucket IO clocks: */ -int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev, - size_t bucket_nr, int rw) +static int __bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev, + size_t bucket_nr, int rw) { struct bch_fs *c = trans->c; - struct btree_iter iter; - struct bkey_i_alloc_v4 *a; - u64 now; - int ret = 0; - a = bch2_trans_start_alloc_update(trans, &iter, POS(dev, bucket_nr)); - ret = PTR_ERR_OR_ZERO(a); + struct btree_iter iter; + struct bkey_i_alloc_v4 *a = + bch2_trans_start_alloc_update_noupdate(trans, &iter, POS(dev, bucket_nr)); + int ret = PTR_ERR_OR_ZERO(a); if (ret) return ret; - now = atomic64_read(&c->io_clock[rw].now); + u64 now = bch2_current_io_time(c, rw); if (a->v.io_time[rw] == now) goto out; @@ -2200,6 +2401,15 @@ out: return ret; } +int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev, + size_t bucket_nr, int rw) +{ + if (bch2_trans_relock(trans)) + bch2_trans_begin(trans); + + return nested_lockrestart_do(trans, __bch2_bucket_io_time_reset(trans, dev, bucket_nr, rw)); +} + /* Startup/shutdown (ro/rw): */ void bch2_recalc_capacity(struct bch_fs *c) @@ -2264,6 +2474,7 @@ void bch2_recalc_capacity(struct bch_fs *c) reserved_sectors = min(reserved_sectors, capacity); + c->reserved = reserved_sectors; c->capacity = capacity - reserved_sectors; c->bucket_size_max = bucket_size_max; @@ -2302,13 +2513,15 @@ static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca) /* device goes ro: */ void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca) { - unsigned i; + lockdep_assert_held(&c->state_lock); /* First, remove device from allocation groups: */ - for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++) + for (unsigned i = 0; i < ARRAY_SIZE(c->rw_devs); i++) clear_bit(ca->dev_idx, c->rw_devs[i].d); + c->rw_devs_change_count++; + /* * Capacity is calculated based off of devices in allocation groups: */ @@ -2337,23 +2550,29 @@ void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca) /* device goes rw: */ void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca) { - unsigned i; + lockdep_assert_held(&c->state_lock); - for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++) + for (unsigned i = 0; i < ARRAY_SIZE(c->rw_devs); i++) if (ca->mi.data_allowed & (1 << i)) set_bit(ca->dev_idx, c->rw_devs[i].d); + + c->rw_devs_change_count++; +} + +void bch2_dev_allocator_background_exit(struct bch_dev *ca) +{ + darray_exit(&ca->discard_buckets_in_flight); } -void bch2_fs_allocator_background_exit(struct bch_fs *c) +void bch2_dev_allocator_background_init(struct bch_dev *ca) { - darray_exit(&c->discard_buckets_in_flight); + mutex_init(&ca->discard_buckets_in_flight_lock); + INIT_WORK(&ca->discard_work, bch2_do_discards_work); + INIT_WORK(&ca->discard_fast_work, bch2_do_discards_fast_work); + INIT_WORK(&ca->invalidate_work, bch2_do_invalidates_work); } void bch2_fs_allocator_background_init(struct bch_fs *c) { spin_lock_init(&c->freelist_lock); - mutex_init(&c->discard_buckets_in_flight_lock); - INIT_WORK(&c->discard_work, bch2_do_discards_work); - INIT_WORK(&c->discard_fast_work, bch2_do_discards_fast_work); - INIT_WORK(&c->invalidate_work, bch2_do_invalidates_work); } |