X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fbkey_methods.c;h=761f5e33b1e69e94ca0aaaa41a9825e496b5840f;hb=f27b135285d032d9570a1accb00412e111b38878;hp=55ef4032b37c159ee9b78fc2b2288a6a7338608a;hpb=34c9be19b376a007041555a7c9a47dfef3d0b1e6;p=bcachefs-tools-debian diff --git a/libbcachefs/bkey_methods.c b/libbcachefs/bkey_methods.c index 55ef403..761f5e3 100644 --- a/libbcachefs/bkey_methods.c +++ b/libbcachefs/bkey_methods.c @@ -1,7 +1,9 @@ // SPDX-License-Identifier: GPL-2.0 #include "bcachefs.h" +#include "backpointers.h" #include "bkey_methods.h" +#include "btree_cache.h" #include "btree_types.h" #include "alloc_background.h" #include "dirent.h" @@ -9,8 +11,12 @@ #include "error.h" #include "extents.h" #include "inode.h" +#include "io_misc.h" +#include "lru.h" #include "quota.h" #include "reflink.h" +#include "snapshot.h" +#include "subvolume.h" #include "xattr.h" const char * const bch2_bkey_types[] = { @@ -20,175 +26,262 @@ const char * const bch2_bkey_types[] = { NULL }; -static const char *deleted_key_invalid(const struct bch_fs *c, - struct bkey_s_c k) +static int deleted_key_invalid(struct bch_fs *c, struct bkey_s_c k, + enum bkey_invalid_flags flags, struct printbuf *err) { - return NULL; + return 0; } -#define bch2_bkey_ops_deleted (struct bkey_ops) { \ +#define bch2_bkey_ops_deleted ((struct bkey_ops) { \ .key_invalid = deleted_key_invalid, \ -} +}) -#define bch2_bkey_ops_discard (struct bkey_ops) { \ +#define bch2_bkey_ops_whiteout ((struct bkey_ops) { \ .key_invalid = deleted_key_invalid, \ -} +}) -static const char *empty_val_key_invalid(const struct bch_fs *c, struct bkey_s_c k) +static int empty_val_key_invalid(struct bch_fs *c, struct bkey_s_c k, + enum bkey_invalid_flags flags, struct printbuf *err) { - if (bkey_val_bytes(k.k)) - return "value size should be zero"; + int ret = 0; - return NULL; + bkey_fsck_err_on(bkey_val_bytes(k.k), c, err, + bkey_val_size_nonzero, + "incorrect value size (%zu != 0)", + bkey_val_bytes(k.k)); +fsck_err: + return ret; } -#define bch2_bkey_ops_error (struct bkey_ops) { \ +#define bch2_bkey_ops_error ((struct bkey_ops) { \ .key_invalid = empty_val_key_invalid, \ -} +}) -static const char *key_type_cookie_invalid(const struct bch_fs *c, - struct bkey_s_c k) +static int key_type_cookie_invalid(struct bch_fs *c, struct bkey_s_c k, + enum bkey_invalid_flags flags, struct printbuf *err) { - if (bkey_val_bytes(k.k) != sizeof(struct bch_cookie)) - return "incorrect value size"; - - return NULL; + return 0; } -#define bch2_bkey_ops_cookie (struct bkey_ops) { \ - .key_invalid = key_type_cookie_invalid, \ -} +#define bch2_bkey_ops_cookie ((struct bkey_ops) { \ + .key_invalid = key_type_cookie_invalid, \ + .min_val_size = 8, \ +}) -#define bch2_bkey_ops_whiteout (struct bkey_ops) { \ +#define bch2_bkey_ops_hash_whiteout ((struct bkey_ops) {\ .key_invalid = empty_val_key_invalid, \ -} +}) -static const char *key_type_inline_data_invalid(const struct bch_fs *c, - struct bkey_s_c k) +static int key_type_inline_data_invalid(struct bch_fs *c, struct bkey_s_c k, + enum bkey_invalid_flags flags, struct printbuf *err) { - return NULL; + return 0; } static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k) { - pr_buf(out, "(%zu bytes)", bkey_val_bytes(k.k)); + struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k); + unsigned datalen = bkey_inline_data_bytes(k.k); + + prt_printf(out, "datalen %u: %*phN", + datalen, min(datalen, 32U), d.v->data); } -#define bch2_bkey_ops_inline_data (struct bkey_ops) { \ +#define bch2_bkey_ops_inline_data ((struct bkey_ops) { \ .key_invalid = key_type_inline_data_invalid, \ .val_to_text = key_type_inline_data_to_text, \ +}) + +static bool key_type_set_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r) +{ + bch2_key_resize(l.k, l.k->size + r.k->size); + return true; } -static const struct bkey_ops bch2_bkey_ops[] = { +#define bch2_bkey_ops_set ((struct bkey_ops) { \ + .key_invalid = empty_val_key_invalid, \ + .key_merge = key_type_set_merge, \ +}) + +const struct bkey_ops bch2_bkey_ops[] = { #define x(name, nr) [KEY_TYPE_##name] = bch2_bkey_ops_##name, BCH_BKEY_TYPES() #undef x }; -const char *bch2_bkey_val_invalid(struct bch_fs *c, struct bkey_s_c k) -{ - if (k.k->type >= KEY_TYPE_MAX) - return "invalid type"; - - return bch2_bkey_ops[k.k->type].key_invalid(c, k); -} +const struct bkey_ops bch2_bkey_null_ops = { +}; -const char *__bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, - enum btree_node_type type) +int bch2_bkey_val_invalid(struct bch_fs *c, struct bkey_s_c k, + enum bkey_invalid_flags flags, + struct printbuf *err) { - if (k.k->u64s < BKEY_U64s) - return "u64s too small"; - - if (type == BKEY_TYPE_BTREE && - bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX) - return "value too big"; - - if (btree_node_type_is_extents(type)) { - if ((k.k->size == 0) != bkey_deleted(k.k)) - return "bad size field"; - - if (k.k->size > k.k->p.offset) - return "size greater than offset"; - } else { - if (k.k->size) - return "nonzero size field"; - } + const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); + int ret = 0; - if (k.k->p.snapshot) - return "nonzero snapshot"; + bkey_fsck_err_on(bkey_val_bytes(k.k) < ops->min_val_size, c, err, + bkey_val_size_too_small, + "bad val size (%zu < %u)", + bkey_val_bytes(k.k), ops->min_val_size); - if (type != BKEY_TYPE_BTREE && - !bkey_cmp(k.k->p, POS_MAX)) - return "POS_MAX key"; + if (!ops->key_invalid) + return 0; - return NULL; + ret = ops->key_invalid(c, k, flags, err); +fsck_err: + return ret; } -const char *bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, - enum btree_node_type type) +static u64 bch2_key_types_allowed[] = { + [BKEY_TYPE_btree] = + BIT_ULL(KEY_TYPE_deleted)| + BIT_ULL(KEY_TYPE_btree_ptr)| + BIT_ULL(KEY_TYPE_btree_ptr_v2), +#define x(name, nr, flags, keys) [BKEY_TYPE_##name] = BIT_ULL(KEY_TYPE_deleted)|keys, + BCH_BTREE_IDS() +#undef x +}; + +const char *bch2_btree_node_type_str(enum btree_node_type type) { - return __bch2_bkey_invalid(c, k, type) ?: - bch2_bkey_val_invalid(c, k); + return type == BKEY_TYPE_btree ? "internal btree node" : bch2_btree_id_str(type - 1); } -const char *bch2_bkey_in_btree_node(struct btree *b, struct bkey_s_c k) +int __bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, + enum btree_node_type type, + enum bkey_invalid_flags flags, + struct printbuf *err) { - if (bkey_cmp(k.k->p, b->data->min_key) < 0) - return "key before start of btree node"; + int ret = 0; + + bkey_fsck_err_on(k.k->u64s < BKEY_U64s, c, err, + bkey_u64s_too_small, + "u64s too small (%u < %zu)", k.k->u64s, BKEY_U64s); + + if (type >= BKEY_TYPE_NR) + return 0; + + bkey_fsck_err_on((flags & BKEY_INVALID_COMMIT) && + !(bch2_key_types_allowed[type] & BIT_ULL(k.k->type)), c, err, + bkey_invalid_type_for_btree, + "invalid key type for btree %s (%s)", + bch2_btree_node_type_str(type), bch2_bkey_types[k.k->type]); + + if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) { + bkey_fsck_err_on(k.k->size == 0, c, err, + bkey_extent_size_zero, + "size == 0"); + + bkey_fsck_err_on(k.k->size > k.k->p.offset, c, err, + bkey_extent_size_greater_than_offset, + "size greater than offset (%u > %llu)", + k.k->size, k.k->p.offset); + } else { + bkey_fsck_err_on(k.k->size, c, err, + bkey_size_nonzero, + "size != 0"); + } - if (bkey_cmp(k.k->p, b->data->max_key) > 0) - return "key past end of btree node"; + if (type != BKEY_TYPE_btree) { + enum btree_id btree = type - 1; + + if (btree_type_has_snapshots(btree)) { + bkey_fsck_err_on(!k.k->p.snapshot, c, err, + bkey_snapshot_zero, + "snapshot == 0"); + } else if (!btree_type_has_snapshot_field(btree)) { + bkey_fsck_err_on(k.k->p.snapshot, c, err, + bkey_snapshot_nonzero, + "nonzero snapshot"); + } else { + /* + * btree uses snapshot field but it's not required to be + * nonzero + */ + } - return NULL; + bkey_fsck_err_on(bkey_eq(k.k->p, POS_MAX), c, err, + bkey_at_pos_max, + "key at POS_MAX"); + } +fsck_err: + return ret; } -void bch2_bkey_debugcheck(struct bch_fs *c, struct btree *b, struct bkey_s_c k) +int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, + enum btree_node_type type, + enum bkey_invalid_flags flags, + struct printbuf *err) { - const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type]; - const char *invalid; + return __bch2_bkey_invalid(c, k, type, flags, err) ?: + bch2_bkey_val_invalid(c, k, flags, err); +} - BUG_ON(!k.k->u64s); +int bch2_bkey_in_btree_node(struct bch_fs *c, struct btree *b, + struct bkey_s_c k, struct printbuf *err) +{ + int ret = 0; - invalid = bch2_bkey_invalid(c, k, btree_node_type(b)) ?: - bch2_bkey_in_btree_node(b, k); - if (invalid) { - char buf[160]; + bkey_fsck_err_on(bpos_lt(k.k->p, b->data->min_key), c, err, + bkey_before_start_of_btree_node, + "key before start of btree node"); - bch2_bkey_val_to_text(&PBUF(buf), c, k); - bch2_fs_inconsistent(c, "invalid bkey %s: %s", buf, invalid); - return; - } - - if (ops->key_debugcheck) - ops->key_debugcheck(c, k); + bkey_fsck_err_on(bpos_gt(k.k->p, b->data->max_key), c, err, + bkey_after_end_of_btree_node, + "key past end of btree node"); +fsck_err: + return ret; } void bch2_bpos_to_text(struct printbuf *out, struct bpos pos) { - if (!bkey_cmp(pos, POS_MIN)) - pr_buf(out, "POS_MIN"); - else if (!bkey_cmp(pos, POS_MAX)) - pr_buf(out, "POS_MAX"); - else - pr_buf(out, "%llu:%llu", pos.inode, pos.offset); + if (bpos_eq(pos, POS_MIN)) + prt_printf(out, "POS_MIN"); + else if (bpos_eq(pos, POS_MAX)) + prt_printf(out, "POS_MAX"); + else if (bpos_eq(pos, SPOS_MAX)) + prt_printf(out, "SPOS_MAX"); + else { + if (pos.inode == U64_MAX) + prt_printf(out, "U64_MAX"); + else + prt_printf(out, "%llu", pos.inode); + prt_printf(out, ":"); + if (pos.offset == U64_MAX) + prt_printf(out, "U64_MAX"); + else + prt_printf(out, "%llu", pos.offset); + prt_printf(out, ":"); + if (pos.snapshot == U32_MAX) + prt_printf(out, "U32_MAX"); + else + prt_printf(out, "%u", pos.snapshot); + } } void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k) { - pr_buf(out, "u64s %u type %s ", k->u64s, - bch2_bkey_types[k->type]); + if (k) { + prt_printf(out, "u64s %u type ", k->u64s); - bch2_bpos_to_text(out, k->p); + if (k->type < KEY_TYPE_MAX) + prt_printf(out, "%s ", bch2_bkey_types[k->type]); + else + prt_printf(out, "%u ", k->type); - pr_buf(out, " snap %u len %u ver %llu", - k->p.snapshot, k->size, k->version.lo); + bch2_bpos_to_text(out, k->p); + + prt_printf(out, " len %u ver %llu", k->size, k->version.lo); + } else { + prt_printf(out, "(null)"); + } } void bch2_val_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k) { - const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type]; + const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); if (likely(ops->val_to_text)) ops->val_to_text(out, c, k); @@ -198,13 +291,16 @@ void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k) { bch2_bkey_to_text(out, k.k); - pr_buf(out, ": "); - bch2_val_to_text(out, c, k); + + if (bkey_val_bytes(k.k)) { + prt_printf(out, ": "); + bch2_val_to_text(out, c, k); + } } void bch2_bkey_swab_val(struct bkey_s k) { - const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type]; + const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); if (ops->swab) ops->swab(k); @@ -212,31 +308,22 @@ void bch2_bkey_swab_val(struct bkey_s k) bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k) { - const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type]; + const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type); return ops->key_normalize ? ops->key_normalize(c, k) : false; } -enum merge_result bch2_bkey_merge(struct bch_fs *c, - struct bkey_s l, struct bkey_s r) +bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r) { - const struct bkey_ops *ops = &bch2_bkey_ops[l.k->type]; - enum merge_result ret; - - if (key_merging_disabled(c) || - !ops->key_merge || - l.k->type != r.k->type || - bversion_cmp(l.k->version, r.k->version) || - bkey_cmp(l.k->p, bkey_start_pos(r.k))) - return BCH_MERGE_NOMERGE; + const struct bkey_ops *ops = bch2_bkey_type_ops(l.k->type); - ret = ops->key_merge(c, l, r); - - if (ret != BCH_MERGE_NOMERGE) - l.k->needs_whiteout |= r.k->needs_whiteout; - return ret; + return ops->key_merge && + bch2_bkey_maybe_mergable(l.k, r.k) && + (u64) l.k->size + r.k->size <= KEY_SIZE_MAX && + !bch2_key_merging_disabled && + ops->key_merge(c, l, r); } static const struct old_bkey_type { @@ -244,18 +331,18 @@ static const struct old_bkey_type { u8 old; u8 new; } bkey_renumber_table[] = { - {BKEY_TYPE_BTREE, 128, KEY_TYPE_btree_ptr }, - {BKEY_TYPE_EXTENTS, 128, KEY_TYPE_extent }, - {BKEY_TYPE_EXTENTS, 129, KEY_TYPE_extent }, - {BKEY_TYPE_EXTENTS, 130, KEY_TYPE_reservation }, - {BKEY_TYPE_INODES, 128, KEY_TYPE_inode }, - {BKEY_TYPE_INODES, 130, KEY_TYPE_inode_generation }, - {BKEY_TYPE_DIRENTS, 128, KEY_TYPE_dirent }, - {BKEY_TYPE_DIRENTS, 129, KEY_TYPE_whiteout }, - {BKEY_TYPE_XATTRS, 128, KEY_TYPE_xattr }, - {BKEY_TYPE_XATTRS, 129, KEY_TYPE_whiteout }, - {BKEY_TYPE_ALLOC, 128, KEY_TYPE_alloc }, - {BKEY_TYPE_QUOTAS, 128, KEY_TYPE_quota }, + {BKEY_TYPE_btree, 128, KEY_TYPE_btree_ptr }, + {BKEY_TYPE_extents, 128, KEY_TYPE_extent }, + {BKEY_TYPE_extents, 129, KEY_TYPE_extent }, + {BKEY_TYPE_extents, 130, KEY_TYPE_reservation }, + {BKEY_TYPE_inodes, 128, KEY_TYPE_inode }, + {BKEY_TYPE_inodes, 130, KEY_TYPE_inode_generation }, + {BKEY_TYPE_dirents, 128, KEY_TYPE_dirent }, + {BKEY_TYPE_dirents, 129, KEY_TYPE_hash_whiteout }, + {BKEY_TYPE_xattrs, 128, KEY_TYPE_xattr }, + {BKEY_TYPE_xattrs, 129, KEY_TYPE_hash_whiteout }, + {BKEY_TYPE_alloc, 128, KEY_TYPE_alloc }, + {BKEY_TYPE_quotas, 128, KEY_TYPE_quota }, }; void bch2_bkey_renumber(enum btree_node_type btree_node_type, @@ -282,15 +369,15 @@ void __bch2_bkey_compat(unsigned level, enum btree_id btree_id, { const struct bkey_ops *ops; struct bkey uk; - struct bkey_s u; + unsigned nr_compat = 5; int i; /* * Do these operations in reverse order in the write path: */ - for (i = 0; i < 4; i++) - switch (!write ? i : 3 - i) { + for (i = 0; i < nr_compat; i++) + switch (!write ? i : nr_compat - 1 - i) { case 0: if (big_endian != CPU_BIG_ENDIAN) bch2_bkey_swab_key(f, k); @@ -301,9 +388,10 @@ void __bch2_bkey_compat(unsigned level, enum btree_id btree_id, break; case 2: if (version < bcachefs_metadata_version_inode_btree_change && - btree_id == BTREE_ID_INODES) { + btree_id == BTREE_ID_inodes) { if (!bkey_packed(k)) { struct bkey_i *u = packed_to_bkey(k); + swap(u->k.p.inode, u->k.p.offset); } else if (f->bits_per_field[BKEY_FIELD_INODE] && f->bits_per_field[BKEY_FIELD_OFFSET]) { @@ -324,6 +412,30 @@ void __bch2_bkey_compat(unsigned level, enum btree_id btree_id, } break; case 3: + if (version < bcachefs_metadata_version_snapshot && + (level || btree_type_has_snapshots(btree_id))) { + struct bkey_i *u = packed_to_bkey(k); + + if (u) { + u->k.p.snapshot = write + ? 0 : U32_MAX; + } else { + u64 min_packed = le64_to_cpu(f->field_offset[BKEY_FIELD_SNAPSHOT]); + u64 max_packed = min_packed + + ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]); + + uk = __bch2_bkey_unpack_key(f, k); + uk.p.snapshot = write + ? min_packed : min_t(u64, U32_MAX, max_packed); + + BUG_ON(!bch2_bkey_pack_key(k, &uk, f)); + } + } + + break; + case 4: { + struct bkey_s u; + if (!bkey_packed(k)) { u = bkey_i_to_s(packed_to_bkey(k)); } else { @@ -335,11 +447,12 @@ void __bch2_bkey_compat(unsigned level, enum btree_id btree_id, if (big_endian != CPU_BIG_ENDIAN) bch2_bkey_swab_val(u); - ops = &bch2_bkey_ops[k->type]; + ops = bch2_bkey_type_ops(k->type); if (ops->compat) ops->compat(btree_id, version, big_endian, write, u); break; + } default: BUG(); }