bcachefs: bch2_str_hash_check_key() now checks inode hash info
Versions of the same inode in different snapshots must have the same hash info; this is critical for lookups to work correctly. We're going to be running the str_hash checks online, at readdir or xattr list time, so we now need str_hash_check_key() to check for inode hash seed mismatches, since it won't be run right after check_inodes(). Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>pull/1131/head
parent
644457ed83
commit
6ea607ca61
|
|
@ -1110,7 +1110,7 @@ static int check_inode(struct btree_trans *trans,
|
|||
if (fsck_err_on(u.bi_hash_seed != snapshot_root->bi_hash_seed ||
|
||||
INODE_STR_HASH(&u) != INODE_STR_HASH(snapshot_root),
|
||||
trans, inode_snapshot_mismatch,
|
||||
"inodes in different snapshots don't match")) {
|
||||
"inode hash info in different snapshots don't match")) {
|
||||
u.bi_hash_seed = snapshot_root->bi_hash_seed;
|
||||
SET_INODE_STR_HASH(&u, INODE_STR_HASH(snapshot_root));
|
||||
do_update = true;
|
||||
|
|
@ -2303,7 +2303,7 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
|
|||
*hash_info = bch2_hash_info_init(c, &i->inode);
|
||||
dir->first_this_inode = false;
|
||||
|
||||
ret = bch2_str_hash_check_key(trans, s, bch2_dirent_hash_desc, hash_info, iter, k);
|
||||
ret = bch2_str_hash_check_key(trans, s, &bch2_dirent_hash_desc, hash_info, iter, k);
|
||||
if (ret < 0)
|
||||
goto err;
|
||||
if (ret) {
|
||||
|
|
@ -2417,7 +2417,7 @@ static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
|
|||
*hash_info = bch2_hash_info_init(c, &i->inode);
|
||||
inode->first_this_inode = false;
|
||||
|
||||
ret = bch2_str_hash_check_key(trans, NULL, bch2_xattr_hash_desc, hash_info, iter, k);
|
||||
ret = bch2_str_hash_check_key(trans, NULL, &bch2_xattr_hash_desc, hash_info, iter, k);
|
||||
bch_err_fn(c, ret);
|
||||
return ret;
|
||||
}
|
||||
|
|
|
|||
|
|
@ -101,38 +101,108 @@ static int hash_pick_winner(struct btree_trans *trans,
|
|||
}
|
||||
}
|
||||
|
||||
int bch2_str_hash_check_key(struct btree_trans *trans,
|
||||
struct snapshots_seen *s,
|
||||
const struct bch_hash_desc desc,
|
||||
struct bch_hash_info *hash_info,
|
||||
struct btree_iter *k_iter, struct bkey_s_c hash_k)
|
||||
static int repair_inode_hash_info(struct btree_trans *trans,
|
||||
struct bch_inode_unpacked *snapshot_root)
|
||||
{
|
||||
struct btree_iter iter;
|
||||
struct bkey_s_c k;
|
||||
int ret = 0;
|
||||
|
||||
for_each_btree_key_reverse_norestart(trans, iter, BTREE_ID_inodes,
|
||||
SPOS(0, snapshot_root->bi_inum, snapshot_root->bi_snapshot - 1),
|
||||
BTREE_ITER_all_snapshots, k, ret) {
|
||||
if (k.k->p.offset != snapshot_root->bi_inum)
|
||||
break;
|
||||
if (!bkey_is_inode(k.k))
|
||||
continue;
|
||||
|
||||
struct bch_inode_unpacked inode;
|
||||
ret = bch2_inode_unpack(k, &inode);
|
||||
if (ret)
|
||||
break;
|
||||
|
||||
if (fsck_err_on(inode.bi_hash_seed != snapshot_root->bi_hash_seed ||
|
||||
INODE_STR_HASH(&inode) != INODE_STR_HASH(snapshot_root),
|
||||
trans, inode_snapshot_mismatch,
|
||||
"inode hash info in different snapshots don't match")) {
|
||||
inode.bi_hash_seed = snapshot_root->bi_hash_seed;
|
||||
SET_INODE_STR_HASH(&inode, INODE_STR_HASH(snapshot_root));
|
||||
ret = __bch2_fsck_write_inode(trans, &inode) ?:
|
||||
bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc) ?:
|
||||
-BCH_ERR_transaction_restart_nested;
|
||||
break;
|
||||
}
|
||||
}
|
||||
fsck_err:
|
||||
bch2_trans_iter_exit(trans, &iter);
|
||||
return ret;
|
||||
}
|
||||
|
||||
/*
|
||||
* All versions of the same inode in different snapshots must have the same hash
|
||||
* seed/type: verify that the hash info we're using matches the root
|
||||
*/
|
||||
static int check_inode_hash_info_matches_root(struct btree_trans *trans, u64 inum,
|
||||
struct bch_hash_info *hash_info)
|
||||
{
|
||||
struct bch_fs *c = trans->c;
|
||||
struct btree_iter iter;
|
||||
struct bkey_s_c k;
|
||||
int ret = 0;
|
||||
|
||||
for_each_btree_key_reverse_norestart(trans, iter, BTREE_ID_inodes, SPOS(0, inum, U32_MAX),
|
||||
BTREE_ITER_all_snapshots, k, ret) {
|
||||
if (k.k->p.offset != inum)
|
||||
break;
|
||||
if (bkey_is_inode(k.k))
|
||||
goto found;
|
||||
}
|
||||
bch_err(c, "%s(): inum %llu not found", __func__, inum);
|
||||
ret = -BCH_ERR_fsck_repair_unimplemented;
|
||||
goto err;
|
||||
found:
|
||||
struct bch_inode_unpacked inode;
|
||||
ret = bch2_inode_unpack(k, &inode);
|
||||
if (ret)
|
||||
goto err;
|
||||
|
||||
struct bch_hash_info hash2 = bch2_hash_info_init(c, &inode);
|
||||
if (memcmp(hash_info, &hash2, sizeof(hash2))) {
|
||||
ret = repair_inode_hash_info(trans, &inode);
|
||||
if (!ret) {
|
||||
bch_err(c, "inode hash info mismatch with root, but mismatch not found");
|
||||
ret = -BCH_ERR_fsck_repair_unimplemented;
|
||||
}
|
||||
}
|
||||
err:
|
||||
bch2_trans_iter_exit(trans, &iter);
|
||||
return ret;
|
||||
}
|
||||
|
||||
int __bch2_str_hash_check_key(struct btree_trans *trans,
|
||||
struct snapshots_seen *s,
|
||||
const struct bch_hash_desc *desc,
|
||||
struct bch_hash_info *hash_info,
|
||||
struct btree_iter *k_iter, struct bkey_s_c hash_k)
|
||||
{
|
||||
struct bch_fs *c = trans->c;
|
||||
struct btree_iter iter = { NULL };
|
||||
struct printbuf buf = PRINTBUF;
|
||||
struct bkey_s_c k;
|
||||
u64 hash;
|
||||
int ret = 0;
|
||||
|
||||
if (hash_k.k->type != desc.key_type)
|
||||
return 0;
|
||||
|
||||
hash = desc.hash_bkey(hash_info, hash_k);
|
||||
|
||||
if (likely(hash == hash_k.k->p.offset))
|
||||
return 0;
|
||||
|
||||
u64 hash = desc->hash_bkey(hash_info, hash_k);
|
||||
if (hash_k.k->p.offset < hash)
|
||||
goto bad_hash;
|
||||
|
||||
for_each_btree_key_norestart(trans, iter, desc.btree_id,
|
||||
for_each_btree_key_norestart(trans, iter, desc->btree_id,
|
||||
SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
|
||||
BTREE_ITER_slots, k, ret) {
|
||||
if (bkey_eq(k.k->p, hash_k.k->p))
|
||||
break;
|
||||
|
||||
if (k.k->type == desc.key_type &&
|
||||
!desc.cmp_bkey(k, hash_k))
|
||||
if (k.k->type == desc->key_type &&
|
||||
!desc->cmp_bkey(k, hash_k))
|
||||
goto duplicate_entries;
|
||||
|
||||
if (bkey_deleted(k.k)) {
|
||||
|
|
@ -145,16 +215,23 @@ out:
|
|||
printbuf_exit(&buf);
|
||||
return ret;
|
||||
bad_hash:
|
||||
/*
|
||||
* Before doing any repair, check hash_info itself:
|
||||
*/
|
||||
ret = check_inode_hash_info_matches_root(trans, hash_k.k->p.inode, hash_info);
|
||||
if (ret)
|
||||
goto out;
|
||||
|
||||
if (fsck_err(trans, hash_table_key_wrong_offset,
|
||||
"hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n %s",
|
||||
bch2_btree_id_str(desc.btree_id), hash_k.k->p.inode, hash_k.k->p.offset, hash,
|
||||
bch2_btree_id_str(desc->btree_id), hash_k.k->p.inode, hash_k.k->p.offset, hash,
|
||||
(printbuf_reset(&buf),
|
||||
bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
|
||||
struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, hash_k);
|
||||
if (IS_ERR(new))
|
||||
return PTR_ERR(new);
|
||||
|
||||
k = bch2_hash_set_or_get_in_snapshot(trans, &iter, desc, hash_info,
|
||||
k = bch2_hash_set_or_get_in_snapshot(trans, &iter, *desc, hash_info,
|
||||
(subvol_inum) { 0, hash_k.k->p.inode },
|
||||
hash_k.k->p.snapshot, new,
|
||||
STR_HASH_must_create|
|
||||
|
|
@ -166,9 +243,9 @@ bad_hash:
|
|||
if (k.k)
|
||||
goto duplicate_entries;
|
||||
|
||||
ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter,
|
||||
ret = bch2_hash_delete_at(trans, *desc, hash_info, k_iter,
|
||||
BTREE_UPDATE_internal_snapshot_node) ?:
|
||||
bch2_fsck_update_backpointers(trans, s, desc, hash_info, new) ?:
|
||||
bch2_fsck_update_backpointers(trans, s, *desc, hash_info, new) ?:
|
||||
bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc) ?:
|
||||
-BCH_ERR_transaction_restart_nested;
|
||||
goto out;
|
||||
|
|
@ -176,7 +253,7 @@ bad_hash:
|
|||
fsck_err:
|
||||
goto out;
|
||||
duplicate_entries:
|
||||
ret = hash_pick_winner(trans, desc, hash_info, hash_k, k);
|
||||
ret = hash_pick_winner(trans, *desc, hash_info, hash_k, k);
|
||||
if (ret < 0)
|
||||
goto out;
|
||||
|
||||
|
|
@ -192,14 +269,14 @@ duplicate_entries:
|
|||
|
||||
switch (ret) {
|
||||
case 0:
|
||||
ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0);
|
||||
ret = bch2_hash_delete_at(trans, *desc, hash_info, k_iter, 0);
|
||||
break;
|
||||
case 1:
|
||||
ret = bch2_hash_delete_at(trans, desc, hash_info, &iter, 0);
|
||||
ret = bch2_hash_delete_at(trans, *desc, hash_info, &iter, 0);
|
||||
break;
|
||||
case 2:
|
||||
ret = fsck_rename_dirent(trans, s, desc, hash_info, bkey_s_c_to_dirent(hash_k)) ?:
|
||||
bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0);
|
||||
ret = fsck_rename_dirent(trans, s, *desc, hash_info, bkey_s_c_to_dirent(hash_k)) ?:
|
||||
bch2_hash_delete_at(trans, *desc, hash_info, k_iter, 0);
|
||||
goto out;
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -394,10 +394,25 @@ int bch2_hash_delete(struct btree_trans *trans,
|
|||
}
|
||||
|
||||
struct snapshots_seen;
|
||||
int bch2_str_hash_check_key(struct btree_trans *,
|
||||
struct snapshots_seen *,
|
||||
const struct bch_hash_desc,
|
||||
struct bch_hash_info *,
|
||||
struct btree_iter *, struct bkey_s_c);
|
||||
int __bch2_str_hash_check_key(struct btree_trans *,
|
||||
struct snapshots_seen *,
|
||||
const struct bch_hash_desc *,
|
||||
struct bch_hash_info *,
|
||||
struct btree_iter *, struct bkey_s_c);
|
||||
|
||||
static inline int bch2_str_hash_check_key(struct btree_trans *trans,
|
||||
struct snapshots_seen *s,
|
||||
const struct bch_hash_desc *desc,
|
||||
struct bch_hash_info *hash_info,
|
||||
struct btree_iter *k_iter, struct bkey_s_c hash_k)
|
||||
{
|
||||
if (hash_k.k->type != desc->key_type)
|
||||
return 0;
|
||||
|
||||
if (likely(desc->hash_bkey(hash_info, hash_k) == hash_k.k->p.offset))
|
||||
return 0;
|
||||
|
||||
return __bch2_str_hash_check_key(trans, s, desc, hash_info, k_iter, hash_k);
|
||||
}
|
||||
|
||||
#endif /* _BCACHEFS_STR_HASH_H */
|
||||
|
|
|
|||
Loading…
Reference in New Issue