X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Ffsck.c;h=ddc2782fc5b15ff4dcd5f5ff00e6d799601f31e6;hb=ee1d223ab2b186051716450050d6b508cd25bbfb;hp=936e6366cb0470da857456b7d4f7e07097881ab2;hpb=8bcd38555cb224b7da722bcdeca5c2d15f3ef284;p=bcachefs-tools-debian diff --git a/libbcachefs/fsck.c b/libbcachefs/fsck.c index 936e636..ddc2782 100644 --- a/libbcachefs/fsck.c +++ b/libbcachefs/fsck.c @@ -1,1493 +1,2450 @@ // SPDX-License-Identifier: GPL-2.0 #include "bcachefs.h" +#include "bkey_buf.h" #include "btree_update.h" +#include "buckets.h" +#include "darray.h" #include "dirent.h" #include "error.h" #include "fs-common.h" #include "fsck.h" #include "inode.h" #include "keylist.h" +#include "subvolume.h" #include "super.h" #include "xattr.h" +#include #include /* struct qstr */ -#include #define QSTR(n) { { { .len = strlen(n) } }, .name = n } -static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum) +/* + * XXX: this is handling transaction restarts without returning + * -BCH_ERR_transaction_restart_nested, this is not how we do things anymore: + */ +static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum, + u32 snapshot) { - struct btree_iter *iter; + struct btree_iter iter; struct bkey_s_c k; u64 sectors = 0; int ret; - for_each_btree_key(trans, iter, BTREE_ID_EXTENTS, - POS(inum, 0), 0, k, ret) { - if (k.k->p.inode != inum) - break; - + for_each_btree_key_upto(trans, iter, BTREE_ID_extents, + SPOS(inum, 0, snapshot), + POS(inum, U64_MAX), + 0, k, ret) if (bkey_extent_is_allocation(k.k)) sectors += k.k->size; - } - bch2_trans_iter_free(trans, iter); + bch2_trans_iter_exit(trans, &iter); return ret ?: sectors; } -static int __remove_dirent(struct btree_trans *trans, - struct bkey_s_c_dirent dirent) +static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum, + u32 snapshot) { - struct bch_fs *c = trans->c; - struct qstr name; - struct bch_inode_unpacked dir_inode; - struct bch_hash_info dir_hash_info; - u64 dir_inum = dirent.k->p.inode; + struct btree_iter iter; + struct bkey_s_c k; + struct bkey_s_c_dirent d; + u64 subdirs = 0; int ret; - char *buf; - - name.len = bch2_dirent_name_bytes(dirent); - buf = bch2_trans_kmalloc(trans, name.len + 1); - if (IS_ERR(buf)) - return PTR_ERR(buf); - - memcpy(buf, dirent.v->d_name, name.len); - buf[name.len] = '\0'; - name.name = buf; - - ret = bch2_inode_find_by_inum_trans(trans, dir_inum, &dir_inode); - if (ret && ret != -EINTR) - bch_err(c, "remove_dirent: err %i looking up directory inode", ret); - if (ret) - return ret; - dir_hash_info = bch2_hash_info_init(c, &dir_inode); + for_each_btree_key_upto(trans, iter, BTREE_ID_dirents, + SPOS(inum, 0, snapshot), + POS(inum, U64_MAX), + 0, k, ret) { + if (k.k->type != KEY_TYPE_dirent) + continue; - ret = bch2_hash_delete(trans, bch2_dirent_hash_desc, - &dir_hash_info, dir_inum, &name); - if (ret && ret != -EINTR) - bch_err(c, "remove_dirent: err %i deleting dirent", ret); - if (ret) - return ret; + d = bkey_s_c_to_dirent(k); + if (d.v->d_type == DT_DIR) + subdirs++; + } + bch2_trans_iter_exit(trans, &iter); - return 0; + return ret ?: subdirs; } -static int remove_dirent(struct btree_trans *trans, - struct bkey_s_c_dirent dirent) +static int __snapshot_lookup_subvol(struct btree_trans *trans, u32 snapshot, + u32 *subvol) { - return __bch2_trans_do(trans, NULL, NULL, - BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW, - __remove_dirent(trans, dirent)); + struct bch_snapshot s; + int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_snapshots, + POS(0, snapshot), 0, + snapshot, &s); + if (!ret) + *subvol = le32_to_cpu(s.subvol); + else if (bch2_err_matches(ret, ENOENT)) + bch_err(trans->c, "snapshot %u not fonud", snapshot); + return ret; + } -static int reattach_inode(struct bch_fs *c, - struct bch_inode_unpacked *lostfound_inode, - u64 inum) +static int __subvol_lookup(struct btree_trans *trans, u32 subvol, + u32 *snapshot, u64 *inum) { - struct bch_inode_unpacked dir_u, inode_u; - char name_buf[20]; - struct qstr name; + struct bch_subvolume s; int ret; - snprintf(name_buf, sizeof(name_buf), "%llu", inum); - name = (struct qstr) QSTR(name_buf); - - ret = bch2_trans_do(c, NULL, NULL, - BTREE_INSERT_LAZY_RW, - bch2_link_trans(&trans, lostfound_inode->bi_inum, - inum, &dir_u, &inode_u, &name)); - if (ret) - bch_err(c, "error %i reattaching inode %llu", ret, inum); + ret = bch2_subvolume_get(trans, subvol, false, 0, &s); + *snapshot = le32_to_cpu(s.snapshot); + *inum = le64_to_cpu(s.inode); return ret; } -struct inode_walker { - bool first_this_inode; - bool have_inode; - u64 cur_inum; - struct bch_inode_unpacked inode; -}; - -static struct inode_walker inode_walker_init(void) +static int subvol_lookup(struct btree_trans *trans, u32 subvol, + u32 *snapshot, u64 *inum) { - return (struct inode_walker) { - .cur_inum = -1, - .have_inode = false, - }; + return lockrestart_do(trans, __subvol_lookup(trans, subvol, snapshot, inum)); } -static int walk_inode(struct btree_trans *trans, - struct inode_walker *w, u64 inum) +static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr, + struct bch_inode_unpacked *inode) { - if (inum != w->cur_inum) { - int ret = bch2_inode_find_by_inum_trans(trans, inum, - &w->inode); + struct btree_iter iter; + struct bkey_s_c k; + int ret; - if (ret && ret != -ENOENT) - return ret; + bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, + POS(0, inode_nr), + BTREE_ITER_ALL_SNAPSHOTS); + k = bch2_btree_iter_peek(&iter); + ret = bkey_err(k); + if (ret) + goto err; - w->have_inode = !ret; - w->cur_inum = inum; - w->first_this_inode = true; - } else { - w->first_this_inode = false; + if (!k.k || !bkey_eq(k.k->p, POS(0, inode_nr))) { + ret = -BCH_ERR_ENOENT_inode; + goto err; } - return 0; + ret = bch2_inode_unpack(k, inode); +err: + if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart)) + bch_err(trans->c, "error fetching inode %llu: %s", + inode_nr, bch2_err_str(ret)); + bch2_trans_iter_exit(trans, &iter); + return ret; } -struct hash_check { - struct bch_hash_info info; - - /* start of current chain of hash collisions: */ - struct btree_iter *chain; +static int __lookup_inode(struct btree_trans *trans, u64 inode_nr, + struct bch_inode_unpacked *inode, + u32 *snapshot) +{ + struct btree_iter iter; + struct bkey_s_c k; + int ret; - /* next offset in current chain of hash collisions: */ - u64 chain_end; -}; + k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes, + SPOS(0, inode_nr, *snapshot), 0); + ret = bkey_err(k); + if (ret) + goto err; -static void hash_check_init(struct hash_check *h) -{ - h->chain = NULL; - h->chain_end = 0; + ret = bkey_is_inode(k.k) + ? bch2_inode_unpack(k, inode) + : -BCH_ERR_ENOENT_inode; + if (!ret) + *snapshot = iter.pos.snapshot; +err: + if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart)) + bch_err(trans->c, "error fetching inode %llu:%u: %s", + inode_nr, *snapshot, bch2_err_str(ret)); + bch2_trans_iter_exit(trans, &iter); + return ret; } -static void hash_stop_chain(struct btree_trans *trans, - struct hash_check *h) +static int lookup_inode(struct btree_trans *trans, u64 inode_nr, + struct bch_inode_unpacked *inode, + u32 *snapshot) { - if (h->chain) - bch2_trans_iter_free(trans, h->chain); - h->chain = NULL; + return lockrestart_do(trans, __lookup_inode(trans, inode_nr, inode, snapshot)); } -static void hash_check_set_inode(struct btree_trans *trans, - struct hash_check *h, - const struct bch_inode_unpacked *bi) +static int __lookup_dirent(struct btree_trans *trans, + struct bch_hash_info hash_info, + subvol_inum dir, struct qstr *name, + u64 *target, unsigned *type) { - h->info = bch2_hash_info_init(trans->c, bi); - hash_stop_chain(trans, h); + struct btree_iter iter; + struct bkey_s_c_dirent d; + int ret; + + ret = bch2_hash_lookup(trans, &iter, bch2_dirent_hash_desc, + &hash_info, dir, name, 0); + if (ret) + return ret; + + d = bkey_s_c_to_dirent(bch2_btree_iter_peek_slot(&iter)); + *target = le64_to_cpu(d.v->d_inum); + *type = d.v->d_type; + bch2_trans_iter_exit(trans, &iter); + return 0; } -static int hash_redo_key(const struct bch_hash_desc desc, - struct btree_trans *trans, struct hash_check *h, - struct btree_iter *k_iter, struct bkey_s_c k, - u64 hashed) +static int __write_inode(struct btree_trans *trans, + struct bch_inode_unpacked *inode, + u32 snapshot) { - struct bkey_i delete; - struct bkey_i *tmp; - - tmp = bch2_trans_kmalloc(trans, bkey_bytes(k.k)); - if (IS_ERR(tmp)) - return PTR_ERR(tmp); + struct bkey_inode_buf *inode_p = + bch2_trans_kmalloc(trans, sizeof(*inode_p)); - bkey_reassemble(tmp, k); + if (IS_ERR(inode_p)) + return PTR_ERR(inode_p); - bkey_init(&delete.k); - delete.k.p = k_iter->pos; - bch2_trans_update(trans, k_iter, &delete, 0); + bch2_inode_pack(inode_p, inode); + inode_p->inode.k.p.snapshot = snapshot; - return bch2_hash_set(trans, desc, &h->info, k_iter->pos.inode, - tmp, BCH_HASH_SET_MUST_CREATE); + return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes, + &inode_p->inode.k_i, + BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE); } -static int fsck_hash_delete_at(struct btree_trans *trans, - const struct bch_hash_desc desc, - struct bch_hash_info *info, - struct btree_iter *iter) +static int write_inode(struct btree_trans *trans, + struct bch_inode_unpacked *inode, + u32 snapshot) { - int ret; -retry: - ret = bch2_hash_delete_at(trans, desc, info, iter) ?: - bch2_trans_commit(trans, NULL, NULL, + int ret = commit_do(trans, NULL, NULL, BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW); - if (ret == -EINTR) { - ret = bch2_btree_iter_traverse(iter); - if (!ret) - goto retry; - } - + BTREE_INSERT_LAZY_RW, + __write_inode(trans, inode, snapshot)); + if (ret) + bch_err(trans->c, "error in fsck: error updating inode: %s", + bch2_err_str(ret)); return ret; } -static int hash_check_duplicates(struct btree_trans *trans, - const struct bch_hash_desc desc, struct hash_check *h, - struct btree_iter *k_iter, struct bkey_s_c k) +static int fsck_inode_rm(struct btree_trans *trans, u64 inum, u32 snapshot) { struct bch_fs *c = trans->c; - struct btree_iter *iter; - struct bkey_s_c k2; - char buf[200]; - int ret = 0; - - if (!bkey_cmp(h->chain->pos, k_iter->pos)) - return 0; + struct btree_iter iter = { NULL }; + struct bkey_i_inode_generation delete; + struct bch_inode_unpacked inode_u; + struct bkey_s_c k; + int ret; - iter = bch2_trans_copy_iter(trans, h->chain); - BUG_ON(IS_ERR(iter)); + do { + ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents, + SPOS(inum, 0, snapshot), + SPOS(inum, U64_MAX, snapshot), + 0, NULL) ?: + bch2_btree_delete_range_trans(trans, BTREE_ID_dirents, + SPOS(inum, 0, snapshot), + SPOS(inum, U64_MAX, snapshot), + 0, NULL) ?: + bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs, + SPOS(inum, 0, snapshot), + SPOS(inum, U64_MAX, snapshot), + 0, NULL); + } while (ret == -BCH_ERR_transaction_restart_nested); + if (ret) + goto err; +retry: + bch2_trans_begin(trans); - for_each_btree_key_continue(iter, 0, k2, ret) { - if (bkey_cmp(k2.k->p, k.k->p) >= 0) - break; + k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes, + SPOS(0, inum, snapshot), BTREE_ITER_INTENT); + ret = bkey_err(k); + if (ret) + goto err; - if (fsck_err_on(k2.k->type == desc.key_type && - !desc.cmp_bkey(k, k2), c, - "duplicate hash table keys:\n%s", - (bch2_bkey_val_to_text(&PBUF(buf), c, - k), buf))) { - ret = fsck_hash_delete_at(trans, desc, &h->info, k_iter); - if (ret) - return ret; - ret = 1; - break; - } + if (!bkey_is_inode(k.k)) { + bch2_fs_inconsistent(c, + "inode %llu:%u not found when deleting", + inum, snapshot); + ret = -EIO; + goto err; } -fsck_err: - bch2_trans_iter_free(trans, iter); - return ret; -} -static void hash_set_chain_start(struct btree_trans *trans, - const struct bch_hash_desc desc, - struct hash_check *h, - struct btree_iter *k_iter, struct bkey_s_c k) -{ - bool hole = (k.k->type != KEY_TYPE_whiteout && - k.k->type != desc.key_type); + bch2_inode_unpack(k, &inode_u); - if (hole || k.k->p.offset > h->chain_end + 1) - hash_stop_chain(trans, h); + /* Subvolume root? */ + if (inode_u.bi_subvol) + bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum); - if (!hole) { - if (!h->chain) { - h->chain = bch2_trans_copy_iter(trans, k_iter); - BUG_ON(IS_ERR(h->chain)); - } + bkey_inode_generation_init(&delete.k_i); + delete.k.p = iter.pos; + delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1); - h->chain_end = k.k->p.offset; - } + ret = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?: + bch2_trans_commit(trans, NULL, NULL, + BTREE_INSERT_NOFAIL); +err: + bch2_trans_iter_exit(trans, &iter); + if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) + goto retry; + + return ret ?: -BCH_ERR_transaction_restart_nested; } -static bool key_has_correct_hash(struct btree_trans *trans, - const struct bch_hash_desc desc, - struct hash_check *h, - struct btree_iter *k_iter, struct bkey_s_c k) +static int __remove_dirent(struct btree_trans *trans, struct bpos pos) { - u64 hash; + struct bch_fs *c = trans->c; + struct btree_iter iter; + struct bch_inode_unpacked dir_inode; + struct bch_hash_info dir_hash_info; + int ret; - hash_set_chain_start(trans, desc, h, k_iter, k); + ret = lookup_first_inode(trans, pos.inode, &dir_inode); + if (ret) + goto err; - if (k.k->type != desc.key_type) - return true; + dir_hash_info = bch2_hash_info_init(c, &dir_inode); - hash = desc.hash_bkey(&h->info, k); + bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT); - return hash >= h->chain->pos.offset && - hash <= k.k->p.offset; + ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc, + &dir_hash_info, &iter, + BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE); + bch2_trans_iter_exit(trans, &iter); +err: + if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart)) + bch_err_fn(c, ret); + return ret; } -static int hash_check_key(struct btree_trans *trans, - const struct bch_hash_desc desc, struct hash_check *h, - struct btree_iter *k_iter, struct bkey_s_c k) +/* Get lost+found, create if it doesn't exist: */ +static int lookup_lostfound(struct btree_trans *trans, u32 subvol, + struct bch_inode_unpacked *lostfound) { struct bch_fs *c = trans->c; - char buf[200]; - u64 hashed; - int ret = 0; + struct bch_inode_unpacked root; + struct bch_hash_info root_hash_info; + struct qstr lostfound_str = QSTR("lost+found"); + subvol_inum root_inum = { .subvol = subvol }; + u64 inum = 0; + unsigned d_type = 0; + u32 snapshot; + int ret; + + ret = __subvol_lookup(trans, subvol, &snapshot, &root_inum.inum); + if (ret) + return ret; - hash_set_chain_start(trans, desc, h, k_iter, k); + ret = __lookup_inode(trans, root_inum.inum, &root, &snapshot); + if (ret) + return ret; - if (k.k->type != desc.key_type) - return 0; + root_hash_info = bch2_hash_info_init(c, &root); - hashed = desc.hash_bkey(&h->info, k); - - if (fsck_err_on(hashed < h->chain->pos.offset || - hashed > k.k->p.offset, c, - "hash table key at wrong offset: btree %u, %llu, " - "hashed to %llu chain starts at %llu\n%s", - desc.btree_id, k.k->p.offset, - hashed, h->chain->pos.offset, - (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf))) { - ret = __bch2_trans_do(trans, NULL, NULL, - BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW, - hash_redo_key(desc, trans, h, k_iter, k, hashed)); - if (ret) { - bch_err(c, "hash_redo_key err %i", ret); - return ret; - } - return 1; + ret = __lookup_dirent(trans, root_hash_info, root_inum, + &lostfound_str, &inum, &d_type); + if (bch2_err_matches(ret, ENOENT)) { + bch_notice(c, "creating lost+found"); + goto create_lostfound; } - ret = hash_check_duplicates(trans, desc, h, k_iter, k); -fsck_err: + if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart)) + bch_err(c, "error looking up lost+found: %s", bch2_err_str(ret)); + if (ret) + return ret; + + if (d_type != DT_DIR) { + bch_err(c, "error looking up lost+found: not a directory"); + return ret; + } + + /* + * The bch2_check_dirents pass has already run, dangling dirents + * shouldn't exist here: + */ + return __lookup_inode(trans, inum, lostfound, &snapshot); + +create_lostfound: + bch2_inode_init_early(c, lostfound); + + ret = bch2_create_trans(trans, root_inum, &root, + lostfound, &lostfound_str, + 0, 0, S_IFDIR|0700, 0, NULL, NULL, + (subvol_inum) { }, 0); + if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart)) + bch_err(c, "error creating lost+found: %s", bch2_err_str(ret)); return ret; } -static int check_dirent_hash(struct btree_trans *trans, struct hash_check *h, - struct btree_iter *iter, struct bkey_s_c *k) +static int __reattach_inode(struct btree_trans *trans, + struct bch_inode_unpacked *inode, + u32 inode_snapshot) { - struct bch_fs *c = trans->c; - struct bkey_i_dirent *d = NULL; - int ret = -EINVAL; - char buf[200]; - unsigned len; - u64 hash; + struct bch_hash_info dir_hash; + struct bch_inode_unpacked lostfound; + char name_buf[20]; + struct qstr name; + u64 dir_offset = 0; + u32 subvol; + int ret; - if (key_has_correct_hash(trans, bch2_dirent_hash_desc, h, iter, *k)) - return 0; + ret = __snapshot_lookup_subvol(trans, inode_snapshot, &subvol); + if (ret) + return ret; - len = bch2_dirent_name_bytes(bkey_s_c_to_dirent(*k)); - BUG_ON(!len); + ret = lookup_lostfound(trans, subvol, &lostfound); + if (ret) + return ret; - memcpy(buf, bkey_s_c_to_dirent(*k).v->d_name, len); - buf[len] = '\0'; + if (S_ISDIR(inode->bi_mode)) { + lostfound.bi_nlink++; - d = kmalloc(bkey_bytes(k->k), GFP_KERNEL); - if (!d) { - bch_err(c, "memory allocation failure"); - return -ENOMEM; + ret = __write_inode(trans, &lostfound, U32_MAX); + if (ret) + return ret; } - bkey_reassemble(&d->k_i, *k); + dir_hash = bch2_hash_info_init(trans->c, &lostfound); - do { - --len; - if (!len) - goto err_redo; + snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum); + name = (struct qstr) QSTR(name_buf); - d->k.u64s = BKEY_U64s + dirent_val_u64s(len); + ret = bch2_dirent_create(trans, + (subvol_inum) { + .subvol = subvol, + .inum = lostfound.bi_inum, + }, + &dir_hash, + inode_d_type(inode), + &name, inode->bi_inum, &dir_offset, + BCH_HASH_SET_MUST_CREATE); + if (ret) + return ret; - BUG_ON(bkey_val_bytes(&d->k) < - offsetof(struct bch_dirent, d_name) + len); + inode->bi_dir = lostfound.bi_inum; + inode->bi_dir_offset = dir_offset; - memset(d->v.d_name + len, 0, - bkey_val_bytes(&d->k) - - offsetof(struct bch_dirent, d_name) - len); + return __write_inode(trans, inode, inode_snapshot); +} - hash = bch2_dirent_hash_desc.hash_bkey(&h->info, - bkey_i_to_s_c(&d->k_i)); - } while (hash < h->chain->pos.offset || - hash > k->k->p.offset); +static int reattach_inode(struct btree_trans *trans, + struct bch_inode_unpacked *inode, + u32 inode_snapshot) +{ + int ret = commit_do(trans, NULL, NULL, + BTREE_INSERT_LAZY_RW| + BTREE_INSERT_NOFAIL, + __reattach_inode(trans, inode, inode_snapshot)); + if (ret) { + bch_err(trans->c, "error reattaching inode %llu: %s", + inode->bi_inum, bch2_err_str(ret)); + return ret; + } - if (fsck_err(c, "dirent with junk at end, was %s (%zu) now %s (%u)", - buf, strlen(buf), d->v.d_name, len)) { - ret = __bch2_trans_do(trans, NULL, NULL, - BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW, - (bch2_trans_update(trans, iter, &d->k_i, 0), 0)); - if (ret) - goto err; + return ret; +} - *k = bch2_btree_iter_peek(iter); +static int remove_backpointer(struct btree_trans *trans, + struct bch_inode_unpacked *inode) +{ + struct btree_iter iter; + struct bkey_s_c_dirent d; + int ret; - BUG_ON(k->k->type != KEY_TYPE_dirent); - } -err: -fsck_err: - kfree(d); + d = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_dirents, + POS(inode->bi_dir, inode->bi_dir_offset), 0, + dirent); + ret = bkey_err(d) ?: + __remove_dirent(trans, d.k->p); + bch2_trans_iter_exit(trans, &iter); return ret; -err_redo: - hash = bch2_dirent_hash_desc.hash_bkey(&h->info, *k); - - if (fsck_err(c, "cannot fix dirent by removing trailing garbage %s (%zu)\n" - "hash table key at wrong offset: btree %u, offset %llu, " - "hashed to %llu chain starts at %llu\n%s", - buf, strlen(buf), BTREE_ID_DIRENTS, - k->k->p.offset, hash, h->chain->pos.offset, - (bch2_bkey_val_to_text(&PBUF(buf), c, - *k), buf))) { - ret = __bch2_trans_do(trans, NULL, NULL, - BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW, - hash_redo_key(bch2_dirent_hash_desc, trans, - h, iter, *k, hash)); - if (ret) - bch_err(c, "hash_redo_key err %i", ret); - else - ret = 1; - } +} + +struct snapshots_seen_entry { + u32 id; + u32 equiv; +}; + +struct snapshots_seen { + struct bpos pos; + DARRAY(struct snapshots_seen_entry) ids; +}; - goto err; +static inline void snapshots_seen_exit(struct snapshots_seen *s) +{ + darray_exit(&s->ids); } -static int bch2_inode_truncate(struct bch_fs *c, u64 inode_nr, u64 new_size) +static inline void snapshots_seen_init(struct snapshots_seen *s) { - return bch2_btree_delete_range(c, BTREE_ID_EXTENTS, - POS(inode_nr, round_up(new_size, block_bytes(c)) >> 9), - POS(inode_nr + 1, 0), NULL); + memset(s, 0, sizeof(*s)); } -static int bch2_fix_overlapping_extent(struct btree_trans *trans, - struct btree_iter *iter, - struct bkey_s_c k, struct bpos cut_at) +static int snapshots_seen_add(struct bch_fs *c, struct snapshots_seen *s, u32 id) { - struct btree_iter *u_iter; - struct bkey_i *u; + struct snapshots_seen_entry *i, n = { id, id }; int ret; - u = bch2_trans_kmalloc(trans, bkey_bytes(k.k)); - ret = PTR_ERR_OR_ZERO(u); - if (ret) - return ret; + darray_for_each(s->ids, i) { + if (n.equiv < i->equiv) + break; - bkey_reassemble(u, k); - bch2_cut_front(cut_at, u); + if (i->equiv == n.equiv) { + bch_err(c, "%s(): adding duplicate snapshot", __func__); + return -EINVAL; + } + } - u_iter = bch2_trans_copy_iter(trans, iter); - ret = PTR_ERR_OR_ZERO(u_iter); + ret = darray_insert_item(&s->ids, i - s->ids.data, n); if (ret) - return ret; - - /* - * We don't want to go through the - * extent_handle_overwrites path: - */ - __bch2_btree_iter_set_pos(u_iter, u->k.p, false); - - /* - * XXX: this is going to leave disk space - * accounting slightly wrong - */ - ret = bch2_trans_update(trans, u_iter, u, 0); - bch2_trans_iter_put(trans, u_iter); + bch_err(c, "error reallocating snapshots_seen table (size %zu)", + s->ids.size); return ret; } -/* - * Walk extents: verify that extents have a corresponding S_ISREG inode, and - * that i_size an i_sectors are consistent - */ -noinline_for_stack -static int check_extents(struct bch_fs *c) +static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s, + enum btree_id btree_id, struct bpos pos) { - struct inode_walker w = inode_walker_init(); - struct btree_trans trans; - struct btree_iter *iter; - struct bkey_s_c k; - struct bkey prev = KEY(0, 0, 0); - u64 i_sectors; + struct snapshots_seen_entry *i, n = { + .id = pos.snapshot, + .equiv = bch2_snapshot_equiv(c, pos.snapshot), + }; int ret = 0; - bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0); - - bch_verbose(c, "checking extents"); - - iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, - POS(BCACHEFS_ROOT_INO, 0), - BTREE_ITER_INTENT); -retry: - for_each_btree_key_continue(iter, 0, k, ret) { - if (bkey_cmp(prev.p, bkey_start_pos(k.k)) > 0) { - char buf1[100]; - char buf2[100]; - - bch2_bkey_to_text(&PBUF(buf1), &prev); - bch2_bkey_to_text(&PBUF(buf2), k.k); - - if (fsck_err(c, "overlapping extents: %s, %s", buf1, buf2)) { - ret = __bch2_trans_do(&trans, NULL, NULL, - BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW, - bch2_fix_overlapping_extent(&trans, - iter, k, prev.p)); - if (ret) - goto err; - } - } - prev = *k.k; + if (!bkey_eq(s->pos, pos)) + s->ids.nr = 0; - ret = walk_inode(&trans, &w, k.k->p.inode); - if (ret) - break; + pos.snapshot = n.equiv; + s->pos = pos; - if (fsck_err_on(!w.have_inode, c, - "extent type %u for missing inode %llu", - k.k->type, k.k->p.inode) || - fsck_err_on(w.have_inode && - !S_ISREG(w.inode.bi_mode) && !S_ISLNK(w.inode.bi_mode), c, - "extent type %u for non regular file, inode %llu mode %o", - k.k->type, k.k->p.inode, w.inode.bi_mode)) { - bch2_trans_unlock(&trans); + darray_for_each(s->ids, i) + if (i->equiv == n.equiv) { + if (fsck_err_on(i->id != n.id, c, + "snapshot deletion did not run correctly:\n" + " duplicate keys in btree %s at %llu:%llu snapshots %u, %u (equiv %u)\n", + bch2_btree_ids[btree_id], + pos.inode, pos.offset, + i->id, n.id, n.equiv)) + return -BCH_ERR_need_snapshot_cleanup; - ret = bch2_inode_truncate(c, k.k->p.inode, 0); - if (ret) - goto err; - continue; + return 0; } - if (fsck_err_on(w.first_this_inode && - w.have_inode && - !(w.inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY) && - w.inode.bi_sectors != - (i_sectors = bch2_count_inode_sectors(&trans, w.cur_inum)), - c, "i_sectors wrong: got %llu, should be %llu", - w.inode.bi_sectors, i_sectors)) { - struct bkey_inode_buf p; + ret = darray_push(&s->ids, n); + if (ret) + bch_err(c, "error reallocating snapshots_seen table (size %zu)", + s->ids.size); +fsck_err: + return ret; +} - w.inode.bi_sectors = i_sectors; +/** + * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor, + * and @ancestor hasn't been overwritten in @seen + * + * That is, returns whether key in @ancestor snapshot is visible in @id snapshot + */ +static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen, + u32 id, u32 ancestor) +{ + ssize_t i; + u32 top = seen->ids.nr ? seen->ids.data[seen->ids.nr - 1].equiv : 0; - bch2_trans_unlock(&trans); + BUG_ON(id > ancestor); + BUG_ON(!bch2_snapshot_is_equiv(c, id)); + BUG_ON(!bch2_snapshot_is_equiv(c, ancestor)); - bch2_inode_pack(&p, &w.inode); + /* @ancestor should be the snapshot most recently added to @seen */ + BUG_ON(ancestor != seen->pos.snapshot); + BUG_ON(ancestor != top); - ret = bch2_btree_insert(c, BTREE_ID_INODES, - &p.inode.k_i, NULL, NULL, - BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW); - if (ret) { - bch_err(c, "error in fsck: error %i updating inode", ret); - goto err; - } + if (id == ancestor) + return true; - /* revalidate iterator: */ - k = bch2_btree_iter_peek(iter); - } + if (!bch2_snapshot_is_ancestor(c, id, ancestor)) + return false; - if (fsck_err_on(w.have_inode && - !(w.inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) && - k.k->type != KEY_TYPE_reservation && - k.k->p.offset > round_up(w.inode.bi_size, block_bytes(c)) >> 9, c, - "extent type %u offset %llu past end of inode %llu, i_size %llu", - k.k->type, k.k->p.offset, k.k->p.inode, w.inode.bi_size)) { - bch2_trans_unlock(&trans); + for (i = seen->ids.nr - 2; + i >= 0 && seen->ids.data[i].equiv >= id; + --i) + if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i].equiv) && + bch2_snapshot_is_ancestor(c, seen->ids.data[i].equiv, ancestor)) + return false; - ret = bch2_inode_truncate(c, k.k->p.inode, - w.inode.bi_size); - if (ret) - goto err; - continue; - } - } -err: -fsck_err: - if (ret == -EINTR) - goto retry; - return bch2_trans_exit(&trans) ?: ret; + return true; } -/* - * Walk dirents: verify that they all have a corresponding S_ISDIR inode, - * validate d_type +/** + * ref_visible - given a key with snapshot id @src that points to a key with + * snapshot id @dst, test whether there is some snapshot in which @dst is + * visible. + * + * This assumes we're visiting @src keys in natural key order. + * + * @s - list of snapshot IDs already seen at @src + * @src - snapshot ID of src key + * @dst - snapshot ID of dst key */ -noinline_for_stack -static int check_dirents(struct bch_fs *c) +static int ref_visible(struct bch_fs *c, struct snapshots_seen *s, + u32 src, u32 dst) { - struct inode_walker w = inode_walker_init(); - struct hash_check h; - struct btree_trans trans; - struct btree_iter *iter; - struct bkey_s_c k; - unsigned name_len; - char buf[200]; - int ret = 0; + return dst <= src + ? key_visible_in_snapshot(c, s, dst, src) + : bch2_snapshot_is_ancestor(c, src, dst); +} - bch_verbose(c, "checking dirents"); +static int ref_visible2(struct bch_fs *c, + u32 src, struct snapshots_seen *src_seen, + u32 dst, struct snapshots_seen *dst_seen) +{ + src = bch2_snapshot_equiv(c, src); + dst = bch2_snapshot_equiv(c, dst); - bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0); + if (dst > src) { + swap(dst, src); + swap(dst_seen, src_seen); + } + return key_visible_in_snapshot(c, src_seen, dst, src); +} - hash_check_init(&h); +#define for_each_visible_inode(_c, _s, _w, _snapshot, _i) \ + for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr && \ + (_i)->snapshot <= (_snapshot); _i++) \ + if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot)) - iter = bch2_trans_get_iter(&trans, BTREE_ID_DIRENTS, - POS(BCACHEFS_ROOT_INO, 0), 0); -retry: - for_each_btree_key_continue(iter, 0, k, ret) { - struct bkey_s_c_dirent d; - struct bch_inode_unpacked target; - bool have_target; - u64 d_inum; +struct inode_walker_entry { + struct bch_inode_unpacked inode; + u32 snapshot; + u64 count; +}; + +struct inode_walker { + bool first_this_inode; + u64 cur_inum; + + DARRAY(struct inode_walker_entry) inodes; +}; + +static void inode_walker_exit(struct inode_walker *w) +{ + darray_exit(&w->inodes); +} + +static struct inode_walker inode_walker_init(void) +{ + return (struct inode_walker) { 0, }; +} + +static int add_inode(struct bch_fs *c, struct inode_walker *w, + struct bkey_s_c inode) +{ + struct bch_inode_unpacked u; + + BUG_ON(bch2_inode_unpack(inode, &u)); + + return darray_push(&w->inodes, ((struct inode_walker_entry) { + .inode = u, + .snapshot = bch2_snapshot_equiv(c, inode.k->p.snapshot), + })); +} + +static int get_inodes_all_snapshots(struct btree_trans *trans, + struct inode_walker *w, u64 inum) +{ + struct bch_fs *c = trans->c; + struct btree_iter iter; + struct bkey_s_c k; + u32 restart_count = trans->restart_count; + int ret; + + if (w->cur_inum == inum) + return 0; + + w->inodes.nr = 0; + + for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, inum), + BTREE_ITER_ALL_SNAPSHOTS, k, ret) { + if (k.k->p.offset != inum) + break; + + if (bkey_is_inode(k.k)) + add_inode(c, w, k); + } + bch2_trans_iter_exit(trans, &iter); + + if (ret) + return ret; + + w->cur_inum = inum; + w->first_this_inode = true; + + if (trans_was_restarted(trans, restart_count)) + return -BCH_ERR_transaction_restart_nested; + + return 0; +} + +static struct inode_walker_entry * +lookup_inode_for_snapshot(struct bch_fs *c, + struct inode_walker *w, u32 snapshot) +{ + struct inode_walker_entry *i; + + snapshot = bch2_snapshot_equiv(c, snapshot); + + darray_for_each(w->inodes, i) + if (bch2_snapshot_is_ancestor(c, snapshot, i->snapshot)) + goto found; + + return NULL; +found: + BUG_ON(snapshot > i->snapshot); + + if (snapshot != i->snapshot) { + struct inode_walker_entry new = *i; + int ret; + + new.snapshot = snapshot; + new.count = 0; + + bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u", + w->cur_inum, snapshot, i->snapshot); + + while (i > w->inodes.data && i[-1].snapshot > snapshot) + --i; + + ret = darray_insert_item(&w->inodes, i - w->inodes.data, new); + if (ret) + return ERR_PTR(ret); + } + + return i; +} + +static struct inode_walker_entry *walk_inode(struct btree_trans *trans, + struct inode_walker *w, struct bpos pos) +{ + int ret = get_inodes_all_snapshots(trans, w, pos.inode); + if (ret) + return ERR_PTR(ret); + + return lookup_inode_for_snapshot(trans->c, w, pos.snapshot); +} + +static int __get_visible_inodes(struct btree_trans *trans, + struct inode_walker *w, + struct snapshots_seen *s, + u64 inum) +{ + struct bch_fs *c = trans->c; + struct btree_iter iter; + struct bkey_s_c k; + int ret; + + w->inodes.nr = 0; + + for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum), + BTREE_ITER_ALL_SNAPSHOTS, k, ret) { + u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot); + + if (k.k->p.offset != inum) + break; + + if (!ref_visible(c, s, s->pos.snapshot, equiv)) + continue; + + if (bkey_is_inode(k.k)) + add_inode(c, w, k); + + if (equiv >= s->pos.snapshot) + break; + } + bch2_trans_iter_exit(trans, &iter); + + return ret; +} + +static int check_key_has_snapshot(struct btree_trans *trans, + struct btree_iter *iter, + struct bkey_s_c k) +{ + struct bch_fs *c = trans->c; + struct printbuf buf = PRINTBUF; + int ret = 0; + + if (mustfix_fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot), c, + "key in missing snapshot: %s", + (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) + ret = bch2_btree_delete_at(trans, iter, + BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1; +fsck_err: + printbuf_exit(&buf); + return ret; +} + +static int hash_redo_key(struct btree_trans *trans, + const struct bch_hash_desc desc, + struct bch_hash_info *hash_info, + struct btree_iter *k_iter, struct bkey_s_c k) +{ + struct bkey_i *delete; + struct bkey_i *tmp; + + delete = bch2_trans_kmalloc(trans, sizeof(*delete)); + if (IS_ERR(delete)) + return PTR_ERR(delete); + + tmp = bch2_bkey_make_mut_noupdate(trans, k); + if (IS_ERR(tmp)) + return PTR_ERR(tmp); + + bkey_init(&delete->k); + delete->k.p = k_iter->pos; + return bch2_btree_iter_traverse(k_iter) ?: + bch2_trans_update(trans, k_iter, delete, 0) ?: + bch2_hash_set_snapshot(trans, desc, hash_info, + (subvol_inum) { 0, k.k->p.inode }, + k.k->p.snapshot, tmp, + BCH_HASH_SET_MUST_CREATE, + BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: + bch2_trans_commit(trans, NULL, NULL, + BTREE_INSERT_NOFAIL| + BTREE_INSERT_LAZY_RW); +} + +static int hash_check_key(struct btree_trans *trans, + 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; + + if (hash_k.k->p.offset < hash) + goto bad_hash; + + 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 (fsck_err_on(k.k->type == desc.key_type && + !desc.cmp_bkey(k, hash_k), c, + "duplicate hash table keys:\n%s", + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, hash_k), + buf.buf))) { + ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1; + break; + } + + if (bkey_deleted(k.k)) { + bch2_trans_iter_exit(trans, &iter); + goto bad_hash; + } + } +out: + bch2_trans_iter_exit(trans, &iter); + printbuf_exit(&buf); + return ret; +bad_hash: + if (fsck_err(c, "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s", + bch2_btree_ids[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))) { + ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k); + if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart)) + bch_err(c, "hash_redo_key err %s", bch2_err_str(ret)); + if (ret) + return ret; + ret = -BCH_ERR_transaction_restart_nested; + } +fsck_err: + goto out; +} + +static int check_inode(struct btree_trans *trans, + struct btree_iter *iter, + struct bkey_s_c k, + struct bch_inode_unpacked *prev, + struct snapshots_seen *s, + bool full) +{ + struct bch_fs *c = trans->c; + struct bch_inode_unpacked u; + bool do_update = false; + int ret; + + ret = check_key_has_snapshot(trans, iter, k); + if (ret < 0) + goto err; + if (ret) + return 0; + + ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p); + if (ret) + goto err; + + /* + * if snapshot id isn't a leaf node, skip it - deletion in + * particular is not atomic, so on the internal snapshot nodes + * we can see inodes marked for deletion after a clean shutdown + */ + if (bch2_snapshot_internal_node(c, k.k->p.snapshot)) + return 0; + + if (!bkey_is_inode(k.k)) + return 0; + + BUG_ON(bch2_inode_unpack(k, &u)); + + if (!full && + !(u.bi_flags & (BCH_INODE_I_SIZE_DIRTY| + BCH_INODE_I_SECTORS_DIRTY| + BCH_INODE_UNLINKED))) + return 0; + + if (prev->bi_inum != u.bi_inum) + *prev = u; + + if (fsck_err_on(prev->bi_hash_seed != u.bi_hash_seed || + inode_d_type(prev) != inode_d_type(&u), c, + "inodes in different snapshots don't match")) { + bch_err(c, "repair not implemented yet"); + return -EINVAL; + } + + if (u.bi_flags & BCH_INODE_UNLINKED && + (!c->sb.clean || + fsck_err(c, "filesystem marked clean, but inode %llu unlinked", + u.bi_inum))) { + bch2_trans_unlock(trans); + bch2_fs_lazy_rw(c); + + ret = fsck_inode_rm(trans, u.bi_inum, iter->pos.snapshot); + if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart)) + bch_err(c, "error in fsck: error while deleting inode: %s", + bch2_err_str(ret)); + return ret; + } + + if (u.bi_flags & BCH_INODE_I_SIZE_DIRTY && + (!c->sb.clean || + fsck_err(c, "filesystem marked clean, but inode %llu has i_size dirty", + u.bi_inum))) { + bch_verbose(c, "truncating inode %llu", u.bi_inum); + + bch2_trans_unlock(trans); + bch2_fs_lazy_rw(c); + + /* + * XXX: need to truncate partial blocks too here - or ideally + * just switch units to bytes and that issue goes away + */ + ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents, + SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9, + iter->pos.snapshot), + POS(u.bi_inum, U64_MAX), + 0, NULL); + if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart)) + bch_err(c, "error in fsck: error truncating inode: %s", + bch2_err_str(ret)); + if (ret) + return ret; + + /* + * We truncated without our normal sector accounting hook, just + * make sure we recalculate it: + */ + u.bi_flags |= BCH_INODE_I_SECTORS_DIRTY; + + u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY; + do_update = true; + } + + if (u.bi_flags & BCH_INODE_I_SECTORS_DIRTY && + (!c->sb.clean || + fsck_err(c, "filesystem marked clean, but inode %llu has i_sectors dirty", + u.bi_inum))) { + s64 sectors; + + bch_verbose(c, "recounting sectors for inode %llu", + u.bi_inum); + + sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot); + if (sectors < 0) { + bch_err(c, "error in fsck: error recounting inode sectors: %s", + bch2_err_str(sectors)); + return sectors; + } + + u.bi_sectors = sectors; + u.bi_flags &= ~BCH_INODE_I_SECTORS_DIRTY; + do_update = true; + } + + if (u.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) { + u.bi_dir = 0; + u.bi_dir_offset = 0; + u.bi_flags &= ~BCH_INODE_BACKPTR_UNTRUSTED; + do_update = true; + } + + if (do_update) { + ret = __write_inode(trans, &u, iter->pos.snapshot); + if (ret) + bch_err(c, "error in fsck: error updating inode: %s", + bch2_err_str(ret)); + } +err: +fsck_err: + if (ret) + bch_err_fn(c, ret); + return ret; +} + +noinline_for_stack +int bch2_check_inodes(struct bch_fs *c) +{ + bool full = c->opts.fsck; + struct btree_trans trans; + struct btree_iter iter; + struct bch_inode_unpacked prev = { 0 }; + struct snapshots_seen s; + struct bkey_s_c k; + int ret; + + snapshots_seen_init(&s); + bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0); + + ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_inodes, + POS_MIN, + BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k, + NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, + check_inode(&trans, &iter, k, &prev, &s, full)); + + bch2_trans_exit(&trans); + snapshots_seen_exit(&s); + if (ret) + bch_err_fn(c, ret); + return ret; +} + +/* + * Checking for overlapping extents needs to be reimplemented + */ +#if 0 +static int fix_overlapping_extent(struct btree_trans *trans, + struct bkey_s_c k, struct bpos cut_at) +{ + struct btree_iter iter; + struct bkey_i *u; + int ret; + + u = bch2_trans_kmalloc(trans, bkey_bytes(k.k)); + ret = PTR_ERR_OR_ZERO(u); + if (ret) + return ret; + + bkey_reassemble(u, k); + bch2_cut_front(cut_at, u); + + + /* + * We don't want to go through the extent_handle_overwrites path: + * + * XXX: this is going to screw up disk accounting, extent triggers + * assume things about extent overwrites - we should be running the + * triggers manually here + */ + bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, u->k.p, + BTREE_ITER_INTENT|BTREE_ITER_NOT_EXTENTS); + + BUG_ON(iter.flags & BTREE_ITER_IS_EXTENTS); + ret = bch2_btree_iter_traverse(&iter) ?: + bch2_trans_update(trans, &iter, u, BTREE_TRIGGER_NORUN) ?: + bch2_trans_commit(trans, NULL, NULL, + BTREE_INSERT_NOFAIL| + BTREE_INSERT_LAZY_RW); + bch2_trans_iter_exit(trans, &iter); + return ret; +} +#endif + +static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans, + struct btree_iter *iter, + struct bpos pos) +{ + return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent); +} + +static bool inode_points_to_dirent(struct bch_inode_unpacked *inode, + struct bkey_s_c_dirent d) +{ + return inode->bi_dir == d.k->p.inode && + inode->bi_dir_offset == d.k->p.offset; +} + +static bool dirent_points_to_inode(struct bkey_s_c_dirent d, + struct bch_inode_unpacked *inode) +{ + return d.v->d_type == DT_SUBVOL + ? le32_to_cpu(d.v->d_child_subvol) == inode->bi_subvol + : le64_to_cpu(d.v->d_inum) == inode->bi_inum; +} + +static int inode_backpointer_exists(struct btree_trans *trans, + struct bch_inode_unpacked *inode, + u32 snapshot) +{ + struct btree_iter iter; + struct bkey_s_c_dirent d; + int ret; + + d = dirent_get_by_pos(trans, &iter, + SPOS(inode->bi_dir, inode->bi_dir_offset, snapshot)); + ret = bkey_err(d); + if (ret) + return bch2_err_matches(ret, ENOENT) ? 0 : ret; + + ret = dirent_points_to_inode(d, inode); + bch2_trans_iter_exit(trans, &iter); + return ret; +} + +static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w) +{ + struct bch_fs *c = trans->c; + struct inode_walker_entry *i; + u32 restart_count = trans->restart_count; + int ret = 0; + s64 count2; + + darray_for_each(w->inodes, i) { + if (i->inode.bi_sectors == i->count) + continue; + + count2 = bch2_count_inode_sectors(trans, w->cur_inum, i->snapshot); + + if (i->count != count2) { + bch_err(c, "fsck counted i_sectors wrong: got %llu should be %llu", + i->count, count2); + i->count = count2; + if (i->inode.bi_sectors == i->count) + continue; + } + + if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY), c, + "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu", + w->cur_inum, i->snapshot, + i->inode.bi_sectors, i->count)) { + i->inode.bi_sectors = i->count; + ret = write_inode(trans, &i->inode, i->snapshot); + if (ret) + break; + } + } +fsck_err: + if (ret) + bch_err_fn(c, ret); + if (!ret && trans_was_restarted(trans, restart_count)) + ret = -BCH_ERR_transaction_restart_nested; + return ret; +} + +struct extent_end { + u32 snapshot; + u64 offset; + struct snapshots_seen seen; +}; + +typedef DARRAY(struct extent_end) extent_ends; + +static int get_print_extent(struct btree_trans *trans, struct bpos pos, struct printbuf *out) +{ + struct btree_iter iter; + struct bkey_s_c k; + int ret; + + k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_extents, pos, + BTREE_ITER_SLOTS| + BTREE_ITER_ALL_SNAPSHOTS| + BTREE_ITER_NOT_EXTENTS); + ret = bkey_err(k); + if (ret) + return ret; - ret = walk_inode(&trans, &w, k.k->p.inode); + bch2_bkey_val_to_text(out, trans->c, k); + bch2_trans_iter_exit(trans, &iter); + return 0; +} + +static int check_overlapping_extents(struct btree_trans *trans, + struct snapshots_seen *seen, + extent_ends *extent_ends, + struct bkey_s_c k, + struct btree_iter *iter) +{ + struct bch_fs *c = trans->c; + struct extent_end *i; + struct printbuf buf = PRINTBUF; + int ret = 0; + + darray_for_each(*extent_ends, i) { + /* duplicate, due to transaction restart: */ + if (i->offset == k.k->p.offset && + i->snapshot == k.k->p.snapshot) + continue; + + if (!ref_visible2(c, + k.k->p.snapshot, seen, + i->snapshot, &i->seen)) + continue; + + if (i->offset <= bkey_start_offset(k.k)) + continue; + + printbuf_reset(&buf); + prt_str(&buf, "overlapping extents:\n "); + bch2_bkey_val_to_text(&buf, c, k); + prt_str(&buf, "\n "); + + ret = get_print_extent(trans, SPOS(k.k->p.inode, i->offset, i->snapshot), &buf); if (ret) break; - if (fsck_err_on(!w.have_inode, c, - "dirent in nonexisting directory:\n%s", - (bch2_bkey_val_to_text(&PBUF(buf), c, - k), buf)) || - fsck_err_on(!S_ISDIR(w.inode.bi_mode), c, - "dirent in non directory inode type %u:\n%s", - mode_to_type(w.inode.bi_mode), - (bch2_bkey_val_to_text(&PBUF(buf), c, - k), buf))) { - ret = bch2_btree_delete_at(&trans, iter, 0); + if (fsck_err(c, "%s", buf.buf)) { + struct bkey_i *update = bch2_trans_kmalloc(trans, bkey_bytes(k.k)); + if ((ret = PTR_ERR_OR_ZERO(update))) + goto err; + bkey_reassemble(update, k); + ret = bch2_trans_update_extent(trans, iter, update, + BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE); + if (ret) + goto err; + } + } +err: +fsck_err: + printbuf_exit(&buf); + return ret; +} + +static int extent_ends_at(extent_ends *extent_ends, + struct snapshots_seen *seen, + struct bkey_s_c k) +{ + struct extent_end *i, n = (struct extent_end) { + .snapshot = k.k->p.snapshot, + .offset = k.k->p.offset, + .seen = *seen, + }; + + n.seen.ids.data = kmemdup(seen->ids.data, + sizeof(seen->ids.data[0]) * seen->ids.size, + GFP_KERNEL); + if (!n.seen.ids.data) + return -BCH_ERR_ENOMEM_fsck_extent_ends_at; + + darray_for_each(*extent_ends, i) { + if (i->snapshot == k.k->p.snapshot) { + snapshots_seen_exit(&i->seen); + *i = n; + return 0; + } + + if (i->snapshot >= k.k->p.snapshot) + break; + } + + return darray_insert_item(extent_ends, i - extent_ends->data, n); +} + +static void extent_ends_reset(extent_ends *extent_ends) +{ + struct extent_end *i; + + darray_for_each(*extent_ends, i) + snapshots_seen_exit(&i->seen); + + extent_ends->nr = 0; +} + +static int check_extent(struct btree_trans *trans, struct btree_iter *iter, + struct bkey_s_c k, + struct inode_walker *inode, + struct snapshots_seen *s, + extent_ends *extent_ends) +{ + struct bch_fs *c = trans->c; + struct inode_walker_entry *i; + struct printbuf buf = PRINTBUF; + struct bpos equiv; + int ret = 0; + + ret = check_key_has_snapshot(trans, iter, k); + if (ret) { + ret = ret < 0 ? ret : 0; + goto out; + } + + equiv = k.k->p; + equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot); + + ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p); + if (ret) + goto err; + + if (k.k->type == KEY_TYPE_whiteout) + goto out; + + if (inode->cur_inum != k.k->p.inode) { + ret = check_i_sectors(trans, inode); + if (ret) + goto err; + + extent_ends_reset(extent_ends); + } + + BUG_ON(!iter->path->should_be_locked); + + ret = check_overlapping_extents(trans, s, extent_ends, k, iter); + if (ret) + goto err; + + ret = extent_ends_at(extent_ends, s, k); + if (ret) + goto err; + + i = walk_inode(trans, inode, equiv); + ret = PTR_ERR_OR_ZERO(i); + if (ret) + goto err; + + if (fsck_err_on(!i, c, + "extent in missing inode:\n %s", + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, k), buf.buf))) { + ret = bch2_btree_delete_at(trans, iter, + BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE); + goto out; + } + + if (!i) + goto out; + + if (fsck_err_on(!S_ISREG(i->inode.bi_mode) && + !S_ISLNK(i->inode.bi_mode), c, + "extent in non regular inode mode %o:\n %s", + i->inode.bi_mode, + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, k), buf.buf))) { + ret = bch2_btree_delete_at(trans, iter, + BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE); + goto out; + } + + /* + * Check inodes in reverse order, from oldest snapshots to newest, so + * that we emit the fewest number of whiteouts necessary: + */ + for (i = inode->inodes.data + inode->inodes.nr - 1; + i >= inode->inodes.data; + --i) { + if (i->snapshot > equiv.snapshot || + !key_visible_in_snapshot(c, s, i->snapshot, equiv.snapshot)) + continue; + + if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) && + k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 && + !bkey_extent_is_reservation(k), c, + "extent type past end of inode %llu:%u, i_size %llu\n %s", + i->inode.bi_inum, i->snapshot, i->inode.bi_size, + (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) { + struct btree_iter iter2; + + bch2_trans_copy_iter(&iter2, iter); + bch2_btree_iter_set_snapshot(&iter2, i->snapshot); + ret = bch2_btree_iter_traverse(&iter2) ?: + bch2_btree_delete_at(trans, &iter2, + BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE); + bch2_trans_iter_exit(trans, &iter2); if (ret) goto err; - continue; + + if (i->snapshot != equiv.snapshot) { + ret = snapshots_seen_add(c, s, i->snapshot); + if (ret) + goto err; + } } + } - if (w.first_this_inode && w.have_inode) - hash_check_set_inode(&trans, &h, &w.inode); + if (bkey_extent_is_allocation(k.k)) + for_each_visible_inode(c, s, inode, equiv.snapshot, i) + i->count += k.k->size; +#if 0 + bch2_bkey_buf_reassemble(&prev, c, k); +#endif - ret = check_dirent_hash(&trans, &h, iter, &k); - if (ret > 0) { - ret = 0; +out: +err: +fsck_err: + printbuf_exit(&buf); + + if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart)) + bch_err_fn(c, ret); + return ret; +} + +/* + * Walk extents: verify that extents have a corresponding S_ISREG inode, and + * that i_size an i_sectors are consistent + */ +int bch2_check_extents(struct bch_fs *c) +{ + struct inode_walker w = inode_walker_init(); + struct snapshots_seen s; + struct btree_trans trans; + struct btree_iter iter; + struct bkey_s_c k; + extent_ends extent_ends = { 0 }; + struct disk_reservation res = { 0 }; + int ret = 0; + + snapshots_seen_init(&s); + bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0); + + ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_extents, + POS(BCACHEFS_ROOT_INO, 0), + BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k, + &res, NULL, + BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, ({ + bch2_disk_reservation_put(c, &res); + check_extent(&trans, &iter, k, &w, &s, &extent_ends); + })); + + bch2_disk_reservation_put(c, &res); + extent_ends_reset(&extent_ends); + darray_exit(&extent_ends); + inode_walker_exit(&w); + bch2_trans_exit(&trans); + snapshots_seen_exit(&s); + + if (ret) + bch_err_fn(c, ret); + return ret; +} + +static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w) +{ + struct bch_fs *c = trans->c; + struct inode_walker_entry *i; + u32 restart_count = trans->restart_count; + int ret = 0; + s64 count2; + + darray_for_each(w->inodes, i) { + if (i->inode.bi_nlink == i->count) continue; + + count2 = bch2_count_subdirs(trans, w->cur_inum, i->snapshot); + if (count2 < 0) + return count2; + + if (i->count != count2) { + bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu", + i->count, count2); + i->count = count2; + if (i->inode.bi_nlink == i->count) + continue; } - if (ret) - goto fsck_err; + if (fsck_err_on(i->inode.bi_nlink != i->count, c, + "directory %llu:%u with wrong i_nlink: got %u, should be %llu", + w->cur_inum, i->snapshot, i->inode.bi_nlink, i->count)) { + i->inode.bi_nlink = i->count; + ret = write_inode(trans, &i->inode, i->snapshot); + if (ret) + break; + } + } +fsck_err: + if (ret) + bch_err_fn(c, ret); + if (!ret && trans_was_restarted(trans, restart_count)) + ret = -BCH_ERR_transaction_restart_nested; + return ret; +} + +static int check_dirent_target(struct btree_trans *trans, + struct btree_iter *iter, + struct bkey_s_c_dirent d, + struct bch_inode_unpacked *target, + u32 target_snapshot) +{ + struct bch_fs *c = trans->c; + struct bkey_i_dirent *n; + bool backpointer_exists = true; + struct printbuf buf = PRINTBUF; + int ret = 0; + + if (!target->bi_dir && + !target->bi_dir_offset) { + target->bi_dir = d.k->p.inode; + target->bi_dir_offset = d.k->p.offset; + + ret = __write_inode(trans, target, target_snapshot); if (ret) - goto fsck_err; + goto err; + } - if (k.k->type != KEY_TYPE_dirent) - continue; + if (!inode_points_to_dirent(target, d)) { + ret = inode_backpointer_exists(trans, target, d.k->p.snapshot); + if (ret < 0) + goto err; - d = bkey_s_c_to_dirent(k); - d_inum = le64_to_cpu(d.v->d_inum); - - name_len = bch2_dirent_name_bytes(d); - - if (fsck_err_on(!name_len, c, "empty dirent") || - fsck_err_on(name_len == 1 && - !memcmp(d.v->d_name, ".", 1), c, - ". dirent") || - fsck_err_on(name_len == 2 && - !memcmp(d.v->d_name, "..", 2), c, - ".. dirent") || - fsck_err_on(name_len == 2 && - !memcmp(d.v->d_name, "..", 2), c, - ".. dirent") || - fsck_err_on(memchr(d.v->d_name, '/', name_len), c, - "dirent name has invalid chars")) { - ret = remove_dirent(&trans, d); + backpointer_exists = ret; + ret = 0; + + if (fsck_err_on(S_ISDIR(target->bi_mode) && + backpointer_exists, c, + "directory %llu with multiple links", + target->bi_inum)) { + ret = __remove_dirent(trans, d.k->p); + goto out; + } + + if (fsck_err_on(backpointer_exists && + !target->bi_nlink, c, + "inode %llu type %s has multiple links but i_nlink 0", + target->bi_inum, bch2_d_types[d.v->d_type])) { + target->bi_nlink++; + target->bi_flags &= ~BCH_INODE_UNLINKED; + + ret = __write_inode(trans, target, target_snapshot); if (ret) goto err; - continue; } - if (fsck_err_on(d_inum == d.k->p.inode, c, - "dirent points to own directory:\n%s", - (bch2_bkey_val_to_text(&PBUF(buf), c, - k), buf))) { - ret = remove_dirent(&trans, d); + if (fsck_err_on(!backpointer_exists, c, + "inode %llu:%u has wrong backpointer:\n" + "got %llu:%llu\n" + "should be %llu:%llu", + target->bi_inum, target_snapshot, + target->bi_dir, + target->bi_dir_offset, + d.k->p.inode, + d.k->p.offset)) { + target->bi_dir = d.k->p.inode; + target->bi_dir_offset = d.k->p.offset; + + ret = __write_inode(trans, target, target_snapshot); if (ret) goto err; - continue; } + } - ret = bch2_inode_find_by_inum_trans(&trans, d_inum, &target); - if (ret && ret != -ENOENT) - break; + if (fsck_err_on(d.v->d_type != inode_d_type(target), c, + "incorrect d_type: got %s, should be %s:\n%s", + bch2_d_type_str(d.v->d_type), + bch2_d_type_str(inode_d_type(target)), + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) { + n = bch2_trans_kmalloc(trans, bkey_bytes(d.k)); + ret = PTR_ERR_OR_ZERO(n); + if (ret) + goto err; + + bkey_reassemble(&n->k_i, d.s_c); + n->v.d_type = inode_d_type(target); + + ret = bch2_trans_update(trans, iter, &n->k_i, 0); + if (ret) + goto err; + + d = dirent_i_to_s_c(n); + } + + if (d.v->d_type == DT_SUBVOL && + target->bi_parent_subvol != le32_to_cpu(d.v->d_parent_subvol) && + (c->sb.version < bcachefs_metadata_version_subvol_dirent || + fsck_err(c, "dirent has wrong d_parent_subvol field: got %u, should be %u", + le32_to_cpu(d.v->d_parent_subvol), + target->bi_parent_subvol))) { + n = bch2_trans_kmalloc(trans, bkey_bytes(d.k)); + ret = PTR_ERR_OR_ZERO(n); + if (ret) + goto err; + + bkey_reassemble(&n->k_i, d.s_c); + n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol); + + ret = bch2_trans_update(trans, iter, &n->k_i, 0); + if (ret) + goto err; + + d = dirent_i_to_s_c(n); + } +out: +err: +fsck_err: + printbuf_exit(&buf); + + if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart)) + bch_err_fn(c, ret); + return ret; +} + +static int check_dirent(struct btree_trans *trans, struct btree_iter *iter, + struct bkey_s_c k, + struct bch_hash_info *hash_info, + struct inode_walker *dir, + struct inode_walker *target, + struct snapshots_seen *s) +{ + struct bch_fs *c = trans->c; + struct bkey_s_c_dirent d; + struct inode_walker_entry *i; + struct printbuf buf = PRINTBUF; + struct bpos equiv; + int ret = 0; + + ret = check_key_has_snapshot(trans, iter, k); + if (ret) { + ret = ret < 0 ? ret : 0; + goto out; + } + + equiv = k.k->p; + equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot); + + ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p); + if (ret) + goto err; + + if (k.k->type == KEY_TYPE_whiteout) + goto out; + + if (dir->cur_inum != k.k->p.inode) { + ret = check_subdir_count(trans, dir); + if (ret) + goto err; + } + + BUG_ON(!iter->path->should_be_locked); + + i = walk_inode(trans, dir, equiv); + ret = PTR_ERR_OR_ZERO(i); + if (ret < 0) + goto err; + + if (dir->first_this_inode) + *hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode); + dir->first_this_inode = false; + + if (fsck_err_on(!i, c, + "dirent in nonexisting directory:\n%s", + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, k), buf.buf))) { + ret = bch2_btree_delete_at(trans, iter, + BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE); + goto out; + } - have_target = !ret; + if (!i) + goto out; + + if (fsck_err_on(!S_ISDIR(i->inode.bi_mode), c, + "dirent in non directory inode type %s:\n%s", + bch2_d_type_str(inode_d_type(&i->inode)), + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, k), buf.buf))) { + ret = bch2_btree_delete_at(trans, iter, 0); + goto out; + } + + ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k); + if (ret < 0) + goto err; + if (ret) { + /* dirent has been deleted */ ret = 0; + goto out; + } - if (fsck_err_on(!have_target, c, - "dirent points to missing inode:\n%s", - (bch2_bkey_val_to_text(&PBUF(buf), c, - k), buf))) { - ret = remove_dirent(&trans, d); - if (ret) - goto err; - continue; + if (k.k->type != KEY_TYPE_dirent) + goto out; + + d = bkey_s_c_to_dirent(k); + + if (d.v->d_type == DT_SUBVOL) { + struct bch_inode_unpacked subvol_root; + u32 target_subvol = le32_to_cpu(d.v->d_child_subvol); + u32 target_snapshot; + u64 target_inum; + + ret = __subvol_lookup(trans, target_subvol, + &target_snapshot, &target_inum); + if (ret && !bch2_err_matches(ret, ENOENT)) + goto err; + + if (fsck_err_on(ret, c, + "dirent points to missing subvolume %u", + le32_to_cpu(d.v->d_child_subvol))) { + ret = __remove_dirent(trans, d.k->p); + goto err; + } + + ret = __lookup_inode(trans, target_inum, + &subvol_root, &target_snapshot); + if (ret && !bch2_err_matches(ret, ENOENT)) + goto err; + + if (fsck_err_on(ret, c, + "subvolume %u points to missing subvolume root %llu", + target_subvol, + target_inum)) { + bch_err(c, "repair not implemented yet"); + ret = -EINVAL; + goto err; } - if (fsck_err_on(have_target && - d.v->d_type != - mode_to_type(target.bi_mode), c, - "incorrect d_type: should be %u:\n%s", - mode_to_type(target.bi_mode), - (bch2_bkey_val_to_text(&PBUF(buf), c, - k), buf))) { - struct bkey_i_dirent *n; - - n = kmalloc(bkey_bytes(d.k), GFP_KERNEL); - if (!n) { - ret = -ENOMEM; + if (fsck_err_on(subvol_root.bi_subvol != target_subvol, c, + "subvol root %llu has wrong bi_subvol field: got %u, should be %u", + target_inum, + subvol_root.bi_subvol, target_subvol)) { + subvol_root.bi_subvol = target_subvol; + ret = __write_inode(trans, &subvol_root, target_snapshot); + if (ret) goto err; - } + } - bkey_reassemble(&n->k_i, d.s_c); - n->v.d_type = mode_to_type(target.bi_mode); + ret = check_dirent_target(trans, iter, d, &subvol_root, + target_snapshot); + if (ret) + goto err; + } else { + ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum)); + if (ret) + goto err; - ret = __bch2_trans_do(&trans, NULL, NULL, - BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW, - (bch2_trans_update(&trans, iter, &n->k_i, 0), 0)); - kfree(n); + if (fsck_err_on(!target->inodes.nr, c, + "dirent points to missing inode: (equiv %u)\n%s", + equiv.snapshot, + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, k), + buf.buf))) { + ret = __remove_dirent(trans, d.k->p); if (ret) goto err; + } + darray_for_each(target->inodes, i) { + ret = check_dirent_target(trans, iter, d, + &i->inode, i->snapshot); + if (ret) + goto err; } } - hash_stop_chain(&trans, &h); + if (d.v->d_type == DT_DIR) + for_each_visible_inode(c, s, dir, equiv.snapshot, i) + i->count++; + +out: err: fsck_err: - if (ret == -EINTR) - goto retry; + printbuf_exit(&buf); - return bch2_trans_exit(&trans) ?: ret; + if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart)) + bch_err_fn(c, ret); + return ret; } /* - * Walk xattrs: verify that they all have a corresponding inode + * Walk dirents: verify that they all have a corresponding S_ISDIR inode, + * validate d_type */ -noinline_for_stack -static int check_xattrs(struct bch_fs *c) +int bch2_check_dirents(struct bch_fs *c) { - struct inode_walker w = inode_walker_init(); - struct hash_check h; + struct inode_walker dir = inode_walker_init(); + struct inode_walker target = inode_walker_init(); + struct snapshots_seen s; + struct bch_hash_info hash_info; struct btree_trans trans; - struct btree_iter *iter; + struct btree_iter iter; struct bkey_s_c k; int ret = 0; - bch_verbose(c, "checking xattrs"); - - hash_check_init(&h); - + snapshots_seen_init(&s); bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0); - iter = bch2_trans_get_iter(&trans, BTREE_ID_XATTRS, - POS(BCACHEFS_ROOT_INO, 0), 0); -retry: - for_each_btree_key_continue(iter, 0, k, ret) { - ret = walk_inode(&trans, &w, k.k->p.inode); - if (ret) - break; - - if (fsck_err_on(!w.have_inode, c, - "xattr for missing inode %llu", - k.k->p.inode)) { - ret = bch2_btree_delete_at(&trans, iter, 0); - if (ret) - goto err; - continue; - } + ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_dirents, + POS(BCACHEFS_ROOT_INO, 0), + BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, + k, + NULL, NULL, + BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, + check_dirent(&trans, &iter, k, &hash_info, &dir, &target, &s)); - if (w.first_this_inode && w.have_inode) - hash_check_set_inode(&trans, &h, &w.inode); + bch2_trans_exit(&trans); + snapshots_seen_exit(&s); + inode_walker_exit(&dir); + inode_walker_exit(&target); - ret = hash_check_key(&trans, bch2_xattr_hash_desc, - &h, iter, k); - if (ret) - goto fsck_err; - } -err: -fsck_err: - if (ret == -EINTR) - goto retry; - return bch2_trans_exit(&trans) ?: ret; + if (ret) + bch_err_fn(c, ret); + return ret; } -/* Get root directory, create if it doesn't exist: */ -static int check_root(struct bch_fs *c, struct bch_inode_unpacked *root_inode) +static int check_xattr(struct btree_trans *trans, struct btree_iter *iter, + struct bkey_s_c k, + struct bch_hash_info *hash_info, + struct inode_walker *inode) { - struct bkey_inode_buf packed; + struct bch_fs *c = trans->c; + struct inode_walker_entry *i; int ret; - bch_verbose(c, "checking root directory"); + ret = check_key_has_snapshot(trans, iter, k); + if (ret) + return ret; - ret = bch2_inode_find_by_inum(c, BCACHEFS_ROOT_INO, root_inode); - if (ret && ret != -ENOENT) + i = walk_inode(trans, inode, k.k->p); + ret = PTR_ERR_OR_ZERO(i); + if (ret) return ret; - if (fsck_err_on(ret, c, "root directory missing")) - goto create_root; + if (inode->first_this_inode) + *hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode); + inode->first_this_inode = false; - if (fsck_err_on(!S_ISDIR(root_inode->bi_mode), c, - "root inode not a directory")) - goto create_root; + if (fsck_err_on(!i, c, + "xattr for missing inode %llu", + k.k->p.inode)) + return bch2_btree_delete_at(trans, iter, 0); - return 0; + if (!i) + return 0; + + ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k); fsck_err: + if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart)) + bch_err_fn(c, ret); return ret; -create_root: - bch2_inode_init(c, root_inode, 0, 0, S_IFDIR|0755, - 0, NULL); - root_inode->bi_inum = BCACHEFS_ROOT_INO; +} + +/* + * Walk xattrs: verify that they all have a corresponding inode + */ +int bch2_check_xattrs(struct bch_fs *c) +{ + struct inode_walker inode = inode_walker_init(); + struct bch_hash_info hash_info; + struct btree_trans trans; + struct btree_iter iter; + struct bkey_s_c k; + int ret = 0; - bch2_inode_pack(&packed, root_inode); + bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0); + + ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_xattrs, + POS(BCACHEFS_ROOT_INO, 0), + BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, + k, + NULL, NULL, + BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, + check_xattr(&trans, &iter, k, &hash_info, &inode)); + + bch2_trans_exit(&trans); - return bch2_btree_insert(c, BTREE_ID_INODES, &packed.inode.k_i, - NULL, NULL, - BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW); + if (ret) + bch_err_fn(c, ret); + return ret; } -/* Get lost+found, create if it doesn't exist: */ -static int check_lostfound(struct bch_fs *c, - struct bch_inode_unpacked *root_inode, - struct bch_inode_unpacked *lostfound_inode) +static int check_root_trans(struct btree_trans *trans) { - struct qstr lostfound = QSTR("lost+found"); - struct bch_hash_info root_hash_info = - bch2_hash_info_init(c, root_inode); + struct bch_fs *c = trans->c; + struct bch_inode_unpacked root_inode; + u32 snapshot; u64 inum; int ret; - bch_verbose(c, "checking lost+found"); + ret = __subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum); + if (ret && !bch2_err_matches(ret, ENOENT)) + return ret; + + if (mustfix_fsck_err_on(ret, c, "root subvol missing")) { + struct bkey_i_subvolume root_subvol; + + snapshot = U32_MAX; + inum = BCACHEFS_ROOT_INO; + + bkey_subvolume_init(&root_subvol.k_i); + root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL; + root_subvol.v.flags = 0; + root_subvol.v.snapshot = cpu_to_le32(snapshot); + root_subvol.v.inode = cpu_to_le64(inum); + ret = commit_do(trans, NULL, NULL, + BTREE_INSERT_NOFAIL| + BTREE_INSERT_LAZY_RW, + __bch2_btree_insert(trans, BTREE_ID_subvolumes, + &root_subvol.k_i, 0)); + if (ret) { + bch_err(c, "error writing root subvol: %s", bch2_err_str(ret)); + goto err; + } - inum = bch2_dirent_lookup(c, BCACHEFS_ROOT_INO, &root_hash_info, - &lostfound); - if (!inum) { - bch_notice(c, "creating lost+found"); - goto create_lostfound; } - ret = bch2_inode_find_by_inum(c, inum, lostfound_inode); - if (ret && ret != -ENOENT) + ret = __lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot); + if (ret && !bch2_err_matches(ret, ENOENT)) return ret; - if (fsck_err_on(ret, c, "lost+found missing")) - goto create_lostfound; - - if (fsck_err_on(!S_ISDIR(lostfound_inode->bi_mode), c, - "lost+found inode not a directory")) - goto create_lostfound; + if (mustfix_fsck_err_on(ret, c, "root directory missing") || + mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode), c, + "root inode not a directory")) { + bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, + 0, NULL); + root_inode.bi_inum = inum; - return 0; + ret = __write_inode(trans, &root_inode, snapshot); + if (ret) + bch_err(c, "error writing root inode: %s", bch2_err_str(ret)); + } +err: fsck_err: return ret; -create_lostfound: - bch2_inode_init_early(c, lostfound_inode); +} + +/* Get root directory, create if it doesn't exist: */ +int bch2_check_root(struct bch_fs *c) +{ + int ret; ret = bch2_trans_do(c, NULL, NULL, - BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW, - bch2_create_trans(&trans, - BCACHEFS_ROOT_INO, root_inode, - lostfound_inode, &lostfound, - 0, 0, S_IFDIR|0700, 0, NULL, NULL)); - if (ret) - bch_err(c, "error creating lost+found: %i", ret); + BTREE_INSERT_NOFAIL| + BTREE_INSERT_LAZY_RW, + check_root_trans(&trans)); + if (ret) + bch_err_fn(c, ret); return ret; } -struct inode_bitmap { - unsigned long *bits; - size_t size; +struct pathbuf_entry { + u64 inum; + u32 snapshot; }; -static inline bool inode_bitmap_test(struct inode_bitmap *b, size_t nr) -{ - return nr < b->size ? test_bit(nr, b->bits) : false; -} +typedef DARRAY(struct pathbuf_entry) pathbuf; -static inline int inode_bitmap_set(struct inode_bitmap *b, size_t nr) +static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot) { - if (nr >= b->size) { - size_t new_size = max_t(size_t, max_t(size_t, - PAGE_SIZE * 8, - b->size * 2), - nr + 1); - void *n; - - new_size = roundup_pow_of_two(new_size); - n = krealloc(b->bits, new_size / 8, GFP_KERNEL|__GFP_ZERO); - if (!n) { - return -ENOMEM; - } + struct pathbuf_entry *i; - b->bits = n; - b->size = new_size; - } + darray_for_each(*p, i) + if (i->inum == inum && + i->snapshot == snapshot) + return true; - __set_bit(nr, b->bits); - return 0; + return false; } -struct pathbuf { - size_t nr; - size_t size; - - struct pathbuf_entry { - u64 inum; - u64 offset; - } *entries; -}; - -static int path_down(struct pathbuf *p, u64 inum) +static int path_down(struct bch_fs *c, pathbuf *p, + u64 inum, u32 snapshot) { - if (p->nr == p->size) { - size_t new_size = max_t(size_t, 256UL, p->size * 2); - void *n = krealloc(p->entries, - new_size * sizeof(p->entries[0]), - GFP_KERNEL); - if (!n) - return -ENOMEM; - - p->entries = n; - p->size = new_size; - }; + int ret = darray_push(p, ((struct pathbuf_entry) { + .inum = inum, + .snapshot = snapshot, + })); - p->entries[p->nr++] = (struct pathbuf_entry) { - .inum = inum, - .offset = 0, - }; - return 0; + if (ret) + bch_err(c, "fsck: error allocating memory for pathbuf, size %zu", + p->size); + return ret; } -noinline_for_stack -static int check_directory_structure(struct bch_fs *c, - struct bch_inode_unpacked *lostfound_inode) +/* + * Check that a given inode is reachable from the root: + * + * XXX: we should also be verifying that inodes are in the right subvolumes + */ +static int check_path(struct btree_trans *trans, + pathbuf *p, + struct bch_inode_unpacked *inode, + u32 snapshot) { - struct inode_bitmap dirs_done = { NULL, 0 }; - struct pathbuf path = { 0, 0, NULL }; - struct pathbuf_entry *e; - struct btree_trans trans; - struct btree_iter *iter; - struct bkey_s_c k; - struct bkey_s_c_dirent dirent; - bool had_unreachable; - u64 d_inum; + struct bch_fs *c = trans->c; int ret = 0; - bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0); + snapshot = bch2_snapshot_equiv(c, snapshot); + p->nr = 0; - bch_verbose(c, "checking directory structure"); + while (!(inode->bi_inum == BCACHEFS_ROOT_INO && + inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)) { + struct btree_iter dirent_iter; + struct bkey_s_c_dirent d; + u32 parent_snapshot = snapshot; - /* DFS: */ -restart_dfs: - had_unreachable = false; + if (inode->bi_subvol) { + u64 inum; - ret = inode_bitmap_set(&dirs_done, BCACHEFS_ROOT_INO); - if (ret) { - bch_err(c, "memory allocation failure in inode_bitmap_set()"); - goto err; - } + ret = subvol_lookup(trans, inode->bi_parent_subvol, + &parent_snapshot, &inum); + if (ret) + break; + } + + ret = lockrestart_do(trans, + PTR_ERR_OR_ZERO((d = dirent_get_by_pos(trans, &dirent_iter, + SPOS(inode->bi_dir, inode->bi_dir_offset, + parent_snapshot))).k)); + if (ret && !bch2_err_matches(ret, ENOENT)) + break; - ret = path_down(&path, BCACHEFS_ROOT_INO); - if (ret) - goto err; + if (!ret && !dirent_points_to_inode(d, inode)) { + bch2_trans_iter_exit(trans, &dirent_iter); + ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode; + } - while (path.nr) { -next: - e = &path.entries[path.nr - 1]; + if (bch2_err_matches(ret, ENOENT)) { + if (fsck_err(c, "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu", + inode->bi_inum, snapshot, + bch2_d_type_str(inode_d_type(inode)), + inode->bi_nlink, + inode->bi_dir, + inode->bi_dir_offset)) + ret = reattach_inode(trans, inode, snapshot); + break; + } - if (e->offset == U64_MAX) - goto up; + bch2_trans_iter_exit(trans, &dirent_iter); - for_each_btree_key(&trans, iter, BTREE_ID_DIRENTS, - POS(e->inum, e->offset + 1), 0, k, ret) { - if (k.k->p.inode != e->inum) - break; + if (!S_ISDIR(inode->bi_mode)) + break; - e->offset = k.k->p.offset; + ret = path_down(c, p, inode->bi_inum, snapshot); + if (ret) { + bch_err(c, "memory allocation failure"); + return ret; + } - if (k.k->type != KEY_TYPE_dirent) - continue; + snapshot = parent_snapshot; - dirent = bkey_s_c_to_dirent(k); + ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot); + if (ret) { + /* Should have been caught in dirents pass */ + bch_err(c, "error looking up parent directory: %i", ret); + break; + } - if (dirent.v->d_type != DT_DIR) - continue; + if (path_is_dup(p, inode->bi_inum, snapshot)) { + struct pathbuf_entry *i; - d_inum = le64_to_cpu(dirent.v->d_inum); + /* XXX print path */ + bch_err(c, "directory structure loop"); - if (fsck_err_on(inode_bitmap_test(&dirs_done, d_inum), c, - "directory %llu has multiple hardlinks", - d_inum)) { - ret = remove_dirent(&trans, dirent); - if (ret) - goto err; - continue; - } + darray_for_each(*p, i) + pr_err("%llu:%u", i->inum, i->snapshot); + pr_err("%llu:%u", inode->bi_inum, snapshot); - ret = inode_bitmap_set(&dirs_done, d_inum); - if (ret) { - bch_err(c, "memory allocation failure in inode_bitmap_set()"); - goto err; - } + if (!fsck_err(c, "directory structure loop")) + return 0; - ret = path_down(&path, d_inum); + ret = commit_do(trans, NULL, NULL, + BTREE_INSERT_NOFAIL| + BTREE_INSERT_LAZY_RW, + remove_backpointer(trans, inode)); if (ret) { - goto err; + bch_err(c, "error removing dirent: %i", ret); + break; } - ret = bch2_trans_iter_free(&trans, iter); - if (ret) { - bch_err(c, "btree error %i in fsck", ret); - goto err; - } - goto next; - } - ret = bch2_trans_iter_free(&trans, iter) ?: ret; - if (ret) { - bch_err(c, "btree error %i in fsck", ret); - goto err; + ret = reattach_inode(trans, inode, snapshot); } -up: - path.nr--; } +fsck_err: + if (ret) + bch_err_fn(c, ret); + return ret; +} - iter = bch2_trans_get_iter(&trans, BTREE_ID_INODES, POS_MIN, 0); -retry: - for_each_btree_key_continue(iter, 0, k, ret) { - if (k.k->type != KEY_TYPE_inode) - continue; +/* + * Check for unreachable inodes, as well as loops in the directory structure: + * After bch2_check_dirents(), if an inode backpointer doesn't exist that means it's + * unreachable: + */ +int bch2_check_directory_structure(struct bch_fs *c) +{ + struct btree_trans trans; + struct btree_iter iter; + struct bkey_s_c k; + struct bch_inode_unpacked u; + pathbuf path = { 0, }; + int ret; - if (!S_ISDIR(le16_to_cpu(bkey_s_c_to_inode(k).v->bi_mode))) - continue; + bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0); - ret = bch2_empty_dir_trans(&trans, k.k->p.inode); - if (ret == -EINTR) - goto retry; - if (!ret) + for_each_btree_key(&trans, iter, BTREE_ID_inodes, POS_MIN, + BTREE_ITER_INTENT| + BTREE_ITER_PREFETCH| + BTREE_ITER_ALL_SNAPSHOTS, k, ret) { + if (!bkey_is_inode(k.k)) continue; - if (fsck_err_on(!inode_bitmap_test(&dirs_done, k.k->p.inode), c, - "unreachable directory found (inum %llu)", - k.k->p.inode)) { - bch2_trans_unlock(&trans); + ret = bch2_inode_unpack(k, &u); + if (ret) { + /* Should have been caught earlier in fsck: */ + bch_err(c, "error unpacking inode %llu: %i", k.k->p.offset, ret); + break; + } - ret = reattach_inode(c, lostfound_inode, k.k->p.inode); - if (ret) { - goto err; - } + if (u.bi_flags & BCH_INODE_UNLINKED) + continue; - had_unreachable = true; - } + ret = check_path(&trans, &path, &u, iter.pos.snapshot); + if (ret) + break; } - bch2_trans_iter_free(&trans, iter); - if (ret) - goto err; + bch2_trans_iter_exit(&trans, &iter); + bch2_trans_exit(&trans); + darray_exit(&path); - if (had_unreachable) { - bch_info(c, "reattached unreachable directories, restarting pass to check for loops"); - kfree(dirs_done.bits); - kfree(path.entries); - memset(&dirs_done, 0, sizeof(dirs_done)); - memset(&path, 0, sizeof(path)); - goto restart_dfs; - } -err: -fsck_err: - ret = bch2_trans_exit(&trans) ?: ret; - kfree(dirs_done.bits); - kfree(path.entries); + if (ret) + bch_err_fn(c, ret); return ret; } -struct nlink { - u32 count; - u32 dir_count; +struct nlink_table { + size_t nr; + size_t size; + + struct nlink { + u64 inum; + u32 snapshot; + u32 count; + } *d; }; -typedef GENRADIX(struct nlink) nlink_table; +static int add_nlink(struct bch_fs *c, struct nlink_table *t, + u64 inum, u32 snapshot) +{ + if (t->nr == t->size) { + size_t new_size = max_t(size_t, 128UL, t->size * 2); + void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL); + + if (!d) { + bch_err(c, "fsck: error allocating memory for nlink_table, size %zu", + new_size); + return -BCH_ERR_ENOMEM_fsck_add_nlink; + } + + if (t->d) + memcpy(d, t->d, t->size * sizeof(t->d[0])); + kvfree(t->d); + + t->d = d; + t->size = new_size; + } + + + t->d[t->nr++] = (struct nlink) { + .inum = inum, + .snapshot = snapshot, + }; + + return 0; +} + +static int nlink_cmp(const void *_l, const void *_r) +{ + const struct nlink *l = _l; + const struct nlink *r = _r; + + return cmp_int(l->inum, r->inum) ?: cmp_int(l->snapshot, r->snapshot); +} -static void inc_link(struct bch_fs *c, nlink_table *links, - u64 range_start, u64 *range_end, - u64 inum, bool dir) +static void inc_link(struct bch_fs *c, struct snapshots_seen *s, + struct nlink_table *links, + u64 range_start, u64 range_end, u64 inum, u32 snapshot) { - struct nlink *link; + struct nlink *link, key = { + .inum = inum, .snapshot = U32_MAX, + }; - if (inum < range_start || inum >= *range_end) + if (inum < range_start || inum >= range_end) return; - link = genradix_ptr_alloc(links, inum - range_start, GFP_KERNEL); - if (!link) { - bch_verbose(c, "allocation failed during fsck - will need another pass"); - *range_end = inum; + link = __inline_bsearch(&key, links->d, links->nr, + sizeof(links->d[0]), nlink_cmp); + if (!link) return; - } - if (dir) - link->dir_count++; - else - link->count++; + while (link > links->d && link[0].inum == link[-1].inum) + --link; + + for (; link < links->d + links->nr && link->inum == inum; link++) + if (ref_visible(c, s, snapshot, link->snapshot)) { + link->count++; + if (link->snapshot >= snapshot) + break; + } } noinline_for_stack -static int bch2_gc_walk_dirents(struct bch_fs *c, nlink_table *links, - u64 range_start, u64 *range_end) +static int check_nlinks_find_hardlinks(struct bch_fs *c, + struct nlink_table *t, + u64 start, u64 *end) { struct btree_trans trans; - struct btree_iter *iter; + struct btree_iter iter; struct bkey_s_c k; - struct bkey_s_c_dirent d; - u64 d_inum; - int ret; + struct bch_inode_unpacked u; + int ret = 0; bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0); - inc_link(c, links, range_start, range_end, BCACHEFS_ROOT_INO, false); + for_each_btree_key(&trans, iter, BTREE_ID_inodes, + POS(0, start), + BTREE_ITER_INTENT| + BTREE_ITER_PREFETCH| + BTREE_ITER_ALL_SNAPSHOTS, k, ret) { + if (!bkey_is_inode(k.k)) + continue; - for_each_btree_key(&trans, iter, BTREE_ID_DIRENTS, POS_MIN, 0, k, ret) { - switch (k.k->type) { - case KEY_TYPE_dirent: - d = bkey_s_c_to_dirent(k); - d_inum = le64_to_cpu(d.v->d_inum); + /* Should never fail, checked by bch2_inode_invalid: */ + BUG_ON(bch2_inode_unpack(k, &u)); - if (d.v->d_type == DT_DIR) - inc_link(c, links, range_start, range_end, - d.k->p.inode, true); + /* + * Backpointer and directory structure checks are sufficient for + * directories, since they can't have hardlinks: + */ + if (S_ISDIR(u.bi_mode)) + continue; - inc_link(c, links, range_start, range_end, - d_inum, false); + if (!u.bi_nlink) + continue; + ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot); + if (ret) { + *end = k.k->p.offset; + ret = 0; break; } - bch2_trans_cond_resched(&trans); } - ret = bch2_trans_exit(&trans) ?: ret; + bch2_trans_iter_exit(&trans, &iter); + bch2_trans_exit(&trans); + if (ret) - bch_err(c, "error in fsck: btree error %i while walking dirents", ret); + bch_err(c, "error in fsck: btree error %i while walking inodes", ret); return ret; } -static int check_inode_nlink(struct bch_fs *c, - struct bch_inode_unpacked *lostfound_inode, - struct bch_inode_unpacked *u, - struct nlink *link, - bool *do_update) +noinline_for_stack +static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links, + u64 range_start, u64 range_end) { - u32 i_nlink = bch2_inode_nlink_get(u); - u32 real_i_nlink = - link->count * nlink_bias(u->bi_mode) + - link->dir_count; - int ret = 0; - - /* - * These should have been caught/fixed by earlier passes, we don't - * repair them here: - */ - if (S_ISDIR(u->bi_mode) && link->count > 1) { - need_fsck_err(c, "directory %llu with multiple hardlinks: %u", - u->bi_inum, link->count); - return 0; - } - - if (S_ISDIR(u->bi_mode) && !link->count) { - need_fsck_err(c, "unreachable directory found (inum %llu)", - u->bi_inum); - return 0; - } + struct btree_trans trans; + struct snapshots_seen s; + struct btree_iter iter; + struct bkey_s_c k; + struct bkey_s_c_dirent d; + int ret; - if (!S_ISDIR(u->bi_mode) && link->dir_count) { - need_fsck_err(c, "non directory with subdirectories", - u->bi_inum); - return 0; - } + snapshots_seen_init(&s); - if (!link->count && - !(u->bi_flags & BCH_INODE_UNLINKED) && - (c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) { - if (fsck_err(c, "unreachable inode %llu not marked as unlinked (type %u)", - u->bi_inum, mode_to_type(u->bi_mode)) == - FSCK_ERR_IGNORE) - return 0; + bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0); - ret = reattach_inode(c, lostfound_inode, u->bi_inum); + for_each_btree_key(&trans, iter, BTREE_ID_dirents, POS_MIN, + BTREE_ITER_INTENT| + BTREE_ITER_PREFETCH| + BTREE_ITER_ALL_SNAPSHOTS, k, ret) { + ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p); if (ret) - return ret; - - link->count = 1; - real_i_nlink = nlink_bias(u->bi_mode) + link->dir_count; - goto set_i_nlink; - } + break; - if (i_nlink < link->count) { - if (fsck_err(c, "inode %llu i_link too small (%u < %u, type %i)", - u->bi_inum, i_nlink, link->count, - mode_to_type(u->bi_mode)) == FSCK_ERR_IGNORE) - return 0; - goto set_i_nlink; - } + switch (k.k->type) { + case KEY_TYPE_dirent: + d = bkey_s_c_to_dirent(k); - if (i_nlink != real_i_nlink && - c->sb.clean) { - if (fsck_err(c, "filesystem marked clean, " - "but inode %llu has wrong i_nlink " - "(type %u i_nlink %u, should be %u)", - u->bi_inum, mode_to_type(u->bi_mode), - i_nlink, real_i_nlink) == FSCK_ERR_IGNORE) - return 0; - goto set_i_nlink; + if (d.v->d_type != DT_DIR && + d.v->d_type != DT_SUBVOL) + inc_link(c, &s, links, range_start, range_end, + le64_to_cpu(d.v->d_inum), + bch2_snapshot_equiv(c, d.k->p.snapshot)); + break; + } } + bch2_trans_iter_exit(&trans, &iter); - if (i_nlink != real_i_nlink && - (c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) { - if (fsck_err(c, "inode %llu has wrong i_nlink " - "(type %u i_nlink %u, should be %u)", - u->bi_inum, mode_to_type(u->bi_mode), - i_nlink, real_i_nlink) == FSCK_ERR_IGNORE) - return 0; - goto set_i_nlink; - } + if (ret) + bch_err(c, "error in fsck: btree error %i while walking dirents", ret); - if (real_i_nlink && i_nlink != real_i_nlink) - bch_verbose(c, "setting inode %llu nlink from %u to %u", - u->bi_inum, i_nlink, real_i_nlink); -set_i_nlink: - if (i_nlink != real_i_nlink) { - bch2_inode_nlink_set(u, real_i_nlink); - *do_update = true; - } -fsck_err: + bch2_trans_exit(&trans); + snapshots_seen_exit(&s); return ret; } -static int check_inode(struct btree_trans *trans, - struct bch_inode_unpacked *lostfound_inode, - struct btree_iter *iter, - struct bkey_s_c_inode inode, - struct nlink *link) +static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter, + struct bkey_s_c k, + struct nlink_table *links, + size_t *idx, u64 range_end) { struct bch_fs *c = trans->c; struct bch_inode_unpacked u; - bool do_update = false; + struct nlink *link = &links->d[*idx]; int ret = 0; - ret = bch2_inode_unpack(inode, &u); - - bch2_trans_unlock(trans); - - if (bch2_fs_inconsistent_on(ret, c, - "error unpacking inode %llu in fsck", - inode.k->p.inode)) - return ret; - - if (link) { - ret = check_inode_nlink(c, lostfound_inode, &u, link, - &do_update); - if (ret) - return ret; - } - - if (u.bi_flags & BCH_INODE_UNLINKED && - (!c->sb.clean || - fsck_err(c, "filesystem marked clean, but inode %llu unlinked", - u.bi_inum))) { - bch_verbose(c, "deleting inode %llu", u.bi_inum); - - ret = bch2_inode_rm(c, u.bi_inum); - if (ret) - bch_err(c, "error in fsck: error %i while deleting inode", ret); - return ret; - } - - if (u.bi_flags & BCH_INODE_I_SIZE_DIRTY && - (!c->sb.clean || - fsck_err(c, "filesystem marked clean, but inode %llu has i_size dirty", - u.bi_inum))) { - bch_verbose(c, "truncating inode %llu", u.bi_inum); - - /* - * XXX: need to truncate partial blocks too here - or ideally - * just switch units to bytes and that issue goes away - */ - - ret = bch2_inode_truncate(c, u.bi_inum, u.bi_size); - if (ret) { - bch_err(c, "error in fsck: error %i truncating inode", ret); - return ret; - } - - /* - * We truncated without our normal sector accounting hook, just - * make sure we recalculate it: - */ - u.bi_flags |= BCH_INODE_I_SECTORS_DIRTY; + if (k.k->p.offset >= range_end) + return 1; - u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY; - do_update = true; - } + if (!bkey_is_inode(k.k)) + return 0; - if (u.bi_flags & BCH_INODE_I_SECTORS_DIRTY && - (!c->sb.clean || - fsck_err(c, "filesystem marked clean, but inode %llu has i_sectors dirty", - u.bi_inum))) { - s64 sectors; + BUG_ON(bch2_inode_unpack(k, &u)); - bch_verbose(c, "recounting sectors for inode %llu", - u.bi_inum); + if (S_ISDIR(u.bi_mode)) + return 0; - sectors = bch2_count_inode_sectors(trans, u.bi_inum); - if (sectors < 0) { - bch_err(c, "error in fsck: error %i recounting inode sectors", - (int) sectors); - return sectors; - } + if (!u.bi_nlink) + return 0; - u.bi_sectors = sectors; - u.bi_flags &= ~BCH_INODE_I_SECTORS_DIRTY; - do_update = true; + while ((cmp_int(link->inum, k.k->p.offset) ?: + cmp_int(link->snapshot, k.k->p.snapshot)) < 0) { + BUG_ON(*idx == links->nr); + link = &links->d[++*idx]; } - if (do_update) { - struct bkey_inode_buf p; - - bch2_inode_pack(&p, &u); - - ret = __bch2_trans_do(trans, NULL, NULL, - BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW, - (bch2_trans_update(trans, iter, &p.inode.k_i, 0), 0)); - if (ret) - bch_err(c, "error in fsck: error %i " - "updating inode", ret); + if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count, c, + "inode %llu type %s has wrong i_nlink (%u, should be %u)", + u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)], + bch2_inode_nlink_get(&u), link->count)) { + bch2_inode_nlink_set(&u, link->count); + ret = __write_inode(trans, &u, k.k->p.snapshot); } fsck_err: return ret; } noinline_for_stack -static int bch2_gc_walk_inodes(struct bch_fs *c, - struct bch_inode_unpacked *lostfound_inode, - nlink_table *links, +static int check_nlinks_update_hardlinks(struct bch_fs *c, + struct nlink_table *links, u64 range_start, u64 range_end) { struct btree_trans trans; - struct btree_iter *iter; + struct btree_iter iter; struct bkey_s_c k; - struct nlink *link, zero_links = { 0, 0 }; - struct genradix_iter nlinks_iter; - int ret = 0, ret2 = 0; - u64 nlinks_pos; + size_t idx = 0; + int ret = 0; bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0); - iter = bch2_trans_get_iter(&trans, BTREE_ID_INODES, - POS(range_start, 0), 0); - nlinks_iter = genradix_iter_init(links, 0); - - while ((k = bch2_btree_iter_peek(iter)).k && - !(ret2 = bkey_err(k))) { -peek_nlinks: link = genradix_iter_peek(&nlinks_iter, links); - - if (!link && (!k.k || iter->pos.inode >= range_end)) - break; - - nlinks_pos = range_start + nlinks_iter.pos; - if (iter->pos.inode > nlinks_pos) { - /* Should have been caught by dirents pass: */ - need_fsck_err_on(link && link->count, c, - "missing inode %llu (nlink %u)", - nlinks_pos, link->count); - genradix_iter_advance(&nlinks_iter, links); - goto peek_nlinks; - } - - if (iter->pos.inode < nlinks_pos || !link) - link = &zero_links; + ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_inodes, + POS(0, range_start), + BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k, + NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, + check_nlinks_update_inode(&trans, &iter, k, links, &idx, range_end)); - if (k.k && k.k->type == KEY_TYPE_inode) { - ret = check_inode(&trans, lostfound_inode, iter, - bkey_s_c_to_inode(k), link); - BUG_ON(ret == -EINTR); - if (ret) - break; - } else { - /* Should have been caught by dirents pass: */ - need_fsck_err_on(link->count, c, - "missing inode %llu (nlink %u)", - nlinks_pos, link->count); - } - - if (nlinks_pos == iter->pos.inode) - genradix_iter_advance(&nlinks_iter, links); - - bch2_btree_iter_next(iter); - bch2_trans_cond_resched(&trans); - } -fsck_err: bch2_trans_exit(&trans); - if (ret2) - bch_err(c, "error in fsck: btree error %i while walking inodes", ret2); + if (ret < 0) { + bch_err(c, "error in fsck: btree error %i while walking inodes", ret); + return ret; + } - return ret ?: ret2; + return 0; } -noinline_for_stack -static int check_inode_nlinks(struct bch_fs *c, - struct bch_inode_unpacked *lostfound_inode) +int bch2_check_nlinks(struct bch_fs *c) { - nlink_table links; + struct nlink_table links = { 0 }; u64 this_iter_range_start, next_iter_range_start = 0; int ret = 0; - bch_verbose(c, "checking inode nlinks"); - - genradix_init(&links); - do { this_iter_range_start = next_iter_range_start; next_iter_range_start = U64_MAX; - ret = bch2_gc_walk_dirents(c, &links, + ret = check_nlinks_find_hardlinks(c, &links, + this_iter_range_start, + &next_iter_range_start); + + ret = check_nlinks_walk_dirents(c, &links, this_iter_range_start, - &next_iter_range_start); + next_iter_range_start); if (ret) break; - ret = bch2_gc_walk_inodes(c, lostfound_inode, &links, + ret = check_nlinks_update_hardlinks(c, &links, this_iter_range_start, next_iter_range_start); if (ret) break; - genradix_free(&links); + links.nr = 0; } while (next_iter_range_start != U64_MAX); - genradix_free(&links); + kvfree(links.d); + if (ret) + bch_err_fn(c, ret); return ret; } -/* - * Checks for inconsistencies that shouldn't happen, unless we have a bug. - * Doesn't fix them yet, mainly because they haven't yet been observed: - */ -int bch2_fsck_full(struct bch_fs *c) +static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter, + struct bkey_s_c k) { - struct bch_inode_unpacked root_inode, lostfound_inode; + struct bkey_s_c_reflink_p p; + struct bkey_i_reflink_p *u; + int ret; - return check_extents(c) ?: - check_dirents(c) ?: - check_xattrs(c) ?: - check_root(c, &root_inode) ?: - check_lostfound(c, &root_inode, &lostfound_inode) ?: - check_directory_structure(c, &lostfound_inode) ?: - check_inode_nlinks(c, &lostfound_inode); -} + if (k.k->type != KEY_TYPE_reflink_p) + return 0; -int bch2_fsck_inode_nlink(struct bch_fs *c) -{ - struct bch_inode_unpacked root_inode, lostfound_inode; + p = bkey_s_c_to_reflink_p(k); + + if (!p.v->front_pad && !p.v->back_pad) + return 0; + + u = bch2_trans_kmalloc(trans, sizeof(*u)); + ret = PTR_ERR_OR_ZERO(u); + if (ret) + return ret; + + bkey_reassemble(&u->k_i, k); + u->v.front_pad = 0; + u->v.back_pad = 0; - return check_root(c, &root_inode) ?: - check_lostfound(c, &root_inode, &lostfound_inode) ?: - check_inode_nlinks(c, &lostfound_inode); + return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN); } -int bch2_fsck_walk_inodes_only(struct bch_fs *c) +int bch2_fix_reflink_p(struct bch_fs *c) { - struct btree_trans trans; - struct btree_iter *iter; + struct btree_iter iter; struct bkey_s_c k; - struct bkey_s_c_inode inode; int ret; - bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0); - - for_each_btree_key(&trans, iter, BTREE_ID_INODES, POS_MIN, 0, k, ret) { - if (k.k->type != KEY_TYPE_inode) - continue; - - inode = bkey_s_c_to_inode(k); + if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix) + return 0; - if (inode.v->bi_flags & - (BCH_INODE_I_SIZE_DIRTY| - BCH_INODE_I_SECTORS_DIRTY| - BCH_INODE_UNLINKED)) { - ret = check_inode(&trans, NULL, iter, inode, NULL); - BUG_ON(ret == -EINTR); - if (ret) - break; - } - } - BUG_ON(ret == -EINTR); + ret = bch2_trans_run(c, + for_each_btree_key_commit(&trans, iter, + BTREE_ID_extents, POS_MIN, + BTREE_ITER_INTENT|BTREE_ITER_PREFETCH| + BTREE_ITER_ALL_SNAPSHOTS, k, + NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW, + fix_reflink_p_key(&trans, &iter, k))); - return bch2_trans_exit(&trans) ?: ret; + if (ret) + bch_err_fn(c, ret); + return ret; }