+static struct inode_walker_entry *
+lookup_inode_for_snapshot(struct bch_fs *c, struct inode_walker *w,
+ u32 snapshot, bool is_whiteout)
+{
+ 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 && !is_whiteout) {
+ struct inode_walker_entry new = *i;
+ size_t pos;
+ 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->last_pos.inode, snapshot, i->snapshot);
+
+ while (i > w->inodes.data && i[-1].snapshot > snapshot)
+ --i;
+
+ pos = i - w->inodes.data;
+ ret = darray_insert_item(&w->inodes, pos, new);
+ if (ret)
+ return ERR_PTR(ret);
+
+ i = w->inodes.data + pos;
+ }
+
+ return i;
+}
+
+static struct inode_walker_entry *walk_inode(struct btree_trans *trans,
+ struct inode_walker *w, struct bpos pos,
+ bool is_whiteout)
+{
+ if (w->last_pos.inode != pos.inode) {
+ int ret = get_inodes_all_snapshots(trans, w, pos.inode);
+ if (ret)
+ return ERR_PTR(ret);
+ } else if (bkey_cmp(w->last_pos, pos)) {
+ darray_for_each(w->inodes, i)
+ i->seen_this_pos = false;
+ }
+
+ w->last_pos = pos;
+
+ return lookup_inode_for_snapshot(trans->c, w, pos.snapshot, is_whiteout);
+}
+
+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,
+ bkey_in_missing_snapshot,
+ "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, BCH_TRANS_COMMIT_no_enospc);
+}
+
+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,
+ hash_table_key_duplicate,
+ "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_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,
+ (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);
+ bch_err_fn(c, ret);
+ if (ret)
+ return ret;
+ ret = -BCH_ERR_transaction_restart_nested;
+ }
+fsck_err:
+ goto out;
+}
+
+static int check_inode_deleted_list(struct btree_trans *trans, struct bpos p)
+{
+ struct btree_iter iter;
+ struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_deleted_inodes, p, 0);
+ int ret = bkey_err(k);
+ if (ret)
+ return ret;
+
+ bch2_trans_iter_exit(trans, &iter);
+ return k.k->type == KEY_TYPE_set;
+}
+
+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 (!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, inode_snapshot_mismatch,
+ "inodes in different snapshots don't match")) {
+ bch_err(c, "repair not implemented yet");
+ return -BCH_ERR_fsck_repair_unimplemented;
+ }
+
+ if ((u.bi_flags & (BCH_INODE_i_size_dirty|BCH_INODE_unlinked)) &&
+ bch2_key_has_snapshot_overwrites(trans, BTREE_ID_inodes, k.k->p)) {
+ struct bpos new_min_pos;
+
+ ret = bch2_propagate_key_to_snapshot_leaves(trans, iter->btree_id, k, &new_min_pos);
+ if (ret)
+ goto err;
+
+ u.bi_flags &= ~BCH_INODE_i_size_dirty|BCH_INODE_unlinked;
+
+ ret = __write_inode(trans, &u, iter->pos.snapshot);
+ bch_err_msg(c, ret, "in fsck updating inode");
+ if (ret)
+ return ret;
+
+ if (!bpos_eq(new_min_pos, POS_MIN))
+ bch2_btree_iter_set_pos(iter, bpos_predecessor(new_min_pos));
+ return 0;
+ }
+
+ if (u.bi_flags & BCH_INODE_unlinked) {
+ ret = check_inode_deleted_list(trans, k.k->p);
+ if (ret < 0)
+ return ret;
+
+ fsck_err_on(ret, c, unlinked_inode_not_on_deleted_list,
+ "inode %llu:%u unlinked, but not on deleted list",
+ u.bi_inum, k.k->p.snapshot);
+ ret = 0;
+ }
+
+ if (u.bi_flags & BCH_INODE_unlinked &&
+ (!c->sb.clean ||
+ fsck_err(c, inode_unlinked_but_clean,
+ "filesystem marked clean, but inode %llu unlinked",
+ u.bi_inum))) {
+ ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
+ bch_err_msg(c, ret, "in fsck deleting inode");
+ return ret;
+ }
+
+ if (u.bi_flags & BCH_INODE_i_size_dirty &&
+ (!c->sb.clean ||
+ fsck_err(c, inode_i_size_dirty_but_clean,
+ "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_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);
+ bch_err_msg(c, ret, "in fsck truncating inode");
+ 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, inode_i_sectors_dirty_but_clean,
+ "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_msg(c, sectors, "in fsck recounting inode 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);
+ bch_err_msg(c, ret, "in fsck updating inode");
+ if (ret)
+ return ret;
+ }
+err:
+fsck_err:
+ bch_err_fn(c, ret);
+ return ret;
+}
+
+int bch2_check_inodes(struct bch_fs *c)
+{
+ bool full = c->opts.fsck;
+ struct bch_inode_unpacked prev = { 0 };
+ struct snapshots_seen s;
+
+ snapshots_seen_init(&s);
+
+ int ret = bch2_trans_run(c,
+ for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
+ POS_MIN,
+ BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
+ NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+ check_inode(trans, &iter, k, &prev, &s, full)));
+
+ snapshots_seen_exit(&s);
+ bch_err_fn(c, ret);
+ return ret;
+}
+
+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 check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
+{
+ struct bch_fs *c = trans->c;
+ 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->last_pos.inode, i->snapshot);
+
+ if (w->recalculate_sums)
+ i->count = count2;
+
+ if (i->count != count2) {
+ bch_err(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
+ w->last_pos.inode, i->snapshot, i->count, count2);
+ return -BCH_ERR_internal_fsck_err;
+ }
+
+ if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_sectors_dirty),
+ c, inode_i_sectors_wrong,
+ "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
+ w->last_pos.inode, i->snapshot,
+ i->inode.bi_sectors, i->count)) {
+ i->inode.bi_sectors = i->count;
+ ret = fsck_write_inode(trans, &i->inode, i->snapshot);
+ if (ret)
+ break;
+ }
+ }
+fsck_err:
+ bch_err_fn(c, ret);
+ return ret ?: trans_was_restarted(trans, restart_count);
+}
+
+struct extent_end {
+ u32 snapshot;
+ u64 offset;
+ struct snapshots_seen seen;