+
+ 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 &&
+ (!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 = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
+ if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
+ 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, "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_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, "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, "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);
+ if (ret) {
+ bch_err_msg(c, ret, "in fsck updating inode");
+ return ret;
+ }
+ }
+err:
+fsck_err:
+ if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
+ 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 = bch2_trans_get(c);
+ 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);
+
+ 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));
+
+ snapshots_seen_exit(&s);
+ bch2_trans_put(trans);
+ if (ret)
+ 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 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->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 %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:
+ if (ret)
+ bch_err_fn(c, ret);
+ return ret ?: trans_was_restarted(trans, restart_count);
+}
+
+struct extent_end {
+ u32 snapshot;
+ u64 offset;
+ struct snapshots_seen seen;
+};
+
+struct extent_ends {
+ struct bpos last_pos;
+ DARRAY(struct extent_end) e;
+};
+
+static void extent_ends_reset(struct extent_ends *extent_ends)
+{
+ struct extent_end *i;
+
+ darray_for_each(extent_ends->e, i)
+ snapshots_seen_exit(&i->seen);
+
+ extent_ends->e.nr = 0;
+}
+
+static void extent_ends_exit(struct extent_ends *extent_ends)
+{
+ extent_ends_reset(extent_ends);
+ darray_exit(&extent_ends->e);
+}
+
+static void extent_ends_init(struct extent_ends *extent_ends)
+{
+ memset(extent_ends, 0, sizeof(*extent_ends));
+}
+
+static int extent_ends_at(struct bch_fs *c,
+ struct extent_ends *extent_ends,
+ struct snapshots_seen *seen,
+ struct bkey_s_c k)
+{
+ struct extent_end *i, n = (struct extent_end) {
+ .offset = k.k->p.offset,
+ .snapshot = k.k->p.snapshot,
+ .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->e, 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->e, i - extent_ends->e.data, n);
+}
+
+static int overlapping_extents_found(struct btree_trans *trans,
+ enum btree_id btree,
+ struct bpos pos1, struct snapshots_seen *pos1_seen,
+ struct bkey pos2,
+ bool *fixed,
+ struct extent_end *extent_end)
+{
+ struct bch_fs *c = trans->c;
+ struct printbuf buf = PRINTBUF;
+ struct btree_iter iter1, iter2 = { NULL };
+ struct bkey_s_c k1, k2;
+ int ret;
+
+ BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
+
+ bch2_trans_iter_init(trans, &iter1, btree, pos1,
+ BTREE_ITER_ALL_SNAPSHOTS|
+ BTREE_ITER_NOT_EXTENTS);
+ k1 = bch2_btree_iter_peek_upto(&iter1, POS(pos1.inode, U64_MAX));
+ ret = bkey_err(k1);
+ if (ret)
+ goto err;
+
+ prt_str(&buf, "\n ");
+ bch2_bkey_val_to_text(&buf, c, k1);
+
+ if (!bpos_eq(pos1, k1.k->p)) {
+ prt_str(&buf, "\n wanted\n ");
+ bch2_bpos_to_text(&buf, pos1);
+ prt_str(&buf, "\n ");
+ bch2_bkey_to_text(&buf, &pos2);
+
+ bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
+ __func__, buf.buf);
+ ret = -BCH_ERR_internal_fsck_err;
+ goto err;
+ }
+
+ bch2_trans_copy_iter(&iter2, &iter1);
+
+ while (1) {
+ bch2_btree_iter_advance(&iter2);
+
+ k2 = bch2_btree_iter_peek_upto(&iter2, POS(pos1.inode, U64_MAX));
+ ret = bkey_err(k2);
+ if (ret)
+ goto err;
+
+ if (bpos_ge(k2.k->p, pos2.p))
+ break;
+ }
+
+ prt_str(&buf, "\n ");
+ bch2_bkey_val_to_text(&buf, c, k2);
+
+ if (bpos_gt(k2.k->p, pos2.p) ||
+ pos2.size != k2.k->size) {
+ bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
+ __func__, buf.buf);
+ ret = -BCH_ERR_internal_fsck_err;
+ goto err;
+ }
+
+ prt_printf(&buf, "\n overwriting %s extent",
+ pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
+
+ if (fsck_err(c, "overlapping extents%s", buf.buf)) {
+ struct btree_iter *old_iter = &iter1;
+ struct disk_reservation res = { 0 };
+
+ if (pos1.snapshot < pos2.p.snapshot) {
+ old_iter = &iter2;
+ swap(k1, k2);
+ }
+
+ trans->extra_journal_res += bch2_bkey_sectors_compressed(k2);
+
+ ret = bch2_trans_update_extent_overwrite(trans, old_iter,
+ BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE,
+ k1, k2) ?:
+ bch2_trans_commit(trans, &res, NULL,
+ BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL);
+ bch2_disk_reservation_put(c, &res);
+
+ if (ret)
+ goto err;
+
+ *fixed = true;
+
+ if (pos1.snapshot == pos2.p.snapshot) {
+ /*
+ * We overwrote the first extent, and did the overwrite
+ * in the same snapshot:
+ */
+ extent_end->offset = bkey_start_offset(&pos2);
+ } else if (pos1.snapshot > pos2.p.snapshot) {
+ /*
+ * We overwrote the first extent in pos2's snapshot:
+ */
+ ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
+ } else {
+ /*
+ * We overwrote the second extent - restart
+ * check_extent() from the top:
+ */
+ ret = -BCH_ERR_transaction_restart_nested;
+ }
+ }
+fsck_err:
+err:
+ bch2_trans_iter_exit(trans, &iter2);
+ bch2_trans_iter_exit(trans, &iter1);
+ printbuf_exit(&buf);
+ return ret;
+}
+
+static int check_overlapping_extents(struct btree_trans *trans,
+ struct snapshots_seen *seen,
+ struct extent_ends *extent_ends,
+ struct bkey_s_c k,
+ u32 equiv,
+ struct btree_iter *iter,
+ bool *fixed)
+{
+ struct bch_fs *c = trans->c;
+ struct extent_end *i;
+ int ret = 0;
+
+ /* transaction restart, running again */
+ if (bpos_eq(extent_ends->last_pos, k.k->p))
+ return 0;
+
+ if (extent_ends->last_pos.inode != k.k->p.inode)
+ extent_ends_reset(extent_ends);
+
+ darray_for_each(extent_ends->e, i) {
+ if (i->offset <= bkey_start_offset(k.k))
+ continue;
+
+ if (!ref_visible2(c,
+ k.k->p.snapshot, seen,
+ i->snapshot, &i->seen))
+ continue;
+
+ ret = overlapping_extents_found(trans, iter->btree_id,
+ SPOS(iter->pos.inode,
+ i->offset,
+ i->snapshot),
+ &i->seen,
+ *k.k, fixed, i);
+ if (ret)
+ goto err;
+ }
+
+ ret = extent_ends_at(c, extent_ends, seen, k);
+ if (ret)
+ goto err;
+
+ extent_ends->last_pos = k.k->p;
+err:
+ return ret;
+}
+
+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,
+ struct extent_ends *extent_ends)
+{
+ struct bch_fs *c = trans->c;
+ struct inode_walker_entry *i;
+ struct printbuf buf = PRINTBUF;
+ struct bpos equiv = k.k->p;
+ int ret = 0;
+
+ equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
+
+ ret = check_key_has_snapshot(trans, iter, k);
+ if (ret) {
+ ret = ret < 0 ? ret : 0;
+ goto out;
+ }
+
+ if (inode->last_pos.inode != k.k->p.inode) {
+ ret = check_i_sectors(trans, inode);