bcachefs: Add missing restart handling to check_topology()

The next patch will add logging of the specific error being corrected in
repair paths to the journal; this means __bch2_fsck_err() can return
transaction restarts in places that previously weren't expecting them.

check_topology() is old code that doesn't use btree iterators for btree
node locking - it'll have to be rewritten in the future to work online.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This commit is contained in:
Kent Overstreet 2025-06-01 20:22:17 -04:00
parent 19272b37aa
commit 47fe65b105

View File

@ -397,7 +397,11 @@ again:
continue;
}
ret = btree_check_node_boundaries(trans, b, prev, cur, pulled_from_scan);
ret = lockrestart_do(trans,
btree_check_node_boundaries(trans, b, prev, cur, pulled_from_scan));
if (ret < 0)
goto err;
if (ret == DID_FILL_FROM_SCAN) {
new_pass = true;
ret = 0;
@ -438,7 +442,8 @@ again:
if (!ret && !IS_ERR_OR_NULL(prev)) {
BUG_ON(cur);
ret = btree_repair_node_end(trans, b, prev, pulled_from_scan);
ret = lockrestart_do(trans,
btree_repair_node_end(trans, b, prev, pulled_from_scan));
if (ret == DID_FILL_FROM_SCAN) {
new_pass = true;
ret = 0;
@ -519,6 +524,46 @@ fsck_err:
bch2_bkey_buf_exit(&prev_k, c);
bch2_bkey_buf_exit(&cur_k, c);
printbuf_exit(&buf);
bch_err_fn(c, ret);
return ret;
}
static int bch2_check_root(struct btree_trans *trans, enum btree_id i,
bool *reconstructed_root)
{
struct bch_fs *c = trans->c;
struct btree_root *r = bch2_btree_id_root(c, i);
struct printbuf buf = PRINTBUF;
int ret = 0;
bch2_btree_id_to_text(&buf, i);
if (r->error) {
bch_info(c, "btree root %s unreadable, must recover from scan", buf.buf);
r->alive = false;
r->error = 0;
if (!bch2_btree_has_scanned_nodes(c, i)) {
__fsck_err(trans,
FSCK_CAN_FIX|(!btree_id_important(i) ? FSCK_AUTOFIX : 0),
btree_root_unreadable_and_scan_found_nothing,
"no nodes found for btree %s, continue?", buf.buf);
bch2_btree_root_alloc_fake_trans(trans, i, 0);
} else {
bch2_btree_root_alloc_fake_trans(trans, i, 1);
bch2_shoot_down_journal_keys(c, i, 1, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
ret = bch2_get_scanned_nodes(c, i, 0, POS_MIN, SPOS_MAX);
if (ret)
goto err;
}
*reconstructed_root = true;
}
err:
fsck_err:
printbuf_exit(&buf);
bch_err_fn(c, ret);
return ret;
}
@ -526,42 +571,18 @@ int bch2_check_topology(struct bch_fs *c)
{
struct btree_trans *trans = bch2_trans_get(c);
struct bpos pulled_from_scan = POS_MIN;
struct printbuf buf = PRINTBUF;
int ret = 0;
bch2_trans_srcu_unlock(trans);
for (unsigned i = 0; i < btree_id_nr_alive(c) && !ret; i++) {
struct btree_root *r = bch2_btree_id_root(c, i);
bool reconstructed_root = false;
recover:
ret = lockrestart_do(trans, bch2_check_root(trans, i, &reconstructed_root));
if (ret)
break;
printbuf_reset(&buf);
bch2_btree_id_to_text(&buf, i);
if (r->error) {
reconstruct_root:
bch_info(c, "btree root %s unreadable, must recover from scan", buf.buf);
r->alive = false;
r->error = 0;
if (!bch2_btree_has_scanned_nodes(c, i)) {
__fsck_err(trans,
FSCK_CAN_FIX|(!btree_id_important(i) ? FSCK_AUTOFIX : 0),
btree_root_unreadable_and_scan_found_nothing,
"no nodes found for btree %s, continue?", buf.buf);
bch2_btree_root_alloc_fake_trans(trans, i, 0);
} else {
bch2_btree_root_alloc_fake_trans(trans, i, 1);
bch2_shoot_down_journal_keys(c, i, 1, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
ret = bch2_get_scanned_nodes(c, i, 0, POS_MIN, SPOS_MAX);
if (ret)
break;
}
reconstructed_root = true;
}
struct btree_root *r = bch2_btree_id_root(c, i);
struct btree *b = r->b;
btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read);
@ -575,17 +596,21 @@ reconstruct_root:
r->b = NULL;
if (!reconstructed_root)
goto reconstruct_root;
if (!reconstructed_root) {
r->error = -EIO;
goto recover;
}
struct printbuf buf = PRINTBUF;
bch2_btree_id_to_text(&buf, i);
bch_err(c, "empty btree root %s", buf.buf);
printbuf_exit(&buf);
bch2_btree_root_alloc_fake_trans(trans, i, 0);
r->alive = false;
ret = 0;
}
}
fsck_err:
printbuf_exit(&buf);
bch2_trans_put(trans);
return ret;
}