Fedora Account System
Red Hat Associate
Red Hat Customer
A flaw was found in corosync. The membership commit token handler in exec/totemsrp.c copies a received memb_commit_token into a fixed-size heap buffer (commit_token_storage[40000]) using the full received message length (msg_len). check_memb_commit_token_sanity() only validates lower bounds (minimum header size and space for addr_entries) and caps addr_entries at PROCESSOR_COUNT_MAX; it does not reject messages larger than commit_token_storage. A crafted commit token with a valid small address list and trailing padding above 40000 bytes can therefore overflow the heap-allocated totemsrp_instance when a node is in MEMB_STATE_GATHER and the token matches membership and ring-sequence gates. Over knet, messages can be up to the default 64 KiB netmtu, allowing roughly 24 KiB of attacker-controlled out-of-bounds write. This is distinct from CVE-2026-35091 (incorrect success return on truncated messages) and CVE-2026-81666 (32-bit integer overflow in required_len). Impact is heap memory corruption and denial of service of the Corosync daemon; code execution was not demonstrated. Reported via PSIRTSUPT-30664 by Adam Bedard. Upstream maintainer Jan Friesse confirmed the issue and provided a proposed fix that adds an upper-bound length check and moves alloca after the sanity check.