REDHAT-BUG-2576161: Medium severity libtirpc libtirpc vulnerability
A flaw was found in libtirpc. The SunRPC UDP/datagram server transport (svcdgrecv / svcudpcreate) reuses a single XDR decode stream sized to the allocated receive buffer. On each datagram, the receive path resets the stream position and calls xdrcallmsg, but it does not rebind the decode bound (xhandy) to the actual received length rlen. Because the XDR memory backend uses xhandy as the remaining length, a shorter request can be decoded together with stale bytes left by a prior datagram. Florian Weimer reproduced this with tst-udp-stale: a full request leaves argument bytes at the tail of the buffer after a shorter reply is encoded; a follow-up truncated AUTHNONE call with no arguments then has svcgetargs consume those leftover bytes and echo them back. A correctly bounded decoder would fail with GARBAGEARGS. This is a logical datagram-boundary violation, not a read past the allocated buffer. The same inherited SunRPC pattern exists in glibc sunrpc/svcudp.c (legacy ABI compatibility; glibc is expected to assign a separate CVE) and in other copies such as krb5 GSSRPC. CVE-2026-104357 is assigned to the libtirpc implementation. Reported by Aisle Research against glibc; Red Hat confirmed the bug is present in libtirpc. Related tracking: OSIM-77274 / RHEL-217046 (glibc report), PSIRTSUPT-22635 (coordination).
Affected Software
Event History
Frequently Asked Questions
Which deployments are exposed?
SunRPC services using libtirpc's UDP/datagram server transport are exposed, specifically the svc_dg_recv or svcudp_create receive path. The issue concerns UDP/datagram handling rather than the allocated receive buffer size.
What does an attacker need to do to trigger the flaw?
An attacker needs to send datagrams to the affected UDP RPC service in a sequence that leaves data in the shared receive buffer and then submits a shorter or truncated request. The later request can have stale bytes from the earlier datagram decoded as its arguments.
What is the practical effect of successful exploitation?
A request that should fail due to missing or truncated arguments can instead be processed using stale argument bytes from a prior datagram. The documented reproduction caused svc_getargs to consume and echo leftover bytes; a correctly bounded decoder would return GARBAGE_ARGS.
How can this be validated?
The issue was reproduced with tst-udp-stale, using a full request followed by a truncated AUTH_NONE call with no arguments. Observing the follow-up call consume or echo data left from the prior request indicates the decoder is not bounded to the received datagram length.