CVE-2026-46114

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> RDMA/rxe: Reject non-8-byte ATOMIC_WRITE payloads<br /> <br /> atomic_write_reply() at drivers/infiniband/sw/rxe/rxe_resp.c<br /> unconditionally dereferences 8 bytes at payload_addr(pkt):<br /> <br /> value = *(u64 *)payload_addr(pkt);<br /> <br /> check_rkey() previously accepted an ATOMIC_WRITE request with pktlen ==<br /> resid == 0 because the length validation only compared pktlen against<br /> resid. A remote initiator that sets the RETH length to 0 therefore reaches<br /> atomic_write_reply() with a zero-byte logical payload, and the responder<br /> reads sizeof(u64) bytes from past the logical end of the packet into<br /> skb-&gt;head tailroom, then writes those 8 bytes into the attacker&amp;#39;s MR via<br /> rxe_mr_do_atomic_write(). That is a remote disclosure of 4 bytes of kernel<br /> tailroom per probe (the other 4 bytes are the packet&amp;#39;s own trailing ICRC).<br /> <br /> IBA oA19-28 defines ATOMIC_WRITE as exactly 8 bytes. Anything else is<br /> protocol-invalid. Hoist a strict length check into check_rkey() so the<br /> responder never reaches the unchecked dereference, and keep the existing<br /> WRITE-family length logic for the normal RDMA WRITE path.<br /> <br /> Reproduced on mainline with an unmodified rxe driver: a sustained<br /> zero-length ATOMIC_WRITE probe repeatedly leaks adjacent skb head-buffer<br /> bytes into the attacker&amp;#39;s MR, including recognisable kernel strings and<br /> partial kernel-direct-map pointer words. With this patch applied the<br /> responder rejects the PDU and the MR stays all-zero.