CVE-2026-98360

In the Linux kernel, the following vulnerability has been resolved:

RDMA/rxe: insert mcg into mcg tree only after rxe mcast add() succeeds

rxe get mcg() publishes a newly allocated multicast group in rxe->mcg tree before programming the backing Ethernet multicast address with rxe mcast add(), which runs outside mcg lock. A local userspace RDMA client reaches this path with ATTACH MCAST on a UD QP; if rxe mcast add() then returns an error (for example -ENODEV when the backing netdev has been removed, or a propagated dev mc add() error), the unwind frees the published group without removing it from the tree. A later lookup of the same MGID dereferences the freed struct rxe mcg from rxe lookup mcg().

Fix this by keeping the new mcg private until rxe mcast add() succeeds. Split the tree publication into rxe publish mcg(), call rxe mcast add() before taking the tree reference, and free the still-private mcg on failure. Because the group is never visible in mcg tree until the multicast address is programmed, no concurrent caller can look it up or attach a QP to a group that is about to be torn down, so the error path needs no conditional unwind. If another caller publishes the same MGID while the address is being programmed, the post-add re-check under mcg lock finds the winner; this caller then drops its private object and balances its own rxe mcast add() with rxe mcast del() before returning the winner.

Reproduced by forcing the rxe mcast add() error return under KASAN: without the change the next attach to the same MGID reports a slab-use-after-free in rxe lookup mcg(); with it the forced failure returns cleanly. A no-injection attach/detach regression, including a two-QP shared join/leave and re-attach, stays KASAN- and leak-clean.