From: Suyuan Wang SO_LOCK_FILTER lets a process lock the filters on a socket before handing it to a less privileged one. Once it is set, the existing attach and detach paths reject changes to the socket's BPF filters with -EPERM, but SO_DETACH_REUSEPORT_BPF does not check SOCK_FILTER_LOCKED. So an unprivileged process that receives a locked socket can still remove the reuseport program, which then changes how packets and connections are spread over the whole reuseport group. Return -EPERM from reuseport_detach_prog() when the filter is locked, like sk_detach_filter() does. Found with a static analysis tool and checked by hand. Fixes: 99f3a064bc2e ("bpf: net: Add SO_DETACH_REUSEPORT_BPF") Cc: stable@vger.kernel.org Co-developed-by: Kaixuan Li Signed-off-by: Kaixuan Li Signed-off-by: Suyuan Wang --- Tested in QEMU (x86_64): root binds a UDP socket with SO_REUSEPORT, attaches a reuseport cBPF program and sets SO_LOCK_FILTER, then a child running as nobody uses the inherited socket. Before: SO_ATTACH_REUSEPORT_CBPF: EPERM SO_DETACH_FILTER: EPERM SO_DETACH_REUSEPORT_BPF: succeeds, the program is gone After: SO_DETACH_REUSEPORT_BPF: EPERM as well --- net/core/sock_reuseport.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/net/core/sock_reuseport.c b/net/core/sock_reuseport.c index b9e5648fa4c5..cdff61161a70 100644 --- a/net/core/sock_reuseport.c +++ b/net/core/sock_reuseport.c @@ -727,6 +727,9 @@ int reuseport_detach_prog(struct sock *sk) struct sock_reuseport *reuse; struct bpf_prog *old_prog; + if (sock_flag(sk, SOCK_FILTER_LOCKED)) + return -EPERM; + old_prog = NULL; spin_lock_bh(&reuseport_lock); reuse = rcu_dereference_protected(sk->sk_reuseport_cb,