On Fri, Jul 14, 2017 at 1:35 PM, Alexander Potapenko <gli...@google.com> wrote: > On Fri, Jul 14, 2017 at 7:04 PM, Neal Cardwell <ncardw...@google.com> wrote: >> On Fri, Jul 14, 2017 at 12:54 PM, Alexander Potapenko <gli...@google.com> >> wrote: >>> KMSAN reported use of uninitialized memory in skb_set_hash_from_sk(), >>> which originated from the TCP request socket created in >>> cookie_v6_check(): >> ... >>> --- a/net/ipv6/syncookies.c >>> +++ b/net/ipv6/syncookies.c >>> @@ -216,6 +216,7 @@ struct sock *cookie_v6_check(struct sock *sk, struct >>> sk_buff *skb) >>> treq->rcv_isn = ntohl(th->seq) - 1; >>> treq->snt_isn = cookie; >>> treq->ts_off = 0; >>> + treq->txhash = 0; >>> >>> /* >>> * We need to lookup the dst_entry to get the correct window size. >> >> I would have thought that the same fix is needed in the corresponding >> line in cookie_v4_check() in net/ipv4/syncookies.c? (I do not see >> txhash being initialized for the IPv4 side.) If it's not needed for >> some reason, then it would be worth a comment in the commit >> description to explain why not. > Most certainly it is needed. I haven't seen reports for that in the > wild and couldn't forge a repro triggering the bug in IPv4, but I'll > give it another shot.
If you force syncookies to be used, with: sysctl -q net.ipv4.tcp_syncookies=2 then every passive IPv4 TCP connection that is accepted by a TCP server app should be using that uninitialized memory, I would think (and thus should be liable to fire the KMSAN use of uninitialized memory warning). Does that work? neal
