Messages in this thread | | | From | Andy Lutomirski <> | Date | Wed, 3 Apr 2013 12:01:36 -0700 | Subject | Re: [ 105/124] af_unix: dont send SCM_CREDENTIAL when dest socket is NULL |
| |
On Wed, Apr 3, 2013 at 11:43 AM, Eric Dumazet <eric.dumazet@gmail.com> wrote: > On Wed, 2013-04-03 at 10:58 -0700, Andy Lutomirski wrote: > >> >> This sounds suspiciously like an SCM_CREDENTIALS bug triggered by a >> race. There's a fix (that needs both a new version from me and a review >> by someone) here: >> >> http://www.spinics.net/lists/netdev/msg229948.html > > Hmm... this is not a stable candidate, IMHO.
Agreed.
> > This has to be fixed (if needed) in a more easy way. > > What about this one liner ? > > CC Eric W. Biederman as he wrote commit > dbe9a4173ea53b72b2c3 > (scm: Don't use struct ucred in NETLINK_CB and struct scm_cookie.) > > diff --git a/include/net/scm.h b/include/net/scm.h > index 975cca0..42359d8 100644 > --- a/include/net/scm.h > +++ b/include/net/scm.h > @@ -120,7 +120,7 @@ static __inline__ void scm_recv(struct socket *sock, struct msghdr *msg, > return; > } > > - if (test_bit(SOCK_PASSCRED, &sock->flags)) { > + if (test_bit(SOCK_PASSCRED, &sock->flags) && scm->creds.pid) { > struct user_namespace *current_ns = current_user_ns(); > struct ucred ucreds = { > .pid = scm->creds.pid, > >
That looks like it's correct. If it gets applied, I'll respin my patches on top of it.
(This approach may be a POSIX violation for all I know, and it's even possible that some really fragile userspace breaks. But I doubt it, and anything that will break as a result is already operating in a highly confused state; hence the original problem.)
--Andy
| |