On Tue, 2017-05-09 at 15:37 -0700, Cong Wang wrote:
> On Tue, May 9, 2017 at 6:29 AM, Eric Dumazet <eric.duma...@gmail.com> wrote:
> > From: Eric Dumazet <eduma...@google.com>
> >
> > syzkaller found a way to trigger double frees from ip_mc_drop_socket()
> >
> > It turns out that leave a copy of parent mc_list at accept() time,
> > which is very bad.
> >
> > Very similar to commit 8b485ce69876 ("tcp: do not inherit
> > fastopen_req from parent")
> >
> > Initial report from Pray3r, completed by Andrey one.
> > Thanks a lot to them !
> >
> > Signed-off-by: Eric Dumazet <eduma...@google.com>
> > Reported-by: Pray3r <pray3...@gmail.com>
> > Reported-by: Andrey Konovalov <andreyk...@google.com>
> > Tested-by: Andrey Konovalov <andreyk...@google.com>
> > ---
> > v2: fix moved into inet_csk_clone_lock() to fix both DCCP and TCP
> >
> > net/ipv4/inet_connection_sock.c | 2 ++
> > 1 file changed, 2 insertions(+)
> >
> > diff --git a/net/ipv4/inet_connection_sock.c
> > b/net/ipv4/inet_connection_sock.c
> > index
> > 5e313c1ac94fc88eca5fe3a0e9e46e551e955ff0..1054d330bf9df3189a21dbb08e27c0e6ad136775
> > 100644
> > --- a/net/ipv4/inet_connection_sock.c
> > +++ b/net/ipv4/inet_connection_sock.c
> > @@ -794,6 +794,8 @@ struct sock *inet_csk_clone_lock(const struct sock *sk,
> > /* listeners have SOCK_RCU_FREE, not the children */
> > sock_reset_flag(newsk, SOCK_RCU_FREE);
> >
> > + inet_sk(newsk)->mc_list = NULL;
> > +
> > newsk->sk_mark = inet_rsk(req)->ir_mark;
> > atomic64_set(&newsk->sk_cookie,
> > atomic64_read(&inet_rsk(req)->ir_cookie));
> >
>
> I think IPv6 needs this too?
>
> diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
> index aeb9497..b3611d9 100644
> --- a/net/ipv6/tcp_ipv6.c
> +++ b/net/ipv6/tcp_ipv6.c
> @@ -1062,6 +1062,7 @@ static struct sock *tcp_v6_syn_recv_sock(const
> struct sock *sk, struct sk_buff *
> newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
> #endif
>
> + newnp->ipv6_mc_list = NULL;
> newnp->ipv6_ac_list = NULL;
> newnp->ipv6_fl_list = NULL;
> newnp->pktoptions = NULL;
Good point, but it looks like you patched on only IPv4 mapped sockets.
And DCCP would need fixes as well.
