On Wed, 2021-01-13 at 11:26 +0100, Eric Dumazet wrote:
>
> On 1/13/21 11:21 AM, Eric Dumazet wrote:
> >
> > On 1/12/21 6:25 PM, Paolo Abeni wrote:
> > > Instead of re-implementing most of inet_shutdown, re-use
> > > such helper, and implement the MPTCP-specific bits at the
> > > 'proto' level.
> > >
> > > The msk-level disconnect() can now be invoked, lets provide a
> > > suitable implementation.
> > >
> > > As a side effect, this fixes bad state management for listener
> > > sockets. The latter could lead to division by 0 oops since
> > > commit ea4ca586b16f ("mptcp: refine MPTCP-level ack scheduling").
> > >
> > > Fixes: 43b54c6ee382 ("mptcp: Use full MPTCP-level disconnect state
> > > machine")
> > > Fixes: ea4ca586b16f ("mptcp: refine MPTCP-level ack scheduling")
> > > Signed-off-by: Paolo Abeni <pab...@redhat.com>
> > > ---
> > > net/mptcp/protocol.c | 62 ++++++++++++--------------------------------
> > > 1 file changed, 17 insertions(+), 45 deletions(-)
> > >
> > > diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
> > > index 2ff8c7caf74f..81faeff8f3bb 100644
> > > --- a/net/mptcp/protocol.c
> > > +++ b/net/mptcp/protocol.c
> > > @@ -2642,11 +2642,12 @@ static void mptcp_copy_inaddrs(struct sock *msk,
> > > const struct sock *ssk)
> > >
> > > static int mptcp_disconnect(struct sock *sk, int flags)
> > > {
> > > - /* Should never be called.
> > > - * inet_stream_connect() calls ->disconnect, but that
> > > - * refers to the subflow socket, not the mptcp one.
> > > - */
> > > - WARN_ON_ONCE(1);
> > > + struct mptcp_subflow_context *subflow;
> > > + struct mptcp_sock *msk = mptcp_sk(sk);
> > > +
> > > + __mptcp_flush_join_list(msk);
> > > + mptcp_for_each_subflow(msk, subflow)
> > > + tcp_disconnect(mptcp_subflow_tcp_sock(subflow), flags);
> >
> > Ouch.
> >
> > tcp_disconnect() is supposed to be called with socket lock being held.
> >
> > Really, CONFIG_LOCKDEP=y should have warned you :/
>
> Or maybe CONFIG_PROVE_RCU=y is needed to catch the bug.
Thank you for catching this!
Yep, CONFIG_PROVE_RCU=y triggers a 'suspicious RCU usage' warning. I
should really enable 'panic_on_warn' in batch tests.
I'll send a patch.
Thanks,
Paolo
