[MPTCP] Re: [PATCH mptcp] squashto: mptcp: protect the rx path with the msk socket spinlock

[Florian Westphal <fw at strlen.de>]

[-- Attachment #1: Type: text/plain, Size: 1035 bytes --]

Paolo Abeni <pabeni(a)redhat.com> wrote:
> > diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
> > index 92236a2ccbea..e62d34034d9e 100644
> > --- a/net/mptcp/protocol.c
> > +++ b/net/mptcp/protocol.c
> > @@ -2707,7 +2707,9 @@ void mptcp_destroy_common(struct mptcp_sock *msk)
> >  
> >  	__mptcp_clear_xmit(sk);
> >  
> > -	skb_queue_splice_tail_init(&sk->sk_receive_queue, &msk->receive_queue);
> > +	/* move to sk_receive_queue, sk_stream_kill_queues will purge it */
> > +	skb_queue_splice_tail_init(&msk->receive_queue, &sk->sk_receive_queue);
> > +
> >  	skb_rbtree_purge(&msk->out_of_order_queue);
> >  	mptcp_token_destroy(msk);
> >  	mptcp_pm_free_anno_list(msk);
> 
> 
> LGTM, thanks for catching this!
> 
> Not sure why self-tests did non complain loudly here ?!?

Probably all test cases drain the rx queue completely before exiting,
we might need something new to cover this.

That being said, packetdrill tests catch it but I suppose those do not
run as frequently as the kselftest infra.