netdev - Re: [PATCH v5 bpf-next 09/12] bpf: tcp: Allow bpf prog to write and parse TCP header option

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Thu, 1 Oct 2020 15:38:42 -0700
From:   Martin KaFai Lau <kafai@...com>
To:     <sdf@...gle.com>
CC:     <bpf@...r.kernel.org>, Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Eric Dumazet <edumazet@...gle.com>, <kernel-team@...com>,
        Lawrence Brakmo <brakmo@...com>,
        Neal Cardwell <ncardwell@...gle.com>, <netdev@...r.kernel.org>,
        Yuchung Cheng <ycheng@...gle.com>
Subject: Re: [PATCH v5 bpf-next 09/12] bpf: tcp: Allow bpf prog to write and
 parse TCP header option

On Wed, Sep 30, 2020 at 08:23:47AM -0700, sdf@...gle.com wrote:
> On 08/20, Martin KaFai Lau wrote:
> > [..]
> > +static inline void bpf_skops_init_child(const struct sock *sk,
> > +					struct sock *child)
> > +{
> > +	tcp_sk(child)->bpf_sock_ops_cb_flags =
> > +		tcp_sk(sk)->bpf_sock_ops_cb_flags &
> > +		(BPF_SOCK_OPS_PARSE_ALL_HDR_OPT_CB_FLAG |
> > +		 BPF_SOCK_OPS_PARSE_UNKNOWN_HDR_OPT_CB_FLAG |
> > +		 BPF_SOCK_OPS_WRITE_HDR_OPT_CB_FLAG);
> > +}
> It looks like it breaks test_tcpbpf_user test in an interesting way, can
> you verify on your side?
> 
> Awhile ago, I've added retries to this test to make it less flaky.
> The test is waiting for 3 BPF_TCP_CLOSE events and now it
> only gets 2 BPF_TCP_CLOSE events.
> 
> IIUC, we used to copy/inherit parent bpf_sock_ops_cb_flags and now
> we are doing only a small subset (bpf tcp header) with the code above.
> 
> I'm still trying to understand whether that's working as intended
> and we need to fix the test or it's a user-visible breakage.
> Thoughts?
Thanks for the report.

Agree.  bpf_skops_init_child() is unnecessary and it will break
existing assumption that the passive established socket
will inherit all cb_flags from the listen socket.  It should just
allow sock_copy() to do its job.

I will post a fix.