netdev - Re: Linux ECN Handling

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]

Message-ID: <CACJspmKojZYnBoQiru8r7U+8bmtiKEZwjuVe7+qZKbYiYxeWQA@mail.gmail.com>
Date:   Wed, 3 Jan 2018 14:21:58 -0800
From:   Steve Ibanez <sibanez@...nford.edu>
To:     Neal Cardwell <ncardwell@...gle.com>
Cc:     Eric Dumazet <edumazet@...gle.com>,
        Yuchung Cheng <ycheng@...gle.com>,
        Daniel Borkmann <daniel@...earbox.net>,
        Netdev <netdev@...r.kernel.org>, Florian Westphal <fw@...len.de>,
        Mohammad Alizadeh <alizadeh@...il.mit.edu>,
        Lawrence Brakmo <brakmo@...com>
Subject: Re: Linux ECN Handling

Hi Neal,

I've attached a pdf of a slide that I made which shows some data from
the kernel log at the receiver as well as the timestamp, SeqNo, and
length of the corresponding segments from the tcpdump trace at the
receiver interface. Hopefully this helps clarify why I think
tcp_transmit_skb() is called at some point before tcp_ack_snd_check()
while processing the CWR segment for which no ACK is sent. Please let
me know if anything is unclear or if you know where that initial call
to tcp_transmit_skb() might be coming from and why it's happening
prematurely.

Best,
-Steve

On Wed, Jan 3, 2018 at 11:39 AM, Neal Cardwell <ncardwell@...gle.com> wrote:
> On Tue, Jan 2, 2018 at 6:57 PM, Steve Ibanez <sibanez@...nford.edu> wrote:
>> Hi Neal,
>>
>> Sorry, my last email was incorrect. It turns out the default tcp
>> congestion control alg that was being used on my client machines was
>> cubic instead of dctcp. That is why tp->processing_cwr field was never
>> set in the tcp_rcv_established function. I've changed the default back
>> to dctcp on all of my machines.
>>
>> I am now logging the value of tp->rcv_nxt at the top of the
>> tcp_transmit_skb() function for all CWR segments. I see that during
>> normal operation, the value of tp->rcv_nxt is equal to the SeqNo in
>> the CWR segment  + length of the CWR segment.
>
> OK, thanks. That makes sense.
>
> This part I didn't understand:
>
>> However, for the unACKed
>> CWR segment, the value of tp->rcv_nxt is just equal to the SeqNo in
>> the CWR segment (i.e. not incremented by the length). And I see that
>> by the time the tcp_ack_snd_check() function is executed, tp->rcv_nxt
>> has been incremented by the length of the unACKed CWR segment.
>
> I would have thought that for the processing of the skb that has the
> CWR, the sequence would be:
>
> (1)  "...the tcp_ack_snd_check() function is executed, tp->rcv_nxt has
> been incremented by the length of the unACKed CWR segment"
>
> (2) then we send the ACK, and the instrumentation at the top of the
> tcp_transmit_skb() function logs that rcv_nxt value (which "has been
> incremented by the length of the unACKed CWR segment").
>
> But you are saying "for the unACKed CWR segment, the value of
> tp->rcv_nxt is just equal to the SeqNo in the CWR segment (i.e. not
> incremented by the length)", which does not seem to match my
> prediction in (2). Apparently I am mis-understanding the sequence.
> Perhaps you can help clear it up for me? :-)
>
> Is it possible that the case where you see "tp->rcv_nxt is just equal
> to the SeqNo in the CWR segment" is a log line that was logged while
> processing the skb that precedes the skb with the CWR?
>
>> The tcp_transmit_skb() function sets the outgoing segment's ack_seq to
>> be tp->rcv_next:
>>
>> th->ack_seq             = htonl(tp->rcv_nxt);
>>
>> So I think the rcv_nxt field is supposed to be incremented before
>> reaching tcp_transmit_skb(). Can you see any reason as to why this
>> field would not be incremented for CWR segments sometimes?
>
> No, so far I haven't been able to think of a reason why rcv_nxt would
> not be incremented for in-order CWR-marked segments...
>
> cheers,
> neal

Download attachment "CWR_log_and_trace.pdf" of type "application/pdf" (34236 bytes)