netdev - Re: tcpdump and Big TCP

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

Message-ID: <CANn89i+Ef7zNz7t6U2_6VEHPDantgyR8d0w3ALOBVVwK0Fe=FQ@mail.gmail.com>
Date: Mon, 2 Oct 2023 19:26:00 +0200
From: Eric Dumazet <edumazet@...gle.com>
To: Xin Long <lucien.xin@...il.com>
Cc: David Ahern <dsahern@...nel.org>, "netdev@...r.kernel.org" <netdev@...r.kernel.org>
Subject: Re: tcpdump and Big TCP

On Mon, Oct 2, 2023 at 7:19 PM Xin Long <lucien.xin@...il.com> wrote:
>
> On Mon, Oct 2, 2023 at 12:25 PM Eric Dumazet <edumazet@...gle.com> wrote:
> >
> > On Mon, Oct 2, 2023 at 6:20 PM David Ahern <dsahern@...nel.org> wrote:
> > >
> > > Eric:
> > >
> > > Looking at the tcpdump source code, it has a GUESS_TSO define that can
> > > be enabled to dump IPv4 packets with tot_len = 0:
> > >
> > >         if (len < hlen) {
> > > #ifdef GUESS_TSO
> > >             if (len) {
> > >                 ND_PRINT("bad-len %u", len);
> > >                 return;
> > >             }
> > >             else {
> > >                 /* we guess that it is a TSO send */
> > >                 len = length;
> > >             }
> > > #else
> > >             ND_PRINT("bad-len %u", len);
> > >             return;
> > > #endif /* GUESS_TSO */
> > >         }
> > >
> > >
> > > The IPv6 version has a similar check but no compile change needed:
> > >         /*
> > >          * RFC 1883 says:
> > >          *
> > >          * The Payload Length field in the IPv6 header must be set to zero
> > >          * in every packet that carries the Jumbo Payload option.  If a
> > >          * packet is received with a valid Jumbo Payload option present and
> > >          * a non-zero IPv6 Payload Length field, an ICMP Parameter Problem
> > >          * message, Code 0, should be sent to the packet's source, pointing
> > >          * to the Option Type field of the Jumbo Payload option.
> > >          *
> > >          * Later versions of the IPv6 spec don't discuss the Jumbo Payload
> > >          * option.
> > >          *
> > >          * If the payload length is 0, we temporarily just set the total
> > >          * length to the remaining data in the packet (which, for Ethernet,
> > >          * could include frame padding, but if it's a Jumbo Payload frame,
> > >          * it shouldn't even be sendable over Ethernet, so we don't worry
> > >          * about that), so we can process the extension headers in order
> > >          * to *find* a Jumbo Payload hop-by-hop option and, when we've
> > >          * processed all the extension headers, check whether we found
> > >          * a Jumbo Payload option, and fail if we haven't.
> > >          */
> > >         if (payload_len != 0) {
> > >                 len = payload_len + sizeof(struct ip6_hdr);
> > >                 if (length < len)
> > >                         ND_PRINT("truncated-ip6 - %u bytes missing!",
> > >                                 len - length);
> > >         } else
> > >                 len = length + sizeof(struct ip6_hdr);
> > >
> > >
> > > Maybe I am missing something, but it appears that no code change to
> > > tcpdump is needed for Linux Big TCP packets other than enabling that
> > > macro when building. I did that in a local build and the large packets
> > > were dumped just fine.
> > >
> Right, wireshark/tshark currently has no problem parsing BIG TCP IPv4 packets.
> I think it enables GUESS_TSO by default.
>
> We also enabled GUESS_TSO in tcpdump for RHEL-9 when BIG TCP IPv4 was
> backported in it.

Make sure to enable this in tcpdump source, so that other distros do
not have to 'guess'.

>
> >
> > My point is that tcpdump should not guess, but look at TP_STATUS_GSO_TCP
> > (and TP_STATUS_CSUM_VALID would also be nice)
> >
> > Otherwise, why add TP_STATUS_GSO_TCP in the first place ?
> That's for more reliable parsing in the future.

We want this. I thought this was obvious.

>
> As currently in libpcap, it doesn't save meta_data(like
> TP_STATUS_CSUM_VALID/GSO_TCP)
> to 'pcap' files, and it requires libpcap APIs change and uses the
> 'pcap-ng' file format.
> I think it will take quite some time to implement in userspace.

Great. Until this is implemented as discussed last year, we will not remove
IPv6 jumbo headers.