netdev - Re: IPv6 Destination Options question

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: <98532586.q66oQjG9bA@cg-notebook>
Date:   Mon, 16 Dec 2019 02:59:01 +0100
From:   Christoph Grenz <christophg+lkml@...nz-bonn.de>
To:     Tom Herbert <tom@...bertland.com>,
        Linux Kernel Network Developers <netdev@...r.kernel.org>
Subject: Re: IPv6 Destination Options question

Hi Tom,

I narrowed the issue down to a possible bug in the kernel-side IPv6 checksum 
calculation:

A raw socket for IPPROTO_MH has checksum calculation/verification enabled by 
default (as if setsockopt(s, IPPROTO_IPV6, IPV6_CHECKSUM, 4) was used).

I get different checksums for the same packet with and without IPV6_DSTOPTS 
ancillary data. The checksum is independent of the content of the Destination 
Options, but adding multiple IPV6_DSTOPTS anciliary data items changes the 
calculated checksum, even if the resulting packet is otherwise identical.

The checksum verification then discards the packets on the receiving side, 
which is why I didn't get any packets on my IPPROTO_MH socket and suspected a 
problem on the receiving side first.

When I calculate the checksum in userspace and set the IPV6_CHECKSUM socket 
option to -1 to disable the in-kernel calculation on the sending side, I 
receive all packets as expected.

It doesn't happen with IPV6_HOPOPTS or IPV6_RTHDR ancillary data, at least as 
far as I checked.

Could you take a look at it before I submit a bug report?

Best regards
Christoph

Am Samstag, 14. Dezember 2019, 22:11:01 CET schrieb Christoph Grenz:
> Hi Tom,
> 
> my receive code boils down to this Python script:
> 
> --------------------------------------------------------------------
> #!/usr/bin/env python3
> 
> from socket import socket, AF_INET6, IPPROTO_IPV6, IPV6_RECVDSTOPTS,
> SOCK_RAW
> 
> IPPROTO_MH = 135 # IPv6 Mobility Header
> 
> sock = socket(AF_INET6, SOCK_RAW, IPPROTO_MH)
> sock.setsockopt(IPPROTO_IPV6, IPV6_RECVDSTOPTS, True)
> 
> sock.bind(('::',  0))
> 
> while True:
> 	packet, ancdata, msg_flags, address = sock.recvmsg(1800, 512)
> 	print(address[0], packet.hex(), ancdata)
> 
> --------------------------------------------------------------------
> 
> Best regards,
> Christoph
> 
> Am Samstag, 14. Dezember 2019, 12:40:16 CET schrieb Tom Herbert:
> > On Sat, Dec 14, 2019 at 8:19 AM Christoph Grenz
> > 
> > <christophg+lkml@...nz-bonn.de> wrote:
> > > Hello,
> > > 
> > > I'm playing around with Mobile IPv6 and noticed a strange behaviour in
> > > the
> > > Linux network system when using IPv6 destination options:
> > > 
> > > I'm able to send destination options on SOCK_DGRAM and SOCK_RAW sockets
> > > with sendmsg() and IPV6_DSTOPTS ancillary data. The sent packets also
> > > look correct in Wireshark.
> > > 
> > > But I'm not able to receive packets with destination options on a socket
> > > with the IPV6_RECVDSTOPTS socket option enabled. Both a packet with a
> > > Home Address Option and a packet with an empty destination options
> > > header
> > > (only containing padding) won't be received on a socket for the payload
> > > protocol.
> > 
> > Christoph, Can you post your receive code?
> > 
> > Thanks
> > 
> > > Only a SOCK_RAW socket for IPPROTO_DSTOPTS receives the packet.
> > > 
> > > I tested this on a vanilla 5.4.0 kernel and got the same behaviour.
> > > Activating dyndbg for everything in net/ipv6 didn't produce any relevant
> > > output in dmesg.
> > > 
> > > Is this expected behaviour or a bug? Or do I maybe need some other
> > > socket
> > > option or a xfrm policy to receive packets with destination options?
> > > 
> > > Best regards
> > > Christoph