netdev - Re: [PATCH 5/6] net: add and use helpers when adjusting gso

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]

Date:   Wed, 28 Feb 2018 17:21:55 +0100
From:   Daniel Borkmann <daniel@...earbox.net>
To:     David Miller <davem@...emloft.net>, dja@...ens.net
Cc:     netdev@...r.kernel.org, alexei.starovoitov@...il.com
Subject: Re: [PATCH 5/6] net: add and use helpers when adjusting gso_size

On 02/28/2018 04:56 PM, David Miller wrote:
> From: Daniel Axtens <dja@...ens.net>
> Date: Wed, 28 Feb 2018 00:04:06 +1100
> 
>> An audit of users of gso_size reveals that an eBPF tc action on a
>> veth pair can be passed an SCTP GSO skb, which has gso_size of
>> GSO_BY_FRAGS.
>>
>> If that action calls bpf_skb_change_proto(), bpf_skb_net_grow()
>> or bpf_skb_net_shrink(), the gso_size will be unconditionally
>> incremented or decremented to some nonsense value.
>>
>> Add helpers that WARN if attempting to change a gso_size of a
>> GSO_BY_FRAGS skb (and leave the value unchanged).
>>
>> Signed-off-by: Daniel Axtens <dja@...ens.net>
> 
> We can't really do this.
> 
> It means that a user loaded eBPF program can trigger logs full of
> warnings merely by using this eBPF helper and generating GSO'd SCTP
> traffic.
> 
> Daniel and Alexei, this is a serious problem.  The eBPF helpers
> mentioned here cannot handle SCTP GSO packets properly, and in fact
> corrupt them if they adjust the gso_size.
> 
> SCTP GSO packets use the GSO_BY_FRAGS scheme and cannot be treated
> the same way we treat normal GSO packets.

Thanks for Cc, I would have missed it. This patch in combination with
patch 6/6 seems okay since we reject it in these cases, which is fine
(although it could be a warn_once), but patch 6/6 is definitely buggy
in that we leave the skb in a half edited state when leaving the helper.
Daniel, want me to fix this up from bpf side and route 5/6 and actual
fix via bpf tree? Otherwise please respin with fixed 6/6.

Thanks,
Daniel