netdev - Re: [PATCH net-next 03/15] udp: do checksum properly in skb_udp_tunnel

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 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:   Sat, 3 Oct 2020 15:40:05 +0800
From:   Xin Long <lucien.xin@...il.com>
To:     Marcelo Ricardo Leitner <marcelo.leitner@...il.com>
Cc:     network dev <netdev@...r.kernel.org>, linux-sctp@...r.kernel.org,
        Neil Horman <nhorman@...driver.com>,
        Michael Tuexen <tuexen@...muenster.de>,
        davem <davem@...emloft.net>, tom@...bertland.com
Subject: Re: [PATCH net-next 03/15] udp: do checksum properly in skb_udp_tunnel_segment

On Sat, Oct 3, 2020 at 12:04 PM Marcelo Ricardo Leitner
<marcelo.leitner@...il.com> wrote:
>
> On Tue, Sep 29, 2020 at 09:48:55PM +0800, Xin Long wrote:
> > This patch fixes two things:
> >
> >   When skb->ip_summed == CHECKSUM_PARTIAL, skb_checksum_help() should be
> >   called do the checksum, instead of gso_make_checksum(), which is used
> >   to do the checksum for current proto after calling skb_segment(), not
> >   after the inner proto's gso_segment().
> >
> >   When offload_csum is disabled, the hardware will not do the checksum
> >   for the current proto, udp. So instead of calling gso_make_checksum(),
> >   it should calculate checksum for udp itself.
>
> Gotta say, this is odd. It is really flipping the two around. What
> about other users of this function, did you test them too?
Not yet, I couldn't found other cases to trigger this.

But I think gso_make_checksum() is not correct to be used here,
as it's trying to calculate the checksum for inner protocol
instead of UDP's. It should be skb_checksum_help(), like on
the xmit path.

>
> It makes sense to be, but would be nice if someone else could review
> this.
Fix the mail of Tom Herbert, and he is the right person to review this.