netdev - Re: [RFC bpf-next v3 6/8] flow_dissector: handle no-skb use case

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: <20190326164529.GB7431@mini-arch.hsd1.ca.comcast.net>
Date:   Tue, 26 Mar 2019 09:45:29 -0700
From:   Stanislav Fomichev <sdf@...ichev.me>
To:     Alexei Starovoitov <alexei.starovoitov@...il.com>
Cc:     Stanislav Fomichev <sdf@...gle.com>, netdev@...r.kernel.org,
        bpf@...r.kernel.org, davem@...emloft.net, ast@...nel.org,
        daniel@...earbox.net, simon.horman@...ronome.com,
        willemb@...gle.com, peterpenkov96@...il.com
Subject: Re: [RFC bpf-next v3 6/8] flow_dissector: handle no-skb use case

On 03/25, Alexei Starovoitov wrote:
> On Sat, Mar 23, 2019 at 09:05:31AM -0700, Stanislav Fomichev wrote:
> > On 03/22, Alexei Starovoitov wrote:
> > > On Fri, Mar 22, 2019 at 06:19:57PM -0700, Stanislav Fomichev wrote:
> > > > Are we ok with breaking api in this case? I'm all in on removing this
> > > > extra information. We can always put it back if somebody complains (and
> > > > manually parse in eth_get_headlen case).
> > > 
> > > Fine. That seems to be the only way forward to clean it all up.
> > > Could you submit patch 1 to bpf tree disallowing vlan fields?
> > > Patch 3 looks like candidate as well?
> > SGTM, will do. Let me also spend some time and do a simple test for
> > the vlan case, to make sure I didn't miss something important.
> > One question here though: would I need to wait for bpf and bpf-next
> > to re-merge to continues the series? Or we can cherry-pick those
> > patches to bpf-next as well (and git will work it out during the
> > merge)?
> > 
> > > > We can still have protocol, because in both skb/skb-less cases we have
> > > > it.
> > > 
> > > proto can work in both cases, but is it needed ? Does program benefit from it?
> > > The kernel side burns extra bytes by copying it and extra branches to handle it.
> > > May be drop it as well?
> > I feel like the program benefits from it, there is no need to go back and
> > re-parse that (and in the skb case, this data is already pulled). I was
> > also thinking about re-purposing flow_keys->n_proto for that (instead
> > of skb->protocol), so it functions as input and output, maybe that's a
> > more clear way to do it.
> 
> Are you saying that skb-less and skb flow dissector progs are looking
> at different positions into the packet ?
No, sorry for confusion, they are both called to parse (optional) L2-vlan
and L3+ headers. However, with-skb case can be called with l2-vlan
parsed (post RFS) or with l2-vlan unparsed (RFS). The vlan is pulled in
__netif_receive_skb_core, but we can still invoke flow dissector prior to
that when doing RFS (get_rps_cpu).

That's why have this 'if skb->vlan_present' check in the bpf_flow.c program
(and then also manually test for ETH_P_8021Q/ETH_P_8021AD).

Let me try to post the patches to bpf tree somewhere this week, we
can discuss the API changes there.

> In case of with-skb it's already after eth header was pulled?
> In such case skb-less should be different program type or
> both should point at the same point.