netdev - Re: [PATCH net-next rfc 5/6] net-tc: convert tc_at to tc_at

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 for Android: free password hash cracker in your pocket

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

Date:   Wed, 4 Jan 2017 14:26:15 -0500
From:   Willem de Bruijn <willemdebruijn.kernel@...il.com>
To:     Daniel Borkmann <daniel@...earbox.net>
Cc:     Network Development <netdev@...r.kernel.org>,
        David Miller <davem@...emloft.net>,
        Florian Westphal <fw@...len.de>, dborkman@...earbox.net,
        Jamal Hadi Salim <jhs@...atatu.com>,
        Alexei Starovoitov <alexei.starovoitov@...il.com>,
        Willem de Bruijn <willemb@...gle.com>
Subject: Re: [PATCH net-next rfc 5/6] net-tc: convert tc_at to tc_at_ingress

>> diff --git a/net/sched/sch_api.c b/net/sched/sch_api.c
>> index ef53ede..be4e18d 100644
>> --- a/net/sched/sch_api.c
>> +++ b/net/sched/sch_api.c
>> @@ -1865,6 +1865,7 @@ int tc_classify(struct sk_buff *skb, const struct
>> tcf_proto *tp,
>>         const struct tcf_proto *old_tp = tp;
>>         int limit = 0;
>>
>> +       skb->tc_at_ingress = !!(tp && tp->q->flags & TCQ_F_INGRESS);
>
>
> I'd prefer if skb->tc_at_ingress is set directly to 0/1 in
> sch_handle_ingress()
> and __dev_queue_xmit() as we do right now, this would avoid above tests in
> fast
> path and it would also avoid to set the same thing in tc_classify() multiple
> times f.e. on egress path walking through multiple qdiscs. I don't see
> anything
> in layers above tc that would read it and expect an AT_STACK-like
> equivalent.
> skb_reset_tc() could thus still remain as you have above in fast-path like
> __netif_receive_skb_core().

I had been thinking about that. After submitting this I noticed that Florian's
patchset had an elegant solution to avoid the branch: set tc_at_ingress in
handle_ing before tc_classify and clear it on the return path.

Then we only set + clear it once on ingress regardless of the depth
of classifiers and do not touch it at all in other code.

https://patchwork.ozlabs.org/patch/472698/

What do you think of that approach?