netdev - Re: [PATCH net-next,v4 08/12] drivers: net: use flow block API

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] [day] [month] [year] [list]

Date:   Fri, 16 Aug 2019 18:00:35 +0100
From:   Edward Cree <ecree@...arflare.com>
To:     Pablo Neira Ayuso <pablo@...filter.org>
CC:     <netdev@...r.kernel.org>, <netfilter-devel@...r.kernel.org>
Subject: Re: [PATCH net-next,v4 08/12] drivers: net: use flow block API

On 16/08/2019 02:04, Pablo Neira Ayuso wrote:
> On Wed, Aug 14, 2019 at 05:17:20PM +0100, Edward Cree wrote:
>> TBH I'm still not clear why you need a flow_block per subsystem, rather than
>>  just having multiple subsystems feed their offload requests through the same
>>  flow_block but with different enum tc_setup_type or enum tc_fl_command or
>>  some other indication that this is "netfilter" rather than "tc" asking for a
>>  tc_cls_flower_offload.
> In tc, the flow_block is set up by when the ingress qdisc is
> registered. The usual scenario for most drivers is to have one single
> flow_block per registered ingress qdisc, this makes a 1:1 mapping
> between ingress qdisc and flow_block.
>
> Still, you can register two or more ingress qdiscs to make them share
> the same policy via 'tc block'. In that case all those qdiscs use one
> single flow_block. This makes a N:1 mapping between these qdisc
> ingress and the flow_block. This policy applies to all ingress qdiscs
> that are part of the same tc block. By 'tc block', I'm refering to the
> tcf_block structure.
>
> In netfilter, there are ingress basechains that are registered per
> device. Each basechain gets a flow_block by when the basechain is
> registered. Shared blocks as in tcf_block are not yet supported, but
> it should not be hard to extend it to make this work.
>
> To reuse the same flow_block as entry point for all subsystems as your
> propose - assuming offloads for two or more subsystems are in place -
> then all of them would need to have the same block sharing
> configuration, which might not be the case, ie. tc ingress might have
> a eth0 and eth1 use the same policy via flow_block, while netfilter
> might have one basechain for eth0 and another for eth1 (no policy
> sharing).
Thank you, that's very helpful.

>> This really needs a design document explaining what all the bits are, how
>>  they fit together, and why they need to be like that.
> I did not design this flow_block abstraction, this concept was already
> in place under a different name and extend it so the ethtool/netfilter
> subsystems to avoid driver code duplication for offloads.
It's more the new implementation that you've created as part of this
 extension that I was asking about, although I agree that the
 abstraction that already existed is in need of documentation too.