netdev - Re: [RFC net-next] ipv6: Attempt to improve options code parsing

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: <aa58193c-0a8f-d11b-fb0c-bc41571e33ac@gmail.com>
Date:   Tue, 3 Aug 2021 18:35:57 +0200
From:   Eric Dumazet <eric.dumazet@...il.com>
To:     Justin Iurman <justin.iurman@...ege.be>,
        Eric Dumazet <eric.dumazet@...il.com>
Cc:     netdev@...r.kernel.org, davem@...emloft.net, kuba@...nel.org,
        yoshfuji@...ux-ipv6.org, dsahern@...nel.org, tom@...bertland.com
Subject: Re: [RFC net-next] ipv6: Attempt to improve options code parsing



On 8/3/21 6:06 PM, Justin Iurman wrote:
>>> As per Eric's comment on a previous patchset that was adding a new HopbyHop
>>> option, i.e. why should a new option appear before or after existing ones in the
>>> list, here is an attempt to suppress such competition. It also improves the
>>> efficiency and fasten the process of matching a Hbh or Dst option, which is
>>> probably something we want regarding the list of new options that could quickly
>>> grow in the future.
>>>
>>> Basically, the two "lists" of options (Hbh and Dst) are replaced by two arrays.
>>> Each array has a size of 256 (for each code point). Each code point points to a
>>> function to process its specific option.
>>>
>>> Thoughts?
>>>
>> Hi Justin
>>
>> I think this still suffers from indirect call costs (CONFIG_RETPOLINE=y),
>> and eventually use more dcache.
> 
> Agree with both. It was the compromise for such a solution, unfortunately.
> 
>> Since we only deal with two sets/arrays, I would simply get rid of them
>> and inline the code using two switch() clauses.
> 
> Indeed, this is the more efficient. However, we still have two "issues":
>  - ip6_parse_tlv will keep growing and code could look ugly at some point

Well, in 10 years there has not been a lot of growth.

>  - there is still a "competition" between options, i.e. "I want to be at the top of the list"

Why would that be ?

A switch() is compiled with no particular order by the compiler.

Code generation depends on case density, and will use bisection-like strategy.

> 
> Anyway, your solution is better than the current one so it's probably the way to go right now.
>