netdev - Re: [PATCH net-next] sctp: define sctp_packet_gso

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]

Date:   Thu, 14 Jun 2018 09:21:52 +0800
From:   Xin Long <lucien.xin@...il.com>
To:     Neil Horman <nhorman@...driver.com>
Cc:     network dev <netdev@...r.kernel.org>, linux-sctp@...r.kernel.org,
        Marcelo Ricardo Leitner <marcelo.leitner@...il.com>,
        davem <davem@...emloft.net>,
        Eric Dumazet <eric.dumazet@...il.com>
Subject: Re: [PATCH net-next] sctp: define sctp_packet_gso_append to build GSO frames

On Thu, Jun 14, 2018 at 8:46 AM, Neil Horman <nhorman@...driver.com> wrote:
> On Thu, Jun 14, 2018 at 07:37:02AM +0800, Xin Long wrote:
>> Now sctp GSO uses skb_gro_receive() to append the data into head
>> skb frag_list. However it actually only needs very few code from
>> skb_gro_receive(). Besides, NAPI_GRO_CB has to be set while most
>> of its members are not needed here.
>>
>> This patch is to add sctp_packet_gso_append() to build GSO frames
>> instead of skb_gro_receive(), and it would avoid many unnecessary
>> checks and make the code clearer.
>>
>> Note that sctp will use page frags instead of frag_list to build
>> GSO frames in another patch. But it may take time, as sctp's GSO
>> frames may have different size. skb_segment() can only split it
>> into the frags with the same size, which would break the border
>> of sctp chunks.
>>
>> Signed-off-by: Xin Long <lucien.xin@...il.com>
> Do you have any performance numbers to compare with and without this patch?
> Adding a function like this implies that any fixes that go into skb_gro_receive
> now need to be evaluated for this function too, which means theres an implied
> overhead in maintaining it.  If we're signing up for that, I'd like to know that
> theres a significant performance benefit.
Hi Neil,

I don't think there's a noticeable performance benefit since it's
just avoided some checks and variables settings.

The new function makes SCTP GSO code clearer and readable,
as skb_gro_receive() should only be used in the GRO code paths,
it's confusing in sctp tx path.

We're doing this, actually because skb_gro_receive() is being
changed now, it would not be suitable for SCTP GSO, see:
https://www.spinics.net/lists/netdev/msg507716.html

And also, we believe page frags will be used to replace frag_list
to build SCTP GSO frames soon. After that, this function will also
be dropped.