netdev - Re: [PATCH net-next 00/17] Change cork to a pointer in sockets

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

Message-ID: <f9bc6eb8fb9c0771d7fff0b4da1a75528c5d6c07.camel@redhat.com>
Date: Mon, 23 Oct 2023 10:24:31 +0200
From: Paolo Abeni <pabeni@...hat.com>
To: Oliver Crumrine <ozlinuxc@...il.com>, netdev@...r.kernel.org, 
	linux-kernel@...r.kernel.org
Cc: davem@...emloft.net
Subject: Re: [PATCH net-next 00/17] Change cork to a pointer in sockets

On Sun, 2023-10-22 at 12:19 -0400, Oliver Crumrine wrote:
> This patch changes the cork field of both the inet_sock and inet6_sk
> structs to a pointer, reducing their size.
> 
> Oliver Crumrine (17):
>   Make cork in inet_sock a pointer.
>   Allocate and free cork in inet_create and inet_release in file
>     net/ipv4/af_inet.c
>   Change cork in ipv6_pinfo to a pointer.
>   Allocate and free cork in inet6_sk.
>   Change occurence of cork in inet_sock to pointer in include/net/ip.h
>   Since cork in inet_sock and inet6_sk will be a pointer, they don't
>     need to be referenced in this function in include/net/ipv6.h
>   Change occurences of cork in inet_sock to pointer in file
>     net/ipv4/af_inet.c
>   Change occurence of cork to pointer in file net/ipv4/datagram.c
>   Change instances of cork in net/ipv4/inet_connection_sock.c to
>     pointer.
>   Change instances of cork to pointer in net/ipv4/ip_output.c
>   Update occurences of cork to pointer in net/ipv4/syncookies.c
>   Change occurences of cork to pointer in net/ipv4/tcp_output.c
>   Change instances of cork to a pointer in file net/ipv4/udp.c
>   Update usages of cork in net/ipv6/ip6_output.c to pointer.
>   Modify occurences of cork in net/ipv6/raw.c to use a pointer
>   Change usages of cork to pointer in net/ipv6/udp.c
>   Change instance of cork to pointer in net/ipv4/tcp_ipv4.c
> 
>  include/linux/ipv6.h            |  2 +-
>  include/net/inet_sock.h         |  2 +-
>  include/net/ip.h                |  2 +-
>  include/net/ipv6.h              |  4 ++--
>  net/ipv4/af_inet.c              |  6 ++++--
>  net/ipv4/datagram.c             |  2 +-
>  net/ipv4/inet_connection_sock.c |  6 +++---
>  net/ipv4/ip_output.c            |  6 +++---
>  net/ipv4/syncookies.c           |  2 +-
>  net/ipv4/tcp_ipv4.c             |  2 +-
>  net/ipv4/tcp_output.c           |  2 +-
>  net/ipv4/udp.c                  |  8 ++++----
>  net/ipv6/af_inet6.c             |  5 +++++
>  net/ipv6/ip6_output.c           | 10 +++++-----
>  net/ipv6/raw.c                  |  4 ++--
>  net/ipv6/udp.c                  |  4 ++--
>  16 files changed, 37 insertions(+), 30 deletions(-)

Could you please explain the rationale behind such changes? 

As the cork struct is still allocated for each inet/inet6 socket, the
total memory used by the the now smaller socket struct and the cork 
struct will be greater then the memory used originally by such socket.
And the double allocation/free will be slower then the original one.

This also adds a bunch of additional pointer de-reference in the xmit
path.

Finally the above change will probably conflict with the goodies
introduced by:

https://lore.kernel.org/netdev/20231017014716.3944813-1-lixiaoyan@google.com/

I'm sorry, but this looks really a no-go.

Before future submissions, please read thoroughly the process
documentation, including the netdev specific bits in maintainer-
netdev.rst: there a bit of issues with the process here (the recipients
list does not include a lot of relevant ones, there are typos there,
the patch series is too long, it breaks the builds ...)

Cheers,

Paolo