Re: [PATCH net-next 0/6] TCP connection repair (v4)
[Posted May 1, 2012 by corbet]
| From: |
| David Miller <davem-AT-davemloft.net> |
| To: |
| xemul-AT-parallels.com |
| Subject: |
| Re: [PATCH net-next 0/6] TCP connection repair (v4) |
| Date: |
| Sat, 21 Apr 2012 15:53:54 -0400 (EDT) |
| Message-ID: |
| <20120421.155354.769351486098851306.davem@davemloft.net> |
| Cc: |
| netdev-AT-vger.kernel.org |
| Archive-link: |
| Article, Thread
|
From: Pavel Emelyanov <xemul@parallels.com>
Date: Thu, 19 Apr 2012 17:38:58 +0400
> Attempt #4 with an API for TCP connection recreation (previous one is
> at http://lists.openwall.net/netdev/2012/03/28/84) re-based on the
> today's net-next tree.
>
> Changes since v3:
>
> * Added repair for TCP options negotiated during 3WHS process, pointed
> out by Li Yu. The explanation of how this happens is in patch #6.
>
> * Named constant for sk_reuse values as proposed by Ben Hutching.
>
> * Off-by-one in repair-queue sockoption caught by Ben.
All applied to net-next, nice work.
Please make the following fix for me. The option recovery code will
result in unaligned accesses, for example you'll do a byte aligned
get_user() for the u16 MSS object in many cases, and this will trap
and cpus such as sparc.
Either add a padding facility or pass more structured data into the
socket option.
Thanks.
(
Log in to post comments)