netdev
[Top] [All Lists]

Re: [PATCH] Improve behaviour of Netlink Sockets

To: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
Subject: Re: [PATCH] Improve behaviour of Netlink Sockets
From: Pablo Neira <pablo@xxxxxxxxxxx>
Date: Wed, 22 Sep 2004 02:24:46 +0200
Cc: hadi@xxxxxxxxxx, "David S. Miller" <davem@xxxxxxxxxx>, netdev@xxxxxxxxxxx
In-reply-to: <20040922000503.GA13218@xxxxxxxxxxxxxxxxxxx>
References: <E1C8way-0000aH-00@xxxxxxxxxxxxxxxxxxxxxxxx> <20040919120249.GA5963@xxxxxxxxxxxxxxxxxxx> <414DF11C.1080505@xxxxxxxxxxx> <20040919215915.GB9573@xxxxxxxxxxxxxxxxxxx> <1095633569.1047.107.camel@xxxxxxxxxxxxxxxx> <20040919231734.GA10124@xxxxxxxxxxxxxxxxxxx> <1095647944.1046.206.camel@xxxxxxxxxxxxxxxx> <20040920025802.GA11567@xxxxxxxxxxxxxxxxxxx> <1095683660.1047.254.camel@xxxxxxxxxxxxxxxx> <414F1E12.6010808@xxxxxxxxxxx> <20040922000503.GA13218@xxxxxxxxxxxxxxxxxxx>
Sender: netdev-bounce@xxxxxxxxxxx
User-agent: Mozilla/5.0 (X11; U; Linux i686; rv:1.6) Gecko/20040528 Debian/1.6-7
Hi Herbert,

Herbert Xu wrote:

On Mon, Sep 20, 2004 at 08:14:42PM +0200, Pablo Neira wrote:
Here a link to the tool that I use to stress netlink sockets.

http://eurodev.net/~pablo/netlinkbench-unicast-1.0.tar.gz

Thanks for the link.  I'm afraid that your kernel module is simply
buggy.

First of all as I explained before the kernel must never wait.  It has
exactly the same effect as extending the receive queue length.

Secondly each of your user-space messages is producing a number of
replies.  This should be done as a dump operation.  If you do it as
a dump operation, then you will never get overruns because the kernel
never sends more than the user can handle.

I'll adapt the module to use dump, I still think that I can reproduce the problem.

thanks for the review, please stay tuned.

regards,
Pablo

<Prev in Thread] Current Thread [Next in Thread>