Re: KPI replacement for sodisconnect?
Re: KPI replacement for sodisconnect?
- Subject: Re: KPI replacement for sodisconnect?
- From: Brian Bergstrand <email@hidden>
- Date: Wed, 18 May 2005 10:36:27 -0500
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On May 18, 2005, at 12:39 AM, Josh Graessley wrote:
>
> sock_shutdown should work fine to disconnect a TCP socket.
>
Yes, I thought of that right after I sent the email. Always seems to
be the case. :)
> There is a KPI hands-on session at WWDC.
>
> Kernel Extensions on Thursday from 2PM to 3:30PM in the Performance
> & Tools lab.
>
Great! I will be there.
> When you say that you are "tracking TCP state changes via access to
> the raw TCP header", do you mean you look at the tcp control block?
> There is no way to do that on Tiger. You can receive notification
> of a number of state changes through a socket filter.
That's what I was afraid of. I was hoping Tiger would not make the
headers opaque as sockaddr (and others) was left in tact, so this is
going to cause some major work.
> What exactly are you trying to accomplish?
Well, w/o going into too much detail, our product uses a complete
abstraction layer for kernel services (malloc, threads, locks, etc)
- -- included in that abstraction is an aysnc IP wrapper model that
provides an abstraction over the socket interface -- all of this is
done to make porting easier to other platforms, and it's actually
making the move to Tiger much easier, as our product has a very large
number of kexts.
The IP abstraction model expects to be able to track TCP transition
states. I'm going to get into the TCP conversion today, so I'll have
a better idea of what Tiger breaks tomorrow. It already sounds like
this is going to be a pain though.
The ability to separately specify whether send/receive should upcall
has caused some headaches already -- Tiger is all or none.
Thanks for you replies, I appreciate it.
>
> -josh
>
> On May 17, 2005, at 10:08 PM, Brian Bergstrand wrote:
>
>
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>> Yes, I didn't mean to imply a re-use of a TCP socket -- just if
>> sock_connect(INADDR_ANY) will cause a disconnect of the current
>> connection. Our current model issues a disconnect because it needs to
>> hang onto the socket for a while longer before actually closing it. I
>> suppose I could do a retain before the close to keep the socket
>> around.
>>
>> I also noticed that soabort is missing a KPI replacement -- we were
>> using this to abort a connection in progress -- although retain/close
>> should work there too.
>>
>
Brian Bergstrand
<http://www.bergstrand.org/brian/> PGP Key ID: 0xB6C7B6A2
-----BEGIN PGP SIGNATURE-----
Version: PGP Desktop 9.0.1 (Build 2130)
iQA/AwUBQothAHnR2Fu2x7aiEQLd9gCgvNKlkLNlwzVwOiG/mEew2SdXQsgAoJxL
roqy4BISdkzv5xG8IVXae4qe
=AQNs
-----END PGP SIGNATURE-----
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Darwin-kernel mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden