Re: kernel dump server
Re: kernel dump server
- Subject: Re: kernel dump server
- From: Peter Lovell <email@hidden>
- Date: Tue, 27 Dec 2005 11:53:35 -0500
Hi Derek,
what I'm seeing is that the client is not sending any arp request at all. It just starts out by sending UDP packets from 1069 to port 1069 on server, using the default router.
And looking at kdp_udp.c, I don't see any ARP resolution code. It just takes whatever is the default gateway (set at kdp_set_gateway_mac, called from rt_setgate in route.c) and sends the request there. SInce we're on a managed network, I'm not at all optimistic about having any configuration changes made on the router.
Thanks for this. I'm off to try talking to the network folks.
Regards.....Peter
On Tuesday, December 27, 2005, at 10:21AM, Derek Kumar <email@hidden> wrote:
>On Dec 26, 2005, at 7:16 PM, Peter Lovell wrote:
>
>> Hi all,
>> I'm setting up a kernel dump server, as described in technote 2118
>> (thanks Quinn).
>>
>> However, I'm having a problem in that the client seems to be unable
>> to find the server. Before banging on this some more in the
>> morning, I thought I'd just check to see if there is any change
>> needed operating in Tiger rather than Panther, with launchd
>> starting to replace the tasks previously handled by xinetd. I don't
>> think there should be, and the server does indeed seem to have a
>> UDP listening port 1069 as I'd expect. The client tries fifteen
>> times (or so) and then just times out.
>>
>> I suspect that there may be local switch/router issues related to
>> UDP (although both are on the same local network) but thought I'd
>> just check the possibility of changes for Tiger.
>I'll second Justin Walker's recommendations, and note a couple of
>known issues, in the event that they apply to your configuration:
>
>A late Panther update added the ability to bind the kernel debugger
>to a non-en0 interface via a kernel boot-arg; however, that also
>appears to have introduced a race where the debugger can occasionally
>match to the wrong interface; you should be able to work around this
>by adding the "kdp_match_name=en0" boot-arg (if you wish to debug
>over en0, which is the default). (Radar 3844652).
>
>Are you on a NAT-ted network? If, for some reason, the ARP resolution
>broadcast for the server fails, the debugger will attempt to route
>the crashdump packets via the cached default route (the debugger's
>"network stack" is a fairly minimalist, polled implementation, and
>doesn't query routing tables and so on, which may or may not be valid
>post-panic), and if the router doesn't forward packets arriving at
>the internal interface addressed to internal nodes, you may see a
>timeout. You can either try and determine why the resolution fails,
>or reconfigure your router as appropriate.
>A future update should address/ameliorate these issues. Feel free to
>contact me with a packet trace if you can't determine the source of
>the problem.
>
>Derek
>
>
>
_______________________________________________
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