Re: rtfree: 0xc741ee88 has 1 refs

Hi,

Per olof Ljungmark wrote:
> Stefan Lambrev wrote:
>> Hi,
>>
>> Sorry to reply to myself, but I found that the problem exist only if 
>> the GW is carp interface, e.g. 10.1.1.1 sits on carp0 on default GW.
>> I'm still testing how to reproduce this in my test lab and will fill 
>> a PR.
>>
>> Stefan Lambrev wrote:
>>> Hi,
>>>
>>> I see rtfree: 0xc741ee88 has 1 refs with freebsd releng_7 (i386) 
>>> from today.
>>>
>>> I think it's easy reproducible. What I have is:
>>>
>>> releng_7 (10.1.1.2) -> default GW (10.1.1.1)
>>> on default GW I have route to 10.10.1.1/24 -> 10.1.1.3
>>>
>>> so everytime when 10.1.1.2 try to contact someone from 10.10.1.1/24 
>>> I see:
>>> rtfree: 0xc741ee88 has 1 refs
>>>
>>> if I add direct route on 10.1.1.2 to 10.10.1.1/24 through 10.1.1.3 
>>> the message will go away.
>>>
>>> Should I ignore this msg for now, or should I expect kernel panic 
>>> soon? :)
>
> Just FYI, I see this on a few boxes including the 7-BETA2 I'm writing 
> this on. None of them has a carp interface though. What I find 
> interesting here is that none of them are able to run a SMP kernel 
> without crashing (no panic, they're just frozen completely).
>
> Perhaps it is a coincidence, I don't know, but I am very interested in 
> your findings and have testbeds if you need.
>
> --per
In the begging I thought this is very easy reproducible, but it isn't :)
In my situation the IP of the default GW sits on carp interface (the 
host have nothing to do with carp)
and when the host receive icmp redirect messages they actually are send 
from the IP of the netwrok card of the GW (not the IP that sits on carp0)
then the host things that those icmp type 5 are spoofed and just ignore 
them and do not add shorter route.
In this situation after reboot the hosts start to moan about rtfree, so 
I played with routes and settings, added manually routes and now when I 
restore everything to previous state
I do not see anymore rtfree warnings. So may be once the route is 
created manually and then removed, wrong call of rtfree is not triggered 
anymore.
I'll reboot the host latter to see if this easy reproducible at least on 
reboot, and if it is, I'll compile debug kernel and will run backtrace.

The other thing that bother me is that icmp redirects are not send from 
the carpIP, but from the real interface's IP? Isn't this a bug, or it is 
normal behavior?

-- 

Best Wishes,
Stefan Lambrev
ICQ# 24134177