[Netkit.users] Connection tracking in netkit

[Massimo Rimondini]

Dear Marco,


>> yes, you are right: I can see this happening on a small sample lab.
>> The conntrack table is apparently empty, even though the network
>> operation is correct and confirms that something is expected to be
>> observed in the table. /proc/sys/net/netfilter/nf_conntrack_count also
>> reads 0.
>>
>> Looks as if you have come across what resembles a kernel bug, although I
>> could not find any other traces of users signaling a similar issue.
>> Things being such, I'm afraid the most concrete suggestion to circumvent
>> the problem would be to compile a different kernel. Instructions are
>> provided in the kernel README file, which is also available here:
>> http://wiki.netkit.org/download/netkit-kernel/README.
>>     
>
> I followed the instructions and compiled a later kernel (2.6.26.6 rather than 
> 2.6.26.5, to be able to reuse the existing config as easily as possible). 
> I followed the instructions to build the new kernel (including patching) and 
> the compilation was successful. I did it on a multilib 64-bit Gentoo system, 
> using ARCH=um SUBARCH=i386.
>   

Good. The build environment is very well suited.

> But now when I try to start a virtual machine I get:
>
> ...
> Checking for the skas3 patch in the host:
>   - /proc/mm...not found: No such file or directory
>   - PTRACE_FAULTINFO...not found
>   - PTRACE_LDT...not found
> UML running in SKAS0 mode
> Adding 11051008 bytes to physical memory to account for exec-shield gap
> *** buffer overflow detected ***: /home/marco/netkit/kernel/netkit-kernel 
> terminated
> ======= Backtrace: =========
> /lib32/libc.so.6(__fortify_fail+0x48)[0xf76c1418]
> /lib32/libc.so.6[0xf76bf460]
> /lib32/libc.so.6[0xf76bed07]
> /lib32/libc.so.6(__snprintf_chk+0x34)[0xf76bebf4]
> /home/marco/netkit/kernel/netkit-kernel[0x807496e]
> [0x0]
> ======= Memory map: ========
> 08048000-083b8000 rwxp 00000000 03:08 4243474                            
> .....
>   

Huh? This sounds like another kernel(+libc) bug... :-(

I think the fastest way of working this around is to try a different
kernel release instead. Although the configuration may slightly differ,
it is reasonably safe to assume the default for newly appeared
parameters. Patches will hopefully still apply (I suggest keeping MPLS
out of the game, though, if it is not a priority for you) and, if they
do not, you should still come up with a mostly usable environment.

I hope you have better luck with the next attempts. If not, feel free to
come back for more clarifications.

Regards,
Massimo.

-------------- parte successiva --------------
Un allegato HTML ? stato rimosso...
URL: http://list.dia.uniroma3.it/pipermail/netkit.users/attachments/20100326/496522b5/attachment.htm