Question: is it normal for ipv6 module to have many refs while there is few
ipv6 connections ?
My system is up for 3 days now. I have been using ACPI sleep and _ifconfig to
alter the ip address_.
lsmod gives:
ipv6 235904 34
and lsof -i6 only shows two ports listening on ipv6.
Even when I close all tcp (including v4) sockets, ipv6 still has around 10
refs in lsmod.
Ten days ago (with -test7) I used to have a broken script for sshd that tried
to bind to ipv6 every 20 mins. The result was that the count for ipv6.ko
would reach as high as 900 w/o any bound sockets.
Does that constitute a leak?
In other news, I haven't managed to trace the bug behind 'lockd', but I guess
it's time I ask about it. After about 4 days uptime [1], both in -test7 and
-test9, lockd BUGs at 'module.h, 296' . After that, nfs becomes unusable[2]
and I have to reboot the kernel. The BUG is hit while I try to mount a nfs
partition, which is being served by another 2.6 machine [3].
I can only report that this has happened. It may be related to the ipv6 leak,
I guess..
[1] Every day, I switch from my home net to my work's net and back. I use
ifconfig new_ip for that.
[2] lockd freezes, the 'mount' process gets <defunct> and nfs cannot be
unloaded.
[3] At work, I mount nfs partitions to a *nix environment. I have not yet seen
lockd freeze there.
On Tue, 4 Nov 2003 18:14:24 +0200
"P. Christeas" <[email protected]> wrote:
> Question: is it normal for ipv6 module to have many refs while there is few
> ipv6 connections ?
> My system is up for 3 days now. I have been using ACPI sleep and _ifconfig to
> alter the ip address_.
> lsmod gives:
> ipv6 235904 34
> and lsof -i6 only shows two ports listening on ipv6.
> Even when I close all tcp (including v4) sockets, ipv6 still has around 10
> refs in lsmod.
This is normal, ipv6 is not safe to unload as a module and therefore
the module reference count is always positive to prevent rmmod ipv6
from succeeding.