2006-10-04 18:02:47

by onovy

[permalink] [raw]
Subject: ip_conntrack_core - possible memory leak in 2.4

hi,

i have there MontaVista based router, with 2.4.17_mvl21-malta-mips_fp_le
kernel. I think, there is memory leak in ip_conntrack code. There are
eta 500 conntrack connection all the time. But after some day i get
"ip_conntrack: table full" in kmsg.
/proc/sys/net/ipv4/netfilter/ip_conntrack_max have 3072 value.
grep ip_conntrack /proc/slabinfo
ip_conntrack 3006 3250 384 319 325 1
^^ there are 3006 allocated conntracks
cat /proc/net/ip_conntrack | wc -l
30
^^ in table are only 30 lines.

Acording to this:
http://lists.netfilter.org/pipermail/netfilter-devel/2004-May/015628.html
i don't think, this is fixed in 2.4 tree, but i can't test it with newer
version.

Thanks
--
S pozdravem/Best regards
Ondrej Novy

Email: [email protected]
Jabber: [email protected]
ICQ: 115-674-713
MSN: [email protected]
Yahoo ID: novy_ondrej
Tel/Cell: +420 777 963 207


2006-10-04 18:11:12

by Stephen Hemminger

[permalink] [raw]
Subject: Re: ip_conntrack_core - possible memory leak in 2.4

On Wed, 4 Oct 2006 20:02:01 +0200
[email protected] wrote:

> hi,
>
> i have there MontaVista based router, with 2.4.17_mvl21-malta-mips_fp_le
> kernel. I think, there is memory leak in ip_conntrack code. There are
> eta 500 conntrack connection all the time. But after some day i get
> "ip_conntrack: table full" in kmsg.
> /proc/sys/net/ipv4/netfilter/ip_conntrack_max have 3072 value.
> grep ip_conntrack /proc/slabinfo
> ip_conntrack 3006 3250 384 319 325 1
> ^^ there are 3006 allocated conntracks
> cat /proc/net/ip_conntrack | wc -l
> 30
> ^^ in table are only 30 lines.
>
> Acording to this:
> http://lists.netfilter.org/pipermail/netfilter-devel/2004-May/015628.html
> i don't think, this is fixed in 2.4 tree, but i can't test it with newer
> version.
>
> Thanks

You paid for an expensive vendor kernel. Use their support, and
make them fix it.

--
Stephen Hemminger <[email protected]>

2006-10-04 18:50:29

by onovy

[permalink] [raw]
Subject: Re: ip_conntrack_core - possible memory leak in 2.4

Wed, Oct 04, 2006 at 11:09:57AM -0700 Stephen Hemminger wrote:
> You paid for an expensive vendor kernel. Use their support, and
> make them fix it.
Yes, you are true, but if this is upstream bug in 2.4, it should be fixed,
should not?
:
--
S pozdravem/Best regards
Ondrej Novy

Email: [email protected]
Jabber: [email protected]
ICQ: 115-674-713
MSN: [email protected]
Yahoo ID: novy_ondrej
Tel/Cell: +420 777 963 207

2006-10-05 20:07:08

by Willy Tarreau

[permalink] [raw]
Subject: Re: ip_conntrack_core - possible memory leak in 2.4


[netfilter-devel list CC'd]

Hello,

On Wed, Oct 04, 2006 at 08:02:01PM +0200, [email protected] wrote:
> hi,
>
> i have there MontaVista based router, with 2.4.17_mvl21-malta-mips_fp_le
> kernel. I think, there is memory leak in ip_conntrack code. There are
> eta 500 conntrack connection all the time. But after some day i get
> "ip_conntrack: table full" in kmsg.
> /proc/sys/net/ipv4/netfilter/ip_conntrack_max have 3072 value.
> grep ip_conntrack /proc/slabinfo
> ip_conntrack 3006 3250 384 319 325 1
> ^^ there are 3006 allocated conntracks
> cat /proc/net/ip_conntrack | wc -l
> 30
> ^^ in table are only 30 lines.
>
> Acording to this:
> http://lists.netfilter.org/pipermail/netfilter-devel/2004-May/015628.html
> i don't think, this is fixed in 2.4 tree, but i can't test it with newer
> version.

Well, I know several old 2.4 netfilter systems running around and which
process between 100 and 200 millions of sessions a day with peak hours
around 4000 sessions/s. They might have been rebooted twice in 3 years,
and they still work without a glitch. So I clearly don't think that the
problem mentionned above is present in plain 2.4. It might be a very
old bug in you rather old kernel, or one specific to some patches in
your distro's kernel (BTW, I would be surprized you wouldn't find
anything more recent than 2.4.17).

Best regards,
Willy

2006-10-05 21:56:00

by onovy

[permalink] [raw]
Subject: Re: ip_conntrack_core - possible memory leak in 2.4

> Well, I know several old 2.4 netfilter systems running around and which
> process between 100 and 200 millions of sessions a day with peak hours
> around 4000 sessions/s. They might have been rebooted twice in 3 years,
> and they still work without a glitch. So I clearly don't think that the
> problem mentionned above is present in plain 2.4. It might be a very
> old bug in you rather old kernel, or one specific to some patches in
> your distro's kernel (BTW, I would be surprized you wouldn't find
> anything more recent than 2.4.17).
If you look at that link, you will see man with same problem as i have at
2.4.26 kernel. I'm just trying to do some diffs and collect some more
information. I can't switch kernel in that device, because i'm not author of
that firmware.

--
S pozdravem/Best regards
Ondrej Novy

Email: [email protected]
Jabber: [email protected]
ICQ: 115-674-713
MSN: [email protected]
Yahoo ID: novy_ondrej
Tel/Cell: +420 777 963 207