QUESTION: Why might Linux suddenly stop replying to pings for no apparent reason?

QUESTION: Why might Linux suddenly stop replying to pings for no apparent reason?

[Terry Phelps]

I have this strange recurring problem with SEVERAL machines, all
running the Oracle "Unbreakable Enterprise Kernel", which is based on
the 3.0.16 kernel.

Here is a quick description, while I still have your attention:

I have server S1, and two desktops, D1 and D2, separated by a router.
The D1 and D2 boxes are side by side, on the same IPv4 subnet,
different from S1's subnet. Maybe once a day, or oftener, I find that
D1 cannot ping S1, but D2 can. There are many possible causes for
that, of course, BUT:

I can SSH to S1 from D2, and S1 can ping both D1 and D2 just fine.
TCPDUMP shows that the ICMP request packets from D1 ARE arriving at
S1. S1 is siimply not replying!

Even stranger:
>From S1, I can traceroute to D2, but cannot traceroute to D1. A
traceroute to D1 gets an ENETDOWN returned from sendto(). But there is
only one NIC in the S1, and it certainly isn't down!

One more thing: If I enter "ip route flush cache" on S1, the problem
clears up immediately.

And another: If I leave D1 pinging S1 every 5 seconds, say, the
problem will NEVER clear up by itself. But if D1 stops pinging S1 for
a few minutes, it works again!

No, there's no firewall or selinux running on any machine involved,
and no firewall between the boxes.

I'm totally confused. Can anyone suggest what to look at?

Re: QUESTION: Why might Linux suddenly stop replying to pings for no apparent reason?

[richard -rw- weinberger]

On Wed, Jul 18, 2012 at 4:42 PM, Terry Phelps <tgphelps50@gmail.com> wrote:
> I have this strange recurring problem with SEVERAL machines, all
> running the Oracle "Unbreakable Enterprise Kernel", which is based on
> the 3.0.16 kernel.

Ask Oracle.

-- 
Thanks,
//richard

Re: QUESTION: Why might Linux suddenly stop replying to pings for no apparent reason?

[Terry Phelps]

On Wed, Jul 18, 2012 at 11:06 AM, richard -rw- weinberger
<richard.weinberger@gmail.com> wrote:
> On Wed, Jul 18, 2012 at 4:42 PM, Terry Phelps <tgphelps50@gmail.com> wrote:
>> I have this strange recurring problem with SEVERAL machines, all
>> running the Oracle "Unbreakable Enterprise Kernel", which is based on
>> the 3.0.16 kernel.
>
> Ask Oracle.

Is there some way to "ask Oracle", in which my question will get to
someone who can possibly help me? I've tried contacting "Oracle Linux
support". They're hopeless. I've posted my question to Oracle forums,
and even general Linux forums, but all I get is boilerplate networking
advice (check ifconfig, check your routes, check your switches, check
autonegotiation, etc.) that doesn't address the specific symptoms I'm
seeing.

If I should be posting my question to somewhere other than the
linux-kernel mailing list, please advise me where that might be. It
sure looks like something in the kernel isn't working properly
somewhere.

Re: QUESTION: Why might Linux suddenly stop replying to pings for no apparent reason?

[Eric Dumazet]

On Wed, 2012-07-18 at 11:42 -0400, Terry Phelps wrote:
> On Wed, Jul 18, 2012 at 11:06 AM, richard -rw- weinberger
> <richard.weinberger@gmail.com> wrote:
> > On Wed, Jul 18, 2012 at 4:42 PM, Terry Phelps <tgphelps50@gmail.com> wrote:
> >> I have this strange recurring problem with SEVERAL machines, all
> >> running the Oracle "Unbreakable Enterprise Kernel", which is based on
> >> the 3.0.16 kernel.
> >
> > Ask Oracle.
> 
> Is there some way to "ask Oracle", in which my question will get to
> someone who can possibly help me? I've tried contacting "Oracle Linux
> support". They're hopeless. I've posted my question to Oracle forums,
> and even general Linux forums, but all I get is boilerplate networking
> advice (check ifconfig, check your routes, check your switches, check
> autonegotiation, etc.) that doesn't address the specific symptoms I'm
> seeing.
> 
> If I should be posting my question to somewhere other than the
> linux-kernel mailing list, please advise me where that might be. It
> sure looks like something in the kernel isn't working properly
> somewhere.

Before saying that, can you try a pristine linux kernel ?

For example linux-3.4.5, or even better the git tree, so that
it will be easier to add knobs and patches.

Obviously we dont know what Oracle added to their kernels, and
dont want to know.

Re: QUESTION: Why might Linux suddenly stop replying to pings for no apparent reason?

[Alan Cox]

> If I should be posting my question to somewhere other than the
> linux-kernel mailing list, please advise me where that might be. It
> sure looks like something in the kernel isn't working properly
> somewhere.

If you can duplicate the problem with an *upstream* kernel then yes it
is. If you only see it on an Oracle kernel then talk to Oracle.

Alan

Re: QUESTION: Why might Linux suddenly stop replying to pings for no apparent reason?

[Terry Phelps]

On Wed, Jul 18, 2012 at 12:14 PM, Alan Cox <alan@lxorguk.ukuu.org.uk> wrote:
>> If I should be posting my question to somewhere other than the
>> linux-kernel mailing list, please advise me where that might be. It
>> sure looks like something in the kernel isn't working properly
>> somewhere.
>
> If you can duplicate the problem with an *upstream* kernel then yes it
> is. If you only see it on an Oracle kernel then talk to Oracle.
>
> Alan

Alan,
Thank you for your help. I am not a kernel hacker by any means, but
will try to do what you suggest.
Should I clone the current kernel git tree, and build the latest
kernel? Or build from some other point in time?

Terry

Re: QUESTION: Why might Linux suddenly stop replying to pings for no apparent reason?

[Alan Cox]

On Wed, 18 Jul 2012 16:30:59 -0400
Terry Phelps <tgphelps50@gmail.com> wrote:

> On Wed, Jul 18, 2012 at 12:14 PM, Alan Cox <alan@lxorguk.ukuu.org.uk> wrote:
> >> If I should be posting my question to somewhere other than the
> >> linux-kernel mailing list, please advise me where that might be. It
> >> sure looks like something in the kernel isn't working properly
> >> somewhere.
> >
> > If you can duplicate the problem with an *upstream* kernel then yes it
> > is. If you only see it on an Oracle kernel then talk to Oracle.
> >
> > Alan
> 
> Alan,
> Thank you for your help. I am not a kernel hacker by any means, but
> will try to do what you suggest.
> Should I clone the current kernel git tree, and build the latest
> kernel? Or build from some other point in time?

I'd grab a modern release kernel - the current 3.4 version is probably
best. Building off the git tree tends to add other "interesting"
variables you don't need.

Alan