[ale] ext3-fs error (RH 3.4.6-2)

[Michael B. Trausch]

On 11/30/2011 01:25 PM, Lightner, Jeff wrote:
> A couple of things:
> 
> 1)  You're not using RH 3.4.6-2 - the message tell you your kernel
> was copiled by that version of gcc.   To see the version of RH you're
> running do "cat /etc/issue" and/or "cat /etc/redhat-release".

Indeed.  2.6.9 was used for RHEL4 from the looks of it, so it's likely
that he's using that (which is ending support soon anyway).

> 2)  The way RedHat does things is it releases a base package from
> upstream then appends it own versioning to that so 2.6.9-42.ELsmp is
> NOT the same as 2.6.9 on any other system as it may have backported
> bug and security fixes in it.   (That being said kernel is handled
> differently than many other packages so you can actually get kernel
> updates from the RedHat yum repositories that might be newer than
> 2.6.9x.

This is generally true regardless of the distribution; most
distributions patch the kernel in some way.  One reason that I prefer
using upstream, vanilla kernels is that it's easier to get support for
them than for distro-kernels (at least, IME, YMMV).

> You should NOT attempt to download and compile a newer
> kernel manually as it would no longer be RHEL supported at that
> point.

Only while the locally-compiled kernel is actually running.  If you have
a problem with the kernel, the first thing to do is to determine if it
is present in the vanilla kernel; if so, file the bug there and file a
bug with the distribution to reference the upstream bug.  Otherwise, if
you cannot reproduce, you have viable information that you can give to
the distributor to say "this problem exists in your kernel version x.y.z
pl eleventyone-foo but not upstream release x.y.z" and that is at least
something to go on.

> If you're using RHEL and paying a subscription fee you can call them
> for support.  If you're NOT paying for a subscription fee and using
> them for support you might want to consider moving to CentOS which is
> a binary compile of RHEL sources.  It doesn't require subscription
> fees but also doesn't have a support number.   (Of course you
> wouldn't want to worry about this until you've solved your base
> issue.)

This would be the one case where it's likely easier to get support for
the distro kernel, though I'd still be inclined to troubleshoot as far
as I can before I start asking for support from the distributor, in the
interest of reducing the amount of back-and-forth communication I have
to do.  What can I say... I'm lazy!

> My thought is as Mike said that it is likely an issue with the disk
> controller or disks themselves.

Possibly, though even so, the kernel shouldn't be attempting to deref a
NULL pointer unless the kernel image itself is somehow corrupted or
modified.  The thing is that in that case, it'd be very likely that the
kernel wouldn't work at all (and in what I'd call a safe/secure system,
it shouldn't because it should be somehow meaningfully signed, but
that's neither here nor there).

If the kernel's not corrupt and there is indeed a problem with the disk
controller or the disk itself, it shouldn't be able to cause the kernel
to crash by deref'ing a NULL pointer; the kernel should be able to catch
such an issue and freeze the FS to save it from any further problems.  A
panic would be warranted, IMHO, but with hopefully a more meaningful
message.

	--- Mike

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 729 bytes
Desc: OpenPGP digital signature
Url : http://mail.ale.org/pipermail/ale/attachments/20111130/228b950d/attachment.bin