[lug] e2fsck emergency

[D. Stimits]

-----Original Message-----
From:	Michael J. Pedersen [SMTP:marvin at keepthetouch.org]
Sent:	Monday, September 04, 2000 10:15 PM
To:	lug at lug.boulder.co.us
Subject:	Re: [lug] e2fsck emergency

On Mon, Sep 04, 2000 at 10:38:02PM -0700, D. Stimits wrote:
> An added question. I've decided if I must reinstall, I'm completely
> removing ext2. I want reiserfs, or any journaling system. However, is 
this
> possible to use during initial install? Will I be required to use ext2 on 
> root? With the SuperMicro i840, I need something far more recoverable. Do 
> any of the distributions now offer a journaling filesystem that I can
> install directly to?

Well, your choices here are fairly slim pickings. First things first, 
though:
Why can't fsck work on that partition?

It is so badly corrupted it can't believe it is ext2:

e2fsck: bad magic number in super-block while trying to open /dev/sda2

The superblock could not be read or does not describe a correct ext2
filesystem. If the device is valid and it really contains an ext2
....

My experience has been that if things die, simply doing 'fsck -y /dev/hda1'
will fix it quite nicely.

This must be a typo, or I have a different fsck. No -y is possible. This is 
on Redhat 6.2. Somewhere I am hoping to find an option to fsck that says 
"yes, this is ext2, and it is so badly corrupted, I want you to do what I 
say no matter how ridiculous, and not exit on me". However, it always 
exits. I have tried various options, including specifically the manual 
ones, totally non-automatic. Maybe I'm missing something, but the compile 
probably had 100 megs in the middle of change at the time of failure...it 
is bad.

If you do go for a reformat/reinstall solution, and insist on a journaling 
fs,
then ReiserFS is one of the best choices. However, you're still running 
beta
code if you do so. And the only distro I'm aware of which supports Reiser 
out
of the box is Mandrake.

Other miscellaneous questions:
How can you be sure it's the hardware? Is there any chance it could be
software related?

The i840 chipset is confirmed to have new IO-APIC problems. I briefly 
exchanged email with Alan Cox, and there is basically no hope if SuperMicro 
won't release info. I spoke, on the phone, with SuperMicro for many many 
hours, day after day...they just have no interest in linux anymore. They 
simply state that it is stable under NT. Unfortunately, it appears to have 
a similar problem under Win 2k, even if NT is stable.

The choice is to run with kernel option "noapic". This means all irq's are 
handled on cpu #0. It changes the behavior of some programs, especially if 
they use spinlocks. Normally they'd be avoided on single cpu; on dual 
running noapic, with only one cpu handling irq's, it makes things far less 
"predictable". Even when it does work, it makes the system more sluggish 
than in windows. Certain hardware operations are still risky even with 
noapic used. I asked Alan Cox if he knew anyone at Intel that might be 
interested in the problem, he said no. I have two choices here, if I want 
to get rid of this kind of problem. Either ask Linus Torvalds what to do, 
or get a new motherboard. SuperMicro has been so blatantly anti-linux, and 
the board still fails under win 2k (I use some studio animation software I 
can't get for linux), that I would prefer to simply dump SuperMicro. Even 
if I manage to get linux support fixed, Win 2K will still die...and we all 
know the possibility of getting microsoft to fix this when SuperMicro has 
no interest.

In the past I've spoke with SuperMicro about other problems, they were 
always 100% quality minded.  They apparently lost their product manager a 
couple of months back, and shoved work onto the other people. Since then 
linux has become officially and vehemently unsupported. I even offered to 
buy a $600 m/b to donate to someone who could do this work for free to 
them...they don't care.

Would you consider staying with ext2, if it is software related, and trying 
a
known to be stable distro? I'd recommend Debian if at all possible. I've 
been
using it personally for about a year now, and the only time I've ever had
downtime was for one of three reasons: Unstable kernel, hardware 
malfunction,
or hardware upgrade.

If it were a pure software problem, yes. But it is the nature of noapic 
under an SMP machine, so this won't help. If I remain using this 
motherboard, I feel I have to have a journaling filesystem as a hedge 
against the IO-APIC problems. What version of Mandrake does reiserfs become 
standard on, 7.0? 7.1? I will try it on a blank drive...but since this is 
ultra 160, and it has performed incredibly well, I don't want to abandon it 
(not to mention the cost). Here the trouble becomes finding a 64 bit pci 
slot or a built-in ultra 160 that isn't i840, isn't rdram, has an AGP slot, 
and isn't SuperMicro.

Should I find a job in the near future, I'll shell out for total 
replacement of ram, cpu's, motherboard, and a separate ultra 160 
controller, all to rid myself of SuperMicro i840 problems.

-----
Michael J. Pedersen
Get GnuPG at http://www.gnupg.org
My GnuPG Key Fingerprint: C31C 7E90 5992 9E5E 9A02 233D D8DD 985E 4E72 4A60
My GnuPG Public Key Available At: http://www.keyserver.net
 << File: ATT00000.att >>