On Sat, 12 Jun 1999, Raider wrote:
> On Thu, 10 Jun 1999 [EMAIL PROTECTED] wrote:
> > > srwx----w- 1 13869 8736 892155456 Jan 9 1987 #61186=
> > > cr----Sr-- 1 11651 9510 59, 145 Sep 4 1971 #61187
> > > b-wx--S--t 1 5441 39578 154, 149 Apr 22 2024 #61189
> > > c---r-SrwT 1 7963 9766 55, 116 Jun 18 2036 #61196
> > This is obviously a y2k problem :-) ^^^^^^^^
>
> Yup. This is what I first thought ;-)
> But here is a bigger problem. I know I run old software.
rh4.2
> is obsolete from a long time. But I don't have the chance to get a
newer
> distribution.
I ran slackware 3.0 with kernel 1.3.18 until last year and had no
trouble with ext2 except hardware problems.
If you still have a 486, check if you have a CMD640 chip on the
motherboard or ide board. The CMD640 VLB will make garbage of your HD
unless you tell the IDE driver to take evasive action:
ide0=cmd640_vlb
I have been bitten by this. Eventually it gets so bad you have to
format the hd and start over. Even mkfs can't work until you format it.
More recent versions of the IDE driver can detect the PCI version, but
not the VLB. There are some other IDE interface chips that cause
similar problems. Have a read of /usr/src/linux/Documentation/ide.txt.
A kernel significantly older than the hardware _could_ cause a problem
like this: you can't program around broken hardware that didn't exist
when you wrote the code. :-). If RH 4.2 uses a 2.0.x kernel, you should
be able to drop 2.0.36 in on top of it and not do any harm. Might even
do a power of good.
I dropped 2.0.30 on top of 1.3.18, and the only things that broke were
ps and top. I don't recommend switching major versions like that, but
for the same major version it should work fine.
> The problem is that e2fsck should check for these
inconsistences.
> there is no '5441' uid and no '39578' gid. And there can't be '154,
149'
> a valid lenght, and dev files (with minors and majors) are only in the
> /dev/ directory AFAIK. Also, in the permissions field... what is that
't'
> or 'T'? These problems should be corrected. Even that louzy scandisk
> does it.
> So far the fastest way to recover is to keep the more important
> stuff in another partition, and the rc files and the rest is backed on
> another partition as well. Lucky me I have from the old computer a
> running Linux (minimal). So I can switch to the old one, format the
> partition with problems, and restore the fs. Than reboot. This takes
> time. And is annoying. And it's so windowish. I have to reboot in
order
> to make changes valid. I hate it! And the thing that drives me mad is
> that windoze didn't crash lately.
>
> > Last time something like this happened to me, the DMA on the
motherboard
> > failed soon after (done anything with floppies, lately? See if they
> > still work) and I had to build this junk-pentium 66. I am still
using
> > the same hd, though.
>
> I had a 486. And I got lucky and made a major upgrade - mb,
proc,
> new hdd, now memories. The rest is the same. And the configuration
> worked well. There was a problem with the partitions. Because I kept
> making the dos partitions in Linux. I did the dos partitions with dos'
> fdisk and all was well - I know in the documentation that comes with
linux
> fdisk it is said to do this. Than that hard drive developed bad
sectors
> which made unavailable a couple of partitions. I had a lot of troble
here
> because those bastards from the service not only that they didn't solve
> the bad sector problem (replacing the hard drive) but they handeled it
bad
> and that seal (with 'warrnty void if removed') was thorn. After some
> arguing they gave me a new hard drive. And I check it constantly for
> problems. So far that is the only problem. If it was something
hardware
> related there should be problems on other partitions.
>
> > If you're _sure_ it's not hardware, the pattern of damage is not
> > inconsistent with the fs being written on by some other OS. Fat fs's
> > don't give a rat's ass what size the partition is. The size of the
fs
> > is determined by entries in the partition boot record, and if that is
> > inconsistent with the partition size, they may well write all over
some
> > other partition, starting with the adjacent one. dosfsck might catch
>
> fdisk (linux) says that every partition (besides the first)
> overlaps the next one. But I chacked the numbers and there is no
There are 2 sets of numbers in the partition table, a 3 byte chs thing,
and a long int bogical block address. The chs thing is rather limited
in what it can describe, especially if you don't (or can't) use "LBA"
mapping in the BIOS. Fdisk checks both, and complains if there is any
inconsistency. What does cfdisk say?
> problem. The windoze way would be to back up the more important data
and
> redo all over again the partitioning. But I can't. And as I said
there
> is no visible overlapping. Also it looks like Linux is messing up the
> stuff. I mean 2 overlaps 3 and so on. With 1 being the first
partition.
> And the first partition is dos. Than 2 is ext2. Than 3 is dos
extended.
> Can be Linux the trouble maker here? Again, I know that fat has the
most
> important part at the beggining - the fat. That would mean that the
other
> 'disks' from windoze should have problems as well. And they don't. I
> should add that I don't know how does ext2 looks from a phisical point
of
> view.
It's a hashed filesystem, split into sections or block groups, each of
which has part of the directory and available space structure. I don't
know where there is any good doco on it.
> So I receive over and over again messages like 'inode xxx has
> wrong block count. clear?' and 'inode xxx has dtime set'. I don't know
> what that means - this is frustrating. Is there a place where I can
get a
> good description of e2fs? I'm interested in other fs descriptions. I
> can't spend too much time now, but I'll do the read as soon as I finish
my
> exams. Oh! And after clearing that stuff (I lose a lot of files in
the
> process) I run again e2fsck (this is a reflex build up from the time
when
> I was using scandisk and it ain't too reliable). I rerun it exactly
after
> first has finished. And I receive that files are cross-linked. So
after
> running 2 or 3 times e2fsck there aren't more errors reported. Can
e2fsck
> be so louzy written so it can't see all the errors first time, or it
makes
> the new ones?
> The errors are on the whole partiton. I don't think there can
be
> any problem because of windoze. And the only partition that linux
fdisk
> doesn't say it overlaps the next one is the first. And /dev/hda1
should
> be the one which does the overwriting.
>
> > this, at least it seems to report files and clusters out of range,
but
>
> I don't think rh4.2 had dosfsck. And using a new version can
be
> incompatible with the kernel.
IIRC, RH 4.2 had a 2.0.x kernel and libc 5. If so, slackware stuff up
to slackware 3.6 should work, if that is still on the slackware ftp
sites. RH 5.0 doesn't seem to have dosfsck, either?! I am really
getting to like slackware. :-). dosfsck isn't very big. I can gzip,
uuencode and mail you a libc5 binary, or dig out the source and mail you
a tar.gz, if you like. Same for bpe.
>
> > to be really sure I would use bpe on /dev/hda1 and see what is in
byte
> > 0xd (sectors per cluster), 0x16-17 (fat sectors), and 0x20-23
(sectors),
> > remebmbering that they are little-endian short and long ints (least
> > significan byte first). Fat sectors counts both fats, so divide it
by
> > 2, multiply it by 256, then by the sectors per cluster, and that is
how
> > many sectors windows thinks it has to write on, not counting the FAT
and
> > boot sector.
>
> Couldn't find bpe...
You could use dd to copy the first block, and dump it with od or
hexdump or so. Or I can mail you bpe. It looks to me like _something_
is writing garbage on /dev/hda2.
> But I don't normally use /dev/hda1. So it is a 300m partition, so
> it can let the linux root partition start before that 1k cylinder. On
it
> there are only the stuff that belong to the windoze directory, and the
> stuff that are too dumb to install some other place than c:. This
means
> that partition is more than half empty, and that is not used very
often.
>
> Raider
> --
> ``Liberate tu-temet ex inferis''
>
Ace the exams!
Lawson
___________________________________________________________________
Get the Internet just the way you want it.
Free software, free e-mail, and free Internet access for a month!
Try Juno Web: http://dl.www.juno.com/dynoget/tagj.
