[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: fc2 issues -- Linux 2.6 issues (not Fedora-specific)



From: Steven Pritchard <steve@silug.org>
> More likely the drive geometry (either in the BIOS or the partition
> table) is a little wacky.  Apparently some of the old compatibility
> cruft was removed from 2.6, and now that's biting people in the ass on
> a regular basis.

- Linux 2.6 Issue

Yep.  Sometimes following specs to the letter is the "Wrong Thing To
Do(TM)".    Mandrake Linux 10 users are running into the same issues.

Linux 2.6 now reads disk geometry from the Extended Int13h BIOS.
This is just following the specs at its best.

- NT Is It's Own Problem

Unfortunately, not only does NT/Win (including XP) continue to read the
Legacy Int13h BIOS, it will simply _ignore_ it in many cases.

Where most people are running into trouble is when their BIOS is
using a non-standard translation.  Then NT/Win just goes off and does
its own 'thang, totally creating a "DOS disk label" (Microsoft calls
this a "MBR disk label" or a "Basic Disk") with a 255/63 geometry,
or similar.

- Poor BIOSes with Non-standard Translation

E.g., IBM ThinkPads use a 240/63 heads/sector translation.

Old Linux 2.4 would read the Legacy Int13h BIOS, see the 240/63
geometry, then inspect the DOS disk label and recognize it was written
with 255/63 by NT.  It would the use the latter.

The new Linux 2.6 kernel reads the Extended Inth13h BIOS and gets
16/63 (common disk geometry for legacy compatbility with pre-1997
BIOSes), and just uses that.  That's why people are seeing partitions
not landing on cylinder boundaries because 16/63 and 240/63 do NOT
line up.

- Nothing New, I've Seen It With Dual-Booting NT With Itself

Ultimately this is a NT-created problem.  I know because trying to
get NT 4.0 and NT 5.x, and even some different versions of NT 5.x (2000,
XP, 2003) to install on the same system is  a test of fruitition when it
comes to such disk geometry.  Microsoft's workaround for this issue is
to use the Logical Disk Manager (LDM) disk label (Microsoft also calls
this a "Dynamic Disk").

LDM solves a lot of issues, including storing its own geometry in the
table, as well as Registry-SID information so you can read NTFS
filesystems from another version of NT that did not create the NTFS
filesystem (and people think writing to NTFS is a risk with Linux?  Ha! 
You're not supposed to write to a NTFS filesystem _unless_ it was
created by the NT that is running, because of the Registry-SID tie-in).

- Supporting LDM disk labels in Linux

LDM has been partially documented by the Linux-NTFS project since
2001-2002.  The Linux kernel, since 2.4.8 (I believe?) has shipped with
LDM disk label support.  Using LILO (which just blindly stores the
"offset" to boot), you _can_ boot a Linux partition on a LDM disk
labelled disk.  This requires you to partition under NT 5.x (or possibly
a newer version of Partition Magic).

Unfortunately, other than to read LDM disk labels, there are no
user-space tools for Linux to modify LDM disk labels.  It would be nice
to at least be able to create "Simple Volumes" for Linux inside of a LDM
disk label in any distro installer.  Additionally, GRUB does not have
LDM disk label support, so you can not use it (unlike LILO, which is
"dumb, boot this address," GRUB is far more intelligent, but it requires
additional support for reading LDM disk labels).

I would like to see this change.  LDM is going to be with us for a
_long_time_ as Microsoft recommends it.  Yes, it would be  better if NT
just had more intelligent disk geometry intelligence.  But LDM is
Microsoft's workaround, and we're all going to be forced into it.  But
it's not so bad, because SIDs are stored in hidden areas, which would
only help the Linux-NTFS driver "safely write" to NTFS filesystems --
just like NT 5.x uses as well (because it is _not_ safe to write to a
NTFS filesystem, even with NT 5.x itself, if it did not create the NTFS
filesystem it is writing to -- _unless_ it has the hidden LDM info).


-- 
Bryan J. Smith, E.I. -- b.j.smith@ieee.org



-
To unsubscribe, send email to majordomo@silug.org with
"unsubscribe silug-discuss" in the body.