[conspire] Comments on setting up disk partitions.

Rick Moen rick at linuxmafia.com
Thu Apr 9 11:06:18 PDT 2009

Quoting Paul Zander (paulz at ieee.org):

> Rick,
> Thanks for your patience.
> To clarify things. There is a legacy limitation of 4 real disk
> partitions. (I am sure that 4 was once seen as a "big number".

Right.  Precisely so.

> Once fdisk has been properly executed, the effective result is sda1,
> sda2, sda3, sda5, sda6,

Well, it _can_ be.

One valid configuration is to eschew putting primary partitions on a
disk entirely, allocate the entire space to a /dev/sda1 extended
partition, and then make data-bearing "logical drive" filesystems
/dev/sda5, /dev/sda6, etc. within that space.  The point is that you
have a hard physical limit of four entries in the MBR partition table.
You can use as few or as many of those four entries as you want, and
(optionally) one of them can be created with the "extended" attribute
to permit creation within that space of "logical drives".

Some partitioning utilities (such as, I think, /sbin/cfdisk) suppress
display of extended partiions if present, and create/handle them
implicitly.  With /sbin/fdisk, everything is (sometimes mind-numbingly) 

Just for your broader perspective, and not relevant to your current
problem:  The type of partition table described, often dubbed the
"IBM/Microsoft" design, originated with MS-DOS and is essentially
unchanged from it -- _and_ is absolutely _not_ the only way to
organise a disk.

For example, MacOS has its own partition-table design, the BSD family 
and Solaris have theirs (dubbed a "disklabel" system, with the
partitions referred to as "slices"), and Intel invented one at the time
they created the Itanium CPU[1] called the GUID Partition Table, part of
their EFI (Extensible Firmware Interface) spec that they intend to
replace the antique PC BIOS.  Linux partitioning tools understand and can
manage _all_ of those partition types, by the way.

(MS-Windows tools, by contrast, understand only the IBM/Microsoft

The only one of those you're likely to come across is the GUID Partition
Table, _if_ you start using extremely large hard disks on Linux, since 
the IBM/Microsoft design has size limits.

>  Not an issue as long as you are mindful when defining the extended
>  partitions and don't look for sda4.

Issues do come up.  ;->  For example, people expect non-destructive
partition-management software (Partition Magic and such) to be able to
move partitions around, swap their order, and so on, which might not be
possible if one of the partitions you want to move is a logical drive
and another is a primary partition.

> I obviously want to share some files, for example, Thunderbird stores its data files to be accessed by which ever OS is running.  

> In previous email, there was a link telling how to tell Thunderbird
> what directory to use instead of the default.  What format for a
> shared partition?

Yeah, dunno.  You might end up having to experiment.

FAT was the old answer.  It's a sucky, fragile, slow,
fragmentation-prone, space-wasting filesystem[2] that can't handle
metadata very well, but at least it's fully documented and not the
object of still-unfinished reverse-engineering the way NTFS is.

You could try ext3, with MS-Windows being retrofitted with the
open-source installable filesystem (IFS) driver for that purpose.
Or you could try NTFS, and use the Linux ntfs-3g driver.

> ...VFAT or FAT32....

"FAT32" is the Microsoft-world name for 32-bit FAT, implicitly with a
1995 kludge to add long-filename support (hiding that data in extra
directory entries with the "volume label" attribute set).  "vfat" is the
Linux-world name for the same thing (there being a fileystem-support
kernel module, and mount-command type, of that name).

[1] Itanium was Intel's market-failing first attempt at a 64-bit CPU,
the design that the wags at http://www.theregister.co.uk/ instantly 
dubbed "Itanic".

[2] Understandable when you realise it's a minor variation on CP/M's 
_floppy_ disk format, and as such was intended for 180 kilobyte
filesystems on machines maxing out at 64 kilobytes of RAM>

conspire mailing list
conspire at linuxmafia.com

More information about the conspire mailing list