Re: Root Drive Mirroring and LVM.

On Tuesday January 27, atossava@cc.helsinki.fi wrote:

Sorry about the crossposting.

I wrote on the Yellow Dog Linux list when somebody asked about software
RAID on YDL about my experiences with it:

quoted

quoted

The one really big gotcha is that the Macintosh partitioning scheme
can't tell the Linux kernel that certain partitions are to be
considered "Linux RAID autodetect" (as in x86 using the DOS partition
table type 0xfd). This means that you can't boot a Mac Linux system
directly from RAID because the kernel won't be able to autostart the
RAID devices. You have to work around this by creating an initial RAM
disk that uses the raidstart command to start your metadevices, then
swaps the initrd out of the way and proceeds to start the real system.

This is not entirely true.  Certainly an initial-ram-disk is one
solution and is (I think) the preferred long-term solution. However
you can also boot from raid with kernel-parameters like:

   md=0,/dev/hda1,/dev/hdc1 boot=/dev/md0

where '0' indicated which md device (md0 in this case), and the
remaining words are the devices to assemble it from.

to which Tim Seufert replied on the same list:

quoted

Hmmm.  That would seem to be a lack in the Linux RAID code, since the
Macintosh partition table has a vastly more flexible partition type
field than DOS: instead of a single byte it's a string.  It would mean
breaking from the convention of using the "Apple_SVR2_UNIX" type for
Linux partitions, but that really is just a convention as far as I know.

Perhaps the PPC Linux developers and the Linux RAID developers should
get together on this and make some decisions so as to make it happen.

I personally think auto-detect is the wrong approach and have no
desire to extend it to other partition types (I cannot remove it from
DOS partitions as that breaks back-compatability).
Just use "md=..."

NeilBrown

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

On Wed, Jan 28, 2004 at 12:42:18PM +1100, Neil Brown wrote:

On Tuesday January 27, atossava@cc.helsinki.fi wrote:

quoted

Sorry about the crossposting.

I wrote on the Yellow Dog Linux list when somebody asked about software
RAID on YDL about my experiences with it:

quoted

quoted

The one really big gotcha is that the Macintosh partitioning scheme
can't tell the Linux kernel that certain partitions are to be
considered "Linux RAID autodetect" (as in x86 using the DOS partition
table type 0xfd). This means that you can't boot a Mac Linux system
directly from RAID because the kernel won't be able to autostart the
RAID devices. You have to work around this by creating an initial RAM
disk that uses the raidstart command to start your metadevices, then
swaps the initrd out of the way and proceeds to start the real system.

This is not entirely true.  Certainly an initial-ram-disk is one
solution and is (I think) the preferred long-term solution. However
you can also boot from raid with kernel-parameters like:

   md=0,/dev/hda1,/dev/hdc1 boot=/dev/md0

where '0' indicated which md device (md0 in this case), and the
remaining words are the devices to assemble it from.

This is said to be broken on latest 2.4.x kernels though. Didn't try
myself though, since i only have a single drive, and don't really want
to mess up with it.

quoted

to which Tim Seufert replied on the same list:

quoted

Hmmm.  That would seem to be a lack in the Linux RAID code, since the
Macintosh partition table has a vastly more flexible partition type
field than DOS: instead of a single byte it's a string.  It would mean
breaking from the convention of using the "Apple_SVR2_UNIX" type for
Linux partitions, but that really is just a convention as far as I know.

Perhaps the PPC Linux developers and the Linux RAID developers should
get together on this and make some decisions so as to make it happen.

I personally think auto-detect is the wrong approach and have no
desire to extend it to other partition types (I cannot remove it from
DOS partitions as that breaks back-compatability).
Just use "md=..."

Ok.

Friendly,

Sven Luther