Unable to eradicate previous version of device information, even with zero-superblock and dd

I've been trying to bring up a RAID10 device, and I'm having some difficulty with automatically-created device names.

mdadm version 2.5.6, Debian Etch.

With metadata=1.2 in my config file,

mdadm --create /dev/md/all --auto=p7 -n 4 --level=10 /dev/sd*2

This does seem to create a RAID array. I see that my /dev/md/ directory is populated with all1 through all7.

On reboot, however, I notice that there's a suddenly a /dev/md127 device. Confused, I attempted to start over many times, but I can't seem to create a non-"all" array and I can't seem to create a simple /dev/md/0 array.

Steps:

To eradicate all prior traces of md configuration, I issue these commands:

mdadm --stop /dev/md/all

which stops.

mdadm --zero-superblock /dev/sd[each drive]2

I went further (after some trouble) and issued

dd if=/dev/zero of=/dev/sd[each drive]2 count=2M

I then issue:

rm /dev/md* /dev/md/*

The ARRAY information is commented out of the config file (mdadm.conf).

On reboot, I see that the devices /dev/md/all, /dev/md/all1, etc. have reappeared, along /dev/md127, /dev/md_127, and /dev/md_d127.

This is very, very puzzling.

Well, I thought I could work around this. I issued

mdadm --create /dev/md/all

with the same paramters as above. I can use cfdisk and fdisk (either one) to create two partitions, /dev/md/all1 and /dev/md/all2.

However,

mkfs.reiserfs /dev/md/all1

claims that the /dev/md/all1 has "no such device or address."

ls -l /dev/md/all gives

brw-rw---- 1 root disk 254, 8129 (date) /dev/md/all1

QUESTIONS:

1. If I create a device called /dev/md/all, should I expect that mdadm will create a device called /dev/md/127, and that mdadm --detail --scan will report it as /dev/md127 or something similar?

2. How can I completely eradicate all traces of previous work, given that zero-superblock and dd on the drives that make up the array doesn't seem to erase previous information?

Moshe Yudkowsky * [email protected] * www.pobox.com/~moshe
 "If you're going to shoot, shoot! Don't talk!"
   		   -- Eli Wallach,"The Good, the Bad, and the Ugly"

That's still happening. However:

Answer:

In order for the md drives to be started on a reboot, upgrade-initramfs
places information about the current configuration into boot
configuration.

In order to eradicate everything, stop all arrays, comment out any ARRAY lines in mdadm.conf, remove all md device files, and then issue

update-initramfs

This cleans out the information that's hidden inside the /boot area. On the next reboot, no extraneous md files are present. It's then possible to issue an mdadm --create /dev/md/all that will create the appropriate md devices automatically with proper major and minor device numbers.

To get the md device started correctly at init time, I seem to require the use of update-initramfs. I will investigate further when I've got some time...

Moshe Yudkowsky * [email protected] * www.pobox.com/~moshe
 "The odds are good, but the goods are odd."
   	     -- Alaskan women, on the high ratio of men to women in Alaska