Red Hat Bugzilla – Bug 336271
Partition names for MPIO devices is inconsistent
Last modified: 2010-01-11 21:31:13 EST
Description of problem:
Version-Release number of selected component (if applicable):
Steps to Reproduce:
1. Map two FCP luns through MPIO with partition tables. Map one to have the device name end in a
number, and one to end in a letter
2. Observe the /dev/mpath and /dev/mapper entries for both
The partition tag varies depending on the last character of the alias or UUID. If the last character is
numeric, the partition tag is 'p' + the partition number. If the last character is alphabetic, the partition
tag is simply the partition number.
# ls -l /dev/mpath
lrwxrwxrwx 1 root root 8 Oct 16 10:50 360a98000433465714b6f447356536236 -> ../dm-12
lrwxrwxrwx 1 root root 8 Oct 16 10:50 360a98000433465714b6f447356536236p1 -> ../dm-18
lrwxrwxrwx 1 root root 8 Oct 16 10:50 360a9800056716c70474a447356796e6f -> ../dm-15
lrwxrwxrwx 1 root root 8 Oct 16 10:50 360a9800056716c70474a447356796e6f1 -> ../dm-19
The partition tag should be consistent without depending on context from the lun name itself.
Because you need a non-numeric separator for whole-disk devices that end in number, it should just
always use 'pX' syntax and not have a special case.
This makes it hard to map the same partition number across luns, as might happen during a SAN lun
clone operation. If the prior UUID and the cloned UUID differ, the partition tags will change as well and
can not be derived from the base lun and partition alone.
With sdXY and hdXY devices, it is always possible to generate the new partition device from the old
device and the new base device via string manipulation alone. Changing the suffix depending on the
base device name in the multipath case breaks that assumption for no good reason.
*** This bug has been marked as a duplicate of 320101 ***