For our acceptance-small configuration we specify devices by-id instead of device name since the
device name occasionally changes. This causes a problem with sgpdd-survey.sh since sgpdd-survey
doesn't translate the alias to the device name.

== test 1: sgpdd-survey, mds, scsidevs == 15:30:35 (1301603435)
Loading modules from
/opt/cray/lustre-cray_gem_s/1.8.4_2.6.32.27_0.2.2_1.0000.5785.16.1-1.0000.25970.22.205/lib64/lustre
lnet.debug=-1
lnet.subsystem_debug=0xffb7e3ff
lnet options: ' accept=all'
=== mds === nid00035 === /dev/disk/by-id/scsi-3600a0b80005121d800001f7a4cd16b41-part1 ===
nid00035: + rslt=/tmp/sgpdd_survey size=1024 crghi=2 thrhi=2
scsidevs="/dev/disk/by-id/scsi-3600a0b80005121d800001f7a4cd16b41-part1"
/opt/cray/lustre-cray_gem_s/1.8.4_2.6.32.27_0.2.2_1.0000.5785.16.1-1.0000.25970.22.205/bin/sgpdd-survey
nid00035: Can't find SG device for /dev/disk/by-id/scsi-3600a0b80005121d800001f7a4cd16b41-part1,
testing for partition
nid00035: Can't find SG device /dev/disk/by-id/scsi-3600a0b80005121d800001f7a4cd16b41-part.

It would be nice if either sgpdd-survey or sgpdd-survey.sh would translate aliases to device names.
The attached patch modifies sgpdd-survey to do a 'readlink -f' on the device name if the device
path is something like "/dev/disk/by-*" and the device path is a symbolic link (maybe this is
redundant?).

This is from Oracle bug 24487.