Linux Storage - Hot removal of a LUN

Linux Storage Solution · Config document · referenced from RDAC problems and LUN removal

Problem 2 of my notes: a LUN had to be removed from the running system without a reboot (hot remove). The notes collect three ways to say "delete" to the SCSI layer through sysfs, the command that shows which disk is which, a quoted English note about taking a path offline, a small script that prints the vendor and model of every block device, and, as the case that was done, the removal of all eight physical paths to the standby array DS5300-SITEB (array ID 1).

The commands

Removing a LUN by its disk name, where sd* is the disk device to be removed from the system. Here sdc and sdd, the two virtual disks of the standby array:

$ echo "1" > /sys/block/sdc/device/delete
$ echo "1" > /sys/block/sdd/device/delete

Removing a LUN by its SCSI address, where h is the HBA number, c the channel on the HBA, t the SCSI target ID and l the LUN. The notes keep the pattern, not a run of it:

$ echo 1 > /sys/class/scsi_device/h:c:t:l/device/delete

The device name, the HBA number, the channel, the target and the LUN are found with one listing. The notes keep no output for it:

$ ls -l /dev/disk/by-*

The notes then quote a note in English, which I give as it stands: "Take the path offline using echo offline > /sys/block/sda/device/state. This will cause any subsequent IO sent to the device on this path to be failed immediately. Device-mapper-multipath will continue to use the remaining paths to the device." It speaks of Device-mapper multipath, not of RDAC, and sda in it is an example name, not the sda of this host, which is LUN 1 of the main array.

The small script "who uses what": for every block device that has a vendor attribute, print its name, vendor and model. It is kept as it was in the notes, with its unclosed typographic quote:

for DSK in `ls /sys/block/`
do
if [ -e "/sys/block/${DSK}/device/vendor" ]
then
VENDOR=`cat /sys/block/${DSK}/device/vendor`
MODEL=`cat /sys/block/${DSK}/device/model`
echo “${DSK} ${VENDOR} ${MODEL}
fi
done

The paths leading to the array with ID 1, DS5300-SITEB:

$ mppUtil -g 1 | grep PathId

PathId: 77000002 (hostId: 0, channelId: 0, targetId: 2)
PathId: 77010002 (hostId: 1, channelId: 0, targetId: 2)
PathId: 77000003 (hostId: 0, channelId: 0, targetId: 3)
PathId: 77010003 (hostId: 1, channelId: 0, targetId: 3)

Deleting every path to the array with ID 1: two hosts, two targets, two LUNs, eight SCSI devices. The lines are kept exactly, with the 1> of the note above; the notes write them in four pairs, one pair per path, separated by blank lines, which the fence leaves out:

$ echo 1>/sys/bus/scsi/devices/0\:0\:2\:1/delete
$ echo 1>/sys/bus/scsi/devices/0\:0\:2\:2/delete
$ echo 1>/sys/bus/scsi/devices/1\:0\:2\:1/delete
$ echo 1>/sys/bus/scsi/devices/1\:0\:2\:2/delete
$ echo 1>/sys/bus/scsi/devices/0\:0\:3\:1/delete
$ echo 1>/sys/bus/scsi/devices/0\:0\:3\:2/delete
$ echo 1>/sys/bus/scsi/devices/1\:0\:3\:1/delete
$ echo 1>/sys/bus/scsi/devices/1\:0\:3\:2/delete

Reading it

The order of the eight lines follows the PathId list: first both LUNs of target 2 on host 0, then target 2 on host 1, then target 3 on hosts 0 and 1. What the notes do not hold: whether sdc and sdd were removed first (the first two lines) and the eight paths afterwards, or whether the two blocks are alternatives; a listing before and after; and whether the driver's mppBusRescan -d ("remove an unmapped or disconnected device", see mppBusRescan commands) was tried instead. The links section of the notes keeps a forum thread titled "RDAC capable of hot removing unmapped LUNs?", so the question was on my mind.

Checked against the Linux kernel 7.2 and Red Hat Enterprise Linux 10