fence_scsi: fix registration handling in device 'off' workflows

fence-agents.spec

@@ -6,13 +6,14 @@
 Name: fence-agents
 Summary: Set of unified programs capable of host isolation ("fencing")
 Version: 4.12.1
-Release: 5
+Release: 6
 License: GPLv2+ and LGPLv2+
 Group: System Environment/Base
 URL: https://github.com/ClusterLabs/fence-agents
 Source0: https://github.com/ClusterLabs/fence-agents/archive//v%{version}/%{name}-%{version}.tar.gz
 Patch0: fence_ipmilan-fix-typo-in-description.patch
 Patch1: fence_scsi-fix-registration-handling-if-ISID-conflic.patch
+Patch2: fence_scsi-fix-registration-handling-in-device-off-workflows.patch
 
 # skipped: pve, raritan, rcd-serial, virsh
 %global allfenceagents %(cat <<EOF
@@ -1128,6 +1129,9 @@ are located on corosync cluster nodes.
 %{_libdir}/fence-virt/cpg.so
 
 %changelog
+* Mon May 13 2024 zouzhimin <zouzhimin@kylinos.cn> - 4.12.1-6
+- fence_scsi: fix registration handling in device 'off' workflows
+
 * Fri Apr 26 2024 zouzhimin <zouzhimin@kylinos.cn> - 4.12.1-5
 - fence_scsi: fix registration handling if ISID conflict
 

fence_scsi-fix-registration-handling-in-device-off-workflows.patch

@@ -0,0 +1,106 @@
+From 34baef58db442148b8e067509d2cdd37b7a91ef4 Mon Sep 17 00:00:00 2001
+From: "sreejit.mohanan" <sreejit.mohanan@nutanix.com>
+Date: Thu, 7 Sep 2023 15:57:51 -0700
+Subject: [PATCH] fence_scsi: fix registration handling in device 'off'
+ workflows
+
+ISID (Initiator Session ID) belonging to I_T Nexus changes for
+RHEL based on the session ID. This means that the connection to
+the device can be set up with different ISID on reconnects.
+
+When a device is powered off, fence_scsi assumes that the client
+has a registration to the device and sends a preempt-and-abort
+request which ends up failing due to reservation conflict.
+
+Fixing this by registering the host key with the device and preempting
+the old registration (if it exists). This should make sure that the
+host is able to preempt the other key successfully.
+---
+ agents/scsi/fence_scsi.py | 29 +++++++++++++++--------------
+ 1 file changed, 15 insertions(+), 14 deletions(-)
+
+diff --git a/agents/scsi/fence_scsi.py b/agents/scsi/fence_scsi.py
+index 42530ceb..519319bf 100644
+--- a/agents/scsi/fence_scsi.py
++++ b/agents/scsi/fence_scsi.py
+@@ -41,7 +41,7 @@ def set_status(conn, options):
+ 		for dev in options["devices"]:
+ 			is_block_device(dev)
+ 
+-			register_dev(options, dev)
++			register_dev(options, dev, options["--key"])
+ 			if options["--key"] not in get_registration_keys(options, dev):
+ 				count += 1
+ 				logging.debug("Failed to register key "\
+@@ -62,7 +62,7 @@ def set_status(conn, options):
+ 			fail_usage("Failed: keys cannot be same. You can not fence yourself.")
+ 		for dev in options["devices"]:
+ 			is_block_device(dev)
+-
++			register_dev(options, dev, host_key)
+ 			if options["--key"] in get_registration_keys(options, dev):
+ 				preempt_abort(options, host_key, dev)
+ 
+@@ -131,11 +131,11 @@ def reset_dev(options, dev):
+ 	return run_cmd(options, options["--sg_turs-path"] + " " + dev)["rc"]
+ 
+ 
+-def register_dev(options, dev):
++def register_dev(options, dev, key):
+ 	dev = os.path.realpath(dev)
+ 	if re.search(r"^dm", dev[5:]):
+ 		for slave in get_mpath_slaves(dev):
+-			register_dev(options, slave)
++			register_dev(options, slave, key)
+ 		return True
+ 
+ 	# Check if any registration exists for the key already. We track this in
+@@ -143,34 +143,35 @@ def register_dev(options, dev):
+ 	# This is needed since the previous registration could be for a
+ 	# different I_T nexus (different ISID).
+ 	registration_key_exists = False
+-	if options["--key"] in get_registration_keys(options, dev):
++	if key in get_registration_keys(options, dev):
++		logging.debug("Registration key exists for device " + dev)
+ 		registration_key_exists = True
+-	if not register_helper(options, options["--key"], dev):
++	if not register_helper(options, dev, key):
+ 		return False
+ 
+ 	if registration_key_exists:
+ 		# If key matches, make sure it matches with the connection that
+ 		# exists right now. To do this, we can issue a preempt with same key
+ 		# which should replace the old invalid entries from the target.
+-		if not preempt(options, options["--key"], dev):
++		if not preempt(options, key, dev, key):
+ 			return False
+ 
+ 		# If there was no reservation, we need to issue another registration
+ 		# since the previous preempt would clear registration made above.
+-		if get_reservation_key(options, dev, False) != options["--key"]:
+-			return register_helper(options, options["--key"], dev)
++		if get_reservation_key(options, dev, False) != key:
++			return register_helper(options, dev, key)
+ 	return True
+ 
+-# cancel registration without aborting tasks
+-def preempt(options, host, dev):
++# helper function to preempt host with 'key' using 'host_key' without aborting tasks
++def preempt(options, host_key, dev, key):
+ 	reset_dev(options,dev)
+-	cmd = options["--sg_persist-path"] + " -n -o -P -T 5 -K " + host + " -S " + options["--key"] + " -d " + dev
++	cmd = options["--sg_persist-path"] + " -n -o -P -T 5 -K " + host_key + " -S " + key + " -d " + dev
+ 	return not bool(run_cmd(options, cmd)["rc"])
+ 
+ # helper function to send the register command
+-def register_helper(options, host, dev):
++def register_helper(options, dev, key):
+ 	reset_dev(options, dev)
+-	cmd = options["--sg_persist-path"] + " -n -o -I -S " + options["--key"] + " -d " + dev
++	cmd = options["--sg_persist-path"] + " -n -o -I -S " + key + " -d " + dev
+ 	cmd += " -Z" if "--aptpl" in options else ""
+ 	return not bool(run_cmd(options, cmd)["rc"])
+ 
+-- 
+2.25.1
+