multipath: add queue_without_daemon config parameter

This patch adds a new multipath.conf default paramter, queue_without_daemon.
If this is set to "no", when multipathd stops, queueing will be turned off for
all devices.  This is useful for devices that set no_path_retry.  If a machine
is shut down while all paths to a device are down, it is possible to hang
waiting for IO to return from the device after multipathd has been stopped.
Without multipathd running, access to the paths cannot be restored, and the
kernel cannot be told to stop queueing IO. Setting queue_without_daemon to "no"
makes multipathd turn off queueing on all devices when it stops, avoiding the
problem.

Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>

7 files changed, 70 insertions, 0 deletions

diff --git a/libmultipath/config.h b/libmultipath/config.h
index 1d66cca..f6f39e6 100644
--- a/libmultipath/config.h
+++ b/libmultipath/config.h
@@ -74,6 +74,7 @@ struct config {
 	int pg_timeout;
 	int max_fds;
 	int force_reload;
+	int queue_without_daemon;
 	int daemon;
 	int flush_on_last_del;
 	int attribute_flags;
diff --git a/libmultipath/dict.c b/libmultipath/dict.c
index 9782599..ab0c831 100644
--- a/libmultipath/dict.c
+++ b/libmultipath/dict.c
@@ -362,6 +362,28 @@ def_no_path_retry_handler(vector strvec)
 }
 
 static int
+def_queue_without_daemon(vector strvec)
+{
+	char * buff;
+
+	buff = set_value(strvec);
+	if (!buff)
+		return 1;
+
+	if (!strncmp(buff, "off", 3) || !strncmp(buff, "no", 2) ||
+	    !strncmp(buff, "0", 1))
+		conf->queue_without_daemon = QUE_NO_DAEMON_OFF;
+	else if (!strncmp(buff, "on", 2) || !strncmp(buff, "yes", 3) ||
+		 !strncmp(buff, "1", 1))
+		conf->queue_without_daemon = QUE_NO_DAEMON_ON;
+	else
+		conf->queue_without_daemon = QUE_NO_DAEMON_UNDEF;
+
+	free(buff);
+	return 0;
+}
+
+static int
 def_pg_timeout_handler(vector strvec)
 {
 	int pg_timeout;
@@ -1944,6 +1966,18 @@ snprint_def_no_path_retry (char * buff, int len, void * data)
 }
 
 static int
+snprint_def_queue_without_daemon (char * buff, int len, void * data)
+{
+	switch (conf->queue_without_daemon) {
+	case QUE_NO_DAEMON_OFF:
+		return snprintf(buff, len, "no");
+	case QUE_NO_DAEMON_ON:
+		return snprintf(buff, len, "yes");
+	}
+	return 0;
+}
+
+static int
 snprint_def_pg_timeout (char * buff, int len, void * data)
 {
 	if (conf->pg_timeout == DEFAULT_PGTIMEOUT)
@@ -2029,6 +2063,7 @@ init_keywords(void)
 	install_keyword("max_fds", &max_fds_handler, &snprint_max_fds);
 	install_keyword("rr_weight", &def_weight_handler, &snprint_def_rr_weight);
 	install_keyword("no_path_retry", &def_no_path_retry_handler, &snprint_def_no_path_retry);
+	install_keyword("queue_without_daemon", &def_queue_without_daemon, &snprint_def_queue_without_daemon);
 	install_keyword("pg_timeout", &def_pg_timeout_handler, &snprint_def_pg_timeout);
 	install_keyword("flush_on_last_del", &def_flush_on_last_del_handler, &snprint_def_flush_on_last_del);
 	install_keyword("user_friendly_names", &names_handler, &snprint_def_user_friendly_names);
diff --git a/libmultipath/structs.h b/libmultipath/structs.h
index dfd0556..c559838 100644
--- a/libmultipath/structs.h
+++ b/libmultipath/structs.h
@@ -63,6 +63,12 @@ enum pgstates {
 	PGSTATE_ACTIVE
 };
 
+enum queue_without_daemon_states {
+	QUE_NO_DAEMON_UNDEF,
+	QUE_NO_DAEMON_OFF,
+	QUE_NO_DAEMON_ON,
+};
+
 enum pgtimeouts {
 	PGTIMEOUT_UNDEF,
 	PGTIMEOUT_NONE
diff --git a/multipath.conf.annotated b/multipath.conf.annotated
index bd04e70..6532622 100644
--- a/multipath.conf.annotated
+++ b/multipath.conf.annotated
@@ -153,6 +153,15 @@
 #	no_path_retry  queue
 #
 #	#
+#	# name    : queue_without_daemon
+#	# scope   : multipathd
+#	# desc    : If set to "no", multipathd will disable queueing for all
+#	#           devices when it is shut down.
+#	# values  : yes|no
+#	# default : yes
+#	queue_without_daemon	no
+#
+#	#
 #	# name    : user_friendly_names
 #	# scope   : multipath
 #	# desc    : If set to "yes", using the bindings file
diff --git a/multipath.conf.synthetic b/multipath.conf.synthetic
index 3e0fd6e..44d1329 100644
--- a/multipath.conf.synthetic
+++ b/multipath.conf.synthetic
@@ -16,6 +16,7 @@
 #	rr_weight		priorities
 #	failback		immediate
 #	no_path_retry		fail
+#	queue_without_daemon    no
 #	user_friendly_names	no
 #	mode			644
 #	uid			0
diff --git a/multipath/multipath.conf.5 b/multipath/multipath.conf.5
index ed9b2d8..1dc1488 100644
--- a/multipath/multipath.conf.5
+++ b/multipath/multipath.conf.5
@@ -251,6 +251,19 @@ will disable the timeout.
 .B dev_loss_tmo
 Specify the number of seconds the scsi layer will wait after a problem has
 been detected on a FC remote port before removing it from the system.
+.TP
+.B queue_without_daemon
+If set to
+.I no
+, when multipathd stops, queueing will be turned off for all devices.
+This is useful for devices that set no_path_retry.  If a machine is
+shut down while all paths to a device are down, it is possible to hang waiting
+for IO to return from the device after multipathd has been stopped. Without
+multipathd running, access to the paths cannot be restored, and the kernel
+cannot be told to stop queueing IO. Setting queue_without_daemon to
+.I no
+, avoids this problem. Default is
+.I yes
 .
 .SH "blacklist section"
 The
diff --git a/multipathd/main.c b/multipathd/main.c
index 3f7eb49..8754144 100644
--- a/multipathd/main.c
+++ b/multipathd/main.c
@@ -1359,6 +1359,8 @@ child (void * param)
 	pthread_t check_thr, uevent_thr, uxlsnr_thr;
 	pthread_attr_t log_attr, misc_attr;
 	struct vectors * vecs;
+	struct multipath * mpp;
+	int i;
 
 	mlockall(MCL_CURRENT | MCL_FUTURE);
 
@@ -1448,6 +1450,9 @@ child (void * param)
 	 */
 	block_signal(SIGHUP, NULL);
 	lock(vecs->lock);
+	if (conf->queue_without_daemon == QUE_NO_DAEMON_OFF)
+		vector_foreach_slot(vecs->mpvec, mpp, i)
+			dm_queue_if_no_path(mpp->alias, 0);
 	remove_maps_and_stop_waiters(vecs);
 	free_pathvec(vecs->pathvec, FREE_PATHS);