Blame SOURCES/0041-multipathd-handle-fpin-events.patch

da32ae
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
da32ae
From: Muneendra Kumar <muneendra.kumar@broadcom.com>
da32ae
Date: Wed, 9 Feb 2022 19:28:10 -0800
da32ae
Subject: [PATCH] multipathd: handle fpin events
da32ae
da32ae
This patch incorporates the functionality to handle
da32ae
FPIN ELS events present as part of FCTransport daemon
da32ae
(available in EPEL8) into the multipathd. This helps us to
da32ae
reduce the response time to react and take the necessary actions
da32ae
on receiving the FPIN events.
da32ae
da32ae
This patch currently support FPIN-Li Events.
da32ae
da32ae
It adds a new thread to listen for ELS frames from driver and on
da32ae
receiving the frame payload, push the payload to a list and notify the
da32ae
fpin_els_li_consumer thread to process it.Once consumer thread is
da32ae
notified, it returns to listen for more ELS frames from driver.
da32ae
da32ae
The consumer thread process the ELS frames and moves the devices paths
da32ae
which are affected due to link integrity to marginal path groups.
da32ae
This also sets the associated portstate to marginal.
da32ae
The paths which are set to marginal path group will be unset
da32ae
on receiving the RSCN events
da32ae
da32ae
[ MW: minor fixup for 32bit compilation ]
da32ae
da32ae
Signed-off-by: Muneendra Kumar <muneendra.kumar@broadcom.com>
da32ae
Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
da32ae
Signed-off-by: Martin Wilck <mwilck@suse.com>
da32ae
Reviewed-by: Martin Wilck <mwilck@suse.com>
da32ae
---
da32ae
 Makefile.inc                      |  13 +
da32ae
 libmultipath/Makefile             |   5 +
da32ae
 libmultipath/dict.c               |  56 +++-
da32ae
 libmultipath/libmultipath.version |   5 +
da32ae
 libmultipath/propsel.c            |  47 ++-
da32ae
 libmultipath/structs.h            |   7 +
da32ae
 multipath/multipath.conf.5        |  19 +-
da32ae
 multipathd/Makefile               |  10 +
da32ae
 multipathd/fpin.h                 |  20 ++
da32ae
 multipathd/fpin_handlers.c        | 540 ++++++++++++++++++++++++++++++
da32ae
 multipathd/main.c                 |  43 ++-
da32ae
 11 files changed, 749 insertions(+), 16 deletions(-)
da32ae
 create mode 100644 multipathd/fpin.h
da32ae
 create mode 100644 multipathd/fpin_handlers.c
da32ae
da32ae
diff --git a/Makefile.inc b/Makefile.inc
da32ae
index 5ac660de..688c4599 100644
da32ae
--- a/Makefile.inc
da32ae
+++ b/Makefile.inc
da32ae
@@ -149,6 +149,19 @@ check_file = $(shell \
da32ae
 	echo "$$found" \
da32ae
 	)
da32ae
 
da32ae
+# Check whether a file contains a variable with name $1 in header file $2
da32ae
+check_var = $(shell \
da32ae
+	if grep -Eq "(^|[[:blank:]])$1([[:blank:]]|=|$$)" "$2"; then \
da32ae
+                found=1; \
da32ae
+                status="yes"; \
da32ae
+        else \
da32ae
+                found=0; \
da32ae
+                status="no"; \
da32ae
+        fi; \
da32ae
+        echo 1>&2 "Checking for ..  $1 in $2 ... $$status"; \
da32ae
+        echo "$$found" \
da32ae
+        )
da32ae
+
da32ae
 %.o:	%.c
da32ae
 	@echo building $@ because of $?
da32ae
 	$(CC) $(CFLAGS) -c -o $@ $<
da32ae
diff --git a/libmultipath/Makefile b/libmultipath/Makefile
da32ae
index 7f3921c5..8a960419 100644
da32ae
--- a/libmultipath/Makefile
da32ae
+++ b/libmultipath/Makefile
da32ae
@@ -45,6 +45,11 @@ ifneq ($(call check_func,dm_hold_control_dev,/usr/include/libdevmapper.h),0)
da32ae
 	CFLAGS += -DLIBDM_API_HOLD_CONTROL
da32ae
 endif
da32ae
 
da32ae
+ifneq ($(call check_var,ELS_DTAG_LNK_INTEGRITY,/usr/include/scsi/fc/fc_els.h),0)
da32ae
+	CFLAGS += -DFPIN_EVENT_HANDLER
da32ae
+endif
da32ae
+
da32ae
+
da32ae
 OBJS = memory.o parser.o vector.o devmapper.o callout.o \
da32ae
 	hwtable.o blacklist.o util.o dmparser.o config.o \
da32ae
 	structs.o discovery.o propsel.o dict.o \
da32ae
diff --git a/libmultipath/dict.c b/libmultipath/dict.c
da32ae
index 1b75be47..eb5a8083 100644
da32ae
--- a/libmultipath/dict.c
da32ae
+++ b/libmultipath/dict.c
da32ae
@@ -513,6 +513,59 @@ snprint_def_find_multipaths(struct config *conf, struct strbuf *buff,
da32ae
 			 find_multipaths_optvals[conf->find_multipaths]);
da32ae
 }
da32ae
 
da32ae
+static const char * const marginal_pathgroups_optvals[] = {
da32ae
+	[MARGINAL_PATHGROUP_OFF] = "off",
da32ae
+	[MARGINAL_PATHGROUP_ON] = "on",
da32ae
+#ifdef FPIN_EVENT_HANDLER
da32ae
+	[MARGINAL_PATHGROUP_FPIN] = "fpin",
da32ae
+#endif
da32ae
+};
da32ae
+
da32ae
+static int
da32ae
+def_marginal_pathgroups_handler(struct config *conf, vector strvec,
da32ae
+			    const char *file, int line_nr)
da32ae
+{
da32ae
+	char *buff;
da32ae
+	unsigned int i;
da32ae
+
da32ae
+	buff = set_value(strvec);
da32ae
+	if (!buff)
da32ae
+		return 1;
da32ae
+	for (i = MARGINAL_PATHGROUP_OFF;
da32ae
+	     i < ARRAY_SIZE(marginal_pathgroups_optvals); i++) {
da32ae
+		if (marginal_pathgroups_optvals[i] != NULL &&
da32ae
+		    !strcmp(buff, marginal_pathgroups_optvals[i])) {
da32ae
+			conf->marginal_pathgroups = i;
da32ae
+			break;
da32ae
+		}
da32ae
+	}
da32ae
+
da32ae
+	if (i >= ARRAY_SIZE(marginal_pathgroups_optvals)) {
da32ae
+		if (strcmp(buff, "no") == 0 || strcmp(buff, "0") == 0)
da32ae
+			conf->marginal_pathgroups = MARGINAL_PATHGROUP_OFF;
da32ae
+		else if (strcmp(buff, "yes") == 0 || strcmp(buff, "1") == 0)
da32ae
+			conf->marginal_pathgroups = MARGINAL_PATHGROUP_ON;
da32ae
+		/* This can only be true if FPIN_EVENT_HANDLER isn't defined,
da32ae
+		 * otherwise this check will have already happened above */
da32ae
+		else if (strcmp(buff, "fpin") == 0)
da32ae
+			condlog(1, "%s line %d, support for \"fpin\" is not compiled in for marginal_pathgroups", file, line_nr);
da32ae
+		else
da32ae
+			condlog(1, "%s line %d, invalid value for marginal_pathgroups: \"%s\"",
da32ae
+				file, line_nr, buff);
da32ae
+	}
da32ae
+	free(buff);
da32ae
+	return 0;
da32ae
+}
da32ae
+
da32ae
+static int
da32ae
+snprint_def_marginal_pathgroups(struct config *conf, struct strbuf *buff,
da32ae
+			    const void *data)
da32ae
+{
da32ae
+	return append_strbuf_quoted(buff,
da32ae
+			 marginal_pathgroups_optvals[conf->marginal_pathgroups]);
da32ae
+}
da32ae
+
da32ae
+
da32ae
 declare_def_handler(selector, set_str)
da32ae
 declare_def_snprint_defstr(selector, print_str, DEFAULT_SELECTOR)
da32ae
 declare_hw_handler(selector, set_str)
da32ae
@@ -1527,9 +1580,6 @@ declare_ovr_snprint(all_tg_pt, print_yes_no_undef)
da32ae
 declare_hw_handler(all_tg_pt, set_yes_no_undef)
da32ae
 declare_hw_snprint(all_tg_pt, print_yes_no_undef)
da32ae
 
da32ae
-declare_def_handler(marginal_pathgroups, set_yes_no)
da32ae
-declare_def_snprint(marginal_pathgroups, print_yes_no)
da32ae
-
da32ae
 declare_def_handler(recheck_wwid, set_yes_no_undef)
da32ae
 declare_def_snprint_defint(recheck_wwid, print_yes_no_undef, DEFAULT_RECHECK_WWID)
da32ae
 declare_ovr_handler(recheck_wwid, set_yes_no_undef)
da32ae
diff --git a/libmultipath/libmultipath.version b/libmultipath/libmultipath.version
da32ae
index 0d89e9e1..1d018eab 100644
da32ae
--- a/libmultipath/libmultipath.version
da32ae
+++ b/libmultipath/libmultipath.version
da32ae
@@ -297,3 +297,8 @@ LIBMULTIPATH_9.1.1 {
da32ae
 global:
da32ae
 	trigger_path_udev_change;
da32ae
 } LIBMULTIPATH_9.1.0;
da32ae
+
da32ae
+LIBMULTIPATH_9.1.2 {
da32ae
+global:
da32ae
+	cleanup_mutex;
da32ae
+} LIBMULTIPATH_9.1.1;
da32ae
diff --git a/libmultipath/propsel.c b/libmultipath/propsel.c
da32ae
index b2876670..677ab9e1 100644
da32ae
--- a/libmultipath/propsel.c
da32ae
+++ b/libmultipath/propsel.c
da32ae
@@ -84,6 +84,8 @@ static const char cmdline_origin[] =
da32ae
 	"(setting: multipath command line [-p] flag)";
da32ae
 static const char autodetect_origin[] =
da32ae
 	"(setting: storage device autodetected)";
da32ae
+static const char fpin_marginal_path_origin[] =
da32ae
+	"(setting: overridden by marginal_path_fpin)";
da32ae
 static const char marginal_path_origin[] =
da32ae
 	"(setting: implied by marginal_path check)";
da32ae
 static const char delay_watch_origin[] =
da32ae
@@ -1036,9 +1038,12 @@ int select_san_path_err_threshold(struct config *conf, struct multipath *mp)
da32ae
 	const char *origin;
da32ae
 	STRBUF_ON_STACK(buff);
da32ae
 
da32ae
-	if (marginal_path_check_enabled(mp)) {
da32ae
+	if (marginal_path_check_enabled(mp) || (conf->marginal_pathgroups == MARGINAL_PATHGROUP_FPIN)) {
da32ae
 		mp->san_path_err_threshold = NU_NO;
da32ae
-		origin = marginal_path_origin;
da32ae
+		if (conf->marginal_pathgroups == MARGINAL_PATHGROUP_FPIN)
da32ae
+			origin = fpin_marginal_path_origin;
da32ae
+		else
da32ae
+			origin = marginal_path_origin;
da32ae
 		goto out;
da32ae
 	}
da32ae
 	mp_set_mpe(san_path_err_threshold);
da32ae
@@ -1059,9 +1064,12 @@ int select_san_path_err_forget_rate(struct config *conf, struct multipath *mp)
da32ae
 	const char *origin;
da32ae
 	STRBUF_ON_STACK(buff);
da32ae
 
da32ae
-	if (marginal_path_check_enabled(mp)) {
da32ae
+	if (marginal_path_check_enabled(mp) || (conf->marginal_pathgroups == MARGINAL_PATHGROUP_FPIN)) {
da32ae
 		mp->san_path_err_forget_rate = NU_NO;
da32ae
-		origin = marginal_path_origin;
da32ae
+		if (conf->marginal_pathgroups == MARGINAL_PATHGROUP_FPIN)
da32ae
+			origin = fpin_marginal_path_origin;
da32ae
+		else
da32ae
+			origin = marginal_path_origin;
da32ae
 		goto out;
da32ae
 	}
da32ae
 	mp_set_mpe(san_path_err_forget_rate);
da32ae
@@ -1083,9 +1091,12 @@ int select_san_path_err_recovery_time(struct config *conf, struct multipath *mp)
da32ae
 	const char *origin;
da32ae
 	STRBUF_ON_STACK(buff);
da32ae
 
da32ae
-	if (marginal_path_check_enabled(mp)) {
da32ae
+	if (marginal_path_check_enabled(mp) || (conf->marginal_pathgroups == MARGINAL_PATHGROUP_FPIN)) {
da32ae
 		mp->san_path_err_recovery_time = NU_NO;
da32ae
-		origin = marginal_path_origin;
da32ae
+		if (conf->marginal_pathgroups == MARGINAL_PATHGROUP_FPIN)
da32ae
+			origin = fpin_marginal_path_origin;
da32ae
+		else
da32ae
+			origin = marginal_path_origin;
da32ae
 		goto out;
da32ae
 	}
da32ae
 	mp_set_mpe(san_path_err_recovery_time);
da32ae
@@ -1107,6 +1118,12 @@ int select_marginal_path_err_sample_time(struct config *conf, struct multipath *
da32ae
 	const char *origin;
da32ae
 	STRBUF_ON_STACK(buff);
da32ae
 
da32ae
+	if (conf->marginal_pathgroups == MARGINAL_PATHGROUP_FPIN) {
da32ae
+		mp->marginal_path_err_sample_time = NU_NO;
da32ae
+		origin = fpin_marginal_path_origin;
da32ae
+		goto out;
da32ae
+	}
da32ae
+
da32ae
 	mp_set_mpe(marginal_path_err_sample_time);
da32ae
 	mp_set_ovr(marginal_path_err_sample_time);
da32ae
 	mp_set_hwe(marginal_path_err_sample_time);
da32ae
@@ -1130,6 +1147,12 @@ int select_marginal_path_err_rate_threshold(struct config *conf, struct multipat
da32ae
 	const char *origin;
da32ae
 	STRBUF_ON_STACK(buff);
da32ae
 
da32ae
+	if (conf->marginal_pathgroups == MARGINAL_PATHGROUP_FPIN) {
da32ae
+		mp->marginal_path_err_rate_threshold = NU_NO;
da32ae
+		origin = fpin_marginal_path_origin;
da32ae
+		goto out;
da32ae
+	}
da32ae
+
da32ae
 	mp_set_mpe(marginal_path_err_rate_threshold);
da32ae
 	mp_set_ovr(marginal_path_err_rate_threshold);
da32ae
 	mp_set_hwe(marginal_path_err_rate_threshold);
da32ae
@@ -1147,6 +1170,12 @@ int select_marginal_path_err_recheck_gap_time(struct config *conf, struct multip
da32ae
 	const char *origin;
da32ae
 	STRBUF_ON_STACK(buff);
da32ae
 
da32ae
+	if (conf->marginal_pathgroups == MARGINAL_PATHGROUP_FPIN) {
da32ae
+		mp->marginal_path_err_recheck_gap_time = NU_NO;
da32ae
+		origin = fpin_marginal_path_origin;
da32ae
+		goto out;
da32ae
+	}
da32ae
+
da32ae
 	mp_set_mpe(marginal_path_err_recheck_gap_time);
da32ae
 	mp_set_ovr(marginal_path_err_recheck_gap_time);
da32ae
 	mp_set_hwe(marginal_path_err_recheck_gap_time);
da32ae
@@ -1165,6 +1194,12 @@ int select_marginal_path_double_failed_time(struct config *conf, struct multipat
da32ae
 	const char *origin;
da32ae
 	STRBUF_ON_STACK(buff);
da32ae
 
da32ae
+	if (conf->marginal_pathgroups == MARGINAL_PATHGROUP_FPIN) {
da32ae
+		mp->marginal_path_double_failed_time = NU_NO;
da32ae
+		origin = fpin_marginal_path_origin;
da32ae
+		goto out;
da32ae
+	}
da32ae
+
da32ae
 	mp_set_mpe(marginal_path_double_failed_time);
da32ae
 	mp_set_ovr(marginal_path_double_failed_time);
da32ae
 	mp_set_hwe(marginal_path_double_failed_time);
da32ae
diff --git a/libmultipath/structs.h b/libmultipath/structs.h
da32ae
index 399540e7..1188363e 100644
da32ae
--- a/libmultipath/structs.h
da32ae
+++ b/libmultipath/structs.h
da32ae
@@ -110,6 +110,12 @@ enum find_multipaths_states {
da32ae
 	__FIND_MULTIPATHS_LAST,
da32ae
 };
da32ae
 
da32ae
+enum marginal_pathgroups_mode {
da32ae
+	MARGINAL_PATHGROUP_OFF = YN_NO,
da32ae
+	MARGINAL_PATHGROUP_ON = YN_YES,
da32ae
+	MARGINAL_PATHGROUP_FPIN,
da32ae
+};
da32ae
+
da32ae
 enum flush_states {
da32ae
 	FLUSH_UNDEF = YNU_UNDEF,
da32ae
 	FLUSH_DISABLED = YNU_NO,
da32ae
@@ -409,6 +415,7 @@ struct multipath {
da32ae
 	unsigned char prflag;
da32ae
 	int all_tg_pt;
da32ae
 	struct gen_multipath generic_mp;
da32ae
+	bool fpin_must_reload;
da32ae
 };
da32ae
 
da32ae
 static inline int marginal_path_check_enabled(const struct multipath *mpp)
da32ae
diff --git a/multipath/multipath.conf.5 b/multipath/multipath.conf.5
da32ae
index 7f85f766..5ed2cd3c 100644
da32ae
--- a/multipath/multipath.conf.5
da32ae
+++ b/multipath/multipath.conf.5
da32ae
@@ -1088,20 +1088,26 @@ The default is: \fBno\fR
da32ae
 .
da32ae
 .TP
da32ae
 .B marginal_pathgroups
da32ae
-If set to \fIno\fR, the \fIdelay_*_checks\fR, \fImarginal_path_*\fR, and
da32ae
+If set to \fIoff\fR, the \fIdelay_*_checks\fR, \fImarginal_path_*\fR, and
da32ae
 \fIsan_path_err_*\fR options will keep marginal, or \(dqshaky\(dq, paths from
da32ae
 being reinstated until they have been monitored for some time. This can cause
da32ae
 situations where all non-marginal paths are down, and no paths are usable
da32ae
 until multipathd detects this and reinstates a marginal path. If the multipath
da32ae
 device is not configured to queue IO in this case, it can cause IO errors to
da32ae
 occur, even though there are marginal paths available.  However, if this
da32ae
-option is set to \fIyes\fR, when one of the marginal path detecting methods
da32ae
+option is set to \fIon\fR, when one of the marginal path detecting methods
da32ae
 determines that a path is marginal, it will be reinstated and placed in a
da32ae
 seperate pathgroup that will only be used after all the non-marginal pathgroups
da32ae
 have been tried first. This prevents the possibility of IO errors occuring
da32ae
 while marginal paths are still usable. After the path has been monitored
da32ae
 for the configured time, and is declared healthy, it will be returned to its
da32ae
-normal pathgroup. See "Shaky paths detection" below for more information.
da32ae
+normal pathgroup.
da32ae
+However if this option is set to \fIfpin\fR multipathd will receive fpin
da32ae
+notifications, set path states to "marginal" accordingly, and regroup paths
da32ae
+as described for "marginal_pathgroups yes". This option can't be used in combination
da32ae
+with other options for "Shaky path detection" (see below).If it is set to fpin,
da32ae
+marginal_path_xyz and san_path_err_xyz parameters are implicitly set to 0.
da32ae
+See "Shaky paths detection" below for more information.
da32ae
 .RS
da32ae
 .TP
da32ae
 The default is: \fBno\fR
da32ae
@@ -1842,6 +1848,13 @@ increase and the threshold is never reached. Ticks are the time between
da32ae
 path checks by multipathd, which is variable and controlled by the
da32ae
 \fIpolling_interval\fR and \fImax_polling_interval\fR parameters.
da32ae
 .
da32ae
+.TP
da32ae
+.B \(dqFPIN \(dq failure tracking
da32ae
+Fibre channel fabrics can notify hosts about fabric-level issues such
da32ae
+as integrity failures or congestion with so-called Fabric Performance
da32ae
+Impact Notifications (FPINs).On receiving the fpin notifications through ELS
da32ae
+multipathd will move the affected path and port states to marginal.
da32ae
+.
da32ae
 .RS 8
da32ae
 .LP
da32ae
 This method is \fBdeprecated\fR in favor of the \(dqmarginal_path\(dq failure
da32ae
diff --git a/multipathd/Makefile b/multipathd/Makefile
da32ae
index 393b6cbb..cd6f7e6d 100644
da32ae
--- a/multipathd/Makefile
da32ae
+++ b/multipathd/Makefile
da32ae
@@ -4,6 +4,10 @@ ifneq ($(call check_func,dm_task_get_errno,/usr/include/libdevmapper.h),0)
da32ae
 	CFLAGS += -DLIBDM_API_GET_ERRNO
da32ae
 endif
da32ae
 
da32ae
+ifneq ($(call check_var,ELS_DTAG_LNK_INTEGRITY,/usr/include/scsi/fc/fc_els.h),0)
da32ae
+	CFLAGS += -DFPIN_EVENT_HANDLER
da32ae
+	FPIN_SUPPORT = 1
da32ae
+endif
da32ae
 #
da32ae
 # debugging stuff
da32ae
 #
da32ae
@@ -34,6 +38,12 @@ endif
da32ae
 OBJS = main.o pidfile.o uxlsnr.o uxclnt.o cli.o cli_handlers.o waiter.o \
da32ae
        dmevents.o init_unwinder.o
da32ae
 
da32ae
+ifeq ($(FPIN_SUPPORT),1)
da32ae
+OBJS += fpin_handlers.o
da32ae
+endif
da32ae
+
da32ae
+
da32ae
+
da32ae
 EXEC = multipathd
da32ae
 
da32ae
 all : $(EXEC)
da32ae
diff --git a/multipathd/fpin.h b/multipathd/fpin.h
da32ae
new file mode 100644
da32ae
index 00000000..bfcc1ce2
da32ae
--- /dev/null
da32ae
+++ b/multipathd/fpin.h
da32ae
@@ -0,0 +1,20 @@
da32ae
+#ifndef __FPIN_H__
da32ae
+#define __FPIN_H__
da32ae
+
da32ae
+#ifdef FPIN_EVENT_HANDLER
da32ae
+void *fpin_fabric_notification_receiver(void *unused);
da32ae
+void *fpin_els_li_consumer(void *data);
da32ae
+void fpin_clean_marginal_dev_list(__attribute__((unused)) void *arg);
da32ae
+#else
da32ae
+static void *fpin_fabric_notification_receiver(__attribute__((unused))void *unused)
da32ae
+{
da32ae
+	return NULL;
da32ae
+}
da32ae
+static void *fpin_els_li_consumer(__attribute__((unused))void *data)
da32ae
+{
da32ae
+	return NULL;
da32ae
+}
da32ae
+/* fpin_clean_marginal_dev_list() is never called */
da32ae
+#endif
da32ae
+
da32ae
+#endif
da32ae
diff --git a/multipathd/fpin_handlers.c b/multipathd/fpin_handlers.c
da32ae
new file mode 100644
da32ae
index 00000000..aaf5655d
da32ae
--- /dev/null
da32ae
+++ b/multipathd/fpin_handlers.c
da32ae
@@ -0,0 +1,540 @@
da32ae
+#include <errno.h>
da32ae
+#include <unistd.h>
da32ae
+#include <sys/types.h>
da32ae
+#include <sys/socket.h>
da32ae
+#include <libudev.h>
da32ae
+#include <scsi/scsi_netlink_fc.h>
da32ae
+#include <scsi/fc/fc_els.h>
da32ae
+
da32ae
+#include "parser.h"
da32ae
+#include "vector.h"
da32ae
+#include "structs.h"
da32ae
+#include "structs_vec.h"
da32ae
+#include "main.h"
da32ae
+#include "debug.h"
da32ae
+#include "util.h"
da32ae
+#include "sysfs.h"
da32ae
+
da32ae
+#include "fpin.h"
da32ae
+#include "devmapper.h"
da32ae
+
da32ae
+static pthread_cond_t fpin_li_cond = PTHREAD_COND_INITIALIZER;
da32ae
+static pthread_mutex_t fpin_li_mutex = PTHREAD_MUTEX_INITIALIZER;
da32ae
+static pthread_mutex_t fpin_li_marginal_dev_mutex = PTHREAD_MUTEX_INITIALIZER;
da32ae
+
da32ae
+static LIST_HEAD(els_marginal_list_head);
da32ae
+static LIST_HEAD(fpin_li_marginal_dev_list_head);
da32ae
+
da32ae
+
da32ae
+#define DEF_RX_BUF_SIZE	4096
da32ae
+#define DEV_NAME_LEN	128
da32ae
+#define FCH_EVT_LINKUP 0x2
da32ae
+#define FCH_EVT_LINK_FPIN 0x501
da32ae
+#define FCH_EVT_RSCN 0x5
da32ae
+
da32ae
+#define list_first_entry(ptr, type, member) \
da32ae
+	list_entry((ptr)->next, type, member)
da32ae
+
da32ae
+/* max ELS frame Size */
da32ae
+#define FC_PAYLOAD_MAXLEN   2048
da32ae
+
da32ae
+struct els_marginal_list {
da32ae
+	uint32_t event_code;
da32ae
+	uint16_t host_num;
da32ae
+	uint16_t length;
da32ae
+	char payload[FC_PAYLOAD_MAXLEN];
da32ae
+	struct list_head node;
da32ae
+};
da32ae
+/* Structure to store the marginal devices info */
da32ae
+struct marginal_dev_list {
da32ae
+	char dev_t[BLK_DEV_SIZE];
da32ae
+	uint32_t host_num;
da32ae
+	struct list_head node;
da32ae
+};
da32ae
+
da32ae
+static void _udev_device_unref(void *p)
da32ae
+{
da32ae
+	udev_device_unref(p);
da32ae
+}
da32ae
+
da32ae
+
da32ae
+/*set/unset the path state to marginal*/
da32ae
+static int fpin_set_pathstate(struct path *pp, bool set)
da32ae
+{
da32ae
+	const char *action = set ? "set" : "unset";
da32ae
+
da32ae
+	if (!pp || !pp->mpp || !pp->mpp->alias)
da32ae
+		return -1;
da32ae
+
da32ae
+	condlog(3, "\n%s: %s  marginal path %s (fpin)",
da32ae
+		action, pp->mpp->alias, pp->dev_t);
da32ae
+	pp->marginal = set;
da32ae
+	pp->mpp->fpin_must_reload = true;
da32ae
+	return 0;
da32ae
+}
da32ae
+
da32ae
+/* This will unset marginal state of a device*/
da32ae
+static void fpin_path_unsetmarginal(char *devname, struct vectors *vecs)
da32ae
+{
da32ae
+	struct path *pp;
da32ae
+
da32ae
+	pp = find_path_by_dev(vecs->pathvec, devname);
da32ae
+	if (!pp)
da32ae
+		pp = find_path_by_devt(vecs->pathvec, devname);
da32ae
+
da32ae
+	fpin_set_pathstate(pp, false);
da32ae
+}
da32ae
+
da32ae
+/*This will set the marginal state of a device*/
da32ae
+static int fpin_path_setmarginal(struct path *pp)
da32ae
+{
da32ae
+	return fpin_set_pathstate(pp, true);
da32ae
+}
da32ae
+
da32ae
+/* Unsets all the devices in the list from marginal state */
da32ae
+static void
da32ae
+fpin_unset_marginal_dev(uint32_t host_num, struct vectors *vecs)
da32ae
+{
da32ae
+	struct marginal_dev_list *tmp_marg = NULL;
da32ae
+	struct marginal_dev_list *marg = NULL;
da32ae
+	struct multipath *mpp;
da32ae
+	int ret = 0;
da32ae
+	int i;
da32ae
+
da32ae
+	pthread_cleanup_push(cleanup_lock, &vecs->lock);
da32ae
+	lock(&vecs->lock);
da32ae
+	pthread_testcancel();
da32ae
+
da32ae
+	pthread_mutex_lock(&fpin_li_marginal_dev_mutex);
da32ae
+	pthread_cleanup_push(cleanup_mutex, &fpin_li_marginal_dev_mutex);
da32ae
+	pthread_testcancel();
da32ae
+	if (list_empty(&fpin_li_marginal_dev_list_head)) {
da32ae
+		condlog(4, "Marginal List is empty\n");
da32ae
+		goto empty;
da32ae
+	}
da32ae
+	list_for_each_entry_safe(marg, tmp_marg, &fpin_li_marginal_dev_list_head, node) {
da32ae
+		if (marg->host_num != host_num)
da32ae
+			continue;
da32ae
+		condlog(4, " unsetting marginal dev: is %s %d\n",
da32ae
+				tmp_marg->dev_t, tmp_marg->host_num);
da32ae
+		fpin_path_unsetmarginal(marg->dev_t, vecs);
da32ae
+		list_del(&marg->node);
da32ae
+		free(marg);
da32ae
+	}
da32ae
+empty:
da32ae
+	pthread_cleanup_pop(1);
da32ae
+	/* walk backwards because reload_and_sync_map() can remove mpp */
da32ae
+	vector_foreach_slot_backwards(vecs->mpvec, mpp, i) {
da32ae
+		if (mpp->fpin_must_reload) {
da32ae
+			ret = reload_and_sync_map(mpp, vecs, 0);
da32ae
+			if (ret == 2)
da32ae
+				condlog(2, "map removed during reload");
da32ae
+			else
da32ae
+				mpp->fpin_must_reload = false;
da32ae
+		}
da32ae
+	}
da32ae
+	pthread_cleanup_pop(1);
da32ae
+}
da32ae
+
da32ae
+/*
da32ae
+ * On Receiving the frame from HBA driver, insert the frame into link
da32ae
+ * integrity frame list which will be picked up later by consumer thread for
da32ae
+ * processing.
da32ae
+ */
da32ae
+static int
da32ae
+fpin_els_add_li_frame(struct fc_nl_event *fc_event)
da32ae
+{
da32ae
+	struct els_marginal_list *els_mrg = NULL;
da32ae
+	int ret = 0;
da32ae
+
da32ae
+	if (fc_event->event_datalen > FC_PAYLOAD_MAXLEN)
da32ae
+		return -EINVAL;
da32ae
+
da32ae
+	pthread_mutex_lock(&fpin_li_mutex);
da32ae
+	pthread_cleanup_push(cleanup_mutex, &fpin_li_mutex);
da32ae
+	pthread_testcancel();
da32ae
+	els_mrg = calloc(1, sizeof(struct els_marginal_list));
da32ae
+	if (els_mrg != NULL) {
da32ae
+		els_mrg->host_num = fc_event->host_no;
da32ae
+		els_mrg->event_code = fc_event->event_code;
da32ae
+		els_mrg->length = fc_event->event_datalen;
da32ae
+		memcpy(els_mrg->payload, &(fc_event->event_data), fc_event->event_datalen);
da32ae
+		list_add_tail(&els_mrg->node, &els_marginal_list_head);
da32ae
+		pthread_cond_signal(&fpin_li_cond);
da32ae
+	} else
da32ae
+		ret = -ENOMEM;
da32ae
+	pthread_cleanup_pop(1);
da32ae
+	return ret;
da32ae
+
da32ae
+}
da32ae
+
da32ae
+/*Sets the rport port_state to marginal*/
da32ae
+static void fpin_set_rport_marginal(struct udev_device *rport_dev)
da32ae
+{
da32ae
+	sysfs_attr_set_value(rport_dev, "port_state",
da32ae
+				"Marginal", strlen("Marginal"));
da32ae
+}
da32ae
+
da32ae
+/*Add the marginal devices info into the list*/
da32ae
+static void
da32ae
+fpin_add_marginal_dev_info(uint32_t host_num, char *devname)
da32ae
+{
da32ae
+	struct marginal_dev_list *newdev = NULL;
da32ae
+
da32ae
+	newdev = calloc(1, sizeof(struct marginal_dev_list));
da32ae
+	if (newdev != NULL) {
da32ae
+		newdev->host_num = host_num;
da32ae
+		strlcpy(newdev->dev_t, devname, BLK_DEV_SIZE);
da32ae
+		condlog(4, "\n%s hostno %d devname %s\n", __func__,
da32ae
+				host_num, newdev->dev_t);
da32ae
+		pthread_mutex_lock(&fpin_li_marginal_dev_mutex);
da32ae
+		list_add_tail(&(newdev->node),
da32ae
+				&fpin_li_marginal_dev_list_head);
da32ae
+		pthread_mutex_unlock(&fpin_li_marginal_dev_mutex);
da32ae
+	}
da32ae
+}
da32ae
+
da32ae
+/*
da32ae
+ * This function goes through the vecs->pathvec, and for
da32ae
+ * each path, check that the host  number,
da32ae
+ * the target WWPN associated with the path matches
da32ae
+ * with the els wwpn and sets the path and port state to
da32ae
+ * Marginal
da32ae
+ */
da32ae
+static int  fpin_chk_wwn_setpath_marginal(uint16_t host_num,  struct vectors *vecs,
da32ae
+		uint64_t els_wwpn)
da32ae
+{
da32ae
+	struct path *pp;
da32ae
+	struct multipath *mpp;
da32ae
+	int i, k;
da32ae
+	char rport_id[42];
da32ae
+	const char *value = NULL;
da32ae
+	struct udev_device *rport_dev = NULL;
da32ae
+	uint64_t wwpn;
da32ae
+	int ret = 0;
da32ae
+
da32ae
+	pthread_cleanup_push(cleanup_lock, &vecs->lock);
da32ae
+	lock(&vecs->lock);
da32ae
+	pthread_testcancel();
da32ae
+
da32ae
+	vector_foreach_slot(vecs->pathvec, pp, k) {
da32ae
+		/* Checks the host number and also for the SCSI FCP */
da32ae
+		if (pp->sg_id.proto_id != SCSI_PROTOCOL_FCP || host_num !=  pp->sg_id.host_no)
da32ae
+			continue;
da32ae
+		sprintf(rport_id, "rport-%d:%d-%d",
da32ae
+				pp->sg_id.host_no, pp->sg_id.channel, pp->sg_id.transport_id);
da32ae
+		rport_dev = udev_device_new_from_subsystem_sysname(udev,
da32ae
+				"fc_remote_ports", rport_id);
da32ae
+		if (!rport_dev) {
da32ae
+			condlog(2, "%s: No fc_remote_port device for '%s'", pp->dev,
da32ae
+					rport_id);
da32ae
+			continue;
da32ae
+		}
da32ae
+		pthread_cleanup_push(_udev_device_unref, rport_dev);
da32ae
+		value = udev_device_get_sysattr_value(rport_dev, "port_name");
da32ae
+		if (!value)
da32ae
+			goto unref;
da32ae
+
da32ae
+		if (value)
da32ae
+			wwpn =  strtol(value, NULL, 16);
da32ae
+		/*
da32ae
+		 * If the port wwpn matches sets the path and port state
da32ae
+		 * to marginal
da32ae
+		 */
da32ae
+		if (wwpn == els_wwpn) {
da32ae
+			ret = fpin_path_setmarginal(pp);
da32ae
+			if (ret < 0)
da32ae
+				goto unref;
da32ae
+			fpin_set_rport_marginal(rport_dev);
da32ae
+			fpin_add_marginal_dev_info(host_num, pp->dev);
da32ae
+		}
da32ae
+unref:
da32ae
+		pthread_cleanup_pop(1);
da32ae
+	}
da32ae
+	/* walk backwards because reload_and_sync_map() can remove mpp */
da32ae
+	vector_foreach_slot_backwards(vecs->mpvec, mpp, i) {
da32ae
+		if (mpp->fpin_must_reload) {
da32ae
+			ret = reload_and_sync_map(mpp, vecs, 0);
da32ae
+			if (ret == 2)
da32ae
+				condlog(2, "map removed during reload");
da32ae
+			else
da32ae
+				mpp->fpin_must_reload = false;
da32ae
+		}
da32ae
+	}
da32ae
+	pthread_cleanup_pop(1);
da32ae
+	return ret;
da32ae
+}
da32ae
+
da32ae
+/*
da32ae
+ * This function loops around all the impacted wwns received as part of els
da32ae
+ * frame and sets the associated path and port states to marginal.
da32ae
+ */
da32ae
+static int
da32ae
+fpin_parse_li_els_setpath_marginal(uint16_t host_num, struct fc_tlv_desc *tlv,
da32ae
+		struct vectors *vecs)
da32ae
+{
da32ae
+	uint32_t wwn_count = 0, iter = 0;
da32ae
+	uint64_t wwpn;
da32ae
+	struct fc_fn_li_desc *li_desc = (struct fc_fn_li_desc *)tlv;
da32ae
+	int count = 0;
da32ae
+	int ret = 0;
da32ae
+
da32ae
+	/* Update the wwn to list */
da32ae
+	wwn_count = be32_to_cpu(li_desc->pname_count);
da32ae
+	condlog(4, "Got wwn count as %d\n", wwn_count);
da32ae
+
da32ae
+	for (iter = 0; iter < wwn_count; iter++) {
da32ae
+		wwpn = be64_to_cpu(li_desc->pname_list[iter]);
da32ae
+		ret = fpin_chk_wwn_setpath_marginal(host_num, vecs, wwpn);
da32ae
+		if (ret < 0)
da32ae
+			condlog(2, "failed to set the path marginal associated with wwpn: 0x%" PRIx64 "\n", wwpn);
da32ae
+
da32ae
+		count++;
da32ae
+	}
da32ae
+	return count;
da32ae
+}
da32ae
+
da32ae
+/*
da32ae
+ * This function process the ELS frame received from HBA driver,
da32ae
+ * and sets the path associated with the port wwn to marginal
da32ae
+ * and also set the port state to marginal.
da32ae
+ */
da32ae
+static int
da32ae
+fpin_process_els_frame(uint16_t host_num, char *fc_payload, struct vectors *vecs)
da32ae
+{
da32ae
+
da32ae
+	int count = -1;
da32ae
+	struct fc_els_fpin *fpin = (struct fc_els_fpin *)fc_payload;
da32ae
+	struct fc_tlv_desc *tlv;
da32ae
+
da32ae
+	tlv = (struct fc_tlv_desc *)&fpin->fpin_desc[0];
da32ae
+
da32ae
+	/*
da32ae
+	 * Parse the els frame and set the affected paths and port
da32ae
+	 * state to marginal
da32ae
+	 */
da32ae
+	count = fpin_parse_li_els_setpath_marginal(host_num, tlv, vecs);
da32ae
+	if (count <= 0)
da32ae
+		condlog(4, "Could not find any WWNs, ret = %d\n",
da32ae
+					count);
da32ae
+	return count;
da32ae
+}
da32ae
+
da32ae
+/*
da32ae
+ * This function process the FPIN ELS frame received from HBA driver,
da32ae
+ * and push the frame to appropriate frame list. Currently we have only FPIN
da32ae
+ * LI frame list.
da32ae
+ */
da32ae
+static int
da32ae
+fpin_handle_els_frame(struct fc_nl_event *fc_event)
da32ae
+{
da32ae
+	int ret = -1;
da32ae
+	uint32_t els_cmd;
da32ae
+	struct fc_els_fpin *fpin = (struct fc_els_fpin *)&fc_event->event_data;
da32ae
+	struct fc_tlv_desc *tlv;
da32ae
+	uint32_t dtag;
da32ae
+
da32ae
+	els_cmd = (uint32_t)fc_event->event_data;
da32ae
+	tlv = (struct fc_tlv_desc *)&fpin->fpin_desc[0];
da32ae
+	dtag = be32_to_cpu(tlv->desc_tag);
da32ae
+	condlog(4, "Got CMD in add as 0x%x fpin_cmd 0x%x dtag 0x%x\n",
da32ae
+			els_cmd, fpin->fpin_cmd, dtag);
da32ae
+
da32ae
+	if ((fc_event->event_code == FCH_EVT_LINK_FPIN) ||
da32ae
+			(fc_event->event_code == FCH_EVT_LINKUP) ||
da32ae
+			(fc_event->event_code == FCH_EVT_RSCN)) {
da32ae
+
da32ae
+		if (els_cmd == ELS_FPIN) {
da32ae
+			/*
da32ae
+			 * Check the type of fpin by checking the tag info
da32ae
+			 * At present we are supporting only LI events
da32ae
+			 */
da32ae
+			if (dtag == ELS_DTAG_LNK_INTEGRITY) {
da32ae
+				/*Push the Payload to FPIN frame queue. */
da32ae
+				ret = fpin_els_add_li_frame(fc_event);
da32ae
+				if (ret != 0)
da32ae
+					condlog(0, "Failed to process LI frame with error %d\n",
da32ae
+							ret);
da32ae
+			} else {
da32ae
+				condlog(4, "Unsupported FPIN received 0x%x\n", dtag);
da32ae
+				return ret;
da32ae
+			}
da32ae
+		} else {
da32ae
+			/*Push the Payload to FPIN frame queue. */
da32ae
+			ret = fpin_els_add_li_frame(fc_event);
da32ae
+			if (ret != 0)
da32ae
+				condlog(0, "Failed to process Linkup/RSCN event with error %d evnt %d\n",
da32ae
+						ret, fc_event->event_code);
da32ae
+		}
da32ae
+	} else
da32ae
+		condlog(4, "Invalid command received: 0x%x\n", els_cmd);
da32ae
+	return ret;
da32ae
+}
da32ae
+
da32ae
+/*cleans the global marginal dev list*/
da32ae
+void fpin_clean_marginal_dev_list(__attribute__((unused)) void *arg)
da32ae
+{
da32ae
+	struct marginal_dev_list *tmp_marg = NULL;
da32ae
+
da32ae
+	pthread_mutex_lock(&fpin_li_marginal_dev_mutex);
da32ae
+	while (!list_empty(&fpin_li_marginal_dev_list_head)) {
da32ae
+		tmp_marg  = list_first_entry(&fpin_li_marginal_dev_list_head,
da32ae
+				struct marginal_dev_list, node);
da32ae
+		list_del(&tmp_marg->node);
da32ae
+		free(tmp_marg);
da32ae
+	}
da32ae
+	pthread_mutex_unlock(&fpin_li_marginal_dev_mutex);
da32ae
+}
da32ae
+
da32ae
+/* Cleans the global els  marginal list */
da32ae
+static void fpin_clean_els_marginal_list(void *arg)
da32ae
+{
da32ae
+	struct list_head *head = (struct list_head *)arg;
da32ae
+	struct els_marginal_list *els_marg;
da32ae
+
da32ae
+	while (!list_empty(head)) {
da32ae
+		els_marg  = list_first_entry(head, struct els_marginal_list,
da32ae
+					     node);
da32ae
+		list_del(&els_marg->node);
da32ae
+		free(els_marg);
da32ae
+	}
da32ae
+}
da32ae
+
da32ae
+static void rcu_unregister(__attribute__((unused)) void *param)
da32ae
+{
da32ae
+	rcu_unregister_thread();
da32ae
+}
da32ae
+/*
da32ae
+ * This is the FPIN ELS consumer thread. The thread sleeps on pthread cond
da32ae
+ * variable unless notified by fpin_fabric_notification_receiver thread.
da32ae
+ * This thread is only to process FPIN-LI ELS frames. A new thread and frame
da32ae
+ * list will be added if any more ELS frames types are to be supported.
da32ae
+ */
da32ae
+void *fpin_els_li_consumer(void *data)
da32ae
+{
da32ae
+	struct list_head marginal_list_head;
da32ae
+	int ret = 0;
da32ae
+	uint16_t host_num;
da32ae
+	struct els_marginal_list *els_marg;
da32ae
+	uint32_t event_code;
da32ae
+	struct vectors *vecs = (struct vectors *)data;
da32ae
+
da32ae
+	pthread_cleanup_push(rcu_unregister, NULL);
da32ae
+	rcu_register_thread();
da32ae
+	pthread_cleanup_push(fpin_clean_marginal_dev_list, NULL);
da32ae
+	INIT_LIST_HEAD(&marginal_list_head);
da32ae
+	pthread_cleanup_push(fpin_clean_els_marginal_list,
da32ae
+				(void *)&marginal_list_head);
da32ae
+	for ( ; ; ) {
da32ae
+		pthread_mutex_lock(&fpin_li_mutex);
da32ae
+		pthread_cleanup_push(cleanup_mutex, &fpin_li_mutex);
da32ae
+		pthread_testcancel();
da32ae
+		while (list_empty(&els_marginal_list_head))
da32ae
+			pthread_cond_wait(&fpin_li_cond, &fpin_li_mutex);
da32ae
+
da32ae
+		if (!list_empty(&els_marginal_list_head)) {
da32ae
+			condlog(4, "Invoke List splice tail\n");
da32ae
+			list_splice_tail_init(&els_marginal_list_head, &marginal_list_head);
da32ae
+		}
da32ae
+		pthread_cleanup_pop(1);
da32ae
+
da32ae
+		while (!list_empty(&marginal_list_head)) {
da32ae
+			els_marg  = list_first_entry(&marginal_list_head,
da32ae
+							struct els_marginal_list, node);
da32ae
+			host_num = els_marg->host_num;
da32ae
+			event_code = els_marg->event_code;
da32ae
+			/* Now finally process FPIN LI ELS Frame */
da32ae
+			condlog(4, "Got a new Payload buffer, processing it\n");
da32ae
+			if ((event_code ==  FCH_EVT_LINKUP) || (event_code == FCH_EVT_RSCN))
da32ae
+				 fpin_unset_marginal_dev(host_num, vecs);
da32ae
+			else {
da32ae
+				ret = fpin_process_els_frame(host_num, els_marg->payload, vecs);
da32ae
+				if (ret <= 0)
da32ae
+					condlog(0, "ELS frame processing failed with ret %d\n", ret);
da32ae
+			}
da32ae
+			list_del(&els_marg->node);
da32ae
+			free(els_marg);
da32ae
+
da32ae
+		}
da32ae
+	}
da32ae
+
da32ae
+	pthread_cleanup_pop(1);
da32ae
+	pthread_cleanup_pop(1);
da32ae
+	pthread_cleanup_pop(1);
da32ae
+	return NULL;
da32ae
+}
da32ae
+
da32ae
+static void receiver_cleanup_list(__attribute__((unused)) void *arg)
da32ae
+{
da32ae
+	pthread_mutex_lock(&fpin_li_mutex);
da32ae
+	fpin_clean_els_marginal_list(&els_marginal_list_head);
da32ae
+	pthread_mutex_unlock(&fpin_li_mutex);
da32ae
+}
da32ae
+
da32ae
+/*
da32ae
+ * Listen for ELS frames from driver. on receiving the frame payload,
da32ae
+ * push the payload to a list, and notify the fpin_els_li_consumer thread to
da32ae
+ * process it. Once consumer thread is notified, return to listen for more ELS
da32ae
+ * frames from driver.
da32ae
+ */
da32ae
+void *fpin_fabric_notification_receiver(__attribute__((unused))void *unused)
da32ae
+{
da32ae
+	int ret;
da32ae
+	long fd;
da32ae
+	uint32_t els_cmd;
da32ae
+	struct fc_nl_event *fc_event = NULL;
da32ae
+	struct sockaddr_nl fc_local;
da32ae
+	unsigned char buf[DEF_RX_BUF_SIZE] __attribute__((aligned(sizeof(uint64_t))));
da32ae
+	size_t plen = 0;
da32ae
+
da32ae
+	pthread_cleanup_push(rcu_unregister, NULL);
da32ae
+	rcu_register_thread();
da32ae
+
da32ae
+	pthread_cleanup_push(receiver_cleanup_list, NULL);
da32ae
+	fd = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_SCSITRANSPORT);
da32ae
+	if (fd < 0) {
da32ae
+		condlog(0, "fc socket error %ld", fd);
da32ae
+		return NULL;
da32ae
+	}
da32ae
+
da32ae
+	pthread_cleanup_push(close_fd, (void *)fd);
da32ae
+	memset(&fc_local, 0, sizeof(fc_local));
da32ae
+	fc_local.nl_family = AF_NETLINK;
da32ae
+	fc_local.nl_groups = ~0;
da32ae
+	fc_local.nl_pid = getpid();
da32ae
+	ret = bind(fd, (struct sockaddr *)&fc_local, sizeof(fc_local));
da32ae
+	if (ret == -1) {
da32ae
+		condlog(0, "fc socket bind error %d\n", ret);
da32ae
+		goto out;
da32ae
+	}
da32ae
+	for ( ; ; ) {
da32ae
+		condlog(4, "Waiting for ELS...\n");
da32ae
+		ret = read(fd, buf, DEF_RX_BUF_SIZE);
da32ae
+		if (ret < 0) {
da32ae
+			condlog(0, "failed to read the els frame (%d)", ret);
da32ae
+			continue;
da32ae
+		}
da32ae
+		condlog(4, "Got a new request %d\n", ret);
da32ae
+		if (!NLMSG_OK((struct nlmsghdr *)buf, (unsigned int)ret)) {
da32ae
+			condlog(0, "bad els frame read (%d)", ret);
da32ae
+			continue;
da32ae
+		}
da32ae
+		/* Push the frame to appropriate frame list */
da32ae
+		plen = NLMSG_PAYLOAD((struct nlmsghdr *)buf, 0);
da32ae
+		fc_event = (struct fc_nl_event *)NLMSG_DATA(buf);
da32ae
+		if (plen < sizeof(*fc_event)) {
da32ae
+			condlog(0, "too short (%d) to be an FC event", ret);
da32ae
+			continue;
da32ae
+		}
da32ae
+		els_cmd = (uint32_t)fc_event->event_data;
da32ae
+		condlog(4, "Got host no as %d, event 0x%x, len %d evntnum %d evntcode %d\n",
da32ae
+				fc_event->host_no, els_cmd, fc_event->event_datalen,
da32ae
+				fc_event->event_num, fc_event->event_code);
da32ae
+		fpin_handle_els_frame(fc_event);
da32ae
+	}
da32ae
+out:
da32ae
+	pthread_cleanup_pop(1);
da32ae
+	pthread_cleanup_pop(1);
da32ae
+	pthread_cleanup_pop(1);
da32ae
+	return NULL;
da32ae
+}
da32ae
diff --git a/multipathd/main.c b/multipathd/main.c
da32ae
index 5def5301..53be9b95 100644
da32ae
--- a/multipathd/main.c
da32ae
+++ b/multipathd/main.c
da32ae
@@ -16,6 +16,7 @@
da32ae
 #include <linux/oom.h>
da32ae
 #include <libudev.h>
da32ae
 #include <urcu.h>
da32ae
+#include "fpin.h"
da32ae
 #ifdef USE_SYSTEMD
da32ae
 #include <systemd/sd-daemon.h>
da32ae
 #endif
da32ae
@@ -130,9 +131,11 @@ static volatile enum daemon_status running_state = DAEMON_INIT;
da32ae
 pid_t daemon_pid;
da32ae
 static pthread_mutex_t config_lock = PTHREAD_MUTEX_INITIALIZER;
da32ae
 static pthread_cond_t config_cond;
da32ae
-static pthread_t check_thr, uevent_thr, uxlsnr_thr, uevq_thr, dmevent_thr;
da32ae
+static pthread_t check_thr, uevent_thr, uxlsnr_thr, uevq_thr, dmevent_thr,
da32ae
+	fpin_thr, fpin_consumer_thr;
da32ae
 static bool check_thr_started, uevent_thr_started, uxlsnr_thr_started,
da32ae
-	uevq_thr_started, dmevent_thr_started;
da32ae
+	uevq_thr_started, dmevent_thr_started, fpin_thr_started,
da32ae
+	fpin_consumer_thr_started;
da32ae
 static int pid_fd = -1;
da32ae
 
da32ae
 static inline enum daemon_status get_running_state(void)
da32ae
@@ -2819,7 +2822,9 @@ reconfigure (struct vectors * vecs)
da32ae
 	conf->sequence_nr = old->sequence_nr + 1;
da32ae
 	rcu_assign_pointer(multipath_conf, conf);
da32ae
 	call_rcu(&old->rcu, rcu_free_config);
da32ae
-
da32ae
+#ifdef FPIN_EVENT_HANDLER
da32ae
+	fpin_clean_marginal_dev_list(NULL);
da32ae
+#endif
da32ae
 	configure(vecs);
da32ae
 
da32ae
 
da32ae
@@ -3060,6 +3065,11 @@ static void cleanup_threads(void)
da32ae
 		pthread_cancel(uevq_thr);
da32ae
 	if (dmevent_thr_started)
da32ae
 		pthread_cancel(dmevent_thr);
da32ae
+	if (fpin_thr_started)
da32ae
+		pthread_cancel(fpin_thr);
da32ae
+	if (fpin_consumer_thr_started)
da32ae
+		pthread_cancel(fpin_consumer_thr);
da32ae
+
da32ae
 
da32ae
 	if (check_thr_started)
da32ae
 		pthread_join(check_thr, NULL);
da32ae
@@ -3071,6 +3081,11 @@ static void cleanup_threads(void)
da32ae
 		pthread_join(uevq_thr, NULL);
da32ae
 	if (dmevent_thr_started)
da32ae
 		pthread_join(dmevent_thr, NULL);
da32ae
+	if (fpin_thr_started)
da32ae
+		pthread_join(fpin_thr, NULL);
da32ae
+	if (fpin_consumer_thr_started)
da32ae
+		pthread_join(fpin_consumer_thr, NULL);
da32ae
+
da32ae
 
da32ae
 	/*
da32ae
 	 * As all threads are joined now, and we're in DAEMON_SHUTDOWN
da32ae
@@ -3168,6 +3183,7 @@ child (__attribute__((unused)) void *param)
da32ae
 	char *envp;
da32ae
 	enum daemon_status state;
da32ae
 	int exit_code = 1;
da32ae
+	int fpin_marginal_paths = 0;
da32ae
 
da32ae
 	init_unwinder();
da32ae
 	mlockall(MCL_CURRENT | MCL_FUTURE);
da32ae
@@ -3246,7 +3262,10 @@ child (__attribute__((unused)) void *param)
da32ae
 
da32ae
 	setscheduler();
da32ae
 	set_oom_adj();
da32ae
-
da32ae
+#ifdef FPIN_EVENT_HANDLER
da32ae
+	if (conf->marginal_pathgroups == MARGINAL_PATHGROUP_FPIN)
da32ae
+		fpin_marginal_paths = 1;
da32ae
+#endif
da32ae
 	/*
da32ae
 	 * Startup done, invalidate configuration
da32ae
 	 */
da32ae
@@ -3314,6 +3333,22 @@ child (__attribute__((unused)) void *param)
da32ae
 		goto failed;
da32ae
 	} else
da32ae
 		uevq_thr_started = true;
da32ae
+
da32ae
+	if (fpin_marginal_paths) {
da32ae
+		if ((rc = pthread_create(&fpin_thr, &misc_attr,
da32ae
+			fpin_fabric_notification_receiver, NULL))) {
da32ae
+			condlog(0, "failed to create the fpin receiver thread: %d", rc);
da32ae
+			goto failed;
da32ae
+		} else
da32ae
+			fpin_thr_started = true;
da32ae
+
da32ae
+		if ((rc = pthread_create(&fpin_consumer_thr,
da32ae
+			&misc_attr, fpin_els_li_consumer, vecs))) {
da32ae
+			condlog(0, "failed to create the fpin consumer thread thread: %d", rc);
da32ae
+			goto failed;
da32ae
+		} else
da32ae
+			fpin_consumer_thr_started = true;
da32ae
+	}
da32ae
 	pthread_attr_destroy(&misc_attr);
da32ae
 
da32ae
 	while (1) {