device-mapper-multipath/0094-libmultipath-Add-max_retries-config-option.patch

From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: Benjamin Marzinski <bmarzins@redhat.com>
Date: Thu, 9 Nov 2023 18:46:11 -0500
Subject: [PATCH] libmultipath: Add max_retries config option

This option lets multipath set a scsi disk's max_retries sysfs value.
Setting this can be helpful for cases where the path checker succeeds,
but IO commands hang and timeout. By default, the SCSI layer will retry
IOs 5 times. Reducing this value will allow multipath to retry the IO
down another path sooner.

Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Reviewed-by: Martin Wilck <mwilck@suse.com>
---
 libmultipath/config.h      |  1 +
 libmultipath/dict.c        | 25 ++++++++++++++++++++++++
 libmultipath/discovery.c   | 40 +++++++++++++++++++++++++++++++++++++-
 libmultipath/structs.h     |  6 ++++++
 multipath/multipath.conf.5 | 14 +++++++++++++
 5 files changed, 85 insertions(+), 1 deletion(-)

diff --git a/libmultipath/config.h b/libmultipath/config.h
index c1e18363..b0ee8241 100644
--- a/libmultipath/config.h
+++ b/libmultipath/config.h
@@ -162,6 +162,7 @@ struct config {
 	int fast_io_fail;
 	unsigned int dev_loss;
 	int eh_deadline;
+	int max_retries;
 	int log_checker_err;
 	int allow_queueing;
 	int allow_usb_devices;
diff --git a/libmultipath/dict.c b/libmultipath/dict.c
index eb2f33a2..0c66c1e1 100644
--- a/libmultipath/dict.c
+++ b/libmultipath/dict.c
@@ -1206,6 +1206,30 @@ declare_hw_snprint(eh_deadline, print_undef_off_zero)
 declare_pc_handler(eh_deadline, set_undef_off_zero)
 declare_pc_snprint(eh_deadline, print_undef_off_zero)
 
+static int
+def_max_retries_handler(struct config *conf, vector strvec, const char *file,
+			int line_nr)
+{
+	char * buff;
+
+	buff = set_value(strvec);
+	if (!buff)
+		return 1;
+
+	if (strcmp(buff, "off") == 0)
+		conf->max_retries = MAX_RETRIES_OFF;
+	else if (strcmp(buff, "0") == 0)
+		conf->max_retries = MAX_RETRIES_ZERO;
+	else
+		do_set_int(strvec, &conf->max_retries, 1, 5, file, line_nr,
+			   buff);
+
+	free(buff);
+	return 0;
+}
+
+declare_def_snprint(max_retries, print_undef_off_zero)
+
 static int
 set_pgpolicy(vector strvec, void *ptr, const char *file, int line_nr)
 {
@@ -2143,6 +2167,7 @@ init_keywords(vector keywords)
 	install_keyword("fast_io_fail_tmo", &def_fast_io_fail_handler, &snprint_def_fast_io_fail);
 	install_keyword("dev_loss_tmo", &def_dev_loss_handler, &snprint_def_dev_loss);
 	install_keyword("eh_deadline", &def_eh_deadline_handler, &snprint_def_eh_deadline);
+	install_keyword("max_retries", &def_max_retries_handler, &snprint_def_max_retries);
 	install_keyword("bindings_file", &def_bindings_file_handler, &snprint_def_bindings_file);
 	install_keyword("wwids_file", &def_wwids_file_handler, &snprint_def_wwids_file);
 	install_keyword("prkeys_file", &def_prkeys_file_handler, &snprint_def_prkeys_file);
diff --git a/libmultipath/discovery.c b/libmultipath/discovery.c
index a592a54e..adf8bbaa 100644
--- a/libmultipath/discovery.c
+++ b/libmultipath/discovery.c
@@ -632,6 +632,42 @@ sysfs_set_eh_deadline(struct path *pp)
 	return (ret <= 0);
 }
 
+static int
+sysfs_set_max_retries(struct config *conf, struct path *pp)
+{
+	struct udev_device *parent;
+	char value[16];
+	STRBUF_ON_STACK(buf);
+	int ret, len;
+
+	if (conf->max_retries == MAX_RETRIES_UNSET)
+		return 0;
+
+	if (!pp->udev || pp->sg_id.host_no < 0)
+		return 1;
+
+	len = sprintf(value, "%d", (conf->max_retries == MAX_RETRIES_OFF)? -1 :
+				   (conf->max_retries == MAX_RETRIES_ZERO)? 0 :
+				   conf->max_retries);
+
+	parent = udev_device_get_parent_with_subsystem_devtype(pp->udev,
+			"scsi", "scsi_device");
+	if (!parent)
+		return 1;
+
+	if (print_strbuf(&buf, "scsi_disk/%i:%i:%i:%" PRIu64 "/max_retries",
+			 pp->sg_id.host_no, pp->sg_id.channel,
+			 pp->sg_id.scsi_id, pp->sg_id.lun) < 0)
+		return 1;
+
+	ret = sysfs_attr_set_value(parent, get_strbuf_str(&buf), value, len);
+	if (len != ret)
+		condlog(3, "%s/%s: failed to set value to %s: %s",
+			udev_device_get_sysname(parent), get_strbuf_str(&buf),
+			value, (ret < 0)? strerror(-ret) : "write underflow");
+	return (len != ret);
+}
+
 static void
 sysfs_set_rport_tmo(struct multipath *mpp, struct path *pp)
 {
@@ -862,13 +898,15 @@ sysfs_set_scsi_tmo (struct config *conf, struct multipath *mpp)
 
 		if (pp->dev_loss == DEV_LOSS_TMO_UNSET &&
 		    pp->fast_io_fail == MP_FAST_IO_FAIL_UNSET &&
-		    pp->eh_deadline == EH_DEADLINE_UNSET)
+		    pp->eh_deadline == EH_DEADLINE_UNSET &&
+		    conf->max_retries == MAX_RETRIES_UNSET)
 			continue;
 
 		if (pp->bus != SYSFS_BUS_SCSI)
 			continue;
 
 		sysfs_set_eh_deadline(pp);
+		sysfs_set_max_retries(conf, pp);
 
 		if (pp->dev_loss == DEV_LOSS_TMO_UNSET &&
 		    pp->fast_io_fail == MP_FAST_IO_FAIL_UNSET)
diff --git a/libmultipath/structs.h b/libmultipath/structs.h
index c1e93e6e..b4252ab5 100644
--- a/libmultipath/structs.h
+++ b/libmultipath/structs.h
@@ -276,6 +276,12 @@ enum eh_deadline_states {
 	EH_DEADLINE_ZERO = UOZ_ZERO,
 };
 
+enum max_retries_states {
+	MAX_RETRIES_UNSET = UOZ_UNDEF,
+	MAX_RETRIES_OFF = UOZ_OFF,
+	MAX_RETRIES_ZERO = UOZ_ZERO,
+};
+
 enum recheck_wwid_states {
 	RECHECK_WWID_UNDEF = YNU_UNDEF,
 	RECHECK_WWID_OFF = YNU_NO,
diff --git a/multipath/multipath.conf.5 b/multipath/multipath.conf.5
index 5e447e67..789f0bfc 100644
--- a/multipath/multipath.conf.5
+++ b/multipath/multipath.conf.5
@@ -743,6 +743,20 @@ The default is: \fB<unset>\fR
 .
 .
 .TP
+.B max_retries
+Specify the maximum number of times the SCSI layer will retry IO commands for
+some types of SCSI errors before returning failure. Setting this can be helpful
+for cases where IO commands hang and timeout. By default, the SCSI layer will
+retry IOs 5 times. Reducing this value will allow multipath to retry the IO
+down another path sooner. Valid values are
+\fB0\fR through \fB5\fR.
+.RS
+.TP
+The default is: \fB<unset>\fR
+.RE
+.
+.
+.TP
 .B bindings_file
 This option is deprecated, and will be removed in a future release.
 The full pathname of the binding file to be used when the user_friendly_names