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@xxxxxxxxxx>
---
libmultipath/config.h | 1 +
libmultipath/dict.c | 25 +++++++++++++++++++++
libmultipath/discovery.c | 41 ++++++++++++++++++++++++++++++++++-
libmultipath/structs.h | 6 +++++
multipath/multipath.conf.5.in | 14 ++++++++++++
5 files changed, 86 insertions(+), 1 deletion(-)
diff --git a/libmultipath/config.h b/libmultipath/config.h
index 8c22ce75..417e5834 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 044067af..e268673f 100644
--- a/libmultipath/dict.c
+++ b/libmultipath/dict.c
@@ -1152,6 +1152,30 @@ 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)
{
@@ -2079,6 +2103,7 @@ 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", &deprecated_bindings_file_handler, &snprint_deprecated);
install_keyword("wwids_file", &deprecated_wwids_file_handler, &snprint_deprecated);
install_keyword("prkeys_file", &deprecated_prkeys_file_handler, &snprint_deprecated);
diff --git a/libmultipath/discovery.c b/libmultipath/discovery.c
index 84ce5fe7..6fd4dabb 100644
--- a/libmultipath/discovery.c
+++ b/libmultipath/discovery.c
@@ -614,6 +614,43 @@ 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)
+ log_sysfs_attr_set_value(3, ret,
+ "%s/%s: failed to set value to %s",
+ udev_device_get_sysname(parent),
+ get_strbuf_str(&buf), value);
+ return (len != ret);
+}
+
static void
sysfs_set_rport_tmo(struct multipath *mpp, struct path *pp)
{
@@ -878,7 +915,8 @@ 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) {
@@ -886,6 +924,7 @@ sysfs_set_scsi_tmo (struct config *conf, struct multipath *mpp)
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 17e13ee7..63551b80 100644
--- a/libmultipath/structs.h
+++ b/libmultipath/structs.h
@@ -295,6 +295,12 @@ 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.in b/multipath/multipath.conf.5.in
index 226d0019..014d6dd1 100644
--- a/multipath/multipath.conf.5.in
+++ b/multipath/multipath.conf.5.in
@@ -793,6 +793,20 @@ 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
(Deprecated) This option is not supported any more, and will be ignored.
.RS
--
2.41.0
