| Message ID | 1434496033-4601-1-git-send-email-cleech@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 06/16/2015 06:07 PM, Chris Leech wrote: > The iSCSI session recovery_tmo setting is writeable in sysfs, but it's > also set every time a connection is established when parameters are set > from iscsid over netlink. That results in the timeout being reset to > the default value after every recovery. > > The DM multipath tools want to use the sysfs interface to lower the > default timeout when there are multiple paths to fail over. It has > caused confusion that we have a writeable sysfs value that seem to keep > resetting itself. > > This patch adds an in-kernel flag that gets set once a sysfs write > occurs, and then ignores netlink parameter setting once it's been > modified via the sysfs interface. My thinking here is that the sysfs > interface is much simpler for external tools to influence the session > timeout, but if we're going to allow it to be modified directly we > should ensure that setting is maintained. > What happened? Why didn't you make it more generic so all future iscsi sysfs settings work the same way like when I reviewed it in the bz? Did it get too messy when we only have the one writeable attr? -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, Jun 17, 2015 at 09:33:04AM -0500, Mike Christie wrote: > On 06/16/2015 06:07 PM, Chris Leech wrote: > > The iSCSI session recovery_tmo setting is writeable in sysfs, but it's > > also set every time a connection is established when parameters are set > > from iscsid over netlink. That results in the timeout being reset to > > the default value after every recovery. > > > > The DM multipath tools want to use the sysfs interface to lower the > > default timeout when there are multiple paths to fail over. It has > > caused confusion that we have a writeable sysfs value that seem to keep > > resetting itself. > > > > This patch adds an in-kernel flag that gets set once a sysfs write > > occurs, and then ignores netlink parameter setting once it's been > > modified via the sysfs interface. My thinking here is that the sysfs > > interface is much simpler for external tools to influence the session > > timeout, but if we're going to allow it to be modified directly we > > should ensure that setting is maintained. > > > > What happened? Why didn't you make it more generic so all future iscsi > sysfs settings work the same way like when I reviewed it in the bz? Did > it get too messy when we only have the one writeable attr? I kept the macro declaration, previous revision had that removed, and modified it so that it could be used for future rw attrs in the same manner. It's possible something could be done for the override check in iscsi_set_param, but there's not a lot of code to share here. - Chris -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 06/16/2015 06:07 PM, Chris Leech wrote: > The iSCSI session recovery_tmo setting is writeable in sysfs, but it's > also set every time a connection is established when parameters are set > from iscsid over netlink. That results in the timeout being reset to > the default value after every recovery. > > The DM multipath tools want to use the sysfs interface to lower the > default timeout when there are multiple paths to fail over. It has > caused confusion that we have a writeable sysfs value that seem to keep > resetting itself. > > This patch adds an in-kernel flag that gets set once a sysfs write > occurs, and then ignores netlink parameter setting once it's been > modified via the sysfs interface. My thinking here is that the sysfs > interface is much simpler for external tools to influence the session > timeout, but if we're going to allow it to be modified directly we > should ensure that setting is maintained. > > Signed-off-by: Chris Leech <cleech@redhat.com> > --- > drivers/scsi/scsi_transport_iscsi.c | 11 ++++++++--- > include/scsi/scsi_transport_iscsi.h | 1 + > 2 files changed, 9 insertions(+), 3 deletions(-) > > diff --git a/drivers/scsi/scsi_transport_iscsi.c b/drivers/scsi/scsi_transport_iscsi.c > index 67d43e3..35ef55f 100644 > --- a/drivers/scsi/scsi_transport_iscsi.c > +++ b/drivers/scsi/scsi_transport_iscsi.c > @@ -2040,6 +2040,7 @@ iscsi_alloc_session(struct Scsi_Host *shost, struct iscsi_transport *transport, > session->transport = transport; > session->creator = -1; > session->recovery_tmo = 120; > + session->recovery_tmo_sysfs_override = false; > session->state = ISCSI_SESSION_FREE; > INIT_DELAYED_WORK(&session->recovery_work, session_recovery_timedout); > INIT_LIST_HEAD(&session->sess_list); > @@ -2784,7 +2785,8 @@ iscsi_set_param(struct iscsi_transport *transport, struct iscsi_uevent *ev) > switch (ev->u.set_param.param) { > case ISCSI_PARAM_SESS_RECOVERY_TMO: > sscanf(data, "%d", &value); > - session->recovery_tmo = value; > + if (!session->recovery_tmo_sysfs_override) > + session->recovery_tmo = value; > break; > default: > err = transport->set_param(conn, ev->u.set_param.param, > @@ -4047,13 +4049,15 @@ store_priv_session_##field(struct device *dev, \ > if ((session->state == ISCSI_SESSION_FREE) || \ > (session->state == ISCSI_SESSION_FAILED)) \ > return -EBUSY; \ > - if (strncmp(buf, "off", 3) == 0) \ > + if (strncmp(buf, "off", 3) == 0) { \ > session->field = -1; \ > - else { \ > + session->field##_sysfs_override = true; \ > + } else { \ > val = simple_strtoul(buf, &cp, 0); \ > if (*cp != '\0' && *cp != '\n') \ > return -EINVAL; \ > session->field = val; \ > + session->field##_sysfs_override = true; \ > } \ > return count; \ > } > @@ -4064,6 +4068,7 @@ store_priv_session_##field(struct device *dev, \ > static ISCSI_CLASS_ATTR(priv_sess, field, S_IRUGO | S_IWUSR, \ > show_priv_session_##field, \ > store_priv_session_##field) > + > iscsi_priv_session_rw_attr(recovery_tmo, "%d"); > > static struct attribute *iscsi_session_attrs[] = { > diff --git a/include/scsi/scsi_transport_iscsi.h b/include/scsi/scsi_transport_iscsi.h > index 2555ee5..6183d20 100644 > --- a/include/scsi/scsi_transport_iscsi.h > +++ b/include/scsi/scsi_transport_iscsi.h > @@ -241,6 +241,7 @@ struct iscsi_cls_session { > > /* recovery fields */ > int recovery_tmo; > + bool recovery_tmo_sysfs_override; > struct delayed_work recovery_work; > > unsigned int target_id; > Looks ok to me. Reviewed-by: Mike Christie <michaelc@cs.wisc.edu> -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/scsi/scsi_transport_iscsi.c b/drivers/scsi/scsi_transport_iscsi.c index 67d43e3..35ef55f 100644 --- a/drivers/scsi/scsi_transport_iscsi.c +++ b/drivers/scsi/scsi_transport_iscsi.c @@ -2040,6 +2040,7 @@ iscsi_alloc_session(struct Scsi_Host *shost, struct iscsi_transport *transport, session->transport = transport; session->creator = -1; session->recovery_tmo = 120; + session->recovery_tmo_sysfs_override = false; session->state = ISCSI_SESSION_FREE; INIT_DELAYED_WORK(&session->recovery_work, session_recovery_timedout); INIT_LIST_HEAD(&session->sess_list); @@ -2784,7 +2785,8 @@ iscsi_set_param(struct iscsi_transport *transport, struct iscsi_uevent *ev) switch (ev->u.set_param.param) { case ISCSI_PARAM_SESS_RECOVERY_TMO: sscanf(data, "%d", &value); - session->recovery_tmo = value; + if (!session->recovery_tmo_sysfs_override) + session->recovery_tmo = value; break; default: err = transport->set_param(conn, ev->u.set_param.param, @@ -4047,13 +4049,15 @@ store_priv_session_##field(struct device *dev, \ if ((session->state == ISCSI_SESSION_FREE) || \ (session->state == ISCSI_SESSION_FAILED)) \ return -EBUSY; \ - if (strncmp(buf, "off", 3) == 0) \ + if (strncmp(buf, "off", 3) == 0) { \ session->field = -1; \ - else { \ + session->field##_sysfs_override = true; \ + } else { \ val = simple_strtoul(buf, &cp, 0); \ if (*cp != '\0' && *cp != '\n') \ return -EINVAL; \ session->field = val; \ + session->field##_sysfs_override = true; \ } \ return count; \ } @@ -4064,6 +4068,7 @@ store_priv_session_##field(struct device *dev, \ static ISCSI_CLASS_ATTR(priv_sess, field, S_IRUGO | S_IWUSR, \ show_priv_session_##field, \ store_priv_session_##field) + iscsi_priv_session_rw_attr(recovery_tmo, "%d"); static struct attribute *iscsi_session_attrs[] = { diff --git a/include/scsi/scsi_transport_iscsi.h b/include/scsi/scsi_transport_iscsi.h index 2555ee5..6183d20 100644 --- a/include/scsi/scsi_transport_iscsi.h +++ b/include/scsi/scsi_transport_iscsi.h @@ -241,6 +241,7 @@ struct iscsi_cls_session { /* recovery fields */ int recovery_tmo; + bool recovery_tmo_sysfs_override; struct delayed_work recovery_work; unsigned int target_id;
The iSCSI session recovery_tmo setting is writeable in sysfs, but it's also set every time a connection is established when parameters are set from iscsid over netlink. That results in the timeout being reset to the default value after every recovery. The DM multipath tools want to use the sysfs interface to lower the default timeout when there are multiple paths to fail over. It has caused confusion that we have a writeable sysfs value that seem to keep resetting itself. This patch adds an in-kernel flag that gets set once a sysfs write occurs, and then ignores netlink parameter setting once it's been modified via the sysfs interface. My thinking here is that the sysfs interface is much simpler for external tools to influence the session timeout, but if we're going to allow it to be modified directly we should ensure that setting is maintained. Signed-off-by: Chris Leech <cleech@redhat.com> --- drivers/scsi/scsi_transport_iscsi.c | 11 ++++++++--- include/scsi/scsi_transport_iscsi.h | 1 + 2 files changed, 9 insertions(+), 3 deletions(-)