[lustre-devel] [PATCH 21/42] lnet: handles unregister/register events

From: Cyril Bordage <[email protected]>

When network is restarted, devices are unregistered and then
registered again. When a device registers using an index that is
different from the previous one (before network was restarted), LNet
ignores it. Consequently, this device stays with link in fatal state.

To fix that, we catch unregistering events to clear the saved index
value, and when a registering event comes, we save the new value.

WC-bug-id: https://jira.whamcloud.com/browse/LU-16378
Lustre-commit: 3c9282a67d73799a0 ("LU-16378 lnet: handles unregister/register events")
Signed-off-by: Cyril Bordage <[email protected]>
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/49375
Reviewed-by: Serguei Smirnov <[email protected]>
Reviewed-by: Amir Shehata <[email protected]>
Reviewed-by: Oleg Drokin <[email protected]>
Signed-off-by: James Simmons <[email protected]>
---
 net/lnet/klnds/socklnd/socklnd.c | 26 ++++++++++++++++++++++++--
 1 file changed, 24 insertions(+), 2 deletions(-)

diff --git a/net/lnet/klnds/socklnd/socklnd.c b/net/lnet/klnds/socklnd/socklnd.c
index d8d1071d40f4..07e056845b24 100644
--- a/net/lnet/klnds/socklnd/socklnd.c
+++ b/net/lnet/klnds/socklnd/socklnd.c
@@ -2010,10 +2010,30 @@ ksocknal_handle_link_state_change(struct net_device *dev,
 		sa = (void *)&ksi->ksni_addr;
 		found_ip = false;
 
-		if (ksi->ksni_index != ifindex ||
-		    strcmp(ksi->ksni_name, dev->name))
+		if (strcmp(ksi->ksni_name, dev->name))
+			continue;
+
+		if (ksi->ksni_index == -1) {
+			if (dev->reg_state != NETREG_REGISTERED)
+				continue;
+			/* A registration just happened: save the new index for
+			 * the device
+			 */
+			ksi->ksni_index = ifindex;
+			goto out;
+		}
+
+		if (ksi->ksni_index != ifindex)
 			continue;
 
+		if (dev->reg_state == NETREG_UNREGISTERING) {
+			/* Device is being unregitering, we need to clear the
+			 * index, it can change when device will be back
+			 */
+			ksi->ksni_index = -1;
+			goto out;
+		}
+
 		ni = net->ksnn_ni;
 
 		in_dev = __in_dev_get_rtnl(dev);
@@ -2108,6 +2128,8 @@ static int ksocknal_device_event(struct notifier_block *unused,
 	case NETDEV_UP:
 	case NETDEV_DOWN:
 	case NETDEV_CHANGE:
+	case NETDEV_REGISTER:
+	case NETDEV_UNREGISTER:
 		ksocknal_handle_link_state_change(dev, operstate);
 		break;
 	}
-- 
2.27.0

_______________________________________________
lustre-devel mailing list
[email protected]
http://lists.lustre.org/listinfo.cgi/lustre-devel-lustre.org