[lustre-devel] [PATCH 19/20] lnet: o2iblnd: detect link state to set fatal error on ni
James Simmons
jsimmons at infradead.org
Fri Oct 14 14:38:10 PDT 2022
From: Serguei Smirnov <ssmirnov at whamcloud.com>
To avoid selecting lnet ni which corresponds to a downed link
for sending, add a mechanism for detecting ip-layer link events
in o2iblnd. On ip link up/down events, find corresponding
ni and toggle ni_fatal_error_on flag. This complements the
existing mechanism for ib-layer link event handling.
WC-bug-id: https://jira.whamcloud.com/browse/LU-16051
Lustre-commit: 30d73908087d5b2f0 ("LU-16051 o2iblnd: detect link state to set fatal error on ni")
Signed-off-by: Serguei Smirnov <ssmirnov at whamcloud.com>
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/48644
Reviewed-by: Frank Sehr <fsehr at whamcloud.com>
Reviewed-by: Chris Horn <chris.horn at hpe.com>
Reviewed-by: Cyril Bordage <cbordage at whamcloud.com>
Reviewed-by: James Simmons <jsimmons at infradead.org>
Reviewed-by: Oleg Drokin <green at whamcloud.com>
Signed-off-by: James Simmons <jsimmons at infradead.org>
---
net/lnet/klnds/o2iblnd/o2iblnd.c | 219 ++++++++++++++++++++++++++-----
1 file changed, 186 insertions(+), 33 deletions(-)
diff --git a/net/lnet/klnds/o2iblnd/o2iblnd.c b/net/lnet/klnds/o2iblnd/o2iblnd.c
index c713528a7e7c..d5ca1a3dd25c 100644
--- a/net/lnet/klnds/o2iblnd/o2iblnd.c
+++ b/net/lnet/klnds/o2iblnd/o2iblnd.c
@@ -2527,6 +2527,184 @@ void kiblnd_destroy_dev(struct kib_dev *dev)
kfree(dev);
}
+static struct kib_dev *
+kiblnd_dev_search(char *ifname)
+{
+ struct kib_dev *alias = NULL;
+ struct kib_dev *dev;
+ char *colon;
+ char *colon2;
+
+ colon = strchr(ifname, ':');
+ list_for_each_entry(dev, &kiblnd_data.kib_devs, ibd_list) {
+ if (strcmp(&dev->ibd_ifname[0], ifname) == 0)
+ return dev;
+
+ if (alias)
+ continue;
+
+ colon2 = strchr(dev->ibd_ifname, ':');
+ if (colon)
+ *colon = 0;
+ if (colon2)
+ *colon2 = 0;
+
+ if (strcmp(&dev->ibd_ifname[0], ifname) == 0)
+ alias = dev;
+
+ if (colon)
+ *colon = ':';
+ if (colon2)
+ *colon2 = ':';
+ }
+ return alias;
+}
+
+static int
+kiblnd_handle_link_state_change(struct net_device *dev,
+ unsigned char operstate)
+{
+ struct lnet_ni *ni = NULL;
+ struct kib_dev *event_kibdev;
+ struct kib_net *net;
+ struct kib_net *cnxt;
+ bool link_down = !(operstate == IF_OPER_UP);
+ struct in_device *in_dev;
+ bool found_ip = false;
+ const struct in_ifaddr *ifa;
+
+ event_kibdev = kiblnd_dev_search(dev->name);
+
+ if (!event_kibdev)
+ goto out;
+
+ list_for_each_entry_safe(net, cnxt, &event_kibdev->ibd_nets, ibn_list) {
+ found_ip = false;
+
+ ni = net->ibn_ni;
+
+ in_dev = __in_dev_get_rtnl(dev);
+ if (!in_dev) {
+ CDEBUG(D_NET, "Interface %s has no IPv4 status.\n",
+ dev->name);
+ CDEBUG(D_NET, "%s: set link fatal state to 1\n",
+ libcfs_nidstr(&net->ibn_ni->ni_nid));
+ atomic_set(&ni->ni_fatal_error_on, 1);
+ continue;
+ }
+ in_dev_for_each_ifa_rtnl(ifa, in_dev) {
+ if (htonl(event_kibdev->ibd_ifip) == ifa->ifa_local)
+ found_ip = true;
+ }
+
+ if (!found_ip) {
+ CDEBUG(D_NET, "Interface %s has no matching ip\n",
+ dev->name);
+ CDEBUG(D_NET, "%s: set link fatal state to 1\n",
+ libcfs_nidstr(&net->ibn_ni->ni_nid));
+ atomic_set(&ni->ni_fatal_error_on, 1);
+ continue;
+ }
+
+ if (link_down) {
+ CDEBUG(D_NET, "%s: set link fatal state to 1\n",
+ libcfs_nidstr(&net->ibn_ni->ni_nid));
+ atomic_set(&ni->ni_fatal_error_on, link_down);
+ } else {
+ CDEBUG(D_NET, "%s: set link fatal state to %u\n",
+ libcfs_nidstr(&net->ibn_ni->ni_nid),
+ (kiblnd_get_link_status(dev) == 0));
+ atomic_set(&ni->ni_fatal_error_on,
+ (kiblnd_get_link_status(dev) == 0));
+ }
+ }
+out:
+ return 0;
+}
+
+static int
+kiblnd_handle_inetaddr_change(struct in_ifaddr *ifa, unsigned long event)
+{
+ struct kib_dev *event_kibdev;
+ struct kib_net *net;
+ struct kib_net *cnxt;
+ struct net_device *event_netdev = ifa->ifa_dev->dev;
+
+ event_kibdev = kiblnd_dev_search(event_netdev->name);
+
+ if (!event_kibdev)
+ goto out;
+
+ if (htonl(event_kibdev->ibd_ifip) != ifa->ifa_local)
+ goto out;
+
+ list_for_each_entry_safe(net, cnxt, &event_kibdev->ibd_nets,
+ ibn_list) {
+ CDEBUG(D_NET, "%s: set link fatal state to %u\n",
+ libcfs_nidstr(&net->ibn_ni->ni_nid),
+ (event == NETDEV_DOWN));
+ atomic_set(&net->ibn_ni->ni_fatal_error_on,
+ (event == NETDEV_DOWN));
+ }
+out:
+ return 0;
+}
+
+/************************************
+ * Net device notifier event handler
+ ************************************/
+static int kiblnd_device_event(struct notifier_block *unused,
+ unsigned long event, void *ptr)
+{
+ struct net_device *dev = netdev_notifier_info_to_dev(ptr);
+ unsigned char operstate;
+
+ operstate = dev->operstate;
+
+ CDEBUG(D_NET, "devevent: status=%ld, iface=%s ifindex %d state %u\n",
+ event, dev->name, dev->ifindex, operstate);
+
+ switch (event) {
+ case NETDEV_UP:
+ case NETDEV_DOWN:
+ case NETDEV_CHANGE:
+ kiblnd_handle_link_state_change(dev, operstate);
+ break;
+ }
+
+ return NOTIFY_OK;
+}
+
+/************************************
+ * Inetaddr notifier event handler
+ ************************************/
+static int kiblnd_inetaddr_event(struct notifier_block *unused,
+ unsigned long event, void *ptr)
+{
+ struct in_ifaddr *ifa = ptr;
+
+ CDEBUG(D_NET, "addrevent: status %ld ip addr %pI4, netmask %pI4.\n",
+ event, &ifa->ifa_address, &ifa->ifa_mask);
+
+ switch (event) {
+ case NETDEV_UP:
+ case NETDEV_DOWN:
+ case NETDEV_CHANGE:
+ kiblnd_handle_inetaddr_change(ifa, event);
+ break;
+ }
+
+ return NOTIFY_OK;
+}
+
+static struct notifier_block kiblnd_dev_notifier_block = {
+ .notifier_call = kiblnd_device_event,
+};
+
+static struct notifier_block kiblnd_inetaddr_notifier_block = {
+ .notifier_call = kiblnd_inetaddr_event,
+};
+
static void kiblnd_base_shutdown(void)
{
struct kib_sched_info *sched;
@@ -2535,6 +2713,11 @@ static void kiblnd_base_shutdown(void)
LASSERT(list_empty(&kiblnd_data.kib_devs));
+ if (kiblnd_data.kib_init == IBLND_INIT_ALL) {
+ unregister_netdevice_notifier(&kiblnd_dev_notifier_block);
+ unregister_inetaddr_notifier(&kiblnd_inetaddr_notifier_block);
+ }
+
switch (kiblnd_data.kib_init) {
default:
LBUG();
@@ -2723,6 +2906,9 @@ static int kiblnd_base_startup(struct net *ns)
goto failed;
}
+ register_netdevice_notifier(&kiblnd_dev_notifier_block);
+ register_inetaddr_notifier(&kiblnd_inetaddr_notifier_block);
+
/* flag everything initialised */
kiblnd_data.kib_init = IBLND_INIT_ALL;
/*****************************************************/
@@ -2799,39 +2985,6 @@ static int kiblnd_dev_start_threads(struct kib_dev *dev, bool newdev, u32 *cpts,
return 0;
}
-static struct kib_dev *
-kiblnd_dev_search(char *ifname)
-{
- struct kib_dev *alias = NULL;
- struct kib_dev *dev;
- char *colon;
- char *colon2;
-
- colon = strchr(ifname, ':');
- list_for_each_entry(dev, &kiblnd_data.kib_devs, ibd_list) {
- if (strcmp(&dev->ibd_ifname[0], ifname) == 0)
- return dev;
-
- if (alias)
- continue;
-
- colon2 = strchr(dev->ibd_ifname, ':');
- if (colon)
- *colon = 0;
- if (colon2)
- *colon2 = 0;
-
- if (strcmp(&dev->ibd_ifname[0], ifname) == 0)
- alias = dev;
-
- if (colon)
- *colon = ':';
- if (colon2)
- *colon2 = ':';
- }
- return alias;
-}
-
static int kiblnd_startup(struct lnet_ni *ni)
{
char *ifname = NULL;
--
2.27.0
More information about the lustre-devel
mailing list