net/smc: introduce list of pnetids for Ethernet devices
authorUrsula Braun <ubraun@linux.ibm.com>
Sat, 26 Sep 2020 10:44:26 +0000 (12:44 +0200)
committerDavid S. Miller <davem@davemloft.net>
Mon, 28 Sep 2020 22:19:03 +0000 (15:19 -0700)
SMCD version 2 allows usage of ISM devices with hardware PNETID
only, if an Ethernet net_device exists with the same hardware PNETID.
This requires to maintain a list of pnetids belonging to
Ethernet net_devices, which is covered by this patch.

Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/smc/smc_netns.h
net/smc/smc_pnet.c
net/smc/smc_pnet.h

index e7a8fc4..0f4f35a 100644 (file)
@@ -16,5 +16,6 @@ extern unsigned int smc_net_id;
 /* per-network namespace private data */
 struct smc_net {
        struct smc_pnettable pnettable;
+       struct smc_pnetids_ndev pnetids_ndev;
 };
 #endif
index 0b0f270..ee1456f 100644 (file)
@@ -29,6 +29,7 @@
 #include "smc_ism.h"
 #include "smc_core.h"
 
+static struct net_device *__pnet_find_base_ndev(struct net_device *ndev);
 static struct net_device *pnet_find_base_ndev(struct net_device *ndev);
 
 static const struct nla_policy smc_pnet_policy[SMC_PNETID_MAX + 1] = {
@@ -712,10 +713,115 @@ static struct genl_family smc_pnet_nl_family __ro_after_init = {
        .n_ops =  ARRAY_SIZE(smc_pnet_ops)
 };
 
+bool smc_pnet_is_ndev_pnetid(struct net *net, u8 *pnetid)
+{
+       struct smc_net *sn = net_generic(net, smc_net_id);
+       struct smc_pnetids_ndev_entry *pe;
+       bool rc = false;
+
+       read_lock(&sn->pnetids_ndev.lock);
+       list_for_each_entry(pe, &sn->pnetids_ndev.list, list) {
+               if (smc_pnet_match(pnetid, pe->pnetid)) {
+                       rc = true;
+                       goto unlock;
+               }
+       }
+
+unlock:
+       read_unlock(&sn->pnetids_ndev.lock);
+       return rc;
+}
+
+static int smc_pnet_add_pnetid(struct net *net, u8 *pnetid)
+{
+       struct smc_net *sn = net_generic(net, smc_net_id);
+       struct smc_pnetids_ndev_entry *pe, *pi;
+
+       pe = kzalloc(sizeof(*pe), GFP_KERNEL);
+       if (!pe)
+               return -ENOMEM;
+
+       write_lock(&sn->pnetids_ndev.lock);
+       list_for_each_entry(pi, &sn->pnetids_ndev.list, list) {
+               if (smc_pnet_match(pnetid, pe->pnetid)) {
+                       refcount_inc(&pi->refcnt);
+                       kfree(pe);
+                       goto unlock;
+               }
+       }
+       refcount_set(&pe->refcnt, 1);
+       memcpy(pe->pnetid, pnetid, SMC_MAX_PNETID_LEN);
+       list_add_tail(&pe->list, &sn->pnetids_ndev.list);
+
+unlock:
+       write_unlock(&sn->pnetids_ndev.lock);
+       return 0;
+}
+
+static void smc_pnet_remove_pnetid(struct net *net, u8 *pnetid)
+{
+       struct smc_net *sn = net_generic(net, smc_net_id);
+       struct smc_pnetids_ndev_entry *pe, *pe2;
+
+       write_lock(&sn->pnetids_ndev.lock);
+       list_for_each_entry_safe(pe, pe2, &sn->pnetids_ndev.list, list) {
+               if (smc_pnet_match(pnetid, pe->pnetid)) {
+                       if (refcount_dec_and_test(&pe->refcnt)) {
+                               list_del(&pe->list);
+                               kfree(pe);
+                       }
+                       break;
+               }
+       }
+       write_unlock(&sn->pnetids_ndev.lock);
+}
+
+static void smc_pnet_add_base_pnetid(struct net *net, struct net_device *dev,
+                                    u8 *ndev_pnetid)
+{
+       struct net_device *base_dev;
+
+       base_dev = __pnet_find_base_ndev(dev);
+       if (base_dev->flags & IFF_UP &&
+           !smc_pnetid_by_dev_port(base_dev->dev.parent, base_dev->dev_port,
+                                   ndev_pnetid)) {
+               /* add to PNETIDs list */
+               smc_pnet_add_pnetid(net, ndev_pnetid);
+       }
+}
+
+/* create initial list of netdevice pnetids */
+static void smc_pnet_create_pnetids_list(struct net *net)
+{
+       u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
+       struct net_device *dev;
+
+       rtnl_lock();
+       for_each_netdev(net, dev)
+               smc_pnet_add_base_pnetid(net, dev, ndev_pnetid);
+       rtnl_unlock();
+}
+
+/* clean up list of netdevice pnetids */
+static void smc_pnet_destroy_pnetids_list(struct net *net)
+{
+       struct smc_net *sn = net_generic(net, smc_net_id);
+       struct smc_pnetids_ndev_entry *pe, *temp_pe;
+
+       write_lock(&sn->pnetids_ndev.lock);
+       list_for_each_entry_safe(pe, temp_pe, &sn->pnetids_ndev.list, list) {
+               list_del(&pe->list);
+               kfree(pe);
+       }
+       write_unlock(&sn->pnetids_ndev.lock);
+}
+
 static int smc_pnet_netdev_event(struct notifier_block *this,
                                 unsigned long event, void *ptr)
 {
        struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
+       struct net *net = dev_net(event_dev);
+       u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
 
        switch (event) {
        case NETDEV_REBOOT:
@@ -725,6 +831,17 @@ static int smc_pnet_netdev_event(struct notifier_block *this,
        case NETDEV_REGISTER:
                smc_pnet_add_by_ndev(event_dev);
                return NOTIFY_OK;
+       case NETDEV_UP:
+               smc_pnet_add_base_pnetid(net, event_dev, ndev_pnetid);
+               return NOTIFY_OK;
+       case NETDEV_DOWN:
+               event_dev = __pnet_find_base_ndev(event_dev);
+               if (!smc_pnetid_by_dev_port(event_dev->dev.parent,
+                                           event_dev->dev_port, ndev_pnetid)) {
+                       /* remove from PNETIDs list */
+                       smc_pnet_remove_pnetid(net, ndev_pnetid);
+               }
+               return NOTIFY_OK;
        default:
                return NOTIFY_DONE;
        }
@@ -739,9 +856,14 @@ int smc_pnet_net_init(struct net *net)
 {
        struct smc_net *sn = net_generic(net, smc_net_id);
        struct smc_pnettable *pnettable = &sn->pnettable;
+       struct smc_pnetids_ndev *pnetids_ndev = &sn->pnetids_ndev;
 
        INIT_LIST_HEAD(&pnettable->pnetlist);
        rwlock_init(&pnettable->lock);
+       INIT_LIST_HEAD(&pnetids_ndev->list);
+       rwlock_init(&pnetids_ndev->lock);
+
+       smc_pnet_create_pnetids_list(net);
 
        return 0;
 }
@@ -756,6 +878,7 @@ int __init smc_pnet_init(void)
        rc = register_netdevice_notifier(&smc_netdev_notifier);
        if (rc)
                genl_unregister_family(&smc_pnet_nl_family);
+
        return rc;
 }
 
@@ -764,6 +887,7 @@ void smc_pnet_net_exit(struct net *net)
 {
        /* flush pnet table */
        smc_pnet_remove_by_pnetid(net, NULL);
+       smc_pnet_destroy_pnetids_list(net);
 }
 
 void smc_pnet_exit(void)
@@ -772,16 +896,11 @@ void smc_pnet_exit(void)
        genl_unregister_family(&smc_pnet_nl_family);
 }
 
-/* Determine one base device for stacked net devices.
- * If the lower device level contains more than one devices
- * (for instance with bonding slaves), just the first device
- * is used to reach a base device.
- */
-static struct net_device *pnet_find_base_ndev(struct net_device *ndev)
+static struct net_device *__pnet_find_base_ndev(struct net_device *ndev)
 {
        int i, nest_lvl;
 
-       rtnl_lock();
+       ASSERT_RTNL();
        nest_lvl = ndev->lower_level;
        for (i = 0; i < nest_lvl; i++) {
                struct list_head *lower = &ndev->adj_list.lower;
@@ -791,6 +910,18 @@ static struct net_device *pnet_find_base_ndev(struct net_device *ndev)
                lower = lower->next;
                ndev = netdev_lower_get_next(ndev, &lower);
        }
+       return ndev;
+}
+
+/* Determine one base device for stacked net devices.
+ * If the lower device level contains more than one devices
+ * (for instance with bonding slaves), just the first device
+ * is used to reach a base device.
+ */
+static struct net_device *pnet_find_base_ndev(struct net_device *ndev)
+{
+       rtnl_lock();
+       ndev = __pnet_find_base_ndev(ndev);
        rtnl_unlock();
        return ndev;
 }
index 811a659..677a47a 100644 (file)
@@ -12,6 +12,8 @@
 #ifndef _SMC_PNET_H
 #define _SMC_PNET_H
 
+#include <net/smc.h>
+
 #if IS_ENABLED(CONFIG_HAVE_PNETID)
 #include <asm/pnet.h>
 #endif
@@ -31,6 +33,17 @@ struct smc_pnettable {
        struct list_head pnetlist;
 };
 
+struct smc_pnetids_ndev {      /* list of pnetids for net devices in UP state*/
+       struct list_head        list;
+       rwlock_t                lock;
+};
+
+struct smc_pnetids_ndev_entry {
+       struct list_head        list;
+       u8                      pnetid[SMC_MAX_PNETID_LEN];
+       refcount_t              refcnt;
+};
+
 static inline int smc_pnetid_by_dev_port(struct device *dev,
                                         unsigned short port, u8 *pnetid)
 {
@@ -52,4 +65,5 @@ int smc_pnetid_by_table_smcd(struct smcd_dev *smcd);
 void smc_pnet_find_alt_roce(struct smc_link_group *lgr,
                            struct smc_init_info *ini,
                            struct smc_ib_device *known_dev);
+bool smc_pnet_is_ndev_pnetid(struct net *net, u8 *pnetid);
 #endif