bond: Bonds never sleep if carrier changes.

[sliver-openvswitch.git] / lib / cfm.c

diff --git a/lib/cfm.c b/lib/cfm.c
index 943cfc1..a605c09 100644 (file)
--- a/lib/cfm.c
+++ b/lib/cfm.c
@@ -45,10 +45,10 @@ struct cfm_internal {
 
     struct timer tx_timer;    /* Send CCM when expired. */
     struct timer fault_timer; /* Check for faults when expired. */
-
-    long long x_recv_time;
 };
 
+static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
+
 static int
 ccm_interval_to_ms(uint8_t interval)
 {
@@ -129,8 +129,9 @@ lookup_remote_mp(const struct hmap *hmap, uint16_t mpid)
 }
 
 /* Allocates a 'cfm' object.  This object should have its 'mpid', 'maid',
- * 'eth_src', and 'interval' filled out.  When changes are made to the 'cfm'
- * object, cfm_configure should be called before using it. */
+ * 'eth_src', and 'interval' filled out.  cfm_configure() should be called
+ * whenever changes are made to 'cfm', and before cfm_run() is called for the
+ * first time. */
 struct cfm *
 cfm_create(void)
 {
@@ -139,7 +140,6 @@ cfm_create(void)
 
     cfmi = xzalloc(sizeof *cfmi);
     cfm  = &cfmi->cfm;
-    cfmi->x_recv_time = LLONG_MIN;
 
     hmap_init(&cfm->remote_mps);
     return cfm;
@@ -168,29 +168,28 @@ cfm_destroy(struct cfm *cfm)
 void
 cfm_run(struct cfm *cfm)
 {
-    long long now = time_msec();
     struct cfm_internal *cfmi = cfm_to_internal(cfm);
 
     if (timer_expired(&cfmi->fault_timer)) {
-        bool fault;
+        long long int interval = cfm_fault_interval(cfmi);
         struct remote_mp *rmp;
-        long long int interval;
-
-        interval = cfm_fault_interval(cfmi);
-        fault = now < cfmi->x_recv_time + interval;
 
+        cfm->fault = false;
         HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) {
-            if (rmp->recv_time < timer_enabled_at(&cfmi->fault_timer, interval)
-                || timer_expired_at(&cfmi->fault_timer, rmp->recv_time)) {
-                rmp->fault = true;
-            }
+            rmp->fault = !rmp->recv;
+            rmp->recv = false;
 
             if (rmp->fault) {
-                fault = true;
+                cfm->fault = true;
+                VLOG_DBG("No CCM from RMP %"PRIu16" in the last %lldms",
+                         rmp->mpid, interval);
             }
         }
 
-        cfm->fault = fault;
+        if (!cfm->fault) {
+            VLOG_DBG("All RMPs received CCMs in the last %lldms", interval);
+        }
+
         timer_set_duration(&cfmi->fault_timer, interval);
     }
 }
@@ -360,9 +359,7 @@ cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p)
     uint8_t ccm_interval;
     struct remote_mp *rmp;
     struct eth_header *eth;
-
-    struct cfm_internal *cfmi        = cfm_to_internal(cfm);
-    static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
+    struct cfm_internal *cfmi = cfm_to_internal(cfm);
 
     eth = p->l2;
     ccm = ofpbuf_at(p, (uint8_t *)p->l3 - (uint8_t *)p->data, CCM_LEN);
@@ -379,8 +376,8 @@ cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p)
     }
 
     /* According to the 802.1ag specification, reception of a CCM with an
-     * incorrect ccm_interval should trigger a fault.  We ignore this
-     * requirement for several reasons.
+     * incorrect ccm_interval, unexpected MAID, or unexpected MPID should
+     * trigger a fault.  We ignore this requirement for several reasons.
      *
      * Faults can cause a controller or Open vSwitch to make potentially
      * expensive changes to the network topology.  It seems prudent to trigger
@@ -388,8 +385,6 @@ cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p)
      * bonds. Furthermore, faults can be maliciously triggered by crafting
      * invalid CCMs. */
     if (memcmp(ccm->maid, cfm->maid, sizeof ccm->maid)) {
-        cfmi->x_recv_time = time_msec();
-        cfm->fault = true;
         VLOG_WARN_RL(&rl, "Received unexpected remote MAID from MAC "
                      ETH_ADDR_FMT, ETH_ADDR_ARGS(eth->eth_src));
     } else {
@@ -399,7 +394,7 @@ cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p)
         rmp = lookup_remote_mp(&cfm->remote_mps, ccm_mpid);
 
         if (rmp) {
-            rmp->recv_time = time_msec();
+            rmp->recv = true;
 
             if (ccm_interval != cfmi->ccm_interval) {
                 VLOG_WARN_RL(&rl, "received a CCM with an invalid interval"
@@ -407,11 +402,12 @@ cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p)
                              rmp->mpid);
             }
         } else {
-            cfmi->x_recv_time = time_msec();
-            cfm->fault = true;
             VLOG_WARN_RL(&rl, "Received unexpected remote MPID %d from MAC "
                          ETH_ADDR_FMT, ccm_mpid, ETH_ADDR_ARGS(eth->eth_src));
         }
+
+        VLOG_DBG("Received CCM (mpid %"PRIu16") (interval %"PRIu8")", ccm_mpid,
+                 ccm_interval);
     }
 }
 
@@ -419,7 +415,6 @@ void
 cfm_dump_ds(const struct cfm *cfm, struct ds *ds)
 {
     const struct cfm_internal *cfmi = cfm_to_internal(cfm);
-    long long int now = time_msec();
     struct remote_mp *rmp;
 
     ds_put_format(ds, "MPID %"PRIu16": %s\n", cfm->mpid,
@@ -431,16 +426,11 @@ cfm_dump_ds(const struct cfm *cfm, struct ds *ds)
     ds_put_format(ds, "\tnext fault check: %lldms\n",
                   timer_msecs_until_expired(&cfmi->fault_timer));
 
-    if (cfmi->x_recv_time != LLONG_MIN) {
-        ds_put_format(ds, "\ttime since bad CCM rx: %lldms\n",
-                      now - cfmi->x_recv_time);
-    }
-
     ds_put_cstr(ds, "\n");
     HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) {
         ds_put_format(ds, "Remote MPID %"PRIu16": %s\n", rmp->mpid,
                       rmp->fault ? "fault" : "");
-        ds_put_format(ds, "\ttime since CCM rx: %lldms\n",
-                      time_msec() - rmp->recv_time);
+        ds_put_format(ds, "\trecv since check: %s",
+                      rmp->recv ? "true" : "false");
     }
 }