X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto-dpif-monitor.c;h=764c7b19939595dc8769cf005814c45302ec0be2;hb=HEAD;hp=a0c3843f1dda3e790b7c9433a590bd18b637236c;hpb=0141e875a14cc30fe9c207a83616861055531d4d;p=sliver-openvswitch.git

diff --git a/ofproto/ofproto-dpif-monitor.c b/ofproto/ofproto-dpif-monitor.c
index a0c3843f1..764c7b199 100644
--- a/ofproto/ofproto-dpif-monitor.c
+++ b/ofproto/ofproto-dpif-monitor.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
+ * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -21,16 +21,31 @@
 
 #include "bfd.h"
 #include "cfm.h"
+#include "guarded-list.h"
 #include "hash.h"
+#include "heap.h"
 #include "hmap.h"
+#include "latch.h"
 #include "ofpbuf.h"
 #include "ofproto-dpif.h"
+#include "ovs-thread.h"
+#include "poll-loop.h"
+#include "seq.h"
+#include "timeval.h"
 #include "util.h"
 #include "vlog.h"
 
+VLOG_DEFINE_THIS_MODULE(ofproto_dpif_monitor);
+
+/* Converts the time in millisecond to heap priority. */
+#define MSEC_TO_PRIO(TIME) (LLONG_MAX - (TIME))
+/* Converts the heap priority to time in millisecond. */
+#define PRIO_TO_MSEC(PRIO) (LLONG_MAX - (PRIO))
+
 /* Monitored port.  It owns references to ofport, bfd, cfm structs. */
 struct mport {
     struct hmap_node hmap_node;       /* In monitor_hmap. */
+    struct heap_node heap_node;       /* In monitor_heap. */
     const struct ofport_dpif *ofport; /* The corresponding ofport. */
 
     struct cfm *cfm;                  /* Reference to cfm. */
@@ -38,25 +53,51 @@ struct mport {
     uint8_t hw_addr[OFP_ETH_ALEN];    /* Hardware address. */
 };
 
+/* Entry of the 'send_soon' list.  Contains the pointer to the
+ * 'ofport_dpif'.  Note, the pointed object is not protected, so
+ * users should always use the mport_find() to convert it to 'mport'. */
+struct send_soon_entry {
+    struct list list_node;            /* In send_soon. */
+    const struct ofport_dpif *ofport;
+};
+
 /* hmap that contains "struct mport"s. */
 static struct hmap monitor_hmap = HMAP_INITIALIZER(&monitor_hmap);
 
-static struct ovs_rwlock monitor_rwlock = OVS_RWLOCK_INITIALIZER;
+/* heap for ordering mport based on bfd/cfm wakeup time. */
+static struct heap monitor_heap;
+
+/* guarded-list for storing the mports that need to send bfd/cfm control
+ * packet soon. */
+static struct guarded_list send_soon = GUARDED_LIST_INITIALIZER(&send_soon);
+
+/* The monitor thread id. */
+static pthread_t monitor_tid;
+/* True if the monitor thread is running. */
+static bool monitor_running;
+
+static struct latch monitor_exit_latch;
+static struct ovs_mutex monitor_mutex = OVS_MUTEX_INITIALIZER;
+
+static void *monitor_main(void *);
+static void monitor_check_send_soon(struct ofpbuf *);
+static void monitor_run(void);
+static void monitor_mport_run(struct mport *, struct ofpbuf *);
 
 static void mport_register(const struct ofport_dpif *, struct bfd *,
                            struct cfm *, uint8_t[ETH_ADDR_LEN])
-    OVS_REQ_WRLOCK(monitor_rwlock);
+    OVS_REQUIRES(monitor_mutex);
 static void mport_unregister(const struct ofport_dpif *)
-    OVS_REQ_WRLOCK(monitor_rwlock);
+    OVS_REQUIRES(monitor_mutex);
 static void mport_update(struct mport *, struct bfd *, struct cfm *,
-                         uint8_t[ETH_ADDR_LEN]) OVS_REQ_WRLOCK(monitor_rwlock);
+                         uint8_t[ETH_ADDR_LEN]) OVS_REQUIRES(monitor_mutex);
 static struct mport *mport_find(const struct ofport_dpif *)
-    OVS_REQ_WRLOCK(monitor_rwlock);
+    OVS_REQUIRES(monitor_mutex);
 
 /* Tries finding and returning the 'mport' from the monitor_hmap.
  * If there is no such 'mport', returns NULL. */
 static struct mport *
-mport_find(const struct ofport_dpif *ofport) OVS_REQ_WRLOCK(monitor_rwlock)
+mport_find(const struct ofport_dpif *ofport) OVS_REQUIRES(monitor_mutex)
 {
     struct mport *node;
 
@@ -69,12 +110,12 @@ mport_find(const struct ofport_dpif *ofport) OVS_REQ_WRLOCK(monitor_rwlock)
     return NULL;
 }
 
-/* Creates a new mport and inserts it into monitor_hmap, if it doesn't exist.
- * Otherwise, just updates its fields. */
+/* Creates a new mport and inserts it into monitor_hmap and monitor_heap,
+ * if it doesn't exist.  Otherwise, just updates its fields. */
 static void
 mport_register(const struct ofport_dpif *ofport, struct bfd *bfd,
                struct cfm *cfm, uint8_t *hw_addr)
-    OVS_REQ_WRLOCK(monitor_rwlock)
+    OVS_REQUIRES(monitor_mutex)
 {
     struct mport *mport = mport_find(ofport);
 
@@ -82,20 +123,22 @@ mport_register(const struct ofport_dpif *ofport, struct bfd *bfd,
         mport = xzalloc(sizeof *mport);
         mport->ofport = ofport;
         hmap_insert(&monitor_hmap, &mport->hmap_node, hash_pointer(ofport, 0));
+        heap_insert(&monitor_heap, &mport->heap_node, 0);
     }
     mport_update(mport, bfd, cfm, hw_addr);
 }
 
-/* Removes mport from monitor_hmap and frees it. */
+/* Removes mport from monitor_hmap and monitor_heap and frees it. */
 static void
 mport_unregister(const struct ofport_dpif *ofport)
-    OVS_REQ_WRLOCK(monitor_rwlock)
+    OVS_REQUIRES(monitor_mutex)
 {
     struct mport *mport = mport_find(ofport);
 
     if (mport) {
         mport_update(mport, NULL, NULL, NULL);
         hmap_remove(&monitor_hmap, &mport->hmap_node);
+        heap_remove(&monitor_heap, &mport->heap_node);
         free(mport);
     }
 }
@@ -103,7 +146,7 @@ mport_unregister(const struct ofport_dpif *ofport)
 /* Updates the fields of an existing mport struct. */
 static void
 mport_update(struct mport *mport, struct bfd *bfd, struct cfm *cfm,
-             uint8_t hw_addr[ETH_ADDR_LEN]) OVS_REQ_WRLOCK(monitor_rwlock)
+             uint8_t hw_addr[ETH_ADDR_LEN]) OVS_REQUIRES(monitor_mutex)
 {
     ovs_assert(mport);
 
@@ -118,82 +161,173 @@ mport_update(struct mport *mport, struct bfd *bfd, struct cfm *cfm,
     if (hw_addr && memcmp(mport->hw_addr, hw_addr, ETH_ADDR_LEN)) {
         memcpy(mport->hw_addr, hw_addr, ETH_ADDR_LEN);
     }
+    /* If bfd/cfm is added or reconfigured, move the mport on top of the heap
+     * so that the monitor thread can run the mport next time it wakes up. */
+    if (mport->bfd || mport->cfm) {
+        heap_change(&monitor_heap, &mport->heap_node, LLONG_MAX);
+    }
 }
 
 
-/* Creates the mport in monitor module if either bfd or cfm
- * is configured.  Otherwise, deletes the mport. */
-void
-ofproto_dpif_monitor_port_update(const struct ofport_dpif *ofport,
-                                 struct bfd *bfd, struct cfm *cfm,
-                                 uint8_t hw_addr[ETH_ADDR_LEN])
+/* The 'main' function for the monitor thread. */
+static void *
+monitor_main(void * args OVS_UNUSED)
 {
-    ovs_rwlock_wrlock(&monitor_rwlock);
-    if (!cfm && !bfd) {
-        mport_unregister(ofport);
-    } else {
-        mport_register(ofport, bfd, cfm, hw_addr);
+    VLOG_INFO("monitor thread created");
+    while (!latch_is_set(&monitor_exit_latch)) {
+        monitor_run();
+        latch_wait(&monitor_exit_latch);
+        poll_block();
     }
-    ovs_rwlock_unlock(&monitor_rwlock);
+    VLOG_INFO("monitor thread terminated");
+    return NULL;
 }
 
-/* Checks the sending of control packets on all mports.  Sends the control
- * packets if needed. */
-void
-ofproto_dpif_monitor_run_fast(void)
+/* The monitor thread should wake up this often to ensure that newly added or
+ * reconfigured monitoring ports are run in a timely manner. */
+#define MONITOR_INTERVAL_MSEC 100
+
+/* Checks the 'send_soon' list and the heap for mports that have timed
+ * out bfd/cfm sessions. */
+static void
+monitor_run(void)
 {
-    struct mport *mport;
-    static uint32_t buf_stub[128 / 4];
+    uint32_t stub[512 / 4];
+    long long int prio_now;
     struct ofpbuf packet;
 
-    ovs_rwlock_rdlock(&monitor_rwlock);
-    HMAP_FOR_EACH (mport, hmap_node, &monitor_hmap) {
-        if (mport->cfm && cfm_should_send_ccm(mport->cfm)) {
-            ofpbuf_use_stub(&packet, buf_stub, sizeof buf_stub);
-            cfm_compose_ccm(mport->cfm, &packet, mport->hw_addr);
-            ofproto_dpif_send_packet(mport->ofport, &packet);
-        }
-        if (mport->bfd && bfd_should_send_packet(mport->bfd)) {
-            ofpbuf_use_stub(&packet, buf_stub, sizeof buf_stub);
-            bfd_put_packet(mport->bfd, &packet, mport->hw_addr);
-            ofproto_dpif_send_packet(mport->ofport, &packet);
-        }
+    ofpbuf_use_stub(&packet, stub, sizeof stub);
+    ovs_mutex_lock(&monitor_mutex);
+
+    /* The monitor_check_send_soon() needs to be run twice.  The first
+     * time is for preventing the same 'mport' from being processed twice
+     * (i.e. once from heap, the other from the 'send_soon' array).
+     * The second run is to cover the case when the control packet is sent
+     * via patch port and the other end needs to send back immediately. */
+    monitor_check_send_soon(&packet);
+
+    prio_now = MSEC_TO_PRIO(time_msec());
+    /* Peeks the top of heap and checks if we should run this mport. */
+    while (!heap_is_empty(&monitor_heap)
+           && heap_max(&monitor_heap)->priority >= prio_now) {
+        struct mport *mport;
+
+        mport = CONTAINER_OF(heap_max(&monitor_heap), struct mport, heap_node);
+        monitor_mport_run(mport, &packet);
+    }
+
+    monitor_check_send_soon(&packet);
+
+    /* Waits on the earliest next wakeup time. */
+    if (!heap_is_empty(&monitor_heap)) {
+        long long int next_timeout, next_mport_wakeup;
+
+        next_timeout = time_msec() + MONITOR_INTERVAL_MSEC;
+        next_mport_wakeup = PRIO_TO_MSEC(heap_max(&monitor_heap)->priority);
+        poll_timer_wait_until(MIN(next_timeout, next_mport_wakeup));
     }
-    ovs_rwlock_unlock(&monitor_rwlock);
+    ovs_mutex_unlock(&monitor_mutex);
+    ofpbuf_uninit(&packet);
 }
 
-/* Executes bfd_run(), cfm_run() on all mports. */
-void
-ofproto_dpif_monitor_run(void)
+/* Checks the 'send_soon' list for any mport that needs to send cfm/bfd
+ * control packet immediately, and calls monitor_mport_run(). */
+static void
+monitor_check_send_soon(struct ofpbuf *packet)
+    OVS_REQUIRES(monitor_mutex)
 {
-    struct mport *mport;
+    while (!guarded_list_is_empty(&send_soon)) {
+        struct send_soon_entry *entry;
+        struct mport *mport;
 
-    ovs_rwlock_rdlock(&monitor_rwlock);
-    HMAP_FOR_EACH (mport, hmap_node, &monitor_hmap) {
-        if (mport->cfm) {
-            cfm_run(mport->cfm);
-        }
-        if (mport->bfd) {
-            bfd_run(mport->bfd);
+        entry = CONTAINER_OF(guarded_list_pop_front(&send_soon),
+                             struct send_soon_entry, list_node);
+        mport = mport_find(entry->ofport);
+        if (mport) {
+            monitor_mport_run(mport, packet);
         }
+        free(entry);
     }
-    ovs_rwlock_unlock(&monitor_rwlock);
 }
 
-/* Executes the bfd_wait() and cfm_wait() functions on all mports. */
+/* Checks the sending of control packet on 'mport'.  Sends the control
+ * packet if needed.  Executes bfd and cfm periodic functions (run, wait)
+ * on 'mport'.  And changes the location of 'mport' in heap based on next
+ * timeout. */
+static void
+monitor_mport_run(struct mport *mport, struct ofpbuf *packet)
+    OVS_REQUIRES(monitor_mutex)
+{
+    long long int next_wake_time;
+
+    if (mport->cfm && cfm_should_send_ccm(mport->cfm)) {
+        ofpbuf_clear(packet);
+        cfm_compose_ccm(mport->cfm, packet, mport->hw_addr);
+        ofproto_dpif_send_packet(mport->ofport, packet);
+    }
+    if (mport->bfd && bfd_should_send_packet(mport->bfd)) {
+        ofpbuf_clear(packet);
+        bfd_put_packet(mport->bfd, packet, mport->hw_addr);
+        ofproto_dpif_send_packet(mport->ofport, packet);
+    }
+    if (mport->cfm) {
+        cfm_run(mport->cfm);
+        cfm_wait(mport->cfm);
+    }
+    if (mport->bfd) {
+        bfd_run(mport->bfd);
+        bfd_wait(mport->bfd);
+    }
+    /* Computes the next wakeup time for this mport. */
+    next_wake_time = MIN(bfd_wake_time(mport->bfd),
+                         cfm_wake_time(mport->cfm));
+    heap_change(&monitor_heap, &mport->heap_node,
+                MSEC_TO_PRIO(next_wake_time));
+}
+
+
+/* Creates the mport in monitor module if either bfd or cfm
+ * is configured.  Otherwise, deletes the mport.
+ * Also checks whether the monitor thread should be started
+ * or terminated. */
 void
-ofproto_dpif_monitor_wait(void)
+ofproto_dpif_monitor_port_update(const struct ofport_dpif *ofport,
+                                 struct bfd *bfd, struct cfm *cfm,
+                                 uint8_t hw_addr[ETH_ADDR_LEN])
 {
-    struct mport *mport;
+    ovs_mutex_lock(&monitor_mutex);
+    if (!cfm && !bfd) {
+        mport_unregister(ofport);
+    } else {
+        mport_register(ofport, bfd, cfm, hw_addr);
+    }
+    ovs_mutex_unlock(&monitor_mutex);
 
-    ovs_rwlock_rdlock(&monitor_rwlock);
-    HMAP_FOR_EACH (mport, hmap_node, &monitor_hmap) {
-        if (mport->cfm) {
-            cfm_wait(mport->cfm);
-        }
-        if (mport->bfd) {
-            bfd_wait(mport->bfd);
-        }
+    /* If the monitor thread is not running and the hmap
+     * is not empty, starts it.  If it is and the hmap is empty,
+     * terminates it. */
+    if (!monitor_running && !hmap_is_empty(&monitor_hmap))  {
+        latch_init(&monitor_exit_latch);
+        monitor_tid = ovs_thread_create("monitor", monitor_main, NULL);
+        monitor_running = true;
+    } else if (monitor_running && hmap_is_empty(&monitor_hmap))  {
+        latch_set(&monitor_exit_latch);
+        xpthread_join(monitor_tid, NULL);
+        latch_destroy(&monitor_exit_latch);
+        monitor_running = false;
     }
-    ovs_rwlock_unlock(&monitor_rwlock);
+}
+
+/* Registers the 'ofport' in the 'send_soon' list.  We cannot directly
+ * insert the corresponding mport to the 'send_soon' list, since the
+ * 'send_soon' list is not updated when the mport is removed.
+ *
+ * Reader of the 'send_soon' list is responsible for freeing the entry. */
+void
+ofproto_dpif_monitor_port_send_soon(const struct ofport_dpif *ofport)
+{
+    struct send_soon_entry *entry = xzalloc(sizeof *entry);
+    entry->ofport = ofport;
+
+    guarded_list_push_back(&send_soon, &entry->list_node, SIZE_MAX);
 }