net: allow GFP_HIGHMEM in __vmalloc()
[linux-2.6.git] / net / core / drop_monitor.c
index 2797b71..36e603c 100644 (file)
 #include <linux/percpu.h>
 #include <linux/timer.h>
 #include <linux/bitops.h>
+#include <linux/slab.h>
 #include <net/genetlink.h>
+#include <net/netevent.h>
 
-#include <trace/skb.h>
+#include <trace/events/skb.h>
+#include <trace/events/napi.h>
 
 #include <asm/unaligned.h>
 
@@ -38,7 +41,8 @@ static void send_dm_alert(struct work_struct *unused);
  * and the work handle that will send up
  * netlink alerts
  */
-struct sock *dm_sock;
+static int trace_state = TRACE_OFF;
+static DEFINE_SPINLOCK(trace_state_lock);
 
 struct per_cpu_dm_data {
        struct work_struct dm_alert_work;
@@ -47,6 +51,14 @@ struct per_cpu_dm_data {
        struct timer_list send_timer;
 };
 
+struct dm_hw_stat_delta {
+       struct net_device *dev;
+       unsigned long last_rx;
+       struct list_head list;
+       struct rcu_head rcu;
+       unsigned long last_drop_val;
+};
+
 static struct genl_family net_drop_monitor_family = {
        .id             = GENL_ID_GENERATE,
        .hdrsize        = 0,
@@ -59,7 +71,8 @@ static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data);
 
 static int dm_hit_limit = 64;
 static int dm_delay = 1;
-
+static unsigned long dm_hw_check_delta = 2*HZ;
+static LIST_HEAD(hw_stats_list);
 
 static void reset_per_cpu_data(struct per_cpu_dm_data *data)
 {
@@ -115,7 +128,7 @@ static void sched_send_work(unsigned long unused)
        schedule_work(&data->dm_alert_work);
 }
 
-static void trace_kfree_skb_hit(struct sk_buff *skb, void *location)
+static void trace_drop_common(struct sk_buff *skb, void *location)
 {
        struct net_dm_alert_msg *msg;
        struct nlmsghdr *nlh;
@@ -159,26 +172,96 @@ out:
        return;
 }
 
+static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb, void *location)
+{
+       trace_drop_common(skb, location);
+}
+
+static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi)
+{
+       struct dm_hw_stat_delta *new_stat;
+
+       /*
+        * Don't check napi structures with no associated device
+        */
+       if (!napi->dev)
+               return;
+
+       rcu_read_lock();
+       list_for_each_entry_rcu(new_stat, &hw_stats_list, list) {
+               /*
+                * only add a note to our monitor buffer if:
+                * 1) this is the dev we received on
+                * 2) its after the last_rx delta
+                * 3) our rx_dropped count has gone up
+                */
+               if ((new_stat->dev == napi->dev)  &&
+                   (time_after(jiffies, new_stat->last_rx + dm_hw_check_delta)) &&
+                   (napi->dev->stats.rx_dropped != new_stat->last_drop_val)) {
+                       trace_drop_common(NULL, NULL);
+                       new_stat->last_drop_val = napi->dev->stats.rx_dropped;
+                       new_stat->last_rx = jiffies;
+                       break;
+               }
+       }
+       rcu_read_unlock();
+}
+
+
+static void free_dm_hw_stat(struct rcu_head *head)
+{
+       struct dm_hw_stat_delta *n;
+       n = container_of(head, struct dm_hw_stat_delta, rcu);
+       kfree(n);
+}
+
 static int set_all_monitor_traces(int state)
 {
        int rc = 0;
+       struct dm_hw_stat_delta *new_stat = NULL;
+       struct dm_hw_stat_delta *temp;
+
+       spin_lock(&trace_state_lock);
+
+       if (state == trace_state) {
+               rc = -EAGAIN;
+               goto out_unlock;
+       }
 
        switch (state) {
        case TRACE_ON:
-               rc |= register_trace_kfree_skb(trace_kfree_skb_hit);
+               rc |= register_trace_kfree_skb(trace_kfree_skb_hit, NULL);
+               rc |= register_trace_napi_poll(trace_napi_poll_hit, NULL);
                break;
        case TRACE_OFF:
-               rc |= unregister_trace_kfree_skb(trace_kfree_skb_hit);
+               rc |= unregister_trace_kfree_skb(trace_kfree_skb_hit, NULL);
+               rc |= unregister_trace_napi_poll(trace_napi_poll_hit, NULL);
 
                tracepoint_synchronize_unregister();
+
+               /*
+                * Clean the device list
+                */
+               list_for_each_entry_safe(new_stat, temp, &hw_stats_list, list) {
+                       if (new_stat->dev == NULL) {
+                               list_del_rcu(&new_stat->list);
+                               call_rcu(&new_stat->rcu, free_dm_hw_stat);
+                       }
+               }
                break;
        default:
                rc = 1;
                break;
        }
 
-       if (rc)
-               return -EINPROGRESS;
+       if (!rc)
+               trace_state = state;
+       else
+               rc = -EINPROGRESS;
+
+out_unlock:
+       spin_unlock(&trace_state_lock);
+
        return rc;
 }
 
@@ -204,6 +287,44 @@ static int net_dm_cmd_trace(struct sk_buff *skb,
        return -ENOTSUPP;
 }
 
+static int dropmon_net_event(struct notifier_block *ev_block,
+                       unsigned long event, void *ptr)
+{
+       struct net_device *dev = ptr;
+       struct dm_hw_stat_delta *new_stat = NULL;
+       struct dm_hw_stat_delta *tmp;
+
+       switch (event) {
+       case NETDEV_REGISTER:
+               new_stat = kzalloc(sizeof(struct dm_hw_stat_delta), GFP_KERNEL);
+
+               if (!new_stat)
+                       goto out;
+
+               new_stat->dev = dev;
+               new_stat->last_rx = jiffies;
+               spin_lock(&trace_state_lock);
+               list_add_rcu(&new_stat->list, &hw_stats_list);
+               spin_unlock(&trace_state_lock);
+               break;
+       case NETDEV_UNREGISTER:
+               spin_lock(&trace_state_lock);
+               list_for_each_entry_safe(new_stat, tmp, &hw_stats_list, list) {
+                       if (new_stat->dev == dev) {
+                               new_stat->dev = NULL;
+                               if (trace_state == TRACE_OFF) {
+                                       list_del_rcu(&new_stat->list);
+                                       call_rcu(&new_stat->rcu, free_dm_hw_stat);
+                                       break;
+                               }
+                       }
+               }
+               spin_unlock(&trace_state_lock);
+               break;
+       }
+out:
+       return NOTIFY_DONE;
+}
 
 static struct genl_ops dropmon_ops[] = {
        {
@@ -220,11 +341,15 @@ static struct genl_ops dropmon_ops[] = {
        },
 };
 
+static struct notifier_block dropmon_net_notifier = {
+       .notifier_call = dropmon_net_event
+};
+
 static int __init init_net_drop_monitor(void)
 {
-       int cpu;
-       int rc, i, ret;
        struct per_cpu_dm_data *data;
+       int cpu, rc;
+
        printk(KERN_INFO "Initalizing network drop monitor service\n");
 
        if (sizeof(void *) > 8) {
@@ -232,21 +357,18 @@ static int __init init_net_drop_monitor(void)
                return -ENOSPC;
        }
 
-       if (genl_register_family(&net_drop_monitor_family) < 0) {
+       rc = genl_register_family_with_ops(&net_drop_monitor_family,
+                                          dropmon_ops,
+                                          ARRAY_SIZE(dropmon_ops));
+       if (rc) {
                printk(KERN_ERR "Could not create drop monitor netlink family\n");
-               return -EFAULT;
+               return rc;
        }
 
-       rc = -EFAULT;
-
-       for (i = 0; i < ARRAY_SIZE(dropmon_ops); i++) {
-               ret = genl_register_ops(&net_drop_monitor_family,
-                                       &dropmon_ops[i]);
-               if (ret) {
-                       printk(KERN_CRIT "failed to register operation %d\n",
-                               dropmon_ops[i].cmd);
-                       goto out_unreg;
-               }
+       rc = register_netdevice_notifier(&dropmon_net_notifier);
+       if (rc < 0) {
+               printk(KERN_CRIT "Failed to register netdevice notifier\n");
+               goto out_unreg;
        }
 
        rc = 0;
@@ -259,6 +381,7 @@ static int __init init_net_drop_monitor(void)
                data->send_timer.data = cpu;
                data->send_timer.function = sched_send_work;
        }
+
        goto out;
 
 out_unreg: