vserver 1.9.5.x5
[linux-2.6.git] / net / core / netpoll.c
index 2f0115e..0860e08 100644 (file)
 #include <linux/interrupt.h>
 #include <linux/netpoll.h>
 #include <linux/sched.h>
+#include <linux/rcupdate.h>
 #include <net/tcp.h>
 #include <net/udp.h>
+#include <asm/unaligned.h>
 
 /*
  * We maintain a small pool of fully-sized skbs, to make sure the
 #define MAX_SKBS 32
 #define MAX_UDP_CHUNK 1460
 
-static spinlock_t skb_list_lock = SPIN_LOCK_UNLOCKED;
+static DEFINE_SPINLOCK(skb_list_lock);
 static int nr_skbs;
 static struct sk_buff *skbs;
 
-static spinlock_t rx_list_lock = SPIN_LOCK_UNLOCKED;
+static DEFINE_SPINLOCK(rx_list_lock);
 static LIST_HEAD(rx_list);
 
-static int trapped;
+static atomic_t trapped;
+static DEFINE_SPINLOCK(netpoll_poll_lock);
+
+#define NETPOLL_RX_ENABLED  1
+#define NETPOLL_RX_DROP     2
 
 #define MAX_SKB_SIZE \
                (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
@@ -59,20 +65,46 @@ static int checksum_udp(struct sk_buff *skb, struct udphdr *uh,
        return csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
 }
 
-void netpoll_poll(struct netpoll *np)
+/*
+ * Check whether delayed processing was scheduled for our current CPU,
+ * and then manually invoke NAPI polling to pump data off the card.
+ *
+ * In cases where there is bi-directional communications, reading only
+ * one message at a time can lead to packets being dropped by the
+ * network adapter, forcing superfluous retries and possibly timeouts.
+ * Thus, we set our budget to greater than 1.
+ */
+static void poll_napi(struct netpoll *np)
 {
-       int budget = 1;
+       int budget = 16;
+       unsigned long flags;
+       struct softnet_data *queue;
+
+       spin_lock_irqsave(&netpoll_poll_lock, flags);
+       queue = &__get_cpu_var(softnet_data);
+       if (test_bit(__LINK_STATE_RX_SCHED, &np->dev->state) &&
+           !list_empty(&queue->poll_list)) {
+               np->dev->netpoll_rx |= NETPOLL_RX_DROP;
+               atomic_inc(&trapped);
 
+               np->dev->poll(np->dev, &budget);
+
+               atomic_dec(&trapped);
+               np->dev->netpoll_rx &= ~NETPOLL_RX_DROP;
+       }
+       spin_unlock_irqrestore(&netpoll_poll_lock, flags);
+}
+
+void netpoll_poll(struct netpoll *np)
+{
        if(!np->dev || !netif_running(np->dev) || !np->dev->poll_controller)
                return;
 
        /* Process pending work on NIC */
        np->dev->poll_controller(np->dev);
+       if (np->dev->poll)
+               poll_napi(np);
 
-       /* If scheduling is stopped, tickle NAPI bits */
-       if(trapped && np->dev->poll &&
-          test_bit(__LINK_STATE_RX_SCHED, &np->dev->state))
-               np->dev->poll(np->dev, &budget);
        zap_completion_queue();
 }
 
@@ -155,7 +187,7 @@ repeat:
        return skb;
 }
 
-void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
+static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 {
        int status;
 
@@ -168,6 +200,18 @@ repeat:
        spin_lock(&np->dev->xmit_lock);
        np->dev->xmit_lock_owner = smp_processor_id();
 
+       /*
+        * network drivers do not expect to be called if the queue is
+        * stopped.
+        */
+       if (netif_queue_stopped(np->dev)) {
+               np->dev->xmit_lock_owner = -1;
+               spin_unlock(&np->dev->xmit_lock);
+
+               netpoll_poll(np);
+               goto repeat;
+       }
+
        status = np->dev->hard_start_xmit(skb, np->dev);
        np->dev->xmit_lock_owner = -1;
        spin_unlock(&np->dev->xmit_lock);
@@ -206,17 +250,17 @@ void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 
        iph = (struct iphdr *)skb_push(skb, sizeof(*iph));
 
-       iph->version  = 4;
-       iph->ihl      = 5;
+       /* iph->version = 4; iph->ihl = 5; */
+       put_unaligned(0x45, (unsigned char *)iph);
        iph->tos      = 0;
-       iph->tot_len  = htons(ip_len);
+       put_unaligned(htons(ip_len), &(iph->tot_len));
        iph->id       = 0;
        iph->frag_off = 0;
        iph->ttl      = 64;
        iph->protocol = IPPROTO_UDP;
        iph->check    = 0;
-       iph->saddr    = htonl(np->local_ip);
-       iph->daddr    = htonl(np->remote_ip);
+       put_unaligned(htonl(np->local_ip), &(iph->saddr));
+       put_unaligned(htonl(np->remote_ip), &(iph->daddr));
        iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 
        eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
@@ -237,14 +281,14 @@ static void arp_reply(struct sk_buff *skb)
        struct sk_buff *send_skb;
        unsigned long flags;
        struct list_head *p;
-       struct netpoll *np = 0;
+       struct netpoll *np = NULL;
 
        spin_lock_irqsave(&rx_list_lock, flags);
        list_for_each(p, &rx_list) {
                np = list_entry(p, struct netpoll, rx_list);
                if ( np->dev == skb->dev )
                        break;
-               np = 0;
+               np = NULL;
        }
        spin_unlock_irqrestore(&rx_list_lock, flags);
 
@@ -337,12 +381,13 @@ int netpoll_rx(struct sk_buff *skb)
                goto out;
 
        /* check if netpoll clients need ARP */
-       if (skb->protocol == __constant_htons(ETH_P_ARP) && trapped) {
+       if (skb->protocol == __constant_htons(ETH_P_ARP) &&
+           atomic_read(&trapped)) {
                arp_reply(skb);
                return 1;
        }
 
-       proto = ntohs(skb->mac.ethernet->h_proto);
+       proto = ntohs(eth_hdr(skb)->h_proto);
        if (proto != ETH_P_IP)
                goto out;
        if (skb->pkt_type == PACKET_OTHERHOST)
@@ -400,7 +445,7 @@ int netpoll_rx(struct sk_buff *skb)
        spin_unlock_irqrestore(&rx_list_lock, flags);
 
 out:
-       return trapped;
+       return atomic_read(&trapped);
 }
 
 int netpoll_parse_options(struct netpoll *np, char *opt)
@@ -411,7 +456,7 @@ int netpoll_parse_options(struct netpoll *np, char *opt)
                if ((delim = strchr(cur, '@')) == NULL)
                        goto parse_failed;
                *delim=0;
-               np->local_port=simple_strtol(cur, 0, 10);
+               np->local_port=simple_strtol(cur, NULL, 10);
                cur=delim;
        }
        cur++;
@@ -446,7 +491,7 @@ int netpoll_parse_options(struct netpoll *np, char *opt)
                if ((delim = strchr(cur, '@')) == NULL)
                        goto parse_failed;
                *delim=0;
-               np->remote_port=simple_strtol(cur, 0, 10);
+               np->remote_port=simple_strtol(cur, NULL, 10);
                cur=delim;
        }
        cur++;
@@ -468,29 +513,29 @@ int netpoll_parse_options(struct netpoll *np, char *opt)
                if ((delim = strchr(cur, ':')) == NULL)
                        goto parse_failed;
                *delim=0;
-               np->remote_mac[0]=simple_strtol(cur, 0, 16);
+               np->remote_mac[0]=simple_strtol(cur, NULL, 16);
                cur=delim+1;
                if ((delim = strchr(cur, ':')) == NULL)
                        goto parse_failed;
                *delim=0;
-               np->remote_mac[1]=simple_strtol(cur, 0, 16);
+               np->remote_mac[1]=simple_strtol(cur, NULL, 16);
                cur=delim+1;
                if ((delim = strchr(cur, ':')) == NULL)
                        goto parse_failed;
                *delim=0;
-               np->remote_mac[2]=simple_strtol(cur, 0, 16);
+               np->remote_mac[2]=simple_strtol(cur, NULL, 16);
                cur=delim+1;
                if ((delim = strchr(cur, ':')) == NULL)
                        goto parse_failed;
                *delim=0;
-               np->remote_mac[3]=simple_strtol(cur, 0, 16);
+               np->remote_mac[3]=simple_strtol(cur, NULL, 16);
                cur=delim+1;
                if ((delim = strchr(cur, ':')) == NULL)
                        goto parse_failed;
                *delim=0;
-               np->remote_mac[4]=simple_strtol(cur, 0, 16);
+               np->remote_mac[4]=simple_strtol(cur, NULL, 16);
                cur=delim+1;
-               np->remote_mac[5]=simple_strtol(cur, 0, 16);
+               np->remote_mac[5]=simple_strtol(cur, NULL, 16);
        }
 
        printk(KERN_INFO "%s: remote ethernet address "
@@ -529,7 +574,7 @@ int netpoll_setup(struct netpoll *np)
                goto release;
        }
 
-       if (!(ndev->flags & IFF_UP)) {
+       if (!netif_running(ndev)) {
                unsigned short oflags;
                unsigned long atmost, atleast;
 
@@ -572,16 +617,18 @@ int netpoll_setup(struct netpoll *np)
                memcpy(np->local_mac, ndev->dev_addr, 6);
 
        if (!np->local_ip) {
-               in_dev = in_dev_get(ndev);
+               rcu_read_lock();
+               in_dev = __in_dev_get(ndev);
 
-               if (!in_dev) {
+               if (!in_dev || !in_dev->ifa_list) {
+                       rcu_read_unlock();
                        printk(KERN_ERR "%s: no IP address for %s, aborting\n",
                               np->name, np->dev_name);
                        goto release;
                }
 
                np->local_ip = ntohl(in_dev->ifa_list->ifa_local);
-               in_dev_put(in_dev);
+               rcu_read_unlock();
                printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
                       np->name, HIPQUAD(np->local_ip));
        }
@@ -591,9 +638,7 @@ int netpoll_setup(struct netpoll *np)
        if(np->rx_hook) {
                unsigned long flags;
 
-#ifdef CONFIG_NETPOLL_RX
-               np->dev->netpoll_rx = 1;
-#endif
+               np->dev->netpoll_rx = NETPOLL_RX_ENABLED;
 
                spin_lock_irqsave(&rx_list_lock, flags);
                list_add(&np->rx_list, &rx_list);
@@ -608,29 +653,31 @@ int netpoll_setup(struct netpoll *np)
 
 void netpoll_cleanup(struct netpoll *np)
 {
-       if(np->rx_hook) {
+       if (np->rx_hook) {
                unsigned long flags;
 
                spin_lock_irqsave(&rx_list_lock, flags);
                list_del(&np->rx_list);
-#ifdef CONFIG_NETPOLL_RX
-               np->dev->netpoll_rx = 0;
-#endif
                spin_unlock_irqrestore(&rx_list_lock, flags);
        }
 
+       if (np->dev)
+               np->dev->netpoll_rx = 0;
        dev_put(np->dev);
-       np->dev = 0;
+       np->dev = NULL;
 }
 
-int netpoll_trap()
+int netpoll_trap(void)
 {
-       return trapped;
+       return atomic_read(&trapped);
 }
 
 void netpoll_set_trap(int trap)
 {
-       trapped = trap;
+       if (trap)
+               atomic_inc(&trapped);
+       else
+               atomic_dec(&trapped);
 }
 
 EXPORT_SYMBOL(netpoll_set_trap);
@@ -638,6 +685,5 @@ EXPORT_SYMBOL(netpoll_trap);
 EXPORT_SYMBOL(netpoll_parse_options);
 EXPORT_SYMBOL(netpoll_setup);
 EXPORT_SYMBOL(netpoll_cleanup);
-EXPORT_SYMBOL(netpoll_send_skb);
 EXPORT_SYMBOL(netpoll_send_udp);
 EXPORT_SYMBOL(netpoll_poll);