net-2.6 : V2 - fix dev_get_valid_name
[safe/jmp/linux-2.6] / net / core / netpoll.c
index 755414c..94825b1 100644 (file)
@@ -9,6 +9,7 @@
  * Copyright (C) 2002  Red Hat, Inc.
  */
 
+#include <linux/moduleparam.h>
 #include <linux/netdevice.h>
 #include <linux/etherdevice.h>
 #include <linux/string.h>
 #include <linux/delay.h>
 #include <linux/rcupdate.h>
 #include <linux/workqueue.h>
+#include <linux/slab.h>
 #include <net/tcp.h>
 #include <net/udp.h>
 #include <asm/unaligned.h>
+#include <trace/events/napi.h>
 
 /*
  * We maintain a small pool of fully-sized skbs, to make sure the
@@ -49,6 +52,9 @@ static atomic_t trapped;
 static void zap_completion_queue(void);
 static void arp_reply(struct sk_buff *skb);
 
+static unsigned int carrier_timeout = 4;
+module_param(carrier_timeout, uint, 0644);
+
 static void queue_process(struct work_struct *work)
 {
        struct netpoll_info *npinfo =
@@ -137,6 +143,7 @@ static int poll_one_napi(struct netpoll_info *npinfo,
        set_bit(NAPI_STATE_NPSVC, &napi->state);
 
        work = napi->poll(napi, budget);
+       trace_napi_poll(napi);
 
        clear_bit(NAPI_STATE_NPSVC, &napi->state);
        atomic_dec(&trapped);
@@ -172,12 +179,15 @@ static void service_arp_queue(struct netpoll_info *npi)
        }
 }
 
-void netpoll_poll(struct netpoll *np)
+void netpoll_poll_dev(struct net_device *dev)
 {
-       struct net_device *dev = np->dev;
-       const struct net_device_ops *ops = dev->netdev_ops;
+       const struct net_device_ops *ops;
 
-       if (!dev || !netif_running(dev) || !ops->ndo_poll_controller)
+       if (!dev || !netif_running(dev))
+               return;
+
+       ops = dev->netdev_ops;
+       if (!ops->ndo_poll_controller)
                return;
 
        /* Process pending work on NIC */
@@ -190,6 +200,11 @@ void netpoll_poll(struct netpoll *np)
        zap_completion_queue();
 }
 
+void netpoll_poll(struct netpoll *np)
+{
+       netpoll_poll_dev(np->dev);
+}
+
 static void refill_skbs(void)
 {
        struct sk_buff *skb;
@@ -271,7 +286,7 @@ static int netpoll_owner_active(struct net_device *dev)
        return 0;
 }
 
-static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
+void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 {
        int status = NETDEV_TX_BUSY;
        unsigned long tries;
@@ -296,8 +311,13 @@ static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
                for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
                     tries > 0; --tries) {
                        if (__netif_tx_trylock(txq)) {
-                               if (!netif_tx_queue_stopped(txq))
+                               if (!netif_tx_queue_stopped(txq)) {
+                                       dev->priv_flags |= IFF_IN_NETPOLL;
                                        status = ops->ndo_start_xmit(skb, dev);
+                                       dev->priv_flags &= ~IFF_IN_NETPOLL;
+                                       if (status == NETDEV_TX_OK)
+                                               txq_trans_update(txq);
+                               }
                                __netif_tx_unlock(txq);
 
                                if (status == NETDEV_TX_OK)
@@ -310,6 +330,11 @@ static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 
                        udelay(USEC_PER_POLL);
                }
+
+               WARN_ONCE(!irqs_disabled(),
+                       "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n",
+                       dev->name, ops->ndo_start_xmit);
+
                local_irq_restore(flags);
        }
 
@@ -345,8 +370,8 @@ void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
        udph->dest = htons(np->remote_port);
        udph->len = htons(udp_len);
        udph->check = 0;
-       udph->check = csum_tcpudp_magic(htonl(np->local_ip),
-                                       htonl(np->remote_ip),
+       udph->check = csum_tcpudp_magic(np->local_ip,
+                                       np->remote_ip,
                                        udp_len, IPPROTO_UDP,
                                        csum_partial(udph, udp_len, 0));
        if (udph->check == 0)
@@ -365,8 +390,8 @@ void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
        iph->ttl      = 64;
        iph->protocol = IPPROTO_UDP;
        iph->check    = 0;
-       put_unaligned(htonl(np->local_ip), &(iph->saddr));
-       put_unaligned(htonl(np->remote_ip), &(iph->daddr));
+       put_unaligned(np->local_ip, &(iph->saddr));
+       put_unaligned(np->remote_ip, &(iph->daddr));
        iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 
        eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
@@ -389,11 +414,24 @@ static void arp_reply(struct sk_buff *skb)
        __be32 sip, tip;
        unsigned char *sha;
        struct sk_buff *send_skb;
-       struct netpoll *np = NULL;
+       struct netpoll *np, *tmp;
+       unsigned long flags;
+       int hits = 0;
 
-       if (npinfo->rx_np && npinfo->rx_np->dev == skb->dev)
-               np = npinfo->rx_np;
-       if (!np)
+       if (list_empty(&npinfo->rx_np))
+               return;
+
+       /* Before checking the packet, we do some early
+          inspection whether this is interesting at all */
+       spin_lock_irqsave(&npinfo->rx_lock, flags);
+       list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
+               if (np->dev == skb->dev)
+                       hits++;
+       }
+       spin_unlock_irqrestore(&npinfo->rx_lock, flags);
+
+       /* No netpoll struct is using this dev */
+       if (!hits)
                return;
 
        /* No arp on this interface */
@@ -419,77 +457,91 @@ static void arp_reply(struct sk_buff *skb)
        arp_ptr += skb->dev->addr_len;
        memcpy(&sip, arp_ptr, 4);
        arp_ptr += 4;
-       /* if we actually cared about dst hw addr, it would get copied here */
+       /* If we actually cared about dst hw addr,
+          it would get copied here */
        arp_ptr += skb->dev->addr_len;
        memcpy(&tip, arp_ptr, 4);
 
        /* Should we ignore arp? */
-       if (tip != htonl(np->local_ip) ||
-           ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
+       if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
                return;
 
        size = arp_hdr_len(skb->dev);
-       send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev),
-                           LL_RESERVED_SPACE(np->dev));
 
-       if (!send_skb)
-               return;
-
-       skb_reset_network_header(send_skb);
-       arp = (struct arphdr *) skb_put(send_skb, size);
-       send_skb->dev = skb->dev;
-       send_skb->protocol = htons(ETH_P_ARP);
+       spin_lock_irqsave(&npinfo->rx_lock, flags);
+       list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
+               if (tip != np->local_ip)
+                       continue;
 
-       /* Fill the device header for the ARP frame */
-       if (dev_hard_header(send_skb, skb->dev, ptype,
-                           sha, np->dev->dev_addr,
-                           send_skb->len) < 0) {
-               kfree_skb(send_skb);
-               return;
-       }
+               send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev),
+                                   LL_RESERVED_SPACE(np->dev));
+               if (!send_skb)
+                       continue;
 
-       /*
-        * Fill out the arp protocol part.
-        *
-        * we only support ethernet device type,
-        * which (according to RFC 1390) should always equal 1 (Ethernet).
-        */
+               skb_reset_network_header(send_skb);
+               arp = (struct arphdr *) skb_put(send_skb, size);
+               send_skb->dev = skb->dev;
+               send_skb->protocol = htons(ETH_P_ARP);
 
-       arp->ar_hrd = htons(np->dev->type);
-       arp->ar_pro = htons(ETH_P_IP);
-       arp->ar_hln = np->dev->addr_len;
-       arp->ar_pln = 4;
-       arp->ar_op = htons(type);
+               /* Fill the device header for the ARP frame */
+               if (dev_hard_header(send_skb, skb->dev, ptype,
+                                   sha, np->dev->dev_addr,
+                                   send_skb->len) < 0) {
+                       kfree_skb(send_skb);
+                       continue;
+               }
 
-       arp_ptr=(unsigned char *)(arp + 1);
-       memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
-       arp_ptr += np->dev->addr_len;
-       memcpy(arp_ptr, &tip, 4);
-       arp_ptr += 4;
-       memcpy(arp_ptr, sha, np->dev->addr_len);
-       arp_ptr += np->dev->addr_len;
-       memcpy(arp_ptr, &sip, 4);
+               /*
+                * Fill out the arp protocol part.
+                *
+                * we only support ethernet device type,
+                * which (according to RFC 1390) should
+                * always equal 1 (Ethernet).
+                */
 
-       netpoll_send_skb(np, send_skb);
+               arp->ar_hrd = htons(np->dev->type);
+               arp->ar_pro = htons(ETH_P_IP);
+               arp->ar_hln = np->dev->addr_len;
+               arp->ar_pln = 4;
+               arp->ar_op = htons(type);
+
+               arp_ptr = (unsigned char *)(arp + 1);
+               memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
+               arp_ptr += np->dev->addr_len;
+               memcpy(arp_ptr, &tip, 4);
+               arp_ptr += 4;
+               memcpy(arp_ptr, sha, np->dev->addr_len);
+               arp_ptr += np->dev->addr_len;
+               memcpy(arp_ptr, &sip, 4);
+
+               netpoll_send_skb(np, send_skb);
+
+               /* If there are several rx_hooks for the same address,
+                  we're fine by sending a single reply */
+               break;
+       }
+       spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 }
 
 int __netpoll_rx(struct sk_buff *skb)
 {
        int proto, len, ulen;
+       int hits = 0;
        struct iphdr *iph;
        struct udphdr *uh;
-       struct netpoll_info *npi = skb->dev->npinfo;
-       struct netpoll *np = npi->rx_np;
+       struct netpoll_info *npinfo = skb->dev->npinfo;
+       struct netpoll *np, *tmp;
 
-       if (!np)
+       if (list_empty(&npinfo->rx_np))
                goto out;
+
        if (skb->dev->type != ARPHRD_ETHER)
                goto out;
 
        /* check if netpoll clients need ARP */
        if (skb->protocol == htons(ETH_P_ARP) &&
            atomic_read(&trapped)) {
-               skb_queue_tail(&npi->arp_tx, skb);
+               skb_queue_tail(&npinfo->arp_tx, skb);
                return 1;
        }
 
@@ -533,16 +585,23 @@ int __netpoll_rx(struct sk_buff *skb)
                goto out;
        if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
                goto out;
-       if (np->local_ip && np->local_ip != ntohl(iph->daddr))
-               goto out;
-       if (np->remote_ip && np->remote_ip != ntohl(iph->saddr))
-               goto out;
-       if (np->local_port && np->local_port != ntohs(uh->dest))
-               goto out;
 
-       np->rx_hook(np, ntohs(uh->source),
-                   (char *)(uh+1),
-                   ulen - sizeof(struct udphdr));
+       list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
+               if (np->local_ip && np->local_ip != iph->daddr)
+                       continue;
+               if (np->remote_ip && np->remote_ip != iph->saddr)
+                       continue;
+               if (np->local_port && np->local_port != ntohs(uh->dest))
+                       continue;
+
+               np->rx_hook(np, ntohs(uh->source),
+                              (char *)(uh+1),
+                              ulen - sizeof(struct udphdr));
+               hits++;
+       }
+
+       if (!hits)
+               goto out;
 
        kfree_skb(skb);
        return 1;
@@ -560,14 +619,14 @@ void netpoll_print_options(struct netpoll *np)
 {
        printk(KERN_INFO "%s: local port %d\n",
                         np->name, np->local_port);
-       printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
-                        np->name, HIPQUAD(np->local_ip));
-       printk(KERN_INFO "%s: interface %s\n",
+       printk(KERN_INFO "%s: local IP %pI4\n",
+                        np->name, &np->local_ip);
+       printk(KERN_INFO "%s: interface '%s'\n",
                         np->name, np->dev_name);
        printk(KERN_INFO "%s: remote port %d\n",
                         np->name, np->remote_port);
-       printk(KERN_INFO "%s: remote IP %d.%d.%d.%d\n",
-                        np->name, HIPQUAD(np->remote_ip));
+       printk(KERN_INFO "%s: remote IP %pI4\n",
+                        np->name, &np->remote_ip);
        printk(KERN_INFO "%s: remote ethernet address %pM\n",
                         np->name, np->remote_mac);
 }
@@ -589,7 +648,7 @@ int netpoll_parse_options(struct netpoll *np, char *opt)
                if ((delim = strchr(cur, '/')) == NULL)
                        goto parse_failed;
                *delim = 0;
-               np->local_ip = ntohl(in_aton(cur));
+               np->local_ip = in_aton(cur);
                cur = delim;
        }
        cur++;
@@ -609,6 +668,9 @@ int netpoll_parse_options(struct netpoll *np, char *opt)
                if ((delim = strchr(cur, '@')) == NULL)
                        goto parse_failed;
                *delim = 0;
+               if (*cur == ' ' || *cur == '\t')
+                       printk(KERN_INFO "%s: warning: whitespace"
+                                       "is not allowed\n", np->name);
                np->remote_port = simple_strtol(cur, NULL, 10);
                cur = delim;
        }
@@ -618,7 +680,7 @@ int netpoll_parse_options(struct netpoll *np, char *opt)
        if ((delim = strchr(cur, '/')) == NULL)
                goto parse_failed;
        *delim = 0;
-       np->remote_ip = ntohl(in_aton(cur));
+       np->remote_ip = in_aton(cur);
        cur = delim + 1;
 
        if (*cur != 0) {
@@ -656,7 +718,7 @@ int netpoll_parse_options(struct netpoll *np, char *opt)
        return 0;
 
  parse_failed:
-       printk(KERN_INFO "%s: couldn't parse config at %s!\n",
+       printk(KERN_INFO "%s: couldn't parse config at '%s'!\n",
               np->name, cur);
        return -1;
 }
@@ -666,6 +728,7 @@ int netpoll_setup(struct netpoll *np)
        struct net_device *ndev = NULL;
        struct in_device *in_dev;
        struct netpoll_info *npinfo;
+       struct netpoll *npe, *tmp;
        unsigned long flags;
        int err;
 
@@ -682,11 +745,11 @@ int netpoll_setup(struct netpoll *np)
                npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
                if (!npinfo) {
                        err = -ENOMEM;
-                       goto release;
+                       goto put;
                }
 
                npinfo->rx_flags = 0;
-               npinfo->rx_np = NULL;
+               INIT_LIST_HEAD(&npinfo->rx_np);
 
                spin_lock_init(&npinfo->rx_lock);
                skb_queue_head_init(&npinfo->arp_tx);
@@ -699,7 +762,10 @@ int netpoll_setup(struct netpoll *np)
                atomic_inc(&npinfo->refcnt);
        }
 
-       if (!ndev->netdev_ops->ndo_poll_controller) {
+       npinfo->netpoll = np;
+
+       if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
+           !ndev->netdev_ops->ndo_poll_controller) {
                printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
                       np->name, np->dev_name);
                err = -ENOTSUPP;
@@ -723,7 +789,7 @@ int netpoll_setup(struct netpoll *np)
                }
 
                atleast = jiffies + HZ/10;
-               atmost = jiffies + 4*HZ;
+               atmost = jiffies + carrier_timeout * HZ;
                while (!netif_carrier_ok(ndev)) {
                        if (time_after(jiffies, atmost)) {
                                printk(KERN_NOTICE
@@ -731,7 +797,7 @@ int netpoll_setup(struct netpoll *np)
                                       np->name);
                                break;
                        }
-                       cond_resched();
+                       msleep(1);
                }
 
                /* If carrier appears to come up instantly, we don't
@@ -759,16 +825,15 @@ int netpoll_setup(struct netpoll *np)
                        goto release;
                }
 
-               np->local_ip = ntohl(in_dev->ifa_list->ifa_local);
+               np->local_ip = in_dev->ifa_list->ifa_local;
                rcu_read_unlock();
-               printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
-                      np->name, HIPQUAD(np->local_ip));
+               printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip);
        }
 
        if (np->rx_hook) {
                spin_lock_irqsave(&npinfo->rx_lock, flags);
                npinfo->rx_flags |= NETPOLL_RX_ENABLED;
-               npinfo->rx_np = np;
+               list_add_tail(&np->rx, &npinfo->rx_np);
                spin_unlock_irqrestore(&npinfo->rx_lock, flags);
        }
 
@@ -784,9 +849,16 @@ int netpoll_setup(struct netpoll *np)
        return 0;
 
  release:
-       if (!ndev->npinfo)
+       if (!ndev->npinfo) {
+               spin_lock_irqsave(&npinfo->rx_lock, flags);
+               list_for_each_entry_safe(npe, tmp, &npinfo->rx_np, rx) {
+                       npe->dev = NULL;
+               }
+               spin_unlock_irqrestore(&npinfo->rx_lock, flags);
+
                kfree(npinfo);
-       np->dev = NULL;
+       }
+put:
        dev_put(ndev);
        return err;
 }
@@ -806,14 +878,16 @@ void netpoll_cleanup(struct netpoll *np)
        if (np->dev) {
                npinfo = np->dev->npinfo;
                if (npinfo) {
-                       if (npinfo->rx_np == np) {
+                       if (!list_empty(&npinfo->rx_np)) {
                                spin_lock_irqsave(&npinfo->rx_lock, flags);
-                               npinfo->rx_np = NULL;
-                               npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
+                               list_del(&np->rx);
+                               if (list_empty(&npinfo->rx_np))
+                                       npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
                                spin_unlock_irqrestore(&npinfo->rx_lock, flags);
                        }
 
                        if (atomic_dec_and_test(&npinfo->refcnt)) {
+                               const struct net_device_ops *ops;
                                skb_queue_purge(&npinfo->arp_tx);
                                skb_queue_purge(&npinfo->txq);
                                cancel_rearming_delayed_work(&npinfo->tx_work);
@@ -821,7 +895,11 @@ void netpoll_cleanup(struct netpoll *np)
                                /* clean after last, unfinished work */
                                __skb_queue_purge(&npinfo->txq);
                                kfree(npinfo);
-                               np->dev->npinfo = NULL;
+                               ops = np->dev->netdev_ops;
+                               if (ops->ndo_netpoll_cleanup)
+                                       ops->ndo_netpoll_cleanup(np->dev);
+                               else
+                                       np->dev->npinfo = NULL;
                        }
                }
 
@@ -844,6 +922,7 @@ void netpoll_set_trap(int trap)
                atomic_dec(&trapped);
 }
 
+EXPORT_SYMBOL(netpoll_send_skb);
 EXPORT_SYMBOL(netpoll_set_trap);
 EXPORT_SYMBOL(netpoll_trap);
 EXPORT_SYMBOL(netpoll_print_options);
@@ -851,4 +930,5 @@ EXPORT_SYMBOL(netpoll_parse_options);
 EXPORT_SYMBOL(netpoll_setup);
 EXPORT_SYMBOL(netpoll_cleanup);
 EXPORT_SYMBOL(netpoll_send_udp);
+EXPORT_SYMBOL(netpoll_poll_dev);
 EXPORT_SYMBOL(netpoll_poll);