ip_gre: percpu stats accounting
authorEric Dumazet <eric.dumazet@gmail.com>
Mon, 27 Sep 2010 03:57:11 +0000 (03:57 +0000)
committerDavid S. Miller <davem@davemloft.net>
Tue, 28 Sep 2010 04:30:43 +0000 (21:30 -0700)
Le lundi 27 septembre 2010 à 14:29 +0100, Ben Hutchings a écrit :

> > diff --git a/net/ipv4/ip_gre.c b/net/ipv4/ip_gre.c
> > index 5d6ddcb..de39b22 100644
> > --- a/net/ipv4/ip_gre.c
> > +++ b/net/ipv4/ip_gre.c
> [...]
> > @@ -377,7 +405,7 @@ static struct ip_tunnel *ipgre_tunnel_locate(struct net *net,
> >   if (parms->name[0])
> >   strlcpy(name, parms->name, IFNAMSIZ);
> >   else
> > - sprintf(name, "gre%%d");
> > + strcpy(name, "gre%d");
> >
> >   dev = alloc_netdev(sizeof(*t), name, ipgre_tunnel_setup);
> >   if (!dev)
> [...]
>
> This is a valid fix, but doesn't belong in this patch!
>

Sorry ? It was not a fix, but at most a cleanup ;)

Anyway I forgot the gretap case...

[PATCH 2/4 v2] ip_gre: percpu stats accounting

Maintain per_cpu tx_bytes, tx_packets, rx_bytes, rx_packets.

Other seldom used fields are kept in netdev->stats structure, possibly
unsafe.

This is a preliminary work to support lockless transmit path, and
correct RX stats, that are already unsafe.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/ip_gre.c

index 5d6ddcb7403b487ea1fa347595685f7f73061803..a1b5d5e030645b945a9b45e3793098701bf9a642 100644 (file)
@@ -165,6 +165,34 @@ struct ipgre_net {
 #define for_each_ip_tunnel_rcu(start) \
        for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
 
+/* often modified stats are per cpu, other are shared (netdev->stats) */
+struct pcpu_tstats {
+       unsigned long   rx_packets;
+       unsigned long   rx_bytes;
+       unsigned long   tx_packets;
+       unsigned long   tx_bytes;
+};
+
+static struct net_device_stats *ipgre_get_stats(struct net_device *dev)
+{
+       struct pcpu_tstats sum = { 0 };
+       int i;
+
+       for_each_possible_cpu(i) {
+               const struct pcpu_tstats *tstats = per_cpu_ptr(dev->tstats, i);
+
+               sum.rx_packets += tstats->rx_packets;
+               sum.rx_bytes   += tstats->rx_bytes;
+               sum.tx_packets += tstats->tx_packets;
+               sum.tx_bytes   += tstats->tx_bytes;
+       }
+       dev->stats.rx_packets = sum.rx_packets;
+       dev->stats.rx_bytes   = sum.rx_bytes;
+       dev->stats.tx_packets = sum.tx_packets;
+       dev->stats.tx_bytes   = sum.tx_bytes;
+       return &dev->stats;
+}
+
 /* Given src, dst and key, find appropriate for input tunnel. */
 
 static struct ip_tunnel * ipgre_tunnel_lookup(struct net_device *dev,
@@ -584,7 +612,7 @@ static int ipgre_rcv(struct sk_buff *skb)
        if ((tunnel = ipgre_tunnel_lookup(skb->dev,
                                          iph->saddr, iph->daddr, key,
                                          gre_proto))) {
-               struct net_device_stats *stats = &tunnel->dev->stats;
+               struct pcpu_tstats *tstats;
 
                secpath_reset(skb);
 
@@ -608,22 +636,22 @@ static int ipgre_rcv(struct sk_buff *skb)
                        /* Looped back packet, drop it! */
                        if (skb_rtable(skb)->fl.iif == 0)
                                goto drop;
-                       stats->multicast++;
+                       tunnel->dev->stats.multicast++;
                        skb->pkt_type = PACKET_BROADCAST;
                }
 #endif
 
                if (((flags&GRE_CSUM) && csum) ||
                    (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
-                       stats->rx_crc_errors++;
-                       stats->rx_errors++;
+                       tunnel->dev->stats.rx_crc_errors++;
+                       tunnel->dev->stats.rx_errors++;
                        goto drop;
                }
                if (tunnel->parms.i_flags&GRE_SEQ) {
                        if (!(flags&GRE_SEQ) ||
                            (tunnel->i_seqno && (s32)(seqno - tunnel->i_seqno) < 0)) {
-                               stats->rx_fifo_errors++;
-                               stats->rx_errors++;
+                               tunnel->dev->stats.rx_fifo_errors++;
+                               tunnel->dev->stats.rx_errors++;
                                goto drop;
                        }
                        tunnel->i_seqno = seqno + 1;
@@ -632,8 +660,8 @@ static int ipgre_rcv(struct sk_buff *skb)
                /* Warning: All skb pointers will be invalidated! */
                if (tunnel->dev->type == ARPHRD_ETHER) {
                        if (!pskb_may_pull(skb, ETH_HLEN)) {
-                               stats->rx_length_errors++;
-                               stats->rx_errors++;
+                               tunnel->dev->stats.rx_length_errors++;
+                               tunnel->dev->stats.rx_errors++;
                                goto drop;
                        }
 
@@ -642,13 +670,17 @@ static int ipgre_rcv(struct sk_buff *skb)
                        skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
                }
 
-               skb_tunnel_rx(skb, tunnel->dev);
+               tstats = this_cpu_ptr(tunnel->dev->tstats);
+               tstats->rx_packets++;
+               tstats->rx_bytes += skb->len;
+
+               __skb_tunnel_rx(skb, tunnel->dev);
 
                skb_reset_network_header(skb);
                ipgre_ecn_decapsulate(iph, skb);
 
                if (netif_rx(skb) == NET_RX_DROP)
-                       stats->rx_dropped++;
+                       tunnel->dev->stats.rx_dropped++;
 
                rcu_read_unlock();
                return 0;
@@ -665,8 +697,7 @@ drop_nolock:
 static netdev_tx_t ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
 {
        struct ip_tunnel *tunnel = netdev_priv(dev);
-       struct net_device_stats *stats = &dev->stats;
-       struct netdev_queue *txq = netdev_get_tx_queue(dev, 0);
+       struct pcpu_tstats *tstats;
        struct iphdr  *old_iph = ip_hdr(skb);
        struct iphdr  *tiph;
        u8     tos;
@@ -694,7 +725,7 @@ static netdev_tx_t ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev
                /* NBMA tunnel */
 
                if (skb_dst(skb) == NULL) {
-                       stats->tx_fifo_errors++;
+                       dev->stats.tx_fifo_errors++;
                        goto tx_error;
                }
 
@@ -740,14 +771,20 @@ static netdev_tx_t ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev
        }
 
        {
-               struct flowi fl = { .oif = tunnel->parms.link,
-                                   .nl_u = { .ip4_u =
-                                             { .daddr = dst,
-                                               .saddr = tiph->saddr,
-                                               .tos = RT_TOS(tos) } },
-                                   .proto = IPPROTO_GRE };
+               struct flowi fl = {
+                       .oif = tunnel->parms.link,
+                       .nl_u = {
+                               .ip4_u = {
+                                       .daddr = dst,
+                                       .saddr = tiph->saddr,
+                                       .tos = RT_TOS(tos)
+                               }
+                       },
+                       .proto = IPPROTO_GRE
+               }
+;
                if (ip_route_output_key(dev_net(dev), &rt, &fl)) {
-                       stats->tx_carrier_errors++;
+                       dev->stats.tx_carrier_errors++;
                        goto tx_error;
                }
        }
@@ -755,7 +792,7 @@ static netdev_tx_t ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev
 
        if (tdev == dev) {
                ip_rt_put(rt);
-               stats->collisions++;
+               dev->stats.collisions++;
                goto tx_error;
        }
 
@@ -818,7 +855,7 @@ static netdev_tx_t ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev
                        dev->needed_headroom = max_headroom;
                if (!new_skb) {
                        ip_rt_put(rt);
-                       txq->tx_dropped++;
+                       dev->stats.tx_dropped++;
                        dev_kfree_skb(skb);
                        return NETDEV_TX_OK;
                }
@@ -885,15 +922,15 @@ static netdev_tx_t ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev
        }
 
        nf_reset(skb);
-
-       IPTUNNEL_XMIT();
+       tstats = this_cpu_ptr(dev->tstats);
+       __IPTUNNEL_XMIT(tstats, &dev->stats);
        return NETDEV_TX_OK;
 
 tx_error_icmp:
        dst_link_failure(skb);
 
 tx_error:
-       stats->tx_errors++;
+       dev->stats.tx_errors++;
        dev_kfree_skb(skb);
        return NETDEV_TX_OK;
 }
@@ -913,13 +950,19 @@ static int ipgre_tunnel_bind_dev(struct net_device *dev)
        /* Guess output device to choose reasonable mtu and needed_headroom */
 
        if (iph->daddr) {
-               struct flowi fl = { .oif = tunnel->parms.link,
-                                   .nl_u = { .ip4_u =
-                                             { .daddr = iph->daddr,
-                                               .saddr = iph->saddr,
-                                               .tos = RT_TOS(iph->tos) } },
-                                   .proto = IPPROTO_GRE };
+               struct flowi fl = {
+                       .oif = tunnel->parms.link,
+                       .nl_u = {
+                               .ip4_u = {
+                                       .daddr = iph->daddr,
+                                       .saddr = iph->saddr,
+                                       .tos = RT_TOS(iph->tos)
+                               }
+                       },
+                       .proto = IPPROTO_GRE
+               };
                struct rtable *rt;
+
                if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
                        tdev = rt->dst.dev;
                        ip_rt_put(rt);
@@ -1171,13 +1214,19 @@ static int ipgre_open(struct net_device *dev)
        struct ip_tunnel *t = netdev_priv(dev);
 
        if (ipv4_is_multicast(t->parms.iph.daddr)) {
-               struct flowi fl = { .oif = t->parms.link,
-                                   .nl_u = { .ip4_u =
-                                             { .daddr = t->parms.iph.daddr,
-                                               .saddr = t->parms.iph.saddr,
-                                               .tos = RT_TOS(t->parms.iph.tos) } },
-                                   .proto = IPPROTO_GRE };
+               struct flowi fl = {
+                       .oif = t->parms.link,
+                       .nl_u = {
+                               .ip4_u = {
+                                       .daddr = t->parms.iph.daddr,
+                                       .saddr = t->parms.iph.saddr,
+                                       .tos = RT_TOS(t->parms.iph.tos)
+                               }
+                       },
+                       .proto = IPPROTO_GRE
+               };
                struct rtable *rt;
+
                if (ip_route_output_key(dev_net(dev), &rt, &fl))
                        return -EADDRNOTAVAIL;
                dev = rt->dst.dev;
@@ -1217,12 +1266,19 @@ static const struct net_device_ops ipgre_netdev_ops = {
        .ndo_start_xmit         = ipgre_tunnel_xmit,
        .ndo_do_ioctl           = ipgre_tunnel_ioctl,
        .ndo_change_mtu         = ipgre_tunnel_change_mtu,
+       .ndo_get_stats          = ipgre_get_stats,
 };
 
+static void ipgre_dev_free(struct net_device *dev)
+{
+       free_percpu(dev->tstats);
+       free_netdev(dev);
+}
+
 static void ipgre_tunnel_setup(struct net_device *dev)
 {
        dev->netdev_ops         = &ipgre_netdev_ops;
-       dev->destructor         = free_netdev;
+       dev->destructor         = ipgre_dev_free;
 
        dev->type               = ARPHRD_IPGRE;
        dev->needed_headroom    = LL_MAX_HEADER + sizeof(struct iphdr) + 4;
@@ -1260,6 +1316,10 @@ static int ipgre_tunnel_init(struct net_device *dev)
        } else
                dev->header_ops = &ipgre_header_ops;
 
+       dev->tstats = alloc_percpu(struct pcpu_tstats);
+       if (!dev->tstats)
+               return -ENOMEM;
+
        return 0;
 }
 
@@ -1446,6 +1506,10 @@ static int ipgre_tap_init(struct net_device *dev)
 
        ipgre_tunnel_bind_dev(dev);
 
+       dev->tstats = alloc_percpu(struct pcpu_tstats);
+       if (!dev->tstats)
+               return -ENOMEM;
+
        return 0;
 }
 
@@ -1456,6 +1520,7 @@ static const struct net_device_ops ipgre_tap_netdev_ops = {
        .ndo_set_mac_address    = eth_mac_addr,
        .ndo_validate_addr      = eth_validate_addr,
        .ndo_change_mtu         = ipgre_tunnel_change_mtu,
+       .ndo_get_stats          = ipgre_get_stats,
 };
 
 static void ipgre_tap_setup(struct net_device *dev)
@@ -1464,7 +1529,7 @@ static void ipgre_tap_setup(struct net_device *dev)
        ether_setup(dev);
 
        dev->netdev_ops         = &ipgre_tap_netdev_ops;
-       dev->destructor         = free_netdev;
+       dev->destructor         = ipgre_dev_free;
 
        dev->iflink             = 0;
        dev->features           |= NETIF_F_NETNS_LOCAL;