X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=datapath%2Fvport-netdev.c;h=dbc8db2ca7817f0e99028e7b55a1d54b2daeaede;hb=ff8d7a5e81625bbb13d33ca73888fc848b02db83;hp=39e963e019a2f77e5a35336971002784b8a08a66;hpb=33ef7695b71a71f2d0bb0e00aab46855afb8c2a6;p=openvswitch diff --git a/datapath/vport-netdev.c b/datapath/vport-netdev.c index 39e963e0..dbc8db2c 100644 --- a/datapath/vport-netdev.c +++ b/datapath/vport-netdev.c @@ -1,11 +1,13 @@ /* - * Copyright (c) 2010 Nicira Networks. + * Copyright (c) 2010, 2011 Nicira Networks. * Distributed under the terms of the GNU GPL version 2. * * Significant portions of this file may be copied from parts of the Linux * kernel, by Linus Torvalds and others. */ +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt + #include #include #include @@ -16,18 +18,45 @@ #include +#include "checksum.h" #include "datapath.h" +#include "vlan.h" #include "vport-internal_dev.h" #include "vport-netdev.h" -#include "compat.h" +#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,37) && \ + !defined(HAVE_VLAN_BUG_WORKAROUND) +#include + +static int vlan_tso __read_mostly = 0; +module_param(vlan_tso, int, 0644); +MODULE_PARM_DESC(vlan_tso, "Enable TSO for VLAN packets"); +#else +#define vlan_tso true +#endif /* If the native device stats aren't 64 bit use the vport stats tracking instead. */ #define USE_VPORT_STATS (sizeof(((struct net_device_stats *)0)->rx_bytes) < sizeof(u64)) static void netdev_port_receive(struct vport *vport, struct sk_buff *skb); -#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) +#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,39) +/* Called with rcu_read_lock and bottom-halves disabled. */ +static rx_handler_result_t netdev_frame_hook(struct sk_buff **pskb) +{ + struct sk_buff *skb = *pskb; + struct vport *vport; + + if (unlikely(skb->pkt_type == PACKET_LOOPBACK)) + return RX_HANDLER_PASS; + + vport = netdev_get_vport(skb->dev); + + netdev_port_receive(vport, skb); + + return RX_HANDLER_CONSUMED; +} +#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) /* Called with rcu_read_lock and bottom-halves disabled. */ static struct sk_buff *netdev_frame_hook(struct sk_buff *skb) { @@ -36,7 +65,7 @@ static struct sk_buff *netdev_frame_hook(struct sk_buff *skb) if (unlikely(skb->pkt_type == PACKET_LOOPBACK)) return skb; - vport = (struct vport *)rcu_dereference(skb->dev->br_port); + vport = netdev_get_vport(skb->dev); netdev_port_receive(vport, skb); @@ -88,13 +117,13 @@ static void netdev_exit(void) } #endif -static struct vport *netdev_create(const char *name, const void __user *config) +static struct vport *netdev_create(const struct vport_parms *parms) { struct vport *vport; struct netdev_vport *netdev_vport; int err; - vport = vport_alloc(sizeof(struct netdev_vport), &netdev_vport_ops); + vport = vport_alloc(sizeof(struct netdev_vport), &netdev_vport_ops, parms); if (IS_ERR(vport)) { err = PTR_ERR(vport); goto error; @@ -102,7 +131,7 @@ static struct vport *netdev_create(const char *name, const void __user *config) netdev_vport = netdev_vport_priv(vport); - netdev_vport->dev = dev_get_by_name(&init_net, name); + netdev_vport->dev = dev_get_by_name(&init_net, parms->name); if (!netdev_vport->dev) { err = -ENODEV; goto error_free_vport; @@ -118,13 +147,24 @@ static struct vport *netdev_create(const char *name, const void __user *config) /* If we are using the vport stats layer initialize it to the current * values so we are roughly consistent with the device stats. */ if (USE_VPORT_STATS) { - struct odp_vport_stats stats; + struct rtnl_link_stats64 stats; err = netdev_get_stats(vport, &stats); if (!err) vport_set_stats(vport, &stats); } + err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook, + vport); + if (err) + goto error_put; + + dev_set_promiscuity(netdev_vport->dev, 1); +#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24) + dev_disable_lro(netdev_vport->dev); +#endif + netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH; + return vport; error_put: @@ -139,38 +179,15 @@ static int netdev_destroy(struct vport *vport) { struct netdev_vport *netdev_vport = netdev_vport_priv(vport); - dev_put(netdev_vport->dev); - vport_free(vport); - - return 0; -} - -static int netdev_attach(struct vport *vport) -{ - struct netdev_vport *netdev_vport = netdev_vport_priv(vport); - int err; - - rcu_assign_pointer(netdev_vport->dev->br_port, - (struct net_bridge_port *)vport); - err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook, - NULL); - if (err) - return err; - - dev_set_promiscuity(netdev_vport->dev, 1); - dev_disable_lro(netdev_vport->dev); - - return 0; -} - -static int netdev_detach(struct vport *vport) -{ - struct netdev_vport *netdev_vport = netdev_vport_priv(vport); - + netdev_vport->dev->priv_flags &= ~IFF_OVS_DATAPATH; netdev_rx_handler_unregister(netdev_vport->dev); - rcu_assign_pointer(netdev_vport->dev->br_port, NULL); dev_set_promiscuity(netdev_vport->dev, -1); + synchronize_rcu(); + + dev_put(netdev_vport->dev); + vport_free(vport); + return 0; } @@ -209,26 +226,10 @@ struct kobject *netdev_get_kobj(const struct vport *vport) return &netdev_vport->dev->NETDEV_DEV_MEMBER.kobj; } -int netdev_get_stats(const struct vport *vport, struct odp_vport_stats *stats) +int netdev_get_stats(const struct vport *vport, struct rtnl_link_stats64 *stats) { const struct netdev_vport *netdev_vport = netdev_vport_priv(vport); - const struct net_device_stats *netdev_stats; - - netdev_stats = dev_get_stats(netdev_vport->dev); - - stats->rx_bytes = netdev_stats->rx_bytes; - stats->rx_packets = netdev_stats->rx_packets; - stats->tx_bytes = netdev_stats->tx_bytes; - stats->tx_packets = netdev_stats->tx_packets; - stats->rx_dropped = netdev_stats->rx_dropped; - stats->rx_errors = netdev_stats->rx_errors; - stats->rx_frame_err = netdev_stats->rx_frame_errors; - stats->rx_over_err = netdev_stats->rx_over_errors; - stats->rx_crc_err = netdev_stats->rx_crc_errors; - stats->tx_dropped = netdev_stats->tx_dropped; - stats->tx_errors = netdev_stats->tx_errors; - stats->collisions = netdev_stats->collisions; - + dev_get_stats(netdev_vport->dev, stats); return 0; } @@ -256,12 +257,6 @@ int netdev_get_ifindex(const struct vport *vport) return netdev_vport->dev->ifindex; } -int netdev_get_iflink(const struct vport *vport) -{ - const struct netdev_vport *netdev_vport = netdev_vport_priv(vport); - return netdev_vport->dev->iflink; -} - int netdev_get_mtu(const struct vport *vport) { const struct netdev_vport *netdev_vport = netdev_vport_priv(vport); @@ -271,6 +266,11 @@ int netdev_get_mtu(const struct vport *vport) /* Must be called with rcu_read_lock. */ static void netdev_port_receive(struct vport *vport, struct sk_buff *skb) { + if (unlikely(!vport)) { + kfree_skb(skb); + return; + } + /* Make our own copy of the packet. Otherwise we will mangle the * packet for anyone who came before us (e.g. tcpdump via AF_PACKET). * (No one comes after us, since we tell handle_bridge() that we took @@ -279,44 +279,155 @@ static void netdev_port_receive(struct vport *vport, struct sk_buff *skb) if (unlikely(!skb)) return; - skb_warn_if_lro(skb); - - /* Push the Ethernet header back on. */ skb_push(skb, ETH_HLEN); - skb_reset_mac_header(skb); - compute_ip_summed(skb, false); + + if (unlikely(compute_ip_summed(skb, false))) { + kfree_skb(skb); + return; + } + vlan_copy_skb_tci(skb); vport_receive(vport, skb); } +static inline unsigned packet_length(const struct sk_buff *skb) +{ + unsigned length = skb->len - ETH_HLEN; + + if (skb->protocol == htons(ETH_P_8021Q)) + length -= VLAN_HLEN; + + return length; +} + +static bool dev_supports_vlan_tx(struct net_device *dev) +{ +#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,37) + /* Software fallback means every device supports vlan_tci on TX. */ + return true; +#elif defined(HAVE_VLAN_BUG_WORKAROUND) + return dev->features & NETIF_F_HW_VLAN_TX; +#else + /* Assume that the driver is buggy. */ + return false; +#endif +} + static int netdev_send(struct vport *vport, struct sk_buff *skb) { struct netdev_vport *netdev_vport = netdev_vport_priv(vport); - int len = skb->len; + int mtu = netdev_vport->dev->mtu; + int len; + + if (unlikely(packet_length(skb) > mtu && !skb_is_gso(skb))) { + if (net_ratelimit()) + pr_warn("%s: dropped over-mtu packet: %d > %d\n", + dp_name(vport->dp), packet_length(skb), mtu); + goto error; + } + + if (unlikely(skb_warn_if_lro(skb))) + goto error; skb->dev = netdev_vport->dev; - forward_ip_summed(skb); + forward_ip_summed(skb, true); + + if (vlan_tx_tag_present(skb) && !dev_supports_vlan_tx(skb->dev)) { + int features = 0; + +#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,26) + features = skb->dev->features & skb->dev->vlan_features; +#endif + + if (!vlan_tso) + features &= ~(NETIF_F_TSO | NETIF_F_TSO6 | + NETIF_F_UFO | NETIF_F_FSO); + + if (skb_is_gso(skb) && + (!skb_gso_ok(skb, features) || + unlikely(skb->ip_summed != CHECKSUM_PARTIAL))) { + struct sk_buff *nskb; + + nskb = skb_gso_segment(skb, features); + if (!nskb) { + if (unlikely(skb_cloned(skb) && + pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) { + kfree_skb(skb); + return 0; + } + + skb_shinfo(skb)->gso_type &= ~SKB_GSO_DODGY; + goto tag; + } + + if (IS_ERR(nskb)) { + kfree_skb(skb); + return 0; + } + consume_skb(skb); + skb = nskb; + + len = 0; + do { + nskb = skb->next; + skb->next = NULL; + + skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb)); + if (likely(skb)) { + len += skb->len; + vlan_set_tci(skb, 0); + dev_queue_xmit(skb); + } + + skb = nskb; + } while (skb); + + return len; + } + +tag: + skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb)); + if (unlikely(!skb)) + return 0; + vlan_set_tci(skb, 0); + } + + len = skb->len; dev_queue_xmit(skb); return len; + +error: + kfree_skb(skb); + vport_record_error(vport, VPORT_E_TX_DROPPED); + return 0; } /* Returns null if this device is not attached to a datapath. */ struct vport *netdev_get_vport(struct net_device *dev) { - return (struct vport *)dev->br_port; +#ifdef IFF_BRIDGE_PORT +#if IFF_BRIDGE_PORT != IFF_OVS_DATAPATH + if (likely(dev->priv_flags & IFF_OVS_DATAPATH)) +#else + if (likely(rcu_access_pointer(dev->rx_handler) == netdev_frame_hook)) +#endif + return (struct vport *)rcu_dereference_rtnl(dev->rx_handler_data); + else + return NULL; +#else + return (struct vport *)rcu_dereference_rtnl(dev->br_port); +#endif } -struct vport_ops netdev_vport_ops = { - .type = "netdev", +const struct vport_ops netdev_vport_ops = { + .type = OVS_VPORT_TYPE_NETDEV, .flags = (VPORT_F_REQUIRED | (USE_VPORT_STATS ? VPORT_F_GEN_STATS : 0)), .init = netdev_init, .exit = netdev_exit, .create = netdev_create, .destroy = netdev_destroy, - .attach = netdev_attach, - .detach = netdev_detach, .set_mtu = netdev_set_mtu, .set_addr = netdev_set_addr, .get_name = netdev_get_name, @@ -327,15 +438,16 @@ struct vport_ops netdev_vport_ops = { .is_running = netdev_is_running, .get_operstate = netdev_get_operstate, .get_ifindex = netdev_get_ifindex, - .get_iflink = netdev_get_iflink, .get_mtu = netdev_get_mtu, .send = netdev_send, }; +#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,36) /* - * Open vSwitch cannot safely coexist with the Linux bridge module on any - * released version of Linux, because there is only a single bridge hook - * function and only a single br_port member in struct net_device. + * In kernels earlier than 2.6.36, Open vSwitch cannot safely coexist with + * the Linux bridge module on any released version of Linux, because there + * is only a single bridge hook function and only a single br_port member + * in struct net_device. * * Declaring and exporting this symbol enforces mutual exclusion. The bridge * module also exports the same symbol, so the module loader will refuse to @@ -347,3 +459,4 @@ struct vport_ops netdev_vport_ops = { */ typeof(br_should_route_hook) br_should_route_hook; EXPORT_SYMBOL(br_should_route_hook); +#endif