X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev-vport.c;h=c6d4db89eeccb64c3d7692918683b41c7d761829;hb=41ca1e0afb4b261a217c9fdaf672ef606e8434f9;hp=13b1d930c371f7ff84d38b75e741caaf21e1cba3;hpb=4c2fa71d662cde318940c4cd555aacd687538510;p=sliver-openvswitch.git diff --git a/lib/netdev-vport.c b/lib/netdev-vport.c index 13b1d930c..c214bf713 100644 --- a/lib/netdev-vport.c +++ b/lib/netdev-vport.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2010 Nicira Networks. + * Copyright (c) 2010, 2011, 2012, 2013 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -20,312 +20,307 @@ #include #include +#include #include #include #include "byte-order.h" +#include "daemon.h" +#include "dirs.h" +#include "dpif.h" +#include "hash.h" +#include "hmap.h" #include "list.h" #include "netdev-provider.h" -#include "openvswitch/datapath-protocol.h" -#include "openvswitch/tunnel.h" +#include "ofpbuf.h" #include "packets.h" +#include "poll-loop.h" +#include "route-table.h" #include "shash.h" #include "socket-util.h" #include "vlog.h" VLOG_DEFINE_THIS_MODULE(netdev_vport); -struct netdev_vport_notifier { - struct netdev_notifier notifier; - struct list list_node; - struct shash_node *shash_node; -}; +#define VXLAN_DST_PORT 4789 +#define LISP_DST_PORT 4341 -struct netdev_dev_vport { - struct netdev_dev netdev_dev; - uint64_t config[VPORT_CONFIG_SIZE / 8]; -}; +#define DEFAULT_TTL 64 struct netdev_vport { - struct netdev netdev; + struct netdev up; + + /* Protects all members below. */ + struct ovs_mutex mutex; + + uint8_t etheraddr[ETH_ADDR_LEN]; + struct netdev_stats stats; + + /* Tunnels. */ + struct netdev_tunnel_config tnl_cfg; + char egress_iface[IFNAMSIZ]; + bool carrier_status; + + /* Patch Ports. */ + char *peer; }; struct vport_class { + const char *dpif_port; struct netdev_class netdev_class; - int (*parse_config)(const struct netdev_dev *, const struct shash *args, - void *config); }; -static struct shash netdev_vport_notifiers = - SHASH_INITIALIZER(&netdev_vport_notifiers); +/* Last read of the route-table's change number. */ +static uint64_t rt_change_seqno; -static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); - -static int netdev_vport_do_ioctl(int cmd, void *arg); -static int netdev_vport_create(const struct netdev_class *, const char *, - const struct shash *, struct netdev_dev **); -static void netdev_vport_poll_notify(const struct netdev *); +static int netdev_vport_construct(struct netdev *); +static int get_patch_config(const struct netdev *netdev, struct smap *args); +static int get_tunnel_config(const struct netdev *, struct smap *args); +static bool tunnel_check_status_change__(struct netdev_vport *); static bool is_vport_class(const struct netdev_class *class) { - return class->create == netdev_vport_create; + return class->construct == netdev_vport_construct; +} + +bool +netdev_vport_is_vport_class(const struct netdev_class *class) +{ + return is_vport_class(class); } static const struct vport_class * vport_class_cast(const struct netdev_class *class) { - assert(is_vport_class(class)); + ovs_assert(is_vport_class(class)); return CONTAINER_OF(class, struct vport_class, netdev_class); } -static struct netdev_dev_vport * -netdev_dev_vport_cast(const struct netdev_dev *netdev_dev) +static struct netdev_vport * +netdev_vport_cast(const struct netdev *netdev) +{ + ovs_assert(is_vport_class(netdev_get_class(netdev))); + return CONTAINER_OF(netdev, struct netdev_vport, up); +} + +static const struct netdev_tunnel_config * +get_netdev_tunnel_config(const struct netdev *netdev) { - assert(is_vport_class(netdev_dev_get_class(netdev_dev))); - return CONTAINER_OF(netdev_dev, struct netdev_dev_vport, netdev_dev); + return &netdev_vport_cast(netdev)->tnl_cfg; } -static struct netdev_vport * -netdev_vport_cast(const struct netdev *netdev) +bool +netdev_vport_is_patch(const struct netdev *netdev) { - struct netdev_dev *netdev_dev = netdev_get_dev(netdev); - assert(is_vport_class(netdev_dev_get_class(netdev_dev))); - return CONTAINER_OF(netdev, struct netdev_vport, netdev); + const struct netdev_class *class = netdev_get_class(netdev); + + return class->get_config == get_patch_config; } -/* If 'netdev' is a vport netdev, copies its kernel configuration into - * 'config'. Otherwise leaves 'config' untouched. */ -void -netdev_vport_get_config(const struct netdev *netdev, void *config) +bool +netdev_vport_is_layer3(const struct netdev *dev) { - const struct netdev_dev *dev = netdev_get_dev(netdev); + const char *type = netdev_get_type(dev); - if (is_vport_class(netdev_dev_get_class(dev))) { - const struct netdev_dev_vport *vport = netdev_dev_vport_cast(dev); - memcpy(config, vport->config, VPORT_CONFIG_SIZE); - } + return (!strcmp("lisp", type)); } -static int -netdev_vport_create(const struct netdev_class *netdev_class, const char *name, - const struct shash *args, - struct netdev_dev **netdev_devp) +static bool +netdev_vport_needs_dst_port(const struct netdev *dev) { - const struct vport_class *vport_class = vport_class_cast(netdev_class); - struct netdev_dev_vport *dev; - int error; + const struct netdev_class *class = netdev_get_class(dev); + const char *type = netdev_get_type(dev); - dev = xmalloc(sizeof *dev); - *netdev_devp = &dev->netdev_dev; - netdev_dev_init(&dev->netdev_dev, name, netdev_class); + return (class->get_config == get_tunnel_config && + (!strcmp("vxlan", type) || !strcmp("lisp", type))); +} - memset(dev->config, 0, sizeof dev->config); - error = vport_class->parse_config(&dev->netdev_dev, args, dev->config); +const char * +netdev_vport_class_get_dpif_port(const struct netdev_class *class) +{ + return is_vport_class(class) ? vport_class_cast(class)->dpif_port : NULL; +} - if (error) { - netdev_dev_uninit(&dev->netdev_dev, true); +const char * +netdev_vport_get_dpif_port(const struct netdev *netdev, + char namebuf[], size_t bufsize) +{ + if (netdev_vport_needs_dst_port(netdev)) { + const struct netdev_vport *vport = netdev_vport_cast(netdev); + const char *type = netdev_get_type(netdev); + + /* + * Note: IFNAMSIZ is 16 bytes long. The maximum length of a VXLAN + * or LISP port name below is 15 or 14 bytes respectively. Still, + * assert here on the size of strlen(type) in case that changes + * in the future. + */ + BUILD_ASSERT(NETDEV_VPORT_NAME_BUFSIZE >= IFNAMSIZ); + ovs_assert(strlen(type) + 10 < IFNAMSIZ); + snprintf(namebuf, bufsize, "%s_sys_%d", type, + ntohs(vport->tnl_cfg.dst_port)); + return namebuf; + } else { + const struct netdev_class *class = netdev_get_class(netdev); + const char *dpif_port = netdev_vport_class_get_dpif_port(class); + return dpif_port ? dpif_port : netdev_get_name(netdev); } - return error; } +char * +netdev_vport_get_dpif_port_strdup(const struct netdev *netdev) +{ + char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; + + return xstrdup(netdev_vport_get_dpif_port(netdev, namebuf, + sizeof namebuf)); +} + +/* Whenever the route-table change number is incremented, + * netdev_vport_route_changed() should be called to update + * the corresponding tunnel interface status. */ static void -netdev_vport_destroy(struct netdev_dev *netdev_dev_) +netdev_vport_route_changed(void) { - struct netdev_dev_vport *netdev_dev = netdev_dev_vport_cast(netdev_dev_); + struct netdev **vports; + size_t i, n_vports; + + vports = netdev_get_vports(&n_vports); + for (i = 0; i < n_vports; i++) { + struct netdev *netdev_ = vports[i]; + struct netdev_vport *netdev = netdev_vport_cast(netdev_); + + ovs_mutex_lock(&netdev->mutex); + /* Finds all tunnel vports. */ + if (netdev->tnl_cfg.ip_dst) { + if (tunnel_check_status_change__(netdev)) { + netdev_change_seq_changed(netdev_); + } + } + netdev_close(netdev_); + ovs_mutex_unlock(&netdev->mutex); + } + + free(vports); +} - free(netdev_dev); +static struct netdev * +netdev_vport_alloc(void) +{ + struct netdev_vport *netdev = xzalloc(sizeof *netdev); + return &netdev->up; } static int -netdev_vport_open(struct netdev_dev *netdev_dev_, int ethertype OVS_UNUSED, - struct netdev **netdevp) +netdev_vport_construct(struct netdev *netdev_) { - struct netdev_vport *netdev; + struct netdev_vport *netdev = netdev_vport_cast(netdev_); + + ovs_mutex_init(&netdev->mutex); + eth_addr_random(netdev->etheraddr); - netdev = xmalloc(sizeof *netdev); - netdev_init(&netdev->netdev, netdev_dev_); + route_table_register(); - *netdevp = &netdev->netdev; return 0; } static void -netdev_vport_close(struct netdev *netdev_) +netdev_vport_destruct(struct netdev *netdev_) { struct netdev_vport *netdev = netdev_vport_cast(netdev_); - free(netdev); + + route_table_unregister(); + free(netdev->peer); + ovs_mutex_destroy(&netdev->mutex); } -static int -netdev_vport_reconfigure(struct netdev_dev *dev_, - const struct shash *args) -{ - const struct netdev_class *netdev_class = netdev_dev_get_class(dev_); - const struct vport_class *vport_class = vport_class_cast(netdev_class); - struct netdev_dev_vport *dev = netdev_dev_vport_cast(dev_); - struct odp_port port; - int error; - - memset(&port, 0, sizeof port); - strncpy(port.devname, netdev_dev_get_name(dev_), sizeof port.devname); - strncpy(port.type, netdev_dev_get_type(dev_), sizeof port.type); - error = vport_class->parse_config(dev_, args, port.config); - if (!error && memcmp(port.config, dev->config, sizeof dev->config)) { - error = netdev_vport_do_ioctl(ODP_VPORT_MOD, &port); - if (!error || error == ENODEV) { - /* Either reconfiguration succeeded or this vport is not installed - * in the kernel (e.g. it hasn't been added to a dpif yet with - * dpif_port_add()). */ - memcpy(dev->config, port.config, sizeof dev->config); - } - } - return error; +static void +netdev_vport_dealloc(struct netdev *netdev_) +{ + struct netdev_vport *netdev = netdev_vport_cast(netdev_); + free(netdev); } static int -netdev_vport_set_etheraddr(struct netdev *netdev, +netdev_vport_set_etheraddr(struct netdev *netdev_, const uint8_t mac[ETH_ADDR_LEN]) { - struct odp_vport_ether vport_ether; - int err; - - ovs_strlcpy(vport_ether.devname, netdev_get_name(netdev), - sizeof vport_ether.devname); - - memcpy(vport_ether.ether_addr, mac, ETH_ADDR_LEN); + struct netdev_vport *netdev = netdev_vport_cast(netdev_); - err = netdev_vport_do_ioctl(ODP_VPORT_ETHER_SET, &vport_ether); - if (err) { - return err; - } + ovs_mutex_lock(&netdev->mutex); + memcpy(netdev->etheraddr, mac, ETH_ADDR_LEN); + ovs_mutex_unlock(&netdev->mutex); + netdev_change_seq_changed(netdev_); - netdev_vport_poll_notify(netdev); return 0; } static int -netdev_vport_get_etheraddr(const struct netdev *netdev, +netdev_vport_get_etheraddr(const struct netdev *netdev_, uint8_t mac[ETH_ADDR_LEN]) { - struct odp_vport_ether vport_ether; - int err; - - ovs_strlcpy(vport_ether.devname, netdev_get_name(netdev), - sizeof vport_ether.devname); + struct netdev_vport *netdev = netdev_vport_cast(netdev_); - err = netdev_vport_do_ioctl(ODP_VPORT_ETHER_GET, &vport_ether); - if (err) { - return err; - } + ovs_mutex_lock(&netdev->mutex); + memcpy(mac, netdev->etheraddr, ETH_ADDR_LEN); + ovs_mutex_unlock(&netdev->mutex); - memcpy(mac, vport_ether.ether_addr, ETH_ADDR_LEN); return 0; } -static int -netdev_vport_get_mtu(const struct netdev *netdev, int *mtup) +/* Checks if the tunnel status has changed and returns a boolean. + * Updates the tunnel status if it has changed. */ +static bool +tunnel_check_status_change__(struct netdev_vport *netdev) + OVS_REQUIRES(netdev->mutex) { - struct odp_vport_mtu vport_mtu; - int err; - - ovs_strlcpy(vport_mtu.devname, netdev_get_name(netdev), - sizeof vport_mtu.devname); - - err = netdev_vport_do_ioctl(ODP_VPORT_MTU_GET, &vport_mtu); - if (err) { - return err; + char iface[IFNAMSIZ]; + bool status = false; + ovs_be32 route; + + iface[0] = '\0'; + route = netdev->tnl_cfg.ip_dst; + if (route_table_get_name(route, iface)) { + struct netdev *egress_netdev; + + if (!netdev_open(iface, "system", &egress_netdev)) { + status = netdev_get_carrier(egress_netdev); + netdev_close(egress_netdev); + } } - *mtup = vport_mtu.mtu; - return 0; -} + if (strcmp(netdev->egress_iface, iface) + || netdev->carrier_status != status) { + ovs_strlcpy(netdev->egress_iface, iface, IFNAMSIZ); + netdev->carrier_status = status; -int -netdev_vport_get_stats(const struct netdev *netdev, struct netdev_stats *stats) -{ - const char *name = netdev_get_name(netdev); - struct odp_vport_stats_req ovsr; - int err; - - ovs_strlcpy(ovsr.devname, name, sizeof ovsr.devname); - err = netdev_vport_do_ioctl(ODP_VPORT_STATS_GET, &ovsr); - if (err) { - return err; - } - - stats->rx_packets = ovsr.stats.rx_packets; - stats->tx_packets = ovsr.stats.tx_packets; - stats->rx_bytes = ovsr.stats.rx_bytes; - stats->tx_bytes = ovsr.stats.tx_bytes; - stats->rx_errors = ovsr.stats.rx_errors; - stats->tx_errors = ovsr.stats.tx_errors; - stats->rx_dropped = ovsr.stats.rx_dropped; - stats->tx_dropped = ovsr.stats.tx_dropped; - stats->multicast = ovsr.stats.multicast; - stats->collisions = ovsr.stats.collisions; - stats->rx_length_errors = ovsr.stats.rx_length_errors; - stats->rx_over_errors = ovsr.stats.rx_over_errors; - stats->rx_crc_errors = ovsr.stats.rx_crc_errors; - stats->rx_frame_errors = ovsr.stats.rx_frame_errors; - stats->rx_fifo_errors = ovsr.stats.rx_fifo_errors; - stats->rx_missed_errors = ovsr.stats.rx_missed_errors; - stats->tx_aborted_errors = ovsr.stats.tx_aborted_errors; - stats->tx_carrier_errors = ovsr.stats.tx_carrier_errors; - stats->tx_fifo_errors = ovsr.stats.tx_fifo_errors; - stats->tx_heartbeat_errors = ovsr.stats.tx_heartbeat_errors; - stats->tx_window_errors = ovsr.stats.tx_window_errors; + return true; + } - return 0; + return false; } -int -netdev_vport_set_stats(struct netdev *netdev, const struct netdev_stats *stats) +static int +tunnel_get_status(const struct netdev *netdev_, struct smap *smap) { - struct odp_vport_stats_req ovsr; - int err; - - ovs_strlcpy(ovsr.devname, netdev_get_name(netdev), sizeof ovsr.devname); - - ovsr.stats.rx_packets = stats->rx_packets; - ovsr.stats.tx_packets = stats->tx_packets; - ovsr.stats.rx_bytes = stats->rx_bytes; - ovsr.stats.tx_bytes = stats->tx_bytes; - ovsr.stats.rx_errors = stats->rx_errors; - ovsr.stats.tx_errors = stats->tx_errors; - ovsr.stats.rx_dropped = stats->rx_dropped; - ovsr.stats.tx_dropped = stats->tx_dropped; - ovsr.stats.multicast = stats->multicast; - ovsr.stats.collisions = stats->collisions; - ovsr.stats.rx_length_errors = stats->rx_length_errors; - ovsr.stats.rx_over_errors = stats->rx_over_errors; - ovsr.stats.rx_crc_errors = stats->rx_crc_errors; - ovsr.stats.rx_frame_errors = stats->rx_frame_errors; - ovsr.stats.rx_fifo_errors = stats->rx_fifo_errors; - ovsr.stats.rx_missed_errors = stats->rx_missed_errors; - ovsr.stats.tx_aborted_errors = stats->tx_aborted_errors; - ovsr.stats.tx_carrier_errors = stats->tx_carrier_errors; - ovsr.stats.tx_fifo_errors = stats->tx_fifo_errors; - ovsr.stats.tx_heartbeat_errors = stats->tx_heartbeat_errors; - ovsr.stats.tx_window_errors = stats->tx_window_errors; + struct netdev_vport *netdev = netdev_vport_cast(netdev_); - err = netdev_vport_do_ioctl(ODP_VPORT_STATS_SET, &ovsr); + if (netdev->egress_iface[0]) { + smap_add(smap, "tunnel_egress_iface", netdev->egress_iface); - /* If the vport layer doesn't know about the device, that doesn't mean it - * doesn't exist (after all were able to open it when netdev_open() was - * called), it just means that it isn't attached and we'll be getting - * stats a different way. */ - if (err == ENODEV) { - err = EOPNOTSUPP; + smap_add(smap, "tunnel_egress_iface_carrier", + netdev->carrier_status ? "up" : "down"); } - return err; + return 0; } static int netdev_vport_update_flags(struct netdev *netdev OVS_UNUSED, - enum netdev_flags off, enum netdev_flags on OVS_UNUSED, - enum netdev_flags *old_flagsp) + enum netdev_flags off, + enum netdev_flags on OVS_UNUSED, + enum netdev_flags *old_flagsp) { if (off & (NETDEV_UP | NETDEV_PROMISC)) { return EOPNOTSUPP; @@ -335,264 +330,458 @@ netdev_vport_update_flags(struct netdev *netdev OVS_UNUSED, return 0; } -static char * -make_poll_name(const struct netdev *netdev) +static void +netdev_vport_run(void) { - return xasprintf("%s:%s", netdev_get_type(netdev), netdev_get_name(netdev)); -} + uint64_t seq; -static int -netdev_vport_poll_add(struct netdev *netdev, - void (*cb)(struct netdev_notifier *), void *aux, - struct netdev_notifier **notifierp) -{ - char *poll_name = make_poll_name(netdev); - struct netdev_vport_notifier *notifier; - struct list *list; - struct shash_node *shash_node; - - shash_node = shash_find_data(&netdev_vport_notifiers, poll_name); - if (!shash_node) { - list = xmalloc(sizeof *list); - list_init(list); - shash_node = shash_add(&netdev_vport_notifiers, poll_name, list); - } else { - list = shash_node->data; + route_table_run(); + seq = route_table_get_change_seq(); + if (rt_change_seqno != seq) { + rt_change_seqno = seq; + netdev_vport_route_changed(); } - - notifier = xmalloc(sizeof *notifier); - netdev_notifier_init(¬ifier->notifier, netdev, cb, aux); - list_push_back(list, ¬ifier->list_node); - notifier->shash_node = shash_node; - - *notifierp = ¬ifier->notifier; - free(poll_name); - - return 0; } static void -netdev_vport_poll_remove(struct netdev_notifier *notifier_) +netdev_vport_wait(void) { - struct netdev_vport_notifier *notifier = - CONTAINER_OF(notifier_, struct netdev_vport_notifier, notifier); - - struct list *list; + uint64_t seq; - list = list_remove(¬ifier->list_node); - if (list_is_empty(list)) { - shash_delete(&netdev_vport_notifiers, notifier->shash_node); - free(list); + route_table_wait(); + seq = route_table_get_change_seq(); + if (rt_change_seqno != seq) { + poll_immediate_wake(); } - - free(notifier); } -/* Helper functions. */ +/* Code specific to tunnel types. */ -static int -netdev_vport_do_ioctl(int cmd, void *arg) +static ovs_be64 +parse_key(const struct smap *args, const char *name, + bool *present, bool *flow) { - static int ioctl_fd = -1; + const char *s; + + *present = false; + *flow = false; - if (ioctl_fd < 0) { - ioctl_fd = open("/dev/net/dp0", O_RDONLY | O_NONBLOCK); - if (ioctl_fd < 0) { - VLOG_ERR_RL(&rl, "failed to open ioctl fd: %s", strerror(errno)); - return errno; + s = smap_get(args, name); + if (!s) { + s = smap_get(args, "key"); + if (!s) { + return 0; } } - return ioctl(ioctl_fd, cmd, arg) ? errno : 0; -} + *present = true; -static void -netdev_vport_poll_notify(const struct netdev *netdev) -{ - char *poll_name = make_poll_name(netdev); - struct list *list = shash_find_data(&netdev_vport_notifiers, - poll_name); - - if (list) { - struct netdev_vport_notifier *notifier; - - LIST_FOR_EACH (notifier, list_node, list) { - struct netdev_notifier *n = ¬ifier->notifier; - n->cb(n); - } + if (!strcmp(s, "flow")) { + *flow = true; + return 0; + } else { + return htonll(strtoull(s, NULL, 0)); } - - free(poll_name); } - -/* Code specific to individual vport types. */ static int -parse_tunnel_config(const struct netdev_dev *dev, const struct shash *args, - void *configp) -{ - const char *name = netdev_dev_get_name(dev); - const char *type = netdev_dev_get_type(dev); - bool is_gre = !strcmp(type, "gre"); - struct tnl_port_config config; - struct shash_node *node; - bool ipsec_mech_set = false; - - memset(&config, 0, sizeof config); - config.flags |= TNL_F_PMTUD; - config.flags |= TNL_F_HDR_CACHE; - - SHASH_FOR_EACH (node, args) { - if (!strcmp(node->name, "remote_ip")) { +set_tunnel_config(struct netdev *dev_, const struct smap *args) +{ + struct netdev_vport *dev = netdev_vport_cast(dev_); + const char *name = netdev_get_name(dev_); + const char *type = netdev_get_type(dev_); + bool ipsec_mech_set, needs_dst_port, has_csum; + struct netdev_tunnel_config tnl_cfg; + struct smap_node *node; + + has_csum = strstr(type, "gre"); + ipsec_mech_set = false; + memset(&tnl_cfg, 0, sizeof tnl_cfg); + + needs_dst_port = netdev_vport_needs_dst_port(dev_); + tnl_cfg.ipsec = strstr(type, "ipsec"); + tnl_cfg.dont_fragment = true; + + SMAP_FOR_EACH (node, args) { + if (!strcmp(node->key, "remote_ip")) { struct in_addr in_addr; - if (lookup_ip(node->data, &in_addr)) { + if (!strcmp(node->value, "flow")) { + tnl_cfg.ip_dst_flow = true; + tnl_cfg.ip_dst = htonl(0); + } else if (lookup_ip(node->value, &in_addr)) { VLOG_WARN("%s: bad %s 'remote_ip'", name, type); + } else if (ip_is_multicast(in_addr.s_addr)) { + VLOG_WARN("%s: multicast remote_ip="IP_FMT" not allowed", + name, IP_ARGS(in_addr.s_addr)); + return EINVAL; } else { - config.daddr = in_addr.s_addr; + tnl_cfg.ip_dst = in_addr.s_addr; } - } else if (!strcmp(node->name, "local_ip")) { + } else if (!strcmp(node->key, "local_ip")) { struct in_addr in_addr; - if (lookup_ip(node->data, &in_addr)) { + if (!strcmp(node->value, "flow")) { + tnl_cfg.ip_src_flow = true; + tnl_cfg.ip_src = htonl(0); + } else if (lookup_ip(node->value, &in_addr)) { VLOG_WARN("%s: bad %s 'local_ip'", name, type); } else { - config.saddr = in_addr.s_addr; + tnl_cfg.ip_src = in_addr.s_addr; } - } else if (!strcmp(node->name, "key") && is_gre) { - if (!strcmp(node->data, "flow")) { - config.flags |= TNL_F_IN_KEY_MATCH; - config.flags |= TNL_F_OUT_KEY_ACTION; + } else if (!strcmp(node->key, "tos")) { + if (!strcmp(node->value, "inherit")) { + tnl_cfg.tos_inherit = true; } else { - uint64_t key = strtoull(node->data, NULL, 0); - config.out_key = config.in_key = htonll(key); + char *endptr; + int tos; + tos = strtol(node->value, &endptr, 0); + if (*endptr == '\0' && tos == (tos & IP_DSCP_MASK)) { + tnl_cfg.tos = tos; + } else { + VLOG_WARN("%s: invalid TOS %s", name, node->value); + } } - } else if (!strcmp(node->name, "in_key") && is_gre) { - if (!strcmp(node->data, "flow")) { - config.flags |= TNL_F_IN_KEY_MATCH; + } else if (!strcmp(node->key, "ttl")) { + if (!strcmp(node->value, "inherit")) { + tnl_cfg.ttl_inherit = true; } else { - config.in_key = htonll(strtoull(node->data, NULL, 0)); + tnl_cfg.ttl = atoi(node->value); } - } else if (!strcmp(node->name, "out_key") && is_gre) { - if (!strcmp(node->data, "flow")) { - config.flags |= TNL_F_OUT_KEY_ACTION; - } else { - config.out_key = htonll(strtoull(node->data, NULL, 0)); + } else if (!strcmp(node->key, "dst_port") && needs_dst_port) { + tnl_cfg.dst_port = htons(atoi(node->value)); + } else if (!strcmp(node->key, "csum") && has_csum) { + if (!strcmp(node->value, "true")) { + tnl_cfg.csum = true; } - } else if (!strcmp(node->name, "tos")) { - if (!strcmp(node->data, "inherit")) { - config.flags |= TNL_F_TOS_INHERIT; - } else { - config.tos = atoi(node->data); + } else if (!strcmp(node->key, "df_default")) { + if (!strcmp(node->value, "false")) { + tnl_cfg.dont_fragment = false; } - } else if (!strcmp(node->name, "ttl")) { - if (!strcmp(node->data, "inherit")) { - config.flags |= TNL_F_TTL_INHERIT; + } else if (!strcmp(node->key, "peer_cert") && tnl_cfg.ipsec) { + if (smap_get(args, "certificate")) { + ipsec_mech_set = true; } else { - config.ttl = atoi(node->data); - } - } else if (!strcmp(node->name, "csum") && is_gre) { - if (!strcmp(node->data, "true")) { - config.flags |= TNL_F_CSUM; - } - } else if (!strcmp(node->name, "pmtud")) { - if (!strcmp(node->data, "false")) { - config.flags &= ~TNL_F_PMTUD; + const char *use_ssl_cert; + + /* If the "use_ssl_cert" is true, then "certificate" and + * "private_key" will be pulled from the SSL table. The + * use of this option is strongly discouraged, since it + * will like be removed when multiple SSL configurations + * are supported by OVS. + */ + use_ssl_cert = smap_get(args, "use_ssl_cert"); + if (!use_ssl_cert || strcmp(use_ssl_cert, "true")) { + VLOG_ERR("%s: 'peer_cert' requires 'certificate' argument", + name); + return EINVAL; + } + ipsec_mech_set = true; } - } else if (!strcmp(node->name, "header_cache")) { - if (!strcmp(node->data, "false")) { - config.flags &= ~TNL_F_HDR_CACHE; - } - } else if (!strcmp(node->name, "ipsec_cert") - || !strcmp(node->name, "ipsec_psk")) { + } else if (!strcmp(node->key, "psk") && tnl_cfg.ipsec) { ipsec_mech_set = true; + } else if (tnl_cfg.ipsec + && (!strcmp(node->key, "certificate") + || !strcmp(node->key, "private_key") + || !strcmp(node->key, "use_ssl_cert"))) { + /* Ignore options not used by the netdev. */ + } else if (!strcmp(node->key, "key") || + !strcmp(node->key, "in_key") || + !strcmp(node->key, "out_key")) { + /* Handled separately below. */ } else { - VLOG_WARN("%s: unknown %s argument '%s'", - name, type, node->name); + VLOG_WARN("%s: unknown %s argument '%s'", name, type, node->key); } } - /* IPsec doesn't work when header caching is enabled. Disable it if the - * IPsec local IP address and authentication mechanism have been defined. */ - if (ipsec_mech_set) { - VLOG_INFO("%s: header caching disabled due to use of IPsec", name); - config.flags &= ~TNL_F_HDR_CACHE; + /* Add a default destination port for VXLAN if none specified. */ + if (!strcmp(type, "vxlan") && !tnl_cfg.dst_port) { + tnl_cfg.dst_port = htons(VXLAN_DST_PORT); + } + + /* Add a default destination port for LISP if none specified. */ + if (!strcmp(type, "lisp") && !tnl_cfg.dst_port) { + tnl_cfg.dst_port = htons(LISP_DST_PORT); + } + + if (tnl_cfg.ipsec) { + static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER; + static pid_t pid = 0; + +#ifndef _WIN32 + ovs_mutex_lock(&mutex); + if (pid <= 0) { + char *file_name = xasprintf("%s/%s", ovs_rundir(), + "ovs-monitor-ipsec.pid"); + pid = read_pidfile(file_name); + free(file_name); + } + ovs_mutex_unlock(&mutex); +#endif + + if (pid < 0) { + VLOG_ERR("%s: IPsec requires the ovs-monitor-ipsec daemon", + name); + return EINVAL; + } + + if (smap_get(args, "peer_cert") && smap_get(args, "psk")) { + VLOG_ERR("%s: cannot define both 'peer_cert' and 'psk'", name); + return EINVAL; + } + + if (!ipsec_mech_set) { + VLOG_ERR("%s: IPsec requires an 'peer_cert' or psk' argument", + name); + return EINVAL; + } } - if (!config.daddr) { - VLOG_WARN("%s: %s type requires valid 'remote_ip' argument", - name, type); + if (!tnl_cfg.ip_dst && !tnl_cfg.ip_dst_flow) { + VLOG_ERR("%s: %s type requires valid 'remote_ip' argument", + name, type); return EINVAL; } + if (tnl_cfg.ip_src_flow && !tnl_cfg.ip_dst_flow) { + VLOG_ERR("%s: %s type requires 'remote_ip=flow' with 'local_ip=flow'", + name, type); + return EINVAL; + } + if (!tnl_cfg.ttl) { + tnl_cfg.ttl = DEFAULT_TTL; + } + + tnl_cfg.in_key = parse_key(args, "in_key", + &tnl_cfg.in_key_present, + &tnl_cfg.in_key_flow); + + tnl_cfg.out_key = parse_key(args, "out_key", + &tnl_cfg.out_key_present, + &tnl_cfg.out_key_flow); + + ovs_mutex_lock(&dev->mutex); + dev->tnl_cfg = tnl_cfg; + tunnel_check_status_change__(dev); + netdev_change_seq_changed(dev_); + ovs_mutex_unlock(&dev->mutex); - BUILD_ASSERT(sizeof config <= VPORT_CONFIG_SIZE); - memcpy(configp, &config, sizeof config); return 0; } static int -parse_patch_config(const struct netdev_dev *dev, const struct shash *args, - void *configp) +get_tunnel_config(const struct netdev *dev, struct smap *args) { - const char *name = netdev_dev_get_name(dev); - const char *peer; + struct netdev_vport *netdev = netdev_vport_cast(dev); + struct netdev_tunnel_config tnl_cfg; - peer = shash_find_data(args, "peer"); - if (!peer) { - VLOG_WARN("%s: patch type requires valid 'peer' argument", name); - return EINVAL; + ovs_mutex_lock(&netdev->mutex); + tnl_cfg = netdev->tnl_cfg; + ovs_mutex_unlock(&netdev->mutex); + + if (tnl_cfg.ip_dst) { + smap_add_format(args, "remote_ip", IP_FMT, IP_ARGS(tnl_cfg.ip_dst)); + } else if (tnl_cfg.ip_dst_flow) { + smap_add(args, "remote_ip", "flow"); + } + + if (tnl_cfg.ip_src) { + smap_add_format(args, "local_ip", IP_FMT, IP_ARGS(tnl_cfg.ip_src)); + } else if (tnl_cfg.ip_src_flow) { + smap_add(args, "local_ip", "flow"); } - if (shash_count(args) > 1) { - VLOG_WARN("%s: patch type takes only a 'peer' argument", name); + if (tnl_cfg.in_key_flow && tnl_cfg.out_key_flow) { + smap_add(args, "key", "flow"); + } else if (tnl_cfg.in_key_present && tnl_cfg.out_key_present + && tnl_cfg.in_key == tnl_cfg.out_key) { + smap_add_format(args, "key", "%"PRIu64, ntohll(tnl_cfg.in_key)); + } else { + if (tnl_cfg.in_key_flow) { + smap_add(args, "in_key", "flow"); + } else if (tnl_cfg.in_key_present) { + smap_add_format(args, "in_key", "%"PRIu64, + ntohll(tnl_cfg.in_key)); + } + + if (tnl_cfg.out_key_flow) { + smap_add(args, "out_key", "flow"); + } else if (tnl_cfg.out_key_present) { + smap_add_format(args, "out_key", "%"PRIu64, + ntohll(tnl_cfg.out_key)); + } + } + + if (tnl_cfg.ttl_inherit) { + smap_add(args, "ttl", "inherit"); + } else if (tnl_cfg.ttl != DEFAULT_TTL) { + smap_add_format(args, "ttl", "%"PRIu8, tnl_cfg.ttl); + } + + if (tnl_cfg.tos_inherit) { + smap_add(args, "tos", "inherit"); + } else if (tnl_cfg.tos) { + smap_add_format(args, "tos", "0x%x", tnl_cfg.tos); + } + + if (tnl_cfg.dst_port) { + uint16_t dst_port = ntohs(tnl_cfg.dst_port); + const char *type = netdev_get_type(dev); + + if ((!strcmp("vxlan", type) && dst_port != VXLAN_DST_PORT) || + (!strcmp("lisp", type) && dst_port != LISP_DST_PORT)) { + smap_add_format(args, "dst_port", "%d", dst_port); + } + } + + if (tnl_cfg.csum) { + smap_add(args, "csum", "true"); + } + + if (!tnl_cfg.dont_fragment) { + smap_add(args, "df_default", "false"); + } + + return 0; +} + +/* Code specific to patch ports. */ + +/* If 'netdev' is a patch port, returns the name of its peer as a malloc()'d + * string that the caller must free. + * + * If 'netdev' is not a patch port, returns NULL. */ +char * +netdev_vport_patch_peer(const struct netdev *netdev_) +{ + char *peer = NULL; + + if (netdev_vport_is_patch(netdev_)) { + struct netdev_vport *netdev = netdev_vport_cast(netdev_); + + ovs_mutex_lock(&netdev->mutex); + if (netdev->peer) { + peer = xstrdup(netdev->peer); + } + ovs_mutex_unlock(&netdev->mutex); + } + + return peer; +} + +void +netdev_vport_inc_rx(const struct netdev *netdev, + const struct dpif_flow_stats *stats) +{ + if (is_vport_class(netdev_get_class(netdev))) { + struct netdev_vport *dev = netdev_vport_cast(netdev); + + ovs_mutex_lock(&dev->mutex); + dev->stats.rx_packets += stats->n_packets; + dev->stats.rx_bytes += stats->n_bytes; + ovs_mutex_unlock(&dev->mutex); + } +} + +void +netdev_vport_inc_tx(const struct netdev *netdev, + const struct dpif_flow_stats *stats) +{ + if (is_vport_class(netdev_get_class(netdev))) { + struct netdev_vport *dev = netdev_vport_cast(netdev); + + ovs_mutex_lock(&dev->mutex); + dev->stats.tx_packets += stats->n_packets; + dev->stats.tx_bytes += stats->n_bytes; + ovs_mutex_unlock(&dev->mutex); + } +} + +static int +get_patch_config(const struct netdev *dev_, struct smap *args) +{ + struct netdev_vport *dev = netdev_vport_cast(dev_); + + ovs_mutex_lock(&dev->mutex); + if (dev->peer) { + smap_add(args, "peer", dev->peer); + } + ovs_mutex_unlock(&dev->mutex); + + return 0; +} + +static int +set_patch_config(struct netdev *dev_, const struct smap *args) +{ + struct netdev_vport *dev = netdev_vport_cast(dev_); + const char *name = netdev_get_name(dev_); + const char *peer; + + peer = smap_get(args, "peer"); + if (!peer) { + VLOG_ERR("%s: patch type requires valid 'peer' argument", name); return EINVAL; } - if (strlen(peer) >= MIN(IFNAMSIZ, VPORT_CONFIG_SIZE)) { - VLOG_WARN("%s: patch 'peer' arg too long", name); + if (smap_count(args) > 1) { + VLOG_ERR("%s: patch type takes only a 'peer' argument", name); return EINVAL; } if (!strcmp(name, peer)) { - VLOG_WARN("%s: patch peer must not be self", name); + VLOG_ERR("%s: patch peer must not be self", name); return EINVAL; } - strncpy(configp, peer, VPORT_CONFIG_SIZE); + ovs_mutex_lock(&dev->mutex); + free(dev->peer); + dev->peer = xstrdup(peer); + netdev_change_seq_changed(dev_); + ovs_mutex_unlock(&dev->mutex); + + return 0; +} + +static int +get_stats(const struct netdev *netdev, struct netdev_stats *stats) +{ + struct netdev_vport *dev = netdev_vport_cast(netdev); + + ovs_mutex_lock(&dev->mutex); + *stats = dev->stats; + ovs_mutex_unlock(&dev->mutex); return 0; } -#define VPORT_FUNCTIONS \ - NULL, /* init */ \ - NULL, /* run */ \ - NULL, /* wait */ \ - \ - netdev_vport_create, \ - netdev_vport_destroy, \ - netdev_vport_reconfigure, \ - \ - netdev_vport_open, \ - netdev_vport_close, \ - \ - NULL, /* enumerate */ \ +#define VPORT_FUNCTIONS(GET_CONFIG, SET_CONFIG, \ + GET_TUNNEL_CONFIG, GET_STATUS) \ + NULL, \ + netdev_vport_run, \ + netdev_vport_wait, \ \ - NULL, /* recv */ \ - NULL, /* recv_wait */ \ - NULL, /* drain */ \ + netdev_vport_alloc, \ + netdev_vport_construct, \ + netdev_vport_destruct, \ + netdev_vport_dealloc, \ + GET_CONFIG, \ + SET_CONFIG, \ + GET_TUNNEL_CONFIG, \ \ NULL, /* send */ \ NULL, /* send_wait */ \ \ netdev_vport_set_etheraddr, \ netdev_vport_get_etheraddr, \ - netdev_vport_get_mtu, \ + NULL, /* get_mtu */ \ + NULL, /* set_mtu */ \ NULL, /* get_ifindex */ \ NULL, /* get_carrier */ \ - netdev_vport_get_stats, \ - netdev_vport_set_stats, \ + NULL, /* get_carrier_resets */ \ + NULL, /* get_miimon */ \ + get_stats, \ + NULL, /* set_stats */ \ \ NULL, /* get_features */ \ NULL, /* set_advertisements */ \ - NULL, /* get_vlan_vid */ \ \ NULL, /* set_policing */ \ NULL, /* get_qos_types */ \ @@ -603,7 +792,9 @@ parse_patch_config(const struct netdev_dev *dev, const struct shash *args, NULL, /* set_queue */ \ NULL, /* delete_queue */ \ NULL, /* get_queue_stats */ \ - NULL, /* dump_queues */ \ + NULL, /* queue_dump_start */ \ + NULL, /* queue_dump_next */ \ + NULL, /* queue_dump_done */ \ NULL, /* dump_queue_stats */ \ \ NULL, /* get_in4 */ \ @@ -611,25 +802,57 @@ parse_patch_config(const struct netdev_dev *dev, const struct shash *args, NULL, /* get_in6 */ \ NULL, /* add_router */ \ NULL, /* get_next_hop */ \ + GET_STATUS, \ NULL, /* arp_lookup */ \ \ netdev_vport_update_flags, \ \ - netdev_vport_poll_add, \ - netdev_vport_poll_remove, + NULL, /* rx_alloc */ \ + NULL, /* rx_construct */ \ + NULL, /* rx_destruct */ \ + NULL, /* rx_dealloc */ \ + NULL, /* rx_recv */ \ + NULL, /* rx_wait */ \ + NULL, /* rx_drain */ + +#define TUNNEL_CLASS(NAME, DPIF_PORT) \ + { DPIF_PORT, \ + { NAME, VPORT_FUNCTIONS(get_tunnel_config, \ + set_tunnel_config, \ + get_netdev_tunnel_config, \ + tunnel_get_status) }} void -netdev_vport_register(void) +netdev_vport_tunnel_register(void) { static const struct vport_class vport_classes[] = { - { { "gre", VPORT_FUNCTIONS }, parse_tunnel_config }, - { { "capwap", VPORT_FUNCTIONS }, parse_tunnel_config }, - { { "patch", VPORT_FUNCTIONS }, parse_patch_config } + TUNNEL_CLASS("gre", "gre_system"), + TUNNEL_CLASS("ipsec_gre", "gre_system"), + TUNNEL_CLASS("gre64", "gre64_system"), + TUNNEL_CLASS("ipsec_gre64", "gre64_system"), + TUNNEL_CLASS("vxlan", "vxlan_system"), + TUNNEL_CLASS("lisp", "lisp_system") }; + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; - int i; + if (ovsthread_once_start(&once)) { + int i; - for (i = 0; i < ARRAY_SIZE(vport_classes); i++) { - netdev_register_provider(&vport_classes[i].netdev_class); + for (i = 0; i < ARRAY_SIZE(vport_classes); i++) { + netdev_register_provider(&vport_classes[i].netdev_class); + } + ovsthread_once_done(&once); } } + +void +netdev_vport_patch_register(void) +{ + static const struct vport_class patch_class = + { NULL, + { "patch", VPORT_FUNCTIONS(get_patch_config, + set_patch_config, + NULL, + NULL) }}; + netdev_register_provider(&patch_class.netdev_class); +}