X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev-vport.c;h=330854493a1903e7afeaedcb134710cf736e7ed2;hb=85b20fd6ee585f462e012fbcc7f966a81edab2ed;hp=5ab058635495f877a65a73ace57fc5afbe0e3e7a;hpb=eb5f3e933b0922d369ad839b2afa1cd7ab764b6e;p=sliver-openvswitch.git diff --git a/lib/netdev-vport.c b/lib/netdev-vport.c index 5ab058635..330854493 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. @@ -20,54 +20,67 @@ #include #include +#include +#include +#include #include #include +#include "byte-order.h" +#include "daemon.h" +#include "dirs.h" +#include "dpif-linux.h" +#include "hash.h" +#include "hmap.h" #include "list.h" +#include "netdev-linux.h" #include "netdev-provider.h" -#include "openvswitch/datapath-protocol.h" +#include "netlink.h" +#include "netlink-notifier.h" +#include "netlink-socket.h" +#include "ofpbuf.h" #include "openvswitch/tunnel.h" #include "packets.h" +#include "route-table.h" #include "shash.h" #include "socket-util.h" +#include "unaligned.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; -}; +/* Default to the OTV port, per the VXLAN IETF draft. */ +#define VXLAN_DST_PORT 8472 + +#define DEFAULT_TTL 64 struct netdev_dev_vport { struct netdev_dev netdev_dev; -}; - -struct netdev_vport { - struct netdev netdev; -}; - -struct vport_info { - const char *devname; - const char *type; - void *config; + struct ofpbuf *options; + unsigned int change_seq; + uint8_t etheraddr[ETH_ADDR_LEN]; + struct netdev_tunnel_config tnl_cfg; }; struct vport_class { - const struct netdev_class netdev_class; - int (*parse_config)(struct vport_info *port, const struct shash *args); + enum ovs_vport_type type; + struct netdev_class netdev_class; + int (*parse_config)(const char *name, const char *type, + const struct smap *args, struct ofpbuf *options, + struct netdev_tunnel_config *tnl_cfg); + int (*unparse_config)(const char *name, const char *type, + const struct nlattr *options, size_t options_len, + struct smap *args); }; -static struct shash netdev_vport_notifiers = - SHASH_INITIALIZER(&netdev_vport_notifiers); - 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 **); + struct netdev_dev **); static void netdev_vport_poll_notify(const struct netdev *); +static int tnl_port_config_from_nlattr(const struct nlattr *options, + size_t options_len, + struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1]); static bool is_vport_class(const struct netdev_class *class) @@ -89,79 +102,115 @@ netdev_dev_vport_cast(const struct netdev_dev *netdev_dev) return CONTAINER_OF(netdev_dev, struct netdev_dev_vport, netdev_dev); } -static struct netdev_vport * -netdev_vport_cast(const struct netdev *netdev) +static struct netdev_dev_vport * +netdev_vport_get_dev(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); + return netdev_dev_vport_cast(netdev_get_dev(netdev)); } -static int -netdev_vport_parse_config(const struct netdev_class *netdev_class, - const char *name, const struct shash *args, - void **configp) -{ - const struct vport_class *c = vport_class_cast(netdev_class); - if (c->parse_config) { - struct vport_info info; - int error; - - info.devname = name; - info.type = netdev_class->type; - error = (c->parse_config)(&info, args); - *configp = info.config; - return error; - } else { - if (!shash_is_empty(args)) { - VLOG_WARN("%s: arguments for %s vports should be empty", - name, netdev_class->type); - } - *configp = NULL; - return 0; - } +static const struct netdev_tunnel_config * +get_netdev_tunnel_config(const struct netdev_dev *netdev_dev) +{ + return &netdev_dev_vport_cast(netdev_dev)->tnl_cfg; } -static int -netdev_vport_create(const struct netdev_class *class, const char *name, - const struct shash *args OVS_UNUSED, - struct netdev_dev **netdev_devp) +/* If 'netdev' is a vport netdev, returns an ofpbuf that contains Netlink + * options to include in OVS_VPORT_ATTR_OPTIONS for configuring that vport. + * Otherwise returns NULL. */ +const struct ofpbuf * +netdev_vport_get_options(const struct netdev *netdev) { - int err; - struct odp_vport_add ova; - struct netdev_dev_vport *netdev_dev; + const struct netdev_dev *dev = netdev_get_dev(netdev); - ovs_strlcpy(ova.port_type, class->type, sizeof ova.port_type); - ovs_strlcpy(ova.devname, name, sizeof ova.devname); - err = netdev_vport_parse_config(class, name, args, &ova.config); - if (err) { - goto exit; - } + return (is_vport_class(netdev_dev_get_class(dev)) + ? netdev_dev_vport_cast(dev)->options + : NULL); +} - err = netdev_vport_do_ioctl(ODP_VPORT_ADD, &ova); +enum ovs_vport_type +netdev_vport_get_vport_type(const struct netdev *netdev) +{ + const struct netdev_dev *dev = netdev_get_dev(netdev); + const struct netdev_class *class = netdev_dev_get_class(dev); + + return (is_vport_class(class) ? vport_class_cast(class)->type + : class == &netdev_internal_class ? OVS_VPORT_TYPE_INTERNAL + : (class == &netdev_linux_class || + class == &netdev_tap_class) ? OVS_VPORT_TYPE_NETDEV + : OVS_VPORT_TYPE_UNSPEC); +} - if (err == EBUSY) { - VLOG_WARN("%s: destroying existing device", name); +static uint32_t +get_u32_or_zero(const struct nlattr *a) +{ + return a ? nl_attr_get_u32(a) : 0; +} - err = netdev_vport_do_ioctl(ODP_VPORT_DEL, ova.devname); - if (err) { - goto exit; +const char * +netdev_vport_get_netdev_type(const struct dpif_linux_vport *vport) +{ + struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1]; + + switch (vport->type) { + case OVS_VPORT_TYPE_UNSPEC: + break; + + case OVS_VPORT_TYPE_NETDEV: + return "system"; + + case OVS_VPORT_TYPE_INTERNAL: + return "internal"; + + case OVS_VPORT_TYPE_PATCH: + return "patch"; + + case OVS_VPORT_TYPE_GRE: + if (tnl_port_config_from_nlattr(vport->options, vport->options_len, + a)) { + break; } + return (get_u32_or_zero(a[OVS_TUNNEL_ATTR_FLAGS]) & TNL_F_IPSEC + ? "ipsec_gre" : "gre"); - err = netdev_vport_do_ioctl(ODP_VPORT_ADD, &ova); - } - if (err) { - goto exit; + case OVS_VPORT_TYPE_GRE64: + if (tnl_port_config_from_nlattr(vport->options, vport->options_len, + a)) { + break; + } + return (get_u32_or_zero(a[OVS_TUNNEL_ATTR_FLAGS]) & TNL_F_IPSEC + ? "ipsec_gre64" : "gre64"); + + case OVS_VPORT_TYPE_CAPWAP: + return "capwap"; + + case OVS_VPORT_TYPE_VXLAN: + return "vxlan"; + + case OVS_VPORT_TYPE_FT_GRE: + case __OVS_VPORT_TYPE_MAX: + break; } - netdev_dev = xmalloc(sizeof *netdev_dev); - netdev_dev_init(&netdev_dev->netdev_dev, name, class); + VLOG_WARN_RL(&rl, "dp%d: port `%s' has unsupported type %u", + vport->dp_ifindex, vport->name, (unsigned int) vport->type); + return "unknown"; +} + +static int +netdev_vport_create(const struct netdev_class *netdev_class, const char *name, + struct netdev_dev **netdev_devp) +{ + struct netdev_dev_vport *dev; - *netdev_devp = &netdev_dev->netdev_dev; + dev = xzalloc(sizeof *dev); + netdev_dev_init(&dev->netdev_dev, name, netdev_class); + dev->change_seq = 1; + eth_addr_random(dev->etheraddr); -exit: - free(ova.config); - return err; + *netdev_devp = &dev->netdev_dev; + route_table_register(); + + return 0; } static void @@ -169,68 +218,107 @@ netdev_vport_destroy(struct netdev_dev *netdev_dev_) { struct netdev_dev_vport *netdev_dev = netdev_dev_vport_cast(netdev_dev_); - netdev_vport_do_ioctl(ODP_VPORT_DEL, - (char *)netdev_dev_get_name(netdev_dev_)); + ofpbuf_delete(netdev_dev->options); + route_table_unregister(); free(netdev_dev); } static int -netdev_vport_open(struct netdev_dev *netdev_dev_, int ethertype OVS_UNUSED, - struct netdev **netdevp) +netdev_vport_open(struct netdev_dev *netdev_dev, struct netdev **netdevp) { - struct netdev_vport *netdev; - - netdev = xmalloc(sizeof *netdev); - netdev_init(&netdev->netdev, netdev_dev_); - - *netdevp = &netdev->netdev; + *netdevp = xmalloc(sizeof **netdevp); + netdev_init(*netdevp, netdev_dev); return 0; } static void -netdev_vport_close(struct netdev *netdev_) +netdev_vport_close(struct netdev *netdev) { - struct netdev_vport *netdev = netdev_vport_cast(netdev_); free(netdev); } static int -netdev_vport_reconfigure(struct netdev_dev *netdev_dev, - const struct shash *args) +netdev_vport_get_config(struct netdev_dev *dev_, struct smap *args) { - const char *name = netdev_dev_get_name(netdev_dev); - struct odp_vport_mod ovm; - int err; + 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_); + const char *name = netdev_dev_get_name(dev_); + int error; + + if (!dev->options) { + struct dpif_linux_vport reply; + struct ofpbuf *buf; + + error = dpif_linux_vport_get(name, &reply, &buf); + if (error) { + VLOG_ERR_RL(&rl, "%s: vport query failed (%s)", + name, strerror(error)); + return error; + } - ovs_strlcpy(ovm.devname, name, sizeof ovm.devname); - err = netdev_vport_parse_config(netdev_dev_get_class(netdev_dev), name, - args, &ovm.config); - if (err) { - return err; + dev->options = ofpbuf_clone_data(reply.options, reply.options_len); + ofpbuf_delete(buf); } - err = netdev_vport_do_ioctl(ODP_VPORT_MOD, &ovm); - free(ovm.config); - return err; + error = vport_class->unparse_config(name, netdev_class->type, + dev->options->data, + dev->options->size, + args); + if (error) { + VLOG_ERR_RL(&rl, "%s: failed to parse kernel config (%s)", + name, strerror(error)); + } + return error; } static int -netdev_vport_set_etheraddr(struct netdev *netdev, - const uint8_t mac[ETH_ADDR_LEN]) +netdev_vport_set_config(struct netdev_dev *dev_, const struct smap *args) { - 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); - - err = netdev_vport_do_ioctl(ODP_VPORT_ETHER_SET, &vport_ether); - if (err) { - return err; + 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_); + const char *name = netdev_dev_get_name(dev_); + struct netdev_tunnel_config tnl_cfg; + struct ofpbuf *options; + int error; + + options = ofpbuf_new(64); + error = vport_class->parse_config(name, netdev_dev_get_type(dev_), + args, options, &tnl_cfg); + if (!error + && (!dev->options + || options->size != dev->options->size + || memcmp(options->data, dev->options->data, options->size))) { + struct dpif_linux_vport vport; + + dpif_linux_vport_init(&vport); + vport.cmd = OVS_VPORT_CMD_SET; + vport.name = name; + vport.options = options->data; + vport.options_len = options->size; + error = dpif_linux_vport_transact(&vport, NULL, NULL); + 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()). */ + ofpbuf_delete(dev->options); + dev->options = options; + dev->tnl_cfg = tnl_cfg; + options = NULL; + error = 0; + } } + ofpbuf_delete(options); + return error; +} + +static int +netdev_vport_set_etheraddr(struct netdev *netdev, + const uint8_t mac[ETH_ADDR_LEN]) +{ + memcpy(netdev_vport_get_dev(netdev)->etheraddr, mac, ETH_ADDR_LEN); netdev_vport_poll_notify(netdev); return 0; } @@ -239,118 +327,82 @@ static int 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); - - err = netdev_vport_do_ioctl(ODP_VPORT_ETHER_GET, &vport_ether); - if (err) { - return err; - } - - memcpy(mac, vport_ether.ether_addr, ETH_ADDR_LEN); + memcpy(mac, netdev_vport_get_dev(netdev)->etheraddr, ETH_ADDR_LEN); return 0; } -static int -netdev_vport_get_mtu(const struct netdev *netdev, int *mtup) +/* Copies 'src' into 'dst', performing format conversion in the process. + * + * 'src' is allowed to be misaligned. */ +static void +netdev_stats_from_ovs_vport_stats(struct netdev_stats *dst, + const struct ovs_vport_stats *src) { - struct odp_vport_mtu vport_mtu; - int err; + dst->rx_packets = get_unaligned_u64(&src->rx_packets); + dst->tx_packets = get_unaligned_u64(&src->tx_packets); + dst->rx_bytes = get_unaligned_u64(&src->rx_bytes); + dst->tx_bytes = get_unaligned_u64(&src->tx_bytes); + dst->rx_errors = get_unaligned_u64(&src->rx_errors); + dst->tx_errors = get_unaligned_u64(&src->tx_errors); + dst->rx_dropped = get_unaligned_u64(&src->rx_dropped); + dst->tx_dropped = get_unaligned_u64(&src->tx_dropped); + dst->multicast = 0; + dst->collisions = 0; + dst->rx_length_errors = 0; + dst->rx_over_errors = 0; + dst->rx_crc_errors = 0; + dst->rx_frame_errors = 0; + dst->rx_fifo_errors = 0; + dst->rx_missed_errors = 0; + dst->tx_aborted_errors = 0; + dst->tx_carrier_errors = 0; + dst->tx_fifo_errors = 0; + dst->tx_heartbeat_errors = 0; + dst->tx_window_errors = 0; +} - ovs_strlcpy(vport_mtu.devname, netdev_get_name(netdev), - sizeof vport_mtu.devname); +int +netdev_vport_get_stats(const struct netdev *netdev, struct netdev_stats *stats) +{ + struct dpif_linux_vport reply; + struct ofpbuf *buf; + int error; - err = netdev_vport_do_ioctl(ODP_VPORT_MTU_GET, &vport_mtu); - if (err) { - return err; + error = dpif_linux_vport_get(netdev_get_name(netdev), &reply, &buf); + if (error) { + return error; + } else if (!reply.stats) { + ofpbuf_delete(buf); + return EOPNOTSUPP; } - *mtup = vport_mtu.mtu; + netdev_stats_from_ovs_vport_stats(stats, reply.stats); + + ofpbuf_delete(buf); + return 0; } -int -netdev_vport_get_stats(const struct netdev *netdev, struct netdev_stats *stats) +static int +tunnel_get_status(const struct netdev *netdev, struct smap *smap) { - 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; + static char iface[IFNAMSIZ]; + ovs_be32 route; - return 0; -} + route = netdev_vport_get_dev(netdev)->tnl_cfg.ip_dst; + if (route_table_get_name(route, iface)) { + struct netdev *egress_netdev; -int -netdev_vport_set_stats(struct netdev *netdev, const struct netdev_stats *stats) -{ - 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; - - err = netdev_vport_do_ioctl(ODP_VPORT_STATS_SET, &ovsr); - - /* 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; - } - - return err; + smap_add(smap, "tunnel_egress_iface", iface); + + if (!netdev_open(iface, "system", &egress_netdev)) { + smap_add(smap, "tunnel_egress_iface_carrier", + netdev_get_carrier(egress_netdev) ? "up" : "down"); + netdev_close(egress_netdev); + } + } + + return 0; } static int @@ -366,243 +418,442 @@ netdev_vport_update_flags(struct netdev *netdev OVS_UNUSED, return 0; } -static char * -make_poll_name(const struct netdev *netdev) +static unsigned int +netdev_vport_change_seq(const struct netdev *netdev) { - return xasprintf("%s:%s", netdev_get_type(netdev), netdev_get_name(netdev)); + return netdev_vport_get_dev(netdev)->change_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; - } - - 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_run(void) +{ + route_table_run(); } 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; - - list = list_remove(¬ifier->list_node); - if (list_is_empty(list)) { - shash_delete(&netdev_vport_notifiers, notifier->shash_node); - free(list); - } - - free(notifier); + route_table_wait(); } /* Helper functions. */ -static int -netdev_vport_do_ioctl(int cmd, void *arg) +static void +netdev_vport_poll_notify(const struct netdev *netdev) { - static int ioctl_fd = -1; + struct netdev_dev_vport *ndv = netdev_vport_get_dev(netdev); - 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; - } + ndv->change_seq++; + if (!ndv->change_seq) { + ndv->change_seq++; } - - return ioctl(ioctl_fd, cmd, arg) ? errno : 0; } + +/* Code specific to individual vport types. */ -static void -netdev_vport_poll_notify(const struct netdev *netdev) +static ovs_be64 +parse_key(const struct smap *args, const char *name, + bool *present, bool *flow) { - char *poll_name = make_poll_name(netdev); - struct list *list = shash_find_data(&netdev_vport_notifiers, - poll_name); + const char *s; - if (list) { - struct netdev_vport_notifier *notifier; + *present = false; + *flow = false; - LIST_FOR_EACH (notifier, list_node, list) { - struct netdev_notifier *n = ¬ifier->notifier; - n->cb(n); + s = smap_get(args, name); + if (!s) { + s = smap_get(args, "key"); + if (!s) { + return 0; } } - free(poll_name); + *present = true; + + if (!strcmp(s, "flow")) { + *flow = true; + return 0; + } else { + return htonll(strtoull(s, NULL, 0)); + } } - -/* Code specific to individual vport types. */ static int -parse_tunnel_config(struct vport_info *port, const struct shash *args) +parse_tunnel_config(const char *name, const char *type, + const struct smap *args, struct ofpbuf *options, + struct netdev_tunnel_config *tnl_cfg_) { - const char *name = port->devname; - bool is_gre = !strcmp(port->type, "gre"); - struct tnl_port_config *config; - struct shash_node *node; - bool ipsec_ip_set = false; - bool ipsec_mech_set = false; + bool ipsec_mech_set, needs_dst_port, has_csum; + struct netdev_tunnel_config tnl_cfg; + struct smap_node *node; + uint8_t flags; + + flags = TNL_F_DF_DEFAULT; + has_csum = strstr(type, "gre"); + ipsec_mech_set = false; + memset(&tnl_cfg, 0, sizeof tnl_cfg); + + if (!strcmp(type, "capwap")) { + VLOG_WARN_ONCE("CAPWAP tunnel support is deprecated."); + } - config = port->config = xzalloc(sizeof *config); - config->flags |= TNL_F_PMTUD; - config->flags |= TNL_F_HDR_CACHE; + needs_dst_port = !strcmp(type, "vxlan"); + tnl_cfg.ipsec = strstr(type, "ipsec"); + if (tnl_cfg.ipsec) { + flags |= TNL_F_IPSEC; + } + tnl_cfg.dont_fragment = true; - SHASH_FOR_EACH (node, args) { - if (!strcmp(node->name, "remote_ip")) { + SMAP_FOR_EACH (node, args) { + if (!strcmp(node->key, "remote_ip")) { struct in_addr in_addr; - if (lookup_ip(node->data, &in_addr)) { - VLOG_WARN("%s: bad %s 'remote_ip'", name, port->type); + if (lookup_ip(node->value, &in_addr)) { + VLOG_WARN("%s: bad %s 'remote_ip'", name, type); } 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)) { - VLOG_WARN("%s: bad %s 'local_ip'", name, port->type); + 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")) { + flags |= TNL_F_TOS_INHERIT; + tnl_cfg.tos_inherit = true; } else { - config->out_key = config->in_key = htonl(atoi(node->data)); + char *endptr; + int tos; + tos = strtol(node->value, &endptr, 0); + if (*endptr == '\0' && tos == (tos & IP_DSCP_MASK)) { + nl_msg_put_u8(options, OVS_TUNNEL_ATTR_TOS, tos); + 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")) { + flags |= TNL_F_TTL_INHERIT; + tnl_cfg.ttl_inherit = true; } else { - config->in_key = htonl(atoi(node->data)); + nl_msg_put_u8(options, OVS_TUNNEL_ATTR_TTL, atoi(node->value)); + 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 = htonl(atoi(node->data)); + } else if (!strcmp(node->key, "dst_port") && needs_dst_port) { + tnl_cfg.dst_port = htons(atoi(node->value)); + nl_msg_put_u16(options, OVS_TUNNEL_ATTR_DST_PORT, + atoi(node->value)); + } else if (!strcmp(node->key, "csum") && has_csum) { + if (!strcmp(node->value, "true")) { + flags |= TNL_F_CSUM; + 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")) { + flags &= ~TNL_F_DF_DEFAULT; + 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; + 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, "pmtud")) { - if (!strcmp(node->data, "false")) { - config->flags &= ~TNL_F_PMTUD; - } - } else if (!strcmp(node->name, "header_cache")) { - if (!strcmp(node->data, "false")) { - config->flags &= ~TNL_F_HDR_CACHE; - } - } else if (!strcmp(node->name, "ipsec_local_ip")) { - ipsec_ip_set = true; - } 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, port->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_ip_set && 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 (needs_dst_port && !tnl_cfg.dst_port) { + nl_msg_put_u16(options, OVS_TUNNEL_ATTR_DST_PORT, VXLAN_DST_PORT); + tnl_cfg.dst_port = htons(VXLAN_DST_PORT); } - if (!config->daddr) { - VLOG_WARN("%s: %s type requires valid 'remote_ip' argument", - name, port->type); + if (tnl_cfg.ipsec) { + static pid_t pid = 0; + if (pid <= 0) { + char *file_name = xasprintf("%s/%s", ovs_rundir(), + "ovs-monitor-ipsec.pid"); + pid = read_pidfile(file_name); + free(file_name); + } + + 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 (!tnl_cfg.ip_dst) { + VLOG_ERR("%s: %s type requires valid 'remote_ip' argument", + name, type); + return EINVAL; + } + nl_msg_put_be32(options, OVS_TUNNEL_ATTR_DST_IPV4, tnl_cfg.ip_dst); + + if (tnl_cfg.ip_src) { + if (ip_is_multicast(tnl_cfg.ip_dst)) { + VLOG_WARN("%s: remote_ip is multicast, ignoring local_ip", name); + tnl_cfg.ip_src = 0; + } else { + nl_msg_put_be32(options, OVS_TUNNEL_ATTR_SRC_IPV4, tnl_cfg.ip_src); + } + } + + 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); + if (tnl_cfg.in_key_present && !tnl_cfg.in_key_flow) { + nl_msg_put_be64(options, OVS_TUNNEL_ATTR_IN_KEY, tnl_cfg.in_key); + } + + tnl_cfg.out_key = parse_key(args, "out_key", + &tnl_cfg.out_key_present, + &tnl_cfg.out_key_flow); + if (tnl_cfg.out_key_present && !tnl_cfg.out_key_flow) { + nl_msg_put_be64(options, OVS_TUNNEL_ATTR_OUT_KEY, tnl_cfg.out_key); + } + nl_msg_put_u32(options, OVS_TUNNEL_ATTR_FLAGS, flags); + + *tnl_cfg_ = tnl_cfg; + + return 0; +} + +static int +tnl_port_config_from_nlattr(const struct nlattr *options, size_t options_len, + struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1]) +{ + static const struct nl_policy ovs_tunnel_policy[] = { + [OVS_TUNNEL_ATTR_FLAGS] = { .type = NL_A_U32, .optional = true }, + [OVS_TUNNEL_ATTR_DST_IPV4] = { .type = NL_A_BE32, .optional = true }, + [OVS_TUNNEL_ATTR_SRC_IPV4] = { .type = NL_A_BE32, .optional = true }, + [OVS_TUNNEL_ATTR_IN_KEY] = { .type = NL_A_BE64, .optional = true }, + [OVS_TUNNEL_ATTR_OUT_KEY] = { .type = NL_A_BE64, .optional = true }, + [OVS_TUNNEL_ATTR_TOS] = { .type = NL_A_U8, .optional = true }, + [OVS_TUNNEL_ATTR_TTL] = { .type = NL_A_U8, .optional = true }, + [OVS_TUNNEL_ATTR_DST_PORT] = { .type = NL_A_U16, .optional = true }, + }; + struct ofpbuf buf; + + ofpbuf_use_const(&buf, options, options_len); + if (!nl_policy_parse(&buf, 0, ovs_tunnel_policy, + a, ARRAY_SIZE(ovs_tunnel_policy))) { return EINVAL; } + return 0; +} + +static uint64_t +get_be64_or_zero(const struct nlattr *a) +{ + return a ? ntohll(nl_attr_get_be64(a)) : 0; +} + +static int +unparse_tunnel_config(const char *name OVS_UNUSED, const char *type OVS_UNUSED, + const struct nlattr *options, size_t options_len, + struct smap *args) +{ + struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1]; + uint32_t flags; + int error; + + error = tnl_port_config_from_nlattr(options, options_len, a); + if (error) { + return error; + } + + if (a[OVS_TUNNEL_ATTR_DST_IPV4]) { + ovs_be32 daddr = nl_attr_get_be32(a[OVS_TUNNEL_ATTR_DST_IPV4]); + smap_add_format(args, "remote_ip", IP_FMT, IP_ARGS(daddr)); + } + + if (a[OVS_TUNNEL_ATTR_SRC_IPV4]) { + ovs_be32 saddr = nl_attr_get_be32(a[OVS_TUNNEL_ATTR_SRC_IPV4]); + smap_add_format(args, "local_ip", IP_FMT, IP_ARGS(saddr)); + } + + if (!a[OVS_TUNNEL_ATTR_IN_KEY] && !a[OVS_TUNNEL_ATTR_OUT_KEY]) { + smap_add(args, "key", "flow"); + } else { + uint64_t in_key = get_be64_or_zero(a[OVS_TUNNEL_ATTR_IN_KEY]); + uint64_t out_key = get_be64_or_zero(a[OVS_TUNNEL_ATTR_OUT_KEY]); + + if (in_key && in_key == out_key) { + smap_add_format(args, "key", "%"PRIu64, in_key); + } else { + if (!a[OVS_TUNNEL_ATTR_IN_KEY]) { + smap_add(args, "in_key", "flow"); + } else if (in_key) { + smap_add_format(args, "in_key", "%"PRIu64, in_key); + } + + if (!a[OVS_TUNNEL_ATTR_OUT_KEY]) { + smap_add(args, "out_key", "flow"); + } else if (out_key) { + smap_add_format(args, "out_key", "%"PRIu64, out_key); + } + } + } + + flags = get_u32_or_zero(a[OVS_TUNNEL_ATTR_FLAGS]); + + if (flags & TNL_F_TTL_INHERIT) { + smap_add(args, "ttl", "inherit"); + } else if (a[OVS_TUNNEL_ATTR_TTL]) { + int ttl = nl_attr_get_u8(a[OVS_TUNNEL_ATTR_TTL]); + smap_add_format(args, "ttl", "%d", ttl); + } + + if (flags & TNL_F_TOS_INHERIT) { + smap_add(args, "tos", "inherit"); + } else if (a[OVS_TUNNEL_ATTR_TOS]) { + int tos = nl_attr_get_u8(a[OVS_TUNNEL_ATTR_TOS]); + smap_add_format(args, "tos", "0x%x", tos); + } + + if (a[OVS_TUNNEL_ATTR_DST_PORT]) { + uint16_t dst_port = nl_attr_get_u16(a[OVS_TUNNEL_ATTR_DST_PORT]); + if (dst_port != VXLAN_DST_PORT) { + smap_add_format(args, "dst_port", "%d", dst_port); + } + } + + if (flags & TNL_F_CSUM) { + smap_add(args, "csum", "true"); + } + if (flags & TNL_F_DF_INHERIT) { + /* Shouldn't happen as "df_inherit" is no longer supported. However, + * for completeness we report it if it's there. */ + smap_add(args, "df_inherit", "true"); + } + if (!(flags & TNL_F_DF_DEFAULT)) { + smap_add(args, "df_default", "false"); + } return 0; } static int -parse_patch_config(struct vport_info *port, const struct shash *args) +parse_patch_config(const char *name, const char *type OVS_UNUSED, + const struct smap *args, struct ofpbuf *options, + struct netdev_tunnel_config *tnl_cfg) { - const char *name = port->devname; const char *peer; - peer = shash_find_data(args, "peer"); + memset(tnl_cfg, 0, sizeof *tnl_cfg); + + peer = smap_get(args, "peer"); if (!peer) { - VLOG_WARN("%s: patch type requires valid 'peer' argument", name); + VLOG_ERR("%s: patch type requires valid 'peer' argument", name); return EINVAL; } - if (shash_count(args) > 1) { - VLOG_WARN("%s: patch type takes only a 'peer' argument", name); + if (smap_count(args) > 1) { + VLOG_ERR("%s: patch type takes only a 'peer' argument", name); return EINVAL; } if (strlen(peer) >= IFNAMSIZ) { - VLOG_WARN("%s: patch 'peer' arg too long", name); + VLOG_ERR("%s: patch 'peer' arg too long", 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; } - port->config = xstrdup(peer); + nl_msg_put_string(options, OVS_PATCH_ATTR_PEER, peer); return 0; } + +static int +unparse_patch_config(const char *name OVS_UNUSED, const char *type OVS_UNUSED, + const struct nlattr *options, size_t options_len, + struct smap *args) +{ + static const struct nl_policy ovs_patch_policy[] = { + [OVS_PATCH_ATTR_PEER] = { .type = NL_A_STRING, + .max_len = IFNAMSIZ, + .optional = false } + }; + + struct nlattr *a[ARRAY_SIZE(ovs_patch_policy)]; + struct ofpbuf buf; + + ofpbuf_use_const(&buf, options, options_len); + if (!nl_policy_parse(&buf, 0, ovs_patch_policy, + a, ARRAY_SIZE(ovs_patch_policy))) { + return EINVAL; + } + + smap_add(args, "peer", nl_attr_get_string(a[OVS_PATCH_ATTR_PEER])); + return 0; +} -#define VPORT_FUNCTIONS \ - NULL, /* init */ \ - NULL, /* run */ \ - NULL, /* wait */ \ +#define VPORT_FUNCTIONS(GET_TUNNEL_CONFIG, GET_STATUS) \ + NULL, \ + netdev_vport_run, \ + netdev_vport_wait, \ \ netdev_vport_create, \ netdev_vport_destroy, \ - netdev_vport_reconfigure, \ + netdev_vport_get_config, \ + netdev_vport_set_config, \ + GET_TUNNEL_CONFIG, \ \ netdev_vport_open, \ netdev_vport_close, \ \ - NULL, /* enumerate */ \ - \ + NULL, /* listen */ \ NULL, /* recv */ \ NULL, /* recv_wait */ \ NULL, /* drain */ \ @@ -612,15 +863,17 @@ parse_patch_config(struct vport_info *port, const struct shash *args) \ netdev_vport_set_etheraddr, \ netdev_vport_get_etheraddr, \ - netdev_vport_get_mtu, \ + NULL, /* get_mtu */ \ + NULL, /* set_mtu */ \ NULL, /* get_ifindex */ \ NULL, /* get_carrier */ \ + NULL, /* get_carrier_resets */ \ + NULL, /* get_miimon */ \ netdev_vport_get_stats, \ - netdev_vport_set_stats, \ + NULL, /* set_stats */ \ \ NULL, /* get_features */ \ NULL, /* set_advertisements */ \ - NULL, /* get_vlan_vid */ \ \ NULL, /* set_policing */ \ NULL, /* get_qos_types */ \ @@ -639,26 +892,38 @@ parse_patch_config(struct vport_info *port, 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, + netdev_vport_change_seq -static const struct vport_class vport_gre_class - = { { "gre", VPORT_FUNCTIONS }, parse_tunnel_config }; - -static const struct vport_class vport_capwap_class - = { { "capwap", VPORT_FUNCTIONS }, parse_tunnel_config }; - -static const struct vport_class vport_patch_class - = { { "patch", VPORT_FUNCTIONS }, parse_patch_config }; +#define TUNNEL_CLASS(NAME, VPORT_TYPE) \ + { VPORT_TYPE, \ + { NAME, VPORT_FUNCTIONS(get_netdev_tunnel_config, \ + tunnel_get_status) }, \ + parse_tunnel_config, unparse_tunnel_config } void netdev_vport_register(void) { - netdev_register_provider(&vport_gre_class.netdev_class); - netdev_register_provider(&vport_capwap_class.netdev_class); - netdev_register_provider(&vport_patch_class.netdev_class); + static const struct vport_class vport_classes[] = { + TUNNEL_CLASS("gre", OVS_VPORT_TYPE_GRE), + TUNNEL_CLASS("ipsec_gre", OVS_VPORT_TYPE_GRE), + TUNNEL_CLASS("gre64", OVS_VPORT_TYPE_GRE64), + TUNNEL_CLASS("ipsec_gre64", OVS_VPORT_TYPE_GRE64), + TUNNEL_CLASS("capwap", OVS_VPORT_TYPE_CAPWAP), + TUNNEL_CLASS("vxlan", OVS_VPORT_TYPE_VXLAN), + + { OVS_VPORT_TYPE_PATCH, + { "patch", VPORT_FUNCTIONS(NULL, NULL) }, + parse_patch_config, unparse_patch_config } + }; + + int i; + + for (i = 0; i < ARRAY_SIZE(vport_classes); i++) { + netdev_register_provider(&vport_classes[i].netdev_class); + } }