X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev-vport.c;h=732fb2a3d644c497a43622dfa6e9997b1c3ce57f;hb=c19e653509deb3f1002c4ed99714e16549ec74e9;hp=b006a677b4224f3c719e8338d15996929b9a2635;hpb=6333182946ffd368eb7623d7408185a521058b46;p=sliver-openvswitch.git diff --git a/lib/netdev-vport.c b/lib/netdev-vport.c index b006a677b..732fb2a3d 100644 --- a/lib/netdev-vport.c +++ b/lib/netdev-vport.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2010 Nicira Networks. + * Copyright (c) 2010, 2011 Nicira Networks. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -26,6 +26,7 @@ #include #include "byte-order.h" +#include "dpif-linux.h" #include "hash.h" #include "hmap.h" #include "list.h" @@ -36,8 +37,8 @@ #include "openvswitch/datapath-protocol.h" #include "openvswitch/tunnel.h" #include "packets.h" +#include "route-table.h" #include "rtnetlink.h" -#include "rtnetlink-route.h" #include "rtnetlink-link.h" #include "shash.h" #include "socket-util.h" @@ -46,16 +47,7 @@ VLOG_DEFINE_THIS_MODULE(netdev_vport); static struct hmap name_map; -static struct hmap route_map; static struct rtnetlink_notifier netdev_vport_link_notifier; -static struct rtnetlink_notifier netdev_vport_route_notifier; - -struct route_node { - struct hmap_node node; /* Node in route_map. */ - int rta_oif; /* Egress interface index. */ - uint32_t rta_dst; /* Destination address in host byte order. */ - unsigned char rtm_dst_len; /* Destination address length. */ -}; struct name_node { struct hmap_node node; /* Node in name_map. */ @@ -72,7 +64,7 @@ struct netdev_vport_notifier { struct netdev_dev_vport { struct netdev_dev netdev_dev; - uint64_t config[VPORT_CONFIG_SIZE / 8]; + struct ofpbuf *options; }; struct netdev_vport { @@ -80,9 +72,13 @@ struct netdev_vport { }; struct vport_class { + enum odp_vport_type type; struct netdev_class netdev_class; - int (*parse_config)(const struct netdev_dev *, const struct shash *args, - void *config); + int (*parse_config)(const char *name, const char *type, + const struct shash *args, struct ofpbuf *options); + int (*unparse_config)(const char *name, const char *type, + const struct nlattr *options, size_t options_len, + struct shash *args); }; static struct shash netdev_vport_notifiers = @@ -90,14 +86,14 @@ static struct shash 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 **); 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[ODP_TUNNEL_ATTR_MAX + 1]); static void netdev_vport_tnl_iface_init(void); -static void netdev_vport_route_change(const struct rtnetlink_route_change *, - void *); static void netdev_vport_link_change(const struct rtnetlink_link_change *, void *); static const char *netdev_vport_get_tnl_iface(const struct netdev *netdev); @@ -130,23 +126,74 @@ netdev_vport_cast(const struct netdev *netdev) return CONTAINER_OF(netdev, struct netdev_vport, netdev); } -/* 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) +/* If 'netdev' is a vport netdev, returns an ofpbuf that contains Netlink + * options to include in ODP_VPORT_ATTR_OPTIONS for configuring that vport. + * Otherwise returns NULL. */ +const struct ofpbuf * +netdev_vport_get_options(const struct netdev *netdev) { const struct netdev_dev *dev = netdev_get_dev(netdev); - 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 (is_vport_class(netdev_dev_get_class(dev)) + ? netdev_dev_vport_cast(dev)->options + : NULL); +} + +enum odp_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 ? ODP_VPORT_TYPE_INTERNAL + : class == &netdev_linux_class ? ODP_VPORT_TYPE_NETDEV + : ODP_VPORT_TYPE_UNSPEC); +} + +const char * +netdev_vport_get_netdev_type(const struct dpif_linux_vport *vport) +{ + struct nlattr *a[ODP_TUNNEL_ATTR_MAX + 1]; + + switch (vport->type) { + case ODP_VPORT_TYPE_UNSPEC: + break; + + case ODP_VPORT_TYPE_NETDEV: + return "system"; + + case ODP_VPORT_TYPE_INTERNAL: + return "internal"; + + case ODP_VPORT_TYPE_PATCH: + return "patch"; + + case ODP_VPORT_TYPE_GRE: + if (tnl_port_config_from_nlattr(vport->options, vport->options_len, + a)) { + break; + } + return (nl_attr_get_u32(a[ODP_TUNNEL_ATTR_FLAGS]) & TNL_F_IPSEC + ? "ipsec_gre" : "gre"); + + case ODP_VPORT_TYPE_CAPWAP: + return "capwap"; + + case __ODP_VPORT_TYPE_MAX: + break; } + + VLOG_WARN_RL(&rl, "dp%d: port `%s' has unsupported type %u", + vport->dp_idx, vport->name, (unsigned int) vport->type); + return "unknown"; } static int netdev_vport_init(void) { netdev_vport_tnl_iface_init(); + route_table_register(); return 0; } @@ -156,19 +203,62 @@ netdev_vport_create(const struct netdev_class *netdev_class, const char *name, struct netdev_dev **netdev_devp) { const struct vport_class *vport_class = vport_class_cast(netdev_class); - struct netdev_dev_vport *dev; + struct ofpbuf *options = NULL; + struct shash fetched_args; int error; - dev = xmalloc(sizeof *dev); - *netdev_devp = &dev->netdev_dev; - netdev_dev_init(&dev->netdev_dev, name, netdev_class); + shash_init(&fetched_args); + + if (!shash_is_empty(args)) { + /* Parse the provided configuration. */ + options = ofpbuf_new(64); + error = vport_class->parse_config(name, netdev_class->type, + args, options); + } else { + /* Fetch an existing configuration from the kernel. + * + * This case could be ambiguous with initializing a new vport with an + * empty configuration, but none of the existing vport classes accept + * an empty configuration. */ + struct dpif_linux_vport reply; + struct ofpbuf *buf; + + error = dpif_linux_vport_get(name, &reply, &buf); + if (!error) { + /* XXX verify correct type */ + error = vport_class->unparse_config(name, netdev_class->type, + reply.options, + reply.options_len, + &fetched_args); + if (error) { + VLOG_ERR_RL(&rl, "%s: failed to parse kernel config (%s)", + name, strerror(error)); + } else { + options = ofpbuf_clone_data(reply.options, reply.options_len); + } + ofpbuf_delete(buf); + } else { + VLOG_ERR_RL(&rl, "%s: vport query failed (%s)", + name, strerror(error)); + } + } + + if (!error) { + struct netdev_dev_vport *dev; - memset(dev->config, 0, sizeof dev->config); - error = vport_class->parse_config(&dev->netdev_dev, args, dev->config); + dev = xmalloc(sizeof *dev); + netdev_dev_init(&dev->netdev_dev, name, + shash_is_empty(&fetched_args) ? args : &fetched_args, + netdev_class); + dev->options = options; - if (error) { - netdev_dev_uninit(&dev->netdev_dev, true); + *netdev_devp = &dev->netdev_dev; + } else { + ofpbuf_delete(options); } + + shash_destroy(&fetched_args); + return error; } @@ -177,6 +267,7 @@ netdev_vport_destroy(struct netdev_dev *netdev_dev_) { struct netdev_dev_vport *netdev_dev = netdev_dev_vport_cast(netdev_dev_); + route_table_unregister(); free(netdev_dev); } @@ -201,28 +292,41 @@ netdev_vport_close(struct netdev *netdev_) } static int -netdev_vport_reconfigure(struct netdev_dev *dev_, - const struct shash *args) +netdev_vport_set_config(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; + const char *name = netdev_dev_get_name(dev_); + struct ofpbuf *options; 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); + options = ofpbuf_new(64); + error = vport_class->parse_config(name, netdev_dev_get_type(dev_), + args, options); + if (!error + && (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 = ODP_VPORT_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()). */ - memcpy(dev->config, port.config, sizeof dev->config); + ofpbuf_delete(dev->options); + dev->options = options; + options = NULL; + error = 0; } } + ofpbuf_delete(options); + return error; } @@ -230,94 +334,94 @@ static int 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); + struct dpif_linux_vport vport; + int error; - memcpy(vport_ether.ether_addr, mac, ETH_ADDR_LEN); + dpif_linux_vport_init(&vport); + vport.cmd = ODP_VPORT_SET; + vport.name = netdev_get_name(netdev); + vport.address = mac; - err = netdev_vport_do_ioctl(ODP_VPORT_ETHER_SET, &vport_ether); - if (err) { - return err; + error = dpif_linux_vport_transact(&vport, NULL, NULL); + if (!error) { + netdev_vport_poll_notify(netdev); } - - netdev_vport_poll_notify(netdev); - return 0; + return error; } 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); + struct dpif_linux_vport reply; + struct ofpbuf *buf; + int error; - err = netdev_vport_do_ioctl(ODP_VPORT_ETHER_GET, &vport_ether); - if (err) { - return err; + error = dpif_linux_vport_get(netdev_get_name(netdev), &reply, &buf); + if (!error) { + if (reply.address) { + memcpy(mac, reply.address, ETH_ADDR_LEN); + } else { + error = EOPNOTSUPP; + } + ofpbuf_delete(buf); } - - memcpy(mac, vport_ether.ether_addr, ETH_ADDR_LEN); - return 0; + return error; } static int netdev_vport_get_mtu(const struct netdev *netdev, int *mtup) { - struct odp_vport_mtu vport_mtu; - int err; - - ovs_strlcpy(vport_mtu.devname, netdev_get_name(netdev), - sizeof vport_mtu.devname); + 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) { + *mtup = reply.mtu; + ofpbuf_delete(buf); } - - *mtup = vport_mtu.mtu; - return 0; + return error; } 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; + struct dpif_linux_vport reply; + struct ofpbuf *buf; + int error; - 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; + error = dpif_linux_vport_get(netdev_get_name(netdev), &reply, &buf); + if (error) { + return error; + } else if (!reply.stats) { + ofpbuf_delete(buf); + return EOPNOTSUPP; + } + + stats->rx_packets = reply.stats->rx_packets; + stats->tx_packets = reply.stats->tx_packets; + stats->rx_bytes = reply.stats->rx_bytes; + stats->tx_bytes = reply.stats->tx_bytes; + stats->rx_errors = reply.stats->rx_errors; + stats->tx_errors = reply.stats->tx_errors; + stats->rx_dropped = reply.stats->rx_dropped; + stats->tx_dropped = reply.stats->tx_dropped; + stats->multicast = reply.stats->multicast; + stats->collisions = reply.stats->collisions; + stats->rx_length_errors = reply.stats->rx_length_errors; + stats->rx_over_errors = reply.stats->rx_over_errors; + stats->rx_crc_errors = reply.stats->rx_crc_errors; + stats->rx_frame_errors = reply.stats->rx_frame_errors; + stats->rx_fifo_errors = reply.stats->rx_fifo_errors; + stats->rx_missed_errors = reply.stats->rx_missed_errors; + stats->tx_aborted_errors = reply.stats->tx_aborted_errors; + stats->tx_carrier_errors = reply.stats->tx_carrier_errors; + stats->tx_fifo_errors = reply.stats->tx_fifo_errors; + stats->tx_heartbeat_errors = reply.stats->tx_heartbeat_errors; + stats->tx_window_errors = reply.stats->tx_window_errors; + + ofpbuf_delete(buf); return 0; } @@ -325,34 +429,38 @@ netdev_vport_get_stats(const struct netdev *netdev, struct netdev_stats *stats) int netdev_vport_set_stats(struct netdev *netdev, const struct netdev_stats *stats) { - struct odp_vport_stats_req ovsr; + struct rtnl_link_stats64 rtnl_stats; + struct dpif_linux_vport vport; 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); + rtnl_stats.rx_packets = stats->rx_packets; + rtnl_stats.tx_packets = stats->tx_packets; + rtnl_stats.rx_bytes = stats->rx_bytes; + rtnl_stats.tx_bytes = stats->tx_bytes; + rtnl_stats.rx_errors = stats->rx_errors; + rtnl_stats.tx_errors = stats->tx_errors; + rtnl_stats.rx_dropped = stats->rx_dropped; + rtnl_stats.tx_dropped = stats->tx_dropped; + rtnl_stats.multicast = stats->multicast; + rtnl_stats.collisions = stats->collisions; + rtnl_stats.rx_length_errors = stats->rx_length_errors; + rtnl_stats.rx_over_errors = stats->rx_over_errors; + rtnl_stats.rx_crc_errors = stats->rx_crc_errors; + rtnl_stats.rx_frame_errors = stats->rx_frame_errors; + rtnl_stats.rx_fifo_errors = stats->rx_fifo_errors; + rtnl_stats.rx_missed_errors = stats->rx_missed_errors; + rtnl_stats.tx_aborted_errors = stats->tx_aborted_errors; + rtnl_stats.tx_carrier_errors = stats->tx_carrier_errors; + rtnl_stats.tx_fifo_errors = stats->tx_fifo_errors; + rtnl_stats.tx_heartbeat_errors = stats->tx_heartbeat_errors; + rtnl_stats.tx_window_errors = stats->tx_window_errors; + + dpif_linux_vport_init(&vport); + vport.cmd = ODP_VPORT_SET; + vport.name = netdev_get_name(netdev); + vport.stats = &rtnl_stats; + + err = dpif_linux_vport_transact(&vport, NULL, NULL); /* 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 @@ -371,7 +479,16 @@ netdev_vport_get_status(const struct netdev *netdev, struct shash *sh) const char *iface = netdev_vport_get_tnl_iface(netdev); if (iface) { + struct netdev *egress_netdev; + shash_add(sh, "tunnel_egress_iface", xstrdup(iface)); + + if (!netdev_open_default(iface, &egress_netdev)) { + shash_add(sh, "tunnel_egress_iface_carrier", + xstrdup(netdev_get_carrier(egress_netdev) + ? "up" : "down")); + netdev_close(egress_netdev); + } } return 0; @@ -447,14 +564,14 @@ static void netdev_vport_run(void) { rtnetlink_link_notifier_run(); - rtnetlink_route_notifier_run(); + route_table_run(); } static void netdev_vport_wait(void) { rtnetlink_link_notifier_wait(); - rtnetlink_route_notifier_wait(); + route_table_wait(); } /* get_tnl_iface() implementation. */ @@ -473,55 +590,23 @@ name_node_lookup(int ifi_index) return NULL; } -static struct route_node * -route_node_lookup(int rta_oif, uint32_t rta_dst, unsigned char rtm_dst_len) -{ - uint32_t hash; - struct route_node *rn; - - hash = hash_3words(rta_oif, rta_dst, rtm_dst_len); - HMAP_FOR_EACH_WITH_HASH(rn, node, hash, &route_map) { - if (rn->rta_oif == rn->rta_oif && - rn->rta_dst == rn->rta_dst && - rn->rtm_dst_len == rn->rtm_dst_len) { - return rn; - } - } - - return NULL; -} - -/* Resets the name or route map depending on the value of 'is_name'. Clears - * the appropriate map, makes an rtnetlink dump request, and calls the change - * callback for each reply from the kernel. One should probably use - * netdev_vport_reset_routes or netdev_vport_reset_names instead. */ +/* Queries the kernel for fresh data to populate the name map with. */ static int -netdev_vport_reset_name_else_route(bool is_name) +netdev_vport_reset_names(void) { int error; - int nlmsg_type; struct nl_dump dump; struct rtgenmsg *rtmsg; struct ofpbuf request, reply; static struct nl_sock *rtnl_sock; + struct name_node *nn, *nn_next; - if (is_name) { - struct name_node *nn, *nn_next; - - HMAP_FOR_EACH_SAFE(nn, nn_next, node, &name_map) { - hmap_remove(&name_map, &nn->node); - free(nn); - } - } else { - struct route_node *rn, *rn_next; - - HMAP_FOR_EACH_SAFE(rn, rn_next, node, &route_map) { - hmap_remove(&route_map, &rn->node); - free(rn); - } + HMAP_FOR_EACH_SAFE(nn, nn_next, node, &name_map) { + hmap_remove(&name_map, &nn->node); + free(nn); } - error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock); + error = nl_sock_create(NETLINK_ROUTE, &rtnl_sock); if (error) { VLOG_WARN_RL(&rl, "Failed to create NETLINK_ROUTE socket"); return error; @@ -529,8 +614,7 @@ netdev_vport_reset_name_else_route(bool is_name) ofpbuf_init(&request, 0); - nlmsg_type = is_name ? RTM_GETLINK : RTM_GETROUTE; - nl_msg_put_nlmsghdr(&request, sizeof *rtmsg, nlmsg_type, NLM_F_REQUEST); + nl_msg_put_nlmsghdr(&request, sizeof *rtmsg, RTM_GETLINK, NLM_F_REQUEST); rtmsg = ofpbuf_put_zeros(&request, sizeof *rtmsg); rtmsg->rtgen_family = AF_INET; @@ -538,76 +622,15 @@ netdev_vport_reset_name_else_route(bool is_name) nl_dump_start(&dump, rtnl_sock, &request); while (nl_dump_next(&dump, &reply)) { - if (is_name) { - struct rtnetlink_link_change change; - - if (rtnetlink_link_parse(&reply, &change)) { - netdev_vport_link_change(&change, NULL); - } - } else { - struct rtnetlink_route_change change; + struct rtnetlink_link_change change; - if (rtnetlink_route_parse(&reply, &change)) { - netdev_vport_route_change(&change, NULL); - } + if (rtnetlink_link_parse(&reply, &change)) { + netdev_vport_link_change(&change, NULL); } } - - error = nl_dump_done(&dump); nl_sock_destroy(rtnl_sock); - return error; -} - -static int -netdev_vport_reset_routes(void) -{ - return netdev_vport_reset_name_else_route(false); -} - -static int -netdev_vport_reset_names(void) -{ - return netdev_vport_reset_name_else_route(true); -} - -static void -netdev_vport_route_change(const struct rtnetlink_route_change *change, - void *aux OVS_UNUSED) -{ - - if (!change) { - netdev_vport_reset_routes(); - } else if (change->nlmsg_type == RTM_NEWROUTE) { - uint32_t hash; - struct route_node *rn; - - if (route_node_lookup(change->rta_oif, change->rta_dst, - change->rtm_dst_len)) { - return; - } - - rn = xzalloc(sizeof *rn); - rn->rta_oif = change->rta_oif; - rn->rta_dst = change->rta_dst; - rn->rtm_dst_len = change->rtm_dst_len; - - hash = hash_3words(rn->rta_oif, rn->rta_dst, rn->rtm_dst_len); - hmap_insert(&route_map, &rn->node, hash); - } else if (change->nlmsg_type == RTM_DELROUTE) { - struct route_node *rn; - - rn = route_node_lookup(change->rta_oif, change->rta_dst, - change->rtm_dst_len); - - if (rn) { - hmap_remove(&route_map, &rn->node); - free(rn); - } - } else { - VLOG_WARN_RL(&rl, "Received unexpected rtnetlink message type %d", - change->nlmsg_type); - } + return nl_dump_done(&dump); } static void @@ -641,10 +664,6 @@ netdev_vport_link_change(const struct rtnetlink_link_change *change, free(nn); } - /* Link deletions do not result in all of the RTM_DELROUTE messages one - * would expect. For now, go ahead and reset route_map whenever a link - * is deleted. */ - netdev_vport_reset_routes(); } else { VLOG_WARN_RL(&rl, "Received unexpected rtnetlink message type %d", change->nlmsg_type); @@ -658,16 +677,11 @@ netdev_vport_tnl_iface_init(void) if (!tnl_iface_is_init) { hmap_init(&name_map); - hmap_init(&route_map); rtnetlink_link_notifier_register(&netdev_vport_link_notifier, - netdev_vport_link_change, NULL); - - rtnetlink_route_notifier_register(&netdev_vport_route_notifier, - netdev_vport_route_change, NULL); + netdev_vport_link_change, NULL); netdev_vport_reset_names(); - netdev_vport_reset_routes(); tnl_iface_is_init = true; } } @@ -675,44 +689,22 @@ netdev_vport_tnl_iface_init(void) static const char * netdev_vport_get_tnl_iface(const struct netdev *netdev) { - int dst_len; + struct nlattr *a[ODP_TUNNEL_ATTR_MAX + 1]; + int ifindex; uint32_t route; struct netdev_dev_vport *ndv; - struct tnl_port_config *config; - struct route_node *rn, *rn_def, *rn_iter; ndv = netdev_dev_vport_cast(netdev_get_dev(netdev)); - config = (struct tnl_port_config *) ndv->config; - route = ntohl(config->daddr); - - dst_len = 0; - rn = NULL; - rn_def = NULL; - - HMAP_FOR_EACH(rn_iter, node, &route_map) { - if (rn_iter->rtm_dst_len == 0 && rn_iter->rta_dst == 0) { - /* Default route. */ - rn_def = rn_iter; - } else if (rn_iter->rtm_dst_len > dst_len) { - uint32_t mask = 0xffffffff << (32 - rn_iter->rtm_dst_len); - if ((route & mask) == (rn_iter->rta_dst & mask)) { - rn = rn_iter; - dst_len = rn_iter->rtm_dst_len; - } - } - } - - if (!rn) { - rn = rn_def; + if (tnl_port_config_from_nlattr(ndv->options->data, ndv->options->size, + a)) { + return NULL; } + route = nl_attr_get_be32(a[ODP_TUNNEL_ATTR_DST_IPV4]); - if (rn) { - uint32_t hash; + if (route_table_get_ifindex(route, &ifindex)) { struct name_node *nn; - - hash = hash_int(rn->rta_oif, 0); - HMAP_FOR_EACH_WITH_HASH(nn, node, hash, &name_map) { - if (nn->ifi_index == rn->rta_oif) { + HMAP_FOR_EACH_WITH_HASH(nn, node, hash_int(ifindex, 0), &name_map) { + if (nn->ifi_index == ifindex) { return nn->ifname; } } @@ -723,22 +715,6 @@ netdev_vport_get_tnl_iface(const struct netdev *netdev) /* Helper functions. */ -static int -netdev_vport_do_ioctl(int cmd, void *arg) -{ - static int ioctl_fd = -1; - - 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; - } - } - - return ioctl(ioctl_fd, cmd, arg) ? errno : 0; -} - static void netdev_vport_poll_notify(const struct netdev *netdev) { @@ -760,32 +736,46 @@ netdev_vport_poll_notify(const struct netdev *netdev) /* Code specific to individual vport types. */ +static void +set_key(const struct shash *args, const char *name, uint16_t type, + struct ofpbuf *options) +{ + const char *s; + + s = shash_find_data(args, name); + if (!s) { + s = shash_find_data(args, "key"); + if (!s) { + s = "0"; + } + } + + if (!strcmp(s, "flow")) { + /* This is the default if no attribute is present. */ + } else { + nl_msg_put_be64(options, type, htonll(strtoull(s, NULL, 0))); + } +} + static int -parse_tunnel_config(const struct netdev_dev *dev, const struct shash *args, - void *configp) +parse_tunnel_config(const char *name, const char *type, + const struct shash *args, struct ofpbuf *options) { - const char *name = netdev_dev_get_name(dev); - const char *type = netdev_dev_get_type(dev); bool is_gre = false; bool is_ipsec = false; - struct tnl_port_config config; struct shash_node *node; bool ipsec_mech_set = false; + ovs_be32 daddr = htonl(0); + uint32_t flags; - memset(&config, 0, sizeof config); - config.flags |= TNL_F_PMTUD; - config.flags |= TNL_F_HDR_CACHE; - + flags = TNL_F_PMTUD | TNL_F_HDR_CACHE; if (!strcmp(type, "gre")) { is_gre = true; } else if (!strcmp(type, "ipsec_gre")) { is_gre = true; is_ipsec = true; - - config.flags |= TNL_F_IPSEC; - - /* IPsec doesn't work when header caching is enabled. */ - config.flags &= ~TNL_F_HDR_CACHE; + flags |= TNL_F_IPSEC; + flags &= ~TNL_F_HDR_CACHE; } SHASH_FOR_EACH (node, args) { @@ -794,58 +784,39 @@ parse_tunnel_config(const struct netdev_dev *dev, const struct shash *args, if (lookup_ip(node->data, &in_addr)) { VLOG_WARN("%s: bad %s 'remote_ip'", name, type); } else { - config.daddr = in_addr.s_addr; + daddr = in_addr.s_addr; } } else if (!strcmp(node->name, "local_ip")) { struct in_addr in_addr; if (lookup_ip(node->data, &in_addr)) { VLOG_WARN("%s: bad %s 'local_ip'", name, type); } else { - config.saddr = 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 { - uint64_t key = strtoull(node->data, NULL, 0); - config.out_key = config.in_key = htonll(key); - } - } else if (!strcmp(node->name, "in_key") && is_gre) { - if (!strcmp(node->data, "flow")) { - config.flags |= TNL_F_IN_KEY_MATCH; - } else { - config.in_key = htonll(strtoull(node->data, NULL, 0)); - } - } 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)); + nl_msg_put_be32(options, ODP_TUNNEL_ATTR_SRC_IPV4, + in_addr.s_addr); } } else if (!strcmp(node->name, "tos")) { if (!strcmp(node->data, "inherit")) { - config.flags |= TNL_F_TOS_INHERIT; + flags |= TNL_F_TOS_INHERIT; } else { - config.tos = atoi(node->data); + nl_msg_put_u8(options, ODP_TUNNEL_ATTR_TOS, atoi(node->data)); } } else if (!strcmp(node->name, "ttl")) { if (!strcmp(node->data, "inherit")) { - config.flags |= TNL_F_TTL_INHERIT; + flags |= TNL_F_TTL_INHERIT; } else { - config.ttl = atoi(node->data); + nl_msg_put_u8(options, ODP_TUNNEL_ATTR_TTL, atoi(node->data)); } } else if (!strcmp(node->name, "csum") && is_gre) { if (!strcmp(node->data, "true")) { - config.flags |= TNL_F_CSUM; + flags |= TNL_F_CSUM; } } else if (!strcmp(node->name, "pmtud")) { if (!strcmp(node->data, "false")) { - config.flags &= ~TNL_F_PMTUD; + flags &= ~TNL_F_PMTUD; } } else if (!strcmp(node->name, "header_cache")) { if (!strcmp(node->data, "false")) { - config.flags &= ~TNL_F_HDR_CACHE; + flags &= ~TNL_F_HDR_CACHE; } } else if (!strcmp(node->name, "peer_cert") && is_ipsec) { if (shash_find(args, "certificate")) { @@ -874,9 +845,12 @@ parse_tunnel_config(const struct netdev_dev *dev, const struct shash *args, || !strcmp(node->name, "private_key") || !strcmp(node->name, "use_ssl_cert"))) { /* Ignore options not used by the netdev. */ + } else if (is_gre && (!strcmp(node->name, "key") && + !strcmp(node->name, "in_key") && + !strcmp(node->name, "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->name); } } @@ -893,22 +867,132 @@ parse_tunnel_config(const struct netdev_dev *dev, const struct shash *args, } } - if (!config.daddr) { + if (is_gre) { + set_key(args, "in_key", ODP_TUNNEL_ATTR_IN_KEY, options); + set_key(args, "out_key", ODP_TUNNEL_ATTR_OUT_KEY, options); + } + + if (!daddr) { VLOG_WARN("%s: %s type requires valid 'remote_ip' argument", name, type); return EINVAL; } + nl_msg_put_be32(options, ODP_TUNNEL_ATTR_DST_IPV4, daddr); + + nl_msg_put_u32(options, ODP_TUNNEL_ATTR_FLAGS, flags); - 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) +tnl_port_config_from_nlattr(const struct nlattr *options, size_t options_len, + struct nlattr *a[ODP_TUNNEL_ATTR_MAX + 1]) +{ + static const struct nl_policy odp_tunnel_policy[] = { + [ODP_TUNNEL_ATTR_FLAGS] = { .type = NL_A_U32 }, + [ODP_TUNNEL_ATTR_DST_IPV4] = { .type = NL_A_BE32 }, + [ODP_TUNNEL_ATTR_SRC_IPV4] = { .type = NL_A_BE32, .optional = true }, + [ODP_TUNNEL_ATTR_IN_KEY] = { .type = NL_A_BE64, .optional = true }, + [ODP_TUNNEL_ATTR_OUT_KEY] = { .type = NL_A_BE64, .optional = true }, + [ODP_TUNNEL_ATTR_TOS] = { .type = NL_A_U8, .optional = true }, + [ODP_TUNNEL_ATTR_TTL] = { .type = NL_A_U8, .optional = true }, + }; + struct ofpbuf buf; + + ofpbuf_use_const(&buf, options, options_len); + if (!nl_policy_parse(&buf, 0, odp_tunnel_policy, + a, ARRAY_SIZE(odp_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 shash *args) +{ + struct nlattr *a[ODP_TUNNEL_ATTR_MAX + 1]; + ovs_be32 daddr; + uint32_t flags; + int error; + + error = tnl_port_config_from_nlattr(options, options_len, a); + if (error) { + return error; + } + + flags = nl_attr_get_u32(a[ODP_TUNNEL_ATTR_FLAGS]); + if (!(flags & TNL_F_HDR_CACHE) == !(flags & TNL_F_IPSEC)) { + smap_add(args, "header_cache", + flags & TNL_F_HDR_CACHE ? "true" : "false"); + } + + daddr = nl_attr_get_be32(a[ODP_TUNNEL_ATTR_DST_IPV4]); + shash_add(args, "remote_ip", xasprintf(IP_FMT, IP_ARGS(&daddr))); + + if (a[ODP_TUNNEL_ATTR_SRC_IPV4]) { + ovs_be32 saddr = nl_attr_get_be32(a[ODP_TUNNEL_ATTR_SRC_IPV4]); + shash_add(args, "local_ip", xasprintf(IP_FMT, IP_ARGS(&saddr))); + } + + if (!a[ODP_TUNNEL_ATTR_IN_KEY] && !a[ODP_TUNNEL_ATTR_OUT_KEY]) { + smap_add(args, "key", "flow"); + } else { + uint64_t in_key = get_be64_or_zero(a[ODP_TUNNEL_ATTR_IN_KEY]); + uint64_t out_key = get_be64_or_zero(a[ODP_TUNNEL_ATTR_OUT_KEY]); + + if (in_key && in_key == out_key) { + shash_add(args, "key", xasprintf("%"PRIu64, in_key)); + } else { + if (!a[ODP_TUNNEL_ATTR_IN_KEY]) { + smap_add(args, "in_key", "flow"); + } else if (in_key) { + shash_add(args, "in_key", xasprintf("%"PRIu64, in_key)); + } + + if (!a[ODP_TUNNEL_ATTR_OUT_KEY]) { + smap_add(args, "out_key", "flow"); + } else if (out_key) { + shash_add(args, "out_key", xasprintf("%"PRIu64, out_key)); + } + } + } + + if (flags & TNL_F_TTL_INHERIT) { + smap_add(args, "tos", "inherit"); + } else if (a[ODP_TUNNEL_ATTR_TTL]) { + int ttl = nl_attr_get_u8(a[ODP_TUNNEL_ATTR_TTL]); + shash_add(args, "tos", xasprintf("%d", ttl)); + } + + if (flags & TNL_F_TOS_INHERIT) { + smap_add(args, "tos", "inherit"); + } else if (a[ODP_TUNNEL_ATTR_TOS]) { + int tos = nl_attr_get_u8(a[ODP_TUNNEL_ATTR_TOS]); + shash_add(args, "tos", xasprintf("%d", tos)); + } + + if (flags & TNL_F_CSUM) { + smap_add(args, "csum", "true"); + } + if (!(flags & TNL_F_PMTUD)) { + smap_add(args, "pmtud", "false"); + } + + return 0; +} + +static int +parse_patch_config(const char *name, const char *type OVS_UNUSED, + const struct shash *args, struct ofpbuf *options) { - const char *name = netdev_dev_get_name(dev); const char *peer; peer = shash_find_data(args, "peer"); @@ -922,7 +1006,7 @@ parse_patch_config(const struct netdev_dev *dev, const struct shash *args, return EINVAL; } - if (strlen(peer) >= MIN(IFNAMSIZ, VPORT_CONFIG_SIZE)) { + if (strlen(peer) >= IFNAMSIZ) { VLOG_WARN("%s: patch 'peer' arg too long", name); return EINVAL; } @@ -932,10 +1016,34 @@ parse_patch_config(const struct netdev_dev *dev, const struct shash *args, return EINVAL; } - strncpy(configp, peer, VPORT_CONFIG_SIZE); + nl_msg_put_string(options, ODP_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 shash *args) +{ + static const struct nl_policy odp_patch_policy[] = { + [ODP_PATCH_ATTR_PEER] = { .type = NL_A_STRING, + .max_len = IFNAMSIZ, + .optional = false } + }; + + struct nlattr *a[ARRAY_SIZE(odp_patch_policy)]; + struct ofpbuf buf; + + ofpbuf_use_const(&buf, options, options_len); + if (!nl_policy_parse(&buf, 0, odp_patch_policy, + a, ARRAY_SIZE(odp_patch_policy))) { + return EINVAL; + } + + smap_add(args, "peer", nl_attr_get_string(a[ODP_PATCH_ATTR_PEER])); + return 0; +} #define VPORT_FUNCTIONS(GET_STATUS) \ netdev_vport_init, \ @@ -944,7 +1052,7 @@ parse_patch_config(const struct netdev_dev *dev, const struct shash *args, \ netdev_vport_create, \ netdev_vport_destroy, \ - netdev_vport_reconfigure, \ + netdev_vport_set_config, \ \ netdev_vport_open, \ netdev_vport_close, \ @@ -1000,13 +1108,21 @@ void netdev_vport_register(void) { static const struct vport_class vport_classes[] = { - { { "gre", VPORT_FUNCTIONS(netdev_vport_get_status) }, - parse_tunnel_config }, - { { "ipsec_gre", VPORT_FUNCTIONS(netdev_vport_get_status) }, - parse_tunnel_config }, - { { "capwap", VPORT_FUNCTIONS(netdev_vport_get_status) }, - parse_tunnel_config }, - { { "patch", VPORT_FUNCTIONS(NULL) }, parse_patch_config } + { ODP_VPORT_TYPE_GRE, + { "gre", VPORT_FUNCTIONS(netdev_vport_get_status) }, + parse_tunnel_config, unparse_tunnel_config }, + + { ODP_VPORT_TYPE_GRE, + { "ipsec_gre", VPORT_FUNCTIONS(netdev_vport_get_status) }, + parse_tunnel_config, unparse_tunnel_config }, + + { ODP_VPORT_TYPE_CAPWAP, + { "capwap", VPORT_FUNCTIONS(netdev_vport_get_status) }, + parse_tunnel_config, unparse_tunnel_config }, + + { ODP_VPORT_TYPE_PATCH, + { "patch", VPORT_FUNCTIONS(NULL) }, + parse_patch_config, unparse_patch_config } }; int i;