/*
- * 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.
#include <errno.h>
#include <fcntl.h>
+#include <sys/socket.h>
+#include <linux/rtnetlink.h>
#include <net/if.h>
#include <sys/ioctl.h>
+#include "byte-order.h"
+#include "hash.h"
+#include "hmap.h"
#include "list.h"
#include "netdev-provider.h"
+#include "netlink.h"
+#include "netlink-socket.h"
+#include "ofpbuf.h"
#include "openvswitch/datapath-protocol.h"
#include "openvswitch/tunnel.h"
#include "packets.h"
+#include "rtnetlink.h"
+#include "route-table.h"
+#include "rtnetlink-link.h"
#include "shash.h"
#include "socket-util.h"
#include "vlog.h"
VLOG_DEFINE_THIS_MODULE(netdev_vport);
+static struct hmap name_map;
+static struct rtnetlink_notifier netdev_vport_link_notifier;
+
+struct name_node {
+ struct hmap_node node; /* Node in name_map. */
+ uint32_t ifi_index; /* Kernel interface index. */
+
+ char ifname[IFNAMSIZ]; /* Interface name. */
+};
+
struct netdev_vport_notifier {
struct netdev_notifier notifier;
struct list list_node;
struct netdev_dev_vport {
struct netdev_dev netdev_dev;
+ uint64_t config[VPORT_CONFIG_SIZE / 8];
};
struct netdev_vport {
struct netdev netdev;
};
-struct vport_info {
- const char *devname;
- const char *type;
- void *config;
-};
-
struct vport_class {
- const struct netdev_class netdev_class;
- int (*parse_config)(struct vport_info *port, const struct shash *args);
+ struct netdev_class netdev_class;
+ int (*parse_config)(const struct netdev_dev *, const struct shash *args,
+ void *config);
};
static struct shash netdev_vport_notifiers =
const struct shash *, struct netdev_dev **);
static void netdev_vport_poll_notify(const struct netdev *);
+static void netdev_vport_tnl_iface_init(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);
+
static bool
is_vport_class(const struct netdev_class *class)
{
return CONTAINER_OF(netdev, struct netdev_vport, 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;
+/* 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)
+{
+ 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);
}
}
static int
-netdev_vport_create(const struct netdev_class *class, const char *name,
- const struct shash *args, struct netdev_dev **netdev_devp)
+netdev_vport_init(void)
{
- int err;
- struct odp_vport_add ova;
- struct netdev_dev_vport *netdev_dev;
-
- 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;
- }
+ netdev_vport_tnl_iface_init();
+ route_table_register();
+ return 0;
+}
- err = netdev_vport_do_ioctl(ODP_VPORT_ADD, &ova);
+static int
+netdev_vport_create(const struct netdev_class *netdev_class, const char *name,
+ const struct shash *args,
+ struct netdev_dev **netdev_devp)
+{
+ const struct vport_class *vport_class = vport_class_cast(netdev_class);
+ struct netdev_dev_vport *dev;
+ int error;
- if (err == EBUSY) {
- VLOG_WARN("%s: destroying existing device", name);
+ dev = xmalloc(sizeof *dev);
+ *netdev_devp = &dev->netdev_dev;
+ netdev_dev_init(&dev->netdev_dev, name, netdev_class);
- err = netdev_vport_do_ioctl(ODP_VPORT_DEL, ova.devname);
- if (err) {
- goto exit;
- }
+ memset(dev->config, 0, sizeof dev->config);
+ error = vport_class->parse_config(&dev->netdev_dev, args, dev->config);
- err = netdev_vport_do_ioctl(ODP_VPORT_ADD, &ova);
- }
- if (err) {
- goto exit;
+ if (error) {
+ netdev_dev_uninit(&dev->netdev_dev, true);
}
-
- netdev_dev = xmalloc(sizeof *netdev_dev);
- netdev_dev_init(&netdev_dev->netdev_dev, name, class);
-
- *netdev_devp = &netdev_dev->netdev_dev;
-
-exit:
- free(ova.config);
- return err;
+ return error;
}
static void
{
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_));
+ route_table_unregister();
free(netdev_dev);
}
}
static int
-netdev_vport_reconfigure(struct netdev_dev *netdev_dev,
+netdev_vport_reconfigure(struct netdev_dev *dev_,
const struct shash *args)
{
- const char *name = netdev_dev_get_name(netdev_dev);
- struct odp_vport_mod ovm;
- int err;
-
- 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;
+ 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);
+ }
}
-
- err = netdev_vport_do_ioctl(ODP_VPORT_MOD, &ovm);
- free(ovm.config);
- return err;
+ return error;
}
static int
return err;
}
+static int
+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;
+}
+
static int
netdev_vport_update_flags(struct netdev *netdev OVS_UNUSED,
enum netdev_flags off, enum netdev_flags on OVS_UNUSED,
free(notifier);
}
+
+static void
+netdev_vport_run(void)
+{
+ rtnetlink_link_notifier_run();
+ route_table_run();
+}
+
+static void
+netdev_vport_wait(void)
+{
+ rtnetlink_link_notifier_wait();
+ route_table_wait();
+}
+\f
+/* get_tnl_iface() implementation. */
+
+static struct name_node *
+name_node_lookup(int ifi_index)
+{
+ struct name_node *nn;
+
+ HMAP_FOR_EACH_WITH_HASH(nn, node, hash_int(ifi_index, 0), &name_map) {
+ if (nn->ifi_index == ifi_index) {
+ return nn;
+ }
+ }
+
+ return NULL;
+}
+
+/* Queries the kernel for fresh data to populate the name map with. */
+static int
+netdev_vport_reset_names(void)
+{
+ int error;
+ struct nl_dump dump;
+ struct rtgenmsg *rtmsg;
+ struct ofpbuf request, reply;
+ static struct nl_sock *rtnl_sock;
+ struct name_node *nn, *nn_next;
+
+ HMAP_FOR_EACH_SAFE(nn, nn_next, node, &name_map) {
+ hmap_remove(&name_map, &nn->node);
+ free(nn);
+ }
+
+ error = nl_sock_create(NETLINK_ROUTE, &rtnl_sock);
+ if (error) {
+ VLOG_WARN_RL(&rl, "Failed to create NETLINK_ROUTE socket");
+ return error;
+ }
+
+ ofpbuf_init(&request, 0);
+
+ nl_msg_put_nlmsghdr(&request, sizeof *rtmsg, RTM_GETLINK, NLM_F_REQUEST);
+
+ rtmsg = ofpbuf_put_zeros(&request, sizeof *rtmsg);
+ rtmsg->rtgen_family = AF_INET;
+
+ nl_dump_start(&dump, rtnl_sock, &request);
+
+ while (nl_dump_next(&dump, &reply)) {
+ struct rtnetlink_link_change change;
+
+ 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 void
+netdev_vport_link_change(const struct rtnetlink_link_change *change,
+ void *aux OVS_UNUSED)
+{
+
+ if (!change) {
+ netdev_vport_reset_names();
+ } else if (change->nlmsg_type == RTM_NEWLINK) {
+ struct name_node *nn;
+
+ if (name_node_lookup(change->ifi_index)) {
+ return;
+ }
+
+ nn = xzalloc(sizeof *nn);
+ nn->ifi_index = change->ifi_index;
+
+ strncpy(nn->ifname, change->ifname, IFNAMSIZ);
+ nn->ifname[IFNAMSIZ - 1] = '\0';
+
+ hmap_insert(&name_map, &nn->node, hash_int(nn->ifi_index, 0));
+ } else if (change->nlmsg_type == RTM_DELLINK) {
+ struct name_node *nn;
+
+ nn = name_node_lookup(change->ifi_index);
+
+ if (nn) {
+ hmap_remove(&name_map, &nn->node);
+ free(nn);
+ }
+
+ } else {
+ VLOG_WARN_RL(&rl, "Received unexpected rtnetlink message type %d",
+ change->nlmsg_type);
+ }
+}
+
+static void
+netdev_vport_tnl_iface_init(void)
+{
+ static bool tnl_iface_is_init = false;
+
+ if (!tnl_iface_is_init) {
+ hmap_init(&name_map);
+
+ rtnetlink_link_notifier_register(&netdev_vport_link_notifier,
+ netdev_vport_link_change, NULL);
+
+ netdev_vport_reset_names();
+ tnl_iface_is_init = true;
+ }
+}
+
+static const char *
+netdev_vport_get_tnl_iface(const struct netdev *netdev)
+{
+ int ifindex;
+ uint32_t route;
+ struct netdev_dev_vport *ndv;
+ struct tnl_port_config *config;
+
+ ndv = netdev_dev_vport_cast(netdev_get_dev(netdev));
+ config = (struct tnl_port_config *) ndv->config;
+ route = config->daddr;
+
+ if (route_table_get_ifindex(route, &ifindex)) {
+ struct name_node *nn;
+ HMAP_FOR_EACH_WITH_HASH(nn, node, hash_int(ifindex, 0), &name_map) {
+ if (nn->ifi_index == ifindex) {
+ return nn->ifname;
+ }
+ }
+ }
+
+ return NULL;
+}
\f
/* Helper functions. */
/* Code specific to individual vport types. */
static int
-parse_tunnel_config(struct vport_info *port, const struct shash *args)
+parse_tunnel_config(const struct netdev_dev *dev, const struct shash *args,
+ void *configp)
{
- const char *name = port->devname;
- bool is_gre = !strcmp(port->type, "gre");
- struct tnl_port_config *config;
+ 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_ip_set = false;
bool ipsec_mech_set = false;
- config = port->config = xzalloc(sizeof *config);
- config->flags |= TNL_F_PMTUD;
- config->flags |= TNL_F_HDR_CACHE;
+ memset(&config, 0, sizeof config);
+ config.flags |= TNL_F_PMTUD;
+ config.flags |= 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;
+ }
SHASH_FOR_EACH (node, args) {
if (!strcmp(node->name, "remote_ip")) {
struct in_addr in_addr;
if (lookup_ip(node->data, &in_addr)) {
- VLOG_WARN("%s: bad %s 'remote_ip'", name, port->type);
+ VLOG_WARN("%s: bad %s 'remote_ip'", name, type);
} else {
- config->daddr = in_addr.s_addr;
+ config.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, port->type);
+ VLOG_WARN("%s: bad %s 'local_ip'", name, type);
} else {
- config->saddr = in_addr.s_addr;
+ 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;
+ config.flags |= TNL_F_IN_KEY_MATCH;
+ config.flags |= TNL_F_OUT_KEY_ACTION;
} else {
- config->out_key = config->in_key = htonl(atoi(node->data));
+ 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;
+ config.flags |= TNL_F_IN_KEY_MATCH;
} else {
- config->in_key = htonl(atoi(node->data));
+ 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;
+ config.flags |= TNL_F_OUT_KEY_ACTION;
} else {
- config->out_key = htonl(atoi(node->data));
+ config.out_key = htonll(strtoull(node->data, NULL, 0));
}
} else if (!strcmp(node->name, "tos")) {
if (!strcmp(node->data, "inherit")) {
- config->flags |= TNL_F_TOS_INHERIT;
+ config.flags |= TNL_F_TOS_INHERIT;
} else {
- config->tos = atoi(node->data);
+ config.tos = atoi(node->data);
}
} else if (!strcmp(node->name, "ttl")) {
if (!strcmp(node->data, "inherit")) {
- config->flags |= TNL_F_TTL_INHERIT;
+ config.flags |= TNL_F_TTL_INHERIT;
} else {
- config->ttl = atoi(node->data);
+ config.ttl = atoi(node->data);
}
} else if (!strcmp(node->name, "csum") && is_gre) {
if (!strcmp(node->data, "true")) {
- config->flags |= TNL_F_CSUM;
+ config.flags |= TNL_F_CSUM;
}
} else if (!strcmp(node->name, "pmtud")) {
if (!strcmp(node->data, "false")) {
- config->flags &= ~TNL_F_PMTUD;
+ config.flags &= ~TNL_F_PMTUD;
}
} else if (!strcmp(node->name, "header_cache")) {
if (!strcmp(node->data, "false")) {
- config->flags &= ~TNL_F_HDR_CACHE;
+ config.flags &= ~TNL_F_HDR_CACHE;
+ }
+ } else if (!strcmp(node->name, "peer_cert") && is_ipsec) {
+ if (shash_find(args, "certificate")) {
+ ipsec_mech_set = true;
+ } else {
+ 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 = shash_find_data(args, "use_ssl_cert");
+ if (!use_ssl_cert || strcmp(use_ssl_cert, "true")) {
+ VLOG_WARN("%s: 'peer_cert' requires 'certificate' argument",
+ name);
+ return EINVAL;
+ }
+ ipsec_mech_set = true;
}
- } 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->name, "psk") && is_ipsec) {
ipsec_mech_set = true;
+ } else if (is_ipsec
+ && (!strcmp(node->name, "certificate")
+ || !strcmp(node->name, "private_key")
+ || !strcmp(node->name, "use_ssl_cert"))) {
+ /* Ignore options not used by the netdev. */
} else {
VLOG_WARN("%s: unknown %s argument '%s'",
- name, port->type, node->name);
+ name, type, node->name);
}
}
- /* 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;
+ if (is_ipsec) {
+ if (shash_find(args, "peer_cert") && shash_find(args, "psk")) {
+ VLOG_WARN("%s: cannot define both 'peer_cert' and 'psk'", name);
+ return EINVAL;
+ }
+
+ if (!ipsec_mech_set) {
+ VLOG_WARN("%s: IPsec requires an 'peer_cert' or psk' argument",
+ name);
+ return EINVAL;
+ }
}
- if (!config->daddr) {
+ if (!config.daddr) {
VLOG_WARN("%s: %s type requires valid 'remote_ip' argument",
- name, port->type);
+ name, type);
return EINVAL;
}
+ BUILD_ASSERT(sizeof config <= VPORT_CONFIG_SIZE);
+ memcpy(configp, &config, sizeof config);
return 0;
}
static int
-parse_patch_config(struct vport_info *port, const struct shash *args)
+parse_patch_config(const struct netdev_dev *dev, const struct shash *args,
+ void *configp)
{
- const char *name = port->devname;
+ const char *name = netdev_dev_get_name(dev);
const char *peer;
peer = shash_find_data(args, "peer");
return EINVAL;
}
- if (strlen(peer) >= IFNAMSIZ) {
+ if (strlen(peer) >= MIN(IFNAMSIZ, VPORT_CONFIG_SIZE)) {
VLOG_WARN("%s: patch 'peer' arg too long", name);
return EINVAL;
}
return EINVAL;
}
- port->config = xstrdup(peer);
+ strncpy(configp, peer, VPORT_CONFIG_SIZE);
return 0;
}
\f
-#define VPORT_FUNCTIONS \
- NULL, /* init */ \
- NULL, /* run */ \
- NULL, /* wait */ \
+#define VPORT_FUNCTIONS(GET_STATUS) \
+ netdev_vport_init, \
+ netdev_vport_run, \
+ netdev_vport_wait, \
\
netdev_vport_create, \
netdev_vport_destroy, \
netdev_vport_get_mtu, \
NULL, /* get_ifindex */ \
NULL, /* get_carrier */ \
+ NULL, /* get_miimon */ \
netdev_vport_get_stats, \
netdev_vport_set_stats, \
\
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,
-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 };
-
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[] = {
+ { { "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 }
+ };
+
+ int i;
+
+ for (i = 0; i < ARRAY_SIZE(vport_classes); i++) {
+ netdev_register_provider(&vport_classes[i].netdev_class);
+ }
}
git://git.onelab.eu / sliver-openvswitch.git / blobdiff