#endif
static struct rtnetlink_notifier netdev_linux_cache_notifier;
-static struct shash cache_map = SHASH_INITIALIZER(&cache_map);
+static int cache_notifier_refcount;
enum {
VALID_IFINDEX = 1 << 0,
int fd;
};
+struct patch_state {
+ char *peer;
+};
+
struct netdev_dev_linux {
struct netdev_dev netdev_dev;
union {
struct tap_state tap;
+ struct patch_state patch;
} state;
};
uint32_t remote_ip;
uint32_t in_key;
uint32_t out_key;
+ uint8_t tos;
bool have_in_key;
bool have_out_key;
bool in_csum;
bool out_csum;
+ bool pmtud;
};
static struct {
* additional log messages. */
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
+static int if_up(const char *name);
static int destroy_gre(const char *name);
static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
int cmd, const char *cmd_name);
{
const char *type = netdev_dev_get_type(netdev_dev);
assert(!strcmp(type, "system") || !strcmp(type, "tap")
- || !strcmp(type, "gre"));
+ || !strcmp(type, "gre") || !strcmp(type, "patch"));
return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
}
{
const char *type = netdev_get_type(netdev);
assert(!strcmp(type, "system") || !strcmp(type, "tap")
- || !strcmp(type, "gre"));
+ || !strcmp(type, "gre") || !strcmp(type, "patch"));
return CONTAINER_OF(netdev, struct netdev_linux, netdev);
}
static void
netdev_linux_cache_cb(const struct rtnetlink_change *change,
- void *aux UNUSED)
+ void *aux OVS_UNUSED)
{
struct netdev_dev_linux *dev;
if (change) {
- dev = shash_find_data(&cache_map, change->ifname);
- if (dev) {
+ struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
+ if (base_dev) {
+ dev = netdev_dev_linux_cast(base_dev);
dev->cache_valid = 0;
}
} else {
+ struct shash device_shash;
struct shash_node *node;
- SHASH_FOR_EACH (node, &cache_map) {
+
+ shash_init(&device_shash);
+ netdev_dev_get_devices(&netdev_linux_class, &device_shash);
+ SHASH_FOR_EACH (node, &device_shash) {
dev = node->data;
dev->cache_valid = 0;
}
+ shash_destroy(&device_shash);
}
}
/* The arguments are marked as unused to prevent warnings on platforms where
* the Netlink interface isn't supported. */
static int
-setup_gre_netlink(const char *name UNUSED, struct gre_config *config UNUSED,
- bool create UNUSED)
+setup_gre_netlink(const char *name OVS_UNUSED,
+ struct gre_config *config OVS_UNUSED, bool create OVS_UNUSED)
{
#ifdef GRE_IOCTL_ONLY
return EOPNOTSUPP;
struct nlattr *info_data_hdr;
uint16_t iflags = 0;
uint16_t oflags = 0;
- uint8_t pmtudisc = 0;
VLOG_DBG("%s: attempting to create gre device using netlink", name);
nl_msg_put_u16(&request, IFLA_GRE_OFLAGS, oflags);
nl_msg_put_u32(&request, IFLA_GRE_LOCAL, config->local_ip);
nl_msg_put_u32(&request, IFLA_GRE_REMOTE, config->remote_ip);
- nl_msg_put_u8(&request, IFLA_GRE_PMTUDISC, pmtudisc);
- nl_msg_put_u8(&request, IFLA_GRE_TTL, 0);
- nl_msg_put_u8(&request, IFLA_GRE_TOS, 0);
+ nl_msg_put_u8(&request, IFLA_GRE_PMTUDISC, config->pmtud);
+ nl_msg_put_u8(&request, IFLA_GRE_TTL, IPDEFTTL);
+ nl_msg_put_u8(&request, IFLA_GRE_TOS, config->tos);
info_data_hdr->nla_len = (char *)ofpbuf_tail(&request)
- (char *)info_data_hdr;
p.iph.protocol = IPPROTO_GRE;
p.iph.saddr = config->local_ip;
p.iph.daddr = config->remote_ip;
+ p.iph.ttl = IPDEFTTL;
+ p.iph.tos = config->tos;
if (config->have_in_key) {
p.i_flags |= GRE_KEY;
p.o_flags |= GRE_CSUM;
}
+ if (config->pmtud) {
+ p.iph.frag_off = htons(IP_DONT_FRAGMENT);
+ }
+
strncpy(ifr.ifr_name, create ? GRE_IOCTL_DEVICE : name, IFNAMSIZ);
ifr.ifr_ifru.ifru_data = (void *)&p;
/* The arguments are marked as unused to prevent warnings on platforms where
* the Netlink interface isn't supported. */
static bool
-check_gre_device_netlink(const char *name UNUSED)
+check_gre_device_netlink(const char *name OVS_UNUSED)
{
#ifdef GRE_IOCTL_ONLY
return false;
memset(&config, 0, sizeof config);
config.in_csum = true;
config.out_csum = true;
+ config.pmtud = true;
SHASH_FOR_EACH (node, args) {
if (!strcmp(node->name, "remote_ip")) {
} else if (!strcmp(node->name, "out_key")) {
config.have_out_key = true;
config.out_key = htonl(atoi(node->data));
+ } else if (!strcmp(node->name, "tos")) {
+ config.tos = atoi(node->data);
} else if (!strcmp(node->name, "csum")) {
if (!strcmp(node->data, "false")) {
config.in_csum = false;
config.out_csum = false;
}
+ } else if (!strcmp(node->name, "pmtud")) {
+ if (!strcmp(node->data, "false")) {
+ config.pmtud = false;
+ }
} else {
VLOG_WARN("unknown gre argument '%s'", node->name);
}
return error;
}
+/* A veth may be created using the 'command' "+<name>,<peer>". A veth may
+ * be destroyed by using the 'command' "-<name>", where <name> can be
+ * either side of the device.
+ */
+static int
+modify_veth(const char *format, ...)
+{
+ FILE *veth_file;
+ va_list args;
+ int retval;
+
+ veth_file = fopen("/sys/class/net/veth_pairs", "w");
+ if (!veth_file) {
+ VLOG_WARN_RL(&rl, "could not open veth device. Are you running a "
+ "supported XenServer with the kernel module loaded?");
+ return ENODEV;
+ }
+ setvbuf(veth_file, NULL, _IONBF, 0);
+
+ va_start(args, format);
+ retval = vfprintf(veth_file, format, args);
+ va_end(args);
+
+ fclose(veth_file);
+ if (retval < 0) {
+ VLOG_WARN_RL(&rl, "could not destroy patch: %s", strerror(errno));
+ return errno;
+ }
+
+ return 0;
+}
+
+static int
+create_patch(const char *name, const char *peer)
+{
+ int retval;
+ struct netdev_dev *peer_nd;
+
+
+ /* Only create the veth if the peer didn't already do it. */
+ peer_nd = netdev_dev_from_name(peer);
+ if (peer_nd) {
+ if (!strcmp("patch", netdev_dev_get_type(peer_nd))) {
+ struct netdev_dev_linux *ndl = netdev_dev_linux_cast(peer_nd);
+ if (!strcmp(name, ndl->state.patch.peer)) {
+ return 0;
+ } else {
+ VLOG_WARN_RL(&rl, "peer '%s' already paired with '%s'",
+ peer, ndl->state.patch.peer);
+ return EINVAL;
+ }
+ } else {
+ VLOG_WARN_RL(&rl, "peer '%s' exists and is not a patch", peer);
+ return EINVAL;
+ }
+ }
+
+ retval = modify_veth("+%s,%s", name, peer);
+ if (retval) {
+ return retval;
+ }
+
+ retval = if_up(name);
+ if (retval) {
+ return retval;
+ }
+
+ retval = if_up(peer);
+ if (retval) {
+ return retval;
+ }
+
+ return 0;
+}
+
+static int
+setup_patch(const char *name, const struct shash *args, char **peer_)
+{
+ const char *peer;
+
+ peer = shash_find_data(args, "peer");
+ if (!peer) {
+ VLOG_WARN("patch type requires valid 'peer' argument");
+ return EINVAL;
+ }
+
+ if (shash_count(args) > 1) {
+ VLOG_WARN("patch type takes only a 'peer' argument");
+ return EINVAL;
+ }
+
+ if (strlen(peer) >= IFNAMSIZ) {
+ VLOG_WARN_RL(&rl, "patch 'peer' arg too long");
+ return EINVAL;
+ }
+
+ *peer_ = xstrdup(peer);
+ return create_patch(name, peer);
+}
+
/* Creates the netdev device of 'type' with 'name'. */
static int
-netdev_linux_create_system(const char *name, const char *type UNUSED,
+netdev_linux_create_system(const char *name, const char *type OVS_UNUSED,
const struct shash *args, struct netdev_dev **netdev_devp)
{
struct netdev_dev_linux *netdev_dev;
VLOG_WARN("%s: arguments for system devices should be empty", name);
}
- if (shash_is_empty(&cache_map)) {
+ if (!cache_notifier_refcount) {
error = rtnetlink_notifier_register(&netdev_linux_cache_notifier,
netdev_linux_cache_cb, NULL);
if (error) {
return error;
}
}
+ cache_notifier_refcount++;
netdev_dev = xzalloc(sizeof *netdev_dev);
- netdev_dev->shash_node = shash_add(&cache_map, name, &netdev_dev);
-
netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class);
+
*netdev_devp = &netdev_dev->netdev_dev;
return 0;
}
* buffers, across all readers. Therefore once data is read it will
* be unavailable to other reads for tap devices. */
static int
-netdev_linux_create_tap(const char *name, const char *type UNUSED,
+netdev_linux_create_tap(const char *name, const char *type OVS_UNUSED,
const struct shash *args, struct netdev_dev **netdev_devp)
{
struct netdev_dev_linux *netdev_dev;
}
static int
-netdev_linux_create_gre(const char *name, const char *type UNUSED,
+netdev_linux_create_gre(const char *name, const char *type OVS_UNUSED,
const struct shash *args, struct netdev_dev **netdev_devp)
{
struct netdev_dev_linux *netdev_dev;
return error;
}
+static int
+netdev_linux_create_patch(const char *name, const char *type OVS_UNUSED,
+ const struct shash *args, struct netdev_dev **netdev_devp)
+{
+ struct netdev_dev_linux *netdev_dev;
+ char *peer = NULL;
+ int error;
+
+ error = setup_patch(name, args, &peer);
+ if (error) {
+ free(peer);
+ return error;
+ }
+
+ netdev_dev = xzalloc(sizeof *netdev_dev);
+ netdev_dev->state.patch.peer = peer;
+ netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_patch_class);
+ *netdev_devp = &netdev_dev->netdev_dev;
+
+ return 0;
+}
+
static int
netdev_linux_reconfigure_gre(struct netdev_dev *netdev_dev_,
const struct shash *args)
/* The arguments are marked as unused to prevent warnings on platforms where
* the Netlink interface isn't supported. */
static int
-destroy_gre_netlink(const char *name UNUSED)
+destroy_gre_netlink(const char *name OVS_UNUSED)
{
#ifdef GRE_IOCTL_ONLY
return EOPNOTSUPP;
}
}
+static void
+destroy_patch(struct netdev_dev_linux *netdev_dev)
+{
+ const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
+ struct patch_state *state = &netdev_dev->state.patch;
+
+ /* Only destroy veth if 'peer' doesn't exist as an existing netdev. */
+ if (!netdev_dev_from_name(state->peer)) {
+ modify_veth("-%s", name);
+ }
+ free(state->peer);
+}
+
/* Destroys the netdev device 'netdev_dev_'. */
static void
netdev_linux_destroy(struct netdev_dev *netdev_dev_)
const char *type = netdev_dev_get_type(netdev_dev_);
if (!strcmp(type, "system")) {
- shash_delete(&cache_map, netdev_dev->shash_node);
+ cache_notifier_refcount--;
- if (shash_is_empty(&cache_map)) {
+ if (!cache_notifier_refcount) {
rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
}
} else if (!strcmp(type, "tap")) {
destroy_tap(netdev_dev);
} else if (!strcmp(type, "gre")) {
destroy_gre(netdev_dev_get_name(&netdev_dev->netdev_dev));
+ } else if (!strcmp(type, "patch")) {
+ destroy_patch(netdev_dev);
}
free(netdev_dev_);
/* Allocate network device. */
netdev = xzalloc(sizeof *netdev);
+ netdev->fd = -1;
netdev_init(&netdev->netdev, netdev_dev_);
error = netdev_get_flags(&netdev->netdev, &flags);
if (error) {
goto error;
}
- } else {
- netdev->fd = -1;
}
*netdevp = &netdev->netdev;
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (netdev->fd >= 0 && strcmp(netdev_get_type(netdev_), "tap")) {
+ if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
close(netdev->fd);
}
free(netdev);
COVERAGE_INC(netdev_get_stats);
if (!(netdev_dev->cache_valid & VALID_IS_INTERNAL)) {
- netdev_dev->is_internal = !strcmp(netdev_get_type(netdev_),
- "tap");
-
+ netdev_dev->is_internal = !strcmp(netdev_get_type(netdev_), "tap");
if (!netdev_dev->is_internal) {
struct ethtool_drvinfo drvinfo;
* will appear to be swapped relative to the other ports since we are the
* one sending the data, not a remote computer. For consistency, we swap
* them back here. */
- if (netdev_dev->is_internal) {
+ if (!error && netdev_dev->is_internal) {
stats->rx_packets = raw_stats.tx_packets;
stats->tx_packets = raw_stats.rx_packets;
stats->rx_bytes = raw_stats.tx_bytes;
COVERAGE_INC(netdev_set_policing);
if (kbits_rate) {
if (!kbits_burst) {
- /* Default to 10 kilobits if not specified. */
- kbits_burst = 10;
+ /* Default to 1000 kilobits if not specified. */
+ kbits_burst = 1000;
}
/* xxx This should be more careful about only adding if it
/* Adds 'router' as a default IP gateway. */
static int
-netdev_linux_add_router(struct netdev *netdev UNUSED, struct in_addr router)
+netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
{
struct in_addr any = { INADDR_ANY };
struct rtentry rt;
static void
netdev_linux_poll_cb(const struct rtnetlink_change *change,
- void *aux UNUSED)
+ void *aux OVS_UNUSED)
{
if (change) {
struct list *list = shash_find_data(&netdev_linux_notifiers,
netdev_linux_poll_add,
netdev_linux_poll_remove,
};
+
+const struct netdev_class netdev_patch_class = {
+ "patch",
+
+ netdev_linux_init,
+ netdev_linux_run,
+ netdev_linux_wait,
+
+ netdev_linux_create_patch,
+ netdev_linux_destroy,
+ NULL, /* reconfigure */
+
+ netdev_linux_open,
+ netdev_linux_close,
+
+ NULL, /* enumerate */
+
+ netdev_linux_recv,
+ netdev_linux_recv_wait,
+ netdev_linux_drain,
+
+ netdev_linux_send,
+ netdev_linux_send_wait,
+
+ netdev_linux_set_etheraddr,
+ netdev_linux_get_etheraddr,
+ netdev_linux_get_mtu,
+ netdev_linux_get_ifindex,
+ netdev_linux_get_carrier,
+ netdev_linux_get_stats,
+
+ netdev_linux_get_features,
+ netdev_linux_set_advertisements,
+ netdev_linux_get_vlan_vid,
+ netdev_linux_set_policing,
+
+ netdev_linux_get_in4,
+ netdev_linux_set_in4,
+ netdev_linux_get_in6,
+ netdev_linux_add_router,
+ netdev_linux_get_next_hop,
+ netdev_linux_arp_lookup,
+
+ netdev_linux_update_flags,
+
+ netdev_linux_poll_add,
+ netdev_linux_poll_remove,
+};
\f
static int
get_stats_via_netlink(int ifindex, struct netdev_stats *stats)