#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,
struct netdev_linux {
struct netdev netdev;
-
- /* File descriptors. For ordinary network devices, the two fds below are
- * the same; for tap devices, they differ. */
- int netdev_fd; /* Network device. */
- int tap_fd; /* TAP character device, if any, otherwise the
- * network device. */
+ int fd;
};
/* An AF_INET socket (used for ioctl operations). */
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 {
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);
}
/* 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;
}
+/* For most types of netdevs we open the device for each call of
+ * netdev_open(). However, this is not the case with tap devices,
+ * since it is only possible to open the device once. In this
+ * situation we share a single file descriptor, and consequently
+ * 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;
/* 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;
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")) {
}
static int
-netdev_linux_open(struct netdev_dev *netdev_dev, int ethertype,
+netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype,
struct netdev **netdevp)
{
+ struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
struct netdev_linux *netdev;
enum netdev_flags flags;
int error;
/* Allocate network device. */
netdev = xzalloc(sizeof *netdev);
- netdev_init(&netdev->netdev, netdev_dev);
- netdev->netdev_fd = -1;
- netdev->tap_fd = -1;
-
- if (!strcmp(netdev_dev_get_type(netdev_dev), "tap")) {
- static const char tap_dev[] = "/dev/net/tun";
- struct ifreq ifr;
-
- /* Open tap device. */
- netdev->tap_fd = open(tap_dev, O_RDWR);
- if (netdev->tap_fd < 0) {
- error = errno;
- VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
- goto error;
- }
-
- /* Create tap device. */
- ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
- strncpy(ifr.ifr_name, netdev_dev_get_name(netdev_dev),
- sizeof ifr.ifr_name);
- if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) {
- VLOG_WARN("%s: creating tap device failed: %s",
- netdev_dev_get_name(netdev_dev),
- strerror(errno));
- error = errno;
- goto error;
- }
-
- /* Make non-blocking. */
- error = set_nonblocking(netdev->tap_fd);
- if (error) {
- goto error;
- }
- }
+ netdev->fd = -1;
+ netdev_init(&netdev->netdev, netdev_dev_);
error = netdev_get_flags(&netdev->netdev, &flags);
if (error == ENODEV) {
goto error;
}
- if (netdev->tap_fd >= 0 || ethertype != NETDEV_ETH_TYPE_NONE) {
+ if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap")) {
+ netdev->fd = netdev_dev->state.tap.fd;
+ } else if (ethertype != NETDEV_ETH_TYPE_NONE) {
struct sockaddr_ll sll;
int protocol;
int ifindex;
protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
: ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
: ethertype);
- netdev->netdev_fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
- if (netdev->netdev_fd < 0) {
+ netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
+ if (netdev->fd < 0) {
error = errno;
goto error;
}
- if (netdev->tap_fd < 0) {
- netdev->tap_fd = netdev->netdev_fd;
- }
/* Set non-blocking mode. */
- error = set_nonblocking(netdev->netdev_fd);
+ error = set_nonblocking(netdev->fd);
if (error) {
goto error;
}
memset(&sll, 0, sizeof sll);
sll.sll_family = AF_PACKET;
sll.sll_ifindex = ifindex;
- if (bind(netdev->netdev_fd,
+ if (bind(netdev->fd,
(struct sockaddr *) &sll, sizeof sll) < 0) {
error = errno;
- VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev),
+ VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_),
strerror(error));
goto error;
}
* packets of the requested type on all system interfaces. We do not
* want to receive that data, but there is no way to avoid it. So we
* must now drain out the receive queue. */
- error = drain_rcvbuf(netdev->netdev_fd);
+ error = drain_rcvbuf(netdev->fd);
if (error) {
goto error;
}
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (netdev->netdev_fd >= 0) {
- close(netdev->netdev_fd);
- }
- if (netdev->tap_fd >= 0 && netdev->netdev_fd != netdev->tap_fd) {
- close(netdev->tap_fd);
+ if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
+ close(netdev->fd);
}
free(netdev);
}
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (netdev->tap_fd < 0) {
+ if (netdev->fd < 0) {
/* Device was opened with NETDEV_ETH_TYPE_NONE. */
return -EAGAIN;
}
for (;;) {
- ssize_t retval = read(netdev->tap_fd, data, size);
+ ssize_t retval = read(netdev->fd, data, size);
if (retval >= 0) {
return retval;
} else if (errno != EINTR) {
netdev_linux_recv_wait(struct netdev *netdev_)
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (netdev->tap_fd >= 0) {
- poll_fd_wait(netdev->tap_fd, POLLIN);
+ if (netdev->fd >= 0) {
+ poll_fd_wait(netdev->fd, POLLIN);
}
}
netdev_linux_drain(struct netdev *netdev_)
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (netdev->tap_fd < 0 && netdev->netdev_fd < 0) {
+ if (netdev->fd < 0) {
return 0;
- } else if (netdev->tap_fd != netdev->netdev_fd) {
+ } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
struct ifreq ifr;
int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
if (error) {
return error;
}
- drain_fd(netdev->tap_fd, ifr.ifr_qlen);
+ drain_fd(netdev->fd, ifr.ifr_qlen);
return 0;
} else {
- return drain_rcvbuf(netdev->netdev_fd);
+ return drain_rcvbuf(netdev->fd);
}
}
/* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
*/
- if (netdev->tap_fd < 0) {
+ if (netdev->fd < 0) {
return EPIPE;
}
for (;;) {
- ssize_t retval = write(netdev->tap_fd, data, size);
+ ssize_t retval = write(netdev->fd, data, size);
if (retval < 0) {
/* The Linux AF_PACKET implementation never blocks waiting for room
* for packets, instead returning ENOBUFS. Translate this into
netdev_linux_send_wait(struct netdev *netdev_)
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (netdev->tap_fd < 0 && netdev->netdev_fd < 0) {
+ if (netdev->fd < 0) {
/* Nothing to do. */
- } else if (netdev->tap_fd == netdev->netdev_fd) {
- poll_fd_wait(netdev->tap_fd, POLLOUT);
+ } else if (strcmp(netdev_get_type(netdev_), "tap")) {
+ poll_fd_wait(netdev->fd, POLLOUT);
} else {
/* TAP device always accepts packets.*/
poll_immediate_wake();
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,