#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). */
* additional log messages. */
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
-static int netdev_linux_do_ethtool(const struct netdev *, struct ethtool_cmd *,
+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);
static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
const char *cmd_name);
{
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);
}
}
uint16_t oflags = 0;
uint8_t pmtudisc = 0;
+ VLOG_DBG("%s: attempting to create gre device using netlink", name);
+
if (!gre_descriptors.nl_sock) {
error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0,
&gre_descriptors.nl_sock);
if (error) {
- VLOG_WARN("couldn't create netlink socket: %s\n", strerror(error));
- gre_descriptors.nl_sock = NULL;
+ VLOG_WARN("couldn't create netlink socket: %s", strerror(error));
goto error;
}
}
error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
ofpbuf_uninit(&request);
if (error) {
- VLOG_WARN("couldn't transact netlink socket: %s\n", strerror(error));
+ VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
goto error;
}
ofpbuf_delete(reply);
struct ip_tunnel_parm p;
struct ifreq ifr;
+ VLOG_DBG("%s: attempting to create gre device using ioctl", name);
+
memset(&p, 0, sizeof p);
strncpy(p.name, name, IFNAMSIZ);
if (!gre_descriptors.ioctl_fd) {
gre_descriptors.ioctl_fd = socket(AF_INET, SOCK_DGRAM, 0);
if (gre_descriptors.ioctl_fd < 0) {
- VLOG_WARN("couldn't create gre ioctl socket: %s\n", strerror(errno));
+ VLOG_WARN("couldn't create gre ioctl socket: %s", strerror(errno));
gre_descriptors.ioctl_fd = 0;
return errno;
}
if (ioctl(gre_descriptors.ioctl_fd, create ? SIOCADDGRETAP : SIOCCHGGRETAP,
&ifr) < 0) {
- VLOG_WARN("couldn't do gre ioctl: %s\n", strerror(errno));
+ VLOG_WARN("couldn't do gre ioctl: %s", strerror(errno));
return errno;
}
return 0;
}
+/* 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)
+{
+#ifdef GRE_IOCTL_ONLY
+ return false;
+#else
+ static const struct nl_policy getlink_policy[] = {
+ [IFLA_LINKINFO] = { .type = NL_A_NESTED, .optional = false },
+ };
+
+ static const struct nl_policy linkinfo_policy[] = {
+ [IFLA_INFO_KIND] = { .type = NL_A_STRING, .optional = false },
+ };
+
+ int error;
+ bool ret = false;
+ struct ofpbuf request, *reply;
+ struct ifinfomsg ifinfomsg;
+ struct nlattr *getlink_attrs[ARRAY_SIZE(getlink_policy)];
+ struct nlattr *linkinfo_attrs[ARRAY_SIZE(linkinfo_policy)];
+ struct ofpbuf linkinfo;
+ const char *device_kind;
+
+ ofpbuf_init(&request, 0);
+
+ nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock,
+ NLMSG_LENGTH(sizeof ifinfomsg), RTM_GETLINK,
+ NLM_F_REQUEST);
+
+ memset(&ifinfomsg, 0, sizeof ifinfomsg);
+ ifinfomsg.ifi_family = AF_UNSPEC;
+ ifinfomsg.ifi_index = do_get_ifindex(name);
+ nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
+
+ error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
+ ofpbuf_uninit(&request);
+ if (error) {
+ VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
+ return false;
+ }
+
+ if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
+ getlink_policy, getlink_attrs,
+ ARRAY_SIZE(getlink_policy))) {
+ VLOG_WARN("received bad rtnl message (getlink policy)");
+ goto error;
+ }
+
+ linkinfo.data = (void *)nl_attr_get(getlink_attrs[IFLA_LINKINFO]);
+ linkinfo.size = nl_attr_get_size(getlink_attrs[IFLA_LINKINFO]);
+ if (!nl_policy_parse(&linkinfo, 0, linkinfo_policy,
+ linkinfo_attrs, ARRAY_SIZE(linkinfo_policy))) {
+ VLOG_WARN("received bad rtnl message (linkinfo policy)");
+ goto error;
+ }
+
+ device_kind = nl_attr_get_string(linkinfo_attrs[IFLA_INFO_KIND]);
+ ret = !strcmp(device_kind, "gretap");
+
+error:
+ ofpbuf_delete(reply);
+ return ret;
+#endif
+}
+
+static bool
+check_gre_device_ioctl(const char *name)
+{
+ struct ethtool_drvinfo drvinfo;
+ int error;
+
+ memset(&drvinfo, 0, sizeof drvinfo);
+ error = netdev_linux_do_ethtool(name, (struct ethtool_cmd *)&drvinfo,
+ ETHTOOL_GDRVINFO, "ETHTOOL_GDRVINFO");
+
+ return !error && !strcmp(drvinfo.driver, "ip_gre")
+ && !strcmp(drvinfo.bus_info, "gretap");
+}
+
static int
setup_gre(const char *name, const struct shash *args, bool create)
{
error = setup_gre_ioctl(name, &config, create);
}
+ if (create && error == EEXIST) {
+ bool gre_device;
+
+ if (gre_descriptors.use_ioctl) {
+ gre_device = check_gre_device_ioctl(name);
+ } else {
+ gre_device = check_gre_device_netlink(name);
+ }
+
+ if (!gre_device) {
+ goto error;
+ }
+
+ VLOG_WARN("replacing existing gre device %s", name);
+ error = destroy_gre(name);
+ if (error) {
+ goto error;
+ }
+
+ if (gre_descriptors.use_ioctl) {
+ error = setup_gre_ioctl(name, &config, create);
+ } else {
+ error = setup_gre_netlink(name, &config, create);
+ }
+ }
+
error:
return error;
}
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,
const struct shash *args, struct netdev_dev **netdev_devp)
/* The arguments are marked as unused to prevent warnings on platforms where
* the Netlink interface isn't supported. */
static int
-destroy_gre_netlink(struct netdev_dev_linux *netdev_dev UNUSED)
+destroy_gre_netlink(const char *name UNUSED)
{
#ifdef GRE_IOCTL_ONLY
return EOPNOTSUPP;
#else
- const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
int error;
struct ofpbuf request, *reply;
struct ifinfomsg ifinfomsg;
error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
ofpbuf_uninit(&request);
if (error) {
- VLOG_WARN("couldn't transact netlink socket: %s\n", strerror(error));
+ VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
goto error;
}
ofpbuf_delete(reply);
}
static int
-destroy_gre_ioctl(struct netdev_dev_linux *netdev_dev)
+destroy_gre_ioctl(const char *name)
{
- const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
struct ip_tunnel_parm p;
struct ifreq ifr;
}
}
+static int
+destroy_gre(const char *name)
+{
+ if (gre_descriptors.use_ioctl) {
+ return destroy_gre_ioctl(name);
+ } else {
+ return destroy_gre_netlink(name);
+ }
+}
+
/* 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")) {
- if (gre_descriptors.use_ioctl) {
- destroy_gre_ioctl(netdev_dev);
- } else {
- destroy_gre_netlink(netdev_dev);
- }
+ destroy_gre(netdev_dev_get_name(&netdev_dev->netdev_dev));
}
free(netdev_dev_);
}
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();
struct ethtool_drvinfo drvinfo;
memset(&drvinfo, 0, sizeof drvinfo);
- error = netdev_linux_do_ethtool(netdev_,
+ error = netdev_linux_do_ethtool(netdev_get_name(netdev_),
(struct ethtool_cmd *)&drvinfo,
ETHTOOL_GDRVINFO,
"ETHTOOL_GDRVINFO");
int error;
memset(&ecmd, 0, sizeof ecmd);
- error = netdev_linux_do_ethtool(netdev, &ecmd,
+ error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
ETHTOOL_GSET, "ETHTOOL_GSET");
if (error) {
return error;
int error;
memset(&ecmd, 0, sizeof ecmd);
- error = netdev_linux_do_ethtool(netdev, &ecmd,
+ error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
ETHTOOL_GSET, "ETHTOOL_GSET");
if (error) {
return error;
if (advertise & OFPPF_PAUSE_ASYM) {
ecmd.advertising |= ADVERTISED_Asym_Pause;
}
- return netdev_linux_do_ethtool(netdev, &ecmd,
+ return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
ETHTOOL_SSET, "ETHTOOL_SSET");
}
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
}
static int
-netdev_linux_do_ethtool(const struct netdev *netdev, struct ethtool_cmd *ecmd,
+netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
int cmd, const char *cmd_name)
{
struct ifreq ifr;
memset(&ifr, 0, sizeof ifr);
- strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
+ strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
ifr.ifr_data = (caddr_t) ecmd;
ecmd->cmd = cmd;
} else {
if (errno != EOPNOTSUPP) {
VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
- "failed: %s", cmd_name, netdev_get_name(netdev),
- strerror(errno));
+ "failed: %s", cmd_name, name, strerror(errno));
} else {
/* The device doesn't support this operation. That's pretty
* common, so there's no point in logging anything. */
projects / openvswitch / blobdiff