2 * Copyright (c) 2009, 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
21 #include <arpa/inet.h>
23 #include <linux/if_tun.h>
25 #include <linux/types.h>
26 #include <linux/ethtool.h>
27 #include <linux/rtnetlink.h>
28 #include <linux/sockios.h>
29 #include <linux/version.h>
30 #include <sys/types.h>
31 #include <sys/ioctl.h>
32 #include <sys/socket.h>
33 #include <netpacket/packet.h>
34 #include <net/ethernet.h>
36 #include <linux/if_tunnel.h>
37 #include <net/if_arp.h>
38 #include <net/if_packet.h>
39 #include <net/route.h>
40 #include <netinet/in.h>
47 #include "dynamic-string.h"
48 #include "fatal-signal.h"
49 #include "netdev-provider.h"
52 #include "openflow/openflow.h"
53 #include "openvswitch/gre.h"
55 #include "poll-loop.h"
56 #include "rtnetlink.h"
57 #include "socket-util.h"
61 #ifndef GRE_IOCTL_ONLY
62 #include <linux/if_link.h>
65 #define THIS_MODULE VLM_netdev_linux
68 /* These were introduced in Linux 2.6.14, so they might be missing if we have
70 #ifndef ADVERTISED_Pause
71 #define ADVERTISED_Pause (1 << 13)
73 #ifndef ADVERTISED_Asym_Pause
74 #define ADVERTISED_Asym_Pause (1 << 14)
77 static struct rtnetlink_notifier netdev_linux_cache_notifier;
78 static struct shash cache_map = SHASH_INITIALIZER(&cache_map);
81 VALID_IFINDEX = 1 << 0,
82 VALID_ETHERADDR = 1 << 1,
86 VALID_CARRIER = 1 << 5,
87 VALID_IS_INTERNAL = 1 << 6
94 struct netdev_dev_linux {
95 struct netdev_dev netdev_dev;
97 struct shash_node *shash_node;
98 unsigned int cache_valid;
101 uint8_t etheraddr[ETH_ADDR_LEN];
102 struct in_addr address, netmask;
109 struct tap_state tap;
113 struct netdev_linux {
114 struct netdev netdev;
116 /* File descriptors. For ordinary network devices, the two fds below are
117 * the same; for tap devices, they differ. */
118 int netdev_fd; /* Network device. */
119 int tap_fd; /* TAP character device, if any, otherwise the
123 /* An AF_INET socket (used for ioctl operations). */
124 static int af_inet_sock = -1;
139 struct nl_sock *nl_sock;
145 struct netdev_linux_notifier {
146 struct netdev_notifier notifier;
150 static struct shash netdev_linux_notifiers =
151 SHASH_INITIALIZER(&netdev_linux_notifiers);
152 static struct rtnetlink_notifier netdev_linux_poll_notifier;
154 /* This is set pretty low because we probably won't learn anything from the
155 * additional log messages. */
156 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
158 static int destroy_gre(const char *name);
159 static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
160 int cmd, const char *cmd_name);
161 static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
162 const char *cmd_name);
163 static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
164 int cmd, const char *cmd_name);
165 static int get_flags(const struct netdev *, int *flagsp);
166 static int set_flags(struct netdev *, int flags);
167 static int do_get_ifindex(const char *netdev_name);
168 static int get_ifindex(const struct netdev *, int *ifindexp);
169 static int do_set_addr(struct netdev *netdev,
170 int ioctl_nr, const char *ioctl_name,
171 struct in_addr addr);
172 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
173 static int set_etheraddr(const char *netdev_name, int hwaddr_family,
174 const uint8_t[ETH_ADDR_LEN]);
175 static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
176 static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
178 static struct netdev_dev_linux *
179 netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
181 const char *type = netdev_dev_get_type(netdev_dev);
182 assert(!strcmp(type, "system") || !strcmp(type, "tap")
183 || !strcmp(type, "gre"));
184 return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
187 static struct netdev_linux *
188 netdev_linux_cast(const struct netdev *netdev)
190 const char *type = netdev_get_type(netdev);
191 assert(!strcmp(type, "system") || !strcmp(type, "tap")
192 || !strcmp(type, "gre"));
193 return CONTAINER_OF(netdev, struct netdev_linux, netdev);
197 netdev_linux_init(void)
199 static int status = -1;
201 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
202 status = af_inet_sock >= 0 ? 0 : errno;
204 VLOG_ERR("failed to create inet socket: %s", strerror(status));
211 netdev_linux_run(void)
213 rtnetlink_notifier_run();
217 netdev_linux_wait(void)
219 rtnetlink_notifier_wait();
223 netdev_linux_cache_cb(const struct rtnetlink_change *change,
226 struct netdev_dev_linux *dev;
228 dev = shash_find_data(&cache_map, change->ifname);
230 dev->cache_valid = 0;
233 struct shash_node *node;
234 SHASH_FOR_EACH (node, &cache_map) {
236 dev->cache_valid = 0;
241 /* The arguments are marked as unused to prevent warnings on platforms where
242 * the Netlink interface isn't supported. */
244 setup_gre_netlink(const char *name UNUSED, struct gre_config *config UNUSED,
247 #ifdef GRE_IOCTL_ONLY
251 struct ofpbuf request, *reply;
252 unsigned int nl_flags;
253 struct ifinfomsg ifinfomsg;
254 struct nlattr *linkinfo_hdr;
255 struct nlattr *info_data_hdr;
258 uint8_t pmtudisc = 0;
260 VLOG_DBG("%s: attempting to create gre device using netlink", name);
262 if (!gre_descriptors.nl_sock) {
263 error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0,
264 &gre_descriptors.nl_sock);
266 VLOG_WARN("couldn't create netlink socket: %s", strerror(error));
271 ofpbuf_init(&request, 0);
273 nl_flags = NLM_F_REQUEST;
275 nl_flags |= NLM_F_CREATE|NLM_F_EXCL;
278 /* We over-reserve space, because we do some pointer arithmetic
279 * and don't want the buffer address shifting under us. */
280 nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock, 2048, RTM_NEWLINK,
283 memset(&ifinfomsg, 0, sizeof ifinfomsg);
284 ifinfomsg.ifi_family = AF_UNSPEC;
285 nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
287 linkinfo_hdr = ofpbuf_tail(&request);
288 nl_msg_put_unspec(&request, IFLA_LINKINFO, NULL, 0);
290 nl_msg_put_unspec(&request, IFLA_INFO_KIND, "gretap", 6);
292 info_data_hdr = ofpbuf_tail(&request);
293 nl_msg_put_unspec(&request, IFLA_INFO_DATA, NULL, 0);
296 if (config->have_in_key) {
299 if (config->have_out_key) {
303 if (config->in_csum) {
306 if (config->out_csum) {
311 nl_msg_put_u32(&request, IFLA_GRE_IKEY, config->in_key);
312 nl_msg_put_u32(&request, IFLA_GRE_OKEY, config->out_key);
313 nl_msg_put_u16(&request, IFLA_GRE_IFLAGS, iflags);
314 nl_msg_put_u16(&request, IFLA_GRE_OFLAGS, oflags);
315 nl_msg_put_u32(&request, IFLA_GRE_LOCAL, config->local_ip);
316 nl_msg_put_u32(&request, IFLA_GRE_REMOTE, config->remote_ip);
317 nl_msg_put_u8(&request, IFLA_GRE_PMTUDISC, pmtudisc);
318 nl_msg_put_u8(&request, IFLA_GRE_TTL, 0);
319 nl_msg_put_u8(&request, IFLA_GRE_TOS, 0);
321 info_data_hdr->nla_len = (char *)ofpbuf_tail(&request)
322 - (char *)info_data_hdr;
323 linkinfo_hdr->nla_len = (char *)ofpbuf_tail(&request)
324 - (char *)linkinfo_hdr;
326 nl_msg_put_string(&request, IFLA_IFNAME, name);
328 error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
329 ofpbuf_uninit(&request);
331 VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
334 ofpbuf_delete(reply);
342 setup_gre_ioctl(const char *name, struct gre_config *config, bool create)
344 struct ip_tunnel_parm p;
347 VLOG_DBG("%s: attempting to create gre device using ioctl", name);
349 memset(&p, 0, sizeof p);
351 strncpy(p.name, name, IFNAMSIZ);
355 p.iph.protocol = IPPROTO_GRE;
356 p.iph.saddr = config->local_ip;
357 p.iph.daddr = config->remote_ip;
359 if (config->have_in_key) {
360 p.i_flags |= GRE_KEY;
361 p.i_key = config->in_key;
363 if (config->have_out_key) {
364 p.o_flags |= GRE_KEY;
365 p.o_key = config->out_key;
368 if (config->in_csum) {
369 p.i_flags |= GRE_CSUM;
371 if (config->out_csum) {
372 p.o_flags |= GRE_CSUM;
375 strncpy(ifr.ifr_name, create ? GRE_IOCTL_DEVICE : name, IFNAMSIZ);
376 ifr.ifr_ifru.ifru_data = (void *)&p;
378 if (!gre_descriptors.ioctl_fd) {
379 gre_descriptors.ioctl_fd = socket(AF_INET, SOCK_DGRAM, 0);
380 if (gre_descriptors.ioctl_fd < 0) {
381 VLOG_WARN("couldn't create gre ioctl socket: %s", strerror(errno));
382 gre_descriptors.ioctl_fd = 0;
387 if (ioctl(gre_descriptors.ioctl_fd, create ? SIOCADDGRETAP : SIOCCHGGRETAP,
389 VLOG_WARN("couldn't do gre ioctl: %s", strerror(errno));
396 /* The arguments are marked as unused to prevent warnings on platforms where
397 * the Netlink interface isn't supported. */
399 check_gre_device_netlink(const char *name UNUSED)
401 #ifdef GRE_IOCTL_ONLY
404 static const struct nl_policy getlink_policy[] = {
405 [IFLA_LINKINFO] = { .type = NL_A_NESTED, .optional = false },
408 static const struct nl_policy linkinfo_policy[] = {
409 [IFLA_INFO_KIND] = { .type = NL_A_STRING, .optional = false },
414 struct ofpbuf request, *reply;
415 struct ifinfomsg ifinfomsg;
416 struct nlattr *getlink_attrs[ARRAY_SIZE(getlink_policy)];
417 struct nlattr *linkinfo_attrs[ARRAY_SIZE(linkinfo_policy)];
418 struct ofpbuf linkinfo;
419 const char *device_kind;
421 ofpbuf_init(&request, 0);
423 nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock,
424 NLMSG_LENGTH(sizeof ifinfomsg), RTM_GETLINK,
427 memset(&ifinfomsg, 0, sizeof ifinfomsg);
428 ifinfomsg.ifi_family = AF_UNSPEC;
429 ifinfomsg.ifi_index = do_get_ifindex(name);
430 nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
432 error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
433 ofpbuf_uninit(&request);
435 VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
439 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
440 getlink_policy, getlink_attrs,
441 ARRAY_SIZE(getlink_policy))) {
442 VLOG_WARN("received bad rtnl message (getlink policy)");
446 linkinfo.data = (void *)nl_attr_get(getlink_attrs[IFLA_LINKINFO]);
447 linkinfo.size = nl_attr_get_size(getlink_attrs[IFLA_LINKINFO]);
448 if (!nl_policy_parse(&linkinfo, 0, linkinfo_policy,
449 linkinfo_attrs, ARRAY_SIZE(linkinfo_policy))) {
450 VLOG_WARN("received bad rtnl message (linkinfo policy)");
454 device_kind = nl_attr_get_string(linkinfo_attrs[IFLA_INFO_KIND]);
455 ret = !strcmp(device_kind, "gretap");
458 ofpbuf_delete(reply);
464 check_gre_device_ioctl(const char *name)
466 struct ethtool_drvinfo drvinfo;
469 memset(&drvinfo, 0, sizeof drvinfo);
470 error = netdev_linux_do_ethtool(name, (struct ethtool_cmd *)&drvinfo,
471 ETHTOOL_GDRVINFO, "ETHTOOL_GDRVINFO");
473 return !error && !strcmp(drvinfo.driver, "ip_gre")
474 && !strcmp(drvinfo.bus_info, "gretap");
478 setup_gre(const char *name, const struct shash *args, bool create)
481 struct in_addr in_addr;
482 struct shash_node *node;
483 struct gre_config config;
485 memset(&config, 0, sizeof config);
486 config.in_csum = true;
487 config.out_csum = true;
489 SHASH_FOR_EACH (node, args) {
490 if (!strcmp(node->name, "remote_ip")) {
491 if (lookup_ip(node->data, &in_addr)) {
492 VLOG_WARN("bad 'remote_ip' for gre device %s ", name);
494 config.remote_ip = in_addr.s_addr;
496 } else if (!strcmp(node->name, "local_ip")) {
497 if (lookup_ip(node->data, &in_addr)) {
498 VLOG_WARN("bad 'local_ip' for gre device %s ", name);
500 config.local_ip = in_addr.s_addr;
502 } else if (!strcmp(node->name, "key")) {
503 config.have_in_key = true;
504 config.have_out_key = true;
505 config.in_key = htonl(atoi(node->data));
506 config.out_key = htonl(atoi(node->data));
507 } else if (!strcmp(node->name, "in_key")) {
508 config.have_in_key = true;
509 config.in_key = htonl(atoi(node->data));
510 } else if (!strcmp(node->name, "out_key")) {
511 config.have_out_key = true;
512 config.out_key = htonl(atoi(node->data));
513 } else if (!strcmp(node->name, "csum")) {
514 if (!strcmp(node->data, "false")) {
515 config.in_csum = false;
516 config.out_csum = false;
519 VLOG_WARN("unknown gre argument '%s'", node->name);
523 if (!config.remote_ip) {
524 VLOG_WARN("gre type requires valid 'remote_ip' argument");
529 if (!gre_descriptors.use_ioctl) {
530 error = setup_gre_netlink(name, &config, create);
531 if (error == EOPNOTSUPP) {
532 gre_descriptors.use_ioctl = true;
535 if (gre_descriptors.use_ioctl) {
536 error = setup_gre_ioctl(name, &config, create);
539 if (create && error == EEXIST) {
542 if (gre_descriptors.use_ioctl) {
543 gre_device = check_gre_device_ioctl(name);
545 gre_device = check_gre_device_netlink(name);
552 VLOG_WARN("replacing existing gre device %s", name);
553 error = destroy_gre(name);
558 if (gre_descriptors.use_ioctl) {
559 error = setup_gre_ioctl(name, &config, create);
561 error = setup_gre_netlink(name, &config, create);
569 /* Creates the netdev device of 'type' with 'name'. */
571 netdev_linux_create_system(const char *name, const char *type UNUSED,
572 const struct shash *args, struct netdev_dev **netdev_devp)
574 struct netdev_dev_linux *netdev_dev;
577 if (!shash_is_empty(args)) {
578 VLOG_WARN("%s: arguments for system devices should be empty", name);
581 if (shash_is_empty(&cache_map)) {
582 error = rtnetlink_notifier_register(&netdev_linux_cache_notifier,
583 netdev_linux_cache_cb, NULL);
589 netdev_dev = xzalloc(sizeof *netdev_dev);
590 netdev_dev->shash_node = shash_add(&cache_map, name, &netdev_dev);
592 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class);
593 *netdev_devp = &netdev_dev->netdev_dev;
598 netdev_linux_create_tap(const char *name, const char *type UNUSED,
599 const struct shash *args, struct netdev_dev **netdev_devp)
601 struct netdev_dev_linux *netdev_dev;
602 struct tap_state *state;
603 static const char tap_dev[] = "/dev/net/tun";
607 if (!shash_is_empty(args)) {
608 VLOG_WARN("%s: arguments for TAP devices should be empty", name);
611 netdev_dev = xzalloc(sizeof *netdev_dev);
612 state = &netdev_dev->state.tap;
614 /* Open tap device. */
615 state->fd = open(tap_dev, O_RDWR);
618 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
622 /* Create tap device. */
623 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
624 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
625 if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
626 VLOG_WARN("%s: creating tap device failed: %s", name,
632 /* Make non-blocking. */
633 error = set_nonblocking(state->fd);
638 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
639 *netdev_devp = &netdev_dev->netdev_dev;
648 if_up(const char *name)
652 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
653 ifr.ifr_flags = IFF_UP;
655 if (ioctl(af_inet_sock, SIOCSIFFLAGS, &ifr) == -1) {
656 VLOG_DBG_RL(&rl, "%s: failed to bring device up: %s",
657 name, strerror(errno));
665 netdev_linux_create_gre(const char *name, const char *type UNUSED,
666 const struct shash *args, struct netdev_dev **netdev_devp)
668 struct netdev_dev_linux *netdev_dev;
671 netdev_dev = xzalloc(sizeof *netdev_dev);
673 error = setup_gre(name, args, true);
683 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_gre_class);
684 *netdev_devp = &netdev_dev->netdev_dev;
693 netdev_linux_reconfigure_gre(struct netdev_dev *netdev_dev_,
694 const struct shash *args)
696 const char *name = netdev_dev_get_name(netdev_dev_);
698 return setup_gre(name, args, false);
701 /* The arguments are marked as unused to prevent warnings on platforms where
702 * the Netlink interface isn't supported. */
704 destroy_gre_netlink(const char *name UNUSED)
706 #ifdef GRE_IOCTL_ONLY
710 struct ofpbuf request, *reply;
711 struct ifinfomsg ifinfomsg;
714 ofpbuf_init(&request, 0);
716 nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock, 0, RTM_DELLINK,
719 memset(&ifinfomsg, 0, sizeof ifinfomsg);
720 ifinfomsg.ifi_family = AF_UNSPEC;
721 nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
723 ifindex = do_get_ifindex(name);
724 nl_msg_put_u32(&request, IFLA_LINK, ifindex);
726 nl_msg_put_string(&request, IFLA_IFNAME, name);
728 error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
729 ofpbuf_uninit(&request);
731 VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
734 ofpbuf_delete(reply);
742 destroy_gre_ioctl(const char *name)
744 struct ip_tunnel_parm p;
747 memset(&p, 0, sizeof p);
748 strncpy(p.name, name, IFNAMSIZ);
750 strncpy(ifr.ifr_name, name, IFNAMSIZ);
751 ifr.ifr_ifru.ifru_data = (void *)&p;
753 if (ioctl(gre_descriptors.ioctl_fd, SIOCDELGRETAP, &ifr) < 0) {
754 VLOG_WARN("couldn't do gre ioctl: %s\n", strerror(errno));
762 destroy_tap(struct netdev_dev_linux *netdev_dev)
764 struct tap_state *state = &netdev_dev->state.tap;
766 if (state->fd >= 0) {
772 destroy_gre(const char *name)
774 if (gre_descriptors.use_ioctl) {
775 return destroy_gre_ioctl(name);
777 return destroy_gre_netlink(name);
781 /* Destroys the netdev device 'netdev_dev_'. */
783 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
785 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
786 const char *type = netdev_dev_get_type(netdev_dev_);
788 if (!strcmp(type, "system")) {
789 shash_delete(&cache_map, netdev_dev->shash_node);
791 if (shash_is_empty(&cache_map)) {
792 rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
794 } else if (!strcmp(type, "tap")) {
795 destroy_tap(netdev_dev);
796 } else if (!strcmp(type, "gre")) {
797 destroy_gre(netdev_dev_get_name(&netdev_dev->netdev_dev));
804 netdev_linux_open(struct netdev_dev *netdev_dev, int ethertype,
805 struct netdev **netdevp)
807 struct netdev_linux *netdev;
808 enum netdev_flags flags;
811 /* Allocate network device. */
812 netdev = xzalloc(sizeof *netdev);
813 netdev_init(&netdev->netdev, netdev_dev);
814 netdev->netdev_fd = -1;
817 if (!strcmp(netdev_dev_get_type(netdev_dev), "tap")) {
818 static const char tap_dev[] = "/dev/net/tun";
821 /* Open tap device. */
822 netdev->tap_fd = open(tap_dev, O_RDWR);
823 if (netdev->tap_fd < 0) {
825 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
829 /* Create tap device. */
830 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
831 strncpy(ifr.ifr_name, netdev_dev_get_name(netdev_dev),
832 sizeof ifr.ifr_name);
833 if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) {
834 VLOG_WARN("%s: creating tap device failed: %s",
835 netdev_dev_get_name(netdev_dev),
841 /* Make non-blocking. */
842 error = set_nonblocking(netdev->tap_fd);
848 error = netdev_get_flags(&netdev->netdev, &flags);
849 if (error == ENODEV) {
853 if (netdev->tap_fd >= 0 || ethertype != NETDEV_ETH_TYPE_NONE) {
854 struct sockaddr_ll sll;
858 /* Create file descriptor. */
859 protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
860 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
862 netdev->netdev_fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
863 if (netdev->netdev_fd < 0) {
867 if (netdev->tap_fd < 0) {
868 netdev->tap_fd = netdev->netdev_fd;
871 /* Set non-blocking mode. */
872 error = set_nonblocking(netdev->netdev_fd);
877 /* Get ethernet device index. */
878 error = get_ifindex(&netdev->netdev, &ifindex);
883 /* Bind to specific ethernet device. */
884 memset(&sll, 0, sizeof sll);
885 sll.sll_family = AF_PACKET;
886 sll.sll_ifindex = ifindex;
887 if (bind(netdev->netdev_fd,
888 (struct sockaddr *) &sll, sizeof sll) < 0) {
890 VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev),
895 /* Between the socket() and bind() calls above, the socket receives all
896 * packets of the requested type on all system interfaces. We do not
897 * want to receive that data, but there is no way to avoid it. So we
898 * must now drain out the receive queue. */
899 error = drain_rcvbuf(netdev->netdev_fd);
905 *netdevp = &netdev->netdev;
909 netdev_uninit(&netdev->netdev, true);
913 /* Closes and destroys 'netdev'. */
915 netdev_linux_close(struct netdev *netdev_)
917 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
919 if (netdev->netdev_fd >= 0) {
920 close(netdev->netdev_fd);
922 if (netdev->tap_fd >= 0 && netdev->netdev_fd != netdev->tap_fd) {
923 close(netdev->tap_fd);
928 /* Initializes 'svec' with a list of the names of all known network devices. */
930 netdev_linux_enumerate(struct svec *svec)
932 struct if_nameindex *names;
934 names = if_nameindex();
938 for (i = 0; names[i].if_name != NULL; i++) {
939 svec_add(svec, names[i].if_name);
941 if_freenameindex(names);
944 VLOG_WARN("could not obtain list of network device names: %s",
951 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
953 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
955 if (netdev->tap_fd < 0) {
956 /* Device was opened with NETDEV_ETH_TYPE_NONE. */
961 ssize_t retval = read(netdev->tap_fd, data, size);
964 } else if (errno != EINTR) {
965 if (errno != EAGAIN) {
966 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
967 strerror(errno), netdev_get_name(netdev_));
974 /* Registers with the poll loop to wake up from the next call to poll_block()
975 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
977 netdev_linux_recv_wait(struct netdev *netdev_)
979 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
980 if (netdev->tap_fd >= 0) {
981 poll_fd_wait(netdev->tap_fd, POLLIN);
985 /* Discards all packets waiting to be received from 'netdev'. */
987 netdev_linux_drain(struct netdev *netdev_)
989 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
990 if (netdev->tap_fd < 0 && netdev->netdev_fd < 0) {
992 } else if (netdev->tap_fd != netdev->netdev_fd) {
994 int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
995 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
999 drain_fd(netdev->tap_fd, ifr.ifr_qlen);
1002 return drain_rcvbuf(netdev->netdev_fd);
1006 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
1007 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
1008 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
1009 * the packet is too big or too small to transmit on the device.
1011 * The caller retains ownership of 'buffer' in all cases.
1013 * The kernel maintains a packet transmission queue, so the caller is not
1014 * expected to do additional queuing of packets. */
1016 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
1018 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1020 /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
1022 if (netdev->tap_fd < 0) {
1027 ssize_t retval = write(netdev->tap_fd, data, size);
1029 /* The Linux AF_PACKET implementation never blocks waiting for room
1030 * for packets, instead returning ENOBUFS. Translate this into
1031 * EAGAIN for the caller. */
1032 if (errno == ENOBUFS) {
1034 } else if (errno == EINTR) {
1036 } else if (errno != EAGAIN) {
1037 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
1038 netdev_get_name(netdev_), strerror(errno));
1041 } else if (retval != size) {
1042 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
1043 "%zu) on %s", retval, size, netdev_get_name(netdev_));
1051 /* Registers with the poll loop to wake up from the next call to poll_block()
1052 * when the packet transmission queue has sufficient room to transmit a packet
1053 * with netdev_send().
1055 * The kernel maintains a packet transmission queue, so the client is not
1056 * expected to do additional queuing of packets. Thus, this function is
1057 * unlikely to ever be used. It is included for completeness. */
1059 netdev_linux_send_wait(struct netdev *netdev_)
1061 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1062 if (netdev->tap_fd < 0 && netdev->netdev_fd < 0) {
1063 /* Nothing to do. */
1064 } else if (netdev->tap_fd == netdev->netdev_fd) {
1065 poll_fd_wait(netdev->tap_fd, POLLOUT);
1067 /* TAP device always accepts packets.*/
1068 poll_immediate_wake();
1072 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
1073 * otherwise a positive errno value. */
1075 netdev_linux_set_etheraddr(struct netdev *netdev_,
1076 const uint8_t mac[ETH_ADDR_LEN])
1078 struct netdev_dev_linux *netdev_dev =
1079 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1082 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)
1083 || !eth_addr_equals(netdev_dev->etheraddr, mac)) {
1084 error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
1086 netdev_dev->cache_valid |= VALID_ETHERADDR;
1087 memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
1095 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
1096 * free the returned buffer. */
1098 netdev_linux_get_etheraddr(const struct netdev *netdev_,
1099 uint8_t mac[ETH_ADDR_LEN])
1101 struct netdev_dev_linux *netdev_dev =
1102 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1103 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
1104 int error = get_etheraddr(netdev_get_name(netdev_),
1105 netdev_dev->etheraddr);
1109 netdev_dev->cache_valid |= VALID_ETHERADDR;
1111 memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
1115 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
1116 * in bytes, not including the hardware header; thus, this is typically 1500
1117 * bytes for Ethernet devices. */
1119 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
1121 struct netdev_dev_linux *netdev_dev =
1122 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1123 if (!(netdev_dev->cache_valid & VALID_MTU)) {
1127 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1128 SIOCGIFMTU, "SIOCGIFMTU");
1132 netdev_dev->mtu = ifr.ifr_mtu;
1133 netdev_dev->cache_valid |= VALID_MTU;
1135 *mtup = netdev_dev->mtu;
1139 /* Returns the ifindex of 'netdev', if successful, as a positive number.
1140 * On failure, returns a negative errno value. */
1142 netdev_linux_get_ifindex(const struct netdev *netdev)
1146 error = get_ifindex(netdev, &ifindex);
1147 return error ? -error : ifindex;
1151 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
1153 struct netdev_dev_linux *netdev_dev =
1154 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1159 if (!(netdev_dev->cache_valid & VALID_CARRIER)) {
1163 fn = xasprintf("/sys/class/net/%s/carrier",
1164 netdev_get_name(netdev_));
1165 fd = open(fn, O_RDONLY);
1168 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error));
1172 retval = read(fd, line, sizeof line);
1175 if (error == EINVAL) {
1176 /* This is the normal return value when we try to check carrier
1177 * if the network device is not up. */
1179 VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error));
1182 } else if (retval == 0) {
1184 VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn);
1188 if (line[0] != '0' && line[0] != '1') {
1190 VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)",
1194 netdev_dev->carrier = line[0] != '0';
1195 netdev_dev->cache_valid |= VALID_CARRIER;
1197 *carrier = netdev_dev->carrier;
1208 /* Check whether we can we use RTM_GETLINK to get network device statistics.
1209 * In pre-2.6.19 kernels, this was only available if wireless extensions were
1212 check_for_working_netlink_stats(void)
1214 /* Decide on the netdev_get_stats() implementation to use. Netlink is
1215 * preferable, so if that works, we'll use it. */
1216 int ifindex = do_get_ifindex("lo");
1218 VLOG_WARN("failed to get ifindex for lo, "
1219 "obtaining netdev stats from proc");
1222 struct netdev_stats stats;
1223 int error = get_stats_via_netlink(ifindex, &stats);
1225 VLOG_DBG("obtaining netdev stats via rtnetlink");
1228 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
1229 "via proc (you are probably running a pre-2.6.19 "
1230 "kernel)", strerror(error));
1236 /* Retrieves current device stats for 'netdev'.
1238 * XXX All of the members of struct netdev_stats are 64 bits wide, but on
1239 * 32-bit architectures the Linux network stats are only 32 bits. */
1241 netdev_linux_get_stats(const struct netdev *netdev_,
1242 struct netdev_stats *stats)
1244 struct netdev_dev_linux *netdev_dev =
1245 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1246 static int use_netlink_stats = -1;
1248 struct netdev_stats raw_stats;
1249 struct netdev_stats *collect_stats = stats;
1251 COVERAGE_INC(netdev_get_stats);
1253 if (!(netdev_dev->cache_valid & VALID_IS_INTERNAL)) {
1254 netdev_dev->is_internal = !strcmp(netdev_get_type(netdev_),
1257 if (!netdev_dev->is_internal) {
1258 struct ethtool_drvinfo drvinfo;
1260 memset(&drvinfo, 0, sizeof drvinfo);
1261 error = netdev_linux_do_ethtool(netdev_get_name(netdev_),
1262 (struct ethtool_cmd *)&drvinfo,
1264 "ETHTOOL_GDRVINFO");
1267 netdev_dev->is_internal = !strcmp(drvinfo.driver,
1272 netdev_dev->cache_valid |= VALID_IS_INTERNAL;
1275 if (netdev_dev->is_internal) {
1276 collect_stats = &raw_stats;
1279 if (use_netlink_stats < 0) {
1280 use_netlink_stats = check_for_working_netlink_stats();
1282 if (use_netlink_stats) {
1285 error = get_ifindex(netdev_, &ifindex);
1287 error = get_stats_via_netlink(ifindex, collect_stats);
1290 error = get_stats_via_proc(netdev_get_name(netdev_), collect_stats);
1293 /* If this port is an internal port then the transmit and receive stats
1294 * will appear to be swapped relative to the other ports since we are the
1295 * one sending the data, not a remote computer. For consistency, we swap
1296 * them back here. */
1297 if (netdev_dev->is_internal) {
1298 stats->rx_packets = raw_stats.tx_packets;
1299 stats->tx_packets = raw_stats.rx_packets;
1300 stats->rx_bytes = raw_stats.tx_bytes;
1301 stats->tx_bytes = raw_stats.rx_bytes;
1302 stats->rx_errors = raw_stats.tx_errors;
1303 stats->tx_errors = raw_stats.rx_errors;
1304 stats->rx_dropped = raw_stats.tx_dropped;
1305 stats->tx_dropped = raw_stats.rx_dropped;
1306 stats->multicast = raw_stats.multicast;
1307 stats->collisions = raw_stats.collisions;
1308 stats->rx_length_errors = 0;
1309 stats->rx_over_errors = 0;
1310 stats->rx_crc_errors = 0;
1311 stats->rx_frame_errors = 0;
1312 stats->rx_fifo_errors = 0;
1313 stats->rx_missed_errors = 0;
1314 stats->tx_aborted_errors = 0;
1315 stats->tx_carrier_errors = 0;
1316 stats->tx_fifo_errors = 0;
1317 stats->tx_heartbeat_errors = 0;
1318 stats->tx_window_errors = 0;
1324 /* Stores the features supported by 'netdev' into each of '*current',
1325 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
1326 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
1327 * successful, otherwise a positive errno value. */
1329 netdev_linux_get_features(struct netdev *netdev,
1330 uint32_t *current, uint32_t *advertised,
1331 uint32_t *supported, uint32_t *peer)
1333 struct ethtool_cmd ecmd;
1336 memset(&ecmd, 0, sizeof ecmd);
1337 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1338 ETHTOOL_GSET, "ETHTOOL_GSET");
1343 /* Supported features. */
1345 if (ecmd.supported & SUPPORTED_10baseT_Half) {
1346 *supported |= OFPPF_10MB_HD;
1348 if (ecmd.supported & SUPPORTED_10baseT_Full) {
1349 *supported |= OFPPF_10MB_FD;
1351 if (ecmd.supported & SUPPORTED_100baseT_Half) {
1352 *supported |= OFPPF_100MB_HD;
1354 if (ecmd.supported & SUPPORTED_100baseT_Full) {
1355 *supported |= OFPPF_100MB_FD;
1357 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
1358 *supported |= OFPPF_1GB_HD;
1360 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
1361 *supported |= OFPPF_1GB_FD;
1363 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
1364 *supported |= OFPPF_10GB_FD;
1366 if (ecmd.supported & SUPPORTED_TP) {
1367 *supported |= OFPPF_COPPER;
1369 if (ecmd.supported & SUPPORTED_FIBRE) {
1370 *supported |= OFPPF_FIBER;
1372 if (ecmd.supported & SUPPORTED_Autoneg) {
1373 *supported |= OFPPF_AUTONEG;
1375 if (ecmd.supported & SUPPORTED_Pause) {
1376 *supported |= OFPPF_PAUSE;
1378 if (ecmd.supported & SUPPORTED_Asym_Pause) {
1379 *supported |= OFPPF_PAUSE_ASYM;
1382 /* Advertised features. */
1384 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
1385 *advertised |= OFPPF_10MB_HD;
1387 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
1388 *advertised |= OFPPF_10MB_FD;
1390 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
1391 *advertised |= OFPPF_100MB_HD;
1393 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
1394 *advertised |= OFPPF_100MB_FD;
1396 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
1397 *advertised |= OFPPF_1GB_HD;
1399 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
1400 *advertised |= OFPPF_1GB_FD;
1402 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
1403 *advertised |= OFPPF_10GB_FD;
1405 if (ecmd.advertising & ADVERTISED_TP) {
1406 *advertised |= OFPPF_COPPER;
1408 if (ecmd.advertising & ADVERTISED_FIBRE) {
1409 *advertised |= OFPPF_FIBER;
1411 if (ecmd.advertising & ADVERTISED_Autoneg) {
1412 *advertised |= OFPPF_AUTONEG;
1414 if (ecmd.advertising & ADVERTISED_Pause) {
1415 *advertised |= OFPPF_PAUSE;
1417 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
1418 *advertised |= OFPPF_PAUSE_ASYM;
1421 /* Current settings. */
1422 if (ecmd.speed == SPEED_10) {
1423 *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD;
1424 } else if (ecmd.speed == SPEED_100) {
1425 *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD;
1426 } else if (ecmd.speed == SPEED_1000) {
1427 *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD;
1428 } else if (ecmd.speed == SPEED_10000) {
1429 *current = OFPPF_10GB_FD;
1434 if (ecmd.port == PORT_TP) {
1435 *current |= OFPPF_COPPER;
1436 } else if (ecmd.port == PORT_FIBRE) {
1437 *current |= OFPPF_FIBER;
1441 *current |= OFPPF_AUTONEG;
1444 /* Peer advertisements. */
1445 *peer = 0; /* XXX */
1450 /* Set the features advertised by 'netdev' to 'advertise'. */
1452 netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
1454 struct ethtool_cmd ecmd;
1457 memset(&ecmd, 0, sizeof ecmd);
1458 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1459 ETHTOOL_GSET, "ETHTOOL_GSET");
1464 ecmd.advertising = 0;
1465 if (advertise & OFPPF_10MB_HD) {
1466 ecmd.advertising |= ADVERTISED_10baseT_Half;
1468 if (advertise & OFPPF_10MB_FD) {
1469 ecmd.advertising |= ADVERTISED_10baseT_Full;
1471 if (advertise & OFPPF_100MB_HD) {
1472 ecmd.advertising |= ADVERTISED_100baseT_Half;
1474 if (advertise & OFPPF_100MB_FD) {
1475 ecmd.advertising |= ADVERTISED_100baseT_Full;
1477 if (advertise & OFPPF_1GB_HD) {
1478 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1480 if (advertise & OFPPF_1GB_FD) {
1481 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1483 if (advertise & OFPPF_10GB_FD) {
1484 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1486 if (advertise & OFPPF_COPPER) {
1487 ecmd.advertising |= ADVERTISED_TP;
1489 if (advertise & OFPPF_FIBER) {
1490 ecmd.advertising |= ADVERTISED_FIBRE;
1492 if (advertise & OFPPF_AUTONEG) {
1493 ecmd.advertising |= ADVERTISED_Autoneg;
1495 if (advertise & OFPPF_PAUSE) {
1496 ecmd.advertising |= ADVERTISED_Pause;
1498 if (advertise & OFPPF_PAUSE_ASYM) {
1499 ecmd.advertising |= ADVERTISED_Asym_Pause;
1501 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1502 ETHTOOL_SSET, "ETHTOOL_SSET");
1505 /* If 'netdev_name' is the name of a VLAN network device (e.g. one created with
1506 * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device
1507 * and returns 0. Otherwise returns a errno value (specifically ENOENT if
1508 * 'netdev_name' is the name of a network device that is not a VLAN device) and
1509 * sets '*vlan_vid' to -1. */
1511 netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1513 const char *netdev_name = netdev_get_name(netdev);
1514 struct ds line = DS_EMPTY_INITIALIZER;
1515 FILE *stream = NULL;
1519 COVERAGE_INC(netdev_get_vlan_vid);
1520 fn = xasprintf("/proc/net/vlan/%s", netdev_name);
1521 stream = fopen(fn, "r");
1527 if (ds_get_line(&line, stream)) {
1528 if (ferror(stream)) {
1530 VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno));
1533 VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn);
1538 if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) {
1540 VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"",
1541 fn, ds_cstr(&line));
1559 #define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress"
1560 #define POLICE_CONFIG_CMD "/sbin/tc filter add dev %s parent ffff: protocol ip prio 50 u32 match ip src 0.0.0.0/0 police rate %dkbit burst %dk mtu 65535 drop flowid :1"
1561 /* We redirect stderr to /dev/null because we often want to remove all
1562 * traffic control configuration on a port so its in a known state. If
1563 * this done when there is no such configuration, tc complains, so we just
1566 #define POLICE_DEL_CMD "/sbin/tc qdisc del dev %s handle ffff: ingress 2>/dev/null"
1568 /* Attempts to set input rate limiting (policing) policy. */
1570 netdev_linux_set_policing(struct netdev *netdev,
1571 uint32_t kbits_rate, uint32_t kbits_burst)
1573 const char *netdev_name = netdev_get_name(netdev);
1576 COVERAGE_INC(netdev_set_policing);
1579 /* Default to 10 kilobits if not specified. */
1583 /* xxx This should be more careful about only adding if it
1584 * xxx actually exists, as opposed to always deleting it. */
1585 snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1586 if (system(command) == -1) {
1587 VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1590 snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name);
1591 if (system(command) != 0) {
1592 VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name);
1596 snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name,
1597 kbits_rate, kbits_burst);
1598 if (system(command) != 0) {
1599 VLOG_WARN_RL(&rl, "%s: problem configuring policing",
1604 snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1605 if (system(command) == -1) {
1606 VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1614 netdev_linux_get_in4(const struct netdev *netdev_,
1615 struct in_addr *address, struct in_addr *netmask)
1617 struct netdev_dev_linux *netdev_dev =
1618 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1620 if (!(netdev_dev->cache_valid & VALID_IN4)) {
1623 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
1624 SIOCGIFADDR, "SIOCGIFADDR");
1629 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
1630 SIOCGIFNETMASK, "SIOCGIFNETMASK");
1635 netdev_dev->cache_valid |= VALID_IN4;
1637 *address = netdev_dev->address;
1638 *netmask = netdev_dev->netmask;
1639 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
1643 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
1644 struct in_addr netmask)
1646 struct netdev_dev_linux *netdev_dev =
1647 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1650 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
1652 netdev_dev->cache_valid |= VALID_IN4;
1653 netdev_dev->address = address;
1654 netdev_dev->netmask = netmask;
1655 if (address.s_addr != INADDR_ANY) {
1656 error = do_set_addr(netdev_, SIOCSIFNETMASK,
1657 "SIOCSIFNETMASK", netmask);
1664 parse_if_inet6_line(const char *line,
1665 struct in6_addr *in6, char ifname[16 + 1])
1667 uint8_t *s6 = in6->s6_addr;
1668 #define X8 "%2"SCNx8
1670 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
1671 "%*x %*x %*x %*x %16s\n",
1672 &s6[0], &s6[1], &s6[2], &s6[3],
1673 &s6[4], &s6[5], &s6[6], &s6[7],
1674 &s6[8], &s6[9], &s6[10], &s6[11],
1675 &s6[12], &s6[13], &s6[14], &s6[15],
1679 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
1680 * 'in6' is non-null) and returns true. Otherwise, returns false. */
1682 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
1684 struct netdev_dev_linux *netdev_dev =
1685 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1686 if (!(netdev_dev->cache_valid & VALID_IN6)) {
1690 netdev_dev->in6 = in6addr_any;
1692 file = fopen("/proc/net/if_inet6", "r");
1694 const char *name = netdev_get_name(netdev_);
1695 while (fgets(line, sizeof line, file)) {
1696 struct in6_addr in6;
1697 char ifname[16 + 1];
1698 if (parse_if_inet6_line(line, &in6, ifname)
1699 && !strcmp(name, ifname))
1701 netdev_dev->in6 = in6;
1707 netdev_dev->cache_valid |= VALID_IN6;
1709 *in6 = netdev_dev->in6;
1714 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
1716 struct sockaddr_in sin;
1717 memset(&sin, 0, sizeof sin);
1718 sin.sin_family = AF_INET;
1719 sin.sin_addr = addr;
1722 memset(sa, 0, sizeof *sa);
1723 memcpy(sa, &sin, sizeof sin);
1727 do_set_addr(struct netdev *netdev,
1728 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
1731 strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
1732 make_in4_sockaddr(&ifr.ifr_addr, addr);
1734 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
1738 /* Adds 'router' as a default IP gateway. */
1740 netdev_linux_add_router(struct netdev *netdev UNUSED, struct in_addr router)
1742 struct in_addr any = { INADDR_ANY };
1746 memset(&rt, 0, sizeof rt);
1747 make_in4_sockaddr(&rt.rt_dst, any);
1748 make_in4_sockaddr(&rt.rt_gateway, router);
1749 make_in4_sockaddr(&rt.rt_genmask, any);
1750 rt.rt_flags = RTF_UP | RTF_GATEWAY;
1751 COVERAGE_INC(netdev_add_router);
1752 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
1754 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
1760 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
1763 static const char fn[] = "/proc/net/route";
1768 *netdev_name = NULL;
1769 stream = fopen(fn, "r");
1770 if (stream == NULL) {
1771 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1776 while (fgets(line, sizeof line, stream)) {
1779 uint32_t dest, gateway, mask;
1780 int refcnt, metric, mtu;
1781 unsigned int flags, use, window, irtt;
1784 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
1786 iface, &dest, &gateway, &flags, &refcnt,
1787 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
1789 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
1793 if (!(flags & RTF_UP)) {
1794 /* Skip routes that aren't up. */
1798 /* The output of 'dest', 'mask', and 'gateway' were given in
1799 * network byte order, so we don't need need any endian
1800 * conversions here. */
1801 if ((dest & mask) == (host->s_addr & mask)) {
1803 /* The host is directly reachable. */
1804 next_hop->s_addr = 0;
1806 /* To reach the host, we must go through a gateway. */
1807 next_hop->s_addr = gateway;
1809 *netdev_name = xstrdup(iface);
1820 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1821 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1822 * returns 0. Otherwise, it returns a positive errno value; in particular,
1823 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
1825 netdev_linux_arp_lookup(const struct netdev *netdev,
1826 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
1829 struct sockaddr_in sin;
1832 memset(&r, 0, sizeof r);
1833 sin.sin_family = AF_INET;
1834 sin.sin_addr.s_addr = ip;
1836 memcpy(&r.arp_pa, &sin, sizeof sin);
1837 r.arp_ha.sa_family = ARPHRD_ETHER;
1839 strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
1840 COVERAGE_INC(netdev_arp_lookup);
1841 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
1843 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
1844 } else if (retval != ENXIO) {
1845 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
1846 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
1852 nd_to_iff_flags(enum netdev_flags nd)
1855 if (nd & NETDEV_UP) {
1858 if (nd & NETDEV_PROMISC) {
1865 iff_to_nd_flags(int iff)
1867 enum netdev_flags nd = 0;
1871 if (iff & IFF_PROMISC) {
1872 nd |= NETDEV_PROMISC;
1878 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
1879 enum netdev_flags on, enum netdev_flags *old_flagsp)
1881 int old_flags, new_flags;
1884 error = get_flags(netdev, &old_flags);
1886 *old_flagsp = iff_to_nd_flags(old_flags);
1887 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
1888 if (new_flags != old_flags) {
1889 error = set_flags(netdev, new_flags);
1896 poll_notify(struct list *list)
1898 struct netdev_linux_notifier *notifier;
1899 LIST_FOR_EACH (notifier, struct netdev_linux_notifier, node, list) {
1900 struct netdev_notifier *n = ¬ifier->notifier;
1906 netdev_linux_poll_cb(const struct rtnetlink_change *change,
1910 struct list *list = shash_find_data(&netdev_linux_notifiers,
1916 struct shash_node *node;
1917 SHASH_FOR_EACH (node, &netdev_linux_notifiers) {
1918 poll_notify(node->data);
1924 netdev_linux_poll_add(struct netdev *netdev,
1925 void (*cb)(struct netdev_notifier *), void *aux,
1926 struct netdev_notifier **notifierp)
1928 const char *netdev_name = netdev_get_name(netdev);
1929 struct netdev_linux_notifier *notifier;
1932 if (shash_is_empty(&netdev_linux_notifiers)) {
1933 int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier,
1934 netdev_linux_poll_cb, NULL);
1940 list = shash_find_data(&netdev_linux_notifiers, netdev_name);
1942 list = xmalloc(sizeof *list);
1944 shash_add(&netdev_linux_notifiers, netdev_name, list);
1947 notifier = xmalloc(sizeof *notifier);
1948 netdev_notifier_init(¬ifier->notifier, netdev, cb, aux);
1949 list_push_back(list, ¬ifier->node);
1950 *notifierp = ¬ifier->notifier;
1955 netdev_linux_poll_remove(struct netdev_notifier *notifier_)
1957 struct netdev_linux_notifier *notifier =
1958 CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier);
1961 /* Remove 'notifier' from its list. */
1962 list = list_remove(¬ifier->node);
1963 if (list_is_empty(list)) {
1964 /* The list is now empty. Remove it from the hash and free it. */
1965 const char *netdev_name = netdev_get_name(notifier->notifier.netdev);
1966 shash_delete(&netdev_linux_notifiers,
1967 shash_find(&netdev_linux_notifiers, netdev_name));
1972 /* If that was the last notifier, unregister. */
1973 if (shash_is_empty(&netdev_linux_notifiers)) {
1974 rtnetlink_notifier_unregister(&netdev_linux_poll_notifier);
1978 const struct netdev_class netdev_linux_class = {
1985 netdev_linux_create_system,
1986 netdev_linux_destroy,
1987 NULL, /* reconfigure */
1992 netdev_linux_enumerate,
1995 netdev_linux_recv_wait,
1999 netdev_linux_send_wait,
2001 netdev_linux_set_etheraddr,
2002 netdev_linux_get_etheraddr,
2003 netdev_linux_get_mtu,
2004 netdev_linux_get_ifindex,
2005 netdev_linux_get_carrier,
2006 netdev_linux_get_stats,
2008 netdev_linux_get_features,
2009 netdev_linux_set_advertisements,
2010 netdev_linux_get_vlan_vid,
2011 netdev_linux_set_policing,
2013 netdev_linux_get_in4,
2014 netdev_linux_set_in4,
2015 netdev_linux_get_in6,
2016 netdev_linux_add_router,
2017 netdev_linux_get_next_hop,
2018 netdev_linux_arp_lookup,
2020 netdev_linux_update_flags,
2022 netdev_linux_poll_add,
2023 netdev_linux_poll_remove,
2026 const struct netdev_class netdev_tap_class = {
2033 netdev_linux_create_tap,
2034 netdev_linux_destroy,
2035 NULL, /* reconfigure */
2040 NULL, /* enumerate */
2043 netdev_linux_recv_wait,
2047 netdev_linux_send_wait,
2049 netdev_linux_set_etheraddr,
2050 netdev_linux_get_etheraddr,
2051 netdev_linux_get_mtu,
2052 netdev_linux_get_ifindex,
2053 netdev_linux_get_carrier,
2054 netdev_linux_get_stats,
2056 netdev_linux_get_features,
2057 netdev_linux_set_advertisements,
2058 netdev_linux_get_vlan_vid,
2059 netdev_linux_set_policing,
2061 netdev_linux_get_in4,
2062 netdev_linux_set_in4,
2063 netdev_linux_get_in6,
2064 netdev_linux_add_router,
2065 netdev_linux_get_next_hop,
2066 netdev_linux_arp_lookup,
2068 netdev_linux_update_flags,
2070 netdev_linux_poll_add,
2071 netdev_linux_poll_remove,
2074 const struct netdev_class netdev_gre_class = {
2081 netdev_linux_create_gre,
2082 netdev_linux_destroy,
2083 netdev_linux_reconfigure_gre,
2088 NULL, /* enumerate */
2091 netdev_linux_recv_wait,
2095 netdev_linux_send_wait,
2097 netdev_linux_set_etheraddr,
2098 netdev_linux_get_etheraddr,
2099 netdev_linux_get_mtu,
2100 netdev_linux_get_ifindex,
2101 netdev_linux_get_carrier,
2102 netdev_linux_get_stats,
2104 netdev_linux_get_features,
2105 netdev_linux_set_advertisements,
2106 netdev_linux_get_vlan_vid,
2107 netdev_linux_set_policing,
2109 netdev_linux_get_in4,
2110 netdev_linux_set_in4,
2111 netdev_linux_get_in6,
2112 netdev_linux_add_router,
2113 netdev_linux_get_next_hop,
2114 netdev_linux_arp_lookup,
2116 netdev_linux_update_flags,
2118 netdev_linux_poll_add,
2119 netdev_linux_poll_remove,
2123 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
2125 /* Policy for RTNLGRP_LINK messages.
2127 * There are *many* more fields in these messages, but currently we only
2128 * care about these fields. */
2129 static const struct nl_policy rtnlgrp_link_policy[] = {
2130 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
2131 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
2132 .min_len = sizeof(struct rtnl_link_stats) },
2136 static struct nl_sock *rtnl_sock;
2137 struct ofpbuf request;
2138 struct ofpbuf *reply;
2139 struct ifinfomsg *ifi;
2140 const struct rtnl_link_stats *rtnl_stats;
2141 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
2145 error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock);
2147 VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",
2153 ofpbuf_init(&request, 0);
2154 nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *ifi,
2155 RTM_GETLINK, NLM_F_REQUEST);
2156 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
2157 ifi->ifi_family = PF_UNSPEC;
2158 ifi->ifi_index = ifindex;
2159 error = nl_sock_transact(rtnl_sock, &request, &reply);
2160 ofpbuf_uninit(&request);
2165 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
2166 rtnlgrp_link_policy,
2167 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
2168 ofpbuf_delete(reply);
2172 if (!attrs[IFLA_STATS]) {
2173 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
2174 ofpbuf_delete(reply);
2178 rtnl_stats = nl_attr_get(attrs[IFLA_STATS]);
2179 stats->rx_packets = rtnl_stats->rx_packets;
2180 stats->tx_packets = rtnl_stats->tx_packets;
2181 stats->rx_bytes = rtnl_stats->rx_bytes;
2182 stats->tx_bytes = rtnl_stats->tx_bytes;
2183 stats->rx_errors = rtnl_stats->rx_errors;
2184 stats->tx_errors = rtnl_stats->tx_errors;
2185 stats->rx_dropped = rtnl_stats->rx_dropped;
2186 stats->tx_dropped = rtnl_stats->tx_dropped;
2187 stats->multicast = rtnl_stats->multicast;
2188 stats->collisions = rtnl_stats->collisions;
2189 stats->rx_length_errors = rtnl_stats->rx_length_errors;
2190 stats->rx_over_errors = rtnl_stats->rx_over_errors;
2191 stats->rx_crc_errors = rtnl_stats->rx_crc_errors;
2192 stats->rx_frame_errors = rtnl_stats->rx_frame_errors;
2193 stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors;
2194 stats->rx_missed_errors = rtnl_stats->rx_missed_errors;
2195 stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors;
2196 stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors;
2197 stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors;
2198 stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors;
2199 stats->tx_window_errors = rtnl_stats->tx_window_errors;
2201 ofpbuf_delete(reply);
2207 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
2209 static const char fn[] = "/proc/net/dev";
2214 stream = fopen(fn, "r");
2216 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
2221 while (fgets(line, sizeof line, stream)) {
2224 #define X64 "%"SCNu64
2227 X64 X64 X64 X64 X64 X64 X64 "%*u"
2228 X64 X64 X64 X64 X64 X64 X64 "%*u",
2234 &stats->rx_fifo_errors,
2235 &stats->rx_frame_errors,
2241 &stats->tx_fifo_errors,
2243 &stats->tx_carrier_errors) != 15) {
2244 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
2245 } else if (!strcmp(devname, netdev_name)) {
2246 stats->rx_length_errors = UINT64_MAX;
2247 stats->rx_over_errors = UINT64_MAX;
2248 stats->rx_crc_errors = UINT64_MAX;
2249 stats->rx_missed_errors = UINT64_MAX;
2250 stats->tx_aborted_errors = UINT64_MAX;
2251 stats->tx_heartbeat_errors = UINT64_MAX;
2252 stats->tx_window_errors = UINT64_MAX;
2258 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
2264 get_flags(const struct netdev *netdev, int *flags)
2269 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS,
2271 *flags = ifr.ifr_flags;
2276 set_flags(struct netdev *netdev, int flags)
2280 ifr.ifr_flags = flags;
2281 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
2286 do_get_ifindex(const char *netdev_name)
2290 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2291 COVERAGE_INC(netdev_get_ifindex);
2292 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
2293 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
2294 netdev_name, strerror(errno));
2297 return ifr.ifr_ifindex;
2301 get_ifindex(const struct netdev *netdev_, int *ifindexp)
2303 struct netdev_dev_linux *netdev_dev =
2304 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2306 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
2307 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
2311 netdev_dev->cache_valid |= VALID_IFINDEX;
2312 netdev_dev->ifindex = ifindex;
2314 *ifindexp = netdev_dev->ifindex;
2319 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
2324 memset(&ifr, 0, sizeof ifr);
2325 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2326 COVERAGE_INC(netdev_get_hwaddr);
2327 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
2328 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
2329 netdev_name, strerror(errno));
2332 hwaddr_family = ifr.ifr_hwaddr.sa_family;
2333 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
2334 VLOG_WARN("%s device has unknown hardware address family %d",
2335 netdev_name, hwaddr_family);
2337 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
2342 set_etheraddr(const char *netdev_name, int hwaddr_family,
2343 const uint8_t mac[ETH_ADDR_LEN])
2347 memset(&ifr, 0, sizeof ifr);
2348 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2349 ifr.ifr_hwaddr.sa_family = hwaddr_family;
2350 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
2351 COVERAGE_INC(netdev_set_hwaddr);
2352 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
2353 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
2354 netdev_name, strerror(errno));
2361 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
2362 int cmd, const char *cmd_name)
2366 memset(&ifr, 0, sizeof ifr);
2367 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
2368 ifr.ifr_data = (caddr_t) ecmd;
2371 COVERAGE_INC(netdev_ethtool);
2372 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
2375 if (errno != EOPNOTSUPP) {
2376 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
2377 "failed: %s", cmd_name, name, strerror(errno));
2379 /* The device doesn't support this operation. That's pretty
2380 * common, so there's no point in logging anything. */
2387 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
2388 const char *cmd_name)
2390 strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
2391 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
2392 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
2400 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
2401 int cmd, const char *cmd_name)
2406 ifr.ifr_addr.sa_family = AF_INET;
2407 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
2409 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
2410 *ip = sin->sin_addr;