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;
118 /* An AF_INET socket (used for ioctl operations). */
119 static int af_inet_sock = -1;
134 struct nl_sock *nl_sock;
140 struct netdev_linux_notifier {
141 struct netdev_notifier notifier;
145 static struct shash netdev_linux_notifiers =
146 SHASH_INITIALIZER(&netdev_linux_notifiers);
147 static struct rtnetlink_notifier netdev_linux_poll_notifier;
149 /* This is set pretty low because we probably won't learn anything from the
150 * additional log messages. */
151 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
153 static int destroy_gre(const char *name);
154 static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
155 int cmd, const char *cmd_name);
156 static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
157 const char *cmd_name);
158 static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
159 int cmd, const char *cmd_name);
160 static int get_flags(const struct netdev *, int *flagsp);
161 static int set_flags(struct netdev *, int flags);
162 static int do_get_ifindex(const char *netdev_name);
163 static int get_ifindex(const struct netdev *, int *ifindexp);
164 static int do_set_addr(struct netdev *netdev,
165 int ioctl_nr, const char *ioctl_name,
166 struct in_addr addr);
167 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
168 static int set_etheraddr(const char *netdev_name, int hwaddr_family,
169 const uint8_t[ETH_ADDR_LEN]);
170 static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
171 static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
173 static struct netdev_dev_linux *
174 netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
176 const char *type = netdev_dev_get_type(netdev_dev);
177 assert(!strcmp(type, "system") || !strcmp(type, "tap")
178 || !strcmp(type, "gre"));
179 return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
182 static struct netdev_linux *
183 netdev_linux_cast(const struct netdev *netdev)
185 const char *type = netdev_get_type(netdev);
186 assert(!strcmp(type, "system") || !strcmp(type, "tap")
187 || !strcmp(type, "gre"));
188 return CONTAINER_OF(netdev, struct netdev_linux, netdev);
192 netdev_linux_init(void)
194 static int status = -1;
196 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
197 status = af_inet_sock >= 0 ? 0 : errno;
199 VLOG_ERR("failed to create inet socket: %s", strerror(status));
206 netdev_linux_run(void)
208 rtnetlink_notifier_run();
212 netdev_linux_wait(void)
214 rtnetlink_notifier_wait();
218 netdev_linux_cache_cb(const struct rtnetlink_change *change,
221 struct netdev_dev_linux *dev;
223 dev = shash_find_data(&cache_map, change->ifname);
225 dev->cache_valid = 0;
228 struct shash_node *node;
229 SHASH_FOR_EACH (node, &cache_map) {
231 dev->cache_valid = 0;
236 /* The arguments are marked as unused to prevent warnings on platforms where
237 * the Netlink interface isn't supported. */
239 setup_gre_netlink(const char *name UNUSED, struct gre_config *config UNUSED,
242 #ifdef GRE_IOCTL_ONLY
246 struct ofpbuf request, *reply;
247 unsigned int nl_flags;
248 struct ifinfomsg ifinfomsg;
249 struct nlattr *linkinfo_hdr;
250 struct nlattr *info_data_hdr;
253 uint8_t pmtudisc = 0;
255 VLOG_DBG("%s: attempting to create gre device using netlink", name);
257 if (!gre_descriptors.nl_sock) {
258 error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0,
259 &gre_descriptors.nl_sock);
261 VLOG_WARN("couldn't create netlink socket: %s", strerror(error));
266 ofpbuf_init(&request, 0);
268 nl_flags = NLM_F_REQUEST;
270 nl_flags |= NLM_F_CREATE|NLM_F_EXCL;
273 /* We over-reserve space, because we do some pointer arithmetic
274 * and don't want the buffer address shifting under us. */
275 nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock, 2048, RTM_NEWLINK,
278 memset(&ifinfomsg, 0, sizeof ifinfomsg);
279 ifinfomsg.ifi_family = AF_UNSPEC;
280 nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
282 linkinfo_hdr = ofpbuf_tail(&request);
283 nl_msg_put_unspec(&request, IFLA_LINKINFO, NULL, 0);
285 nl_msg_put_unspec(&request, IFLA_INFO_KIND, "gretap", 6);
287 info_data_hdr = ofpbuf_tail(&request);
288 nl_msg_put_unspec(&request, IFLA_INFO_DATA, NULL, 0);
291 if (config->have_in_key) {
294 if (config->have_out_key) {
298 if (config->in_csum) {
301 if (config->out_csum) {
306 nl_msg_put_u32(&request, IFLA_GRE_IKEY, config->in_key);
307 nl_msg_put_u32(&request, IFLA_GRE_OKEY, config->out_key);
308 nl_msg_put_u16(&request, IFLA_GRE_IFLAGS, iflags);
309 nl_msg_put_u16(&request, IFLA_GRE_OFLAGS, oflags);
310 nl_msg_put_u32(&request, IFLA_GRE_LOCAL, config->local_ip);
311 nl_msg_put_u32(&request, IFLA_GRE_REMOTE, config->remote_ip);
312 nl_msg_put_u8(&request, IFLA_GRE_PMTUDISC, pmtudisc);
313 nl_msg_put_u8(&request, IFLA_GRE_TTL, 0);
314 nl_msg_put_u8(&request, IFLA_GRE_TOS, 0);
316 info_data_hdr->nla_len = (char *)ofpbuf_tail(&request)
317 - (char *)info_data_hdr;
318 linkinfo_hdr->nla_len = (char *)ofpbuf_tail(&request)
319 - (char *)linkinfo_hdr;
321 nl_msg_put_string(&request, IFLA_IFNAME, name);
323 error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
324 ofpbuf_uninit(&request);
326 VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
329 ofpbuf_delete(reply);
337 setup_gre_ioctl(const char *name, struct gre_config *config, bool create)
339 struct ip_tunnel_parm p;
342 VLOG_DBG("%s: attempting to create gre device using ioctl", name);
344 memset(&p, 0, sizeof p);
346 strncpy(p.name, name, IFNAMSIZ);
350 p.iph.protocol = IPPROTO_GRE;
351 p.iph.saddr = config->local_ip;
352 p.iph.daddr = config->remote_ip;
354 if (config->have_in_key) {
355 p.i_flags |= GRE_KEY;
356 p.i_key = config->in_key;
358 if (config->have_out_key) {
359 p.o_flags |= GRE_KEY;
360 p.o_key = config->out_key;
363 if (config->in_csum) {
364 p.i_flags |= GRE_CSUM;
366 if (config->out_csum) {
367 p.o_flags |= GRE_CSUM;
370 strncpy(ifr.ifr_name, create ? GRE_IOCTL_DEVICE : name, IFNAMSIZ);
371 ifr.ifr_ifru.ifru_data = (void *)&p;
373 if (!gre_descriptors.ioctl_fd) {
374 gre_descriptors.ioctl_fd = socket(AF_INET, SOCK_DGRAM, 0);
375 if (gre_descriptors.ioctl_fd < 0) {
376 VLOG_WARN("couldn't create gre ioctl socket: %s", strerror(errno));
377 gre_descriptors.ioctl_fd = 0;
382 if (ioctl(gre_descriptors.ioctl_fd, create ? SIOCADDGRETAP : SIOCCHGGRETAP,
384 VLOG_WARN("couldn't do gre ioctl: %s", strerror(errno));
391 /* The arguments are marked as unused to prevent warnings on platforms where
392 * the Netlink interface isn't supported. */
394 check_gre_device_netlink(const char *name UNUSED)
396 #ifdef GRE_IOCTL_ONLY
399 static const struct nl_policy getlink_policy[] = {
400 [IFLA_LINKINFO] = { .type = NL_A_NESTED, .optional = false },
403 static const struct nl_policy linkinfo_policy[] = {
404 [IFLA_INFO_KIND] = { .type = NL_A_STRING, .optional = false },
409 struct ofpbuf request, *reply;
410 struct ifinfomsg ifinfomsg;
411 struct nlattr *getlink_attrs[ARRAY_SIZE(getlink_policy)];
412 struct nlattr *linkinfo_attrs[ARRAY_SIZE(linkinfo_policy)];
413 struct ofpbuf linkinfo;
414 const char *device_kind;
416 ofpbuf_init(&request, 0);
418 nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock,
419 NLMSG_LENGTH(sizeof ifinfomsg), RTM_GETLINK,
422 memset(&ifinfomsg, 0, sizeof ifinfomsg);
423 ifinfomsg.ifi_family = AF_UNSPEC;
424 ifinfomsg.ifi_index = do_get_ifindex(name);
425 nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
427 error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
428 ofpbuf_uninit(&request);
430 VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
434 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
435 getlink_policy, getlink_attrs,
436 ARRAY_SIZE(getlink_policy))) {
437 VLOG_WARN("received bad rtnl message (getlink policy)");
441 linkinfo.data = (void *)nl_attr_get(getlink_attrs[IFLA_LINKINFO]);
442 linkinfo.size = nl_attr_get_size(getlink_attrs[IFLA_LINKINFO]);
443 if (!nl_policy_parse(&linkinfo, 0, linkinfo_policy,
444 linkinfo_attrs, ARRAY_SIZE(linkinfo_policy))) {
445 VLOG_WARN("received bad rtnl message (linkinfo policy)");
449 device_kind = nl_attr_get_string(linkinfo_attrs[IFLA_INFO_KIND]);
450 ret = !strcmp(device_kind, "gretap");
453 ofpbuf_delete(reply);
459 check_gre_device_ioctl(const char *name)
461 struct ethtool_drvinfo drvinfo;
464 memset(&drvinfo, 0, sizeof drvinfo);
465 error = netdev_linux_do_ethtool(name, (struct ethtool_cmd *)&drvinfo,
466 ETHTOOL_GDRVINFO, "ETHTOOL_GDRVINFO");
468 return !error && !strcmp(drvinfo.driver, "ip_gre")
469 && !strcmp(drvinfo.bus_info, "gretap");
473 setup_gre(const char *name, const struct shash *args, bool create)
476 struct in_addr in_addr;
477 struct shash_node *node;
478 struct gre_config config;
480 memset(&config, 0, sizeof config);
481 config.in_csum = true;
482 config.out_csum = true;
484 SHASH_FOR_EACH (node, args) {
485 if (!strcmp(node->name, "remote_ip")) {
486 if (lookup_ip(node->data, &in_addr)) {
487 VLOG_WARN("bad 'remote_ip' for gre device %s ", name);
489 config.remote_ip = in_addr.s_addr;
491 } else if (!strcmp(node->name, "local_ip")) {
492 if (lookup_ip(node->data, &in_addr)) {
493 VLOG_WARN("bad 'local_ip' for gre device %s ", name);
495 config.local_ip = in_addr.s_addr;
497 } else if (!strcmp(node->name, "key")) {
498 config.have_in_key = true;
499 config.have_out_key = true;
500 config.in_key = htonl(atoi(node->data));
501 config.out_key = htonl(atoi(node->data));
502 } else if (!strcmp(node->name, "in_key")) {
503 config.have_in_key = true;
504 config.in_key = htonl(atoi(node->data));
505 } else if (!strcmp(node->name, "out_key")) {
506 config.have_out_key = true;
507 config.out_key = htonl(atoi(node->data));
508 } else if (!strcmp(node->name, "csum")) {
509 if (!strcmp(node->data, "false")) {
510 config.in_csum = false;
511 config.out_csum = false;
514 VLOG_WARN("unknown gre argument '%s'", node->name);
518 if (!config.remote_ip) {
519 VLOG_WARN("gre type requires valid 'remote_ip' argument");
524 if (!gre_descriptors.use_ioctl) {
525 error = setup_gre_netlink(name, &config, create);
526 if (error == EOPNOTSUPP) {
527 gre_descriptors.use_ioctl = true;
530 if (gre_descriptors.use_ioctl) {
531 error = setup_gre_ioctl(name, &config, create);
534 if (create && error == EEXIST) {
537 if (gre_descriptors.use_ioctl) {
538 gre_device = check_gre_device_ioctl(name);
540 gre_device = check_gre_device_netlink(name);
547 VLOG_WARN("replacing existing gre device %s", name);
548 error = destroy_gre(name);
553 if (gre_descriptors.use_ioctl) {
554 error = setup_gre_ioctl(name, &config, create);
556 error = setup_gre_netlink(name, &config, create);
564 /* Creates the netdev device of 'type' with 'name'. */
566 netdev_linux_create_system(const char *name, const char *type UNUSED,
567 const struct shash *args, struct netdev_dev **netdev_devp)
569 struct netdev_dev_linux *netdev_dev;
572 if (!shash_is_empty(args)) {
573 VLOG_WARN("%s: arguments for system devices should be empty", name);
576 if (shash_is_empty(&cache_map)) {
577 error = rtnetlink_notifier_register(&netdev_linux_cache_notifier,
578 netdev_linux_cache_cb, NULL);
584 netdev_dev = xzalloc(sizeof *netdev_dev);
585 netdev_dev->shash_node = shash_add(&cache_map, name, &netdev_dev);
587 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class);
588 *netdev_devp = &netdev_dev->netdev_dev;
592 /* For most types of netdevs we open the device for each call of
593 * netdev_open(). However, this is not the case with tap devices,
594 * since it is only possible to open the device once. In this
595 * situation we share a single file descriptor, and consequently
596 * buffers, across all readers. Therefore once data is read it will
597 * be unavailable to other reads for tap devices. */
599 netdev_linux_create_tap(const char *name, const char *type UNUSED,
600 const struct shash *args, struct netdev_dev **netdev_devp)
602 struct netdev_dev_linux *netdev_dev;
603 struct tap_state *state;
604 static const char tap_dev[] = "/dev/net/tun";
608 if (!shash_is_empty(args)) {
609 VLOG_WARN("%s: arguments for TAP devices should be empty", name);
612 netdev_dev = xzalloc(sizeof *netdev_dev);
613 state = &netdev_dev->state.tap;
615 /* Open tap device. */
616 state->fd = open(tap_dev, O_RDWR);
619 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
623 /* Create tap device. */
624 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
625 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
626 if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
627 VLOG_WARN("%s: creating tap device failed: %s", name,
633 /* Make non-blocking. */
634 error = set_nonblocking(state->fd);
639 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
640 *netdev_devp = &netdev_dev->netdev_dev;
649 if_up(const char *name)
653 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
654 ifr.ifr_flags = IFF_UP;
656 if (ioctl(af_inet_sock, SIOCSIFFLAGS, &ifr) == -1) {
657 VLOG_DBG_RL(&rl, "%s: failed to bring device up: %s",
658 name, strerror(errno));
666 netdev_linux_create_gre(const char *name, const char *type UNUSED,
667 const struct shash *args, struct netdev_dev **netdev_devp)
669 struct netdev_dev_linux *netdev_dev;
672 netdev_dev = xzalloc(sizeof *netdev_dev);
674 error = setup_gre(name, args, true);
684 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_gre_class);
685 *netdev_devp = &netdev_dev->netdev_dev;
694 netdev_linux_reconfigure_gre(struct netdev_dev *netdev_dev_,
695 const struct shash *args)
697 const char *name = netdev_dev_get_name(netdev_dev_);
699 return setup_gre(name, args, false);
702 /* The arguments are marked as unused to prevent warnings on platforms where
703 * the Netlink interface isn't supported. */
705 destroy_gre_netlink(const char *name UNUSED)
707 #ifdef GRE_IOCTL_ONLY
711 struct ofpbuf request, *reply;
712 struct ifinfomsg ifinfomsg;
715 ofpbuf_init(&request, 0);
717 nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock, 0, RTM_DELLINK,
720 memset(&ifinfomsg, 0, sizeof ifinfomsg);
721 ifinfomsg.ifi_family = AF_UNSPEC;
722 nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
724 ifindex = do_get_ifindex(name);
725 nl_msg_put_u32(&request, IFLA_LINK, ifindex);
727 nl_msg_put_string(&request, IFLA_IFNAME, name);
729 error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
730 ofpbuf_uninit(&request);
732 VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
735 ofpbuf_delete(reply);
743 destroy_gre_ioctl(const char *name)
745 struct ip_tunnel_parm p;
748 memset(&p, 0, sizeof p);
749 strncpy(p.name, name, IFNAMSIZ);
751 strncpy(ifr.ifr_name, name, IFNAMSIZ);
752 ifr.ifr_ifru.ifru_data = (void *)&p;
754 if (ioctl(gre_descriptors.ioctl_fd, SIOCDELGRETAP, &ifr) < 0) {
755 VLOG_WARN("couldn't do gre ioctl: %s\n", strerror(errno));
763 destroy_tap(struct netdev_dev_linux *netdev_dev)
765 struct tap_state *state = &netdev_dev->state.tap;
767 if (state->fd >= 0) {
773 destroy_gre(const char *name)
775 if (gre_descriptors.use_ioctl) {
776 return destroy_gre_ioctl(name);
778 return destroy_gre_netlink(name);
782 /* Destroys the netdev device 'netdev_dev_'. */
784 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
786 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
787 const char *type = netdev_dev_get_type(netdev_dev_);
789 if (!strcmp(type, "system")) {
790 shash_delete(&cache_map, netdev_dev->shash_node);
792 if (shash_is_empty(&cache_map)) {
793 rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
795 } else if (!strcmp(type, "tap")) {
796 destroy_tap(netdev_dev);
797 } else if (!strcmp(type, "gre")) {
798 destroy_gre(netdev_dev_get_name(&netdev_dev->netdev_dev));
805 netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype,
806 struct netdev **netdevp)
808 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
809 struct netdev_linux *netdev;
810 enum netdev_flags flags;
813 /* Allocate network device. */
814 netdev = xzalloc(sizeof *netdev);
815 netdev_init(&netdev->netdev, netdev_dev_);
817 error = netdev_get_flags(&netdev->netdev, &flags);
818 if (error == ENODEV) {
822 if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap")) {
823 netdev->fd = netdev_dev->state.tap.fd;
825 } else if (ethertype != NETDEV_ETH_TYPE_NONE) {
826 struct sockaddr_ll sll;
830 /* Create file descriptor. */
831 protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
832 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
834 netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
835 if (netdev->fd < 0) {
840 /* Set non-blocking mode. */
841 error = set_nonblocking(netdev->fd);
846 /* Get ethernet device index. */
847 error = get_ifindex(&netdev->netdev, &ifindex);
852 /* Bind to specific ethernet device. */
853 memset(&sll, 0, sizeof sll);
854 sll.sll_family = AF_PACKET;
855 sll.sll_ifindex = ifindex;
857 (struct sockaddr *) &sll, sizeof sll) < 0) {
859 VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_),
864 /* Between the socket() and bind() calls above, the socket receives all
865 * packets of the requested type on all system interfaces. We do not
866 * want to receive that data, but there is no way to avoid it. So we
867 * must now drain out the receive queue. */
868 error = drain_rcvbuf(netdev->fd);
874 *netdevp = &netdev->netdev;
878 netdev_uninit(&netdev->netdev, true);
882 /* Closes and destroys 'netdev'. */
884 netdev_linux_close(struct netdev *netdev_)
886 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
888 if (netdev->fd >= 0 && strcmp(netdev_get_type(netdev_), "tap")) {
894 /* Initializes 'svec' with a list of the names of all known network devices. */
896 netdev_linux_enumerate(struct svec *svec)
898 struct if_nameindex *names;
900 names = if_nameindex();
904 for (i = 0; names[i].if_name != NULL; i++) {
905 svec_add(svec, names[i].if_name);
907 if_freenameindex(names);
910 VLOG_WARN("could not obtain list of network device names: %s",
917 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
919 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
921 if (netdev->fd < 0) {
922 /* Device was opened with NETDEV_ETH_TYPE_NONE. */
927 ssize_t retval = read(netdev->fd, data, size);
930 } else if (errno != EINTR) {
931 if (errno != EAGAIN) {
932 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
933 strerror(errno), netdev_get_name(netdev_));
940 /* Registers with the poll loop to wake up from the next call to poll_block()
941 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
943 netdev_linux_recv_wait(struct netdev *netdev_)
945 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
946 if (netdev->fd >= 0) {
947 poll_fd_wait(netdev->fd, POLLIN);
951 /* Discards all packets waiting to be received from 'netdev'. */
953 netdev_linux_drain(struct netdev *netdev_)
955 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
956 if (netdev->fd < 0) {
958 } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
960 int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
961 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
965 drain_fd(netdev->fd, ifr.ifr_qlen);
968 return drain_rcvbuf(netdev->fd);
972 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
973 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
974 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
975 * the packet is too big or too small to transmit on the device.
977 * The caller retains ownership of 'buffer' in all cases.
979 * The kernel maintains a packet transmission queue, so the caller is not
980 * expected to do additional queuing of packets. */
982 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
984 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
986 /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
988 if (netdev->fd < 0) {
993 ssize_t retval = write(netdev->fd, data, size);
995 /* The Linux AF_PACKET implementation never blocks waiting for room
996 * for packets, instead returning ENOBUFS. Translate this into
997 * EAGAIN for the caller. */
998 if (errno == ENOBUFS) {
1000 } else if (errno == EINTR) {
1002 } else if (errno != EAGAIN) {
1003 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
1004 netdev_get_name(netdev_), strerror(errno));
1007 } else if (retval != size) {
1008 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
1009 "%zu) on %s", retval, size, netdev_get_name(netdev_));
1017 /* Registers with the poll loop to wake up from the next call to poll_block()
1018 * when the packet transmission queue has sufficient room to transmit a packet
1019 * with netdev_send().
1021 * The kernel maintains a packet transmission queue, so the client is not
1022 * expected to do additional queuing of packets. Thus, this function is
1023 * unlikely to ever be used. It is included for completeness. */
1025 netdev_linux_send_wait(struct netdev *netdev_)
1027 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1028 if (netdev->fd < 0) {
1029 /* Nothing to do. */
1030 } else if (strcmp(netdev_get_type(netdev_), "tap")) {
1031 poll_fd_wait(netdev->fd, POLLOUT);
1033 /* TAP device always accepts packets.*/
1034 poll_immediate_wake();
1038 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
1039 * otherwise a positive errno value. */
1041 netdev_linux_set_etheraddr(struct netdev *netdev_,
1042 const uint8_t mac[ETH_ADDR_LEN])
1044 struct netdev_dev_linux *netdev_dev =
1045 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1048 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)
1049 || !eth_addr_equals(netdev_dev->etheraddr, mac)) {
1050 error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
1052 netdev_dev->cache_valid |= VALID_ETHERADDR;
1053 memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
1061 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
1062 * free the returned buffer. */
1064 netdev_linux_get_etheraddr(const struct netdev *netdev_,
1065 uint8_t mac[ETH_ADDR_LEN])
1067 struct netdev_dev_linux *netdev_dev =
1068 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1069 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
1070 int error = get_etheraddr(netdev_get_name(netdev_),
1071 netdev_dev->etheraddr);
1075 netdev_dev->cache_valid |= VALID_ETHERADDR;
1077 memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
1081 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
1082 * in bytes, not including the hardware header; thus, this is typically 1500
1083 * bytes for Ethernet devices. */
1085 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
1087 struct netdev_dev_linux *netdev_dev =
1088 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1089 if (!(netdev_dev->cache_valid & VALID_MTU)) {
1093 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1094 SIOCGIFMTU, "SIOCGIFMTU");
1098 netdev_dev->mtu = ifr.ifr_mtu;
1099 netdev_dev->cache_valid |= VALID_MTU;
1101 *mtup = netdev_dev->mtu;
1105 /* Returns the ifindex of 'netdev', if successful, as a positive number.
1106 * On failure, returns a negative errno value. */
1108 netdev_linux_get_ifindex(const struct netdev *netdev)
1112 error = get_ifindex(netdev, &ifindex);
1113 return error ? -error : ifindex;
1117 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
1119 struct netdev_dev_linux *netdev_dev =
1120 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1125 if (!(netdev_dev->cache_valid & VALID_CARRIER)) {
1129 fn = xasprintf("/sys/class/net/%s/carrier",
1130 netdev_get_name(netdev_));
1131 fd = open(fn, O_RDONLY);
1134 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error));
1138 retval = read(fd, line, sizeof line);
1141 if (error == EINVAL) {
1142 /* This is the normal return value when we try to check carrier
1143 * if the network device is not up. */
1145 VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error));
1148 } else if (retval == 0) {
1150 VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn);
1154 if (line[0] != '0' && line[0] != '1') {
1156 VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)",
1160 netdev_dev->carrier = line[0] != '0';
1161 netdev_dev->cache_valid |= VALID_CARRIER;
1163 *carrier = netdev_dev->carrier;
1174 /* Check whether we can we use RTM_GETLINK to get network device statistics.
1175 * In pre-2.6.19 kernels, this was only available if wireless extensions were
1178 check_for_working_netlink_stats(void)
1180 /* Decide on the netdev_get_stats() implementation to use. Netlink is
1181 * preferable, so if that works, we'll use it. */
1182 int ifindex = do_get_ifindex("lo");
1184 VLOG_WARN("failed to get ifindex for lo, "
1185 "obtaining netdev stats from proc");
1188 struct netdev_stats stats;
1189 int error = get_stats_via_netlink(ifindex, &stats);
1191 VLOG_DBG("obtaining netdev stats via rtnetlink");
1194 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
1195 "via proc (you are probably running a pre-2.6.19 "
1196 "kernel)", strerror(error));
1202 /* Retrieves current device stats for 'netdev'.
1204 * XXX All of the members of struct netdev_stats are 64 bits wide, but on
1205 * 32-bit architectures the Linux network stats are only 32 bits. */
1207 netdev_linux_get_stats(const struct netdev *netdev_,
1208 struct netdev_stats *stats)
1210 struct netdev_dev_linux *netdev_dev =
1211 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1212 static int use_netlink_stats = -1;
1214 struct netdev_stats raw_stats;
1215 struct netdev_stats *collect_stats = stats;
1217 COVERAGE_INC(netdev_get_stats);
1219 if (!(netdev_dev->cache_valid & VALID_IS_INTERNAL)) {
1220 netdev_dev->is_internal = !strcmp(netdev_get_type(netdev_),
1223 if (!netdev_dev->is_internal) {
1224 struct ethtool_drvinfo drvinfo;
1226 memset(&drvinfo, 0, sizeof drvinfo);
1227 error = netdev_linux_do_ethtool(netdev_get_name(netdev_),
1228 (struct ethtool_cmd *)&drvinfo,
1230 "ETHTOOL_GDRVINFO");
1233 netdev_dev->is_internal = !strcmp(drvinfo.driver,
1238 netdev_dev->cache_valid |= VALID_IS_INTERNAL;
1241 if (netdev_dev->is_internal) {
1242 collect_stats = &raw_stats;
1245 if (use_netlink_stats < 0) {
1246 use_netlink_stats = check_for_working_netlink_stats();
1248 if (use_netlink_stats) {
1251 error = get_ifindex(netdev_, &ifindex);
1253 error = get_stats_via_netlink(ifindex, collect_stats);
1256 error = get_stats_via_proc(netdev_get_name(netdev_), collect_stats);
1259 /* If this port is an internal port then the transmit and receive stats
1260 * will appear to be swapped relative to the other ports since we are the
1261 * one sending the data, not a remote computer. For consistency, we swap
1262 * them back here. */
1263 if (netdev_dev->is_internal) {
1264 stats->rx_packets = raw_stats.tx_packets;
1265 stats->tx_packets = raw_stats.rx_packets;
1266 stats->rx_bytes = raw_stats.tx_bytes;
1267 stats->tx_bytes = raw_stats.rx_bytes;
1268 stats->rx_errors = raw_stats.tx_errors;
1269 stats->tx_errors = raw_stats.rx_errors;
1270 stats->rx_dropped = raw_stats.tx_dropped;
1271 stats->tx_dropped = raw_stats.rx_dropped;
1272 stats->multicast = raw_stats.multicast;
1273 stats->collisions = raw_stats.collisions;
1274 stats->rx_length_errors = 0;
1275 stats->rx_over_errors = 0;
1276 stats->rx_crc_errors = 0;
1277 stats->rx_frame_errors = 0;
1278 stats->rx_fifo_errors = 0;
1279 stats->rx_missed_errors = 0;
1280 stats->tx_aborted_errors = 0;
1281 stats->tx_carrier_errors = 0;
1282 stats->tx_fifo_errors = 0;
1283 stats->tx_heartbeat_errors = 0;
1284 stats->tx_window_errors = 0;
1290 /* Stores the features supported by 'netdev' into each of '*current',
1291 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
1292 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
1293 * successful, otherwise a positive errno value. */
1295 netdev_linux_get_features(struct netdev *netdev,
1296 uint32_t *current, uint32_t *advertised,
1297 uint32_t *supported, uint32_t *peer)
1299 struct ethtool_cmd ecmd;
1302 memset(&ecmd, 0, sizeof ecmd);
1303 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1304 ETHTOOL_GSET, "ETHTOOL_GSET");
1309 /* Supported features. */
1311 if (ecmd.supported & SUPPORTED_10baseT_Half) {
1312 *supported |= OFPPF_10MB_HD;
1314 if (ecmd.supported & SUPPORTED_10baseT_Full) {
1315 *supported |= OFPPF_10MB_FD;
1317 if (ecmd.supported & SUPPORTED_100baseT_Half) {
1318 *supported |= OFPPF_100MB_HD;
1320 if (ecmd.supported & SUPPORTED_100baseT_Full) {
1321 *supported |= OFPPF_100MB_FD;
1323 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
1324 *supported |= OFPPF_1GB_HD;
1326 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
1327 *supported |= OFPPF_1GB_FD;
1329 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
1330 *supported |= OFPPF_10GB_FD;
1332 if (ecmd.supported & SUPPORTED_TP) {
1333 *supported |= OFPPF_COPPER;
1335 if (ecmd.supported & SUPPORTED_FIBRE) {
1336 *supported |= OFPPF_FIBER;
1338 if (ecmd.supported & SUPPORTED_Autoneg) {
1339 *supported |= OFPPF_AUTONEG;
1341 if (ecmd.supported & SUPPORTED_Pause) {
1342 *supported |= OFPPF_PAUSE;
1344 if (ecmd.supported & SUPPORTED_Asym_Pause) {
1345 *supported |= OFPPF_PAUSE_ASYM;
1348 /* Advertised features. */
1350 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
1351 *advertised |= OFPPF_10MB_HD;
1353 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
1354 *advertised |= OFPPF_10MB_FD;
1356 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
1357 *advertised |= OFPPF_100MB_HD;
1359 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
1360 *advertised |= OFPPF_100MB_FD;
1362 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
1363 *advertised |= OFPPF_1GB_HD;
1365 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
1366 *advertised |= OFPPF_1GB_FD;
1368 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
1369 *advertised |= OFPPF_10GB_FD;
1371 if (ecmd.advertising & ADVERTISED_TP) {
1372 *advertised |= OFPPF_COPPER;
1374 if (ecmd.advertising & ADVERTISED_FIBRE) {
1375 *advertised |= OFPPF_FIBER;
1377 if (ecmd.advertising & ADVERTISED_Autoneg) {
1378 *advertised |= OFPPF_AUTONEG;
1380 if (ecmd.advertising & ADVERTISED_Pause) {
1381 *advertised |= OFPPF_PAUSE;
1383 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
1384 *advertised |= OFPPF_PAUSE_ASYM;
1387 /* Current settings. */
1388 if (ecmd.speed == SPEED_10) {
1389 *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD;
1390 } else if (ecmd.speed == SPEED_100) {
1391 *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD;
1392 } else if (ecmd.speed == SPEED_1000) {
1393 *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD;
1394 } else if (ecmd.speed == SPEED_10000) {
1395 *current = OFPPF_10GB_FD;
1400 if (ecmd.port == PORT_TP) {
1401 *current |= OFPPF_COPPER;
1402 } else if (ecmd.port == PORT_FIBRE) {
1403 *current |= OFPPF_FIBER;
1407 *current |= OFPPF_AUTONEG;
1410 /* Peer advertisements. */
1411 *peer = 0; /* XXX */
1416 /* Set the features advertised by 'netdev' to 'advertise'. */
1418 netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
1420 struct ethtool_cmd ecmd;
1423 memset(&ecmd, 0, sizeof ecmd);
1424 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1425 ETHTOOL_GSET, "ETHTOOL_GSET");
1430 ecmd.advertising = 0;
1431 if (advertise & OFPPF_10MB_HD) {
1432 ecmd.advertising |= ADVERTISED_10baseT_Half;
1434 if (advertise & OFPPF_10MB_FD) {
1435 ecmd.advertising |= ADVERTISED_10baseT_Full;
1437 if (advertise & OFPPF_100MB_HD) {
1438 ecmd.advertising |= ADVERTISED_100baseT_Half;
1440 if (advertise & OFPPF_100MB_FD) {
1441 ecmd.advertising |= ADVERTISED_100baseT_Full;
1443 if (advertise & OFPPF_1GB_HD) {
1444 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1446 if (advertise & OFPPF_1GB_FD) {
1447 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1449 if (advertise & OFPPF_10GB_FD) {
1450 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1452 if (advertise & OFPPF_COPPER) {
1453 ecmd.advertising |= ADVERTISED_TP;
1455 if (advertise & OFPPF_FIBER) {
1456 ecmd.advertising |= ADVERTISED_FIBRE;
1458 if (advertise & OFPPF_AUTONEG) {
1459 ecmd.advertising |= ADVERTISED_Autoneg;
1461 if (advertise & OFPPF_PAUSE) {
1462 ecmd.advertising |= ADVERTISED_Pause;
1464 if (advertise & OFPPF_PAUSE_ASYM) {
1465 ecmd.advertising |= ADVERTISED_Asym_Pause;
1467 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1468 ETHTOOL_SSET, "ETHTOOL_SSET");
1471 /* If 'netdev_name' is the name of a VLAN network device (e.g. one created with
1472 * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device
1473 * and returns 0. Otherwise returns a errno value (specifically ENOENT if
1474 * 'netdev_name' is the name of a network device that is not a VLAN device) and
1475 * sets '*vlan_vid' to -1. */
1477 netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1479 const char *netdev_name = netdev_get_name(netdev);
1480 struct ds line = DS_EMPTY_INITIALIZER;
1481 FILE *stream = NULL;
1485 COVERAGE_INC(netdev_get_vlan_vid);
1486 fn = xasprintf("/proc/net/vlan/%s", netdev_name);
1487 stream = fopen(fn, "r");
1493 if (ds_get_line(&line, stream)) {
1494 if (ferror(stream)) {
1496 VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno));
1499 VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn);
1504 if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) {
1506 VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"",
1507 fn, ds_cstr(&line));
1525 #define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress"
1526 #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"
1527 /* We redirect stderr to /dev/null because we often want to remove all
1528 * traffic control configuration on a port so its in a known state. If
1529 * this done when there is no such configuration, tc complains, so we just
1532 #define POLICE_DEL_CMD "/sbin/tc qdisc del dev %s handle ffff: ingress 2>/dev/null"
1534 /* Attempts to set input rate limiting (policing) policy. */
1536 netdev_linux_set_policing(struct netdev *netdev,
1537 uint32_t kbits_rate, uint32_t kbits_burst)
1539 const char *netdev_name = netdev_get_name(netdev);
1542 COVERAGE_INC(netdev_set_policing);
1545 /* Default to 10 kilobits if not specified. */
1549 /* xxx This should be more careful about only adding if it
1550 * xxx actually exists, as opposed to always deleting it. */
1551 snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1552 if (system(command) == -1) {
1553 VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1556 snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name);
1557 if (system(command) != 0) {
1558 VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name);
1562 snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name,
1563 kbits_rate, kbits_burst);
1564 if (system(command) != 0) {
1565 VLOG_WARN_RL(&rl, "%s: problem configuring policing",
1570 snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1571 if (system(command) == -1) {
1572 VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1580 netdev_linux_get_in4(const struct netdev *netdev_,
1581 struct in_addr *address, struct in_addr *netmask)
1583 struct netdev_dev_linux *netdev_dev =
1584 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1586 if (!(netdev_dev->cache_valid & VALID_IN4)) {
1589 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
1590 SIOCGIFADDR, "SIOCGIFADDR");
1595 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
1596 SIOCGIFNETMASK, "SIOCGIFNETMASK");
1601 netdev_dev->cache_valid |= VALID_IN4;
1603 *address = netdev_dev->address;
1604 *netmask = netdev_dev->netmask;
1605 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
1609 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
1610 struct in_addr netmask)
1612 struct netdev_dev_linux *netdev_dev =
1613 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1616 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
1618 netdev_dev->cache_valid |= VALID_IN4;
1619 netdev_dev->address = address;
1620 netdev_dev->netmask = netmask;
1621 if (address.s_addr != INADDR_ANY) {
1622 error = do_set_addr(netdev_, SIOCSIFNETMASK,
1623 "SIOCSIFNETMASK", netmask);
1630 parse_if_inet6_line(const char *line,
1631 struct in6_addr *in6, char ifname[16 + 1])
1633 uint8_t *s6 = in6->s6_addr;
1634 #define X8 "%2"SCNx8
1636 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
1637 "%*x %*x %*x %*x %16s\n",
1638 &s6[0], &s6[1], &s6[2], &s6[3],
1639 &s6[4], &s6[5], &s6[6], &s6[7],
1640 &s6[8], &s6[9], &s6[10], &s6[11],
1641 &s6[12], &s6[13], &s6[14], &s6[15],
1645 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
1646 * 'in6' is non-null) and returns true. Otherwise, returns false. */
1648 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
1650 struct netdev_dev_linux *netdev_dev =
1651 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1652 if (!(netdev_dev->cache_valid & VALID_IN6)) {
1656 netdev_dev->in6 = in6addr_any;
1658 file = fopen("/proc/net/if_inet6", "r");
1660 const char *name = netdev_get_name(netdev_);
1661 while (fgets(line, sizeof line, file)) {
1662 struct in6_addr in6;
1663 char ifname[16 + 1];
1664 if (parse_if_inet6_line(line, &in6, ifname)
1665 && !strcmp(name, ifname))
1667 netdev_dev->in6 = in6;
1673 netdev_dev->cache_valid |= VALID_IN6;
1675 *in6 = netdev_dev->in6;
1680 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
1682 struct sockaddr_in sin;
1683 memset(&sin, 0, sizeof sin);
1684 sin.sin_family = AF_INET;
1685 sin.sin_addr = addr;
1688 memset(sa, 0, sizeof *sa);
1689 memcpy(sa, &sin, sizeof sin);
1693 do_set_addr(struct netdev *netdev,
1694 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
1697 strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
1698 make_in4_sockaddr(&ifr.ifr_addr, addr);
1700 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
1704 /* Adds 'router' as a default IP gateway. */
1706 netdev_linux_add_router(struct netdev *netdev UNUSED, struct in_addr router)
1708 struct in_addr any = { INADDR_ANY };
1712 memset(&rt, 0, sizeof rt);
1713 make_in4_sockaddr(&rt.rt_dst, any);
1714 make_in4_sockaddr(&rt.rt_gateway, router);
1715 make_in4_sockaddr(&rt.rt_genmask, any);
1716 rt.rt_flags = RTF_UP | RTF_GATEWAY;
1717 COVERAGE_INC(netdev_add_router);
1718 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
1720 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
1726 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
1729 static const char fn[] = "/proc/net/route";
1734 *netdev_name = NULL;
1735 stream = fopen(fn, "r");
1736 if (stream == NULL) {
1737 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1742 while (fgets(line, sizeof line, stream)) {
1745 uint32_t dest, gateway, mask;
1746 int refcnt, metric, mtu;
1747 unsigned int flags, use, window, irtt;
1750 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
1752 iface, &dest, &gateway, &flags, &refcnt,
1753 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
1755 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
1759 if (!(flags & RTF_UP)) {
1760 /* Skip routes that aren't up. */
1764 /* The output of 'dest', 'mask', and 'gateway' were given in
1765 * network byte order, so we don't need need any endian
1766 * conversions here. */
1767 if ((dest & mask) == (host->s_addr & mask)) {
1769 /* The host is directly reachable. */
1770 next_hop->s_addr = 0;
1772 /* To reach the host, we must go through a gateway. */
1773 next_hop->s_addr = gateway;
1775 *netdev_name = xstrdup(iface);
1786 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1787 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1788 * returns 0. Otherwise, it returns a positive errno value; in particular,
1789 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
1791 netdev_linux_arp_lookup(const struct netdev *netdev,
1792 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
1795 struct sockaddr_in sin;
1798 memset(&r, 0, sizeof r);
1799 sin.sin_family = AF_INET;
1800 sin.sin_addr.s_addr = ip;
1802 memcpy(&r.arp_pa, &sin, sizeof sin);
1803 r.arp_ha.sa_family = ARPHRD_ETHER;
1805 strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
1806 COVERAGE_INC(netdev_arp_lookup);
1807 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
1809 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
1810 } else if (retval != ENXIO) {
1811 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
1812 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
1818 nd_to_iff_flags(enum netdev_flags nd)
1821 if (nd & NETDEV_UP) {
1824 if (nd & NETDEV_PROMISC) {
1831 iff_to_nd_flags(int iff)
1833 enum netdev_flags nd = 0;
1837 if (iff & IFF_PROMISC) {
1838 nd |= NETDEV_PROMISC;
1844 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
1845 enum netdev_flags on, enum netdev_flags *old_flagsp)
1847 int old_flags, new_flags;
1850 error = get_flags(netdev, &old_flags);
1852 *old_flagsp = iff_to_nd_flags(old_flags);
1853 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
1854 if (new_flags != old_flags) {
1855 error = set_flags(netdev, new_flags);
1862 poll_notify(struct list *list)
1864 struct netdev_linux_notifier *notifier;
1865 LIST_FOR_EACH (notifier, struct netdev_linux_notifier, node, list) {
1866 struct netdev_notifier *n = ¬ifier->notifier;
1872 netdev_linux_poll_cb(const struct rtnetlink_change *change,
1876 struct list *list = shash_find_data(&netdev_linux_notifiers,
1882 struct shash_node *node;
1883 SHASH_FOR_EACH (node, &netdev_linux_notifiers) {
1884 poll_notify(node->data);
1890 netdev_linux_poll_add(struct netdev *netdev,
1891 void (*cb)(struct netdev_notifier *), void *aux,
1892 struct netdev_notifier **notifierp)
1894 const char *netdev_name = netdev_get_name(netdev);
1895 struct netdev_linux_notifier *notifier;
1898 if (shash_is_empty(&netdev_linux_notifiers)) {
1899 int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier,
1900 netdev_linux_poll_cb, NULL);
1906 list = shash_find_data(&netdev_linux_notifiers, netdev_name);
1908 list = xmalloc(sizeof *list);
1910 shash_add(&netdev_linux_notifiers, netdev_name, list);
1913 notifier = xmalloc(sizeof *notifier);
1914 netdev_notifier_init(¬ifier->notifier, netdev, cb, aux);
1915 list_push_back(list, ¬ifier->node);
1916 *notifierp = ¬ifier->notifier;
1921 netdev_linux_poll_remove(struct netdev_notifier *notifier_)
1923 struct netdev_linux_notifier *notifier =
1924 CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier);
1927 /* Remove 'notifier' from its list. */
1928 list = list_remove(¬ifier->node);
1929 if (list_is_empty(list)) {
1930 /* The list is now empty. Remove it from the hash and free it. */
1931 const char *netdev_name = netdev_get_name(notifier->notifier.netdev);
1932 shash_delete(&netdev_linux_notifiers,
1933 shash_find(&netdev_linux_notifiers, netdev_name));
1938 /* If that was the last notifier, unregister. */
1939 if (shash_is_empty(&netdev_linux_notifiers)) {
1940 rtnetlink_notifier_unregister(&netdev_linux_poll_notifier);
1944 const struct netdev_class netdev_linux_class = {
1951 netdev_linux_create_system,
1952 netdev_linux_destroy,
1953 NULL, /* reconfigure */
1958 netdev_linux_enumerate,
1961 netdev_linux_recv_wait,
1965 netdev_linux_send_wait,
1967 netdev_linux_set_etheraddr,
1968 netdev_linux_get_etheraddr,
1969 netdev_linux_get_mtu,
1970 netdev_linux_get_ifindex,
1971 netdev_linux_get_carrier,
1972 netdev_linux_get_stats,
1974 netdev_linux_get_features,
1975 netdev_linux_set_advertisements,
1976 netdev_linux_get_vlan_vid,
1977 netdev_linux_set_policing,
1979 netdev_linux_get_in4,
1980 netdev_linux_set_in4,
1981 netdev_linux_get_in6,
1982 netdev_linux_add_router,
1983 netdev_linux_get_next_hop,
1984 netdev_linux_arp_lookup,
1986 netdev_linux_update_flags,
1988 netdev_linux_poll_add,
1989 netdev_linux_poll_remove,
1992 const struct netdev_class netdev_tap_class = {
1999 netdev_linux_create_tap,
2000 netdev_linux_destroy,
2001 NULL, /* reconfigure */
2006 NULL, /* enumerate */
2009 netdev_linux_recv_wait,
2013 netdev_linux_send_wait,
2015 netdev_linux_set_etheraddr,
2016 netdev_linux_get_etheraddr,
2017 netdev_linux_get_mtu,
2018 netdev_linux_get_ifindex,
2019 netdev_linux_get_carrier,
2020 netdev_linux_get_stats,
2022 netdev_linux_get_features,
2023 netdev_linux_set_advertisements,
2024 netdev_linux_get_vlan_vid,
2025 netdev_linux_set_policing,
2027 netdev_linux_get_in4,
2028 netdev_linux_set_in4,
2029 netdev_linux_get_in6,
2030 netdev_linux_add_router,
2031 netdev_linux_get_next_hop,
2032 netdev_linux_arp_lookup,
2034 netdev_linux_update_flags,
2036 netdev_linux_poll_add,
2037 netdev_linux_poll_remove,
2040 const struct netdev_class netdev_gre_class = {
2047 netdev_linux_create_gre,
2048 netdev_linux_destroy,
2049 netdev_linux_reconfigure_gre,
2054 NULL, /* enumerate */
2057 netdev_linux_recv_wait,
2061 netdev_linux_send_wait,
2063 netdev_linux_set_etheraddr,
2064 netdev_linux_get_etheraddr,
2065 netdev_linux_get_mtu,
2066 netdev_linux_get_ifindex,
2067 netdev_linux_get_carrier,
2068 netdev_linux_get_stats,
2070 netdev_linux_get_features,
2071 netdev_linux_set_advertisements,
2072 netdev_linux_get_vlan_vid,
2073 netdev_linux_set_policing,
2075 netdev_linux_get_in4,
2076 netdev_linux_set_in4,
2077 netdev_linux_get_in6,
2078 netdev_linux_add_router,
2079 netdev_linux_get_next_hop,
2080 netdev_linux_arp_lookup,
2082 netdev_linux_update_flags,
2084 netdev_linux_poll_add,
2085 netdev_linux_poll_remove,
2089 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
2091 /* Policy for RTNLGRP_LINK messages.
2093 * There are *many* more fields in these messages, but currently we only
2094 * care about these fields. */
2095 static const struct nl_policy rtnlgrp_link_policy[] = {
2096 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
2097 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
2098 .min_len = sizeof(struct rtnl_link_stats) },
2102 static struct nl_sock *rtnl_sock;
2103 struct ofpbuf request;
2104 struct ofpbuf *reply;
2105 struct ifinfomsg *ifi;
2106 const struct rtnl_link_stats *rtnl_stats;
2107 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
2111 error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock);
2113 VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",
2119 ofpbuf_init(&request, 0);
2120 nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *ifi,
2121 RTM_GETLINK, NLM_F_REQUEST);
2122 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
2123 ifi->ifi_family = PF_UNSPEC;
2124 ifi->ifi_index = ifindex;
2125 error = nl_sock_transact(rtnl_sock, &request, &reply);
2126 ofpbuf_uninit(&request);
2131 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
2132 rtnlgrp_link_policy,
2133 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
2134 ofpbuf_delete(reply);
2138 if (!attrs[IFLA_STATS]) {
2139 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
2140 ofpbuf_delete(reply);
2144 rtnl_stats = nl_attr_get(attrs[IFLA_STATS]);
2145 stats->rx_packets = rtnl_stats->rx_packets;
2146 stats->tx_packets = rtnl_stats->tx_packets;
2147 stats->rx_bytes = rtnl_stats->rx_bytes;
2148 stats->tx_bytes = rtnl_stats->tx_bytes;
2149 stats->rx_errors = rtnl_stats->rx_errors;
2150 stats->tx_errors = rtnl_stats->tx_errors;
2151 stats->rx_dropped = rtnl_stats->rx_dropped;
2152 stats->tx_dropped = rtnl_stats->tx_dropped;
2153 stats->multicast = rtnl_stats->multicast;
2154 stats->collisions = rtnl_stats->collisions;
2155 stats->rx_length_errors = rtnl_stats->rx_length_errors;
2156 stats->rx_over_errors = rtnl_stats->rx_over_errors;
2157 stats->rx_crc_errors = rtnl_stats->rx_crc_errors;
2158 stats->rx_frame_errors = rtnl_stats->rx_frame_errors;
2159 stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors;
2160 stats->rx_missed_errors = rtnl_stats->rx_missed_errors;
2161 stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors;
2162 stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors;
2163 stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors;
2164 stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors;
2165 stats->tx_window_errors = rtnl_stats->tx_window_errors;
2167 ofpbuf_delete(reply);
2173 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
2175 static const char fn[] = "/proc/net/dev";
2180 stream = fopen(fn, "r");
2182 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
2187 while (fgets(line, sizeof line, stream)) {
2190 #define X64 "%"SCNu64
2193 X64 X64 X64 X64 X64 X64 X64 "%*u"
2194 X64 X64 X64 X64 X64 X64 X64 "%*u",
2200 &stats->rx_fifo_errors,
2201 &stats->rx_frame_errors,
2207 &stats->tx_fifo_errors,
2209 &stats->tx_carrier_errors) != 15) {
2210 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
2211 } else if (!strcmp(devname, netdev_name)) {
2212 stats->rx_length_errors = UINT64_MAX;
2213 stats->rx_over_errors = UINT64_MAX;
2214 stats->rx_crc_errors = UINT64_MAX;
2215 stats->rx_missed_errors = UINT64_MAX;
2216 stats->tx_aborted_errors = UINT64_MAX;
2217 stats->tx_heartbeat_errors = UINT64_MAX;
2218 stats->tx_window_errors = UINT64_MAX;
2224 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
2230 get_flags(const struct netdev *netdev, int *flags)
2235 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS,
2237 *flags = ifr.ifr_flags;
2242 set_flags(struct netdev *netdev, int flags)
2246 ifr.ifr_flags = flags;
2247 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
2252 do_get_ifindex(const char *netdev_name)
2256 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2257 COVERAGE_INC(netdev_get_ifindex);
2258 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
2259 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
2260 netdev_name, strerror(errno));
2263 return ifr.ifr_ifindex;
2267 get_ifindex(const struct netdev *netdev_, int *ifindexp)
2269 struct netdev_dev_linux *netdev_dev =
2270 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2272 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
2273 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
2277 netdev_dev->cache_valid |= VALID_IFINDEX;
2278 netdev_dev->ifindex = ifindex;
2280 *ifindexp = netdev_dev->ifindex;
2285 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
2290 memset(&ifr, 0, sizeof ifr);
2291 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2292 COVERAGE_INC(netdev_get_hwaddr);
2293 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
2294 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
2295 netdev_name, strerror(errno));
2298 hwaddr_family = ifr.ifr_hwaddr.sa_family;
2299 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
2300 VLOG_WARN("%s device has unknown hardware address family %d",
2301 netdev_name, hwaddr_family);
2303 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
2308 set_etheraddr(const char *netdev_name, int hwaddr_family,
2309 const uint8_t mac[ETH_ADDR_LEN])
2313 memset(&ifr, 0, sizeof ifr);
2314 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2315 ifr.ifr_hwaddr.sa_family = hwaddr_family;
2316 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
2317 COVERAGE_INC(netdev_set_hwaddr);
2318 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
2319 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
2320 netdev_name, strerror(errno));
2327 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
2328 int cmd, const char *cmd_name)
2332 memset(&ifr, 0, sizeof ifr);
2333 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
2334 ifr.ifr_data = (caddr_t) ecmd;
2337 COVERAGE_INC(netdev_ethtool);
2338 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
2341 if (errno != EOPNOTSUPP) {
2342 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
2343 "failed: %s", cmd_name, name, strerror(errno));
2345 /* The device doesn't support this operation. That's pretty
2346 * common, so there's no point in logging anything. */
2353 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
2354 const char *cmd_name)
2356 strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
2357 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
2358 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
2366 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
2367 int cmd, const char *cmd_name)
2372 ifr.ifr_addr.sa_family = AF_INET;
2373 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
2375 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
2376 *ip = sin->sin_addr;