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 int cache_notifier_refcount;
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;
136 struct nl_sock *nl_sock;
142 struct netdev_linux_notifier {
143 struct netdev_notifier notifier;
147 static struct shash netdev_linux_notifiers =
148 SHASH_INITIALIZER(&netdev_linux_notifiers);
149 static struct rtnetlink_notifier netdev_linux_poll_notifier;
151 /* This is set pretty low because we probably won't learn anything from the
152 * additional log messages. */
153 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
155 static int destroy_gre(const char *name);
156 static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
157 int cmd, const char *cmd_name);
158 static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
159 const char *cmd_name);
160 static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
161 int cmd, const char *cmd_name);
162 static int get_flags(const struct netdev *, int *flagsp);
163 static int set_flags(struct netdev *, int flags);
164 static int do_get_ifindex(const char *netdev_name);
165 static int get_ifindex(const struct netdev *, int *ifindexp);
166 static int do_set_addr(struct netdev *netdev,
167 int ioctl_nr, const char *ioctl_name,
168 struct in_addr addr);
169 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
170 static int set_etheraddr(const char *netdev_name, int hwaddr_family,
171 const uint8_t[ETH_ADDR_LEN]);
172 static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
173 static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
175 static struct netdev_dev_linux *
176 netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
178 const char *type = netdev_dev_get_type(netdev_dev);
179 assert(!strcmp(type, "system") || !strcmp(type, "tap")
180 || !strcmp(type, "gre"));
181 return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
184 static struct netdev_linux *
185 netdev_linux_cast(const struct netdev *netdev)
187 const char *type = netdev_get_type(netdev);
188 assert(!strcmp(type, "system") || !strcmp(type, "tap")
189 || !strcmp(type, "gre"));
190 return CONTAINER_OF(netdev, struct netdev_linux, netdev);
194 netdev_linux_init(void)
196 static int status = -1;
198 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
199 status = af_inet_sock >= 0 ? 0 : errno;
201 VLOG_ERR("failed to create inet socket: %s", strerror(status));
208 netdev_linux_run(void)
210 rtnetlink_notifier_run();
214 netdev_linux_wait(void)
216 rtnetlink_notifier_wait();
220 netdev_linux_cache_cb(const struct rtnetlink_change *change,
221 void *aux OVS_UNUSED)
223 struct netdev_dev_linux *dev;
225 struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
227 dev = netdev_dev_linux_cast(base_dev);
228 dev->cache_valid = 0;
231 struct shash device_shash;
232 struct shash_node *node;
234 shash_init(&device_shash);
235 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
236 SHASH_FOR_EACH (node, &device_shash) {
238 dev->cache_valid = 0;
240 shash_destroy(&device_shash);
244 /* The arguments are marked as unused to prevent warnings on platforms where
245 * the Netlink interface isn't supported. */
247 setup_gre_netlink(const char *name OVS_UNUSED,
248 struct gre_config *config OVS_UNUSED, bool create OVS_UNUSED)
250 #ifdef GRE_IOCTL_ONLY
254 struct ofpbuf request, *reply;
255 unsigned int nl_flags;
256 struct ifinfomsg ifinfomsg;
257 struct nlattr *linkinfo_hdr;
258 struct nlattr *info_data_hdr;
262 VLOG_DBG("%s: attempting to create gre device using netlink", name);
264 if (!gre_descriptors.nl_sock) {
265 error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0,
266 &gre_descriptors.nl_sock);
268 VLOG_WARN("couldn't create netlink socket: %s", strerror(error));
273 ofpbuf_init(&request, 0);
275 nl_flags = NLM_F_REQUEST;
277 nl_flags |= NLM_F_CREATE|NLM_F_EXCL;
280 /* We over-reserve space, because we do some pointer arithmetic
281 * and don't want the buffer address shifting under us. */
282 nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock, 2048, RTM_NEWLINK,
285 memset(&ifinfomsg, 0, sizeof ifinfomsg);
286 ifinfomsg.ifi_family = AF_UNSPEC;
287 nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
289 linkinfo_hdr = ofpbuf_tail(&request);
290 nl_msg_put_unspec(&request, IFLA_LINKINFO, NULL, 0);
292 nl_msg_put_unspec(&request, IFLA_INFO_KIND, "gretap", 6);
294 info_data_hdr = ofpbuf_tail(&request);
295 nl_msg_put_unspec(&request, IFLA_INFO_DATA, NULL, 0);
298 if (config->have_in_key) {
301 if (config->have_out_key) {
305 if (config->in_csum) {
308 if (config->out_csum) {
313 nl_msg_put_u32(&request, IFLA_GRE_IKEY, config->in_key);
314 nl_msg_put_u32(&request, IFLA_GRE_OKEY, config->out_key);
315 nl_msg_put_u16(&request, IFLA_GRE_IFLAGS, iflags);
316 nl_msg_put_u16(&request, IFLA_GRE_OFLAGS, oflags);
317 nl_msg_put_u32(&request, IFLA_GRE_LOCAL, config->local_ip);
318 nl_msg_put_u32(&request, IFLA_GRE_REMOTE, config->remote_ip);
319 nl_msg_put_u8(&request, IFLA_GRE_PMTUDISC, config->pmtud);
320 nl_msg_put_u8(&request, IFLA_GRE_TTL, IPDEFTTL);
321 nl_msg_put_u8(&request, IFLA_GRE_TOS, config->tos);
323 info_data_hdr->nla_len = (char *)ofpbuf_tail(&request)
324 - (char *)info_data_hdr;
325 linkinfo_hdr->nla_len = (char *)ofpbuf_tail(&request)
326 - (char *)linkinfo_hdr;
328 nl_msg_put_string(&request, IFLA_IFNAME, name);
330 error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
331 ofpbuf_uninit(&request);
333 VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
336 ofpbuf_delete(reply);
344 setup_gre_ioctl(const char *name, struct gre_config *config, bool create)
346 struct ip_tunnel_parm p;
349 VLOG_DBG("%s: attempting to create gre device using ioctl", name);
351 memset(&p, 0, sizeof p);
353 strncpy(p.name, name, IFNAMSIZ);
357 p.iph.protocol = IPPROTO_GRE;
358 p.iph.saddr = config->local_ip;
359 p.iph.daddr = config->remote_ip;
360 p.iph.ttl = IPDEFTTL;
361 p.iph.tos = config->tos;
363 if (config->have_in_key) {
364 p.i_flags |= GRE_KEY;
365 p.i_key = config->in_key;
367 if (config->have_out_key) {
368 p.o_flags |= GRE_KEY;
369 p.o_key = config->out_key;
372 if (config->in_csum) {
373 p.i_flags |= GRE_CSUM;
375 if (config->out_csum) {
376 p.o_flags |= GRE_CSUM;
380 p.iph.frag_off = htons(IP_DONT_FRAGMENT);
383 strncpy(ifr.ifr_name, create ? GRE_IOCTL_DEVICE : name, IFNAMSIZ);
384 ifr.ifr_ifru.ifru_data = (void *)&p;
386 if (!gre_descriptors.ioctl_fd) {
387 gre_descriptors.ioctl_fd = socket(AF_INET, SOCK_DGRAM, 0);
388 if (gre_descriptors.ioctl_fd < 0) {
389 VLOG_WARN("couldn't create gre ioctl socket: %s", strerror(errno));
390 gre_descriptors.ioctl_fd = 0;
395 if (ioctl(gre_descriptors.ioctl_fd, create ? SIOCADDGRETAP : SIOCCHGGRETAP,
397 VLOG_WARN("couldn't do gre ioctl: %s", strerror(errno));
404 /* The arguments are marked as unused to prevent warnings on platforms where
405 * the Netlink interface isn't supported. */
407 check_gre_device_netlink(const char *name OVS_UNUSED)
409 #ifdef GRE_IOCTL_ONLY
412 static const struct nl_policy getlink_policy[] = {
413 [IFLA_LINKINFO] = { .type = NL_A_NESTED, .optional = false },
416 static const struct nl_policy linkinfo_policy[] = {
417 [IFLA_INFO_KIND] = { .type = NL_A_STRING, .optional = false },
422 struct ofpbuf request, *reply;
423 struct ifinfomsg ifinfomsg;
424 struct nlattr *getlink_attrs[ARRAY_SIZE(getlink_policy)];
425 struct nlattr *linkinfo_attrs[ARRAY_SIZE(linkinfo_policy)];
426 struct ofpbuf linkinfo;
427 const char *device_kind;
429 ofpbuf_init(&request, 0);
431 nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock,
432 NLMSG_LENGTH(sizeof ifinfomsg), RTM_GETLINK,
435 memset(&ifinfomsg, 0, sizeof ifinfomsg);
436 ifinfomsg.ifi_family = AF_UNSPEC;
437 ifinfomsg.ifi_index = do_get_ifindex(name);
438 nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
440 error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
441 ofpbuf_uninit(&request);
443 VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
447 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
448 getlink_policy, getlink_attrs,
449 ARRAY_SIZE(getlink_policy))) {
450 VLOG_WARN("received bad rtnl message (getlink policy)");
454 linkinfo.data = (void *)nl_attr_get(getlink_attrs[IFLA_LINKINFO]);
455 linkinfo.size = nl_attr_get_size(getlink_attrs[IFLA_LINKINFO]);
456 if (!nl_policy_parse(&linkinfo, 0, linkinfo_policy,
457 linkinfo_attrs, ARRAY_SIZE(linkinfo_policy))) {
458 VLOG_WARN("received bad rtnl message (linkinfo policy)");
462 device_kind = nl_attr_get_string(linkinfo_attrs[IFLA_INFO_KIND]);
463 ret = !strcmp(device_kind, "gretap");
466 ofpbuf_delete(reply);
472 check_gre_device_ioctl(const char *name)
474 struct ethtool_drvinfo drvinfo;
477 memset(&drvinfo, 0, sizeof drvinfo);
478 error = netdev_linux_do_ethtool(name, (struct ethtool_cmd *)&drvinfo,
479 ETHTOOL_GDRVINFO, "ETHTOOL_GDRVINFO");
481 return !error && !strcmp(drvinfo.driver, "ip_gre")
482 && !strcmp(drvinfo.bus_info, "gretap");
486 setup_gre(const char *name, const struct shash *args, bool create)
489 struct in_addr in_addr;
490 struct shash_node *node;
491 struct gre_config config;
493 memset(&config, 0, sizeof config);
494 config.in_csum = true;
495 config.out_csum = true;
498 SHASH_FOR_EACH (node, args) {
499 if (!strcmp(node->name, "remote_ip")) {
500 if (lookup_ip(node->data, &in_addr)) {
501 VLOG_WARN("bad 'remote_ip' for gre device %s ", name);
503 config.remote_ip = in_addr.s_addr;
505 } else if (!strcmp(node->name, "local_ip")) {
506 if (lookup_ip(node->data, &in_addr)) {
507 VLOG_WARN("bad 'local_ip' for gre device %s ", name);
509 config.local_ip = in_addr.s_addr;
511 } else if (!strcmp(node->name, "key")) {
512 config.have_in_key = true;
513 config.have_out_key = true;
514 config.in_key = htonl(atoi(node->data));
515 config.out_key = htonl(atoi(node->data));
516 } else if (!strcmp(node->name, "in_key")) {
517 config.have_in_key = true;
518 config.in_key = htonl(atoi(node->data));
519 } else if (!strcmp(node->name, "out_key")) {
520 config.have_out_key = true;
521 config.out_key = htonl(atoi(node->data));
522 } else if (!strcmp(node->name, "tos")) {
523 config.tos = atoi(node->data);
524 } else if (!strcmp(node->name, "csum")) {
525 if (!strcmp(node->data, "false")) {
526 config.in_csum = false;
527 config.out_csum = false;
529 } else if (!strcmp(node->name, "pmtud")) {
530 if (!strcmp(node->data, "false")) {
531 config.pmtud = false;
534 VLOG_WARN("unknown gre argument '%s'", node->name);
538 if (!config.remote_ip) {
539 VLOG_WARN("gre type requires valid 'remote_ip' argument");
544 if (!gre_descriptors.use_ioctl) {
545 error = setup_gre_netlink(name, &config, create);
546 if (error == EOPNOTSUPP) {
547 gre_descriptors.use_ioctl = true;
550 if (gre_descriptors.use_ioctl) {
551 error = setup_gre_ioctl(name, &config, create);
554 if (create && error == EEXIST) {
557 if (gre_descriptors.use_ioctl) {
558 gre_device = check_gre_device_ioctl(name);
560 gre_device = check_gre_device_netlink(name);
567 VLOG_WARN("replacing existing gre device %s", name);
568 error = destroy_gre(name);
573 if (gre_descriptors.use_ioctl) {
574 error = setup_gre_ioctl(name, &config, create);
576 error = setup_gre_netlink(name, &config, create);
584 /* Creates the netdev device of 'type' with 'name'. */
586 netdev_linux_create_system(const char *name, const char *type OVS_UNUSED,
587 const struct shash *args, struct netdev_dev **netdev_devp)
589 struct netdev_dev_linux *netdev_dev;
592 if (!shash_is_empty(args)) {
593 VLOG_WARN("%s: arguments for system devices should be empty", name);
596 if (!cache_notifier_refcount) {
597 error = rtnetlink_notifier_register(&netdev_linux_cache_notifier,
598 netdev_linux_cache_cb, NULL);
603 cache_notifier_refcount++;
605 netdev_dev = xzalloc(sizeof *netdev_dev);
606 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class);
608 *netdev_devp = &netdev_dev->netdev_dev;
612 /* For most types of netdevs we open the device for each call of
613 * netdev_open(). However, this is not the case with tap devices,
614 * since it is only possible to open the device once. In this
615 * situation we share a single file descriptor, and consequently
616 * buffers, across all readers. Therefore once data is read it will
617 * be unavailable to other reads for tap devices. */
619 netdev_linux_create_tap(const char *name, const char *type OVS_UNUSED,
620 const struct shash *args, struct netdev_dev **netdev_devp)
622 struct netdev_dev_linux *netdev_dev;
623 struct tap_state *state;
624 static const char tap_dev[] = "/dev/net/tun";
628 if (!shash_is_empty(args)) {
629 VLOG_WARN("%s: arguments for TAP devices should be empty", name);
632 netdev_dev = xzalloc(sizeof *netdev_dev);
633 state = &netdev_dev->state.tap;
635 /* Open tap device. */
636 state->fd = open(tap_dev, O_RDWR);
639 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
643 /* Create tap device. */
644 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
645 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
646 if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
647 VLOG_WARN("%s: creating tap device failed: %s", name,
653 /* Make non-blocking. */
654 error = set_nonblocking(state->fd);
659 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
660 *netdev_devp = &netdev_dev->netdev_dev;
669 if_up(const char *name)
673 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
674 ifr.ifr_flags = IFF_UP;
676 if (ioctl(af_inet_sock, SIOCSIFFLAGS, &ifr) == -1) {
677 VLOG_DBG_RL(&rl, "%s: failed to bring device up: %s",
678 name, strerror(errno));
686 netdev_linux_create_gre(const char *name, const char *type OVS_UNUSED,
687 const struct shash *args, struct netdev_dev **netdev_devp)
689 struct netdev_dev_linux *netdev_dev;
692 netdev_dev = xzalloc(sizeof *netdev_dev);
694 error = setup_gre(name, args, true);
704 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_gre_class);
705 *netdev_devp = &netdev_dev->netdev_dev;
714 netdev_linux_reconfigure_gre(struct netdev_dev *netdev_dev_,
715 const struct shash *args)
717 const char *name = netdev_dev_get_name(netdev_dev_);
719 return setup_gre(name, args, false);
722 /* The arguments are marked as unused to prevent warnings on platforms where
723 * the Netlink interface isn't supported. */
725 destroy_gre_netlink(const char *name OVS_UNUSED)
727 #ifdef GRE_IOCTL_ONLY
731 struct ofpbuf request, *reply;
732 struct ifinfomsg ifinfomsg;
735 ofpbuf_init(&request, 0);
737 nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock, 0, RTM_DELLINK,
740 memset(&ifinfomsg, 0, sizeof ifinfomsg);
741 ifinfomsg.ifi_family = AF_UNSPEC;
742 nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
744 ifindex = do_get_ifindex(name);
745 nl_msg_put_u32(&request, IFLA_LINK, ifindex);
747 nl_msg_put_string(&request, IFLA_IFNAME, name);
749 error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
750 ofpbuf_uninit(&request);
752 VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
755 ofpbuf_delete(reply);
763 destroy_gre_ioctl(const char *name)
765 struct ip_tunnel_parm p;
768 memset(&p, 0, sizeof p);
769 strncpy(p.name, name, IFNAMSIZ);
771 strncpy(ifr.ifr_name, name, IFNAMSIZ);
772 ifr.ifr_ifru.ifru_data = (void *)&p;
774 if (ioctl(gre_descriptors.ioctl_fd, SIOCDELGRETAP, &ifr) < 0) {
775 VLOG_WARN("couldn't do gre ioctl: %s\n", strerror(errno));
783 destroy_tap(struct netdev_dev_linux *netdev_dev)
785 struct tap_state *state = &netdev_dev->state.tap;
787 if (state->fd >= 0) {
793 destroy_gre(const char *name)
795 if (gre_descriptors.use_ioctl) {
796 return destroy_gre_ioctl(name);
798 return destroy_gre_netlink(name);
802 /* Destroys the netdev device 'netdev_dev_'. */
804 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
806 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
807 const char *type = netdev_dev_get_type(netdev_dev_);
809 if (!strcmp(type, "system")) {
810 cache_notifier_refcount--;
812 if (!cache_notifier_refcount) {
813 rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
815 } else if (!strcmp(type, "tap")) {
816 destroy_tap(netdev_dev);
817 } else if (!strcmp(type, "gre")) {
818 destroy_gre(netdev_dev_get_name(&netdev_dev->netdev_dev));
825 netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype,
826 struct netdev **netdevp)
828 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
829 struct netdev_linux *netdev;
830 enum netdev_flags flags;
833 /* Allocate network device. */
834 netdev = xzalloc(sizeof *netdev);
836 netdev_init(&netdev->netdev, netdev_dev_);
838 error = netdev_get_flags(&netdev->netdev, &flags);
839 if (error == ENODEV) {
843 if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap")) {
844 netdev->fd = netdev_dev->state.tap.fd;
845 } else if (ethertype != NETDEV_ETH_TYPE_NONE) {
846 struct sockaddr_ll sll;
850 /* Create file descriptor. */
851 protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
852 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
854 netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
855 if (netdev->fd < 0) {
860 /* Set non-blocking mode. */
861 error = set_nonblocking(netdev->fd);
866 /* Get ethernet device index. */
867 error = get_ifindex(&netdev->netdev, &ifindex);
872 /* Bind to specific ethernet device. */
873 memset(&sll, 0, sizeof sll);
874 sll.sll_family = AF_PACKET;
875 sll.sll_ifindex = ifindex;
877 (struct sockaddr *) &sll, sizeof sll) < 0) {
879 VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_),
884 /* Between the socket() and bind() calls above, the socket receives all
885 * packets of the requested type on all system interfaces. We do not
886 * want to receive that data, but there is no way to avoid it. So we
887 * must now drain out the receive queue. */
888 error = drain_rcvbuf(netdev->fd);
894 *netdevp = &netdev->netdev;
898 netdev_uninit(&netdev->netdev, true);
902 /* Closes and destroys 'netdev'. */
904 netdev_linux_close(struct netdev *netdev_)
906 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
908 if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
914 /* Initializes 'svec' with a list of the names of all known network devices. */
916 netdev_linux_enumerate(struct svec *svec)
918 struct if_nameindex *names;
920 names = if_nameindex();
924 for (i = 0; names[i].if_name != NULL; i++) {
925 svec_add(svec, names[i].if_name);
927 if_freenameindex(names);
930 VLOG_WARN("could not obtain list of network device names: %s",
937 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
939 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
941 if (netdev->fd < 0) {
942 /* Device was opened with NETDEV_ETH_TYPE_NONE. */
947 ssize_t retval = read(netdev->fd, data, size);
950 } else if (errno != EINTR) {
951 if (errno != EAGAIN) {
952 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
953 strerror(errno), netdev_get_name(netdev_));
960 /* Registers with the poll loop to wake up from the next call to poll_block()
961 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
963 netdev_linux_recv_wait(struct netdev *netdev_)
965 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
966 if (netdev->fd >= 0) {
967 poll_fd_wait(netdev->fd, POLLIN);
971 /* Discards all packets waiting to be received from 'netdev'. */
973 netdev_linux_drain(struct netdev *netdev_)
975 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
976 if (netdev->fd < 0) {
978 } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
980 int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
981 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
985 drain_fd(netdev->fd, ifr.ifr_qlen);
988 return drain_rcvbuf(netdev->fd);
992 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
993 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
994 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
995 * the packet is too big or too small to transmit on the device.
997 * The caller retains ownership of 'buffer' in all cases.
999 * The kernel maintains a packet transmission queue, so the caller is not
1000 * expected to do additional queuing of packets. */
1002 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
1004 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1006 /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
1008 if (netdev->fd < 0) {
1013 ssize_t retval = write(netdev->fd, data, size);
1015 /* The Linux AF_PACKET implementation never blocks waiting for room
1016 * for packets, instead returning ENOBUFS. Translate this into
1017 * EAGAIN for the caller. */
1018 if (errno == ENOBUFS) {
1020 } else if (errno == EINTR) {
1022 } else if (errno != EAGAIN) {
1023 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
1024 netdev_get_name(netdev_), strerror(errno));
1027 } else if (retval != size) {
1028 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
1029 "%zu) on %s", retval, size, netdev_get_name(netdev_));
1037 /* Registers with the poll loop to wake up from the next call to poll_block()
1038 * when the packet transmission queue has sufficient room to transmit a packet
1039 * with netdev_send().
1041 * The kernel maintains a packet transmission queue, so the client is not
1042 * expected to do additional queuing of packets. Thus, this function is
1043 * unlikely to ever be used. It is included for completeness. */
1045 netdev_linux_send_wait(struct netdev *netdev_)
1047 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1048 if (netdev->fd < 0) {
1049 /* Nothing to do. */
1050 } else if (strcmp(netdev_get_type(netdev_), "tap")) {
1051 poll_fd_wait(netdev->fd, POLLOUT);
1053 /* TAP device always accepts packets.*/
1054 poll_immediate_wake();
1058 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
1059 * otherwise a positive errno value. */
1061 netdev_linux_set_etheraddr(struct netdev *netdev_,
1062 const uint8_t mac[ETH_ADDR_LEN])
1064 struct netdev_dev_linux *netdev_dev =
1065 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1068 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)
1069 || !eth_addr_equals(netdev_dev->etheraddr, mac)) {
1070 error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
1072 netdev_dev->cache_valid |= VALID_ETHERADDR;
1073 memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
1081 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
1082 * free the returned buffer. */
1084 netdev_linux_get_etheraddr(const struct netdev *netdev_,
1085 uint8_t mac[ETH_ADDR_LEN])
1087 struct netdev_dev_linux *netdev_dev =
1088 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1089 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
1090 int error = get_etheraddr(netdev_get_name(netdev_),
1091 netdev_dev->etheraddr);
1095 netdev_dev->cache_valid |= VALID_ETHERADDR;
1097 memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
1101 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
1102 * in bytes, not including the hardware header; thus, this is typically 1500
1103 * bytes for Ethernet devices. */
1105 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
1107 struct netdev_dev_linux *netdev_dev =
1108 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1109 if (!(netdev_dev->cache_valid & VALID_MTU)) {
1113 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1114 SIOCGIFMTU, "SIOCGIFMTU");
1118 netdev_dev->mtu = ifr.ifr_mtu;
1119 netdev_dev->cache_valid |= VALID_MTU;
1121 *mtup = netdev_dev->mtu;
1125 /* Returns the ifindex of 'netdev', if successful, as a positive number.
1126 * On failure, returns a negative errno value. */
1128 netdev_linux_get_ifindex(const struct netdev *netdev)
1132 error = get_ifindex(netdev, &ifindex);
1133 return error ? -error : ifindex;
1137 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
1139 struct netdev_dev_linux *netdev_dev =
1140 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1145 if (!(netdev_dev->cache_valid & VALID_CARRIER)) {
1149 fn = xasprintf("/sys/class/net/%s/carrier",
1150 netdev_get_name(netdev_));
1151 fd = open(fn, O_RDONLY);
1154 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error));
1158 retval = read(fd, line, sizeof line);
1161 if (error == EINVAL) {
1162 /* This is the normal return value when we try to check carrier
1163 * if the network device is not up. */
1165 VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error));
1168 } else if (retval == 0) {
1170 VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn);
1174 if (line[0] != '0' && line[0] != '1') {
1176 VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)",
1180 netdev_dev->carrier = line[0] != '0';
1181 netdev_dev->cache_valid |= VALID_CARRIER;
1183 *carrier = netdev_dev->carrier;
1194 /* Check whether we can we use RTM_GETLINK to get network device statistics.
1195 * In pre-2.6.19 kernels, this was only available if wireless extensions were
1198 check_for_working_netlink_stats(void)
1200 /* Decide on the netdev_get_stats() implementation to use. Netlink is
1201 * preferable, so if that works, we'll use it. */
1202 int ifindex = do_get_ifindex("lo");
1204 VLOG_WARN("failed to get ifindex for lo, "
1205 "obtaining netdev stats from proc");
1208 struct netdev_stats stats;
1209 int error = get_stats_via_netlink(ifindex, &stats);
1211 VLOG_DBG("obtaining netdev stats via rtnetlink");
1214 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
1215 "via proc (you are probably running a pre-2.6.19 "
1216 "kernel)", strerror(error));
1222 /* Retrieves current device stats for 'netdev'.
1224 * XXX All of the members of struct netdev_stats are 64 bits wide, but on
1225 * 32-bit architectures the Linux network stats are only 32 bits. */
1227 netdev_linux_get_stats(const struct netdev *netdev_,
1228 struct netdev_stats *stats)
1230 struct netdev_dev_linux *netdev_dev =
1231 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1232 static int use_netlink_stats = -1;
1234 struct netdev_stats raw_stats;
1235 struct netdev_stats *collect_stats = stats;
1237 COVERAGE_INC(netdev_get_stats);
1239 if (!(netdev_dev->cache_valid & VALID_IS_INTERNAL)) {
1240 netdev_dev->is_internal = !strcmp(netdev_get_type(netdev_), "tap");
1241 if (!netdev_dev->is_internal) {
1242 struct ethtool_drvinfo drvinfo;
1244 memset(&drvinfo, 0, sizeof drvinfo);
1245 error = netdev_linux_do_ethtool(netdev_get_name(netdev_),
1246 (struct ethtool_cmd *)&drvinfo,
1248 "ETHTOOL_GDRVINFO");
1251 netdev_dev->is_internal = !strcmp(drvinfo.driver,
1256 netdev_dev->cache_valid |= VALID_IS_INTERNAL;
1259 if (netdev_dev->is_internal) {
1260 collect_stats = &raw_stats;
1263 if (use_netlink_stats < 0) {
1264 use_netlink_stats = check_for_working_netlink_stats();
1266 if (use_netlink_stats) {
1269 error = get_ifindex(netdev_, &ifindex);
1271 error = get_stats_via_netlink(ifindex, collect_stats);
1274 error = get_stats_via_proc(netdev_get_name(netdev_), collect_stats);
1277 /* If this port is an internal port then the transmit and receive stats
1278 * will appear to be swapped relative to the other ports since we are the
1279 * one sending the data, not a remote computer. For consistency, we swap
1280 * them back here. */
1281 if (!error && netdev_dev->is_internal) {
1282 stats->rx_packets = raw_stats.tx_packets;
1283 stats->tx_packets = raw_stats.rx_packets;
1284 stats->rx_bytes = raw_stats.tx_bytes;
1285 stats->tx_bytes = raw_stats.rx_bytes;
1286 stats->rx_errors = raw_stats.tx_errors;
1287 stats->tx_errors = raw_stats.rx_errors;
1288 stats->rx_dropped = raw_stats.tx_dropped;
1289 stats->tx_dropped = raw_stats.rx_dropped;
1290 stats->multicast = raw_stats.multicast;
1291 stats->collisions = raw_stats.collisions;
1292 stats->rx_length_errors = 0;
1293 stats->rx_over_errors = 0;
1294 stats->rx_crc_errors = 0;
1295 stats->rx_frame_errors = 0;
1296 stats->rx_fifo_errors = 0;
1297 stats->rx_missed_errors = 0;
1298 stats->tx_aborted_errors = 0;
1299 stats->tx_carrier_errors = 0;
1300 stats->tx_fifo_errors = 0;
1301 stats->tx_heartbeat_errors = 0;
1302 stats->tx_window_errors = 0;
1308 /* Stores the features supported by 'netdev' into each of '*current',
1309 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
1310 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
1311 * successful, otherwise a positive errno value. */
1313 netdev_linux_get_features(struct netdev *netdev,
1314 uint32_t *current, uint32_t *advertised,
1315 uint32_t *supported, uint32_t *peer)
1317 struct ethtool_cmd ecmd;
1320 memset(&ecmd, 0, sizeof ecmd);
1321 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1322 ETHTOOL_GSET, "ETHTOOL_GSET");
1327 /* Supported features. */
1329 if (ecmd.supported & SUPPORTED_10baseT_Half) {
1330 *supported |= OFPPF_10MB_HD;
1332 if (ecmd.supported & SUPPORTED_10baseT_Full) {
1333 *supported |= OFPPF_10MB_FD;
1335 if (ecmd.supported & SUPPORTED_100baseT_Half) {
1336 *supported |= OFPPF_100MB_HD;
1338 if (ecmd.supported & SUPPORTED_100baseT_Full) {
1339 *supported |= OFPPF_100MB_FD;
1341 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
1342 *supported |= OFPPF_1GB_HD;
1344 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
1345 *supported |= OFPPF_1GB_FD;
1347 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
1348 *supported |= OFPPF_10GB_FD;
1350 if (ecmd.supported & SUPPORTED_TP) {
1351 *supported |= OFPPF_COPPER;
1353 if (ecmd.supported & SUPPORTED_FIBRE) {
1354 *supported |= OFPPF_FIBER;
1356 if (ecmd.supported & SUPPORTED_Autoneg) {
1357 *supported |= OFPPF_AUTONEG;
1359 if (ecmd.supported & SUPPORTED_Pause) {
1360 *supported |= OFPPF_PAUSE;
1362 if (ecmd.supported & SUPPORTED_Asym_Pause) {
1363 *supported |= OFPPF_PAUSE_ASYM;
1366 /* Advertised features. */
1368 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
1369 *advertised |= OFPPF_10MB_HD;
1371 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
1372 *advertised |= OFPPF_10MB_FD;
1374 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
1375 *advertised |= OFPPF_100MB_HD;
1377 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
1378 *advertised |= OFPPF_100MB_FD;
1380 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
1381 *advertised |= OFPPF_1GB_HD;
1383 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
1384 *advertised |= OFPPF_1GB_FD;
1386 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
1387 *advertised |= OFPPF_10GB_FD;
1389 if (ecmd.advertising & ADVERTISED_TP) {
1390 *advertised |= OFPPF_COPPER;
1392 if (ecmd.advertising & ADVERTISED_FIBRE) {
1393 *advertised |= OFPPF_FIBER;
1395 if (ecmd.advertising & ADVERTISED_Autoneg) {
1396 *advertised |= OFPPF_AUTONEG;
1398 if (ecmd.advertising & ADVERTISED_Pause) {
1399 *advertised |= OFPPF_PAUSE;
1401 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
1402 *advertised |= OFPPF_PAUSE_ASYM;
1405 /* Current settings. */
1406 if (ecmd.speed == SPEED_10) {
1407 *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD;
1408 } else if (ecmd.speed == SPEED_100) {
1409 *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD;
1410 } else if (ecmd.speed == SPEED_1000) {
1411 *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD;
1412 } else if (ecmd.speed == SPEED_10000) {
1413 *current = OFPPF_10GB_FD;
1418 if (ecmd.port == PORT_TP) {
1419 *current |= OFPPF_COPPER;
1420 } else if (ecmd.port == PORT_FIBRE) {
1421 *current |= OFPPF_FIBER;
1425 *current |= OFPPF_AUTONEG;
1428 /* Peer advertisements. */
1429 *peer = 0; /* XXX */
1434 /* Set the features advertised by 'netdev' to 'advertise'. */
1436 netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
1438 struct ethtool_cmd ecmd;
1441 memset(&ecmd, 0, sizeof ecmd);
1442 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1443 ETHTOOL_GSET, "ETHTOOL_GSET");
1448 ecmd.advertising = 0;
1449 if (advertise & OFPPF_10MB_HD) {
1450 ecmd.advertising |= ADVERTISED_10baseT_Half;
1452 if (advertise & OFPPF_10MB_FD) {
1453 ecmd.advertising |= ADVERTISED_10baseT_Full;
1455 if (advertise & OFPPF_100MB_HD) {
1456 ecmd.advertising |= ADVERTISED_100baseT_Half;
1458 if (advertise & OFPPF_100MB_FD) {
1459 ecmd.advertising |= ADVERTISED_100baseT_Full;
1461 if (advertise & OFPPF_1GB_HD) {
1462 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1464 if (advertise & OFPPF_1GB_FD) {
1465 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1467 if (advertise & OFPPF_10GB_FD) {
1468 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1470 if (advertise & OFPPF_COPPER) {
1471 ecmd.advertising |= ADVERTISED_TP;
1473 if (advertise & OFPPF_FIBER) {
1474 ecmd.advertising |= ADVERTISED_FIBRE;
1476 if (advertise & OFPPF_AUTONEG) {
1477 ecmd.advertising |= ADVERTISED_Autoneg;
1479 if (advertise & OFPPF_PAUSE) {
1480 ecmd.advertising |= ADVERTISED_Pause;
1482 if (advertise & OFPPF_PAUSE_ASYM) {
1483 ecmd.advertising |= ADVERTISED_Asym_Pause;
1485 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1486 ETHTOOL_SSET, "ETHTOOL_SSET");
1489 /* If 'netdev_name' is the name of a VLAN network device (e.g. one created with
1490 * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device
1491 * and returns 0. Otherwise returns a errno value (specifically ENOENT if
1492 * 'netdev_name' is the name of a network device that is not a VLAN device) and
1493 * sets '*vlan_vid' to -1. */
1495 netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1497 const char *netdev_name = netdev_get_name(netdev);
1498 struct ds line = DS_EMPTY_INITIALIZER;
1499 FILE *stream = NULL;
1503 COVERAGE_INC(netdev_get_vlan_vid);
1504 fn = xasprintf("/proc/net/vlan/%s", netdev_name);
1505 stream = fopen(fn, "r");
1511 if (ds_get_line(&line, stream)) {
1512 if (ferror(stream)) {
1514 VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno));
1517 VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn);
1522 if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) {
1524 VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"",
1525 fn, ds_cstr(&line));
1543 #define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress"
1544 #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"
1545 /* We redirect stderr to /dev/null because we often want to remove all
1546 * traffic control configuration on a port so its in a known state. If
1547 * this done when there is no such configuration, tc complains, so we just
1550 #define POLICE_DEL_CMD "/sbin/tc qdisc del dev %s handle ffff: ingress 2>/dev/null"
1552 /* Attempts to set input rate limiting (policing) policy. */
1554 netdev_linux_set_policing(struct netdev *netdev,
1555 uint32_t kbits_rate, uint32_t kbits_burst)
1557 const char *netdev_name = netdev_get_name(netdev);
1560 COVERAGE_INC(netdev_set_policing);
1563 /* Default to 1000 kilobits if not specified. */
1567 /* xxx This should be more careful about only adding if it
1568 * xxx actually exists, as opposed to always deleting it. */
1569 snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1570 if (system(command) == -1) {
1571 VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1574 snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name);
1575 if (system(command) != 0) {
1576 VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name);
1580 snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name,
1581 kbits_rate, kbits_burst);
1582 if (system(command) != 0) {
1583 VLOG_WARN_RL(&rl, "%s: problem configuring policing",
1588 snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1589 if (system(command) == -1) {
1590 VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1598 netdev_linux_get_in4(const struct netdev *netdev_,
1599 struct in_addr *address, struct in_addr *netmask)
1601 struct netdev_dev_linux *netdev_dev =
1602 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1604 if (!(netdev_dev->cache_valid & VALID_IN4)) {
1607 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
1608 SIOCGIFADDR, "SIOCGIFADDR");
1613 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
1614 SIOCGIFNETMASK, "SIOCGIFNETMASK");
1619 netdev_dev->cache_valid |= VALID_IN4;
1621 *address = netdev_dev->address;
1622 *netmask = netdev_dev->netmask;
1623 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
1627 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
1628 struct in_addr netmask)
1630 struct netdev_dev_linux *netdev_dev =
1631 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1634 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
1636 netdev_dev->cache_valid |= VALID_IN4;
1637 netdev_dev->address = address;
1638 netdev_dev->netmask = netmask;
1639 if (address.s_addr != INADDR_ANY) {
1640 error = do_set_addr(netdev_, SIOCSIFNETMASK,
1641 "SIOCSIFNETMASK", netmask);
1648 parse_if_inet6_line(const char *line,
1649 struct in6_addr *in6, char ifname[16 + 1])
1651 uint8_t *s6 = in6->s6_addr;
1652 #define X8 "%2"SCNx8
1654 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
1655 "%*x %*x %*x %*x %16s\n",
1656 &s6[0], &s6[1], &s6[2], &s6[3],
1657 &s6[4], &s6[5], &s6[6], &s6[7],
1658 &s6[8], &s6[9], &s6[10], &s6[11],
1659 &s6[12], &s6[13], &s6[14], &s6[15],
1663 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
1664 * 'in6' is non-null) and returns true. Otherwise, returns false. */
1666 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
1668 struct netdev_dev_linux *netdev_dev =
1669 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1670 if (!(netdev_dev->cache_valid & VALID_IN6)) {
1674 netdev_dev->in6 = in6addr_any;
1676 file = fopen("/proc/net/if_inet6", "r");
1678 const char *name = netdev_get_name(netdev_);
1679 while (fgets(line, sizeof line, file)) {
1680 struct in6_addr in6;
1681 char ifname[16 + 1];
1682 if (parse_if_inet6_line(line, &in6, ifname)
1683 && !strcmp(name, ifname))
1685 netdev_dev->in6 = in6;
1691 netdev_dev->cache_valid |= VALID_IN6;
1693 *in6 = netdev_dev->in6;
1698 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
1700 struct sockaddr_in sin;
1701 memset(&sin, 0, sizeof sin);
1702 sin.sin_family = AF_INET;
1703 sin.sin_addr = addr;
1706 memset(sa, 0, sizeof *sa);
1707 memcpy(sa, &sin, sizeof sin);
1711 do_set_addr(struct netdev *netdev,
1712 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
1715 strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
1716 make_in4_sockaddr(&ifr.ifr_addr, addr);
1718 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
1722 /* Adds 'router' as a default IP gateway. */
1724 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
1726 struct in_addr any = { INADDR_ANY };
1730 memset(&rt, 0, sizeof rt);
1731 make_in4_sockaddr(&rt.rt_dst, any);
1732 make_in4_sockaddr(&rt.rt_gateway, router);
1733 make_in4_sockaddr(&rt.rt_genmask, any);
1734 rt.rt_flags = RTF_UP | RTF_GATEWAY;
1735 COVERAGE_INC(netdev_add_router);
1736 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
1738 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
1744 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
1747 static const char fn[] = "/proc/net/route";
1752 *netdev_name = NULL;
1753 stream = fopen(fn, "r");
1754 if (stream == NULL) {
1755 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1760 while (fgets(line, sizeof line, stream)) {
1763 uint32_t dest, gateway, mask;
1764 int refcnt, metric, mtu;
1765 unsigned int flags, use, window, irtt;
1768 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
1770 iface, &dest, &gateway, &flags, &refcnt,
1771 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
1773 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
1777 if (!(flags & RTF_UP)) {
1778 /* Skip routes that aren't up. */
1782 /* The output of 'dest', 'mask', and 'gateway' were given in
1783 * network byte order, so we don't need need any endian
1784 * conversions here. */
1785 if ((dest & mask) == (host->s_addr & mask)) {
1787 /* The host is directly reachable. */
1788 next_hop->s_addr = 0;
1790 /* To reach the host, we must go through a gateway. */
1791 next_hop->s_addr = gateway;
1793 *netdev_name = xstrdup(iface);
1804 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1805 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1806 * returns 0. Otherwise, it returns a positive errno value; in particular,
1807 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
1809 netdev_linux_arp_lookup(const struct netdev *netdev,
1810 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
1813 struct sockaddr_in sin;
1816 memset(&r, 0, sizeof r);
1817 sin.sin_family = AF_INET;
1818 sin.sin_addr.s_addr = ip;
1820 memcpy(&r.arp_pa, &sin, sizeof sin);
1821 r.arp_ha.sa_family = ARPHRD_ETHER;
1823 strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
1824 COVERAGE_INC(netdev_arp_lookup);
1825 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
1827 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
1828 } else if (retval != ENXIO) {
1829 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
1830 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
1836 nd_to_iff_flags(enum netdev_flags nd)
1839 if (nd & NETDEV_UP) {
1842 if (nd & NETDEV_PROMISC) {
1849 iff_to_nd_flags(int iff)
1851 enum netdev_flags nd = 0;
1855 if (iff & IFF_PROMISC) {
1856 nd |= NETDEV_PROMISC;
1862 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
1863 enum netdev_flags on, enum netdev_flags *old_flagsp)
1865 int old_flags, new_flags;
1868 error = get_flags(netdev, &old_flags);
1870 *old_flagsp = iff_to_nd_flags(old_flags);
1871 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
1872 if (new_flags != old_flags) {
1873 error = set_flags(netdev, new_flags);
1880 poll_notify(struct list *list)
1882 struct netdev_linux_notifier *notifier;
1883 LIST_FOR_EACH (notifier, struct netdev_linux_notifier, node, list) {
1884 struct netdev_notifier *n = ¬ifier->notifier;
1890 netdev_linux_poll_cb(const struct rtnetlink_change *change,
1891 void *aux OVS_UNUSED)
1894 struct list *list = shash_find_data(&netdev_linux_notifiers,
1900 struct shash_node *node;
1901 SHASH_FOR_EACH (node, &netdev_linux_notifiers) {
1902 poll_notify(node->data);
1908 netdev_linux_poll_add(struct netdev *netdev,
1909 void (*cb)(struct netdev_notifier *), void *aux,
1910 struct netdev_notifier **notifierp)
1912 const char *netdev_name = netdev_get_name(netdev);
1913 struct netdev_linux_notifier *notifier;
1916 if (shash_is_empty(&netdev_linux_notifiers)) {
1917 int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier,
1918 netdev_linux_poll_cb, NULL);
1924 list = shash_find_data(&netdev_linux_notifiers, netdev_name);
1926 list = xmalloc(sizeof *list);
1928 shash_add(&netdev_linux_notifiers, netdev_name, list);
1931 notifier = xmalloc(sizeof *notifier);
1932 netdev_notifier_init(¬ifier->notifier, netdev, cb, aux);
1933 list_push_back(list, ¬ifier->node);
1934 *notifierp = ¬ifier->notifier;
1939 netdev_linux_poll_remove(struct netdev_notifier *notifier_)
1941 struct netdev_linux_notifier *notifier =
1942 CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier);
1945 /* Remove 'notifier' from its list. */
1946 list = list_remove(¬ifier->node);
1947 if (list_is_empty(list)) {
1948 /* The list is now empty. Remove it from the hash and free it. */
1949 const char *netdev_name = netdev_get_name(notifier->notifier.netdev);
1950 shash_delete(&netdev_linux_notifiers,
1951 shash_find(&netdev_linux_notifiers, netdev_name));
1956 /* If that was the last notifier, unregister. */
1957 if (shash_is_empty(&netdev_linux_notifiers)) {
1958 rtnetlink_notifier_unregister(&netdev_linux_poll_notifier);
1962 const struct netdev_class netdev_linux_class = {
1969 netdev_linux_create_system,
1970 netdev_linux_destroy,
1971 NULL, /* reconfigure */
1976 netdev_linux_enumerate,
1979 netdev_linux_recv_wait,
1983 netdev_linux_send_wait,
1985 netdev_linux_set_etheraddr,
1986 netdev_linux_get_etheraddr,
1987 netdev_linux_get_mtu,
1988 netdev_linux_get_ifindex,
1989 netdev_linux_get_carrier,
1990 netdev_linux_get_stats,
1992 netdev_linux_get_features,
1993 netdev_linux_set_advertisements,
1994 netdev_linux_get_vlan_vid,
1995 netdev_linux_set_policing,
1997 netdev_linux_get_in4,
1998 netdev_linux_set_in4,
1999 netdev_linux_get_in6,
2000 netdev_linux_add_router,
2001 netdev_linux_get_next_hop,
2002 netdev_linux_arp_lookup,
2004 netdev_linux_update_flags,
2006 netdev_linux_poll_add,
2007 netdev_linux_poll_remove,
2010 const struct netdev_class netdev_tap_class = {
2017 netdev_linux_create_tap,
2018 netdev_linux_destroy,
2019 NULL, /* reconfigure */
2024 NULL, /* enumerate */
2027 netdev_linux_recv_wait,
2031 netdev_linux_send_wait,
2033 netdev_linux_set_etheraddr,
2034 netdev_linux_get_etheraddr,
2035 netdev_linux_get_mtu,
2036 netdev_linux_get_ifindex,
2037 netdev_linux_get_carrier,
2038 netdev_linux_get_stats,
2040 netdev_linux_get_features,
2041 netdev_linux_set_advertisements,
2042 netdev_linux_get_vlan_vid,
2043 netdev_linux_set_policing,
2045 netdev_linux_get_in4,
2046 netdev_linux_set_in4,
2047 netdev_linux_get_in6,
2048 netdev_linux_add_router,
2049 netdev_linux_get_next_hop,
2050 netdev_linux_arp_lookup,
2052 netdev_linux_update_flags,
2054 netdev_linux_poll_add,
2055 netdev_linux_poll_remove,
2058 const struct netdev_class netdev_gre_class = {
2065 netdev_linux_create_gre,
2066 netdev_linux_destroy,
2067 netdev_linux_reconfigure_gre,
2072 NULL, /* enumerate */
2075 netdev_linux_recv_wait,
2079 netdev_linux_send_wait,
2081 netdev_linux_set_etheraddr,
2082 netdev_linux_get_etheraddr,
2083 netdev_linux_get_mtu,
2084 netdev_linux_get_ifindex,
2085 netdev_linux_get_carrier,
2086 netdev_linux_get_stats,
2088 netdev_linux_get_features,
2089 netdev_linux_set_advertisements,
2090 netdev_linux_get_vlan_vid,
2091 netdev_linux_set_policing,
2093 netdev_linux_get_in4,
2094 netdev_linux_set_in4,
2095 netdev_linux_get_in6,
2096 netdev_linux_add_router,
2097 netdev_linux_get_next_hop,
2098 netdev_linux_arp_lookup,
2100 netdev_linux_update_flags,
2102 netdev_linux_poll_add,
2103 netdev_linux_poll_remove,
2107 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
2109 /* Policy for RTNLGRP_LINK messages.
2111 * There are *many* more fields in these messages, but currently we only
2112 * care about these fields. */
2113 static const struct nl_policy rtnlgrp_link_policy[] = {
2114 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
2115 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
2116 .min_len = sizeof(struct rtnl_link_stats) },
2120 static struct nl_sock *rtnl_sock;
2121 struct ofpbuf request;
2122 struct ofpbuf *reply;
2123 struct ifinfomsg *ifi;
2124 const struct rtnl_link_stats *rtnl_stats;
2125 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
2129 error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock);
2131 VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",
2137 ofpbuf_init(&request, 0);
2138 nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *ifi,
2139 RTM_GETLINK, NLM_F_REQUEST);
2140 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
2141 ifi->ifi_family = PF_UNSPEC;
2142 ifi->ifi_index = ifindex;
2143 error = nl_sock_transact(rtnl_sock, &request, &reply);
2144 ofpbuf_uninit(&request);
2149 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
2150 rtnlgrp_link_policy,
2151 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
2152 ofpbuf_delete(reply);
2156 if (!attrs[IFLA_STATS]) {
2157 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
2158 ofpbuf_delete(reply);
2162 rtnl_stats = nl_attr_get(attrs[IFLA_STATS]);
2163 stats->rx_packets = rtnl_stats->rx_packets;
2164 stats->tx_packets = rtnl_stats->tx_packets;
2165 stats->rx_bytes = rtnl_stats->rx_bytes;
2166 stats->tx_bytes = rtnl_stats->tx_bytes;
2167 stats->rx_errors = rtnl_stats->rx_errors;
2168 stats->tx_errors = rtnl_stats->tx_errors;
2169 stats->rx_dropped = rtnl_stats->rx_dropped;
2170 stats->tx_dropped = rtnl_stats->tx_dropped;
2171 stats->multicast = rtnl_stats->multicast;
2172 stats->collisions = rtnl_stats->collisions;
2173 stats->rx_length_errors = rtnl_stats->rx_length_errors;
2174 stats->rx_over_errors = rtnl_stats->rx_over_errors;
2175 stats->rx_crc_errors = rtnl_stats->rx_crc_errors;
2176 stats->rx_frame_errors = rtnl_stats->rx_frame_errors;
2177 stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors;
2178 stats->rx_missed_errors = rtnl_stats->rx_missed_errors;
2179 stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors;
2180 stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors;
2181 stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors;
2182 stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors;
2183 stats->tx_window_errors = rtnl_stats->tx_window_errors;
2185 ofpbuf_delete(reply);
2191 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
2193 static const char fn[] = "/proc/net/dev";
2198 stream = fopen(fn, "r");
2200 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
2205 while (fgets(line, sizeof line, stream)) {
2208 #define X64 "%"SCNu64
2211 X64 X64 X64 X64 X64 X64 X64 "%*u"
2212 X64 X64 X64 X64 X64 X64 X64 "%*u",
2218 &stats->rx_fifo_errors,
2219 &stats->rx_frame_errors,
2225 &stats->tx_fifo_errors,
2227 &stats->tx_carrier_errors) != 15) {
2228 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
2229 } else if (!strcmp(devname, netdev_name)) {
2230 stats->rx_length_errors = UINT64_MAX;
2231 stats->rx_over_errors = UINT64_MAX;
2232 stats->rx_crc_errors = UINT64_MAX;
2233 stats->rx_missed_errors = UINT64_MAX;
2234 stats->tx_aborted_errors = UINT64_MAX;
2235 stats->tx_heartbeat_errors = UINT64_MAX;
2236 stats->tx_window_errors = UINT64_MAX;
2242 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
2248 get_flags(const struct netdev *netdev, int *flags)
2253 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS,
2255 *flags = ifr.ifr_flags;
2260 set_flags(struct netdev *netdev, int flags)
2264 ifr.ifr_flags = flags;
2265 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
2270 do_get_ifindex(const char *netdev_name)
2274 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2275 COVERAGE_INC(netdev_get_ifindex);
2276 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
2277 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
2278 netdev_name, strerror(errno));
2281 return ifr.ifr_ifindex;
2285 get_ifindex(const struct netdev *netdev_, int *ifindexp)
2287 struct netdev_dev_linux *netdev_dev =
2288 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2290 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
2291 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
2295 netdev_dev->cache_valid |= VALID_IFINDEX;
2296 netdev_dev->ifindex = ifindex;
2298 *ifindexp = netdev_dev->ifindex;
2303 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
2308 memset(&ifr, 0, sizeof ifr);
2309 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2310 COVERAGE_INC(netdev_get_hwaddr);
2311 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
2312 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
2313 netdev_name, strerror(errno));
2316 hwaddr_family = ifr.ifr_hwaddr.sa_family;
2317 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
2318 VLOG_WARN("%s device has unknown hardware address family %d",
2319 netdev_name, hwaddr_family);
2321 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
2326 set_etheraddr(const char *netdev_name, int hwaddr_family,
2327 const uint8_t mac[ETH_ADDR_LEN])
2331 memset(&ifr, 0, sizeof ifr);
2332 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2333 ifr.ifr_hwaddr.sa_family = hwaddr_family;
2334 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
2335 COVERAGE_INC(netdev_set_hwaddr);
2336 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
2337 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
2338 netdev_name, strerror(errno));
2345 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
2346 int cmd, const char *cmd_name)
2350 memset(&ifr, 0, sizeof ifr);
2351 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
2352 ifr.ifr_data = (caddr_t) ecmd;
2355 COVERAGE_INC(netdev_ethtool);
2356 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
2359 if (errno != EOPNOTSUPP) {
2360 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
2361 "failed: %s", cmd_name, name, strerror(errno));
2363 /* The device doesn't support this operation. That's pretty
2364 * common, so there's no point in logging anything. */
2371 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
2372 const char *cmd_name)
2374 strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
2375 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
2376 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
2384 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
2385 int cmd, const char *cmd_name)
2390 ifr.ifr_addr.sa_family = AF_INET;
2391 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
2393 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
2394 *ip = sin->sin_addr;