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;
135 struct nl_sock *nl_sock;
141 struct netdev_linux_notifier {
142 struct netdev_notifier notifier;
146 static struct shash netdev_linux_notifiers =
147 SHASH_INITIALIZER(&netdev_linux_notifiers);
148 static struct rtnetlink_notifier netdev_linux_poll_notifier;
150 /* This is set pretty low because we probably won't learn anything from the
151 * additional log messages. */
152 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
154 static int destroy_gre(const char *name);
155 static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
156 int cmd, const char *cmd_name);
157 static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
158 const char *cmd_name);
159 static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
160 int cmd, const char *cmd_name);
161 static int get_flags(const struct netdev *, int *flagsp);
162 static int set_flags(struct netdev *, int flags);
163 static int do_get_ifindex(const char *netdev_name);
164 static int get_ifindex(const struct netdev *, int *ifindexp);
165 static int do_set_addr(struct netdev *netdev,
166 int ioctl_nr, const char *ioctl_name,
167 struct in_addr addr);
168 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
169 static int set_etheraddr(const char *netdev_name, int hwaddr_family,
170 const uint8_t[ETH_ADDR_LEN]);
171 static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
172 static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
174 static struct netdev_dev_linux *
175 netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
177 const char *type = netdev_dev_get_type(netdev_dev);
178 assert(!strcmp(type, "system") || !strcmp(type, "tap")
179 || !strcmp(type, "gre"));
180 return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
183 static struct netdev_linux *
184 netdev_linux_cast(const struct netdev *netdev)
186 const char *type = netdev_get_type(netdev);
187 assert(!strcmp(type, "system") || !strcmp(type, "tap")
188 || !strcmp(type, "gre"));
189 return CONTAINER_OF(netdev, struct netdev_linux, netdev);
193 netdev_linux_init(void)
195 static int status = -1;
197 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
198 status = af_inet_sock >= 0 ? 0 : errno;
200 VLOG_ERR("failed to create inet socket: %s", strerror(status));
207 netdev_linux_run(void)
209 rtnetlink_notifier_run();
213 netdev_linux_wait(void)
215 rtnetlink_notifier_wait();
219 netdev_linux_cache_cb(const struct rtnetlink_change *change,
220 void *aux OVS_UNUSED)
222 struct netdev_dev_linux *dev;
224 struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
226 dev = netdev_dev_linux_cast(base_dev);
227 dev->cache_valid = 0;
230 struct shash device_shash;
231 struct shash_node *node;
233 shash_init(&device_shash);
234 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
235 SHASH_FOR_EACH (node, &device_shash) {
237 dev->cache_valid = 0;
239 shash_destroy(&device_shash);
243 /* The arguments are marked as unused to prevent warnings on platforms where
244 * the Netlink interface isn't supported. */
246 setup_gre_netlink(const char *name OVS_UNUSED,
247 struct gre_config *config OVS_UNUSED, bool create OVS_UNUSED)
249 #ifdef GRE_IOCTL_ONLY
253 struct ofpbuf request, *reply;
254 unsigned int nl_flags;
255 struct ifinfomsg ifinfomsg;
256 struct nlattr *linkinfo_hdr;
257 struct nlattr *info_data_hdr;
260 uint8_t pmtudisc = 0;
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, pmtudisc);
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;
379 strncpy(ifr.ifr_name, create ? GRE_IOCTL_DEVICE : name, IFNAMSIZ);
380 ifr.ifr_ifru.ifru_data = (void *)&p;
382 if (!gre_descriptors.ioctl_fd) {
383 gre_descriptors.ioctl_fd = socket(AF_INET, SOCK_DGRAM, 0);
384 if (gre_descriptors.ioctl_fd < 0) {
385 VLOG_WARN("couldn't create gre ioctl socket: %s", strerror(errno));
386 gre_descriptors.ioctl_fd = 0;
391 if (ioctl(gre_descriptors.ioctl_fd, create ? SIOCADDGRETAP : SIOCCHGGRETAP,
393 VLOG_WARN("couldn't do gre ioctl: %s", strerror(errno));
400 /* The arguments are marked as unused to prevent warnings on platforms where
401 * the Netlink interface isn't supported. */
403 check_gre_device_netlink(const char *name OVS_UNUSED)
405 #ifdef GRE_IOCTL_ONLY
408 static const struct nl_policy getlink_policy[] = {
409 [IFLA_LINKINFO] = { .type = NL_A_NESTED, .optional = false },
412 static const struct nl_policy linkinfo_policy[] = {
413 [IFLA_INFO_KIND] = { .type = NL_A_STRING, .optional = false },
418 struct ofpbuf request, *reply;
419 struct ifinfomsg ifinfomsg;
420 struct nlattr *getlink_attrs[ARRAY_SIZE(getlink_policy)];
421 struct nlattr *linkinfo_attrs[ARRAY_SIZE(linkinfo_policy)];
422 struct ofpbuf linkinfo;
423 const char *device_kind;
425 ofpbuf_init(&request, 0);
427 nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock,
428 NLMSG_LENGTH(sizeof ifinfomsg), RTM_GETLINK,
431 memset(&ifinfomsg, 0, sizeof ifinfomsg);
432 ifinfomsg.ifi_family = AF_UNSPEC;
433 ifinfomsg.ifi_index = do_get_ifindex(name);
434 nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
436 error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
437 ofpbuf_uninit(&request);
439 VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
443 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
444 getlink_policy, getlink_attrs,
445 ARRAY_SIZE(getlink_policy))) {
446 VLOG_WARN("received bad rtnl message (getlink policy)");
450 linkinfo.data = (void *)nl_attr_get(getlink_attrs[IFLA_LINKINFO]);
451 linkinfo.size = nl_attr_get_size(getlink_attrs[IFLA_LINKINFO]);
452 if (!nl_policy_parse(&linkinfo, 0, linkinfo_policy,
453 linkinfo_attrs, ARRAY_SIZE(linkinfo_policy))) {
454 VLOG_WARN("received bad rtnl message (linkinfo policy)");
458 device_kind = nl_attr_get_string(linkinfo_attrs[IFLA_INFO_KIND]);
459 ret = !strcmp(device_kind, "gretap");
462 ofpbuf_delete(reply);
468 check_gre_device_ioctl(const char *name)
470 struct ethtool_drvinfo drvinfo;
473 memset(&drvinfo, 0, sizeof drvinfo);
474 error = netdev_linux_do_ethtool(name, (struct ethtool_cmd *)&drvinfo,
475 ETHTOOL_GDRVINFO, "ETHTOOL_GDRVINFO");
477 return !error && !strcmp(drvinfo.driver, "ip_gre")
478 && !strcmp(drvinfo.bus_info, "gretap");
482 setup_gre(const char *name, const struct shash *args, bool create)
485 struct in_addr in_addr;
486 struct shash_node *node;
487 struct gre_config config;
489 memset(&config, 0, sizeof config);
490 config.in_csum = true;
491 config.out_csum = true;
493 SHASH_FOR_EACH (node, args) {
494 if (!strcmp(node->name, "remote_ip")) {
495 if (lookup_ip(node->data, &in_addr)) {
496 VLOG_WARN("bad 'remote_ip' for gre device %s ", name);
498 config.remote_ip = in_addr.s_addr;
500 } else if (!strcmp(node->name, "local_ip")) {
501 if (lookup_ip(node->data, &in_addr)) {
502 VLOG_WARN("bad 'local_ip' for gre device %s ", name);
504 config.local_ip = in_addr.s_addr;
506 } else if (!strcmp(node->name, "key")) {
507 config.have_in_key = true;
508 config.have_out_key = true;
509 config.in_key = htonl(atoi(node->data));
510 config.out_key = htonl(atoi(node->data));
511 } else if (!strcmp(node->name, "in_key")) {
512 config.have_in_key = true;
513 config.in_key = htonl(atoi(node->data));
514 } else if (!strcmp(node->name, "out_key")) {
515 config.have_out_key = true;
516 config.out_key = htonl(atoi(node->data));
517 } else if (!strcmp(node->name, "tos")) {
518 config.tos = atoi(node->data);
519 } else if (!strcmp(node->name, "csum")) {
520 if (!strcmp(node->data, "false")) {
521 config.in_csum = false;
522 config.out_csum = false;
525 VLOG_WARN("unknown gre argument '%s'", node->name);
529 if (!config.remote_ip) {
530 VLOG_WARN("gre type requires valid 'remote_ip' argument");
535 if (!gre_descriptors.use_ioctl) {
536 error = setup_gre_netlink(name, &config, create);
537 if (error == EOPNOTSUPP) {
538 gre_descriptors.use_ioctl = true;
541 if (gre_descriptors.use_ioctl) {
542 error = setup_gre_ioctl(name, &config, create);
545 if (create && error == EEXIST) {
548 if (gre_descriptors.use_ioctl) {
549 gre_device = check_gre_device_ioctl(name);
551 gre_device = check_gre_device_netlink(name);
558 VLOG_WARN("replacing existing gre device %s", name);
559 error = destroy_gre(name);
564 if (gre_descriptors.use_ioctl) {
565 error = setup_gre_ioctl(name, &config, create);
567 error = setup_gre_netlink(name, &config, create);
575 /* Creates the netdev device of 'type' with 'name'. */
577 netdev_linux_create_system(const char *name, const char *type OVS_UNUSED,
578 const struct shash *args, struct netdev_dev **netdev_devp)
580 struct netdev_dev_linux *netdev_dev;
583 if (!shash_is_empty(args)) {
584 VLOG_WARN("%s: arguments for system devices should be empty", name);
587 if (!cache_notifier_refcount) {
588 error = rtnetlink_notifier_register(&netdev_linux_cache_notifier,
589 netdev_linux_cache_cb, NULL);
594 cache_notifier_refcount++;
596 netdev_dev = xzalloc(sizeof *netdev_dev);
597 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class);
599 *netdev_devp = &netdev_dev->netdev_dev;
603 /* For most types of netdevs we open the device for each call of
604 * netdev_open(). However, this is not the case with tap devices,
605 * since it is only possible to open the device once. In this
606 * situation we share a single file descriptor, and consequently
607 * buffers, across all readers. Therefore once data is read it will
608 * be unavailable to other reads for tap devices. */
610 netdev_linux_create_tap(const char *name, const char *type OVS_UNUSED,
611 const struct shash *args, struct netdev_dev **netdev_devp)
613 struct netdev_dev_linux *netdev_dev;
614 struct tap_state *state;
615 static const char tap_dev[] = "/dev/net/tun";
619 if (!shash_is_empty(args)) {
620 VLOG_WARN("%s: arguments for TAP devices should be empty", name);
623 netdev_dev = xzalloc(sizeof *netdev_dev);
624 state = &netdev_dev->state.tap;
626 /* Open tap device. */
627 state->fd = open(tap_dev, O_RDWR);
630 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
634 /* Create tap device. */
635 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
636 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
637 if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
638 VLOG_WARN("%s: creating tap device failed: %s", name,
644 /* Make non-blocking. */
645 error = set_nonblocking(state->fd);
650 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
651 *netdev_devp = &netdev_dev->netdev_dev;
660 if_up(const char *name)
664 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
665 ifr.ifr_flags = IFF_UP;
667 if (ioctl(af_inet_sock, SIOCSIFFLAGS, &ifr) == -1) {
668 VLOG_DBG_RL(&rl, "%s: failed to bring device up: %s",
669 name, strerror(errno));
677 netdev_linux_create_gre(const char *name, const char *type OVS_UNUSED,
678 const struct shash *args, struct netdev_dev **netdev_devp)
680 struct netdev_dev_linux *netdev_dev;
683 netdev_dev = xzalloc(sizeof *netdev_dev);
685 error = setup_gre(name, args, true);
695 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_gre_class);
696 *netdev_devp = &netdev_dev->netdev_dev;
705 netdev_linux_reconfigure_gre(struct netdev_dev *netdev_dev_,
706 const struct shash *args)
708 const char *name = netdev_dev_get_name(netdev_dev_);
710 return setup_gre(name, args, false);
713 /* The arguments are marked as unused to prevent warnings on platforms where
714 * the Netlink interface isn't supported. */
716 destroy_gre_netlink(const char *name OVS_UNUSED)
718 #ifdef GRE_IOCTL_ONLY
722 struct ofpbuf request, *reply;
723 struct ifinfomsg ifinfomsg;
726 ofpbuf_init(&request, 0);
728 nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock, 0, RTM_DELLINK,
731 memset(&ifinfomsg, 0, sizeof ifinfomsg);
732 ifinfomsg.ifi_family = AF_UNSPEC;
733 nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
735 ifindex = do_get_ifindex(name);
736 nl_msg_put_u32(&request, IFLA_LINK, ifindex);
738 nl_msg_put_string(&request, IFLA_IFNAME, name);
740 error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
741 ofpbuf_uninit(&request);
743 VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
746 ofpbuf_delete(reply);
754 destroy_gre_ioctl(const char *name)
756 struct ip_tunnel_parm p;
759 memset(&p, 0, sizeof p);
760 strncpy(p.name, name, IFNAMSIZ);
762 strncpy(ifr.ifr_name, name, IFNAMSIZ);
763 ifr.ifr_ifru.ifru_data = (void *)&p;
765 if (ioctl(gre_descriptors.ioctl_fd, SIOCDELGRETAP, &ifr) < 0) {
766 VLOG_WARN("couldn't do gre ioctl: %s\n", strerror(errno));
774 destroy_tap(struct netdev_dev_linux *netdev_dev)
776 struct tap_state *state = &netdev_dev->state.tap;
778 if (state->fd >= 0) {
784 destroy_gre(const char *name)
786 if (gre_descriptors.use_ioctl) {
787 return destroy_gre_ioctl(name);
789 return destroy_gre_netlink(name);
793 /* Destroys the netdev device 'netdev_dev_'. */
795 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
797 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
798 const char *type = netdev_dev_get_type(netdev_dev_);
800 if (!strcmp(type, "system")) {
801 cache_notifier_refcount--;
803 if (!cache_notifier_refcount) {
804 rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
806 } else if (!strcmp(type, "tap")) {
807 destroy_tap(netdev_dev);
808 } else if (!strcmp(type, "gre")) {
809 destroy_gre(netdev_dev_get_name(&netdev_dev->netdev_dev));
816 netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype,
817 struct netdev **netdevp)
819 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
820 struct netdev_linux *netdev;
821 enum netdev_flags flags;
824 /* Allocate network device. */
825 netdev = xzalloc(sizeof *netdev);
827 netdev_init(&netdev->netdev, netdev_dev_);
829 error = netdev_get_flags(&netdev->netdev, &flags);
830 if (error == ENODEV) {
834 if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap")) {
835 netdev->fd = netdev_dev->state.tap.fd;
836 } else if (ethertype != NETDEV_ETH_TYPE_NONE) {
837 struct sockaddr_ll sll;
841 /* Create file descriptor. */
842 protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
843 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
845 netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
846 if (netdev->fd < 0) {
851 /* Set non-blocking mode. */
852 error = set_nonblocking(netdev->fd);
857 /* Get ethernet device index. */
858 error = get_ifindex(&netdev->netdev, &ifindex);
863 /* Bind to specific ethernet device. */
864 memset(&sll, 0, sizeof sll);
865 sll.sll_family = AF_PACKET;
866 sll.sll_ifindex = ifindex;
868 (struct sockaddr *) &sll, sizeof sll) < 0) {
870 VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_),
875 /* Between the socket() and bind() calls above, the socket receives all
876 * packets of the requested type on all system interfaces. We do not
877 * want to receive that data, but there is no way to avoid it. So we
878 * must now drain out the receive queue. */
879 error = drain_rcvbuf(netdev->fd);
885 *netdevp = &netdev->netdev;
889 netdev_uninit(&netdev->netdev, true);
893 /* Closes and destroys 'netdev'. */
895 netdev_linux_close(struct netdev *netdev_)
897 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
899 if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
905 /* Initializes 'svec' with a list of the names of all known network devices. */
907 netdev_linux_enumerate(struct svec *svec)
909 struct if_nameindex *names;
911 names = if_nameindex();
915 for (i = 0; names[i].if_name != NULL; i++) {
916 svec_add(svec, names[i].if_name);
918 if_freenameindex(names);
921 VLOG_WARN("could not obtain list of network device names: %s",
928 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
930 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
932 if (netdev->fd < 0) {
933 /* Device was opened with NETDEV_ETH_TYPE_NONE. */
938 ssize_t retval = read(netdev->fd, data, size);
941 } else if (errno != EINTR) {
942 if (errno != EAGAIN) {
943 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
944 strerror(errno), netdev_get_name(netdev_));
951 /* Registers with the poll loop to wake up from the next call to poll_block()
952 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
954 netdev_linux_recv_wait(struct netdev *netdev_)
956 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
957 if (netdev->fd >= 0) {
958 poll_fd_wait(netdev->fd, POLLIN);
962 /* Discards all packets waiting to be received from 'netdev'. */
964 netdev_linux_drain(struct netdev *netdev_)
966 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
967 if (netdev->fd < 0) {
969 } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
971 int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
972 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
976 drain_fd(netdev->fd, ifr.ifr_qlen);
979 return drain_rcvbuf(netdev->fd);
983 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
984 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
985 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
986 * the packet is too big or too small to transmit on the device.
988 * The caller retains ownership of 'buffer' in all cases.
990 * The kernel maintains a packet transmission queue, so the caller is not
991 * expected to do additional queuing of packets. */
993 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
995 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
997 /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
999 if (netdev->fd < 0) {
1004 ssize_t retval = write(netdev->fd, data, size);
1006 /* The Linux AF_PACKET implementation never blocks waiting for room
1007 * for packets, instead returning ENOBUFS. Translate this into
1008 * EAGAIN for the caller. */
1009 if (errno == ENOBUFS) {
1011 } else if (errno == EINTR) {
1013 } else if (errno != EAGAIN) {
1014 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
1015 netdev_get_name(netdev_), strerror(errno));
1018 } else if (retval != size) {
1019 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
1020 "%zu) on %s", retval, size, netdev_get_name(netdev_));
1028 /* Registers with the poll loop to wake up from the next call to poll_block()
1029 * when the packet transmission queue has sufficient room to transmit a packet
1030 * with netdev_send().
1032 * The kernel maintains a packet transmission queue, so the client is not
1033 * expected to do additional queuing of packets. Thus, this function is
1034 * unlikely to ever be used. It is included for completeness. */
1036 netdev_linux_send_wait(struct netdev *netdev_)
1038 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1039 if (netdev->fd < 0) {
1040 /* Nothing to do. */
1041 } else if (strcmp(netdev_get_type(netdev_), "tap")) {
1042 poll_fd_wait(netdev->fd, POLLOUT);
1044 /* TAP device always accepts packets.*/
1045 poll_immediate_wake();
1049 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
1050 * otherwise a positive errno value. */
1052 netdev_linux_set_etheraddr(struct netdev *netdev_,
1053 const uint8_t mac[ETH_ADDR_LEN])
1055 struct netdev_dev_linux *netdev_dev =
1056 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1059 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)
1060 || !eth_addr_equals(netdev_dev->etheraddr, mac)) {
1061 error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
1063 netdev_dev->cache_valid |= VALID_ETHERADDR;
1064 memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
1072 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
1073 * free the returned buffer. */
1075 netdev_linux_get_etheraddr(const struct netdev *netdev_,
1076 uint8_t mac[ETH_ADDR_LEN])
1078 struct netdev_dev_linux *netdev_dev =
1079 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1080 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
1081 int error = get_etheraddr(netdev_get_name(netdev_),
1082 netdev_dev->etheraddr);
1086 netdev_dev->cache_valid |= VALID_ETHERADDR;
1088 memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
1092 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
1093 * in bytes, not including the hardware header; thus, this is typically 1500
1094 * bytes for Ethernet devices. */
1096 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
1098 struct netdev_dev_linux *netdev_dev =
1099 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1100 if (!(netdev_dev->cache_valid & VALID_MTU)) {
1104 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1105 SIOCGIFMTU, "SIOCGIFMTU");
1109 netdev_dev->mtu = ifr.ifr_mtu;
1110 netdev_dev->cache_valid |= VALID_MTU;
1112 *mtup = netdev_dev->mtu;
1116 /* Returns the ifindex of 'netdev', if successful, as a positive number.
1117 * On failure, returns a negative errno value. */
1119 netdev_linux_get_ifindex(const struct netdev *netdev)
1123 error = get_ifindex(netdev, &ifindex);
1124 return error ? -error : ifindex;
1128 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
1130 struct netdev_dev_linux *netdev_dev =
1131 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1136 if (!(netdev_dev->cache_valid & VALID_CARRIER)) {
1140 fn = xasprintf("/sys/class/net/%s/carrier",
1141 netdev_get_name(netdev_));
1142 fd = open(fn, O_RDONLY);
1145 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error));
1149 retval = read(fd, line, sizeof line);
1152 if (error == EINVAL) {
1153 /* This is the normal return value when we try to check carrier
1154 * if the network device is not up. */
1156 VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error));
1159 } else if (retval == 0) {
1161 VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn);
1165 if (line[0] != '0' && line[0] != '1') {
1167 VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)",
1171 netdev_dev->carrier = line[0] != '0';
1172 netdev_dev->cache_valid |= VALID_CARRIER;
1174 *carrier = netdev_dev->carrier;
1185 /* Check whether we can we use RTM_GETLINK to get network device statistics.
1186 * In pre-2.6.19 kernels, this was only available if wireless extensions were
1189 check_for_working_netlink_stats(void)
1191 /* Decide on the netdev_get_stats() implementation to use. Netlink is
1192 * preferable, so if that works, we'll use it. */
1193 int ifindex = do_get_ifindex("lo");
1195 VLOG_WARN("failed to get ifindex for lo, "
1196 "obtaining netdev stats from proc");
1199 struct netdev_stats stats;
1200 int error = get_stats_via_netlink(ifindex, &stats);
1202 VLOG_DBG("obtaining netdev stats via rtnetlink");
1205 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
1206 "via proc (you are probably running a pre-2.6.19 "
1207 "kernel)", strerror(error));
1213 /* Retrieves current device stats for 'netdev'.
1215 * XXX All of the members of struct netdev_stats are 64 bits wide, but on
1216 * 32-bit architectures the Linux network stats are only 32 bits. */
1218 netdev_linux_get_stats(const struct netdev *netdev_,
1219 struct netdev_stats *stats)
1221 struct netdev_dev_linux *netdev_dev =
1222 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1223 static int use_netlink_stats = -1;
1225 struct netdev_stats raw_stats;
1226 struct netdev_stats *collect_stats = stats;
1228 COVERAGE_INC(netdev_get_stats);
1230 if (!(netdev_dev->cache_valid & VALID_IS_INTERNAL)) {
1231 netdev_dev->is_internal = !strcmp(netdev_get_type(netdev_), "tap");
1232 if (!netdev_dev->is_internal) {
1233 struct ethtool_drvinfo drvinfo;
1235 memset(&drvinfo, 0, sizeof drvinfo);
1236 error = netdev_linux_do_ethtool(netdev_get_name(netdev_),
1237 (struct ethtool_cmd *)&drvinfo,
1239 "ETHTOOL_GDRVINFO");
1242 netdev_dev->is_internal = !strcmp(drvinfo.driver,
1247 netdev_dev->cache_valid |= VALID_IS_INTERNAL;
1250 if (netdev_dev->is_internal) {
1251 collect_stats = &raw_stats;
1254 if (use_netlink_stats < 0) {
1255 use_netlink_stats = check_for_working_netlink_stats();
1257 if (use_netlink_stats) {
1260 error = get_ifindex(netdev_, &ifindex);
1262 error = get_stats_via_netlink(ifindex, collect_stats);
1265 error = get_stats_via_proc(netdev_get_name(netdev_), collect_stats);
1268 /* If this port is an internal port then the transmit and receive stats
1269 * will appear to be swapped relative to the other ports since we are the
1270 * one sending the data, not a remote computer. For consistency, we swap
1271 * them back here. */
1272 if (!error && netdev_dev->is_internal) {
1273 stats->rx_packets = raw_stats.tx_packets;
1274 stats->tx_packets = raw_stats.rx_packets;
1275 stats->rx_bytes = raw_stats.tx_bytes;
1276 stats->tx_bytes = raw_stats.rx_bytes;
1277 stats->rx_errors = raw_stats.tx_errors;
1278 stats->tx_errors = raw_stats.rx_errors;
1279 stats->rx_dropped = raw_stats.tx_dropped;
1280 stats->tx_dropped = raw_stats.rx_dropped;
1281 stats->multicast = raw_stats.multicast;
1282 stats->collisions = raw_stats.collisions;
1283 stats->rx_length_errors = 0;
1284 stats->rx_over_errors = 0;
1285 stats->rx_crc_errors = 0;
1286 stats->rx_frame_errors = 0;
1287 stats->rx_fifo_errors = 0;
1288 stats->rx_missed_errors = 0;
1289 stats->tx_aborted_errors = 0;
1290 stats->tx_carrier_errors = 0;
1291 stats->tx_fifo_errors = 0;
1292 stats->tx_heartbeat_errors = 0;
1293 stats->tx_window_errors = 0;
1299 /* Stores the features supported by 'netdev' into each of '*current',
1300 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
1301 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
1302 * successful, otherwise a positive errno value. */
1304 netdev_linux_get_features(struct netdev *netdev,
1305 uint32_t *current, uint32_t *advertised,
1306 uint32_t *supported, uint32_t *peer)
1308 struct ethtool_cmd ecmd;
1311 memset(&ecmd, 0, sizeof ecmd);
1312 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1313 ETHTOOL_GSET, "ETHTOOL_GSET");
1318 /* Supported features. */
1320 if (ecmd.supported & SUPPORTED_10baseT_Half) {
1321 *supported |= OFPPF_10MB_HD;
1323 if (ecmd.supported & SUPPORTED_10baseT_Full) {
1324 *supported |= OFPPF_10MB_FD;
1326 if (ecmd.supported & SUPPORTED_100baseT_Half) {
1327 *supported |= OFPPF_100MB_HD;
1329 if (ecmd.supported & SUPPORTED_100baseT_Full) {
1330 *supported |= OFPPF_100MB_FD;
1332 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
1333 *supported |= OFPPF_1GB_HD;
1335 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
1336 *supported |= OFPPF_1GB_FD;
1338 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
1339 *supported |= OFPPF_10GB_FD;
1341 if (ecmd.supported & SUPPORTED_TP) {
1342 *supported |= OFPPF_COPPER;
1344 if (ecmd.supported & SUPPORTED_FIBRE) {
1345 *supported |= OFPPF_FIBER;
1347 if (ecmd.supported & SUPPORTED_Autoneg) {
1348 *supported |= OFPPF_AUTONEG;
1350 if (ecmd.supported & SUPPORTED_Pause) {
1351 *supported |= OFPPF_PAUSE;
1353 if (ecmd.supported & SUPPORTED_Asym_Pause) {
1354 *supported |= OFPPF_PAUSE_ASYM;
1357 /* Advertised features. */
1359 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
1360 *advertised |= OFPPF_10MB_HD;
1362 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
1363 *advertised |= OFPPF_10MB_FD;
1365 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
1366 *advertised |= OFPPF_100MB_HD;
1368 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
1369 *advertised |= OFPPF_100MB_FD;
1371 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
1372 *advertised |= OFPPF_1GB_HD;
1374 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
1375 *advertised |= OFPPF_1GB_FD;
1377 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
1378 *advertised |= OFPPF_10GB_FD;
1380 if (ecmd.advertising & ADVERTISED_TP) {
1381 *advertised |= OFPPF_COPPER;
1383 if (ecmd.advertising & ADVERTISED_FIBRE) {
1384 *advertised |= OFPPF_FIBER;
1386 if (ecmd.advertising & ADVERTISED_Autoneg) {
1387 *advertised |= OFPPF_AUTONEG;
1389 if (ecmd.advertising & ADVERTISED_Pause) {
1390 *advertised |= OFPPF_PAUSE;
1392 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
1393 *advertised |= OFPPF_PAUSE_ASYM;
1396 /* Current settings. */
1397 if (ecmd.speed == SPEED_10) {
1398 *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD;
1399 } else if (ecmd.speed == SPEED_100) {
1400 *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD;
1401 } else if (ecmd.speed == SPEED_1000) {
1402 *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD;
1403 } else if (ecmd.speed == SPEED_10000) {
1404 *current = OFPPF_10GB_FD;
1409 if (ecmd.port == PORT_TP) {
1410 *current |= OFPPF_COPPER;
1411 } else if (ecmd.port == PORT_FIBRE) {
1412 *current |= OFPPF_FIBER;
1416 *current |= OFPPF_AUTONEG;
1419 /* Peer advertisements. */
1420 *peer = 0; /* XXX */
1425 /* Set the features advertised by 'netdev' to 'advertise'. */
1427 netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
1429 struct ethtool_cmd ecmd;
1432 memset(&ecmd, 0, sizeof ecmd);
1433 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1434 ETHTOOL_GSET, "ETHTOOL_GSET");
1439 ecmd.advertising = 0;
1440 if (advertise & OFPPF_10MB_HD) {
1441 ecmd.advertising |= ADVERTISED_10baseT_Half;
1443 if (advertise & OFPPF_10MB_FD) {
1444 ecmd.advertising |= ADVERTISED_10baseT_Full;
1446 if (advertise & OFPPF_100MB_HD) {
1447 ecmd.advertising |= ADVERTISED_100baseT_Half;
1449 if (advertise & OFPPF_100MB_FD) {
1450 ecmd.advertising |= ADVERTISED_100baseT_Full;
1452 if (advertise & OFPPF_1GB_HD) {
1453 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1455 if (advertise & OFPPF_1GB_FD) {
1456 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1458 if (advertise & OFPPF_10GB_FD) {
1459 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1461 if (advertise & OFPPF_COPPER) {
1462 ecmd.advertising |= ADVERTISED_TP;
1464 if (advertise & OFPPF_FIBER) {
1465 ecmd.advertising |= ADVERTISED_FIBRE;
1467 if (advertise & OFPPF_AUTONEG) {
1468 ecmd.advertising |= ADVERTISED_Autoneg;
1470 if (advertise & OFPPF_PAUSE) {
1471 ecmd.advertising |= ADVERTISED_Pause;
1473 if (advertise & OFPPF_PAUSE_ASYM) {
1474 ecmd.advertising |= ADVERTISED_Asym_Pause;
1476 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1477 ETHTOOL_SSET, "ETHTOOL_SSET");
1480 /* If 'netdev_name' is the name of a VLAN network device (e.g. one created with
1481 * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device
1482 * and returns 0. Otherwise returns a errno value (specifically ENOENT if
1483 * 'netdev_name' is the name of a network device that is not a VLAN device) and
1484 * sets '*vlan_vid' to -1. */
1486 netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1488 const char *netdev_name = netdev_get_name(netdev);
1489 struct ds line = DS_EMPTY_INITIALIZER;
1490 FILE *stream = NULL;
1494 COVERAGE_INC(netdev_get_vlan_vid);
1495 fn = xasprintf("/proc/net/vlan/%s", netdev_name);
1496 stream = fopen(fn, "r");
1502 if (ds_get_line(&line, stream)) {
1503 if (ferror(stream)) {
1505 VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno));
1508 VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn);
1513 if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) {
1515 VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"",
1516 fn, ds_cstr(&line));
1534 #define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress"
1535 #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"
1536 /* We redirect stderr to /dev/null because we often want to remove all
1537 * traffic control configuration on a port so its in a known state. If
1538 * this done when there is no such configuration, tc complains, so we just
1541 #define POLICE_DEL_CMD "/sbin/tc qdisc del dev %s handle ffff: ingress 2>/dev/null"
1543 /* Attempts to set input rate limiting (policing) policy. */
1545 netdev_linux_set_policing(struct netdev *netdev,
1546 uint32_t kbits_rate, uint32_t kbits_burst)
1548 const char *netdev_name = netdev_get_name(netdev);
1551 COVERAGE_INC(netdev_set_policing);
1554 /* Default to 1000 kilobits if not specified. */
1558 /* xxx This should be more careful about only adding if it
1559 * xxx actually exists, as opposed to always deleting it. */
1560 snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1561 if (system(command) == -1) {
1562 VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1565 snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name);
1566 if (system(command) != 0) {
1567 VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name);
1571 snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name,
1572 kbits_rate, kbits_burst);
1573 if (system(command) != 0) {
1574 VLOG_WARN_RL(&rl, "%s: problem configuring policing",
1579 snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1580 if (system(command) == -1) {
1581 VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1589 netdev_linux_get_in4(const struct netdev *netdev_,
1590 struct in_addr *address, struct in_addr *netmask)
1592 struct netdev_dev_linux *netdev_dev =
1593 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1595 if (!(netdev_dev->cache_valid & VALID_IN4)) {
1598 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
1599 SIOCGIFADDR, "SIOCGIFADDR");
1604 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
1605 SIOCGIFNETMASK, "SIOCGIFNETMASK");
1610 netdev_dev->cache_valid |= VALID_IN4;
1612 *address = netdev_dev->address;
1613 *netmask = netdev_dev->netmask;
1614 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
1618 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
1619 struct in_addr netmask)
1621 struct netdev_dev_linux *netdev_dev =
1622 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1625 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
1627 netdev_dev->cache_valid |= VALID_IN4;
1628 netdev_dev->address = address;
1629 netdev_dev->netmask = netmask;
1630 if (address.s_addr != INADDR_ANY) {
1631 error = do_set_addr(netdev_, SIOCSIFNETMASK,
1632 "SIOCSIFNETMASK", netmask);
1639 parse_if_inet6_line(const char *line,
1640 struct in6_addr *in6, char ifname[16 + 1])
1642 uint8_t *s6 = in6->s6_addr;
1643 #define X8 "%2"SCNx8
1645 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
1646 "%*x %*x %*x %*x %16s\n",
1647 &s6[0], &s6[1], &s6[2], &s6[3],
1648 &s6[4], &s6[5], &s6[6], &s6[7],
1649 &s6[8], &s6[9], &s6[10], &s6[11],
1650 &s6[12], &s6[13], &s6[14], &s6[15],
1654 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
1655 * 'in6' is non-null) and returns true. Otherwise, returns false. */
1657 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
1659 struct netdev_dev_linux *netdev_dev =
1660 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1661 if (!(netdev_dev->cache_valid & VALID_IN6)) {
1665 netdev_dev->in6 = in6addr_any;
1667 file = fopen("/proc/net/if_inet6", "r");
1669 const char *name = netdev_get_name(netdev_);
1670 while (fgets(line, sizeof line, file)) {
1671 struct in6_addr in6;
1672 char ifname[16 + 1];
1673 if (parse_if_inet6_line(line, &in6, ifname)
1674 && !strcmp(name, ifname))
1676 netdev_dev->in6 = in6;
1682 netdev_dev->cache_valid |= VALID_IN6;
1684 *in6 = netdev_dev->in6;
1689 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
1691 struct sockaddr_in sin;
1692 memset(&sin, 0, sizeof sin);
1693 sin.sin_family = AF_INET;
1694 sin.sin_addr = addr;
1697 memset(sa, 0, sizeof *sa);
1698 memcpy(sa, &sin, sizeof sin);
1702 do_set_addr(struct netdev *netdev,
1703 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
1706 strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
1707 make_in4_sockaddr(&ifr.ifr_addr, addr);
1709 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
1713 /* Adds 'router' as a default IP gateway. */
1715 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
1717 struct in_addr any = { INADDR_ANY };
1721 memset(&rt, 0, sizeof rt);
1722 make_in4_sockaddr(&rt.rt_dst, any);
1723 make_in4_sockaddr(&rt.rt_gateway, router);
1724 make_in4_sockaddr(&rt.rt_genmask, any);
1725 rt.rt_flags = RTF_UP | RTF_GATEWAY;
1726 COVERAGE_INC(netdev_add_router);
1727 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
1729 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
1735 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
1738 static const char fn[] = "/proc/net/route";
1743 *netdev_name = NULL;
1744 stream = fopen(fn, "r");
1745 if (stream == NULL) {
1746 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1751 while (fgets(line, sizeof line, stream)) {
1754 uint32_t dest, gateway, mask;
1755 int refcnt, metric, mtu;
1756 unsigned int flags, use, window, irtt;
1759 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
1761 iface, &dest, &gateway, &flags, &refcnt,
1762 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
1764 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
1768 if (!(flags & RTF_UP)) {
1769 /* Skip routes that aren't up. */
1773 /* The output of 'dest', 'mask', and 'gateway' were given in
1774 * network byte order, so we don't need need any endian
1775 * conversions here. */
1776 if ((dest & mask) == (host->s_addr & mask)) {
1778 /* The host is directly reachable. */
1779 next_hop->s_addr = 0;
1781 /* To reach the host, we must go through a gateway. */
1782 next_hop->s_addr = gateway;
1784 *netdev_name = xstrdup(iface);
1795 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1796 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1797 * returns 0. Otherwise, it returns a positive errno value; in particular,
1798 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
1800 netdev_linux_arp_lookup(const struct netdev *netdev,
1801 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
1804 struct sockaddr_in sin;
1807 memset(&r, 0, sizeof r);
1808 sin.sin_family = AF_INET;
1809 sin.sin_addr.s_addr = ip;
1811 memcpy(&r.arp_pa, &sin, sizeof sin);
1812 r.arp_ha.sa_family = ARPHRD_ETHER;
1814 strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
1815 COVERAGE_INC(netdev_arp_lookup);
1816 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
1818 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
1819 } else if (retval != ENXIO) {
1820 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
1821 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
1827 nd_to_iff_flags(enum netdev_flags nd)
1830 if (nd & NETDEV_UP) {
1833 if (nd & NETDEV_PROMISC) {
1840 iff_to_nd_flags(int iff)
1842 enum netdev_flags nd = 0;
1846 if (iff & IFF_PROMISC) {
1847 nd |= NETDEV_PROMISC;
1853 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
1854 enum netdev_flags on, enum netdev_flags *old_flagsp)
1856 int old_flags, new_flags;
1859 error = get_flags(netdev, &old_flags);
1861 *old_flagsp = iff_to_nd_flags(old_flags);
1862 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
1863 if (new_flags != old_flags) {
1864 error = set_flags(netdev, new_flags);
1871 poll_notify(struct list *list)
1873 struct netdev_linux_notifier *notifier;
1874 LIST_FOR_EACH (notifier, struct netdev_linux_notifier, node, list) {
1875 struct netdev_notifier *n = ¬ifier->notifier;
1881 netdev_linux_poll_cb(const struct rtnetlink_change *change,
1882 void *aux OVS_UNUSED)
1885 struct list *list = shash_find_data(&netdev_linux_notifiers,
1891 struct shash_node *node;
1892 SHASH_FOR_EACH (node, &netdev_linux_notifiers) {
1893 poll_notify(node->data);
1899 netdev_linux_poll_add(struct netdev *netdev,
1900 void (*cb)(struct netdev_notifier *), void *aux,
1901 struct netdev_notifier **notifierp)
1903 const char *netdev_name = netdev_get_name(netdev);
1904 struct netdev_linux_notifier *notifier;
1907 if (shash_is_empty(&netdev_linux_notifiers)) {
1908 int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier,
1909 netdev_linux_poll_cb, NULL);
1915 list = shash_find_data(&netdev_linux_notifiers, netdev_name);
1917 list = xmalloc(sizeof *list);
1919 shash_add(&netdev_linux_notifiers, netdev_name, list);
1922 notifier = xmalloc(sizeof *notifier);
1923 netdev_notifier_init(¬ifier->notifier, netdev, cb, aux);
1924 list_push_back(list, ¬ifier->node);
1925 *notifierp = ¬ifier->notifier;
1930 netdev_linux_poll_remove(struct netdev_notifier *notifier_)
1932 struct netdev_linux_notifier *notifier =
1933 CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier);
1936 /* Remove 'notifier' from its list. */
1937 list = list_remove(¬ifier->node);
1938 if (list_is_empty(list)) {
1939 /* The list is now empty. Remove it from the hash and free it. */
1940 const char *netdev_name = netdev_get_name(notifier->notifier.netdev);
1941 shash_delete(&netdev_linux_notifiers,
1942 shash_find(&netdev_linux_notifiers, netdev_name));
1947 /* If that was the last notifier, unregister. */
1948 if (shash_is_empty(&netdev_linux_notifiers)) {
1949 rtnetlink_notifier_unregister(&netdev_linux_poll_notifier);
1953 const struct netdev_class netdev_linux_class = {
1960 netdev_linux_create_system,
1961 netdev_linux_destroy,
1962 NULL, /* reconfigure */
1967 netdev_linux_enumerate,
1970 netdev_linux_recv_wait,
1974 netdev_linux_send_wait,
1976 netdev_linux_set_etheraddr,
1977 netdev_linux_get_etheraddr,
1978 netdev_linux_get_mtu,
1979 netdev_linux_get_ifindex,
1980 netdev_linux_get_carrier,
1981 netdev_linux_get_stats,
1983 netdev_linux_get_features,
1984 netdev_linux_set_advertisements,
1985 netdev_linux_get_vlan_vid,
1986 netdev_linux_set_policing,
1988 netdev_linux_get_in4,
1989 netdev_linux_set_in4,
1990 netdev_linux_get_in6,
1991 netdev_linux_add_router,
1992 netdev_linux_get_next_hop,
1993 netdev_linux_arp_lookup,
1995 netdev_linux_update_flags,
1997 netdev_linux_poll_add,
1998 netdev_linux_poll_remove,
2001 const struct netdev_class netdev_tap_class = {
2008 netdev_linux_create_tap,
2009 netdev_linux_destroy,
2010 NULL, /* reconfigure */
2015 NULL, /* enumerate */
2018 netdev_linux_recv_wait,
2022 netdev_linux_send_wait,
2024 netdev_linux_set_etheraddr,
2025 netdev_linux_get_etheraddr,
2026 netdev_linux_get_mtu,
2027 netdev_linux_get_ifindex,
2028 netdev_linux_get_carrier,
2029 netdev_linux_get_stats,
2031 netdev_linux_get_features,
2032 netdev_linux_set_advertisements,
2033 netdev_linux_get_vlan_vid,
2034 netdev_linux_set_policing,
2036 netdev_linux_get_in4,
2037 netdev_linux_set_in4,
2038 netdev_linux_get_in6,
2039 netdev_linux_add_router,
2040 netdev_linux_get_next_hop,
2041 netdev_linux_arp_lookup,
2043 netdev_linux_update_flags,
2045 netdev_linux_poll_add,
2046 netdev_linux_poll_remove,
2049 const struct netdev_class netdev_gre_class = {
2056 netdev_linux_create_gre,
2057 netdev_linux_destroy,
2058 netdev_linux_reconfigure_gre,
2063 NULL, /* enumerate */
2066 netdev_linux_recv_wait,
2070 netdev_linux_send_wait,
2072 netdev_linux_set_etheraddr,
2073 netdev_linux_get_etheraddr,
2074 netdev_linux_get_mtu,
2075 netdev_linux_get_ifindex,
2076 netdev_linux_get_carrier,
2077 netdev_linux_get_stats,
2079 netdev_linux_get_features,
2080 netdev_linux_set_advertisements,
2081 netdev_linux_get_vlan_vid,
2082 netdev_linux_set_policing,
2084 netdev_linux_get_in4,
2085 netdev_linux_set_in4,
2086 netdev_linux_get_in6,
2087 netdev_linux_add_router,
2088 netdev_linux_get_next_hop,
2089 netdev_linux_arp_lookup,
2091 netdev_linux_update_flags,
2093 netdev_linux_poll_add,
2094 netdev_linux_poll_remove,
2098 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
2100 /* Policy for RTNLGRP_LINK messages.
2102 * There are *many* more fields in these messages, but currently we only
2103 * care about these fields. */
2104 static const struct nl_policy rtnlgrp_link_policy[] = {
2105 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
2106 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
2107 .min_len = sizeof(struct rtnl_link_stats) },
2111 static struct nl_sock *rtnl_sock;
2112 struct ofpbuf request;
2113 struct ofpbuf *reply;
2114 struct ifinfomsg *ifi;
2115 const struct rtnl_link_stats *rtnl_stats;
2116 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
2120 error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock);
2122 VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",
2128 ofpbuf_init(&request, 0);
2129 nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *ifi,
2130 RTM_GETLINK, NLM_F_REQUEST);
2131 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
2132 ifi->ifi_family = PF_UNSPEC;
2133 ifi->ifi_index = ifindex;
2134 error = nl_sock_transact(rtnl_sock, &request, &reply);
2135 ofpbuf_uninit(&request);
2140 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
2141 rtnlgrp_link_policy,
2142 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
2143 ofpbuf_delete(reply);
2147 if (!attrs[IFLA_STATS]) {
2148 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
2149 ofpbuf_delete(reply);
2153 rtnl_stats = nl_attr_get(attrs[IFLA_STATS]);
2154 stats->rx_packets = rtnl_stats->rx_packets;
2155 stats->tx_packets = rtnl_stats->tx_packets;
2156 stats->rx_bytes = rtnl_stats->rx_bytes;
2157 stats->tx_bytes = rtnl_stats->tx_bytes;
2158 stats->rx_errors = rtnl_stats->rx_errors;
2159 stats->tx_errors = rtnl_stats->tx_errors;
2160 stats->rx_dropped = rtnl_stats->rx_dropped;
2161 stats->tx_dropped = rtnl_stats->tx_dropped;
2162 stats->multicast = rtnl_stats->multicast;
2163 stats->collisions = rtnl_stats->collisions;
2164 stats->rx_length_errors = rtnl_stats->rx_length_errors;
2165 stats->rx_over_errors = rtnl_stats->rx_over_errors;
2166 stats->rx_crc_errors = rtnl_stats->rx_crc_errors;
2167 stats->rx_frame_errors = rtnl_stats->rx_frame_errors;
2168 stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors;
2169 stats->rx_missed_errors = rtnl_stats->rx_missed_errors;
2170 stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors;
2171 stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors;
2172 stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors;
2173 stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors;
2174 stats->tx_window_errors = rtnl_stats->tx_window_errors;
2176 ofpbuf_delete(reply);
2182 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
2184 static const char fn[] = "/proc/net/dev";
2189 stream = fopen(fn, "r");
2191 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
2196 while (fgets(line, sizeof line, stream)) {
2199 #define X64 "%"SCNu64
2202 X64 X64 X64 X64 X64 X64 X64 "%*u"
2203 X64 X64 X64 X64 X64 X64 X64 "%*u",
2209 &stats->rx_fifo_errors,
2210 &stats->rx_frame_errors,
2216 &stats->tx_fifo_errors,
2218 &stats->tx_carrier_errors) != 15) {
2219 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
2220 } else if (!strcmp(devname, netdev_name)) {
2221 stats->rx_length_errors = UINT64_MAX;
2222 stats->rx_over_errors = UINT64_MAX;
2223 stats->rx_crc_errors = UINT64_MAX;
2224 stats->rx_missed_errors = UINT64_MAX;
2225 stats->tx_aborted_errors = UINT64_MAX;
2226 stats->tx_heartbeat_errors = UINT64_MAX;
2227 stats->tx_window_errors = UINT64_MAX;
2233 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
2239 get_flags(const struct netdev *netdev, int *flags)
2244 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS,
2246 *flags = ifr.ifr_flags;
2251 set_flags(struct netdev *netdev, int flags)
2255 ifr.ifr_flags = flags;
2256 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
2261 do_get_ifindex(const char *netdev_name)
2265 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2266 COVERAGE_INC(netdev_get_ifindex);
2267 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
2268 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
2269 netdev_name, strerror(errno));
2272 return ifr.ifr_ifindex;
2276 get_ifindex(const struct netdev *netdev_, int *ifindexp)
2278 struct netdev_dev_linux *netdev_dev =
2279 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2281 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
2282 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
2286 netdev_dev->cache_valid |= VALID_IFINDEX;
2287 netdev_dev->ifindex = ifindex;
2289 *ifindexp = netdev_dev->ifindex;
2294 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
2299 memset(&ifr, 0, sizeof ifr);
2300 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2301 COVERAGE_INC(netdev_get_hwaddr);
2302 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
2303 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
2304 netdev_name, strerror(errno));
2307 hwaddr_family = ifr.ifr_hwaddr.sa_family;
2308 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
2309 VLOG_WARN("%s device has unknown hardware address family %d",
2310 netdev_name, hwaddr_family);
2312 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
2317 set_etheraddr(const char *netdev_name, int hwaddr_family,
2318 const uint8_t mac[ETH_ADDR_LEN])
2322 memset(&ifr, 0, sizeof ifr);
2323 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2324 ifr.ifr_hwaddr.sa_family = hwaddr_family;
2325 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
2326 COVERAGE_INC(netdev_set_hwaddr);
2327 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
2328 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
2329 netdev_name, strerror(errno));
2336 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
2337 int cmd, const char *cmd_name)
2341 memset(&ifr, 0, sizeof ifr);
2342 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
2343 ifr.ifr_data = (caddr_t) ecmd;
2346 COVERAGE_INC(netdev_ethtool);
2347 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
2350 if (errno != EOPNOTSUPP) {
2351 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
2352 "failed: %s", cmd_name, name, strerror(errno));
2354 /* The device doesn't support this operation. That's pretty
2355 * common, so there's no point in logging anything. */
2362 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
2363 const char *cmd_name)
2365 strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
2366 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
2367 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
2375 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
2376 int cmd, const char *cmd_name)
2381 ifr.ifr_addr.sa_family = AF_INET;
2382 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
2384 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
2385 *ip = sin->sin_addr;