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/pkt_sched.h>
28 #include <linux/rtnetlink.h>
29 #include <linux/sockios.h>
30 #include <linux/version.h>
31 #include <sys/types.h>
32 #include <sys/ioctl.h>
33 #include <sys/socket.h>
34 #include <netpacket/packet.h>
35 #include <net/ethernet.h>
37 #include <linux/if_tunnel.h>
38 #include <net/if_arp.h>
39 #include <net/if_packet.h>
40 #include <net/route.h>
41 #include <netinet/in.h>
48 #include "dynamic-string.h"
49 #include "fatal-signal.h"
50 #include "netdev-provider.h"
51 #include "netdev-vport.h"
54 #include "openflow/openflow.h"
55 #include "openvswitch/internal_dev.h"
56 #include "openvswitch/gre.h"
58 #include "poll-loop.h"
59 #include "rtnetlink.h"
60 #include "socket-util.h"
64 #define THIS_MODULE VLM_netdev_linux
67 /* These were introduced in Linux 2.6.14, so they might be missing if we have
69 #ifndef ADVERTISED_Pause
70 #define ADVERTISED_Pause (1 << 13)
72 #ifndef ADVERTISED_Asym_Pause
73 #define ADVERTISED_Asym_Pause (1 << 14)
76 static struct rtnetlink_notifier netdev_linux_cache_notifier;
77 static int cache_notifier_refcount;
80 VALID_IFINDEX = 1 << 0,
81 VALID_ETHERADDR = 1 << 1,
85 VALID_CARRIER = 1 << 5,
86 VALID_IS_PSEUDO = 1 << 6, /* Represents is_internal and is_tap. */
87 VALID_POLICING = 1 << 7,
88 VALID_HAVE_VPORT_STATS = 1 << 8
96 struct netdev_dev_linux {
97 struct netdev_dev netdev_dev;
99 struct shash_node *shash_node;
100 unsigned int cache_valid;
102 /* The following are figured out "on demand" only. They are only valid
103 * when the corresponding VALID_* bit in 'cache_valid' is set. */
105 uint8_t etheraddr[ETH_ADDR_LEN];
106 struct in_addr address, netmask;
110 bool is_internal; /* Is this an openvswitch internal device? */
111 bool is_tap; /* Is this a tuntap device? */
112 uint32_t kbits_rate; /* Policing data. */
113 uint32_t kbits_burst;
114 bool have_vport_stats;
117 struct tap_state tap;
121 struct netdev_linux {
122 struct netdev netdev;
126 /* An AF_INET socket (used for ioctl operations). */
127 static int af_inet_sock = -1;
129 struct netdev_linux_notifier {
130 struct netdev_notifier notifier;
134 static struct shash netdev_linux_notifiers =
135 SHASH_INITIALIZER(&netdev_linux_notifiers);
136 static struct rtnetlink_notifier netdev_linux_poll_notifier;
138 /* This is set pretty low because we probably won't learn anything from the
139 * additional log messages. */
140 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
142 static int netdev_linux_init(void);
144 static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
145 int cmd, const char *cmd_name);
146 static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
147 const char *cmd_name);
148 static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
149 int cmd, const char *cmd_name);
150 static int get_flags(const struct netdev *, int *flagsp);
151 static int set_flags(struct netdev *, int flags);
152 static int do_get_ifindex(const char *netdev_name);
153 static int get_ifindex(const struct netdev *, int *ifindexp);
154 static int do_set_addr(struct netdev *netdev,
155 int ioctl_nr, const char *ioctl_name,
156 struct in_addr addr);
157 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
158 static int set_etheraddr(const char *netdev_name, int hwaddr_family,
159 const uint8_t[ETH_ADDR_LEN]);
160 static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
161 static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
162 static int get_rtnl_sock(struct nl_sock **);
165 is_netdev_linux_class(const struct netdev_class *netdev_class)
167 return netdev_class->init == netdev_linux_init;
170 static struct netdev_dev_linux *
171 netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
173 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
174 assert(is_netdev_linux_class(netdev_class));
176 return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
179 static struct netdev_linux *
180 netdev_linux_cast(const struct netdev *netdev)
182 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
183 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
184 assert(is_netdev_linux_class(netdev_class));
186 return CONTAINER_OF(netdev, struct netdev_linux, netdev);
190 netdev_linux_init(void)
192 static int status = -1;
194 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
195 status = af_inet_sock >= 0 ? 0 : errno;
197 VLOG_ERR("failed to create inet socket: %s", strerror(status));
204 netdev_linux_run(void)
206 rtnetlink_notifier_run();
210 netdev_linux_wait(void)
212 rtnetlink_notifier_wait();
216 netdev_linux_cache_cb(const struct rtnetlink_change *change,
217 void *aux OVS_UNUSED)
219 struct netdev_dev_linux *dev;
221 struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
223 const struct netdev_class *netdev_class =
224 netdev_dev_get_class(base_dev);
226 if (is_netdev_linux_class(netdev_class)) {
227 dev = netdev_dev_linux_cast(base_dev);
228 dev->cache_valid = 0;
232 struct shash device_shash;
233 struct shash_node *node;
235 shash_init(&device_shash);
236 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
237 SHASH_FOR_EACH (node, &device_shash) {
239 dev->cache_valid = 0;
241 shash_destroy(&device_shash);
245 /* Creates the netdev device of 'type' with 'name'. */
247 netdev_linux_create_system(const char *name, const char *type OVS_UNUSED,
248 const struct shash *args, struct netdev_dev **netdev_devp)
250 struct netdev_dev_linux *netdev_dev;
253 if (!shash_is_empty(args)) {
254 VLOG_WARN("%s: arguments for system devices should be empty", name);
257 if (!cache_notifier_refcount) {
258 error = rtnetlink_notifier_register(&netdev_linux_cache_notifier,
259 netdev_linux_cache_cb, NULL);
264 cache_notifier_refcount++;
266 netdev_dev = xzalloc(sizeof *netdev_dev);
267 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class);
269 *netdev_devp = &netdev_dev->netdev_dev;
273 /* For most types of netdevs we open the device for each call of
274 * netdev_open(). However, this is not the case with tap devices,
275 * since it is only possible to open the device once. In this
276 * situation we share a single file descriptor, and consequently
277 * buffers, across all readers. Therefore once data is read it will
278 * be unavailable to other reads for tap devices. */
280 netdev_linux_create_tap(const char *name, const char *type OVS_UNUSED,
281 const struct shash *args, struct netdev_dev **netdev_devp)
283 struct netdev_dev_linux *netdev_dev;
284 struct tap_state *state;
285 static const char tap_dev[] = "/dev/net/tun";
289 if (!shash_is_empty(args)) {
290 VLOG_WARN("%s: arguments for TAP devices should be empty", name);
293 netdev_dev = xzalloc(sizeof *netdev_dev);
294 state = &netdev_dev->state.tap;
296 /* Open tap device. */
297 state->fd = open(tap_dev, O_RDWR);
300 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
304 /* Create tap device. */
305 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
306 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
307 if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
308 VLOG_WARN("%s: creating tap device failed: %s", name,
314 /* Make non-blocking. */
315 error = set_nonblocking(state->fd);
320 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
321 *netdev_devp = &netdev_dev->netdev_dev;
330 destroy_tap(struct netdev_dev_linux *netdev_dev)
332 struct tap_state *state = &netdev_dev->state.tap;
334 if (state->fd >= 0) {
339 /* Destroys the netdev device 'netdev_dev_'. */
341 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
343 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
344 const char *type = netdev_dev_get_type(netdev_dev_);
346 if (!strcmp(type, "system")) {
347 cache_notifier_refcount--;
349 if (!cache_notifier_refcount) {
350 rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
352 } else if (!strcmp(type, "tap")) {
353 destroy_tap(netdev_dev);
360 netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype,
361 struct netdev **netdevp)
363 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
364 struct netdev_linux *netdev;
365 enum netdev_flags flags;
368 /* Allocate network device. */
369 netdev = xzalloc(sizeof *netdev);
371 netdev_init(&netdev->netdev, netdev_dev_);
373 error = netdev_get_flags(&netdev->netdev, &flags);
374 if (error == ENODEV) {
378 if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap") &&
379 !netdev_dev->state.tap.opened) {
381 /* We assume that the first user of the tap device is the primary user
382 * and give them the tap FD. Subsequent users probably just expect
383 * this to be a system device so open it normally to avoid send/receive
384 * directions appearing to be reversed. */
385 netdev->fd = netdev_dev->state.tap.fd;
386 netdev_dev->state.tap.opened = true;
387 } else if (ethertype != NETDEV_ETH_TYPE_NONE) {
388 struct sockaddr_ll sll;
392 /* Create file descriptor. */
393 protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
394 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
396 netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
397 if (netdev->fd < 0) {
402 /* Set non-blocking mode. */
403 error = set_nonblocking(netdev->fd);
408 /* Get ethernet device index. */
409 error = get_ifindex(&netdev->netdev, &ifindex);
414 /* Bind to specific ethernet device. */
415 memset(&sll, 0, sizeof sll);
416 sll.sll_family = AF_PACKET;
417 sll.sll_ifindex = ifindex;
419 (struct sockaddr *) &sll, sizeof sll) < 0) {
421 VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_),
426 /* Between the socket() and bind() calls above, the socket receives all
427 * packets of the requested type on all system interfaces. We do not
428 * want to receive that data, but there is no way to avoid it. So we
429 * must now drain out the receive queue. */
430 error = drain_rcvbuf(netdev->fd);
436 *netdevp = &netdev->netdev;
440 netdev_uninit(&netdev->netdev, true);
444 /* Closes and destroys 'netdev'. */
446 netdev_linux_close(struct netdev *netdev_)
448 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
450 if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
456 /* Initializes 'svec' with a list of the names of all known network devices. */
458 netdev_linux_enumerate(struct svec *svec)
460 struct if_nameindex *names;
462 names = if_nameindex();
466 for (i = 0; names[i].if_name != NULL; i++) {
467 svec_add(svec, names[i].if_name);
469 if_freenameindex(names);
472 VLOG_WARN("could not obtain list of network device names: %s",
479 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
481 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
483 if (netdev->fd < 0) {
484 /* Device was opened with NETDEV_ETH_TYPE_NONE. */
489 ssize_t retval = read(netdev->fd, data, size);
492 } else if (errno != EINTR) {
493 if (errno != EAGAIN) {
494 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
495 strerror(errno), netdev_get_name(netdev_));
502 /* Registers with the poll loop to wake up from the next call to poll_block()
503 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
505 netdev_linux_recv_wait(struct netdev *netdev_)
507 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
508 if (netdev->fd >= 0) {
509 poll_fd_wait(netdev->fd, POLLIN);
513 /* Discards all packets waiting to be received from 'netdev'. */
515 netdev_linux_drain(struct netdev *netdev_)
517 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
518 if (netdev->fd < 0) {
520 } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
522 int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
523 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
527 drain_fd(netdev->fd, ifr.ifr_qlen);
530 return drain_rcvbuf(netdev->fd);
534 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
535 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
536 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
537 * the packet is too big or too small to transmit on the device.
539 * The caller retains ownership of 'buffer' in all cases.
541 * The kernel maintains a packet transmission queue, so the caller is not
542 * expected to do additional queuing of packets. */
544 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
546 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
548 /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
550 if (netdev->fd < 0) {
555 ssize_t retval = write(netdev->fd, data, size);
557 /* The Linux AF_PACKET implementation never blocks waiting for room
558 * for packets, instead returning ENOBUFS. Translate this into
559 * EAGAIN for the caller. */
560 if (errno == ENOBUFS) {
562 } else if (errno == EINTR) {
564 } else if (errno != EAGAIN) {
565 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
566 netdev_get_name(netdev_), strerror(errno));
569 } else if (retval != size) {
570 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
571 "%zu) on %s", retval, size, netdev_get_name(netdev_));
579 /* Registers with the poll loop to wake up from the next call to poll_block()
580 * when the packet transmission queue has sufficient room to transmit a packet
581 * with netdev_send().
583 * The kernel maintains a packet transmission queue, so the client is not
584 * expected to do additional queuing of packets. Thus, this function is
585 * unlikely to ever be used. It is included for completeness. */
587 netdev_linux_send_wait(struct netdev *netdev_)
589 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
590 if (netdev->fd < 0) {
592 } else if (strcmp(netdev_get_type(netdev_), "tap")) {
593 poll_fd_wait(netdev->fd, POLLOUT);
595 /* TAP device always accepts packets.*/
596 poll_immediate_wake();
600 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
601 * otherwise a positive errno value. */
603 netdev_linux_set_etheraddr(struct netdev *netdev_,
604 const uint8_t mac[ETH_ADDR_LEN])
606 struct netdev_dev_linux *netdev_dev =
607 netdev_dev_linux_cast(netdev_get_dev(netdev_));
610 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)
611 || !eth_addr_equals(netdev_dev->etheraddr, mac)) {
612 error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
614 netdev_dev->cache_valid |= VALID_ETHERADDR;
615 memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
623 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
624 * free the returned buffer. */
626 netdev_linux_get_etheraddr(const struct netdev *netdev_,
627 uint8_t mac[ETH_ADDR_LEN])
629 struct netdev_dev_linux *netdev_dev =
630 netdev_dev_linux_cast(netdev_get_dev(netdev_));
631 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
632 int error = get_etheraddr(netdev_get_name(netdev_),
633 netdev_dev->etheraddr);
637 netdev_dev->cache_valid |= VALID_ETHERADDR;
639 memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
643 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
644 * in bytes, not including the hardware header; thus, this is typically 1500
645 * bytes for Ethernet devices. */
647 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
649 struct netdev_dev_linux *netdev_dev =
650 netdev_dev_linux_cast(netdev_get_dev(netdev_));
651 if (!(netdev_dev->cache_valid & VALID_MTU)) {
655 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
656 SIOCGIFMTU, "SIOCGIFMTU");
660 netdev_dev->mtu = ifr.ifr_mtu;
661 netdev_dev->cache_valid |= VALID_MTU;
663 *mtup = netdev_dev->mtu;
667 /* Returns the ifindex of 'netdev', if successful, as a positive number.
668 * On failure, returns a negative errno value. */
670 netdev_linux_get_ifindex(const struct netdev *netdev)
674 error = get_ifindex(netdev, &ifindex);
675 return error ? -error : ifindex;
679 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
681 struct netdev_dev_linux *netdev_dev =
682 netdev_dev_linux_cast(netdev_get_dev(netdev_));
687 if (!(netdev_dev->cache_valid & VALID_CARRIER)) {
691 fn = xasprintf("/sys/class/net/%s/carrier",
692 netdev_get_name(netdev_));
693 fd = open(fn, O_RDONLY);
696 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error));
700 retval = read(fd, line, sizeof line);
703 if (error == EINVAL) {
704 /* This is the normal return value when we try to check carrier
705 * if the network device is not up. */
707 VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error));
710 } else if (retval == 0) {
712 VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn);
716 if (line[0] != '0' && line[0] != '1') {
718 VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)",
722 netdev_dev->carrier = line[0] != '0';
723 netdev_dev->cache_valid |= VALID_CARRIER;
725 *carrier = netdev_dev->carrier;
736 /* Check whether we can we use RTM_GETLINK to get network device statistics.
737 * In pre-2.6.19 kernels, this was only available if wireless extensions were
740 check_for_working_netlink_stats(void)
742 /* Decide on the netdev_get_stats() implementation to use. Netlink is
743 * preferable, so if that works, we'll use it. */
744 int ifindex = do_get_ifindex("lo");
746 VLOG_WARN("failed to get ifindex for lo, "
747 "obtaining netdev stats from proc");
750 struct netdev_stats stats;
751 int error = get_stats_via_netlink(ifindex, &stats);
753 VLOG_DBG("obtaining netdev stats via rtnetlink");
756 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
757 "via proc (you are probably running a pre-2.6.19 "
758 "kernel)", strerror(error));
764 /* Brings the 'is_internal' and 'is_tap' members of 'netdev_dev' up-to-date. */
766 netdev_linux_update_is_pseudo(struct netdev_dev_linux *netdev_dev)
768 if (!(netdev_dev->cache_valid & VALID_IS_PSEUDO)) {
769 const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
770 const char *type = netdev_dev_get_type(&netdev_dev->netdev_dev);
772 netdev_dev->is_tap = !strcmp(type, "tap");
773 netdev_dev->is_internal = false;
774 if (!netdev_dev->is_tap) {
775 struct ethtool_drvinfo drvinfo;
778 memset(&drvinfo, 0, sizeof drvinfo);
779 error = netdev_linux_do_ethtool(name,
780 (struct ethtool_cmd *)&drvinfo,
784 if (!error && !strcmp(drvinfo.driver, "openvswitch")) {
785 netdev_dev->is_internal = true;
789 netdev_dev->cache_valid |= VALID_IS_PSEUDO;
794 swap_uint64(uint64_t *a, uint64_t *b)
801 /* Retrieves current device stats for 'netdev'. */
803 netdev_linux_get_stats(const struct netdev *netdev_,
804 struct netdev_stats *stats)
806 struct netdev_dev_linux *netdev_dev =
807 netdev_dev_linux_cast(netdev_get_dev(netdev_));
808 static int use_netlink_stats = -1;
811 COVERAGE_INC(netdev_get_stats);
813 if (netdev_dev->have_vport_stats ||
814 !(netdev_dev->cache_valid & VALID_HAVE_VPORT_STATS)) {
816 error = netdev_vport_get_stats(netdev_, stats);
817 netdev_dev->have_vport_stats = !error;
818 netdev_dev->cache_valid |= VALID_HAVE_VPORT_STATS;
821 if (!netdev_dev->have_vport_stats) {
822 if (use_netlink_stats < 0) {
823 use_netlink_stats = check_for_working_netlink_stats();
825 if (use_netlink_stats) {
828 error = get_ifindex(netdev_, &ifindex);
830 error = get_stats_via_netlink(ifindex, stats);
833 error = get_stats_via_proc(netdev_get_name(netdev_), stats);
837 /* If this port is an internal port then the transmit and receive stats
838 * will appear to be swapped relative to the other ports since we are the
839 * one sending the data, not a remote computer. For consistency, we swap
840 * them back here. This does not apply if we are getting stats from the
841 * vport layer because it always tracks stats from the perspective of the
843 netdev_linux_update_is_pseudo(netdev_dev);
844 if (!error && !netdev_dev->have_vport_stats &&
845 (netdev_dev->is_internal || netdev_dev->is_tap)) {
846 swap_uint64(&stats->rx_packets, &stats->tx_packets);
847 swap_uint64(&stats->rx_bytes, &stats->tx_bytes);
848 swap_uint64(&stats->rx_errors, &stats->tx_errors);
849 swap_uint64(&stats->rx_dropped, &stats->tx_dropped);
850 stats->rx_length_errors = 0;
851 stats->rx_over_errors = 0;
852 stats->rx_crc_errors = 0;
853 stats->rx_frame_errors = 0;
854 stats->rx_fifo_errors = 0;
855 stats->rx_missed_errors = 0;
856 stats->tx_aborted_errors = 0;
857 stats->tx_carrier_errors = 0;
858 stats->tx_fifo_errors = 0;
859 stats->tx_heartbeat_errors = 0;
860 stats->tx_window_errors = 0;
867 netdev_linux_set_stats(struct netdev *netdev,
868 const struct netdev_stats *stats)
870 struct netdev_dev_linux *netdev_dev =
871 netdev_dev_linux_cast(netdev_get_dev(netdev));
872 struct internal_dev_stats dp_dev_stats;
875 /* We must reject this call if 'netdev' is not an Open vSwitch internal
876 * port, because the ioctl that we are about to execute is in the "device
877 * private ioctls" range, which means that executing it on a device that
878 * is not the type we expect could do any random thing.
880 * (Amusingly, these ioctl numbers are commented "THESE IOCTLS ARE
881 * _DEPRECATED_ AND WILL DISAPPEAR IN 2.5.X" in linux/sockios.h. I guess
882 * DaveM is a little behind on that.) */
883 netdev_linux_update_is_pseudo(netdev_dev);
884 if (!netdev_dev->is_internal) {
888 /* This actually only sets the *offset* that the dp_dev applies, but in our
889 * usage for fake bond devices the dp_dev never has any traffic of it own
890 * so it has the same effect. */
891 dp_dev_stats.rx_packets = stats->rx_packets;
892 dp_dev_stats.rx_bytes = stats->rx_bytes;
893 dp_dev_stats.tx_packets = stats->tx_packets;
894 dp_dev_stats.tx_bytes = stats->tx_bytes;
895 ifr.ifr_data = (void *) &dp_dev_stats;
896 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr,
897 INTERNAL_DEV_SET_STATS,
898 "INTERNAL_DEV_SET_STATS");
901 /* Stores the features supported by 'netdev' into each of '*current',
902 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
903 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
904 * successful, otherwise a positive errno value. */
906 netdev_linux_get_features(struct netdev *netdev,
907 uint32_t *current, uint32_t *advertised,
908 uint32_t *supported, uint32_t *peer)
910 struct ethtool_cmd ecmd;
913 memset(&ecmd, 0, sizeof ecmd);
914 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
915 ETHTOOL_GSET, "ETHTOOL_GSET");
920 /* Supported features. */
922 if (ecmd.supported & SUPPORTED_10baseT_Half) {
923 *supported |= OFPPF_10MB_HD;
925 if (ecmd.supported & SUPPORTED_10baseT_Full) {
926 *supported |= OFPPF_10MB_FD;
928 if (ecmd.supported & SUPPORTED_100baseT_Half) {
929 *supported |= OFPPF_100MB_HD;
931 if (ecmd.supported & SUPPORTED_100baseT_Full) {
932 *supported |= OFPPF_100MB_FD;
934 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
935 *supported |= OFPPF_1GB_HD;
937 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
938 *supported |= OFPPF_1GB_FD;
940 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
941 *supported |= OFPPF_10GB_FD;
943 if (ecmd.supported & SUPPORTED_TP) {
944 *supported |= OFPPF_COPPER;
946 if (ecmd.supported & SUPPORTED_FIBRE) {
947 *supported |= OFPPF_FIBER;
949 if (ecmd.supported & SUPPORTED_Autoneg) {
950 *supported |= OFPPF_AUTONEG;
952 if (ecmd.supported & SUPPORTED_Pause) {
953 *supported |= OFPPF_PAUSE;
955 if (ecmd.supported & SUPPORTED_Asym_Pause) {
956 *supported |= OFPPF_PAUSE_ASYM;
959 /* Advertised features. */
961 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
962 *advertised |= OFPPF_10MB_HD;
964 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
965 *advertised |= OFPPF_10MB_FD;
967 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
968 *advertised |= OFPPF_100MB_HD;
970 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
971 *advertised |= OFPPF_100MB_FD;
973 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
974 *advertised |= OFPPF_1GB_HD;
976 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
977 *advertised |= OFPPF_1GB_FD;
979 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
980 *advertised |= OFPPF_10GB_FD;
982 if (ecmd.advertising & ADVERTISED_TP) {
983 *advertised |= OFPPF_COPPER;
985 if (ecmd.advertising & ADVERTISED_FIBRE) {
986 *advertised |= OFPPF_FIBER;
988 if (ecmd.advertising & ADVERTISED_Autoneg) {
989 *advertised |= OFPPF_AUTONEG;
991 if (ecmd.advertising & ADVERTISED_Pause) {
992 *advertised |= OFPPF_PAUSE;
994 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
995 *advertised |= OFPPF_PAUSE_ASYM;
998 /* Current settings. */
999 if (ecmd.speed == SPEED_10) {
1000 *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD;
1001 } else if (ecmd.speed == SPEED_100) {
1002 *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD;
1003 } else if (ecmd.speed == SPEED_1000) {
1004 *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD;
1005 } else if (ecmd.speed == SPEED_10000) {
1006 *current = OFPPF_10GB_FD;
1011 if (ecmd.port == PORT_TP) {
1012 *current |= OFPPF_COPPER;
1013 } else if (ecmd.port == PORT_FIBRE) {
1014 *current |= OFPPF_FIBER;
1018 *current |= OFPPF_AUTONEG;
1021 /* Peer advertisements. */
1022 *peer = 0; /* XXX */
1027 /* Set the features advertised by 'netdev' to 'advertise'. */
1029 netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
1031 struct ethtool_cmd ecmd;
1034 memset(&ecmd, 0, sizeof ecmd);
1035 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1036 ETHTOOL_GSET, "ETHTOOL_GSET");
1041 ecmd.advertising = 0;
1042 if (advertise & OFPPF_10MB_HD) {
1043 ecmd.advertising |= ADVERTISED_10baseT_Half;
1045 if (advertise & OFPPF_10MB_FD) {
1046 ecmd.advertising |= ADVERTISED_10baseT_Full;
1048 if (advertise & OFPPF_100MB_HD) {
1049 ecmd.advertising |= ADVERTISED_100baseT_Half;
1051 if (advertise & OFPPF_100MB_FD) {
1052 ecmd.advertising |= ADVERTISED_100baseT_Full;
1054 if (advertise & OFPPF_1GB_HD) {
1055 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1057 if (advertise & OFPPF_1GB_FD) {
1058 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1060 if (advertise & OFPPF_10GB_FD) {
1061 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1063 if (advertise & OFPPF_COPPER) {
1064 ecmd.advertising |= ADVERTISED_TP;
1066 if (advertise & OFPPF_FIBER) {
1067 ecmd.advertising |= ADVERTISED_FIBRE;
1069 if (advertise & OFPPF_AUTONEG) {
1070 ecmd.advertising |= ADVERTISED_Autoneg;
1072 if (advertise & OFPPF_PAUSE) {
1073 ecmd.advertising |= ADVERTISED_Pause;
1075 if (advertise & OFPPF_PAUSE_ASYM) {
1076 ecmd.advertising |= ADVERTISED_Asym_Pause;
1078 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1079 ETHTOOL_SSET, "ETHTOOL_SSET");
1082 /* If 'netdev_name' is the name of a VLAN network device (e.g. one created with
1083 * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device
1084 * and returns 0. Otherwise returns a errno value (specifically ENOENT if
1085 * 'netdev_name' is the name of a network device that is not a VLAN device) and
1086 * sets '*vlan_vid' to -1. */
1088 netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1090 const char *netdev_name = netdev_get_name(netdev);
1091 struct ds line = DS_EMPTY_INITIALIZER;
1092 FILE *stream = NULL;
1096 COVERAGE_INC(netdev_get_vlan_vid);
1097 fn = xasprintf("/proc/net/vlan/%s", netdev_name);
1098 stream = fopen(fn, "r");
1104 if (ds_get_line(&line, stream)) {
1105 if (ferror(stream)) {
1107 VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno));
1110 VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn);
1115 if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) {
1117 VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"",
1118 fn, ds_cstr(&line));
1136 #define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress"
1137 #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"
1139 /* Remove ingress policing from 'netdev'. Returns 0 if successful, otherwise a
1140 * positive errno value.
1142 * This function is equivalent to running
1143 * /sbin/tc qdisc del dev %s handle ffff: ingress
1144 * but it is much, much faster.
1147 netdev_linux_remove_policing(struct netdev *netdev)
1149 struct netdev_dev_linux *netdev_dev =
1150 netdev_dev_linux_cast(netdev_get_dev(netdev));
1151 const char *netdev_name = netdev_get_name(netdev);
1153 struct ofpbuf request;
1154 struct ofpbuf *reply;
1155 struct tcmsg *tcmsg;
1156 struct nl_sock *rtnl_sock;
1160 error = get_ifindex(netdev, &ifindex);
1165 error = get_rtnl_sock(&rtnl_sock);
1170 ofpbuf_init(&request, 0);
1171 nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *tcmsg,
1172 RTM_DELQDISC, NLM_F_REQUEST);
1173 tcmsg = ofpbuf_put_zeros(&request, sizeof *tcmsg);
1174 tcmsg->tcm_family = AF_UNSPEC;
1175 tcmsg->tcm_ifindex = ifindex;
1176 tcmsg->tcm_handle = 0xffff0000;
1177 tcmsg->tcm_parent = TC_H_INGRESS;
1178 nl_msg_put_string(&request, TCA_KIND, "ingress");
1179 nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0);
1180 error = nl_sock_transact(rtnl_sock, &request, &reply);
1181 ofpbuf_uninit(&request);
1182 ofpbuf_delete(reply);
1183 if (error && error != ENOENT && error != EINVAL) {
1184 VLOG_WARN_RL(&rl, "%s: removing policing failed: %s",
1185 netdev_name, strerror(error));
1189 netdev_dev->kbits_rate = 0;
1190 netdev_dev->kbits_burst = 0;
1191 netdev_dev->cache_valid |= VALID_POLICING;
1195 /* Attempts to set input rate limiting (policing) policy. */
1197 netdev_linux_set_policing(struct netdev *netdev,
1198 uint32_t kbits_rate, uint32_t kbits_burst)
1200 struct netdev_dev_linux *netdev_dev =
1201 netdev_dev_linux_cast(netdev_get_dev(netdev));
1202 const char *netdev_name = netdev_get_name(netdev);
1205 COVERAGE_INC(netdev_set_policing);
1207 kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */
1208 : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */
1209 : kbits_burst); /* Stick with user-specified value. */
1211 if (netdev_dev->cache_valid & VALID_POLICING
1212 && netdev_dev->kbits_rate == kbits_rate
1213 && netdev_dev->kbits_burst == kbits_burst) {
1214 /* Assume that settings haven't changed since we last set them. */
1218 netdev_linux_remove_policing(netdev);
1220 snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name);
1221 if (system(command) != 0) {
1222 VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name);
1226 snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name,
1227 kbits_rate, kbits_burst);
1228 if (system(command) != 0) {
1229 VLOG_WARN_RL(&rl, "%s: problem configuring policing",
1234 netdev_dev->kbits_rate = kbits_rate;
1235 netdev_dev->kbits_burst = kbits_burst;
1236 netdev_dev->cache_valid |= VALID_POLICING;
1243 netdev_linux_get_in4(const struct netdev *netdev_,
1244 struct in_addr *address, struct in_addr *netmask)
1246 struct netdev_dev_linux *netdev_dev =
1247 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1249 if (!(netdev_dev->cache_valid & VALID_IN4)) {
1252 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
1253 SIOCGIFADDR, "SIOCGIFADDR");
1258 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
1259 SIOCGIFNETMASK, "SIOCGIFNETMASK");
1264 netdev_dev->cache_valid |= VALID_IN4;
1266 *address = netdev_dev->address;
1267 *netmask = netdev_dev->netmask;
1268 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
1272 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
1273 struct in_addr netmask)
1275 struct netdev_dev_linux *netdev_dev =
1276 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1279 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
1281 netdev_dev->cache_valid |= VALID_IN4;
1282 netdev_dev->address = address;
1283 netdev_dev->netmask = netmask;
1284 if (address.s_addr != INADDR_ANY) {
1285 error = do_set_addr(netdev_, SIOCSIFNETMASK,
1286 "SIOCSIFNETMASK", netmask);
1293 parse_if_inet6_line(const char *line,
1294 struct in6_addr *in6, char ifname[16 + 1])
1296 uint8_t *s6 = in6->s6_addr;
1297 #define X8 "%2"SCNx8
1299 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
1300 "%*x %*x %*x %*x %16s\n",
1301 &s6[0], &s6[1], &s6[2], &s6[3],
1302 &s6[4], &s6[5], &s6[6], &s6[7],
1303 &s6[8], &s6[9], &s6[10], &s6[11],
1304 &s6[12], &s6[13], &s6[14], &s6[15],
1308 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
1309 * 'in6' is non-null) and returns true. Otherwise, returns false. */
1311 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
1313 struct netdev_dev_linux *netdev_dev =
1314 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1315 if (!(netdev_dev->cache_valid & VALID_IN6)) {
1319 netdev_dev->in6 = in6addr_any;
1321 file = fopen("/proc/net/if_inet6", "r");
1323 const char *name = netdev_get_name(netdev_);
1324 while (fgets(line, sizeof line, file)) {
1325 struct in6_addr in6;
1326 char ifname[16 + 1];
1327 if (parse_if_inet6_line(line, &in6, ifname)
1328 && !strcmp(name, ifname))
1330 netdev_dev->in6 = in6;
1336 netdev_dev->cache_valid |= VALID_IN6;
1338 *in6 = netdev_dev->in6;
1343 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
1345 struct sockaddr_in sin;
1346 memset(&sin, 0, sizeof sin);
1347 sin.sin_family = AF_INET;
1348 sin.sin_addr = addr;
1351 memset(sa, 0, sizeof *sa);
1352 memcpy(sa, &sin, sizeof sin);
1356 do_set_addr(struct netdev *netdev,
1357 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
1360 strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
1361 make_in4_sockaddr(&ifr.ifr_addr, addr);
1363 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
1367 /* Adds 'router' as a default IP gateway. */
1369 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
1371 struct in_addr any = { INADDR_ANY };
1375 memset(&rt, 0, sizeof rt);
1376 make_in4_sockaddr(&rt.rt_dst, any);
1377 make_in4_sockaddr(&rt.rt_gateway, router);
1378 make_in4_sockaddr(&rt.rt_genmask, any);
1379 rt.rt_flags = RTF_UP | RTF_GATEWAY;
1380 COVERAGE_INC(netdev_add_router);
1381 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
1383 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
1389 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
1392 static const char fn[] = "/proc/net/route";
1397 *netdev_name = NULL;
1398 stream = fopen(fn, "r");
1399 if (stream == NULL) {
1400 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1405 while (fgets(line, sizeof line, stream)) {
1408 uint32_t dest, gateway, mask;
1409 int refcnt, metric, mtu;
1410 unsigned int flags, use, window, irtt;
1413 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
1415 iface, &dest, &gateway, &flags, &refcnt,
1416 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
1418 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
1422 if (!(flags & RTF_UP)) {
1423 /* Skip routes that aren't up. */
1427 /* The output of 'dest', 'mask', and 'gateway' were given in
1428 * network byte order, so we don't need need any endian
1429 * conversions here. */
1430 if ((dest & mask) == (host->s_addr & mask)) {
1432 /* The host is directly reachable. */
1433 next_hop->s_addr = 0;
1435 /* To reach the host, we must go through a gateway. */
1436 next_hop->s_addr = gateway;
1438 *netdev_name = xstrdup(iface);
1449 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1450 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1451 * returns 0. Otherwise, it returns a positive errno value; in particular,
1452 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
1454 netdev_linux_arp_lookup(const struct netdev *netdev,
1455 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
1458 struct sockaddr_in sin;
1461 memset(&r, 0, sizeof r);
1462 sin.sin_family = AF_INET;
1463 sin.sin_addr.s_addr = ip;
1465 memcpy(&r.arp_pa, &sin, sizeof sin);
1466 r.arp_ha.sa_family = ARPHRD_ETHER;
1468 strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
1469 COVERAGE_INC(netdev_arp_lookup);
1470 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
1472 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
1473 } else if (retval != ENXIO) {
1474 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
1475 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
1481 nd_to_iff_flags(enum netdev_flags nd)
1484 if (nd & NETDEV_UP) {
1487 if (nd & NETDEV_PROMISC) {
1494 iff_to_nd_flags(int iff)
1496 enum netdev_flags nd = 0;
1500 if (iff & IFF_PROMISC) {
1501 nd |= NETDEV_PROMISC;
1507 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
1508 enum netdev_flags on, enum netdev_flags *old_flagsp)
1510 int old_flags, new_flags;
1513 error = get_flags(netdev, &old_flags);
1515 *old_flagsp = iff_to_nd_flags(old_flags);
1516 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
1517 if (new_flags != old_flags) {
1518 error = set_flags(netdev, new_flags);
1525 poll_notify(struct list *list)
1527 struct netdev_linux_notifier *notifier;
1528 LIST_FOR_EACH (notifier, struct netdev_linux_notifier, node, list) {
1529 struct netdev_notifier *n = ¬ifier->notifier;
1535 netdev_linux_poll_cb(const struct rtnetlink_change *change,
1536 void *aux OVS_UNUSED)
1539 struct list *list = shash_find_data(&netdev_linux_notifiers,
1545 struct shash_node *node;
1546 SHASH_FOR_EACH (node, &netdev_linux_notifiers) {
1547 poll_notify(node->data);
1553 netdev_linux_poll_add(struct netdev *netdev,
1554 void (*cb)(struct netdev_notifier *), void *aux,
1555 struct netdev_notifier **notifierp)
1557 const char *netdev_name = netdev_get_name(netdev);
1558 struct netdev_linux_notifier *notifier;
1561 if (shash_is_empty(&netdev_linux_notifiers)) {
1562 int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier,
1563 netdev_linux_poll_cb, NULL);
1569 list = shash_find_data(&netdev_linux_notifiers, netdev_name);
1571 list = xmalloc(sizeof *list);
1573 shash_add(&netdev_linux_notifiers, netdev_name, list);
1576 notifier = xmalloc(sizeof *notifier);
1577 netdev_notifier_init(¬ifier->notifier, netdev, cb, aux);
1578 list_push_back(list, ¬ifier->node);
1579 *notifierp = ¬ifier->notifier;
1584 netdev_linux_poll_remove(struct netdev_notifier *notifier_)
1586 struct netdev_linux_notifier *notifier =
1587 CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier);
1590 /* Remove 'notifier' from its list. */
1591 list = list_remove(¬ifier->node);
1592 if (list_is_empty(list)) {
1593 /* The list is now empty. Remove it from the hash and free it. */
1594 const char *netdev_name = netdev_get_name(notifier->notifier.netdev);
1595 shash_delete(&netdev_linux_notifiers,
1596 shash_find(&netdev_linux_notifiers, netdev_name));
1601 /* If that was the last notifier, unregister. */
1602 if (shash_is_empty(&netdev_linux_notifiers)) {
1603 rtnetlink_notifier_unregister(&netdev_linux_poll_notifier);
1607 const struct netdev_class netdev_linux_class = {
1614 netdev_linux_create_system,
1615 netdev_linux_destroy,
1616 NULL, /* reconfigure */
1621 netdev_linux_enumerate,
1624 netdev_linux_recv_wait,
1628 netdev_linux_send_wait,
1630 netdev_linux_set_etheraddr,
1631 netdev_linux_get_etheraddr,
1632 netdev_linux_get_mtu,
1633 netdev_linux_get_ifindex,
1634 netdev_linux_get_carrier,
1635 netdev_linux_get_stats,
1636 netdev_linux_set_stats,
1638 netdev_linux_get_features,
1639 netdev_linux_set_advertisements,
1640 netdev_linux_get_vlan_vid,
1641 netdev_linux_set_policing,
1643 netdev_linux_get_in4,
1644 netdev_linux_set_in4,
1645 netdev_linux_get_in6,
1646 netdev_linux_add_router,
1647 netdev_linux_get_next_hop,
1648 netdev_linux_arp_lookup,
1650 netdev_linux_update_flags,
1652 netdev_linux_poll_add,
1653 netdev_linux_poll_remove,
1656 const struct netdev_class netdev_tap_class = {
1663 netdev_linux_create_tap,
1664 netdev_linux_destroy,
1665 NULL, /* reconfigure */
1670 NULL, /* enumerate */
1673 netdev_linux_recv_wait,
1677 netdev_linux_send_wait,
1679 netdev_linux_set_etheraddr,
1680 netdev_linux_get_etheraddr,
1681 netdev_linux_get_mtu,
1682 netdev_linux_get_ifindex,
1683 netdev_linux_get_carrier,
1684 netdev_linux_get_stats,
1685 NULL, /* set_stats */
1687 netdev_linux_get_features,
1688 netdev_linux_set_advertisements,
1689 netdev_linux_get_vlan_vid,
1690 netdev_linux_set_policing,
1692 netdev_linux_get_in4,
1693 netdev_linux_set_in4,
1694 netdev_linux_get_in6,
1695 netdev_linux_add_router,
1696 netdev_linux_get_next_hop,
1697 netdev_linux_arp_lookup,
1699 netdev_linux_update_flags,
1701 netdev_linux_poll_add,
1702 netdev_linux_poll_remove,
1707 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
1709 /* Policy for RTNLGRP_LINK messages.
1711 * There are *many* more fields in these messages, but currently we only
1712 * care about these fields. */
1713 static const struct nl_policy rtnlgrp_link_policy[] = {
1714 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
1715 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
1716 .min_len = sizeof(struct rtnl_link_stats) },
1719 struct nl_sock *rtnl_sock;
1720 struct ofpbuf request;
1721 struct ofpbuf *reply;
1722 struct ifinfomsg *ifi;
1723 const struct rtnl_link_stats *rtnl_stats;
1724 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
1727 error = get_rtnl_sock(&rtnl_sock);
1732 ofpbuf_init(&request, 0);
1733 nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *ifi,
1734 RTM_GETLINK, NLM_F_REQUEST);
1735 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
1736 ifi->ifi_family = PF_UNSPEC;
1737 ifi->ifi_index = ifindex;
1738 error = nl_sock_transact(rtnl_sock, &request, &reply);
1739 ofpbuf_uninit(&request);
1744 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
1745 rtnlgrp_link_policy,
1746 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
1747 ofpbuf_delete(reply);
1751 if (!attrs[IFLA_STATS]) {
1752 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
1753 ofpbuf_delete(reply);
1757 rtnl_stats = nl_attr_get(attrs[IFLA_STATS]);
1758 stats->rx_packets = rtnl_stats->rx_packets;
1759 stats->tx_packets = rtnl_stats->tx_packets;
1760 stats->rx_bytes = rtnl_stats->rx_bytes;
1761 stats->tx_bytes = rtnl_stats->tx_bytes;
1762 stats->rx_errors = rtnl_stats->rx_errors;
1763 stats->tx_errors = rtnl_stats->tx_errors;
1764 stats->rx_dropped = rtnl_stats->rx_dropped;
1765 stats->tx_dropped = rtnl_stats->tx_dropped;
1766 stats->multicast = rtnl_stats->multicast;
1767 stats->collisions = rtnl_stats->collisions;
1768 stats->rx_length_errors = rtnl_stats->rx_length_errors;
1769 stats->rx_over_errors = rtnl_stats->rx_over_errors;
1770 stats->rx_crc_errors = rtnl_stats->rx_crc_errors;
1771 stats->rx_frame_errors = rtnl_stats->rx_frame_errors;
1772 stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors;
1773 stats->rx_missed_errors = rtnl_stats->rx_missed_errors;
1774 stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors;
1775 stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors;
1776 stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors;
1777 stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors;
1778 stats->tx_window_errors = rtnl_stats->tx_window_errors;
1780 ofpbuf_delete(reply);
1786 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
1788 static const char fn[] = "/proc/net/dev";
1793 stream = fopen(fn, "r");
1795 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1800 while (fgets(line, sizeof line, stream)) {
1803 #define X64 "%"SCNu64
1806 X64 X64 X64 X64 X64 X64 X64 "%*u"
1807 X64 X64 X64 X64 X64 X64 X64 "%*u",
1813 &stats->rx_fifo_errors,
1814 &stats->rx_frame_errors,
1820 &stats->tx_fifo_errors,
1822 &stats->tx_carrier_errors) != 15) {
1823 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
1824 } else if (!strcmp(devname, netdev_name)) {
1825 stats->rx_length_errors = UINT64_MAX;
1826 stats->rx_over_errors = UINT64_MAX;
1827 stats->rx_crc_errors = UINT64_MAX;
1828 stats->rx_missed_errors = UINT64_MAX;
1829 stats->tx_aborted_errors = UINT64_MAX;
1830 stats->tx_heartbeat_errors = UINT64_MAX;
1831 stats->tx_window_errors = UINT64_MAX;
1837 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
1843 get_flags(const struct netdev *netdev, int *flags)
1848 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS,
1850 *flags = ifr.ifr_flags;
1855 set_flags(struct netdev *netdev, int flags)
1859 ifr.ifr_flags = flags;
1860 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
1865 do_get_ifindex(const char *netdev_name)
1869 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
1870 COVERAGE_INC(netdev_get_ifindex);
1871 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
1872 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
1873 netdev_name, strerror(errno));
1876 return ifr.ifr_ifindex;
1880 get_ifindex(const struct netdev *netdev_, int *ifindexp)
1882 struct netdev_dev_linux *netdev_dev =
1883 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1885 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
1886 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
1890 netdev_dev->cache_valid |= VALID_IFINDEX;
1891 netdev_dev->ifindex = ifindex;
1893 *ifindexp = netdev_dev->ifindex;
1898 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
1903 memset(&ifr, 0, sizeof ifr);
1904 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
1905 COVERAGE_INC(netdev_get_hwaddr);
1906 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
1907 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
1908 netdev_name, strerror(errno));
1911 hwaddr_family = ifr.ifr_hwaddr.sa_family;
1912 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
1913 VLOG_WARN("%s device has unknown hardware address family %d",
1914 netdev_name, hwaddr_family);
1916 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
1921 set_etheraddr(const char *netdev_name, int hwaddr_family,
1922 const uint8_t mac[ETH_ADDR_LEN])
1926 memset(&ifr, 0, sizeof ifr);
1927 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
1928 ifr.ifr_hwaddr.sa_family = hwaddr_family;
1929 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
1930 COVERAGE_INC(netdev_set_hwaddr);
1931 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
1932 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
1933 netdev_name, strerror(errno));
1940 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
1941 int cmd, const char *cmd_name)
1945 memset(&ifr, 0, sizeof ifr);
1946 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
1947 ifr.ifr_data = (caddr_t) ecmd;
1950 COVERAGE_INC(netdev_ethtool);
1951 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
1954 if (errno != EOPNOTSUPP) {
1955 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
1956 "failed: %s", cmd_name, name, strerror(errno));
1958 /* The device doesn't support this operation. That's pretty
1959 * common, so there's no point in logging anything. */
1966 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
1967 const char *cmd_name)
1969 strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
1970 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
1971 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
1979 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
1980 int cmd, const char *cmd_name)
1985 ifr.ifr_addr.sa_family = AF_INET;
1986 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
1988 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
1989 *ip = sin->sin_addr;
1994 /* Obtains a Netlink routing socket that is not subscribed to any multicast
1995 * groups. Returns 0 if successful, otherwise a positive errno value. Stores
1996 * the socket in '*rtnl_sockp' if successful, otherwise a null pointer. */
1998 get_rtnl_sock(struct nl_sock **rtnl_sockp)
2000 static struct nl_sock *sock;
2004 error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &sock);
2006 VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",