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/internal_dev.h"
54 #include "openvswitch/gre.h"
56 #include "poll-loop.h"
57 #include "rtnetlink.h"
58 #include "socket-util.h"
62 #define THIS_MODULE VLM_netdev_linux
65 /* These were introduced in Linux 2.6.14, so they might be missing if we have
67 #ifndef ADVERTISED_Pause
68 #define ADVERTISED_Pause (1 << 13)
70 #ifndef ADVERTISED_Asym_Pause
71 #define ADVERTISED_Asym_Pause (1 << 14)
74 static struct rtnetlink_notifier netdev_linux_cache_notifier;
75 static int cache_notifier_refcount;
78 VALID_IFINDEX = 1 << 0,
79 VALID_ETHERADDR = 1 << 1,
83 VALID_CARRIER = 1 << 5,
84 VALID_IS_PSEUDO = 1 << 6 /* Represents is_internal and is_tap. */
95 struct netdev_dev_linux {
96 struct netdev_dev netdev_dev;
98 struct shash_node *shash_node;
99 unsigned int cache_valid;
101 /* The following are figured out "on demand" only. They are only valid
102 * when the corresponding VALID_* bit in 'cache_valid' is set. */
104 uint8_t etheraddr[ETH_ADDR_LEN];
105 struct in_addr address, netmask;
109 bool is_internal; /* Is this an openvswitch internal device? */
110 bool is_tap; /* Is this a tuntap device? */
113 struct tap_state tap;
114 struct patch_state patch;
118 struct netdev_linux {
119 struct netdev netdev;
123 /* An AF_INET socket (used for ioctl operations). */
124 static int af_inet_sock = -1;
126 struct netdev_linux_notifier {
127 struct netdev_notifier notifier;
131 static struct shash netdev_linux_notifiers =
132 SHASH_INITIALIZER(&netdev_linux_notifiers);
133 static struct rtnetlink_notifier netdev_linux_poll_notifier;
135 /* This is set pretty low because we probably won't learn anything from the
136 * additional log messages. */
137 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
139 static int netdev_linux_init(void);
141 static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
142 int cmd, const char *cmd_name);
143 static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
144 const char *cmd_name);
145 static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
146 int cmd, const char *cmd_name);
147 static int get_flags(const struct netdev *, int *flagsp);
148 static int set_flags(struct netdev *, int flags);
149 static int do_get_ifindex(const char *netdev_name);
150 static int get_ifindex(const struct netdev *, int *ifindexp);
151 static int do_set_addr(struct netdev *netdev,
152 int ioctl_nr, const char *ioctl_name,
153 struct in_addr addr);
154 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
155 static int set_etheraddr(const char *netdev_name, int hwaddr_family,
156 const uint8_t[ETH_ADDR_LEN]);
157 static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
158 static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
161 is_netdev_linux_class(const struct netdev_class *netdev_class)
163 return netdev_class->init == netdev_linux_init;
166 static struct netdev_dev_linux *
167 netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
169 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
170 assert(is_netdev_linux_class(netdev_class));
172 return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
175 static struct netdev_linux *
176 netdev_linux_cast(const struct netdev *netdev)
178 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
179 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
180 assert(is_netdev_linux_class(netdev_class));
182 return CONTAINER_OF(netdev, struct netdev_linux, netdev);
186 netdev_linux_init(void)
188 static int status = -1;
190 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
191 status = af_inet_sock >= 0 ? 0 : errno;
193 VLOG_ERR("failed to create inet socket: %s", strerror(status));
200 netdev_linux_run(void)
202 rtnetlink_notifier_run();
206 netdev_linux_wait(void)
208 rtnetlink_notifier_wait();
212 netdev_linux_cache_cb(const struct rtnetlink_change *change,
213 void *aux OVS_UNUSED)
215 struct netdev_dev_linux *dev;
217 struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
219 const struct netdev_class *netdev_class =
220 netdev_dev_get_class(base_dev);
222 if (is_netdev_linux_class(netdev_class)) {
223 dev = netdev_dev_linux_cast(base_dev);
224 dev->cache_valid = 0;
228 struct shash device_shash;
229 struct shash_node *node;
231 shash_init(&device_shash);
232 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
233 SHASH_FOR_EACH (node, &device_shash) {
235 dev->cache_valid = 0;
237 shash_destroy(&device_shash);
242 if_up(const char *name)
246 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
247 ifr.ifr_flags = IFF_UP;
249 if (ioctl(af_inet_sock, SIOCSIFFLAGS, &ifr) == -1) {
250 VLOG_DBG_RL(&rl, "%s: failed to bring device up: %s",
251 name, strerror(errno));
258 /* A veth may be created using the 'command' "+<name>,<peer>". A veth may
259 * be destroyed by using the 'command' "-<name>", where <name> can be
260 * either side of the device.
263 modify_veth(const char *format, ...)
269 veth_file = fopen("/sys/class/net/veth_pairs", "w");
271 VLOG_WARN_RL(&rl, "could not open veth device. Are you running a "
272 "supported XenServer with the kernel module loaded?");
275 setvbuf(veth_file, NULL, _IONBF, 0);
277 va_start(args, format);
278 retval = vfprintf(veth_file, format, args);
283 VLOG_WARN_RL(&rl, "could not destroy patch: %s", strerror(errno));
291 create_patch(const char *name, const char *peer)
294 struct netdev_dev *peer_nd;
297 /* Only create the veth if the peer didn't already do it. */
298 peer_nd = netdev_dev_from_name(peer);
300 if (!strcmp("patch", netdev_dev_get_type(peer_nd))) {
301 struct netdev_dev_linux *ndl = netdev_dev_linux_cast(peer_nd);
302 if (!strcmp(name, ndl->state.patch.peer)) {
305 VLOG_WARN_RL(&rl, "peer '%s' already paired with '%s'",
306 peer, ndl->state.patch.peer);
310 VLOG_WARN_RL(&rl, "peer '%s' exists and is not a patch", peer);
315 retval = modify_veth("+%s,%s", name, peer);
320 retval = if_up(name);
325 retval = if_up(peer);
334 setup_patch(const char *name, const struct shash *args, char **peer_)
338 peer = shash_find_data(args, "peer");
340 VLOG_WARN("patch type requires valid 'peer' argument");
344 if (shash_count(args) > 1) {
345 VLOG_WARN("patch type takes only a 'peer' argument");
349 if (strlen(peer) >= IFNAMSIZ) {
350 VLOG_WARN_RL(&rl, "patch 'peer' arg too long");
354 *peer_ = xstrdup(peer);
355 return create_patch(name, peer);
358 /* Creates the netdev device of 'type' with 'name'. */
360 netdev_linux_create_system(const char *name, const char *type OVS_UNUSED,
361 const struct shash *args, struct netdev_dev **netdev_devp)
363 struct netdev_dev_linux *netdev_dev;
366 if (!shash_is_empty(args)) {
367 VLOG_WARN("%s: arguments for system devices should be empty", name);
370 if (!cache_notifier_refcount) {
371 error = rtnetlink_notifier_register(&netdev_linux_cache_notifier,
372 netdev_linux_cache_cb, NULL);
377 cache_notifier_refcount++;
379 netdev_dev = xzalloc(sizeof *netdev_dev);
380 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class);
382 *netdev_devp = &netdev_dev->netdev_dev;
386 /* For most types of netdevs we open the device for each call of
387 * netdev_open(). However, this is not the case with tap devices,
388 * since it is only possible to open the device once. In this
389 * situation we share a single file descriptor, and consequently
390 * buffers, across all readers. Therefore once data is read it will
391 * be unavailable to other reads for tap devices. */
393 netdev_linux_create_tap(const char *name, const char *type OVS_UNUSED,
394 const struct shash *args, struct netdev_dev **netdev_devp)
396 struct netdev_dev_linux *netdev_dev;
397 struct tap_state *state;
398 static const char tap_dev[] = "/dev/net/tun";
402 if (!shash_is_empty(args)) {
403 VLOG_WARN("%s: arguments for TAP devices should be empty", name);
406 netdev_dev = xzalloc(sizeof *netdev_dev);
407 state = &netdev_dev->state.tap;
409 /* Open tap device. */
410 state->fd = open(tap_dev, O_RDWR);
413 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
417 /* Create tap device. */
418 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
419 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
420 if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
421 VLOG_WARN("%s: creating tap device failed: %s", name,
427 /* Make non-blocking. */
428 error = set_nonblocking(state->fd);
433 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
434 *netdev_devp = &netdev_dev->netdev_dev;
443 netdev_linux_create_patch(const char *name, const char *type OVS_UNUSED,
444 const struct shash *args, struct netdev_dev **netdev_devp)
446 struct netdev_dev_linux *netdev_dev;
450 error = setup_patch(name, args, &peer);
456 netdev_dev = xzalloc(sizeof *netdev_dev);
457 netdev_dev->state.patch.peer = peer;
458 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_patch_class);
459 *netdev_devp = &netdev_dev->netdev_dev;
465 destroy_tap(struct netdev_dev_linux *netdev_dev)
467 struct tap_state *state = &netdev_dev->state.tap;
469 if (state->fd >= 0) {
475 destroy_patch(struct netdev_dev_linux *netdev_dev)
477 const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
478 struct patch_state *state = &netdev_dev->state.patch;
480 /* Only destroy veth if 'peer' doesn't exist as an existing netdev. */
481 if (!netdev_dev_from_name(state->peer)) {
482 modify_veth("-%s", name);
487 /* Destroys the netdev device 'netdev_dev_'. */
489 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
491 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
492 const char *type = netdev_dev_get_type(netdev_dev_);
494 if (!strcmp(type, "system")) {
495 cache_notifier_refcount--;
497 if (!cache_notifier_refcount) {
498 rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
500 } else if (!strcmp(type, "tap")) {
501 destroy_tap(netdev_dev);
502 } else if (!strcmp(type, "patch")) {
503 destroy_patch(netdev_dev);
510 netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype,
511 struct netdev **netdevp)
513 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
514 struct netdev_linux *netdev;
515 enum netdev_flags flags;
518 /* Allocate network device. */
519 netdev = xzalloc(sizeof *netdev);
521 netdev_init(&netdev->netdev, netdev_dev_);
523 error = netdev_get_flags(&netdev->netdev, &flags);
524 if (error == ENODEV) {
528 if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap")) {
529 netdev->fd = netdev_dev->state.tap.fd;
530 } else if (ethertype != NETDEV_ETH_TYPE_NONE) {
531 struct sockaddr_ll sll;
535 /* Create file descriptor. */
536 protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
537 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
539 netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
540 if (netdev->fd < 0) {
545 /* Set non-blocking mode. */
546 error = set_nonblocking(netdev->fd);
551 /* Get ethernet device index. */
552 error = get_ifindex(&netdev->netdev, &ifindex);
557 /* Bind to specific ethernet device. */
558 memset(&sll, 0, sizeof sll);
559 sll.sll_family = AF_PACKET;
560 sll.sll_ifindex = ifindex;
562 (struct sockaddr *) &sll, sizeof sll) < 0) {
564 VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_),
569 /* Between the socket() and bind() calls above, the socket receives all
570 * packets of the requested type on all system interfaces. We do not
571 * want to receive that data, but there is no way to avoid it. So we
572 * must now drain out the receive queue. */
573 error = drain_rcvbuf(netdev->fd);
579 *netdevp = &netdev->netdev;
583 netdev_uninit(&netdev->netdev, true);
587 /* Closes and destroys 'netdev'. */
589 netdev_linux_close(struct netdev *netdev_)
591 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
593 if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
599 /* Initializes 'svec' with a list of the names of all known network devices. */
601 netdev_linux_enumerate(struct svec *svec)
603 struct if_nameindex *names;
605 names = if_nameindex();
609 for (i = 0; names[i].if_name != NULL; i++) {
610 svec_add(svec, names[i].if_name);
612 if_freenameindex(names);
615 VLOG_WARN("could not obtain list of network device names: %s",
622 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
624 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
626 if (netdev->fd < 0) {
627 /* Device was opened with NETDEV_ETH_TYPE_NONE. */
632 ssize_t retval = read(netdev->fd, data, size);
635 } else if (errno != EINTR) {
636 if (errno != EAGAIN) {
637 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
638 strerror(errno), netdev_get_name(netdev_));
645 /* Registers with the poll loop to wake up from the next call to poll_block()
646 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
648 netdev_linux_recv_wait(struct netdev *netdev_)
650 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
651 if (netdev->fd >= 0) {
652 poll_fd_wait(netdev->fd, POLLIN);
656 /* Discards all packets waiting to be received from 'netdev'. */
658 netdev_linux_drain(struct netdev *netdev_)
660 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
661 if (netdev->fd < 0) {
663 } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
665 int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
666 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
670 drain_fd(netdev->fd, ifr.ifr_qlen);
673 return drain_rcvbuf(netdev->fd);
677 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
678 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
679 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
680 * the packet is too big or too small to transmit on the device.
682 * The caller retains ownership of 'buffer' in all cases.
684 * The kernel maintains a packet transmission queue, so the caller is not
685 * expected to do additional queuing of packets. */
687 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
689 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
691 /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
693 if (netdev->fd < 0) {
698 ssize_t retval = write(netdev->fd, data, size);
700 /* The Linux AF_PACKET implementation never blocks waiting for room
701 * for packets, instead returning ENOBUFS. Translate this into
702 * EAGAIN for the caller. */
703 if (errno == ENOBUFS) {
705 } else if (errno == EINTR) {
707 } else if (errno != EAGAIN) {
708 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
709 netdev_get_name(netdev_), strerror(errno));
712 } else if (retval != size) {
713 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
714 "%zu) on %s", retval, size, netdev_get_name(netdev_));
722 /* Registers with the poll loop to wake up from the next call to poll_block()
723 * when the packet transmission queue has sufficient room to transmit a packet
724 * with netdev_send().
726 * The kernel maintains a packet transmission queue, so the client is not
727 * expected to do additional queuing of packets. Thus, this function is
728 * unlikely to ever be used. It is included for completeness. */
730 netdev_linux_send_wait(struct netdev *netdev_)
732 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
733 if (netdev->fd < 0) {
735 } else if (strcmp(netdev_get_type(netdev_), "tap")) {
736 poll_fd_wait(netdev->fd, POLLOUT);
738 /* TAP device always accepts packets.*/
739 poll_immediate_wake();
743 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
744 * otherwise a positive errno value. */
746 netdev_linux_set_etheraddr(struct netdev *netdev_,
747 const uint8_t mac[ETH_ADDR_LEN])
749 struct netdev_dev_linux *netdev_dev =
750 netdev_dev_linux_cast(netdev_get_dev(netdev_));
753 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)
754 || !eth_addr_equals(netdev_dev->etheraddr, mac)) {
755 error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
757 netdev_dev->cache_valid |= VALID_ETHERADDR;
758 memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
766 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
767 * free the returned buffer. */
769 netdev_linux_get_etheraddr(const struct netdev *netdev_,
770 uint8_t mac[ETH_ADDR_LEN])
772 struct netdev_dev_linux *netdev_dev =
773 netdev_dev_linux_cast(netdev_get_dev(netdev_));
774 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
775 int error = get_etheraddr(netdev_get_name(netdev_),
776 netdev_dev->etheraddr);
780 netdev_dev->cache_valid |= VALID_ETHERADDR;
782 memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
786 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
787 * in bytes, not including the hardware header; thus, this is typically 1500
788 * bytes for Ethernet devices. */
790 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
792 struct netdev_dev_linux *netdev_dev =
793 netdev_dev_linux_cast(netdev_get_dev(netdev_));
794 if (!(netdev_dev->cache_valid & VALID_MTU)) {
798 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
799 SIOCGIFMTU, "SIOCGIFMTU");
803 netdev_dev->mtu = ifr.ifr_mtu;
804 netdev_dev->cache_valid |= VALID_MTU;
806 *mtup = netdev_dev->mtu;
810 /* Returns the ifindex of 'netdev', if successful, as a positive number.
811 * On failure, returns a negative errno value. */
813 netdev_linux_get_ifindex(const struct netdev *netdev)
817 error = get_ifindex(netdev, &ifindex);
818 return error ? -error : ifindex;
822 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
824 struct netdev_dev_linux *netdev_dev =
825 netdev_dev_linux_cast(netdev_get_dev(netdev_));
830 if (!(netdev_dev->cache_valid & VALID_CARRIER)) {
834 fn = xasprintf("/sys/class/net/%s/carrier",
835 netdev_get_name(netdev_));
836 fd = open(fn, O_RDONLY);
839 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error));
843 retval = read(fd, line, sizeof line);
846 if (error == EINVAL) {
847 /* This is the normal return value when we try to check carrier
848 * if the network device is not up. */
850 VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error));
853 } else if (retval == 0) {
855 VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn);
859 if (line[0] != '0' && line[0] != '1') {
861 VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)",
865 netdev_dev->carrier = line[0] != '0';
866 netdev_dev->cache_valid |= VALID_CARRIER;
868 *carrier = netdev_dev->carrier;
879 /* Check whether we can we use RTM_GETLINK to get network device statistics.
880 * In pre-2.6.19 kernels, this was only available if wireless extensions were
883 check_for_working_netlink_stats(void)
885 /* Decide on the netdev_get_stats() implementation to use. Netlink is
886 * preferable, so if that works, we'll use it. */
887 int ifindex = do_get_ifindex("lo");
889 VLOG_WARN("failed to get ifindex for lo, "
890 "obtaining netdev stats from proc");
893 struct netdev_stats stats;
894 int error = get_stats_via_netlink(ifindex, &stats);
896 VLOG_DBG("obtaining netdev stats via rtnetlink");
899 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
900 "via proc (you are probably running a pre-2.6.19 "
901 "kernel)", strerror(error));
907 /* Brings the 'is_internal' and 'is_tap' members of 'netdev_dev' up-to-date. */
909 netdev_linux_update_is_pseudo(struct netdev_dev_linux *netdev_dev)
911 if (!(netdev_dev->cache_valid & VALID_IS_PSEUDO)) {
912 const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
913 const char *type = netdev_dev_get_type(&netdev_dev->netdev_dev);
915 netdev_dev->is_tap = !strcmp(type, "tap");
916 netdev_dev->is_internal = false;
917 if (!netdev_dev->is_tap) {
918 struct ethtool_drvinfo drvinfo;
921 memset(&drvinfo, 0, sizeof drvinfo);
922 error = netdev_linux_do_ethtool(name,
923 (struct ethtool_cmd *)&drvinfo,
927 if (!error && !strcmp(drvinfo.driver, "openvswitch")) {
928 netdev_dev->is_internal = true;
932 netdev_dev->cache_valid |= VALID_IS_PSEUDO;
936 /* Retrieves current device stats for 'netdev'.
938 * XXX All of the members of struct netdev_stats are 64 bits wide, but on
939 * 32-bit architectures the Linux network stats are only 32 bits. */
941 netdev_linux_get_stats(const struct netdev *netdev_,
942 struct netdev_stats *stats)
944 struct netdev_dev_linux *netdev_dev =
945 netdev_dev_linux_cast(netdev_get_dev(netdev_));
946 static int use_netlink_stats = -1;
948 struct netdev_stats raw_stats;
949 struct netdev_stats *collect_stats = stats;
951 COVERAGE_INC(netdev_get_stats);
953 netdev_linux_update_is_pseudo(netdev_dev);
954 if (netdev_dev->is_internal) {
955 collect_stats = &raw_stats;
958 if (use_netlink_stats < 0) {
959 use_netlink_stats = check_for_working_netlink_stats();
961 if (use_netlink_stats) {
964 error = get_ifindex(netdev_, &ifindex);
966 error = get_stats_via_netlink(ifindex, collect_stats);
969 error = get_stats_via_proc(netdev_get_name(netdev_), collect_stats);
972 /* If this port is an internal port then the transmit and receive stats
973 * will appear to be swapped relative to the other ports since we are the
974 * one sending the data, not a remote computer. For consistency, we swap
976 if (!error && (netdev_dev->is_internal || netdev_dev->is_tap)) {
977 stats->rx_packets = raw_stats.tx_packets;
978 stats->tx_packets = raw_stats.rx_packets;
979 stats->rx_bytes = raw_stats.tx_bytes;
980 stats->tx_bytes = raw_stats.rx_bytes;
981 stats->rx_errors = raw_stats.tx_errors;
982 stats->tx_errors = raw_stats.rx_errors;
983 stats->rx_dropped = raw_stats.tx_dropped;
984 stats->tx_dropped = raw_stats.rx_dropped;
985 stats->multicast = raw_stats.multicast;
986 stats->collisions = raw_stats.collisions;
987 stats->rx_length_errors = 0;
988 stats->rx_over_errors = 0;
989 stats->rx_crc_errors = 0;
990 stats->rx_frame_errors = 0;
991 stats->rx_fifo_errors = 0;
992 stats->rx_missed_errors = 0;
993 stats->tx_aborted_errors = 0;
994 stats->tx_carrier_errors = 0;
995 stats->tx_fifo_errors = 0;
996 stats->tx_heartbeat_errors = 0;
997 stats->tx_window_errors = 0;
1004 netdev_linux_set_stats(struct netdev *netdev,
1005 const struct netdev_stats *stats)
1007 struct netdev_dev_linux *netdev_dev =
1008 netdev_dev_linux_cast(netdev_get_dev(netdev));
1009 struct internal_dev_stats dp_dev_stats;
1012 /* We must reject this call if 'netdev' is not an Open vSwitch internal
1013 * port, because the ioctl that we are about to execute is in the "device
1014 * private ioctls" range, which means that executing it on a device that
1015 * is not the type we expect could do any random thing.
1017 * (Amusingly, these ioctl numbers are commented "THESE IOCTLS ARE
1018 * _DEPRECATED_ AND WILL DISAPPEAR IN 2.5.X" in linux/sockios.h. I guess
1019 * DaveM is a little behind on that.) */
1020 netdev_linux_update_is_pseudo(netdev_dev);
1021 if (!netdev_dev->is_internal) {
1025 /* This actually only sets the *offset* that the dp_dev applies, but in our
1026 * usage for fake bond devices the dp_dev never has any traffic of it own
1027 * so it has the same effect. */
1028 dp_dev_stats.rx_packets = stats->rx_packets;
1029 dp_dev_stats.rx_bytes = stats->rx_bytes;
1030 dp_dev_stats.tx_packets = stats->tx_packets;
1031 dp_dev_stats.tx_bytes = stats->tx_bytes;
1032 ifr.ifr_data = (void *) &dp_dev_stats;
1033 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr,
1034 INTERNAL_DEV_SET_STATS,
1035 "INTERNAL_DEV_SET_STATS");
1038 /* Stores the features supported by 'netdev' into each of '*current',
1039 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
1040 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
1041 * successful, otherwise a positive errno value. */
1043 netdev_linux_get_features(struct netdev *netdev,
1044 uint32_t *current, uint32_t *advertised,
1045 uint32_t *supported, uint32_t *peer)
1047 struct ethtool_cmd ecmd;
1050 memset(&ecmd, 0, sizeof ecmd);
1051 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1052 ETHTOOL_GSET, "ETHTOOL_GSET");
1057 /* Supported features. */
1059 if (ecmd.supported & SUPPORTED_10baseT_Half) {
1060 *supported |= OFPPF_10MB_HD;
1062 if (ecmd.supported & SUPPORTED_10baseT_Full) {
1063 *supported |= OFPPF_10MB_FD;
1065 if (ecmd.supported & SUPPORTED_100baseT_Half) {
1066 *supported |= OFPPF_100MB_HD;
1068 if (ecmd.supported & SUPPORTED_100baseT_Full) {
1069 *supported |= OFPPF_100MB_FD;
1071 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
1072 *supported |= OFPPF_1GB_HD;
1074 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
1075 *supported |= OFPPF_1GB_FD;
1077 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
1078 *supported |= OFPPF_10GB_FD;
1080 if (ecmd.supported & SUPPORTED_TP) {
1081 *supported |= OFPPF_COPPER;
1083 if (ecmd.supported & SUPPORTED_FIBRE) {
1084 *supported |= OFPPF_FIBER;
1086 if (ecmd.supported & SUPPORTED_Autoneg) {
1087 *supported |= OFPPF_AUTONEG;
1089 if (ecmd.supported & SUPPORTED_Pause) {
1090 *supported |= OFPPF_PAUSE;
1092 if (ecmd.supported & SUPPORTED_Asym_Pause) {
1093 *supported |= OFPPF_PAUSE_ASYM;
1096 /* Advertised features. */
1098 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
1099 *advertised |= OFPPF_10MB_HD;
1101 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
1102 *advertised |= OFPPF_10MB_FD;
1104 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
1105 *advertised |= OFPPF_100MB_HD;
1107 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
1108 *advertised |= OFPPF_100MB_FD;
1110 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
1111 *advertised |= OFPPF_1GB_HD;
1113 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
1114 *advertised |= OFPPF_1GB_FD;
1116 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
1117 *advertised |= OFPPF_10GB_FD;
1119 if (ecmd.advertising & ADVERTISED_TP) {
1120 *advertised |= OFPPF_COPPER;
1122 if (ecmd.advertising & ADVERTISED_FIBRE) {
1123 *advertised |= OFPPF_FIBER;
1125 if (ecmd.advertising & ADVERTISED_Autoneg) {
1126 *advertised |= OFPPF_AUTONEG;
1128 if (ecmd.advertising & ADVERTISED_Pause) {
1129 *advertised |= OFPPF_PAUSE;
1131 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
1132 *advertised |= OFPPF_PAUSE_ASYM;
1135 /* Current settings. */
1136 if (ecmd.speed == SPEED_10) {
1137 *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD;
1138 } else if (ecmd.speed == SPEED_100) {
1139 *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD;
1140 } else if (ecmd.speed == SPEED_1000) {
1141 *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD;
1142 } else if (ecmd.speed == SPEED_10000) {
1143 *current = OFPPF_10GB_FD;
1148 if (ecmd.port == PORT_TP) {
1149 *current |= OFPPF_COPPER;
1150 } else if (ecmd.port == PORT_FIBRE) {
1151 *current |= OFPPF_FIBER;
1155 *current |= OFPPF_AUTONEG;
1158 /* Peer advertisements. */
1159 *peer = 0; /* XXX */
1164 /* Set the features advertised by 'netdev' to 'advertise'. */
1166 netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
1168 struct ethtool_cmd ecmd;
1171 memset(&ecmd, 0, sizeof ecmd);
1172 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1173 ETHTOOL_GSET, "ETHTOOL_GSET");
1178 ecmd.advertising = 0;
1179 if (advertise & OFPPF_10MB_HD) {
1180 ecmd.advertising |= ADVERTISED_10baseT_Half;
1182 if (advertise & OFPPF_10MB_FD) {
1183 ecmd.advertising |= ADVERTISED_10baseT_Full;
1185 if (advertise & OFPPF_100MB_HD) {
1186 ecmd.advertising |= ADVERTISED_100baseT_Half;
1188 if (advertise & OFPPF_100MB_FD) {
1189 ecmd.advertising |= ADVERTISED_100baseT_Full;
1191 if (advertise & OFPPF_1GB_HD) {
1192 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1194 if (advertise & OFPPF_1GB_FD) {
1195 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1197 if (advertise & OFPPF_10GB_FD) {
1198 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1200 if (advertise & OFPPF_COPPER) {
1201 ecmd.advertising |= ADVERTISED_TP;
1203 if (advertise & OFPPF_FIBER) {
1204 ecmd.advertising |= ADVERTISED_FIBRE;
1206 if (advertise & OFPPF_AUTONEG) {
1207 ecmd.advertising |= ADVERTISED_Autoneg;
1209 if (advertise & OFPPF_PAUSE) {
1210 ecmd.advertising |= ADVERTISED_Pause;
1212 if (advertise & OFPPF_PAUSE_ASYM) {
1213 ecmd.advertising |= ADVERTISED_Asym_Pause;
1215 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1216 ETHTOOL_SSET, "ETHTOOL_SSET");
1219 /* If 'netdev_name' is the name of a VLAN network device (e.g. one created with
1220 * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device
1221 * and returns 0. Otherwise returns a errno value (specifically ENOENT if
1222 * 'netdev_name' is the name of a network device that is not a VLAN device) and
1223 * sets '*vlan_vid' to -1. */
1225 netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1227 const char *netdev_name = netdev_get_name(netdev);
1228 struct ds line = DS_EMPTY_INITIALIZER;
1229 FILE *stream = NULL;
1233 COVERAGE_INC(netdev_get_vlan_vid);
1234 fn = xasprintf("/proc/net/vlan/%s", netdev_name);
1235 stream = fopen(fn, "r");
1241 if (ds_get_line(&line, stream)) {
1242 if (ferror(stream)) {
1244 VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno));
1247 VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn);
1252 if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) {
1254 VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"",
1255 fn, ds_cstr(&line));
1273 #define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress"
1274 #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"
1275 /* We redirect stderr to /dev/null because we often want to remove all
1276 * traffic control configuration on a port so its in a known state. If
1277 * this done when there is no such configuration, tc complains, so we just
1280 #define POLICE_DEL_CMD "/sbin/tc qdisc del dev %s handle ffff: ingress 2>/dev/null"
1282 /* Attempts to set input rate limiting (policing) policy. */
1284 netdev_linux_set_policing(struct netdev *netdev,
1285 uint32_t kbits_rate, uint32_t kbits_burst)
1287 const char *netdev_name = netdev_get_name(netdev);
1290 COVERAGE_INC(netdev_set_policing);
1293 /* Default to 1000 kilobits if not specified. */
1297 /* xxx This should be more careful about only adding if it
1298 * xxx actually exists, as opposed to always deleting it. */
1299 snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1300 if (system(command) == -1) {
1301 VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1304 snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name);
1305 if (system(command) != 0) {
1306 VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name);
1310 snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name,
1311 kbits_rate, kbits_burst);
1312 if (system(command) != 0) {
1313 VLOG_WARN_RL(&rl, "%s: problem configuring policing",
1318 snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1319 if (system(command) == -1) {
1320 VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1328 netdev_linux_get_in4(const struct netdev *netdev_,
1329 struct in_addr *address, struct in_addr *netmask)
1331 struct netdev_dev_linux *netdev_dev =
1332 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1334 if (!(netdev_dev->cache_valid & VALID_IN4)) {
1337 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
1338 SIOCGIFADDR, "SIOCGIFADDR");
1343 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
1344 SIOCGIFNETMASK, "SIOCGIFNETMASK");
1349 netdev_dev->cache_valid |= VALID_IN4;
1351 *address = netdev_dev->address;
1352 *netmask = netdev_dev->netmask;
1353 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
1357 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
1358 struct in_addr netmask)
1360 struct netdev_dev_linux *netdev_dev =
1361 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1364 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
1366 netdev_dev->cache_valid |= VALID_IN4;
1367 netdev_dev->address = address;
1368 netdev_dev->netmask = netmask;
1369 if (address.s_addr != INADDR_ANY) {
1370 error = do_set_addr(netdev_, SIOCSIFNETMASK,
1371 "SIOCSIFNETMASK", netmask);
1378 parse_if_inet6_line(const char *line,
1379 struct in6_addr *in6, char ifname[16 + 1])
1381 uint8_t *s6 = in6->s6_addr;
1382 #define X8 "%2"SCNx8
1384 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
1385 "%*x %*x %*x %*x %16s\n",
1386 &s6[0], &s6[1], &s6[2], &s6[3],
1387 &s6[4], &s6[5], &s6[6], &s6[7],
1388 &s6[8], &s6[9], &s6[10], &s6[11],
1389 &s6[12], &s6[13], &s6[14], &s6[15],
1393 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
1394 * 'in6' is non-null) and returns true. Otherwise, returns false. */
1396 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
1398 struct netdev_dev_linux *netdev_dev =
1399 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1400 if (!(netdev_dev->cache_valid & VALID_IN6)) {
1404 netdev_dev->in6 = in6addr_any;
1406 file = fopen("/proc/net/if_inet6", "r");
1408 const char *name = netdev_get_name(netdev_);
1409 while (fgets(line, sizeof line, file)) {
1410 struct in6_addr in6;
1411 char ifname[16 + 1];
1412 if (parse_if_inet6_line(line, &in6, ifname)
1413 && !strcmp(name, ifname))
1415 netdev_dev->in6 = in6;
1421 netdev_dev->cache_valid |= VALID_IN6;
1423 *in6 = netdev_dev->in6;
1428 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
1430 struct sockaddr_in sin;
1431 memset(&sin, 0, sizeof sin);
1432 sin.sin_family = AF_INET;
1433 sin.sin_addr = addr;
1436 memset(sa, 0, sizeof *sa);
1437 memcpy(sa, &sin, sizeof sin);
1441 do_set_addr(struct netdev *netdev,
1442 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
1445 strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
1446 make_in4_sockaddr(&ifr.ifr_addr, addr);
1448 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
1452 /* Adds 'router' as a default IP gateway. */
1454 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
1456 struct in_addr any = { INADDR_ANY };
1460 memset(&rt, 0, sizeof rt);
1461 make_in4_sockaddr(&rt.rt_dst, any);
1462 make_in4_sockaddr(&rt.rt_gateway, router);
1463 make_in4_sockaddr(&rt.rt_genmask, any);
1464 rt.rt_flags = RTF_UP | RTF_GATEWAY;
1465 COVERAGE_INC(netdev_add_router);
1466 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
1468 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
1474 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
1477 static const char fn[] = "/proc/net/route";
1482 *netdev_name = NULL;
1483 stream = fopen(fn, "r");
1484 if (stream == NULL) {
1485 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1490 while (fgets(line, sizeof line, stream)) {
1493 uint32_t dest, gateway, mask;
1494 int refcnt, metric, mtu;
1495 unsigned int flags, use, window, irtt;
1498 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
1500 iface, &dest, &gateway, &flags, &refcnt,
1501 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
1503 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
1507 if (!(flags & RTF_UP)) {
1508 /* Skip routes that aren't up. */
1512 /* The output of 'dest', 'mask', and 'gateway' were given in
1513 * network byte order, so we don't need need any endian
1514 * conversions here. */
1515 if ((dest & mask) == (host->s_addr & mask)) {
1517 /* The host is directly reachable. */
1518 next_hop->s_addr = 0;
1520 /* To reach the host, we must go through a gateway. */
1521 next_hop->s_addr = gateway;
1523 *netdev_name = xstrdup(iface);
1534 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1535 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1536 * returns 0. Otherwise, it returns a positive errno value; in particular,
1537 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
1539 netdev_linux_arp_lookup(const struct netdev *netdev,
1540 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
1543 struct sockaddr_in sin;
1546 memset(&r, 0, sizeof r);
1547 sin.sin_family = AF_INET;
1548 sin.sin_addr.s_addr = ip;
1550 memcpy(&r.arp_pa, &sin, sizeof sin);
1551 r.arp_ha.sa_family = ARPHRD_ETHER;
1553 strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
1554 COVERAGE_INC(netdev_arp_lookup);
1555 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
1557 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
1558 } else if (retval != ENXIO) {
1559 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
1560 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
1566 nd_to_iff_flags(enum netdev_flags nd)
1569 if (nd & NETDEV_UP) {
1572 if (nd & NETDEV_PROMISC) {
1579 iff_to_nd_flags(int iff)
1581 enum netdev_flags nd = 0;
1585 if (iff & IFF_PROMISC) {
1586 nd |= NETDEV_PROMISC;
1592 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
1593 enum netdev_flags on, enum netdev_flags *old_flagsp)
1595 int old_flags, new_flags;
1598 error = get_flags(netdev, &old_flags);
1600 *old_flagsp = iff_to_nd_flags(old_flags);
1601 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
1602 if (new_flags != old_flags) {
1603 error = set_flags(netdev, new_flags);
1610 poll_notify(struct list *list)
1612 struct netdev_linux_notifier *notifier;
1613 LIST_FOR_EACH (notifier, struct netdev_linux_notifier, node, list) {
1614 struct netdev_notifier *n = ¬ifier->notifier;
1620 netdev_linux_poll_cb(const struct rtnetlink_change *change,
1621 void *aux OVS_UNUSED)
1624 struct list *list = shash_find_data(&netdev_linux_notifiers,
1630 struct shash_node *node;
1631 SHASH_FOR_EACH (node, &netdev_linux_notifiers) {
1632 poll_notify(node->data);
1638 netdev_linux_poll_add(struct netdev *netdev,
1639 void (*cb)(struct netdev_notifier *), void *aux,
1640 struct netdev_notifier **notifierp)
1642 const char *netdev_name = netdev_get_name(netdev);
1643 struct netdev_linux_notifier *notifier;
1646 if (shash_is_empty(&netdev_linux_notifiers)) {
1647 int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier,
1648 netdev_linux_poll_cb, NULL);
1654 list = shash_find_data(&netdev_linux_notifiers, netdev_name);
1656 list = xmalloc(sizeof *list);
1658 shash_add(&netdev_linux_notifiers, netdev_name, list);
1661 notifier = xmalloc(sizeof *notifier);
1662 netdev_notifier_init(¬ifier->notifier, netdev, cb, aux);
1663 list_push_back(list, ¬ifier->node);
1664 *notifierp = ¬ifier->notifier;
1669 netdev_linux_poll_remove(struct netdev_notifier *notifier_)
1671 struct netdev_linux_notifier *notifier =
1672 CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier);
1675 /* Remove 'notifier' from its list. */
1676 list = list_remove(¬ifier->node);
1677 if (list_is_empty(list)) {
1678 /* The list is now empty. Remove it from the hash and free it. */
1679 const char *netdev_name = netdev_get_name(notifier->notifier.netdev);
1680 shash_delete(&netdev_linux_notifiers,
1681 shash_find(&netdev_linux_notifiers, netdev_name));
1686 /* If that was the last notifier, unregister. */
1687 if (shash_is_empty(&netdev_linux_notifiers)) {
1688 rtnetlink_notifier_unregister(&netdev_linux_poll_notifier);
1692 const struct netdev_class netdev_linux_class = {
1699 netdev_linux_create_system,
1700 netdev_linux_destroy,
1701 NULL, /* reconfigure */
1706 netdev_linux_enumerate,
1709 netdev_linux_recv_wait,
1713 netdev_linux_send_wait,
1715 netdev_linux_set_etheraddr,
1716 netdev_linux_get_etheraddr,
1717 netdev_linux_get_mtu,
1718 netdev_linux_get_ifindex,
1719 netdev_linux_get_carrier,
1720 netdev_linux_get_stats,
1721 netdev_linux_set_stats,
1723 netdev_linux_get_features,
1724 netdev_linux_set_advertisements,
1725 netdev_linux_get_vlan_vid,
1726 netdev_linux_set_policing,
1728 netdev_linux_get_in4,
1729 netdev_linux_set_in4,
1730 netdev_linux_get_in6,
1731 netdev_linux_add_router,
1732 netdev_linux_get_next_hop,
1733 netdev_linux_arp_lookup,
1735 netdev_linux_update_flags,
1737 netdev_linux_poll_add,
1738 netdev_linux_poll_remove,
1741 const struct netdev_class netdev_tap_class = {
1748 netdev_linux_create_tap,
1749 netdev_linux_destroy,
1750 NULL, /* reconfigure */
1755 NULL, /* enumerate */
1758 netdev_linux_recv_wait,
1762 netdev_linux_send_wait,
1764 netdev_linux_set_etheraddr,
1765 netdev_linux_get_etheraddr,
1766 netdev_linux_get_mtu,
1767 netdev_linux_get_ifindex,
1768 netdev_linux_get_carrier,
1769 netdev_linux_get_stats,
1770 NULL, /* set_stats */
1772 netdev_linux_get_features,
1773 netdev_linux_set_advertisements,
1774 netdev_linux_get_vlan_vid,
1775 netdev_linux_set_policing,
1777 netdev_linux_get_in4,
1778 netdev_linux_set_in4,
1779 netdev_linux_get_in6,
1780 netdev_linux_add_router,
1781 netdev_linux_get_next_hop,
1782 netdev_linux_arp_lookup,
1784 netdev_linux_update_flags,
1786 netdev_linux_poll_add,
1787 netdev_linux_poll_remove,
1790 const struct netdev_class netdev_patch_class = {
1797 netdev_linux_create_patch,
1798 netdev_linux_destroy,
1799 NULL, /* reconfigure */
1804 NULL, /* enumerate */
1807 netdev_linux_recv_wait,
1811 netdev_linux_send_wait,
1813 netdev_linux_set_etheraddr,
1814 netdev_linux_get_etheraddr,
1815 netdev_linux_get_mtu,
1816 netdev_linux_get_ifindex,
1817 netdev_linux_get_carrier,
1818 netdev_linux_get_stats,
1819 NULL, /* set_stats */
1821 netdev_linux_get_features,
1822 netdev_linux_set_advertisements,
1823 netdev_linux_get_vlan_vid,
1824 netdev_linux_set_policing,
1826 netdev_linux_get_in4,
1827 netdev_linux_set_in4,
1828 netdev_linux_get_in6,
1829 netdev_linux_add_router,
1830 netdev_linux_get_next_hop,
1831 netdev_linux_arp_lookup,
1833 netdev_linux_update_flags,
1835 netdev_linux_poll_add,
1836 netdev_linux_poll_remove,
1841 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
1843 /* Policy for RTNLGRP_LINK messages.
1845 * There are *many* more fields in these messages, but currently we only
1846 * care about these fields. */
1847 static const struct nl_policy rtnlgrp_link_policy[] = {
1848 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
1849 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
1850 .min_len = sizeof(struct rtnl_link_stats) },
1854 static struct nl_sock *rtnl_sock;
1855 struct ofpbuf request;
1856 struct ofpbuf *reply;
1857 struct ifinfomsg *ifi;
1858 const struct rtnl_link_stats *rtnl_stats;
1859 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
1863 error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock);
1865 VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",
1871 ofpbuf_init(&request, 0);
1872 nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *ifi,
1873 RTM_GETLINK, NLM_F_REQUEST);
1874 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
1875 ifi->ifi_family = PF_UNSPEC;
1876 ifi->ifi_index = ifindex;
1877 error = nl_sock_transact(rtnl_sock, &request, &reply);
1878 ofpbuf_uninit(&request);
1883 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
1884 rtnlgrp_link_policy,
1885 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
1886 ofpbuf_delete(reply);
1890 if (!attrs[IFLA_STATS]) {
1891 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
1892 ofpbuf_delete(reply);
1896 rtnl_stats = nl_attr_get(attrs[IFLA_STATS]);
1897 stats->rx_packets = rtnl_stats->rx_packets;
1898 stats->tx_packets = rtnl_stats->tx_packets;
1899 stats->rx_bytes = rtnl_stats->rx_bytes;
1900 stats->tx_bytes = rtnl_stats->tx_bytes;
1901 stats->rx_errors = rtnl_stats->rx_errors;
1902 stats->tx_errors = rtnl_stats->tx_errors;
1903 stats->rx_dropped = rtnl_stats->rx_dropped;
1904 stats->tx_dropped = rtnl_stats->tx_dropped;
1905 stats->multicast = rtnl_stats->multicast;
1906 stats->collisions = rtnl_stats->collisions;
1907 stats->rx_length_errors = rtnl_stats->rx_length_errors;
1908 stats->rx_over_errors = rtnl_stats->rx_over_errors;
1909 stats->rx_crc_errors = rtnl_stats->rx_crc_errors;
1910 stats->rx_frame_errors = rtnl_stats->rx_frame_errors;
1911 stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors;
1912 stats->rx_missed_errors = rtnl_stats->rx_missed_errors;
1913 stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors;
1914 stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors;
1915 stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors;
1916 stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors;
1917 stats->tx_window_errors = rtnl_stats->tx_window_errors;
1919 ofpbuf_delete(reply);
1925 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
1927 static const char fn[] = "/proc/net/dev";
1932 stream = fopen(fn, "r");
1934 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1939 while (fgets(line, sizeof line, stream)) {
1942 #define X64 "%"SCNu64
1945 X64 X64 X64 X64 X64 X64 X64 "%*u"
1946 X64 X64 X64 X64 X64 X64 X64 "%*u",
1952 &stats->rx_fifo_errors,
1953 &stats->rx_frame_errors,
1959 &stats->tx_fifo_errors,
1961 &stats->tx_carrier_errors) != 15) {
1962 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
1963 } else if (!strcmp(devname, netdev_name)) {
1964 stats->rx_length_errors = UINT64_MAX;
1965 stats->rx_over_errors = UINT64_MAX;
1966 stats->rx_crc_errors = UINT64_MAX;
1967 stats->rx_missed_errors = UINT64_MAX;
1968 stats->tx_aborted_errors = UINT64_MAX;
1969 stats->tx_heartbeat_errors = UINT64_MAX;
1970 stats->tx_window_errors = UINT64_MAX;
1976 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
1982 get_flags(const struct netdev *netdev, int *flags)
1987 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS,
1989 *flags = ifr.ifr_flags;
1994 set_flags(struct netdev *netdev, int flags)
1998 ifr.ifr_flags = flags;
1999 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
2004 do_get_ifindex(const char *netdev_name)
2008 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2009 COVERAGE_INC(netdev_get_ifindex);
2010 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
2011 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
2012 netdev_name, strerror(errno));
2015 return ifr.ifr_ifindex;
2019 get_ifindex(const struct netdev *netdev_, int *ifindexp)
2021 struct netdev_dev_linux *netdev_dev =
2022 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2024 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
2025 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
2029 netdev_dev->cache_valid |= VALID_IFINDEX;
2030 netdev_dev->ifindex = ifindex;
2032 *ifindexp = netdev_dev->ifindex;
2037 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
2042 memset(&ifr, 0, sizeof ifr);
2043 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2044 COVERAGE_INC(netdev_get_hwaddr);
2045 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
2046 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
2047 netdev_name, strerror(errno));
2050 hwaddr_family = ifr.ifr_hwaddr.sa_family;
2051 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
2052 VLOG_WARN("%s device has unknown hardware address family %d",
2053 netdev_name, hwaddr_family);
2055 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
2060 set_etheraddr(const char *netdev_name, int hwaddr_family,
2061 const uint8_t mac[ETH_ADDR_LEN])
2065 memset(&ifr, 0, sizeof ifr);
2066 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2067 ifr.ifr_hwaddr.sa_family = hwaddr_family;
2068 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
2069 COVERAGE_INC(netdev_set_hwaddr);
2070 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
2071 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
2072 netdev_name, strerror(errno));
2079 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
2080 int cmd, const char *cmd_name)
2084 memset(&ifr, 0, sizeof ifr);
2085 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
2086 ifr.ifr_data = (caddr_t) ecmd;
2089 COVERAGE_INC(netdev_ethtool);
2090 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
2093 if (errno != EOPNOTSUPP) {
2094 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
2095 "failed: %s", cmd_name, name, strerror(errno));
2097 /* The device doesn't support this operation. That's pretty
2098 * common, so there's no point in logging anything. */
2105 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
2106 const char *cmd_name)
2108 strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
2109 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
2110 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
2118 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
2119 int cmd, const char *cmd_name)
2124 ifr.ifr_addr.sa_family = AF_INET;
2125 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
2127 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
2128 *ip = sin->sin_addr;