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);
159 static int get_rtnl_sock(struct nl_sock **);
162 is_netdev_linux_class(const struct netdev_class *netdev_class)
164 return netdev_class->init == netdev_linux_init;
167 static struct netdev_dev_linux *
168 netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
170 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
171 assert(is_netdev_linux_class(netdev_class));
173 return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
176 static struct netdev_linux *
177 netdev_linux_cast(const struct netdev *netdev)
179 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
180 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
181 assert(is_netdev_linux_class(netdev_class));
183 return CONTAINER_OF(netdev, struct netdev_linux, netdev);
187 netdev_linux_init(void)
189 static int status = -1;
191 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
192 status = af_inet_sock >= 0 ? 0 : errno;
194 VLOG_ERR("failed to create inet socket: %s", strerror(status));
201 netdev_linux_run(void)
203 rtnetlink_notifier_run();
207 netdev_linux_wait(void)
209 rtnetlink_notifier_wait();
213 netdev_linux_cache_cb(const struct rtnetlink_change *change,
214 void *aux OVS_UNUSED)
216 struct netdev_dev_linux *dev;
218 struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
220 const struct netdev_class *netdev_class =
221 netdev_dev_get_class(base_dev);
223 if (is_netdev_linux_class(netdev_class)) {
224 dev = netdev_dev_linux_cast(base_dev);
225 dev->cache_valid = 0;
229 struct shash device_shash;
230 struct shash_node *node;
232 shash_init(&device_shash);
233 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
234 SHASH_FOR_EACH (node, &device_shash) {
236 dev->cache_valid = 0;
238 shash_destroy(&device_shash);
243 if_up(const char *name)
247 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
248 ifr.ifr_flags = IFF_UP;
250 if (ioctl(af_inet_sock, SIOCSIFFLAGS, &ifr) == -1) {
251 VLOG_DBG_RL(&rl, "%s: failed to bring device up: %s",
252 name, strerror(errno));
259 /* A veth may be created using the 'command' "+<name>,<peer>". A veth may
260 * be destroyed by using the 'command' "-<name>", where <name> can be
261 * either side of the device.
264 modify_veth(const char *format, ...)
270 veth_file = fopen("/sys/class/net/veth_pairs", "w");
272 VLOG_WARN_RL(&rl, "could not open veth device. Are you running a "
273 "supported XenServer with the kernel module loaded?");
276 setvbuf(veth_file, NULL, _IONBF, 0);
278 va_start(args, format);
279 retval = vfprintf(veth_file, format, args);
284 VLOG_WARN_RL(&rl, "could not destroy patch: %s", strerror(errno));
292 create_patch(const char *name, const char *peer)
295 struct netdev_dev *peer_nd;
298 /* Only create the veth if the peer didn't already do it. */
299 peer_nd = netdev_dev_from_name(peer);
301 if (!strcmp("patch", netdev_dev_get_type(peer_nd))) {
302 struct netdev_dev_linux *ndl = netdev_dev_linux_cast(peer_nd);
303 if (!strcmp(name, ndl->state.patch.peer)) {
306 VLOG_WARN_RL(&rl, "peer '%s' already paired with '%s'",
307 peer, ndl->state.patch.peer);
311 VLOG_WARN_RL(&rl, "peer '%s' exists and is not a patch", peer);
316 retval = modify_veth("+%s,%s", name, peer);
321 retval = if_up(name);
326 retval = if_up(peer);
335 setup_patch(const char *name, const struct shash *args, char **peer_)
339 peer = shash_find_data(args, "peer");
341 VLOG_WARN("patch type requires valid 'peer' argument");
345 if (shash_count(args) > 1) {
346 VLOG_WARN("patch type takes only a 'peer' argument");
350 if (strlen(peer) >= IFNAMSIZ) {
351 VLOG_WARN_RL(&rl, "patch 'peer' arg too long");
355 *peer_ = xstrdup(peer);
356 return create_patch(name, peer);
359 /* Creates the netdev device of 'type' with 'name'. */
361 netdev_linux_create_system(const char *name, const char *type OVS_UNUSED,
362 const struct shash *args, struct netdev_dev **netdev_devp)
364 struct netdev_dev_linux *netdev_dev;
367 if (!shash_is_empty(args)) {
368 VLOG_WARN("%s: arguments for system devices should be empty", name);
371 if (!cache_notifier_refcount) {
372 error = rtnetlink_notifier_register(&netdev_linux_cache_notifier,
373 netdev_linux_cache_cb, NULL);
378 cache_notifier_refcount++;
380 netdev_dev = xzalloc(sizeof *netdev_dev);
381 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class);
383 *netdev_devp = &netdev_dev->netdev_dev;
387 /* For most types of netdevs we open the device for each call of
388 * netdev_open(). However, this is not the case with tap devices,
389 * since it is only possible to open the device once. In this
390 * situation we share a single file descriptor, and consequently
391 * buffers, across all readers. Therefore once data is read it will
392 * be unavailable to other reads for tap devices. */
394 netdev_linux_create_tap(const char *name, const char *type OVS_UNUSED,
395 const struct shash *args, struct netdev_dev **netdev_devp)
397 struct netdev_dev_linux *netdev_dev;
398 struct tap_state *state;
399 static const char tap_dev[] = "/dev/net/tun";
403 if (!shash_is_empty(args)) {
404 VLOG_WARN("%s: arguments for TAP devices should be empty", name);
407 netdev_dev = xzalloc(sizeof *netdev_dev);
408 state = &netdev_dev->state.tap;
410 /* Open tap device. */
411 state->fd = open(tap_dev, O_RDWR);
414 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
418 /* Create tap device. */
419 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
420 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
421 if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
422 VLOG_WARN("%s: creating tap device failed: %s", name,
428 /* Make non-blocking. */
429 error = set_nonblocking(state->fd);
434 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
435 *netdev_devp = &netdev_dev->netdev_dev;
444 netdev_linux_create_patch(const char *name, const char *type OVS_UNUSED,
445 const struct shash *args, struct netdev_dev **netdev_devp)
447 struct netdev_dev_linux *netdev_dev;
451 error = setup_patch(name, args, &peer);
457 netdev_dev = xzalloc(sizeof *netdev_dev);
458 netdev_dev->state.patch.peer = peer;
459 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_patch_class);
460 *netdev_devp = &netdev_dev->netdev_dev;
466 destroy_tap(struct netdev_dev_linux *netdev_dev)
468 struct tap_state *state = &netdev_dev->state.tap;
470 if (state->fd >= 0) {
476 destroy_patch(struct netdev_dev_linux *netdev_dev)
478 const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
479 struct patch_state *state = &netdev_dev->state.patch;
481 /* Only destroy veth if 'peer' doesn't exist as an existing netdev. */
482 if (!netdev_dev_from_name(state->peer)) {
483 modify_veth("-%s", name);
488 /* Destroys the netdev device 'netdev_dev_'. */
490 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
492 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
493 const char *type = netdev_dev_get_type(netdev_dev_);
495 if (!strcmp(type, "system")) {
496 cache_notifier_refcount--;
498 if (!cache_notifier_refcount) {
499 rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
501 } else if (!strcmp(type, "tap")) {
502 destroy_tap(netdev_dev);
503 } else if (!strcmp(type, "patch")) {
504 destroy_patch(netdev_dev);
511 netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype,
512 struct netdev **netdevp)
514 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
515 struct netdev_linux *netdev;
516 enum netdev_flags flags;
519 /* Allocate network device. */
520 netdev = xzalloc(sizeof *netdev);
522 netdev_init(&netdev->netdev, netdev_dev_);
524 error = netdev_get_flags(&netdev->netdev, &flags);
525 if (error == ENODEV) {
529 if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap")) {
530 netdev->fd = netdev_dev->state.tap.fd;
531 } else if (ethertype != NETDEV_ETH_TYPE_NONE) {
532 struct sockaddr_ll sll;
536 /* Create file descriptor. */
537 protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
538 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
540 netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
541 if (netdev->fd < 0) {
546 /* Set non-blocking mode. */
547 error = set_nonblocking(netdev->fd);
552 /* Get ethernet device index. */
553 error = get_ifindex(&netdev->netdev, &ifindex);
558 /* Bind to specific ethernet device. */
559 memset(&sll, 0, sizeof sll);
560 sll.sll_family = AF_PACKET;
561 sll.sll_ifindex = ifindex;
563 (struct sockaddr *) &sll, sizeof sll) < 0) {
565 VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_),
570 /* Between the socket() and bind() calls above, the socket receives all
571 * packets of the requested type on all system interfaces. We do not
572 * want to receive that data, but there is no way to avoid it. So we
573 * must now drain out the receive queue. */
574 error = drain_rcvbuf(netdev->fd);
580 *netdevp = &netdev->netdev;
584 netdev_uninit(&netdev->netdev, true);
588 /* Closes and destroys 'netdev'. */
590 netdev_linux_close(struct netdev *netdev_)
592 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
594 if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
600 /* Initializes 'svec' with a list of the names of all known network devices. */
602 netdev_linux_enumerate(struct svec *svec)
604 struct if_nameindex *names;
606 names = if_nameindex();
610 for (i = 0; names[i].if_name != NULL; i++) {
611 svec_add(svec, names[i].if_name);
613 if_freenameindex(names);
616 VLOG_WARN("could not obtain list of network device names: %s",
623 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
625 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
627 if (netdev->fd < 0) {
628 /* Device was opened with NETDEV_ETH_TYPE_NONE. */
633 ssize_t retval = read(netdev->fd, data, size);
636 } else if (errno != EINTR) {
637 if (errno != EAGAIN) {
638 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
639 strerror(errno), netdev_get_name(netdev_));
646 /* Registers with the poll loop to wake up from the next call to poll_block()
647 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
649 netdev_linux_recv_wait(struct netdev *netdev_)
651 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
652 if (netdev->fd >= 0) {
653 poll_fd_wait(netdev->fd, POLLIN);
657 /* Discards all packets waiting to be received from 'netdev'. */
659 netdev_linux_drain(struct netdev *netdev_)
661 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
662 if (netdev->fd < 0) {
664 } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
666 int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
667 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
671 drain_fd(netdev->fd, ifr.ifr_qlen);
674 return drain_rcvbuf(netdev->fd);
678 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
679 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
680 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
681 * the packet is too big or too small to transmit on the device.
683 * The caller retains ownership of 'buffer' in all cases.
685 * The kernel maintains a packet transmission queue, so the caller is not
686 * expected to do additional queuing of packets. */
688 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
690 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
692 /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
694 if (netdev->fd < 0) {
699 ssize_t retval = write(netdev->fd, data, size);
701 /* The Linux AF_PACKET implementation never blocks waiting for room
702 * for packets, instead returning ENOBUFS. Translate this into
703 * EAGAIN for the caller. */
704 if (errno == ENOBUFS) {
706 } else if (errno == EINTR) {
708 } else if (errno != EAGAIN) {
709 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
710 netdev_get_name(netdev_), strerror(errno));
713 } else if (retval != size) {
714 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
715 "%zu) on %s", retval, size, netdev_get_name(netdev_));
723 /* Registers with the poll loop to wake up from the next call to poll_block()
724 * when the packet transmission queue has sufficient room to transmit a packet
725 * with netdev_send().
727 * The kernel maintains a packet transmission queue, so the client is not
728 * expected to do additional queuing of packets. Thus, this function is
729 * unlikely to ever be used. It is included for completeness. */
731 netdev_linux_send_wait(struct netdev *netdev_)
733 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
734 if (netdev->fd < 0) {
736 } else if (strcmp(netdev_get_type(netdev_), "tap")) {
737 poll_fd_wait(netdev->fd, POLLOUT);
739 /* TAP device always accepts packets.*/
740 poll_immediate_wake();
744 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
745 * otherwise a positive errno value. */
747 netdev_linux_set_etheraddr(struct netdev *netdev_,
748 const uint8_t mac[ETH_ADDR_LEN])
750 struct netdev_dev_linux *netdev_dev =
751 netdev_dev_linux_cast(netdev_get_dev(netdev_));
754 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)
755 || !eth_addr_equals(netdev_dev->etheraddr, mac)) {
756 error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
758 netdev_dev->cache_valid |= VALID_ETHERADDR;
759 memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
767 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
768 * free the returned buffer. */
770 netdev_linux_get_etheraddr(const struct netdev *netdev_,
771 uint8_t mac[ETH_ADDR_LEN])
773 struct netdev_dev_linux *netdev_dev =
774 netdev_dev_linux_cast(netdev_get_dev(netdev_));
775 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
776 int error = get_etheraddr(netdev_get_name(netdev_),
777 netdev_dev->etheraddr);
781 netdev_dev->cache_valid |= VALID_ETHERADDR;
783 memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
787 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
788 * in bytes, not including the hardware header; thus, this is typically 1500
789 * bytes for Ethernet devices. */
791 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
793 struct netdev_dev_linux *netdev_dev =
794 netdev_dev_linux_cast(netdev_get_dev(netdev_));
795 if (!(netdev_dev->cache_valid & VALID_MTU)) {
799 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
800 SIOCGIFMTU, "SIOCGIFMTU");
804 netdev_dev->mtu = ifr.ifr_mtu;
805 netdev_dev->cache_valid |= VALID_MTU;
807 *mtup = netdev_dev->mtu;
811 /* Returns the ifindex of 'netdev', if successful, as a positive number.
812 * On failure, returns a negative errno value. */
814 netdev_linux_get_ifindex(const struct netdev *netdev)
818 error = get_ifindex(netdev, &ifindex);
819 return error ? -error : ifindex;
823 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
825 struct netdev_dev_linux *netdev_dev =
826 netdev_dev_linux_cast(netdev_get_dev(netdev_));
831 if (!(netdev_dev->cache_valid & VALID_CARRIER)) {
835 fn = xasprintf("/sys/class/net/%s/carrier",
836 netdev_get_name(netdev_));
837 fd = open(fn, O_RDONLY);
840 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error));
844 retval = read(fd, line, sizeof line);
847 if (error == EINVAL) {
848 /* This is the normal return value when we try to check carrier
849 * if the network device is not up. */
851 VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error));
854 } else if (retval == 0) {
856 VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn);
860 if (line[0] != '0' && line[0] != '1') {
862 VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)",
866 netdev_dev->carrier = line[0] != '0';
867 netdev_dev->cache_valid |= VALID_CARRIER;
869 *carrier = netdev_dev->carrier;
880 /* Check whether we can we use RTM_GETLINK to get network device statistics.
881 * In pre-2.6.19 kernels, this was only available if wireless extensions were
884 check_for_working_netlink_stats(void)
886 /* Decide on the netdev_get_stats() implementation to use. Netlink is
887 * preferable, so if that works, we'll use it. */
888 int ifindex = do_get_ifindex("lo");
890 VLOG_WARN("failed to get ifindex for lo, "
891 "obtaining netdev stats from proc");
894 struct netdev_stats stats;
895 int error = get_stats_via_netlink(ifindex, &stats);
897 VLOG_DBG("obtaining netdev stats via rtnetlink");
900 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
901 "via proc (you are probably running a pre-2.6.19 "
902 "kernel)", strerror(error));
908 /* Brings the 'is_internal' and 'is_tap' members of 'netdev_dev' up-to-date. */
910 netdev_linux_update_is_pseudo(struct netdev_dev_linux *netdev_dev)
912 if (!(netdev_dev->cache_valid & VALID_IS_PSEUDO)) {
913 const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
914 const char *type = netdev_dev_get_type(&netdev_dev->netdev_dev);
916 netdev_dev->is_tap = !strcmp(type, "tap");
917 netdev_dev->is_internal = false;
918 if (!netdev_dev->is_tap) {
919 struct ethtool_drvinfo drvinfo;
922 memset(&drvinfo, 0, sizeof drvinfo);
923 error = netdev_linux_do_ethtool(name,
924 (struct ethtool_cmd *)&drvinfo,
928 if (!error && !strcmp(drvinfo.driver, "openvswitch")) {
929 netdev_dev->is_internal = true;
933 netdev_dev->cache_valid |= VALID_IS_PSEUDO;
937 /* Retrieves current device stats for 'netdev'.
939 * XXX All of the members of struct netdev_stats are 64 bits wide, but on
940 * 32-bit architectures the Linux network stats are only 32 bits. */
942 netdev_linux_get_stats(const struct netdev *netdev_,
943 struct netdev_stats *stats)
945 struct netdev_dev_linux *netdev_dev =
946 netdev_dev_linux_cast(netdev_get_dev(netdev_));
947 static int use_netlink_stats = -1;
949 struct netdev_stats raw_stats;
950 struct netdev_stats *collect_stats = stats;
952 COVERAGE_INC(netdev_get_stats);
954 netdev_linux_update_is_pseudo(netdev_dev);
955 if (netdev_dev->is_internal) {
956 collect_stats = &raw_stats;
959 if (use_netlink_stats < 0) {
960 use_netlink_stats = check_for_working_netlink_stats();
962 if (use_netlink_stats) {
965 error = get_ifindex(netdev_, &ifindex);
967 error = get_stats_via_netlink(ifindex, collect_stats);
970 error = get_stats_via_proc(netdev_get_name(netdev_), collect_stats);
973 /* If this port is an internal port then the transmit and receive stats
974 * will appear to be swapped relative to the other ports since we are the
975 * one sending the data, not a remote computer. For consistency, we swap
977 if (!error && (netdev_dev->is_internal || netdev_dev->is_tap)) {
978 stats->rx_packets = raw_stats.tx_packets;
979 stats->tx_packets = raw_stats.rx_packets;
980 stats->rx_bytes = raw_stats.tx_bytes;
981 stats->tx_bytes = raw_stats.rx_bytes;
982 stats->rx_errors = raw_stats.tx_errors;
983 stats->tx_errors = raw_stats.rx_errors;
984 stats->rx_dropped = raw_stats.tx_dropped;
985 stats->tx_dropped = raw_stats.rx_dropped;
986 stats->multicast = raw_stats.multicast;
987 stats->collisions = raw_stats.collisions;
988 stats->rx_length_errors = 0;
989 stats->rx_over_errors = 0;
990 stats->rx_crc_errors = 0;
991 stats->rx_frame_errors = 0;
992 stats->rx_fifo_errors = 0;
993 stats->rx_missed_errors = 0;
994 stats->tx_aborted_errors = 0;
995 stats->tx_carrier_errors = 0;
996 stats->tx_fifo_errors = 0;
997 stats->tx_heartbeat_errors = 0;
998 stats->tx_window_errors = 0;
1005 netdev_linux_set_stats(struct netdev *netdev,
1006 const struct netdev_stats *stats)
1008 struct netdev_dev_linux *netdev_dev =
1009 netdev_dev_linux_cast(netdev_get_dev(netdev));
1010 struct internal_dev_stats dp_dev_stats;
1013 /* We must reject this call if 'netdev' is not an Open vSwitch internal
1014 * port, because the ioctl that we are about to execute is in the "device
1015 * private ioctls" range, which means that executing it on a device that
1016 * is not the type we expect could do any random thing.
1018 * (Amusingly, these ioctl numbers are commented "THESE IOCTLS ARE
1019 * _DEPRECATED_ AND WILL DISAPPEAR IN 2.5.X" in linux/sockios.h. I guess
1020 * DaveM is a little behind on that.) */
1021 netdev_linux_update_is_pseudo(netdev_dev);
1022 if (!netdev_dev->is_internal) {
1026 /* This actually only sets the *offset* that the dp_dev applies, but in our
1027 * usage for fake bond devices the dp_dev never has any traffic of it own
1028 * so it has the same effect. */
1029 dp_dev_stats.rx_packets = stats->rx_packets;
1030 dp_dev_stats.rx_bytes = stats->rx_bytes;
1031 dp_dev_stats.tx_packets = stats->tx_packets;
1032 dp_dev_stats.tx_bytes = stats->tx_bytes;
1033 ifr.ifr_data = (void *) &dp_dev_stats;
1034 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr,
1035 INTERNAL_DEV_SET_STATS,
1036 "INTERNAL_DEV_SET_STATS");
1039 /* Stores the features supported by 'netdev' into each of '*current',
1040 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
1041 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
1042 * successful, otherwise a positive errno value. */
1044 netdev_linux_get_features(struct netdev *netdev,
1045 uint32_t *current, uint32_t *advertised,
1046 uint32_t *supported, uint32_t *peer)
1048 struct ethtool_cmd ecmd;
1051 memset(&ecmd, 0, sizeof ecmd);
1052 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1053 ETHTOOL_GSET, "ETHTOOL_GSET");
1058 /* Supported features. */
1060 if (ecmd.supported & SUPPORTED_10baseT_Half) {
1061 *supported |= OFPPF_10MB_HD;
1063 if (ecmd.supported & SUPPORTED_10baseT_Full) {
1064 *supported |= OFPPF_10MB_FD;
1066 if (ecmd.supported & SUPPORTED_100baseT_Half) {
1067 *supported |= OFPPF_100MB_HD;
1069 if (ecmd.supported & SUPPORTED_100baseT_Full) {
1070 *supported |= OFPPF_100MB_FD;
1072 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
1073 *supported |= OFPPF_1GB_HD;
1075 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
1076 *supported |= OFPPF_1GB_FD;
1078 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
1079 *supported |= OFPPF_10GB_FD;
1081 if (ecmd.supported & SUPPORTED_TP) {
1082 *supported |= OFPPF_COPPER;
1084 if (ecmd.supported & SUPPORTED_FIBRE) {
1085 *supported |= OFPPF_FIBER;
1087 if (ecmd.supported & SUPPORTED_Autoneg) {
1088 *supported |= OFPPF_AUTONEG;
1090 if (ecmd.supported & SUPPORTED_Pause) {
1091 *supported |= OFPPF_PAUSE;
1093 if (ecmd.supported & SUPPORTED_Asym_Pause) {
1094 *supported |= OFPPF_PAUSE_ASYM;
1097 /* Advertised features. */
1099 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
1100 *advertised |= OFPPF_10MB_HD;
1102 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
1103 *advertised |= OFPPF_10MB_FD;
1105 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
1106 *advertised |= OFPPF_100MB_HD;
1108 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
1109 *advertised |= OFPPF_100MB_FD;
1111 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
1112 *advertised |= OFPPF_1GB_HD;
1114 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
1115 *advertised |= OFPPF_1GB_FD;
1117 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
1118 *advertised |= OFPPF_10GB_FD;
1120 if (ecmd.advertising & ADVERTISED_TP) {
1121 *advertised |= OFPPF_COPPER;
1123 if (ecmd.advertising & ADVERTISED_FIBRE) {
1124 *advertised |= OFPPF_FIBER;
1126 if (ecmd.advertising & ADVERTISED_Autoneg) {
1127 *advertised |= OFPPF_AUTONEG;
1129 if (ecmd.advertising & ADVERTISED_Pause) {
1130 *advertised |= OFPPF_PAUSE;
1132 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
1133 *advertised |= OFPPF_PAUSE_ASYM;
1136 /* Current settings. */
1137 if (ecmd.speed == SPEED_10) {
1138 *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD;
1139 } else if (ecmd.speed == SPEED_100) {
1140 *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD;
1141 } else if (ecmd.speed == SPEED_1000) {
1142 *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD;
1143 } else if (ecmd.speed == SPEED_10000) {
1144 *current = OFPPF_10GB_FD;
1149 if (ecmd.port == PORT_TP) {
1150 *current |= OFPPF_COPPER;
1151 } else if (ecmd.port == PORT_FIBRE) {
1152 *current |= OFPPF_FIBER;
1156 *current |= OFPPF_AUTONEG;
1159 /* Peer advertisements. */
1160 *peer = 0; /* XXX */
1165 /* Set the features advertised by 'netdev' to 'advertise'. */
1167 netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
1169 struct ethtool_cmd ecmd;
1172 memset(&ecmd, 0, sizeof ecmd);
1173 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1174 ETHTOOL_GSET, "ETHTOOL_GSET");
1179 ecmd.advertising = 0;
1180 if (advertise & OFPPF_10MB_HD) {
1181 ecmd.advertising |= ADVERTISED_10baseT_Half;
1183 if (advertise & OFPPF_10MB_FD) {
1184 ecmd.advertising |= ADVERTISED_10baseT_Full;
1186 if (advertise & OFPPF_100MB_HD) {
1187 ecmd.advertising |= ADVERTISED_100baseT_Half;
1189 if (advertise & OFPPF_100MB_FD) {
1190 ecmd.advertising |= ADVERTISED_100baseT_Full;
1192 if (advertise & OFPPF_1GB_HD) {
1193 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1195 if (advertise & OFPPF_1GB_FD) {
1196 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1198 if (advertise & OFPPF_10GB_FD) {
1199 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1201 if (advertise & OFPPF_COPPER) {
1202 ecmd.advertising |= ADVERTISED_TP;
1204 if (advertise & OFPPF_FIBER) {
1205 ecmd.advertising |= ADVERTISED_FIBRE;
1207 if (advertise & OFPPF_AUTONEG) {
1208 ecmd.advertising |= ADVERTISED_Autoneg;
1210 if (advertise & OFPPF_PAUSE) {
1211 ecmd.advertising |= ADVERTISED_Pause;
1213 if (advertise & OFPPF_PAUSE_ASYM) {
1214 ecmd.advertising |= ADVERTISED_Asym_Pause;
1216 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1217 ETHTOOL_SSET, "ETHTOOL_SSET");
1220 /* If 'netdev_name' is the name of a VLAN network device (e.g. one created with
1221 * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device
1222 * and returns 0. Otherwise returns a errno value (specifically ENOENT if
1223 * 'netdev_name' is the name of a network device that is not a VLAN device) and
1224 * sets '*vlan_vid' to -1. */
1226 netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1228 const char *netdev_name = netdev_get_name(netdev);
1229 struct ds line = DS_EMPTY_INITIALIZER;
1230 FILE *stream = NULL;
1234 COVERAGE_INC(netdev_get_vlan_vid);
1235 fn = xasprintf("/proc/net/vlan/%s", netdev_name);
1236 stream = fopen(fn, "r");
1242 if (ds_get_line(&line, stream)) {
1243 if (ferror(stream)) {
1245 VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno));
1248 VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn);
1253 if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) {
1255 VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"",
1256 fn, ds_cstr(&line));
1274 #define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress"
1275 #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"
1276 /* We redirect stderr to /dev/null because we often want to remove all
1277 * traffic control configuration on a port so its in a known state. If
1278 * this done when there is no such configuration, tc complains, so we just
1281 #define POLICE_DEL_CMD "/sbin/tc qdisc del dev %s handle ffff: ingress 2>/dev/null"
1283 /* Remove ingress policing from 'netdev'. Returns 0 if successful, otherwise a
1284 * positive errno value. */
1286 netdev_linux_remove_policing(struct netdev *netdev)
1288 const char *netdev_name = netdev_get_name(netdev);
1291 /* xxx This should be more careful about only adding if it
1292 * xxx actually exists, as opposed to always deleting it. */
1293 snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1294 if (system(command) == -1) {
1295 VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1301 /* Attempts to set input rate limiting (policing) policy. */
1303 netdev_linux_set_policing(struct netdev *netdev,
1304 uint32_t kbits_rate, uint32_t kbits_burst)
1306 const char *netdev_name = netdev_get_name(netdev);
1309 COVERAGE_INC(netdev_set_policing);
1311 netdev_linux_remove_policing(netdev);
1314 /* Default to 1000 kilobits if not specified. */
1318 snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name);
1319 if (system(command) != 0) {
1320 VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name);
1324 snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name,
1325 kbits_rate, kbits_burst);
1326 if (system(command) != 0) {
1327 VLOG_WARN_RL(&rl, "%s: problem configuring policing",
1337 netdev_linux_get_in4(const struct netdev *netdev_,
1338 struct in_addr *address, struct in_addr *netmask)
1340 struct netdev_dev_linux *netdev_dev =
1341 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1343 if (!(netdev_dev->cache_valid & VALID_IN4)) {
1346 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
1347 SIOCGIFADDR, "SIOCGIFADDR");
1352 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
1353 SIOCGIFNETMASK, "SIOCGIFNETMASK");
1358 netdev_dev->cache_valid |= VALID_IN4;
1360 *address = netdev_dev->address;
1361 *netmask = netdev_dev->netmask;
1362 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
1366 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
1367 struct in_addr netmask)
1369 struct netdev_dev_linux *netdev_dev =
1370 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1373 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
1375 netdev_dev->cache_valid |= VALID_IN4;
1376 netdev_dev->address = address;
1377 netdev_dev->netmask = netmask;
1378 if (address.s_addr != INADDR_ANY) {
1379 error = do_set_addr(netdev_, SIOCSIFNETMASK,
1380 "SIOCSIFNETMASK", netmask);
1387 parse_if_inet6_line(const char *line,
1388 struct in6_addr *in6, char ifname[16 + 1])
1390 uint8_t *s6 = in6->s6_addr;
1391 #define X8 "%2"SCNx8
1393 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
1394 "%*x %*x %*x %*x %16s\n",
1395 &s6[0], &s6[1], &s6[2], &s6[3],
1396 &s6[4], &s6[5], &s6[6], &s6[7],
1397 &s6[8], &s6[9], &s6[10], &s6[11],
1398 &s6[12], &s6[13], &s6[14], &s6[15],
1402 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
1403 * 'in6' is non-null) and returns true. Otherwise, returns false. */
1405 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
1407 struct netdev_dev_linux *netdev_dev =
1408 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1409 if (!(netdev_dev->cache_valid & VALID_IN6)) {
1413 netdev_dev->in6 = in6addr_any;
1415 file = fopen("/proc/net/if_inet6", "r");
1417 const char *name = netdev_get_name(netdev_);
1418 while (fgets(line, sizeof line, file)) {
1419 struct in6_addr in6;
1420 char ifname[16 + 1];
1421 if (parse_if_inet6_line(line, &in6, ifname)
1422 && !strcmp(name, ifname))
1424 netdev_dev->in6 = in6;
1430 netdev_dev->cache_valid |= VALID_IN6;
1432 *in6 = netdev_dev->in6;
1437 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
1439 struct sockaddr_in sin;
1440 memset(&sin, 0, sizeof sin);
1441 sin.sin_family = AF_INET;
1442 sin.sin_addr = addr;
1445 memset(sa, 0, sizeof *sa);
1446 memcpy(sa, &sin, sizeof sin);
1450 do_set_addr(struct netdev *netdev,
1451 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
1454 strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
1455 make_in4_sockaddr(&ifr.ifr_addr, addr);
1457 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
1461 /* Adds 'router' as a default IP gateway. */
1463 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
1465 struct in_addr any = { INADDR_ANY };
1469 memset(&rt, 0, sizeof rt);
1470 make_in4_sockaddr(&rt.rt_dst, any);
1471 make_in4_sockaddr(&rt.rt_gateway, router);
1472 make_in4_sockaddr(&rt.rt_genmask, any);
1473 rt.rt_flags = RTF_UP | RTF_GATEWAY;
1474 COVERAGE_INC(netdev_add_router);
1475 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
1477 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
1483 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
1486 static const char fn[] = "/proc/net/route";
1491 *netdev_name = NULL;
1492 stream = fopen(fn, "r");
1493 if (stream == NULL) {
1494 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1499 while (fgets(line, sizeof line, stream)) {
1502 uint32_t dest, gateway, mask;
1503 int refcnt, metric, mtu;
1504 unsigned int flags, use, window, irtt;
1507 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
1509 iface, &dest, &gateway, &flags, &refcnt,
1510 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
1512 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
1516 if (!(flags & RTF_UP)) {
1517 /* Skip routes that aren't up. */
1521 /* The output of 'dest', 'mask', and 'gateway' were given in
1522 * network byte order, so we don't need need any endian
1523 * conversions here. */
1524 if ((dest & mask) == (host->s_addr & mask)) {
1526 /* The host is directly reachable. */
1527 next_hop->s_addr = 0;
1529 /* To reach the host, we must go through a gateway. */
1530 next_hop->s_addr = gateway;
1532 *netdev_name = xstrdup(iface);
1543 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1544 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1545 * returns 0. Otherwise, it returns a positive errno value; in particular,
1546 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
1548 netdev_linux_arp_lookup(const struct netdev *netdev,
1549 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
1552 struct sockaddr_in sin;
1555 memset(&r, 0, sizeof r);
1556 sin.sin_family = AF_INET;
1557 sin.sin_addr.s_addr = ip;
1559 memcpy(&r.arp_pa, &sin, sizeof sin);
1560 r.arp_ha.sa_family = ARPHRD_ETHER;
1562 strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
1563 COVERAGE_INC(netdev_arp_lookup);
1564 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
1566 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
1567 } else if (retval != ENXIO) {
1568 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
1569 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
1575 nd_to_iff_flags(enum netdev_flags nd)
1578 if (nd & NETDEV_UP) {
1581 if (nd & NETDEV_PROMISC) {
1588 iff_to_nd_flags(int iff)
1590 enum netdev_flags nd = 0;
1594 if (iff & IFF_PROMISC) {
1595 nd |= NETDEV_PROMISC;
1601 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
1602 enum netdev_flags on, enum netdev_flags *old_flagsp)
1604 int old_flags, new_flags;
1607 error = get_flags(netdev, &old_flags);
1609 *old_flagsp = iff_to_nd_flags(old_flags);
1610 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
1611 if (new_flags != old_flags) {
1612 error = set_flags(netdev, new_flags);
1619 poll_notify(struct list *list)
1621 struct netdev_linux_notifier *notifier;
1622 LIST_FOR_EACH (notifier, struct netdev_linux_notifier, node, list) {
1623 struct netdev_notifier *n = ¬ifier->notifier;
1629 netdev_linux_poll_cb(const struct rtnetlink_change *change,
1630 void *aux OVS_UNUSED)
1633 struct list *list = shash_find_data(&netdev_linux_notifiers,
1639 struct shash_node *node;
1640 SHASH_FOR_EACH (node, &netdev_linux_notifiers) {
1641 poll_notify(node->data);
1647 netdev_linux_poll_add(struct netdev *netdev,
1648 void (*cb)(struct netdev_notifier *), void *aux,
1649 struct netdev_notifier **notifierp)
1651 const char *netdev_name = netdev_get_name(netdev);
1652 struct netdev_linux_notifier *notifier;
1655 if (shash_is_empty(&netdev_linux_notifiers)) {
1656 int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier,
1657 netdev_linux_poll_cb, NULL);
1663 list = shash_find_data(&netdev_linux_notifiers, netdev_name);
1665 list = xmalloc(sizeof *list);
1667 shash_add(&netdev_linux_notifiers, netdev_name, list);
1670 notifier = xmalloc(sizeof *notifier);
1671 netdev_notifier_init(¬ifier->notifier, netdev, cb, aux);
1672 list_push_back(list, ¬ifier->node);
1673 *notifierp = ¬ifier->notifier;
1678 netdev_linux_poll_remove(struct netdev_notifier *notifier_)
1680 struct netdev_linux_notifier *notifier =
1681 CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier);
1684 /* Remove 'notifier' from its list. */
1685 list = list_remove(¬ifier->node);
1686 if (list_is_empty(list)) {
1687 /* The list is now empty. Remove it from the hash and free it. */
1688 const char *netdev_name = netdev_get_name(notifier->notifier.netdev);
1689 shash_delete(&netdev_linux_notifiers,
1690 shash_find(&netdev_linux_notifiers, netdev_name));
1695 /* If that was the last notifier, unregister. */
1696 if (shash_is_empty(&netdev_linux_notifiers)) {
1697 rtnetlink_notifier_unregister(&netdev_linux_poll_notifier);
1701 const struct netdev_class netdev_linux_class = {
1708 netdev_linux_create_system,
1709 netdev_linux_destroy,
1710 NULL, /* reconfigure */
1715 netdev_linux_enumerate,
1718 netdev_linux_recv_wait,
1722 netdev_linux_send_wait,
1724 netdev_linux_set_etheraddr,
1725 netdev_linux_get_etheraddr,
1726 netdev_linux_get_mtu,
1727 netdev_linux_get_ifindex,
1728 netdev_linux_get_carrier,
1729 netdev_linux_get_stats,
1730 netdev_linux_set_stats,
1732 netdev_linux_get_features,
1733 netdev_linux_set_advertisements,
1734 netdev_linux_get_vlan_vid,
1735 netdev_linux_set_policing,
1737 netdev_linux_get_in4,
1738 netdev_linux_set_in4,
1739 netdev_linux_get_in6,
1740 netdev_linux_add_router,
1741 netdev_linux_get_next_hop,
1742 netdev_linux_arp_lookup,
1744 netdev_linux_update_flags,
1746 netdev_linux_poll_add,
1747 netdev_linux_poll_remove,
1750 const struct netdev_class netdev_tap_class = {
1757 netdev_linux_create_tap,
1758 netdev_linux_destroy,
1759 NULL, /* reconfigure */
1764 NULL, /* enumerate */
1767 netdev_linux_recv_wait,
1771 netdev_linux_send_wait,
1773 netdev_linux_set_etheraddr,
1774 netdev_linux_get_etheraddr,
1775 netdev_linux_get_mtu,
1776 netdev_linux_get_ifindex,
1777 netdev_linux_get_carrier,
1778 netdev_linux_get_stats,
1779 NULL, /* set_stats */
1781 netdev_linux_get_features,
1782 netdev_linux_set_advertisements,
1783 netdev_linux_get_vlan_vid,
1784 netdev_linux_set_policing,
1786 netdev_linux_get_in4,
1787 netdev_linux_set_in4,
1788 netdev_linux_get_in6,
1789 netdev_linux_add_router,
1790 netdev_linux_get_next_hop,
1791 netdev_linux_arp_lookup,
1793 netdev_linux_update_flags,
1795 netdev_linux_poll_add,
1796 netdev_linux_poll_remove,
1799 const struct netdev_class netdev_patch_class = {
1806 netdev_linux_create_patch,
1807 netdev_linux_destroy,
1808 NULL, /* reconfigure */
1813 NULL, /* enumerate */
1816 netdev_linux_recv_wait,
1820 netdev_linux_send_wait,
1822 netdev_linux_set_etheraddr,
1823 netdev_linux_get_etheraddr,
1824 netdev_linux_get_mtu,
1825 netdev_linux_get_ifindex,
1826 netdev_linux_get_carrier,
1827 netdev_linux_get_stats,
1828 NULL, /* set_stats */
1830 netdev_linux_get_features,
1831 netdev_linux_set_advertisements,
1832 netdev_linux_get_vlan_vid,
1833 netdev_linux_set_policing,
1835 netdev_linux_get_in4,
1836 netdev_linux_set_in4,
1837 netdev_linux_get_in6,
1838 netdev_linux_add_router,
1839 netdev_linux_get_next_hop,
1840 netdev_linux_arp_lookup,
1842 netdev_linux_update_flags,
1844 netdev_linux_poll_add,
1845 netdev_linux_poll_remove,
1850 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
1852 /* Policy for RTNLGRP_LINK messages.
1854 * There are *many* more fields in these messages, but currently we only
1855 * care about these fields. */
1856 static const struct nl_policy rtnlgrp_link_policy[] = {
1857 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
1858 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
1859 .min_len = sizeof(struct rtnl_link_stats) },
1862 struct nl_sock *rtnl_sock;
1863 struct ofpbuf request;
1864 struct ofpbuf *reply;
1865 struct ifinfomsg *ifi;
1866 const struct rtnl_link_stats *rtnl_stats;
1867 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
1870 error = get_rtnl_sock(&rtnl_sock);
1875 ofpbuf_init(&request, 0);
1876 nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *ifi,
1877 RTM_GETLINK, NLM_F_REQUEST);
1878 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
1879 ifi->ifi_family = PF_UNSPEC;
1880 ifi->ifi_index = ifindex;
1881 error = nl_sock_transact(rtnl_sock, &request, &reply);
1882 ofpbuf_uninit(&request);
1887 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
1888 rtnlgrp_link_policy,
1889 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
1890 ofpbuf_delete(reply);
1894 if (!attrs[IFLA_STATS]) {
1895 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
1896 ofpbuf_delete(reply);
1900 rtnl_stats = nl_attr_get(attrs[IFLA_STATS]);
1901 stats->rx_packets = rtnl_stats->rx_packets;
1902 stats->tx_packets = rtnl_stats->tx_packets;
1903 stats->rx_bytes = rtnl_stats->rx_bytes;
1904 stats->tx_bytes = rtnl_stats->tx_bytes;
1905 stats->rx_errors = rtnl_stats->rx_errors;
1906 stats->tx_errors = rtnl_stats->tx_errors;
1907 stats->rx_dropped = rtnl_stats->rx_dropped;
1908 stats->tx_dropped = rtnl_stats->tx_dropped;
1909 stats->multicast = rtnl_stats->multicast;
1910 stats->collisions = rtnl_stats->collisions;
1911 stats->rx_length_errors = rtnl_stats->rx_length_errors;
1912 stats->rx_over_errors = rtnl_stats->rx_over_errors;
1913 stats->rx_crc_errors = rtnl_stats->rx_crc_errors;
1914 stats->rx_frame_errors = rtnl_stats->rx_frame_errors;
1915 stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors;
1916 stats->rx_missed_errors = rtnl_stats->rx_missed_errors;
1917 stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors;
1918 stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors;
1919 stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors;
1920 stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors;
1921 stats->tx_window_errors = rtnl_stats->tx_window_errors;
1923 ofpbuf_delete(reply);
1929 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
1931 static const char fn[] = "/proc/net/dev";
1936 stream = fopen(fn, "r");
1938 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1943 while (fgets(line, sizeof line, stream)) {
1946 #define X64 "%"SCNu64
1949 X64 X64 X64 X64 X64 X64 X64 "%*u"
1950 X64 X64 X64 X64 X64 X64 X64 "%*u",
1956 &stats->rx_fifo_errors,
1957 &stats->rx_frame_errors,
1963 &stats->tx_fifo_errors,
1965 &stats->tx_carrier_errors) != 15) {
1966 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
1967 } else if (!strcmp(devname, netdev_name)) {
1968 stats->rx_length_errors = UINT64_MAX;
1969 stats->rx_over_errors = UINT64_MAX;
1970 stats->rx_crc_errors = UINT64_MAX;
1971 stats->rx_missed_errors = UINT64_MAX;
1972 stats->tx_aborted_errors = UINT64_MAX;
1973 stats->tx_heartbeat_errors = UINT64_MAX;
1974 stats->tx_window_errors = UINT64_MAX;
1980 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
1986 get_flags(const struct netdev *netdev, int *flags)
1991 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS,
1993 *flags = ifr.ifr_flags;
1998 set_flags(struct netdev *netdev, int flags)
2002 ifr.ifr_flags = flags;
2003 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
2008 do_get_ifindex(const char *netdev_name)
2012 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2013 COVERAGE_INC(netdev_get_ifindex);
2014 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
2015 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
2016 netdev_name, strerror(errno));
2019 return ifr.ifr_ifindex;
2023 get_ifindex(const struct netdev *netdev_, int *ifindexp)
2025 struct netdev_dev_linux *netdev_dev =
2026 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2028 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
2029 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
2033 netdev_dev->cache_valid |= VALID_IFINDEX;
2034 netdev_dev->ifindex = ifindex;
2036 *ifindexp = netdev_dev->ifindex;
2041 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
2046 memset(&ifr, 0, sizeof ifr);
2047 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2048 COVERAGE_INC(netdev_get_hwaddr);
2049 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
2050 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
2051 netdev_name, strerror(errno));
2054 hwaddr_family = ifr.ifr_hwaddr.sa_family;
2055 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
2056 VLOG_WARN("%s device has unknown hardware address family %d",
2057 netdev_name, hwaddr_family);
2059 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
2064 set_etheraddr(const char *netdev_name, int hwaddr_family,
2065 const uint8_t mac[ETH_ADDR_LEN])
2069 memset(&ifr, 0, sizeof ifr);
2070 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2071 ifr.ifr_hwaddr.sa_family = hwaddr_family;
2072 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
2073 COVERAGE_INC(netdev_set_hwaddr);
2074 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
2075 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
2076 netdev_name, strerror(errno));
2083 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
2084 int cmd, const char *cmd_name)
2088 memset(&ifr, 0, sizeof ifr);
2089 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
2090 ifr.ifr_data = (caddr_t) ecmd;
2093 COVERAGE_INC(netdev_ethtool);
2094 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
2097 if (errno != EOPNOTSUPP) {
2098 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
2099 "failed: %s", cmd_name, name, strerror(errno));
2101 /* The device doesn't support this operation. That's pretty
2102 * common, so there's no point in logging anything. */
2109 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
2110 const char *cmd_name)
2112 strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
2113 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
2114 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
2122 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
2123 int cmd, const char *cmd_name)
2128 ifr.ifr_addr.sa_family = AF_INET;
2129 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
2131 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
2132 *ip = sin->sin_addr;
2137 /* Obtains a Netlink routing socket that is not subscribed to any multicast
2138 * groups. Returns 0 if successful, otherwise a positive errno value. Stores
2139 * the socket in '*rtnl_sockp' if successful, otherwise a null pointer. */
2141 get_rtnl_sock(struct nl_sock **rtnl_sockp)
2143 static struct nl_sock *sock;
2147 error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &sock);
2149 VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",