2 * Copyright (c) 2009, 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
21 #include <arpa/inet.h>
23 #include <linux/if_tun.h>
25 #include <linux/types.h>
26 #include <linux/ethtool.h>
27 #include <linux/pkt_sched.h>
28 #include <linux/rtnetlink.h>
29 #include <linux/sockios.h>
30 #include <linux/version.h>
31 #include <sys/types.h>
32 #include <sys/ioctl.h>
33 #include <sys/socket.h>
34 #include <netpacket/packet.h>
35 #include <net/ethernet.h>
37 #include <linux/if_tunnel.h>
38 #include <net/if_arp.h>
39 #include <net/if_packet.h>
40 #include <net/route.h>
41 #include <netinet/in.h>
48 #include "dynamic-string.h"
49 #include "fatal-signal.h"
50 #include "netdev-provider.h"
51 #include "netdev-vport.h"
54 #include "openflow/openflow.h"
55 #include "openvswitch/gre.h"
57 #include "poll-loop.h"
58 #include "rtnetlink.h"
59 #include "socket-util.h"
63 #define THIS_MODULE VLM_netdev_linux
66 /* These were introduced in Linux 2.6.14, so they might be missing if we have
68 #ifndef ADVERTISED_Pause
69 #define ADVERTISED_Pause (1 << 13)
71 #ifndef ADVERTISED_Asym_Pause
72 #define ADVERTISED_Asym_Pause (1 << 14)
75 static struct rtnetlink_notifier netdev_linux_cache_notifier;
76 static int cache_notifier_refcount;
79 VALID_IFINDEX = 1 << 0,
80 VALID_ETHERADDR = 1 << 1,
84 VALID_CARRIER = 1 << 5,
85 VALID_IS_PSEUDO = 1 << 6, /* Represents is_internal and is_tap. */
86 VALID_POLICING = 1 << 7,
87 VALID_HAVE_VPORT_STATS = 1 << 8
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? */
111 uint32_t kbits_rate; /* Policing data. */
112 uint32_t kbits_burst;
113 bool have_vport_stats;
116 struct tap_state tap;
120 struct netdev_linux {
121 struct netdev netdev;
125 /* An AF_INET socket (used for ioctl operations). */
126 static int af_inet_sock = -1;
128 struct netdev_linux_notifier {
129 struct netdev_notifier notifier;
133 static struct shash netdev_linux_notifiers =
134 SHASH_INITIALIZER(&netdev_linux_notifiers);
135 static struct rtnetlink_notifier netdev_linux_poll_notifier;
137 /* This is set pretty low because we probably won't learn anything from the
138 * additional log messages. */
139 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
141 static int netdev_linux_init(void);
143 static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
144 int cmd, const char *cmd_name);
145 static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
146 const char *cmd_name);
147 static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
148 int cmd, const char *cmd_name);
149 static int get_flags(const struct netdev *, int *flagsp);
150 static int set_flags(struct netdev *, int flags);
151 static int do_get_ifindex(const char *netdev_name);
152 static int get_ifindex(const struct netdev *, int *ifindexp);
153 static int do_set_addr(struct netdev *netdev,
154 int ioctl_nr, const char *ioctl_name,
155 struct in_addr addr);
156 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
157 static int set_etheraddr(const char *netdev_name, int hwaddr_family,
158 const uint8_t[ETH_ADDR_LEN]);
159 static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
160 static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
161 static int get_rtnl_sock(struct nl_sock **);
164 is_netdev_linux_class(const struct netdev_class *netdev_class)
166 return netdev_class->init == netdev_linux_init;
169 static struct netdev_dev_linux *
170 netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
172 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
173 assert(is_netdev_linux_class(netdev_class));
175 return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
178 static struct netdev_linux *
179 netdev_linux_cast(const struct netdev *netdev)
181 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
182 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
183 assert(is_netdev_linux_class(netdev_class));
185 return CONTAINER_OF(netdev, struct netdev_linux, netdev);
189 netdev_linux_init(void)
191 static int status = -1;
193 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
194 status = af_inet_sock >= 0 ? 0 : errno;
196 VLOG_ERR("failed to create inet socket: %s", strerror(status));
203 netdev_linux_run(void)
205 rtnetlink_notifier_run();
209 netdev_linux_wait(void)
211 rtnetlink_notifier_wait();
215 netdev_linux_cache_cb(const struct rtnetlink_change *change,
216 void *aux OVS_UNUSED)
218 struct netdev_dev_linux *dev;
220 struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
222 const struct netdev_class *netdev_class =
223 netdev_dev_get_class(base_dev);
225 if (is_netdev_linux_class(netdev_class)) {
226 dev = netdev_dev_linux_cast(base_dev);
227 dev->cache_valid = 0;
231 struct shash device_shash;
232 struct shash_node *node;
234 shash_init(&device_shash);
235 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
236 SHASH_FOR_EACH (node, &device_shash) {
238 dev->cache_valid = 0;
240 shash_destroy(&device_shash);
244 /* Creates the netdev device of 'type' with 'name'. */
246 netdev_linux_create_system(const char *name, const char *type OVS_UNUSED,
247 const struct shash *args, struct netdev_dev **netdev_devp)
249 struct netdev_dev_linux *netdev_dev;
252 if (!shash_is_empty(args)) {
253 VLOG_WARN("%s: arguments for system devices should be empty", name);
256 if (!cache_notifier_refcount) {
257 error = rtnetlink_notifier_register(&netdev_linux_cache_notifier,
258 netdev_linux_cache_cb, NULL);
263 cache_notifier_refcount++;
265 netdev_dev = xzalloc(sizeof *netdev_dev);
266 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class);
268 *netdev_devp = &netdev_dev->netdev_dev;
272 /* For most types of netdevs we open the device for each call of
273 * netdev_open(). However, this is not the case with tap devices,
274 * since it is only possible to open the device once. In this
275 * situation we share a single file descriptor, and consequently
276 * buffers, across all readers. Therefore once data is read it will
277 * be unavailable to other reads for tap devices. */
279 netdev_linux_create_tap(const char *name, const char *type OVS_UNUSED,
280 const struct shash *args, struct netdev_dev **netdev_devp)
282 struct netdev_dev_linux *netdev_dev;
283 struct tap_state *state;
284 static const char tap_dev[] = "/dev/net/tun";
288 if (!shash_is_empty(args)) {
289 VLOG_WARN("%s: arguments for TAP devices should be empty", name);
292 netdev_dev = xzalloc(sizeof *netdev_dev);
293 state = &netdev_dev->state.tap;
295 /* Open tap device. */
296 state->fd = open(tap_dev, O_RDWR);
299 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
303 /* Create tap device. */
304 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
305 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
306 if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
307 VLOG_WARN("%s: creating tap device failed: %s", name,
313 /* Make non-blocking. */
314 error = set_nonblocking(state->fd);
319 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
320 *netdev_devp = &netdev_dev->netdev_dev;
329 destroy_tap(struct netdev_dev_linux *netdev_dev)
331 struct tap_state *state = &netdev_dev->state.tap;
333 if (state->fd >= 0) {
338 /* Destroys the netdev device 'netdev_dev_'. */
340 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
342 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
343 const char *type = netdev_dev_get_type(netdev_dev_);
345 if (!strcmp(type, "system")) {
346 cache_notifier_refcount--;
348 if (!cache_notifier_refcount) {
349 rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
351 } else if (!strcmp(type, "tap")) {
352 destroy_tap(netdev_dev);
359 netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype,
360 struct netdev **netdevp)
362 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
363 struct netdev_linux *netdev;
364 enum netdev_flags flags;
367 /* Allocate network device. */
368 netdev = xzalloc(sizeof *netdev);
370 netdev_init(&netdev->netdev, netdev_dev_);
372 error = netdev_get_flags(&netdev->netdev, &flags);
373 if (error == ENODEV) {
377 if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap") &&
378 !netdev_dev->state.tap.opened) {
380 /* We assume that the first user of the tap device is the primary user
381 * and give them the tap FD. Subsequent users probably just expect
382 * this to be a system device so open it normally to avoid send/receive
383 * directions appearing to be reversed. */
384 netdev->fd = netdev_dev->state.tap.fd;
385 netdev_dev->state.tap.opened = true;
386 } else if (ethertype != NETDEV_ETH_TYPE_NONE) {
387 struct sockaddr_ll sll;
391 /* Create file descriptor. */
392 protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
393 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
395 netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
396 if (netdev->fd < 0) {
401 /* Set non-blocking mode. */
402 error = set_nonblocking(netdev->fd);
407 /* Get ethernet device index. */
408 error = get_ifindex(&netdev->netdev, &ifindex);
413 /* Bind to specific ethernet device. */
414 memset(&sll, 0, sizeof sll);
415 sll.sll_family = AF_PACKET;
416 sll.sll_ifindex = ifindex;
418 (struct sockaddr *) &sll, sizeof sll) < 0) {
420 VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_),
425 /* Between the socket() and bind() calls above, the socket receives all
426 * packets of the requested type on all system interfaces. We do not
427 * want to receive that data, but there is no way to avoid it. So we
428 * must now drain out the receive queue. */
429 error = drain_rcvbuf(netdev->fd);
435 *netdevp = &netdev->netdev;
439 netdev_uninit(&netdev->netdev, true);
443 /* Closes and destroys 'netdev'. */
445 netdev_linux_close(struct netdev *netdev_)
447 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
449 if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
455 /* Initializes 'svec' with a list of the names of all known network devices. */
457 netdev_linux_enumerate(struct svec *svec)
459 struct if_nameindex *names;
461 names = if_nameindex();
465 for (i = 0; names[i].if_name != NULL; i++) {
466 svec_add(svec, names[i].if_name);
468 if_freenameindex(names);
471 VLOG_WARN("could not obtain list of network device names: %s",
478 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
480 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
482 if (netdev->fd < 0) {
483 /* Device was opened with NETDEV_ETH_TYPE_NONE. */
488 ssize_t retval = read(netdev->fd, data, size);
491 } else if (errno != EINTR) {
492 if (errno != EAGAIN) {
493 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
494 strerror(errno), netdev_get_name(netdev_));
501 /* Registers with the poll loop to wake up from the next call to poll_block()
502 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
504 netdev_linux_recv_wait(struct netdev *netdev_)
506 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
507 if (netdev->fd >= 0) {
508 poll_fd_wait(netdev->fd, POLLIN);
512 /* Discards all packets waiting to be received from 'netdev'. */
514 netdev_linux_drain(struct netdev *netdev_)
516 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
517 if (netdev->fd < 0) {
519 } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
521 int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
522 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
526 drain_fd(netdev->fd, ifr.ifr_qlen);
529 return drain_rcvbuf(netdev->fd);
533 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
534 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
535 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
536 * the packet is too big or too small to transmit on the device.
538 * The caller retains ownership of 'buffer' in all cases.
540 * The kernel maintains a packet transmission queue, so the caller is not
541 * expected to do additional queuing of packets. */
543 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
545 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
547 /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
549 if (netdev->fd < 0) {
554 ssize_t retval = write(netdev->fd, data, size);
556 /* The Linux AF_PACKET implementation never blocks waiting for room
557 * for packets, instead returning ENOBUFS. Translate this into
558 * EAGAIN for the caller. */
559 if (errno == ENOBUFS) {
561 } else if (errno == EINTR) {
563 } else if (errno != EAGAIN) {
564 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
565 netdev_get_name(netdev_), strerror(errno));
568 } else if (retval != size) {
569 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
570 "%zu) on %s", retval, size, netdev_get_name(netdev_));
578 /* Registers with the poll loop to wake up from the next call to poll_block()
579 * when the packet transmission queue has sufficient room to transmit a packet
580 * with netdev_send().
582 * The kernel maintains a packet transmission queue, so the client is not
583 * expected to do additional queuing of packets. Thus, this function is
584 * unlikely to ever be used. It is included for completeness. */
586 netdev_linux_send_wait(struct netdev *netdev_)
588 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
589 if (netdev->fd < 0) {
591 } else if (strcmp(netdev_get_type(netdev_), "tap")) {
592 poll_fd_wait(netdev->fd, POLLOUT);
594 /* TAP device always accepts packets.*/
595 poll_immediate_wake();
599 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
600 * otherwise a positive errno value. */
602 netdev_linux_set_etheraddr(struct netdev *netdev_,
603 const uint8_t mac[ETH_ADDR_LEN])
605 struct netdev_dev_linux *netdev_dev =
606 netdev_dev_linux_cast(netdev_get_dev(netdev_));
609 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)
610 || !eth_addr_equals(netdev_dev->etheraddr, mac)) {
611 error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
613 netdev_dev->cache_valid |= VALID_ETHERADDR;
614 memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
622 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
623 * free the returned buffer. */
625 netdev_linux_get_etheraddr(const struct netdev *netdev_,
626 uint8_t mac[ETH_ADDR_LEN])
628 struct netdev_dev_linux *netdev_dev =
629 netdev_dev_linux_cast(netdev_get_dev(netdev_));
630 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
631 int error = get_etheraddr(netdev_get_name(netdev_),
632 netdev_dev->etheraddr);
636 netdev_dev->cache_valid |= VALID_ETHERADDR;
638 memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
642 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
643 * in bytes, not including the hardware header; thus, this is typically 1500
644 * bytes for Ethernet devices. */
646 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
648 struct netdev_dev_linux *netdev_dev =
649 netdev_dev_linux_cast(netdev_get_dev(netdev_));
650 if (!(netdev_dev->cache_valid & VALID_MTU)) {
654 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
655 SIOCGIFMTU, "SIOCGIFMTU");
659 netdev_dev->mtu = ifr.ifr_mtu;
660 netdev_dev->cache_valid |= VALID_MTU;
662 *mtup = netdev_dev->mtu;
666 /* Returns the ifindex of 'netdev', if successful, as a positive number.
667 * On failure, returns a negative errno value. */
669 netdev_linux_get_ifindex(const struct netdev *netdev)
673 error = get_ifindex(netdev, &ifindex);
674 return error ? -error : ifindex;
678 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
680 struct netdev_dev_linux *netdev_dev =
681 netdev_dev_linux_cast(netdev_get_dev(netdev_));
686 if (!(netdev_dev->cache_valid & VALID_CARRIER)) {
690 fn = xasprintf("/sys/class/net/%s/carrier",
691 netdev_get_name(netdev_));
692 fd = open(fn, O_RDONLY);
695 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error));
699 retval = read(fd, line, sizeof line);
702 if (error == EINVAL) {
703 /* This is the normal return value when we try to check carrier
704 * if the network device is not up. */
706 VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error));
709 } else if (retval == 0) {
711 VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn);
715 if (line[0] != '0' && line[0] != '1') {
717 VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)",
721 netdev_dev->carrier = line[0] != '0';
722 netdev_dev->cache_valid |= VALID_CARRIER;
724 *carrier = netdev_dev->carrier;
735 /* Check whether we can we use RTM_GETLINK to get network device statistics.
736 * In pre-2.6.19 kernels, this was only available if wireless extensions were
739 check_for_working_netlink_stats(void)
741 /* Decide on the netdev_get_stats() implementation to use. Netlink is
742 * preferable, so if that works, we'll use it. */
743 int ifindex = do_get_ifindex("lo");
745 VLOG_WARN("failed to get ifindex for lo, "
746 "obtaining netdev stats from proc");
749 struct netdev_stats stats;
750 int error = get_stats_via_netlink(ifindex, &stats);
752 VLOG_DBG("obtaining netdev stats via rtnetlink");
755 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
756 "via proc (you are probably running a pre-2.6.19 "
757 "kernel)", strerror(error));
763 /* Brings the 'is_internal' and 'is_tap' members of 'netdev_dev' up-to-date. */
765 netdev_linux_update_is_pseudo(struct netdev_dev_linux *netdev_dev)
767 if (!(netdev_dev->cache_valid & VALID_IS_PSEUDO)) {
768 const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
769 const char *type = netdev_dev_get_type(&netdev_dev->netdev_dev);
771 netdev_dev->is_tap = !strcmp(type, "tap");
772 netdev_dev->is_internal = false;
773 if (!netdev_dev->is_tap) {
774 struct ethtool_drvinfo drvinfo;
777 memset(&drvinfo, 0, sizeof drvinfo);
778 error = netdev_linux_do_ethtool(name,
779 (struct ethtool_cmd *)&drvinfo,
783 if (!error && !strcmp(drvinfo.driver, "openvswitch")) {
784 netdev_dev->is_internal = true;
788 netdev_dev->cache_valid |= VALID_IS_PSEUDO;
793 swap_uint64(uint64_t *a, uint64_t *b)
800 /* Retrieves current device stats for 'netdev'. */
802 netdev_linux_get_stats(const struct netdev *netdev_,
803 struct netdev_stats *stats)
805 struct netdev_dev_linux *netdev_dev =
806 netdev_dev_linux_cast(netdev_get_dev(netdev_));
807 static int use_netlink_stats = -1;
810 COVERAGE_INC(netdev_get_stats);
812 if (netdev_dev->have_vport_stats ||
813 !(netdev_dev->cache_valid & VALID_HAVE_VPORT_STATS)) {
815 error = netdev_vport_get_stats(netdev_, stats);
816 netdev_dev->have_vport_stats = !error;
817 netdev_dev->cache_valid |= VALID_HAVE_VPORT_STATS;
820 if (!netdev_dev->have_vport_stats) {
821 if (use_netlink_stats < 0) {
822 use_netlink_stats = check_for_working_netlink_stats();
824 if (use_netlink_stats) {
827 error = get_ifindex(netdev_, &ifindex);
829 error = get_stats_via_netlink(ifindex, stats);
832 error = get_stats_via_proc(netdev_get_name(netdev_), stats);
836 /* If this port is an internal port then the transmit and receive stats
837 * will appear to be swapped relative to the other ports since we are the
838 * one sending the data, not a remote computer. For consistency, we swap
839 * them back here. This does not apply if we are getting stats from the
840 * vport layer because it always tracks stats from the perspective of the
842 netdev_linux_update_is_pseudo(netdev_dev);
843 if (!error && !netdev_dev->have_vport_stats &&
844 (netdev_dev->is_internal || netdev_dev->is_tap)) {
845 swap_uint64(&stats->rx_packets, &stats->tx_packets);
846 swap_uint64(&stats->rx_bytes, &stats->tx_bytes);
847 swap_uint64(&stats->rx_errors, &stats->tx_errors);
848 swap_uint64(&stats->rx_dropped, &stats->tx_dropped);
849 stats->rx_length_errors = 0;
850 stats->rx_over_errors = 0;
851 stats->rx_crc_errors = 0;
852 stats->rx_frame_errors = 0;
853 stats->rx_fifo_errors = 0;
854 stats->rx_missed_errors = 0;
855 stats->tx_aborted_errors = 0;
856 stats->tx_carrier_errors = 0;
857 stats->tx_fifo_errors = 0;
858 stats->tx_heartbeat_errors = 0;
859 stats->tx_window_errors = 0;
865 /* Stores the features supported by 'netdev' into each of '*current',
866 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
867 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
868 * successful, otherwise a positive errno value. */
870 netdev_linux_get_features(struct netdev *netdev,
871 uint32_t *current, uint32_t *advertised,
872 uint32_t *supported, uint32_t *peer)
874 struct ethtool_cmd ecmd;
877 memset(&ecmd, 0, sizeof ecmd);
878 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
879 ETHTOOL_GSET, "ETHTOOL_GSET");
884 /* Supported features. */
886 if (ecmd.supported & SUPPORTED_10baseT_Half) {
887 *supported |= OFPPF_10MB_HD;
889 if (ecmd.supported & SUPPORTED_10baseT_Full) {
890 *supported |= OFPPF_10MB_FD;
892 if (ecmd.supported & SUPPORTED_100baseT_Half) {
893 *supported |= OFPPF_100MB_HD;
895 if (ecmd.supported & SUPPORTED_100baseT_Full) {
896 *supported |= OFPPF_100MB_FD;
898 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
899 *supported |= OFPPF_1GB_HD;
901 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
902 *supported |= OFPPF_1GB_FD;
904 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
905 *supported |= OFPPF_10GB_FD;
907 if (ecmd.supported & SUPPORTED_TP) {
908 *supported |= OFPPF_COPPER;
910 if (ecmd.supported & SUPPORTED_FIBRE) {
911 *supported |= OFPPF_FIBER;
913 if (ecmd.supported & SUPPORTED_Autoneg) {
914 *supported |= OFPPF_AUTONEG;
916 if (ecmd.supported & SUPPORTED_Pause) {
917 *supported |= OFPPF_PAUSE;
919 if (ecmd.supported & SUPPORTED_Asym_Pause) {
920 *supported |= OFPPF_PAUSE_ASYM;
923 /* Advertised features. */
925 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
926 *advertised |= OFPPF_10MB_HD;
928 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
929 *advertised |= OFPPF_10MB_FD;
931 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
932 *advertised |= OFPPF_100MB_HD;
934 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
935 *advertised |= OFPPF_100MB_FD;
937 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
938 *advertised |= OFPPF_1GB_HD;
940 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
941 *advertised |= OFPPF_1GB_FD;
943 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
944 *advertised |= OFPPF_10GB_FD;
946 if (ecmd.advertising & ADVERTISED_TP) {
947 *advertised |= OFPPF_COPPER;
949 if (ecmd.advertising & ADVERTISED_FIBRE) {
950 *advertised |= OFPPF_FIBER;
952 if (ecmd.advertising & ADVERTISED_Autoneg) {
953 *advertised |= OFPPF_AUTONEG;
955 if (ecmd.advertising & ADVERTISED_Pause) {
956 *advertised |= OFPPF_PAUSE;
958 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
959 *advertised |= OFPPF_PAUSE_ASYM;
962 /* Current settings. */
963 if (ecmd.speed == SPEED_10) {
964 *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD;
965 } else if (ecmd.speed == SPEED_100) {
966 *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD;
967 } else if (ecmd.speed == SPEED_1000) {
968 *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD;
969 } else if (ecmd.speed == SPEED_10000) {
970 *current = OFPPF_10GB_FD;
975 if (ecmd.port == PORT_TP) {
976 *current |= OFPPF_COPPER;
977 } else if (ecmd.port == PORT_FIBRE) {
978 *current |= OFPPF_FIBER;
982 *current |= OFPPF_AUTONEG;
985 /* Peer advertisements. */
991 /* Set the features advertised by 'netdev' to 'advertise'. */
993 netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
995 struct ethtool_cmd ecmd;
998 memset(&ecmd, 0, sizeof ecmd);
999 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1000 ETHTOOL_GSET, "ETHTOOL_GSET");
1005 ecmd.advertising = 0;
1006 if (advertise & OFPPF_10MB_HD) {
1007 ecmd.advertising |= ADVERTISED_10baseT_Half;
1009 if (advertise & OFPPF_10MB_FD) {
1010 ecmd.advertising |= ADVERTISED_10baseT_Full;
1012 if (advertise & OFPPF_100MB_HD) {
1013 ecmd.advertising |= ADVERTISED_100baseT_Half;
1015 if (advertise & OFPPF_100MB_FD) {
1016 ecmd.advertising |= ADVERTISED_100baseT_Full;
1018 if (advertise & OFPPF_1GB_HD) {
1019 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1021 if (advertise & OFPPF_1GB_FD) {
1022 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1024 if (advertise & OFPPF_10GB_FD) {
1025 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1027 if (advertise & OFPPF_COPPER) {
1028 ecmd.advertising |= ADVERTISED_TP;
1030 if (advertise & OFPPF_FIBER) {
1031 ecmd.advertising |= ADVERTISED_FIBRE;
1033 if (advertise & OFPPF_AUTONEG) {
1034 ecmd.advertising |= ADVERTISED_Autoneg;
1036 if (advertise & OFPPF_PAUSE) {
1037 ecmd.advertising |= ADVERTISED_Pause;
1039 if (advertise & OFPPF_PAUSE_ASYM) {
1040 ecmd.advertising |= ADVERTISED_Asym_Pause;
1042 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1043 ETHTOOL_SSET, "ETHTOOL_SSET");
1046 /* If 'netdev_name' is the name of a VLAN network device (e.g. one created with
1047 * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device
1048 * and returns 0. Otherwise returns a errno value (specifically ENOENT if
1049 * 'netdev_name' is the name of a network device that is not a VLAN device) and
1050 * sets '*vlan_vid' to -1. */
1052 netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1054 const char *netdev_name = netdev_get_name(netdev);
1055 struct ds line = DS_EMPTY_INITIALIZER;
1056 FILE *stream = NULL;
1060 COVERAGE_INC(netdev_get_vlan_vid);
1061 fn = xasprintf("/proc/net/vlan/%s", netdev_name);
1062 stream = fopen(fn, "r");
1068 if (ds_get_line(&line, stream)) {
1069 if (ferror(stream)) {
1071 VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno));
1074 VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn);
1079 if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) {
1081 VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"",
1082 fn, ds_cstr(&line));
1100 #define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress"
1101 #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"
1103 /* Remove ingress policing from 'netdev'. Returns 0 if successful, otherwise a
1104 * positive errno value.
1106 * This function is equivalent to running
1107 * /sbin/tc qdisc del dev %s handle ffff: ingress
1108 * but it is much, much faster.
1111 netdev_linux_remove_policing(struct netdev *netdev)
1113 struct netdev_dev_linux *netdev_dev =
1114 netdev_dev_linux_cast(netdev_get_dev(netdev));
1115 const char *netdev_name = netdev_get_name(netdev);
1117 struct ofpbuf request;
1118 struct ofpbuf *reply;
1119 struct tcmsg *tcmsg;
1120 struct nl_sock *rtnl_sock;
1124 error = get_ifindex(netdev, &ifindex);
1129 error = get_rtnl_sock(&rtnl_sock);
1134 ofpbuf_init(&request, 0);
1135 nl_msg_put_nlmsghdr(&request, sizeof *tcmsg, RTM_DELQDISC, NLM_F_REQUEST);
1136 tcmsg = ofpbuf_put_zeros(&request, sizeof *tcmsg);
1137 tcmsg->tcm_family = AF_UNSPEC;
1138 tcmsg->tcm_ifindex = ifindex;
1139 tcmsg->tcm_handle = 0xffff0000;
1140 tcmsg->tcm_parent = TC_H_INGRESS;
1141 nl_msg_put_string(&request, TCA_KIND, "ingress");
1142 nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0);
1143 error = nl_sock_transact(rtnl_sock, &request, &reply);
1144 ofpbuf_uninit(&request);
1145 ofpbuf_delete(reply);
1146 if (error && error != ENOENT && error != EINVAL) {
1147 VLOG_WARN_RL(&rl, "%s: removing policing failed: %s",
1148 netdev_name, strerror(error));
1152 netdev_dev->kbits_rate = 0;
1153 netdev_dev->kbits_burst = 0;
1154 netdev_dev->cache_valid |= VALID_POLICING;
1158 /* Attempts to set input rate limiting (policing) policy. */
1160 netdev_linux_set_policing(struct netdev *netdev,
1161 uint32_t kbits_rate, uint32_t kbits_burst)
1163 struct netdev_dev_linux *netdev_dev =
1164 netdev_dev_linux_cast(netdev_get_dev(netdev));
1165 const char *netdev_name = netdev_get_name(netdev);
1168 COVERAGE_INC(netdev_set_policing);
1170 kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */
1171 : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */
1172 : kbits_burst); /* Stick with user-specified value. */
1174 if (netdev_dev->cache_valid & VALID_POLICING
1175 && netdev_dev->kbits_rate == kbits_rate
1176 && netdev_dev->kbits_burst == kbits_burst) {
1177 /* Assume that settings haven't changed since we last set them. */
1181 netdev_linux_remove_policing(netdev);
1183 snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name);
1184 if (system(command) != 0) {
1185 VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name);
1189 snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name,
1190 kbits_rate, kbits_burst);
1191 if (system(command) != 0) {
1192 VLOG_WARN_RL(&rl, "%s: problem configuring policing",
1197 netdev_dev->kbits_rate = kbits_rate;
1198 netdev_dev->kbits_burst = kbits_burst;
1199 netdev_dev->cache_valid |= VALID_POLICING;
1206 netdev_linux_get_in4(const struct netdev *netdev_,
1207 struct in_addr *address, struct in_addr *netmask)
1209 struct netdev_dev_linux *netdev_dev =
1210 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1212 if (!(netdev_dev->cache_valid & VALID_IN4)) {
1215 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
1216 SIOCGIFADDR, "SIOCGIFADDR");
1221 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
1222 SIOCGIFNETMASK, "SIOCGIFNETMASK");
1227 netdev_dev->cache_valid |= VALID_IN4;
1229 *address = netdev_dev->address;
1230 *netmask = netdev_dev->netmask;
1231 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
1235 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
1236 struct in_addr netmask)
1238 struct netdev_dev_linux *netdev_dev =
1239 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1242 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
1244 netdev_dev->cache_valid |= VALID_IN4;
1245 netdev_dev->address = address;
1246 netdev_dev->netmask = netmask;
1247 if (address.s_addr != INADDR_ANY) {
1248 error = do_set_addr(netdev_, SIOCSIFNETMASK,
1249 "SIOCSIFNETMASK", netmask);
1256 parse_if_inet6_line(const char *line,
1257 struct in6_addr *in6, char ifname[16 + 1])
1259 uint8_t *s6 = in6->s6_addr;
1260 #define X8 "%2"SCNx8
1262 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
1263 "%*x %*x %*x %*x %16s\n",
1264 &s6[0], &s6[1], &s6[2], &s6[3],
1265 &s6[4], &s6[5], &s6[6], &s6[7],
1266 &s6[8], &s6[9], &s6[10], &s6[11],
1267 &s6[12], &s6[13], &s6[14], &s6[15],
1271 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
1272 * 'in6' is non-null) and returns true. Otherwise, returns false. */
1274 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
1276 struct netdev_dev_linux *netdev_dev =
1277 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1278 if (!(netdev_dev->cache_valid & VALID_IN6)) {
1282 netdev_dev->in6 = in6addr_any;
1284 file = fopen("/proc/net/if_inet6", "r");
1286 const char *name = netdev_get_name(netdev_);
1287 while (fgets(line, sizeof line, file)) {
1288 struct in6_addr in6;
1289 char ifname[16 + 1];
1290 if (parse_if_inet6_line(line, &in6, ifname)
1291 && !strcmp(name, ifname))
1293 netdev_dev->in6 = in6;
1299 netdev_dev->cache_valid |= VALID_IN6;
1301 *in6 = netdev_dev->in6;
1306 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
1308 struct sockaddr_in sin;
1309 memset(&sin, 0, sizeof sin);
1310 sin.sin_family = AF_INET;
1311 sin.sin_addr = addr;
1314 memset(sa, 0, sizeof *sa);
1315 memcpy(sa, &sin, sizeof sin);
1319 do_set_addr(struct netdev *netdev,
1320 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
1323 strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
1324 make_in4_sockaddr(&ifr.ifr_addr, addr);
1326 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
1330 /* Adds 'router' as a default IP gateway. */
1332 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
1334 struct in_addr any = { INADDR_ANY };
1338 memset(&rt, 0, sizeof rt);
1339 make_in4_sockaddr(&rt.rt_dst, any);
1340 make_in4_sockaddr(&rt.rt_gateway, router);
1341 make_in4_sockaddr(&rt.rt_genmask, any);
1342 rt.rt_flags = RTF_UP | RTF_GATEWAY;
1343 COVERAGE_INC(netdev_add_router);
1344 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
1346 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
1352 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
1355 static const char fn[] = "/proc/net/route";
1360 *netdev_name = NULL;
1361 stream = fopen(fn, "r");
1362 if (stream == NULL) {
1363 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1368 while (fgets(line, sizeof line, stream)) {
1371 uint32_t dest, gateway, mask;
1372 int refcnt, metric, mtu;
1373 unsigned int flags, use, window, irtt;
1376 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
1378 iface, &dest, &gateway, &flags, &refcnt,
1379 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
1381 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
1385 if (!(flags & RTF_UP)) {
1386 /* Skip routes that aren't up. */
1390 /* The output of 'dest', 'mask', and 'gateway' were given in
1391 * network byte order, so we don't need need any endian
1392 * conversions here. */
1393 if ((dest & mask) == (host->s_addr & mask)) {
1395 /* The host is directly reachable. */
1396 next_hop->s_addr = 0;
1398 /* To reach the host, we must go through a gateway. */
1399 next_hop->s_addr = gateway;
1401 *netdev_name = xstrdup(iface);
1412 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1413 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1414 * returns 0. Otherwise, it returns a positive errno value; in particular,
1415 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
1417 netdev_linux_arp_lookup(const struct netdev *netdev,
1418 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
1421 struct sockaddr_in sin;
1424 memset(&r, 0, sizeof r);
1425 sin.sin_family = AF_INET;
1426 sin.sin_addr.s_addr = ip;
1428 memcpy(&r.arp_pa, &sin, sizeof sin);
1429 r.arp_ha.sa_family = ARPHRD_ETHER;
1431 strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
1432 COVERAGE_INC(netdev_arp_lookup);
1433 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
1435 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
1436 } else if (retval != ENXIO) {
1437 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
1438 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
1444 nd_to_iff_flags(enum netdev_flags nd)
1447 if (nd & NETDEV_UP) {
1450 if (nd & NETDEV_PROMISC) {
1457 iff_to_nd_flags(int iff)
1459 enum netdev_flags nd = 0;
1463 if (iff & IFF_PROMISC) {
1464 nd |= NETDEV_PROMISC;
1470 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
1471 enum netdev_flags on, enum netdev_flags *old_flagsp)
1473 int old_flags, new_flags;
1476 error = get_flags(netdev, &old_flags);
1478 *old_flagsp = iff_to_nd_flags(old_flags);
1479 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
1480 if (new_flags != old_flags) {
1481 error = set_flags(netdev, new_flags);
1488 poll_notify(struct list *list)
1490 struct netdev_linux_notifier *notifier;
1491 LIST_FOR_EACH (notifier, struct netdev_linux_notifier, node, list) {
1492 struct netdev_notifier *n = ¬ifier->notifier;
1498 netdev_linux_poll_cb(const struct rtnetlink_change *change,
1499 void *aux OVS_UNUSED)
1502 struct list *list = shash_find_data(&netdev_linux_notifiers,
1508 struct shash_node *node;
1509 SHASH_FOR_EACH (node, &netdev_linux_notifiers) {
1510 poll_notify(node->data);
1516 netdev_linux_poll_add(struct netdev *netdev,
1517 void (*cb)(struct netdev_notifier *), void *aux,
1518 struct netdev_notifier **notifierp)
1520 const char *netdev_name = netdev_get_name(netdev);
1521 struct netdev_linux_notifier *notifier;
1524 if (shash_is_empty(&netdev_linux_notifiers)) {
1525 int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier,
1526 netdev_linux_poll_cb, NULL);
1532 list = shash_find_data(&netdev_linux_notifiers, netdev_name);
1534 list = xmalloc(sizeof *list);
1536 shash_add(&netdev_linux_notifiers, netdev_name, list);
1539 notifier = xmalloc(sizeof *notifier);
1540 netdev_notifier_init(¬ifier->notifier, netdev, cb, aux);
1541 list_push_back(list, ¬ifier->node);
1542 *notifierp = ¬ifier->notifier;
1547 netdev_linux_poll_remove(struct netdev_notifier *notifier_)
1549 struct netdev_linux_notifier *notifier =
1550 CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier);
1553 /* Remove 'notifier' from its list. */
1554 list = list_remove(¬ifier->node);
1555 if (list_is_empty(list)) {
1556 /* The list is now empty. Remove it from the hash and free it. */
1557 const char *netdev_name = netdev_get_name(notifier->notifier.netdev);
1558 shash_delete(&netdev_linux_notifiers,
1559 shash_find(&netdev_linux_notifiers, netdev_name));
1564 /* If that was the last notifier, unregister. */
1565 if (shash_is_empty(&netdev_linux_notifiers)) {
1566 rtnetlink_notifier_unregister(&netdev_linux_poll_notifier);
1570 const struct netdev_class netdev_linux_class = {
1577 netdev_linux_create_system,
1578 netdev_linux_destroy,
1579 NULL, /* reconfigure */
1584 netdev_linux_enumerate,
1587 netdev_linux_recv_wait,
1591 netdev_linux_send_wait,
1593 netdev_linux_set_etheraddr,
1594 netdev_linux_get_etheraddr,
1595 netdev_linux_get_mtu,
1596 netdev_linux_get_ifindex,
1597 netdev_linux_get_carrier,
1598 netdev_linux_get_stats,
1599 netdev_vport_set_stats,
1601 netdev_linux_get_features,
1602 netdev_linux_set_advertisements,
1603 netdev_linux_get_vlan_vid,
1604 netdev_linux_set_policing,
1606 netdev_linux_get_in4,
1607 netdev_linux_set_in4,
1608 netdev_linux_get_in6,
1609 netdev_linux_add_router,
1610 netdev_linux_get_next_hop,
1611 netdev_linux_arp_lookup,
1613 netdev_linux_update_flags,
1615 netdev_linux_poll_add,
1616 netdev_linux_poll_remove,
1619 const struct netdev_class netdev_tap_class = {
1626 netdev_linux_create_tap,
1627 netdev_linux_destroy,
1628 NULL, /* reconfigure */
1633 NULL, /* enumerate */
1636 netdev_linux_recv_wait,
1640 netdev_linux_send_wait,
1642 netdev_linux_set_etheraddr,
1643 netdev_linux_get_etheraddr,
1644 netdev_linux_get_mtu,
1645 netdev_linux_get_ifindex,
1646 netdev_linux_get_carrier,
1647 netdev_linux_get_stats,
1648 NULL, /* set_stats */
1650 netdev_linux_get_features,
1651 netdev_linux_set_advertisements,
1652 netdev_linux_get_vlan_vid,
1653 netdev_linux_set_policing,
1655 netdev_linux_get_in4,
1656 netdev_linux_set_in4,
1657 netdev_linux_get_in6,
1658 netdev_linux_add_router,
1659 netdev_linux_get_next_hop,
1660 netdev_linux_arp_lookup,
1662 netdev_linux_update_flags,
1664 netdev_linux_poll_add,
1665 netdev_linux_poll_remove,
1670 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
1672 /* Policy for RTNLGRP_LINK messages.
1674 * There are *many* more fields in these messages, but currently we only
1675 * care about these fields. */
1676 static const struct nl_policy rtnlgrp_link_policy[] = {
1677 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
1678 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
1679 .min_len = sizeof(struct rtnl_link_stats) },
1682 struct nl_sock *rtnl_sock;
1683 struct ofpbuf request;
1684 struct ofpbuf *reply;
1685 struct ifinfomsg *ifi;
1686 const struct rtnl_link_stats *rtnl_stats;
1687 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
1690 error = get_rtnl_sock(&rtnl_sock);
1695 ofpbuf_init(&request, 0);
1696 nl_msg_put_nlmsghdr(&request, sizeof *ifi, RTM_GETLINK, NLM_F_REQUEST);
1697 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
1698 ifi->ifi_family = PF_UNSPEC;
1699 ifi->ifi_index = ifindex;
1700 error = nl_sock_transact(rtnl_sock, &request, &reply);
1701 ofpbuf_uninit(&request);
1706 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
1707 rtnlgrp_link_policy,
1708 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
1709 ofpbuf_delete(reply);
1713 if (!attrs[IFLA_STATS]) {
1714 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
1715 ofpbuf_delete(reply);
1719 rtnl_stats = nl_attr_get(attrs[IFLA_STATS]);
1720 stats->rx_packets = rtnl_stats->rx_packets;
1721 stats->tx_packets = rtnl_stats->tx_packets;
1722 stats->rx_bytes = rtnl_stats->rx_bytes;
1723 stats->tx_bytes = rtnl_stats->tx_bytes;
1724 stats->rx_errors = rtnl_stats->rx_errors;
1725 stats->tx_errors = rtnl_stats->tx_errors;
1726 stats->rx_dropped = rtnl_stats->rx_dropped;
1727 stats->tx_dropped = rtnl_stats->tx_dropped;
1728 stats->multicast = rtnl_stats->multicast;
1729 stats->collisions = rtnl_stats->collisions;
1730 stats->rx_length_errors = rtnl_stats->rx_length_errors;
1731 stats->rx_over_errors = rtnl_stats->rx_over_errors;
1732 stats->rx_crc_errors = rtnl_stats->rx_crc_errors;
1733 stats->rx_frame_errors = rtnl_stats->rx_frame_errors;
1734 stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors;
1735 stats->rx_missed_errors = rtnl_stats->rx_missed_errors;
1736 stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors;
1737 stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors;
1738 stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors;
1739 stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors;
1740 stats->tx_window_errors = rtnl_stats->tx_window_errors;
1742 ofpbuf_delete(reply);
1748 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
1750 static const char fn[] = "/proc/net/dev";
1755 stream = fopen(fn, "r");
1757 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1762 while (fgets(line, sizeof line, stream)) {
1765 #define X64 "%"SCNu64
1768 X64 X64 X64 X64 X64 X64 X64 "%*u"
1769 X64 X64 X64 X64 X64 X64 X64 "%*u",
1775 &stats->rx_fifo_errors,
1776 &stats->rx_frame_errors,
1782 &stats->tx_fifo_errors,
1784 &stats->tx_carrier_errors) != 15) {
1785 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
1786 } else if (!strcmp(devname, netdev_name)) {
1787 stats->rx_length_errors = UINT64_MAX;
1788 stats->rx_over_errors = UINT64_MAX;
1789 stats->rx_crc_errors = UINT64_MAX;
1790 stats->rx_missed_errors = UINT64_MAX;
1791 stats->tx_aborted_errors = UINT64_MAX;
1792 stats->tx_heartbeat_errors = UINT64_MAX;
1793 stats->tx_window_errors = UINT64_MAX;
1799 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
1805 get_flags(const struct netdev *netdev, int *flags)
1810 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS,
1812 *flags = ifr.ifr_flags;
1817 set_flags(struct netdev *netdev, int flags)
1821 ifr.ifr_flags = flags;
1822 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
1827 do_get_ifindex(const char *netdev_name)
1831 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
1832 COVERAGE_INC(netdev_get_ifindex);
1833 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
1834 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
1835 netdev_name, strerror(errno));
1838 return ifr.ifr_ifindex;
1842 get_ifindex(const struct netdev *netdev_, int *ifindexp)
1844 struct netdev_dev_linux *netdev_dev =
1845 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1847 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
1848 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
1852 netdev_dev->cache_valid |= VALID_IFINDEX;
1853 netdev_dev->ifindex = ifindex;
1855 *ifindexp = netdev_dev->ifindex;
1860 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
1865 memset(&ifr, 0, sizeof ifr);
1866 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
1867 COVERAGE_INC(netdev_get_hwaddr);
1868 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
1869 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
1870 netdev_name, strerror(errno));
1873 hwaddr_family = ifr.ifr_hwaddr.sa_family;
1874 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
1875 VLOG_WARN("%s device has unknown hardware address family %d",
1876 netdev_name, hwaddr_family);
1878 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
1883 set_etheraddr(const char *netdev_name, int hwaddr_family,
1884 const uint8_t mac[ETH_ADDR_LEN])
1888 memset(&ifr, 0, sizeof ifr);
1889 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
1890 ifr.ifr_hwaddr.sa_family = hwaddr_family;
1891 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
1892 COVERAGE_INC(netdev_set_hwaddr);
1893 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
1894 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
1895 netdev_name, strerror(errno));
1902 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
1903 int cmd, const char *cmd_name)
1907 memset(&ifr, 0, sizeof ifr);
1908 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
1909 ifr.ifr_data = (caddr_t) ecmd;
1912 COVERAGE_INC(netdev_ethtool);
1913 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
1916 if (errno != EOPNOTSUPP) {
1917 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
1918 "failed: %s", cmd_name, name, strerror(errno));
1920 /* The device doesn't support this operation. That's pretty
1921 * common, so there's no point in logging anything. */
1928 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
1929 const char *cmd_name)
1931 strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
1932 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
1933 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
1941 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
1942 int cmd, const char *cmd_name)
1947 ifr.ifr_addr.sa_family = AF_INET;
1948 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
1950 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
1951 *ip = sin->sin_addr;
1956 /* Obtains a Netlink routing socket that is not subscribed to any multicast
1957 * groups. Returns 0 if successful, otherwise a positive errno value. Stores
1958 * the socket in '*rtnl_sockp' if successful, otherwise a null pointer. */
1960 get_rtnl_sock(struct nl_sock **rtnl_sockp)
1962 static struct nl_sock *sock;
1966 error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &sock);
1968 VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",