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"
53 #include "openflow/openflow.h"
54 #include "openvswitch/internal_dev.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
94 struct netdev_dev_linux {
95 struct netdev_dev netdev_dev;
97 struct shash_node *shash_node;
98 unsigned int cache_valid;
100 /* The following are figured out "on demand" only. They are only valid
101 * when the corresponding VALID_* bit in 'cache_valid' is set. */
103 uint8_t etheraddr[ETH_ADDR_LEN];
104 struct in_addr address, netmask;
108 bool is_internal; /* Is this an openvswitch internal device? */
109 bool is_tap; /* Is this a tuntap device? */
110 uint32_t kbits_rate; /* Policing data. */
111 uint32_t kbits_burst;
114 struct tap_state tap;
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);
242 /* Creates the netdev device of 'type' with 'name'. */
244 netdev_linux_create_system(const char *name, const char *type OVS_UNUSED,
245 const struct shash *args, struct netdev_dev **netdev_devp)
247 struct netdev_dev_linux *netdev_dev;
250 if (!shash_is_empty(args)) {
251 VLOG_WARN("%s: arguments for system devices should be empty", name);
254 if (!cache_notifier_refcount) {
255 error = rtnetlink_notifier_register(&netdev_linux_cache_notifier,
256 netdev_linux_cache_cb, NULL);
261 cache_notifier_refcount++;
263 netdev_dev = xzalloc(sizeof *netdev_dev);
264 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class);
266 *netdev_devp = &netdev_dev->netdev_dev;
270 /* For most types of netdevs we open the device for each call of
271 * netdev_open(). However, this is not the case with tap devices,
272 * since it is only possible to open the device once. In this
273 * situation we share a single file descriptor, and consequently
274 * buffers, across all readers. Therefore once data is read it will
275 * be unavailable to other reads for tap devices. */
277 netdev_linux_create_tap(const char *name, const char *type OVS_UNUSED,
278 const struct shash *args, struct netdev_dev **netdev_devp)
280 struct netdev_dev_linux *netdev_dev;
281 struct tap_state *state;
282 static const char tap_dev[] = "/dev/net/tun";
286 if (!shash_is_empty(args)) {
287 VLOG_WARN("%s: arguments for TAP devices should be empty", name);
290 netdev_dev = xzalloc(sizeof *netdev_dev);
291 state = &netdev_dev->state.tap;
293 /* Open tap device. */
294 state->fd = open(tap_dev, O_RDWR);
297 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
301 /* Create tap device. */
302 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
303 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
304 if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
305 VLOG_WARN("%s: creating tap device failed: %s", name,
311 /* Make non-blocking. */
312 error = set_nonblocking(state->fd);
317 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
318 *netdev_devp = &netdev_dev->netdev_dev;
327 destroy_tap(struct netdev_dev_linux *netdev_dev)
329 struct tap_state *state = &netdev_dev->state.tap;
331 if (state->fd >= 0) {
336 /* Destroys the netdev device 'netdev_dev_'. */
338 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
340 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
341 const char *type = netdev_dev_get_type(netdev_dev_);
343 if (!strcmp(type, "system")) {
344 cache_notifier_refcount--;
346 if (!cache_notifier_refcount) {
347 rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
349 } else if (!strcmp(type, "tap")) {
350 destroy_tap(netdev_dev);
357 netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype,
358 struct netdev **netdevp)
360 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
361 struct netdev_linux *netdev;
362 enum netdev_flags flags;
365 /* Allocate network device. */
366 netdev = xzalloc(sizeof *netdev);
368 netdev_init(&netdev->netdev, netdev_dev_);
370 error = netdev_get_flags(&netdev->netdev, &flags);
371 if (error == ENODEV) {
375 if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap") &&
376 !netdev_dev->state.tap.opened) {
378 /* We assume that the first user of the tap device is the primary user
379 * and give them the tap FD. Subsequent users probably just expect
380 * this to be a system device so open it normally to avoid send/receive
381 * directions appearing to be reversed. */
382 netdev->fd = netdev_dev->state.tap.fd;
383 netdev_dev->state.tap.opened = true;
384 } else if (ethertype != NETDEV_ETH_TYPE_NONE) {
385 struct sockaddr_ll sll;
389 /* Create file descriptor. */
390 protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
391 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
393 netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
394 if (netdev->fd < 0) {
399 /* Set non-blocking mode. */
400 error = set_nonblocking(netdev->fd);
405 /* Get ethernet device index. */
406 error = get_ifindex(&netdev->netdev, &ifindex);
411 /* Bind to specific ethernet device. */
412 memset(&sll, 0, sizeof sll);
413 sll.sll_family = AF_PACKET;
414 sll.sll_ifindex = ifindex;
416 (struct sockaddr *) &sll, sizeof sll) < 0) {
418 VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_),
423 /* Between the socket() and bind() calls above, the socket receives all
424 * packets of the requested type on all system interfaces. We do not
425 * want to receive that data, but there is no way to avoid it. So we
426 * must now drain out the receive queue. */
427 error = drain_rcvbuf(netdev->fd);
433 *netdevp = &netdev->netdev;
437 netdev_uninit(&netdev->netdev, true);
441 /* Closes and destroys 'netdev'. */
443 netdev_linux_close(struct netdev *netdev_)
445 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
447 if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
453 /* Initializes 'svec' with a list of the names of all known network devices. */
455 netdev_linux_enumerate(struct svec *svec)
457 struct if_nameindex *names;
459 names = if_nameindex();
463 for (i = 0; names[i].if_name != NULL; i++) {
464 svec_add(svec, names[i].if_name);
466 if_freenameindex(names);
469 VLOG_WARN("could not obtain list of network device names: %s",
476 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
478 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
480 if (netdev->fd < 0) {
481 /* Device was opened with NETDEV_ETH_TYPE_NONE. */
486 ssize_t retval = read(netdev->fd, data, size);
489 } else if (errno != EINTR) {
490 if (errno != EAGAIN) {
491 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
492 strerror(errno), netdev_get_name(netdev_));
499 /* Registers with the poll loop to wake up from the next call to poll_block()
500 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
502 netdev_linux_recv_wait(struct netdev *netdev_)
504 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
505 if (netdev->fd >= 0) {
506 poll_fd_wait(netdev->fd, POLLIN);
510 /* Discards all packets waiting to be received from 'netdev'. */
512 netdev_linux_drain(struct netdev *netdev_)
514 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
515 if (netdev->fd < 0) {
517 } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
519 int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
520 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
524 drain_fd(netdev->fd, ifr.ifr_qlen);
527 return drain_rcvbuf(netdev->fd);
531 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
532 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
533 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
534 * the packet is too big or too small to transmit on the device.
536 * The caller retains ownership of 'buffer' in all cases.
538 * The kernel maintains a packet transmission queue, so the caller is not
539 * expected to do additional queuing of packets. */
541 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
543 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
545 /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
547 if (netdev->fd < 0) {
552 ssize_t retval = write(netdev->fd, data, size);
554 /* The Linux AF_PACKET implementation never blocks waiting for room
555 * for packets, instead returning ENOBUFS. Translate this into
556 * EAGAIN for the caller. */
557 if (errno == ENOBUFS) {
559 } else if (errno == EINTR) {
561 } else if (errno != EAGAIN) {
562 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
563 netdev_get_name(netdev_), strerror(errno));
566 } else if (retval != size) {
567 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
568 "%zu) on %s", retval, size, netdev_get_name(netdev_));
576 /* Registers with the poll loop to wake up from the next call to poll_block()
577 * when the packet transmission queue has sufficient room to transmit a packet
578 * with netdev_send().
580 * The kernel maintains a packet transmission queue, so the client is not
581 * expected to do additional queuing of packets. Thus, this function is
582 * unlikely to ever be used. It is included for completeness. */
584 netdev_linux_send_wait(struct netdev *netdev_)
586 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
587 if (netdev->fd < 0) {
589 } else if (strcmp(netdev_get_type(netdev_), "tap")) {
590 poll_fd_wait(netdev->fd, POLLOUT);
592 /* TAP device always accepts packets.*/
593 poll_immediate_wake();
597 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
598 * otherwise a positive errno value. */
600 netdev_linux_set_etheraddr(struct netdev *netdev_,
601 const uint8_t mac[ETH_ADDR_LEN])
603 struct netdev_dev_linux *netdev_dev =
604 netdev_dev_linux_cast(netdev_get_dev(netdev_));
607 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)
608 || !eth_addr_equals(netdev_dev->etheraddr, mac)) {
609 error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
611 netdev_dev->cache_valid |= VALID_ETHERADDR;
612 memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
620 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
621 * free the returned buffer. */
623 netdev_linux_get_etheraddr(const struct netdev *netdev_,
624 uint8_t mac[ETH_ADDR_LEN])
626 struct netdev_dev_linux *netdev_dev =
627 netdev_dev_linux_cast(netdev_get_dev(netdev_));
628 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
629 int error = get_etheraddr(netdev_get_name(netdev_),
630 netdev_dev->etheraddr);
634 netdev_dev->cache_valid |= VALID_ETHERADDR;
636 memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
640 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
641 * in bytes, not including the hardware header; thus, this is typically 1500
642 * bytes for Ethernet devices. */
644 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
646 struct netdev_dev_linux *netdev_dev =
647 netdev_dev_linux_cast(netdev_get_dev(netdev_));
648 if (!(netdev_dev->cache_valid & VALID_MTU)) {
652 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
653 SIOCGIFMTU, "SIOCGIFMTU");
657 netdev_dev->mtu = ifr.ifr_mtu;
658 netdev_dev->cache_valid |= VALID_MTU;
660 *mtup = netdev_dev->mtu;
664 /* Returns the ifindex of 'netdev', if successful, as a positive number.
665 * On failure, returns a negative errno value. */
667 netdev_linux_get_ifindex(const struct netdev *netdev)
671 error = get_ifindex(netdev, &ifindex);
672 return error ? -error : ifindex;
676 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
678 struct netdev_dev_linux *netdev_dev =
679 netdev_dev_linux_cast(netdev_get_dev(netdev_));
684 if (!(netdev_dev->cache_valid & VALID_CARRIER)) {
688 fn = xasprintf("/sys/class/net/%s/carrier",
689 netdev_get_name(netdev_));
690 fd = open(fn, O_RDONLY);
693 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error));
697 retval = read(fd, line, sizeof line);
700 if (error == EINVAL) {
701 /* This is the normal return value when we try to check carrier
702 * if the network device is not up. */
704 VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error));
707 } else if (retval == 0) {
709 VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn);
713 if (line[0] != '0' && line[0] != '1') {
715 VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)",
719 netdev_dev->carrier = line[0] != '0';
720 netdev_dev->cache_valid |= VALID_CARRIER;
722 *carrier = netdev_dev->carrier;
733 /* Check whether we can we use RTM_GETLINK to get network device statistics.
734 * In pre-2.6.19 kernels, this was only available if wireless extensions were
737 check_for_working_netlink_stats(void)
739 /* Decide on the netdev_get_stats() implementation to use. Netlink is
740 * preferable, so if that works, we'll use it. */
741 int ifindex = do_get_ifindex("lo");
743 VLOG_WARN("failed to get ifindex for lo, "
744 "obtaining netdev stats from proc");
747 struct netdev_stats stats;
748 int error = get_stats_via_netlink(ifindex, &stats);
750 VLOG_DBG("obtaining netdev stats via rtnetlink");
753 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
754 "via proc (you are probably running a pre-2.6.19 "
755 "kernel)", strerror(error));
761 /* Brings the 'is_internal' and 'is_tap' members of 'netdev_dev' up-to-date. */
763 netdev_linux_update_is_pseudo(struct netdev_dev_linux *netdev_dev)
765 if (!(netdev_dev->cache_valid & VALID_IS_PSEUDO)) {
766 const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
767 const char *type = netdev_dev_get_type(&netdev_dev->netdev_dev);
769 netdev_dev->is_tap = !strcmp(type, "tap");
770 netdev_dev->is_internal = false;
771 if (!netdev_dev->is_tap) {
772 struct ethtool_drvinfo drvinfo;
775 memset(&drvinfo, 0, sizeof drvinfo);
776 error = netdev_linux_do_ethtool(name,
777 (struct ethtool_cmd *)&drvinfo,
781 if (!error && !strcmp(drvinfo.driver, "openvswitch")) {
782 netdev_dev->is_internal = true;
786 netdev_dev->cache_valid |= VALID_IS_PSEUDO;
791 swap_uint64(uint64_t *a, uint64_t *b)
798 /* Retrieves current device stats for 'netdev'.
800 * XXX All of the members of struct netdev_stats are 64 bits wide, but on
801 * 32-bit architectures the Linux network stats are only 32 bits. */
803 netdev_linux_get_stats(const struct netdev *netdev_,
804 struct netdev_stats *stats)
806 struct netdev_dev_linux *netdev_dev =
807 netdev_dev_linux_cast(netdev_get_dev(netdev_));
808 static int use_netlink_stats = -1;
811 COVERAGE_INC(netdev_get_stats);
813 if (use_netlink_stats < 0) {
814 use_netlink_stats = check_for_working_netlink_stats();
816 if (use_netlink_stats) {
819 error = get_ifindex(netdev_, &ifindex);
821 error = get_stats_via_netlink(ifindex, stats);
824 error = get_stats_via_proc(netdev_get_name(netdev_), stats);
827 /* If this port is an internal port then the transmit and receive stats
828 * will appear to be swapped relative to the other ports since we are the
829 * one sending the data, not a remote computer. For consistency, we swap
831 netdev_linux_update_is_pseudo(netdev_dev);
832 if (!error && (netdev_dev->is_internal || netdev_dev->is_tap)) {
833 swap_uint64(&stats->rx_packets, &stats->tx_packets);
834 swap_uint64(&stats->rx_bytes, &stats->tx_bytes);
835 swap_uint64(&stats->rx_errors, &stats->tx_errors);
836 swap_uint64(&stats->rx_dropped, &stats->tx_dropped);
837 stats->rx_length_errors = 0;
838 stats->rx_over_errors = 0;
839 stats->rx_crc_errors = 0;
840 stats->rx_frame_errors = 0;
841 stats->rx_fifo_errors = 0;
842 stats->rx_missed_errors = 0;
843 stats->tx_aborted_errors = 0;
844 stats->tx_carrier_errors = 0;
845 stats->tx_fifo_errors = 0;
846 stats->tx_heartbeat_errors = 0;
847 stats->tx_window_errors = 0;
854 netdev_linux_set_stats(struct netdev *netdev,
855 const struct netdev_stats *stats)
857 struct netdev_dev_linux *netdev_dev =
858 netdev_dev_linux_cast(netdev_get_dev(netdev));
859 struct internal_dev_stats dp_dev_stats;
862 /* We must reject this call if 'netdev' is not an Open vSwitch internal
863 * port, because the ioctl that we are about to execute is in the "device
864 * private ioctls" range, which means that executing it on a device that
865 * is not the type we expect could do any random thing.
867 * (Amusingly, these ioctl numbers are commented "THESE IOCTLS ARE
868 * _DEPRECATED_ AND WILL DISAPPEAR IN 2.5.X" in linux/sockios.h. I guess
869 * DaveM is a little behind on that.) */
870 netdev_linux_update_is_pseudo(netdev_dev);
871 if (!netdev_dev->is_internal) {
875 /* This actually only sets the *offset* that the dp_dev applies, but in our
876 * usage for fake bond devices the dp_dev never has any traffic of it own
877 * so it has the same effect. */
878 dp_dev_stats.rx_packets = stats->rx_packets;
879 dp_dev_stats.rx_bytes = stats->rx_bytes;
880 dp_dev_stats.tx_packets = stats->tx_packets;
881 dp_dev_stats.tx_bytes = stats->tx_bytes;
882 ifr.ifr_data = (void *) &dp_dev_stats;
883 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr,
884 INTERNAL_DEV_SET_STATS,
885 "INTERNAL_DEV_SET_STATS");
888 /* Stores the features supported by 'netdev' into each of '*current',
889 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
890 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
891 * successful, otherwise a positive errno value. */
893 netdev_linux_get_features(struct netdev *netdev,
894 uint32_t *current, uint32_t *advertised,
895 uint32_t *supported, uint32_t *peer)
897 struct ethtool_cmd ecmd;
900 memset(&ecmd, 0, sizeof ecmd);
901 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
902 ETHTOOL_GSET, "ETHTOOL_GSET");
907 /* Supported features. */
909 if (ecmd.supported & SUPPORTED_10baseT_Half) {
910 *supported |= OFPPF_10MB_HD;
912 if (ecmd.supported & SUPPORTED_10baseT_Full) {
913 *supported |= OFPPF_10MB_FD;
915 if (ecmd.supported & SUPPORTED_100baseT_Half) {
916 *supported |= OFPPF_100MB_HD;
918 if (ecmd.supported & SUPPORTED_100baseT_Full) {
919 *supported |= OFPPF_100MB_FD;
921 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
922 *supported |= OFPPF_1GB_HD;
924 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
925 *supported |= OFPPF_1GB_FD;
927 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
928 *supported |= OFPPF_10GB_FD;
930 if (ecmd.supported & SUPPORTED_TP) {
931 *supported |= OFPPF_COPPER;
933 if (ecmd.supported & SUPPORTED_FIBRE) {
934 *supported |= OFPPF_FIBER;
936 if (ecmd.supported & SUPPORTED_Autoneg) {
937 *supported |= OFPPF_AUTONEG;
939 if (ecmd.supported & SUPPORTED_Pause) {
940 *supported |= OFPPF_PAUSE;
942 if (ecmd.supported & SUPPORTED_Asym_Pause) {
943 *supported |= OFPPF_PAUSE_ASYM;
946 /* Advertised features. */
948 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
949 *advertised |= OFPPF_10MB_HD;
951 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
952 *advertised |= OFPPF_10MB_FD;
954 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
955 *advertised |= OFPPF_100MB_HD;
957 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
958 *advertised |= OFPPF_100MB_FD;
960 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
961 *advertised |= OFPPF_1GB_HD;
963 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
964 *advertised |= OFPPF_1GB_FD;
966 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
967 *advertised |= OFPPF_10GB_FD;
969 if (ecmd.advertising & ADVERTISED_TP) {
970 *advertised |= OFPPF_COPPER;
972 if (ecmd.advertising & ADVERTISED_FIBRE) {
973 *advertised |= OFPPF_FIBER;
975 if (ecmd.advertising & ADVERTISED_Autoneg) {
976 *advertised |= OFPPF_AUTONEG;
978 if (ecmd.advertising & ADVERTISED_Pause) {
979 *advertised |= OFPPF_PAUSE;
981 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
982 *advertised |= OFPPF_PAUSE_ASYM;
985 /* Current settings. */
986 if (ecmd.speed == SPEED_10) {
987 *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD;
988 } else if (ecmd.speed == SPEED_100) {
989 *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD;
990 } else if (ecmd.speed == SPEED_1000) {
991 *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD;
992 } else if (ecmd.speed == SPEED_10000) {
993 *current = OFPPF_10GB_FD;
998 if (ecmd.port == PORT_TP) {
999 *current |= OFPPF_COPPER;
1000 } else if (ecmd.port == PORT_FIBRE) {
1001 *current |= OFPPF_FIBER;
1005 *current |= OFPPF_AUTONEG;
1008 /* Peer advertisements. */
1009 *peer = 0; /* XXX */
1014 /* Set the features advertised by 'netdev' to 'advertise'. */
1016 netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
1018 struct ethtool_cmd ecmd;
1021 memset(&ecmd, 0, sizeof ecmd);
1022 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1023 ETHTOOL_GSET, "ETHTOOL_GSET");
1028 ecmd.advertising = 0;
1029 if (advertise & OFPPF_10MB_HD) {
1030 ecmd.advertising |= ADVERTISED_10baseT_Half;
1032 if (advertise & OFPPF_10MB_FD) {
1033 ecmd.advertising |= ADVERTISED_10baseT_Full;
1035 if (advertise & OFPPF_100MB_HD) {
1036 ecmd.advertising |= ADVERTISED_100baseT_Half;
1038 if (advertise & OFPPF_100MB_FD) {
1039 ecmd.advertising |= ADVERTISED_100baseT_Full;
1041 if (advertise & OFPPF_1GB_HD) {
1042 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1044 if (advertise & OFPPF_1GB_FD) {
1045 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1047 if (advertise & OFPPF_10GB_FD) {
1048 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1050 if (advertise & OFPPF_COPPER) {
1051 ecmd.advertising |= ADVERTISED_TP;
1053 if (advertise & OFPPF_FIBER) {
1054 ecmd.advertising |= ADVERTISED_FIBRE;
1056 if (advertise & OFPPF_AUTONEG) {
1057 ecmd.advertising |= ADVERTISED_Autoneg;
1059 if (advertise & OFPPF_PAUSE) {
1060 ecmd.advertising |= ADVERTISED_Pause;
1062 if (advertise & OFPPF_PAUSE_ASYM) {
1063 ecmd.advertising |= ADVERTISED_Asym_Pause;
1065 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1066 ETHTOOL_SSET, "ETHTOOL_SSET");
1069 /* If 'netdev_name' is the name of a VLAN network device (e.g. one created with
1070 * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device
1071 * and returns 0. Otherwise returns a errno value (specifically ENOENT if
1072 * 'netdev_name' is the name of a network device that is not a VLAN device) and
1073 * sets '*vlan_vid' to -1. */
1075 netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1077 const char *netdev_name = netdev_get_name(netdev);
1078 struct ds line = DS_EMPTY_INITIALIZER;
1079 FILE *stream = NULL;
1083 COVERAGE_INC(netdev_get_vlan_vid);
1084 fn = xasprintf("/proc/net/vlan/%s", netdev_name);
1085 stream = fopen(fn, "r");
1091 if (ds_get_line(&line, stream)) {
1092 if (ferror(stream)) {
1094 VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno));
1097 VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn);
1102 if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) {
1104 VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"",
1105 fn, ds_cstr(&line));
1123 #define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress"
1124 #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"
1126 /* Remove ingress policing from 'netdev'. Returns 0 if successful, otherwise a
1127 * positive errno value.
1129 * This function is equivalent to running
1130 * /sbin/tc qdisc del dev %s handle ffff: ingress
1131 * but it is much, much faster.
1134 netdev_linux_remove_policing(struct netdev *netdev)
1136 struct netdev_dev_linux *netdev_dev =
1137 netdev_dev_linux_cast(netdev_get_dev(netdev));
1138 const char *netdev_name = netdev_get_name(netdev);
1140 struct ofpbuf request;
1141 struct ofpbuf *reply;
1142 struct tcmsg *tcmsg;
1143 struct nl_sock *rtnl_sock;
1147 error = get_ifindex(netdev, &ifindex);
1152 error = get_rtnl_sock(&rtnl_sock);
1157 ofpbuf_init(&request, 0);
1158 nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *tcmsg,
1159 RTM_DELQDISC, NLM_F_REQUEST);
1160 tcmsg = ofpbuf_put_zeros(&request, sizeof *tcmsg);
1161 tcmsg->tcm_family = AF_UNSPEC;
1162 tcmsg->tcm_ifindex = ifindex;
1163 tcmsg->tcm_handle = 0xffff0000;
1164 tcmsg->tcm_parent = TC_H_INGRESS;
1165 nl_msg_put_string(&request, TCA_KIND, "ingress");
1166 nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0);
1167 error = nl_sock_transact(rtnl_sock, &request, &reply);
1168 ofpbuf_uninit(&request);
1169 ofpbuf_delete(reply);
1170 if (error && error != ENOENT && error != EINVAL) {
1171 VLOG_WARN_RL(&rl, "%s: removing policing failed: %s",
1172 netdev_name, strerror(error));
1176 netdev_dev->kbits_rate = 0;
1177 netdev_dev->kbits_burst = 0;
1178 netdev_dev->cache_valid |= VALID_POLICING;
1182 /* Attempts to set input rate limiting (policing) policy. */
1184 netdev_linux_set_policing(struct netdev *netdev,
1185 uint32_t kbits_rate, uint32_t kbits_burst)
1187 struct netdev_dev_linux *netdev_dev =
1188 netdev_dev_linux_cast(netdev_get_dev(netdev));
1189 const char *netdev_name = netdev_get_name(netdev);
1192 COVERAGE_INC(netdev_set_policing);
1194 kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */
1195 : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */
1196 : kbits_burst); /* Stick with user-specified value. */
1198 if (netdev_dev->cache_valid & VALID_POLICING
1199 && netdev_dev->kbits_rate == kbits_rate
1200 && netdev_dev->kbits_burst == kbits_burst) {
1201 /* Assume that settings haven't changed since we last set them. */
1205 netdev_linux_remove_policing(netdev);
1207 snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name);
1208 if (system(command) != 0) {
1209 VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name);
1213 snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name,
1214 kbits_rate, kbits_burst);
1215 if (system(command) != 0) {
1216 VLOG_WARN_RL(&rl, "%s: problem configuring policing",
1221 netdev_dev->kbits_rate = kbits_rate;
1222 netdev_dev->kbits_burst = kbits_burst;
1223 netdev_dev->cache_valid |= VALID_POLICING;
1230 netdev_linux_get_in4(const struct netdev *netdev_,
1231 struct in_addr *address, struct in_addr *netmask)
1233 struct netdev_dev_linux *netdev_dev =
1234 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1236 if (!(netdev_dev->cache_valid & VALID_IN4)) {
1239 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
1240 SIOCGIFADDR, "SIOCGIFADDR");
1245 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
1246 SIOCGIFNETMASK, "SIOCGIFNETMASK");
1251 netdev_dev->cache_valid |= VALID_IN4;
1253 *address = netdev_dev->address;
1254 *netmask = netdev_dev->netmask;
1255 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
1259 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
1260 struct in_addr netmask)
1262 struct netdev_dev_linux *netdev_dev =
1263 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1266 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
1268 netdev_dev->cache_valid |= VALID_IN4;
1269 netdev_dev->address = address;
1270 netdev_dev->netmask = netmask;
1271 if (address.s_addr != INADDR_ANY) {
1272 error = do_set_addr(netdev_, SIOCSIFNETMASK,
1273 "SIOCSIFNETMASK", netmask);
1280 parse_if_inet6_line(const char *line,
1281 struct in6_addr *in6, char ifname[16 + 1])
1283 uint8_t *s6 = in6->s6_addr;
1284 #define X8 "%2"SCNx8
1286 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
1287 "%*x %*x %*x %*x %16s\n",
1288 &s6[0], &s6[1], &s6[2], &s6[3],
1289 &s6[4], &s6[5], &s6[6], &s6[7],
1290 &s6[8], &s6[9], &s6[10], &s6[11],
1291 &s6[12], &s6[13], &s6[14], &s6[15],
1295 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
1296 * 'in6' is non-null) and returns true. Otherwise, returns false. */
1298 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
1300 struct netdev_dev_linux *netdev_dev =
1301 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1302 if (!(netdev_dev->cache_valid & VALID_IN6)) {
1306 netdev_dev->in6 = in6addr_any;
1308 file = fopen("/proc/net/if_inet6", "r");
1310 const char *name = netdev_get_name(netdev_);
1311 while (fgets(line, sizeof line, file)) {
1312 struct in6_addr in6;
1313 char ifname[16 + 1];
1314 if (parse_if_inet6_line(line, &in6, ifname)
1315 && !strcmp(name, ifname))
1317 netdev_dev->in6 = in6;
1323 netdev_dev->cache_valid |= VALID_IN6;
1325 *in6 = netdev_dev->in6;
1330 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
1332 struct sockaddr_in sin;
1333 memset(&sin, 0, sizeof sin);
1334 sin.sin_family = AF_INET;
1335 sin.sin_addr = addr;
1338 memset(sa, 0, sizeof *sa);
1339 memcpy(sa, &sin, sizeof sin);
1343 do_set_addr(struct netdev *netdev,
1344 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
1347 strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
1348 make_in4_sockaddr(&ifr.ifr_addr, addr);
1350 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
1354 /* Adds 'router' as a default IP gateway. */
1356 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
1358 struct in_addr any = { INADDR_ANY };
1362 memset(&rt, 0, sizeof rt);
1363 make_in4_sockaddr(&rt.rt_dst, any);
1364 make_in4_sockaddr(&rt.rt_gateway, router);
1365 make_in4_sockaddr(&rt.rt_genmask, any);
1366 rt.rt_flags = RTF_UP | RTF_GATEWAY;
1367 COVERAGE_INC(netdev_add_router);
1368 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
1370 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
1376 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
1379 static const char fn[] = "/proc/net/route";
1384 *netdev_name = NULL;
1385 stream = fopen(fn, "r");
1386 if (stream == NULL) {
1387 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1392 while (fgets(line, sizeof line, stream)) {
1395 uint32_t dest, gateway, mask;
1396 int refcnt, metric, mtu;
1397 unsigned int flags, use, window, irtt;
1400 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
1402 iface, &dest, &gateway, &flags, &refcnt,
1403 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
1405 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
1409 if (!(flags & RTF_UP)) {
1410 /* Skip routes that aren't up. */
1414 /* The output of 'dest', 'mask', and 'gateway' were given in
1415 * network byte order, so we don't need need any endian
1416 * conversions here. */
1417 if ((dest & mask) == (host->s_addr & mask)) {
1419 /* The host is directly reachable. */
1420 next_hop->s_addr = 0;
1422 /* To reach the host, we must go through a gateway. */
1423 next_hop->s_addr = gateway;
1425 *netdev_name = xstrdup(iface);
1436 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1437 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1438 * returns 0. Otherwise, it returns a positive errno value; in particular,
1439 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
1441 netdev_linux_arp_lookup(const struct netdev *netdev,
1442 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
1445 struct sockaddr_in sin;
1448 memset(&r, 0, sizeof r);
1449 sin.sin_family = AF_INET;
1450 sin.sin_addr.s_addr = ip;
1452 memcpy(&r.arp_pa, &sin, sizeof sin);
1453 r.arp_ha.sa_family = ARPHRD_ETHER;
1455 strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
1456 COVERAGE_INC(netdev_arp_lookup);
1457 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
1459 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
1460 } else if (retval != ENXIO) {
1461 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
1462 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
1468 nd_to_iff_flags(enum netdev_flags nd)
1471 if (nd & NETDEV_UP) {
1474 if (nd & NETDEV_PROMISC) {
1481 iff_to_nd_flags(int iff)
1483 enum netdev_flags nd = 0;
1487 if (iff & IFF_PROMISC) {
1488 nd |= NETDEV_PROMISC;
1494 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
1495 enum netdev_flags on, enum netdev_flags *old_flagsp)
1497 int old_flags, new_flags;
1500 error = get_flags(netdev, &old_flags);
1502 *old_flagsp = iff_to_nd_flags(old_flags);
1503 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
1504 if (new_flags != old_flags) {
1505 error = set_flags(netdev, new_flags);
1512 poll_notify(struct list *list)
1514 struct netdev_linux_notifier *notifier;
1515 LIST_FOR_EACH (notifier, struct netdev_linux_notifier, node, list) {
1516 struct netdev_notifier *n = ¬ifier->notifier;
1522 netdev_linux_poll_cb(const struct rtnetlink_change *change,
1523 void *aux OVS_UNUSED)
1526 struct list *list = shash_find_data(&netdev_linux_notifiers,
1532 struct shash_node *node;
1533 SHASH_FOR_EACH (node, &netdev_linux_notifiers) {
1534 poll_notify(node->data);
1540 netdev_linux_poll_add(struct netdev *netdev,
1541 void (*cb)(struct netdev_notifier *), void *aux,
1542 struct netdev_notifier **notifierp)
1544 const char *netdev_name = netdev_get_name(netdev);
1545 struct netdev_linux_notifier *notifier;
1548 if (shash_is_empty(&netdev_linux_notifiers)) {
1549 int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier,
1550 netdev_linux_poll_cb, NULL);
1556 list = shash_find_data(&netdev_linux_notifiers, netdev_name);
1558 list = xmalloc(sizeof *list);
1560 shash_add(&netdev_linux_notifiers, netdev_name, list);
1563 notifier = xmalloc(sizeof *notifier);
1564 netdev_notifier_init(¬ifier->notifier, netdev, cb, aux);
1565 list_push_back(list, ¬ifier->node);
1566 *notifierp = ¬ifier->notifier;
1571 netdev_linux_poll_remove(struct netdev_notifier *notifier_)
1573 struct netdev_linux_notifier *notifier =
1574 CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier);
1577 /* Remove 'notifier' from its list. */
1578 list = list_remove(¬ifier->node);
1579 if (list_is_empty(list)) {
1580 /* The list is now empty. Remove it from the hash and free it. */
1581 const char *netdev_name = netdev_get_name(notifier->notifier.netdev);
1582 shash_delete(&netdev_linux_notifiers,
1583 shash_find(&netdev_linux_notifiers, netdev_name));
1588 /* If that was the last notifier, unregister. */
1589 if (shash_is_empty(&netdev_linux_notifiers)) {
1590 rtnetlink_notifier_unregister(&netdev_linux_poll_notifier);
1594 const struct netdev_class netdev_linux_class = {
1601 netdev_linux_create_system,
1602 netdev_linux_destroy,
1603 NULL, /* reconfigure */
1608 netdev_linux_enumerate,
1611 netdev_linux_recv_wait,
1615 netdev_linux_send_wait,
1617 netdev_linux_set_etheraddr,
1618 netdev_linux_get_etheraddr,
1619 netdev_linux_get_mtu,
1620 netdev_linux_get_ifindex,
1621 netdev_linux_get_carrier,
1622 netdev_linux_get_stats,
1623 netdev_linux_set_stats,
1625 netdev_linux_get_features,
1626 netdev_linux_set_advertisements,
1627 netdev_linux_get_vlan_vid,
1628 netdev_linux_set_policing,
1630 netdev_linux_get_in4,
1631 netdev_linux_set_in4,
1632 netdev_linux_get_in6,
1633 netdev_linux_add_router,
1634 netdev_linux_get_next_hop,
1635 netdev_linux_arp_lookup,
1637 netdev_linux_update_flags,
1639 netdev_linux_poll_add,
1640 netdev_linux_poll_remove,
1643 const struct netdev_class netdev_tap_class = {
1650 netdev_linux_create_tap,
1651 netdev_linux_destroy,
1652 NULL, /* reconfigure */
1657 NULL, /* enumerate */
1660 netdev_linux_recv_wait,
1664 netdev_linux_send_wait,
1666 netdev_linux_set_etheraddr,
1667 netdev_linux_get_etheraddr,
1668 netdev_linux_get_mtu,
1669 netdev_linux_get_ifindex,
1670 netdev_linux_get_carrier,
1671 netdev_linux_get_stats,
1672 NULL, /* set_stats */
1674 netdev_linux_get_features,
1675 netdev_linux_set_advertisements,
1676 netdev_linux_get_vlan_vid,
1677 netdev_linux_set_policing,
1679 netdev_linux_get_in4,
1680 netdev_linux_set_in4,
1681 netdev_linux_get_in6,
1682 netdev_linux_add_router,
1683 netdev_linux_get_next_hop,
1684 netdev_linux_arp_lookup,
1686 netdev_linux_update_flags,
1688 netdev_linux_poll_add,
1689 netdev_linux_poll_remove,
1694 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
1696 /* Policy for RTNLGRP_LINK messages.
1698 * There are *many* more fields in these messages, but currently we only
1699 * care about these fields. */
1700 static const struct nl_policy rtnlgrp_link_policy[] = {
1701 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
1702 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
1703 .min_len = sizeof(struct rtnl_link_stats) },
1706 struct nl_sock *rtnl_sock;
1707 struct ofpbuf request;
1708 struct ofpbuf *reply;
1709 struct ifinfomsg *ifi;
1710 const struct rtnl_link_stats *rtnl_stats;
1711 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
1714 error = get_rtnl_sock(&rtnl_sock);
1719 ofpbuf_init(&request, 0);
1720 nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *ifi,
1721 RTM_GETLINK, NLM_F_REQUEST);
1722 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
1723 ifi->ifi_family = PF_UNSPEC;
1724 ifi->ifi_index = ifindex;
1725 error = nl_sock_transact(rtnl_sock, &request, &reply);
1726 ofpbuf_uninit(&request);
1731 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
1732 rtnlgrp_link_policy,
1733 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
1734 ofpbuf_delete(reply);
1738 if (!attrs[IFLA_STATS]) {
1739 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
1740 ofpbuf_delete(reply);
1744 rtnl_stats = nl_attr_get(attrs[IFLA_STATS]);
1745 stats->rx_packets = rtnl_stats->rx_packets;
1746 stats->tx_packets = rtnl_stats->tx_packets;
1747 stats->rx_bytes = rtnl_stats->rx_bytes;
1748 stats->tx_bytes = rtnl_stats->tx_bytes;
1749 stats->rx_errors = rtnl_stats->rx_errors;
1750 stats->tx_errors = rtnl_stats->tx_errors;
1751 stats->rx_dropped = rtnl_stats->rx_dropped;
1752 stats->tx_dropped = rtnl_stats->tx_dropped;
1753 stats->multicast = rtnl_stats->multicast;
1754 stats->collisions = rtnl_stats->collisions;
1755 stats->rx_length_errors = rtnl_stats->rx_length_errors;
1756 stats->rx_over_errors = rtnl_stats->rx_over_errors;
1757 stats->rx_crc_errors = rtnl_stats->rx_crc_errors;
1758 stats->rx_frame_errors = rtnl_stats->rx_frame_errors;
1759 stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors;
1760 stats->rx_missed_errors = rtnl_stats->rx_missed_errors;
1761 stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors;
1762 stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors;
1763 stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors;
1764 stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors;
1765 stats->tx_window_errors = rtnl_stats->tx_window_errors;
1767 ofpbuf_delete(reply);
1773 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
1775 static const char fn[] = "/proc/net/dev";
1780 stream = fopen(fn, "r");
1782 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1787 while (fgets(line, sizeof line, stream)) {
1790 #define X64 "%"SCNu64
1793 X64 X64 X64 X64 X64 X64 X64 "%*u"
1794 X64 X64 X64 X64 X64 X64 X64 "%*u",
1800 &stats->rx_fifo_errors,
1801 &stats->rx_frame_errors,
1807 &stats->tx_fifo_errors,
1809 &stats->tx_carrier_errors) != 15) {
1810 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
1811 } else if (!strcmp(devname, netdev_name)) {
1812 stats->rx_length_errors = UINT64_MAX;
1813 stats->rx_over_errors = UINT64_MAX;
1814 stats->rx_crc_errors = UINT64_MAX;
1815 stats->rx_missed_errors = UINT64_MAX;
1816 stats->tx_aborted_errors = UINT64_MAX;
1817 stats->tx_heartbeat_errors = UINT64_MAX;
1818 stats->tx_window_errors = UINT64_MAX;
1824 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
1830 get_flags(const struct netdev *netdev, int *flags)
1835 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS,
1837 *flags = ifr.ifr_flags;
1842 set_flags(struct netdev *netdev, int flags)
1846 ifr.ifr_flags = flags;
1847 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
1852 do_get_ifindex(const char *netdev_name)
1856 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
1857 COVERAGE_INC(netdev_get_ifindex);
1858 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
1859 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
1860 netdev_name, strerror(errno));
1863 return ifr.ifr_ifindex;
1867 get_ifindex(const struct netdev *netdev_, int *ifindexp)
1869 struct netdev_dev_linux *netdev_dev =
1870 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1872 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
1873 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
1877 netdev_dev->cache_valid |= VALID_IFINDEX;
1878 netdev_dev->ifindex = ifindex;
1880 *ifindexp = netdev_dev->ifindex;
1885 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
1890 memset(&ifr, 0, sizeof ifr);
1891 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
1892 COVERAGE_INC(netdev_get_hwaddr);
1893 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
1894 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
1895 netdev_name, strerror(errno));
1898 hwaddr_family = ifr.ifr_hwaddr.sa_family;
1899 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
1900 VLOG_WARN("%s device has unknown hardware address family %d",
1901 netdev_name, hwaddr_family);
1903 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
1908 set_etheraddr(const char *netdev_name, int hwaddr_family,
1909 const uint8_t mac[ETH_ADDR_LEN])
1913 memset(&ifr, 0, sizeof ifr);
1914 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
1915 ifr.ifr_hwaddr.sa_family = hwaddr_family;
1916 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
1917 COVERAGE_INC(netdev_set_hwaddr);
1918 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
1919 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
1920 netdev_name, strerror(errno));
1927 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
1928 int cmd, const char *cmd_name)
1932 memset(&ifr, 0, sizeof ifr);
1933 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
1934 ifr.ifr_data = (caddr_t) ecmd;
1937 COVERAGE_INC(netdev_ethtool);
1938 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
1941 if (errno != EOPNOTSUPP) {
1942 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
1943 "failed: %s", cmd_name, name, strerror(errno));
1945 /* The device doesn't support this operation. That's pretty
1946 * common, so there's no point in logging anything. */
1953 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
1954 const char *cmd_name)
1956 strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
1957 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
1958 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
1966 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
1967 int cmd, const char *cmd_name)
1972 ifr.ifr_addr.sa_family = AF_INET;
1973 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
1975 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
1976 *ip = sin->sin_addr;
1981 /* Obtains a Netlink routing socket that is not subscribed to any multicast
1982 * groups. Returns 0 if successful, otherwise a positive errno value. Stores
1983 * the socket in '*rtnl_sockp' if successful, otherwise a null pointer. */
1985 get_rtnl_sock(struct nl_sock **rtnl_sockp)
1987 static struct nl_sock *sock;
1991 error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &sock);
1993 VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",