2 * Copyright (c) 2008, 2009 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.
23 #include <arpa/inet.h>
25 #include <linux/if_tun.h>
26 #include <linux/types.h>
27 #include <linux/ethtool.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 <net/if_arp.h>
38 #include <net/if_packet.h>
39 #include <net/route.h>
40 #include <netinet/in.h>
46 #include "dynamic-string.h"
47 #include "fatal-signal.h"
51 #include "openflow/openflow.h"
53 #include "poll-loop.h"
54 #include "socket-util.h"
57 /* linux/if.h defines IFF_LOWER_UP, net/if.h doesn't.
58 * net/if.h defines if_nameindex(), linux/if.h doesn't.
59 * We can't include both headers, so define IFF_LOWER_UP ourselves. */
61 #define IFF_LOWER_UP 0x10000
64 /* These were introduced in Linux 2.6.14, so they might be missing if we have
66 #ifndef ADVERTISED_Pause
67 #define ADVERTISED_Pause (1 << 13)
69 #ifndef ADVERTISED_Asym_Pause
70 #define ADVERTISED_Asym_Pause (1 << 14)
73 #define THIS_MODULE VLM_netdev
80 /* File descriptors. For ordinary network devices, the two fds below are
81 * the same; for tap devices, they differ. */
82 int netdev_fd; /* Network device. */
83 int tap_fd; /* TAP character device, if any, otherwise the
86 /* Cached network device information. */
87 int ifindex; /* -1 if not known. */
88 uint8_t etheraddr[ETH_ADDR_LEN];
95 int save_flags; /* Initial device flags. */
96 int changed_flags; /* Flags that we changed. */
99 /* Policy for RTNLGRP_LINK messages.
101 * There are *many* more fields in these messages, but currently we only care
102 * about interface names. */
103 static const struct nl_policy rtnlgrp_link_policy[] = {
104 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
105 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
106 .min_len = sizeof(struct rtnl_link_stats) },
109 /* All open network devices. */
110 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
112 /* An AF_INET socket (used for ioctl operations). */
113 static int af_inet_sock = -1;
115 /* NETLINK_ROUTE socket. */
116 static struct nl_sock *rtnl_sock;
118 /* Can we use RTM_GETLINK to get network device statistics? (In pre-2.6.19
119 * kernels, this was only available if wireless extensions were enabled.) */
120 static bool use_netlink_stats;
122 /* This is set pretty low because we probably won't learn anything from the
123 * additional log messages. */
124 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
126 static void init_netdev(void);
127 static int do_open_netdev(const char *name, int ethertype, int tap_fd,
128 struct netdev **netdev_);
129 static int restore_flags(struct netdev *netdev);
130 static int get_flags(const char *netdev_name, int *flagsp);
131 static int set_flags(const char *netdev_name, int flags);
132 static int do_get_ifindex(const char *netdev_name);
133 static int get_ifindex(const struct netdev *, int *ifindexp);
134 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN],
135 int *hwaddr_familyp);
136 static int set_etheraddr(const char *netdev_name, int hwaddr_family,
137 const uint8_t[ETH_ADDR_LEN]);
139 /* Obtains the IPv6 address for 'name' into 'in6'. */
141 get_ipv6_address(const char *name, struct in6_addr *in6)
146 file = fopen("/proc/net/if_inet6", "r");
148 /* This most likely indicates that the host doesn't have IPv6 support,
149 * so it's not really a failure condition.*/
154 while (fgets(line, sizeof line, file)) {
155 uint8_t *s6 = in6->s6_addr;
159 if (sscanf(line, " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
160 "%*x %*x %*x %*x %16s\n",
161 &s6[0], &s6[1], &s6[2], &s6[3],
162 &s6[4], &s6[5], &s6[6], &s6[7],
163 &s6[8], &s6[9], &s6[10], &s6[11],
164 &s6[12], &s6[13], &s6[14], &s6[15],
166 && !strcmp(name, ifname))
178 do_ethtool(struct netdev *netdev, struct ethtool_cmd *ecmd,
179 int cmd, const char *cmd_name)
183 memset(&ifr, 0, sizeof ifr);
184 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
185 ifr.ifr_data = (caddr_t) ecmd;
188 COVERAGE_INC(netdev_ethtool);
189 if (ioctl(netdev->netdev_fd, SIOCETHTOOL, &ifr) == 0) {
192 if (errno != EOPNOTSUPP) {
193 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
194 "failed: %s", cmd_name, netdev->name,
197 /* The device doesn't support this operation. That's pretty
198 * common, so there's no point in logging anything. */
205 do_get_features(struct netdev *netdev,
206 uint32_t *current, uint32_t *advertised,
207 uint32_t *supported, uint32_t *peer)
209 struct ethtool_cmd ecmd;
217 memset(&ecmd, 0, sizeof ecmd);
218 error = do_ethtool(netdev, &ecmd, ETHTOOL_GSET, "ETHTOOL_GSET");
223 if (ecmd.supported & SUPPORTED_10baseT_Half) {
224 *supported |= OFPPF_10MB_HD;
226 if (ecmd.supported & SUPPORTED_10baseT_Full) {
227 *supported |= OFPPF_10MB_FD;
229 if (ecmd.supported & SUPPORTED_100baseT_Half) {
230 *supported |= OFPPF_100MB_HD;
232 if (ecmd.supported & SUPPORTED_100baseT_Full) {
233 *supported |= OFPPF_100MB_FD;
235 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
236 *supported |= OFPPF_1GB_HD;
238 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
239 *supported |= OFPPF_1GB_FD;
241 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
242 *supported |= OFPPF_10GB_FD;
244 if (ecmd.supported & SUPPORTED_TP) {
245 *supported |= OFPPF_COPPER;
247 if (ecmd.supported & SUPPORTED_FIBRE) {
248 *supported |= OFPPF_FIBER;
250 if (ecmd.supported & SUPPORTED_Autoneg) {
251 *supported |= OFPPF_AUTONEG;
253 if (ecmd.supported & SUPPORTED_Pause) {
254 *supported |= OFPPF_PAUSE;
256 if (ecmd.supported & SUPPORTED_Asym_Pause) {
257 *supported |= OFPPF_PAUSE_ASYM;
260 /* Set the advertised features */
261 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
262 *advertised |= OFPPF_10MB_HD;
264 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
265 *advertised |= OFPPF_10MB_FD;
267 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
268 *advertised |= OFPPF_100MB_HD;
270 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
271 *advertised |= OFPPF_100MB_FD;
273 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
274 *advertised |= OFPPF_1GB_HD;
276 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
277 *advertised |= OFPPF_1GB_FD;
279 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
280 *advertised |= OFPPF_10GB_FD;
282 if (ecmd.advertising & ADVERTISED_TP) {
283 *advertised |= OFPPF_COPPER;
285 if (ecmd.advertising & ADVERTISED_FIBRE) {
286 *advertised |= OFPPF_FIBER;
288 if (ecmd.advertising & ADVERTISED_Autoneg) {
289 *advertised |= OFPPF_AUTONEG;
291 if (ecmd.advertising & ADVERTISED_Pause) {
292 *advertised |= OFPPF_PAUSE;
294 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
295 *advertised |= OFPPF_PAUSE_ASYM;
298 /* Set the current features */
299 if (ecmd.speed == SPEED_10) {
300 *current = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
302 else if (ecmd.speed == SPEED_100) {
303 *current = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
305 else if (ecmd.speed == SPEED_1000) {
306 *current = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
308 else if (ecmd.speed == SPEED_10000) {
309 *current = OFPPF_10GB_FD;
312 if (ecmd.port == PORT_TP) {
313 *current |= OFPPF_COPPER;
315 else if (ecmd.port == PORT_FIBRE) {
316 *current |= OFPPF_FIBER;
320 *current |= OFPPF_AUTONEG;
325 /* Opens the network device named 'name' (e.g. "eth0") and returns zero if
326 * successful, otherwise a positive errno value. On success, sets '*netdevp'
327 * to the new network device, otherwise to null.
329 * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
330 * capture frames of that type received on the device. It may also be one of
331 * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those
334 netdev_open(const char *name, int ethertype, struct netdev **netdevp)
336 if (!strncmp(name, "tap:", 4)) {
337 return netdev_open_tap(name + 4, netdevp);
339 return do_open_netdev(name, ethertype, -1, netdevp);
343 /* Opens a TAP virtual network device. If 'name' is a nonnull, non-empty
344 * string, attempts to assign that name to the TAP device (failing if the name
345 * is already in use); otherwise, a name is automatically assigned. Returns
346 * zero if successful, otherwise a positive errno value. On success, sets
347 * '*netdevp' to the new network device, otherwise to null. */
349 netdev_open_tap(const char *name, struct netdev **netdevp)
351 static const char tap_dev[] = "/dev/net/tun";
356 tap_fd = open(tap_dev, O_RDWR);
358 ovs_error(errno, "opening \"%s\" failed", tap_dev);
362 memset(&ifr, 0, sizeof ifr);
363 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
365 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
367 if (ioctl(tap_fd, TUNSETIFF, &ifr) < 0) {
369 ovs_error(error, "ioctl(TUNSETIFF) on \"%s\" failed", tap_dev);
374 error = set_nonblocking(tap_fd);
376 ovs_error(error, "set_nonblocking on \"%s\" failed", tap_dev);
381 error = do_open_netdev(ifr.ifr_name, NETDEV_ETH_TYPE_NONE, tap_fd,
390 do_open_netdev(const char *name, int ethertype, int tap_fd,
391 struct netdev **netdev_)
394 struct sockaddr_ll sll;
397 uint8_t etheraddr[ETH_ADDR_LEN];
403 struct netdev *netdev;
407 COVERAGE_INC(netdev_open);
409 /* Create raw socket. */
410 netdev_fd = socket(PF_PACKET, SOCK_RAW,
411 htons(ethertype == NETDEV_ETH_TYPE_NONE ? 0
412 : ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
413 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
419 if (ethertype != NETDEV_ETH_TYPE_NONE) {
420 /* Set non-blocking mode. */
421 error = set_nonblocking(netdev_fd);
423 goto error_already_set;
426 /* Get ethernet device index. */
427 ifindex = do_get_ifindex(name);
432 /* Bind to specific ethernet device. */
433 memset(&sll, 0, sizeof sll);
434 sll.sll_family = AF_PACKET;
435 sll.sll_ifindex = ifindex;
436 if (bind(netdev_fd, (struct sockaddr *) &sll, sizeof sll) < 0) {
437 VLOG_ERR("bind to %s failed: %s", name, strerror(errno));
441 /* Between the socket() and bind() calls above, the socket receives all
442 * packets of the requested type on all system interfaces. We do not
443 * want to receive that data, but there is no way to avoid it. So we
444 * must now drain out the receive queue. */
445 error = drain_rcvbuf(netdev_fd);
447 goto error_already_set;
451 /* Get MAC address. */
452 error = get_etheraddr(name, etheraddr, &hwaddr_family);
454 goto error_already_set;
458 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
459 if (ioctl(netdev_fd, SIOCGIFMTU, &ifr) < 0) {
460 VLOG_ERR("ioctl(SIOCGIFMTU) on %s device failed: %s",
461 name, strerror(errno));
466 /* Get TX queue length. */
467 if (ioctl(netdev_fd, SIOCGIFTXQLEN, &ifr) < 0) {
468 VLOG_ERR("ioctl(SIOCGIFTXQLEN) on %s device failed: %s",
469 name, strerror(errno));
472 txqlen = ifr.ifr_qlen;
474 get_ipv6_address(name, &in6);
476 /* Allocate network device. */
477 netdev = xmalloc(sizeof *netdev);
478 netdev->name = xstrdup(name);
479 netdev->ifindex = ifindex;
480 netdev->txqlen = txqlen;
481 netdev->hwaddr_family = hwaddr_family;
482 netdev->netdev_fd = netdev_fd;
483 netdev->tap_fd = tap_fd < 0 ? netdev_fd : tap_fd;
484 memcpy(netdev->etheraddr, etheraddr, sizeof etheraddr);
488 /* Save flags to restore at close or exit. */
489 error = get_flags(netdev->name, &netdev->save_flags);
491 goto error_already_set;
493 netdev->changed_flags = 0;
494 fatal_signal_block();
495 list_push_back(&netdev_list, &netdev->node);
496 fatal_signal_unblock();
512 /* Closes and destroys 'netdev'. */
514 netdev_close(struct netdev *netdev)
517 /* Bring down interface and drop promiscuous mode, if we brought up
518 * the interface or enabled promiscuous mode. */
520 fatal_signal_block();
521 error = restore_flags(netdev);
522 list_remove(&netdev->node);
523 fatal_signal_unblock();
525 VLOG_WARN("failed to restore network device flags on %s: %s",
526 netdev->name, strerror(error));
531 close(netdev->netdev_fd);
532 if (netdev->netdev_fd != netdev->tap_fd) {
533 close(netdev->tap_fd);
539 /* Pads 'buffer' out with zero-bytes to the minimum valid length of an
540 * Ethernet packet, if necessary. */
542 pad_to_minimum_length(struct ofpbuf *buffer)
544 if (buffer->size < ETH_TOTAL_MIN) {
545 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
549 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
550 * must have initialized with sufficient room for the packet. The space
551 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
552 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
553 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
554 * need not be included.)
556 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
557 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
558 * positive errno value. Returns EAGAIN immediately if no packet is ready to
562 netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
566 assert(buffer->size == 0);
567 assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
569 n_bytes = read(netdev->tap_fd,
570 ofpbuf_tail(buffer), ofpbuf_tailroom(buffer));
571 } while (n_bytes < 0 && errno == EINTR);
573 if (errno != EAGAIN) {
574 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
575 strerror(errno), netdev->name);
579 COVERAGE_INC(netdev_received);
580 buffer->size += n_bytes;
582 /* When the kernel internally sends out an Ethernet frame on an
583 * interface, it gives us a copy *before* padding the frame to the
584 * minimum length. Thus, when it sends out something like an ARP
585 * request, we see a too-short frame. So pad it out to the minimum
587 pad_to_minimum_length(buffer);
592 /* Registers with the poll loop to wake up from the next call to poll_block()
593 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
595 netdev_recv_wait(struct netdev *netdev)
597 poll_fd_wait(netdev->tap_fd, POLLIN);
600 /* Discards all packets waiting to be received from 'netdev'. */
602 netdev_drain(struct netdev *netdev)
604 if (netdev->tap_fd != netdev->netdev_fd) {
605 drain_fd(netdev->tap_fd, netdev->txqlen);
608 return drain_rcvbuf(netdev->netdev_fd);
612 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
613 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
614 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
615 * the packet is too big or too small to transmit on the device.
617 * The caller retains ownership of 'buffer' in all cases.
619 * The kernel maintains a packet transmission queue, so the caller is not
620 * expected to do additional queuing of packets. */
622 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
627 n_bytes = write(netdev->tap_fd, buffer->data, buffer->size);
628 } while (n_bytes < 0 && errno == EINTR);
631 /* The Linux AF_PACKET implementation never blocks waiting for room
632 * for packets, instead returning ENOBUFS. Translate this into EAGAIN
634 if (errno == ENOBUFS) {
636 } else if (errno != EAGAIN) {
637 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
638 netdev->name, strerror(errno));
641 } else if (n_bytes != buffer->size) {
643 "send partial Ethernet packet (%d bytes of %zu) on %s",
644 (int) n_bytes, buffer->size, netdev->name);
647 COVERAGE_INC(netdev_sent);
652 /* Registers with the poll loop to wake up from the next call to poll_block()
653 * when the packet transmission queue has sufficient room to transmit a packet
654 * with netdev_send().
656 * The kernel maintains a packet transmission queue, so the client is not
657 * expected to do additional queuing of packets. Thus, this function is
658 * unlikely to ever be used. It is included for completeness. */
660 netdev_send_wait(struct netdev *netdev)
662 if (netdev->tap_fd == netdev->netdev_fd) {
663 poll_fd_wait(netdev->tap_fd, POLLOUT);
665 /* TAP device always accepts packets.*/
666 poll_immediate_wake();
670 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
671 * otherwise a positive errno value. */
673 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
675 int error = set_etheraddr(netdev->name, netdev->hwaddr_family, mac);
677 memcpy(netdev->etheraddr, mac, ETH_ADDR_LEN);
683 netdev_nodev_set_etheraddr(const char *name, const uint8_t mac[ETH_ADDR_LEN])
686 return set_etheraddr(name, ARPHRD_ETHER, mac);
689 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
690 * free the returned buffer. */
692 netdev_get_etheraddr(const struct netdev *netdev)
694 return netdev->etheraddr;
697 /* Returns the name of the network device that 'netdev' represents,
698 * e.g. "eth0". The caller must not modify or free the returned string. */
700 netdev_get_name(const struct netdev *netdev)
705 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
706 * in bytes, not including the hardware header; thus, this is typically 1500
707 * bytes for Ethernet devices. */
709 netdev_get_mtu(const struct netdev *netdev)
714 /* Stores the features supported by 'netdev' into each of '*current',
715 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
716 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
717 * successful, otherwise a positive errno value. On failure, all of the
718 * passed-in values are set to 0. */
720 netdev_get_features(struct netdev *netdev,
721 uint32_t *current, uint32_t *advertised,
722 uint32_t *supported, uint32_t *peer)
725 return do_get_features(netdev,
726 current ? current : &dummy[0],
727 advertised ? advertised : &dummy[1],
728 supported ? supported : &dummy[2],
729 peer ? peer : &dummy[3]);
732 /* Set the features advertised by 'netdev' to 'advertise'. */
734 netdev_set_advertisements(struct netdev *netdev, uint32_t advertise)
736 struct ethtool_cmd ecmd;
739 memset(&ecmd, 0, sizeof ecmd);
740 error = do_ethtool(netdev, &ecmd, ETHTOOL_GSET, "ETHTOOL_GSET");
745 ecmd.advertising = 0;
746 if (advertise & OFPPF_10MB_HD) {
747 ecmd.advertising |= ADVERTISED_10baseT_Half;
749 if (advertise & OFPPF_10MB_FD) {
750 ecmd.advertising |= ADVERTISED_10baseT_Full;
752 if (advertise & OFPPF_100MB_HD) {
753 ecmd.advertising |= ADVERTISED_100baseT_Half;
755 if (advertise & OFPPF_100MB_FD) {
756 ecmd.advertising |= ADVERTISED_100baseT_Full;
758 if (advertise & OFPPF_1GB_HD) {
759 ecmd.advertising |= ADVERTISED_1000baseT_Half;
761 if (advertise & OFPPF_1GB_FD) {
762 ecmd.advertising |= ADVERTISED_1000baseT_Full;
764 if (advertise & OFPPF_10GB_FD) {
765 ecmd.advertising |= ADVERTISED_10000baseT_Full;
767 if (advertise & OFPPF_COPPER) {
768 ecmd.advertising |= ADVERTISED_TP;
770 if (advertise & OFPPF_FIBER) {
771 ecmd.advertising |= ADVERTISED_FIBRE;
773 if (advertise & OFPPF_AUTONEG) {
774 ecmd.advertising |= ADVERTISED_Autoneg;
776 if (advertise & OFPPF_PAUSE) {
777 ecmd.advertising |= ADVERTISED_Pause;
779 if (advertise & OFPPF_PAUSE_ASYM) {
780 ecmd.advertising |= ADVERTISED_Asym_Pause;
782 return do_ethtool(netdev, &ecmd, ETHTOOL_SSET, "ETHTOOL_SSET");
785 /* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address (if
786 * 'in4' is non-null) and returns true. Otherwise, returns false. */
788 netdev_get_in4(const struct netdev *netdev, struct in_addr *in4)
791 struct in_addr ip = { INADDR_ANY };
793 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
794 ifr.ifr_addr.sa_family = AF_INET;
795 COVERAGE_INC(netdev_get_in4);
796 if (ioctl(af_inet_sock, SIOCGIFADDR, &ifr) == 0) {
797 struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
800 VLOG_DBG_RL(&rl, "%s: ioctl(SIOCGIFADDR) failed: %s",
801 netdev->name, strerror(errno));
806 return ip.s_addr != INADDR_ANY;
810 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
812 struct sockaddr_in sin;
813 memset(&sin, 0, sizeof sin);
814 sin.sin_family = AF_INET;
818 memset(sa, 0, sizeof *sa);
819 memcpy(sa, &sin, sizeof sin);
823 do_set_addr(struct netdev *netdev, int sock,
824 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
829 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
830 make_in4_sockaddr(&ifr.ifr_addr, addr);
831 COVERAGE_INC(netdev_set_in4);
832 error = ioctl(sock, ioctl_nr, &ifr) < 0 ? errno : 0;
834 VLOG_WARN("ioctl(%s): %s", ioctl_name, strerror(error));
839 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
840 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
841 * positive errno value. */
843 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
847 error = do_set_addr(netdev, af_inet_sock,
848 SIOCSIFADDR, "SIOCSIFADDR", addr);
849 if (!error && addr.s_addr != INADDR_ANY) {
850 error = do_set_addr(netdev, af_inet_sock,
851 SIOCSIFNETMASK, "SIOCSIFNETMASK", mask);
856 /* Adds 'router' as a default IP gateway. */
858 netdev_add_router(struct in_addr router)
860 struct in_addr any = { INADDR_ANY };
864 memset(&rt, 0, sizeof rt);
865 make_in4_sockaddr(&rt.rt_dst, any);
866 make_in4_sockaddr(&rt.rt_gateway, router);
867 make_in4_sockaddr(&rt.rt_genmask, any);
868 rt.rt_flags = RTF_UP | RTF_GATEWAY;
869 COVERAGE_INC(netdev_add_router);
870 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
872 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
877 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
878 * 'in6' is non-null) and returns true. Otherwise, returns false. */
880 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
885 return memcmp(&netdev->in6, &in6addr_any, sizeof netdev->in6) != 0;
888 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
889 * Returns 0 if successful, otherwise a positive errno value. On failure,
890 * stores 0 into '*flagsp'. */
892 netdev_get_flags(const struct netdev *netdev, enum netdev_flags *flagsp)
894 return netdev_nodev_get_flags(netdev->name, flagsp);
898 nd_to_iff_flags(enum netdev_flags nd)
901 if (nd & NETDEV_UP) {
904 if (nd & NETDEV_PROMISC) {
910 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
911 * 'on'. If 'permanent' is true, the changes will persist; otherwise, they
912 * will be reverted when 'netdev' is closed or the program exits. Returns 0 if
913 * successful, otherwise a positive errno value. */
915 do_update_flags(struct netdev *netdev, enum netdev_flags off,
916 enum netdev_flags on, bool permanent)
918 int old_flags, new_flags;
921 error = get_flags(netdev->name, &old_flags);
926 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
928 netdev->changed_flags |= new_flags ^ old_flags;
930 if (new_flags != old_flags) {
931 error = set_flags(netdev->name, new_flags);
936 /* Sets the flags for 'netdev' to 'flags'.
937 * If 'permanent' is true, the changes will persist; otherwise, they
938 * will be reverted when 'netdev' is closed or the program exits.
939 * Returns 0 if successful, otherwise a positive errno value. */
941 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
944 return do_update_flags(netdev, -1, flags, permanent);
947 /* Turns on the specified 'flags' on 'netdev'.
948 * If 'permanent' is true, the changes will persist; otherwise, they
949 * will be reverted when 'netdev' is closed or the program exits.
950 * Returns 0 if successful, otherwise a positive errno value. */
952 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
955 return do_update_flags(netdev, 0, flags, permanent);
958 /* Turns off the specified 'flags' on 'netdev'.
959 * If 'permanent' is true, the changes will persist; otherwise, they
960 * will be reverted when 'netdev' is closed or the program exits.
961 * Returns 0 if successful, otherwise a positive errno value. */
963 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
966 return do_update_flags(netdev, flags, 0, permanent);
969 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
970 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
971 * returns 0. Otherwise, it returns a positive errno value; in particular,
972 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
974 netdev_arp_lookup(const struct netdev *netdev,
975 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
978 struct sockaddr_in *pa;
981 memset(&r, 0, sizeof r);
982 pa = (struct sockaddr_in *) &r.arp_pa;
983 pa->sin_family = AF_INET;
984 pa->sin_addr.s_addr = ip;
986 r.arp_ha.sa_family = ARPHRD_ETHER;
988 strncpy(r.arp_dev, netdev->name, sizeof r.arp_dev);
989 COVERAGE_INC(netdev_arp_lookup);
990 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
992 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
993 } else if (retval != ENXIO) {
994 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
995 netdev->name, IP_ARGS(&ip), strerror(retval));
1001 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
1003 struct ofpbuf request;
1004 struct ofpbuf *reply;
1005 struct ifinfomsg *ifi;
1006 const struct rtnl_link_stats *rtnl_stats;
1007 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
1010 ofpbuf_init(&request, 0);
1011 nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *ifi,
1012 RTM_GETLINK, NLM_F_REQUEST);
1013 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
1014 ifi->ifi_family = PF_UNSPEC;
1015 ifi->ifi_index = ifindex;
1016 error = nl_sock_transact(rtnl_sock, &request, &reply);
1017 ofpbuf_uninit(&request);
1022 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
1023 rtnlgrp_link_policy,
1024 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
1025 ofpbuf_delete(reply);
1029 if (!attrs[IFLA_STATS]) {
1030 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
1034 rtnl_stats = nl_attr_get(attrs[IFLA_STATS]);
1035 stats->rx_packets = rtnl_stats->rx_packets;
1036 stats->tx_packets = rtnl_stats->tx_packets;
1037 stats->rx_bytes = rtnl_stats->rx_bytes;
1038 stats->tx_bytes = rtnl_stats->tx_bytes;
1039 stats->rx_errors = rtnl_stats->rx_errors;
1040 stats->tx_errors = rtnl_stats->tx_errors;
1041 stats->rx_dropped = rtnl_stats->rx_dropped;
1042 stats->tx_dropped = rtnl_stats->tx_dropped;
1043 stats->multicast = rtnl_stats->multicast;
1044 stats->collisions = rtnl_stats->collisions;
1045 stats->rx_length_errors = rtnl_stats->rx_length_errors;
1046 stats->rx_over_errors = rtnl_stats->rx_over_errors;
1047 stats->rx_crc_errors = rtnl_stats->rx_crc_errors;
1048 stats->rx_frame_errors = rtnl_stats->rx_frame_errors;
1049 stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors;
1050 stats->rx_missed_errors = rtnl_stats->rx_missed_errors;
1051 stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors;
1052 stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors;
1053 stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors;
1054 stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors;
1055 stats->tx_window_errors = rtnl_stats->tx_window_errors;
1061 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
1063 static const char fn[] = "/proc/net/dev";
1068 stream = fopen(fn, "r");
1070 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1075 while (fgets(line, sizeof line, stream)) {
1078 #define X64 "%"SCNu64
1081 X64 X64 X64 X64 X64 X64 X64 "%*u"
1082 X64 X64 X64 X64 X64 X64 X64 "%*u",
1088 &stats->rx_fifo_errors,
1089 &stats->rx_frame_errors,
1095 &stats->tx_fifo_errors,
1097 &stats->tx_carrier_errors) != 15) {
1098 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
1099 } else if (!strcmp(devname, netdev_name)) {
1100 stats->rx_length_errors = UINT64_MAX;
1101 stats->rx_over_errors = UINT64_MAX;
1102 stats->rx_crc_errors = UINT64_MAX;
1103 stats->rx_missed_errors = UINT64_MAX;
1104 stats->tx_aborted_errors = UINT64_MAX;
1105 stats->tx_heartbeat_errors = UINT64_MAX;
1106 stats->tx_window_errors = UINT64_MAX;
1112 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
1117 /* Sets 'carrier' to true if carrier is active (link light is on) on
1120 netdev_get_carrier(const struct netdev *netdev, bool *carrier)
1122 return netdev_nodev_get_carrier(netdev->name, carrier);
1126 netdev_nodev_get_carrier(const char *netdev_name, bool *carrier)
1136 fn = xasprintf("/sys/class/net/%s/carrier", netdev_name);
1137 fd = open(fn, O_RDONLY);
1140 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error));
1144 retval = read(fd, line, sizeof line);
1147 if (error == EINVAL) {
1148 /* This is the normal return value when we try to check carrier if
1149 * the network device is not up. */
1151 VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error));
1154 } else if (retval == 0) {
1156 VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn);
1160 if (line[0] != '0' && line[0] != '1') {
1162 VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)", fn, line[0]);
1165 *carrier = line[0] != '0';
1175 /* Retrieves current device stats for 'netdev'. */
1177 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
1181 COVERAGE_INC(netdev_get_stats);
1182 if (use_netlink_stats) {
1185 error = get_ifindex(netdev, &ifindex);
1187 error = get_stats_via_netlink(ifindex, stats);
1190 error = get_stats_via_proc(netdev->name, stats);
1194 memset(stats, 0xff, sizeof *stats);
1199 #define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress"
1200 #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"
1201 /* We redirect stderr to /dev/null because we often want to remove all
1202 * traffic control configuration on a port so its in a known state. If
1203 * this done when there is no such configuration, tc complains, so we just
1206 #define POLICE_DEL_CMD "/sbin/tc qdisc del dev %s handle ffff: ingress 2>/dev/null"
1208 /* Attempts to set input rate limiting (policing) policy. */
1210 netdev_nodev_set_policing(const char *netdev_name, uint32_t kbits_rate,
1211 uint32_t kbits_burst)
1217 COVERAGE_INC(netdev_set_policing);
1220 /* Default to 10 kilobits if not specified. */
1224 /* xxx This should be more careful about only adding if it
1225 * xxx actually exists, as opposed to always deleting it. */
1226 snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1227 if (system(command) == -1) {
1228 VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1231 snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name);
1232 if (system(command) != 0) {
1233 VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name);
1237 snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name,
1238 kbits_rate, kbits_burst);
1239 if (system(command) != 0) {
1240 VLOG_WARN_RL(&rl, "%s: problem configuring policing",
1245 snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1246 if (system(command) == -1) {
1247 VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1255 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
1256 uint32_t kbits_burst)
1258 return netdev_nodev_set_policing(netdev->name, kbits_rate, kbits_burst);
1261 /* Initializes 'svec' with a list of the names of all known network devices. */
1263 netdev_enumerate(struct svec *svec)
1265 struct if_nameindex *names;
1268 names = if_nameindex();
1272 for (i = 0; names[i].if_name != NULL; i++) {
1273 svec_add(svec, names[i].if_name);
1275 if_freenameindex(names);
1277 VLOG_WARN("could not obtain list of network device names: %s",
1282 /* Obtains the current flags for the network device named 'netdev_name' and
1283 * stores them into '*flagsp'. Returns 0 if successful, otherwise a positive
1284 * errno value. On error, stores 0 into '*flagsp'.
1286 * If only device flags are needed, this is more efficient than calling
1287 * netdev_open(), netdev_get_flags(), netdev_close(). */
1289 netdev_nodev_get_flags(const char *netdev_name, enum netdev_flags *flagsp)
1296 error = get_flags(netdev_name, &flags);
1301 if (flags & IFF_UP) {
1302 *flagsp |= NETDEV_UP;
1304 if (flags & IFF_PROMISC) {
1305 *flagsp |= NETDEV_PROMISC;
1311 netdev_nodev_get_etheraddr(const char *netdev_name, uint8_t mac[6])
1315 return get_etheraddr(netdev_name, mac, NULL);
1318 /* If 'netdev_name' is the name of a VLAN network device (e.g. one created with
1319 * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device
1320 * and returns 0. Otherwise returns a errno value (specifically ENOENT if
1321 * 'netdev_name' is the name of a network device that is not a VLAN device) and
1322 * sets '*vlan_vid' to -1. */
1324 netdev_get_vlan_vid(const char *netdev_name, int *vlan_vid)
1326 struct ds line = DS_EMPTY_INITIALIZER;
1327 FILE *stream = NULL;
1331 COVERAGE_INC(netdev_get_vlan_vid);
1332 fn = xasprintf("/proc/net/vlan/%s", netdev_name);
1333 stream = fopen(fn, "r");
1339 if (ds_get_line(&line, stream)) {
1340 if (ferror(stream)) {
1342 VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno));
1345 VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn);
1350 if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) {
1352 VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"",
1353 fn, ds_cstr(&line));
1371 static void restore_all_flags(void *aux);
1373 /* Set up a signal hook to restore network device flags on program
1385 fatal_signal_add_hook(restore_all_flags, NULL, true);
1387 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
1388 if (af_inet_sock < 0) {
1389 ovs_fatal(errno, "socket(AF_INET)");
1392 error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock);
1394 ovs_fatal(error, "socket(AF_NETLINK, NETLINK_ROUTE)");
1397 /* Decide on the netdev_get_stats() implementation to use. Netlink is
1398 * preferable, so if that works, we'll use it. */
1399 ifindex = do_get_ifindex("lo");
1401 VLOG_WARN("failed to get ifindex for lo, "
1402 "obtaining netdev stats from proc");
1403 use_netlink_stats = false;
1405 struct netdev_stats stats;
1406 error = get_stats_via_netlink(ifindex, &stats);
1408 VLOG_DBG("obtaining netdev stats via rtnetlink");
1409 use_netlink_stats = true;
1411 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
1412 "via proc (you are probably running a pre-2.6.19 "
1413 "kernel)", strerror(error));
1414 use_netlink_stats = false;
1420 /* Restore the network device flags on 'netdev' to those that were active
1421 * before we changed them. Returns 0 if successful, otherwise a positive
1424 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
1426 restore_flags(struct netdev *netdev)
1431 /* Get current flags. */
1432 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
1433 COVERAGE_INC(netdev_get_flags);
1434 if (ioctl(netdev->netdev_fd, SIOCGIFFLAGS, &ifr) < 0) {
1438 /* Restore flags that we might have changed, if necessary. */
1439 restore_flags = netdev->changed_flags & (IFF_PROMISC | IFF_UP);
1440 if ((ifr.ifr_flags ^ netdev->save_flags) & restore_flags) {
1441 ifr.ifr_flags &= ~restore_flags;
1442 ifr.ifr_flags |= netdev->save_flags & restore_flags;
1443 COVERAGE_INC(netdev_set_flags);
1444 if (ioctl(netdev->netdev_fd, SIOCSIFFLAGS, &ifr) < 0) {
1452 /* Retores all the flags on all network devices that we modified. Called from
1453 * a signal handler, so it does not attempt to report error conditions. */
1455 restore_all_flags(void *aux UNUSED)
1457 struct netdev *netdev;
1458 LIST_FOR_EACH (netdev, struct netdev, node, &netdev_list) {
1459 restore_flags(netdev);
1464 get_flags(const char *netdev_name, int *flags)
1467 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
1468 COVERAGE_INC(netdev_get_flags);
1469 if (ioctl(af_inet_sock, SIOCGIFFLAGS, &ifr) < 0) {
1470 VLOG_ERR("ioctl(SIOCGIFFLAGS) on %s device failed: %s",
1471 netdev_name, strerror(errno));
1474 *flags = ifr.ifr_flags;
1479 set_flags(const char *netdev_name, int flags)
1482 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
1483 ifr.ifr_flags = flags;
1484 COVERAGE_INC(netdev_set_flags);
1485 if (ioctl(af_inet_sock, SIOCSIFFLAGS, &ifr) < 0) {
1486 VLOG_ERR("ioctl(SIOCSIFFLAGS) on %s device failed: %s",
1487 netdev_name, strerror(errno));
1494 do_get_ifindex(const char *netdev_name)
1498 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
1499 COVERAGE_INC(netdev_get_ifindex);
1500 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
1501 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
1502 netdev_name, strerror(errno));
1505 return ifr.ifr_ifindex;
1509 get_ifindex(const struct netdev *netdev, int *ifindexp)
1512 if (netdev->ifindex < 0) {
1513 int ifindex = do_get_ifindex(netdev->name);
1517 ((struct netdev *) netdev)->ifindex = ifindex;
1519 *ifindexp = netdev->ifindex;
1524 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN],
1525 int *hwaddr_familyp)
1529 memset(&ifr, 0, sizeof ifr);
1530 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
1531 COVERAGE_INC(netdev_get_hwaddr);
1532 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
1533 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
1534 netdev_name, strerror(errno));
1537 if (hwaddr_familyp) {
1538 int hwaddr_family = ifr.ifr_hwaddr.sa_family;
1539 *hwaddr_familyp = hwaddr_family;
1540 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
1541 VLOG_WARN("%s device has unknown hardware address family %d",
1542 netdev_name, hwaddr_family);
1545 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
1550 set_etheraddr(const char *netdev_name, int hwaddr_family,
1551 const uint8_t mac[ETH_ADDR_LEN])
1555 memset(&ifr, 0, sizeof ifr);
1556 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
1557 ifr.ifr_hwaddr.sa_family = hwaddr_family;
1558 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
1559 COVERAGE_INC(netdev_set_hwaddr);
1560 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
1561 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
1562 netdev_name, strerror(errno));