1 /* Copyright (c) 2008 The Board of Trustees of The Leland Stanford
4 * We are making the OpenFlow specification and associated documentation
5 * (Software) available for public use and benefit with the expectation
6 * that others will use, modify and enhance the Software and contribute
7 * those enhancements back to the community. However, since we would
8 * like to make the Software available for broadest use, with as few
9 * restrictions as possible permission is hereby granted, free of
10 * charge, to any person obtaining a copy of this Software to deal in
11 * the Software under the copyrights without restriction, including
12 * without limitation the rights to use, copy, modify, merge, publish,
13 * distribute, sublicense, and/or sell copies of the Software, and to
14 * permit persons to whom the Software is furnished to do so, subject to
15 * the following conditions:
17 * The above copyright notice and this permission notice shall be
18 * included in all copies or substantial portions of the Software.
20 * THE SOFTWRE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
21 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
23 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
24 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
25 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
26 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
29 * The name and trademarks of copyright holder(s) may NOT be used in
30 * advertising or publicity pertaining to the Software or any
31 * derivatives without specific, written prior permission.
40 #include <arpa/inet.h>
42 #include <linux/if_tun.h>
43 #include <linux/types.h>
44 #include <linux/ethtool.h>
45 #include <linux/sockios.h>
46 #include <linux/version.h>
47 #include <sys/types.h>
48 #include <sys/ioctl.h>
49 #include <sys/socket.h>
50 #include <netpacket/packet.h>
51 #include <net/ethernet.h>
53 #include <net/if_arp.h>
54 #include <net/if_packet.h>
55 #include <net/route.h>
56 #include <netinet/in.h>
61 #include "fatal-signal.h"
66 #include "poll-loop.h"
67 #include "socket-util.h"
69 #define THIS_MODULE VLM_netdev
76 /* File descriptors. For ordinary network devices, the two fds below are
77 * the same; for tap devices, they differ. */
78 int netdev_fd; /* Network device. */
79 int tap_fd; /* TAP character device, if any, otherwise the
82 /* Cached network device information. */
84 uint8_t etheraddr[ETH_ADDR_LEN];
90 /* Bitmaps of OFPPF_* that describe features. All bits disabled if
91 * unsupported or unavailable. */
92 uint32_t curr; /* Current features. */
93 uint32_t advertised; /* Features being advertised by the port. */
94 uint32_t supported; /* Features supported by the port. */
95 uint32_t peer; /* Features advertised by the peer. */
97 int save_flags; /* Initial device flags. */
98 int changed_flags; /* Flags that we changed. */
101 /* All open network devices. */
102 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
104 /* An AF_INET socket (used for ioctl operations). */
105 static int af_inet_sock = -1;
107 /* This is set pretty low because we probably won't learn anything from the
108 * additional log messages. */
109 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
111 static void init_netdev(void);
112 static int do_open_netdev(const char *name, int ethertype, int tap_fd,
113 struct netdev **netdev_);
114 static int restore_flags(struct netdev *netdev);
115 static int get_flags(const struct netdev *, int fd, int *flagsp);
116 static int set_flags(struct netdev *, int fd, int flags);
118 /* Obtains the IPv6 address for 'name' into 'in6'. */
120 get_ipv6_address(const char *name, struct in6_addr *in6)
125 file = fopen("/proc/net/if_inet6", "r");
127 /* This most likely indicates that the host doesn't have IPv6 support,
128 * so it's not really a failure condition.*/
133 while (fgets(line, sizeof line, file)) {
134 uint8_t *s6 = in6->s6_addr;
138 if (sscanf(line, " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
139 "%*x %*x %*x %*x %16s\n",
140 &s6[0], &s6[1], &s6[2], &s6[3],
141 &s6[4], &s6[5], &s6[6], &s6[7],
142 &s6[8], &s6[9], &s6[10], &s6[11],
143 &s6[12], &s6[13], &s6[14], &s6[15],
145 && !strcmp(name, ifname))
156 do_ethtool(struct netdev *netdev)
159 struct ethtool_cmd ecmd;
162 netdev->supported = 0;
163 netdev->advertised = 0;
166 memset(&ifr, 0, sizeof ifr);
167 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
168 ifr.ifr_data = (caddr_t) &ecmd;
170 memset(&ecmd, 0, sizeof ecmd);
171 ecmd.cmd = ETHTOOL_GSET;
172 if (ioctl(netdev->netdev_fd, SIOCETHTOOL, &ifr) == 0) {
173 if (ecmd.supported & SUPPORTED_10baseT_Half) {
174 netdev->supported |= OFPPF_10MB_HD;
176 if (ecmd.supported & SUPPORTED_10baseT_Full) {
177 netdev->supported |= OFPPF_10MB_FD;
179 if (ecmd.supported & SUPPORTED_100baseT_Half) {
180 netdev->supported |= OFPPF_100MB_HD;
182 if (ecmd.supported & SUPPORTED_100baseT_Full) {
183 netdev->supported |= OFPPF_100MB_FD;
185 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
186 netdev->supported |= OFPPF_1GB_HD;
188 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
189 netdev->supported |= OFPPF_1GB_FD;
191 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
192 netdev->supported |= OFPPF_10GB_FD;
194 if (ecmd.supported & SUPPORTED_TP) {
195 netdev->supported |= OFPPF_COPPER;
197 if (ecmd.supported & SUPPORTED_FIBRE) {
198 netdev->supported |= OFPPF_FIBER;
200 if (ecmd.supported & SUPPORTED_Autoneg) {
201 netdev->supported |= OFPPF_AUTONEG;
203 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
204 if (ecmd.supported & SUPPORTED_Pause) {
205 netdev->supported |= OFPPF_PAUSE;
207 if (ecmd.supported & SUPPORTED_Asym_Pause) {
208 netdev->supported |= OFPPF_PAUSE_ASYM;
210 #endif /* kernel >= 2.6.14 */
212 /* Set the advertised features */
213 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
214 netdev->advertised |= OFPPF_10MB_HD;
216 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
217 netdev->advertised |= OFPPF_10MB_FD;
219 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
220 netdev->advertised |= OFPPF_100MB_HD;
222 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
223 netdev->advertised |= OFPPF_100MB_FD;
225 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
226 netdev->advertised |= OFPPF_1GB_HD;
228 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
229 netdev->advertised |= OFPPF_1GB_FD;
231 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
232 netdev->advertised |= OFPPF_10GB_FD;
234 if (ecmd.advertising & ADVERTISED_TP) {
235 netdev->advertised |= OFPPF_COPPER;
237 if (ecmd.advertising & ADVERTISED_FIBRE) {
238 netdev->advertised |= OFPPF_FIBER;
240 if (ecmd.advertising & ADVERTISED_Autoneg) {
241 netdev->advertised |= OFPPF_AUTONEG;
243 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
244 if (ecmd.advertising & ADVERTISED_Pause) {
245 netdev->advertised |= OFPPF_PAUSE;
247 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
248 netdev->advertised |= OFPPF_PAUSE_ASYM;
250 #endif /* kernel >= 2.6.14 */
252 /* Set the current features */
253 if (ecmd.speed == SPEED_10) {
254 netdev->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
256 else if (ecmd.speed == SPEED_100) {
257 netdev->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
259 else if (ecmd.speed == SPEED_1000) {
260 netdev->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
262 else if (ecmd.speed == SPEED_10000) {
263 netdev->curr = OFPPF_10GB_FD;
266 if (ecmd.port == PORT_TP) {
267 netdev->curr |= OFPPF_COPPER;
269 else if (ecmd.port == PORT_FIBRE) {
270 netdev->curr |= OFPPF_FIBER;
274 netdev->curr |= OFPPF_AUTONEG;
277 VLOG_DBG("ioctl(SIOCETHTOOL) failed: %s", strerror(errno));
281 /* Opens the network device named 'name' (e.g. "eth0") and returns zero if
282 * successful, otherwise a positive errno value. On success, sets '*netdevp'
283 * to the new network device, otherwise to null.
285 * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
286 * capture frames of that type received on the device. It may also be one of
287 * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those
290 netdev_open(const char *name, int ethertype, struct netdev **netdevp)
292 if (!strncmp(name, "tap:", 4)) {
293 return netdev_open_tap(name + 4, netdevp);
295 return do_open_netdev(name, ethertype, -1, netdevp);
299 /* Opens a TAP virtual network device. If 'name' is a nonnull, non-empty
300 * string, attempts to assign that name to the TAP device (failing if the name
301 * is already in use); otherwise, a name is automatically assigned. Returns
302 * zero if successful, otherwise a positive errno value. On success, sets
303 * '*netdevp' to the new network device, otherwise to null. */
305 netdev_open_tap(const char *name, struct netdev **netdevp)
307 static const char tap_dev[] = "/dev/net/tun";
312 tap_fd = open(tap_dev, O_RDWR);
314 ofp_error(errno, "opening \"%s\" failed", tap_dev);
318 memset(&ifr, 0, sizeof ifr);
319 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
321 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
323 if (ioctl(tap_fd, TUNSETIFF, &ifr) < 0) {
325 ofp_error(error, "ioctl(TUNSETIFF) on \"%s\" failed", tap_dev);
330 error = set_nonblocking(tap_fd);
332 ofp_error(error, "set_nonblocking on \"%s\" failed", tap_dev);
337 error = do_open_netdev(ifr.ifr_name, NETDEV_ETH_TYPE_NONE, tap_fd,
346 do_open_netdev(const char *name, int ethertype, int tap_fd,
347 struct netdev **netdev_)
350 struct sockaddr_ll sll;
352 unsigned int ifindex;
353 uint8_t etheraddr[ETH_ADDR_LEN];
358 struct netdev *netdev;
363 /* Create raw socket. */
364 netdev_fd = socket(PF_PACKET, SOCK_RAW,
365 htons(ethertype == NETDEV_ETH_TYPE_NONE ? 0
366 : ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
367 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
373 /* Set non-blocking mode. */
374 error = set_nonblocking(netdev_fd);
376 goto error_already_set;
379 /* Get ethernet device index. */
380 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
381 if (ioctl(netdev_fd, SIOCGIFINDEX, &ifr) < 0) {
382 VLOG_ERR("ioctl(SIOCGIFINDEX) on %s device failed: %s",
383 name, strerror(errno));
386 ifindex = ifr.ifr_ifindex;
388 /* Bind to specific ethernet device. */
389 memset(&sll, 0, sizeof sll);
390 sll.sll_family = AF_PACKET;
391 sll.sll_ifindex = ifindex;
392 if (bind(netdev_fd, (struct sockaddr *) &sll, sizeof sll) < 0) {
393 VLOG_ERR("bind to %s failed: %s", name, strerror(errno));
397 if (ethertype != NETDEV_ETH_TYPE_NONE) {
398 /* Between the socket() and bind() calls above, the socket receives all
399 * packets of the requested type on all system interfaces. We do not
400 * want to receive that data, but there is no way to avoid it. So we
401 * must now drain out the receive queue. */
402 error = drain_rcvbuf(netdev_fd);
408 /* Get MAC address. */
409 if (ioctl(netdev_fd, SIOCGIFHWADDR, &ifr) < 0) {
410 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
411 name, strerror(errno));
414 if (ifr.ifr_hwaddr.sa_family != AF_UNSPEC
415 && ifr.ifr_hwaddr.sa_family != ARPHRD_ETHER) {
416 VLOG_WARN("%s device has unknown hardware address family %d",
417 name, (int) ifr.ifr_hwaddr.sa_family);
419 memcpy(etheraddr, ifr.ifr_hwaddr.sa_data, sizeof etheraddr);
422 if (ioctl(netdev_fd, SIOCGIFMTU, &ifr) < 0) {
423 VLOG_ERR("ioctl(SIOCGIFMTU) on %s device failed: %s",
424 name, strerror(errno));
429 /* Get TX queue length. */
430 if (ioctl(netdev_fd, SIOCGIFTXQLEN, &ifr) < 0) {
431 VLOG_ERR("ioctl(SIOCGIFTXQLEN) on %s device failed: %s",
432 name, strerror(errno));
435 txqlen = ifr.ifr_qlen;
437 get_ipv6_address(name, &in6);
439 /* Allocate network device. */
440 netdev = xmalloc(sizeof *netdev);
441 netdev->name = xstrdup(name);
442 netdev->ifindex = ifindex;
443 netdev->txqlen = txqlen;
444 netdev->netdev_fd = netdev_fd;
445 netdev->tap_fd = tap_fd < 0 ? netdev_fd : tap_fd;
446 memcpy(netdev->etheraddr, etheraddr, sizeof etheraddr);
450 /* Get speed, features. */
453 /* Save flags to restore at close or exit. */
454 error = get_flags(netdev, netdev_fd, &netdev->save_flags);
456 goto error_already_set;
458 netdev->changed_flags = 0;
459 fatal_signal_block();
460 list_push_back(&netdev_list, &netdev->node);
461 fatal_signal_unblock();
477 /* Closes and destroys 'netdev'. */
479 netdev_close(struct netdev *netdev)
482 /* Bring down interface and drop promiscuous mode, if we brought up
483 * the interface or enabled promiscuous mode. */
485 fatal_signal_block();
486 error = restore_flags(netdev);
487 list_remove(&netdev->node);
488 fatal_signal_unblock();
490 VLOG_WARN("failed to restore network device flags on %s: %s",
491 netdev->name, strerror(error));
496 close(netdev->netdev_fd);
497 if (netdev->netdev_fd != netdev->tap_fd) {
498 close(netdev->tap_fd);
504 /* Pads 'buffer' out with zero-bytes to the minimum valid length of an
505 * Ethernet packet, if necessary. */
507 pad_to_minimum_length(struct ofpbuf *buffer)
509 if (buffer->size < ETH_TOTAL_MIN) {
510 size_t shortage = ETH_TOTAL_MIN - buffer->size;
511 memset(ofpbuf_put_uninit(buffer, shortage), 0, shortage);
515 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
516 * must have initialized with sufficient room for the packet. The space
517 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
518 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
519 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
520 * need not be included.)
522 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
523 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
524 * positive errno value. Returns EAGAIN immediately if no packet is ready to
528 netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
532 assert(buffer->size == 0);
533 assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
535 n_bytes = read(netdev->tap_fd,
536 ofpbuf_tail(buffer), ofpbuf_tailroom(buffer));
537 } while (n_bytes < 0 && errno == EINTR);
539 if (errno != EAGAIN) {
540 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
541 strerror(errno), netdev->name);
545 buffer->size += n_bytes;
547 /* When the kernel internally sends out an Ethernet frame on an
548 * interface, it gives us a copy *before* padding the frame to the
549 * minimum length. Thus, when it sends out something like an ARP
550 * request, we see a too-short frame. So pad it out to the minimum
552 pad_to_minimum_length(buffer);
557 /* Registers with the poll loop to wake up from the next call to poll_block()
558 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
560 netdev_recv_wait(struct netdev *netdev)
562 poll_fd_wait(netdev->tap_fd, POLLIN);
565 /* Discards all packets waiting to be received from 'netdev'. */
567 netdev_drain(struct netdev *netdev)
569 if (netdev->tap_fd != netdev->netdev_fd) {
570 drain_fd(netdev->tap_fd, netdev->txqlen);
573 return drain_rcvbuf(netdev->netdev_fd);
577 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
578 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
579 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
580 * the packet is too big or too small to transmit on the device.
582 * The caller retains ownership of 'buffer' in all cases.
584 * The kernel maintains a packet transmission queue, so the caller is not
585 * expected to do additional queuing of packets. */
587 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
592 n_bytes = write(netdev->tap_fd, buffer->data, buffer->size);
593 } while (n_bytes < 0 && errno == EINTR);
596 /* The Linux AF_PACKET implementation never blocks waiting for room
597 * for packets, instead returning ENOBUFS. Translate this into EAGAIN
599 if (errno == ENOBUFS) {
601 } else if (errno != EAGAIN) {
602 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
603 netdev->name, strerror(errno));
606 } else if (n_bytes != buffer->size) {
608 "send partial Ethernet packet (%d bytes of %zu) on %s",
609 (int) n_bytes, buffer->size, netdev->name);
616 /* Registers with the poll loop to wake up from the next call to poll_block()
617 * when the packet transmission queue has sufficient room to transmit a packet
618 * with netdev_send().
620 * The kernel maintains a packet transmission queue, so the client is not
621 * expected to do additional queuing of packets. Thus, this function is
622 * unlikely to ever be used. It is included for completeness. */
624 netdev_send_wait(struct netdev *netdev)
626 if (netdev->tap_fd == netdev->netdev_fd) {
627 poll_fd_wait(netdev->tap_fd, POLLOUT);
629 /* TAP device always accepts packets.*/
630 poll_immediate_wake();
634 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
635 * free the returned buffer. */
637 netdev_get_etheraddr(const struct netdev *netdev)
639 return netdev->etheraddr;
642 /* Returns the name of the network device that 'netdev' represents,
643 * e.g. "eth0". The caller must not modify or free the returned string. */
645 netdev_get_name(const struct netdev *netdev)
650 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
651 * in bytes, not including the hardware header; thus, this is typically 1500
652 * bytes for Ethernet devices. */
654 netdev_get_mtu(const struct netdev *netdev)
659 /* Checks the link status. Returns 1 or 0 to indicate the link is active
660 * or not, respectively. Any other return value indicates an error. */
662 netdev_get_link_status(const struct netdev *netdev)
665 struct ethtool_value edata;
667 memset(&ifr, 0, sizeof ifr);
668 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
669 ifr.ifr_data = (caddr_t) &edata;
671 memset(&edata, 0, sizeof edata);
672 edata.cmd = ETHTOOL_GLINK;
673 if (ioctl(netdev->netdev_fd, SIOCETHTOOL, &ifr) == 0) {
684 /* Returns the features supported by 'netdev' of type 'type', as a bitmap
685 * of bits from enum ofp_phy_features, in host byte order. */
687 netdev_get_features(struct netdev *netdev, int type)
691 case NETDEV_FEAT_CURRENT:
693 case NETDEV_FEAT_ADVERTISED:
694 return netdev->advertised;
695 case NETDEV_FEAT_SUPPORTED:
696 return netdev->supported;
697 case NETDEV_FEAT_PEER:
700 VLOG_WARN("Unknown feature type: %d\n", type);
705 /* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address (if
706 * 'in4' is non-null) and returns true. Otherwise, returns false. */
708 netdev_get_in4(const struct netdev *netdev, struct in_addr *in4)
711 struct in_addr ip = { INADDR_ANY };
713 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
714 ifr.ifr_addr.sa_family = AF_INET;
715 if (ioctl(af_inet_sock, SIOCGIFADDR, &ifr) == 0) {
716 struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
719 VLOG_DBG_RL(&rl, "%s: ioctl(SIOCGIFADDR) failed: %s",
720 netdev->name, strerror(errno));
725 return ip.s_addr != INADDR_ANY;
729 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
731 struct sockaddr_in sin;
732 memset(&sin, 0, sizeof sin);
733 sin.sin_family = AF_INET;
737 memset(sa, 0, sizeof *sa);
738 memcpy(sa, &sin, sizeof sin);
742 do_set_addr(struct netdev *netdev, int sock,
743 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
748 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
749 make_in4_sockaddr(&ifr.ifr_addr, addr);
750 error = ioctl(sock, ioctl_nr, &ifr) < 0 ? errno : 0;
752 VLOG_WARN("ioctl(%s): %s", ioctl_name, strerror(error));
757 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
758 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
759 * positive errno value. */
761 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
765 error = do_set_addr(netdev, af_inet_sock,
766 SIOCSIFADDR, "SIOCSIFADDR", addr);
767 if (!error && addr.s_addr != INADDR_ANY) {
768 error = do_set_addr(netdev, af_inet_sock,
769 SIOCSIFNETMASK, "SIOCSIFNETMASK", mask);
774 /* Adds 'router' as a default gateway for 'netdev''s IP address. */
776 netdev_add_router(struct netdev *netdev, struct in_addr router)
778 struct in_addr any = { INADDR_ANY };
782 memset(&rt, 0, sizeof rt);
783 make_in4_sockaddr(&rt.rt_dst, any);
784 make_in4_sockaddr(&rt.rt_gateway, router);
785 make_in4_sockaddr(&rt.rt_genmask, any);
786 rt.rt_flags = RTF_UP | RTF_GATEWAY;
787 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
789 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
794 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
795 * 'in6' is non-null) and returns true. Otherwise, returns false. */
797 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
802 return memcmp(&netdev->in6, &in6addr_any, sizeof netdev->in6) != 0;
805 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
806 * Returns 0 if successful, otherwise a positive errno value. */
808 netdev_get_flags(const struct netdev *netdev, enum netdev_flags *flagsp)
812 error = get_flags(netdev, netdev->netdev_fd, &flags);
818 if (flags & IFF_UP) {
819 *flagsp |= NETDEV_UP;
821 if (flags & IFF_PROMISC) {
822 *flagsp |= NETDEV_PROMISC;
828 nd_to_iff_flags(enum netdev_flags nd)
831 if (nd & NETDEV_UP) {
834 if (nd & NETDEV_PROMISC) {
840 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
841 * 'on'. If 'permanent' is true, the changes will persist; otherwise, they
842 * will be reverted when 'netdev' is closed or the program exits. Returns 0 if
843 * successful, otherwise a positive errno value. */
845 do_update_flags(struct netdev *netdev, int fd, enum netdev_flags off,
846 enum netdev_flags on, bool permanent)
848 int old_flags, new_flags;
851 error = get_flags(netdev, fd, &old_flags);
856 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
858 netdev->changed_flags |= new_flags ^ old_flags;
860 if (new_flags != old_flags) {
861 error = set_flags(netdev, fd, new_flags);
866 /* Sets the flags for 'netdev' to 'flags'.
867 * If 'permanent' is true, the changes will persist; otherwise, they
868 * will be reverted when 'netdev' is closed or the program exits.
869 * Returns 0 if successful, otherwise a positive errno value. */
871 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
874 return do_update_flags(netdev, netdev->netdev_fd, -1, flags, permanent);
877 /* Turns on the specified 'flags' on 'netdev'.
878 * If 'permanent' is true, the changes will persist; otherwise, they
879 * will be reverted when 'netdev' is closed or the program exits.
880 * Returns 0 if successful, otherwise a positive errno value. */
882 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
885 return do_update_flags(netdev, netdev->netdev_fd, 0, flags, permanent);
888 /* Turns off the specified 'flags' on 'netdev'.
889 * If 'permanent' is true, the changes will persist; otherwise, they
890 * will be reverted when 'netdev' is closed or the program exits.
891 * Returns 0 if successful, otherwise a positive errno value. */
893 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
896 return do_update_flags(netdev, netdev->netdev_fd, flags, 0, permanent);
899 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
900 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
901 * returns 0. Otherwise, it returns a positive errno value; in particular,
902 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
904 netdev_arp_lookup(const struct netdev *netdev,
905 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
908 struct sockaddr_in *pa;
911 memset(&r, 0, sizeof r);
912 pa = (struct sockaddr_in *) &r.arp_pa;
913 pa->sin_family = AF_INET;
914 pa->sin_addr.s_addr = ip;
916 r.arp_ha.sa_family = ARPHRD_ETHER;
918 strncpy(r.arp_dev, netdev->name, sizeof r.arp_dev);
919 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
921 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
922 } else if (retval != ENXIO) {
923 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
924 netdev->name, IP_ARGS(&ip), strerror(retval));
929 static void restore_all_flags(void *aux);
931 /* Set up a signal hook to restore network device flags on program
939 fatal_signal_add_hook(restore_all_flags, NULL, true);
940 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
941 if (af_inet_sock < 0) {
942 ofp_fatal(errno, "socket(AF_INET)");
947 /* Restore the network device flags on 'netdev' to those that were active
948 * before we changed them. Returns 0 if successful, otherwise a positive
951 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
953 restore_flags(struct netdev *netdev)
958 /* Get current flags. */
959 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
960 if (ioctl(netdev->netdev_fd, SIOCGIFFLAGS, &ifr) < 0) {
964 /* Restore flags that we might have changed, if necessary. */
965 restore_flags = netdev->changed_flags & (IFF_PROMISC | IFF_UP);
966 if ((ifr.ifr_flags ^ netdev->save_flags) & restore_flags) {
967 ifr.ifr_flags &= ~restore_flags;
968 ifr.ifr_flags |= netdev->save_flags & restore_flags;
969 if (ioctl(netdev->netdev_fd, SIOCSIFFLAGS, &ifr) < 0) {
977 /* Retores all the flags on all network devices that we modified. Called from
978 * a signal handler, so it does not attempt to report error conditions. */
980 restore_all_flags(void *aux UNUSED)
982 struct netdev *netdev;
983 LIST_FOR_EACH (netdev, struct netdev, node, &netdev_list) {
984 restore_flags(netdev);
989 get_flags(const struct netdev *netdev, int fd, int *flags)
992 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
993 if (ioctl(fd, SIOCGIFFLAGS, &ifr) < 0) {
994 VLOG_ERR("ioctl(SIOCGIFFLAGS) on %s device failed: %s",
995 netdev->name, strerror(errno));
998 *flags = ifr.ifr_flags;
1003 set_flags(struct netdev *netdev, int fd, int flags)
1006 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
1007 ifr.ifr_flags = flags;
1008 if (ioctl(netdev->netdev_fd, SIOCSIFFLAGS, &ifr) < 0) {
1009 VLOG_ERR("ioctl(SIOCSIFFLAGS) on %s device failed: %s",
1010 netdev->name, strerror(errno));