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"
64 #include "openflow/openflow.h"
66 #include "poll-loop.h"
67 #include "socket-util.h"
70 #define THIS_MODULE VLM_netdev
77 /* File descriptors. For ordinary network devices, the two fds below are
78 * the same; for tap devices, they differ. */
79 int netdev_fd; /* Network device. */
80 int tap_fd; /* TAP character device, if any, otherwise the
83 /* Cached network device information. */
85 uint8_t etheraddr[ETH_ADDR_LEN];
92 /* Bitmaps of OFPPF_* that describe features. All bits disabled if
93 * unsupported or unavailable. */
94 uint32_t curr; /* Current features. */
95 uint32_t advertised; /* Features being advertised by the port. */
96 uint32_t supported; /* Features supported by the port. */
97 uint32_t peer; /* Features advertised by the peer. */
99 int save_flags; /* Initial device flags. */
100 int changed_flags; /* Flags that we changed. */
103 /* All open network devices. */
104 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
106 /* An AF_INET socket (used for ioctl operations). */
107 static int af_inet_sock = -1;
109 /* This is set pretty low because we probably won't learn anything from the
110 * additional log messages. */
111 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
113 static void init_netdev(void);
114 static int do_open_netdev(const char *name, int ethertype, int tap_fd,
115 struct netdev **netdev_);
116 static int restore_flags(struct netdev *netdev);
117 static int get_flags(const struct netdev *, int fd, int *flagsp);
118 static int set_flags(struct netdev *, int fd, int flags);
120 /* Obtains the IPv6 address for 'name' into 'in6'. */
122 get_ipv6_address(const char *name, struct in6_addr *in6)
127 file = fopen("/proc/net/if_inet6", "r");
129 /* This most likely indicates that the host doesn't have IPv6 support,
130 * so it's not really a failure condition.*/
135 while (fgets(line, sizeof line, file)) {
136 uint8_t *s6 = in6->s6_addr;
140 if (sscanf(line, " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
141 "%*x %*x %*x %*x %16s\n",
142 &s6[0], &s6[1], &s6[2], &s6[3],
143 &s6[4], &s6[5], &s6[6], &s6[7],
144 &s6[8], &s6[9], &s6[10], &s6[11],
145 &s6[12], &s6[13], &s6[14], &s6[15],
147 && !strcmp(name, ifname))
159 do_ethtool(struct netdev *netdev)
162 struct ethtool_cmd ecmd;
165 netdev->supported = 0;
166 netdev->advertised = 0;
169 memset(&ifr, 0, sizeof ifr);
170 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
171 ifr.ifr_data = (caddr_t) &ecmd;
173 memset(&ecmd, 0, sizeof ecmd);
174 ecmd.cmd = ETHTOOL_GSET;
175 if (ioctl(netdev->netdev_fd, SIOCETHTOOL, &ifr) == 0) {
176 if (ecmd.supported & SUPPORTED_10baseT_Half) {
177 netdev->supported |= OFPPF_10MB_HD;
179 if (ecmd.supported & SUPPORTED_10baseT_Full) {
180 netdev->supported |= OFPPF_10MB_FD;
182 if (ecmd.supported & SUPPORTED_100baseT_Half) {
183 netdev->supported |= OFPPF_100MB_HD;
185 if (ecmd.supported & SUPPORTED_100baseT_Full) {
186 netdev->supported |= OFPPF_100MB_FD;
188 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
189 netdev->supported |= OFPPF_1GB_HD;
191 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
192 netdev->supported |= OFPPF_1GB_FD;
194 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
195 netdev->supported |= OFPPF_10GB_FD;
197 if (ecmd.supported & SUPPORTED_TP) {
198 netdev->supported |= OFPPF_COPPER;
200 if (ecmd.supported & SUPPORTED_FIBRE) {
201 netdev->supported |= OFPPF_FIBER;
203 if (ecmd.supported & SUPPORTED_Autoneg) {
204 netdev->supported |= OFPPF_AUTONEG;
206 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
207 if (ecmd.supported & SUPPORTED_Pause) {
208 netdev->supported |= OFPPF_PAUSE;
210 if (ecmd.supported & SUPPORTED_Asym_Pause) {
211 netdev->supported |= OFPPF_PAUSE_ASYM;
213 #endif /* kernel >= 2.6.14 */
215 /* Set the advertised features */
216 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
217 netdev->advertised |= OFPPF_10MB_HD;
219 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
220 netdev->advertised |= OFPPF_10MB_FD;
222 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
223 netdev->advertised |= OFPPF_100MB_HD;
225 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
226 netdev->advertised |= OFPPF_100MB_FD;
228 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
229 netdev->advertised |= OFPPF_1GB_HD;
231 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
232 netdev->advertised |= OFPPF_1GB_FD;
234 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
235 netdev->advertised |= OFPPF_10GB_FD;
237 if (ecmd.advertising & ADVERTISED_TP) {
238 netdev->advertised |= OFPPF_COPPER;
240 if (ecmd.advertising & ADVERTISED_FIBRE) {
241 netdev->advertised |= OFPPF_FIBER;
243 if (ecmd.advertising & ADVERTISED_Autoneg) {
244 netdev->advertised |= OFPPF_AUTONEG;
246 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
247 if (ecmd.advertising & ADVERTISED_Pause) {
248 netdev->advertised |= OFPPF_PAUSE;
250 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
251 netdev->advertised |= OFPPF_PAUSE_ASYM;
253 #endif /* kernel >= 2.6.14 */
255 /* Set the current features */
256 if (ecmd.speed == SPEED_10) {
257 netdev->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
259 else if (ecmd.speed == SPEED_100) {
260 netdev->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
262 else if (ecmd.speed == SPEED_1000) {
263 netdev->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
265 else if (ecmd.speed == SPEED_10000) {
266 netdev->curr = OFPPF_10GB_FD;
269 if (ecmd.port == PORT_TP) {
270 netdev->curr |= OFPPF_COPPER;
272 else if (ecmd.port == PORT_FIBRE) {
273 netdev->curr |= OFPPF_FIBER;
277 netdev->curr |= OFPPF_AUTONEG;
280 VLOG_DBG("ioctl(SIOCETHTOOL) failed: %s", strerror(errno));
284 /* Opens the network device named 'name' (e.g. "eth0") and returns zero if
285 * successful, otherwise a positive errno value. On success, sets '*netdevp'
286 * to the new network device, otherwise to null.
288 * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
289 * capture frames of that type received on the device. It may also be one of
290 * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those
293 netdev_open(const char *name, int ethertype, struct netdev **netdevp)
295 if (!strncmp(name, "tap:", 4)) {
296 return netdev_open_tap(name + 4, netdevp);
298 return do_open_netdev(name, ethertype, -1, netdevp);
302 /* Opens a TAP virtual network device. If 'name' is a nonnull, non-empty
303 * string, attempts to assign that name to the TAP device (failing if the name
304 * is already in use); otherwise, a name is automatically assigned. Returns
305 * zero if successful, otherwise a positive errno value. On success, sets
306 * '*netdevp' to the new network device, otherwise to null. */
308 netdev_open_tap(const char *name, struct netdev **netdevp)
310 static const char tap_dev[] = "/dev/net/tun";
315 tap_fd = open(tap_dev, O_RDWR);
317 ofp_error(errno, "opening \"%s\" failed", tap_dev);
321 memset(&ifr, 0, sizeof ifr);
322 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
324 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
326 if (ioctl(tap_fd, TUNSETIFF, &ifr) < 0) {
328 ofp_error(error, "ioctl(TUNSETIFF) on \"%s\" failed", tap_dev);
333 error = set_nonblocking(tap_fd);
335 ofp_error(error, "set_nonblocking on \"%s\" failed", tap_dev);
340 error = do_open_netdev(ifr.ifr_name, NETDEV_ETH_TYPE_NONE, tap_fd,
349 do_open_netdev(const char *name, int ethertype, int tap_fd,
350 struct netdev **netdev_)
353 struct sockaddr_ll sll;
355 unsigned int ifindex;
356 uint8_t etheraddr[ETH_ADDR_LEN];
362 struct netdev *netdev;
367 /* Create raw socket. */
368 netdev_fd = socket(PF_PACKET, SOCK_RAW,
369 htons(ethertype == NETDEV_ETH_TYPE_NONE ? 0
370 : ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
371 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
377 /* Set non-blocking mode. */
378 error = set_nonblocking(netdev_fd);
380 goto error_already_set;
383 /* Get ethernet device index. */
384 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
385 if (ioctl(netdev_fd, SIOCGIFINDEX, &ifr) < 0) {
386 VLOG_ERR("ioctl(SIOCGIFINDEX) on %s device failed: %s",
387 name, strerror(errno));
390 ifindex = ifr.ifr_ifindex;
392 /* Bind to specific ethernet device. */
393 memset(&sll, 0, sizeof sll);
394 sll.sll_family = AF_PACKET;
395 sll.sll_ifindex = ifindex;
396 if (bind(netdev_fd, (struct sockaddr *) &sll, sizeof sll) < 0) {
397 VLOG_ERR("bind to %s failed: %s", name, strerror(errno));
401 if (ethertype != NETDEV_ETH_TYPE_NONE) {
402 /* Between the socket() and bind() calls above, the socket receives all
403 * packets of the requested type on all system interfaces. We do not
404 * want to receive that data, but there is no way to avoid it. So we
405 * must now drain out the receive queue. */
406 error = drain_rcvbuf(netdev_fd);
412 /* Get MAC address. */
413 if (ioctl(netdev_fd, SIOCGIFHWADDR, &ifr) < 0) {
414 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
415 name, strerror(errno));
418 hwaddr_family = ifr.ifr_hwaddr.sa_family;
419 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
420 VLOG_WARN("%s device has unknown hardware address family %d",
421 name, hwaddr_family);
423 memcpy(etheraddr, ifr.ifr_hwaddr.sa_data, sizeof etheraddr);
426 if (ioctl(netdev_fd, SIOCGIFMTU, &ifr) < 0) {
427 VLOG_ERR("ioctl(SIOCGIFMTU) on %s device failed: %s",
428 name, strerror(errno));
433 /* Get TX queue length. */
434 if (ioctl(netdev_fd, SIOCGIFTXQLEN, &ifr) < 0) {
435 VLOG_ERR("ioctl(SIOCGIFTXQLEN) on %s device failed: %s",
436 name, strerror(errno));
439 txqlen = ifr.ifr_qlen;
441 get_ipv6_address(name, &in6);
443 /* Allocate network device. */
444 netdev = xmalloc(sizeof *netdev);
445 netdev->name = xstrdup(name);
446 netdev->ifindex = ifindex;
447 netdev->txqlen = txqlen;
448 netdev->hwaddr_family = hwaddr_family;
449 netdev->netdev_fd = netdev_fd;
450 netdev->tap_fd = tap_fd < 0 ? netdev_fd : tap_fd;
451 memcpy(netdev->etheraddr, etheraddr, sizeof etheraddr);
455 /* Get speed, features. */
458 /* Save flags to restore at close or exit. */
459 error = get_flags(netdev, netdev_fd, &netdev->save_flags);
461 goto error_already_set;
463 netdev->changed_flags = 0;
464 fatal_signal_block();
465 list_push_back(&netdev_list, &netdev->node);
466 fatal_signal_unblock();
482 /* Closes and destroys 'netdev'. */
484 netdev_close(struct netdev *netdev)
487 /* Bring down interface and drop promiscuous mode, if we brought up
488 * the interface or enabled promiscuous mode. */
490 fatal_signal_block();
491 error = restore_flags(netdev);
492 list_remove(&netdev->node);
493 fatal_signal_unblock();
495 VLOG_WARN("failed to restore network device flags on %s: %s",
496 netdev->name, strerror(error));
501 close(netdev->netdev_fd);
502 if (netdev->netdev_fd != netdev->tap_fd) {
503 close(netdev->tap_fd);
509 /* Pads 'buffer' out with zero-bytes to the minimum valid length of an
510 * Ethernet packet, if necessary. */
512 pad_to_minimum_length(struct ofpbuf *buffer)
514 if (buffer->size < ETH_TOTAL_MIN) {
515 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
519 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
520 * must have initialized with sufficient room for the packet. The space
521 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
522 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
523 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
524 * need not be included.)
526 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
527 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
528 * positive errno value. Returns EAGAIN immediately if no packet is ready to
532 netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
536 assert(buffer->size == 0);
537 assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
539 n_bytes = read(netdev->tap_fd,
540 ofpbuf_tail(buffer), ofpbuf_tailroom(buffer));
541 } while (n_bytes < 0 && errno == EINTR);
543 if (errno != EAGAIN) {
544 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
545 strerror(errno), netdev->name);
549 buffer->size += n_bytes;
551 /* When the kernel internally sends out an Ethernet frame on an
552 * interface, it gives us a copy *before* padding the frame to the
553 * minimum length. Thus, when it sends out something like an ARP
554 * request, we see a too-short frame. So pad it out to the minimum
556 pad_to_minimum_length(buffer);
561 /* Registers with the poll loop to wake up from the next call to poll_block()
562 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
564 netdev_recv_wait(struct netdev *netdev)
566 poll_fd_wait(netdev->tap_fd, POLLIN);
569 /* Discards all packets waiting to be received from 'netdev'. */
571 netdev_drain(struct netdev *netdev)
573 if (netdev->tap_fd != netdev->netdev_fd) {
574 drain_fd(netdev->tap_fd, netdev->txqlen);
577 return drain_rcvbuf(netdev->netdev_fd);
581 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
582 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
583 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
584 * the packet is too big or too small to transmit on the device.
586 * The caller retains ownership of 'buffer' in all cases.
588 * The kernel maintains a packet transmission queue, so the caller is not
589 * expected to do additional queuing of packets. */
591 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
596 n_bytes = write(netdev->tap_fd, buffer->data, buffer->size);
597 } while (n_bytes < 0 && errno == EINTR);
600 /* The Linux AF_PACKET implementation never blocks waiting for room
601 * for packets, instead returning ENOBUFS. Translate this into EAGAIN
603 if (errno == ENOBUFS) {
605 } else if (errno != EAGAIN) {
606 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
607 netdev->name, strerror(errno));
610 } else if (n_bytes != buffer->size) {
612 "send partial Ethernet packet (%d bytes of %zu) on %s",
613 (int) n_bytes, buffer->size, netdev->name);
620 /* Registers with the poll loop to wake up from the next call to poll_block()
621 * when the packet transmission queue has sufficient room to transmit a packet
622 * with netdev_send().
624 * The kernel maintains a packet transmission queue, so the client is not
625 * expected to do additional queuing of packets. Thus, this function is
626 * unlikely to ever be used. It is included for completeness. */
628 netdev_send_wait(struct netdev *netdev)
630 if (netdev->tap_fd == netdev->netdev_fd) {
631 poll_fd_wait(netdev->tap_fd, POLLOUT);
633 /* TAP device always accepts packets.*/
634 poll_immediate_wake();
638 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
639 * otherwise a positive errno value. */
641 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
645 memset(&ifr, 0, sizeof ifr);
646 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
647 ifr.ifr_hwaddr.sa_family = netdev->hwaddr_family;
648 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
649 if (ioctl(netdev->netdev_fd, SIOCSIFHWADDR, &ifr) < 0) {
650 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
651 netdev->name, strerror(errno));
654 memcpy(netdev->etheraddr, mac, ETH_ADDR_LEN);
658 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
659 * free the returned buffer. */
661 netdev_get_etheraddr(const struct netdev *netdev)
663 return netdev->etheraddr;
666 /* Returns the name of the network device that 'netdev' represents,
667 * e.g. "eth0". The caller must not modify or free the returned string. */
669 netdev_get_name(const struct netdev *netdev)
674 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
675 * in bytes, not including the hardware header; thus, this is typically 1500
676 * bytes for Ethernet devices. */
678 netdev_get_mtu(const struct netdev *netdev)
683 /* Checks the link status. Returns 1 or 0 to indicate the link is active
684 * or not, respectively. Any other return value indicates an error. */
686 netdev_get_link_status(const struct netdev *netdev)
689 struct ethtool_value edata;
691 memset(&ifr, 0, sizeof ifr);
692 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
693 ifr.ifr_data = (caddr_t) &edata;
695 memset(&edata, 0, sizeof edata);
696 edata.cmd = ETHTOOL_GLINK;
697 if (ioctl(netdev->netdev_fd, SIOCETHTOOL, &ifr) == 0) {
708 /* Returns the features supported by 'netdev' of type 'type', as a bitmap
709 * of bits from enum ofp_phy_features, in host byte order. */
711 netdev_get_features(struct netdev *netdev, int type)
715 case NETDEV_FEAT_CURRENT:
717 case NETDEV_FEAT_ADVERTISED:
718 return netdev->advertised;
719 case NETDEV_FEAT_SUPPORTED:
720 return netdev->supported;
721 case NETDEV_FEAT_PEER:
724 VLOG_WARN("Unknown feature type: %d\n", type);
729 /* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address (if
730 * 'in4' is non-null) and returns true. Otherwise, returns false. */
732 netdev_get_in4(const struct netdev *netdev, struct in_addr *in4)
735 struct in_addr ip = { INADDR_ANY };
737 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
738 ifr.ifr_addr.sa_family = AF_INET;
739 if (ioctl(af_inet_sock, SIOCGIFADDR, &ifr) == 0) {
740 struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
743 VLOG_DBG_RL(&rl, "%s: ioctl(SIOCGIFADDR) failed: %s",
744 netdev->name, strerror(errno));
749 return ip.s_addr != INADDR_ANY;
753 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
755 struct sockaddr_in sin;
756 memset(&sin, 0, sizeof sin);
757 sin.sin_family = AF_INET;
761 memset(sa, 0, sizeof *sa);
762 memcpy(sa, &sin, sizeof sin);
766 do_set_addr(struct netdev *netdev, int sock,
767 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
772 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
773 make_in4_sockaddr(&ifr.ifr_addr, addr);
774 error = ioctl(sock, ioctl_nr, &ifr) < 0 ? errno : 0;
776 VLOG_WARN("ioctl(%s): %s", ioctl_name, strerror(error));
781 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
782 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
783 * positive errno value. */
785 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
789 error = do_set_addr(netdev, af_inet_sock,
790 SIOCSIFADDR, "SIOCSIFADDR", addr);
791 if (!error && addr.s_addr != INADDR_ANY) {
792 error = do_set_addr(netdev, af_inet_sock,
793 SIOCSIFNETMASK, "SIOCSIFNETMASK", mask);
798 /* Adds 'router' as a default gateway for 'netdev''s IP address. */
800 netdev_add_router(struct netdev *netdev, struct in_addr router)
802 struct in_addr any = { INADDR_ANY };
806 memset(&rt, 0, sizeof rt);
807 make_in4_sockaddr(&rt.rt_dst, any);
808 make_in4_sockaddr(&rt.rt_gateway, router);
809 make_in4_sockaddr(&rt.rt_genmask, any);
810 rt.rt_flags = RTF_UP | RTF_GATEWAY;
811 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
813 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
818 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
819 * 'in6' is non-null) and returns true. Otherwise, returns false. */
821 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
826 return memcmp(&netdev->in6, &in6addr_any, sizeof netdev->in6) != 0;
829 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
830 * Returns 0 if successful, otherwise a positive errno value. */
832 netdev_get_flags(const struct netdev *netdev, enum netdev_flags *flagsp)
836 error = get_flags(netdev, netdev->netdev_fd, &flags);
842 if (flags & IFF_UP) {
843 *flagsp |= NETDEV_UP;
845 if (flags & IFF_PROMISC) {
846 *flagsp |= NETDEV_PROMISC;
852 nd_to_iff_flags(enum netdev_flags nd)
855 if (nd & NETDEV_UP) {
858 if (nd & NETDEV_PROMISC) {
864 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
865 * 'on'. If 'permanent' is true, the changes will persist; otherwise, they
866 * will be reverted when 'netdev' is closed or the program exits. Returns 0 if
867 * successful, otherwise a positive errno value. */
869 do_update_flags(struct netdev *netdev, int fd, enum netdev_flags off,
870 enum netdev_flags on, bool permanent)
872 int old_flags, new_flags;
875 error = get_flags(netdev, fd, &old_flags);
880 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
882 netdev->changed_flags |= new_flags ^ old_flags;
884 if (new_flags != old_flags) {
885 error = set_flags(netdev, fd, new_flags);
890 /* Sets the flags for 'netdev' to 'flags'.
891 * If 'permanent' is true, the changes will persist; otherwise, they
892 * will be reverted when 'netdev' is closed or the program exits.
893 * Returns 0 if successful, otherwise a positive errno value. */
895 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
898 return do_update_flags(netdev, netdev->netdev_fd, -1, flags, permanent);
901 /* Turns on the specified 'flags' on 'netdev'.
902 * If 'permanent' is true, the changes will persist; otherwise, they
903 * will be reverted when 'netdev' is closed or the program exits.
904 * Returns 0 if successful, otherwise a positive errno value. */
906 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
909 return do_update_flags(netdev, netdev->netdev_fd, 0, flags, permanent);
912 /* Turns off the specified 'flags' on 'netdev'.
913 * If 'permanent' is true, the changes will persist; otherwise, they
914 * will be reverted when 'netdev' is closed or the program exits.
915 * Returns 0 if successful, otherwise a positive errno value. */
917 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
920 return do_update_flags(netdev, netdev->netdev_fd, flags, 0, permanent);
923 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
924 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
925 * returns 0. Otherwise, it returns a positive errno value; in particular,
926 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
928 netdev_arp_lookup(const struct netdev *netdev,
929 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
932 struct sockaddr_in *pa;
935 memset(&r, 0, sizeof r);
936 pa = (struct sockaddr_in *) &r.arp_pa;
937 pa->sin_family = AF_INET;
938 pa->sin_addr.s_addr = ip;
940 r.arp_ha.sa_family = ARPHRD_ETHER;
942 strncpy(r.arp_dev, netdev->name, sizeof r.arp_dev);
943 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
945 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
946 } else if (retval != ENXIO) {
947 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
948 netdev->name, IP_ARGS(&ip), strerror(retval));
953 /* Initializes 'svec' with a list of the names of all known network devices. */
955 netdev_enumerate(struct svec *svec)
957 struct if_nameindex *names;
960 names = if_nameindex();
964 for (i = 0; names[i].if_name != NULL; i++) {
965 svec_add(svec, names[i].if_name);
967 if_freenameindex(names);
969 VLOG_WARN("could not obtain list of network device names: %s",
974 static void restore_all_flags(void *aux);
976 /* Set up a signal hook to restore network device flags on program
984 fatal_signal_add_hook(restore_all_flags, NULL, true);
985 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
986 if (af_inet_sock < 0) {
987 ofp_fatal(errno, "socket(AF_INET)");
992 /* Restore the network device flags on 'netdev' to those that were active
993 * before we changed them. Returns 0 if successful, otherwise a positive
996 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
998 restore_flags(struct netdev *netdev)
1003 /* Get current flags. */
1004 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
1005 if (ioctl(netdev->netdev_fd, SIOCGIFFLAGS, &ifr) < 0) {
1009 /* Restore flags that we might have changed, if necessary. */
1010 restore_flags = netdev->changed_flags & (IFF_PROMISC | IFF_UP);
1011 if ((ifr.ifr_flags ^ netdev->save_flags) & restore_flags) {
1012 ifr.ifr_flags &= ~restore_flags;
1013 ifr.ifr_flags |= netdev->save_flags & restore_flags;
1014 if (ioctl(netdev->netdev_fd, SIOCSIFFLAGS, &ifr) < 0) {
1022 /* Retores all the flags on all network devices that we modified. Called from
1023 * a signal handler, so it does not attempt to report error conditions. */
1025 restore_all_flags(void *aux UNUSED)
1027 struct netdev *netdev;
1028 LIST_FOR_EACH (netdev, struct netdev, node, &netdev_list) {
1029 restore_flags(netdev);
1034 get_flags(const struct netdev *netdev, int fd, int *flags)
1037 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
1038 if (ioctl(fd, SIOCGIFFLAGS, &ifr) < 0) {
1039 VLOG_ERR("ioctl(SIOCGIFFLAGS) on %s device failed: %s",
1040 netdev->name, strerror(errno));
1043 *flags = ifr.ifr_flags;
1048 set_flags(struct netdev *netdev, int fd, int flags)
1051 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
1052 ifr.ifr_flags = flags;
1053 if (ioctl(netdev->netdev_fd, SIOCSIFFLAGS, &ifr) < 0) {
1054 VLOG_ERR("ioctl(SIOCSIFFLAGS) on %s device failed: %s",
1055 netdev->name, strerror(errno));