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 SOFTWARE 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.
38 #include <arpa/inet.h>
40 #include <linux/types.h>
41 #include <linux/ethtool.h>
42 #include <linux/sockios.h>
43 #include <sys/types.h>
44 #include <sys/ioctl.h>
45 #include <sys/socket.h>
46 #include <netpacket/packet.h>
47 #include <net/ethernet.h>
49 #include <net/if_arp.h>
50 #include <net/if_packet.h>
51 #include <netinet/in.h>
57 #include "fatal-signal.h"
61 #include "poll-loop.h"
62 #include "socket-util.h"
64 #define THIS_MODULE VLM_netdev
72 uint8_t etheraddr[ETH_ADDR_LEN];
81 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
83 /* An AF_INET socket (used for ioctl operations). */
84 static int af_inet_sock = -1;
86 static void init_netdev(void);
87 static int restore_flags(struct netdev *netdev);
88 static int get_flags(const struct netdev *, int *flagsp);
89 static int set_flags(struct netdev *, int flags);
91 /* Obtains the IPv4 address for 'name' into 'in4'. Returns true if
94 get_ipv4_address(const char *name, struct in_addr *in4)
98 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
99 ifr.ifr_addr.sa_family = AF_INET;
100 if (ioctl(af_inet_sock, SIOCGIFADDR, &ifr) == 0) {
101 struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
102 *in4 = sin->sin_addr;
104 in4->s_addr = INADDR_ANY;
110 /* Obtains the IPv6 address for 'name' into 'in6'. */
112 get_ipv6_address(const char *name, struct in6_addr *in6)
117 file = fopen("/proc/net/if_inet6", "r");
119 /* This most likely indicates that the host doesn't have IPv6 support,
120 * so it's not really a failure condition.*/
125 while (fgets(line, sizeof line, file)) {
126 uint8_t *s6 = in6->s6_addr;
130 if (sscanf(line, " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
131 "%*x %*x %*x %*x %16s\n",
132 &s6[0], &s6[1], &s6[2], &s6[3],
133 &s6[4], &s6[5], &s6[6], &s6[7],
134 &s6[8], &s6[9], &s6[10], &s6[11],
135 &s6[12], &s6[13], &s6[14], &s6[15],
137 && !strcmp(name, ifname))
148 do_ethtool(struct netdev *netdev)
151 struct ethtool_cmd ecmd;
154 netdev->features = 0;
156 memset(&ifr, 0, sizeof ifr);
157 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
158 ifr.ifr_data = (caddr_t) &ecmd;
160 memset(&ecmd, 0, sizeof ecmd);
161 ecmd.cmd = ETHTOOL_GSET;
162 if (ioctl(netdev->fd, SIOCETHTOOL, &ifr) == 0) {
163 if (ecmd.supported & SUPPORTED_10baseT_Half) {
164 netdev->features |= OFPPF_10MB_HD;
166 if (ecmd.supported & SUPPORTED_10baseT_Full) {
167 netdev->features |= OFPPF_10MB_FD;
169 if (ecmd.supported & SUPPORTED_100baseT_Half) {
170 netdev->features |= OFPPF_100MB_HD;
172 if (ecmd.supported & SUPPORTED_100baseT_Full) {
173 netdev->features |= OFPPF_100MB_FD;
175 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
176 netdev->features |= OFPPF_1GB_HD;
178 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
179 netdev->features |= OFPPF_1GB_FD;
181 /* 10Gbps half-duplex doesn't exist... */
182 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
183 netdev->features |= OFPPF_10GB_FD;
186 switch (ecmd.speed) {
196 netdev->speed = 1000;
200 netdev->speed = 2500;
204 netdev->speed = 10000;
208 VLOG_DBG("ioctl(SIOCETHTOOL) failed: %s", strerror(errno));
212 /* Opens the network device named 'name' (e.g. "eth0") and returns zero if
213 * successful, otherwise a positive errno value. On success, sets '*netdev'
214 * to the new network device, otherwise to null.
216 * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
217 * capture frames of that type received on the device. It may also be one of
218 * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those
221 netdev_open(const char *name, int ethertype, struct netdev **netdev_)
224 struct sockaddr_ll sll;
226 unsigned int ifindex;
227 uint8_t etheraddr[ETH_ADDR_LEN];
232 struct netdev *netdev;
237 /* Create raw socket. */
238 fd = socket(PF_PACKET, SOCK_RAW,
239 htons(ethertype == NETDEV_ETH_TYPE_NONE ? 0
240 : ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
241 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
247 /* Get ethernet device index. */
248 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
249 if (ioctl(fd, SIOCGIFINDEX, &ifr) < 0) {
250 VLOG_ERR("ioctl(SIOCGIFINDEX) on %s device failed: %s",
251 name, strerror(errno));
254 ifindex = ifr.ifr_ifindex;
256 /* Bind to specific ethernet device. */
257 memset(&sll, 0, sizeof sll);
258 sll.sll_family = AF_PACKET;
259 sll.sll_ifindex = ifindex;
260 if (bind(fd, (struct sockaddr *) &sll, sizeof sll) < 0) {
261 VLOG_ERR("bind to %s failed: %s", name, strerror(errno));
265 if (ethertype != NETDEV_ETH_TYPE_NONE) {
266 /* Between the socket() and bind() calls above, the socket receives all
267 * packets of the requested type on all system interfaces. We do not
268 * want to receive that data, but there is no way to avoid it. So we
269 * must now drain out the receive queue. */
270 error = drain_rcvbuf(fd);
276 /* Get MAC address. */
277 if (ioctl(fd, SIOCGIFHWADDR, &ifr) < 0) {
278 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
279 name, strerror(errno));
282 if (ifr.ifr_hwaddr.sa_family != AF_UNSPEC
283 && ifr.ifr_hwaddr.sa_family != ARPHRD_ETHER) {
284 VLOG_WARN("%s device has unknown hardware address family %d",
285 name, (int) ifr.ifr_hwaddr.sa_family);
287 memcpy(etheraddr, ifr.ifr_hwaddr.sa_data, sizeof etheraddr);
290 if (ioctl(fd, SIOCGIFMTU, &ifr) < 0) {
291 VLOG_ERR("ioctl(SIOCGIFMTU) on %s device failed: %s",
292 name, strerror(errno));
297 if (!get_ipv4_address(name, &in4)) {
300 get_ipv6_address(name, &in6);
302 /* Allocate network device. */
303 netdev = xmalloc(sizeof *netdev);
304 netdev->name = xstrdup(name);
305 netdev->ifindex = ifindex;
307 memcpy(netdev->etheraddr, etheraddr, sizeof etheraddr);
312 /* Get speed, features. */
315 /* Save flags to restore at close or exit. */
316 error = get_flags(netdev, &netdev->save_flags);
320 fatal_signal_block();
321 list_push_back(&netdev_list, &netdev->node);
322 fatal_signal_unblock();
335 /* Closes and destroys 'netdev'. */
337 netdev_close(struct netdev *netdev)
340 /* Bring down interface and drop promiscuous mode, if we brought up
341 * the interface or enabled promiscuous mode. */
343 fatal_signal_block();
344 error = restore_flags(netdev);
345 list_remove(&netdev->node);
346 fatal_signal_unblock();
348 VLOG_WARN("failed to restore network device flags on %s: %s",
349 netdev->name, strerror(error));
359 /* Pads 'buffer' out with zero-bytes to the minimum valid length of an
360 * Ethernet packet, if necessary. */
362 pad_to_minimum_length(struct buffer *buffer)
364 if (buffer->size < ETH_TOTAL_MIN) {
365 size_t shortage = ETH_TOTAL_MIN - buffer->size;
366 memset(buffer_put_uninit(buffer, shortage), 0, shortage);
370 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
371 * must have initialized with sufficient room for the packet. The space
372 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
373 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
374 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
375 * need not be included.)
377 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
378 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
379 * positive errno value. Returns EAGAIN immediately if no packet is ready to
383 netdev_recv(struct netdev *netdev, struct buffer *buffer)
387 assert(buffer->size == 0);
388 assert(buffer_tailroom(buffer) >= ETH_TOTAL_MIN);
390 n_bytes = recv(netdev->fd,
391 buffer_tail(buffer), buffer_tailroom(buffer),
393 } while (n_bytes < 0 && errno == EINTR);
395 if (errno != EAGAIN) {
396 VLOG_WARN("error receiving Ethernet packet on %s: %s",
397 strerror(errno), netdev->name);
401 buffer->size += n_bytes;
403 /* When the kernel internally sends out an Ethernet frame on an
404 * interface, it gives us a copy *before* padding the frame to the
405 * minimum length. Thus, when it sends out something like an ARP
406 * request, we see a too-short frame. So pad it out to the minimum
408 pad_to_minimum_length(buffer);
413 /* Registers with the poll loop to wake up from the next call to poll_block()
414 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
416 netdev_recv_wait(struct netdev *netdev)
418 poll_fd_wait(netdev->fd, POLLIN);
421 /* Discards all packets waiting to be received from 'netdev'. */
423 netdev_drain(struct netdev *netdev)
425 drain_rcvbuf(netdev->fd);
428 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
429 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
430 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
431 * the packet is too big or too small to transmit on the device.
433 * The caller retains ownership of 'buffer' in all cases.
435 * The kernel maintains a packet transmission queue, so the caller is not
436 * expected to do additional queuing of packets. */
438 netdev_send(struct netdev *netdev, const struct buffer *buffer)
441 const struct eth_header *eh;
443 /* Pull out the Ethernet header. */
444 if (buffer->size < ETH_HEADER_LEN) {
445 VLOG_WARN("cannot send %zu-byte frame on %s",
446 buffer->size, netdev->name);
449 eh = buffer_at_assert(buffer, 0, sizeof *eh);
452 n_bytes = sendto(netdev->fd, buffer->data, buffer->size, 0, NULL, 0);
453 } while (n_bytes < 0 && errno == EINTR);
456 /* The Linux AF_PACKET implementation never blocks waiting for room
457 * for packets, instead returning ENOBUFS. Translate this into EAGAIN
459 if (errno == ENOBUFS) {
461 } else if (errno != EAGAIN) {
462 VLOG_WARN("error sending Ethernet packet on %s: %s",
463 netdev->name, strerror(errno));
466 } else if (n_bytes != buffer->size) {
467 VLOG_WARN("send partial Ethernet packet (%d bytes of %zu) on %s",
468 (int) n_bytes, buffer->size, netdev->name);
475 /* Registers with the poll loop to wake up from the next call to poll_block()
476 * when the packet transmission queue has sufficient room to transmit a packet
477 * with netdev_send().
479 * The kernel maintains a packet transmission queue, so the client is not
480 * expected to do additional queuing of packets. Thus, this function is
481 * unlikely to ever be used. It is included for completeness. */
483 netdev_send_wait(struct netdev *netdev)
485 poll_fd_wait(netdev->fd, POLLOUT);
488 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
489 * free the returned buffer. */
491 netdev_get_etheraddr(const struct netdev *netdev)
493 return netdev->etheraddr;
496 /* Returns the name of the network device that 'netdev' represents,
497 * e.g. "eth0". The caller must not modify or free the returned string. */
499 netdev_get_name(const struct netdev *netdev)
504 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
505 * in bytes, not including the hardware header; thus, this is typically 1500
506 * bytes for Ethernet devices. */
508 netdev_get_mtu(const struct netdev *netdev)
513 /* Returns the current speed of the network device that 'netdev' represents, in
514 * megabits per second, or 0 if the speed is unknown. */
516 netdev_get_speed(const struct netdev *netdev)
518 return netdev->speed;
521 /* Returns the features supported by 'netdev', as a bitmap of bits from enum
522 * ofp_phy_port, in host byte order. */
524 netdev_get_features(const struct netdev *netdev)
526 return netdev->features;
529 /* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address and
530 * returns true. Otherwise, returns false. */
532 netdev_get_in4(const struct netdev *netdev, struct in_addr *in4)
535 return in4->s_addr != INADDR_ANY;
538 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
539 * returns true. Otherwise, returns false. */
541 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
544 return memcmp(in6, &in6addr_any, sizeof *in6) != 0;
547 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
548 * Returns 0 if successful, otherwise a positive errno value. */
550 netdev_get_flags(const struct netdev *netdev, enum netdev_flags *flagsp)
554 error = get_flags(netdev, &flags);
560 if (flags & IFF_UP) {
561 *flagsp |= NETDEV_UP;
563 if (flags & IFF_PROMISC) {
564 *flagsp |= NETDEV_PROMISC;
570 nd_to_iff_flags(enum netdev_flags nd)
573 if (nd & NETDEV_UP) {
576 if (nd & NETDEV_PROMISC) {
582 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
583 * 'on'. Returns 0 if successful, otherwise a positive errno value. */
585 do_update_flags(struct netdev *netdev, enum netdev_flags off,
586 enum netdev_flags on)
588 int old_flags, new_flags;
591 error = get_flags(netdev, &old_flags);
596 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
597 if (new_flags != old_flags) {
598 error = set_flags(netdev, new_flags);
603 /* Sets the flags for 'netdev' to 'flags'.
604 * Returns 0 if successful, otherwise a positive errno value. */
606 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags)
608 return do_update_flags(netdev, -1, flags);
611 /* Turns on the specified 'flags' on 'netdev'.
612 * Returns 0 if successful, otherwise a positive errno value. */
614 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags)
616 return do_update_flags(netdev, 0, flags);
619 /* Turns off the specified 'flags' on 'netdev'.
620 * Returns 0 if successful, otherwise a positive errno value. */
622 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags)
624 return do_update_flags(netdev, flags, 0);
627 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
628 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
629 * returns 0. Otherwise, it returns a positive errno value; in particular,
630 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
632 netdev_arp_lookup(const struct netdev *netdev,
633 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
636 struct sockaddr_in *pa;
639 memset(&r, 0, sizeof r);
640 pa = (struct sockaddr_in *) &r.arp_pa;
641 pa->sin_family = AF_INET;
642 pa->sin_addr.s_addr = ip;
644 r.arp_ha.sa_family = ARPHRD_ETHER;
646 strncpy(r.arp_dev, netdev->name, sizeof r.arp_dev);
647 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
649 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
650 } else if (retval != ENXIO) {
651 VLOG_WARN("%s: could not look up ARP entry for "IP_FMT": %s",
652 netdev->name, IP_ARGS(&ip), strerror(retval));
657 static void restore_all_flags(void *aux);
659 /* Set up a signal hook to restore network device flags on program
667 fatal_signal_add_hook(restore_all_flags, NULL);
668 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
669 if (af_inet_sock < 0) {
670 fatal(errno, "socket(AF_INET)");
675 /* Restore the network device flags on 'netdev' to those that were active
676 * before we changed them. Returns 0 if successful, otherwise a positive
679 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
681 restore_flags(struct netdev *netdev)
685 /* Get current flags. */
686 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
687 if (ioctl(netdev->fd, SIOCGIFFLAGS, &ifr) < 0) {
691 /* Restore flags that we might have changed, if necessary. */
692 if ((ifr.ifr_flags ^ netdev->save_flags) & (IFF_PROMISC | IFF_UP)) {
693 ifr.ifr_flags &= ~(IFF_PROMISC | IFF_UP);
694 ifr.ifr_flags |= netdev->save_flags & (IFF_PROMISC | IFF_UP);
695 if (ioctl(netdev->fd, SIOCSIFFLAGS, &ifr) < 0) {
703 /* Retores all the flags on all network devices that we modified. Called from
704 * a signal handler, so it does not attempt to report error conditions. */
706 restore_all_flags(void *aux UNUSED)
708 struct netdev *netdev;
709 LIST_FOR_EACH (netdev, struct netdev, node, &netdev_list) {
710 restore_flags(netdev);
715 get_flags(const struct netdev *netdev, int *flags)
718 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
719 if (ioctl(netdev->fd, SIOCGIFFLAGS, &ifr) < 0) {
720 VLOG_ERR("ioctl(SIOCGIFFLAGS) on %s device failed: %s",
721 netdev->name, strerror(errno));
724 *flags = ifr.ifr_flags;
729 set_flags(struct netdev *netdev, int flags)
732 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
733 ifr.ifr_flags = flags;
734 if (ioctl(netdev->fd, SIOCSIFFLAGS, &ifr) < 0) {
735 VLOG_ERR("ioctl(SIOCSIFFLAGS) on %s device failed: %s",
736 netdev->name, strerror(errno));