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.
39 #include <arpa/inet.h>
41 #include <linux/types.h>
42 #include <linux/ethtool.h>
43 #include <linux/sockios.h>
44 #include <sys/types.h>
45 #include <sys/ioctl.h>
46 #include <sys/socket.h>
47 #include <netpacket/packet.h>
48 #include <net/ethernet.h>
50 #include <net/if_arp.h>
51 #include <net/if_packet.h>
52 #include <net/route.h>
53 #include <netinet/in.h>
59 #include "fatal-signal.h"
63 #include "poll-loop.h"
64 #include "socket-util.h"
66 #define THIS_MODULE VLM_netdev
74 uint8_t etheraddr[ETH_ADDR_LEN];
80 int save_flags; /* Initial device flags. */
81 int changed_flags; /* Flags that we changed. */
84 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
86 /* An AF_INET socket (used for ioctl operations). */
87 static int af_inet_sock = -1;
89 static void init_netdev(void);
90 static int restore_flags(struct netdev *netdev);
91 static int get_flags(const struct netdev *, int *flagsp);
92 static int set_flags(struct netdev *, int flags);
94 /* Obtains the IPv4 address for 'name' into 'in4'. Returns true if
97 get_ipv4_address(const char *name, struct in_addr *in4)
101 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
102 ifr.ifr_addr.sa_family = AF_INET;
103 if (ioctl(af_inet_sock, SIOCGIFADDR, &ifr) == 0) {
104 struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
105 *in4 = sin->sin_addr;
107 in4->s_addr = INADDR_ANY;
113 /* Obtains the IPv6 address for 'name' into 'in6'. */
115 get_ipv6_address(const char *name, struct in6_addr *in6)
120 file = fopen("/proc/net/if_inet6", "r");
122 /* This most likely indicates that the host doesn't have IPv6 support,
123 * so it's not really a failure condition.*/
128 while (fgets(line, sizeof line, file)) {
129 uint8_t *s6 = in6->s6_addr;
133 if (sscanf(line, " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
134 "%*x %*x %*x %*x %16s\n",
135 &s6[0], &s6[1], &s6[2], &s6[3],
136 &s6[4], &s6[5], &s6[6], &s6[7],
137 &s6[8], &s6[9], &s6[10], &s6[11],
138 &s6[12], &s6[13], &s6[14], &s6[15],
140 && !strcmp(name, ifname))
151 do_ethtool(struct netdev *netdev)
154 struct ethtool_cmd ecmd;
157 netdev->features = 0;
159 memset(&ifr, 0, sizeof ifr);
160 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
161 ifr.ifr_data = (caddr_t) &ecmd;
163 memset(&ecmd, 0, sizeof ecmd);
164 ecmd.cmd = ETHTOOL_GSET;
165 if (ioctl(netdev->fd, SIOCETHTOOL, &ifr) == 0) {
166 if (ecmd.supported & SUPPORTED_10baseT_Half) {
167 netdev->features |= OFPPF_10MB_HD;
169 if (ecmd.supported & SUPPORTED_10baseT_Full) {
170 netdev->features |= OFPPF_10MB_FD;
172 if (ecmd.supported & SUPPORTED_100baseT_Half) {
173 netdev->features |= OFPPF_100MB_HD;
175 if (ecmd.supported & SUPPORTED_100baseT_Full) {
176 netdev->features |= OFPPF_100MB_FD;
178 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
179 netdev->features |= OFPPF_1GB_HD;
181 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
182 netdev->features |= OFPPF_1GB_FD;
184 /* 10Gbps half-duplex doesn't exist... */
185 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
186 netdev->features |= OFPPF_10GB_FD;
189 switch (ecmd.speed) {
199 netdev->speed = 1000;
203 netdev->speed = 2500;
207 netdev->speed = 10000;
211 VLOG_DBG("ioctl(SIOCETHTOOL) failed: %s", strerror(errno));
215 /* Opens the network device named 'name' (e.g. "eth0") and returns zero if
216 * successful, otherwise a positive errno value. On success, sets '*netdev'
217 * to the new network device, otherwise to null.
219 * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
220 * capture frames of that type received on the device. It may also be one of
221 * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those
224 netdev_open(const char *name, int ethertype, struct netdev **netdev_)
227 struct sockaddr_ll sll;
229 unsigned int ifindex;
230 uint8_t etheraddr[ETH_ADDR_LEN];
235 struct netdev *netdev;
240 /* Create raw socket. */
241 fd = socket(PF_PACKET, SOCK_RAW,
242 htons(ethertype == NETDEV_ETH_TYPE_NONE ? 0
243 : ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
244 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
250 /* Get ethernet device index. */
251 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
252 if (ioctl(fd, SIOCGIFINDEX, &ifr) < 0) {
253 VLOG_ERR("ioctl(SIOCGIFINDEX) on %s device failed: %s",
254 name, strerror(errno));
257 ifindex = ifr.ifr_ifindex;
259 /* Bind to specific ethernet device. */
260 memset(&sll, 0, sizeof sll);
261 sll.sll_family = AF_PACKET;
262 sll.sll_ifindex = ifindex;
263 if (bind(fd, (struct sockaddr *) &sll, sizeof sll) < 0) {
264 VLOG_ERR("bind to %s failed: %s", name, strerror(errno));
268 if (ethertype != NETDEV_ETH_TYPE_NONE) {
269 /* Between the socket() and bind() calls above, the socket receives all
270 * packets of the requested type on all system interfaces. We do not
271 * want to receive that data, but there is no way to avoid it. So we
272 * must now drain out the receive queue. */
273 error = drain_rcvbuf(fd);
279 /* Get MAC address. */
280 if (ioctl(fd, SIOCGIFHWADDR, &ifr) < 0) {
281 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
282 name, strerror(errno));
285 if (ifr.ifr_hwaddr.sa_family != AF_UNSPEC
286 && ifr.ifr_hwaddr.sa_family != ARPHRD_ETHER) {
287 VLOG_WARN("%s device has unknown hardware address family %d",
288 name, (int) ifr.ifr_hwaddr.sa_family);
290 memcpy(etheraddr, ifr.ifr_hwaddr.sa_data, sizeof etheraddr);
293 if (ioctl(fd, SIOCGIFMTU, &ifr) < 0) {
294 VLOG_ERR("ioctl(SIOCGIFMTU) on %s device failed: %s",
295 name, strerror(errno));
300 if (!get_ipv4_address(name, &in4)) {
303 get_ipv6_address(name, &in6);
305 /* Allocate network device. */
306 netdev = xmalloc(sizeof *netdev);
307 netdev->name = xstrdup(name);
308 netdev->ifindex = ifindex;
310 memcpy(netdev->etheraddr, etheraddr, sizeof etheraddr);
315 /* Get speed, features. */
318 /* Save flags to restore at close or exit. */
319 error = get_flags(netdev, &netdev->save_flags);
323 netdev->changed_flags = 0;
324 fatal_signal_block();
325 list_push_back(&netdev_list, &netdev->node);
326 fatal_signal_unblock();
339 /* Closes and destroys 'netdev'. */
341 netdev_close(struct netdev *netdev)
344 /* Bring down interface and drop promiscuous mode, if we brought up
345 * the interface or enabled promiscuous mode. */
347 fatal_signal_block();
348 error = restore_flags(netdev);
349 list_remove(&netdev->node);
350 fatal_signal_unblock();
352 VLOG_WARN("failed to restore network device flags on %s: %s",
353 netdev->name, strerror(error));
363 /* Pads 'buffer' out with zero-bytes to the minimum valid length of an
364 * Ethernet packet, if necessary. */
366 pad_to_minimum_length(struct buffer *buffer)
368 if (buffer->size < ETH_TOTAL_MIN) {
369 size_t shortage = ETH_TOTAL_MIN - buffer->size;
370 memset(buffer_put_uninit(buffer, shortage), 0, shortage);
374 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
375 * must have initialized with sufficient room for the packet. The space
376 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
377 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
378 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
379 * need not be included.)
381 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
382 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
383 * positive errno value. Returns EAGAIN immediately if no packet is ready to
387 netdev_recv(struct netdev *netdev, struct buffer *buffer)
391 assert(buffer->size == 0);
392 assert(buffer_tailroom(buffer) >= ETH_TOTAL_MIN);
394 n_bytes = recv(netdev->fd,
395 buffer_tail(buffer), buffer_tailroom(buffer),
397 } while (n_bytes < 0 && errno == EINTR);
399 if (errno != EAGAIN) {
400 VLOG_WARN("error receiving Ethernet packet on %s: %s",
401 strerror(errno), netdev->name);
405 buffer->size += n_bytes;
407 /* When the kernel internally sends out an Ethernet frame on an
408 * interface, it gives us a copy *before* padding the frame to the
409 * minimum length. Thus, when it sends out something like an ARP
410 * request, we see a too-short frame. So pad it out to the minimum
412 pad_to_minimum_length(buffer);
417 /* Registers with the poll loop to wake up from the next call to poll_block()
418 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
420 netdev_recv_wait(struct netdev *netdev)
422 poll_fd_wait(netdev->fd, POLLIN);
425 /* Discards all packets waiting to be received from 'netdev'. */
427 netdev_drain(struct netdev *netdev)
429 drain_rcvbuf(netdev->fd);
432 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
433 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
434 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
435 * the packet is too big or too small to transmit on the device.
437 * The caller retains ownership of 'buffer' in all cases.
439 * The kernel maintains a packet transmission queue, so the caller is not
440 * expected to do additional queuing of packets. */
442 netdev_send(struct netdev *netdev, const struct buffer *buffer)
445 const struct eth_header *eh;
447 /* Pull out the Ethernet header. */
448 if (buffer->size < ETH_HEADER_LEN) {
449 VLOG_WARN("cannot send %zu-byte frame on %s",
450 buffer->size, netdev->name);
453 eh = buffer_at_assert(buffer, 0, sizeof *eh);
456 n_bytes = sendto(netdev->fd, buffer->data, buffer->size, 0, NULL, 0);
457 } while (n_bytes < 0 && errno == EINTR);
460 /* The Linux AF_PACKET implementation never blocks waiting for room
461 * for packets, instead returning ENOBUFS. Translate this into EAGAIN
463 if (errno == ENOBUFS) {
465 } else if (errno != EAGAIN) {
466 VLOG_WARN("error sending Ethernet packet on %s: %s",
467 netdev->name, strerror(errno));
470 } else if (n_bytes != buffer->size) {
471 VLOG_WARN("send partial Ethernet packet (%d bytes of %zu) on %s",
472 (int) n_bytes, buffer->size, netdev->name);
479 /* Registers with the poll loop to wake up from the next call to poll_block()
480 * when the packet transmission queue has sufficient room to transmit a packet
481 * with netdev_send().
483 * The kernel maintains a packet transmission queue, so the client is not
484 * expected to do additional queuing of packets. Thus, this function is
485 * unlikely to ever be used. It is included for completeness. */
487 netdev_send_wait(struct netdev *netdev)
489 poll_fd_wait(netdev->fd, POLLOUT);
492 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
493 * free the returned buffer. */
495 netdev_get_etheraddr(const struct netdev *netdev)
497 return netdev->etheraddr;
500 /* Returns the name of the network device that 'netdev' represents,
501 * e.g. "eth0". The caller must not modify or free the returned string. */
503 netdev_get_name(const struct netdev *netdev)
508 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
509 * in bytes, not including the hardware header; thus, this is typically 1500
510 * bytes for Ethernet devices. */
512 netdev_get_mtu(const struct netdev *netdev)
517 /* Returns the current speed of the network device that 'netdev' represents, in
518 * megabits per second, or 0 if the speed is unknown. */
520 netdev_get_speed(const struct netdev *netdev)
522 return netdev->speed;
525 /* Returns the features supported by 'netdev', as a bitmap of bits from enum
526 * ofp_phy_port, in host byte order. */
528 netdev_get_features(const struct netdev *netdev)
530 return netdev->features;
533 /* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address and
534 * returns true. Otherwise, returns false. */
536 netdev_get_in4(const struct netdev *netdev, struct in_addr *in4)
539 return in4->s_addr != INADDR_ANY;
543 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
545 struct sockaddr_in sin;
546 memset(&sin, 0, sizeof sin);
547 sin.sin_family = AF_INET;
551 memset(sa, 0, sizeof *sa);
552 memcpy(sa, &sin, sizeof sin);
556 do_set_addr(struct netdev *netdev, int sock,
557 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
562 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
563 make_in4_sockaddr(&ifr.ifr_addr, addr);
564 error = ioctl(sock, ioctl_nr, &ifr) < 0 ? errno : 0;
566 VLOG_WARN("ioctl(%s): %s", ioctl_name, strerror(error));
571 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
572 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
573 * positive errno value. */
575 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
579 error = do_set_addr(netdev, af_inet_sock,
580 SIOCSIFADDR, "SIOCSIFADDR", addr);
583 if (addr.s_addr != INADDR_ANY) {
584 error = do_set_addr(netdev, af_inet_sock,
585 SIOCSIFNETMASK, "SIOCSIFNETMASK", mask);
591 /* Adds 'router' as a default gateway for 'netdev''s IP address. */
593 netdev_add_router(struct netdev *netdev, struct in_addr router)
595 struct in_addr any = { INADDR_ANY };
599 memset(&rt, 0, sizeof rt);
600 make_in4_sockaddr(&rt.rt_dst, any);
601 make_in4_sockaddr(&rt.rt_gateway, router);
602 make_in4_sockaddr(&rt.rt_genmask, any);
603 rt.rt_flags = RTF_UP | RTF_GATEWAY;
604 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
606 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
611 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
612 * returns true. Otherwise, returns false. */
614 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
617 return memcmp(in6, &in6addr_any, sizeof *in6) != 0;
620 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
621 * Returns 0 if successful, otherwise a positive errno value. */
623 netdev_get_flags(const struct netdev *netdev, enum netdev_flags *flagsp)
627 error = get_flags(netdev, &flags);
633 if (flags & IFF_UP) {
634 *flagsp |= NETDEV_UP;
636 if (flags & IFF_PROMISC) {
637 *flagsp |= NETDEV_PROMISC;
643 nd_to_iff_flags(enum netdev_flags nd)
646 if (nd & NETDEV_UP) {
649 if (nd & NETDEV_PROMISC) {
655 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
656 * 'on'. If 'permanent' is true, the changes will persist; otherwise, they
657 * will be reverted when 'netdev' is closed or the program exits. Returns 0 if
658 * successful, otherwise a positive errno value. */
660 do_update_flags(struct netdev *netdev, enum netdev_flags off,
661 enum netdev_flags on, bool permanent)
663 int old_flags, new_flags;
666 error = get_flags(netdev, &old_flags);
671 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
673 netdev->changed_flags |= new_flags ^ old_flags;
675 if (new_flags != old_flags) {
676 error = set_flags(netdev, new_flags);
681 /* Sets the flags for 'netdev' to 'flags'.
682 * If 'permanent' is true, the changes will persist; otherwise, they
683 * will be reverted when 'netdev' is closed or the program exits.
684 * Returns 0 if successful, otherwise a positive errno value. */
686 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
689 return do_update_flags(netdev, -1, flags, permanent);
692 /* Turns on the specified 'flags' on 'netdev'.
693 * If 'permanent' is true, the changes will persist; otherwise, they
694 * will be reverted when 'netdev' is closed or the program exits.
695 * Returns 0 if successful, otherwise a positive errno value. */
697 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
700 return do_update_flags(netdev, 0, flags, permanent);
703 /* Turns off the specified 'flags' on 'netdev'.
704 * If 'permanent' is true, the changes will persist; otherwise, they
705 * will be reverted when 'netdev' is closed or the program exits.
706 * Returns 0 if successful, otherwise a positive errno value. */
708 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
711 return do_update_flags(netdev, flags, 0, permanent);
714 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
715 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
716 * returns 0. Otherwise, it returns a positive errno value; in particular,
717 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
719 netdev_arp_lookup(const struct netdev *netdev,
720 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
723 struct sockaddr_in *pa;
726 memset(&r, 0, sizeof r);
727 pa = (struct sockaddr_in *) &r.arp_pa;
728 pa->sin_family = AF_INET;
729 pa->sin_addr.s_addr = ip;
731 r.arp_ha.sa_family = ARPHRD_ETHER;
733 strncpy(r.arp_dev, netdev->name, sizeof r.arp_dev);
734 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
736 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
737 } else if (retval != ENXIO) {
738 VLOG_WARN("%s: could not look up ARP entry for "IP_FMT": %s",
739 netdev->name, IP_ARGS(&ip), strerror(retval));
744 static void restore_all_flags(void *aux);
746 /* Set up a signal hook to restore network device flags on program
754 fatal_signal_add_hook(restore_all_flags, NULL);
755 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
756 if (af_inet_sock < 0) {
757 fatal(errno, "socket(AF_INET)");
762 /* Restore the network device flags on 'netdev' to those that were active
763 * before we changed them. Returns 0 if successful, otherwise a positive
766 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
768 restore_flags(struct netdev *netdev)
773 /* Get current flags. */
774 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
775 if (ioctl(netdev->fd, SIOCGIFFLAGS, &ifr) < 0) {
779 /* Restore flags that we might have changed, if necessary. */
780 restore_flags = netdev->changed_flags & (IFF_PROMISC | IFF_UP);
781 if ((ifr.ifr_flags ^ netdev->save_flags) & restore_flags) {
782 ifr.ifr_flags &= ~restore_flags;
783 ifr.ifr_flags |= netdev->save_flags & restore_flags;
784 if (ioctl(netdev->fd, SIOCSIFFLAGS, &ifr) < 0) {
792 /* Retores all the flags on all network devices that we modified. Called from
793 * a signal handler, so it does not attempt to report error conditions. */
795 restore_all_flags(void *aux UNUSED)
797 struct netdev *netdev;
798 LIST_FOR_EACH (netdev, struct netdev, node, &netdev_list) {
799 restore_flags(netdev);
804 get_flags(const struct netdev *netdev, int *flags)
807 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
808 if (ioctl(netdev->fd, SIOCGIFFLAGS, &ifr) < 0) {
809 VLOG_ERR("ioctl(SIOCGIFFLAGS) on %s device failed: %s",
810 netdev->name, strerror(errno));
813 *flags = ifr.ifr_flags;
818 set_flags(struct netdev *netdev, int flags)
821 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
822 ifr.ifr_flags = flags;
823 if (ioctl(netdev->fd, SIOCSIFFLAGS, &ifr) < 0) {
824 VLOG_ERR("ioctl(SIOCSIFFLAGS) on %s device failed: %s",
825 netdev->name, strerror(errno));