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"
63 #define THIS_MODULE VLM_netdev
70 uint8_t etheraddr[ETH_ADDR_LEN];
79 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
81 static void init_netdev(void);
82 static int restore_flags(struct netdev *netdev);
83 static int get_flags(const struct netdev *, int *flagsp);
84 static int set_flags(struct netdev *, int flags);
86 /* Obtains the IPv4 address for 'name' into 'in4'. Returns true if
89 get_ipv4_address(const char *name, struct in_addr *in4)
94 sock = socket(AF_INET, SOCK_DGRAM, 0);
96 VLOG_WARN("socket(AF_INET): %s", strerror(errno));
100 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
101 ifr.ifr_addr.sa_family = AF_INET;
102 if (ioctl(sock, SIOCGIFADDR, &ifr) == 0) {
103 struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
104 *in4 = sin->sin_addr;
106 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 netdev_open(const char *name, struct netdev **netdev_)
224 unsigned int ifindex;
225 socklen_t rcvbuf_len;
227 uint8_t etheraddr[ETH_ADDR_LEN];
232 struct netdev *netdev;
237 /* Create raw socket.
239 * We have to use SOCK_PACKET, despite its deprecation, because only
240 * SOCK_PACKET lets us set the hardware source address of outgoing
242 fd = socket(PF_PACKET, SOCK_PACKET, htons(ETH_P_ALL));
247 /* Bind to specific ethernet device. */
248 memset(&sa, 0, sizeof sa);
249 sa.sa_family = AF_UNSPEC;
250 strncpy((char *) sa.sa_data, name, sizeof sa.sa_data);
251 if (bind(fd, &sa, sizeof sa) < 0) {
252 VLOG_ERR("bind to %s failed: %s", name, strerror(errno));
256 /* Between the socket() and bind() calls above, the socket receives all
257 * packets on all system interfaces. We do not want to receive that
258 * data, but there is no way to avoid it. So we must now drain out the
259 * receive queue. There is no way to know how long the receive queue is,
260 * but we know that the total number of bytes queued does not exceed the
261 * receive buffer size, so we pull packets until none are left or we've
262 * read that many bytes. */
263 rcvbuf_len = sizeof rcvbuf;
264 if (getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, &rcvbuf_len) < 0) {
265 VLOG_ERR("getsockopt(SO_RCVBUF) on %s device failed: %s",
266 name, strerror(errno));
271 ssize_t n_bytes = recv(fd, &buffer, 1, MSG_TRUNC | MSG_DONTWAIT);
278 /* Get ethernet device index. */
279 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
280 if (ioctl(fd, SIOCGIFINDEX, &ifr) < 0) {
281 VLOG_ERR("ioctl(SIOCGIFINDEX) on %s device failed: %s",
282 name, strerror(errno));
285 ifindex = ifr.ifr_ifindex;
287 /* Get MAC address. */
288 if (ioctl(fd, SIOCGIFHWADDR, &ifr) < 0) {
289 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
290 name, strerror(errno));
293 if (ifr.ifr_hwaddr.sa_family != AF_UNSPEC
294 && ifr.ifr_hwaddr.sa_family != ARPHRD_ETHER) {
295 VLOG_WARN("%s device has unknown hardware address family %d",
296 name, (int) ifr.ifr_hwaddr.sa_family);
298 memcpy(etheraddr, ifr.ifr_hwaddr.sa_data, sizeof etheraddr);
301 if (ioctl(fd, SIOCGIFMTU, &ifr) < 0) {
302 VLOG_ERR("ioctl(SIOCGIFMTU) on %s device failed: %s",
303 name, strerror(errno));
308 if (!get_ipv4_address(name, &in4)) {
311 get_ipv6_address(name, &in6);
313 /* Allocate network device. */
314 netdev = xmalloc(sizeof *netdev);
315 netdev->name = xstrdup(name);
317 memcpy(netdev->etheraddr, etheraddr, sizeof etheraddr);
322 /* Get speed, features. */
325 /* Save flags to restore at close or exit. */
326 error = get_flags(netdev, &netdev->save_flags);
330 fatal_signal_block();
331 list_push_back(&netdev_list, &netdev->node);
332 fatal_signal_unblock();
345 /* Closes and destroys 'netdev'. */
347 netdev_close(struct netdev *netdev)
350 /* Bring down interface and drop promiscuous mode, if we brought up
351 * the interface or enabled promiscuous mode. */
353 fatal_signal_block();
354 error = restore_flags(netdev);
355 list_remove(&netdev->node);
356 fatal_signal_unblock();
358 VLOG_WARN("failed to restore network device flags on %s: %s",
359 netdev->name, strerror(error));
369 /* Pads 'buffer' out with zero-bytes to the minimum valid length of an
370 * Ethernet packet, if necessary. */
372 pad_to_minimum_length(struct buffer *buffer)
374 if (buffer->size < ETH_TOTAL_MIN) {
375 size_t shortage = ETH_TOTAL_MIN - buffer->size;
376 memset(buffer_put_uninit(buffer, shortage), 0, shortage);
380 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
381 * must have initialized with sufficient room for the packet. The space
382 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
383 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
384 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
385 * need not be included.)
387 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
388 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
389 * positive errno value. Returns EAGAIN immediately if no packet is ready to
393 netdev_recv(struct netdev *netdev, struct buffer *buffer)
397 assert(buffer->size == 0);
398 assert(buffer_tailroom(buffer) >= ETH_TOTAL_MIN);
400 n_bytes = recv(netdev->fd,
401 buffer_tail(buffer), buffer_tailroom(buffer),
403 } while (n_bytes < 0 && errno == EINTR);
405 if (errno != EAGAIN) {
406 VLOG_WARN("error receiving Ethernet packet on %s: %s",
407 strerror(errno), netdev->name);
411 buffer->size += n_bytes;
413 /* When the kernel internally sends out an Ethernet frame on an
414 * interface, it gives us a copy *before* padding the frame to the
415 * minimum length. Thus, when it sends out something like an ARP
416 * request, we see a too-short frame. So pad it out to the minimum
418 pad_to_minimum_length(buffer);
423 /* Registers with the poll loop to wake up from the next call to poll_block()
424 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
426 netdev_recv_wait(struct netdev *netdev)
428 poll_fd_wait(netdev->fd, POLLIN);
431 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
432 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
433 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
434 * the packet is too big to transmit on the device.
436 * The kernel maintains a packet transmission queue, so the caller is not
437 * expected to do additional queuing of packets. */
439 netdev_send(struct netdev *netdev, struct buffer *buffer)
442 const struct eth_header *eh;
443 struct sockaddr_pkt spkt;
445 /* Ensure packet is long enough. (Although all incoming packets are at
446 * least ETH_TOTAL_MIN bytes long, we could have trimmed some data off a
447 * minimum-size packet, e.g. by dropping a vlan header.)
449 * The kernel does not require this, but it ensures that we always access
450 * valid memory in grabbing the sockaddr below. */
451 pad_to_minimum_length(buffer);
453 /* Construct packet sockaddr, which SOCK_PACKET requires. */
454 spkt.spkt_family = AF_PACKET;
455 strncpy((char *) spkt.spkt_device, netdev->name, sizeof spkt.spkt_device);
456 eh = buffer_at_assert(buffer, 0, sizeof *eh);
457 spkt.spkt_protocol = eh->eth_type;
460 n_bytes = sendto(netdev->fd, buffer->data, buffer->size, 0,
461 (const struct sockaddr *) &spkt, sizeof spkt);
462 } while (n_bytes < 0 && errno == EINTR);
465 /* The Linux AF_PACKET implementation never blocks waiting for room
466 * for packets, instead returning ENOBUFS. Translate this into EAGAIN
468 if (errno == ENOBUFS) {
470 } else if (errno != EAGAIN) {
471 VLOG_WARN("error sending Ethernet packet on %s: %s",
472 netdev->name, strerror(errno));
475 } else if (n_bytes != buffer->size) {
476 VLOG_WARN("send partial Ethernet packet (%d bytes of %zu) on %s",
477 (int) n_bytes, buffer->size, netdev->name);
484 /* Registers with the poll loop to wake up from the next call to poll_block()
485 * when the packet transmission queue has sufficient room to transmit a packet
486 * with netdev_send().
488 * The kernel maintains a packet transmission queue, so the client is not
489 * expected to do additional queuing of packets. Thus, this function is
490 * unlikely to ever be used. It is included for completeness. */
492 netdev_send_wait(struct netdev *netdev)
494 poll_fd_wait(netdev->fd, POLLOUT);
497 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
498 * free the returned buffer. */
500 netdev_get_etheraddr(const struct netdev *netdev)
502 return netdev->etheraddr;
505 /* Returns the name of the network device that 'netdev' represents,
506 * e.g. "eth0". The caller must not modify or free the returned string. */
508 netdev_get_name(const struct netdev *netdev)
513 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
514 * in bytes, not including the hardware header; thus, this is typically 1500
515 * bytes for Ethernet devices. */
517 netdev_get_mtu(const struct netdev *netdev)
522 /* Returns the current speed of the network device that 'netdev' represents, in
523 * megabits per second, or 0 if the speed is unknown. */
525 netdev_get_speed(const struct netdev *netdev)
527 return netdev->speed;
530 /* Returns the features supported by 'netdev', as a bitmap of bits from enum
531 * ofp_phy_port, in host byte order. */
533 netdev_get_features(const struct netdev *netdev)
535 return netdev->features;
538 /* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address and
539 * returns true. Otherwise, returns false. */
541 netdev_get_in4(const struct netdev *netdev, struct in_addr *in4)
544 return in4->s_addr != INADDR_ANY;
547 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
548 * returns true. Otherwise, returns false. */
550 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
553 return memcmp(in6, &in6addr_any, sizeof *in6) != 0;
556 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
557 * Returns 0 if successful, otherwise a positive errno value. */
559 netdev_get_flags(const struct netdev *netdev, enum netdev_flags *flagsp)
563 error = get_flags(netdev, &flags);
569 if (flags & IFF_UP) {
570 *flagsp |= NETDEV_UP;
572 if (flags & IFF_PROMISC) {
573 *flagsp |= NETDEV_PROMISC;
578 /* Sets the flags for 'netdev' to 'nd_flags'.
579 * Returns 0 if successful, otherwise a positive errno value. */
581 netdev_set_flags(struct netdev *netdev, enum netdev_flags nd_flags)
583 int old_flags, new_flags;
586 error = get_flags(netdev, &old_flags);
591 new_flags = old_flags & ~(IFF_UP | IFF_PROMISC);
592 if (nd_flags & NETDEV_UP) {
595 if (nd_flags & NETDEV_PROMISC) {
596 new_flags |= IFF_PROMISC;
598 if (new_flags != old_flags) {
599 error = set_flags(netdev, new_flags);
604 static void restore_all_flags(void *aux);
606 /* Set up a signal hook to restore network device flags on program
614 fatal_signal_add_hook(restore_all_flags, NULL);
618 /* Restore the network device flags on 'netdev' to those that were active
619 * before we changed them. Returns 0 if successful, otherwise a positive
622 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
624 restore_flags(struct netdev *netdev)
628 /* Get current flags. */
629 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
630 if (ioctl(netdev->fd, SIOCGIFFLAGS, &ifr) < 0) {
634 /* Restore flags that we might have changed, if necessary. */
635 if ((ifr.ifr_flags ^ netdev->save_flags) & (IFF_PROMISC | IFF_UP)) {
636 ifr.ifr_flags &= ~(IFF_PROMISC | IFF_UP);
637 ifr.ifr_flags |= netdev->save_flags & (IFF_PROMISC | IFF_UP);
638 if (ioctl(netdev->fd, SIOCSIFFLAGS, &ifr) < 0) {
646 /* Retores all the flags on all network devices that we modified. Called from
647 * a signal handler, so it does not attempt to report error conditions. */
649 restore_all_flags(void *aux UNUSED)
651 struct netdev *netdev;
652 LIST_FOR_EACH (netdev, struct netdev, node, &netdev_list) {
653 restore_flags(netdev);
658 get_flags(const struct netdev *netdev, int *flags)
661 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
662 if (ioctl(netdev->fd, SIOCGIFFLAGS, &ifr) < 0) {
663 VLOG_ERR("ioctl(SIOCGIFFLAGS) on %s device failed: %s",
664 netdev->name, strerror(errno));
667 *flags = ifr.ifr_flags;
672 set_flags(struct netdev *netdev, int flags)
675 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
676 ifr.ifr_flags = flags;
677 if (ioctl(netdev->fd, SIOCSIFFLAGS, &ifr) < 0) {
678 VLOG_ERR("ioctl(SIOCSIFFLAGS) on %s device failed: %s",
679 netdev->name, strerror(errno));