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];
77 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
79 static void init_netdev(void);
80 static int restore_flags(struct netdev *netdev);
82 /* Check whether device NAME has an IPv4 address assigned to it and, if so, log
85 check_ipv4_address(const char *name)
90 sock = socket(AF_INET, SOCK_DGRAM, 0);
92 VLOG_WARN("socket(AF_INET): %s", strerror(errno));
96 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
97 ifr.ifr_addr.sa_family = AF_INET;
98 if (ioctl(sock, SIOCGIFADDR, &ifr) == 0) {
99 VLOG_ERR("%s device has assigned IP address %s", name,
100 inet_ntoa(((struct sockaddr_in*) &ifr.ifr_addr)->sin_addr));
106 /* Check whether device NAME has an IPv6 address assigned to it and, if so, log
109 check_ipv6_address(const char *name)
114 file = fopen("/proc/net/if_inet6", "r");
119 while (fgets(line, sizeof line, file)) {
121 uint8_t *s6 = in6.s6_addr;
125 if (sscanf(line, " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
126 "%*x %*x %*x %*x %16s\n",
127 &s6[0], &s6[1], &s6[2], &s6[3],
128 &s6[4], &s6[5], &s6[6], &s6[7],
129 &s6[8], &s6[9], &s6[10], &s6[11],
130 &s6[12], &s6[13], &s6[14], &s6[15],
132 && !strcmp(name, ifname))
134 char in6_name[INET6_ADDRSTRLEN + 1];
135 inet_ntop(AF_INET6, &in6, in6_name, sizeof in6_name);
136 VLOG_ERR("%s device has assigned IPv6 address %s",
145 do_ethtool(struct netdev *netdev)
148 struct ethtool_cmd ecmd;
151 netdev->features = 0;
153 memset(&ifr, 0, sizeof ifr);
154 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
155 ifr.ifr_data = (caddr_t) &ecmd;
157 memset(&ecmd, 0, sizeof ecmd);
158 ecmd.cmd = ETHTOOL_GSET;
159 if (ioctl(netdev->fd, SIOCETHTOOL, &ifr) == 0) {
160 if (ecmd.supported & SUPPORTED_10baseT_Half) {
161 netdev->features |= OFPPF_10MB_HD;
163 if (ecmd.supported & SUPPORTED_10baseT_Full) {
164 netdev->features |= OFPPF_10MB_FD;
166 if (ecmd.supported & SUPPORTED_100baseT_Half) {
167 netdev->features |= OFPPF_100MB_HD;
169 if (ecmd.supported & SUPPORTED_100baseT_Full) {
170 netdev->features |= OFPPF_100MB_FD;
172 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
173 netdev->features |= OFPPF_1GB_HD;
175 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
176 netdev->features |= OFPPF_1GB_FD;
178 /* 10Gbps half-duplex doesn't exist... */
179 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
180 netdev->features |= OFPPF_10GB_FD;
183 switch (ecmd.speed) {
193 netdev->speed = 1000;
197 netdev->speed = 2500;
201 netdev->speed = 10000;
205 VLOG_DBG("ioctl(SIOCETHTOOL) failed: %s", strerror(errno));
209 /* Opens the network device named 'name' (e.g. "eth0") and returns zero if
210 * successful, otherwise a positive errno value. On success, sets '*netdev'
211 * to the new network device, otherwise to null. */
213 netdev_open(const char *name, struct netdev **netdev_)
218 unsigned int ifindex;
219 socklen_t rcvbuf_len;
221 uint8_t etheraddr[ETH_ADDR_LEN];
224 struct netdev *netdev;
229 /* Create raw socket.
231 * We have to use SOCK_PACKET, despite its deprecation, because only
232 * SOCK_PACKET lets us set the hardware source address of outgoing
234 fd = socket(PF_PACKET, SOCK_PACKET, htons(ETH_P_ALL));
239 /* Bind to specific ethernet device. */
240 memset(&sa, 0, sizeof sa);
241 sa.sa_family = AF_UNSPEC;
242 strncpy((char *) sa.sa_data, name, sizeof sa.sa_data);
243 if (bind(fd, &sa, sizeof sa) < 0) {
244 VLOG_ERR("bind to %s failed: %s", name, strerror(errno));
248 /* Between the socket() and bind() calls above, the socket receives all
249 * packets on all system interfaces. We do not want to receive that
250 * data, but there is no way to avoid it. So we must now drain out the
251 * receive queue. There is no way to know how long the receive queue is,
252 * but we know that the total number of bytes queued does not exceed the
253 * receive buffer size, so we pull packets until none are left or we've
254 * read that many bytes. */
255 rcvbuf_len = sizeof rcvbuf;
256 if (getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, &rcvbuf_len) < 0) {
257 VLOG_ERR("getsockopt(SO_RCVBUF) on %s device failed: %s",
258 name, strerror(errno));
263 ssize_t n_bytes = recv(fd, &buffer, 1, MSG_TRUNC | MSG_DONTWAIT);
270 /* Get ethernet device index. */
271 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
272 if (ioctl(fd, SIOCGIFINDEX, &ifr) < 0) {
273 VLOG_ERR("ioctl(SIOCGIFINDEX) on %s device failed: %s",
274 name, strerror(errno));
277 ifindex = ifr.ifr_ifindex;
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 /* Allocate network device. */
301 netdev = xmalloc(sizeof *netdev);
302 netdev->name = xstrdup(name);
304 memcpy(netdev->etheraddr, etheraddr, sizeof etheraddr);
307 /* Get speed, features. */
310 /* Save flags to restore at close or exit. */
311 if (ioctl(fd, SIOCGIFFLAGS, &ifr) < 0) {
312 VLOG_ERR("ioctl(SIOCGIFFLAGS) on %s device failed: %s",
313 name, strerror(errno));
316 netdev->save_flags = ifr.ifr_flags;
317 fatal_signal_block();
318 list_push_back(&netdev_list, &netdev->node);
319 fatal_signal_unblock();
321 /* Bring up interface and set promiscuous mode. */
322 ifr.ifr_flags |= IFF_PROMISC | IFF_UP;
323 if (ioctl(fd, SIOCSIFFLAGS, &ifr) < 0) {
325 VLOG_ERR("failed to set promiscuous mode on %s device: %s",
326 name, strerror(errno));
327 netdev_close(netdev);
331 /* Complain to administrator if any IP addresses are assigned to the
332 * interface. We warn about this because packets received for that IP
333 * address will be processed both by the kernel TCP/IP stack and by us as a
334 * switch, which produces poor results. */
335 check_ipv4_address(name);
336 check_ipv6_address(name);
348 /* Closes and destroys 'netdev'. */
350 netdev_close(struct netdev *netdev)
353 /* Bring down interface and drop promiscuous mode, if we brought up
354 * the interface or enabled promiscuous mode. */
356 fatal_signal_block();
357 error = restore_flags(netdev);
358 list_remove(&netdev->node);
359 fatal_signal_unblock();
361 VLOG_WARN("failed to restore network device flags on %s: %s",
362 netdev->name, strerror(error));
372 /* Pads 'buffer' out with zero-bytes to the minimum valid length of an
373 * Ethernet packet, if necessary. */
375 pad_to_minimum_length(struct buffer *buffer)
377 if (buffer->size < ETH_TOTAL_MIN) {
378 size_t shortage = ETH_TOTAL_MIN - buffer->size;
379 memset(buffer_put_uninit(buffer, shortage), 0, shortage);
383 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
384 * must have initialized with sufficient room for the packet. The space
385 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
386 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
387 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
388 * need not be included.)
390 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
391 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
392 * positive errno value. Returns EAGAIN immediately if no packet is ready to
396 netdev_recv(struct netdev *netdev, struct buffer *buffer)
400 assert(buffer->size == 0);
401 assert(buffer_tailroom(buffer) >= ETH_TOTAL_MIN);
403 n_bytes = recv(netdev->fd,
404 buffer_tail(buffer), buffer_tailroom(buffer),
406 } while (n_bytes < 0 && errno == EINTR);
408 if (errno != EAGAIN) {
409 VLOG_WARN("error receiving Ethernet packet on %s: %s",
410 strerror(errno), netdev->name);
414 buffer->size += n_bytes;
416 /* When the kernel internally sends out an Ethernet frame on an
417 * interface, it gives us a copy *before* padding the frame to the
418 * minimum length. Thus, when it sends out something like an ARP
419 * request, we see a too-short frame. So pad it out to the minimum
421 pad_to_minimum_length(buffer);
426 /* Registers with the poll loop to wake up from the next call to poll_block()
427 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
429 netdev_recv_wait(struct netdev *netdev)
431 poll_fd_wait(netdev->fd, POLLIN);
434 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
435 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
436 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
437 * the packet is too big to transmit on the device.
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, struct buffer *buffer)
445 const struct eth_header *eh;
446 struct sockaddr_pkt spkt;
448 /* Ensure packet is long enough. (Although all incoming packets are at
449 * least ETH_TOTAL_MIN bytes long, we could have trimmed some data off a
450 * minimum-size packet, e.g. by dropping a vlan header.)
452 * The kernel does not require this, but it ensures that we always access
453 * valid memory in grabbing the sockaddr below. */
454 pad_to_minimum_length(buffer);
456 /* Construct packet sockaddr, which SOCK_PACKET requires. */
457 spkt.spkt_family = AF_PACKET;
458 strncpy((char *) spkt.spkt_device, netdev->name, sizeof spkt.spkt_device);
459 eh = buffer_at_assert(buffer, 0, sizeof *eh);
460 spkt.spkt_protocol = eh->eth_type;
463 n_bytes = sendto(netdev->fd, buffer->data, buffer->size, 0,
464 (const struct sockaddr *) &spkt, sizeof spkt);
465 } while (n_bytes < 0 && errno == EINTR);
468 /* The Linux AF_PACKET implementation never blocks waiting for room
469 * for packets, instead returning ENOBUFS. Translate this into EAGAIN
471 if (errno == ENOBUFS) {
473 } else if (errno != EAGAIN) {
474 VLOG_WARN("error sending Ethernet packet on %s: %s",
475 netdev->name, strerror(errno));
478 } else if (n_bytes != buffer->size) {
479 VLOG_WARN("send partial Ethernet packet (%d bytes of %zu) on %s",
480 (int) n_bytes, buffer->size, netdev->name);
487 /* Registers with the poll loop to wake up from the next call to poll_block()
488 * when the packet transmission queue has sufficient room to transmit a packet
489 * with netdev_send().
491 * The kernel maintains a packet transmission queue, so the client is not
492 * expected to do additional queuing of packets. Thus, this function is
493 * unlikely to ever be used. It is included for completeness. */
495 netdev_send_wait(struct netdev *netdev)
497 poll_fd_wait(netdev->fd, POLLOUT);
500 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
501 * free the returned buffer. */
503 netdev_get_etheraddr(const struct netdev *netdev)
505 return netdev->etheraddr;
508 /* Returns the name of the network device that 'netdev' represents,
509 * e.g. "eth0". The caller must not modify or free the returned string. */
511 netdev_get_name(const struct netdev *netdev)
516 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
517 * in bytes, not including the hardware header; thus, this is typically 1500
518 * bytes for Ethernet devices. */
520 netdev_get_mtu(const struct netdev *netdev)
525 /* Returns the current speed of the network device that 'netdev' represents, in
526 * megabits per second, or 0 if the speed is unknown. */
528 netdev_get_speed(const struct netdev *netdev)
530 return netdev->speed;
533 /* Returns the features supported by 'netdev', as a bitmap of bits from enum
534 * ofp_phy_port, in host byte order. */
536 netdev_get_features(const struct netdev *netdev)
538 return netdev->features;
541 static void restore_all_flags(void *aux);
543 /* Set up a signal hook to restore network device flags on program
551 fatal_signal_add_hook(restore_all_flags, NULL);
555 /* Restore the network device flags on 'netdev' to those that were active
556 * before we changed them. Returns 0 if successful, otherwise a positive
559 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
561 restore_flags(struct netdev *netdev)
565 /* Get current flags. */
566 strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
567 if (ioctl(netdev->fd, SIOCGIFFLAGS, &ifr) < 0) {
571 /* Restore flags that we might have changed, if necessary. */
572 if ((ifr.ifr_flags ^ netdev->save_flags) & (IFF_PROMISC | IFF_UP)) {
573 ifr.ifr_flags &= ~(IFF_PROMISC | IFF_UP);
574 ifr.ifr_flags |= netdev->save_flags & (IFF_PROMISC | IFF_UP);
575 if (ioctl(netdev->fd, SIOCSIFFLAGS, &ifr) < 0) {
583 /* Retores all the flags on all network devices that we modified. Called from
584 * a signal handler, so it does not attempt to report error conditions. */
586 restore_all_flags(void *aux UNUSED)
588 struct netdev *netdev;
589 LIST_FOR_EACH (netdev, struct netdev, node, &netdev_list) {
590 restore_flags(netdev);