2 * Copyright (c) 2008, 2009 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
23 #include <netinet/in.h>
29 #include "dynamic-string.h"
30 #include "fatal-signal.h"
32 #include "netdev-provider.h"
35 #include "poll-loop.h"
39 #define THIS_MODULE VLM_netdev
42 static const struct netdev_class *netdev_classes[] = {
46 enum { N_NETDEV_CLASSES = ARRAY_SIZE(netdev_classes) };
48 /* All open network devices. */
49 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
51 /* This is set pretty low because we probably won't learn anything from the
52 * additional log messages. */
53 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
55 static void restore_all_flags(void *aux);
56 static int restore_flags(struct netdev *netdev);
58 /* Attempts to initialize the netdev module. Returns 0 if successful,
59 * otherwise a positive errno value.
61 * Calling this function is optional. If not called explicitly, it will
62 * automatically be called upon the first attempt to open a network device. */
64 netdev_initialize(void)
66 static int status = -1;
70 fatal_signal_add_hook(restore_all_flags, NULL, true);
73 for (i = 0; i < N_NETDEV_CLASSES; i++) {
74 const struct netdev_class *class = netdev_classes[i];
76 int retval = class->init();
78 VLOG_ERR("failed to initialize %s network device "
79 "class: %s", class->name, strerror(retval));
90 /* Performs periodic work needed by all the various kinds of netdevs.
92 * If your program opens any netdevs, it must call this function within its
98 for (i = 0; i < N_NETDEV_CLASSES; i++) {
99 const struct netdev_class *class = netdev_classes[i];
106 /* Arranges for poll_block() to wake up when netdev_run() needs to be called.
108 * If your program opens any netdevs, it must call this function within its
114 for (i = 0; i < N_NETDEV_CLASSES; i++) {
115 const struct netdev_class *class = netdev_classes[i];
122 /* Opens the network device named 'name' (e.g. "eth0") and returns zero if
123 * successful, otherwise a positive errno value. On success, sets '*netdevp'
124 * to the new network device, otherwise to null.
126 * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
127 * capture frames of that type received on the device. It may also be one of
128 * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those
131 netdev_open(const char *name_, int ethertype, struct netdev **netdevp)
133 char *name = xstrdup(name_);
134 char *prefix, *suffix, *colon;
135 struct netdev *netdev = NULL;
139 error = netdev_initialize();
144 colon = strchr(name, ':');
154 for (i = 0; i < N_NETDEV_CLASSES; i++) {
155 const struct netdev_class *class = netdev_classes[i];
156 if (!strcmp(prefix, class->prefix)) {
157 error = class->open(name_, suffix, ethertype, &netdev);
161 error = EAFNOSUPPORT;
164 *netdevp = error ? NULL : netdev;
168 /* Closes and destroys 'netdev'. */
170 netdev_close(struct netdev *netdev)
176 /* Restore flags that we changed, if any. */
177 fatal_signal_block();
178 error = restore_flags(netdev);
179 list_remove(&netdev->node);
180 fatal_signal_unblock();
182 VLOG_WARN("failed to restore network device flags on %s: %s",
183 netdev->name, strerror(error));
188 netdev->class->close(netdev);
193 /* Returns true if a network device named 'name' exists and may be opened,
194 * otherwise false. */
196 netdev_exists(const char *name)
198 struct netdev *netdev;
201 error = netdev_open(name, NETDEV_ETH_TYPE_NONE, &netdev);
203 netdev_close(netdev);
206 if (error != ENODEV) {
207 VLOG_WARN("failed to open network device %s: %s",
208 name, strerror(error));
214 /* Initializes 'svec' with a list of the names of all known network devices. */
216 netdev_enumerate(struct svec *svec)
223 error = netdev_initialize();
229 for (i = 0; i < N_NETDEV_CLASSES; i++) {
230 const struct netdev_class *class = netdev_classes[i];
231 if (class->enumerate) {
232 int retval = class->enumerate(svec);
234 VLOG_WARN("failed to enumerate %s network devices: %s",
235 class->name, strerror(retval));
245 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
246 * must have initialized with sufficient room for the packet. The space
247 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
248 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
249 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
250 * need not be included.)
252 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
253 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
254 * positive errno value. Returns EAGAIN immediately if no packet is ready to
258 netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
262 assert(buffer->size == 0);
263 assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
265 retval = netdev->class->recv(netdev,
266 buffer->data, ofpbuf_tailroom(buffer));
268 COVERAGE_INC(netdev_received);
269 buffer->size += retval;
270 if (buffer->size < ETH_TOTAL_MIN) {
271 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
279 /* Registers with the poll loop to wake up from the next call to poll_block()
280 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
282 netdev_recv_wait(struct netdev *netdev)
284 netdev->class->recv_wait(netdev);
287 /* Discards all packets waiting to be received from 'netdev'. */
289 netdev_drain(struct netdev *netdev)
291 return netdev->class->drain(netdev);
294 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
295 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
296 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
297 * the packet is too big or too small to transmit on the device.
299 * The caller retains ownership of 'buffer' in all cases.
301 * The kernel maintains a packet transmission queue, so the caller is not
302 * expected to do additional queuing of packets. */
304 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
306 int error = netdev->class->send(netdev, buffer->data, buffer->size);
308 COVERAGE_INC(netdev_sent);
313 /* Registers with the poll loop to wake up from the next call to poll_block()
314 * when the packet transmission queue has sufficient room to transmit a packet
315 * with netdev_send().
317 * The kernel maintains a packet transmission queue, so the client is not
318 * expected to do additional queuing of packets. Thus, this function is
319 * unlikely to ever be used. It is included for completeness. */
321 netdev_send_wait(struct netdev *netdev)
323 return netdev->class->send_wait(netdev);
326 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
327 * otherwise a positive errno value. */
329 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
331 return netdev->class->set_etheraddr(netdev, mac);
334 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
335 * the MAC address into 'mac'. On failure, returns a positive errno value and
336 * clears 'mac' to all-zeros. */
338 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
340 return netdev->class->get_etheraddr(netdev, mac);
343 /* Returns the name of the network device that 'netdev' represents,
344 * e.g. "eth0". The caller must not modify or free the returned string. */
346 netdev_get_name(const struct netdev *netdev)
351 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
352 * (and received) packets, in bytes, not including the hardware header; thus,
353 * this is typically 1500 bytes for Ethernet devices.
355 * If successful, returns 0 and stores the MTU size in '*mtup'. On failure,
356 * returns a positive errno value and stores ETH_PAYLOAD_MAX (1500) in
359 netdev_get_mtu(const struct netdev *netdev, int *mtup)
361 int error = netdev->class->get_mtu(netdev, mtup);
363 VLOG_WARN_RL(&rl, "failed to retrieve MTU for network device %s: %s",
364 netdev_get_name(netdev), strerror(error));
365 *mtup = ETH_PAYLOAD_MAX;
370 /* Stores the features supported by 'netdev' into each of '*current',
371 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
372 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
373 * successful, otherwise a positive errno value. On failure, all of the
374 * passed-in values are set to 0. */
376 netdev_get_features(struct netdev *netdev,
377 uint32_t *current, uint32_t *advertised,
378 uint32_t *supported, uint32_t *peer)
381 return netdev->class->get_features(netdev,
382 current ? current : &dummy[0],
383 advertised ? advertised : &dummy[1],
384 supported ? supported : &dummy[2],
385 peer ? peer : &dummy[3]);
388 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
389 * successful, otherwise a positive errno value. */
391 netdev_set_advertisements(struct netdev *netdev, uint32_t advertise)
393 return (netdev->class->set_advertisements
394 ? netdev->class->set_advertisements(netdev, advertise)
398 /* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address and
399 * returns 0. Otherwise, returns a positive errno value and sets '*in4' to 0
402 * The following error values have well-defined meanings:
404 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
406 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
408 * 'in4' may be null, in which case the address itself is not reported. */
410 netdev_get_in4(const struct netdev *netdev, struct in_addr *in4)
412 struct in_addr dummy;
415 error = (netdev->class->get_in4
416 ? netdev->class->get_in4(netdev, in4 ? in4 : &dummy)
424 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
425 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
426 * positive errno value. */
428 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
430 return (netdev->class->set_in4
431 ? netdev->class->set_in4(netdev, addr, mask)
435 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
438 netdev_add_router(struct netdev *netdev, struct in_addr router)
440 COVERAGE_INC(netdev_add_router);
441 return (netdev->class->add_router
442 ? netdev->class->add_router(netdev, router)
446 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
447 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
448 * all-zero-bits (in6addr_any).
450 * The following error values have well-defined meanings:
452 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
454 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
456 * 'in6' may be null, in which case the address itself is not reported. */
458 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
460 struct in6_addr dummy;
463 error = (netdev->class->get_in6
464 ? netdev->class->get_in6(netdev, in6 ? in6 : &dummy)
467 memset(in6, 0, sizeof *in6);
472 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
473 * 'on'. If 'permanent' is true, the changes will persist; otherwise, they
474 * will be reverted when 'netdev' is closed or the program exits. Returns 0 if
475 * successful, otherwise a positive errno value. */
477 do_update_flags(struct netdev *netdev, enum netdev_flags off,
478 enum netdev_flags on, enum netdev_flags *old_flagsp,
481 enum netdev_flags old_flags;
484 error = netdev->class->update_flags(netdev, off & ~on, on, &old_flags);
486 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
487 off || on ? "set" : "get", netdev_get_name(netdev),
490 } else if ((off || on) && !permanent) {
491 enum netdev_flags new_flags = (old_flags & ~off) | on;
492 enum netdev_flags changed_flags = old_flags ^ new_flags;
494 if (!netdev->changed_flags) {
495 netdev->save_flags = old_flags;
497 netdev->changed_flags |= changed_flags;
501 *old_flagsp = old_flags;
506 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
507 * Returns 0 if successful, otherwise a positive errno value. On failure,
508 * stores 0 into '*flagsp'. */
510 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
512 struct netdev *netdev = (struct netdev *) netdev_;
513 return do_update_flags(netdev, 0, 0, flagsp, false);
516 /* Sets the flags for 'netdev' to 'flags'.
517 * If 'permanent' is true, the changes will persist; otherwise, they
518 * will be reverted when 'netdev' is closed or the program exits.
519 * Returns 0 if successful, otherwise a positive errno value. */
521 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
524 return do_update_flags(netdev, -1, flags, NULL, permanent);
527 /* Turns on the specified 'flags' on 'netdev'.
528 * If 'permanent' is true, the changes will persist; otherwise, they
529 * will be reverted when 'netdev' is closed or the program exits.
530 * Returns 0 if successful, otherwise a positive errno value. */
532 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
535 return do_update_flags(netdev, 0, flags, NULL, permanent);
538 /* Turns off the specified 'flags' on 'netdev'.
539 * If 'permanent' is true, the changes will persist; otherwise, they
540 * will be reverted when 'netdev' is closed or the program exits.
541 * Returns 0 if successful, otherwise a positive errno value. */
543 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
546 return do_update_flags(netdev, flags, 0, NULL, permanent);
549 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
550 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
551 * returns 0. Otherwise, it returns a positive errno value; in particular,
552 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
554 netdev_arp_lookup(const struct netdev *netdev,
555 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
557 int error = (netdev->class->arp_lookup
558 ? netdev->class->arp_lookup(netdev, ip, mac)
561 memset(mac, 0, ETH_ADDR_LEN);
566 /* Sets 'carrier' to true if carrier is active (link light is on) on
569 netdev_get_carrier(const struct netdev *netdev, bool *carrier)
571 int error = (netdev->class->get_carrier
572 ? netdev->class->get_carrier(netdev, carrier)
580 /* Retrieves current device stats for 'netdev'. */
582 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
586 COVERAGE_INC(netdev_get_stats);
587 error = (netdev->class->get_stats
588 ? netdev->class->get_stats(netdev, stats)
591 memset(stats, 0xff, sizeof *stats);
596 /* Attempts to set input rate limiting (policing) policy, such that up to
597 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
598 * size of 'kbits' kb. */
600 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
601 uint32_t kbits_burst)
603 return (netdev->class->set_policing
604 ? netdev->class->set_policing(netdev, kbits_rate, kbits_burst)
608 /* If 'netdev' is a VLAN network device (e.g. one created with vconfig(8)),
609 * sets '*vlan_vid' to the VLAN VID associated with that device and returns 0.
610 * Otherwise returns a errno value (specifically ENOENT if 'netdev_name' is the
611 * name of a network device that is not a VLAN device) and sets '*vlan_vid' to
614 netdev_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
616 int error = (netdev->class->get_vlan_vid
617 ? netdev->class->get_vlan_vid(netdev, vlan_vid)
625 /* Returns a network device that has 'in4' as its IP address, if one exists,
626 * otherwise a null pointer. */
628 netdev_find_dev_by_in4(const struct in_addr *in4)
630 struct netdev *netdev;
631 struct svec dev_list;
634 netdev_enumerate(&dev_list);
635 for (i = 0; i < dev_list.n; i++) {
636 const char *name = dev_list.names[i];
637 struct in_addr dev_in4;
639 if (!netdev_open(name, NETDEV_ETH_TYPE_NONE, &netdev)
640 && !netdev_get_in4(netdev, &dev_in4)
641 && dev_in4.s_addr == in4->s_addr) {
644 netdev_close(netdev);
649 svec_destroy(&dev_list);
653 /* Initializes 'netdev' as a netdev named 'name' of the specified 'class'.
655 * This function adds 'netdev' to a netdev-owned linked list, so it is very
656 * important that 'netdev' only be freed after calling netdev_close(). */
658 netdev_init(struct netdev *netdev, const char *name,
659 const struct netdev_class *class)
661 netdev->class = class;
662 netdev->name = xstrdup(name);
663 netdev->save_flags = 0;
664 netdev->changed_flags = 0;
665 list_push_back(&netdev_list, &netdev->node);
668 /* Initializes 'notifier' as a netdev notifier for 'netdev', for which
669 * notification will consist of calling 'cb', with auxiliary data 'aux'. */
671 netdev_notifier_init(struct netdev_notifier *notifier, struct netdev *netdev,
672 void (*cb)(struct netdev_notifier *), void *aux)
674 notifier->netdev = netdev;
679 /* Tracks changes in the status of a set of network devices. */
680 struct netdev_monitor {
681 struct shash polled_netdevs;
682 struct shash changed_netdevs;
685 /* Creates and returns a new structure for monitor changes in the status of
686 * network devices. */
687 struct netdev_monitor *
688 netdev_monitor_create(void)
690 struct netdev_monitor *monitor = xmalloc(sizeof *monitor);
691 shash_init(&monitor->polled_netdevs);
692 shash_init(&monitor->changed_netdevs);
696 /* Destroys 'monitor'. */
698 netdev_monitor_destroy(struct netdev_monitor *monitor)
701 struct shash_node *node;
703 SHASH_FOR_EACH (node, &monitor->polled_netdevs) {
704 struct netdev_notifier *notifier = node->data;
705 notifier->netdev->class->poll_remove(notifier);
708 shash_destroy(&monitor->polled_netdevs);
709 shash_destroy(&monitor->changed_netdevs);
715 netdev_monitor_cb(struct netdev_notifier *notifier)
717 struct netdev_monitor *monitor = notifier->aux;
718 const char *name = netdev_get_name(notifier->netdev);
719 if (!shash_find(&monitor->changed_netdevs, name)) {
720 shash_add(&monitor->changed_netdevs, name, NULL);
724 /* Attempts to add 'netdev' as a netdev monitored by 'monitor'. Returns 0 if
725 * successful, otherwise a positive errno value.
727 * Adding a given 'netdev' to a monitor multiple times is equivalent to adding
730 netdev_monitor_add(struct netdev_monitor *monitor, struct netdev *netdev)
732 const char *netdev_name = netdev_get_name(netdev);
734 if (!shash_find(&monitor->polled_netdevs, netdev_name)
735 && netdev->class->poll_add)
737 struct netdev_notifier *notifier;
738 error = netdev->class->poll_add(netdev, netdev_monitor_cb, monitor,
741 assert(notifier->netdev == netdev);
742 shash_add(&monitor->polled_netdevs, netdev_name, notifier);
748 /* Removes 'netdev' from the set of netdevs monitored by 'monitor'. (This has
749 * no effect if 'netdev' is not in the set of devices monitored by
752 netdev_monitor_remove(struct netdev_monitor *monitor, struct netdev *netdev)
754 const char *netdev_name = netdev_get_name(netdev);
755 struct shash_node *node;
757 node = shash_find(&monitor->polled_netdevs, netdev_name);
759 /* Cancel future notifications. */
760 struct netdev_notifier *notifier = node->data;
761 netdev->class->poll_remove(notifier);
762 shash_delete(&monitor->polled_netdevs, node);
764 /* Drop any pending notification. */
765 node = shash_find(&monitor->changed_netdevs, netdev_name);
767 shash_delete(&monitor->changed_netdevs, node);
772 /* Checks for changes to netdevs in the set monitored by 'monitor'. If any of
773 * the attributes (Ethernet address, carrier status, speed or peer-advertised
774 * speed, flags, etc.) of a network device monitored by 'monitor' has changed,
775 * sets '*devnamep' to the name of a device that has changed and returns 0.
776 * The caller is responsible for freeing '*devnamep' (with free()).
778 * If no devices have changed, sets '*devnamep' to NULL and returns EAGAIN.
781 netdev_monitor_poll(struct netdev_monitor *monitor, char **devnamep)
783 struct shash_node *node = shash_first(&monitor->changed_netdevs);
788 *devnamep = xstrdup(node->name);
789 shash_delete(&monitor->changed_netdevs, node);
794 /* Registers with the poll loop to wake up from the next call to poll_block()
795 * when netdev_monitor_poll(monitor) would indicate that a device has
798 netdev_monitor_poll_wait(const struct netdev_monitor *monitor)
800 if (!shash_is_empty(&monitor->changed_netdevs)) {
801 poll_immediate_wake();
803 /* XXX Nothing needed here for netdev_linux, but maybe other netdev
804 * classes need help. */
808 /* Restore the network device flags on 'netdev' to those that were active
809 * before we changed them. Returns 0 if successful, otherwise a positive
812 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
814 restore_flags(struct netdev *netdev)
816 if (netdev->changed_flags) {
817 enum netdev_flags restore = netdev->save_flags & netdev->changed_flags;
818 enum netdev_flags old_flags;
819 return netdev->class->update_flags(netdev,
820 netdev->changed_flags & ~restore,
821 restore, &old_flags);
826 /* Retores all the flags on all network devices that we modified. Called from
827 * a signal handler, so it does not attempt to report error conditions. */
829 restore_all_flags(void *aux UNUSED)
831 struct netdev *netdev;
832 LIST_FOR_EACH (netdev, struct netdev, node, &netdev_list) {
833 restore_flags(netdev);