2 * Copyright (c) 2008, 2009, 2010 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"
33 #include "netdev-provider.h"
34 #include "netdev-vport.h"
36 #include "openflow/openflow.h"
38 #include "poll-loop.h"
43 VLOG_DEFINE_THIS_MODULE(netdev)
45 static struct shash netdev_classes = SHASH_INITIALIZER(&netdev_classes);
47 /* All created network devices. */
48 static struct shash netdev_dev_shash = SHASH_INITIALIZER(&netdev_dev_shash);
50 /* All open network devices. */
51 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
53 /* This is set pretty low because we probably won't learn anything from the
54 * additional log messages. */
55 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
57 static void close_all_netdevs(void *aux OVS_UNUSED);
58 static int restore_flags(struct netdev *netdev);
59 void update_device_args(struct netdev_dev *, const struct shash *args);
62 netdev_initialize(void)
69 fatal_signal_add_hook(close_all_netdevs, NULL, NULL, true);
72 netdev_register_provider(&netdev_linux_class);
73 netdev_register_provider(&netdev_tap_class);
74 netdev_vport_register();
79 /* Performs periodic work needed by all the various kinds of netdevs.
81 * If your program opens any netdevs, it must call this function within its
86 struct shash_node *node;
87 SHASH_FOR_EACH(node, &netdev_classes) {
88 const struct netdev_class *netdev_class = node->data;
89 if (netdev_class->run) {
95 /* Arranges for poll_block() to wake up when netdev_run() needs to be called.
97 * If your program opens any netdevs, it must call this function within its
102 struct shash_node *node;
103 SHASH_FOR_EACH(node, &netdev_classes) {
104 const struct netdev_class *netdev_class = node->data;
105 if (netdev_class->wait) {
106 netdev_class->wait();
111 /* Initializes and registers a new netdev provider. After successful
112 * registration, new netdevs of that type can be opened using netdev_open(). */
114 netdev_register_provider(const struct netdev_class *new_class)
116 if (shash_find(&netdev_classes, new_class->type)) {
117 VLOG_WARN("attempted to register duplicate netdev provider: %s",
122 if (new_class->init) {
123 int error = new_class->init();
125 VLOG_ERR("failed to initialize %s network device class: %s",
126 new_class->type, strerror(error));
131 shash_add(&netdev_classes, new_class->type, new_class);
136 /* Unregisters a netdev provider. 'type' must have been previously
137 * registered and not currently be in use by any netdevs. After unregistration
138 * new netdevs of that type cannot be opened using netdev_open(). */
140 netdev_unregister_provider(const char *type)
142 struct shash_node *del_node, *netdev_dev_node;
144 del_node = shash_find(&netdev_classes, type);
146 VLOG_WARN("attempted to unregister a netdev provider that is not "
147 "registered: %s", type);
151 SHASH_FOR_EACH(netdev_dev_node, &netdev_dev_shash) {
152 struct netdev_dev *netdev_dev = netdev_dev_node->data;
153 if (!strcmp(netdev_dev->netdev_class->type, type)) {
154 VLOG_WARN("attempted to unregister in use netdev provider: %s",
160 shash_delete(&netdev_classes, del_node);
165 /* Clears 'types' and enumerates the types of all currently registered netdev
166 * providers into it. The caller must first initialize the svec. */
168 netdev_enumerate_types(struct svec *types)
170 struct shash_node *node;
175 SHASH_FOR_EACH(node, &netdev_classes) {
176 const struct netdev_class *netdev_class = node->data;
177 svec_add(types, netdev_class->type);
181 /* Compares 'args' to those used to those used by 'dev'. Returns true
182 * if the arguments are the same, false otherwise. Does not update the
183 * values stored in 'dev'. */
185 compare_device_args(const struct netdev_dev *dev, const struct shash *args)
187 const struct shash_node **new_args;
191 if (shash_count(args) != dev->n_args) {
195 new_args = shash_sort(args);
196 for (i = 0; i < dev->n_args; i++) {
197 if (strcmp(dev->args[i].key, new_args[i]->name) ||
198 strcmp(dev->args[i].value, new_args[i]->data)) {
210 compare_args(const void *a_, const void *b_)
212 const struct arg *a = a_;
213 const struct arg *b = b_;
214 return strcmp(a->key, b->key);
218 update_device_args(struct netdev_dev *dev, const struct shash *args)
220 struct shash_node *node;
224 for (i = 0; i < dev->n_args; i++) {
225 free(dev->args[i].key);
226 free(dev->args[i].value);
233 if (!args || shash_is_empty(args)) {
237 dev->n_args = shash_count(args);
238 dev->args = xmalloc(dev->n_args * sizeof *dev->args);
241 SHASH_FOR_EACH(node, args) {
242 dev->args[i].key = xstrdup(node->name);
243 dev->args[i].value = xstrdup(node->data);
247 qsort(dev->args, dev->n_args, sizeof *dev->args, compare_args);
251 create_device(struct netdev_options *options, struct netdev_dev **netdev_devp)
253 struct netdev_class *netdev_class;
255 if (!options->type || strlen(options->type) == 0) {
256 /* Default to system. */
257 options->type = "system";
260 netdev_class = shash_find_data(&netdev_classes, options->type);
265 return netdev_class->create(netdev_class, options->name, options->args,
269 /* Opens the network device named 'name' (e.g. "eth0") and returns zero if
270 * successful, otherwise a positive errno value. On success, sets '*netdevp'
271 * to the new network device, otherwise to null.
273 * If this is the first time the device has been opened, then create is called
274 * before opening. The device is created using the given type and arguments.
276 * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
277 * capture frames of that type received on the device. It may also be one of
278 * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those
281 netdev_open(struct netdev_options *options, struct netdev **netdevp)
283 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
284 struct netdev_dev *netdev_dev;
290 if (!options->args) {
291 options->args = &empty_args;
294 netdev_dev = shash_find_data(&netdev_dev_shash, options->name);
297 error = create_device(options, &netdev_dev);
299 if (error == EAFNOSUPPORT) {
300 VLOG_WARN("could not create netdev %s of unknown type %s",
301 options->name, options->type);
305 update_device_args(netdev_dev, options->args);
307 } else if (!shash_is_empty(options->args) &&
308 !compare_device_args(netdev_dev, options->args)) {
310 VLOG_WARN("%s: attempted to open already open netdev with "
311 "different arguments", options->name);
315 error = netdev_dev->netdev_class->open(netdev_dev, options->ethertype,
319 netdev_dev->ref_cnt++;
321 if (!netdev_dev->ref_cnt) {
322 netdev_dev_uninit(netdev_dev, true);
330 netdev_open_default(const char *name, struct netdev **netdevp)
332 struct netdev_options options;
334 memset(&options, 0, sizeof options);
336 options.ethertype = NETDEV_ETH_TYPE_NONE;
338 return netdev_open(&options, netdevp);
341 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
342 * or NULL if none are needed. */
344 netdev_reconfigure(struct netdev *netdev, const struct shash *args)
346 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
347 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
353 if (netdev_dev->netdev_class->reconfigure) {
354 if (!compare_device_args(netdev_dev, args)) {
355 update_device_args(netdev_dev, args);
356 return netdev_dev->netdev_class->reconfigure(netdev_dev, args);
358 } else if (!shash_is_empty(args)) {
359 VLOG_WARN("%s: arguments provided to device that does not have a "
360 "reconfigure function", netdev_get_name(netdev));
366 /* Closes and destroys 'netdev'. */
368 netdev_close(struct netdev *netdev)
371 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
373 assert(netdev_dev->ref_cnt);
374 netdev_dev->ref_cnt--;
375 netdev_uninit(netdev, true);
377 /* If the reference count for the netdev device is zero, destroy it. */
378 if (!netdev_dev->ref_cnt) {
379 netdev_dev_uninit(netdev_dev, true);
384 /* Returns true if a network device named 'name' exists and may be opened,
385 * otherwise false. */
387 netdev_exists(const char *name)
389 struct netdev *netdev;
392 error = netdev_open_default(name, &netdev);
394 netdev_close(netdev);
397 if (error != ENODEV) {
398 VLOG_WARN("failed to open network device %s: %s",
399 name, strerror(error));
405 /* Returns true if a network device named 'name' is currently opened,
406 * otherwise false. */
408 netdev_is_open(const char *name)
410 return !!shash_find_data(&netdev_dev_shash, name);
413 /* Clears 'svec' and enumerates the names of all known network devices. */
415 netdev_enumerate(struct svec *svec)
417 struct shash_node *node;
423 SHASH_FOR_EACH(node, &netdev_classes) {
424 const struct netdev_class *netdev_class = node->data;
425 if (netdev_class->enumerate) {
426 int retval = netdev_class->enumerate(svec);
428 VLOG_WARN("failed to enumerate %s network devices: %s",
429 netdev_class->type, strerror(retval));
440 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
441 * must have initialized with sufficient room for the packet. The space
442 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
443 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
444 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
445 * need not be included.)
447 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
448 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
449 * positive errno value. Returns EAGAIN immediately if no packet is ready to
452 * Some network devices may not implement support for this function. In such
453 * cases this function will always return EOPNOTSUPP. */
455 netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
457 int (*recv)(struct netdev *, void *, size_t);
460 assert(buffer->size == 0);
461 assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
463 recv = netdev_get_dev(netdev)->netdev_class->recv;
465 ? (recv)(netdev, buffer->data, ofpbuf_tailroom(buffer))
468 COVERAGE_INC(netdev_received);
469 buffer->size += retval;
470 if (buffer->size < ETH_TOTAL_MIN) {
471 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
479 /* Registers with the poll loop to wake up from the next call to poll_block()
480 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
482 netdev_recv_wait(struct netdev *netdev)
484 void (*recv_wait)(struct netdev *);
486 recv_wait = netdev_get_dev(netdev)->netdev_class->recv_wait;
492 /* Discards all packets waiting to be received from 'netdev'. */
494 netdev_drain(struct netdev *netdev)
496 int (*drain)(struct netdev *);
498 drain = netdev_get_dev(netdev)->netdev_class->drain;
499 return drain ? drain(netdev) : 0;
502 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
503 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
504 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
505 * the packet is too big or too small to transmit on the device.
507 * The caller retains ownership of 'buffer' in all cases.
509 * The kernel maintains a packet transmission queue, so the caller is not
510 * expected to do additional queuing of packets.
512 * Some network devices may not implement support for this function. In such
513 * cases this function will always return EOPNOTSUPP. */
515 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
517 int (*send)(struct netdev *, const void *, size_t);
520 send = netdev_get_dev(netdev)->netdev_class->send;
521 error = send ? (send)(netdev, buffer->data, buffer->size) : EOPNOTSUPP;
523 COVERAGE_INC(netdev_sent);
528 /* Registers with the poll loop to wake up from the next call to poll_block()
529 * when the packet transmission queue has sufficient room to transmit a packet
530 * with netdev_send().
532 * The kernel maintains a packet transmission queue, so the client is not
533 * expected to do additional queuing of packets. Thus, this function is
534 * unlikely to ever be used. It is included for completeness. */
536 netdev_send_wait(struct netdev *netdev)
538 void (*send_wait)(struct netdev *);
540 send_wait = netdev_get_dev(netdev)->netdev_class->send_wait;
546 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
547 * otherwise a positive errno value. */
549 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
551 return netdev_get_dev(netdev)->netdev_class->set_etheraddr(netdev, mac);
554 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
555 * the MAC address into 'mac'. On failure, returns a positive errno value and
556 * clears 'mac' to all-zeros. */
558 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
560 return netdev_get_dev(netdev)->netdev_class->get_etheraddr(netdev, mac);
563 /* Returns the name of the network device that 'netdev' represents,
564 * e.g. "eth0". The caller must not modify or free the returned string. */
566 netdev_get_name(const struct netdev *netdev)
568 return netdev_get_dev(netdev)->name;
571 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
572 * (and received) packets, in bytes, not including the hardware header; thus,
573 * this is typically 1500 bytes for Ethernet devices.
575 * If successful, returns 0 and stores the MTU size in '*mtup'. On failure,
576 * returns a positive errno value and stores ETH_PAYLOAD_MAX (1500) in
579 netdev_get_mtu(const struct netdev *netdev, int *mtup)
581 int error = netdev_get_dev(netdev)->netdev_class->get_mtu(netdev, mtup);
583 VLOG_WARN_RL(&rl, "failed to retrieve MTU for network device %s: %s",
584 netdev_get_name(netdev), strerror(error));
585 *mtup = ETH_PAYLOAD_MAX;
590 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
591 * failure, returns a negative errno value.
593 * The desired semantics of the ifindex value are a combination of those
594 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
595 * value should be unique within a host and remain stable at least until
596 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
597 * but many systems do not follow this rule anyhow.
599 * Some network devices may not implement support for this function. In such
600 * cases this function will always return -EOPNOTSUPP.
603 netdev_get_ifindex(const struct netdev *netdev)
605 int (*get_ifindex)(const struct netdev *);
607 get_ifindex = netdev_get_dev(netdev)->netdev_class->get_ifindex;
609 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
612 /* Stores the features supported by 'netdev' into each of '*current',
613 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
614 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
615 * successful, otherwise a positive errno value. On failure, all of the
616 * passed-in values are set to 0.
618 * Some network devices may not implement support for this function. In such
619 * cases this function will always return EOPNOTSUPP. */
621 netdev_get_features(struct netdev *netdev,
622 uint32_t *current, uint32_t *advertised,
623 uint32_t *supported, uint32_t *peer)
625 int (*get_features)(struct netdev *netdev,
626 uint32_t *current, uint32_t *advertised,
627 uint32_t *supported, uint32_t *peer);
635 advertised = &dummy[1];
638 supported = &dummy[2];
644 get_features = netdev_get_dev(netdev)->netdev_class->get_features;
646 ? get_features(netdev, current, advertised, supported, peer)
649 *current = *advertised = *supported = *peer = 0;
654 /* Returns the maximum speed of a network connection that has the "enum
655 * ofp_port_features" bits in 'features', in bits per second. If no bits that
656 * indicate a speed are set in 'features', assumes 100Mbps. */
658 netdev_features_to_bps(uint32_t features)
661 F_10000MB = OFPPF_10GB_FD,
662 F_1000MB = OFPPF_1GB_HD | OFPPF_1GB_FD,
663 F_100MB = OFPPF_100MB_HD | OFPPF_100MB_FD,
664 F_10MB = OFPPF_10MB_HD | OFPPF_10MB_FD
667 return ( features & F_10000MB ? UINT64_C(10000000000)
668 : features & F_1000MB ? UINT64_C(1000000000)
669 : features & F_100MB ? UINT64_C(100000000)
670 : features & F_10MB ? UINT64_C(10000000)
671 : UINT64_C(100000000));
674 /* Returns true if any of the "enum ofp_port_features" bits that indicate a
675 * full-duplex link are set in 'features', otherwise false. */
677 netdev_features_is_full_duplex(uint32_t features)
679 return (features & (OFPPF_10MB_FD | OFPPF_100MB_FD | OFPPF_1GB_FD
680 | OFPPF_10GB_FD)) != 0;
683 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
684 * successful, otherwise a positive errno value. */
686 netdev_set_advertisements(struct netdev *netdev, uint32_t advertise)
688 return (netdev_get_dev(netdev)->netdev_class->set_advertisements
689 ? netdev_get_dev(netdev)->netdev_class->set_advertisements(
694 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
695 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
696 * errno value and sets '*address' to 0 (INADDR_ANY).
698 * The following error values have well-defined meanings:
700 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
702 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
704 * 'address' or 'netmask' or both may be null, in which case the address or
705 * netmask is not reported. */
707 netdev_get_in4(const struct netdev *netdev,
708 struct in_addr *address_, struct in_addr *netmask_)
710 struct in_addr address;
711 struct in_addr netmask;
714 error = (netdev_get_dev(netdev)->netdev_class->get_in4
715 ? netdev_get_dev(netdev)->netdev_class->get_in4(netdev,
719 address_->s_addr = error ? 0 : address.s_addr;
722 netmask_->s_addr = error ? 0 : netmask.s_addr;
727 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
728 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
729 * positive errno value. */
731 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
733 return (netdev_get_dev(netdev)->netdev_class->set_in4
734 ? netdev_get_dev(netdev)->netdev_class->set_in4(netdev, addr, mask)
738 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
741 netdev_add_router(struct netdev *netdev, struct in_addr router)
743 COVERAGE_INC(netdev_add_router);
744 return (netdev_get_dev(netdev)->netdev_class->add_router
745 ? netdev_get_dev(netdev)->netdev_class->add_router(netdev, router)
749 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
750 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
751 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
752 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
753 * a directly connected network) in '*next_hop' and a copy of the name of the
754 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
755 * responsible for freeing '*netdev_name' (by calling free()). */
757 netdev_get_next_hop(const struct netdev *netdev,
758 const struct in_addr *host, struct in_addr *next_hop,
761 int error = (netdev_get_dev(netdev)->netdev_class->get_next_hop
762 ? netdev_get_dev(netdev)->netdev_class->get_next_hop(
763 host, next_hop, netdev_name)
766 next_hop->s_addr = 0;
772 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
773 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
774 * all-zero-bits (in6addr_any).
776 * The following error values have well-defined meanings:
778 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
780 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
782 * 'in6' may be null, in which case the address itself is not reported. */
784 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
786 struct in6_addr dummy;
789 error = (netdev_get_dev(netdev)->netdev_class->get_in6
790 ? netdev_get_dev(netdev)->netdev_class->get_in6(netdev,
794 memset(in6, 0, sizeof *in6);
799 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
800 * 'on'. If 'permanent' is true, the changes will persist; otherwise, they
801 * will be reverted when 'netdev' is closed or the program exits. Returns 0 if
802 * successful, otherwise a positive errno value. */
804 do_update_flags(struct netdev *netdev, enum netdev_flags off,
805 enum netdev_flags on, enum netdev_flags *old_flagsp,
808 enum netdev_flags old_flags;
811 error = netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
812 off & ~on, on, &old_flags);
814 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
815 off || on ? "set" : "get", netdev_get_name(netdev),
818 } else if ((off || on) && !permanent) {
819 enum netdev_flags new_flags = (old_flags & ~off) | on;
820 enum netdev_flags changed_flags = old_flags ^ new_flags;
822 if (!netdev->changed_flags) {
823 netdev->save_flags = old_flags;
825 netdev->changed_flags |= changed_flags;
829 *old_flagsp = old_flags;
834 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
835 * Returns 0 if successful, otherwise a positive errno value. On failure,
836 * stores 0 into '*flagsp'. */
838 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
840 struct netdev *netdev = (struct netdev *) netdev_;
841 return do_update_flags(netdev, 0, 0, flagsp, false);
844 /* Sets the flags for 'netdev' to 'flags'.
845 * If 'permanent' is true, the changes will persist; otherwise, they
846 * will be reverted when 'netdev' is closed or the program exits.
847 * Returns 0 if successful, otherwise a positive errno value. */
849 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
852 return do_update_flags(netdev, -1, flags, NULL, permanent);
855 /* Turns on the specified 'flags' on 'netdev'.
856 * If 'permanent' is true, the changes will persist; otherwise, they
857 * will be reverted when 'netdev' is closed or the program exits.
858 * Returns 0 if successful, otherwise a positive errno value. */
860 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
863 return do_update_flags(netdev, 0, flags, NULL, permanent);
866 /* Turns off the specified 'flags' on 'netdev'.
867 * If 'permanent' is true, the changes will persist; otherwise, they
868 * will be reverted when 'netdev' is closed or the program exits.
869 * Returns 0 if successful, otherwise a positive errno value. */
871 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
874 return do_update_flags(netdev, flags, 0, NULL, permanent);
877 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
878 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
879 * returns 0. Otherwise, it returns a positive errno value; in particular,
880 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
882 netdev_arp_lookup(const struct netdev *netdev,
883 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
885 int error = (netdev_get_dev(netdev)->netdev_class->arp_lookup
886 ? netdev_get_dev(netdev)->netdev_class->arp_lookup(netdev,
890 memset(mac, 0, ETH_ADDR_LEN);
895 /* Sets 'carrier' to true if carrier is active (link light is on) on
898 netdev_get_carrier(const struct netdev *netdev, bool *carrier)
900 int error = (netdev_get_dev(netdev)->netdev_class->get_carrier
901 ? netdev_get_dev(netdev)->netdev_class->get_carrier(netdev,
910 /* Retrieves current device stats for 'netdev'. */
912 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
916 COVERAGE_INC(netdev_get_stats);
917 error = (netdev_get_dev(netdev)->netdev_class->get_stats
918 ? netdev_get_dev(netdev)->netdev_class->get_stats(netdev, stats)
921 memset(stats, 0xff, sizeof *stats);
926 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
927 * Returns 0 if successful, otherwise a positive errno value.
929 * This will probably fail for most network devices. Some devices might only
930 * allow setting their stats to 0. */
932 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
934 return (netdev_get_dev(netdev)->netdev_class->set_stats
935 ? netdev_get_dev(netdev)->netdev_class->set_stats(netdev, stats)
939 /* Attempts to set input rate limiting (policing) policy, such that up to
940 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
941 * size of 'kbits' kb. */
943 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
944 uint32_t kbits_burst)
946 return (netdev_get_dev(netdev)->netdev_class->set_policing
947 ? netdev_get_dev(netdev)->netdev_class->set_policing(netdev,
948 kbits_rate, kbits_burst)
952 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
953 * empty if 'netdev' does not support QoS. Any names added to 'types' should
954 * be documented as valid for the "type" column in the "QoS" table in
955 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
957 * Every network device supports disabling QoS with a type of "", but this type
958 * will not be added to 'types'.
960 * The caller must initialize 'types' (e.g. with svec_init()) before calling
961 * this function. The caller is responsible for destroying 'types' (e.g. with
962 * svec_destroy()) when it is no longer needed.
964 * Returns 0 if successful, otherwise a positive errno value. */
966 netdev_get_qos_types(const struct netdev *netdev, struct svec *types)
968 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
969 return (class->get_qos_types
970 ? class->get_qos_types(netdev, types)
974 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
975 * which should be "" or one of the types returned by netdev_get_qos_types()
976 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
977 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
978 * 'caps' to all zeros. */
980 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
981 struct netdev_qos_capabilities *caps)
983 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
986 int retval = (class->get_qos_capabilities
987 ? class->get_qos_capabilities(netdev, type, caps)
990 memset(caps, 0, sizeof *caps);
994 /* Every netdev supports turning off QoS. */
995 memset(caps, 0, sizeof *caps);
1000 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1001 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1002 * the number of queues (zero on failure) in '*n_queuesp'.
1004 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1006 netdev_get_n_queues(const struct netdev *netdev,
1007 const char *type, unsigned int *n_queuesp)
1009 struct netdev_qos_capabilities caps;
1012 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1013 *n_queuesp = caps.n_queues;
1017 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1018 * stores the name of the current form of QoS into '*typep', stores any details
1019 * of configuration as string key-value pairs in 'details', and returns 0. On
1020 * failure, sets '*typep' to NULL and returns a positive errno value.
1022 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1024 * The caller must initialize 'details' as an empty shash (e.g. with
1025 * shash_init()) before calling this function. The caller must free 'details',
1026 * including 'data' members, when it is no longer needed (e.g. with
1027 * shash_destroy_free_data()).
1029 * The caller must not modify or free '*typep'.
1031 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1032 * 'netdev'. The contents of 'details' should be documented as valid for
1033 * '*typep' in the "other_config" column in the "QoS" table in
1034 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1036 netdev_get_qos(const struct netdev *netdev,
1037 const char **typep, struct shash *details)
1039 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1042 if (class->get_qos) {
1043 retval = class->get_qos(netdev, typep, details);
1046 shash_clear_free_data(details);
1050 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1056 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1057 * with details of configuration from 'details'. Returns 0 if successful,
1058 * otherwise a positive errno value. On error, the previous QoS configuration
1061 * When this function changes the type of QoS (not just 'details'), this also
1062 * resets all queue configuration for 'netdev' to their defaults (which depend
1063 * on the specific type of QoS). Otherwise, the queue configuration for
1064 * 'netdev' is unchanged.
1066 * 'type' should be "" (to disable QoS) or one of the types returned by
1067 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1068 * documented as valid for the given 'type' in the "other_config" column in the
1069 * "QoS" table in vswitchd/vswitch.xml (which is built as
1070 * ovs-vswitchd.conf.db(8)).
1072 * NULL may be specified for 'details' if there are no configuration
1075 netdev_set_qos(struct netdev *netdev,
1076 const char *type, const struct shash *details)
1078 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1084 if (class->set_qos) {
1086 static struct shash empty = SHASH_INITIALIZER(&empty);
1089 return class->set_qos(netdev, type, details);
1091 return *type ? EOPNOTSUPP : 0;
1095 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1096 * successful, adds that information as string key-value pairs to 'details'.
1097 * Returns 0 if successful, otherwise a positive errno value.
1099 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1100 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1102 * The returned contents of 'details' should be documented as valid for the
1103 * given 'type' in the "other_config" column in the "Queue" table in
1104 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1106 * The caller must initialize 'details' (e.g. with shash_init()) before calling
1107 * this function. The caller must free 'details', including 'data' members,
1108 * when it is no longer needed (e.g. with shash_destroy_free_data()). */
1110 netdev_get_queue(const struct netdev *netdev,
1111 unsigned int queue_id, struct shash *details)
1113 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1116 retval = (class->get_queue
1117 ? class->get_queue(netdev, queue_id, details)
1120 shash_clear_free_data(details);
1125 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1126 * string pairs in 'details'. The contents of 'details' should be documented
1127 * as valid for the given 'type' in the "other_config" column in the "Queue"
1128 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1129 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1130 * given queue's configuration should be unmodified.
1132 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1133 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1135 * This function does not modify 'details', and the caller retains ownership of
1138 netdev_set_queue(struct netdev *netdev,
1139 unsigned int queue_id, const struct shash *details)
1141 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1142 return (class->set_queue
1143 ? class->set_queue(netdev, queue_id, details)
1147 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1148 * of QoS may have a fixed set of queues, in which case attempts to delete them
1149 * will fail with EOPNOTSUPP.
1151 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1152 * given queue will be unmodified.
1154 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1155 * the current form of QoS (e.g. as returned by
1156 * netdev_get_n_queues(netdev)). */
1158 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1160 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1161 return (class->delete_queue
1162 ? class->delete_queue(netdev, queue_id)
1166 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1167 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1168 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1169 * positive errno value and fills 'stats' with all-1-bits. */
1171 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1172 struct netdev_queue_stats *stats)
1174 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1177 retval = (class->get_queue_stats
1178 ? class->get_queue_stats(netdev, queue_id, stats)
1181 memset(stats, 0xff, sizeof *stats);
1186 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1187 * its configuration, and the 'aux' specified by the caller. The order of
1188 * iteration is unspecified, but (when successful) each queue is visited
1191 * Calling this function may be more efficient than calling netdev_get_queue()
1194 * 'cb' must not modify or free the 'details' argument passed in.
1196 * Returns 0 if successful, otherwise a positive errno value. On error, some
1197 * configured queues may not have been included in the iteration. */
1199 netdev_dump_queues(const struct netdev *netdev,
1200 netdev_dump_queues_cb *cb, void *aux)
1202 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1203 return (class->dump_queues
1204 ? class->dump_queues(netdev, cb, aux)
1208 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1209 * its statistics, and the 'aux' specified by the caller. The order of
1210 * iteration is unspecified, but (when successful) each queue is visited
1213 * Calling this function may be more efficient than calling
1214 * netdev_get_queue_stats() for every queue.
1216 * 'cb' must not modify or free the statistics passed in.
1218 * Returns 0 if successful, otherwise a positive errno value. On error, some
1219 * configured queues may not have been included in the iteration. */
1221 netdev_dump_queue_stats(const struct netdev *netdev,
1222 netdev_dump_queue_stats_cb *cb, void *aux)
1224 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1225 return (class->dump_queue_stats
1226 ? class->dump_queue_stats(netdev, cb, aux)
1230 /* If 'netdev' is a VLAN network device (e.g. one created with vconfig(8)),
1231 * sets '*vlan_vid' to the VLAN VID associated with that device and returns 0.
1232 * Otherwise returns a errno value (specifically ENOENT if 'netdev_name' is the
1233 * name of a network device that is not a VLAN device) and sets '*vlan_vid' to
1236 netdev_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1238 int error = (netdev_get_dev(netdev)->netdev_class->get_vlan_vid
1239 ? netdev_get_dev(netdev)->netdev_class->get_vlan_vid(netdev,
1248 /* Returns a network device that has 'in4' as its IP address, if one exists,
1249 * otherwise a null pointer. */
1251 netdev_find_dev_by_in4(const struct in_addr *in4)
1253 struct netdev *netdev;
1254 struct svec dev_list = SVEC_EMPTY_INITIALIZER;
1257 netdev_enumerate(&dev_list);
1258 for (i = 0; i < dev_list.n; i++) {
1259 const char *name = dev_list.names[i];
1260 struct in_addr dev_in4;
1262 if (!netdev_open_default(name, &netdev)
1263 && !netdev_get_in4(netdev, &dev_in4, NULL)
1264 && dev_in4.s_addr == in4->s_addr) {
1267 netdev_close(netdev);
1272 svec_destroy(&dev_list);
1276 /* Initializes 'netdev_dev' as a netdev device named 'name' of the
1277 * specified 'netdev_class'.
1279 * This function adds 'netdev_dev' to a netdev-owned shash, so it is
1280 * very important that 'netdev_dev' only be freed after calling
1281 * the refcount drops to zero. */
1283 netdev_dev_init(struct netdev_dev *netdev_dev, const char *name,
1284 const struct netdev_class *netdev_class)
1286 assert(!shash_find(&netdev_dev_shash, name));
1288 memset(netdev_dev, 0, sizeof *netdev_dev);
1289 netdev_dev->netdev_class = netdev_class;
1290 netdev_dev->name = xstrdup(name);
1291 netdev_dev->node = shash_add(&netdev_dev_shash, name, netdev_dev);
1294 /* Undoes the results of initialization.
1296 * Normally this function does not need to be called as netdev_close has
1297 * the same effect when the refcount drops to zero.
1298 * However, it may be called by providers due to an error on creation
1299 * that occurs after initialization. It this case netdev_close() would
1300 * never be called. */
1302 netdev_dev_uninit(struct netdev_dev *netdev_dev, bool destroy)
1304 char *name = netdev_dev->name;
1306 assert(!netdev_dev->ref_cnt);
1308 shash_delete(&netdev_dev_shash, netdev_dev->node);
1309 update_device_args(netdev_dev, NULL);
1312 netdev_dev->netdev_class->destroy(netdev_dev);
1317 /* Returns the class type of 'netdev_dev'.
1319 * The caller must not free the returned value. */
1321 netdev_dev_get_type(const struct netdev_dev *netdev_dev)
1323 return netdev_dev->netdev_class->type;
1326 /* Returns the class associated with 'netdev_dev'. */
1327 const struct netdev_class *
1328 netdev_dev_get_class(const struct netdev_dev *netdev_dev)
1330 return netdev_dev->netdev_class;
1333 /* Returns the name of 'netdev_dev'.
1335 * The caller must not free the returned value. */
1337 netdev_dev_get_name(const struct netdev_dev *netdev_dev)
1339 return netdev_dev->name;
1342 /* Returns the netdev_dev with 'name' or NULL if there is none.
1344 * The caller must not free the returned value. */
1346 netdev_dev_from_name(const char *name)
1348 return shash_find_data(&netdev_dev_shash, name);
1351 /* Fills 'device_list' with devices that match 'netdev_class'.
1353 * The caller is responsible for initializing and destroying 'device_list'
1354 * but the contained netdev_devs must not be freed. */
1356 netdev_dev_get_devices(const struct netdev_class *netdev_class,
1357 struct shash *device_list)
1359 struct shash_node *node;
1360 SHASH_FOR_EACH (node, &netdev_dev_shash) {
1361 struct netdev_dev *dev = node->data;
1363 if (dev->netdev_class == netdev_class) {
1364 shash_add(device_list, node->name, node->data);
1369 /* Initializes 'netdev' as a instance of the netdev_dev.
1371 * This function adds 'netdev' to a netdev-owned linked list, so it is very
1372 * important that 'netdev' only be freed after calling netdev_close(). */
1374 netdev_init(struct netdev *netdev, struct netdev_dev *netdev_dev)
1376 memset(netdev, 0, sizeof *netdev);
1377 netdev->netdev_dev = netdev_dev;
1378 list_push_back(&netdev_list, &netdev->node);
1381 /* Undoes the results of initialization.
1383 * Normally this function only needs to be called from netdev_close().
1384 * However, it may be called by providers due to an error on opening
1385 * that occurs after initialization. It this case netdev_close() would
1386 * never be called. */
1388 netdev_uninit(struct netdev *netdev, bool close)
1390 /* Restore flags that we changed, if any. */
1391 int error = restore_flags(netdev);
1392 list_remove(&netdev->node);
1394 VLOG_WARN("failed to restore network device flags on %s: %s",
1395 netdev_get_name(netdev), strerror(error));
1399 netdev_get_dev(netdev)->netdev_class->close(netdev);
1404 /* Returns the class type of 'netdev'.
1406 * The caller must not free the returned value. */
1408 netdev_get_type(const struct netdev *netdev)
1410 return netdev_get_dev(netdev)->netdev_class->type;
1414 netdev_get_dev(const struct netdev *netdev)
1416 return netdev->netdev_dev;
1419 /* Initializes 'notifier' as a netdev notifier for 'netdev', for which
1420 * notification will consist of calling 'cb', with auxiliary data 'aux'. */
1422 netdev_notifier_init(struct netdev_notifier *notifier, struct netdev *netdev,
1423 void (*cb)(struct netdev_notifier *), void *aux)
1425 notifier->netdev = netdev;
1427 notifier->aux = aux;
1430 /* Tracks changes in the status of a set of network devices. */
1431 struct netdev_monitor {
1432 struct shash polled_netdevs;
1433 struct shash changed_netdevs;
1436 /* Creates and returns a new structure for monitor changes in the status of
1437 * network devices. */
1438 struct netdev_monitor *
1439 netdev_monitor_create(void)
1441 struct netdev_monitor *monitor = xmalloc(sizeof *monitor);
1442 shash_init(&monitor->polled_netdevs);
1443 shash_init(&monitor->changed_netdevs);
1447 /* Destroys 'monitor'. */
1449 netdev_monitor_destroy(struct netdev_monitor *monitor)
1452 struct shash_node *node;
1454 SHASH_FOR_EACH (node, &monitor->polled_netdevs) {
1455 struct netdev_notifier *notifier = node->data;
1456 netdev_get_dev(notifier->netdev)->netdev_class->poll_remove(
1460 shash_destroy(&monitor->polled_netdevs);
1461 shash_destroy(&monitor->changed_netdevs);
1467 netdev_monitor_cb(struct netdev_notifier *notifier)
1469 struct netdev_monitor *monitor = notifier->aux;
1470 const char *name = netdev_get_name(notifier->netdev);
1471 shash_add_once(&monitor->changed_netdevs, name, NULL);
1474 /* Attempts to add 'netdev' as a netdev monitored by 'monitor'. Returns 0 if
1475 * successful, otherwise a positive errno value.
1477 * Adding a given 'netdev' to a monitor multiple times is equivalent to adding
1480 netdev_monitor_add(struct netdev_monitor *monitor, struct netdev *netdev)
1482 const char *netdev_name = netdev_get_name(netdev);
1484 if (!shash_find(&monitor->polled_netdevs, netdev_name)
1485 && netdev_get_dev(netdev)->netdev_class->poll_add)
1487 struct netdev_notifier *notifier;
1488 error = netdev_get_dev(netdev)->netdev_class->poll_add(netdev,
1489 netdev_monitor_cb, monitor, ¬ifier);
1491 assert(notifier->netdev == netdev);
1492 shash_add(&monitor->polled_netdevs, netdev_name, notifier);
1498 /* Removes 'netdev' from the set of netdevs monitored by 'monitor'. (This has
1499 * no effect if 'netdev' is not in the set of devices monitored by
1502 netdev_monitor_remove(struct netdev_monitor *monitor, struct netdev *netdev)
1504 const char *netdev_name = netdev_get_name(netdev);
1505 struct shash_node *node;
1507 node = shash_find(&monitor->polled_netdevs, netdev_name);
1509 /* Cancel future notifications. */
1510 struct netdev_notifier *notifier = node->data;
1511 netdev_get_dev(netdev)->netdev_class->poll_remove(notifier);
1512 shash_delete(&monitor->polled_netdevs, node);
1514 /* Drop any pending notification. */
1515 node = shash_find(&monitor->changed_netdevs, netdev_name);
1517 shash_delete(&monitor->changed_netdevs, node);
1522 /* Checks for changes to netdevs in the set monitored by 'monitor'. If any of
1523 * the attributes (Ethernet address, carrier status, speed or peer-advertised
1524 * speed, flags, etc.) of a network device monitored by 'monitor' has changed,
1525 * sets '*devnamep' to the name of a device that has changed and returns 0.
1526 * The caller is responsible for freeing '*devnamep' (with free()).
1528 * If no devices have changed, sets '*devnamep' to NULL and returns EAGAIN. */
1530 netdev_monitor_poll(struct netdev_monitor *monitor, char **devnamep)
1532 struct shash_node *node = shash_first(&monitor->changed_netdevs);
1537 *devnamep = shash_steal(&monitor->changed_netdevs, node);
1542 /* Registers with the poll loop to wake up from the next call to poll_block()
1543 * when netdev_monitor_poll(monitor) would indicate that a device has
1546 netdev_monitor_poll_wait(const struct netdev_monitor *monitor)
1548 if (!shash_is_empty(&monitor->changed_netdevs)) {
1549 poll_immediate_wake();
1551 /* XXX Nothing needed here for netdev_linux, but maybe other netdev
1552 * classes need help. */
1556 /* Restore the network device flags on 'netdev' to those that were active
1557 * before we changed them. Returns 0 if successful, otherwise a positive
1560 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
1562 restore_flags(struct netdev *netdev)
1564 if (netdev->changed_flags) {
1565 enum netdev_flags restore = netdev->save_flags & netdev->changed_flags;
1566 enum netdev_flags old_flags;
1567 return netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
1568 netdev->changed_flags & ~restore,
1569 restore, &old_flags);
1574 /* Close all netdevs on shutdown so they can do any needed cleanup such as
1575 * destroying devices, restoring flags, etc. */
1577 close_all_netdevs(void *aux OVS_UNUSED)
1579 struct netdev *netdev, *next;
1580 LIST_FOR_EACH_SAFE(netdev, next, node, &netdev_list) {
1581 netdev_close(netdev);