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"
35 #include "openflow/openflow.h"
37 #include "poll-loop.h"
42 VLOG_DEFINE_THIS_MODULE(netdev)
44 static const struct netdev_class *base_netdev_classes[] = {
54 static struct shash netdev_classes = SHASH_INITIALIZER(&netdev_classes);
56 /* All created network devices. */
57 static struct shash netdev_dev_shash = SHASH_INITIALIZER(&netdev_dev_shash);
59 /* All open network devices. */
60 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
62 /* This is set pretty low because we probably won't learn anything from the
63 * additional log messages. */
64 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
66 static void close_all_netdevs(void *aux OVS_UNUSED);
67 static int restore_flags(struct netdev *netdev);
68 void update_device_args(struct netdev_dev *, const struct shash *args);
71 netdev_initialize(void)
73 static int status = -1;
78 fatal_signal_add_hook(close_all_netdevs, NULL, NULL, true);
81 for (i = 0; i < ARRAY_SIZE(base_netdev_classes); i++) {
82 netdev_register_provider(base_netdev_classes[i]);
87 /* Performs periodic work needed by all the various kinds of netdevs.
89 * If your program opens any netdevs, it must call this function within its
94 struct shash_node *node;
95 SHASH_FOR_EACH(node, &netdev_classes) {
96 const struct netdev_class *netdev_class = node->data;
97 if (netdev_class->run) {
103 /* Arranges for poll_block() to wake up when netdev_run() needs to be called.
105 * If your program opens any netdevs, it must call this function within its
110 struct shash_node *node;
111 SHASH_FOR_EACH(node, &netdev_classes) {
112 const struct netdev_class *netdev_class = node->data;
113 if (netdev_class->wait) {
114 netdev_class->wait();
119 /* Initializes and registers a new netdev provider. After successful
120 * registration, new netdevs of that type can be opened using netdev_open(). */
122 netdev_register_provider(const struct netdev_class *new_class)
124 struct netdev_class *new_provider;
126 if (shash_find(&netdev_classes, new_class->type)) {
127 VLOG_WARN("attempted to register duplicate netdev provider: %s",
132 if (new_class->init) {
133 int error = new_class->init();
135 VLOG_ERR("failed to initialize %s network device class: %s",
136 new_class->type, strerror(error));
141 new_provider = xmalloc(sizeof *new_provider);
142 memcpy(new_provider, new_class, sizeof *new_provider);
144 shash_add(&netdev_classes, new_class->type, new_provider);
149 /* Unregisters a netdev provider. 'type' must have been previously
150 * registered and not currently be in use by any netdevs. After unregistration
151 * new netdevs of that type cannot be opened using netdev_open(). */
153 netdev_unregister_provider(const char *type)
155 struct shash_node *del_node, *netdev_dev_node;
157 del_node = shash_find(&netdev_classes, type);
159 VLOG_WARN("attempted to unregister a netdev provider that is not "
160 "registered: %s", type);
164 SHASH_FOR_EACH(netdev_dev_node, &netdev_dev_shash) {
165 struct netdev_dev *netdev_dev = netdev_dev_node->data;
166 if (!strcmp(netdev_dev->netdev_class->type, type)) {
167 VLOG_WARN("attempted to unregister in use netdev provider: %s",
173 shash_delete(&netdev_classes, del_node);
174 free(del_node->data);
179 /* Clears 'types' and enumerates the types of all currently registered netdev
180 * providers into it. The caller must first initialize the svec. */
182 netdev_enumerate_types(struct svec *types)
184 struct shash_node *node;
189 SHASH_FOR_EACH(node, &netdev_classes) {
190 const struct netdev_class *netdev_class = node->data;
191 svec_add(types, netdev_class->type);
195 /* Compares 'args' to those used to those used by 'dev'. Returns true
196 * if the arguments are the same, false otherwise. Does not update the
197 * values stored in 'dev'. */
199 compare_device_args(const struct netdev_dev *dev, const struct shash *args)
201 const struct shash_node **new_args;
205 if (shash_count(args) != dev->n_args) {
209 new_args = shash_sort(args);
210 for (i = 0; i < dev->n_args; i++) {
211 if (strcmp(dev->args[i].key, new_args[i]->name) ||
212 strcmp(dev->args[i].value, new_args[i]->data)) {
224 compare_args(const void *a_, const void *b_)
226 const struct arg *a = a_;
227 const struct arg *b = b_;
228 return strcmp(a->key, b->key);
232 update_device_args(struct netdev_dev *dev, const struct shash *args)
234 struct shash_node *node;
238 for (i = 0; i < dev->n_args; i++) {
239 free(dev->args[i].key);
240 free(dev->args[i].value);
247 if (!args || shash_is_empty(args)) {
251 dev->n_args = shash_count(args);
252 dev->args = xmalloc(dev->n_args * sizeof *dev->args);
255 SHASH_FOR_EACH(node, args) {
256 dev->args[i].key = xstrdup(node->name);
257 dev->args[i].value = xstrdup(node->data);
261 qsort(dev->args, dev->n_args, sizeof *dev->args, compare_args);
265 create_device(struct netdev_options *options, struct netdev_dev **netdev_devp)
267 struct netdev_class *netdev_class;
269 if (!options->type || strlen(options->type) == 0) {
270 /* Default to system. */
271 options->type = "system";
274 netdev_class = shash_find_data(&netdev_classes, options->type);
279 return netdev_class->create(options->name, options->type, options->args,
283 /* Opens the network device named 'name' (e.g. "eth0") and returns zero if
284 * successful, otherwise a positive errno value. On success, sets '*netdevp'
285 * to the new network device, otherwise to null.
287 * If this is the first time the device has been opened, then create is called
288 * before opening. The device is created using the given type and arguments.
290 * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
291 * capture frames of that type received on the device. It may also be one of
292 * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those
296 netdev_open(struct netdev_options *options, struct netdev **netdevp)
298 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
299 struct netdev_dev *netdev_dev;
305 if (!options->args) {
306 options->args = &empty_args;
309 netdev_dev = shash_find_data(&netdev_dev_shash, options->name);
312 error = create_device(options, &netdev_dev);
314 if (error == EAFNOSUPPORT) {
315 VLOG_WARN("could not create netdev %s of unknown type %s",
316 options->name, options->type);
320 update_device_args(netdev_dev, options->args);
322 } else if (!shash_is_empty(options->args) &&
323 !compare_device_args(netdev_dev, options->args)) {
325 VLOG_WARN("%s: attempted to open already open netdev with "
326 "different arguments", options->name);
330 error = netdev_dev->netdev_class->open(netdev_dev, options->ethertype,
334 netdev_dev->ref_cnt++;
336 if (!netdev_dev->ref_cnt) {
337 netdev_dev_uninit(netdev_dev, true);
345 netdev_open_default(const char *name, struct netdev **netdevp)
347 struct netdev_options options;
349 memset(&options, 0, sizeof options);
351 options.ethertype = NETDEV_ETH_TYPE_NONE;
353 return netdev_open(&options, netdevp);
356 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
357 * or NULL if none are needed. */
359 netdev_reconfigure(struct netdev *netdev, const struct shash *args)
361 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
362 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
368 if (netdev_dev->netdev_class->reconfigure) {
369 if (!compare_device_args(netdev_dev, args)) {
370 update_device_args(netdev_dev, args);
371 return netdev_dev->netdev_class->reconfigure(netdev_dev, args);
373 } else if (!shash_is_empty(args)) {
374 VLOG_WARN("%s: arguments provided to device that does not have a "
375 "reconfigure function", netdev_get_name(netdev));
381 /* Closes and destroys 'netdev'. */
383 netdev_close(struct netdev *netdev)
386 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
388 assert(netdev_dev->ref_cnt);
389 netdev_dev->ref_cnt--;
390 netdev_uninit(netdev, true);
392 /* If the reference count for the netdev device is zero, destroy it. */
393 if (!netdev_dev->ref_cnt) {
394 netdev_dev_uninit(netdev_dev, true);
399 /* Returns true if a network device named 'name' exists and may be opened,
400 * otherwise false. */
402 netdev_exists(const char *name)
404 struct netdev *netdev;
407 error = netdev_open_default(name, &netdev);
409 netdev_close(netdev);
412 if (error != ENODEV) {
413 VLOG_WARN("failed to open network device %s: %s",
414 name, strerror(error));
420 /* Returns true if a network device named 'name' is currently opened,
421 * otherwise false. */
423 netdev_is_open(const char *name)
425 return !!shash_find_data(&netdev_dev_shash, name);
428 /* Clears 'svec' and enumerates the names of all known network devices. */
430 netdev_enumerate(struct svec *svec)
432 struct shash_node *node;
438 SHASH_FOR_EACH(node, &netdev_classes) {
439 const struct netdev_class *netdev_class = node->data;
440 if (netdev_class->enumerate) {
441 int retval = netdev_class->enumerate(svec);
443 VLOG_WARN("failed to enumerate %s network devices: %s",
444 netdev_class->type, strerror(retval));
455 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
456 * must have initialized with sufficient room for the packet. The space
457 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
458 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
459 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
460 * need not be included.)
462 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
463 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
464 * positive errno value. Returns EAGAIN immediately if no packet is ready to
467 * Some network devices may not implement support for this function. In such
468 * cases this function will always return EOPNOTSUPP.
471 netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
473 int (*recv)(struct netdev *, void *, size_t);
476 assert(buffer->size == 0);
477 assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
479 recv = netdev_get_dev(netdev)->netdev_class->recv;
481 ? (recv)(netdev, buffer->data, ofpbuf_tailroom(buffer))
484 COVERAGE_INC(netdev_received);
485 buffer->size += retval;
486 if (buffer->size < ETH_TOTAL_MIN) {
487 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
495 /* Registers with the poll loop to wake up from the next call to poll_block()
496 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
498 netdev_recv_wait(struct netdev *netdev)
500 void (*recv_wait)(struct netdev *);
502 recv_wait = netdev_get_dev(netdev)->netdev_class->recv_wait;
508 /* Discards all packets waiting to be received from 'netdev'. */
510 netdev_drain(struct netdev *netdev)
512 int (*drain)(struct netdev *);
514 drain = netdev_get_dev(netdev)->netdev_class->drain;
515 return drain ? drain(netdev) : 0;
518 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
519 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
520 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
521 * the packet is too big or too small to transmit on the device.
523 * The caller retains ownership of 'buffer' in all cases.
525 * The kernel maintains a packet transmission queue, so the caller is not
526 * expected to do additional queuing of packets.
528 * Some network devices may not implement support for this function. In such
529 * cases this function will always return EOPNOTSUPP. */
531 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
533 int (*send)(struct netdev *, const void *, size_t);
536 send = netdev_get_dev(netdev)->netdev_class->send;
537 error = send ? (send)(netdev, buffer->data, buffer->size) : EOPNOTSUPP;
539 COVERAGE_INC(netdev_sent);
544 /* Registers with the poll loop to wake up from the next call to poll_block()
545 * when the packet transmission queue has sufficient room to transmit a packet
546 * with netdev_send().
548 * The kernel maintains a packet transmission queue, so the client is not
549 * expected to do additional queuing of packets. Thus, this function is
550 * unlikely to ever be used. It is included for completeness. */
552 netdev_send_wait(struct netdev *netdev)
554 void (*send_wait)(struct netdev *);
556 send_wait = netdev_get_dev(netdev)->netdev_class->send_wait;
562 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
563 * otherwise a positive errno value. */
565 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
567 return netdev_get_dev(netdev)->netdev_class->set_etheraddr(netdev, mac);
570 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
571 * the MAC address into 'mac'. On failure, returns a positive errno value and
572 * clears 'mac' to all-zeros. */
574 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
576 return netdev_get_dev(netdev)->netdev_class->get_etheraddr(netdev, mac);
579 /* Returns the name of the network device that 'netdev' represents,
580 * e.g. "eth0". The caller must not modify or free the returned string. */
582 netdev_get_name(const struct netdev *netdev)
584 return netdev_get_dev(netdev)->name;
587 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
588 * (and received) packets, in bytes, not including the hardware header; thus,
589 * this is typically 1500 bytes for Ethernet devices.
591 * If successful, returns 0 and stores the MTU size in '*mtup'. On failure,
592 * returns a positive errno value and stores ETH_PAYLOAD_MAX (1500) in
595 netdev_get_mtu(const struct netdev *netdev, int *mtup)
597 int error = netdev_get_dev(netdev)->netdev_class->get_mtu(netdev, mtup);
599 VLOG_WARN_RL(&rl, "failed to retrieve MTU for network device %s: %s",
600 netdev_get_name(netdev), strerror(error));
601 *mtup = ETH_PAYLOAD_MAX;
606 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
607 * failure, returns a negative errno value.
609 * The desired semantics of the ifindex value are a combination of those
610 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
611 * value should be unique within a host and remain stable at least until
612 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
613 * but many systems do not follow this rule anyhow.
615 * Some network devices may not implement support for this function. In such
616 * cases this function will always return -EOPNOTSUPP.
619 netdev_get_ifindex(const struct netdev *netdev)
621 int (*get_ifindex)(const struct netdev *);
623 get_ifindex = netdev_get_dev(netdev)->netdev_class->get_ifindex;
625 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
628 /* Stores the features supported by 'netdev' into each of '*current',
629 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
630 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
631 * successful, otherwise a positive errno value. On failure, all of the
632 * passed-in values are set to 0.
634 * Some network devices may not implement support for this function. In such
635 * cases this function will always return EOPNOTSUPP.
638 netdev_get_features(struct netdev *netdev,
639 uint32_t *current, uint32_t *advertised,
640 uint32_t *supported, uint32_t *peer)
642 int (*get_features)(struct netdev *netdev,
643 uint32_t *current, uint32_t *advertised,
644 uint32_t *supported, uint32_t *peer);
652 advertised = &dummy[1];
655 supported = &dummy[2];
661 get_features = netdev_get_dev(netdev)->netdev_class->get_features;
663 ? get_features(netdev, current, advertised, supported, peer)
666 *current = *advertised = *supported = *peer = 0;
671 /* Returns the maximum speed of a network connection that has the "enum
672 * ofp_port_features" bits in 'features', in bits per second. If no bits that
673 * indicate a speed are set in 'features', assumes 100Mbps. */
675 netdev_features_to_bps(uint32_t features)
678 F_10000MB = OFPPF_10GB_FD,
679 F_1000MB = OFPPF_1GB_HD | OFPPF_1GB_FD,
680 F_100MB = OFPPF_100MB_HD | OFPPF_100MB_FD,
681 F_10MB = OFPPF_10MB_HD | OFPPF_10MB_FD
684 return ( features & F_10000MB ? UINT64_C(10000000000)
685 : features & F_1000MB ? UINT64_C(1000000000)
686 : features & F_100MB ? UINT64_C(100000000)
687 : features & F_10MB ? UINT64_C(10000000)
688 : UINT64_C(100000000));
691 /* Returns true if any of the "enum ofp_port_features" bits that indicate a
692 * full-duplex link are set in 'features', otherwise false. */
694 netdev_features_is_full_duplex(uint32_t features)
696 return (features & (OFPPF_10MB_FD | OFPPF_100MB_FD | OFPPF_1GB_FD
697 | OFPPF_10GB_FD)) != 0;
700 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
701 * successful, otherwise a positive errno value. */
703 netdev_set_advertisements(struct netdev *netdev, uint32_t advertise)
705 return (netdev_get_dev(netdev)->netdev_class->set_advertisements
706 ? netdev_get_dev(netdev)->netdev_class->set_advertisements(
711 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
712 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
713 * errno value and sets '*address' to 0 (INADDR_ANY).
715 * The following error values have well-defined meanings:
717 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
719 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
721 * 'address' or 'netmask' or both may be null, in which case the address or netmask
722 * is not reported. */
724 netdev_get_in4(const struct netdev *netdev,
725 struct in_addr *address_, struct in_addr *netmask_)
727 struct in_addr address;
728 struct in_addr netmask;
731 error = (netdev_get_dev(netdev)->netdev_class->get_in4
732 ? netdev_get_dev(netdev)->netdev_class->get_in4(netdev,
736 address_->s_addr = error ? 0 : address.s_addr;
739 netmask_->s_addr = error ? 0 : netmask.s_addr;
744 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
745 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
746 * positive errno value. */
748 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
750 return (netdev_get_dev(netdev)->netdev_class->set_in4
751 ? netdev_get_dev(netdev)->netdev_class->set_in4(netdev, addr, mask)
755 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
758 netdev_add_router(struct netdev *netdev, struct in_addr router)
760 COVERAGE_INC(netdev_add_router);
761 return (netdev_get_dev(netdev)->netdev_class->add_router
762 ? netdev_get_dev(netdev)->netdev_class->add_router(netdev, router)
766 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
767 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
768 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
769 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
770 * a directly connected network) in '*next_hop' and a copy of the name of the
771 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
772 * responsible for freeing '*netdev_name' (by calling free()). */
774 netdev_get_next_hop(const struct netdev *netdev,
775 const struct in_addr *host, struct in_addr *next_hop,
778 int error = (netdev_get_dev(netdev)->netdev_class->get_next_hop
779 ? netdev_get_dev(netdev)->netdev_class->get_next_hop(
780 host, next_hop, netdev_name)
783 next_hop->s_addr = 0;
789 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
790 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
791 * all-zero-bits (in6addr_any).
793 * The following error values have well-defined meanings:
795 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
797 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
799 * 'in6' may be null, in which case the address itself is not reported. */
801 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
803 struct in6_addr dummy;
806 error = (netdev_get_dev(netdev)->netdev_class->get_in6
807 ? netdev_get_dev(netdev)->netdev_class->get_in6(netdev,
811 memset(in6, 0, sizeof *in6);
816 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
817 * 'on'. If 'permanent' is true, the changes will persist; otherwise, they
818 * will be reverted when 'netdev' is closed or the program exits. Returns 0 if
819 * successful, otherwise a positive errno value. */
821 do_update_flags(struct netdev *netdev, enum netdev_flags off,
822 enum netdev_flags on, enum netdev_flags *old_flagsp,
825 enum netdev_flags old_flags;
828 error = netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
829 off & ~on, on, &old_flags);
831 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
832 off || on ? "set" : "get", netdev_get_name(netdev),
835 } else if ((off || on) && !permanent) {
836 enum netdev_flags new_flags = (old_flags & ~off) | on;
837 enum netdev_flags changed_flags = old_flags ^ new_flags;
839 if (!netdev->changed_flags) {
840 netdev->save_flags = old_flags;
842 netdev->changed_flags |= changed_flags;
846 *old_flagsp = old_flags;
851 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
852 * Returns 0 if successful, otherwise a positive errno value. On failure,
853 * stores 0 into '*flagsp'. */
855 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
857 struct netdev *netdev = (struct netdev *) netdev_;
858 return do_update_flags(netdev, 0, 0, flagsp, false);
861 /* Sets the flags for 'netdev' to 'flags'.
862 * If 'permanent' is true, the changes will persist; otherwise, they
863 * will be reverted when 'netdev' is closed or the program exits.
864 * Returns 0 if successful, otherwise a positive errno value. */
866 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
869 return do_update_flags(netdev, -1, flags, NULL, permanent);
872 /* Turns on the specified 'flags' on 'netdev'.
873 * If 'permanent' is true, the changes will persist; otherwise, they
874 * will be reverted when 'netdev' is closed or the program exits.
875 * Returns 0 if successful, otherwise a positive errno value. */
877 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
880 return do_update_flags(netdev, 0, flags, NULL, permanent);
883 /* Turns off the specified 'flags' on 'netdev'.
884 * If 'permanent' is true, the changes will persist; otherwise, they
885 * will be reverted when 'netdev' is closed or the program exits.
886 * Returns 0 if successful, otherwise a positive errno value. */
888 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
891 return do_update_flags(netdev, flags, 0, NULL, permanent);
894 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
895 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
896 * returns 0. Otherwise, it returns a positive errno value; in particular,
897 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
899 netdev_arp_lookup(const struct netdev *netdev,
900 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
902 int error = (netdev_get_dev(netdev)->netdev_class->arp_lookup
903 ? netdev_get_dev(netdev)->netdev_class->arp_lookup(netdev,
907 memset(mac, 0, ETH_ADDR_LEN);
912 /* Sets 'carrier' to true if carrier is active (link light is on) on
915 netdev_get_carrier(const struct netdev *netdev, bool *carrier)
917 int error = (netdev_get_dev(netdev)->netdev_class->get_carrier
918 ? netdev_get_dev(netdev)->netdev_class->get_carrier(netdev,
927 /* Retrieves current device stats for 'netdev'. */
929 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
933 COVERAGE_INC(netdev_get_stats);
934 error = (netdev_get_dev(netdev)->netdev_class->get_stats
935 ? netdev_get_dev(netdev)->netdev_class->get_stats(netdev, stats)
938 memset(stats, 0xff, sizeof *stats);
943 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
944 * Returns 0 if successful, otherwise a positive errno value.
946 * This will probably fail for most network devices. Some devices might only
947 * allow setting their stats to 0. */
949 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
951 return (netdev_get_dev(netdev)->netdev_class->set_stats
952 ? netdev_get_dev(netdev)->netdev_class->set_stats(netdev, stats)
956 /* Attempts to set input rate limiting (policing) policy, such that up to
957 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
958 * size of 'kbits' kb. */
960 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
961 uint32_t kbits_burst)
963 return (netdev_get_dev(netdev)->netdev_class->set_policing
964 ? netdev_get_dev(netdev)->netdev_class->set_policing(netdev,
965 kbits_rate, kbits_burst)
969 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
970 * empty if 'netdev' does not support QoS. Any names added to 'types' should
971 * be documented as valid for the "type" column in the "QoS" table in
972 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
974 * Every network device supports disabling QoS with a type of "", but this type
975 * will not be added to 'types'.
977 * The caller must initialize 'types' (e.g. with svec_init()) before calling
978 * this function. The caller is responsible for destroying 'types' (e.g. with
979 * svec_destroy()) when it is no longer needed.
981 * Returns 0 if successful, otherwise a positive errno value. */
983 netdev_get_qos_types(const struct netdev *netdev, struct svec *types)
985 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
986 return (class->get_qos_types
987 ? class->get_qos_types(netdev, types)
991 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
992 * which should be "" or one of the types returned by netdev_get_qos_types()
993 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
994 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
995 * 'caps' to all zeros. */
997 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
998 struct netdev_qos_capabilities *caps)
1000 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1003 int retval = (class->get_qos_capabilities
1004 ? class->get_qos_capabilities(netdev, type, caps)
1007 memset(caps, 0, sizeof *caps);
1011 /* Every netdev supports turning off QoS. */
1012 memset(caps, 0, sizeof *caps);
1017 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1018 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1019 * the number of queues (zero on failure) in '*n_queuesp'.
1021 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1023 netdev_get_n_queues(const struct netdev *netdev,
1024 const char *type, unsigned int *n_queuesp)
1026 struct netdev_qos_capabilities caps;
1029 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1030 *n_queuesp = caps.n_queues;
1034 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1035 * stores the name of the current form of QoS into '*typep', stores any details
1036 * of configuration as string key-value pairs in 'details', and returns 0. On
1037 * failure, sets '*typep' to NULL and returns a positive errno value.
1039 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1041 * The caller must initialize 'details' as an empty shash (e.g. with
1042 * shash_init()) before calling this function. The caller must free 'details',
1043 * including 'data' members, when it is no longer needed (e.g. with
1044 * shash_destroy_free_data()).
1046 * The caller must not modify or free '*typep'.
1048 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1049 * 'netdev'. The contents of 'details' should be documented as valid for
1050 * '*typep' in the "other_config" column in the "QoS" table in
1051 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1053 netdev_get_qos(const struct netdev *netdev,
1054 const char **typep, struct shash *details)
1056 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1059 if (class->get_qos) {
1060 retval = class->get_qos(netdev, typep, details);
1063 shash_clear_free_data(details);
1067 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1073 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1074 * with details of configuration from 'details'. Returns 0 if successful,
1075 * otherwise a positive errno value. On error, the previous QoS configuration
1078 * When this function changes the type of QoS (not just 'details'), this also
1079 * resets all queue configuration for 'netdev' to their defaults (which depend
1080 * on the specific type of QoS). Otherwise, the queue configuration for
1081 * 'netdev' is unchanged.
1083 * 'type' should be "" (to disable QoS) or one of the types returned by
1084 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1085 * documented as valid for the given 'type' in the "other_config" column in the
1086 * "QoS" table in vswitchd/vswitch.xml (which is built as
1087 * ovs-vswitchd.conf.db(8)).
1089 * NULL may be specified for 'details' if there are no configuration
1092 netdev_set_qos(struct netdev *netdev,
1093 const char *type, const struct shash *details)
1095 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1101 if (class->set_qos) {
1103 static struct shash empty = SHASH_INITIALIZER(&empty);
1106 return class->set_qos(netdev, type, details);
1108 return *type ? EOPNOTSUPP : 0;
1112 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1113 * successful, adds that information as string key-value pairs to 'details'.
1114 * Returns 0 if successful, otherwise a positive errno value.
1116 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1117 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1119 * The returned contents of 'details' should be documented as valid for the
1120 * given 'type' in the "other_config" column in the "Queue" table in
1121 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1123 * The caller must initialize 'details' (e.g. with shash_init()) before calling
1124 * this function. The caller must free 'details', including 'data' members,
1125 * when it is no longer needed (e.g. with shash_destroy_free_data()). */
1127 netdev_get_queue(const struct netdev *netdev,
1128 unsigned int queue_id, struct shash *details)
1130 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1133 retval = (class->get_queue
1134 ? class->get_queue(netdev, queue_id, details)
1137 shash_clear_free_data(details);
1142 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1143 * string pairs in 'details'. The contents of 'details' should be documented
1144 * as valid for the given 'type' in the "other_config" column in the "Queue"
1145 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1146 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1147 * given queue's configuration should be unmodified.
1149 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1150 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1152 * This function does not modify 'details', and the caller retains ownership of
1156 netdev_set_queue(struct netdev *netdev,
1157 unsigned int queue_id, const struct shash *details)
1159 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1160 return (class->set_queue
1161 ? class->set_queue(netdev, queue_id, details)
1165 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1166 * of QoS may have a fixed set of queues, in which case attempts to delete them
1167 * will fail with EOPNOTSUPP.
1169 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1170 * given queue will be unmodified.
1172 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1173 * the current form of QoS (e.g. as returned by
1174 * netdev_get_n_queues(netdev)). */
1176 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1178 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1179 return (class->delete_queue
1180 ? class->delete_queue(netdev, queue_id)
1184 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1185 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1186 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1187 * positive errno value and fills 'stats' with all-1-bits. */
1189 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1190 struct netdev_queue_stats *stats)
1192 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1195 retval = (class->get_queue_stats
1196 ? class->get_queue_stats(netdev, queue_id, stats)
1199 memset(stats, 0xff, sizeof *stats);
1204 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1205 * its configuration, and the 'aux' specified by the caller. The order of
1206 * iteration is unspecified, but (when successful) each queue is visited
1209 * Calling this function may be more efficient than calling netdev_get_queue()
1212 * 'cb' must not modify or free the 'details' argument passed in.
1214 * Returns 0 if successful, otherwise a positive errno value. On error, some
1215 * configured queues may not have been included in the iteration. */
1217 netdev_dump_queues(const struct netdev *netdev,
1218 netdev_dump_queues_cb *cb, void *aux)
1220 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1221 return (class->dump_queues
1222 ? class->dump_queues(netdev, cb, aux)
1226 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1227 * its statistics, and the 'aux' specified by the caller. The order of
1228 * iteration is unspecified, but (when successful) each queue is visited
1231 * Calling this function may be more efficient than calling
1232 * netdev_get_queue_stats() for every queue.
1234 * 'cb' must not modify or free the statistics passed in.
1236 * Returns 0 if successful, otherwise a positive errno value. On error, some
1237 * configured queues may not have been included in the iteration. */
1239 netdev_dump_queue_stats(const struct netdev *netdev,
1240 netdev_dump_queue_stats_cb *cb, void *aux)
1242 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1243 return (class->dump_queue_stats
1244 ? class->dump_queue_stats(netdev, cb, aux)
1248 /* If 'netdev' is a VLAN network device (e.g. one created with vconfig(8)),
1249 * sets '*vlan_vid' to the VLAN VID associated with that device and returns 0.
1250 * Otherwise returns a errno value (specifically ENOENT if 'netdev_name' is the
1251 * name of a network device that is not a VLAN device) and sets '*vlan_vid' to
1254 netdev_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1256 int error = (netdev_get_dev(netdev)->netdev_class->get_vlan_vid
1257 ? netdev_get_dev(netdev)->netdev_class->get_vlan_vid(netdev,
1266 /* Returns a network device that has 'in4' as its IP address, if one exists,
1267 * otherwise a null pointer. */
1269 netdev_find_dev_by_in4(const struct in_addr *in4)
1271 struct netdev *netdev;
1272 struct svec dev_list = SVEC_EMPTY_INITIALIZER;
1275 netdev_enumerate(&dev_list);
1276 for (i = 0; i < dev_list.n; i++) {
1277 const char *name = dev_list.names[i];
1278 struct in_addr dev_in4;
1280 if (!netdev_open_default(name, &netdev)
1281 && !netdev_get_in4(netdev, &dev_in4, NULL)
1282 && dev_in4.s_addr == in4->s_addr) {
1285 netdev_close(netdev);
1290 svec_destroy(&dev_list);
1294 /* Initializes 'netdev_dev' as a netdev device named 'name' of the
1295 * specified 'netdev_class'.
1297 * This function adds 'netdev_dev' to a netdev-owned shash, so it is
1298 * very important that 'netdev_dev' only be freed after calling
1299 * the refcount drops to zero. */
1301 netdev_dev_init(struct netdev_dev *netdev_dev, const char *name,
1302 const struct netdev_class *netdev_class)
1304 assert(!shash_find(&netdev_dev_shash, name));
1306 memset(netdev_dev, 0, sizeof *netdev_dev);
1307 netdev_dev->netdev_class = netdev_class;
1308 netdev_dev->name = xstrdup(name);
1309 netdev_dev->node = shash_add(&netdev_dev_shash, name, netdev_dev);
1312 /* Undoes the results of initialization.
1314 * Normally this function does not need to be called as netdev_close has
1315 * the same effect when the refcount drops to zero.
1316 * However, it may be called by providers due to an error on creation
1317 * that occurs after initialization. It this case netdev_close() would
1318 * never be called. */
1320 netdev_dev_uninit(struct netdev_dev *netdev_dev, bool destroy)
1322 char *name = netdev_dev->name;
1324 assert(!netdev_dev->ref_cnt);
1326 shash_delete(&netdev_dev_shash, netdev_dev->node);
1327 update_device_args(netdev_dev, NULL);
1330 netdev_dev->netdev_class->destroy(netdev_dev);
1335 /* Returns the class type of 'netdev_dev'.
1337 * The caller must not free the returned value. */
1339 netdev_dev_get_type(const struct netdev_dev *netdev_dev)
1341 return netdev_dev->netdev_class->type;
1344 /* Returns the class associated with 'netdev_dev'. */
1345 const struct netdev_class *
1346 netdev_dev_get_class(const struct netdev_dev *netdev_dev)
1348 return netdev_dev->netdev_class;
1351 /* Returns the name of 'netdev_dev'.
1353 * The caller must not free the returned value. */
1355 netdev_dev_get_name(const struct netdev_dev *netdev_dev)
1357 return netdev_dev->name;
1360 /* Returns the netdev_dev with 'name' or NULL if there is none.
1362 * The caller must not free the returned value. */
1364 netdev_dev_from_name(const char *name)
1366 return shash_find_data(&netdev_dev_shash, name);
1369 /* Fills 'device_list' with devices that match 'netdev_class'.
1371 * The caller is responsible for initializing and destroying 'device_list'
1372 * but the contained netdev_devs must not be freed. */
1374 netdev_dev_get_devices(const struct netdev_class *netdev_class,
1375 struct shash *device_list)
1377 struct shash_node *node;
1378 SHASH_FOR_EACH (node, &netdev_dev_shash) {
1379 struct netdev_dev *dev = node->data;
1381 if (dev->netdev_class == netdev_class) {
1382 shash_add(device_list, node->name, node->data);
1387 /* Initializes 'netdev' as a instance of the netdev_dev.
1389 * This function adds 'netdev' to a netdev-owned linked list, so it is very
1390 * important that 'netdev' only be freed after calling netdev_close(). */
1392 netdev_init(struct netdev *netdev, struct netdev_dev *netdev_dev)
1394 memset(netdev, 0, sizeof *netdev);
1395 netdev->netdev_dev = netdev_dev;
1396 list_push_back(&netdev_list, &netdev->node);
1399 /* Undoes the results of initialization.
1401 * Normally this function only needs to be called from netdev_close().
1402 * However, it may be called by providers due to an error on opening
1403 * that occurs after initialization. It this case netdev_close() would
1404 * never be called. */
1406 netdev_uninit(struct netdev *netdev, bool close)
1408 /* Restore flags that we changed, if any. */
1409 int error = restore_flags(netdev);
1410 list_remove(&netdev->node);
1412 VLOG_WARN("failed to restore network device flags on %s: %s",
1413 netdev_get_name(netdev), strerror(error));
1417 netdev_get_dev(netdev)->netdev_class->close(netdev);
1422 /* Returns the class type of 'netdev'.
1424 * The caller must not free the returned value. */
1426 netdev_get_type(const struct netdev *netdev)
1428 return netdev_get_dev(netdev)->netdev_class->type;
1432 netdev_get_dev(const struct netdev *netdev)
1434 return netdev->netdev_dev;
1437 /* Initializes 'notifier' as a netdev notifier for 'netdev', for which
1438 * notification will consist of calling 'cb', with auxiliary data 'aux'. */
1440 netdev_notifier_init(struct netdev_notifier *notifier, struct netdev *netdev,
1441 void (*cb)(struct netdev_notifier *), void *aux)
1443 notifier->netdev = netdev;
1445 notifier->aux = aux;
1448 /* Tracks changes in the status of a set of network devices. */
1449 struct netdev_monitor {
1450 struct shash polled_netdevs;
1451 struct shash changed_netdevs;
1454 /* Creates and returns a new structure for monitor changes in the status of
1455 * network devices. */
1456 struct netdev_monitor *
1457 netdev_monitor_create(void)
1459 struct netdev_monitor *monitor = xmalloc(sizeof *monitor);
1460 shash_init(&monitor->polled_netdevs);
1461 shash_init(&monitor->changed_netdevs);
1465 /* Destroys 'monitor'. */
1467 netdev_monitor_destroy(struct netdev_monitor *monitor)
1470 struct shash_node *node;
1472 SHASH_FOR_EACH (node, &monitor->polled_netdevs) {
1473 struct netdev_notifier *notifier = node->data;
1474 netdev_get_dev(notifier->netdev)->netdev_class->poll_remove(
1478 shash_destroy(&monitor->polled_netdevs);
1479 shash_destroy(&monitor->changed_netdevs);
1485 netdev_monitor_cb(struct netdev_notifier *notifier)
1487 struct netdev_monitor *monitor = notifier->aux;
1488 const char *name = netdev_get_name(notifier->netdev);
1489 shash_add_once(&monitor->changed_netdevs, name, NULL);
1492 /* Attempts to add 'netdev' as a netdev monitored by 'monitor'. Returns 0 if
1493 * successful, otherwise a positive errno value.
1495 * Adding a given 'netdev' to a monitor multiple times is equivalent to adding
1498 netdev_monitor_add(struct netdev_monitor *monitor, struct netdev *netdev)
1500 const char *netdev_name = netdev_get_name(netdev);
1502 if (!shash_find(&monitor->polled_netdevs, netdev_name)
1503 && netdev_get_dev(netdev)->netdev_class->poll_add)
1505 struct netdev_notifier *notifier;
1506 error = netdev_get_dev(netdev)->netdev_class->poll_add(netdev,
1507 netdev_monitor_cb, monitor, ¬ifier);
1509 assert(notifier->netdev == netdev);
1510 shash_add(&monitor->polled_netdevs, netdev_name, notifier);
1516 /* Removes 'netdev' from the set of netdevs monitored by 'monitor'. (This has
1517 * no effect if 'netdev' is not in the set of devices monitored by
1520 netdev_monitor_remove(struct netdev_monitor *monitor, struct netdev *netdev)
1522 const char *netdev_name = netdev_get_name(netdev);
1523 struct shash_node *node;
1525 node = shash_find(&monitor->polled_netdevs, netdev_name);
1527 /* Cancel future notifications. */
1528 struct netdev_notifier *notifier = node->data;
1529 netdev_get_dev(netdev)->netdev_class->poll_remove(notifier);
1530 shash_delete(&monitor->polled_netdevs, node);
1532 /* Drop any pending notification. */
1533 node = shash_find(&monitor->changed_netdevs, netdev_name);
1535 shash_delete(&monitor->changed_netdevs, node);
1540 /* Checks for changes to netdevs in the set monitored by 'monitor'. If any of
1541 * the attributes (Ethernet address, carrier status, speed or peer-advertised
1542 * speed, flags, etc.) of a network device monitored by 'monitor' has changed,
1543 * sets '*devnamep' to the name of a device that has changed and returns 0.
1544 * The caller is responsible for freeing '*devnamep' (with free()).
1546 * If no devices have changed, sets '*devnamep' to NULL and returns EAGAIN.
1549 netdev_monitor_poll(struct netdev_monitor *monitor, char **devnamep)
1551 struct shash_node *node = shash_first(&monitor->changed_netdevs);
1556 *devnamep = xstrdup(node->name);
1557 shash_delete(&monitor->changed_netdevs, node);
1562 /* Registers with the poll loop to wake up from the next call to poll_block()
1563 * when netdev_monitor_poll(monitor) would indicate that a device has
1566 netdev_monitor_poll_wait(const struct netdev_monitor *monitor)
1568 if (!shash_is_empty(&monitor->changed_netdevs)) {
1569 poll_immediate_wake();
1571 /* XXX Nothing needed here for netdev_linux, but maybe other netdev
1572 * classes need help. */
1576 /* Restore the network device flags on 'netdev' to those that were active
1577 * before we changed them. Returns 0 if successful, otherwise a positive
1580 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
1582 restore_flags(struct netdev *netdev)
1584 if (netdev->changed_flags) {
1585 enum netdev_flags restore = netdev->save_flags & netdev->changed_flags;
1586 enum netdev_flags old_flags;
1587 return netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
1588 netdev->changed_flags & ~restore,
1589 restore, &old_flags);
1594 /* Close all netdevs on shutdown so they can do any needed cleanup such as
1595 * destroying devices, restoring flags, etc. */
1597 close_all_netdevs(void *aux OVS_UNUSED)
1599 struct netdev *netdev, *next;
1600 LIST_FOR_EACH_SAFE(netdev, next, struct netdev, node, &netdev_list) {
1601 netdev_close(netdev);