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
295 netdev_open(struct netdev_options *options, struct netdev **netdevp)
297 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
298 struct netdev_dev *netdev_dev;
304 if (!options->args) {
305 options->args = &empty_args;
308 netdev_dev = shash_find_data(&netdev_dev_shash, options->name);
311 error = create_device(options, &netdev_dev);
313 if (error == EAFNOSUPPORT) {
314 VLOG_WARN("could not create netdev %s of unknown type %s",
315 options->name, options->type);
319 update_device_args(netdev_dev, options->args);
321 } else if (!shash_is_empty(options->args) &&
322 !compare_device_args(netdev_dev, options->args)) {
324 VLOG_WARN("%s: attempted to open already open netdev with "
325 "different arguments", options->name);
329 error = netdev_dev->netdev_class->open(netdev_dev, options->ethertype,
333 netdev_dev->ref_cnt++;
335 if (!netdev_dev->ref_cnt) {
336 netdev_dev_uninit(netdev_dev, true);
344 netdev_open_default(const char *name, struct netdev **netdevp)
346 struct netdev_options options;
348 memset(&options, 0, sizeof options);
350 options.ethertype = NETDEV_ETH_TYPE_NONE;
352 return netdev_open(&options, netdevp);
355 /* Increments the reference count on 'netdev'. Returns 'netdev'. */
357 netdev_reopen(struct netdev *netdev)
363 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
364 * or NULL if none are needed. */
366 netdev_reconfigure(struct netdev *netdev, const struct shash *args)
368 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
369 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
375 if (netdev_dev->netdev_class->reconfigure) {
376 if (!compare_device_args(netdev_dev, args)) {
377 update_device_args(netdev_dev, args);
378 return netdev_dev->netdev_class->reconfigure(netdev_dev, args);
380 } else if (!shash_is_empty(args)) {
381 VLOG_WARN("%s: arguments provided to device that does not have a "
382 "reconfigure function", netdev_get_name(netdev));
388 /* Decrements the reference count on 'netdev'. If the reference count reaches
389 * zero, closes and destroys 'netdev'. */
391 netdev_close(struct netdev *netdev)
393 assert(!netdev || netdev->ref_cnt > 0);
394 if (netdev && !--netdev->ref_cnt) {
395 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
397 assert(netdev_dev->ref_cnt);
398 netdev_dev->ref_cnt--;
399 netdev_uninit(netdev, true);
401 /* If the reference count for the netdev device is zero, destroy it. */
402 if (!netdev_dev->ref_cnt) {
403 netdev_dev_uninit(netdev_dev, true);
408 /* Returns true if a network device named 'name' exists and may be opened,
409 * otherwise false. */
411 netdev_exists(const char *name)
413 struct netdev *netdev;
416 error = netdev_open_default(name, &netdev);
418 netdev_close(netdev);
421 if (error != ENODEV) {
422 VLOG_WARN("failed to open network device %s: %s",
423 name, strerror(error));
429 /* Returns true if a network device named 'name' is currently opened,
430 * otherwise false. */
432 netdev_is_open(const char *name)
434 return !!shash_find_data(&netdev_dev_shash, name);
437 /* Clears 'svec' and enumerates the names of all known network devices. */
439 netdev_enumerate(struct svec *svec)
441 struct shash_node *node;
447 SHASH_FOR_EACH(node, &netdev_classes) {
448 const struct netdev_class *netdev_class = node->data;
449 if (netdev_class->enumerate) {
450 int retval = netdev_class->enumerate(svec);
452 VLOG_WARN("failed to enumerate %s network devices: %s",
453 netdev_class->type, strerror(retval));
464 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
465 * must have initialized with sufficient room for the packet. The space
466 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
467 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
468 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
469 * need not be included.)
471 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
472 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
473 * positive errno value. Returns EAGAIN immediately if no packet is ready to
476 * Some network devices may not implement support for this function. In such
477 * cases this function will always return EOPNOTSUPP. */
479 netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
481 int (*recv)(struct netdev *, void *, size_t);
484 assert(buffer->size == 0);
485 assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
487 recv = netdev_get_dev(netdev)->netdev_class->recv;
489 ? (recv)(netdev, buffer->data, ofpbuf_tailroom(buffer))
492 COVERAGE_INC(netdev_received);
493 buffer->size += retval;
494 if (buffer->size < ETH_TOTAL_MIN) {
495 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
503 /* Registers with the poll loop to wake up from the next call to poll_block()
504 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
506 netdev_recv_wait(struct netdev *netdev)
508 void (*recv_wait)(struct netdev *);
510 recv_wait = netdev_get_dev(netdev)->netdev_class->recv_wait;
516 /* Discards all packets waiting to be received from 'netdev'. */
518 netdev_drain(struct netdev *netdev)
520 int (*drain)(struct netdev *);
522 drain = netdev_get_dev(netdev)->netdev_class->drain;
523 return drain ? drain(netdev) : 0;
526 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
527 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
528 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
529 * the packet is too big or too small to transmit on the device.
531 * The caller retains ownership of 'buffer' in all cases.
533 * The kernel maintains a packet transmission queue, so the caller is not
534 * expected to do additional queuing of packets.
536 * Some network devices may not implement support for this function. In such
537 * cases this function will always return EOPNOTSUPP. */
539 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
541 int (*send)(struct netdev *, const void *, size_t);
544 send = netdev_get_dev(netdev)->netdev_class->send;
545 error = send ? (send)(netdev, buffer->data, buffer->size) : EOPNOTSUPP;
547 COVERAGE_INC(netdev_sent);
552 /* Registers with the poll loop to wake up from the next call to poll_block()
553 * when the packet transmission queue has sufficient room to transmit a packet
554 * with netdev_send().
556 * The kernel maintains a packet transmission queue, so the client is not
557 * expected to do additional queuing of packets. Thus, this function is
558 * unlikely to ever be used. It is included for completeness. */
560 netdev_send_wait(struct netdev *netdev)
562 void (*send_wait)(struct netdev *);
564 send_wait = netdev_get_dev(netdev)->netdev_class->send_wait;
570 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
571 * otherwise a positive errno value. */
573 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
575 return netdev_get_dev(netdev)->netdev_class->set_etheraddr(netdev, mac);
578 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
579 * the MAC address into 'mac'. On failure, returns a positive errno value and
580 * clears 'mac' to all-zeros. */
582 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
584 return netdev_get_dev(netdev)->netdev_class->get_etheraddr(netdev, mac);
587 /* Returns the name of the network device that 'netdev' represents,
588 * e.g. "eth0". The caller must not modify or free the returned string. */
590 netdev_get_name(const struct netdev *netdev)
592 return netdev_get_dev(netdev)->name;
595 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
596 * (and received) packets, in bytes, not including the hardware header; thus,
597 * this is typically 1500 bytes for Ethernet devices.
599 * If successful, returns 0 and stores the MTU size in '*mtup'. On failure,
600 * returns a positive errno value and stores ETH_PAYLOAD_MAX (1500) in
603 netdev_get_mtu(const struct netdev *netdev, int *mtup)
605 int error = netdev_get_dev(netdev)->netdev_class->get_mtu(netdev, mtup);
607 VLOG_WARN_RL(&rl, "failed to retrieve MTU for network device %s: %s",
608 netdev_get_name(netdev), strerror(error));
609 *mtup = ETH_PAYLOAD_MAX;
614 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
615 * failure, returns a negative errno value.
617 * The desired semantics of the ifindex value are a combination of those
618 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
619 * value should be unique within a host and remain stable at least until
620 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
621 * but many systems do not follow this rule anyhow.
623 * Some network devices may not implement support for this function. In such
624 * cases this function will always return -EOPNOTSUPP.
627 netdev_get_ifindex(const struct netdev *netdev)
629 int (*get_ifindex)(const struct netdev *);
631 get_ifindex = netdev_get_dev(netdev)->netdev_class->get_ifindex;
633 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
636 /* Stores the features supported by 'netdev' into each of '*current',
637 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
638 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
639 * successful, otherwise a positive errno value. On failure, all of the
640 * passed-in values are set to 0.
642 * Some network devices may not implement support for this function. In such
643 * cases this function will always return EOPNOTSUPP. */
645 netdev_get_features(struct netdev *netdev,
646 uint32_t *current, uint32_t *advertised,
647 uint32_t *supported, uint32_t *peer)
649 int (*get_features)(struct netdev *netdev,
650 uint32_t *current, uint32_t *advertised,
651 uint32_t *supported, uint32_t *peer);
659 advertised = &dummy[1];
662 supported = &dummy[2];
668 get_features = netdev_get_dev(netdev)->netdev_class->get_features;
670 ? get_features(netdev, current, advertised, supported, peer)
673 *current = *advertised = *supported = *peer = 0;
678 /* Returns the maximum speed of a network connection that has the "enum
679 * ofp_port_features" bits in 'features', in bits per second. If no bits that
680 * indicate a speed are set in 'features', assumes 100Mbps. */
682 netdev_features_to_bps(uint32_t features)
685 F_10000MB = OFPPF_10GB_FD,
686 F_1000MB = OFPPF_1GB_HD | OFPPF_1GB_FD,
687 F_100MB = OFPPF_100MB_HD | OFPPF_100MB_FD,
688 F_10MB = OFPPF_10MB_HD | OFPPF_10MB_FD
691 return ( features & F_10000MB ? UINT64_C(10000000000)
692 : features & F_1000MB ? UINT64_C(1000000000)
693 : features & F_100MB ? UINT64_C(100000000)
694 : features & F_10MB ? UINT64_C(10000000)
695 : UINT64_C(100000000));
698 /* Returns true if any of the "enum ofp_port_features" bits that indicate a
699 * full-duplex link are set in 'features', otherwise false. */
701 netdev_features_is_full_duplex(uint32_t features)
703 return (features & (OFPPF_10MB_FD | OFPPF_100MB_FD | OFPPF_1GB_FD
704 | OFPPF_10GB_FD)) != 0;
707 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
708 * successful, otherwise a positive errno value. */
710 netdev_set_advertisements(struct netdev *netdev, uint32_t advertise)
712 return (netdev_get_dev(netdev)->netdev_class->set_advertisements
713 ? netdev_get_dev(netdev)->netdev_class->set_advertisements(
718 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
719 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
720 * errno value and sets '*address' to 0 (INADDR_ANY).
722 * The following error values have well-defined meanings:
724 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
726 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
728 * 'address' or 'netmask' or both may be null, in which case the address or
729 * netmask is not reported. */
731 netdev_get_in4(const struct netdev *netdev,
732 struct in_addr *address_, struct in_addr *netmask_)
734 struct in_addr address;
735 struct in_addr netmask;
738 error = (netdev_get_dev(netdev)->netdev_class->get_in4
739 ? netdev_get_dev(netdev)->netdev_class->get_in4(netdev,
743 address_->s_addr = error ? 0 : address.s_addr;
746 netmask_->s_addr = error ? 0 : netmask.s_addr;
751 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
752 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
753 * positive errno value. */
755 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
757 return (netdev_get_dev(netdev)->netdev_class->set_in4
758 ? netdev_get_dev(netdev)->netdev_class->set_in4(netdev, addr, mask)
762 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
765 netdev_add_router(struct netdev *netdev, struct in_addr router)
767 COVERAGE_INC(netdev_add_router);
768 return (netdev_get_dev(netdev)->netdev_class->add_router
769 ? netdev_get_dev(netdev)->netdev_class->add_router(netdev, router)
773 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
774 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
775 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
776 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
777 * a directly connected network) in '*next_hop' and a copy of the name of the
778 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
779 * responsible for freeing '*netdev_name' (by calling free()). */
781 netdev_get_next_hop(const struct netdev *netdev,
782 const struct in_addr *host, struct in_addr *next_hop,
785 int error = (netdev_get_dev(netdev)->netdev_class->get_next_hop
786 ? netdev_get_dev(netdev)->netdev_class->get_next_hop(
787 host, next_hop, netdev_name)
790 next_hop->s_addr = 0;
796 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
797 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
798 * all-zero-bits (in6addr_any).
800 * The following error values have well-defined meanings:
802 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
804 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
806 * 'in6' may be null, in which case the address itself is not reported. */
808 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
810 struct in6_addr dummy;
813 error = (netdev_get_dev(netdev)->netdev_class->get_in6
814 ? netdev_get_dev(netdev)->netdev_class->get_in6(netdev,
818 memset(in6, 0, sizeof *in6);
823 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
824 * 'on'. If 'permanent' is true, the changes will persist; otherwise, they
825 * will be reverted when 'netdev' is closed or the program exits. Returns 0 if
826 * successful, otherwise a positive errno value. */
828 do_update_flags(struct netdev *netdev, enum netdev_flags off,
829 enum netdev_flags on, enum netdev_flags *old_flagsp,
832 enum netdev_flags old_flags;
835 error = netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
836 off & ~on, on, &old_flags);
838 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
839 off || on ? "set" : "get", netdev_get_name(netdev),
842 } else if ((off || on) && !permanent) {
843 enum netdev_flags new_flags = (old_flags & ~off) | on;
844 enum netdev_flags changed_flags = old_flags ^ new_flags;
846 if (!netdev->changed_flags) {
847 netdev->save_flags = old_flags;
849 netdev->changed_flags |= changed_flags;
853 *old_flagsp = old_flags;
858 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
859 * Returns 0 if successful, otherwise a positive errno value. On failure,
860 * stores 0 into '*flagsp'. */
862 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
864 struct netdev *netdev = (struct netdev *) netdev_;
865 return do_update_flags(netdev, 0, 0, flagsp, false);
868 /* Sets the flags for 'netdev' to 'flags'.
869 * If 'permanent' is true, the changes will persist; otherwise, they
870 * will be reverted when 'netdev' is closed or the program exits.
871 * Returns 0 if successful, otherwise a positive errno value. */
873 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
876 return do_update_flags(netdev, -1, flags, NULL, permanent);
879 /* Turns on the specified 'flags' on 'netdev'.
880 * If 'permanent' is true, the changes will persist; otherwise, they
881 * will be reverted when 'netdev' is closed or the program exits.
882 * Returns 0 if successful, otherwise a positive errno value. */
884 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
887 return do_update_flags(netdev, 0, flags, NULL, permanent);
890 /* Turns off the specified 'flags' on 'netdev'.
891 * If 'permanent' is true, the changes will persist; otherwise, they
892 * will be reverted when 'netdev' is closed or the program exits.
893 * Returns 0 if successful, otherwise a positive errno value. */
895 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
898 return do_update_flags(netdev, flags, 0, NULL, permanent);
901 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
902 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
903 * returns 0. Otherwise, it returns a positive errno value; in particular,
904 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
906 netdev_arp_lookup(const struct netdev *netdev,
907 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
909 int error = (netdev_get_dev(netdev)->netdev_class->arp_lookup
910 ? netdev_get_dev(netdev)->netdev_class->arp_lookup(netdev,
914 memset(mac, 0, ETH_ADDR_LEN);
919 /* Sets 'carrier' to true if carrier is active (link light is on) on
922 netdev_get_carrier(const struct netdev *netdev, bool *carrier)
924 int error = (netdev_get_dev(netdev)->netdev_class->get_carrier
925 ? netdev_get_dev(netdev)->netdev_class->get_carrier(netdev,
934 /* Retrieves current device stats for 'netdev'. */
936 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
940 COVERAGE_INC(netdev_get_stats);
941 error = (netdev_get_dev(netdev)->netdev_class->get_stats
942 ? netdev_get_dev(netdev)->netdev_class->get_stats(netdev, stats)
945 memset(stats, 0xff, sizeof *stats);
950 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
951 * Returns 0 if successful, otherwise a positive errno value.
953 * This will probably fail for most network devices. Some devices might only
954 * allow setting their stats to 0. */
956 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
958 return (netdev_get_dev(netdev)->netdev_class->set_stats
959 ? netdev_get_dev(netdev)->netdev_class->set_stats(netdev, stats)
963 /* Attempts to set input rate limiting (policing) policy, such that up to
964 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
965 * size of 'kbits' kb. */
967 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
968 uint32_t kbits_burst)
970 return (netdev_get_dev(netdev)->netdev_class->set_policing
971 ? netdev_get_dev(netdev)->netdev_class->set_policing(netdev,
972 kbits_rate, kbits_burst)
976 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
977 * empty if 'netdev' does not support QoS. Any names added to 'types' should
978 * be documented as valid for the "type" column in the "QoS" table in
979 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
981 * Every network device supports disabling QoS with a type of "", but this type
982 * will not be added to 'types'.
984 * The caller must initialize 'types' (e.g. with svec_init()) before calling
985 * this function. The caller is responsible for destroying 'types' (e.g. with
986 * svec_destroy()) when it is no longer needed.
988 * Returns 0 if successful, otherwise a positive errno value. */
990 netdev_get_qos_types(const struct netdev *netdev, struct svec *types)
992 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
993 return (class->get_qos_types
994 ? class->get_qos_types(netdev, types)
998 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
999 * which should be "" or one of the types returned by netdev_get_qos_types()
1000 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
1001 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
1002 * 'caps' to all zeros. */
1004 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
1005 struct netdev_qos_capabilities *caps)
1007 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1010 int retval = (class->get_qos_capabilities
1011 ? class->get_qos_capabilities(netdev, type, caps)
1014 memset(caps, 0, sizeof *caps);
1018 /* Every netdev supports turning off QoS. */
1019 memset(caps, 0, sizeof *caps);
1024 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1025 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1026 * the number of queues (zero on failure) in '*n_queuesp'.
1028 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1030 netdev_get_n_queues(const struct netdev *netdev,
1031 const char *type, unsigned int *n_queuesp)
1033 struct netdev_qos_capabilities caps;
1036 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1037 *n_queuesp = caps.n_queues;
1041 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1042 * stores the name of the current form of QoS into '*typep', stores any details
1043 * of configuration as string key-value pairs in 'details', and returns 0. On
1044 * failure, sets '*typep' to NULL and returns a positive errno value.
1046 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1048 * The caller must initialize 'details' as an empty shash (e.g. with
1049 * shash_init()) before calling this function. The caller must free 'details',
1050 * including 'data' members, when it is no longer needed (e.g. with
1051 * shash_destroy_free_data()).
1053 * The caller must not modify or free '*typep'.
1055 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1056 * 'netdev'. The contents of 'details' should be documented as valid for
1057 * '*typep' in the "other_config" column in the "QoS" table in
1058 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1060 netdev_get_qos(const struct netdev *netdev,
1061 const char **typep, struct shash *details)
1063 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1066 if (class->get_qos) {
1067 retval = class->get_qos(netdev, typep, details);
1070 shash_clear_free_data(details);
1074 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1080 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1081 * with details of configuration from 'details'. Returns 0 if successful,
1082 * otherwise a positive errno value. On error, the previous QoS configuration
1085 * When this function changes the type of QoS (not just 'details'), this also
1086 * resets all queue configuration for 'netdev' to their defaults (which depend
1087 * on the specific type of QoS). Otherwise, the queue configuration for
1088 * 'netdev' is unchanged.
1090 * 'type' should be "" (to disable QoS) or one of the types returned by
1091 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1092 * documented as valid for the given 'type' in the "other_config" column in the
1093 * "QoS" table in vswitchd/vswitch.xml (which is built as
1094 * ovs-vswitchd.conf.db(8)).
1096 * NULL may be specified for 'details' if there are no configuration
1099 netdev_set_qos(struct netdev *netdev,
1100 const char *type, const struct shash *details)
1102 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1108 if (class->set_qos) {
1110 static struct shash empty = SHASH_INITIALIZER(&empty);
1113 return class->set_qos(netdev, type, details);
1115 return *type ? EOPNOTSUPP : 0;
1119 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1120 * successful, adds that information as string key-value pairs to 'details'.
1121 * Returns 0 if successful, otherwise a positive errno value.
1123 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1124 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1126 * The returned contents of 'details' should be documented as valid for the
1127 * given 'type' in the "other_config" column in the "Queue" table in
1128 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1130 * The caller must initialize 'details' (e.g. with shash_init()) before calling
1131 * this function. The caller must free 'details', including 'data' members,
1132 * when it is no longer needed (e.g. with shash_destroy_free_data()). */
1134 netdev_get_queue(const struct netdev *netdev,
1135 unsigned int queue_id, struct shash *details)
1137 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1140 retval = (class->get_queue
1141 ? class->get_queue(netdev, queue_id, details)
1144 shash_clear_free_data(details);
1149 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1150 * string pairs in 'details'. The contents of 'details' should be documented
1151 * as valid for the given 'type' in the "other_config" column in the "Queue"
1152 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1153 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1154 * given queue's configuration should be unmodified.
1156 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1157 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1159 * This function does not modify 'details', and the caller retains ownership of
1162 netdev_set_queue(struct netdev *netdev,
1163 unsigned int queue_id, const struct shash *details)
1165 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1166 return (class->set_queue
1167 ? class->set_queue(netdev, queue_id, details)
1171 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1172 * of QoS may have a fixed set of queues, in which case attempts to delete them
1173 * will fail with EOPNOTSUPP.
1175 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1176 * given queue will be unmodified.
1178 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1179 * the current form of QoS (e.g. as returned by
1180 * netdev_get_n_queues(netdev)). */
1182 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1184 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1185 return (class->delete_queue
1186 ? class->delete_queue(netdev, queue_id)
1190 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1191 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1192 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1193 * positive errno value and fills 'stats' with all-1-bits. */
1195 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1196 struct netdev_queue_stats *stats)
1198 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1201 retval = (class->get_queue_stats
1202 ? class->get_queue_stats(netdev, queue_id, stats)
1205 memset(stats, 0xff, sizeof *stats);
1210 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1211 * its configuration, and the 'aux' specified by the caller. The order of
1212 * iteration is unspecified, but (when successful) each queue is visited
1215 * Calling this function may be more efficient than calling netdev_get_queue()
1218 * 'cb' must not modify or free the 'details' argument passed in.
1220 * Returns 0 if successful, otherwise a positive errno value. On error, some
1221 * configured queues may not have been included in the iteration. */
1223 netdev_dump_queues(const struct netdev *netdev,
1224 netdev_dump_queues_cb *cb, void *aux)
1226 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1227 return (class->dump_queues
1228 ? class->dump_queues(netdev, cb, aux)
1232 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1233 * its statistics, and the 'aux' specified by the caller. The order of
1234 * iteration is unspecified, but (when successful) each queue is visited
1237 * Calling this function may be more efficient than calling
1238 * netdev_get_queue_stats() for every queue.
1240 * 'cb' must not modify or free the statistics passed in.
1242 * Returns 0 if successful, otherwise a positive errno value. On error, some
1243 * configured queues may not have been included in the iteration. */
1245 netdev_dump_queue_stats(const struct netdev *netdev,
1246 netdev_dump_queue_stats_cb *cb, void *aux)
1248 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1249 return (class->dump_queue_stats
1250 ? class->dump_queue_stats(netdev, cb, aux)
1254 /* If 'netdev' is a VLAN network device (e.g. one created with vconfig(8)),
1255 * sets '*vlan_vid' to the VLAN VID associated with that device and returns 0.
1256 * Otherwise returns a errno value (specifically ENOENT if 'netdev_name' is the
1257 * name of a network device that is not a VLAN device) and sets '*vlan_vid' to
1260 netdev_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1262 int error = (netdev_get_dev(netdev)->netdev_class->get_vlan_vid
1263 ? netdev_get_dev(netdev)->netdev_class->get_vlan_vid(netdev,
1272 /* Returns a network device that has 'in4' as its IP address, if one exists,
1273 * otherwise a null pointer. */
1275 netdev_find_dev_by_in4(const struct in_addr *in4)
1277 struct netdev *netdev;
1278 struct svec dev_list = SVEC_EMPTY_INITIALIZER;
1281 netdev_enumerate(&dev_list);
1282 for (i = 0; i < dev_list.n; i++) {
1283 const char *name = dev_list.names[i];
1284 struct in_addr dev_in4;
1286 if (!netdev_open_default(name, &netdev)
1287 && !netdev_get_in4(netdev, &dev_in4, NULL)
1288 && dev_in4.s_addr == in4->s_addr) {
1291 netdev_close(netdev);
1296 svec_destroy(&dev_list);
1300 /* Initializes 'netdev_dev' as a netdev device named 'name' of the
1301 * specified 'netdev_class'.
1303 * This function adds 'netdev_dev' to a netdev-owned shash, so it is
1304 * very important that 'netdev_dev' only be freed after calling
1305 * the refcount drops to zero. */
1307 netdev_dev_init(struct netdev_dev *netdev_dev, const char *name,
1308 const struct netdev_class *netdev_class)
1310 assert(!shash_find(&netdev_dev_shash, name));
1312 memset(netdev_dev, 0, sizeof *netdev_dev);
1313 netdev_dev->netdev_class = netdev_class;
1314 netdev_dev->name = xstrdup(name);
1315 netdev_dev->node = shash_add(&netdev_dev_shash, name, netdev_dev);
1318 /* Undoes the results of initialization.
1320 * Normally this function does not need to be called as netdev_close has
1321 * the same effect when the refcount drops to zero.
1322 * However, it may be called by providers due to an error on creation
1323 * that occurs after initialization. It this case netdev_close() would
1324 * never be called. */
1326 netdev_dev_uninit(struct netdev_dev *netdev_dev, bool destroy)
1328 char *name = netdev_dev->name;
1330 assert(!netdev_dev->ref_cnt);
1332 shash_delete(&netdev_dev_shash, netdev_dev->node);
1333 update_device_args(netdev_dev, NULL);
1336 netdev_dev->netdev_class->destroy(netdev_dev);
1341 /* Returns the class type of 'netdev_dev'.
1343 * The caller must not free the returned value. */
1345 netdev_dev_get_type(const struct netdev_dev *netdev_dev)
1347 return netdev_dev->netdev_class->type;
1350 /* Returns the class associated with 'netdev_dev'. */
1351 const struct netdev_class *
1352 netdev_dev_get_class(const struct netdev_dev *netdev_dev)
1354 return netdev_dev->netdev_class;
1357 /* Returns the name of 'netdev_dev'.
1359 * The caller must not free the returned value. */
1361 netdev_dev_get_name(const struct netdev_dev *netdev_dev)
1363 return netdev_dev->name;
1366 /* Returns the netdev_dev with 'name' or NULL if there is none.
1368 * The caller must not free the returned value. */
1370 netdev_dev_from_name(const char *name)
1372 return shash_find_data(&netdev_dev_shash, name);
1375 /* Fills 'device_list' with devices that match 'netdev_class'.
1377 * The caller is responsible for initializing and destroying 'device_list'
1378 * but the contained netdev_devs must not be freed. */
1380 netdev_dev_get_devices(const struct netdev_class *netdev_class,
1381 struct shash *device_list)
1383 struct shash_node *node;
1384 SHASH_FOR_EACH (node, &netdev_dev_shash) {
1385 struct netdev_dev *dev = node->data;
1387 if (dev->netdev_class == netdev_class) {
1388 shash_add(device_list, node->name, node->data);
1393 /* Initializes 'netdev' as a instance of the netdev_dev.
1395 * This function adds 'netdev' to a netdev-owned linked list, so it is very
1396 * important that 'netdev' only be freed after calling netdev_close(). */
1398 netdev_init(struct netdev *netdev, struct netdev_dev *netdev_dev)
1400 memset(netdev, 0, sizeof *netdev);
1401 netdev->netdev_dev = netdev_dev;
1402 netdev->ref_cnt = 1;
1403 list_push_back(&netdev_list, &netdev->node);
1406 /* Undoes the results of initialization.
1408 * Normally this function only needs to be called from netdev_close().
1409 * However, it may be called by providers due to an error on opening
1410 * that occurs after initialization. In this case netdev_close() would
1411 * never be called. */
1413 netdev_uninit(struct netdev *netdev, bool close)
1415 /* Restore flags that we changed, if any. */
1416 int error = restore_flags(netdev);
1417 list_remove(&netdev->node);
1419 VLOG_WARN("failed to restore network device flags on %s: %s",
1420 netdev_get_name(netdev), strerror(error));
1424 netdev_get_dev(netdev)->netdev_class->close(netdev);
1429 /* Returns the class type of 'netdev'.
1431 * The caller must not free the returned value. */
1433 netdev_get_type(const struct netdev *netdev)
1435 return netdev_get_dev(netdev)->netdev_class->type;
1439 netdev_get_dev(const struct netdev *netdev)
1441 return netdev->netdev_dev;
1444 /* Initializes 'notifier' as a netdev notifier for 'netdev', for which
1445 * notification will consist of calling 'cb', with auxiliary data 'aux'. */
1447 netdev_notifier_init(struct netdev_notifier *notifier, struct netdev *netdev,
1448 void (*cb)(struct netdev_notifier *), void *aux)
1450 notifier->netdev = netdev;
1452 notifier->aux = aux;
1455 /* Tracks changes in the status of a set of network devices. */
1456 struct netdev_monitor {
1457 struct shash polled_netdevs;
1458 struct shash changed_netdevs;
1461 /* Creates and returns a new structure for monitor changes in the status of
1462 * network devices. */
1463 struct netdev_monitor *
1464 netdev_monitor_create(void)
1466 struct netdev_monitor *monitor = xmalloc(sizeof *monitor);
1467 shash_init(&monitor->polled_netdevs);
1468 shash_init(&monitor->changed_netdevs);
1472 /* Destroys 'monitor'. */
1474 netdev_monitor_destroy(struct netdev_monitor *monitor)
1477 struct shash_node *node;
1479 SHASH_FOR_EACH (node, &monitor->polled_netdevs) {
1480 struct netdev_notifier *notifier = node->data;
1481 netdev_get_dev(notifier->netdev)->netdev_class->poll_remove(
1485 shash_destroy(&monitor->polled_netdevs);
1486 shash_destroy(&monitor->changed_netdevs);
1492 netdev_monitor_cb(struct netdev_notifier *notifier)
1494 struct netdev_monitor *monitor = notifier->aux;
1495 const char *name = netdev_get_name(notifier->netdev);
1496 shash_add_once(&monitor->changed_netdevs, name, NULL);
1499 /* Attempts to add 'netdev' as a netdev monitored by 'monitor'. Returns 0 if
1500 * successful, otherwise a positive errno value.
1502 * Adding a given 'netdev' to a monitor multiple times is equivalent to adding
1505 netdev_monitor_add(struct netdev_monitor *monitor, struct netdev *netdev)
1507 const char *netdev_name = netdev_get_name(netdev);
1509 if (!shash_find(&monitor->polled_netdevs, netdev_name)
1510 && netdev_get_dev(netdev)->netdev_class->poll_add)
1512 struct netdev_notifier *notifier;
1513 error = netdev_get_dev(netdev)->netdev_class->poll_add(netdev,
1514 netdev_monitor_cb, monitor, ¬ifier);
1516 assert(notifier->netdev == netdev);
1517 shash_add(&monitor->polled_netdevs, netdev_name, notifier);
1523 /* Removes 'netdev' from the set of netdevs monitored by 'monitor'. (This has
1524 * no effect if 'netdev' is not in the set of devices monitored by
1527 netdev_monitor_remove(struct netdev_monitor *monitor, struct netdev *netdev)
1529 const char *netdev_name = netdev_get_name(netdev);
1530 struct shash_node *node;
1532 node = shash_find(&monitor->polled_netdevs, netdev_name);
1534 /* Cancel future notifications. */
1535 struct netdev_notifier *notifier = node->data;
1536 netdev_get_dev(netdev)->netdev_class->poll_remove(notifier);
1537 shash_delete(&monitor->polled_netdevs, node);
1539 /* Drop any pending notification. */
1540 node = shash_find(&monitor->changed_netdevs, netdev_name);
1542 shash_delete(&monitor->changed_netdevs, node);
1547 /* Checks for changes to netdevs in the set monitored by 'monitor'. If any of
1548 * the attributes (Ethernet address, carrier status, speed or peer-advertised
1549 * speed, flags, etc.) of a network device monitored by 'monitor' has changed,
1550 * sets '*devnamep' to the name of a device that has changed and returns 0.
1551 * The caller is responsible for freeing '*devnamep' (with free()).
1553 * If no devices have changed, sets '*devnamep' to NULL and returns EAGAIN. */
1555 netdev_monitor_poll(struct netdev_monitor *monitor, char **devnamep)
1557 struct shash_node *node = shash_first(&monitor->changed_netdevs);
1562 *devnamep = shash_steal(&monitor->changed_netdevs, node);
1567 /* Registers with the poll loop to wake up from the next call to poll_block()
1568 * when netdev_monitor_poll(monitor) would indicate that a device has
1571 netdev_monitor_poll_wait(const struct netdev_monitor *monitor)
1573 if (!shash_is_empty(&monitor->changed_netdevs)) {
1574 poll_immediate_wake();
1576 /* XXX Nothing needed here for netdev_linux, but maybe other netdev
1577 * classes need help. */
1581 /* Restore the network device flags on 'netdev' to those that were active
1582 * before we changed them. Returns 0 if successful, otherwise a positive
1585 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
1587 restore_flags(struct netdev *netdev)
1589 if (netdev->changed_flags) {
1590 enum netdev_flags restore = netdev->save_flags & netdev->changed_flags;
1591 enum netdev_flags old_flags;
1592 return netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
1593 netdev->changed_flags & ~restore,
1594 restore, &old_flags);
1599 /* Close all netdevs on shutdown so they can do any needed cleanup such as
1600 * destroying devices, restoring flags, etc. */
1602 close_all_netdevs(void *aux OVS_UNUSED)
1604 struct netdev *netdev, *next;
1605 LIST_FOR_EACH_SAFE(netdev, next, node, &netdev_list) {
1606 netdev_close(netdev);