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 COVERAGE_DEFINE(netdev_received);
46 COVERAGE_DEFINE(netdev_sent);
47 COVERAGE_DEFINE(netdev_add_router);
48 COVERAGE_DEFINE(netdev_get_stats);
50 static struct shash netdev_classes = SHASH_INITIALIZER(&netdev_classes);
52 /* All created network devices. */
53 static struct shash netdev_dev_shash = SHASH_INITIALIZER(&netdev_dev_shash);
55 /* All open network devices. */
56 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
58 /* This is set pretty low because we probably won't learn anything from the
59 * additional log messages. */
60 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
62 static void close_all_netdevs(void *aux OVS_UNUSED);
63 static int restore_flags(struct netdev *netdev);
64 void update_device_args(struct netdev_dev *, const struct shash *args);
67 netdev_initialize(void)
74 fatal_signal_add_hook(close_all_netdevs, NULL, NULL, true);
77 netdev_register_provider(&netdev_linux_class);
78 netdev_register_provider(&netdev_tap_class);
79 netdev_vport_register();
84 /* Performs periodic work needed by all the various kinds of netdevs.
86 * If your program opens any netdevs, it must call this function within its
91 struct shash_node *node;
92 SHASH_FOR_EACH(node, &netdev_classes) {
93 const struct netdev_class *netdev_class = node->data;
94 if (netdev_class->run) {
100 /* Arranges for poll_block() to wake up when netdev_run() needs to be called.
102 * If your program opens any netdevs, it must call this function within its
107 struct shash_node *node;
108 SHASH_FOR_EACH(node, &netdev_classes) {
109 const struct netdev_class *netdev_class = node->data;
110 if (netdev_class->wait) {
111 netdev_class->wait();
116 /* Initializes and registers a new netdev provider. After successful
117 * registration, new netdevs of that type can be opened using netdev_open(). */
119 netdev_register_provider(const struct netdev_class *new_class)
121 if (shash_find(&netdev_classes, new_class->type)) {
122 VLOG_WARN("attempted to register duplicate netdev provider: %s",
127 if (new_class->init) {
128 int error = new_class->init();
130 VLOG_ERR("failed to initialize %s network device class: %s",
131 new_class->type, strerror(error));
136 shash_add(&netdev_classes, new_class->type, new_class);
141 /* Unregisters a netdev provider. 'type' must have been previously
142 * registered and not currently be in use by any netdevs. After unregistration
143 * new netdevs of that type cannot be opened using netdev_open(). */
145 netdev_unregister_provider(const char *type)
147 struct shash_node *del_node, *netdev_dev_node;
149 del_node = shash_find(&netdev_classes, type);
151 VLOG_WARN("attempted to unregister a netdev provider that is not "
152 "registered: %s", type);
156 SHASH_FOR_EACH(netdev_dev_node, &netdev_dev_shash) {
157 struct netdev_dev *netdev_dev = netdev_dev_node->data;
158 if (!strcmp(netdev_dev->netdev_class->type, type)) {
159 VLOG_WARN("attempted to unregister in use netdev provider: %s",
165 shash_delete(&netdev_classes, del_node);
170 /* Clears 'types' and enumerates the types of all currently registered netdev
171 * providers into it. The caller must first initialize the svec. */
173 netdev_enumerate_types(struct svec *types)
175 struct shash_node *node;
180 SHASH_FOR_EACH(node, &netdev_classes) {
181 const struct netdev_class *netdev_class = node->data;
182 svec_add(types, netdev_class->type);
186 /* Compares 'args' to those used to those used by 'dev'. Returns true
187 * if the arguments are the same, false otherwise. Does not update the
188 * values stored in 'dev'. */
190 compare_device_args(const struct netdev_dev *dev, const struct shash *args)
192 const struct shash_node **new_args;
196 if (shash_count(args) != dev->n_args) {
200 new_args = shash_sort(args);
201 for (i = 0; i < dev->n_args; i++) {
202 if (strcmp(dev->args[i].key, new_args[i]->name) ||
203 strcmp(dev->args[i].value, new_args[i]->data)) {
215 compare_args(const void *a_, const void *b_)
217 const struct arg *a = a_;
218 const struct arg *b = b_;
219 return strcmp(a->key, b->key);
223 update_device_args(struct netdev_dev *dev, const struct shash *args)
225 struct shash_node *node;
229 for (i = 0; i < dev->n_args; i++) {
230 free(dev->args[i].key);
231 free(dev->args[i].value);
238 if (!args || shash_is_empty(args)) {
242 dev->n_args = shash_count(args);
243 dev->args = xmalloc(dev->n_args * sizeof *dev->args);
246 SHASH_FOR_EACH(node, args) {
247 dev->args[i].key = xstrdup(node->name);
248 dev->args[i].value = xstrdup(node->data);
252 qsort(dev->args, dev->n_args, sizeof *dev->args, compare_args);
256 create_device(struct netdev_options *options, struct netdev_dev **netdev_devp)
258 struct netdev_class *netdev_class;
261 if (!options->type || strlen(options->type) == 0) {
262 /* Default to system. */
263 options->type = "system";
266 netdev_class = shash_find_data(&netdev_classes, options->type);
271 error = netdev_class->create(netdev_class, options->name, options->args,
273 assert(error || (*netdev_devp)->netdev_class == netdev_class);
277 /* Opens the network device named 'name' (e.g. "eth0") and returns zero if
278 * successful, otherwise a positive errno value. On success, sets '*netdevp'
279 * to the new network device, otherwise to null.
281 * If this is the first time the device has been opened, then create is called
282 * before opening. The device is created using the given type and arguments.
284 * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
285 * capture frames of that type received on the device. It may also be one of
286 * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those
289 netdev_open(struct netdev_options *options, struct netdev **netdevp)
291 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
292 struct netdev_dev *netdev_dev;
298 if (!options->args) {
299 options->args = &empty_args;
302 netdev_dev = shash_find_data(&netdev_dev_shash, options->name);
305 error = create_device(options, &netdev_dev);
307 if (error == EAFNOSUPPORT) {
308 VLOG_WARN("could not create netdev %s of unknown type %s",
309 options->name, options->type);
313 update_device_args(netdev_dev, options->args);
315 } else if (!shash_is_empty(options->args) &&
316 !compare_device_args(netdev_dev, options->args)) {
318 VLOG_WARN("%s: attempted to open already open netdev with "
319 "different arguments", options->name);
323 error = netdev_dev->netdev_class->open(netdev_dev, options->ethertype,
327 netdev_dev->ref_cnt++;
329 if (!netdev_dev->ref_cnt) {
330 netdev_dev_uninit(netdev_dev, true);
338 netdev_open_default(const char *name, struct netdev **netdevp)
340 struct netdev_options options;
342 memset(&options, 0, sizeof options);
344 options.ethertype = NETDEV_ETH_TYPE_NONE;
346 return netdev_open(&options, netdevp);
349 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
350 * or NULL if none are needed. */
352 netdev_reconfigure(struct netdev *netdev, const struct shash *args)
354 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
355 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
361 if (netdev_dev->netdev_class->reconfigure) {
362 if (!compare_device_args(netdev_dev, args)) {
363 update_device_args(netdev_dev, args);
364 return netdev_dev->netdev_class->reconfigure(netdev_dev, args);
366 } else if (!shash_is_empty(args)) {
367 VLOG_WARN("%s: arguments provided to device that does not have a "
368 "reconfigure function", netdev_get_name(netdev));
374 /* Closes and destroys 'netdev'. */
376 netdev_close(struct netdev *netdev)
379 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
381 assert(netdev_dev->ref_cnt);
382 netdev_dev->ref_cnt--;
383 netdev_uninit(netdev, true);
385 /* If the reference count for the netdev device is zero, destroy it. */
386 if (!netdev_dev->ref_cnt) {
387 netdev_dev_uninit(netdev_dev, true);
392 /* Returns true if a network device named 'name' exists and may be opened,
393 * otherwise false. */
395 netdev_exists(const char *name)
397 struct netdev *netdev;
400 error = netdev_open_default(name, &netdev);
402 netdev_close(netdev);
405 if (error != ENODEV) {
406 VLOG_WARN("failed to open network device %s: %s",
407 name, strerror(error));
413 /* Returns true if a network device named 'name' is currently opened,
414 * otherwise false. */
416 netdev_is_open(const char *name)
418 return !!shash_find_data(&netdev_dev_shash, name);
421 /* Clears 'svec' and enumerates the names of all known network devices. */
423 netdev_enumerate(struct svec *svec)
425 struct shash_node *node;
431 SHASH_FOR_EACH(node, &netdev_classes) {
432 const struct netdev_class *netdev_class = node->data;
433 if (netdev_class->enumerate) {
434 int retval = netdev_class->enumerate(svec);
436 VLOG_WARN("failed to enumerate %s network devices: %s",
437 netdev_class->type, strerror(retval));
448 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
449 * must have initialized with sufficient room for the packet. The space
450 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
451 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
452 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
453 * need not be included.)
455 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
456 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
457 * positive errno value. Returns EAGAIN immediately if no packet is ready to
460 * Some network devices may not implement support for this function. In such
461 * cases this function will always return EOPNOTSUPP. */
463 netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
465 int (*recv)(struct netdev *, void *, size_t);
468 assert(buffer->size == 0);
469 assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
471 recv = netdev_get_dev(netdev)->netdev_class->recv;
473 ? (recv)(netdev, buffer->data, ofpbuf_tailroom(buffer))
476 COVERAGE_INC(netdev_received);
477 buffer->size += retval;
478 if (buffer->size < ETH_TOTAL_MIN) {
479 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
487 /* Registers with the poll loop to wake up from the next call to poll_block()
488 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
490 netdev_recv_wait(struct netdev *netdev)
492 void (*recv_wait)(struct netdev *);
494 recv_wait = netdev_get_dev(netdev)->netdev_class->recv_wait;
500 /* Discards all packets waiting to be received from 'netdev'. */
502 netdev_drain(struct netdev *netdev)
504 int (*drain)(struct netdev *);
506 drain = netdev_get_dev(netdev)->netdev_class->drain;
507 return drain ? drain(netdev) : 0;
510 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
511 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
512 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
513 * the packet is too big or too small to transmit on the device.
515 * The caller retains ownership of 'buffer' in all cases.
517 * The kernel maintains a packet transmission queue, so the caller is not
518 * expected to do additional queuing of packets.
520 * Some network devices may not implement support for this function. In such
521 * cases this function will always return EOPNOTSUPP. */
523 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
525 int (*send)(struct netdev *, const void *, size_t);
528 send = netdev_get_dev(netdev)->netdev_class->send;
529 error = send ? (send)(netdev, buffer->data, buffer->size) : EOPNOTSUPP;
531 COVERAGE_INC(netdev_sent);
536 /* Registers with the poll loop to wake up from the next call to poll_block()
537 * when the packet transmission queue has sufficient room to transmit a packet
538 * with netdev_send().
540 * The kernel maintains a packet transmission queue, so the client is not
541 * expected to do additional queuing of packets. Thus, this function is
542 * unlikely to ever be used. It is included for completeness. */
544 netdev_send_wait(struct netdev *netdev)
546 void (*send_wait)(struct netdev *);
548 send_wait = netdev_get_dev(netdev)->netdev_class->send_wait;
554 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
555 * otherwise a positive errno value. */
557 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
559 return netdev_get_dev(netdev)->netdev_class->set_etheraddr(netdev, mac);
562 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
563 * the MAC address into 'mac'. On failure, returns a positive errno value and
564 * clears 'mac' to all-zeros. */
566 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
568 return netdev_get_dev(netdev)->netdev_class->get_etheraddr(netdev, mac);
571 /* Returns the name of the network device that 'netdev' represents,
572 * e.g. "eth0". The caller must not modify or free the returned string. */
574 netdev_get_name(const struct netdev *netdev)
576 return netdev_get_dev(netdev)->name;
579 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
580 * (and received) packets, in bytes, not including the hardware header; thus,
581 * this is typically 1500 bytes for Ethernet devices.
583 * If successful, returns 0 and stores the MTU size in '*mtup'. On failure,
584 * returns a positive errno value and stores ETH_PAYLOAD_MAX (1500) in
587 netdev_get_mtu(const struct netdev *netdev, int *mtup)
589 int error = netdev_get_dev(netdev)->netdev_class->get_mtu(netdev, mtup);
591 VLOG_WARN_RL(&rl, "failed to retrieve MTU for network device %s: %s",
592 netdev_get_name(netdev), strerror(error));
593 *mtup = ETH_PAYLOAD_MAX;
598 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
599 * failure, returns a negative errno value.
601 * The desired semantics of the ifindex value are a combination of those
602 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
603 * value should be unique within a host and remain stable at least until
604 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
605 * but many systems do not follow this rule anyhow.
607 * Some network devices may not implement support for this function. In such
608 * cases this function will always return -EOPNOTSUPP.
611 netdev_get_ifindex(const struct netdev *netdev)
613 int (*get_ifindex)(const struct netdev *);
615 get_ifindex = netdev_get_dev(netdev)->netdev_class->get_ifindex;
617 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
620 /* Stores the features supported by 'netdev' into each of '*current',
621 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
622 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
623 * successful, otherwise a positive errno value. On failure, all of the
624 * passed-in values are set to 0.
626 * Some network devices may not implement support for this function. In such
627 * cases this function will always return EOPNOTSUPP. */
629 netdev_get_features(struct netdev *netdev,
630 uint32_t *current, uint32_t *advertised,
631 uint32_t *supported, uint32_t *peer)
633 int (*get_features)(struct netdev *netdev,
634 uint32_t *current, uint32_t *advertised,
635 uint32_t *supported, uint32_t *peer);
643 advertised = &dummy[1];
646 supported = &dummy[2];
652 get_features = netdev_get_dev(netdev)->netdev_class->get_features;
654 ? get_features(netdev, current, advertised, supported, peer)
657 *current = *advertised = *supported = *peer = 0;
662 /* Returns the maximum speed of a network connection that has the "enum
663 * ofp_port_features" bits in 'features', in bits per second. If no bits that
664 * indicate a speed are set in 'features', assumes 100Mbps. */
666 netdev_features_to_bps(uint32_t features)
669 F_10000MB = OFPPF_10GB_FD,
670 F_1000MB = OFPPF_1GB_HD | OFPPF_1GB_FD,
671 F_100MB = OFPPF_100MB_HD | OFPPF_100MB_FD,
672 F_10MB = OFPPF_10MB_HD | OFPPF_10MB_FD
675 return ( features & F_10000MB ? UINT64_C(10000000000)
676 : features & F_1000MB ? UINT64_C(1000000000)
677 : features & F_100MB ? UINT64_C(100000000)
678 : features & F_10MB ? UINT64_C(10000000)
679 : UINT64_C(100000000));
682 /* Returns true if any of the "enum ofp_port_features" bits that indicate a
683 * full-duplex link are set in 'features', otherwise false. */
685 netdev_features_is_full_duplex(uint32_t features)
687 return (features & (OFPPF_10MB_FD | OFPPF_100MB_FD | OFPPF_1GB_FD
688 | OFPPF_10GB_FD)) != 0;
691 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
692 * successful, otherwise a positive errno value. */
694 netdev_set_advertisements(struct netdev *netdev, uint32_t advertise)
696 return (netdev_get_dev(netdev)->netdev_class->set_advertisements
697 ? netdev_get_dev(netdev)->netdev_class->set_advertisements(
702 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
703 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
704 * errno value and sets '*address' to 0 (INADDR_ANY).
706 * The following error values have well-defined meanings:
708 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
710 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
712 * 'address' or 'netmask' or both may be null, in which case the address or
713 * netmask is not reported. */
715 netdev_get_in4(const struct netdev *netdev,
716 struct in_addr *address_, struct in_addr *netmask_)
718 struct in_addr address;
719 struct in_addr netmask;
722 error = (netdev_get_dev(netdev)->netdev_class->get_in4
723 ? netdev_get_dev(netdev)->netdev_class->get_in4(netdev,
727 address_->s_addr = error ? 0 : address.s_addr;
730 netmask_->s_addr = error ? 0 : netmask.s_addr;
735 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
736 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
737 * positive errno value. */
739 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
741 return (netdev_get_dev(netdev)->netdev_class->set_in4
742 ? netdev_get_dev(netdev)->netdev_class->set_in4(netdev, addr, mask)
746 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
749 netdev_add_router(struct netdev *netdev, struct in_addr router)
751 COVERAGE_INC(netdev_add_router);
752 return (netdev_get_dev(netdev)->netdev_class->add_router
753 ? netdev_get_dev(netdev)->netdev_class->add_router(netdev, router)
757 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
758 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
759 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
760 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
761 * a directly connected network) in '*next_hop' and a copy of the name of the
762 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
763 * responsible for freeing '*netdev_name' (by calling free()). */
765 netdev_get_next_hop(const struct netdev *netdev,
766 const struct in_addr *host, struct in_addr *next_hop,
769 int error = (netdev_get_dev(netdev)->netdev_class->get_next_hop
770 ? netdev_get_dev(netdev)->netdev_class->get_next_hop(
771 host, next_hop, netdev_name)
774 next_hop->s_addr = 0;
780 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
781 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
782 * all-zero-bits (in6addr_any).
784 * The following error values have well-defined meanings:
786 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
788 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
790 * 'in6' may be null, in which case the address itself is not reported. */
792 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
794 struct in6_addr dummy;
797 error = (netdev_get_dev(netdev)->netdev_class->get_in6
798 ? netdev_get_dev(netdev)->netdev_class->get_in6(netdev,
802 memset(in6, 0, sizeof *in6);
807 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
808 * 'on'. If 'permanent' is true, the changes will persist; otherwise, they
809 * will be reverted when 'netdev' is closed or the program exits. Returns 0 if
810 * successful, otherwise a positive errno value. */
812 do_update_flags(struct netdev *netdev, enum netdev_flags off,
813 enum netdev_flags on, enum netdev_flags *old_flagsp,
816 enum netdev_flags old_flags;
819 error = netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
820 off & ~on, on, &old_flags);
822 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
823 off || on ? "set" : "get", netdev_get_name(netdev),
826 } else if ((off || on) && !permanent) {
827 enum netdev_flags new_flags = (old_flags & ~off) | on;
828 enum netdev_flags changed_flags = old_flags ^ new_flags;
830 if (!netdev->changed_flags) {
831 netdev->save_flags = old_flags;
833 netdev->changed_flags |= changed_flags;
837 *old_flagsp = old_flags;
842 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
843 * Returns 0 if successful, otherwise a positive errno value. On failure,
844 * stores 0 into '*flagsp'. */
846 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
848 struct netdev *netdev = (struct netdev *) netdev_;
849 return do_update_flags(netdev, 0, 0, flagsp, false);
852 /* Sets the flags for 'netdev' to 'flags'.
853 * If 'permanent' is true, the changes will persist; otherwise, they
854 * will be reverted when 'netdev' is closed or the program exits.
855 * Returns 0 if successful, otherwise a positive errno value. */
857 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
860 return do_update_flags(netdev, -1, flags, NULL, permanent);
863 /* Turns on the specified 'flags' on 'netdev'.
864 * If 'permanent' is true, the changes will persist; otherwise, they
865 * will be reverted when 'netdev' is closed or the program exits.
866 * Returns 0 if successful, otherwise a positive errno value. */
868 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
871 return do_update_flags(netdev, 0, flags, NULL, permanent);
874 /* Turns off the specified 'flags' on 'netdev'.
875 * If 'permanent' is true, the changes will persist; otherwise, they
876 * will be reverted when 'netdev' is closed or the program exits.
877 * Returns 0 if successful, otherwise a positive errno value. */
879 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
882 return do_update_flags(netdev, flags, 0, NULL, permanent);
885 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
886 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
887 * returns 0. Otherwise, it returns a positive errno value; in particular,
888 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
890 netdev_arp_lookup(const struct netdev *netdev,
891 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
893 int error = (netdev_get_dev(netdev)->netdev_class->arp_lookup
894 ? netdev_get_dev(netdev)->netdev_class->arp_lookup(netdev,
898 memset(mac, 0, ETH_ADDR_LEN);
903 /* Returns true if carrier is active (link light is on) on 'netdev'. */
905 netdev_get_carrier(const struct netdev *netdev)
908 enum netdev_flags flags;
911 netdev_get_flags(netdev, &flags);
912 if (!(flags & NETDEV_UP)) {
916 if (!netdev_get_dev(netdev)->netdev_class->get_carrier) {
920 error = netdev_get_dev(netdev)->netdev_class->get_carrier(netdev,
923 VLOG_DBG("%s: failed to get network device carrier status, assuming "
924 "down: %s", netdev_get_name(netdev), strerror(error));
931 /* Retrieves current device stats for 'netdev'. */
933 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
937 COVERAGE_INC(netdev_get_stats);
938 error = (netdev_get_dev(netdev)->netdev_class->get_stats
939 ? netdev_get_dev(netdev)->netdev_class->get_stats(netdev, stats)
942 memset(stats, 0xff, sizeof *stats);
947 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
948 * Returns 0 if successful, otherwise a positive errno value.
950 * This will probably fail for most network devices. Some devices might only
951 * allow setting their stats to 0. */
953 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
955 return (netdev_get_dev(netdev)->netdev_class->set_stats
956 ? netdev_get_dev(netdev)->netdev_class->set_stats(netdev, stats)
960 /* Attempts to set input rate limiting (policing) policy, such that up to
961 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
962 * size of 'kbits' kb. */
964 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
965 uint32_t kbits_burst)
967 return (netdev_get_dev(netdev)->netdev_class->set_policing
968 ? netdev_get_dev(netdev)->netdev_class->set_policing(netdev,
969 kbits_rate, kbits_burst)
973 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
974 * empty if 'netdev' does not support QoS. Any names added to 'types' should
975 * be documented as valid for the "type" column in the "QoS" table in
976 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
978 * Every network device supports disabling QoS with a type of "", but this type
979 * will not be added to 'types'.
981 * The caller must initialize 'types' (e.g. with svec_init()) before calling
982 * this function. The caller is responsible for destroying 'types' (e.g. with
983 * svec_destroy()) when it is no longer needed.
985 * Returns 0 if successful, otherwise a positive errno value. */
987 netdev_get_qos_types(const struct netdev *netdev, struct svec *types)
989 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
990 return (class->get_qos_types
991 ? class->get_qos_types(netdev, types)
995 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
996 * which should be "" or one of the types returned by netdev_get_qos_types()
997 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
998 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
999 * 'caps' to all zeros. */
1001 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
1002 struct netdev_qos_capabilities *caps)
1004 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1007 int retval = (class->get_qos_capabilities
1008 ? class->get_qos_capabilities(netdev, type, caps)
1011 memset(caps, 0, sizeof *caps);
1015 /* Every netdev supports turning off QoS. */
1016 memset(caps, 0, sizeof *caps);
1021 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1022 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1023 * the number of queues (zero on failure) in '*n_queuesp'.
1025 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1027 netdev_get_n_queues(const struct netdev *netdev,
1028 const char *type, unsigned int *n_queuesp)
1030 struct netdev_qos_capabilities caps;
1033 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1034 *n_queuesp = caps.n_queues;
1038 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1039 * stores the name of the current form of QoS into '*typep', stores any details
1040 * of configuration as string key-value pairs in 'details', and returns 0. On
1041 * failure, sets '*typep' to NULL and returns a positive errno value.
1043 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1045 * The caller must initialize 'details' as an empty shash (e.g. with
1046 * shash_init()) before calling this function. The caller must free 'details',
1047 * including 'data' members, when it is no longer needed (e.g. with
1048 * shash_destroy_free_data()).
1050 * The caller must not modify or free '*typep'.
1052 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1053 * 'netdev'. The contents of 'details' should be documented as valid for
1054 * '*typep' in the "other_config" column in the "QoS" table in
1055 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1057 netdev_get_qos(const struct netdev *netdev,
1058 const char **typep, struct shash *details)
1060 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1063 if (class->get_qos) {
1064 retval = class->get_qos(netdev, typep, details);
1067 shash_clear_free_data(details);
1071 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1077 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1078 * with details of configuration from 'details'. Returns 0 if successful,
1079 * otherwise a positive errno value. On error, the previous QoS configuration
1082 * When this function changes the type of QoS (not just 'details'), this also
1083 * resets all queue configuration for 'netdev' to their defaults (which depend
1084 * on the specific type of QoS). Otherwise, the queue configuration for
1085 * 'netdev' is unchanged.
1087 * 'type' should be "" (to disable QoS) or one of the types returned by
1088 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1089 * documented as valid for the given 'type' in the "other_config" column in the
1090 * "QoS" table in vswitchd/vswitch.xml (which is built as
1091 * ovs-vswitchd.conf.db(8)).
1093 * NULL may be specified for 'details' if there are no configuration
1096 netdev_set_qos(struct netdev *netdev,
1097 const char *type, const struct shash *details)
1099 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1105 if (class->set_qos) {
1107 static struct shash empty = SHASH_INITIALIZER(&empty);
1110 return class->set_qos(netdev, type, details);
1112 return *type ? EOPNOTSUPP : 0;
1116 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1117 * successful, adds that information as string key-value pairs to 'details'.
1118 * Returns 0 if successful, otherwise a positive errno value.
1120 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1121 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1123 * The returned contents of 'details' should be documented as valid for the
1124 * given 'type' in the "other_config" column in the "Queue" table in
1125 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1127 * The caller must initialize 'details' (e.g. with shash_init()) before calling
1128 * this function. The caller must free 'details', including 'data' members,
1129 * when it is no longer needed (e.g. with shash_destroy_free_data()). */
1131 netdev_get_queue(const struct netdev *netdev,
1132 unsigned int queue_id, struct shash *details)
1134 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1137 retval = (class->get_queue
1138 ? class->get_queue(netdev, queue_id, details)
1141 shash_clear_free_data(details);
1146 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1147 * string pairs in 'details'. The contents of 'details' should be documented
1148 * as valid for the given 'type' in the "other_config" column in the "Queue"
1149 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1150 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1151 * given queue's configuration should be unmodified.
1153 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1154 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1156 * This function does not modify 'details', and the caller retains ownership of
1159 netdev_set_queue(struct netdev *netdev,
1160 unsigned int queue_id, const struct shash *details)
1162 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1163 return (class->set_queue
1164 ? class->set_queue(netdev, queue_id, details)
1168 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1169 * of QoS may have a fixed set of queues, in which case attempts to delete them
1170 * will fail with EOPNOTSUPP.
1172 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1173 * given queue will be unmodified.
1175 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1176 * the current form of QoS (e.g. as returned by
1177 * netdev_get_n_queues(netdev)). */
1179 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1181 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1182 return (class->delete_queue
1183 ? class->delete_queue(netdev, queue_id)
1187 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1188 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1189 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1190 * positive errno value and fills 'stats' with all-1-bits. */
1192 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1193 struct netdev_queue_stats *stats)
1195 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1198 retval = (class->get_queue_stats
1199 ? class->get_queue_stats(netdev, queue_id, stats)
1202 memset(stats, 0xff, sizeof *stats);
1207 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1208 * its configuration, and the 'aux' specified by the caller. The order of
1209 * iteration is unspecified, but (when successful) each queue is visited
1212 * Calling this function may be more efficient than calling netdev_get_queue()
1215 * 'cb' must not modify or free the 'details' argument passed in.
1217 * Returns 0 if successful, otherwise a positive errno value. On error, some
1218 * configured queues may not have been included in the iteration. */
1220 netdev_dump_queues(const struct netdev *netdev,
1221 netdev_dump_queues_cb *cb, void *aux)
1223 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1224 return (class->dump_queues
1225 ? class->dump_queues(netdev, cb, aux)
1229 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1230 * its statistics, and the 'aux' specified by the caller. The order of
1231 * iteration is unspecified, but (when successful) each queue is visited
1234 * Calling this function may be more efficient than calling
1235 * netdev_get_queue_stats() for every queue.
1237 * 'cb' must not modify or free the statistics passed in.
1239 * Returns 0 if successful, otherwise a positive errno value. On error, some
1240 * configured queues may not have been included in the iteration. */
1242 netdev_dump_queue_stats(const struct netdev *netdev,
1243 netdev_dump_queue_stats_cb *cb, void *aux)
1245 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1246 return (class->dump_queue_stats
1247 ? class->dump_queue_stats(netdev, cb, aux)
1251 /* If 'netdev' is a VLAN network device (e.g. one created with vconfig(8)),
1252 * sets '*vlan_vid' to the VLAN VID associated with that device and returns 0.
1253 * Otherwise returns a errno value (specifically ENOENT if 'netdev_name' is the
1254 * name of a network device that is not a VLAN device) and sets '*vlan_vid' to
1257 netdev_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1259 int error = (netdev_get_dev(netdev)->netdev_class->get_vlan_vid
1260 ? netdev_get_dev(netdev)->netdev_class->get_vlan_vid(netdev,
1269 /* Returns a network device that has 'in4' as its IP address, if one exists,
1270 * otherwise a null pointer. */
1272 netdev_find_dev_by_in4(const struct in_addr *in4)
1274 struct netdev *netdev;
1275 struct svec dev_list = SVEC_EMPTY_INITIALIZER;
1278 netdev_enumerate(&dev_list);
1279 for (i = 0; i < dev_list.n; i++) {
1280 const char *name = dev_list.names[i];
1281 struct in_addr dev_in4;
1283 if (!netdev_open_default(name, &netdev)
1284 && !netdev_get_in4(netdev, &dev_in4, NULL)
1285 && dev_in4.s_addr == in4->s_addr) {
1288 netdev_close(netdev);
1293 svec_destroy(&dev_list);
1297 /* Initializes 'netdev_dev' as a netdev device named 'name' of the
1298 * specified 'netdev_class'.
1300 * This function adds 'netdev_dev' to a netdev-owned shash, so it is
1301 * very important that 'netdev_dev' only be freed after calling
1302 * the refcount drops to zero. */
1304 netdev_dev_init(struct netdev_dev *netdev_dev, const char *name,
1305 const struct netdev_class *netdev_class)
1307 assert(!shash_find(&netdev_dev_shash, name));
1309 memset(netdev_dev, 0, sizeof *netdev_dev);
1310 netdev_dev->netdev_class = netdev_class;
1311 netdev_dev->name = xstrdup(name);
1312 netdev_dev->node = shash_add(&netdev_dev_shash, name, netdev_dev);
1315 /* Undoes the results of initialization.
1317 * Normally this function does not need to be called as netdev_close has
1318 * the same effect when the refcount drops to zero.
1319 * However, it may be called by providers due to an error on creation
1320 * that occurs after initialization. It this case netdev_close() would
1321 * never be called. */
1323 netdev_dev_uninit(struct netdev_dev *netdev_dev, bool destroy)
1325 char *name = netdev_dev->name;
1327 assert(!netdev_dev->ref_cnt);
1329 shash_delete(&netdev_dev_shash, netdev_dev->node);
1330 update_device_args(netdev_dev, NULL);
1333 netdev_dev->netdev_class->destroy(netdev_dev);
1338 /* Returns the class type of 'netdev_dev'.
1340 * The caller must not free the returned value. */
1342 netdev_dev_get_type(const struct netdev_dev *netdev_dev)
1344 return netdev_dev->netdev_class->type;
1347 /* Returns the class associated with 'netdev_dev'. */
1348 const struct netdev_class *
1349 netdev_dev_get_class(const struct netdev_dev *netdev_dev)
1351 return netdev_dev->netdev_class;
1354 /* Returns the name of 'netdev_dev'.
1356 * The caller must not free the returned value. */
1358 netdev_dev_get_name(const struct netdev_dev *netdev_dev)
1360 return netdev_dev->name;
1363 /* Returns the netdev_dev with 'name' or NULL if there is none.
1365 * The caller must not free the returned value. */
1367 netdev_dev_from_name(const char *name)
1369 return shash_find_data(&netdev_dev_shash, name);
1372 /* Fills 'device_list' with devices that match 'netdev_class'.
1374 * The caller is responsible for initializing and destroying 'device_list'
1375 * but the contained netdev_devs must not be freed. */
1377 netdev_dev_get_devices(const struct netdev_class *netdev_class,
1378 struct shash *device_list)
1380 struct shash_node *node;
1381 SHASH_FOR_EACH (node, &netdev_dev_shash) {
1382 struct netdev_dev *dev = node->data;
1384 if (dev->netdev_class == netdev_class) {
1385 shash_add(device_list, node->name, node->data);
1390 /* Initializes 'netdev' as a instance of the netdev_dev.
1392 * This function adds 'netdev' to a netdev-owned linked list, so it is very
1393 * important that 'netdev' only be freed after calling netdev_close(). */
1395 netdev_init(struct netdev *netdev, struct netdev_dev *netdev_dev)
1397 memset(netdev, 0, sizeof *netdev);
1398 netdev->netdev_dev = netdev_dev;
1399 list_push_back(&netdev_list, &netdev->node);
1402 /* Undoes the results of initialization.
1404 * Normally this function only needs to be called from netdev_close().
1405 * However, it may be called by providers due to an error on opening
1406 * that occurs after initialization. It this case netdev_close() would
1407 * never be called. */
1409 netdev_uninit(struct netdev *netdev, bool close)
1411 /* Restore flags that we changed, if any. */
1412 int error = restore_flags(netdev);
1413 list_remove(&netdev->node);
1415 VLOG_WARN("failed to restore network device flags on %s: %s",
1416 netdev_get_name(netdev), strerror(error));
1420 netdev_get_dev(netdev)->netdev_class->close(netdev);
1425 /* Returns the class type of 'netdev'.
1427 * The caller must not free the returned value. */
1429 netdev_get_type(const struct netdev *netdev)
1431 return netdev_get_dev(netdev)->netdev_class->type;
1435 netdev_get_dev(const struct netdev *netdev)
1437 return netdev->netdev_dev;
1440 /* Initializes 'notifier' as a netdev notifier for 'netdev', for which
1441 * notification will consist of calling 'cb', with auxiliary data 'aux'. */
1443 netdev_notifier_init(struct netdev_notifier *notifier, struct netdev *netdev,
1444 void (*cb)(struct netdev_notifier *), void *aux)
1446 notifier->netdev = netdev;
1448 notifier->aux = aux;
1451 /* Tracks changes in the status of a set of network devices. */
1452 struct netdev_monitor {
1453 struct shash polled_netdevs;
1454 struct shash changed_netdevs;
1457 /* Creates and returns a new structure for monitor changes in the status of
1458 * network devices. */
1459 struct netdev_monitor *
1460 netdev_monitor_create(void)
1462 struct netdev_monitor *monitor = xmalloc(sizeof *monitor);
1463 shash_init(&monitor->polled_netdevs);
1464 shash_init(&monitor->changed_netdevs);
1468 /* Destroys 'monitor'. */
1470 netdev_monitor_destroy(struct netdev_monitor *monitor)
1473 struct shash_node *node;
1475 SHASH_FOR_EACH (node, &monitor->polled_netdevs) {
1476 struct netdev_notifier *notifier = node->data;
1477 netdev_get_dev(notifier->netdev)->netdev_class->poll_remove(
1481 shash_destroy(&monitor->polled_netdevs);
1482 shash_destroy(&monitor->changed_netdevs);
1488 netdev_monitor_cb(struct netdev_notifier *notifier)
1490 struct netdev_monitor *monitor = notifier->aux;
1491 const char *name = netdev_get_name(notifier->netdev);
1492 shash_add_once(&monitor->changed_netdevs, name, NULL);
1495 /* Attempts to add 'netdev' as a netdev monitored by 'monitor'. Returns 0 if
1496 * successful, otherwise a positive errno value.
1498 * Adding a given 'netdev' to a monitor multiple times is equivalent to adding
1501 netdev_monitor_add(struct netdev_monitor *monitor, struct netdev *netdev)
1503 const char *netdev_name = netdev_get_name(netdev);
1505 if (!shash_find(&monitor->polled_netdevs, netdev_name)
1506 && netdev_get_dev(netdev)->netdev_class->poll_add)
1508 struct netdev_notifier *notifier;
1509 error = netdev_get_dev(netdev)->netdev_class->poll_add(netdev,
1510 netdev_monitor_cb, monitor, ¬ifier);
1512 assert(notifier->netdev == netdev);
1513 shash_add(&monitor->polled_netdevs, netdev_name, notifier);
1519 /* Removes 'netdev' from the set of netdevs monitored by 'monitor'. (This has
1520 * no effect if 'netdev' is not in the set of devices monitored by
1523 netdev_monitor_remove(struct netdev_monitor *monitor, struct netdev *netdev)
1525 const char *netdev_name = netdev_get_name(netdev);
1526 struct shash_node *node;
1528 node = shash_find(&monitor->polled_netdevs, netdev_name);
1530 /* Cancel future notifications. */
1531 struct netdev_notifier *notifier = node->data;
1532 netdev_get_dev(netdev)->netdev_class->poll_remove(notifier);
1533 shash_delete(&monitor->polled_netdevs, node);
1535 /* Drop any pending notification. */
1536 node = shash_find(&monitor->changed_netdevs, netdev_name);
1538 shash_delete(&monitor->changed_netdevs, node);
1543 /* Checks for changes to netdevs in the set monitored by 'monitor'. If any of
1544 * the attributes (Ethernet address, carrier status, speed or peer-advertised
1545 * speed, flags, etc.) of a network device monitored by 'monitor' has changed,
1546 * sets '*devnamep' to the name of a device that has changed and returns 0.
1547 * The caller is responsible for freeing '*devnamep' (with free()).
1549 * If no devices have changed, sets '*devnamep' to NULL and returns EAGAIN. */
1551 netdev_monitor_poll(struct netdev_monitor *monitor, char **devnamep)
1553 struct shash_node *node = shash_first(&monitor->changed_netdevs);
1558 *devnamep = shash_steal(&monitor->changed_netdevs, node);
1563 /* Registers with the poll loop to wake up from the next call to poll_block()
1564 * when netdev_monitor_poll(monitor) would indicate that a device has
1567 netdev_monitor_poll_wait(const struct netdev_monitor *monitor)
1569 if (!shash_is_empty(&monitor->changed_netdevs)) {
1570 poll_immediate_wake();
1572 /* XXX Nothing needed here for netdev_linux, but maybe other netdev
1573 * classes need help. */
1577 /* Restore the network device flags on 'netdev' to those that were active
1578 * before we changed them. Returns 0 if successful, otherwise a positive
1581 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
1583 restore_flags(struct netdev *netdev)
1585 if (netdev->changed_flags) {
1586 enum netdev_flags restore = netdev->save_flags & netdev->changed_flags;
1587 enum netdev_flags old_flags;
1588 return netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
1589 netdev->changed_flags & ~restore,
1590 restore, &old_flags);
1595 /* Close all netdevs on shutdown so they can do any needed cleanup such as
1596 * destroying devices, restoring flags, etc. */
1598 close_all_netdevs(void *aux OVS_UNUSED)
1600 struct netdev *netdev, *next;
1601 LIST_FOR_EACH_SAFE(netdev, next, node, &netdev_list) {
1602 netdev_close(netdev);