2 * Copyright (c) 2008, 2009, 2010, 2011 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"
44 VLOG_DEFINE_THIS_MODULE(netdev);
46 COVERAGE_DEFINE(netdev_received);
47 COVERAGE_DEFINE(netdev_sent);
48 COVERAGE_DEFINE(netdev_add_router);
49 COVERAGE_DEFINE(netdev_get_stats);
51 static struct shash netdev_classes = SHASH_INITIALIZER(&netdev_classes);
53 /* All created network devices. */
54 static struct shash netdev_dev_shash = SHASH_INITIALIZER(&netdev_dev_shash);
56 /* All open network devices. */
57 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
59 /* This is set pretty low because we probably won't learn anything from the
60 * additional log messages. */
61 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
63 static void close_all_netdevs(void *aux OVS_UNUSED);
64 static int restore_flags(struct netdev *netdev);
65 void update_device_args(struct netdev_dev *, const struct shash *args);
68 netdev_initialize(void)
75 fatal_signal_add_hook(close_all_netdevs, NULL, NULL, true);
78 netdev_register_provider(&netdev_linux_class);
79 netdev_register_provider(&netdev_internal_class);
80 netdev_register_provider(&netdev_tap_class);
81 netdev_vport_register();
86 /* Performs periodic work needed by all the various kinds of netdevs.
88 * If your program opens any netdevs, it must call this function within its
93 struct shash_node *node;
94 SHASH_FOR_EACH(node, &netdev_classes) {
95 const struct netdev_class *netdev_class = node->data;
96 if (netdev_class->run) {
102 /* Arranges for poll_block() to wake up when netdev_run() needs to be called.
104 * If your program opens any netdevs, it must call this function within its
109 struct shash_node *node;
110 SHASH_FOR_EACH(node, &netdev_classes) {
111 const struct netdev_class *netdev_class = node->data;
112 if (netdev_class->wait) {
113 netdev_class->wait();
118 /* Initializes and registers a new netdev provider. After successful
119 * registration, new netdevs of that type can be opened using netdev_open(). */
121 netdev_register_provider(const struct netdev_class *new_class)
123 if (shash_find(&netdev_classes, new_class->type)) {
124 VLOG_WARN("attempted to register duplicate netdev provider: %s",
129 if (new_class->init) {
130 int error = new_class->init();
132 VLOG_ERR("failed to initialize %s network device class: %s",
133 new_class->type, strerror(error));
138 shash_add(&netdev_classes, new_class->type, new_class);
143 /* Unregisters a netdev provider. 'type' must have been previously
144 * registered and not currently be in use by any netdevs. After unregistration
145 * new netdevs of that type cannot be opened using netdev_open(). */
147 netdev_unregister_provider(const char *type)
149 struct shash_node *del_node, *netdev_dev_node;
151 del_node = shash_find(&netdev_classes, type);
153 VLOG_WARN("attempted to unregister a netdev provider that is not "
154 "registered: %s", type);
158 SHASH_FOR_EACH(netdev_dev_node, &netdev_dev_shash) {
159 struct netdev_dev *netdev_dev = netdev_dev_node->data;
160 if (!strcmp(netdev_dev->netdev_class->type, type)) {
161 VLOG_WARN("attempted to unregister in use netdev provider: %s",
167 shash_delete(&netdev_classes, del_node);
172 const struct netdev_class *
173 netdev_lookup_provider(const char *type)
176 return shash_find_data(&netdev_classes, type && type[0] ? type : "system");
179 /* Clears 'types' and enumerates the types of all currently registered netdev
180 * providers into it. The caller must first initialize the sset. */
182 netdev_enumerate_types(struct sset *types)
184 struct shash_node *node;
189 SHASH_FOR_EACH(node, &netdev_classes) {
190 const struct netdev_class *netdev_class = node->data;
191 sset_add(types, netdev_class->type);
196 update_device_args(struct netdev_dev *dev, const struct shash *args)
198 smap_destroy(&dev->args);
199 smap_clone(&dev->args, args);
202 /* Opens the network device named 'name' (e.g. "eth0") and returns zero if
203 * successful, otherwise a positive errno value. On success, sets '*netdevp'
204 * to the new network device, otherwise to null.
206 * If this is the first time the device has been opened, then create is called
207 * before opening. The device is created using the given type and arguments.
209 * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
210 * capture frames of that type received on the device. It may also be one of
211 * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those
214 netdev_open(struct netdev_options *options, struct netdev **netdevp)
216 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
217 struct netdev_dev *netdev_dev;
223 if (!options->args) {
224 options->args = &empty_args;
227 netdev_dev = shash_find_data(&netdev_dev_shash, options->name);
230 const struct netdev_class *class;
232 class = netdev_lookup_provider(options->type);
234 VLOG_WARN("could not create netdev %s of unknown type %s",
235 options->name, options->type);
238 error = class->create(class, options->name, options->args,
243 assert(netdev_dev->netdev_class == class);
245 } else if (!shash_is_empty(options->args) &&
246 !smap_equal(&netdev_dev->args, options->args)) {
248 VLOG_WARN("%s: attempted to open already open netdev with "
249 "different arguments", options->name);
253 error = netdev_dev->netdev_class->open(netdev_dev, options->ethertype,
257 netdev_dev->ref_cnt++;
259 if (!netdev_dev->ref_cnt) {
260 netdev_dev_uninit(netdev_dev, true);
268 netdev_open_default(const char *name, struct netdev **netdevp)
270 struct netdev_options options;
272 memset(&options, 0, sizeof options);
274 options.ethertype = NETDEV_ETH_TYPE_NONE;
276 return netdev_open(&options, netdevp);
279 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
280 * or NULL if none are needed. */
282 netdev_set_config(struct netdev *netdev, const struct shash *args)
284 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
285 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
291 if (netdev_dev->netdev_class->set_config) {
292 if (!smap_equal(&netdev_dev->args, args)) {
293 update_device_args(netdev_dev, args);
294 return netdev_dev->netdev_class->set_config(netdev_dev, args);
296 } else if (!shash_is_empty(args)) {
297 VLOG_WARN("%s: arguments provided to device whose configuration "
298 "cannot be changed", netdev_get_name(netdev));
304 /* Returns the current configuration for 'netdev'. This is either the
305 * configuration passed to netdev_open() or netdev_set_config(), or it is a
306 * configuration retrieved from the device itself if no configuration was
307 * passed to those functions.
309 * 'netdev' retains ownership of the returned configuration. */
311 netdev_get_config(const struct netdev *netdev)
313 return &netdev_get_dev(netdev)->args;
316 /* Closes and destroys 'netdev'. */
318 netdev_close(struct netdev *netdev)
321 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
323 assert(netdev_dev->ref_cnt);
324 netdev_dev->ref_cnt--;
325 netdev_uninit(netdev, true);
327 /* If the reference count for the netdev device is zero, destroy it. */
328 if (!netdev_dev->ref_cnt) {
329 netdev_dev_uninit(netdev_dev, true);
334 /* Returns true if a network device named 'name' exists and may be opened,
335 * otherwise false. */
337 netdev_exists(const char *name)
339 struct netdev *netdev;
342 error = netdev_open_default(name, &netdev);
344 netdev_close(netdev);
347 if (error != ENODEV) {
348 VLOG_WARN("failed to open network device %s: %s",
349 name, strerror(error));
355 /* Returns true if a network device named 'name' is currently opened,
356 * otherwise false. */
358 netdev_is_open(const char *name)
360 return !!shash_find_data(&netdev_dev_shash, name);
363 /* Clears 'sset' and enumerates the names of all known network devices. */
365 netdev_enumerate(struct sset *sset)
367 struct shash_node *node;
373 SHASH_FOR_EACH(node, &netdev_classes) {
374 const struct netdev_class *netdev_class = node->data;
375 if (netdev_class->enumerate) {
376 int retval = netdev_class->enumerate(sset);
378 VLOG_WARN("failed to enumerate %s network devices: %s",
379 netdev_class->type, strerror(retval));
390 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
391 * must have initialized with sufficient room for the packet. The space
392 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
393 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
394 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
395 * need not be included.)
397 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
398 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
399 * positive errno value. Returns EAGAIN immediately if no packet is ready to
402 * Some network devices may not implement support for this function. In such
403 * cases this function will always return EOPNOTSUPP. */
405 netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
407 int (*recv)(struct netdev *, void *, size_t);
410 assert(buffer->size == 0);
411 assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
413 recv = netdev_get_dev(netdev)->netdev_class->recv;
415 ? (recv)(netdev, buffer->data, ofpbuf_tailroom(buffer))
418 COVERAGE_INC(netdev_received);
419 buffer->size += retval;
420 if (buffer->size < ETH_TOTAL_MIN) {
421 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
429 /* Registers with the poll loop to wake up from the next call to poll_block()
430 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
432 netdev_recv_wait(struct netdev *netdev)
434 void (*recv_wait)(struct netdev *);
436 recv_wait = netdev_get_dev(netdev)->netdev_class->recv_wait;
442 /* Discards all packets waiting to be received from 'netdev'. */
444 netdev_drain(struct netdev *netdev)
446 int (*drain)(struct netdev *);
448 drain = netdev_get_dev(netdev)->netdev_class->drain;
449 return drain ? drain(netdev) : 0;
452 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
453 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
454 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
455 * the packet is too big or too small to transmit on the device.
457 * The caller retains ownership of 'buffer' in all cases.
459 * The kernel maintains a packet transmission queue, so the caller is not
460 * expected to do additional queuing of packets.
462 * Some network devices may not implement support for this function. In such
463 * cases this function will always return EOPNOTSUPP. */
465 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
467 int (*send)(struct netdev *, const void *, size_t);
470 send = netdev_get_dev(netdev)->netdev_class->send;
471 error = send ? (send)(netdev, buffer->data, buffer->size) : EOPNOTSUPP;
473 COVERAGE_INC(netdev_sent);
478 /* Registers with the poll loop to wake up from the next call to poll_block()
479 * when the packet transmission queue has sufficient room to transmit a packet
480 * with netdev_send().
482 * The kernel maintains a packet transmission queue, so the client is not
483 * expected to do additional queuing of packets. Thus, this function is
484 * unlikely to ever be used. It is included for completeness. */
486 netdev_send_wait(struct netdev *netdev)
488 void (*send_wait)(struct netdev *);
490 send_wait = netdev_get_dev(netdev)->netdev_class->send_wait;
496 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
497 * otherwise a positive errno value. */
499 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
501 return netdev_get_dev(netdev)->netdev_class->set_etheraddr(netdev, mac);
504 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
505 * the MAC address into 'mac'. On failure, returns a positive errno value and
506 * clears 'mac' to all-zeros. */
508 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
510 return netdev_get_dev(netdev)->netdev_class->get_etheraddr(netdev, mac);
513 /* Returns the name of the network device that 'netdev' represents,
514 * e.g. "eth0". The caller must not modify or free the returned string. */
516 netdev_get_name(const struct netdev *netdev)
518 return netdev_get_dev(netdev)->name;
521 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
522 * (and received) packets, in bytes, not including the hardware header; thus,
523 * this is typically 1500 bytes for Ethernet devices.
525 * If successful, returns 0 and stores the MTU size in '*mtup'. Stores INT_MAX
526 * in '*mtup' if 'netdev' does not have an MTU (as e.g. some tunnels do not).On
527 * failure, returns a positive errno value and stores ETH_PAYLOAD_MAX (1500) in
530 netdev_get_mtu(const struct netdev *netdev, int *mtup)
532 int error = netdev_get_dev(netdev)->netdev_class->get_mtu(netdev, mtup);
534 VLOG_WARN_RL(&rl, "failed to retrieve MTU for network device %s: %s",
535 netdev_get_name(netdev), strerror(error));
536 *mtup = ETH_PAYLOAD_MAX;
541 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
542 * failure, returns a negative errno value.
544 * The desired semantics of the ifindex value are a combination of those
545 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
546 * value should be unique within a host and remain stable at least until
547 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
548 * but many systems do not follow this rule anyhow.
550 * Some network devices may not implement support for this function. In such
551 * cases this function will always return -EOPNOTSUPP.
554 netdev_get_ifindex(const struct netdev *netdev)
556 int (*get_ifindex)(const struct netdev *);
558 get_ifindex = netdev_get_dev(netdev)->netdev_class->get_ifindex;
560 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
563 /* Stores the features supported by 'netdev' into each of '*current',
564 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
565 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
566 * successful, otherwise a positive errno value. On failure, all of the
567 * passed-in values are set to 0.
569 * Some network devices may not implement support for this function. In such
570 * cases this function will always return EOPNOTSUPP. */
572 netdev_get_features(const struct netdev *netdev,
573 uint32_t *current, uint32_t *advertised,
574 uint32_t *supported, uint32_t *peer)
576 int (*get_features)(const struct netdev *netdev,
577 uint32_t *current, uint32_t *advertised,
578 uint32_t *supported, uint32_t *peer);
586 advertised = &dummy[1];
589 supported = &dummy[2];
595 get_features = netdev_get_dev(netdev)->netdev_class->get_features;
597 ? get_features(netdev, current, advertised, supported, peer)
600 *current = *advertised = *supported = *peer = 0;
605 /* Returns the maximum speed of a network connection that has the "enum
606 * ofp_port_features" bits in 'features', in bits per second. If no bits that
607 * indicate a speed are set in 'features', assumes 100Mbps. */
609 netdev_features_to_bps(uint32_t features)
612 F_10000MB = OFPPF_10GB_FD,
613 F_1000MB = OFPPF_1GB_HD | OFPPF_1GB_FD,
614 F_100MB = OFPPF_100MB_HD | OFPPF_100MB_FD,
615 F_10MB = OFPPF_10MB_HD | OFPPF_10MB_FD
618 return ( features & F_10000MB ? UINT64_C(10000000000)
619 : features & F_1000MB ? UINT64_C(1000000000)
620 : features & F_100MB ? UINT64_C(100000000)
621 : features & F_10MB ? UINT64_C(10000000)
622 : UINT64_C(100000000));
625 /* Returns true if any of the "enum ofp_port_features" bits that indicate a
626 * full-duplex link are set in 'features', otherwise false. */
628 netdev_features_is_full_duplex(uint32_t features)
630 return (features & (OFPPF_10MB_FD | OFPPF_100MB_FD | OFPPF_1GB_FD
631 | OFPPF_10GB_FD)) != 0;
634 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
635 * successful, otherwise a positive errno value. */
637 netdev_set_advertisements(struct netdev *netdev, uint32_t advertise)
639 return (netdev_get_dev(netdev)->netdev_class->set_advertisements
640 ? netdev_get_dev(netdev)->netdev_class->set_advertisements(
645 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
646 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
647 * errno value and sets '*address' to 0 (INADDR_ANY).
649 * The following error values have well-defined meanings:
651 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
653 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
655 * 'address' or 'netmask' or both may be null, in which case the address or
656 * netmask is not reported. */
658 netdev_get_in4(const struct netdev *netdev,
659 struct in_addr *address_, struct in_addr *netmask_)
661 struct in_addr address;
662 struct in_addr netmask;
665 error = (netdev_get_dev(netdev)->netdev_class->get_in4
666 ? netdev_get_dev(netdev)->netdev_class->get_in4(netdev,
670 address_->s_addr = error ? 0 : address.s_addr;
673 netmask_->s_addr = error ? 0 : netmask.s_addr;
678 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
679 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
680 * positive errno value. */
682 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
684 return (netdev_get_dev(netdev)->netdev_class->set_in4
685 ? netdev_get_dev(netdev)->netdev_class->set_in4(netdev, addr, mask)
689 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
692 netdev_add_router(struct netdev *netdev, struct in_addr router)
694 COVERAGE_INC(netdev_add_router);
695 return (netdev_get_dev(netdev)->netdev_class->add_router
696 ? netdev_get_dev(netdev)->netdev_class->add_router(netdev, router)
700 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
701 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
702 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
703 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
704 * a directly connected network) in '*next_hop' and a copy of the name of the
705 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
706 * responsible for freeing '*netdev_name' (by calling free()). */
708 netdev_get_next_hop(const struct netdev *netdev,
709 const struct in_addr *host, struct in_addr *next_hop,
712 int error = (netdev_get_dev(netdev)->netdev_class->get_next_hop
713 ? netdev_get_dev(netdev)->netdev_class->get_next_hop(
714 host, next_hop, netdev_name)
717 next_hop->s_addr = 0;
723 /* Populates 'sh' with status information.
725 * Populates 'sh' with 'netdev' specific status information. This information
726 * may be used to populate the status column of the Interface table as defined
727 * in ovs-vswitchd.conf.db(5). */
729 netdev_get_status(const struct netdev *netdev, struct shash *sh)
731 struct netdev_dev *dev = netdev_get_dev(netdev);
733 return (dev->netdev_class->get_status
734 ? dev->netdev_class->get_status(netdev, sh)
738 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
739 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
740 * all-zero-bits (in6addr_any).
742 * The following error values have well-defined meanings:
744 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
746 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
748 * 'in6' may be null, in which case the address itself is not reported. */
750 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
752 struct in6_addr dummy;
755 error = (netdev_get_dev(netdev)->netdev_class->get_in6
756 ? netdev_get_dev(netdev)->netdev_class->get_in6(netdev,
760 memset(in6, 0, sizeof *in6);
765 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
766 * 'on'. If 'permanent' is true, the changes will persist; otherwise, they
767 * will be reverted when 'netdev' is closed or the program exits. Returns 0 if
768 * successful, otherwise a positive errno value. */
770 do_update_flags(struct netdev *netdev, enum netdev_flags off,
771 enum netdev_flags on, enum netdev_flags *old_flagsp,
774 enum netdev_flags old_flags;
777 error = netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
778 off & ~on, on, &old_flags);
780 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
781 off || on ? "set" : "get", netdev_get_name(netdev),
784 } else if ((off || on) && !permanent) {
785 enum netdev_flags new_flags = (old_flags & ~off) | on;
786 enum netdev_flags changed_flags = old_flags ^ new_flags;
788 if (!netdev->changed_flags) {
789 netdev->save_flags = old_flags;
791 netdev->changed_flags |= changed_flags;
795 *old_flagsp = old_flags;
800 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
801 * Returns 0 if successful, otherwise a positive errno value. On failure,
802 * stores 0 into '*flagsp'. */
804 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
806 struct netdev *netdev = (struct netdev *) netdev_;
807 return do_update_flags(netdev, 0, 0, flagsp, false);
810 /* Sets the flags for 'netdev' to 'flags'.
811 * If 'permanent' is true, the changes will persist; otherwise, they
812 * will be reverted when 'netdev' is closed or the program exits.
813 * Returns 0 if successful, otherwise a positive errno value. */
815 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
818 return do_update_flags(netdev, -1, flags, NULL, permanent);
821 /* Turns on the specified 'flags' on 'netdev'.
822 * If 'permanent' is true, the changes will persist; otherwise, they
823 * will be reverted when 'netdev' is closed or the program exits.
824 * Returns 0 if successful, otherwise a positive errno value. */
826 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
829 return do_update_flags(netdev, 0, flags, NULL, permanent);
832 /* Turns off the specified 'flags' on 'netdev'.
833 * If 'permanent' is true, the changes will persist; otherwise, they
834 * will be reverted when 'netdev' is closed or the program exits.
835 * Returns 0 if successful, otherwise a positive errno value. */
837 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
840 return do_update_flags(netdev, flags, 0, NULL, permanent);
843 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
844 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
845 * returns 0. Otherwise, it returns a positive errno value; in particular,
846 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
848 netdev_arp_lookup(const struct netdev *netdev,
849 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
851 int error = (netdev_get_dev(netdev)->netdev_class->arp_lookup
852 ? netdev_get_dev(netdev)->netdev_class->arp_lookup(netdev,
856 memset(mac, 0, ETH_ADDR_LEN);
861 /* Returns true if carrier is active (link light is on) on 'netdev'. */
863 netdev_get_carrier(const struct netdev *netdev)
866 enum netdev_flags flags;
869 netdev_get_flags(netdev, &flags);
870 if (!(flags & NETDEV_UP)) {
874 if (!netdev_get_dev(netdev)->netdev_class->get_carrier) {
878 error = netdev_get_dev(netdev)->netdev_class->get_carrier(netdev,
881 VLOG_DBG("%s: failed to get network device carrier status, assuming "
882 "down: %s", netdev_get_name(netdev), strerror(error));
889 /* Returns true if 'netdev' is up according to its MII. */
891 netdev_get_miimon(const struct netdev *netdev)
894 enum netdev_flags flags;
897 netdev_get_flags(netdev, &flags);
898 if (!(flags & NETDEV_UP)) {
902 if (!netdev_get_dev(netdev)->netdev_class->get_miimon) {
906 error = netdev_get_dev(netdev)->netdev_class->get_miimon(netdev, &miimon);
909 VLOG_DBG("%s: failed to get network device MII status, assuming "
910 "down: %s", netdev_get_name(netdev), strerror(error));
917 /* Retrieves current device stats for 'netdev'. */
919 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
923 COVERAGE_INC(netdev_get_stats);
924 error = (netdev_get_dev(netdev)->netdev_class->get_stats
925 ? netdev_get_dev(netdev)->netdev_class->get_stats(netdev, stats)
928 memset(stats, 0xff, sizeof *stats);
933 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
934 * Returns 0 if successful, otherwise a positive errno value.
936 * This will probably fail for most network devices. Some devices might only
937 * allow setting their stats to 0. */
939 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
941 return (netdev_get_dev(netdev)->netdev_class->set_stats
942 ? netdev_get_dev(netdev)->netdev_class->set_stats(netdev, stats)
946 /* Attempts to set input rate limiting (policing) policy, such that up to
947 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
948 * size of 'kbits' kb. */
950 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
951 uint32_t kbits_burst)
953 return (netdev_get_dev(netdev)->netdev_class->set_policing
954 ? netdev_get_dev(netdev)->netdev_class->set_policing(netdev,
955 kbits_rate, kbits_burst)
959 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
960 * empty if 'netdev' does not support QoS. Any names added to 'types' should
961 * be documented as valid for the "type" column in the "QoS" table in
962 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
964 * Every network device supports disabling QoS with a type of "", but this type
965 * will not be added to 'types'.
967 * The caller must initialize 'types' (e.g. with sset_init()) before calling
968 * this function. The caller is responsible for destroying 'types' (e.g. with
969 * sset_destroy()) when it is no longer needed.
971 * Returns 0 if successful, otherwise a positive errno value. */
973 netdev_get_qos_types(const struct netdev *netdev, struct sset *types)
975 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
976 return (class->get_qos_types
977 ? class->get_qos_types(netdev, types)
981 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
982 * which should be "" or one of the types returned by netdev_get_qos_types()
983 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
984 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
985 * 'caps' to all zeros. */
987 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
988 struct netdev_qos_capabilities *caps)
990 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
993 int retval = (class->get_qos_capabilities
994 ? class->get_qos_capabilities(netdev, type, caps)
997 memset(caps, 0, sizeof *caps);
1001 /* Every netdev supports turning off QoS. */
1002 memset(caps, 0, sizeof *caps);
1007 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1008 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1009 * the number of queues (zero on failure) in '*n_queuesp'.
1011 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1013 netdev_get_n_queues(const struct netdev *netdev,
1014 const char *type, unsigned int *n_queuesp)
1016 struct netdev_qos_capabilities caps;
1019 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1020 *n_queuesp = caps.n_queues;
1024 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1025 * stores the name of the current form of QoS into '*typep', stores any details
1026 * of configuration as string key-value pairs in 'details', and returns 0. On
1027 * failure, sets '*typep' to NULL and returns a positive errno value.
1029 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1031 * The caller must initialize 'details' as an empty shash (e.g. with
1032 * shash_init()) before calling this function. The caller must free 'details',
1033 * including 'data' members, when it is no longer needed (e.g. with
1034 * shash_destroy_free_data()).
1036 * The caller must not modify or free '*typep'.
1038 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1039 * 'netdev'. The contents of 'details' should be documented as valid for
1040 * '*typep' in the "other_config" column in the "QoS" table in
1041 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1043 netdev_get_qos(const struct netdev *netdev,
1044 const char **typep, struct shash *details)
1046 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1049 if (class->get_qos) {
1050 retval = class->get_qos(netdev, typep, details);
1053 shash_clear_free_data(details);
1057 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1063 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1064 * with details of configuration from 'details'. Returns 0 if successful,
1065 * otherwise a positive errno value. On error, the previous QoS configuration
1068 * When this function changes the type of QoS (not just 'details'), this also
1069 * resets all queue configuration for 'netdev' to their defaults (which depend
1070 * on the specific type of QoS). Otherwise, the queue configuration for
1071 * 'netdev' is unchanged.
1073 * 'type' should be "" (to disable QoS) or one of the types returned by
1074 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1075 * documented as valid for the given 'type' in the "other_config" column in the
1076 * "QoS" table in vswitchd/vswitch.xml (which is built as
1077 * ovs-vswitchd.conf.db(8)).
1079 * NULL may be specified for 'details' if there are no configuration
1082 netdev_set_qos(struct netdev *netdev,
1083 const char *type, const struct shash *details)
1085 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1091 if (class->set_qos) {
1093 static struct shash empty = SHASH_INITIALIZER(&empty);
1096 return class->set_qos(netdev, type, details);
1098 return *type ? EOPNOTSUPP : 0;
1102 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1103 * successful, adds that information as string key-value pairs to 'details'.
1104 * Returns 0 if successful, otherwise a positive errno value.
1106 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1107 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1109 * The returned contents of 'details' should be documented as valid for the
1110 * given 'type' in the "other_config" column in the "Queue" table in
1111 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1113 * The caller must initialize 'details' (e.g. with shash_init()) before calling
1114 * this function. The caller must free 'details', including 'data' members,
1115 * when it is no longer needed (e.g. with shash_destroy_free_data()). */
1117 netdev_get_queue(const struct netdev *netdev,
1118 unsigned int queue_id, struct shash *details)
1120 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1123 retval = (class->get_queue
1124 ? class->get_queue(netdev, queue_id, details)
1127 shash_clear_free_data(details);
1132 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1133 * string pairs in 'details'. The contents of 'details' should be documented
1134 * as valid for the given 'type' in the "other_config" column in the "Queue"
1135 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1136 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1137 * given queue's configuration should be unmodified.
1139 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1140 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1142 * This function does not modify 'details', and the caller retains ownership of
1145 netdev_set_queue(struct netdev *netdev,
1146 unsigned int queue_id, const struct shash *details)
1148 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1149 return (class->set_queue
1150 ? class->set_queue(netdev, queue_id, details)
1154 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1155 * of QoS may have a fixed set of queues, in which case attempts to delete them
1156 * will fail with EOPNOTSUPP.
1158 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1159 * given queue will be unmodified.
1161 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1162 * the current form of QoS (e.g. as returned by
1163 * netdev_get_n_queues(netdev)). */
1165 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1167 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1168 return (class->delete_queue
1169 ? class->delete_queue(netdev, queue_id)
1173 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1174 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1175 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1176 * positive errno value and fills 'stats' with all-1-bits. */
1178 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1179 struct netdev_queue_stats *stats)
1181 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1184 retval = (class->get_queue_stats
1185 ? class->get_queue_stats(netdev, queue_id, stats)
1188 memset(stats, 0xff, sizeof *stats);
1193 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1194 * its configuration, and the 'aux' specified by the caller. The order of
1195 * iteration is unspecified, but (when successful) each queue is visited
1198 * Calling this function may be more efficient than calling netdev_get_queue()
1201 * 'cb' must not modify or free the 'details' argument passed in.
1203 * Returns 0 if successful, otherwise a positive errno value. On error, some
1204 * configured queues may not have been included in the iteration. */
1206 netdev_dump_queues(const struct netdev *netdev,
1207 netdev_dump_queues_cb *cb, void *aux)
1209 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1210 return (class->dump_queues
1211 ? class->dump_queues(netdev, cb, aux)
1215 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1216 * its statistics, and the 'aux' specified by the caller. The order of
1217 * iteration is unspecified, but (when successful) each queue is visited
1220 * Calling this function may be more efficient than calling
1221 * netdev_get_queue_stats() for every queue.
1223 * 'cb' must not modify or free the statistics passed in.
1225 * Returns 0 if successful, otherwise a positive errno value. On error, some
1226 * configured queues may not have been included in the iteration. */
1228 netdev_dump_queue_stats(const struct netdev *netdev,
1229 netdev_dump_queue_stats_cb *cb, void *aux)
1231 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1232 return (class->dump_queue_stats
1233 ? class->dump_queue_stats(netdev, cb, aux)
1237 /* If 'netdev' is a VLAN network device (e.g. one created with vconfig(8)),
1238 * sets '*vlan_vid' to the VLAN VID associated with that device and returns 0.
1239 * Otherwise returns a errno value (specifically ENOENT if 'netdev_name' is the
1240 * name of a network device that is not a VLAN device) and sets '*vlan_vid' to
1243 netdev_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1245 int error = (netdev_get_dev(netdev)->netdev_class->get_vlan_vid
1246 ? netdev_get_dev(netdev)->netdev_class->get_vlan_vid(netdev,
1255 /* Returns a network device that has 'in4' as its IP address, if one exists,
1256 * otherwise a null pointer. */
1258 netdev_find_dev_by_in4(const struct in_addr *in4)
1260 struct netdev *netdev;
1261 struct sset dev_list = SSET_INITIALIZER(&dev_list);
1264 netdev_enumerate(&dev_list);
1265 SSET_FOR_EACH (name, &dev_list) {
1266 struct in_addr dev_in4;
1268 if (!netdev_open_default(name, &netdev)
1269 && !netdev_get_in4(netdev, &dev_in4, NULL)
1270 && dev_in4.s_addr == in4->s_addr) {
1273 netdev_close(netdev);
1278 sset_destroy(&dev_list);
1282 /* Initializes 'netdev_dev' as a netdev device named 'name' of the specified
1283 * 'netdev_class'. This function is ordinarily called from a netdev provider's
1284 * 'create' function.
1286 * 'args' should be the arguments that were passed to the netdev provider's
1287 * 'create'. If an empty set of arguments was passed, and 'name' is the name
1288 * of a network device that existed before the 'create' call, then 'args' may
1289 * instead be the configuration for that existing device.
1291 * This function adds 'netdev_dev' to a netdev-owned shash, so it is
1292 * very important that 'netdev_dev' only be freed after calling
1293 * the refcount drops to zero. */
1295 netdev_dev_init(struct netdev_dev *netdev_dev, const char *name,
1296 const struct shash *args,
1297 const struct netdev_class *netdev_class)
1299 assert(!shash_find(&netdev_dev_shash, name));
1301 memset(netdev_dev, 0, sizeof *netdev_dev);
1302 netdev_dev->netdev_class = netdev_class;
1303 netdev_dev->name = xstrdup(name);
1304 netdev_dev->node = shash_add(&netdev_dev_shash, name, netdev_dev);
1305 smap_clone(&netdev_dev->args, args);
1308 /* Undoes the results of initialization.
1310 * Normally this function does not need to be called as netdev_close has
1311 * the same effect when the refcount drops to zero.
1312 * However, it may be called by providers due to an error on creation
1313 * that occurs after initialization. It this case netdev_close() would
1314 * never be called. */
1316 netdev_dev_uninit(struct netdev_dev *netdev_dev, bool destroy)
1318 char *name = netdev_dev->name;
1320 assert(!netdev_dev->ref_cnt);
1322 shash_delete(&netdev_dev_shash, netdev_dev->node);
1323 smap_destroy(&netdev_dev->args);
1326 netdev_dev->netdev_class->destroy(netdev_dev);
1331 /* Returns the class type of 'netdev_dev'.
1333 * The caller must not free the returned value. */
1335 netdev_dev_get_type(const struct netdev_dev *netdev_dev)
1337 return netdev_dev->netdev_class->type;
1340 /* Returns the class associated with 'netdev_dev'. */
1341 const struct netdev_class *
1342 netdev_dev_get_class(const struct netdev_dev *netdev_dev)
1344 return netdev_dev->netdev_class;
1347 /* Returns the name of 'netdev_dev'.
1349 * The caller must not free the returned value. */
1351 netdev_dev_get_name(const struct netdev_dev *netdev_dev)
1353 return netdev_dev->name;
1356 /* Returns the netdev_dev with 'name' or NULL if there is none.
1358 * The caller must not free the returned value. */
1360 netdev_dev_from_name(const char *name)
1362 return shash_find_data(&netdev_dev_shash, name);
1365 /* Fills 'device_list' with devices that match 'netdev_class'.
1367 * The caller is responsible for initializing and destroying 'device_list'
1368 * but the contained netdev_devs must not be freed. */
1370 netdev_dev_get_devices(const struct netdev_class *netdev_class,
1371 struct shash *device_list)
1373 struct shash_node *node;
1374 SHASH_FOR_EACH (node, &netdev_dev_shash) {
1375 struct netdev_dev *dev = node->data;
1377 if (dev->netdev_class == netdev_class) {
1378 shash_add(device_list, node->name, node->data);
1383 /* Initializes 'netdev' as a instance of the netdev_dev.
1385 * This function adds 'netdev' to a netdev-owned linked list, so it is very
1386 * important that 'netdev' only be freed after calling netdev_close(). */
1388 netdev_init(struct netdev *netdev, struct netdev_dev *netdev_dev)
1390 memset(netdev, 0, sizeof *netdev);
1391 netdev->netdev_dev = netdev_dev;
1392 list_push_back(&netdev_list, &netdev->node);
1395 /* Undoes the results of initialization.
1397 * Normally this function only needs to be called from netdev_close().
1398 * However, it may be called by providers due to an error on opening
1399 * that occurs after initialization. It this case netdev_close() would
1400 * never be called. */
1402 netdev_uninit(struct netdev *netdev, bool close)
1404 /* Restore flags that we changed, if any. */
1405 int error = restore_flags(netdev);
1406 list_remove(&netdev->node);
1408 VLOG_WARN("failed to restore network device flags on %s: %s",
1409 netdev_get_name(netdev), strerror(error));
1413 netdev_get_dev(netdev)->netdev_class->close(netdev);
1418 /* Returns the class type of 'netdev'.
1420 * The caller must not free the returned value. */
1422 netdev_get_type(const struct netdev *netdev)
1424 return netdev_get_dev(netdev)->netdev_class->type;
1428 netdev_get_dev(const struct netdev *netdev)
1430 return netdev->netdev_dev;
1433 /* Initializes 'notifier' as a netdev notifier for 'netdev', for which
1434 * notification will consist of calling 'cb', with auxiliary data 'aux'. */
1436 netdev_notifier_init(struct netdev_notifier *notifier, struct netdev *netdev,
1437 void (*cb)(struct netdev_notifier *), void *aux)
1439 notifier->netdev = netdev;
1441 notifier->aux = aux;
1444 /* Tracks changes in the status of a set of network devices. */
1445 struct netdev_monitor {
1446 struct shash polled_netdevs;
1447 struct sset changed_netdevs;
1450 /* Creates and returns a new structure for monitor changes in the status of
1451 * network devices. */
1452 struct netdev_monitor *
1453 netdev_monitor_create(void)
1455 struct netdev_monitor *monitor = xmalloc(sizeof *monitor);
1456 shash_init(&monitor->polled_netdevs);
1457 sset_init(&monitor->changed_netdevs);
1461 /* Destroys 'monitor'. */
1463 netdev_monitor_destroy(struct netdev_monitor *monitor)
1466 struct shash_node *node;
1468 SHASH_FOR_EACH (node, &monitor->polled_netdevs) {
1469 struct netdev_notifier *notifier = node->data;
1470 netdev_get_dev(notifier->netdev)->netdev_class->poll_remove(
1474 shash_destroy(&monitor->polled_netdevs);
1475 sset_destroy(&monitor->changed_netdevs);
1481 netdev_monitor_cb(struct netdev_notifier *notifier)
1483 struct netdev_monitor *monitor = notifier->aux;
1484 const char *name = netdev_get_name(notifier->netdev);
1485 sset_add(&monitor->changed_netdevs, name);
1488 /* Attempts to add 'netdev' as a netdev monitored by 'monitor'. Returns 0 if
1489 * successful, otherwise a positive errno value.
1491 * Adding a given 'netdev' to a monitor multiple times is equivalent to adding
1494 netdev_monitor_add(struct netdev_monitor *monitor, struct netdev *netdev)
1496 const char *netdev_name = netdev_get_name(netdev);
1498 if (!shash_find(&monitor->polled_netdevs, netdev_name)
1499 && netdev_get_dev(netdev)->netdev_class->poll_add)
1501 struct netdev_notifier *notifier;
1502 error = netdev_get_dev(netdev)->netdev_class->poll_add(netdev,
1503 netdev_monitor_cb, monitor, ¬ifier);
1505 assert(notifier->netdev == netdev);
1506 shash_add(&monitor->polled_netdevs, netdev_name, notifier);
1512 /* Removes 'netdev' from the set of netdevs monitored by 'monitor'. (This has
1513 * no effect if 'netdev' is not in the set of devices monitored by
1516 netdev_monitor_remove(struct netdev_monitor *monitor, struct netdev *netdev)
1518 const char *netdev_name = netdev_get_name(netdev);
1519 struct shash_node *node;
1521 node = shash_find(&monitor->polled_netdevs, netdev_name);
1523 /* Cancel future notifications. */
1524 struct netdev_notifier *notifier = node->data;
1525 netdev_get_dev(netdev)->netdev_class->poll_remove(notifier);
1526 shash_delete(&monitor->polled_netdevs, node);
1528 /* Drop any pending notification. */
1529 sset_find_and_delete(&monitor->changed_netdevs, netdev_name);
1533 /* Checks for changes to netdevs in the set monitored by 'monitor'. If any of
1534 * the attributes (Ethernet address, carrier status, speed or peer-advertised
1535 * speed, flags, etc.) of a network device monitored by 'monitor' has changed,
1536 * sets '*devnamep' to the name of a device that has changed and returns 0.
1537 * The caller is responsible for freeing '*devnamep' (with free()).
1539 * If no devices have changed, sets '*devnamep' to NULL and returns EAGAIN. */
1541 netdev_monitor_poll(struct netdev_monitor *monitor, char **devnamep)
1543 if (sset_is_empty(&monitor->changed_netdevs)) {
1547 *devnamep = sset_pop(&monitor->changed_netdevs);
1552 /* Registers with the poll loop to wake up from the next call to poll_block()
1553 * when netdev_monitor_poll(monitor) would indicate that a device has
1556 netdev_monitor_poll_wait(const struct netdev_monitor *monitor)
1558 if (!sset_is_empty(&monitor->changed_netdevs)) {
1559 poll_immediate_wake();
1561 /* XXX Nothing needed here for netdev_linux, but maybe other netdev
1562 * classes need help. */
1566 /* Restore the network device flags on 'netdev' to those that were active
1567 * before we changed them. Returns 0 if successful, otherwise a positive
1570 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
1572 restore_flags(struct netdev *netdev)
1574 if (netdev->changed_flags) {
1575 enum netdev_flags restore = netdev->save_flags & netdev->changed_flags;
1576 enum netdev_flags old_flags;
1577 return netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
1578 netdev->changed_flags & ~restore,
1579 restore, &old_flags);
1584 /* Close all netdevs on shutdown so they can do any needed cleanup such as
1585 * destroying devices, restoring flags, etc. */
1587 close_all_netdevs(void *aux OVS_UNUSED)
1589 struct netdev *netdev, *next;
1590 LIST_FOR_EACH_SAFE(netdev, next, node, &netdev_list) {
1591 netdev_close(netdev);