2 * Copyright (c) 2008, 2009, 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
23 #include <netinet/in.h>
29 #include "dynamic-string.h"
30 #include "fatal-signal.h"
33 #include "netdev-provider.h"
34 #include "netdev-vport.h"
36 #include "openflow/openflow.h"
38 #include "poll-loop.h"
43 VLOG_DEFINE_THIS_MODULE(netdev);
45 static struct shash netdev_classes = SHASH_INITIALIZER(&netdev_classes);
47 /* All created network devices. */
48 static struct shash netdev_dev_shash = SHASH_INITIALIZER(&netdev_dev_shash);
50 /* All open network devices. */
51 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
53 /* This is set pretty low because we probably won't learn anything from the
54 * additional log messages. */
55 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
57 static void close_all_netdevs(void *aux OVS_UNUSED);
58 static int restore_flags(struct netdev *netdev);
59 void update_device_args(struct netdev_dev *, const struct shash *args);
62 netdev_initialize(void)
69 fatal_signal_add_hook(close_all_netdevs, NULL, NULL, true);
72 netdev_register_provider(&netdev_linux_class);
73 netdev_register_provider(&netdev_tap_class);
74 netdev_vport_register();
79 /* Performs periodic work needed by all the various kinds of netdevs.
81 * If your program opens any netdevs, it must call this function within its
86 struct shash_node *node;
87 SHASH_FOR_EACH(node, &netdev_classes) {
88 const struct netdev_class *netdev_class = node->data;
89 if (netdev_class->run) {
95 /* Arranges for poll_block() to wake up when netdev_run() needs to be called.
97 * If your program opens any netdevs, it must call this function within its
102 struct shash_node *node;
103 SHASH_FOR_EACH(node, &netdev_classes) {
104 const struct netdev_class *netdev_class = node->data;
105 if (netdev_class->wait) {
106 netdev_class->wait();
111 /* Initializes and registers a new netdev provider. After successful
112 * registration, new netdevs of that type can be opened using netdev_open(). */
114 netdev_register_provider(const struct netdev_class *new_class)
116 if (shash_find(&netdev_classes, new_class->type)) {
117 VLOG_WARN("attempted to register duplicate netdev provider: %s",
122 if (new_class->init) {
123 int error = new_class->init();
125 VLOG_ERR("failed to initialize %s network device class: %s",
126 new_class->type, strerror(error));
131 shash_add(&netdev_classes, new_class->type, new_class);
136 /* Unregisters a netdev provider. 'type' must have been previously
137 * registered and not currently be in use by any netdevs. After unregistration
138 * new netdevs of that type cannot be opened using netdev_open(). */
140 netdev_unregister_provider(const char *type)
142 struct shash_node *del_node, *netdev_dev_node;
144 del_node = shash_find(&netdev_classes, type);
146 VLOG_WARN("attempted to unregister a netdev provider that is not "
147 "registered: %s", type);
151 SHASH_FOR_EACH(netdev_dev_node, &netdev_dev_shash) {
152 struct netdev_dev *netdev_dev = netdev_dev_node->data;
153 if (!strcmp(netdev_dev->netdev_class->type, type)) {
154 VLOG_WARN("attempted to unregister in use netdev provider: %s",
160 shash_delete(&netdev_classes, del_node);
165 /* Clears 'types' and enumerates the types of all currently registered netdev
166 * providers into it. The caller must first initialize the svec. */
168 netdev_enumerate_types(struct svec *types)
170 struct shash_node *node;
175 SHASH_FOR_EACH(node, &netdev_classes) {
176 const struct netdev_class *netdev_class = node->data;
177 svec_add(types, netdev_class->type);
181 /* Compares 'args' to those used to those used by 'dev'. Returns true
182 * if the arguments are the same, false otherwise. Does not update the
183 * values stored in 'dev'. */
185 compare_device_args(const struct netdev_dev *dev, const struct shash *args)
187 const struct shash_node **new_args;
191 if (shash_count(args) != dev->n_args) {
195 new_args = shash_sort(args);
196 for (i = 0; i < dev->n_args; i++) {
197 if (strcmp(dev->args[i].key, new_args[i]->name) ||
198 strcmp(dev->args[i].value, new_args[i]->data)) {
210 compare_args(const void *a_, const void *b_)
212 const struct arg *a = a_;
213 const struct arg *b = b_;
214 return strcmp(a->key, b->key);
218 update_device_args(struct netdev_dev *dev, const struct shash *args)
220 struct shash_node *node;
224 for (i = 0; i < dev->n_args; i++) {
225 free(dev->args[i].key);
226 free(dev->args[i].value);
233 if (!args || shash_is_empty(args)) {
237 dev->n_args = shash_count(args);
238 dev->args = xmalloc(dev->n_args * sizeof *dev->args);
241 SHASH_FOR_EACH(node, args) {
242 dev->args[i].key = xstrdup(node->name);
243 dev->args[i].value = xstrdup(node->data);
247 qsort(dev->args, dev->n_args, sizeof *dev->args, compare_args);
251 create_device(struct netdev_options *options, struct netdev_dev **netdev_devp)
253 struct netdev_class *netdev_class;
256 if (!options->type || strlen(options->type) == 0) {
257 /* Default to system. */
258 options->type = "system";
261 netdev_class = shash_find_data(&netdev_classes, options->type);
266 error = netdev_class->create(netdev_class, options->name, options->args,
268 assert(error || (*netdev_devp)->netdev_class == netdev_class);
272 /* Opens the network device named 'name' (e.g. "eth0") and returns zero if
273 * successful, otherwise a positive errno value. On success, sets '*netdevp'
274 * to the new network device, otherwise to null.
276 * If this is the first time the device has been opened, then create is called
277 * before opening. The device is created using the given type and arguments.
279 * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
280 * capture frames of that type received on the device. It may also be one of
281 * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those
284 netdev_open(struct netdev_options *options, struct netdev **netdevp)
286 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
287 struct netdev_dev *netdev_dev;
293 if (!options->args) {
294 options->args = &empty_args;
297 netdev_dev = shash_find_data(&netdev_dev_shash, options->name);
300 error = create_device(options, &netdev_dev);
302 if (error == EAFNOSUPPORT) {
303 VLOG_WARN("could not create netdev %s of unknown type %s",
304 options->name, options->type);
308 update_device_args(netdev_dev, options->args);
310 } else if (!shash_is_empty(options->args) &&
311 !compare_device_args(netdev_dev, options->args)) {
313 VLOG_WARN("%s: attempted to open already open netdev with "
314 "different arguments", options->name);
318 error = netdev_dev->netdev_class->open(netdev_dev, options->ethertype,
322 netdev_dev->ref_cnt++;
324 if (!netdev_dev->ref_cnt) {
325 netdev_dev_uninit(netdev_dev, true);
333 netdev_open_default(const char *name, struct netdev **netdevp)
335 struct netdev_options options;
337 memset(&options, 0, sizeof options);
339 options.ethertype = NETDEV_ETH_TYPE_NONE;
341 return netdev_open(&options, netdevp);
344 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
345 * or NULL if none are needed. */
347 netdev_reconfigure(struct netdev *netdev, const struct shash *args)
349 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
350 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
356 if (netdev_dev->netdev_class->reconfigure) {
357 if (!compare_device_args(netdev_dev, args)) {
358 update_device_args(netdev_dev, args);
359 return netdev_dev->netdev_class->reconfigure(netdev_dev, args);
361 } else if (!shash_is_empty(args)) {
362 VLOG_WARN("%s: arguments provided to device that does not have a "
363 "reconfigure function", netdev_get_name(netdev));
369 /* Closes and destroys 'netdev'. */
371 netdev_close(struct netdev *netdev)
374 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
376 assert(netdev_dev->ref_cnt);
377 netdev_dev->ref_cnt--;
378 netdev_uninit(netdev, true);
380 /* If the reference count for the netdev device is zero, destroy it. */
381 if (!netdev_dev->ref_cnt) {
382 netdev_dev_uninit(netdev_dev, true);
387 /* Returns true if a network device named 'name' exists and may be opened,
388 * otherwise false. */
390 netdev_exists(const char *name)
392 struct netdev *netdev;
395 error = netdev_open_default(name, &netdev);
397 netdev_close(netdev);
400 if (error != ENODEV) {
401 VLOG_WARN("failed to open network device %s: %s",
402 name, strerror(error));
408 /* Returns true if a network device named 'name' is currently opened,
409 * otherwise false. */
411 netdev_is_open(const char *name)
413 return !!shash_find_data(&netdev_dev_shash, name);
416 /* Clears 'svec' and enumerates the names of all known network devices. */
418 netdev_enumerate(struct svec *svec)
420 struct shash_node *node;
426 SHASH_FOR_EACH(node, &netdev_classes) {
427 const struct netdev_class *netdev_class = node->data;
428 if (netdev_class->enumerate) {
429 int retval = netdev_class->enumerate(svec);
431 VLOG_WARN("failed to enumerate %s network devices: %s",
432 netdev_class->type, strerror(retval));
443 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
444 * must have initialized with sufficient room for the packet. The space
445 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
446 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
447 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
448 * need not be included.)
450 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
451 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
452 * positive errno value. Returns EAGAIN immediately if no packet is ready to
455 * Some network devices may not implement support for this function. In such
456 * cases this function will always return EOPNOTSUPP. */
458 netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
460 int (*recv)(struct netdev *, void *, size_t);
463 assert(buffer->size == 0);
464 assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
466 recv = netdev_get_dev(netdev)->netdev_class->recv;
468 ? (recv)(netdev, buffer->data, ofpbuf_tailroom(buffer))
471 COVERAGE_INC(netdev_received);
472 buffer->size += retval;
473 if (buffer->size < ETH_TOTAL_MIN) {
474 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
482 /* Registers with the poll loop to wake up from the next call to poll_block()
483 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
485 netdev_recv_wait(struct netdev *netdev)
487 void (*recv_wait)(struct netdev *);
489 recv_wait = netdev_get_dev(netdev)->netdev_class->recv_wait;
495 /* Discards all packets waiting to be received from 'netdev'. */
497 netdev_drain(struct netdev *netdev)
499 int (*drain)(struct netdev *);
501 drain = netdev_get_dev(netdev)->netdev_class->drain;
502 return drain ? drain(netdev) : 0;
505 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
506 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
507 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
508 * the packet is too big or too small to transmit on the device.
510 * The caller retains ownership of 'buffer' in all cases.
512 * The kernel maintains a packet transmission queue, so the caller is not
513 * expected to do additional queuing of packets.
515 * Some network devices may not implement support for this function. In such
516 * cases this function will always return EOPNOTSUPP. */
518 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
520 int (*send)(struct netdev *, const void *, size_t);
523 send = netdev_get_dev(netdev)->netdev_class->send;
524 error = send ? (send)(netdev, buffer->data, buffer->size) : EOPNOTSUPP;
526 COVERAGE_INC(netdev_sent);
531 /* Registers with the poll loop to wake up from the next call to poll_block()
532 * when the packet transmission queue has sufficient room to transmit a packet
533 * with netdev_send().
535 * The kernel maintains a packet transmission queue, so the client is not
536 * expected to do additional queuing of packets. Thus, this function is
537 * unlikely to ever be used. It is included for completeness. */
539 netdev_send_wait(struct netdev *netdev)
541 void (*send_wait)(struct netdev *);
543 send_wait = netdev_get_dev(netdev)->netdev_class->send_wait;
549 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
550 * otherwise a positive errno value. */
552 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
554 return netdev_get_dev(netdev)->netdev_class->set_etheraddr(netdev, mac);
557 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
558 * the MAC address into 'mac'. On failure, returns a positive errno value and
559 * clears 'mac' to all-zeros. */
561 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
563 return netdev_get_dev(netdev)->netdev_class->get_etheraddr(netdev, mac);
566 /* Returns the name of the network device that 'netdev' represents,
567 * e.g. "eth0". The caller must not modify or free the returned string. */
569 netdev_get_name(const struct netdev *netdev)
571 return netdev_get_dev(netdev)->name;
574 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
575 * (and received) packets, in bytes, not including the hardware header; thus,
576 * this is typically 1500 bytes for Ethernet devices.
578 * If successful, returns 0 and stores the MTU size in '*mtup'. On failure,
579 * returns a positive errno value and stores ETH_PAYLOAD_MAX (1500) in
582 netdev_get_mtu(const struct netdev *netdev, int *mtup)
584 int error = netdev_get_dev(netdev)->netdev_class->get_mtu(netdev, mtup);
586 VLOG_WARN_RL(&rl, "failed to retrieve MTU for network device %s: %s",
587 netdev_get_name(netdev), strerror(error));
588 *mtup = ETH_PAYLOAD_MAX;
593 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
594 * failure, returns a negative errno value.
596 * The desired semantics of the ifindex value are a combination of those
597 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
598 * value should be unique within a host and remain stable at least until
599 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
600 * but many systems do not follow this rule anyhow.
602 * Some network devices may not implement support for this function. In such
603 * cases this function will always return -EOPNOTSUPP.
606 netdev_get_ifindex(const struct netdev *netdev)
608 int (*get_ifindex)(const struct netdev *);
610 get_ifindex = netdev_get_dev(netdev)->netdev_class->get_ifindex;
612 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
615 /* Stores the features supported by 'netdev' into each of '*current',
616 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
617 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
618 * successful, otherwise a positive errno value. On failure, all of the
619 * passed-in values are set to 0.
621 * Some network devices may not implement support for this function. In such
622 * cases this function will always return EOPNOTSUPP. */
624 netdev_get_features(struct netdev *netdev,
625 uint32_t *current, uint32_t *advertised,
626 uint32_t *supported, uint32_t *peer)
628 int (*get_features)(struct netdev *netdev,
629 uint32_t *current, uint32_t *advertised,
630 uint32_t *supported, uint32_t *peer);
638 advertised = &dummy[1];
641 supported = &dummy[2];
647 get_features = netdev_get_dev(netdev)->netdev_class->get_features;
649 ? get_features(netdev, current, advertised, supported, peer)
652 *current = *advertised = *supported = *peer = 0;
657 /* Returns the maximum speed of a network connection that has the "enum
658 * ofp_port_features" bits in 'features', in bits per second. If no bits that
659 * indicate a speed are set in 'features', assumes 100Mbps. */
661 netdev_features_to_bps(uint32_t features)
664 F_10000MB = OFPPF_10GB_FD,
665 F_1000MB = OFPPF_1GB_HD | OFPPF_1GB_FD,
666 F_100MB = OFPPF_100MB_HD | OFPPF_100MB_FD,
667 F_10MB = OFPPF_10MB_HD | OFPPF_10MB_FD
670 return ( features & F_10000MB ? UINT64_C(10000000000)
671 : features & F_1000MB ? UINT64_C(1000000000)
672 : features & F_100MB ? UINT64_C(100000000)
673 : features & F_10MB ? UINT64_C(10000000)
674 : UINT64_C(100000000));
677 /* Returns true if any of the "enum ofp_port_features" bits that indicate a
678 * full-duplex link are set in 'features', otherwise false. */
680 netdev_features_is_full_duplex(uint32_t features)
682 return (features & (OFPPF_10MB_FD | OFPPF_100MB_FD | OFPPF_1GB_FD
683 | OFPPF_10GB_FD)) != 0;
686 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
687 * successful, otherwise a positive errno value. */
689 netdev_set_advertisements(struct netdev *netdev, uint32_t advertise)
691 return (netdev_get_dev(netdev)->netdev_class->set_advertisements
692 ? netdev_get_dev(netdev)->netdev_class->set_advertisements(
697 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
698 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
699 * errno value and sets '*address' to 0 (INADDR_ANY).
701 * The following error values have well-defined meanings:
703 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
705 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
707 * 'address' or 'netmask' or both may be null, in which case the address or
708 * netmask is not reported. */
710 netdev_get_in4(const struct netdev *netdev,
711 struct in_addr *address_, struct in_addr *netmask_)
713 struct in_addr address;
714 struct in_addr netmask;
717 error = (netdev_get_dev(netdev)->netdev_class->get_in4
718 ? netdev_get_dev(netdev)->netdev_class->get_in4(netdev,
722 address_->s_addr = error ? 0 : address.s_addr;
725 netmask_->s_addr = error ? 0 : netmask.s_addr;
730 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
731 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
732 * positive errno value. */
734 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
736 return (netdev_get_dev(netdev)->netdev_class->set_in4
737 ? netdev_get_dev(netdev)->netdev_class->set_in4(netdev, addr, mask)
741 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
744 netdev_add_router(struct netdev *netdev, struct in_addr router)
746 COVERAGE_INC(netdev_add_router);
747 return (netdev_get_dev(netdev)->netdev_class->add_router
748 ? netdev_get_dev(netdev)->netdev_class->add_router(netdev, router)
752 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
753 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
754 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
755 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
756 * a directly connected network) in '*next_hop' and a copy of the name of the
757 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
758 * responsible for freeing '*netdev_name' (by calling free()). */
760 netdev_get_next_hop(const struct netdev *netdev,
761 const struct in_addr *host, struct in_addr *next_hop,
764 int error = (netdev_get_dev(netdev)->netdev_class->get_next_hop
765 ? netdev_get_dev(netdev)->netdev_class->get_next_hop(
766 host, next_hop, netdev_name)
769 next_hop->s_addr = 0;
775 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
776 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
777 * all-zero-bits (in6addr_any).
779 * The following error values have well-defined meanings:
781 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
783 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
785 * 'in6' may be null, in which case the address itself is not reported. */
787 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
789 struct in6_addr dummy;
792 error = (netdev_get_dev(netdev)->netdev_class->get_in6
793 ? netdev_get_dev(netdev)->netdev_class->get_in6(netdev,
797 memset(in6, 0, sizeof *in6);
802 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
803 * 'on'. If 'permanent' is true, the changes will persist; otherwise, they
804 * will be reverted when 'netdev' is closed or the program exits. Returns 0 if
805 * successful, otherwise a positive errno value. */
807 do_update_flags(struct netdev *netdev, enum netdev_flags off,
808 enum netdev_flags on, enum netdev_flags *old_flagsp,
811 enum netdev_flags old_flags;
814 error = netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
815 off & ~on, on, &old_flags);
817 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
818 off || on ? "set" : "get", netdev_get_name(netdev),
821 } else if ((off || on) && !permanent) {
822 enum netdev_flags new_flags = (old_flags & ~off) | on;
823 enum netdev_flags changed_flags = old_flags ^ new_flags;
825 if (!netdev->changed_flags) {
826 netdev->save_flags = old_flags;
828 netdev->changed_flags |= changed_flags;
832 *old_flagsp = old_flags;
837 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
838 * Returns 0 if successful, otherwise a positive errno value. On failure,
839 * stores 0 into '*flagsp'. */
841 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
843 struct netdev *netdev = (struct netdev *) netdev_;
844 return do_update_flags(netdev, 0, 0, flagsp, false);
847 /* Sets the flags for 'netdev' to 'flags'.
848 * If 'permanent' is true, the changes will persist; otherwise, they
849 * will be reverted when 'netdev' is closed or the program exits.
850 * Returns 0 if successful, otherwise a positive errno value. */
852 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
855 return do_update_flags(netdev, -1, flags, NULL, permanent);
858 /* Turns on the specified 'flags' on 'netdev'.
859 * If 'permanent' is true, the changes will persist; otherwise, they
860 * will be reverted when 'netdev' is closed or the program exits.
861 * Returns 0 if successful, otherwise a positive errno value. */
863 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
866 return do_update_flags(netdev, 0, flags, NULL, permanent);
869 /* Turns off the specified 'flags' on 'netdev'.
870 * If 'permanent' is true, the changes will persist; otherwise, they
871 * will be reverted when 'netdev' is closed or the program exits.
872 * Returns 0 if successful, otherwise a positive errno value. */
874 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
877 return do_update_flags(netdev, flags, 0, NULL, permanent);
880 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
881 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
882 * returns 0. Otherwise, it returns a positive errno value; in particular,
883 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
885 netdev_arp_lookup(const struct netdev *netdev,
886 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
888 int error = (netdev_get_dev(netdev)->netdev_class->arp_lookup
889 ? netdev_get_dev(netdev)->netdev_class->arp_lookup(netdev,
893 memset(mac, 0, ETH_ADDR_LEN);
898 /* Returns true if carrier is active (link light is on) on 'netdev'. */
900 netdev_get_carrier(const struct netdev *netdev)
903 enum netdev_flags flags;
906 netdev_get_flags(netdev, &flags);
907 if (!(flags & NETDEV_UP)) {
911 if (!netdev_get_dev(netdev)->netdev_class->get_carrier) {
915 error = netdev_get_dev(netdev)->netdev_class->get_carrier(netdev,
918 VLOG_DBG("%s: failed to get network device carrier status, assuming "
919 "down: %s", netdev_get_name(netdev), strerror(error));
926 /* Retrieves current device stats for 'netdev'. */
928 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
932 COVERAGE_INC(netdev_get_stats);
933 error = (netdev_get_dev(netdev)->netdev_class->get_stats
934 ? netdev_get_dev(netdev)->netdev_class->get_stats(netdev, stats)
937 memset(stats, 0xff, sizeof *stats);
942 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
943 * Returns 0 if successful, otherwise a positive errno value.
945 * This will probably fail for most network devices. Some devices might only
946 * allow setting their stats to 0. */
948 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
950 return (netdev_get_dev(netdev)->netdev_class->set_stats
951 ? netdev_get_dev(netdev)->netdev_class->set_stats(netdev, stats)
955 /* Attempts to set input rate limiting (policing) policy, such that up to
956 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
957 * size of 'kbits' kb. */
959 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
960 uint32_t kbits_burst)
962 return (netdev_get_dev(netdev)->netdev_class->set_policing
963 ? netdev_get_dev(netdev)->netdev_class->set_policing(netdev,
964 kbits_rate, kbits_burst)
968 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
969 * empty if 'netdev' does not support QoS. Any names added to 'types' should
970 * be documented as valid for the "type" column in the "QoS" table in
971 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
973 * Every network device supports disabling QoS with a type of "", but this type
974 * will not be added to 'types'.
976 * The caller must initialize 'types' (e.g. with svec_init()) before calling
977 * this function. The caller is responsible for destroying 'types' (e.g. with
978 * svec_destroy()) when it is no longer needed.
980 * Returns 0 if successful, otherwise a positive errno value. */
982 netdev_get_qos_types(const struct netdev *netdev, struct svec *types)
984 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
985 return (class->get_qos_types
986 ? class->get_qos_types(netdev, types)
990 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
991 * which should be "" or one of the types returned by netdev_get_qos_types()
992 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
993 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
994 * 'caps' to all zeros. */
996 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
997 struct netdev_qos_capabilities *caps)
999 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1002 int retval = (class->get_qos_capabilities
1003 ? class->get_qos_capabilities(netdev, type, caps)
1006 memset(caps, 0, sizeof *caps);
1010 /* Every netdev supports turning off QoS. */
1011 memset(caps, 0, sizeof *caps);
1016 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1017 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1018 * the number of queues (zero on failure) in '*n_queuesp'.
1020 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1022 netdev_get_n_queues(const struct netdev *netdev,
1023 const char *type, unsigned int *n_queuesp)
1025 struct netdev_qos_capabilities caps;
1028 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1029 *n_queuesp = caps.n_queues;
1033 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1034 * stores the name of the current form of QoS into '*typep', stores any details
1035 * of configuration as string key-value pairs in 'details', and returns 0. On
1036 * failure, sets '*typep' to NULL and returns a positive errno value.
1038 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1040 * The caller must initialize 'details' as an empty shash (e.g. with
1041 * shash_init()) before calling this function. The caller must free 'details',
1042 * including 'data' members, when it is no longer needed (e.g. with
1043 * shash_destroy_free_data()).
1045 * The caller must not modify or free '*typep'.
1047 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1048 * 'netdev'. The contents of 'details' should be documented as valid for
1049 * '*typep' in the "other_config" column in the "QoS" table in
1050 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1052 netdev_get_qos(const struct netdev *netdev,
1053 const char **typep, struct shash *details)
1055 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1058 if (class->get_qos) {
1059 retval = class->get_qos(netdev, typep, details);
1062 shash_clear_free_data(details);
1066 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1072 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1073 * with details of configuration from 'details'. Returns 0 if successful,
1074 * otherwise a positive errno value. On error, the previous QoS configuration
1077 * When this function changes the type of QoS (not just 'details'), this also
1078 * resets all queue configuration for 'netdev' to their defaults (which depend
1079 * on the specific type of QoS). Otherwise, the queue configuration for
1080 * 'netdev' is unchanged.
1082 * 'type' should be "" (to disable QoS) or one of the types returned by
1083 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1084 * documented as valid for the given 'type' in the "other_config" column in the
1085 * "QoS" table in vswitchd/vswitch.xml (which is built as
1086 * ovs-vswitchd.conf.db(8)).
1088 * NULL may be specified for 'details' if there are no configuration
1091 netdev_set_qos(struct netdev *netdev,
1092 const char *type, const struct shash *details)
1094 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1100 if (class->set_qos) {
1102 static struct shash empty = SHASH_INITIALIZER(&empty);
1105 return class->set_qos(netdev, type, details);
1107 return *type ? EOPNOTSUPP : 0;
1111 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1112 * successful, adds that information as string key-value pairs to 'details'.
1113 * Returns 0 if successful, otherwise a positive errno value.
1115 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1116 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1118 * The returned contents of 'details' should be documented as valid for the
1119 * given 'type' in the "other_config" column in the "Queue" table in
1120 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1122 * The caller must initialize 'details' (e.g. with shash_init()) before calling
1123 * this function. The caller must free 'details', including 'data' members,
1124 * when it is no longer needed (e.g. with shash_destroy_free_data()). */
1126 netdev_get_queue(const struct netdev *netdev,
1127 unsigned int queue_id, struct shash *details)
1129 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1132 retval = (class->get_queue
1133 ? class->get_queue(netdev, queue_id, details)
1136 shash_clear_free_data(details);
1141 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1142 * string pairs in 'details'. The contents of 'details' should be documented
1143 * as valid for the given 'type' in the "other_config" column in the "Queue"
1144 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1145 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1146 * given queue's configuration should be unmodified.
1148 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1149 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1151 * This function does not modify 'details', and the caller retains ownership of
1154 netdev_set_queue(struct netdev *netdev,
1155 unsigned int queue_id, const struct shash *details)
1157 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1158 return (class->set_queue
1159 ? class->set_queue(netdev, queue_id, details)
1163 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1164 * of QoS may have a fixed set of queues, in which case attempts to delete them
1165 * will fail with EOPNOTSUPP.
1167 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1168 * given queue will be unmodified.
1170 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1171 * the current form of QoS (e.g. as returned by
1172 * netdev_get_n_queues(netdev)). */
1174 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1176 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1177 return (class->delete_queue
1178 ? class->delete_queue(netdev, queue_id)
1182 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1183 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1184 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1185 * positive errno value and fills 'stats' with all-1-bits. */
1187 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1188 struct netdev_queue_stats *stats)
1190 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1193 retval = (class->get_queue_stats
1194 ? class->get_queue_stats(netdev, queue_id, stats)
1197 memset(stats, 0xff, sizeof *stats);
1202 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1203 * its configuration, and the 'aux' specified by the caller. The order of
1204 * iteration is unspecified, but (when successful) each queue is visited
1207 * Calling this function may be more efficient than calling netdev_get_queue()
1210 * 'cb' must not modify or free the 'details' argument passed in.
1212 * Returns 0 if successful, otherwise a positive errno value. On error, some
1213 * configured queues may not have been included in the iteration. */
1215 netdev_dump_queues(const struct netdev *netdev,
1216 netdev_dump_queues_cb *cb, void *aux)
1218 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1219 return (class->dump_queues
1220 ? class->dump_queues(netdev, cb, aux)
1224 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1225 * its statistics, and the 'aux' specified by the caller. The order of
1226 * iteration is unspecified, but (when successful) each queue is visited
1229 * Calling this function may be more efficient than calling
1230 * netdev_get_queue_stats() for every queue.
1232 * 'cb' must not modify or free the statistics passed in.
1234 * Returns 0 if successful, otherwise a positive errno value. On error, some
1235 * configured queues may not have been included in the iteration. */
1237 netdev_dump_queue_stats(const struct netdev *netdev,
1238 netdev_dump_queue_stats_cb *cb, void *aux)
1240 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1241 return (class->dump_queue_stats
1242 ? class->dump_queue_stats(netdev, cb, aux)
1246 /* If 'netdev' is a VLAN network device (e.g. one created with vconfig(8)),
1247 * sets '*vlan_vid' to the VLAN VID associated with that device and returns 0.
1248 * Otherwise returns a errno value (specifically ENOENT if 'netdev_name' is the
1249 * name of a network device that is not a VLAN device) and sets '*vlan_vid' to
1252 netdev_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1254 int error = (netdev_get_dev(netdev)->netdev_class->get_vlan_vid
1255 ? netdev_get_dev(netdev)->netdev_class->get_vlan_vid(netdev,
1264 /* Returns a network device that has 'in4' as its IP address, if one exists,
1265 * otherwise a null pointer. */
1267 netdev_find_dev_by_in4(const struct in_addr *in4)
1269 struct netdev *netdev;
1270 struct svec dev_list = SVEC_EMPTY_INITIALIZER;
1273 netdev_enumerate(&dev_list);
1274 for (i = 0; i < dev_list.n; i++) {
1275 const char *name = dev_list.names[i];
1276 struct in_addr dev_in4;
1278 if (!netdev_open_default(name, &netdev)
1279 && !netdev_get_in4(netdev, &dev_in4, NULL)
1280 && dev_in4.s_addr == in4->s_addr) {
1283 netdev_close(netdev);
1288 svec_destroy(&dev_list);
1292 /* Initializes 'netdev_dev' as a netdev device named 'name' of the
1293 * specified 'netdev_class'.
1295 * This function adds 'netdev_dev' to a netdev-owned shash, so it is
1296 * very important that 'netdev_dev' only be freed after calling
1297 * the refcount drops to zero. */
1299 netdev_dev_init(struct netdev_dev *netdev_dev, const char *name,
1300 const struct netdev_class *netdev_class)
1302 assert(!shash_find(&netdev_dev_shash, name));
1304 memset(netdev_dev, 0, sizeof *netdev_dev);
1305 netdev_dev->netdev_class = netdev_class;
1306 netdev_dev->name = xstrdup(name);
1307 netdev_dev->node = shash_add(&netdev_dev_shash, name, netdev_dev);
1310 /* Undoes the results of initialization.
1312 * Normally this function does not need to be called as netdev_close has
1313 * the same effect when the refcount drops to zero.
1314 * However, it may be called by providers due to an error on creation
1315 * that occurs after initialization. It this case netdev_close() would
1316 * never be called. */
1318 netdev_dev_uninit(struct netdev_dev *netdev_dev, bool destroy)
1320 char *name = netdev_dev->name;
1322 assert(!netdev_dev->ref_cnt);
1324 shash_delete(&netdev_dev_shash, netdev_dev->node);
1325 update_device_args(netdev_dev, NULL);
1328 netdev_dev->netdev_class->destroy(netdev_dev);
1333 /* Returns the class type of 'netdev_dev'.
1335 * The caller must not free the returned value. */
1337 netdev_dev_get_type(const struct netdev_dev *netdev_dev)
1339 return netdev_dev->netdev_class->type;
1342 /* Returns the class associated with 'netdev_dev'. */
1343 const struct netdev_class *
1344 netdev_dev_get_class(const struct netdev_dev *netdev_dev)
1346 return netdev_dev->netdev_class;
1349 /* Returns the name of 'netdev_dev'.
1351 * The caller must not free the returned value. */
1353 netdev_dev_get_name(const struct netdev_dev *netdev_dev)
1355 return netdev_dev->name;
1358 /* Returns the netdev_dev with 'name' or NULL if there is none.
1360 * The caller must not free the returned value. */
1362 netdev_dev_from_name(const char *name)
1364 return shash_find_data(&netdev_dev_shash, name);
1367 /* Fills 'device_list' with devices that match 'netdev_class'.
1369 * The caller is responsible for initializing and destroying 'device_list'
1370 * but the contained netdev_devs must not be freed. */
1372 netdev_dev_get_devices(const struct netdev_class *netdev_class,
1373 struct shash *device_list)
1375 struct shash_node *node;
1376 SHASH_FOR_EACH (node, &netdev_dev_shash) {
1377 struct netdev_dev *dev = node->data;
1379 if (dev->netdev_class == netdev_class) {
1380 shash_add(device_list, node->name, node->data);
1385 /* Initializes 'netdev' as a instance of the netdev_dev.
1387 * This function adds 'netdev' to a netdev-owned linked list, so it is very
1388 * important that 'netdev' only be freed after calling netdev_close(). */
1390 netdev_init(struct netdev *netdev, struct netdev_dev *netdev_dev)
1392 memset(netdev, 0, sizeof *netdev);
1393 netdev->netdev_dev = netdev_dev;
1394 list_push_back(&netdev_list, &netdev->node);
1397 /* Undoes the results of initialization.
1399 * Normally this function only needs to be called from netdev_close().
1400 * However, it may be called by providers due to an error on opening
1401 * that occurs after initialization. It this case netdev_close() would
1402 * never be called. */
1404 netdev_uninit(struct netdev *netdev, bool close)
1406 /* Restore flags that we changed, if any. */
1407 int error = restore_flags(netdev);
1408 list_remove(&netdev->node);
1410 VLOG_WARN("failed to restore network device flags on %s: %s",
1411 netdev_get_name(netdev), strerror(error));
1415 netdev_get_dev(netdev)->netdev_class->close(netdev);
1420 /* Returns the class type of 'netdev'.
1422 * The caller must not free the returned value. */
1424 netdev_get_type(const struct netdev *netdev)
1426 return netdev_get_dev(netdev)->netdev_class->type;
1430 netdev_get_dev(const struct netdev *netdev)
1432 return netdev->netdev_dev;
1435 /* Initializes 'notifier' as a netdev notifier for 'netdev', for which
1436 * notification will consist of calling 'cb', with auxiliary data 'aux'. */
1438 netdev_notifier_init(struct netdev_notifier *notifier, struct netdev *netdev,
1439 void (*cb)(struct netdev_notifier *), void *aux)
1441 notifier->netdev = netdev;
1443 notifier->aux = aux;
1446 /* Tracks changes in the status of a set of network devices. */
1447 struct netdev_monitor {
1448 struct shash polled_netdevs;
1449 struct shash changed_netdevs;
1452 /* Creates and returns a new structure for monitor changes in the status of
1453 * network devices. */
1454 struct netdev_monitor *
1455 netdev_monitor_create(void)
1457 struct netdev_monitor *monitor = xmalloc(sizeof *monitor);
1458 shash_init(&monitor->polled_netdevs);
1459 shash_init(&monitor->changed_netdevs);
1463 /* Destroys 'monitor'. */
1465 netdev_monitor_destroy(struct netdev_monitor *monitor)
1468 struct shash_node *node;
1470 SHASH_FOR_EACH (node, &monitor->polled_netdevs) {
1471 struct netdev_notifier *notifier = node->data;
1472 netdev_get_dev(notifier->netdev)->netdev_class->poll_remove(
1476 shash_destroy(&monitor->polled_netdevs);
1477 shash_destroy(&monitor->changed_netdevs);
1483 netdev_monitor_cb(struct netdev_notifier *notifier)
1485 struct netdev_monitor *monitor = notifier->aux;
1486 const char *name = netdev_get_name(notifier->netdev);
1487 shash_add_once(&monitor->changed_netdevs, name, NULL);
1490 /* Attempts to add 'netdev' as a netdev monitored by 'monitor'. Returns 0 if
1491 * successful, otherwise a positive errno value.
1493 * Adding a given 'netdev' to a monitor multiple times is equivalent to adding
1496 netdev_monitor_add(struct netdev_monitor *monitor, struct netdev *netdev)
1498 const char *netdev_name = netdev_get_name(netdev);
1500 if (!shash_find(&monitor->polled_netdevs, netdev_name)
1501 && netdev_get_dev(netdev)->netdev_class->poll_add)
1503 struct netdev_notifier *notifier;
1504 error = netdev_get_dev(netdev)->netdev_class->poll_add(netdev,
1505 netdev_monitor_cb, monitor, ¬ifier);
1507 assert(notifier->netdev == netdev);
1508 shash_add(&monitor->polled_netdevs, netdev_name, notifier);
1514 /* Removes 'netdev' from the set of netdevs monitored by 'monitor'. (This has
1515 * no effect if 'netdev' is not in the set of devices monitored by
1518 netdev_monitor_remove(struct netdev_monitor *monitor, struct netdev *netdev)
1520 const char *netdev_name = netdev_get_name(netdev);
1521 struct shash_node *node;
1523 node = shash_find(&monitor->polled_netdevs, netdev_name);
1525 /* Cancel future notifications. */
1526 struct netdev_notifier *notifier = node->data;
1527 netdev_get_dev(netdev)->netdev_class->poll_remove(notifier);
1528 shash_delete(&monitor->polled_netdevs, node);
1530 /* Drop any pending notification. */
1531 node = shash_find(&monitor->changed_netdevs, netdev_name);
1533 shash_delete(&monitor->changed_netdevs, node);
1538 /* Checks for changes to netdevs in the set monitored by 'monitor'. If any of
1539 * the attributes (Ethernet address, carrier status, speed or peer-advertised
1540 * speed, flags, etc.) of a network device monitored by 'monitor' has changed,
1541 * sets '*devnamep' to the name of a device that has changed and returns 0.
1542 * The caller is responsible for freeing '*devnamep' (with free()).
1544 * If no devices have changed, sets '*devnamep' to NULL and returns EAGAIN. */
1546 netdev_monitor_poll(struct netdev_monitor *monitor, char **devnamep)
1548 struct shash_node *node = shash_first(&monitor->changed_netdevs);
1553 *devnamep = shash_steal(&monitor->changed_netdevs, node);
1558 /* Registers with the poll loop to wake up from the next call to poll_block()
1559 * when netdev_monitor_poll(monitor) would indicate that a device has
1562 netdev_monitor_poll_wait(const struct netdev_monitor *monitor)
1564 if (!shash_is_empty(&monitor->changed_netdevs)) {
1565 poll_immediate_wake();
1567 /* XXX Nothing needed here for netdev_linux, but maybe other netdev
1568 * classes need help. */
1572 /* Restore the network device flags on 'netdev' to those that were active
1573 * before we changed them. Returns 0 if successful, otherwise a positive
1576 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
1578 restore_flags(struct netdev *netdev)
1580 if (netdev->changed_flags) {
1581 enum netdev_flags restore = netdev->save_flags & netdev->changed_flags;
1582 enum netdev_flags old_flags;
1583 return netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
1584 netdev->changed_flags & ~restore,
1585 restore, &old_flags);
1590 /* Close all netdevs on shutdown so they can do any needed cleanup such as
1591 * destroying devices, restoring flags, etc. */
1593 close_all_netdevs(void *aux OVS_UNUSED)
1595 struct netdev *netdev, *next;
1596 LIST_FOR_EACH_SAFE(netdev, next, node, &netdev_list) {
1597 netdev_close(netdev);