2 * Copyright (c) 2008, 2009, 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
23 #include <netinet/in.h>
29 #include "dynamic-string.h"
30 #include "fatal-signal.h"
33 #include "netdev-provider.h"
35 #include "openflow/openflow.h"
37 #include "poll-loop.h"
41 #define THIS_MODULE VLM_netdev
44 static const struct netdev_class *base_netdev_classes[] = {
53 static struct shash netdev_classes = SHASH_INITIALIZER(&netdev_classes);
55 /* All created network devices. */
56 static struct shash netdev_dev_shash = SHASH_INITIALIZER(&netdev_dev_shash);
58 /* All open network devices. */
59 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
61 /* This is set pretty low because we probably won't learn anything from the
62 * additional log messages. */
63 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
65 static void close_all_netdevs(void *aux OVS_UNUSED);
66 static int restore_flags(struct netdev *netdev);
67 void update_device_args(struct netdev_dev *, const struct shash *args);
70 netdev_initialize(void)
72 static int status = -1;
77 fatal_signal_add_hook(close_all_netdevs, NULL, NULL, true);
80 for (i = 0; i < ARRAY_SIZE(base_netdev_classes); i++) {
81 netdev_register_provider(base_netdev_classes[i]);
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 struct netdev_class *new_provider;
125 if (shash_find(&netdev_classes, new_class->type)) {
126 VLOG_WARN("attempted to register duplicate netdev provider: %s",
131 if (new_class->init) {
132 int error = new_class->init();
134 VLOG_ERR("failed to initialize %s network device class: %s",
135 new_class->type, strerror(error));
140 new_provider = xmalloc(sizeof *new_provider);
141 memcpy(new_provider, new_class, sizeof *new_provider);
143 shash_add(&netdev_classes, new_class->type, new_provider);
148 /* Unregisters a netdev provider. 'type' must have been previously
149 * registered and not currently be in use by any netdevs. After unregistration
150 * new netdevs of that type cannot be opened using netdev_open(). */
152 netdev_unregister_provider(const char *type)
154 struct shash_node *del_node, *netdev_dev_node;
156 del_node = shash_find(&netdev_classes, type);
158 VLOG_WARN("attempted to unregister a netdev provider that is not "
159 "registered: %s", type);
163 SHASH_FOR_EACH(netdev_dev_node, &netdev_dev_shash) {
164 struct netdev_dev *netdev_dev = netdev_dev_node->data;
165 if (!strcmp(netdev_dev->netdev_class->type, type)) {
166 VLOG_WARN("attempted to unregister in use netdev provider: %s",
172 shash_delete(&netdev_classes, del_node);
173 free(del_node->data);
178 /* Clears 'types' and enumerates the types of all currently registered netdev
179 * providers into it. The caller must first initialize the svec. */
181 netdev_enumerate_types(struct svec *types)
183 struct shash_node *node;
188 SHASH_FOR_EACH(node, &netdev_classes) {
189 const struct netdev_class *netdev_class = node->data;
190 svec_add(types, netdev_class->type);
194 /* Compares 'args' to those used to those used by 'dev'. Returns true
195 * if the arguments are the same, false otherwise. Does not update the
196 * values stored in 'dev'. */
198 compare_device_args(const struct netdev_dev *dev, const struct shash *args)
200 const struct shash_node **new_args;
204 if (shash_count(args) != dev->n_args) {
208 new_args = shash_sort(args);
209 for (i = 0; i < dev->n_args; i++) {
210 if (strcmp(dev->args[i].key, new_args[i]->name) ||
211 strcmp(dev->args[i].value, new_args[i]->data)) {
223 compare_args(const void *a_, const void *b_)
225 const struct arg *a = a_;
226 const struct arg *b = b_;
227 return strcmp(a->key, b->key);
231 update_device_args(struct netdev_dev *dev, const struct shash *args)
233 struct shash_node *node;
237 for (i = 0; i < dev->n_args; i++) {
238 free(dev->args[i].key);
239 free(dev->args[i].value);
246 if (!args || shash_is_empty(args)) {
250 dev->n_args = shash_count(args);
251 dev->args = xmalloc(dev->n_args * sizeof *dev->args);
254 SHASH_FOR_EACH(node, args) {
255 dev->args[i].key = xstrdup(node->name);
256 dev->args[i].value = xstrdup(node->data);
260 qsort(dev->args, dev->n_args, sizeof *dev->args, compare_args);
264 create_device(struct netdev_options *options, struct netdev_dev **netdev_devp)
266 struct netdev_class *netdev_class;
268 if (!options->type || strlen(options->type) == 0) {
269 /* Default to system. */
270 options->type = "system";
273 netdev_class = shash_find_data(&netdev_classes, options->type);
275 VLOG_WARN("could not create netdev %s of unknown type %s",
276 options->name, options->type);
280 return netdev_class->create(options->name, options->type, options->args,
284 /* Opens the network device named 'name' (e.g. "eth0") and returns zero if
285 * successful, otherwise a positive errno value. On success, sets '*netdevp'
286 * to the new network device, otherwise to null.
288 * If this is the first time the device has been opened, then create is called
289 * before opening. The device is created using the given type and arguments.
291 * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
292 * capture frames of that type received on the device. It may also be one of
293 * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those
297 netdev_open(struct netdev_options *options, struct netdev **netdevp)
299 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
300 struct netdev_dev *netdev_dev;
306 if (!options->args) {
307 options->args = &empty_args;
310 netdev_dev = shash_find_data(&netdev_dev_shash, options->name);
313 error = create_device(options, &netdev_dev);
317 update_device_args(netdev_dev, options->args);
319 } else if (!shash_is_empty(options->args) &&
320 !compare_device_args(netdev_dev, options->args)) {
322 VLOG_WARN("%s: attempted to open already open netdev with "
323 "different arguments", options->name);
327 error = netdev_dev->netdev_class->open(netdev_dev, options->ethertype,
331 netdev_dev->ref_cnt++;
333 if (!netdev_dev->ref_cnt) {
334 netdev_dev_uninit(netdev_dev, true);
342 netdev_open_default(const char *name, struct netdev **netdevp)
344 struct netdev_options options;
346 memset(&options, 0, sizeof options);
348 options.ethertype = NETDEV_ETH_TYPE_NONE;
350 return netdev_open(&options, netdevp);
353 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
354 * or NULL if none are needed. */
356 netdev_reconfigure(struct netdev *netdev, const struct shash *args)
358 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
359 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
365 if (netdev_dev->netdev_class->reconfigure) {
366 if (!compare_device_args(netdev_dev, args)) {
367 update_device_args(netdev_dev, args);
368 return netdev_dev->netdev_class->reconfigure(netdev_dev, args);
370 } else if (!shash_is_empty(args)) {
371 VLOG_WARN("%s: arguments provided to device that does not have a "
372 "reconfigure function", netdev_get_name(netdev));
378 /* Closes and destroys 'netdev'. */
380 netdev_close(struct netdev *netdev)
383 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
385 assert(netdev_dev->ref_cnt);
386 netdev_dev->ref_cnt--;
387 netdev_uninit(netdev, true);
389 /* If the reference count for the netdev device is zero, destroy it. */
390 if (!netdev_dev->ref_cnt) {
391 netdev_dev_uninit(netdev_dev, true);
396 /* Returns true if a network device named 'name' exists and may be opened,
397 * otherwise false. */
399 netdev_exists(const char *name)
401 struct netdev *netdev;
404 error = netdev_open_default(name, &netdev);
406 netdev_close(netdev);
409 if (error != ENODEV) {
410 VLOG_WARN("failed to open network device %s: %s",
411 name, strerror(error));
417 /* Returns true if a network device named 'name' is currently opened,
418 * otherwise false. */
420 netdev_is_open(const char *name)
422 return !!shash_find_data(&netdev_dev_shash, name);
425 /* Clears 'svec' and enumerates the names of all known network devices. */
427 netdev_enumerate(struct svec *svec)
429 struct shash_node *node;
435 SHASH_FOR_EACH(node, &netdev_classes) {
436 const struct netdev_class *netdev_class = node->data;
437 if (netdev_class->enumerate) {
438 int retval = netdev_class->enumerate(svec);
440 VLOG_WARN("failed to enumerate %s network devices: %s",
441 netdev_class->type, strerror(retval));
452 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
453 * must have initialized with sufficient room for the packet. The space
454 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
455 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
456 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
457 * need not be included.)
459 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
460 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
461 * positive errno value. Returns EAGAIN immediately if no packet is ready to
464 * Some network devices may not implement support for this function. In such
465 * cases this function will always return EOPNOTSUPP.
468 netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
470 int (*recv)(struct netdev *, void *, size_t);
473 assert(buffer->size == 0);
474 assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
476 recv = netdev_get_dev(netdev)->netdev_class->recv;
478 ? (recv)(netdev, buffer->data, ofpbuf_tailroom(buffer))
481 COVERAGE_INC(netdev_received);
482 buffer->size += retval;
483 if (buffer->size < ETH_TOTAL_MIN) {
484 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
492 /* Registers with the poll loop to wake up from the next call to poll_block()
493 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
495 netdev_recv_wait(struct netdev *netdev)
497 void (*recv_wait)(struct netdev *);
499 recv_wait = netdev_get_dev(netdev)->netdev_class->recv_wait;
505 /* Discards all packets waiting to be received from 'netdev'. */
507 netdev_drain(struct netdev *netdev)
509 int (*drain)(struct netdev *);
511 drain = netdev_get_dev(netdev)->netdev_class->drain;
512 return drain ? drain(netdev) : 0;
515 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
516 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
517 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
518 * the packet is too big or too small to transmit on the device.
520 * The caller retains ownership of 'buffer' in all cases.
522 * The kernel maintains a packet transmission queue, so the caller is not
523 * expected to do additional queuing of packets.
525 * Some network devices may not implement support for this function. In such
526 * cases this function will always return EOPNOTSUPP. */
528 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
530 int (*send)(struct netdev *, const void *, size_t);
533 send = netdev_get_dev(netdev)->netdev_class->send;
534 error = send ? (send)(netdev, buffer->data, buffer->size) : EOPNOTSUPP;
536 COVERAGE_INC(netdev_sent);
541 /* Registers with the poll loop to wake up from the next call to poll_block()
542 * when the packet transmission queue has sufficient room to transmit a packet
543 * with netdev_send().
545 * The kernel maintains a packet transmission queue, so the client is not
546 * expected to do additional queuing of packets. Thus, this function is
547 * unlikely to ever be used. It is included for completeness. */
549 netdev_send_wait(struct netdev *netdev)
551 void (*send_wait)(struct netdev *);
553 send_wait = netdev_get_dev(netdev)->netdev_class->send_wait;
559 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
560 * otherwise a positive errno value. */
562 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
564 return netdev_get_dev(netdev)->netdev_class->set_etheraddr(netdev, mac);
567 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
568 * the MAC address into 'mac'. On failure, returns a positive errno value and
569 * clears 'mac' to all-zeros. */
571 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
573 return netdev_get_dev(netdev)->netdev_class->get_etheraddr(netdev, mac);
576 /* Returns the name of the network device that 'netdev' represents,
577 * e.g. "eth0". The caller must not modify or free the returned string. */
579 netdev_get_name(const struct netdev *netdev)
581 return netdev_get_dev(netdev)->name;
584 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
585 * (and received) packets, in bytes, not including the hardware header; thus,
586 * this is typically 1500 bytes for Ethernet devices.
588 * If successful, returns 0 and stores the MTU size in '*mtup'. On failure,
589 * returns a positive errno value and stores ETH_PAYLOAD_MAX (1500) in
592 netdev_get_mtu(const struct netdev *netdev, int *mtup)
594 int error = netdev_get_dev(netdev)->netdev_class->get_mtu(netdev, mtup);
596 VLOG_WARN_RL(&rl, "failed to retrieve MTU for network device %s: %s",
597 netdev_get_name(netdev), strerror(error));
598 *mtup = ETH_PAYLOAD_MAX;
603 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
604 * failure, returns a negative errno value.
606 * The desired semantics of the ifindex value are a combination of those
607 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
608 * value should be unique within a host and remain stable at least until
609 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
610 * but many systems do not follow this rule anyhow.
612 * Some network devices may not implement support for this function. In such
613 * cases this function will always return -EOPNOTSUPP.
616 netdev_get_ifindex(const struct netdev *netdev)
618 int (*get_ifindex)(const struct netdev *);
620 get_ifindex = netdev_get_dev(netdev)->netdev_class->get_ifindex;
622 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
625 /* Stores the features supported by 'netdev' into each of '*current',
626 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
627 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
628 * successful, otherwise a positive errno value. On failure, all of the
629 * passed-in values are set to 0.
631 * Some network devices may not implement support for this function. In such
632 * cases this function will always return EOPNOTSUPP.
635 netdev_get_features(struct netdev *netdev,
636 uint32_t *current, uint32_t *advertised,
637 uint32_t *supported, uint32_t *peer)
639 int (*get_features)(struct netdev *netdev,
640 uint32_t *current, uint32_t *advertised,
641 uint32_t *supported, uint32_t *peer);
649 advertised = &dummy[1];
652 supported = &dummy[2];
658 get_features = netdev_get_dev(netdev)->netdev_class->get_features;
660 ? get_features(netdev, current, advertised, supported, peer)
663 *current = *advertised = *supported = *peer = 0;
668 /* Returns the maximum speed of a network connection that has the "enum
669 * ofp_port_features" bits in 'features', in bits per second. If no bits that
670 * indicate a speed are set in 'features', assumes 100Mbps. */
672 netdev_features_to_bps(uint32_t features)
675 F_10000MB = OFPPF_10GB_FD,
676 F_1000MB = OFPPF_1GB_HD | OFPPF_1GB_FD,
677 F_100MB = OFPPF_100MB_HD | OFPPF_100MB_FD,
678 F_10MB = OFPPF_10MB_HD | OFPPF_10MB_FD
681 return ( features & F_10000MB ? UINT64_C(10000000000)
682 : features & F_1000MB ? UINT64_C(1000000000)
683 : features & F_100MB ? UINT64_C(100000000)
684 : features & F_10MB ? UINT64_C(10000000)
685 : UINT64_C(100000000));
688 /* Returns true if any of the "enum ofp_port_features" bits that indicate a
689 * full-duplex link are set in 'features', otherwise false. */
691 netdev_features_is_full_duplex(uint32_t features)
693 return (features & (OFPPF_10MB_FD | OFPPF_100MB_FD | OFPPF_1GB_FD
694 | OFPPF_10GB_FD)) != 0;
697 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
698 * successful, otherwise a positive errno value. */
700 netdev_set_advertisements(struct netdev *netdev, uint32_t advertise)
702 return (netdev_get_dev(netdev)->netdev_class->set_advertisements
703 ? netdev_get_dev(netdev)->netdev_class->set_advertisements(
708 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
709 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
710 * errno value and sets '*address' to 0 (INADDR_ANY).
712 * The following error values have well-defined meanings:
714 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
716 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
718 * 'address' or 'netmask' or both may be null, in which case the address or netmask
719 * is not reported. */
721 netdev_get_in4(const struct netdev *netdev,
722 struct in_addr *address_, struct in_addr *netmask_)
724 struct in_addr address;
725 struct in_addr netmask;
728 error = (netdev_get_dev(netdev)->netdev_class->get_in4
729 ? netdev_get_dev(netdev)->netdev_class->get_in4(netdev,
733 address_->s_addr = error ? 0 : address.s_addr;
736 netmask_->s_addr = error ? 0 : netmask.s_addr;
741 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
742 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
743 * positive errno value. */
745 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
747 return (netdev_get_dev(netdev)->netdev_class->set_in4
748 ? netdev_get_dev(netdev)->netdev_class->set_in4(netdev, addr, mask)
752 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
755 netdev_add_router(struct netdev *netdev, struct in_addr router)
757 COVERAGE_INC(netdev_add_router);
758 return (netdev_get_dev(netdev)->netdev_class->add_router
759 ? netdev_get_dev(netdev)->netdev_class->add_router(netdev, router)
763 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
764 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
765 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
766 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
767 * a directly connected network) in '*next_hop' and a copy of the name of the
768 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
769 * responsible for freeing '*netdev_name' (by calling free()). */
771 netdev_get_next_hop(const struct netdev *netdev,
772 const struct in_addr *host, struct in_addr *next_hop,
775 int error = (netdev_get_dev(netdev)->netdev_class->get_next_hop
776 ? netdev_get_dev(netdev)->netdev_class->get_next_hop(
777 host, next_hop, netdev_name)
780 next_hop->s_addr = 0;
786 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
787 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
788 * all-zero-bits (in6addr_any).
790 * The following error values have well-defined meanings:
792 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
794 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
796 * 'in6' may be null, in which case the address itself is not reported. */
798 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
800 struct in6_addr dummy;
803 error = (netdev_get_dev(netdev)->netdev_class->get_in6
804 ? netdev_get_dev(netdev)->netdev_class->get_in6(netdev,
808 memset(in6, 0, sizeof *in6);
813 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
814 * 'on'. If 'permanent' is true, the changes will persist; otherwise, they
815 * will be reverted when 'netdev' is closed or the program exits. Returns 0 if
816 * successful, otherwise a positive errno value. */
818 do_update_flags(struct netdev *netdev, enum netdev_flags off,
819 enum netdev_flags on, enum netdev_flags *old_flagsp,
822 enum netdev_flags old_flags;
825 error = netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
826 off & ~on, on, &old_flags);
828 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
829 off || on ? "set" : "get", netdev_get_name(netdev),
832 } else if ((off || on) && !permanent) {
833 enum netdev_flags new_flags = (old_flags & ~off) | on;
834 enum netdev_flags changed_flags = old_flags ^ new_flags;
836 if (!netdev->changed_flags) {
837 netdev->save_flags = old_flags;
839 netdev->changed_flags |= changed_flags;
843 *old_flagsp = old_flags;
848 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
849 * Returns 0 if successful, otherwise a positive errno value. On failure,
850 * stores 0 into '*flagsp'. */
852 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
854 struct netdev *netdev = (struct netdev *) netdev_;
855 return do_update_flags(netdev, 0, 0, flagsp, false);
858 /* Sets the flags for 'netdev' to 'flags'.
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_set_flags(struct netdev *netdev, enum netdev_flags flags,
866 return do_update_flags(netdev, -1, flags, NULL, permanent);
869 /* Turns on 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_on(struct netdev *netdev, enum netdev_flags flags,
877 return do_update_flags(netdev, 0, flags, NULL, permanent);
880 /* Turns off the specified 'flags' on 'netdev'.
881 * If 'permanent' is true, the changes will persist; otherwise, they
882 * will be reverted when 'netdev' is closed or the program exits.
883 * Returns 0 if successful, otherwise a positive errno value. */
885 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
888 return do_update_flags(netdev, flags, 0, NULL, permanent);
891 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
892 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
893 * returns 0. Otherwise, it returns a positive errno value; in particular,
894 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
896 netdev_arp_lookup(const struct netdev *netdev,
897 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
899 int error = (netdev_get_dev(netdev)->netdev_class->arp_lookup
900 ? netdev_get_dev(netdev)->netdev_class->arp_lookup(netdev,
904 memset(mac, 0, ETH_ADDR_LEN);
909 /* Sets 'carrier' to true if carrier is active (link light is on) on
912 netdev_get_carrier(const struct netdev *netdev, bool *carrier)
914 int error = (netdev_get_dev(netdev)->netdev_class->get_carrier
915 ? netdev_get_dev(netdev)->netdev_class->get_carrier(netdev,
924 /* Retrieves current device stats for 'netdev'. */
926 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
930 COVERAGE_INC(netdev_get_stats);
931 error = (netdev_get_dev(netdev)->netdev_class->get_stats
932 ? netdev_get_dev(netdev)->netdev_class->get_stats(netdev, stats)
935 memset(stats, 0xff, sizeof *stats);
940 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
941 * Returns 0 if successful, otherwise a positive errno value.
943 * This will probably fail for most network devices. Some devices might only
944 * allow setting their stats to 0. */
946 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
948 return (netdev_get_dev(netdev)->netdev_class->set_stats
949 ? netdev_get_dev(netdev)->netdev_class->set_stats(netdev, stats)
953 /* Attempts to set input rate limiting (policing) policy, such that up to
954 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
955 * size of 'kbits' kb. */
957 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
958 uint32_t kbits_burst)
960 return (netdev_get_dev(netdev)->netdev_class->set_policing
961 ? netdev_get_dev(netdev)->netdev_class->set_policing(netdev,
962 kbits_rate, kbits_burst)
966 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
967 * empty if 'netdev' does not support QoS. Any names added to 'types' should
968 * be documented as valid for the "type" column in the "QoS" table in
969 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
971 * Every network device supports disabling QoS with a type of "", but this type
972 * will not be added to 'types'.
974 * The caller must initialize 'types' (e.g. with svec_init()) before calling
975 * this function. The caller is responsible for destroying 'types' (e.g. with
976 * svec_destroy()) when it is no longer needed.
978 * Returns 0 if successful, otherwise a positive errno value. */
980 netdev_get_qos_types(const struct netdev *netdev, struct svec *types)
982 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
983 return (class->get_qos_types
984 ? class->get_qos_types(netdev, types)
988 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
989 * which should be "" or one of the types returned by netdev_get_qos_types()
990 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
991 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
992 * 'caps' to all zeros. */
994 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
995 struct netdev_qos_capabilities *caps)
997 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1000 int retval = (class->get_qos_capabilities
1001 ? class->get_qos_capabilities(netdev, type, caps)
1004 memset(caps, 0, sizeof *caps);
1008 /* Every netdev supports turning off QoS. */
1009 memset(caps, 0, sizeof *caps);
1014 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1015 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1016 * the number of queues (zero on failure) in '*n_queuesp'.
1018 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1020 netdev_get_n_queues(const struct netdev *netdev,
1021 const char *type, unsigned int *n_queuesp)
1023 struct netdev_qos_capabilities caps;
1026 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1027 *n_queuesp = caps.n_queues;
1031 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1032 * stores the name of the current form of QoS into '*typep', stores any details
1033 * of configuration as string key-value pairs in 'details', and returns 0. On
1034 * failure, sets '*typep' to NULL and returns a positive errno value.
1036 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1038 * The caller must initialize 'details' as an empty shash (e.g. with
1039 * shash_init()) before calling this function. The caller must free 'details',
1040 * including 'data' members, when it is no longer needed (e.g. with
1041 * shash_destroy_free_data()).
1043 * The caller must not modify or free '*typep'.
1045 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1046 * 'netdev'. The contents of 'details' should be documented as valid for
1047 * '*typep' in the "other_config" column in the "QoS" table in
1048 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1050 netdev_get_qos(const struct netdev *netdev,
1051 const char **typep, struct shash *details)
1053 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1056 if (class->get_qos) {
1057 retval = class->get_qos(netdev, typep, details);
1060 shash_clear_free_data(details);
1064 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1070 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1071 * with details of configuration from 'details'. Returns 0 if successful,
1072 * otherwise a positive errno value. On error, the previous QoS configuration
1075 * When this function changes the type of QoS (not just 'details'), this also
1076 * resets all queue configuration for 'netdev' to their defaults (which depend
1077 * on the specific type of QoS). Otherwise, the queue configuration for
1078 * 'netdev' is unchanged.
1080 * 'type' should be "" (to disable QoS) or one of the types returned by
1081 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1082 * documented as valid for the given 'type' in the "other_config" column in the
1083 * "QoS" table in vswitchd/vswitch.xml (which is built as
1084 * ovs-vswitchd.conf.db(8)).
1086 * NULL may be specified for 'details' if there are no configuration
1089 netdev_set_qos(struct netdev *netdev,
1090 const char *type, const struct shash *details)
1092 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1098 if (class->set_qos) {
1100 static struct shash empty = SHASH_INITIALIZER(&empty);
1103 return class->set_qos(netdev, type, details);
1105 return *type ? EOPNOTSUPP : 0;
1109 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1110 * successful, adds that information as string key-value pairs to 'details'.
1111 * Returns 0 if successful, otherwise a positive errno value.
1113 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1114 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1116 * The returned contents of 'details' should be documented as valid for the
1117 * given 'type' in the "other_config" column in the "Queue" table in
1118 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1120 * The caller must initialize 'details' (e.g. with shash_init()) before calling
1121 * this function. The caller must free 'details', including 'data' members,
1122 * when it is no longer needed (e.g. with shash_destroy_free_data()). */
1124 netdev_get_queue(const struct netdev *netdev,
1125 unsigned int queue_id, struct shash *details)
1127 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1130 retval = (class->get_queue
1131 ? class->get_queue(netdev, queue_id, details)
1134 shash_clear_free_data(details);
1139 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1140 * string pairs in 'details'. The contents of 'details' should be documented
1141 * as valid for the given 'type' in the "other_config" column in the "Queue"
1142 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1143 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1144 * given queue's configuration should be unmodified.
1146 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1147 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1149 * This function does not modify 'details', and the caller retains ownership of
1153 netdev_set_queue(struct netdev *netdev,
1154 unsigned int queue_id, const struct shash *details)
1156 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1157 return (class->set_queue
1158 ? class->set_queue(netdev, queue_id, details)
1162 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1163 * of QoS may have a fixed set of queues, in which case attempts to delete them
1164 * will fail with EOPNOTSUPP.
1166 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1167 * given queue will be unmodified.
1169 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1170 * the current form of QoS (e.g. as returned by
1171 * netdev_get_n_queues(netdev)). */
1173 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1175 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1176 return (class->delete_queue
1177 ? class->delete_queue(netdev, queue_id)
1181 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1182 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1183 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1184 * positive errno value and fills 'stats' with all-1-bits. */
1186 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1187 struct netdev_queue_stats *stats)
1189 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1192 retval = (class->get_queue_stats
1193 ? class->get_queue_stats(netdev, queue_id, stats)
1196 memset(stats, 0xff, sizeof *stats);
1201 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1202 * its configuration, and the 'aux' specified by the caller. The order of
1203 * iteration is unspecified, but (when successful) each queue is visited
1206 * Calling this function may be more efficient than calling netdev_get_queue()
1209 * 'cb' must not modify or free the 'details' argument passed in.
1211 * Returns 0 if successful, otherwise a positive errno value. On error, some
1212 * configured queues may not have been included in the iteration. */
1214 netdev_dump_queues(const struct netdev *netdev,
1215 netdev_dump_queues_cb *cb, void *aux)
1217 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1218 return (class->dump_queues
1219 ? class->dump_queues(netdev, cb, aux)
1223 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1224 * its statistics, and the 'aux' specified by the caller. The order of
1225 * iteration is unspecified, but (when successful) each queue is visited
1228 * Calling this function may be more efficient than calling
1229 * netdev_get_queue_stats() for every queue.
1231 * 'cb' must not modify or free the statistics passed in.
1233 * Returns 0 if successful, otherwise a positive errno value. On error, some
1234 * configured queues may not have been included in the iteration. */
1236 netdev_dump_queue_stats(const struct netdev *netdev,
1237 netdev_dump_queue_stats_cb *cb, void *aux)
1239 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1240 return (class->dump_queue_stats
1241 ? class->dump_queue_stats(netdev, cb, aux)
1245 /* If 'netdev' is a VLAN network device (e.g. one created with vconfig(8)),
1246 * sets '*vlan_vid' to the VLAN VID associated with that device and returns 0.
1247 * Otherwise returns a errno value (specifically ENOENT if 'netdev_name' is the
1248 * name of a network device that is not a VLAN device) and sets '*vlan_vid' to
1251 netdev_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1253 int error = (netdev_get_dev(netdev)->netdev_class->get_vlan_vid
1254 ? netdev_get_dev(netdev)->netdev_class->get_vlan_vid(netdev,
1263 /* Returns a network device that has 'in4' as its IP address, if one exists,
1264 * otherwise a null pointer. */
1266 netdev_find_dev_by_in4(const struct in_addr *in4)
1268 struct netdev *netdev;
1269 struct svec dev_list = SVEC_EMPTY_INITIALIZER;
1272 netdev_enumerate(&dev_list);
1273 for (i = 0; i < dev_list.n; i++) {
1274 const char *name = dev_list.names[i];
1275 struct in_addr dev_in4;
1277 if (!netdev_open_default(name, &netdev)
1278 && !netdev_get_in4(netdev, &dev_in4, NULL)
1279 && dev_in4.s_addr == in4->s_addr) {
1282 netdev_close(netdev);
1287 svec_destroy(&dev_list);
1291 /* Initializes 'netdev_dev' as a netdev device named 'name' of the
1292 * specified 'netdev_class'.
1294 * This function adds 'netdev_dev' to a netdev-owned shash, so it is
1295 * very important that 'netdev_dev' only be freed after calling
1296 * the refcount drops to zero. */
1298 netdev_dev_init(struct netdev_dev *netdev_dev, const char *name,
1299 const struct netdev_class *netdev_class)
1301 assert(!shash_find(&netdev_dev_shash, name));
1303 memset(netdev_dev, 0, sizeof *netdev_dev);
1304 netdev_dev->netdev_class = netdev_class;
1305 netdev_dev->name = xstrdup(name);
1306 netdev_dev->node = shash_add(&netdev_dev_shash, name, netdev_dev);
1309 /* Undoes the results of initialization.
1311 * Normally this function does not need to be called as netdev_close has
1312 * the same effect when the refcount drops to zero.
1313 * However, it may be called by providers due to an error on creation
1314 * that occurs after initialization. It this case netdev_close() would
1315 * never be called. */
1317 netdev_dev_uninit(struct netdev_dev *netdev_dev, bool destroy)
1319 char *name = netdev_dev->name;
1321 assert(!netdev_dev->ref_cnt);
1323 shash_delete(&netdev_dev_shash, netdev_dev->node);
1324 update_device_args(netdev_dev, NULL);
1327 netdev_dev->netdev_class->destroy(netdev_dev);
1332 /* Returns the class type of 'netdev_dev'.
1334 * The caller must not free the returned value. */
1336 netdev_dev_get_type(const struct netdev_dev *netdev_dev)
1338 return netdev_dev->netdev_class->type;
1341 /* Returns the class associated with 'netdev_dev'. */
1342 const struct netdev_class *
1343 netdev_dev_get_class(const struct netdev_dev *netdev_dev)
1345 return netdev_dev->netdev_class;
1348 /* Returns the name of 'netdev_dev'.
1350 * The caller must not free the returned value. */
1352 netdev_dev_get_name(const struct netdev_dev *netdev_dev)
1354 return netdev_dev->name;
1357 /* Returns the netdev_dev with 'name' or NULL if there is none.
1359 * The caller must not free the returned value. */
1361 netdev_dev_from_name(const char *name)
1363 return shash_find_data(&netdev_dev_shash, name);
1366 /* Fills 'device_list' with devices that match 'netdev_class'.
1368 * The caller is responsible for initializing and destroying 'device_list'
1369 * but the contained netdev_devs must not be freed. */
1371 netdev_dev_get_devices(const struct netdev_class *netdev_class,
1372 struct shash *device_list)
1374 struct shash_node *node;
1375 SHASH_FOR_EACH (node, &netdev_dev_shash) {
1376 struct netdev_dev *dev = node->data;
1378 if (dev->netdev_class == netdev_class) {
1379 shash_add(device_list, node->name, node->data);
1384 /* Initializes 'netdev' as a instance of the netdev_dev.
1386 * This function adds 'netdev' to a netdev-owned linked list, so it is very
1387 * important that 'netdev' only be freed after calling netdev_close(). */
1389 netdev_init(struct netdev *netdev, struct netdev_dev *netdev_dev)
1391 memset(netdev, 0, sizeof *netdev);
1392 netdev->netdev_dev = netdev_dev;
1393 list_push_back(&netdev_list, &netdev->node);
1396 /* Undoes the results of initialization.
1398 * Normally this function only needs to be called from netdev_close().
1399 * However, it may be called by providers due to an error on opening
1400 * that occurs after initialization. It this case netdev_close() would
1401 * never be called. */
1403 netdev_uninit(struct netdev *netdev, bool close)
1405 /* Restore flags that we changed, if any. */
1406 int error = restore_flags(netdev);
1407 list_remove(&netdev->node);
1409 VLOG_WARN("failed to restore network device flags on %s: %s",
1410 netdev_get_name(netdev), strerror(error));
1414 netdev_get_dev(netdev)->netdev_class->close(netdev);
1419 /* Returns the class type of 'netdev'.
1421 * The caller must not free the returned value. */
1423 netdev_get_type(const struct netdev *netdev)
1425 return netdev_get_dev(netdev)->netdev_class->type;
1429 netdev_get_dev(const struct netdev *netdev)
1431 return netdev->netdev_dev;
1434 /* Initializes 'notifier' as a netdev notifier for 'netdev', for which
1435 * notification will consist of calling 'cb', with auxiliary data 'aux'. */
1437 netdev_notifier_init(struct netdev_notifier *notifier, struct netdev *netdev,
1438 void (*cb)(struct netdev_notifier *), void *aux)
1440 notifier->netdev = netdev;
1442 notifier->aux = aux;
1445 /* Tracks changes in the status of a set of network devices. */
1446 struct netdev_monitor {
1447 struct shash polled_netdevs;
1448 struct shash changed_netdevs;
1451 /* Creates and returns a new structure for monitor changes in the status of
1452 * network devices. */
1453 struct netdev_monitor *
1454 netdev_monitor_create(void)
1456 struct netdev_monitor *monitor = xmalloc(sizeof *monitor);
1457 shash_init(&monitor->polled_netdevs);
1458 shash_init(&monitor->changed_netdevs);
1462 /* Destroys 'monitor'. */
1464 netdev_monitor_destroy(struct netdev_monitor *monitor)
1467 struct shash_node *node;
1469 SHASH_FOR_EACH (node, &monitor->polled_netdevs) {
1470 struct netdev_notifier *notifier = node->data;
1471 netdev_get_dev(notifier->netdev)->netdev_class->poll_remove(
1475 shash_destroy(&monitor->polled_netdevs);
1476 shash_destroy(&monitor->changed_netdevs);
1482 netdev_monitor_cb(struct netdev_notifier *notifier)
1484 struct netdev_monitor *monitor = notifier->aux;
1485 const char *name = netdev_get_name(notifier->netdev);
1486 if (!shash_find(&monitor->changed_netdevs, name)) {
1487 shash_add(&monitor->changed_netdevs, name, NULL);
1491 /* Attempts to add 'netdev' as a netdev monitored by 'monitor'. Returns 0 if
1492 * successful, otherwise a positive errno value.
1494 * Adding a given 'netdev' to a monitor multiple times is equivalent to adding
1497 netdev_monitor_add(struct netdev_monitor *monitor, struct netdev *netdev)
1499 const char *netdev_name = netdev_get_name(netdev);
1501 if (!shash_find(&monitor->polled_netdevs, netdev_name)
1502 && netdev_get_dev(netdev)->netdev_class->poll_add)
1504 struct netdev_notifier *notifier;
1505 error = netdev_get_dev(netdev)->netdev_class->poll_add(netdev,
1506 netdev_monitor_cb, monitor, ¬ifier);
1508 assert(notifier->netdev == netdev);
1509 shash_add(&monitor->polled_netdevs, netdev_name, notifier);
1515 /* Removes 'netdev' from the set of netdevs monitored by 'monitor'. (This has
1516 * no effect if 'netdev' is not in the set of devices monitored by
1519 netdev_monitor_remove(struct netdev_monitor *monitor, struct netdev *netdev)
1521 const char *netdev_name = netdev_get_name(netdev);
1522 struct shash_node *node;
1524 node = shash_find(&monitor->polled_netdevs, netdev_name);
1526 /* Cancel future notifications. */
1527 struct netdev_notifier *notifier = node->data;
1528 netdev_get_dev(netdev)->netdev_class->poll_remove(notifier);
1529 shash_delete(&monitor->polled_netdevs, node);
1531 /* Drop any pending notification. */
1532 node = shash_find(&monitor->changed_netdevs, netdev_name);
1534 shash_delete(&monitor->changed_netdevs, node);
1539 /* Checks for changes to netdevs in the set monitored by 'monitor'. If any of
1540 * the attributes (Ethernet address, carrier status, speed or peer-advertised
1541 * speed, flags, etc.) of a network device monitored by 'monitor' has changed,
1542 * sets '*devnamep' to the name of a device that has changed and returns 0.
1543 * The caller is responsible for freeing '*devnamep' (with free()).
1545 * If no devices have changed, sets '*devnamep' to NULL and returns EAGAIN.
1548 netdev_monitor_poll(struct netdev_monitor *monitor, char **devnamep)
1550 struct shash_node *node = shash_first(&monitor->changed_netdevs);
1555 *devnamep = xstrdup(node->name);
1556 shash_delete(&monitor->changed_netdevs, node);
1561 /* Registers with the poll loop to wake up from the next call to poll_block()
1562 * when netdev_monitor_poll(monitor) would indicate that a device has
1565 netdev_monitor_poll_wait(const struct netdev_monitor *monitor)
1567 if (!shash_is_empty(&monitor->changed_netdevs)) {
1568 poll_immediate_wake();
1570 /* XXX Nothing needed here for netdev_linux, but maybe other netdev
1571 * classes need help. */
1575 /* Restore the network device flags on 'netdev' to those that were active
1576 * before we changed them. Returns 0 if successful, otherwise a positive
1579 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
1581 restore_flags(struct netdev *netdev)
1583 if (netdev->changed_flags) {
1584 enum netdev_flags restore = netdev->save_flags & netdev->changed_flags;
1585 enum netdev_flags old_flags;
1586 return netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
1587 netdev->changed_flags & ~restore,
1588 restore, &old_flags);
1593 /* Close all netdevs on shutdown so they can do any needed cleanup such as
1594 * destroying devices, restoring flags, etc. */
1596 close_all_netdevs(void *aux OVS_UNUSED)
1598 struct netdev *netdev, *next;
1599 LIST_FOR_EACH_SAFE(netdev, next, struct netdev, node, &netdev_list) {
1600 netdev_close(netdev);