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"
43 VLOG_DEFINE_THIS_MODULE(netdev);
45 COVERAGE_DEFINE(netdev_received);
46 COVERAGE_DEFINE(netdev_sent);
47 COVERAGE_DEFINE(netdev_add_router);
48 COVERAGE_DEFINE(netdev_get_stats);
50 static struct shash netdev_classes = SHASH_INITIALIZER(&netdev_classes);
52 /* All created network devices. */
53 static struct shash netdev_dev_shash = SHASH_INITIALIZER(&netdev_dev_shash);
55 /* All open network devices. */
56 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
58 /* This is set pretty low because we probably won't learn anything from the
59 * additional log messages. */
60 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
62 static void close_all_netdevs(void *aux OVS_UNUSED);
63 static int restore_flags(struct netdev *netdev);
64 void update_device_args(struct netdev_dev *, const struct shash *args);
67 netdev_initialize(void)
74 fatal_signal_add_hook(close_all_netdevs, NULL, NULL, true);
77 netdev_register_provider(&netdev_linux_class);
78 netdev_register_provider(&netdev_internal_class);
79 netdev_register_provider(&netdev_tap_class);
80 netdev_vport_register();
85 /* Performs periodic work needed by all the various kinds of netdevs.
87 * If your program opens any netdevs, it must call this function within its
92 struct shash_node *node;
93 SHASH_FOR_EACH(node, &netdev_classes) {
94 const struct netdev_class *netdev_class = node->data;
95 if (netdev_class->run) {
101 /* Arranges for poll_block() to wake up when netdev_run() needs to be called.
103 * If your program opens any netdevs, it must call this function within its
108 struct shash_node *node;
109 SHASH_FOR_EACH(node, &netdev_classes) {
110 const struct netdev_class *netdev_class = node->data;
111 if (netdev_class->wait) {
112 netdev_class->wait();
117 /* Initializes and registers a new netdev provider. After successful
118 * registration, new netdevs of that type can be opened using netdev_open(). */
120 netdev_register_provider(const struct netdev_class *new_class)
122 if (shash_find(&netdev_classes, new_class->type)) {
123 VLOG_WARN("attempted to register duplicate netdev provider: %s",
128 if (new_class->init) {
129 int error = new_class->init();
131 VLOG_ERR("failed to initialize %s network device class: %s",
132 new_class->type, strerror(error));
137 shash_add(&netdev_classes, new_class->type, new_class);
142 /* Unregisters a netdev provider. 'type' must have been previously
143 * registered and not currently be in use by any netdevs. After unregistration
144 * new netdevs of that type cannot be opened using netdev_open(). */
146 netdev_unregister_provider(const char *type)
148 struct shash_node *del_node, *netdev_dev_node;
150 del_node = shash_find(&netdev_classes, type);
152 VLOG_WARN("attempted to unregister a netdev provider that is not "
153 "registered: %s", type);
157 SHASH_FOR_EACH(netdev_dev_node, &netdev_dev_shash) {
158 struct netdev_dev *netdev_dev = netdev_dev_node->data;
159 if (!strcmp(netdev_dev->netdev_class->type, type)) {
160 VLOG_WARN("attempted to unregister in use netdev provider: %s",
166 shash_delete(&netdev_classes, del_node);
171 const struct netdev_class *
172 netdev_lookup_provider(const char *type)
175 return shash_find_data(&netdev_classes, type && type[0] ? type : "system");
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);
195 update_device_args(struct netdev_dev *dev, const struct shash *args)
197 smap_destroy(&dev->args);
198 smap_clone(&dev->args, args);
201 /* Opens the network device named 'name' (e.g. "eth0") and returns zero if
202 * successful, otherwise a positive errno value. On success, sets '*netdevp'
203 * to the new network device, otherwise to null.
205 * If this is the first time the device has been opened, then create is called
206 * before opening. The device is created using the given type and arguments.
208 * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
209 * capture frames of that type received on the device. It may also be one of
210 * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those
213 netdev_open(struct netdev_options *options, struct netdev **netdevp)
215 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
216 struct netdev_dev *netdev_dev;
222 if (!options->args) {
223 options->args = &empty_args;
226 netdev_dev = shash_find_data(&netdev_dev_shash, options->name);
229 const struct netdev_class *class;
231 class = netdev_lookup_provider(options->type);
233 VLOG_WARN("could not create netdev %s of unknown type %s",
234 options->name, options->type);
237 error = class->create(class, options->name, options->args,
242 assert(netdev_dev->netdev_class == class);
244 } else if (!shash_is_empty(options->args) &&
245 !smap_equal(&netdev_dev->args, options->args)) {
247 VLOG_WARN("%s: attempted to open already open netdev with "
248 "different arguments", options->name);
252 error = netdev_dev->netdev_class->open(netdev_dev, options->ethertype,
256 netdev_dev->ref_cnt++;
258 if (!netdev_dev->ref_cnt) {
259 netdev_dev_uninit(netdev_dev, true);
267 netdev_open_default(const char *name, struct netdev **netdevp)
269 struct netdev_options options;
271 memset(&options, 0, sizeof options);
273 options.ethertype = NETDEV_ETH_TYPE_NONE;
275 return netdev_open(&options, netdevp);
278 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
279 * or NULL if none are needed. */
281 netdev_set_config(struct netdev *netdev, const struct shash *args)
283 struct shash empty_args = SHASH_INITIALIZER(&empty_args);
284 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
290 if (netdev_dev->netdev_class->set_config) {
291 if (!smap_equal(&netdev_dev->args, args)) {
292 update_device_args(netdev_dev, args);
293 return netdev_dev->netdev_class->set_config(netdev_dev, args);
295 } else if (!shash_is_empty(args)) {
296 VLOG_WARN("%s: arguments provided to device whose configuration "
297 "cannot be changed", netdev_get_name(netdev));
303 /* Returns the current configuration for 'netdev'. This is either the
304 * configuration passed to netdev_open() or netdev_set_config(), or it is a
305 * configuration retrieved from the device itself if no configuration was
306 * passed to those functions.
308 * 'netdev' retains ownership of the returned configuration. */
310 netdev_get_config(const struct netdev *netdev)
312 return &netdev_get_dev(netdev)->args;
315 /* Closes and destroys 'netdev'. */
317 netdev_close(struct netdev *netdev)
320 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
322 assert(netdev_dev->ref_cnt);
323 netdev_dev->ref_cnt--;
324 netdev_uninit(netdev, true);
326 /* If the reference count for the netdev device is zero, destroy it. */
327 if (!netdev_dev->ref_cnt) {
328 netdev_dev_uninit(netdev_dev, true);
333 /* Returns true if a network device named 'name' exists and may be opened,
334 * otherwise false. */
336 netdev_exists(const char *name)
338 struct netdev *netdev;
341 error = netdev_open_default(name, &netdev);
343 netdev_close(netdev);
346 if (error != ENODEV) {
347 VLOG_WARN("failed to open network device %s: %s",
348 name, strerror(error));
354 /* Returns true if a network device named 'name' is currently opened,
355 * otherwise false. */
357 netdev_is_open(const char *name)
359 return !!shash_find_data(&netdev_dev_shash, name);
362 /* Clears 'svec' and enumerates the names of all known network devices. */
364 netdev_enumerate(struct svec *svec)
366 struct shash_node *node;
372 SHASH_FOR_EACH(node, &netdev_classes) {
373 const struct netdev_class *netdev_class = node->data;
374 if (netdev_class->enumerate) {
375 int retval = netdev_class->enumerate(svec);
377 VLOG_WARN("failed to enumerate %s network devices: %s",
378 netdev_class->type, strerror(retval));
389 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
390 * must have initialized with sufficient room for the packet. The space
391 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
392 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
393 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
394 * need not be included.)
396 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
397 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
398 * positive errno value. Returns EAGAIN immediately if no packet is ready to
401 * Some network devices may not implement support for this function. In such
402 * cases this function will always return EOPNOTSUPP. */
404 netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
406 int (*recv)(struct netdev *, void *, size_t);
409 assert(buffer->size == 0);
410 assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
412 recv = netdev_get_dev(netdev)->netdev_class->recv;
414 ? (recv)(netdev, buffer->data, ofpbuf_tailroom(buffer))
417 COVERAGE_INC(netdev_received);
418 buffer->size += retval;
419 if (buffer->size < ETH_TOTAL_MIN) {
420 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
428 /* Registers with the poll loop to wake up from the next call to poll_block()
429 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
431 netdev_recv_wait(struct netdev *netdev)
433 void (*recv_wait)(struct netdev *);
435 recv_wait = netdev_get_dev(netdev)->netdev_class->recv_wait;
441 /* Discards all packets waiting to be received from 'netdev'. */
443 netdev_drain(struct netdev *netdev)
445 int (*drain)(struct netdev *);
447 drain = netdev_get_dev(netdev)->netdev_class->drain;
448 return drain ? drain(netdev) : 0;
451 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
452 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
453 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
454 * the packet is too big or too small to transmit on the device.
456 * The caller retains ownership of 'buffer' in all cases.
458 * The kernel maintains a packet transmission queue, so the caller is not
459 * expected to do additional queuing of packets.
461 * Some network devices may not implement support for this function. In such
462 * cases this function will always return EOPNOTSUPP. */
464 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
466 int (*send)(struct netdev *, const void *, size_t);
469 send = netdev_get_dev(netdev)->netdev_class->send;
470 error = send ? (send)(netdev, buffer->data, buffer->size) : EOPNOTSUPP;
472 COVERAGE_INC(netdev_sent);
477 /* Registers with the poll loop to wake up from the next call to poll_block()
478 * when the packet transmission queue has sufficient room to transmit a packet
479 * with netdev_send().
481 * The kernel maintains a packet transmission queue, so the client is not
482 * expected to do additional queuing of packets. Thus, this function is
483 * unlikely to ever be used. It is included for completeness. */
485 netdev_send_wait(struct netdev *netdev)
487 void (*send_wait)(struct netdev *);
489 send_wait = netdev_get_dev(netdev)->netdev_class->send_wait;
495 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
496 * otherwise a positive errno value. */
498 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
500 return netdev_get_dev(netdev)->netdev_class->set_etheraddr(netdev, mac);
503 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
504 * the MAC address into 'mac'. On failure, returns a positive errno value and
505 * clears 'mac' to all-zeros. */
507 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
509 return netdev_get_dev(netdev)->netdev_class->get_etheraddr(netdev, mac);
512 /* Returns the name of the network device that 'netdev' represents,
513 * e.g. "eth0". The caller must not modify or free the returned string. */
515 netdev_get_name(const struct netdev *netdev)
517 return netdev_get_dev(netdev)->name;
520 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
521 * (and received) packets, in bytes, not including the hardware header; thus,
522 * this is typically 1500 bytes for Ethernet devices.
524 * If successful, returns 0 and stores the MTU size in '*mtup'. Stores INT_MAX
525 * in '*mtup' if 'netdev' does not have an MTU (as e.g. some tunnels do not).On
526 * failure, returns a positive errno value and stores ETH_PAYLOAD_MAX (1500) in
529 netdev_get_mtu(const struct netdev *netdev, int *mtup)
531 int error = netdev_get_dev(netdev)->netdev_class->get_mtu(netdev, mtup);
533 VLOG_WARN_RL(&rl, "failed to retrieve MTU for network device %s: %s",
534 netdev_get_name(netdev), strerror(error));
535 *mtup = ETH_PAYLOAD_MAX;
540 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
541 * failure, returns a negative errno value.
543 * The desired semantics of the ifindex value are a combination of those
544 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
545 * value should be unique within a host and remain stable at least until
546 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
547 * but many systems do not follow this rule anyhow.
549 * Some network devices may not implement support for this function. In such
550 * cases this function will always return -EOPNOTSUPP.
553 netdev_get_ifindex(const struct netdev *netdev)
555 int (*get_ifindex)(const struct netdev *);
557 get_ifindex = netdev_get_dev(netdev)->netdev_class->get_ifindex;
559 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
562 /* Stores the features supported by 'netdev' into each of '*current',
563 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
564 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
565 * successful, otherwise a positive errno value. On failure, all of the
566 * passed-in values are set to 0.
568 * Some network devices may not implement support for this function. In such
569 * cases this function will always return EOPNOTSUPP. */
571 netdev_get_features(const struct netdev *netdev,
572 uint32_t *current, uint32_t *advertised,
573 uint32_t *supported, uint32_t *peer)
575 int (*get_features)(const struct netdev *netdev,
576 uint32_t *current, uint32_t *advertised,
577 uint32_t *supported, uint32_t *peer);
585 advertised = &dummy[1];
588 supported = &dummy[2];
594 get_features = netdev_get_dev(netdev)->netdev_class->get_features;
596 ? get_features(netdev, current, advertised, supported, peer)
599 *current = *advertised = *supported = *peer = 0;
604 /* Returns the maximum speed of a network connection that has the "enum
605 * ofp_port_features" bits in 'features', in bits per second. If no bits that
606 * indicate a speed are set in 'features', assumes 100Mbps. */
608 netdev_features_to_bps(uint32_t features)
611 F_10000MB = OFPPF_10GB_FD,
612 F_1000MB = OFPPF_1GB_HD | OFPPF_1GB_FD,
613 F_100MB = OFPPF_100MB_HD | OFPPF_100MB_FD,
614 F_10MB = OFPPF_10MB_HD | OFPPF_10MB_FD
617 return ( features & F_10000MB ? UINT64_C(10000000000)
618 : features & F_1000MB ? UINT64_C(1000000000)
619 : features & F_100MB ? UINT64_C(100000000)
620 : features & F_10MB ? UINT64_C(10000000)
621 : UINT64_C(100000000));
624 /* Returns true if any of the "enum ofp_port_features" bits that indicate a
625 * full-duplex link are set in 'features', otherwise false. */
627 netdev_features_is_full_duplex(uint32_t features)
629 return (features & (OFPPF_10MB_FD | OFPPF_100MB_FD | OFPPF_1GB_FD
630 | OFPPF_10GB_FD)) != 0;
633 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
634 * successful, otherwise a positive errno value. */
636 netdev_set_advertisements(struct netdev *netdev, uint32_t advertise)
638 return (netdev_get_dev(netdev)->netdev_class->set_advertisements
639 ? netdev_get_dev(netdev)->netdev_class->set_advertisements(
644 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
645 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
646 * errno value and sets '*address' to 0 (INADDR_ANY).
648 * The following error values have well-defined meanings:
650 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
652 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
654 * 'address' or 'netmask' or both may be null, in which case the address or
655 * netmask is not reported. */
657 netdev_get_in4(const struct netdev *netdev,
658 struct in_addr *address_, struct in_addr *netmask_)
660 struct in_addr address;
661 struct in_addr netmask;
664 error = (netdev_get_dev(netdev)->netdev_class->get_in4
665 ? netdev_get_dev(netdev)->netdev_class->get_in4(netdev,
669 address_->s_addr = error ? 0 : address.s_addr;
672 netmask_->s_addr = error ? 0 : netmask.s_addr;
677 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
678 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
679 * positive errno value. */
681 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
683 return (netdev_get_dev(netdev)->netdev_class->set_in4
684 ? netdev_get_dev(netdev)->netdev_class->set_in4(netdev, addr, mask)
688 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
691 netdev_add_router(struct netdev *netdev, struct in_addr router)
693 COVERAGE_INC(netdev_add_router);
694 return (netdev_get_dev(netdev)->netdev_class->add_router
695 ? netdev_get_dev(netdev)->netdev_class->add_router(netdev, router)
699 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
700 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
701 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
702 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
703 * a directly connected network) in '*next_hop' and a copy of the name of the
704 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
705 * responsible for freeing '*netdev_name' (by calling free()). */
707 netdev_get_next_hop(const struct netdev *netdev,
708 const struct in_addr *host, struct in_addr *next_hop,
711 int error = (netdev_get_dev(netdev)->netdev_class->get_next_hop
712 ? netdev_get_dev(netdev)->netdev_class->get_next_hop(
713 host, next_hop, netdev_name)
716 next_hop->s_addr = 0;
722 /* Populates 'sh' with status information.
724 * Populates 'sh' with 'netdev' specific status information. This information
725 * may be used to populate the status column of the Interface table as defined
726 * in ovs-vswitchd.conf.db(5). */
728 netdev_get_status(const struct netdev *netdev, struct shash *sh)
730 struct netdev_dev *dev = netdev_get_dev(netdev);
732 return (dev->netdev_class->get_status
733 ? dev->netdev_class->get_status(netdev, sh)
737 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
738 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
739 * all-zero-bits (in6addr_any).
741 * The following error values have well-defined meanings:
743 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
745 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
747 * 'in6' may be null, in which case the address itself is not reported. */
749 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
751 struct in6_addr dummy;
754 error = (netdev_get_dev(netdev)->netdev_class->get_in6
755 ? netdev_get_dev(netdev)->netdev_class->get_in6(netdev,
759 memset(in6, 0, sizeof *in6);
764 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
765 * 'on'. If 'permanent' is true, the changes will persist; otherwise, they
766 * will be reverted when 'netdev' is closed or the program exits. Returns 0 if
767 * successful, otherwise a positive errno value. */
769 do_update_flags(struct netdev *netdev, enum netdev_flags off,
770 enum netdev_flags on, enum netdev_flags *old_flagsp,
773 enum netdev_flags old_flags;
776 error = netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
777 off & ~on, on, &old_flags);
779 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
780 off || on ? "set" : "get", netdev_get_name(netdev),
783 } else if ((off || on) && !permanent) {
784 enum netdev_flags new_flags = (old_flags & ~off) | on;
785 enum netdev_flags changed_flags = old_flags ^ new_flags;
787 if (!netdev->changed_flags) {
788 netdev->save_flags = old_flags;
790 netdev->changed_flags |= changed_flags;
794 *old_flagsp = old_flags;
799 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
800 * Returns 0 if successful, otherwise a positive errno value. On failure,
801 * stores 0 into '*flagsp'. */
803 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
805 struct netdev *netdev = (struct netdev *) netdev_;
806 return do_update_flags(netdev, 0, 0, flagsp, false);
809 /* Sets the flags for 'netdev' to 'flags'.
810 * If 'permanent' is true, the changes will persist; otherwise, they
811 * will be reverted when 'netdev' is closed or the program exits.
812 * Returns 0 if successful, otherwise a positive errno value. */
814 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
817 return do_update_flags(netdev, -1, flags, NULL, permanent);
820 /* Turns on the specified 'flags' on 'netdev'.
821 * If 'permanent' is true, the changes will persist; otherwise, they
822 * will be reverted when 'netdev' is closed or the program exits.
823 * Returns 0 if successful, otherwise a positive errno value. */
825 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
828 return do_update_flags(netdev, 0, flags, NULL, permanent);
831 /* Turns off the specified 'flags' on 'netdev'.
832 * If 'permanent' is true, the changes will persist; otherwise, they
833 * will be reverted when 'netdev' is closed or the program exits.
834 * Returns 0 if successful, otherwise a positive errno value. */
836 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
839 return do_update_flags(netdev, flags, 0, NULL, permanent);
842 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
843 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
844 * returns 0. Otherwise, it returns a positive errno value; in particular,
845 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
847 netdev_arp_lookup(const struct netdev *netdev,
848 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
850 int error = (netdev_get_dev(netdev)->netdev_class->arp_lookup
851 ? netdev_get_dev(netdev)->netdev_class->arp_lookup(netdev,
855 memset(mac, 0, ETH_ADDR_LEN);
860 /* Returns true if carrier is active (link light is on) on 'netdev'. */
862 netdev_get_carrier(const struct netdev *netdev)
865 enum netdev_flags flags;
868 netdev_get_flags(netdev, &flags);
869 if (!(flags & NETDEV_UP)) {
873 if (!netdev_get_dev(netdev)->netdev_class->get_carrier) {
877 error = netdev_get_dev(netdev)->netdev_class->get_carrier(netdev,
880 VLOG_DBG("%s: failed to get network device carrier status, assuming "
881 "down: %s", netdev_get_name(netdev), strerror(error));
888 /* Returns true if 'netdev' is up according to its MII. */
890 netdev_get_miimon(const struct netdev *netdev)
893 enum netdev_flags flags;
896 netdev_get_flags(netdev, &flags);
897 if (!(flags & NETDEV_UP)) {
901 if (!netdev_get_dev(netdev)->netdev_class->get_miimon) {
905 error = netdev_get_dev(netdev)->netdev_class->get_miimon(netdev, &miimon);
908 VLOG_DBG("%s: failed to get network device MII status, assuming "
909 "down: %s", netdev_get_name(netdev), strerror(error));
916 /* Retrieves current device stats for 'netdev'. */
918 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
922 COVERAGE_INC(netdev_get_stats);
923 error = (netdev_get_dev(netdev)->netdev_class->get_stats
924 ? netdev_get_dev(netdev)->netdev_class->get_stats(netdev, stats)
927 memset(stats, 0xff, sizeof *stats);
932 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
933 * Returns 0 if successful, otherwise a positive errno value.
935 * This will probably fail for most network devices. Some devices might only
936 * allow setting their stats to 0. */
938 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
940 return (netdev_get_dev(netdev)->netdev_class->set_stats
941 ? netdev_get_dev(netdev)->netdev_class->set_stats(netdev, stats)
945 /* Attempts to set input rate limiting (policing) policy, such that up to
946 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
947 * size of 'kbits' kb. */
949 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
950 uint32_t kbits_burst)
952 return (netdev_get_dev(netdev)->netdev_class->set_policing
953 ? netdev_get_dev(netdev)->netdev_class->set_policing(netdev,
954 kbits_rate, kbits_burst)
958 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
959 * empty if 'netdev' does not support QoS. Any names added to 'types' should
960 * be documented as valid for the "type" column in the "QoS" table in
961 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
963 * Every network device supports disabling QoS with a type of "", but this type
964 * will not be added to 'types'.
966 * The caller must initialize 'types' (e.g. with svec_init()) before calling
967 * this function. The caller is responsible for destroying 'types' (e.g. with
968 * svec_destroy()) when it is no longer needed.
970 * Returns 0 if successful, otherwise a positive errno value. */
972 netdev_get_qos_types(const struct netdev *netdev, struct svec *types)
974 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
975 return (class->get_qos_types
976 ? class->get_qos_types(netdev, types)
980 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
981 * which should be "" or one of the types returned by netdev_get_qos_types()
982 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
983 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
984 * 'caps' to all zeros. */
986 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
987 struct netdev_qos_capabilities *caps)
989 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
992 int retval = (class->get_qos_capabilities
993 ? class->get_qos_capabilities(netdev, type, caps)
996 memset(caps, 0, sizeof *caps);
1000 /* Every netdev supports turning off QoS. */
1001 memset(caps, 0, sizeof *caps);
1006 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1007 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1008 * the number of queues (zero on failure) in '*n_queuesp'.
1010 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1012 netdev_get_n_queues(const struct netdev *netdev,
1013 const char *type, unsigned int *n_queuesp)
1015 struct netdev_qos_capabilities caps;
1018 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1019 *n_queuesp = caps.n_queues;
1023 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1024 * stores the name of the current form of QoS into '*typep', stores any details
1025 * of configuration as string key-value pairs in 'details', and returns 0. On
1026 * failure, sets '*typep' to NULL and returns a positive errno value.
1028 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1030 * The caller must initialize 'details' as an empty shash (e.g. with
1031 * shash_init()) before calling this function. The caller must free 'details',
1032 * including 'data' members, when it is no longer needed (e.g. with
1033 * shash_destroy_free_data()).
1035 * The caller must not modify or free '*typep'.
1037 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1038 * 'netdev'. The contents of 'details' should be documented as valid for
1039 * '*typep' in the "other_config" column in the "QoS" table in
1040 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1042 netdev_get_qos(const struct netdev *netdev,
1043 const char **typep, struct shash *details)
1045 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1048 if (class->get_qos) {
1049 retval = class->get_qos(netdev, typep, details);
1052 shash_clear_free_data(details);
1056 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1062 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1063 * with details of configuration from 'details'. Returns 0 if successful,
1064 * otherwise a positive errno value. On error, the previous QoS configuration
1067 * When this function changes the type of QoS (not just 'details'), this also
1068 * resets all queue configuration for 'netdev' to their defaults (which depend
1069 * on the specific type of QoS). Otherwise, the queue configuration for
1070 * 'netdev' is unchanged.
1072 * 'type' should be "" (to disable QoS) or one of the types returned by
1073 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1074 * documented as valid for the given 'type' in the "other_config" column in the
1075 * "QoS" table in vswitchd/vswitch.xml (which is built as
1076 * ovs-vswitchd.conf.db(8)).
1078 * NULL may be specified for 'details' if there are no configuration
1081 netdev_set_qos(struct netdev *netdev,
1082 const char *type, const struct shash *details)
1084 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1090 if (class->set_qos) {
1092 static struct shash empty = SHASH_INITIALIZER(&empty);
1095 return class->set_qos(netdev, type, details);
1097 return *type ? EOPNOTSUPP : 0;
1101 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1102 * successful, adds that information as string key-value pairs to 'details'.
1103 * Returns 0 if successful, otherwise a positive errno value.
1105 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1106 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1108 * The returned contents of 'details' should be documented as valid for the
1109 * given 'type' in the "other_config" column in the "Queue" table in
1110 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1112 * The caller must initialize 'details' (e.g. with shash_init()) before calling
1113 * this function. The caller must free 'details', including 'data' members,
1114 * when it is no longer needed (e.g. with shash_destroy_free_data()). */
1116 netdev_get_queue(const struct netdev *netdev,
1117 unsigned int queue_id, struct shash *details)
1119 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1122 retval = (class->get_queue
1123 ? class->get_queue(netdev, queue_id, details)
1126 shash_clear_free_data(details);
1131 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1132 * string pairs in 'details'. The contents of 'details' should be documented
1133 * as valid for the given 'type' in the "other_config" column in the "Queue"
1134 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1135 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1136 * given queue's configuration should be unmodified.
1138 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1139 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1141 * This function does not modify 'details', and the caller retains ownership of
1144 netdev_set_queue(struct netdev *netdev,
1145 unsigned int queue_id, const struct shash *details)
1147 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1148 return (class->set_queue
1149 ? class->set_queue(netdev, queue_id, details)
1153 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1154 * of QoS may have a fixed set of queues, in which case attempts to delete them
1155 * will fail with EOPNOTSUPP.
1157 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1158 * given queue will be unmodified.
1160 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1161 * the current form of QoS (e.g. as returned by
1162 * netdev_get_n_queues(netdev)). */
1164 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1166 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1167 return (class->delete_queue
1168 ? class->delete_queue(netdev, queue_id)
1172 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1173 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1174 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1175 * positive errno value and fills 'stats' with all-1-bits. */
1177 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1178 struct netdev_queue_stats *stats)
1180 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1183 retval = (class->get_queue_stats
1184 ? class->get_queue_stats(netdev, queue_id, stats)
1187 memset(stats, 0xff, sizeof *stats);
1192 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1193 * its configuration, and the 'aux' specified by the caller. The order of
1194 * iteration is unspecified, but (when successful) each queue is visited
1197 * Calling this function may be more efficient than calling netdev_get_queue()
1200 * 'cb' must not modify or free the 'details' argument passed in.
1202 * Returns 0 if successful, otherwise a positive errno value. On error, some
1203 * configured queues may not have been included in the iteration. */
1205 netdev_dump_queues(const struct netdev *netdev,
1206 netdev_dump_queues_cb *cb, void *aux)
1208 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1209 return (class->dump_queues
1210 ? class->dump_queues(netdev, cb, aux)
1214 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1215 * its statistics, and the 'aux' specified by the caller. The order of
1216 * iteration is unspecified, but (when successful) each queue is visited
1219 * Calling this function may be more efficient than calling
1220 * netdev_get_queue_stats() for every queue.
1222 * 'cb' must not modify or free the statistics passed in.
1224 * Returns 0 if successful, otherwise a positive errno value. On error, some
1225 * configured queues may not have been included in the iteration. */
1227 netdev_dump_queue_stats(const struct netdev *netdev,
1228 netdev_dump_queue_stats_cb *cb, void *aux)
1230 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1231 return (class->dump_queue_stats
1232 ? class->dump_queue_stats(netdev, cb, aux)
1236 /* If 'netdev' is a VLAN network device (e.g. one created with vconfig(8)),
1237 * sets '*vlan_vid' to the VLAN VID associated with that device and returns 0.
1238 * Otherwise returns a errno value (specifically ENOENT if 'netdev_name' is the
1239 * name of a network device that is not a VLAN device) and sets '*vlan_vid' to
1242 netdev_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1244 int error = (netdev_get_dev(netdev)->netdev_class->get_vlan_vid
1245 ? netdev_get_dev(netdev)->netdev_class->get_vlan_vid(netdev,
1254 /* Returns a network device that has 'in4' as its IP address, if one exists,
1255 * otherwise a null pointer. */
1257 netdev_find_dev_by_in4(const struct in_addr *in4)
1259 struct netdev *netdev;
1260 struct svec dev_list = SVEC_EMPTY_INITIALIZER;
1263 netdev_enumerate(&dev_list);
1264 for (i = 0; i < dev_list.n; i++) {
1265 const char *name = dev_list.names[i];
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 svec_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 shash 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 shash_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 shash_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 shash_add_once(&monitor->changed_netdevs, name, NULL);
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 node = shash_find(&monitor->changed_netdevs, netdev_name);
1531 shash_delete(&monitor->changed_netdevs, node);
1536 /* Checks for changes to netdevs in the set monitored by 'monitor'. If any of
1537 * the attributes (Ethernet address, carrier status, speed or peer-advertised
1538 * speed, flags, etc.) of a network device monitored by 'monitor' has changed,
1539 * sets '*devnamep' to the name of a device that has changed and returns 0.
1540 * The caller is responsible for freeing '*devnamep' (with free()).
1542 * If no devices have changed, sets '*devnamep' to NULL and returns EAGAIN. */
1544 netdev_monitor_poll(struct netdev_monitor *monitor, char **devnamep)
1546 struct shash_node *node = shash_first(&monitor->changed_netdevs);
1551 *devnamep = shash_steal(&monitor->changed_netdevs, node);
1556 /* Registers with the poll loop to wake up from the next call to poll_block()
1557 * when netdev_monitor_poll(monitor) would indicate that a device has
1560 netdev_monitor_poll_wait(const struct netdev_monitor *monitor)
1562 if (!shash_is_empty(&monitor->changed_netdevs)) {
1563 poll_immediate_wake();
1565 /* XXX Nothing needed here for netdev_linux, but maybe other netdev
1566 * classes need help. */
1570 /* Restore the network device flags on 'netdev' to those that were active
1571 * before we changed them. Returns 0 if successful, otherwise a positive
1574 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
1576 restore_flags(struct netdev *netdev)
1578 if (netdev->changed_flags) {
1579 enum netdev_flags restore = netdev->save_flags & netdev->changed_flags;
1580 enum netdev_flags old_flags;
1581 return netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
1582 netdev->changed_flags & ~restore,
1583 restore, &old_flags);
1588 /* Close all netdevs on shutdown so they can do any needed cleanup such as
1589 * destroying devices, restoring flags, etc. */
1591 close_all_netdevs(void *aux OVS_UNUSED)
1593 struct netdev *netdev, *next;
1594 LIST_FOR_EACH_SAFE(netdev, next, node, &netdev_list) {
1595 netdev_close(netdev);