2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
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.
22 #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"
45 VLOG_DEFINE_THIS_MODULE(netdev);
47 COVERAGE_DEFINE(netdev_received);
48 COVERAGE_DEFINE(netdev_sent);
49 COVERAGE_DEFINE(netdev_add_router);
50 COVERAGE_DEFINE(netdev_get_stats);
52 struct netdev_saved_flags {
53 struct netdev *netdev;
54 struct list node; /* In struct netdev's saved_flags_list. */
55 enum netdev_flags saved_flags;
56 enum netdev_flags saved_values;
59 /* Protects 'netdev_shash' and the mutable members of struct netdev. */
60 static struct ovs_mutex netdev_mutex = OVS_MUTEX_INITIALIZER;
62 /* All created network devices. */
63 static struct shash netdev_shash OVS_GUARDED_BY(netdev_mutex)
64 = SHASH_INITIALIZER(&netdev_shash);
66 /* Protects 'netdev_classes' against insertions or deletions.
68 * This is not an rwlock for performance reasons but to allow recursive
69 * acquisition when calling into providers. For example, netdev_run() calls
70 * into provider 'run' functions, which might reasonably want to call one of
71 * the netdev functions that takes netdev_class_rwlock read-only. */
72 static struct ovs_rwlock netdev_class_rwlock OVS_ACQ_BEFORE(netdev_mutex)
73 = OVS_RWLOCK_INITIALIZER;
75 /* Contains 'struct netdev_registered_class'es. */
76 static struct hmap netdev_classes OVS_GUARDED_BY(netdev_class_rwlock)
77 = HMAP_INITIALIZER(&netdev_classes);
79 struct netdev_registered_class {
80 struct hmap_node hmap_node; /* In 'netdev_classes', by class->type. */
81 const struct netdev_class *class;
82 atomic_int ref_cnt; /* Number of 'struct netdev's of this class. */
85 /* This is set pretty low because we probably won't learn anything from the
86 * additional log messages. */
87 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
89 static void restore_all_flags(void *aux OVS_UNUSED);
90 void update_device_args(struct netdev *, const struct shash *args);
93 netdev_initialize(void)
94 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
96 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
98 if (ovsthread_once_start(&once)) {
99 fatal_signal_add_hook(restore_all_flags, NULL, NULL, true);
100 netdev_vport_patch_register();
102 #ifdef LINUX_DATAPATH
103 netdev_register_provider(&netdev_linux_class);
104 netdev_register_provider(&netdev_internal_class);
105 netdev_register_provider(&netdev_tap_class);
106 netdev_vport_tunnel_register();
108 #if defined(__FreeBSD__) || defined(__NetBSD__)
109 netdev_register_provider(&netdev_tap_class);
110 netdev_register_provider(&netdev_bsd_class);
112 netdev_register_provider(&netdev_tunnel_class);
113 netdev_register_provider(&netdev_pltap_class);
115 ovsthread_once_done(&once);
119 /* Performs periodic work needed by all the various kinds of netdevs.
121 * If your program opens any netdevs, it must call this function within its
125 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
127 struct netdev_registered_class *rc;
129 ovs_rwlock_rdlock(&netdev_class_rwlock);
130 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
133 ovs_rwlock_unlock(&netdev_class_rwlock);
136 /* Arranges for poll_block() to wake up when netdev_run() needs to be called.
138 * If your program opens any netdevs, it must call this function within its
142 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
144 struct netdev_registered_class *rc;
146 ovs_rwlock_rdlock(&netdev_class_rwlock);
147 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
150 ovs_rwlock_unlock(&netdev_class_rwlock);
153 static struct netdev_registered_class *
154 netdev_lookup_class(const char *type)
155 OVS_REQ_RDLOCK(netdev_class_rwlock)
157 struct netdev_registered_class *rc;
159 HMAP_FOR_EACH_WITH_HASH (rc, hmap_node, hash_string(type, 0),
161 if (!strcmp(type, rc->class->type)) {
168 /* Initializes and registers a new netdev provider. After successful
169 * registration, new netdevs of that type can be opened using netdev_open(). */
171 netdev_register_provider(const struct netdev_class *new_class)
172 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
176 ovs_rwlock_wrlock(&netdev_class_rwlock);
177 if (netdev_lookup_class(new_class->type)) {
178 VLOG_WARN("attempted to register duplicate netdev provider: %s",
182 error = new_class->init ? new_class->init() : 0;
184 struct netdev_registered_class *rc;
186 rc = xmalloc(sizeof *rc);
187 hmap_insert(&netdev_classes, &rc->hmap_node,
188 hash_string(new_class->type, 0));
189 rc->class = new_class;
190 atomic_init(&rc->ref_cnt, 0);
192 VLOG_ERR("failed to initialize %s network device class: %s",
193 new_class->type, ovs_strerror(error));
196 ovs_rwlock_unlock(&netdev_class_rwlock);
201 /* Unregisters a netdev provider. 'type' must have been previously
202 * registered and not currently be in use by any netdevs. After unregistration
203 * new netdevs of that type cannot be opened using netdev_open(). */
205 netdev_unregister_provider(const char *type)
206 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
208 struct netdev_registered_class *rc;
211 ovs_rwlock_wrlock(&netdev_class_rwlock);
212 rc = netdev_lookup_class(type);
214 VLOG_WARN("attempted to unregister a netdev provider that is not "
215 "registered: %s", type);
216 error = EAFNOSUPPORT;
220 atomic_read(&rc->ref_cnt, &ref_cnt);
222 hmap_remove(&netdev_classes, &rc->hmap_node);
226 VLOG_WARN("attempted to unregister in use netdev provider: %s",
231 ovs_rwlock_unlock(&netdev_class_rwlock);
236 /* Clears 'types' and enumerates the types of all currently registered netdev
237 * providers into it. The caller must first initialize the sset. */
239 netdev_enumerate_types(struct sset *types)
240 OVS_EXCLUDED(netdev_mutex)
242 struct netdev_registered_class *rc;
247 ovs_rwlock_rdlock(&netdev_class_rwlock);
248 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
249 sset_add(types, rc->class->type);
251 ovs_rwlock_unlock(&netdev_class_rwlock);
254 /* Check that the network device name is not the same as any of the registered
255 * vport providers' dpif_port name (dpif_port is NULL if the vport provider
256 * does not define it) or the datapath internal port name (e.g. ovs-system).
258 * Returns true if there is a name conflict, false otherwise. */
260 netdev_is_reserved_name(const char *name)
261 OVS_EXCLUDED(netdev_mutex)
263 struct netdev_registered_class *rc;
267 ovs_rwlock_rdlock(&netdev_class_rwlock);
268 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
269 const char *dpif_port = netdev_vport_class_get_dpif_port(rc->class);
270 if (dpif_port && !strcmp(dpif_port, name)) {
271 ovs_rwlock_unlock(&netdev_class_rwlock);
275 ovs_rwlock_unlock(&netdev_class_rwlock);
277 if (!strncmp(name, "ovs-", 4)) {
282 dp_enumerate_types(&types);
283 SSET_FOR_EACH (type, &types) {
284 if (!strcmp(name+4, type)) {
285 sset_destroy(&types);
289 sset_destroy(&types);
295 /* Opens the network device named 'name' (e.g. "eth0") of the specified 'type'
296 * (e.g. "system") and returns zero if successful, otherwise a positive errno
297 * value. On success, sets '*netdevp' to the new network device, otherwise to
300 * Some network devices may need to be configured (with netdev_set_config())
301 * before they can be used. */
303 netdev_open(const char *name, const char *type, struct netdev **netdevp)
304 OVS_EXCLUDED(netdev_mutex)
306 struct netdev *netdev;
311 ovs_rwlock_rdlock(&netdev_class_rwlock);
312 ovs_mutex_lock(&netdev_mutex);
313 netdev = shash_find_data(&netdev_shash, name);
315 struct netdev_registered_class *rc;
317 rc = netdev_lookup_class(type && type[0] ? type : "system");
319 netdev = rc->class->alloc();
321 memset(netdev, 0, sizeof *netdev);
322 netdev->netdev_class = rc->class;
323 netdev->name = xstrdup(name);
324 netdev->node = shash_add(&netdev_shash, name, netdev);
325 list_init(&netdev->saved_flags_list);
327 error = rc->class->construct(netdev);
331 atomic_add(&rc->ref_cnt, 1, &old_ref_cnt);
334 ovs_assert(list_is_empty(&netdev->saved_flags_list));
335 shash_delete(&netdev_shash, netdev->node);
336 rc->class->dealloc(netdev);
342 VLOG_WARN("could not create netdev %s of unknown type %s",
344 error = EAFNOSUPPORT;
350 ovs_mutex_unlock(&netdev_mutex);
351 ovs_rwlock_unlock(&netdev_class_rwlock);
362 /* Returns a reference to 'netdev_' for the caller to own. Returns null if
363 * 'netdev_' is null. */
365 netdev_ref(const struct netdev *netdev_)
366 OVS_EXCLUDED(netdev_mutex)
368 struct netdev *netdev = CONST_CAST(struct netdev *, netdev_);
371 ovs_mutex_lock(&netdev_mutex);
372 ovs_assert(netdev->ref_cnt > 0);
374 ovs_mutex_unlock(&netdev_mutex);
379 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
380 * or NULL if none are needed. */
382 netdev_set_config(struct netdev *netdev, const struct smap *args)
383 OVS_EXCLUDED(netdev_mutex)
385 if (netdev->netdev_class->set_config) {
386 const struct smap no_args = SMAP_INITIALIZER(&no_args);
387 return netdev->netdev_class->set_config(netdev,
388 args ? args : &no_args);
389 } else if (args && !smap_is_empty(args)) {
390 VLOG_WARN("%s: arguments provided to device that is not configurable",
391 netdev_get_name(netdev));
397 /* Returns the current configuration for 'netdev' in 'args'. The caller must
398 * have already initialized 'args' with smap_init(). Returns 0 on success, in
399 * which case 'args' will be filled with 'netdev''s configuration. On failure
400 * returns a positive errno value, in which case 'args' will be empty.
402 * The caller owns 'args' and its contents and must eventually free them with
405 netdev_get_config(const struct netdev *netdev, struct smap *args)
406 OVS_EXCLUDED(netdev_mutex)
411 if (netdev->netdev_class->get_config) {
412 error = netdev->netdev_class->get_config(netdev, args);
423 const struct netdev_tunnel_config *
424 netdev_get_tunnel_config(const struct netdev *netdev)
425 OVS_EXCLUDED(netdev_mutex)
427 if (netdev->netdev_class->get_tunnel_config) {
428 return netdev->netdev_class->get_tunnel_config(netdev);
435 netdev_unref(struct netdev *dev)
436 OVS_RELEASES(netdev_mutex)
438 ovs_assert(dev->ref_cnt);
439 if (!--dev->ref_cnt) {
440 const struct netdev_class *class = dev->netdev_class;
441 struct netdev_registered_class *rc;
444 dev->netdev_class->destruct(dev);
446 shash_delete(&netdev_shash, dev->node);
448 dev->netdev_class->dealloc(dev);
449 ovs_mutex_unlock(&netdev_mutex);
451 ovs_rwlock_rdlock(&netdev_class_rwlock);
452 rc = netdev_lookup_class(class->type);
453 atomic_sub(&rc->ref_cnt, 1, &old_ref_cnt);
454 ovs_assert(old_ref_cnt > 0);
455 ovs_rwlock_unlock(&netdev_class_rwlock);
457 ovs_mutex_unlock(&netdev_mutex);
461 /* Closes and destroys 'netdev'. */
463 netdev_close(struct netdev *netdev)
464 OVS_EXCLUDED(netdev_mutex)
467 ovs_mutex_lock(&netdev_mutex);
468 netdev_unref(netdev);
472 /* Parses 'netdev_name_', which is of the form [type@]name into its component
473 * pieces. 'name' and 'type' must be freed by the caller. */
475 netdev_parse_name(const char *netdev_name_, char **name, char **type)
477 char *netdev_name = xstrdup(netdev_name_);
480 separator = strchr(netdev_name, '@');
484 *name = xstrdup(separator + 1);
487 *type = xstrdup("system");
492 netdev_rx_open(struct netdev *netdev, struct netdev_rx **rxp)
493 OVS_EXCLUDED(netdev_mutex)
497 if (netdev->netdev_class->rx_alloc) {
498 struct netdev_rx *rx = netdev->netdev_class->rx_alloc();
501 error = netdev->netdev_class->rx_construct(rx);
503 ovs_mutex_lock(&netdev_mutex);
505 ovs_mutex_unlock(&netdev_mutex);
510 netdev->netdev_class->rx_dealloc(rx);
523 netdev_rx_close(struct netdev_rx *rx)
524 OVS_EXCLUDED(netdev_mutex)
527 struct netdev *netdev = rx->netdev;
528 netdev->netdev_class->rx_destruct(rx);
529 netdev->netdev_class->rx_dealloc(rx);
530 netdev_close(netdev);
535 netdev_rx_recv(struct netdev_rx *rx, struct ofpbuf *buffer)
539 ovs_assert(buffer->size == 0);
540 ovs_assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
542 retval = rx->netdev->netdev_class->rx_recv(rx, buffer->data,
543 ofpbuf_tailroom(buffer));
545 COVERAGE_INC(netdev_received);
546 buffer->size += retval;
547 if (buffer->size < ETH_TOTAL_MIN) {
548 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
557 netdev_rx_wait(struct netdev_rx *rx)
559 rx->netdev->netdev_class->rx_wait(rx);
563 netdev_rx_drain(struct netdev_rx *rx)
565 return (rx->netdev->netdev_class->rx_drain
566 ? rx->netdev->netdev_class->rx_drain(rx)
570 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
571 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
572 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
573 * the packet is too big or too small to transmit on the device.
575 * The caller retains ownership of 'buffer' in all cases.
577 * The kernel maintains a packet transmission queue, so the caller is not
578 * expected to do additional queuing of packets.
580 * Some network devices may not implement support for this function. In such
581 * cases this function will always return EOPNOTSUPP. */
583 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
587 error = (netdev->netdev_class->send
588 ? netdev->netdev_class->send(netdev, buffer->data, buffer->size)
591 COVERAGE_INC(netdev_sent);
596 /* Registers with the poll loop to wake up from the next call to poll_block()
597 * when the packet transmission queue has sufficient room to transmit a packet
598 * with netdev_send().
600 * The kernel maintains a packet transmission queue, so the client is not
601 * expected to do additional queuing of packets. Thus, this function is
602 * unlikely to ever be used. It is included for completeness. */
604 netdev_send_wait(struct netdev *netdev)
606 if (netdev->netdev_class->send_wait) {
607 netdev->netdev_class->send_wait(netdev);
611 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
612 * otherwise a positive errno value. */
614 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
616 return netdev->netdev_class->set_etheraddr(netdev, mac);
619 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
620 * the MAC address into 'mac'. On failure, returns a positive errno value and
621 * clears 'mac' to all-zeros. */
623 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
625 return netdev->netdev_class->get_etheraddr(netdev, mac);
628 /* Returns the name of the network device that 'netdev' represents,
629 * e.g. "eth0". The caller must not modify or free the returned string. */
631 netdev_get_name(const struct netdev *netdev)
636 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
637 * (and received) packets, in bytes, not including the hardware header; thus,
638 * this is typically 1500 bytes for Ethernet devices.
640 * If successful, returns 0 and stores the MTU size in '*mtup'. Returns
641 * EOPNOTSUPP if 'netdev' does not have an MTU (as e.g. some tunnels do not).
642 * On other failure, returns a positive errno value. On failure, sets '*mtup'
645 netdev_get_mtu(const struct netdev *netdev, int *mtup)
647 const struct netdev_class *class = netdev->netdev_class;
650 error = class->get_mtu ? class->get_mtu(netdev, mtup) : EOPNOTSUPP;
653 if (error != EOPNOTSUPP) {
654 VLOG_DBG_RL(&rl, "failed to retrieve MTU for network device %s: "
655 "%s", netdev_get_name(netdev), ovs_strerror(error));
661 /* Sets the MTU of 'netdev'. The MTU is the maximum size of transmitted
662 * (and received) packets, in bytes.
664 * If successful, returns 0. Returns EOPNOTSUPP if 'netdev' does not have an
665 * MTU (as e.g. some tunnels do not). On other failure, returns a positive
668 netdev_set_mtu(const struct netdev *netdev, int mtu)
670 const struct netdev_class *class = netdev->netdev_class;
673 error = class->set_mtu ? class->set_mtu(netdev, mtu) : EOPNOTSUPP;
674 if (error && error != EOPNOTSUPP) {
675 VLOG_DBG_RL(&rl, "failed to set MTU for network device %s: %s",
676 netdev_get_name(netdev), ovs_strerror(error));
682 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
683 * failure, returns a negative errno value.
685 * The desired semantics of the ifindex value are a combination of those
686 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
687 * value should be unique within a host and remain stable at least until
688 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
689 * but many systems do not follow this rule anyhow.
691 * Some network devices may not implement support for this function. In such
692 * cases this function will always return -EOPNOTSUPP.
695 netdev_get_ifindex(const struct netdev *netdev)
697 int (*get_ifindex)(const struct netdev *);
699 get_ifindex = netdev->netdev_class->get_ifindex;
701 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
704 /* Stores the features supported by 'netdev' into each of '*current',
705 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
706 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
707 * successful, otherwise a positive errno value. On failure, all of the
708 * passed-in values are set to 0.
710 * Some network devices may not implement support for this function. In such
711 * cases this function will always return EOPNOTSUPP. */
713 netdev_get_features(const struct netdev *netdev,
714 enum netdev_features *current,
715 enum netdev_features *advertised,
716 enum netdev_features *supported,
717 enum netdev_features *peer)
719 int (*get_features)(const struct netdev *netdev,
720 enum netdev_features *current,
721 enum netdev_features *advertised,
722 enum netdev_features *supported,
723 enum netdev_features *peer);
724 enum netdev_features dummy[4];
731 advertised = &dummy[1];
734 supported = &dummy[2];
740 get_features = netdev->netdev_class->get_features;
742 ? get_features(netdev, current, advertised, supported,
746 *current = *advertised = *supported = *peer = 0;
751 /* Returns the maximum speed of a network connection that has the NETDEV_F_*
752 * bits in 'features', in bits per second. If no bits that indicate a speed
753 * are set in 'features', returns 'default_bps'. */
755 netdev_features_to_bps(enum netdev_features features,
756 uint64_t default_bps)
759 F_1000000MB = NETDEV_F_1TB_FD,
760 F_100000MB = NETDEV_F_100GB_FD,
761 F_40000MB = NETDEV_F_40GB_FD,
762 F_10000MB = NETDEV_F_10GB_FD,
763 F_1000MB = NETDEV_F_1GB_HD | NETDEV_F_1GB_FD,
764 F_100MB = NETDEV_F_100MB_HD | NETDEV_F_100MB_FD,
765 F_10MB = NETDEV_F_10MB_HD | NETDEV_F_10MB_FD
768 return ( features & F_1000000MB ? UINT64_C(1000000000000)
769 : features & F_100000MB ? UINT64_C(100000000000)
770 : features & F_40000MB ? UINT64_C(40000000000)
771 : features & F_10000MB ? UINT64_C(10000000000)
772 : features & F_1000MB ? UINT64_C(1000000000)
773 : features & F_100MB ? UINT64_C(100000000)
774 : features & F_10MB ? UINT64_C(10000000)
778 /* Returns true if any of the NETDEV_F_* bits that indicate a full-duplex link
779 * are set in 'features', otherwise false. */
781 netdev_features_is_full_duplex(enum netdev_features features)
783 return (features & (NETDEV_F_10MB_FD | NETDEV_F_100MB_FD | NETDEV_F_1GB_FD
784 | NETDEV_F_10GB_FD | NETDEV_F_40GB_FD
785 | NETDEV_F_100GB_FD | NETDEV_F_1TB_FD)) != 0;
788 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
789 * successful, otherwise a positive errno value. */
791 netdev_set_advertisements(struct netdev *netdev,
792 enum netdev_features advertise)
794 return (netdev->netdev_class->set_advertisements
795 ? netdev->netdev_class->set_advertisements(
800 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
801 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
802 * errno value and sets '*address' to 0 (INADDR_ANY).
804 * The following error values have well-defined meanings:
806 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
808 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
810 * 'address' or 'netmask' or both may be null, in which case the address or
811 * netmask is not reported. */
813 netdev_get_in4(const struct netdev *netdev,
814 struct in_addr *address_, struct in_addr *netmask_)
816 struct in_addr address;
817 struct in_addr netmask;
820 error = (netdev->netdev_class->get_in4
821 ? netdev->netdev_class->get_in4(netdev,
825 address_->s_addr = error ? 0 : address.s_addr;
828 netmask_->s_addr = error ? 0 : netmask.s_addr;
833 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
834 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
835 * positive errno value. */
837 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
839 return (netdev->netdev_class->set_in4
840 ? netdev->netdev_class->set_in4(netdev, addr, mask)
844 /* Obtains ad IPv4 address from device name and save the address in
845 * in4. Returns 0 if successful, otherwise a positive errno value.
848 netdev_get_in4_by_name(const char *device_name, struct in_addr *in4)
850 struct netdev *netdev;
853 error = netdev_open(device_name, "system", &netdev);
855 in4->s_addr = htonl(0);
859 error = netdev_get_in4(netdev, in4, NULL);
860 netdev_close(netdev);
864 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
867 netdev_add_router(struct netdev *netdev, struct in_addr router)
869 COVERAGE_INC(netdev_add_router);
870 return (netdev->netdev_class->add_router
871 ? netdev->netdev_class->add_router(netdev, router)
875 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
876 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
877 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
878 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
879 * a directly connected network) in '*next_hop' and a copy of the name of the
880 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
881 * responsible for freeing '*netdev_name' (by calling free()). */
883 netdev_get_next_hop(const struct netdev *netdev,
884 const struct in_addr *host, struct in_addr *next_hop,
887 int error = (netdev->netdev_class->get_next_hop
888 ? netdev->netdev_class->get_next_hop(
889 host, next_hop, netdev_name)
892 next_hop->s_addr = 0;
898 /* Populates 'smap' with status information.
900 * Populates 'smap' with 'netdev' specific status information. This
901 * information may be used to populate the status column of the Interface table
902 * as defined in ovs-vswitchd.conf.db(5). */
904 netdev_get_status(const struct netdev *netdev, struct smap *smap)
906 return (netdev->netdev_class->get_status
907 ? netdev->netdev_class->get_status(netdev, smap)
911 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
912 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
913 * all-zero-bits (in6addr_any).
915 * The following error values have well-defined meanings:
917 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
919 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
921 * 'in6' may be null, in which case the address itself is not reported. */
923 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
925 struct in6_addr dummy;
928 error = (netdev->netdev_class->get_in6
929 ? netdev->netdev_class->get_in6(netdev,
933 memset(in6, 0, sizeof *in6);
938 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
939 * 'on'. Returns 0 if successful, otherwise a positive errno value. */
941 do_update_flags(struct netdev *netdev, enum netdev_flags off,
942 enum netdev_flags on, enum netdev_flags *old_flagsp,
943 struct netdev_saved_flags **sfp)
944 OVS_EXCLUDED(netdev_mutex)
946 struct netdev_saved_flags *sf = NULL;
947 enum netdev_flags old_flags;
950 error = netdev->netdev_class->update_flags(netdev, off & ~on, on,
953 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
954 off || on ? "set" : "get", netdev_get_name(netdev),
955 ovs_strerror(error));
957 } else if ((off || on) && sfp) {
958 enum netdev_flags new_flags = (old_flags & ~off) | on;
959 enum netdev_flags changed_flags = old_flags ^ new_flags;
961 ovs_mutex_lock(&netdev_mutex);
962 *sfp = sf = xmalloc(sizeof *sf);
964 list_push_front(&netdev->saved_flags_list, &sf->node);
965 sf->saved_flags = changed_flags;
966 sf->saved_values = changed_flags & new_flags;
969 ovs_mutex_unlock(&netdev_mutex);
974 *old_flagsp = old_flags;
983 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
984 * Returns 0 if successful, otherwise a positive errno value. On failure,
985 * stores 0 into '*flagsp'. */
987 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
989 struct netdev *netdev = CONST_CAST(struct netdev *, netdev_);
990 return do_update_flags(netdev, 0, 0, flagsp, NULL);
993 /* Sets the flags for 'netdev' to 'flags'.
994 * Returns 0 if successful, otherwise a positive errno value. */
996 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
997 struct netdev_saved_flags **sfp)
999 return do_update_flags(netdev, -1, flags, NULL, sfp);
1002 /* Turns on the specified 'flags' on 'netdev':
1004 * - On success, returns 0. If 'sfp' is nonnull, sets '*sfp' to a newly
1005 * allocated 'struct netdev_saved_flags *' that may be passed to
1006 * netdev_restore_flags() to restore the original values of 'flags' on
1007 * 'netdev' (this will happen automatically at program termination if
1008 * netdev_restore_flags() is never called) , or to NULL if no flags were
1011 * - On failure, returns a positive errno value. If 'sfp' is nonnull, sets
1012 * '*sfp' to NULL. */
1014 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
1015 struct netdev_saved_flags **sfp)
1017 return do_update_flags(netdev, 0, flags, NULL, sfp);
1020 /* Turns off the specified 'flags' on 'netdev'. See netdev_turn_flags_on() for
1021 * details of the interface. */
1023 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
1024 struct netdev_saved_flags **sfp)
1026 return do_update_flags(netdev, flags, 0, NULL, sfp);
1029 /* Restores the flags that were saved in 'sf', and destroys 'sf'.
1030 * Does nothing if 'sf' is NULL. */
1032 netdev_restore_flags(struct netdev_saved_flags *sf)
1033 OVS_EXCLUDED(netdev_mutex)
1036 struct netdev *netdev = sf->netdev;
1037 enum netdev_flags old_flags;
1039 netdev->netdev_class->update_flags(netdev,
1040 sf->saved_flags & sf->saved_values,
1041 sf->saved_flags & ~sf->saved_values,
1044 ovs_mutex_lock(&netdev_mutex);
1045 list_remove(&sf->node);
1047 netdev_unref(netdev);
1051 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1052 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1053 * returns 0. Otherwise, it returns a positive errno value; in particular,
1054 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
1056 netdev_arp_lookup(const struct netdev *netdev,
1057 ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
1059 int error = (netdev->netdev_class->arp_lookup
1060 ? netdev->netdev_class->arp_lookup(netdev, ip, mac)
1063 memset(mac, 0, ETH_ADDR_LEN);
1068 /* Returns true if carrier is active (link light is on) on 'netdev'. */
1070 netdev_get_carrier(const struct netdev *netdev)
1073 enum netdev_flags flags;
1076 netdev_get_flags(netdev, &flags);
1077 if (!(flags & NETDEV_UP)) {
1081 if (!netdev->netdev_class->get_carrier) {
1085 error = netdev->netdev_class->get_carrier(netdev, &carrier);
1087 VLOG_DBG("%s: failed to get network device carrier status, assuming "
1088 "down: %s", netdev_get_name(netdev), ovs_strerror(error));
1095 /* Returns the number of times 'netdev''s carrier has changed. */
1097 netdev_get_carrier_resets(const struct netdev *netdev)
1099 return (netdev->netdev_class->get_carrier_resets
1100 ? netdev->netdev_class->get_carrier_resets(netdev)
1104 /* Attempts to force netdev_get_carrier() to poll 'netdev''s MII registers for
1105 * link status instead of checking 'netdev''s carrier. 'netdev''s MII
1106 * registers will be polled once ever 'interval' milliseconds. If 'netdev'
1107 * does not support MII, another method may be used as a fallback. If
1108 * 'interval' is less than or equal to zero, reverts netdev_get_carrier() to
1109 * its normal behavior.
1111 * Returns 0 if successful, otherwise a positive errno value. */
1113 netdev_set_miimon_interval(struct netdev *netdev, long long int interval)
1115 return (netdev->netdev_class->set_miimon_interval
1116 ? netdev->netdev_class->set_miimon_interval(netdev, interval)
1120 /* Retrieves current device stats for 'netdev'. */
1122 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
1126 COVERAGE_INC(netdev_get_stats);
1127 error = (netdev->netdev_class->get_stats
1128 ? netdev->netdev_class->get_stats(netdev, stats)
1131 memset(stats, 0xff, sizeof *stats);
1136 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
1137 * Returns 0 if successful, otherwise a positive errno value.
1139 * This will probably fail for most network devices. Some devices might only
1140 * allow setting their stats to 0. */
1142 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
1144 return (netdev->netdev_class->set_stats
1145 ? netdev->netdev_class->set_stats(netdev, stats)
1149 /* Attempts to set input rate limiting (policing) policy, such that up to
1150 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
1151 * size of 'kbits' kb. */
1153 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
1154 uint32_t kbits_burst)
1156 return (netdev->netdev_class->set_policing
1157 ? netdev->netdev_class->set_policing(netdev,
1158 kbits_rate, kbits_burst)
1162 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
1163 * empty if 'netdev' does not support QoS. Any names added to 'types' should
1164 * be documented as valid for the "type" column in the "QoS" table in
1165 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1167 * Every network device supports disabling QoS with a type of "", but this type
1168 * will not be added to 'types'.
1170 * The caller must initialize 'types' (e.g. with sset_init()) before calling
1171 * this function. The caller is responsible for destroying 'types' (e.g. with
1172 * sset_destroy()) when it is no longer needed.
1174 * Returns 0 if successful, otherwise a positive errno value. */
1176 netdev_get_qos_types(const struct netdev *netdev, struct sset *types)
1178 const struct netdev_class *class = netdev->netdev_class;
1179 return (class->get_qos_types
1180 ? class->get_qos_types(netdev, types)
1184 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
1185 * which should be "" or one of the types returned by netdev_get_qos_types()
1186 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
1187 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
1188 * 'caps' to all zeros. */
1190 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
1191 struct netdev_qos_capabilities *caps)
1193 const struct netdev_class *class = netdev->netdev_class;
1196 int retval = (class->get_qos_capabilities
1197 ? class->get_qos_capabilities(netdev, type, caps)
1200 memset(caps, 0, sizeof *caps);
1204 /* Every netdev supports turning off QoS. */
1205 memset(caps, 0, sizeof *caps);
1210 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1211 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1212 * the number of queues (zero on failure) in '*n_queuesp'.
1214 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1216 netdev_get_n_queues(const struct netdev *netdev,
1217 const char *type, unsigned int *n_queuesp)
1219 struct netdev_qos_capabilities caps;
1222 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1223 *n_queuesp = caps.n_queues;
1227 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1228 * stores the name of the current form of QoS into '*typep', stores any details
1229 * of configuration as string key-value pairs in 'details', and returns 0. On
1230 * failure, sets '*typep' to NULL and returns a positive errno value.
1232 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1234 * The caller must initialize 'details' as an empty smap (e.g. with
1235 * smap_init()) before calling this function. The caller must free 'details'
1236 * when it is no longer needed (e.g. with smap_destroy()).
1238 * The caller must not modify or free '*typep'.
1240 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1241 * 'netdev'. The contents of 'details' should be documented as valid for
1242 * '*typep' in the "other_config" column in the "QoS" table in
1243 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1245 netdev_get_qos(const struct netdev *netdev,
1246 const char **typep, struct smap *details)
1248 const struct netdev_class *class = netdev->netdev_class;
1251 if (class->get_qos) {
1252 retval = class->get_qos(netdev, typep, details);
1255 smap_clear(details);
1259 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1265 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1266 * with details of configuration from 'details'. Returns 0 if successful,
1267 * otherwise a positive errno value. On error, the previous QoS configuration
1270 * When this function changes the type of QoS (not just 'details'), this also
1271 * resets all queue configuration for 'netdev' to their defaults (which depend
1272 * on the specific type of QoS). Otherwise, the queue configuration for
1273 * 'netdev' is unchanged.
1275 * 'type' should be "" (to disable QoS) or one of the types returned by
1276 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1277 * documented as valid for the given 'type' in the "other_config" column in the
1278 * "QoS" table in vswitchd/vswitch.xml (which is built as
1279 * ovs-vswitchd.conf.db(8)).
1281 * NULL may be specified for 'details' if there are no configuration
1284 netdev_set_qos(struct netdev *netdev,
1285 const char *type, const struct smap *details)
1287 const struct netdev_class *class = netdev->netdev_class;
1293 if (class->set_qos) {
1295 static const struct smap empty = SMAP_INITIALIZER(&empty);
1298 return class->set_qos(netdev, type, details);
1300 return *type ? EOPNOTSUPP : 0;
1304 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1305 * successful, adds that information as string key-value pairs to 'details'.
1306 * Returns 0 if successful, otherwise a positive errno value.
1308 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1309 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1311 * The returned contents of 'details' should be documented as valid for the
1312 * given 'type' in the "other_config" column in the "Queue" table in
1313 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1315 * The caller must initialize 'details' (e.g. with smap_init()) before calling
1316 * this function. The caller must free 'details' when it is no longer needed
1317 * (e.g. with smap_destroy()). */
1319 netdev_get_queue(const struct netdev *netdev,
1320 unsigned int queue_id, struct smap *details)
1322 const struct netdev_class *class = netdev->netdev_class;
1325 retval = (class->get_queue
1326 ? class->get_queue(netdev, queue_id, details)
1329 smap_clear(details);
1334 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1335 * string pairs in 'details'. The contents of 'details' should be documented
1336 * as valid for the given 'type' in the "other_config" column in the "Queue"
1337 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1338 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1339 * given queue's configuration should be unmodified.
1341 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1342 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1344 * This function does not modify 'details', and the caller retains ownership of
1347 netdev_set_queue(struct netdev *netdev,
1348 unsigned int queue_id, const struct smap *details)
1350 const struct netdev_class *class = netdev->netdev_class;
1351 return (class->set_queue
1352 ? class->set_queue(netdev, queue_id, details)
1356 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1357 * of QoS may have a fixed set of queues, in which case attempts to delete them
1358 * will fail with EOPNOTSUPP.
1360 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1361 * given queue will be unmodified.
1363 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1364 * the current form of QoS (e.g. as returned by
1365 * netdev_get_n_queues(netdev)). */
1367 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1369 const struct netdev_class *class = netdev->netdev_class;
1370 return (class->delete_queue
1371 ? class->delete_queue(netdev, queue_id)
1375 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1376 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1377 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1378 * positive errno value and fills 'stats' with values indicating unsupported
1381 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1382 struct netdev_queue_stats *stats)
1384 const struct netdev_class *class = netdev->netdev_class;
1387 retval = (class->get_queue_stats
1388 ? class->get_queue_stats(netdev, queue_id, stats)
1391 stats->tx_bytes = UINT64_MAX;
1392 stats->tx_packets = UINT64_MAX;
1393 stats->tx_errors = UINT64_MAX;
1394 stats->created = LLONG_MIN;
1399 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1400 * its configuration, and the 'aux' specified by the caller. The order of
1401 * iteration is unspecified, but (when successful) each queue is visited
1404 * Calling this function may be more efficient than calling netdev_get_queue()
1407 * 'cb' must not modify or free the 'details' argument passed in. It may
1408 * delete or modify the queue passed in as its 'queue_id' argument. It may
1409 * modify but must not delete any other queue within 'netdev'. 'cb' should not
1410 * add new queues because this may cause some queues to be visited twice or not
1413 * Returns 0 if successful, otherwise a positive errno value. On error, some
1414 * configured queues may not have been included in the iteration. */
1416 netdev_dump_queues(const struct netdev *netdev,
1417 netdev_dump_queues_cb *cb, void *aux)
1419 const struct netdev_class *class = netdev->netdev_class;
1420 return (class->dump_queues
1421 ? class->dump_queues(netdev, cb, aux)
1425 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1426 * its statistics, and the 'aux' specified by the caller. The order of
1427 * iteration is unspecified, but (when successful) each queue is visited
1430 * Calling this function may be more efficient than calling
1431 * netdev_get_queue_stats() for every queue.
1433 * 'cb' must not modify or free the statistics passed in.
1435 * Returns 0 if successful, otherwise a positive errno value. On error, some
1436 * configured queues may not have been included in the iteration. */
1438 netdev_dump_queue_stats(const struct netdev *netdev,
1439 netdev_dump_queue_stats_cb *cb, void *aux)
1441 const struct netdev_class *class = netdev->netdev_class;
1442 return (class->dump_queue_stats
1443 ? class->dump_queue_stats(netdev, cb, aux)
1447 /* Returns a sequence number which indicates changes in one of 'netdev''s
1448 * properties. The returned sequence will be nonzero so that callers have a
1449 * value which they may use as a reset when tracking 'netdev'.
1451 * The returned sequence number will change whenever 'netdev''s flags,
1452 * features, ethernet address, or carrier changes. It may change for other
1453 * reasons as well, or no reason at all. */
1455 netdev_change_seq(const struct netdev *netdev)
1457 return netdev->netdev_class->change_seq(netdev);
1460 /* Returns the class type of 'netdev'.
1462 * The caller must not free the returned value. */
1464 netdev_get_type(const struct netdev *netdev)
1466 return netdev->netdev_class->type;
1469 /* Returns the class associated with 'netdev'. */
1470 const struct netdev_class *
1471 netdev_get_class(const struct netdev *netdev)
1473 return netdev->netdev_class;
1476 /* Returns the netdev with 'name' or NULL if there is none.
1478 * The caller must free the returned netdev with netdev_close(). */
1480 netdev_from_name(const char *name)
1481 OVS_EXCLUDED(netdev_mutex)
1483 struct netdev *netdev;
1485 ovs_mutex_lock(&netdev_mutex);
1486 netdev = shash_find_data(&netdev_shash, name);
1490 ovs_mutex_unlock(&netdev_mutex);
1495 /* Fills 'device_list' with devices that match 'netdev_class'.
1497 * The caller is responsible for initializing and destroying 'device_list' and
1498 * must close each device on the list. */
1500 netdev_get_devices(const struct netdev_class *netdev_class,
1501 struct shash *device_list)
1502 OVS_EXCLUDED(netdev_mutex)
1504 struct shash_node *node;
1506 ovs_mutex_lock(&netdev_mutex);
1507 SHASH_FOR_EACH (node, &netdev_shash) {
1508 struct netdev *dev = node->data;
1510 if (dev->netdev_class == netdev_class) {
1512 shash_add(device_list, node->name, node->data);
1515 ovs_mutex_unlock(&netdev_mutex);
1519 netdev_get_type_from_name(const char *name)
1521 struct netdev *dev = netdev_from_name(name);
1522 const char *type = dev ? netdev_get_type(dev) : NULL;
1528 netdev_rx_get_netdev(const struct netdev_rx *rx)
1530 ovs_assert(rx->netdev->ref_cnt > 0);
1535 netdev_rx_get_name(const struct netdev_rx *rx)
1537 return netdev_get_name(netdev_rx_get_netdev(rx));
1541 restore_all_flags(void *aux OVS_UNUSED)
1543 struct shash_node *node;
1545 SHASH_FOR_EACH (node, &netdev_shash) {
1546 struct netdev *netdev = node->data;
1547 const struct netdev_saved_flags *sf;
1548 enum netdev_flags saved_values;
1549 enum netdev_flags saved_flags;
1551 saved_values = saved_flags = 0;
1552 LIST_FOR_EACH (sf, node, &netdev->saved_flags_list) {
1553 saved_flags |= sf->saved_flags;
1554 saved_values &= ~sf->saved_flags;
1555 saved_values |= sf->saved_flags & sf->saved_values;
1558 enum netdev_flags old_flags;
1560 netdev->netdev_class->update_flags(netdev,
1561 saved_flags & saved_values,
1562 saved_flags & ~saved_values,