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) {
131 if (rc->class->run) {
135 ovs_rwlock_unlock(&netdev_class_rwlock);
138 /* Arranges for poll_block() to wake up when netdev_run() needs to be called.
140 * If your program opens any netdevs, it must call this function within its
144 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
146 struct netdev_registered_class *rc;
148 ovs_rwlock_rdlock(&netdev_class_rwlock);
149 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
150 if (rc->class->wait) {
154 ovs_rwlock_unlock(&netdev_class_rwlock);
157 static struct netdev_registered_class *
158 netdev_lookup_class(const char *type)
159 OVS_REQ_RDLOCK(netdev_class_rwlock)
161 struct netdev_registered_class *rc;
163 HMAP_FOR_EACH_WITH_HASH (rc, hmap_node, hash_string(type, 0),
165 if (!strcmp(type, rc->class->type)) {
172 /* Initializes and registers a new netdev provider. After successful
173 * registration, new netdevs of that type can be opened using netdev_open(). */
175 netdev_register_provider(const struct netdev_class *new_class)
176 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
180 ovs_rwlock_wrlock(&netdev_class_rwlock);
181 if (netdev_lookup_class(new_class->type)) {
182 VLOG_WARN("attempted to register duplicate netdev provider: %s",
186 error = new_class->init ? new_class->init() : 0;
188 struct netdev_registered_class *rc;
190 rc = xmalloc(sizeof *rc);
191 hmap_insert(&netdev_classes, &rc->hmap_node,
192 hash_string(new_class->type, 0));
193 rc->class = new_class;
194 atomic_init(&rc->ref_cnt, 0);
196 VLOG_ERR("failed to initialize %s network device class: %s",
197 new_class->type, ovs_strerror(error));
200 ovs_rwlock_unlock(&netdev_class_rwlock);
205 /* Unregisters a netdev provider. 'type' must have been previously
206 * registered and not currently be in use by any netdevs. After unregistration
207 * new netdevs of that type cannot be opened using netdev_open(). */
209 netdev_unregister_provider(const char *type)
210 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
212 struct netdev_registered_class *rc;
215 ovs_rwlock_wrlock(&netdev_class_rwlock);
216 rc = netdev_lookup_class(type);
218 VLOG_WARN("attempted to unregister a netdev provider that is not "
219 "registered: %s", type);
220 error = EAFNOSUPPORT;
224 atomic_read(&rc->ref_cnt, &ref_cnt);
226 hmap_remove(&netdev_classes, &rc->hmap_node);
230 VLOG_WARN("attempted to unregister in use netdev provider: %s",
235 ovs_rwlock_unlock(&netdev_class_rwlock);
240 /* Clears 'types' and enumerates the types of all currently registered netdev
241 * providers into it. The caller must first initialize the sset. */
243 netdev_enumerate_types(struct sset *types)
244 OVS_EXCLUDED(netdev_mutex)
246 struct netdev_registered_class *rc;
251 ovs_rwlock_rdlock(&netdev_class_rwlock);
252 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
253 sset_add(types, rc->class->type);
255 ovs_rwlock_unlock(&netdev_class_rwlock);
258 /* Check that the network device name is not the same as any of the registered
259 * vport providers' dpif_port name (dpif_port is NULL if the vport provider
260 * does not define it) or the datapath internal port name (e.g. ovs-system).
262 * Returns true if there is a name conflict, false otherwise. */
264 netdev_is_reserved_name(const char *name)
265 OVS_EXCLUDED(netdev_mutex)
267 struct netdev_registered_class *rc;
271 ovs_rwlock_rdlock(&netdev_class_rwlock);
272 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
273 const char *dpif_port = netdev_vport_class_get_dpif_port(rc->class);
274 if (dpif_port && !strcmp(dpif_port, name)) {
275 ovs_rwlock_unlock(&netdev_class_rwlock);
279 ovs_rwlock_unlock(&netdev_class_rwlock);
281 if (!strncmp(name, "ovs-", 4)) {
286 dp_enumerate_types(&types);
287 SSET_FOR_EACH (type, &types) {
288 if (!strcmp(name+4, type)) {
289 sset_destroy(&types);
293 sset_destroy(&types);
299 /* Opens the network device named 'name' (e.g. "eth0") of the specified 'type'
300 * (e.g. "system") and returns zero if successful, otherwise a positive errno
301 * value. On success, sets '*netdevp' to the new network device, otherwise to
304 * Some network devices may need to be configured (with netdev_set_config())
305 * before they can be used. */
307 netdev_open(const char *name, const char *type, struct netdev **netdevp)
308 OVS_EXCLUDED(netdev_mutex)
310 struct netdev *netdev;
315 ovs_rwlock_rdlock(&netdev_class_rwlock);
316 ovs_mutex_lock(&netdev_mutex);
317 netdev = shash_find_data(&netdev_shash, name);
319 struct netdev_registered_class *rc;
321 rc = netdev_lookup_class(type && type[0] ? type : "system");
323 netdev = rc->class->alloc();
325 memset(netdev, 0, sizeof *netdev);
326 netdev->netdev_class = rc->class;
327 netdev->name = xstrdup(name);
328 netdev->node = shash_add(&netdev_shash, name, netdev);
329 list_init(&netdev->saved_flags_list);
331 error = rc->class->construct(netdev);
335 atomic_add(&rc->ref_cnt, 1, &old_ref_cnt);
338 ovs_assert(list_is_empty(&netdev->saved_flags_list));
339 shash_delete(&netdev_shash, netdev->node);
340 rc->class->dealloc(netdev);
346 VLOG_WARN("could not create netdev %s of unknown type %s",
348 error = EAFNOSUPPORT;
354 ovs_mutex_unlock(&netdev_mutex);
355 ovs_rwlock_unlock(&netdev_class_rwlock);
366 /* Returns a reference to 'netdev_' for the caller to own. Returns null if
367 * 'netdev_' is null. */
369 netdev_ref(const struct netdev *netdev_)
370 OVS_EXCLUDED(netdev_mutex)
372 struct netdev *netdev = CONST_CAST(struct netdev *, netdev_);
375 ovs_mutex_lock(&netdev_mutex);
376 ovs_assert(netdev->ref_cnt > 0);
378 ovs_mutex_unlock(&netdev_mutex);
383 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
384 * or NULL if none are needed. */
386 netdev_set_config(struct netdev *netdev, const struct smap *args)
387 OVS_EXCLUDED(netdev_mutex)
389 if (netdev->netdev_class->set_config) {
390 const struct smap no_args = SMAP_INITIALIZER(&no_args);
391 return netdev->netdev_class->set_config(netdev,
392 args ? args : &no_args);
393 } else if (args && !smap_is_empty(args)) {
394 VLOG_WARN("%s: arguments provided to device that is not configurable",
395 netdev_get_name(netdev));
401 /* Returns the current configuration for 'netdev' in 'args'. The caller must
402 * have already initialized 'args' with smap_init(). Returns 0 on success, in
403 * which case 'args' will be filled with 'netdev''s configuration. On failure
404 * returns a positive errno value, in which case 'args' will be empty.
406 * The caller owns 'args' and its contents and must eventually free them with
409 netdev_get_config(const struct netdev *netdev, struct smap *args)
410 OVS_EXCLUDED(netdev_mutex)
415 if (netdev->netdev_class->get_config) {
416 error = netdev->netdev_class->get_config(netdev, args);
427 const struct netdev_tunnel_config *
428 netdev_get_tunnel_config(const struct netdev *netdev)
429 OVS_EXCLUDED(netdev_mutex)
431 if (netdev->netdev_class->get_tunnel_config) {
432 return netdev->netdev_class->get_tunnel_config(netdev);
439 netdev_unref(struct netdev *dev)
440 OVS_RELEASES(netdev_mutex)
442 ovs_assert(dev->ref_cnt);
443 if (!--dev->ref_cnt) {
444 const struct netdev_class *class = dev->netdev_class;
445 struct netdev_registered_class *rc;
448 dev->netdev_class->destruct(dev);
450 shash_delete(&netdev_shash, dev->node);
452 dev->netdev_class->dealloc(dev);
453 ovs_mutex_unlock(&netdev_mutex);
455 ovs_rwlock_rdlock(&netdev_class_rwlock);
456 rc = netdev_lookup_class(class->type);
457 atomic_sub(&rc->ref_cnt, 1, &old_ref_cnt);
458 ovs_assert(old_ref_cnt > 0);
459 ovs_rwlock_unlock(&netdev_class_rwlock);
461 ovs_mutex_unlock(&netdev_mutex);
465 /* Closes and destroys 'netdev'. */
467 netdev_close(struct netdev *netdev)
468 OVS_EXCLUDED(netdev_mutex)
471 ovs_mutex_lock(&netdev_mutex);
472 netdev_unref(netdev);
476 /* Parses 'netdev_name_', which is of the form [type@]name into its component
477 * pieces. 'name' and 'type' must be freed by the caller. */
479 netdev_parse_name(const char *netdev_name_, char **name, char **type)
481 char *netdev_name = xstrdup(netdev_name_);
484 separator = strchr(netdev_name, '@');
488 *name = xstrdup(separator + 1);
491 *type = xstrdup("system");
496 netdev_rx_open(struct netdev *netdev, struct netdev_rx **rxp)
497 OVS_EXCLUDED(netdev_mutex)
501 if (netdev->netdev_class->rx_alloc) {
502 struct netdev_rx *rx = netdev->netdev_class->rx_alloc();
505 error = netdev->netdev_class->rx_construct(rx);
507 ovs_mutex_lock(&netdev_mutex);
509 ovs_mutex_unlock(&netdev_mutex);
514 netdev->netdev_class->rx_dealloc(rx);
527 netdev_rx_close(struct netdev_rx *rx)
528 OVS_EXCLUDED(netdev_mutex)
531 struct netdev *netdev = rx->netdev;
532 netdev->netdev_class->rx_destruct(rx);
533 netdev->netdev_class->rx_dealloc(rx);
534 netdev_close(netdev);
539 netdev_rx_recv(struct netdev_rx *rx, struct ofpbuf *buffer)
543 ovs_assert(buffer->size == 0);
544 ovs_assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
546 retval = rx->netdev->netdev_class->rx_recv(rx, buffer->data,
547 ofpbuf_tailroom(buffer));
549 COVERAGE_INC(netdev_received);
550 buffer->size += retval;
551 if (buffer->size < ETH_TOTAL_MIN) {
552 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
561 netdev_rx_wait(struct netdev_rx *rx)
563 rx->netdev->netdev_class->rx_wait(rx);
567 netdev_rx_drain(struct netdev_rx *rx)
569 return (rx->netdev->netdev_class->rx_drain
570 ? rx->netdev->netdev_class->rx_drain(rx)
574 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
575 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
576 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
577 * the packet is too big or too small to transmit on the device.
579 * The caller retains ownership of 'buffer' in all cases.
581 * The kernel maintains a packet transmission queue, so the caller is not
582 * expected to do additional queuing of packets.
584 * Some network devices may not implement support for this function. In such
585 * cases this function will always return EOPNOTSUPP. */
587 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
591 error = (netdev->netdev_class->send
592 ? netdev->netdev_class->send(netdev, buffer->data, buffer->size)
595 COVERAGE_INC(netdev_sent);
600 /* Registers with the poll loop to wake up from the next call to poll_block()
601 * when the packet transmission queue has sufficient room to transmit a packet
602 * with netdev_send().
604 * The kernel maintains a packet transmission queue, so the client is not
605 * expected to do additional queuing of packets. Thus, this function is
606 * unlikely to ever be used. It is included for completeness. */
608 netdev_send_wait(struct netdev *netdev)
610 if (netdev->netdev_class->send_wait) {
611 netdev->netdev_class->send_wait(netdev);
615 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
616 * otherwise a positive errno value. */
618 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
620 return netdev->netdev_class->set_etheraddr(netdev, mac);
623 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
624 * the MAC address into 'mac'. On failure, returns a positive errno value and
625 * clears 'mac' to all-zeros. */
627 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
629 return netdev->netdev_class->get_etheraddr(netdev, mac);
632 /* Returns the name of the network device that 'netdev' represents,
633 * e.g. "eth0". The caller must not modify or free the returned string. */
635 netdev_get_name(const struct netdev *netdev)
640 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
641 * (and received) packets, in bytes, not including the hardware header; thus,
642 * this is typically 1500 bytes for Ethernet devices.
644 * If successful, returns 0 and stores the MTU size in '*mtup'. Returns
645 * EOPNOTSUPP if 'netdev' does not have an MTU (as e.g. some tunnels do not).
646 * On other failure, returns a positive errno value. On failure, sets '*mtup'
649 netdev_get_mtu(const struct netdev *netdev, int *mtup)
651 const struct netdev_class *class = netdev->netdev_class;
654 error = class->get_mtu ? class->get_mtu(netdev, mtup) : EOPNOTSUPP;
657 if (error != EOPNOTSUPP) {
658 VLOG_DBG_RL(&rl, "failed to retrieve MTU for network device %s: "
659 "%s", netdev_get_name(netdev), ovs_strerror(error));
665 /* Sets the MTU of 'netdev'. The MTU is the maximum size of transmitted
666 * (and received) packets, in bytes.
668 * If successful, returns 0. Returns EOPNOTSUPP if 'netdev' does not have an
669 * MTU (as e.g. some tunnels do not). On other failure, returns a positive
672 netdev_set_mtu(const struct netdev *netdev, int mtu)
674 const struct netdev_class *class = netdev->netdev_class;
677 error = class->set_mtu ? class->set_mtu(netdev, mtu) : EOPNOTSUPP;
678 if (error && error != EOPNOTSUPP) {
679 VLOG_DBG_RL(&rl, "failed to set MTU for network device %s: %s",
680 netdev_get_name(netdev), ovs_strerror(error));
686 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
687 * failure, returns a negative errno value.
689 * The desired semantics of the ifindex value are a combination of those
690 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
691 * value should be unique within a host and remain stable at least until
692 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
693 * but many systems do not follow this rule anyhow.
695 * Some network devices may not implement support for this function. In such
696 * cases this function will always return -EOPNOTSUPP.
699 netdev_get_ifindex(const struct netdev *netdev)
701 int (*get_ifindex)(const struct netdev *);
703 get_ifindex = netdev->netdev_class->get_ifindex;
705 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
708 /* Stores the features supported by 'netdev' into each of '*current',
709 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
710 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
711 * successful, otherwise a positive errno value. On failure, all of the
712 * passed-in values are set to 0.
714 * Some network devices may not implement support for this function. In such
715 * cases this function will always return EOPNOTSUPP. */
717 netdev_get_features(const struct netdev *netdev,
718 enum netdev_features *current,
719 enum netdev_features *advertised,
720 enum netdev_features *supported,
721 enum netdev_features *peer)
723 int (*get_features)(const struct netdev *netdev,
724 enum netdev_features *current,
725 enum netdev_features *advertised,
726 enum netdev_features *supported,
727 enum netdev_features *peer);
728 enum netdev_features dummy[4];
735 advertised = &dummy[1];
738 supported = &dummy[2];
744 get_features = netdev->netdev_class->get_features;
746 ? get_features(netdev, current, advertised, supported,
750 *current = *advertised = *supported = *peer = 0;
755 /* Returns the maximum speed of a network connection that has the NETDEV_F_*
756 * bits in 'features', in bits per second. If no bits that indicate a speed
757 * are set in 'features', returns 'default_bps'. */
759 netdev_features_to_bps(enum netdev_features features,
760 uint64_t default_bps)
763 F_1000000MB = NETDEV_F_1TB_FD,
764 F_100000MB = NETDEV_F_100GB_FD,
765 F_40000MB = NETDEV_F_40GB_FD,
766 F_10000MB = NETDEV_F_10GB_FD,
767 F_1000MB = NETDEV_F_1GB_HD | NETDEV_F_1GB_FD,
768 F_100MB = NETDEV_F_100MB_HD | NETDEV_F_100MB_FD,
769 F_10MB = NETDEV_F_10MB_HD | NETDEV_F_10MB_FD
772 return ( features & F_1000000MB ? UINT64_C(1000000000000)
773 : features & F_100000MB ? UINT64_C(100000000000)
774 : features & F_40000MB ? UINT64_C(40000000000)
775 : features & F_10000MB ? UINT64_C(10000000000)
776 : features & F_1000MB ? UINT64_C(1000000000)
777 : features & F_100MB ? UINT64_C(100000000)
778 : features & F_10MB ? UINT64_C(10000000)
782 /* Returns true if any of the NETDEV_F_* bits that indicate a full-duplex link
783 * are set in 'features', otherwise false. */
785 netdev_features_is_full_duplex(enum netdev_features features)
787 return (features & (NETDEV_F_10MB_FD | NETDEV_F_100MB_FD | NETDEV_F_1GB_FD
788 | NETDEV_F_10GB_FD | NETDEV_F_40GB_FD
789 | NETDEV_F_100GB_FD | NETDEV_F_1TB_FD)) != 0;
792 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
793 * successful, otherwise a positive errno value. */
795 netdev_set_advertisements(struct netdev *netdev,
796 enum netdev_features advertise)
798 return (netdev->netdev_class->set_advertisements
799 ? netdev->netdev_class->set_advertisements(
804 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
805 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
806 * errno value and sets '*address' to 0 (INADDR_ANY).
808 * The following error values have well-defined meanings:
810 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
812 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
814 * 'address' or 'netmask' or both may be null, in which case the address or
815 * netmask is not reported. */
817 netdev_get_in4(const struct netdev *netdev,
818 struct in_addr *address_, struct in_addr *netmask_)
820 struct in_addr address;
821 struct in_addr netmask;
824 error = (netdev->netdev_class->get_in4
825 ? netdev->netdev_class->get_in4(netdev,
829 address_->s_addr = error ? 0 : address.s_addr;
832 netmask_->s_addr = error ? 0 : netmask.s_addr;
837 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
838 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
839 * positive errno value. */
841 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
843 return (netdev->netdev_class->set_in4
844 ? netdev->netdev_class->set_in4(netdev, addr, mask)
848 /* Obtains ad IPv4 address from device name and save the address in
849 * in4. Returns 0 if successful, otherwise a positive errno value.
852 netdev_get_in4_by_name(const char *device_name, struct in_addr *in4)
854 struct netdev *netdev;
857 error = netdev_open(device_name, "system", &netdev);
859 in4->s_addr = htonl(0);
863 error = netdev_get_in4(netdev, in4, NULL);
864 netdev_close(netdev);
868 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
871 netdev_add_router(struct netdev *netdev, struct in_addr router)
873 COVERAGE_INC(netdev_add_router);
874 return (netdev->netdev_class->add_router
875 ? netdev->netdev_class->add_router(netdev, router)
879 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
880 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
881 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
882 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
883 * a directly connected network) in '*next_hop' and a copy of the name of the
884 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
885 * responsible for freeing '*netdev_name' (by calling free()). */
887 netdev_get_next_hop(const struct netdev *netdev,
888 const struct in_addr *host, struct in_addr *next_hop,
891 int error = (netdev->netdev_class->get_next_hop
892 ? netdev->netdev_class->get_next_hop(
893 host, next_hop, netdev_name)
896 next_hop->s_addr = 0;
902 /* Populates 'smap' with status information.
904 * Populates 'smap' with 'netdev' specific status information. This
905 * information may be used to populate the status column of the Interface table
906 * as defined in ovs-vswitchd.conf.db(5). */
908 netdev_get_status(const struct netdev *netdev, struct smap *smap)
910 return (netdev->netdev_class->get_status
911 ? netdev->netdev_class->get_status(netdev, smap)
915 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
916 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
917 * all-zero-bits (in6addr_any).
919 * The following error values have well-defined meanings:
921 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
923 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
925 * 'in6' may be null, in which case the address itself is not reported. */
927 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
929 struct in6_addr dummy;
932 error = (netdev->netdev_class->get_in6
933 ? netdev->netdev_class->get_in6(netdev,
937 memset(in6, 0, sizeof *in6);
942 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
943 * 'on'. Returns 0 if successful, otherwise a positive errno value. */
945 do_update_flags(struct netdev *netdev, enum netdev_flags off,
946 enum netdev_flags on, enum netdev_flags *old_flagsp,
947 struct netdev_saved_flags **sfp)
948 OVS_EXCLUDED(netdev_mutex)
950 struct netdev_saved_flags *sf = NULL;
951 enum netdev_flags old_flags;
954 error = netdev->netdev_class->update_flags(netdev, off & ~on, on,
957 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
958 off || on ? "set" : "get", netdev_get_name(netdev),
959 ovs_strerror(error));
961 } else if ((off || on) && sfp) {
962 enum netdev_flags new_flags = (old_flags & ~off) | on;
963 enum netdev_flags changed_flags = old_flags ^ new_flags;
965 ovs_mutex_lock(&netdev_mutex);
966 *sfp = sf = xmalloc(sizeof *sf);
968 list_push_front(&netdev->saved_flags_list, &sf->node);
969 sf->saved_flags = changed_flags;
970 sf->saved_values = changed_flags & new_flags;
973 ovs_mutex_unlock(&netdev_mutex);
978 *old_flagsp = old_flags;
987 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
988 * Returns 0 if successful, otherwise a positive errno value. On failure,
989 * stores 0 into '*flagsp'. */
991 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
993 struct netdev *netdev = CONST_CAST(struct netdev *, netdev_);
994 return do_update_flags(netdev, 0, 0, flagsp, NULL);
997 /* Sets the flags for 'netdev' to 'flags'.
998 * Returns 0 if successful, otherwise a positive errno value. */
1000 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
1001 struct netdev_saved_flags **sfp)
1003 return do_update_flags(netdev, -1, flags, NULL, sfp);
1006 /* Turns on the specified 'flags' on 'netdev':
1008 * - On success, returns 0. If 'sfp' is nonnull, sets '*sfp' to a newly
1009 * allocated 'struct netdev_saved_flags *' that may be passed to
1010 * netdev_restore_flags() to restore the original values of 'flags' on
1011 * 'netdev' (this will happen automatically at program termination if
1012 * netdev_restore_flags() is never called) , or to NULL if no flags were
1015 * - On failure, returns a positive errno value. If 'sfp' is nonnull, sets
1016 * '*sfp' to NULL. */
1018 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
1019 struct netdev_saved_flags **sfp)
1021 return do_update_flags(netdev, 0, flags, NULL, sfp);
1024 /* Turns off the specified 'flags' on 'netdev'. See netdev_turn_flags_on() for
1025 * details of the interface. */
1027 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
1028 struct netdev_saved_flags **sfp)
1030 return do_update_flags(netdev, flags, 0, NULL, sfp);
1033 /* Restores the flags that were saved in 'sf', and destroys 'sf'.
1034 * Does nothing if 'sf' is NULL. */
1036 netdev_restore_flags(struct netdev_saved_flags *sf)
1037 OVS_EXCLUDED(netdev_mutex)
1040 struct netdev *netdev = sf->netdev;
1041 enum netdev_flags old_flags;
1043 netdev->netdev_class->update_flags(netdev,
1044 sf->saved_flags & sf->saved_values,
1045 sf->saved_flags & ~sf->saved_values,
1048 ovs_mutex_lock(&netdev_mutex);
1049 list_remove(&sf->node);
1051 netdev_unref(netdev);
1055 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1056 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1057 * returns 0. Otherwise, it returns a positive errno value; in particular,
1058 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
1060 netdev_arp_lookup(const struct netdev *netdev,
1061 ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
1063 int error = (netdev->netdev_class->arp_lookup
1064 ? netdev->netdev_class->arp_lookup(netdev, ip, mac)
1067 memset(mac, 0, ETH_ADDR_LEN);
1072 /* Returns true if carrier is active (link light is on) on 'netdev'. */
1074 netdev_get_carrier(const struct netdev *netdev)
1077 enum netdev_flags flags;
1080 netdev_get_flags(netdev, &flags);
1081 if (!(flags & NETDEV_UP)) {
1085 if (!netdev->netdev_class->get_carrier) {
1089 error = netdev->netdev_class->get_carrier(netdev, &carrier);
1091 VLOG_DBG("%s: failed to get network device carrier status, assuming "
1092 "down: %s", netdev_get_name(netdev), ovs_strerror(error));
1099 /* Returns the number of times 'netdev''s carrier has changed. */
1101 netdev_get_carrier_resets(const struct netdev *netdev)
1103 return (netdev->netdev_class->get_carrier_resets
1104 ? netdev->netdev_class->get_carrier_resets(netdev)
1108 /* Attempts to force netdev_get_carrier() to poll 'netdev''s MII registers for
1109 * link status instead of checking 'netdev''s carrier. 'netdev''s MII
1110 * registers will be polled once ever 'interval' milliseconds. If 'netdev'
1111 * does not support MII, another method may be used as a fallback. If
1112 * 'interval' is less than or equal to zero, reverts netdev_get_carrier() to
1113 * its normal behavior.
1115 * Returns 0 if successful, otherwise a positive errno value. */
1117 netdev_set_miimon_interval(struct netdev *netdev, long long int interval)
1119 return (netdev->netdev_class->set_miimon_interval
1120 ? netdev->netdev_class->set_miimon_interval(netdev, interval)
1124 /* Retrieves current device stats for 'netdev'. */
1126 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
1130 COVERAGE_INC(netdev_get_stats);
1131 error = (netdev->netdev_class->get_stats
1132 ? netdev->netdev_class->get_stats(netdev, stats)
1135 memset(stats, 0xff, sizeof *stats);
1140 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
1141 * Returns 0 if successful, otherwise a positive errno value.
1143 * This will probably fail for most network devices. Some devices might only
1144 * allow setting their stats to 0. */
1146 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
1148 return (netdev->netdev_class->set_stats
1149 ? netdev->netdev_class->set_stats(netdev, stats)
1153 /* Attempts to set input rate limiting (policing) policy, such that up to
1154 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
1155 * size of 'kbits' kb. */
1157 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
1158 uint32_t kbits_burst)
1160 return (netdev->netdev_class->set_policing
1161 ? netdev->netdev_class->set_policing(netdev,
1162 kbits_rate, kbits_burst)
1166 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
1167 * empty if 'netdev' does not support QoS. Any names added to 'types' should
1168 * be documented as valid for the "type" column in the "QoS" table in
1169 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1171 * Every network device supports disabling QoS with a type of "", but this type
1172 * will not be added to 'types'.
1174 * The caller must initialize 'types' (e.g. with sset_init()) before calling
1175 * this function. The caller is responsible for destroying 'types' (e.g. with
1176 * sset_destroy()) when it is no longer needed.
1178 * Returns 0 if successful, otherwise a positive errno value. */
1180 netdev_get_qos_types(const struct netdev *netdev, struct sset *types)
1182 const struct netdev_class *class = netdev->netdev_class;
1183 return (class->get_qos_types
1184 ? class->get_qos_types(netdev, types)
1188 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
1189 * which should be "" or one of the types returned by netdev_get_qos_types()
1190 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
1191 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
1192 * 'caps' to all zeros. */
1194 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
1195 struct netdev_qos_capabilities *caps)
1197 const struct netdev_class *class = netdev->netdev_class;
1200 int retval = (class->get_qos_capabilities
1201 ? class->get_qos_capabilities(netdev, type, caps)
1204 memset(caps, 0, sizeof *caps);
1208 /* Every netdev supports turning off QoS. */
1209 memset(caps, 0, sizeof *caps);
1214 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1215 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1216 * the number of queues (zero on failure) in '*n_queuesp'.
1218 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1220 netdev_get_n_queues(const struct netdev *netdev,
1221 const char *type, unsigned int *n_queuesp)
1223 struct netdev_qos_capabilities caps;
1226 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1227 *n_queuesp = caps.n_queues;
1231 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1232 * stores the name of the current form of QoS into '*typep', stores any details
1233 * of configuration as string key-value pairs in 'details', and returns 0. On
1234 * failure, sets '*typep' to NULL and returns a positive errno value.
1236 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1238 * The caller must initialize 'details' as an empty smap (e.g. with
1239 * smap_init()) before calling this function. The caller must free 'details'
1240 * when it is no longer needed (e.g. with smap_destroy()).
1242 * The caller must not modify or free '*typep'.
1244 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1245 * 'netdev'. The contents of 'details' should be documented as valid for
1246 * '*typep' in the "other_config" column in the "QoS" table in
1247 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1249 netdev_get_qos(const struct netdev *netdev,
1250 const char **typep, struct smap *details)
1252 const struct netdev_class *class = netdev->netdev_class;
1255 if (class->get_qos) {
1256 retval = class->get_qos(netdev, typep, details);
1259 smap_clear(details);
1263 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1269 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1270 * with details of configuration from 'details'. Returns 0 if successful,
1271 * otherwise a positive errno value. On error, the previous QoS configuration
1274 * When this function changes the type of QoS (not just 'details'), this also
1275 * resets all queue configuration for 'netdev' to their defaults (which depend
1276 * on the specific type of QoS). Otherwise, the queue configuration for
1277 * 'netdev' is unchanged.
1279 * 'type' should be "" (to disable QoS) or one of the types returned by
1280 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1281 * documented as valid for the given 'type' in the "other_config" column in the
1282 * "QoS" table in vswitchd/vswitch.xml (which is built as
1283 * ovs-vswitchd.conf.db(8)).
1285 * NULL may be specified for 'details' if there are no configuration
1288 netdev_set_qos(struct netdev *netdev,
1289 const char *type, const struct smap *details)
1291 const struct netdev_class *class = netdev->netdev_class;
1297 if (class->set_qos) {
1299 static const struct smap empty = SMAP_INITIALIZER(&empty);
1302 return class->set_qos(netdev, type, details);
1304 return *type ? EOPNOTSUPP : 0;
1308 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1309 * successful, adds that information as string key-value pairs to 'details'.
1310 * Returns 0 if successful, otherwise a positive errno value.
1312 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1313 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1315 * The returned contents of 'details' should be documented as valid for the
1316 * given 'type' in the "other_config" column in the "Queue" table in
1317 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1319 * The caller must initialize 'details' (e.g. with smap_init()) before calling
1320 * this function. The caller must free 'details' when it is no longer needed
1321 * (e.g. with smap_destroy()). */
1323 netdev_get_queue(const struct netdev *netdev,
1324 unsigned int queue_id, struct smap *details)
1326 const struct netdev_class *class = netdev->netdev_class;
1329 retval = (class->get_queue
1330 ? class->get_queue(netdev, queue_id, details)
1333 smap_clear(details);
1338 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1339 * string pairs in 'details'. The contents of 'details' should be documented
1340 * as valid for the given 'type' in the "other_config" column in the "Queue"
1341 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1342 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1343 * given queue's configuration should be unmodified.
1345 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1346 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1348 * This function does not modify 'details', and the caller retains ownership of
1351 netdev_set_queue(struct netdev *netdev,
1352 unsigned int queue_id, const struct smap *details)
1354 const struct netdev_class *class = netdev->netdev_class;
1355 return (class->set_queue
1356 ? class->set_queue(netdev, queue_id, details)
1360 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1361 * of QoS may have a fixed set of queues, in which case attempts to delete them
1362 * will fail with EOPNOTSUPP.
1364 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1365 * given queue will be unmodified.
1367 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1368 * the current form of QoS (e.g. as returned by
1369 * netdev_get_n_queues(netdev)). */
1371 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1373 const struct netdev_class *class = netdev->netdev_class;
1374 return (class->delete_queue
1375 ? class->delete_queue(netdev, queue_id)
1379 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1380 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1381 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1382 * positive errno value and fills 'stats' with values indicating unsupported
1385 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1386 struct netdev_queue_stats *stats)
1388 const struct netdev_class *class = netdev->netdev_class;
1391 retval = (class->get_queue_stats
1392 ? class->get_queue_stats(netdev, queue_id, stats)
1395 stats->tx_bytes = UINT64_MAX;
1396 stats->tx_packets = UINT64_MAX;
1397 stats->tx_errors = UINT64_MAX;
1398 stats->created = LLONG_MIN;
1403 /* Initializes 'dump' to begin dumping the queues in a netdev.
1405 * This function provides no status indication. An error status for the entire
1406 * dump operation is provided when it is completed by calling
1407 * netdev_queue_dump_done().
1410 netdev_queue_dump_start(struct netdev_queue_dump *dump,
1411 const struct netdev *netdev)
1413 dump->netdev = netdev_ref(netdev);
1414 if (netdev->netdev_class->queue_dump_start) {
1415 dump->error = netdev->netdev_class->queue_dump_start(netdev,
1418 dump->error = EOPNOTSUPP;
1422 /* Attempts to retrieve another queue from 'dump', which must have been
1423 * initialized with netdev_queue_dump_start(). On success, stores a new queue
1424 * ID into '*queue_id', fills 'details' with configuration details for the
1425 * queue, and returns true. On failure, returns false.
1427 * Queues are not necessarily dumped in increasing order of queue ID (or any
1428 * other predictable order).
1430 * Failure might indicate an actual error or merely that the last queue has
1431 * been dumped. An error status for the entire dump operation is provided when
1432 * it is completed by calling netdev_queue_dump_done().
1434 * The returned contents of 'details' should be documented as valid for the
1435 * given 'type' in the "other_config" column in the "Queue" table in
1436 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1438 * The caller must initialize 'details' (e.g. with smap_init()) before calling
1439 * this function. This function will clear and replace its contents. The
1440 * caller must free 'details' when it is no longer needed (e.g. with
1441 * smap_destroy()). */
1443 netdev_queue_dump_next(struct netdev_queue_dump *dump,
1444 unsigned int *queue_id, struct smap *details)
1446 const struct netdev *netdev = dump->netdev;
1452 dump->error = netdev->netdev_class->queue_dump_next(netdev, dump->state,
1456 netdev->netdev_class->queue_dump_done(netdev, dump->state);
1462 /* Completes queue table dump operation 'dump', which must have been
1463 * initialized with netdev_queue_dump_start(). Returns 0 if the dump operation
1464 * was error-free, otherwise a positive errno value describing the problem. */
1466 netdev_queue_dump_done(struct netdev_queue_dump *dump)
1468 const struct netdev *netdev = dump->netdev;
1469 if (!dump->error && netdev->netdev_class->queue_dump_done) {
1470 dump->error = netdev->netdev_class->queue_dump_done(netdev,
1473 netdev_close(dump->netdev);
1474 return dump->error == EOF ? 0 : dump->error;
1477 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1478 * its statistics, and the 'aux' specified by the caller. The order of
1479 * iteration is unspecified, but (when successful) each queue is visited
1482 * Calling this function may be more efficient than calling
1483 * netdev_get_queue_stats() for every queue.
1485 * 'cb' must not modify or free the statistics passed in.
1487 * Returns 0 if successful, otherwise a positive errno value. On error, some
1488 * configured queues may not have been included in the iteration. */
1490 netdev_dump_queue_stats(const struct netdev *netdev,
1491 netdev_dump_queue_stats_cb *cb, void *aux)
1493 const struct netdev_class *class = netdev->netdev_class;
1494 return (class->dump_queue_stats
1495 ? class->dump_queue_stats(netdev, cb, aux)
1499 /* Returns a sequence number which indicates changes in one of 'netdev''s
1500 * properties. The returned sequence will be nonzero so that callers have a
1501 * value which they may use as a reset when tracking 'netdev'.
1503 * The returned sequence number will change whenever 'netdev''s flags,
1504 * features, ethernet address, or carrier changes. It may change for other
1505 * reasons as well, or no reason at all. */
1507 netdev_change_seq(const struct netdev *netdev)
1509 return netdev->netdev_class->change_seq(netdev);
1512 /* Returns the class type of 'netdev'.
1514 * The caller must not free the returned value. */
1516 netdev_get_type(const struct netdev *netdev)
1518 return netdev->netdev_class->type;
1521 /* Returns the class associated with 'netdev'. */
1522 const struct netdev_class *
1523 netdev_get_class(const struct netdev *netdev)
1525 return netdev->netdev_class;
1528 /* Returns the netdev with 'name' or NULL if there is none.
1530 * The caller must free the returned netdev with netdev_close(). */
1532 netdev_from_name(const char *name)
1533 OVS_EXCLUDED(netdev_mutex)
1535 struct netdev *netdev;
1537 ovs_mutex_lock(&netdev_mutex);
1538 netdev = shash_find_data(&netdev_shash, name);
1542 ovs_mutex_unlock(&netdev_mutex);
1547 /* Fills 'device_list' with devices that match 'netdev_class'.
1549 * The caller is responsible for initializing and destroying 'device_list' and
1550 * must close each device on the list. */
1552 netdev_get_devices(const struct netdev_class *netdev_class,
1553 struct shash *device_list)
1554 OVS_EXCLUDED(netdev_mutex)
1556 struct shash_node *node;
1558 ovs_mutex_lock(&netdev_mutex);
1559 SHASH_FOR_EACH (node, &netdev_shash) {
1560 struct netdev *dev = node->data;
1562 if (dev->netdev_class == netdev_class) {
1564 shash_add(device_list, node->name, node->data);
1567 ovs_mutex_unlock(&netdev_mutex);
1571 netdev_get_type_from_name(const char *name)
1573 struct netdev *dev = netdev_from_name(name);
1574 const char *type = dev ? netdev_get_type(dev) : NULL;
1580 netdev_rx_get_netdev(const struct netdev_rx *rx)
1582 ovs_assert(rx->netdev->ref_cnt > 0);
1587 netdev_rx_get_name(const struct netdev_rx *rx)
1589 return netdev_get_name(netdev_rx_get_netdev(rx));
1593 restore_all_flags(void *aux OVS_UNUSED)
1595 struct shash_node *node;
1597 SHASH_FOR_EACH (node, &netdev_shash) {
1598 struct netdev *netdev = node->data;
1599 const struct netdev_saved_flags *sf;
1600 enum netdev_flags saved_values;
1601 enum netdev_flags saved_flags;
1603 saved_values = saved_flags = 0;
1604 LIST_FOR_EACH (sf, node, &netdev->saved_flags_list) {
1605 saved_flags |= sf->saved_flags;
1606 saved_values &= ~sf->saved_flags;
1607 saved_values |= sf->saved_flags & sf->saved_values;
1610 enum netdev_flags old_flags;
1612 netdev->netdev_class->update_flags(netdev,
1613 saved_flags & saved_values,
1614 saved_flags & ~saved_values,