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>
27 #include "connectivity.h"
30 #include "dynamic-string.h"
31 #include "fatal-signal.h"
34 #include "netdev-provider.h"
35 #include "netdev-vport.h"
37 #include "openflow/openflow.h"
39 #include "poll-loop.h"
47 VLOG_DEFINE_THIS_MODULE(netdev);
49 COVERAGE_DEFINE(netdev_received);
50 COVERAGE_DEFINE(netdev_sent);
51 COVERAGE_DEFINE(netdev_add_router);
52 COVERAGE_DEFINE(netdev_get_stats);
54 struct netdev_saved_flags {
55 struct netdev *netdev;
56 struct list node; /* In struct netdev's saved_flags_list. */
57 enum netdev_flags saved_flags;
58 enum netdev_flags saved_values;
61 /* Protects 'netdev_shash' and the mutable members of struct netdev. */
62 static struct ovs_mutex netdev_mutex = OVS_MUTEX_INITIALIZER;
64 /* All created network devices. */
65 static struct shash netdev_shash OVS_GUARDED_BY(netdev_mutex)
66 = SHASH_INITIALIZER(&netdev_shash);
68 /* Protects 'netdev_classes' against insertions or deletions.
70 * This is not an rwlock for performance reasons but to allow recursive
71 * acquisition when calling into providers. For example, netdev_run() calls
72 * into provider 'run' functions, which might reasonably want to call one of
73 * the netdev functions that takes netdev_class_rwlock read-only. */
74 static struct ovs_rwlock netdev_class_rwlock OVS_ACQ_BEFORE(netdev_mutex)
75 = OVS_RWLOCK_INITIALIZER;
77 /* Contains 'struct netdev_registered_class'es. */
78 static struct hmap netdev_classes OVS_GUARDED_BY(netdev_class_rwlock)
79 = HMAP_INITIALIZER(&netdev_classes);
81 struct netdev_registered_class {
82 struct hmap_node hmap_node; /* In 'netdev_classes', by class->type. */
83 const struct netdev_class *class;
84 atomic_int ref_cnt; /* Number of 'struct netdev's of this class. */
87 /* This is set pretty low because we probably won't learn anything from the
88 * additional log messages. */
89 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
91 static void restore_all_flags(void *aux OVS_UNUSED);
92 void update_device_args(struct netdev *, const struct shash *args);
95 netdev_initialize(void)
96 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
98 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
100 if (ovsthread_once_start(&once)) {
101 fatal_signal_add_hook(restore_all_flags, NULL, NULL, true);
102 netdev_vport_patch_register();
104 #ifdef LINUX_DATAPATH
105 netdev_register_provider(&netdev_linux_class);
106 netdev_register_provider(&netdev_internal_class);
107 netdev_register_provider(&netdev_tap_class);
108 netdev_vport_tunnel_register();
110 #if defined(__FreeBSD__) || defined(__NetBSD__)
111 netdev_register_provider(&netdev_tap_class);
112 netdev_register_provider(&netdev_bsd_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);
227 atomic_destroy(&rc->ref_cnt);
231 VLOG_WARN("attempted to unregister in use netdev provider: %s",
236 ovs_rwlock_unlock(&netdev_class_rwlock);
241 /* Clears 'types' and enumerates the types of all currently registered netdev
242 * providers into it. The caller must first initialize the sset. */
244 netdev_enumerate_types(struct sset *types)
245 OVS_EXCLUDED(netdev_mutex)
247 struct netdev_registered_class *rc;
252 ovs_rwlock_rdlock(&netdev_class_rwlock);
253 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
254 sset_add(types, rc->class->type);
256 ovs_rwlock_unlock(&netdev_class_rwlock);
259 /* Check that the network device name is not the same as any of the registered
260 * vport providers' dpif_port name (dpif_port is NULL if the vport provider
261 * does not define it) or the datapath internal port name (e.g. ovs-system).
263 * Returns true if there is a name conflict, false otherwise. */
265 netdev_is_reserved_name(const char *name)
266 OVS_EXCLUDED(netdev_mutex)
268 struct netdev_registered_class *rc;
272 ovs_rwlock_rdlock(&netdev_class_rwlock);
273 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
274 const char *dpif_port = netdev_vport_class_get_dpif_port(rc->class);
275 if (dpif_port && !strcmp(dpif_port, name)) {
276 ovs_rwlock_unlock(&netdev_class_rwlock);
280 ovs_rwlock_unlock(&netdev_class_rwlock);
282 if (!strncmp(name, "ovs-", 4)) {
287 dp_enumerate_types(&types);
288 SSET_FOR_EACH (type, &types) {
289 if (!strcmp(name+4, type)) {
290 sset_destroy(&types);
294 sset_destroy(&types);
300 /* Opens the network device named 'name' (e.g. "eth0") of the specified 'type'
301 * (e.g. "system") and returns zero if successful, otherwise a positive errno
302 * value. On success, sets '*netdevp' to the new network device, otherwise to
305 * Some network devices may need to be configured (with netdev_set_config())
306 * before they can be used. */
308 netdev_open(const char *name, const char *type, struct netdev **netdevp)
309 OVS_EXCLUDED(netdev_mutex)
311 struct netdev *netdev;
316 ovs_rwlock_rdlock(&netdev_class_rwlock);
317 ovs_mutex_lock(&netdev_mutex);
318 netdev = shash_find_data(&netdev_shash, name);
320 struct netdev_registered_class *rc;
322 rc = netdev_lookup_class(type && type[0] ? type : "system");
324 netdev = rc->class->alloc();
326 memset(netdev, 0, sizeof *netdev);
327 netdev->netdev_class = rc->class;
328 netdev->name = xstrdup(name);
329 netdev->node = shash_add(&netdev_shash, name, netdev);
330 list_init(&netdev->saved_flags_list);
332 error = rc->class->construct(netdev);
336 atomic_add(&rc->ref_cnt, 1, &old_ref_cnt);
337 seq_change(connectivity_seq_get());
340 ovs_assert(list_is_empty(&netdev->saved_flags_list));
341 shash_delete(&netdev_shash, netdev->node);
342 rc->class->dealloc(netdev);
348 VLOG_WARN("could not create netdev %s of unknown type %s",
350 error = EAFNOSUPPORT;
356 ovs_mutex_unlock(&netdev_mutex);
357 ovs_rwlock_unlock(&netdev_class_rwlock);
368 /* Returns a reference to 'netdev_' for the caller to own. Returns null if
369 * 'netdev_' is null. */
371 netdev_ref(const struct netdev *netdev_)
372 OVS_EXCLUDED(netdev_mutex)
374 struct netdev *netdev = CONST_CAST(struct netdev *, netdev_);
377 ovs_mutex_lock(&netdev_mutex);
378 ovs_assert(netdev->ref_cnt > 0);
380 ovs_mutex_unlock(&netdev_mutex);
385 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
386 * or NULL if none are needed. */
388 netdev_set_config(struct netdev *netdev, const struct smap *args)
389 OVS_EXCLUDED(netdev_mutex)
391 if (netdev->netdev_class->set_config) {
392 const struct smap no_args = SMAP_INITIALIZER(&no_args);
395 error = netdev->netdev_class->set_config(netdev,
396 args ? args : &no_args);
398 VLOG_WARN("%s: could not set configuration (%s)",
399 netdev_get_name(netdev), ovs_strerror(error));
402 } else if (args && !smap_is_empty(args)) {
403 VLOG_WARN("%s: arguments provided to device that is not configurable",
404 netdev_get_name(netdev));
409 /* Returns the current configuration for 'netdev' in 'args'. The caller must
410 * have already initialized 'args' with smap_init(). Returns 0 on success, in
411 * which case 'args' will be filled with 'netdev''s configuration. On failure
412 * returns a positive errno value, in which case 'args' will be empty.
414 * The caller owns 'args' and its contents and must eventually free them with
417 netdev_get_config(const struct netdev *netdev, struct smap *args)
418 OVS_EXCLUDED(netdev_mutex)
423 if (netdev->netdev_class->get_config) {
424 error = netdev->netdev_class->get_config(netdev, args);
435 const struct netdev_tunnel_config *
436 netdev_get_tunnel_config(const struct netdev *netdev)
437 OVS_EXCLUDED(netdev_mutex)
439 if (netdev->netdev_class->get_tunnel_config) {
440 return netdev->netdev_class->get_tunnel_config(netdev);
447 netdev_unref(struct netdev *dev)
448 OVS_RELEASES(netdev_mutex)
450 ovs_assert(dev->ref_cnt);
451 if (!--dev->ref_cnt) {
452 const struct netdev_class *class = dev->netdev_class;
453 struct netdev_registered_class *rc;
456 dev->netdev_class->destruct(dev);
458 shash_delete(&netdev_shash, dev->node);
460 dev->netdev_class->dealloc(dev);
461 ovs_mutex_unlock(&netdev_mutex);
463 ovs_rwlock_rdlock(&netdev_class_rwlock);
464 rc = netdev_lookup_class(class->type);
465 atomic_sub(&rc->ref_cnt, 1, &old_ref_cnt);
466 ovs_assert(old_ref_cnt > 0);
467 ovs_rwlock_unlock(&netdev_class_rwlock);
469 ovs_mutex_unlock(&netdev_mutex);
473 /* Closes and destroys 'netdev'. */
475 netdev_close(struct netdev *netdev)
476 OVS_EXCLUDED(netdev_mutex)
479 ovs_mutex_lock(&netdev_mutex);
480 netdev_unref(netdev);
484 /* Parses 'netdev_name_', which is of the form [type@]name into its component
485 * pieces. 'name' and 'type' must be freed by the caller. */
487 netdev_parse_name(const char *netdev_name_, char **name, char **type)
489 char *netdev_name = xstrdup(netdev_name_);
492 separator = strchr(netdev_name, '@');
496 *name = xstrdup(separator + 1);
499 *type = xstrdup("system");
504 netdev_rx_open(struct netdev *netdev, struct netdev_rx **rxp)
505 OVS_EXCLUDED(netdev_mutex)
509 if (netdev->netdev_class->rx_alloc) {
510 struct netdev_rx *rx = netdev->netdev_class->rx_alloc();
513 error = netdev->netdev_class->rx_construct(rx);
515 ovs_mutex_lock(&netdev_mutex);
517 ovs_mutex_unlock(&netdev_mutex);
522 netdev->netdev_class->rx_dealloc(rx);
535 netdev_rx_close(struct netdev_rx *rx)
536 OVS_EXCLUDED(netdev_mutex)
539 struct netdev *netdev = rx->netdev;
540 netdev->netdev_class->rx_destruct(rx);
541 netdev->netdev_class->rx_dealloc(rx);
542 netdev_close(netdev);
547 netdev_rx_recv(struct netdev_rx *rx, struct ofpbuf *buffer)
551 ovs_assert(buffer->size == 0);
552 ovs_assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
554 retval = rx->netdev->netdev_class->rx_recv(rx, buffer);
556 COVERAGE_INC(netdev_received);
557 if (buffer->size < ETH_TOTAL_MIN) {
558 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
567 netdev_rx_wait(struct netdev_rx *rx)
569 rx->netdev->netdev_class->rx_wait(rx);
573 netdev_rx_drain(struct netdev_rx *rx)
575 return (rx->netdev->netdev_class->rx_drain
576 ? rx->netdev->netdev_class->rx_drain(rx)
580 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
581 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
582 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
583 * the packet is too big or too small to transmit on the device.
585 * The caller retains ownership of 'buffer' in all cases.
587 * The kernel maintains a packet transmission queue, so the caller is not
588 * expected to do additional queuing of packets.
590 * Some network devices may not implement support for this function. In such
591 * cases this function will always return EOPNOTSUPP. */
593 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
597 error = (netdev->netdev_class->send
598 ? netdev->netdev_class->send(netdev, buffer->data, buffer->size)
601 COVERAGE_INC(netdev_sent);
606 /* Registers with the poll loop to wake up from the next call to poll_block()
607 * when the packet transmission queue has sufficient room to transmit a packet
608 * with netdev_send().
610 * The kernel maintains a packet transmission queue, so the client is not
611 * expected to do additional queuing of packets. Thus, this function is
612 * unlikely to ever be used. It is included for completeness. */
614 netdev_send_wait(struct netdev *netdev)
616 if (netdev->netdev_class->send_wait) {
617 netdev->netdev_class->send_wait(netdev);
621 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
622 * otherwise a positive errno value. */
624 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
626 return netdev->netdev_class->set_etheraddr(netdev, mac);
629 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
630 * the MAC address into 'mac'. On failure, returns a positive errno value and
631 * clears 'mac' to all-zeros. */
633 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
635 return netdev->netdev_class->get_etheraddr(netdev, mac);
638 /* Returns the name of the network device that 'netdev' represents,
639 * e.g. "eth0". The caller must not modify or free the returned string. */
641 netdev_get_name(const struct netdev *netdev)
646 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
647 * (and received) packets, in bytes, not including the hardware header; thus,
648 * this is typically 1500 bytes for Ethernet devices.
650 * If successful, returns 0 and stores the MTU size in '*mtup'. Returns
651 * EOPNOTSUPP if 'netdev' does not have an MTU (as e.g. some tunnels do not).
652 * On other failure, returns a positive errno value. On failure, sets '*mtup'
655 netdev_get_mtu(const struct netdev *netdev, int *mtup)
657 const struct netdev_class *class = netdev->netdev_class;
660 error = class->get_mtu ? class->get_mtu(netdev, mtup) : EOPNOTSUPP;
663 if (error != EOPNOTSUPP) {
664 VLOG_DBG_RL(&rl, "failed to retrieve MTU for network device %s: "
665 "%s", netdev_get_name(netdev), ovs_strerror(error));
671 /* Sets the MTU of 'netdev'. The MTU is the maximum size of transmitted
672 * (and received) packets, in bytes.
674 * If successful, returns 0. Returns EOPNOTSUPP if 'netdev' does not have an
675 * MTU (as e.g. some tunnels do not). On other failure, returns a positive
678 netdev_set_mtu(const struct netdev *netdev, int mtu)
680 const struct netdev_class *class = netdev->netdev_class;
683 error = class->set_mtu ? class->set_mtu(netdev, mtu) : EOPNOTSUPP;
684 if (error && error != EOPNOTSUPP) {
685 VLOG_DBG_RL(&rl, "failed to set MTU for network device %s: %s",
686 netdev_get_name(netdev), ovs_strerror(error));
692 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
693 * failure, returns a negative errno value.
695 * The desired semantics of the ifindex value are a combination of those
696 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
697 * value should be unique within a host and remain stable at least until
698 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
699 * but many systems do not follow this rule anyhow.
701 * Some network devices may not implement support for this function. In such
702 * cases this function will always return -EOPNOTSUPP.
705 netdev_get_ifindex(const struct netdev *netdev)
707 int (*get_ifindex)(const struct netdev *);
709 get_ifindex = netdev->netdev_class->get_ifindex;
711 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
714 /* Stores the features supported by 'netdev' into each of '*current',
715 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
716 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
717 * successful, otherwise a positive errno value. On failure, all of the
718 * passed-in values are set to 0.
720 * Some network devices may not implement support for this function. In such
721 * cases this function will always return EOPNOTSUPP. */
723 netdev_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)
729 int (*get_features)(const struct netdev *netdev,
730 enum netdev_features *current,
731 enum netdev_features *advertised,
732 enum netdev_features *supported,
733 enum netdev_features *peer);
734 enum netdev_features dummy[4];
741 advertised = &dummy[1];
744 supported = &dummy[2];
750 get_features = netdev->netdev_class->get_features;
752 ? get_features(netdev, current, advertised, supported,
756 *current = *advertised = *supported = *peer = 0;
761 /* Returns the maximum speed of a network connection that has the NETDEV_F_*
762 * bits in 'features', in bits per second. If no bits that indicate a speed
763 * are set in 'features', returns 'default_bps'. */
765 netdev_features_to_bps(enum netdev_features features,
766 uint64_t default_bps)
769 F_1000000MB = NETDEV_F_1TB_FD,
770 F_100000MB = NETDEV_F_100GB_FD,
771 F_40000MB = NETDEV_F_40GB_FD,
772 F_10000MB = NETDEV_F_10GB_FD,
773 F_1000MB = NETDEV_F_1GB_HD | NETDEV_F_1GB_FD,
774 F_100MB = NETDEV_F_100MB_HD | NETDEV_F_100MB_FD,
775 F_10MB = NETDEV_F_10MB_HD | NETDEV_F_10MB_FD
778 return ( features & F_1000000MB ? UINT64_C(1000000000000)
779 : features & F_100000MB ? UINT64_C(100000000000)
780 : features & F_40000MB ? UINT64_C(40000000000)
781 : features & F_10000MB ? UINT64_C(10000000000)
782 : features & F_1000MB ? UINT64_C(1000000000)
783 : features & F_100MB ? UINT64_C(100000000)
784 : features & F_10MB ? UINT64_C(10000000)
788 /* Returns true if any of the NETDEV_F_* bits that indicate a full-duplex link
789 * are set in 'features', otherwise false. */
791 netdev_features_is_full_duplex(enum netdev_features features)
793 return (features & (NETDEV_F_10MB_FD | NETDEV_F_100MB_FD | NETDEV_F_1GB_FD
794 | NETDEV_F_10GB_FD | NETDEV_F_40GB_FD
795 | NETDEV_F_100GB_FD | NETDEV_F_1TB_FD)) != 0;
798 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
799 * successful, otherwise a positive errno value. */
801 netdev_set_advertisements(struct netdev *netdev,
802 enum netdev_features advertise)
804 return (netdev->netdev_class->set_advertisements
805 ? netdev->netdev_class->set_advertisements(
810 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
811 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
812 * errno value and sets '*address' to 0 (INADDR_ANY).
814 * The following error values have well-defined meanings:
816 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
818 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
820 * 'address' or 'netmask' or both may be null, in which case the address or
821 * netmask is not reported. */
823 netdev_get_in4(const struct netdev *netdev,
824 struct in_addr *address_, struct in_addr *netmask_)
826 struct in_addr address;
827 struct in_addr netmask;
830 error = (netdev->netdev_class->get_in4
831 ? netdev->netdev_class->get_in4(netdev,
835 address_->s_addr = error ? 0 : address.s_addr;
838 netmask_->s_addr = error ? 0 : netmask.s_addr;
843 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
844 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
845 * positive errno value. */
847 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
849 return (netdev->netdev_class->set_in4
850 ? netdev->netdev_class->set_in4(netdev, addr, mask)
854 /* Obtains ad IPv4 address from device name and save the address in
855 * in4. Returns 0 if successful, otherwise a positive errno value.
858 netdev_get_in4_by_name(const char *device_name, struct in_addr *in4)
860 struct netdev *netdev;
863 error = netdev_open(device_name, "system", &netdev);
865 in4->s_addr = htonl(0);
869 error = netdev_get_in4(netdev, in4, NULL);
870 netdev_close(netdev);
874 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
877 netdev_add_router(struct netdev *netdev, struct in_addr router)
879 COVERAGE_INC(netdev_add_router);
880 return (netdev->netdev_class->add_router
881 ? netdev->netdev_class->add_router(netdev, router)
885 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
886 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
887 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
888 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
889 * a directly connected network) in '*next_hop' and a copy of the name of the
890 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
891 * responsible for freeing '*netdev_name' (by calling free()). */
893 netdev_get_next_hop(const struct netdev *netdev,
894 const struct in_addr *host, struct in_addr *next_hop,
897 int error = (netdev->netdev_class->get_next_hop
898 ? netdev->netdev_class->get_next_hop(
899 host, next_hop, netdev_name)
902 next_hop->s_addr = 0;
908 /* Populates 'smap' with status information.
910 * Populates 'smap' with 'netdev' specific status information. This
911 * information may be used to populate the status column of the Interface table
912 * as defined in ovs-vswitchd.conf.db(5). */
914 netdev_get_status(const struct netdev *netdev, struct smap *smap)
916 return (netdev->netdev_class->get_status
917 ? netdev->netdev_class->get_status(netdev, smap)
921 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
922 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
923 * all-zero-bits (in6addr_any).
925 * The following error values have well-defined meanings:
927 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
929 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
931 * 'in6' may be null, in which case the address itself is not reported. */
933 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
935 struct in6_addr dummy;
938 error = (netdev->netdev_class->get_in6
939 ? netdev->netdev_class->get_in6(netdev,
943 memset(in6, 0, sizeof *in6);
948 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
949 * 'on'. Returns 0 if successful, otherwise a positive errno value. */
951 do_update_flags(struct netdev *netdev, enum netdev_flags off,
952 enum netdev_flags on, enum netdev_flags *old_flagsp,
953 struct netdev_saved_flags **sfp)
954 OVS_EXCLUDED(netdev_mutex)
956 struct netdev_saved_flags *sf = NULL;
957 enum netdev_flags old_flags;
960 error = netdev->netdev_class->update_flags(netdev, off & ~on, on,
963 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
964 off || on ? "set" : "get", netdev_get_name(netdev),
965 ovs_strerror(error));
967 } else if ((off || on) && sfp) {
968 enum netdev_flags new_flags = (old_flags & ~off) | on;
969 enum netdev_flags changed_flags = old_flags ^ new_flags;
971 ovs_mutex_lock(&netdev_mutex);
972 *sfp = sf = xmalloc(sizeof *sf);
974 list_push_front(&netdev->saved_flags_list, &sf->node);
975 sf->saved_flags = changed_flags;
976 sf->saved_values = changed_flags & new_flags;
979 ovs_mutex_unlock(&netdev_mutex);
984 *old_flagsp = old_flags;
993 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
994 * Returns 0 if successful, otherwise a positive errno value. On failure,
995 * stores 0 into '*flagsp'. */
997 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
999 struct netdev *netdev = CONST_CAST(struct netdev *, netdev_);
1000 return do_update_flags(netdev, 0, 0, flagsp, NULL);
1003 /* Sets the flags for 'netdev' to 'flags'.
1004 * Returns 0 if successful, otherwise a positive errno value. */
1006 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
1007 struct netdev_saved_flags **sfp)
1009 return do_update_flags(netdev, -1, flags, NULL, sfp);
1012 /* Turns on the specified 'flags' on 'netdev':
1014 * - On success, returns 0. If 'sfp' is nonnull, sets '*sfp' to a newly
1015 * allocated 'struct netdev_saved_flags *' that may be passed to
1016 * netdev_restore_flags() to restore the original values of 'flags' on
1017 * 'netdev' (this will happen automatically at program termination if
1018 * netdev_restore_flags() is never called) , or to NULL if no flags were
1021 * - On failure, returns a positive errno value. If 'sfp' is nonnull, sets
1022 * '*sfp' to NULL. */
1024 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
1025 struct netdev_saved_flags **sfp)
1027 return do_update_flags(netdev, 0, flags, NULL, sfp);
1030 /* Turns off the specified 'flags' on 'netdev'. See netdev_turn_flags_on() for
1031 * details of the interface. */
1033 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
1034 struct netdev_saved_flags **sfp)
1036 return do_update_flags(netdev, flags, 0, NULL, sfp);
1039 /* Restores the flags that were saved in 'sf', and destroys 'sf'.
1040 * Does nothing if 'sf' is NULL. */
1042 netdev_restore_flags(struct netdev_saved_flags *sf)
1043 OVS_EXCLUDED(netdev_mutex)
1046 struct netdev *netdev = sf->netdev;
1047 enum netdev_flags old_flags;
1049 netdev->netdev_class->update_flags(netdev,
1050 sf->saved_flags & sf->saved_values,
1051 sf->saved_flags & ~sf->saved_values,
1054 ovs_mutex_lock(&netdev_mutex);
1055 list_remove(&sf->node);
1057 netdev_unref(netdev);
1061 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1062 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1063 * returns 0. Otherwise, it returns a positive errno value; in particular,
1064 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
1066 netdev_arp_lookup(const struct netdev *netdev,
1067 ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
1069 int error = (netdev->netdev_class->arp_lookup
1070 ? netdev->netdev_class->arp_lookup(netdev, ip, mac)
1073 memset(mac, 0, ETH_ADDR_LEN);
1078 /* Returns true if carrier is active (link light is on) on 'netdev'. */
1080 netdev_get_carrier(const struct netdev *netdev)
1083 enum netdev_flags flags;
1086 netdev_get_flags(netdev, &flags);
1087 if (!(flags & NETDEV_UP)) {
1091 if (!netdev->netdev_class->get_carrier) {
1095 error = netdev->netdev_class->get_carrier(netdev, &carrier);
1097 VLOG_DBG("%s: failed to get network device carrier status, assuming "
1098 "down: %s", netdev_get_name(netdev), ovs_strerror(error));
1105 /* Returns the number of times 'netdev''s carrier has changed. */
1107 netdev_get_carrier_resets(const struct netdev *netdev)
1109 return (netdev->netdev_class->get_carrier_resets
1110 ? netdev->netdev_class->get_carrier_resets(netdev)
1114 /* Attempts to force netdev_get_carrier() to poll 'netdev''s MII registers for
1115 * link status instead of checking 'netdev''s carrier. 'netdev''s MII
1116 * registers will be polled once ever 'interval' milliseconds. If 'netdev'
1117 * does not support MII, another method may be used as a fallback. If
1118 * 'interval' is less than or equal to zero, reverts netdev_get_carrier() to
1119 * its normal behavior.
1121 * Returns 0 if successful, otherwise a positive errno value. */
1123 netdev_set_miimon_interval(struct netdev *netdev, long long int interval)
1125 return (netdev->netdev_class->set_miimon_interval
1126 ? netdev->netdev_class->set_miimon_interval(netdev, interval)
1130 /* Retrieves current device stats for 'netdev'. */
1132 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
1136 COVERAGE_INC(netdev_get_stats);
1137 error = (netdev->netdev_class->get_stats
1138 ? netdev->netdev_class->get_stats(netdev, stats)
1141 memset(stats, 0xff, sizeof *stats);
1146 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
1147 * Returns 0 if successful, otherwise a positive errno value.
1149 * This will probably fail for most network devices. Some devices might only
1150 * allow setting their stats to 0. */
1152 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
1154 return (netdev->netdev_class->set_stats
1155 ? netdev->netdev_class->set_stats(netdev, stats)
1159 /* Attempts to set input rate limiting (policing) policy, such that up to
1160 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
1161 * size of 'kbits' kb. */
1163 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
1164 uint32_t kbits_burst)
1166 return (netdev->netdev_class->set_policing
1167 ? netdev->netdev_class->set_policing(netdev,
1168 kbits_rate, kbits_burst)
1172 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
1173 * empty if 'netdev' does not support QoS. Any names added to 'types' should
1174 * be documented as valid for the "type" column in the "QoS" table in
1175 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1177 * Every network device supports disabling QoS with a type of "", but this type
1178 * will not be added to 'types'.
1180 * The caller must initialize 'types' (e.g. with sset_init()) before calling
1181 * this function. The caller is responsible for destroying 'types' (e.g. with
1182 * sset_destroy()) when it is no longer needed.
1184 * Returns 0 if successful, otherwise a positive errno value. */
1186 netdev_get_qos_types(const struct netdev *netdev, struct sset *types)
1188 const struct netdev_class *class = netdev->netdev_class;
1189 return (class->get_qos_types
1190 ? class->get_qos_types(netdev, types)
1194 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
1195 * which should be "" or one of the types returned by netdev_get_qos_types()
1196 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
1197 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
1198 * 'caps' to all zeros. */
1200 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
1201 struct netdev_qos_capabilities *caps)
1203 const struct netdev_class *class = netdev->netdev_class;
1206 int retval = (class->get_qos_capabilities
1207 ? class->get_qos_capabilities(netdev, type, caps)
1210 memset(caps, 0, sizeof *caps);
1214 /* Every netdev supports turning off QoS. */
1215 memset(caps, 0, sizeof *caps);
1220 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1221 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1222 * the number of queues (zero on failure) in '*n_queuesp'.
1224 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1226 netdev_get_n_queues(const struct netdev *netdev,
1227 const char *type, unsigned int *n_queuesp)
1229 struct netdev_qos_capabilities caps;
1232 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1233 *n_queuesp = caps.n_queues;
1237 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1238 * stores the name of the current form of QoS into '*typep', stores any details
1239 * of configuration as string key-value pairs in 'details', and returns 0. On
1240 * failure, sets '*typep' to NULL and returns a positive errno value.
1242 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1244 * The caller must initialize 'details' as an empty smap (e.g. with
1245 * smap_init()) before calling this function. The caller must free 'details'
1246 * when it is no longer needed (e.g. with smap_destroy()).
1248 * The caller must not modify or free '*typep'.
1250 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1251 * 'netdev'. The contents of 'details' should be documented as valid for
1252 * '*typep' in the "other_config" column in the "QoS" table in
1253 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1255 netdev_get_qos(const struct netdev *netdev,
1256 const char **typep, struct smap *details)
1258 const struct netdev_class *class = netdev->netdev_class;
1261 if (class->get_qos) {
1262 retval = class->get_qos(netdev, typep, details);
1265 smap_clear(details);
1269 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1275 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1276 * with details of configuration from 'details'. Returns 0 if successful,
1277 * otherwise a positive errno value. On error, the previous QoS configuration
1280 * When this function changes the type of QoS (not just 'details'), this also
1281 * resets all queue configuration for 'netdev' to their defaults (which depend
1282 * on the specific type of QoS). Otherwise, the queue configuration for
1283 * 'netdev' is unchanged.
1285 * 'type' should be "" (to disable QoS) or one of the types returned by
1286 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1287 * documented as valid for the given 'type' in the "other_config" column in the
1288 * "QoS" table in vswitchd/vswitch.xml (which is built as
1289 * ovs-vswitchd.conf.db(8)).
1291 * NULL may be specified for 'details' if there are no configuration
1294 netdev_set_qos(struct netdev *netdev,
1295 const char *type, const struct smap *details)
1297 const struct netdev_class *class = netdev->netdev_class;
1303 if (class->set_qos) {
1305 static const struct smap empty = SMAP_INITIALIZER(&empty);
1308 return class->set_qos(netdev, type, details);
1310 return *type ? EOPNOTSUPP : 0;
1314 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1315 * successful, adds that information as string key-value pairs to 'details'.
1316 * Returns 0 if successful, otherwise a positive errno value.
1318 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1319 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1321 * The returned contents of 'details' should be documented as valid for the
1322 * given 'type' in the "other_config" column in the "Queue" table in
1323 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1325 * The caller must initialize 'details' (e.g. with smap_init()) before calling
1326 * this function. The caller must free 'details' when it is no longer needed
1327 * (e.g. with smap_destroy()). */
1329 netdev_get_queue(const struct netdev *netdev,
1330 unsigned int queue_id, struct smap *details)
1332 const struct netdev_class *class = netdev->netdev_class;
1335 retval = (class->get_queue
1336 ? class->get_queue(netdev, queue_id, details)
1339 smap_clear(details);
1344 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1345 * string pairs in 'details'. The contents of 'details' should be documented
1346 * as valid for the given 'type' in the "other_config" column in the "Queue"
1347 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1348 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1349 * given queue's configuration should be unmodified.
1351 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1352 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1354 * This function does not modify 'details', and the caller retains ownership of
1357 netdev_set_queue(struct netdev *netdev,
1358 unsigned int queue_id, const struct smap *details)
1360 const struct netdev_class *class = netdev->netdev_class;
1361 return (class->set_queue
1362 ? class->set_queue(netdev, queue_id, details)
1366 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1367 * of QoS may have a fixed set of queues, in which case attempts to delete them
1368 * will fail with EOPNOTSUPP.
1370 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1371 * given queue will be unmodified.
1373 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1374 * the current form of QoS (e.g. as returned by
1375 * netdev_get_n_queues(netdev)). */
1377 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1379 const struct netdev_class *class = netdev->netdev_class;
1380 return (class->delete_queue
1381 ? class->delete_queue(netdev, queue_id)
1385 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1386 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1387 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1388 * positive errno value and fills 'stats' with values indicating unsupported
1391 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1392 struct netdev_queue_stats *stats)
1394 const struct netdev_class *class = netdev->netdev_class;
1397 retval = (class->get_queue_stats
1398 ? class->get_queue_stats(netdev, queue_id, stats)
1401 stats->tx_bytes = UINT64_MAX;
1402 stats->tx_packets = UINT64_MAX;
1403 stats->tx_errors = UINT64_MAX;
1404 stats->created = LLONG_MIN;
1409 /* Initializes 'dump' to begin dumping the queues in a netdev.
1411 * This function provides no status indication. An error status for the entire
1412 * dump operation is provided when it is completed by calling
1413 * netdev_queue_dump_done().
1416 netdev_queue_dump_start(struct netdev_queue_dump *dump,
1417 const struct netdev *netdev)
1419 dump->netdev = netdev_ref(netdev);
1420 if (netdev->netdev_class->queue_dump_start) {
1421 dump->error = netdev->netdev_class->queue_dump_start(netdev,
1424 dump->error = EOPNOTSUPP;
1428 /* Attempts to retrieve another queue from 'dump', which must have been
1429 * initialized with netdev_queue_dump_start(). On success, stores a new queue
1430 * ID into '*queue_id', fills 'details' with configuration details for the
1431 * queue, and returns true. On failure, returns false.
1433 * Queues are not necessarily dumped in increasing order of queue ID (or any
1434 * other predictable order).
1436 * Failure might indicate an actual error or merely that the last queue has
1437 * been dumped. An error status for the entire dump operation is provided when
1438 * it is completed by calling netdev_queue_dump_done().
1440 * The returned contents of 'details' should be documented as valid for the
1441 * given 'type' in the "other_config" column in the "Queue" table in
1442 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1444 * The caller must initialize 'details' (e.g. with smap_init()) before calling
1445 * this function. This function will clear and replace its contents. The
1446 * caller must free 'details' when it is no longer needed (e.g. with
1447 * smap_destroy()). */
1449 netdev_queue_dump_next(struct netdev_queue_dump *dump,
1450 unsigned int *queue_id, struct smap *details)
1452 const struct netdev *netdev = dump->netdev;
1458 dump->error = netdev->netdev_class->queue_dump_next(netdev, dump->state,
1462 netdev->netdev_class->queue_dump_done(netdev, dump->state);
1468 /* Completes queue table dump operation 'dump', which must have been
1469 * initialized with netdev_queue_dump_start(). Returns 0 if the dump operation
1470 * was error-free, otherwise a positive errno value describing the problem. */
1472 netdev_queue_dump_done(struct netdev_queue_dump *dump)
1474 const struct netdev *netdev = dump->netdev;
1475 if (!dump->error && netdev->netdev_class->queue_dump_done) {
1476 dump->error = netdev->netdev_class->queue_dump_done(netdev,
1479 netdev_close(dump->netdev);
1480 return dump->error == EOF ? 0 : dump->error;
1483 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1484 * its statistics, and the 'aux' specified by the caller. The order of
1485 * iteration is unspecified, but (when successful) each queue is visited
1488 * Calling this function may be more efficient than calling
1489 * netdev_get_queue_stats() for every queue.
1491 * 'cb' must not modify or free the statistics passed in.
1493 * Returns 0 if successful, otherwise a positive errno value. On error, some
1494 * configured queues may not have been included in the iteration. */
1496 netdev_dump_queue_stats(const struct netdev *netdev,
1497 netdev_dump_queue_stats_cb *cb, void *aux)
1499 const struct netdev_class *class = netdev->netdev_class;
1500 return (class->dump_queue_stats
1501 ? class->dump_queue_stats(netdev, cb, aux)
1506 /* Returns the class type of 'netdev'.
1508 * The caller must not free the returned value. */
1510 netdev_get_type(const struct netdev *netdev)
1512 return netdev->netdev_class->type;
1515 /* Returns the class associated with 'netdev'. */
1516 const struct netdev_class *
1517 netdev_get_class(const struct netdev *netdev)
1519 return netdev->netdev_class;
1522 /* Returns the netdev with 'name' or NULL if there is none.
1524 * The caller must free the returned netdev with netdev_close(). */
1526 netdev_from_name(const char *name)
1527 OVS_EXCLUDED(netdev_mutex)
1529 struct netdev *netdev;
1531 ovs_mutex_lock(&netdev_mutex);
1532 netdev = shash_find_data(&netdev_shash, name);
1536 ovs_mutex_unlock(&netdev_mutex);
1541 /* Fills 'device_list' with devices that match 'netdev_class'.
1543 * The caller is responsible for initializing and destroying 'device_list' and
1544 * must close each device on the list. */
1546 netdev_get_devices(const struct netdev_class *netdev_class,
1547 struct shash *device_list)
1548 OVS_EXCLUDED(netdev_mutex)
1550 struct shash_node *node;
1552 ovs_mutex_lock(&netdev_mutex);
1553 SHASH_FOR_EACH (node, &netdev_shash) {
1554 struct netdev *dev = node->data;
1556 if (dev->netdev_class == netdev_class) {
1558 shash_add(device_list, node->name, node->data);
1561 ovs_mutex_unlock(&netdev_mutex);
1565 netdev_get_type_from_name(const char *name)
1567 struct netdev *dev = netdev_from_name(name);
1568 const char *type = dev ? netdev_get_type(dev) : NULL;
1574 netdev_rx_get_netdev(const struct netdev_rx *rx)
1576 ovs_assert(rx->netdev->ref_cnt > 0);
1581 netdev_rx_get_name(const struct netdev_rx *rx)
1583 return netdev_get_name(netdev_rx_get_netdev(rx));
1587 restore_all_flags(void *aux OVS_UNUSED)
1589 struct shash_node *node;
1591 SHASH_FOR_EACH (node, &netdev_shash) {
1592 struct netdev *netdev = node->data;
1593 const struct netdev_saved_flags *sf;
1594 enum netdev_flags saved_values;
1595 enum netdev_flags saved_flags;
1597 saved_values = saved_flags = 0;
1598 LIST_FOR_EACH (sf, node, &netdev->saved_flags_list) {
1599 saved_flags |= sf->saved_flags;
1600 saved_values &= ~sf->saved_flags;
1601 saved_values |= sf->saved_flags & sf->saved_values;
1604 enum netdev_flags old_flags;
1606 netdev->netdev_class->update_flags(netdev,
1607 saved_flags & saved_values,
1608 saved_flags & ~saved_values,