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);
114 netdev_register_provider(&netdev_tunnel_class);
115 netdev_register_provider(&netdev_pltap_class);
117 ovsthread_once_done(&once);
121 /* Performs periodic work needed by all the various kinds of netdevs.
123 * If your program opens any netdevs, it must call this function within its
127 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
129 struct netdev_registered_class *rc;
131 ovs_rwlock_rdlock(&netdev_class_rwlock);
132 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
133 if (rc->class->run) {
137 ovs_rwlock_unlock(&netdev_class_rwlock);
140 /* Arranges for poll_block() to wake up when netdev_run() needs to be called.
142 * If your program opens any netdevs, it must call this function within its
146 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
148 struct netdev_registered_class *rc;
150 ovs_rwlock_rdlock(&netdev_class_rwlock);
151 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
152 if (rc->class->wait) {
156 ovs_rwlock_unlock(&netdev_class_rwlock);
159 static struct netdev_registered_class *
160 netdev_lookup_class(const char *type)
161 OVS_REQ_RDLOCK(netdev_class_rwlock)
163 struct netdev_registered_class *rc;
165 HMAP_FOR_EACH_WITH_HASH (rc, hmap_node, hash_string(type, 0),
167 if (!strcmp(type, rc->class->type)) {
174 /* Initializes and registers a new netdev provider. After successful
175 * registration, new netdevs of that type can be opened using netdev_open(). */
177 netdev_register_provider(const struct netdev_class *new_class)
178 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
182 ovs_rwlock_wrlock(&netdev_class_rwlock);
183 if (netdev_lookup_class(new_class->type)) {
184 VLOG_WARN("attempted to register duplicate netdev provider: %s",
188 error = new_class->init ? new_class->init() : 0;
190 struct netdev_registered_class *rc;
192 rc = xmalloc(sizeof *rc);
193 hmap_insert(&netdev_classes, &rc->hmap_node,
194 hash_string(new_class->type, 0));
195 rc->class = new_class;
196 atomic_init(&rc->ref_cnt, 0);
198 VLOG_ERR("failed to initialize %s network device class: %s",
199 new_class->type, ovs_strerror(error));
202 ovs_rwlock_unlock(&netdev_class_rwlock);
207 /* Unregisters a netdev provider. 'type' must have been previously
208 * registered and not currently be in use by any netdevs. After unregistration
209 * new netdevs of that type cannot be opened using netdev_open(). */
211 netdev_unregister_provider(const char *type)
212 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
214 struct netdev_registered_class *rc;
217 ovs_rwlock_wrlock(&netdev_class_rwlock);
218 rc = netdev_lookup_class(type);
220 VLOG_WARN("attempted to unregister a netdev provider that is not "
221 "registered: %s", type);
222 error = EAFNOSUPPORT;
226 atomic_read(&rc->ref_cnt, &ref_cnt);
228 hmap_remove(&netdev_classes, &rc->hmap_node);
232 VLOG_WARN("attempted to unregister in use netdev provider: %s",
237 ovs_rwlock_unlock(&netdev_class_rwlock);
242 /* Clears 'types' and enumerates the types of all currently registered netdev
243 * providers into it. The caller must first initialize the sset. */
245 netdev_enumerate_types(struct sset *types)
246 OVS_EXCLUDED(netdev_mutex)
248 struct netdev_registered_class *rc;
253 ovs_rwlock_rdlock(&netdev_class_rwlock);
254 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
255 sset_add(types, rc->class->type);
257 ovs_rwlock_unlock(&netdev_class_rwlock);
260 /* Check that the network device name is not the same as any of the registered
261 * vport providers' dpif_port name (dpif_port is NULL if the vport provider
262 * does not define it) or the datapath internal port name (e.g. ovs-system).
264 * Returns true if there is a name conflict, false otherwise. */
266 netdev_is_reserved_name(const char *name)
267 OVS_EXCLUDED(netdev_mutex)
269 struct netdev_registered_class *rc;
273 ovs_rwlock_rdlock(&netdev_class_rwlock);
274 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
275 const char *dpif_port = netdev_vport_class_get_dpif_port(rc->class);
276 if (dpif_port && !strcmp(dpif_port, name)) {
277 ovs_rwlock_unlock(&netdev_class_rwlock);
281 ovs_rwlock_unlock(&netdev_class_rwlock);
283 if (!strncmp(name, "ovs-", 4)) {
288 dp_enumerate_types(&types);
289 SSET_FOR_EACH (type, &types) {
290 if (!strcmp(name+4, type)) {
291 sset_destroy(&types);
295 sset_destroy(&types);
301 /* Opens the network device named 'name' (e.g. "eth0") of the specified 'type'
302 * (e.g. "system") and returns zero if successful, otherwise a positive errno
303 * value. On success, sets '*netdevp' to the new network device, otherwise to
306 * Some network devices may need to be configured (with netdev_set_config())
307 * before they can be used. */
309 netdev_open(const char *name, const char *type, struct netdev **netdevp)
310 OVS_EXCLUDED(netdev_mutex)
312 struct netdev *netdev;
317 ovs_rwlock_rdlock(&netdev_class_rwlock);
318 ovs_mutex_lock(&netdev_mutex);
319 netdev = shash_find_data(&netdev_shash, name);
321 struct netdev_registered_class *rc;
323 rc = netdev_lookup_class(type && type[0] ? type : "system");
325 netdev = rc->class->alloc();
327 memset(netdev, 0, sizeof *netdev);
328 netdev->netdev_class = rc->class;
329 netdev->name = xstrdup(name);
330 netdev->node = shash_add(&netdev_shash, name, netdev);
331 list_init(&netdev->saved_flags_list);
333 error = rc->class->construct(netdev);
337 atomic_add(&rc->ref_cnt, 1, &old_ref_cnt);
338 seq_change(connectivity_seq_get());
341 ovs_assert(list_is_empty(&netdev->saved_flags_list));
342 shash_delete(&netdev_shash, netdev->node);
343 rc->class->dealloc(netdev);
349 VLOG_WARN("could not create netdev %s of unknown type %s",
351 error = EAFNOSUPPORT;
357 ovs_mutex_unlock(&netdev_mutex);
358 ovs_rwlock_unlock(&netdev_class_rwlock);
369 /* Returns a reference to 'netdev_' for the caller to own. Returns null if
370 * 'netdev_' is null. */
372 netdev_ref(const struct netdev *netdev_)
373 OVS_EXCLUDED(netdev_mutex)
375 struct netdev *netdev = CONST_CAST(struct netdev *, netdev_);
378 ovs_mutex_lock(&netdev_mutex);
379 ovs_assert(netdev->ref_cnt > 0);
381 ovs_mutex_unlock(&netdev_mutex);
386 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
387 * or NULL if none are needed. */
389 netdev_set_config(struct netdev *netdev, const struct smap *args)
390 OVS_EXCLUDED(netdev_mutex)
392 if (netdev->netdev_class->set_config) {
393 const struct smap no_args = SMAP_INITIALIZER(&no_args);
396 error = netdev->netdev_class->set_config(netdev,
397 args ? args : &no_args);
399 VLOG_WARN("%s: could not set configuration (%s)",
400 netdev_get_name(netdev), ovs_strerror(error));
403 } else if (args && !smap_is_empty(args)) {
404 VLOG_WARN("%s: arguments provided to device that is not configurable",
405 netdev_get_name(netdev));
410 /* Returns the current configuration for 'netdev' in 'args'. The caller must
411 * have already initialized 'args' with smap_init(). Returns 0 on success, in
412 * which case 'args' will be filled with 'netdev''s configuration. On failure
413 * returns a positive errno value, in which case 'args' will be empty.
415 * The caller owns 'args' and its contents and must eventually free them with
418 netdev_get_config(const struct netdev *netdev, struct smap *args)
419 OVS_EXCLUDED(netdev_mutex)
424 if (netdev->netdev_class->get_config) {
425 error = netdev->netdev_class->get_config(netdev, args);
436 const struct netdev_tunnel_config *
437 netdev_get_tunnel_config(const struct netdev *netdev)
438 OVS_EXCLUDED(netdev_mutex)
440 if (netdev->netdev_class->get_tunnel_config) {
441 return netdev->netdev_class->get_tunnel_config(netdev);
448 netdev_unref(struct netdev *dev)
449 OVS_RELEASES(netdev_mutex)
451 ovs_assert(dev->ref_cnt);
452 if (!--dev->ref_cnt) {
453 const struct netdev_class *class = dev->netdev_class;
454 struct netdev_registered_class *rc;
457 dev->netdev_class->destruct(dev);
459 shash_delete(&netdev_shash, dev->node);
461 dev->netdev_class->dealloc(dev);
462 ovs_mutex_unlock(&netdev_mutex);
464 ovs_rwlock_rdlock(&netdev_class_rwlock);
465 rc = netdev_lookup_class(class->type);
466 atomic_sub(&rc->ref_cnt, 1, &old_ref_cnt);
467 ovs_assert(old_ref_cnt > 0);
468 ovs_rwlock_unlock(&netdev_class_rwlock);
470 ovs_mutex_unlock(&netdev_mutex);
474 /* Closes and destroys 'netdev'. */
476 netdev_close(struct netdev *netdev)
477 OVS_EXCLUDED(netdev_mutex)
480 ovs_mutex_lock(&netdev_mutex);
481 netdev_unref(netdev);
485 /* Parses 'netdev_name_', which is of the form [type@]name into its component
486 * pieces. 'name' and 'type' must be freed by the caller. */
488 netdev_parse_name(const char *netdev_name_, char **name, char **type)
490 char *netdev_name = xstrdup(netdev_name_);
493 separator = strchr(netdev_name, '@');
497 *name = xstrdup(separator + 1);
500 *type = xstrdup("system");
505 netdev_rx_open(struct netdev *netdev, struct netdev_rx **rxp)
506 OVS_EXCLUDED(netdev_mutex)
510 if (netdev->netdev_class->rx_alloc) {
511 struct netdev_rx *rx = netdev->netdev_class->rx_alloc();
514 error = netdev->netdev_class->rx_construct(rx);
516 ovs_mutex_lock(&netdev_mutex);
518 ovs_mutex_unlock(&netdev_mutex);
523 netdev->netdev_class->rx_dealloc(rx);
536 netdev_rx_close(struct netdev_rx *rx)
537 OVS_EXCLUDED(netdev_mutex)
540 struct netdev *netdev = rx->netdev;
541 netdev->netdev_class->rx_destruct(rx);
542 netdev->netdev_class->rx_dealloc(rx);
543 netdev_close(netdev);
548 netdev_rx_recv(struct netdev_rx *rx, struct ofpbuf *buffer)
552 ovs_assert(buffer->size == 0);
553 ovs_assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
555 retval = rx->netdev->netdev_class->rx_recv(rx, buffer->data,
556 ofpbuf_tailroom(buffer));
558 COVERAGE_INC(netdev_received);
559 buffer->size += retval;
560 if (buffer->size < ETH_TOTAL_MIN) {
561 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
570 netdev_rx_wait(struct netdev_rx *rx)
572 rx->netdev->netdev_class->rx_wait(rx);
576 netdev_rx_drain(struct netdev_rx *rx)
578 return (rx->netdev->netdev_class->rx_drain
579 ? rx->netdev->netdev_class->rx_drain(rx)
583 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
584 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
585 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
586 * the packet is too big or too small to transmit on the device.
588 * The caller retains ownership of 'buffer' in all cases.
590 * The kernel maintains a packet transmission queue, so the caller is not
591 * expected to do additional queuing of packets.
593 * Some network devices may not implement support for this function. In such
594 * cases this function will always return EOPNOTSUPP. */
596 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
600 error = (netdev->netdev_class->send
601 ? netdev->netdev_class->send(netdev, buffer->data, buffer->size)
604 COVERAGE_INC(netdev_sent);
609 /* Registers with the poll loop to wake up from the next call to poll_block()
610 * when the packet transmission queue has sufficient room to transmit a packet
611 * with netdev_send().
613 * The kernel maintains a packet transmission queue, so the client is not
614 * expected to do additional queuing of packets. Thus, this function is
615 * unlikely to ever be used. It is included for completeness. */
617 netdev_send_wait(struct netdev *netdev)
619 if (netdev->netdev_class->send_wait) {
620 netdev->netdev_class->send_wait(netdev);
624 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
625 * otherwise a positive errno value. */
627 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
629 return netdev->netdev_class->set_etheraddr(netdev, mac);
632 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
633 * the MAC address into 'mac'. On failure, returns a positive errno value and
634 * clears 'mac' to all-zeros. */
636 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
638 return netdev->netdev_class->get_etheraddr(netdev, mac);
641 /* Returns the name of the network device that 'netdev' represents,
642 * e.g. "eth0". The caller must not modify or free the returned string. */
644 netdev_get_name(const struct netdev *netdev)
649 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
650 * (and received) packets, in bytes, not including the hardware header; thus,
651 * this is typically 1500 bytes for Ethernet devices.
653 * If successful, returns 0 and stores the MTU size in '*mtup'. Returns
654 * EOPNOTSUPP if 'netdev' does not have an MTU (as e.g. some tunnels do not).
655 * On other failure, returns a positive errno value. On failure, sets '*mtup'
658 netdev_get_mtu(const struct netdev *netdev, int *mtup)
660 const struct netdev_class *class = netdev->netdev_class;
663 error = class->get_mtu ? class->get_mtu(netdev, mtup) : EOPNOTSUPP;
666 if (error != EOPNOTSUPP) {
667 VLOG_DBG_RL(&rl, "failed to retrieve MTU for network device %s: "
668 "%s", netdev_get_name(netdev), ovs_strerror(error));
674 /* Sets the MTU of 'netdev'. The MTU is the maximum size of transmitted
675 * (and received) packets, in bytes.
677 * If successful, returns 0. Returns EOPNOTSUPP if 'netdev' does not have an
678 * MTU (as e.g. some tunnels do not). On other failure, returns a positive
681 netdev_set_mtu(const struct netdev *netdev, int mtu)
683 const struct netdev_class *class = netdev->netdev_class;
686 error = class->set_mtu ? class->set_mtu(netdev, mtu) : EOPNOTSUPP;
687 if (error && error != EOPNOTSUPP) {
688 VLOG_DBG_RL(&rl, "failed to set MTU for network device %s: %s",
689 netdev_get_name(netdev), ovs_strerror(error));
695 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
696 * failure, returns a negative errno value.
698 * The desired semantics of the ifindex value are a combination of those
699 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
700 * value should be unique within a host and remain stable at least until
701 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
702 * but many systems do not follow this rule anyhow.
704 * Some network devices may not implement support for this function. In such
705 * cases this function will always return -EOPNOTSUPP.
708 netdev_get_ifindex(const struct netdev *netdev)
710 int (*get_ifindex)(const struct netdev *);
712 get_ifindex = netdev->netdev_class->get_ifindex;
714 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
717 /* Stores the features supported by 'netdev' into each of '*current',
718 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
719 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
720 * successful, otherwise a positive errno value. On failure, all of the
721 * passed-in values are set to 0.
723 * Some network devices may not implement support for this function. In such
724 * cases this function will always return EOPNOTSUPP. */
726 netdev_get_features(const struct netdev *netdev,
727 enum netdev_features *current,
728 enum netdev_features *advertised,
729 enum netdev_features *supported,
730 enum netdev_features *peer)
732 int (*get_features)(const struct netdev *netdev,
733 enum netdev_features *current,
734 enum netdev_features *advertised,
735 enum netdev_features *supported,
736 enum netdev_features *peer);
737 enum netdev_features dummy[4];
744 advertised = &dummy[1];
747 supported = &dummy[2];
753 get_features = netdev->netdev_class->get_features;
755 ? get_features(netdev, current, advertised, supported,
759 *current = *advertised = *supported = *peer = 0;
764 /* Returns the maximum speed of a network connection that has the NETDEV_F_*
765 * bits in 'features', in bits per second. If no bits that indicate a speed
766 * are set in 'features', returns 'default_bps'. */
768 netdev_features_to_bps(enum netdev_features features,
769 uint64_t default_bps)
772 F_1000000MB = NETDEV_F_1TB_FD,
773 F_100000MB = NETDEV_F_100GB_FD,
774 F_40000MB = NETDEV_F_40GB_FD,
775 F_10000MB = NETDEV_F_10GB_FD,
776 F_1000MB = NETDEV_F_1GB_HD | NETDEV_F_1GB_FD,
777 F_100MB = NETDEV_F_100MB_HD | NETDEV_F_100MB_FD,
778 F_10MB = NETDEV_F_10MB_HD | NETDEV_F_10MB_FD
781 return ( features & F_1000000MB ? UINT64_C(1000000000000)
782 : features & F_100000MB ? UINT64_C(100000000000)
783 : features & F_40000MB ? UINT64_C(40000000000)
784 : features & F_10000MB ? UINT64_C(10000000000)
785 : features & F_1000MB ? UINT64_C(1000000000)
786 : features & F_100MB ? UINT64_C(100000000)
787 : features & F_10MB ? UINT64_C(10000000)
791 /* Returns true if any of the NETDEV_F_* bits that indicate a full-duplex link
792 * are set in 'features', otherwise false. */
794 netdev_features_is_full_duplex(enum netdev_features features)
796 return (features & (NETDEV_F_10MB_FD | NETDEV_F_100MB_FD | NETDEV_F_1GB_FD
797 | NETDEV_F_10GB_FD | NETDEV_F_40GB_FD
798 | NETDEV_F_100GB_FD | NETDEV_F_1TB_FD)) != 0;
801 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
802 * successful, otherwise a positive errno value. */
804 netdev_set_advertisements(struct netdev *netdev,
805 enum netdev_features advertise)
807 return (netdev->netdev_class->set_advertisements
808 ? netdev->netdev_class->set_advertisements(
813 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
814 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
815 * errno value and sets '*address' to 0 (INADDR_ANY).
817 * The following error values have well-defined meanings:
819 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
821 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
823 * 'address' or 'netmask' or both may be null, in which case the address or
824 * netmask is not reported. */
826 netdev_get_in4(const struct netdev *netdev,
827 struct in_addr *address_, struct in_addr *netmask_)
829 struct in_addr address;
830 struct in_addr netmask;
833 error = (netdev->netdev_class->get_in4
834 ? netdev->netdev_class->get_in4(netdev,
838 address_->s_addr = error ? 0 : address.s_addr;
841 netmask_->s_addr = error ? 0 : netmask.s_addr;
846 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
847 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
848 * positive errno value. */
850 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
852 return (netdev->netdev_class->set_in4
853 ? netdev->netdev_class->set_in4(netdev, addr, mask)
857 /* Obtains ad IPv4 address from device name and save the address in
858 * in4. Returns 0 if successful, otherwise a positive errno value.
861 netdev_get_in4_by_name(const char *device_name, struct in_addr *in4)
863 struct netdev *netdev;
866 error = netdev_open(device_name, "system", &netdev);
868 in4->s_addr = htonl(0);
872 error = netdev_get_in4(netdev, in4, NULL);
873 netdev_close(netdev);
877 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
880 netdev_add_router(struct netdev *netdev, struct in_addr router)
882 COVERAGE_INC(netdev_add_router);
883 return (netdev->netdev_class->add_router
884 ? netdev->netdev_class->add_router(netdev, router)
888 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
889 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
890 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
891 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
892 * a directly connected network) in '*next_hop' and a copy of the name of the
893 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
894 * responsible for freeing '*netdev_name' (by calling free()). */
896 netdev_get_next_hop(const struct netdev *netdev,
897 const struct in_addr *host, struct in_addr *next_hop,
900 int error = (netdev->netdev_class->get_next_hop
901 ? netdev->netdev_class->get_next_hop(
902 host, next_hop, netdev_name)
905 next_hop->s_addr = 0;
911 /* Populates 'smap' with status information.
913 * Populates 'smap' with 'netdev' specific status information. This
914 * information may be used to populate the status column of the Interface table
915 * as defined in ovs-vswitchd.conf.db(5). */
917 netdev_get_status(const struct netdev *netdev, struct smap *smap)
919 return (netdev->netdev_class->get_status
920 ? netdev->netdev_class->get_status(netdev, smap)
924 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
925 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
926 * all-zero-bits (in6addr_any).
928 * The following error values have well-defined meanings:
930 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
932 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
934 * 'in6' may be null, in which case the address itself is not reported. */
936 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
938 struct in6_addr dummy;
941 error = (netdev->netdev_class->get_in6
942 ? netdev->netdev_class->get_in6(netdev,
946 memset(in6, 0, sizeof *in6);
951 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
952 * 'on'. Returns 0 if successful, otherwise a positive errno value. */
954 do_update_flags(struct netdev *netdev, enum netdev_flags off,
955 enum netdev_flags on, enum netdev_flags *old_flagsp,
956 struct netdev_saved_flags **sfp)
957 OVS_EXCLUDED(netdev_mutex)
959 struct netdev_saved_flags *sf = NULL;
960 enum netdev_flags old_flags;
963 error = netdev->netdev_class->update_flags(netdev, off & ~on, on,
966 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
967 off || on ? "set" : "get", netdev_get_name(netdev),
968 ovs_strerror(error));
970 } else if ((off || on) && sfp) {
971 enum netdev_flags new_flags = (old_flags & ~off) | on;
972 enum netdev_flags changed_flags = old_flags ^ new_flags;
974 ovs_mutex_lock(&netdev_mutex);
975 *sfp = sf = xmalloc(sizeof *sf);
977 list_push_front(&netdev->saved_flags_list, &sf->node);
978 sf->saved_flags = changed_flags;
979 sf->saved_values = changed_flags & new_flags;
982 ovs_mutex_unlock(&netdev_mutex);
987 *old_flagsp = old_flags;
996 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
997 * Returns 0 if successful, otherwise a positive errno value. On failure,
998 * stores 0 into '*flagsp'. */
1000 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
1002 struct netdev *netdev = CONST_CAST(struct netdev *, netdev_);
1003 return do_update_flags(netdev, 0, 0, flagsp, NULL);
1006 /* Sets the flags for 'netdev' to 'flags'.
1007 * Returns 0 if successful, otherwise a positive errno value. */
1009 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
1010 struct netdev_saved_flags **sfp)
1012 return do_update_flags(netdev, -1, flags, NULL, sfp);
1015 /* Turns on the specified 'flags' on 'netdev':
1017 * - On success, returns 0. If 'sfp' is nonnull, sets '*sfp' to a newly
1018 * allocated 'struct netdev_saved_flags *' that may be passed to
1019 * netdev_restore_flags() to restore the original values of 'flags' on
1020 * 'netdev' (this will happen automatically at program termination if
1021 * netdev_restore_flags() is never called) , or to NULL if no flags were
1024 * - On failure, returns a positive errno value. If 'sfp' is nonnull, sets
1025 * '*sfp' to NULL. */
1027 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
1028 struct netdev_saved_flags **sfp)
1030 return do_update_flags(netdev, 0, flags, NULL, sfp);
1033 /* Turns off the specified 'flags' on 'netdev'. See netdev_turn_flags_on() for
1034 * details of the interface. */
1036 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
1037 struct netdev_saved_flags **sfp)
1039 return do_update_flags(netdev, flags, 0, NULL, sfp);
1042 /* Restores the flags that were saved in 'sf', and destroys 'sf'.
1043 * Does nothing if 'sf' is NULL. */
1045 netdev_restore_flags(struct netdev_saved_flags *sf)
1046 OVS_EXCLUDED(netdev_mutex)
1049 struct netdev *netdev = sf->netdev;
1050 enum netdev_flags old_flags;
1052 netdev->netdev_class->update_flags(netdev,
1053 sf->saved_flags & sf->saved_values,
1054 sf->saved_flags & ~sf->saved_values,
1057 ovs_mutex_lock(&netdev_mutex);
1058 list_remove(&sf->node);
1060 netdev_unref(netdev);
1064 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1065 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1066 * returns 0. Otherwise, it returns a positive errno value; in particular,
1067 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
1069 netdev_arp_lookup(const struct netdev *netdev,
1070 ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
1072 int error = (netdev->netdev_class->arp_lookup
1073 ? netdev->netdev_class->arp_lookup(netdev, ip, mac)
1076 memset(mac, 0, ETH_ADDR_LEN);
1081 /* Returns true if carrier is active (link light is on) on 'netdev'. */
1083 netdev_get_carrier(const struct netdev *netdev)
1086 enum netdev_flags flags;
1089 netdev_get_flags(netdev, &flags);
1090 if (!(flags & NETDEV_UP)) {
1094 if (!netdev->netdev_class->get_carrier) {
1098 error = netdev->netdev_class->get_carrier(netdev, &carrier);
1100 VLOG_DBG("%s: failed to get network device carrier status, assuming "
1101 "down: %s", netdev_get_name(netdev), ovs_strerror(error));
1108 /* Returns the number of times 'netdev''s carrier has changed. */
1110 netdev_get_carrier_resets(const struct netdev *netdev)
1112 return (netdev->netdev_class->get_carrier_resets
1113 ? netdev->netdev_class->get_carrier_resets(netdev)
1117 /* Attempts to force netdev_get_carrier() to poll 'netdev''s MII registers for
1118 * link status instead of checking 'netdev''s carrier. 'netdev''s MII
1119 * registers will be polled once ever 'interval' milliseconds. If 'netdev'
1120 * does not support MII, another method may be used as a fallback. If
1121 * 'interval' is less than or equal to zero, reverts netdev_get_carrier() to
1122 * its normal behavior.
1124 * Returns 0 if successful, otherwise a positive errno value. */
1126 netdev_set_miimon_interval(struct netdev *netdev, long long int interval)
1128 return (netdev->netdev_class->set_miimon_interval
1129 ? netdev->netdev_class->set_miimon_interval(netdev, interval)
1133 /* Retrieves current device stats for 'netdev'. */
1135 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
1139 COVERAGE_INC(netdev_get_stats);
1140 error = (netdev->netdev_class->get_stats
1141 ? netdev->netdev_class->get_stats(netdev, stats)
1144 memset(stats, 0xff, sizeof *stats);
1149 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
1150 * Returns 0 if successful, otherwise a positive errno value.
1152 * This will probably fail for most network devices. Some devices might only
1153 * allow setting their stats to 0. */
1155 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
1157 return (netdev->netdev_class->set_stats
1158 ? netdev->netdev_class->set_stats(netdev, stats)
1162 /* Attempts to set input rate limiting (policing) policy, such that up to
1163 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
1164 * size of 'kbits' kb. */
1166 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
1167 uint32_t kbits_burst)
1169 return (netdev->netdev_class->set_policing
1170 ? netdev->netdev_class->set_policing(netdev,
1171 kbits_rate, kbits_burst)
1175 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
1176 * empty if 'netdev' does not support QoS. Any names added to 'types' should
1177 * be documented as valid for the "type" column in the "QoS" table in
1178 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1180 * Every network device supports disabling QoS with a type of "", but this type
1181 * will not be added to 'types'.
1183 * The caller must initialize 'types' (e.g. with sset_init()) before calling
1184 * this function. The caller is responsible for destroying 'types' (e.g. with
1185 * sset_destroy()) when it is no longer needed.
1187 * Returns 0 if successful, otherwise a positive errno value. */
1189 netdev_get_qos_types(const struct netdev *netdev, struct sset *types)
1191 const struct netdev_class *class = netdev->netdev_class;
1192 return (class->get_qos_types
1193 ? class->get_qos_types(netdev, types)
1197 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
1198 * which should be "" or one of the types returned by netdev_get_qos_types()
1199 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
1200 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
1201 * 'caps' to all zeros. */
1203 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
1204 struct netdev_qos_capabilities *caps)
1206 const struct netdev_class *class = netdev->netdev_class;
1209 int retval = (class->get_qos_capabilities
1210 ? class->get_qos_capabilities(netdev, type, caps)
1213 memset(caps, 0, sizeof *caps);
1217 /* Every netdev supports turning off QoS. */
1218 memset(caps, 0, sizeof *caps);
1223 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1224 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1225 * the number of queues (zero on failure) in '*n_queuesp'.
1227 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1229 netdev_get_n_queues(const struct netdev *netdev,
1230 const char *type, unsigned int *n_queuesp)
1232 struct netdev_qos_capabilities caps;
1235 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1236 *n_queuesp = caps.n_queues;
1240 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1241 * stores the name of the current form of QoS into '*typep', stores any details
1242 * of configuration as string key-value pairs in 'details', and returns 0. On
1243 * failure, sets '*typep' to NULL and returns a positive errno value.
1245 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1247 * The caller must initialize 'details' as an empty smap (e.g. with
1248 * smap_init()) before calling this function. The caller must free 'details'
1249 * when it is no longer needed (e.g. with smap_destroy()).
1251 * The caller must not modify or free '*typep'.
1253 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1254 * 'netdev'. The contents of 'details' should be documented as valid for
1255 * '*typep' in the "other_config" column in the "QoS" table in
1256 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1258 netdev_get_qos(const struct netdev *netdev,
1259 const char **typep, struct smap *details)
1261 const struct netdev_class *class = netdev->netdev_class;
1264 if (class->get_qos) {
1265 retval = class->get_qos(netdev, typep, details);
1268 smap_clear(details);
1272 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1278 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1279 * with details of configuration from 'details'. Returns 0 if successful,
1280 * otherwise a positive errno value. On error, the previous QoS configuration
1283 * When this function changes the type of QoS (not just 'details'), this also
1284 * resets all queue configuration for 'netdev' to their defaults (which depend
1285 * on the specific type of QoS). Otherwise, the queue configuration for
1286 * 'netdev' is unchanged.
1288 * 'type' should be "" (to disable QoS) or one of the types returned by
1289 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1290 * documented as valid for the given 'type' in the "other_config" column in the
1291 * "QoS" table in vswitchd/vswitch.xml (which is built as
1292 * ovs-vswitchd.conf.db(8)).
1294 * NULL may be specified for 'details' if there are no configuration
1297 netdev_set_qos(struct netdev *netdev,
1298 const char *type, const struct smap *details)
1300 const struct netdev_class *class = netdev->netdev_class;
1306 if (class->set_qos) {
1308 static const struct smap empty = SMAP_INITIALIZER(&empty);
1311 return class->set_qos(netdev, type, details);
1313 return *type ? EOPNOTSUPP : 0;
1317 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1318 * successful, adds that information as string key-value pairs to 'details'.
1319 * Returns 0 if successful, otherwise a positive errno value.
1321 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1322 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1324 * The returned contents of 'details' should be documented as valid for the
1325 * given 'type' in the "other_config" column in the "Queue" table in
1326 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1328 * The caller must initialize 'details' (e.g. with smap_init()) before calling
1329 * this function. The caller must free 'details' when it is no longer needed
1330 * (e.g. with smap_destroy()). */
1332 netdev_get_queue(const struct netdev *netdev,
1333 unsigned int queue_id, struct smap *details)
1335 const struct netdev_class *class = netdev->netdev_class;
1338 retval = (class->get_queue
1339 ? class->get_queue(netdev, queue_id, details)
1342 smap_clear(details);
1347 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1348 * string pairs in 'details'. The contents of 'details' should be documented
1349 * as valid for the given 'type' in the "other_config" column in the "Queue"
1350 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1351 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1352 * given queue's configuration should be unmodified.
1354 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1355 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1357 * This function does not modify 'details', and the caller retains ownership of
1360 netdev_set_queue(struct netdev *netdev,
1361 unsigned int queue_id, const struct smap *details)
1363 const struct netdev_class *class = netdev->netdev_class;
1364 return (class->set_queue
1365 ? class->set_queue(netdev, queue_id, details)
1369 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1370 * of QoS may have a fixed set of queues, in which case attempts to delete them
1371 * will fail with EOPNOTSUPP.
1373 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1374 * given queue will be unmodified.
1376 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1377 * the current form of QoS (e.g. as returned by
1378 * netdev_get_n_queues(netdev)). */
1380 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1382 const struct netdev_class *class = netdev->netdev_class;
1383 return (class->delete_queue
1384 ? class->delete_queue(netdev, queue_id)
1388 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1389 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1390 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1391 * positive errno value and fills 'stats' with values indicating unsupported
1394 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1395 struct netdev_queue_stats *stats)
1397 const struct netdev_class *class = netdev->netdev_class;
1400 retval = (class->get_queue_stats
1401 ? class->get_queue_stats(netdev, queue_id, stats)
1404 stats->tx_bytes = UINT64_MAX;
1405 stats->tx_packets = UINT64_MAX;
1406 stats->tx_errors = UINT64_MAX;
1407 stats->created = LLONG_MIN;
1412 /* Initializes 'dump' to begin dumping the queues in a netdev.
1414 * This function provides no status indication. An error status for the entire
1415 * dump operation is provided when it is completed by calling
1416 * netdev_queue_dump_done().
1419 netdev_queue_dump_start(struct netdev_queue_dump *dump,
1420 const struct netdev *netdev)
1422 dump->netdev = netdev_ref(netdev);
1423 if (netdev->netdev_class->queue_dump_start) {
1424 dump->error = netdev->netdev_class->queue_dump_start(netdev,
1427 dump->error = EOPNOTSUPP;
1431 /* Attempts to retrieve another queue from 'dump', which must have been
1432 * initialized with netdev_queue_dump_start(). On success, stores a new queue
1433 * ID into '*queue_id', fills 'details' with configuration details for the
1434 * queue, and returns true. On failure, returns false.
1436 * Queues are not necessarily dumped in increasing order of queue ID (or any
1437 * other predictable order).
1439 * Failure might indicate an actual error or merely that the last queue has
1440 * been dumped. An error status for the entire dump operation is provided when
1441 * it is completed by calling netdev_queue_dump_done().
1443 * The returned contents of 'details' should be documented as valid for the
1444 * given 'type' in the "other_config" column in the "Queue" table in
1445 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1447 * The caller must initialize 'details' (e.g. with smap_init()) before calling
1448 * this function. This function will clear and replace its contents. The
1449 * caller must free 'details' when it is no longer needed (e.g. with
1450 * smap_destroy()). */
1452 netdev_queue_dump_next(struct netdev_queue_dump *dump,
1453 unsigned int *queue_id, struct smap *details)
1455 const struct netdev *netdev = dump->netdev;
1461 dump->error = netdev->netdev_class->queue_dump_next(netdev, dump->state,
1465 netdev->netdev_class->queue_dump_done(netdev, dump->state);
1471 /* Completes queue table dump operation 'dump', which must have been
1472 * initialized with netdev_queue_dump_start(). Returns 0 if the dump operation
1473 * was error-free, otherwise a positive errno value describing the problem. */
1475 netdev_queue_dump_done(struct netdev_queue_dump *dump)
1477 const struct netdev *netdev = dump->netdev;
1478 if (!dump->error && netdev->netdev_class->queue_dump_done) {
1479 dump->error = netdev->netdev_class->queue_dump_done(netdev,
1482 netdev_close(dump->netdev);
1483 return dump->error == EOF ? 0 : dump->error;
1486 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1487 * its statistics, and the 'aux' specified by the caller. The order of
1488 * iteration is unspecified, but (when successful) each queue is visited
1491 * Calling this function may be more efficient than calling
1492 * netdev_get_queue_stats() for every queue.
1494 * 'cb' must not modify or free the statistics passed in.
1496 * Returns 0 if successful, otherwise a positive errno value. On error, some
1497 * configured queues may not have been included in the iteration. */
1499 netdev_dump_queue_stats(const struct netdev *netdev,
1500 netdev_dump_queue_stats_cb *cb, void *aux)
1502 const struct netdev_class *class = netdev->netdev_class;
1503 return (class->dump_queue_stats
1504 ? class->dump_queue_stats(netdev, cb, aux)
1509 /* Returns the class type of 'netdev'.
1511 * The caller must not free the returned value. */
1513 netdev_get_type(const struct netdev *netdev)
1515 return netdev->netdev_class->type;
1518 /* Returns the class associated with 'netdev'. */
1519 const struct netdev_class *
1520 netdev_get_class(const struct netdev *netdev)
1522 return netdev->netdev_class;
1525 /* Returns the netdev with 'name' or NULL if there is none.
1527 * The caller must free the returned netdev with netdev_close(). */
1529 netdev_from_name(const char *name)
1530 OVS_EXCLUDED(netdev_mutex)
1532 struct netdev *netdev;
1534 ovs_mutex_lock(&netdev_mutex);
1535 netdev = shash_find_data(&netdev_shash, name);
1539 ovs_mutex_unlock(&netdev_mutex);
1544 /* Fills 'device_list' with devices that match 'netdev_class'.
1546 * The caller is responsible for initializing and destroying 'device_list' and
1547 * must close each device on the list. */
1549 netdev_get_devices(const struct netdev_class *netdev_class,
1550 struct shash *device_list)
1551 OVS_EXCLUDED(netdev_mutex)
1553 struct shash_node *node;
1555 ovs_mutex_lock(&netdev_mutex);
1556 SHASH_FOR_EACH (node, &netdev_shash) {
1557 struct netdev *dev = node->data;
1559 if (dev->netdev_class == netdev_class) {
1561 shash_add(device_list, node->name, node->data);
1564 ovs_mutex_unlock(&netdev_mutex);
1568 netdev_get_type_from_name(const char *name)
1570 struct netdev *dev = netdev_from_name(name);
1571 const char *type = dev ? netdev_get_type(dev) : NULL;
1577 netdev_rx_get_netdev(const struct netdev_rx *rx)
1579 ovs_assert(rx->netdev->ref_cnt > 0);
1584 netdev_rx_get_name(const struct netdev_rx *rx)
1586 return netdev_get_name(netdev_rx_get_netdev(rx));
1590 restore_all_flags(void *aux OVS_UNUSED)
1592 struct shash_node *node;
1594 SHASH_FOR_EACH (node, &netdev_shash) {
1595 struct netdev *netdev = node->data;
1596 const struct netdev_saved_flags *sf;
1597 enum netdev_flags saved_values;
1598 enum netdev_flags saved_flags;
1600 saved_values = saved_flags = 0;
1601 LIST_FOR_EACH (sf, node, &netdev->saved_flags_list) {
1602 saved_flags |= sf->saved_flags;
1603 saved_values &= ~sf->saved_flags;
1604 saved_values |= sf->saved_flags & sf->saved_values;
1607 enum netdev_flags old_flags;
1609 netdev->netdev_class->update_flags(netdev,
1610 saved_flags & saved_values,
1611 saved_flags & ~saved_values,