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);
113 ovsthread_once_done(&once);
117 /* Performs periodic work needed by all the various kinds of netdevs.
119 * If your program opens any netdevs, it must call this function within its
123 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
125 struct netdev_registered_class *rc;
127 ovs_rwlock_rdlock(&netdev_class_rwlock);
128 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
129 if (rc->class->run) {
133 ovs_rwlock_unlock(&netdev_class_rwlock);
136 /* Arranges for poll_block() to wake up when netdev_run() needs to be called.
138 * If your program opens any netdevs, it must call this function within its
142 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
144 struct netdev_registered_class *rc;
146 ovs_rwlock_rdlock(&netdev_class_rwlock);
147 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
148 if (rc->class->wait) {
152 ovs_rwlock_unlock(&netdev_class_rwlock);
155 static struct netdev_registered_class *
156 netdev_lookup_class(const char *type)
157 OVS_REQ_RDLOCK(netdev_class_rwlock)
159 struct netdev_registered_class *rc;
161 HMAP_FOR_EACH_WITH_HASH (rc, hmap_node, hash_string(type, 0),
163 if (!strcmp(type, rc->class->type)) {
170 /* Initializes and registers a new netdev provider. After successful
171 * registration, new netdevs of that type can be opened using netdev_open(). */
173 netdev_register_provider(const struct netdev_class *new_class)
174 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
178 ovs_rwlock_wrlock(&netdev_class_rwlock);
179 if (netdev_lookup_class(new_class->type)) {
180 VLOG_WARN("attempted to register duplicate netdev provider: %s",
184 error = new_class->init ? new_class->init() : 0;
186 struct netdev_registered_class *rc;
188 rc = xmalloc(sizeof *rc);
189 hmap_insert(&netdev_classes, &rc->hmap_node,
190 hash_string(new_class->type, 0));
191 rc->class = new_class;
192 atomic_init(&rc->ref_cnt, 0);
194 VLOG_ERR("failed to initialize %s network device class: %s",
195 new_class->type, ovs_strerror(error));
198 ovs_rwlock_unlock(&netdev_class_rwlock);
203 /* Unregisters a netdev provider. 'type' must have been previously
204 * registered and not currently be in use by any netdevs. After unregistration
205 * new netdevs of that type cannot be opened using netdev_open(). */
207 netdev_unregister_provider(const char *type)
208 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
210 struct netdev_registered_class *rc;
213 ovs_rwlock_wrlock(&netdev_class_rwlock);
214 rc = netdev_lookup_class(type);
216 VLOG_WARN("attempted to unregister a netdev provider that is not "
217 "registered: %s", type);
218 error = EAFNOSUPPORT;
222 atomic_read(&rc->ref_cnt, &ref_cnt);
224 hmap_remove(&netdev_classes, &rc->hmap_node);
228 VLOG_WARN("attempted to unregister in use netdev provider: %s",
233 ovs_rwlock_unlock(&netdev_class_rwlock);
238 /* Clears 'types' and enumerates the types of all currently registered netdev
239 * providers into it. The caller must first initialize the sset. */
241 netdev_enumerate_types(struct sset *types)
242 OVS_EXCLUDED(netdev_mutex)
244 struct netdev_registered_class *rc;
249 ovs_rwlock_rdlock(&netdev_class_rwlock);
250 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
251 sset_add(types, rc->class->type);
253 ovs_rwlock_unlock(&netdev_class_rwlock);
256 /* Check that the network device name is not the same as any of the registered
257 * vport providers' dpif_port name (dpif_port is NULL if the vport provider
258 * does not define it) or the datapath internal port name (e.g. ovs-system).
260 * Returns true if there is a name conflict, false otherwise. */
262 netdev_is_reserved_name(const char *name)
263 OVS_EXCLUDED(netdev_mutex)
265 struct netdev_registered_class *rc;
269 ovs_rwlock_rdlock(&netdev_class_rwlock);
270 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
271 const char *dpif_port = netdev_vport_class_get_dpif_port(rc->class);
272 if (dpif_port && !strcmp(dpif_port, name)) {
273 ovs_rwlock_unlock(&netdev_class_rwlock);
277 ovs_rwlock_unlock(&netdev_class_rwlock);
279 if (!strncmp(name, "ovs-", 4)) {
284 dp_enumerate_types(&types);
285 SSET_FOR_EACH (type, &types) {
286 if (!strcmp(name+4, type)) {
287 sset_destroy(&types);
291 sset_destroy(&types);
297 /* Opens the network device named 'name' (e.g. "eth0") of the specified 'type'
298 * (e.g. "system") and returns zero if successful, otherwise a positive errno
299 * value. On success, sets '*netdevp' to the new network device, otherwise to
302 * Some network devices may need to be configured (with netdev_set_config())
303 * before they can be used. */
305 netdev_open(const char *name, const char *type, struct netdev **netdevp)
306 OVS_EXCLUDED(netdev_mutex)
308 struct netdev *netdev;
313 ovs_rwlock_rdlock(&netdev_class_rwlock);
314 ovs_mutex_lock(&netdev_mutex);
315 netdev = shash_find_data(&netdev_shash, name);
317 struct netdev_registered_class *rc;
319 rc = netdev_lookup_class(type && type[0] ? type : "system");
321 netdev = rc->class->alloc();
323 memset(netdev, 0, sizeof *netdev);
324 netdev->netdev_class = rc->class;
325 netdev->name = xstrdup(name);
326 netdev->node = shash_add(&netdev_shash, name, netdev);
327 list_init(&netdev->saved_flags_list);
329 error = rc->class->construct(netdev);
333 atomic_add(&rc->ref_cnt, 1, &old_ref_cnt);
336 ovs_assert(list_is_empty(&netdev->saved_flags_list));
337 shash_delete(&netdev_shash, netdev->node);
338 rc->class->dealloc(netdev);
344 VLOG_WARN("could not create netdev %s of unknown type %s",
346 error = EAFNOSUPPORT;
352 ovs_mutex_unlock(&netdev_mutex);
353 ovs_rwlock_unlock(&netdev_class_rwlock);
364 /* Returns a reference to 'netdev_' for the caller to own. Returns null if
365 * 'netdev_' is null. */
367 netdev_ref(const struct netdev *netdev_)
368 OVS_EXCLUDED(netdev_mutex)
370 struct netdev *netdev = CONST_CAST(struct netdev *, netdev_);
373 ovs_mutex_lock(&netdev_mutex);
374 ovs_assert(netdev->ref_cnt > 0);
376 ovs_mutex_unlock(&netdev_mutex);
381 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
382 * or NULL if none are needed. */
384 netdev_set_config(struct netdev *netdev, const struct smap *args)
385 OVS_EXCLUDED(netdev_mutex)
387 if (netdev->netdev_class->set_config) {
388 const struct smap no_args = SMAP_INITIALIZER(&no_args);
389 return netdev->netdev_class->set_config(netdev,
390 args ? args : &no_args);
391 } else if (args && !smap_is_empty(args)) {
392 VLOG_WARN("%s: arguments provided to device that is not configurable",
393 netdev_get_name(netdev));
399 /* Returns the current configuration for 'netdev' in 'args'. The caller must
400 * have already initialized 'args' with smap_init(). Returns 0 on success, in
401 * which case 'args' will be filled with 'netdev''s configuration. On failure
402 * returns a positive errno value, in which case 'args' will be empty.
404 * The caller owns 'args' and its contents and must eventually free them with
407 netdev_get_config(const struct netdev *netdev, struct smap *args)
408 OVS_EXCLUDED(netdev_mutex)
413 if (netdev->netdev_class->get_config) {
414 error = netdev->netdev_class->get_config(netdev, args);
425 const struct netdev_tunnel_config *
426 netdev_get_tunnel_config(const struct netdev *netdev)
427 OVS_EXCLUDED(netdev_mutex)
429 if (netdev->netdev_class->get_tunnel_config) {
430 return netdev->netdev_class->get_tunnel_config(netdev);
437 netdev_unref(struct netdev *dev)
438 OVS_RELEASES(netdev_mutex)
440 ovs_assert(dev->ref_cnt);
441 if (!--dev->ref_cnt) {
442 const struct netdev_class *class = dev->netdev_class;
443 struct netdev_registered_class *rc;
446 dev->netdev_class->destruct(dev);
448 shash_delete(&netdev_shash, dev->node);
450 dev->netdev_class->dealloc(dev);
451 ovs_mutex_unlock(&netdev_mutex);
453 ovs_rwlock_rdlock(&netdev_class_rwlock);
454 rc = netdev_lookup_class(class->type);
455 atomic_sub(&rc->ref_cnt, 1, &old_ref_cnt);
456 ovs_assert(old_ref_cnt > 0);
457 ovs_rwlock_unlock(&netdev_class_rwlock);
459 ovs_mutex_unlock(&netdev_mutex);
463 /* Closes and destroys 'netdev'. */
465 netdev_close(struct netdev *netdev)
466 OVS_EXCLUDED(netdev_mutex)
469 ovs_mutex_lock(&netdev_mutex);
470 netdev_unref(netdev);
474 /* Parses 'netdev_name_', which is of the form [type@]name into its component
475 * pieces. 'name' and 'type' must be freed by the caller. */
477 netdev_parse_name(const char *netdev_name_, char **name, char **type)
479 char *netdev_name = xstrdup(netdev_name_);
482 separator = strchr(netdev_name, '@');
486 *name = xstrdup(separator + 1);
489 *type = xstrdup("system");
494 netdev_rx_open(struct netdev *netdev, struct netdev_rx **rxp)
495 OVS_EXCLUDED(netdev_mutex)
499 if (netdev->netdev_class->rx_alloc) {
500 struct netdev_rx *rx = netdev->netdev_class->rx_alloc();
503 error = netdev->netdev_class->rx_construct(rx);
505 ovs_mutex_lock(&netdev_mutex);
507 ovs_mutex_unlock(&netdev_mutex);
512 netdev->netdev_class->rx_dealloc(rx);
525 netdev_rx_close(struct netdev_rx *rx)
526 OVS_EXCLUDED(netdev_mutex)
529 struct netdev *netdev = rx->netdev;
530 netdev->netdev_class->rx_destruct(rx);
531 netdev->netdev_class->rx_dealloc(rx);
532 netdev_close(netdev);
537 netdev_rx_recv(struct netdev_rx *rx, struct ofpbuf *buffer)
541 ovs_assert(buffer->size == 0);
542 ovs_assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
544 retval = rx->netdev->netdev_class->rx_recv(rx, buffer->data,
545 ofpbuf_tailroom(buffer));
547 COVERAGE_INC(netdev_received);
548 buffer->size += retval;
549 if (buffer->size < ETH_TOTAL_MIN) {
550 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
559 netdev_rx_wait(struct netdev_rx *rx)
561 rx->netdev->netdev_class->rx_wait(rx);
565 netdev_rx_drain(struct netdev_rx *rx)
567 return (rx->netdev->netdev_class->rx_drain
568 ? rx->netdev->netdev_class->rx_drain(rx)
572 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
573 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
574 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
575 * the packet is too big or too small to transmit on the device.
577 * The caller retains ownership of 'buffer' in all cases.
579 * The kernel maintains a packet transmission queue, so the caller is not
580 * expected to do additional queuing of packets.
582 * Some network devices may not implement support for this function. In such
583 * cases this function will always return EOPNOTSUPP. */
585 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
589 error = (netdev->netdev_class->send
590 ? netdev->netdev_class->send(netdev, buffer->data, buffer->size)
593 COVERAGE_INC(netdev_sent);
598 /* Registers with the poll loop to wake up from the next call to poll_block()
599 * when the packet transmission queue has sufficient room to transmit a packet
600 * with netdev_send().
602 * The kernel maintains a packet transmission queue, so the client is not
603 * expected to do additional queuing of packets. Thus, this function is
604 * unlikely to ever be used. It is included for completeness. */
606 netdev_send_wait(struct netdev *netdev)
608 if (netdev->netdev_class->send_wait) {
609 netdev->netdev_class->send_wait(netdev);
613 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
614 * otherwise a positive errno value. */
616 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
618 return netdev->netdev_class->set_etheraddr(netdev, mac);
621 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
622 * the MAC address into 'mac'. On failure, returns a positive errno value and
623 * clears 'mac' to all-zeros. */
625 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
627 return netdev->netdev_class->get_etheraddr(netdev, mac);
630 /* Returns the name of the network device that 'netdev' represents,
631 * e.g. "eth0". The caller must not modify or free the returned string. */
633 netdev_get_name(const struct netdev *netdev)
638 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
639 * (and received) packets, in bytes, not including the hardware header; thus,
640 * this is typically 1500 bytes for Ethernet devices.
642 * If successful, returns 0 and stores the MTU size in '*mtup'. Returns
643 * EOPNOTSUPP if 'netdev' does not have an MTU (as e.g. some tunnels do not).
644 * On other failure, returns a positive errno value. On failure, sets '*mtup'
647 netdev_get_mtu(const struct netdev *netdev, int *mtup)
649 const struct netdev_class *class = netdev->netdev_class;
652 error = class->get_mtu ? class->get_mtu(netdev, mtup) : EOPNOTSUPP;
655 if (error != EOPNOTSUPP) {
656 VLOG_DBG_RL(&rl, "failed to retrieve MTU for network device %s: "
657 "%s", netdev_get_name(netdev), ovs_strerror(error));
663 /* Sets the MTU of 'netdev'. The MTU is the maximum size of transmitted
664 * (and received) packets, in bytes.
666 * If successful, returns 0. Returns EOPNOTSUPP if 'netdev' does not have an
667 * MTU (as e.g. some tunnels do not). On other failure, returns a positive
670 netdev_set_mtu(const struct netdev *netdev, int mtu)
672 const struct netdev_class *class = netdev->netdev_class;
675 error = class->set_mtu ? class->set_mtu(netdev, mtu) : EOPNOTSUPP;
676 if (error && error != EOPNOTSUPP) {
677 VLOG_DBG_RL(&rl, "failed to set MTU for network device %s: %s",
678 netdev_get_name(netdev), ovs_strerror(error));
684 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
685 * failure, returns a negative errno value.
687 * The desired semantics of the ifindex value are a combination of those
688 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
689 * value should be unique within a host and remain stable at least until
690 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
691 * but many systems do not follow this rule anyhow.
693 * Some network devices may not implement support for this function. In such
694 * cases this function will always return -EOPNOTSUPP.
697 netdev_get_ifindex(const struct netdev *netdev)
699 int (*get_ifindex)(const struct netdev *);
701 get_ifindex = netdev->netdev_class->get_ifindex;
703 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
706 /* Stores the features supported by 'netdev' into each of '*current',
707 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
708 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
709 * successful, otherwise a positive errno value. On failure, all of the
710 * passed-in values are set to 0.
712 * Some network devices may not implement support for this function. In such
713 * cases this function will always return EOPNOTSUPP. */
715 netdev_get_features(const struct netdev *netdev,
716 enum netdev_features *current,
717 enum netdev_features *advertised,
718 enum netdev_features *supported,
719 enum netdev_features *peer)
721 int (*get_features)(const struct netdev *netdev,
722 enum netdev_features *current,
723 enum netdev_features *advertised,
724 enum netdev_features *supported,
725 enum netdev_features *peer);
726 enum netdev_features dummy[4];
733 advertised = &dummy[1];
736 supported = &dummy[2];
742 get_features = netdev->netdev_class->get_features;
744 ? get_features(netdev, current, advertised, supported,
748 *current = *advertised = *supported = *peer = 0;
753 /* Returns the maximum speed of a network connection that has the NETDEV_F_*
754 * bits in 'features', in bits per second. If no bits that indicate a speed
755 * are set in 'features', returns 'default_bps'. */
757 netdev_features_to_bps(enum netdev_features features,
758 uint64_t default_bps)
761 F_1000000MB = NETDEV_F_1TB_FD,
762 F_100000MB = NETDEV_F_100GB_FD,
763 F_40000MB = NETDEV_F_40GB_FD,
764 F_10000MB = NETDEV_F_10GB_FD,
765 F_1000MB = NETDEV_F_1GB_HD | NETDEV_F_1GB_FD,
766 F_100MB = NETDEV_F_100MB_HD | NETDEV_F_100MB_FD,
767 F_10MB = NETDEV_F_10MB_HD | NETDEV_F_10MB_FD
770 return ( features & F_1000000MB ? UINT64_C(1000000000000)
771 : features & F_100000MB ? UINT64_C(100000000000)
772 : features & F_40000MB ? UINT64_C(40000000000)
773 : features & F_10000MB ? UINT64_C(10000000000)
774 : features & F_1000MB ? UINT64_C(1000000000)
775 : features & F_100MB ? UINT64_C(100000000)
776 : features & F_10MB ? UINT64_C(10000000)
780 /* Returns true if any of the NETDEV_F_* bits that indicate a full-duplex link
781 * are set in 'features', otherwise false. */
783 netdev_features_is_full_duplex(enum netdev_features features)
785 return (features & (NETDEV_F_10MB_FD | NETDEV_F_100MB_FD | NETDEV_F_1GB_FD
786 | NETDEV_F_10GB_FD | NETDEV_F_40GB_FD
787 | NETDEV_F_100GB_FD | NETDEV_F_1TB_FD)) != 0;
790 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
791 * successful, otherwise a positive errno value. */
793 netdev_set_advertisements(struct netdev *netdev,
794 enum netdev_features advertise)
796 return (netdev->netdev_class->set_advertisements
797 ? netdev->netdev_class->set_advertisements(
802 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
803 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
804 * errno value and sets '*address' to 0 (INADDR_ANY).
806 * The following error values have well-defined meanings:
808 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
810 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
812 * 'address' or 'netmask' or both may be null, in which case the address or
813 * netmask is not reported. */
815 netdev_get_in4(const struct netdev *netdev,
816 struct in_addr *address_, struct in_addr *netmask_)
818 struct in_addr address;
819 struct in_addr netmask;
822 error = (netdev->netdev_class->get_in4
823 ? netdev->netdev_class->get_in4(netdev,
827 address_->s_addr = error ? 0 : address.s_addr;
830 netmask_->s_addr = error ? 0 : netmask.s_addr;
835 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
836 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
837 * positive errno value. */
839 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
841 return (netdev->netdev_class->set_in4
842 ? netdev->netdev_class->set_in4(netdev, addr, mask)
846 /* Obtains ad IPv4 address from device name and save the address in
847 * in4. Returns 0 if successful, otherwise a positive errno value.
850 netdev_get_in4_by_name(const char *device_name, struct in_addr *in4)
852 struct netdev *netdev;
855 error = netdev_open(device_name, "system", &netdev);
857 in4->s_addr = htonl(0);
861 error = netdev_get_in4(netdev, in4, NULL);
862 netdev_close(netdev);
866 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
869 netdev_add_router(struct netdev *netdev, struct in_addr router)
871 COVERAGE_INC(netdev_add_router);
872 return (netdev->netdev_class->add_router
873 ? netdev->netdev_class->add_router(netdev, router)
877 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
878 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
879 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
880 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
881 * a directly connected network) in '*next_hop' and a copy of the name of the
882 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
883 * responsible for freeing '*netdev_name' (by calling free()). */
885 netdev_get_next_hop(const struct netdev *netdev,
886 const struct in_addr *host, struct in_addr *next_hop,
889 int error = (netdev->netdev_class->get_next_hop
890 ? netdev->netdev_class->get_next_hop(
891 host, next_hop, netdev_name)
894 next_hop->s_addr = 0;
900 /* Populates 'smap' with status information.
902 * Populates 'smap' with 'netdev' specific status information. This
903 * information may be used to populate the status column of the Interface table
904 * as defined in ovs-vswitchd.conf.db(5). */
906 netdev_get_status(const struct netdev *netdev, struct smap *smap)
908 return (netdev->netdev_class->get_status
909 ? netdev->netdev_class->get_status(netdev, smap)
913 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
914 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
915 * all-zero-bits (in6addr_any).
917 * The following error values have well-defined meanings:
919 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
921 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
923 * 'in6' may be null, in which case the address itself is not reported. */
925 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
927 struct in6_addr dummy;
930 error = (netdev->netdev_class->get_in6
931 ? netdev->netdev_class->get_in6(netdev,
935 memset(in6, 0, sizeof *in6);
940 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
941 * 'on'. Returns 0 if successful, otherwise a positive errno value. */
943 do_update_flags(struct netdev *netdev, enum netdev_flags off,
944 enum netdev_flags on, enum netdev_flags *old_flagsp,
945 struct netdev_saved_flags **sfp)
946 OVS_EXCLUDED(netdev_mutex)
948 struct netdev_saved_flags *sf = NULL;
949 enum netdev_flags old_flags;
952 error = netdev->netdev_class->update_flags(netdev, off & ~on, on,
955 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
956 off || on ? "set" : "get", netdev_get_name(netdev),
957 ovs_strerror(error));
959 } else if ((off || on) && sfp) {
960 enum netdev_flags new_flags = (old_flags & ~off) | on;
961 enum netdev_flags changed_flags = old_flags ^ new_flags;
963 ovs_mutex_lock(&netdev_mutex);
964 *sfp = sf = xmalloc(sizeof *sf);
966 list_push_front(&netdev->saved_flags_list, &sf->node);
967 sf->saved_flags = changed_flags;
968 sf->saved_values = changed_flags & new_flags;
971 ovs_mutex_unlock(&netdev_mutex);
976 *old_flagsp = old_flags;
985 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
986 * Returns 0 if successful, otherwise a positive errno value. On failure,
987 * stores 0 into '*flagsp'. */
989 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
991 struct netdev *netdev = CONST_CAST(struct netdev *, netdev_);
992 return do_update_flags(netdev, 0, 0, flagsp, NULL);
995 /* Sets the flags for 'netdev' to 'flags'.
996 * Returns 0 if successful, otherwise a positive errno value. */
998 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
999 struct netdev_saved_flags **sfp)
1001 return do_update_flags(netdev, -1, flags, NULL, sfp);
1004 /* Turns on the specified 'flags' on 'netdev':
1006 * - On success, returns 0. If 'sfp' is nonnull, sets '*sfp' to a newly
1007 * allocated 'struct netdev_saved_flags *' that may be passed to
1008 * netdev_restore_flags() to restore the original values of 'flags' on
1009 * 'netdev' (this will happen automatically at program termination if
1010 * netdev_restore_flags() is never called) , or to NULL if no flags were
1013 * - On failure, returns a positive errno value. If 'sfp' is nonnull, sets
1014 * '*sfp' to NULL. */
1016 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
1017 struct netdev_saved_flags **sfp)
1019 return do_update_flags(netdev, 0, flags, NULL, sfp);
1022 /* Turns off the specified 'flags' on 'netdev'. See netdev_turn_flags_on() for
1023 * details of the interface. */
1025 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
1026 struct netdev_saved_flags **sfp)
1028 return do_update_flags(netdev, flags, 0, NULL, sfp);
1031 /* Restores the flags that were saved in 'sf', and destroys 'sf'.
1032 * Does nothing if 'sf' is NULL. */
1034 netdev_restore_flags(struct netdev_saved_flags *sf)
1035 OVS_EXCLUDED(netdev_mutex)
1038 struct netdev *netdev = sf->netdev;
1039 enum netdev_flags old_flags;
1041 netdev->netdev_class->update_flags(netdev,
1042 sf->saved_flags & sf->saved_values,
1043 sf->saved_flags & ~sf->saved_values,
1046 ovs_mutex_lock(&netdev_mutex);
1047 list_remove(&sf->node);
1049 netdev_unref(netdev);
1053 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1054 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1055 * returns 0. Otherwise, it returns a positive errno value; in particular,
1056 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
1058 netdev_arp_lookup(const struct netdev *netdev,
1059 ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
1061 int error = (netdev->netdev_class->arp_lookup
1062 ? netdev->netdev_class->arp_lookup(netdev, ip, mac)
1065 memset(mac, 0, ETH_ADDR_LEN);
1070 /* Returns true if carrier is active (link light is on) on 'netdev'. */
1072 netdev_get_carrier(const struct netdev *netdev)
1075 enum netdev_flags flags;
1078 netdev_get_flags(netdev, &flags);
1079 if (!(flags & NETDEV_UP)) {
1083 if (!netdev->netdev_class->get_carrier) {
1087 error = netdev->netdev_class->get_carrier(netdev, &carrier);
1089 VLOG_DBG("%s: failed to get network device carrier status, assuming "
1090 "down: %s", netdev_get_name(netdev), ovs_strerror(error));
1097 /* Returns the number of times 'netdev''s carrier has changed. */
1099 netdev_get_carrier_resets(const struct netdev *netdev)
1101 return (netdev->netdev_class->get_carrier_resets
1102 ? netdev->netdev_class->get_carrier_resets(netdev)
1106 /* Attempts to force netdev_get_carrier() to poll 'netdev''s MII registers for
1107 * link status instead of checking 'netdev''s carrier. 'netdev''s MII
1108 * registers will be polled once ever 'interval' milliseconds. If 'netdev'
1109 * does not support MII, another method may be used as a fallback. If
1110 * 'interval' is less than or equal to zero, reverts netdev_get_carrier() to
1111 * its normal behavior.
1113 * Returns 0 if successful, otherwise a positive errno value. */
1115 netdev_set_miimon_interval(struct netdev *netdev, long long int interval)
1117 return (netdev->netdev_class->set_miimon_interval
1118 ? netdev->netdev_class->set_miimon_interval(netdev, interval)
1122 /* Retrieves current device stats for 'netdev'. */
1124 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
1128 COVERAGE_INC(netdev_get_stats);
1129 error = (netdev->netdev_class->get_stats
1130 ? netdev->netdev_class->get_stats(netdev, stats)
1133 memset(stats, 0xff, sizeof *stats);
1138 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
1139 * Returns 0 if successful, otherwise a positive errno value.
1141 * This will probably fail for most network devices. Some devices might only
1142 * allow setting their stats to 0. */
1144 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
1146 return (netdev->netdev_class->set_stats
1147 ? netdev->netdev_class->set_stats(netdev, stats)
1151 /* Attempts to set input rate limiting (policing) policy, such that up to
1152 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
1153 * size of 'kbits' kb. */
1155 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
1156 uint32_t kbits_burst)
1158 return (netdev->netdev_class->set_policing
1159 ? netdev->netdev_class->set_policing(netdev,
1160 kbits_rate, kbits_burst)
1164 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
1165 * empty if 'netdev' does not support QoS. Any names added to 'types' should
1166 * be documented as valid for the "type" column in the "QoS" table in
1167 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1169 * Every network device supports disabling QoS with a type of "", but this type
1170 * will not be added to 'types'.
1172 * The caller must initialize 'types' (e.g. with sset_init()) before calling
1173 * this function. The caller is responsible for destroying 'types' (e.g. with
1174 * sset_destroy()) when it is no longer needed.
1176 * Returns 0 if successful, otherwise a positive errno value. */
1178 netdev_get_qos_types(const struct netdev *netdev, struct sset *types)
1180 const struct netdev_class *class = netdev->netdev_class;
1181 return (class->get_qos_types
1182 ? class->get_qos_types(netdev, types)
1186 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
1187 * which should be "" or one of the types returned by netdev_get_qos_types()
1188 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
1189 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
1190 * 'caps' to all zeros. */
1192 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
1193 struct netdev_qos_capabilities *caps)
1195 const struct netdev_class *class = netdev->netdev_class;
1198 int retval = (class->get_qos_capabilities
1199 ? class->get_qos_capabilities(netdev, type, caps)
1202 memset(caps, 0, sizeof *caps);
1206 /* Every netdev supports turning off QoS. */
1207 memset(caps, 0, sizeof *caps);
1212 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1213 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1214 * the number of queues (zero on failure) in '*n_queuesp'.
1216 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1218 netdev_get_n_queues(const struct netdev *netdev,
1219 const char *type, unsigned int *n_queuesp)
1221 struct netdev_qos_capabilities caps;
1224 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1225 *n_queuesp = caps.n_queues;
1229 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1230 * stores the name of the current form of QoS into '*typep', stores any details
1231 * of configuration as string key-value pairs in 'details', and returns 0. On
1232 * failure, sets '*typep' to NULL and returns a positive errno value.
1234 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1236 * The caller must initialize 'details' as an empty smap (e.g. with
1237 * smap_init()) before calling this function. The caller must free 'details'
1238 * when it is no longer needed (e.g. with smap_destroy()).
1240 * The caller must not modify or free '*typep'.
1242 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1243 * 'netdev'. The contents of 'details' should be documented as valid for
1244 * '*typep' in the "other_config" column in the "QoS" table in
1245 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1247 netdev_get_qos(const struct netdev *netdev,
1248 const char **typep, struct smap *details)
1250 const struct netdev_class *class = netdev->netdev_class;
1253 if (class->get_qos) {
1254 retval = class->get_qos(netdev, typep, details);
1257 smap_clear(details);
1261 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1267 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1268 * with details of configuration from 'details'. Returns 0 if successful,
1269 * otherwise a positive errno value. On error, the previous QoS configuration
1272 * When this function changes the type of QoS (not just 'details'), this also
1273 * resets all queue configuration for 'netdev' to their defaults (which depend
1274 * on the specific type of QoS). Otherwise, the queue configuration for
1275 * 'netdev' is unchanged.
1277 * 'type' should be "" (to disable QoS) or one of the types returned by
1278 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1279 * documented as valid for the given 'type' in the "other_config" column in the
1280 * "QoS" table in vswitchd/vswitch.xml (which is built as
1281 * ovs-vswitchd.conf.db(8)).
1283 * NULL may be specified for 'details' if there are no configuration
1286 netdev_set_qos(struct netdev *netdev,
1287 const char *type, const struct smap *details)
1289 const struct netdev_class *class = netdev->netdev_class;
1295 if (class->set_qos) {
1297 static const struct smap empty = SMAP_INITIALIZER(&empty);
1300 return class->set_qos(netdev, type, details);
1302 return *type ? EOPNOTSUPP : 0;
1306 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1307 * successful, adds that information as string key-value pairs to 'details'.
1308 * Returns 0 if successful, otherwise a positive errno value.
1310 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1311 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1313 * The returned contents of 'details' should be documented as valid for the
1314 * given 'type' in the "other_config" column in the "Queue" table in
1315 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1317 * The caller must initialize 'details' (e.g. with smap_init()) before calling
1318 * this function. The caller must free 'details' when it is no longer needed
1319 * (e.g. with smap_destroy()). */
1321 netdev_get_queue(const struct netdev *netdev,
1322 unsigned int queue_id, struct smap *details)
1324 const struct netdev_class *class = netdev->netdev_class;
1327 retval = (class->get_queue
1328 ? class->get_queue(netdev, queue_id, details)
1331 smap_clear(details);
1336 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1337 * string pairs in 'details'. The contents of 'details' should be documented
1338 * as valid for the given 'type' in the "other_config" column in the "Queue"
1339 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1340 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1341 * given queue's configuration should be unmodified.
1343 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1344 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1346 * This function does not modify 'details', and the caller retains ownership of
1349 netdev_set_queue(struct netdev *netdev,
1350 unsigned int queue_id, const struct smap *details)
1352 const struct netdev_class *class = netdev->netdev_class;
1353 return (class->set_queue
1354 ? class->set_queue(netdev, queue_id, details)
1358 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1359 * of QoS may have a fixed set of queues, in which case attempts to delete them
1360 * will fail with EOPNOTSUPP.
1362 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1363 * given queue will be unmodified.
1365 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1366 * the current form of QoS (e.g. as returned by
1367 * netdev_get_n_queues(netdev)). */
1369 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1371 const struct netdev_class *class = netdev->netdev_class;
1372 return (class->delete_queue
1373 ? class->delete_queue(netdev, queue_id)
1377 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1378 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1379 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1380 * positive errno value and fills 'stats' with values indicating unsupported
1383 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1384 struct netdev_queue_stats *stats)
1386 const struct netdev_class *class = netdev->netdev_class;
1389 retval = (class->get_queue_stats
1390 ? class->get_queue_stats(netdev, queue_id, stats)
1393 stats->tx_bytes = UINT64_MAX;
1394 stats->tx_packets = UINT64_MAX;
1395 stats->tx_errors = UINT64_MAX;
1396 stats->created = LLONG_MIN;
1401 /* Initializes 'dump' to begin dumping the queues in a netdev.
1403 * This function provides no status indication. An error status for the entire
1404 * dump operation is provided when it is completed by calling
1405 * netdev_queue_dump_done().
1408 netdev_queue_dump_start(struct netdev_queue_dump *dump,
1409 const struct netdev *netdev)
1411 dump->netdev = netdev_ref(netdev);
1412 if (netdev->netdev_class->queue_dump_start) {
1413 dump->error = netdev->netdev_class->queue_dump_start(netdev,
1416 dump->error = EOPNOTSUPP;
1420 /* Attempts to retrieve another queue from 'dump', which must have been
1421 * initialized with netdev_queue_dump_start(). On success, stores a new queue
1422 * ID into '*queue_id', fills 'details' with configuration details for the
1423 * queue, and returns true. On failure, returns false.
1425 * Queues are not necessarily dumped in increasing order of queue ID (or any
1426 * other predictable order).
1428 * Failure might indicate an actual error or merely that the last queue has
1429 * been dumped. An error status for the entire dump operation is provided when
1430 * it is completed by calling netdev_queue_dump_done().
1432 * The returned contents of 'details' should be documented as valid for the
1433 * given 'type' in the "other_config" column in the "Queue" table in
1434 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1436 * The caller must initialize 'details' (e.g. with smap_init()) before calling
1437 * this function. This function will clear and replace its contents. The
1438 * caller must free 'details' when it is no longer needed (e.g. with
1439 * smap_destroy()). */
1441 netdev_queue_dump_next(struct netdev_queue_dump *dump,
1442 unsigned int *queue_id, struct smap *details)
1444 const struct netdev *netdev = dump->netdev;
1450 dump->error = netdev->netdev_class->queue_dump_next(netdev, dump->state,
1454 netdev->netdev_class->queue_dump_done(netdev, dump->state);
1460 /* Completes queue table dump operation 'dump', which must have been
1461 * initialized with netdev_queue_dump_start(). Returns 0 if the dump operation
1462 * was error-free, otherwise a positive errno value describing the problem. */
1464 netdev_queue_dump_done(struct netdev_queue_dump *dump)
1466 const struct netdev *netdev = dump->netdev;
1467 if (!dump->error && netdev->netdev_class->queue_dump_done) {
1468 dump->error = netdev->netdev_class->queue_dump_done(netdev,
1471 netdev_close(dump->netdev);
1472 return dump->error == EOF ? 0 : dump->error;
1475 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1476 * its statistics, and the 'aux' specified by the caller. The order of
1477 * iteration is unspecified, but (when successful) each queue is visited
1480 * Calling this function may be more efficient than calling
1481 * netdev_get_queue_stats() for every queue.
1483 * 'cb' must not modify or free the statistics passed in.
1485 * Returns 0 if successful, otherwise a positive errno value. On error, some
1486 * configured queues may not have been included in the iteration. */
1488 netdev_dump_queue_stats(const struct netdev *netdev,
1489 netdev_dump_queue_stats_cb *cb, void *aux)
1491 const struct netdev_class *class = netdev->netdev_class;
1492 return (class->dump_queue_stats
1493 ? class->dump_queue_stats(netdev, cb, aux)
1497 /* Returns a sequence number which indicates changes in one of 'netdev''s
1498 * properties. The returned sequence will be nonzero so that callers have a
1499 * value which they may use as a reset when tracking 'netdev'.
1501 * The returned sequence number will change whenever 'netdev''s flags,
1502 * features, ethernet address, or carrier changes. It may change for other
1503 * reasons as well, or no reason at all. */
1505 netdev_change_seq(const struct netdev *netdev)
1507 return netdev->netdev_class->change_seq(netdev);
1510 /* Returns the class type of 'netdev'.
1512 * The caller must not free the returned value. */
1514 netdev_get_type(const struct netdev *netdev)
1516 return netdev->netdev_class->type;
1519 /* Returns the class associated with 'netdev'. */
1520 const struct netdev_class *
1521 netdev_get_class(const struct netdev *netdev)
1523 return netdev->netdev_class;
1526 /* Returns the netdev with 'name' or NULL if there is none.
1528 * The caller must free the returned netdev with netdev_close(). */
1530 netdev_from_name(const char *name)
1531 OVS_EXCLUDED(netdev_mutex)
1533 struct netdev *netdev;
1535 ovs_mutex_lock(&netdev_mutex);
1536 netdev = shash_find_data(&netdev_shash, name);
1540 ovs_mutex_unlock(&netdev_mutex);
1545 /* Fills 'device_list' with devices that match 'netdev_class'.
1547 * The caller is responsible for initializing and destroying 'device_list' and
1548 * must close each device on the list. */
1550 netdev_get_devices(const struct netdev_class *netdev_class,
1551 struct shash *device_list)
1552 OVS_EXCLUDED(netdev_mutex)
1554 struct shash_node *node;
1556 ovs_mutex_lock(&netdev_mutex);
1557 SHASH_FOR_EACH (node, &netdev_shash) {
1558 struct netdev *dev = node->data;
1560 if (dev->netdev_class == netdev_class) {
1562 shash_add(device_list, node->name, node->data);
1565 ovs_mutex_unlock(&netdev_mutex);
1569 netdev_get_type_from_name(const char *name)
1571 struct netdev *dev = netdev_from_name(name);
1572 const char *type = dev ? netdev_get_type(dev) : NULL;
1578 netdev_rx_get_netdev(const struct netdev_rx *rx)
1580 ovs_assert(rx->netdev->ref_cnt > 0);
1585 netdev_rx_get_name(const struct netdev_rx *rx)
1587 return netdev_get_name(netdev_rx_get_netdev(rx));
1591 restore_all_flags(void *aux OVS_UNUSED)
1593 struct shash_node *node;
1595 SHASH_FOR_EACH (node, &netdev_shash) {
1596 struct netdev *netdev = node->data;
1597 const struct netdev_saved_flags *sf;
1598 enum netdev_flags saved_values;
1599 enum netdev_flags saved_flags;
1601 saved_values = saved_flags = 0;
1602 LIST_FOR_EACH (sf, node, &netdev->saved_flags_list) {
1603 saved_flags |= sf->saved_flags;
1604 saved_values &= ~sf->saved_flags;
1605 saved_values |= sf->saved_flags & sf->saved_values;
1608 enum netdev_flags old_flags;
1610 netdev->netdev_class->update_flags(netdev,
1611 saved_flags & saved_values,
1612 saved_flags & ~saved_values,