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) {
131 ovs_rwlock_unlock(&netdev_class_rwlock);
134 /* Arranges for poll_block() to wake up when netdev_run() needs to be called.
136 * If your program opens any netdevs, it must call this function within its
140 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
142 struct netdev_registered_class *rc;
144 ovs_rwlock_rdlock(&netdev_class_rwlock);
145 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
148 ovs_rwlock_unlock(&netdev_class_rwlock);
151 static struct netdev_registered_class *
152 netdev_lookup_class(const char *type)
153 OVS_REQ_RDLOCK(netdev_class_rwlock)
155 struct netdev_registered_class *rc;
157 HMAP_FOR_EACH_WITH_HASH (rc, hmap_node, hash_string(type, 0),
159 if (!strcmp(type, rc->class->type)) {
166 /* Initializes and registers a new netdev provider. After successful
167 * registration, new netdevs of that type can be opened using netdev_open(). */
169 netdev_register_provider(const struct netdev_class *new_class)
170 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
174 ovs_rwlock_wrlock(&netdev_class_rwlock);
175 if (netdev_lookup_class(new_class->type)) {
176 VLOG_WARN("attempted to register duplicate netdev provider: %s",
180 error = new_class->init ? new_class->init() : 0;
182 struct netdev_registered_class *rc;
184 rc = xmalloc(sizeof *rc);
185 hmap_insert(&netdev_classes, &rc->hmap_node,
186 hash_string(new_class->type, 0));
187 rc->class = new_class;
188 atomic_init(&rc->ref_cnt, 0);
190 VLOG_ERR("failed to initialize %s network device class: %s",
191 new_class->type, ovs_strerror(error));
194 ovs_rwlock_unlock(&netdev_class_rwlock);
199 /* Unregisters a netdev provider. 'type' must have been previously
200 * registered and not currently be in use by any netdevs. After unregistration
201 * new netdevs of that type cannot be opened using netdev_open(). */
203 netdev_unregister_provider(const char *type)
204 OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex)
206 struct netdev_registered_class *rc;
209 ovs_rwlock_wrlock(&netdev_class_rwlock);
210 rc = netdev_lookup_class(type);
212 VLOG_WARN("attempted to unregister a netdev provider that is not "
213 "registered: %s", type);
214 error = EAFNOSUPPORT;
218 atomic_read(&rc->ref_cnt, &ref_cnt);
220 hmap_remove(&netdev_classes, &rc->hmap_node);
224 VLOG_WARN("attempted to unregister in use netdev provider: %s",
229 ovs_rwlock_unlock(&netdev_class_rwlock);
234 /* Clears 'types' and enumerates the types of all currently registered netdev
235 * providers into it. The caller must first initialize the sset. */
237 netdev_enumerate_types(struct sset *types)
238 OVS_EXCLUDED(netdev_mutex)
240 struct netdev_registered_class *rc;
245 ovs_rwlock_rdlock(&netdev_class_rwlock);
246 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
247 sset_add(types, rc->class->type);
249 ovs_rwlock_unlock(&netdev_class_rwlock);
252 /* Check that the network device name is not the same as any of the registered
253 * vport providers' dpif_port name (dpif_port is NULL if the vport provider
254 * does not define it) or the datapath internal port name (e.g. ovs-system).
256 * Returns true if there is a name conflict, false otherwise. */
258 netdev_is_reserved_name(const char *name)
259 OVS_EXCLUDED(netdev_mutex)
261 struct netdev_registered_class *rc;
265 ovs_rwlock_rdlock(&netdev_class_rwlock);
266 HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) {
267 const char *dpif_port = netdev_vport_class_get_dpif_port(rc->class);
268 if (dpif_port && !strcmp(dpif_port, name)) {
269 ovs_rwlock_unlock(&netdev_class_rwlock);
273 ovs_rwlock_unlock(&netdev_class_rwlock);
275 if (!strncmp(name, "ovs-", 4)) {
280 dp_enumerate_types(&types);
281 SSET_FOR_EACH (type, &types) {
282 if (!strcmp(name+4, type)) {
283 sset_destroy(&types);
287 sset_destroy(&types);
293 /* Opens the network device named 'name' (e.g. "eth0") of the specified 'type'
294 * (e.g. "system") and returns zero if successful, otherwise a positive errno
295 * value. On success, sets '*netdevp' to the new network device, otherwise to
298 * Some network devices may need to be configured (with netdev_set_config())
299 * before they can be used. */
301 netdev_open(const char *name, const char *type, struct netdev **netdevp)
302 OVS_EXCLUDED(netdev_mutex)
304 struct netdev *netdev;
309 ovs_rwlock_rdlock(&netdev_class_rwlock);
310 ovs_mutex_lock(&netdev_mutex);
311 netdev = shash_find_data(&netdev_shash, name);
313 struct netdev_registered_class *rc;
315 rc = netdev_lookup_class(type && type[0] ? type : "system");
317 netdev = rc->class->alloc();
319 memset(netdev, 0, sizeof *netdev);
320 netdev->netdev_class = rc->class;
321 netdev->name = xstrdup(name);
322 netdev->node = shash_add(&netdev_shash, name, netdev);
323 list_init(&netdev->saved_flags_list);
325 error = rc->class->construct(netdev);
329 atomic_add(&rc->ref_cnt, 1, &old_ref_cnt);
331 rc->class->dealloc(netdev);
337 VLOG_WARN("could not create netdev %s of unknown type %s",
339 error = EAFNOSUPPORT;
345 ovs_mutex_unlock(&netdev_mutex);
346 ovs_rwlock_unlock(&netdev_class_rwlock);
357 /* Returns a reference to 'netdev_' for the caller to own. Returns null if
358 * 'netdev_' is null. */
360 netdev_ref(const struct netdev *netdev_)
361 OVS_EXCLUDED(netdev_mutex)
363 struct netdev *netdev = CONST_CAST(struct netdev *, netdev_);
366 ovs_mutex_lock(&netdev_mutex);
367 ovs_assert(netdev->ref_cnt > 0);
369 ovs_mutex_unlock(&netdev_mutex);
374 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
375 * or NULL if none are needed. */
377 netdev_set_config(struct netdev *netdev, const struct smap *args)
378 OVS_EXCLUDED(netdev_mutex)
380 if (netdev->netdev_class->set_config) {
381 const struct smap no_args = SMAP_INITIALIZER(&no_args);
382 return netdev->netdev_class->set_config(netdev,
383 args ? args : &no_args);
384 } else if (args && !smap_is_empty(args)) {
385 VLOG_WARN("%s: arguments provided to device that is not configurable",
386 netdev_get_name(netdev));
392 /* Returns the current configuration for 'netdev' in 'args'. The caller must
393 * have already initialized 'args' with smap_init(). Returns 0 on success, in
394 * which case 'args' will be filled with 'netdev''s configuration. On failure
395 * returns a positive errno value, in which case 'args' will be empty.
397 * The caller owns 'args' and its contents and must eventually free them with
400 netdev_get_config(const struct netdev *netdev, struct smap *args)
401 OVS_EXCLUDED(netdev_mutex)
406 if (netdev->netdev_class->get_config) {
407 error = netdev->netdev_class->get_config(netdev, args);
418 const struct netdev_tunnel_config *
419 netdev_get_tunnel_config(const struct netdev *netdev)
420 OVS_EXCLUDED(netdev_mutex)
422 if (netdev->netdev_class->get_tunnel_config) {
423 return netdev->netdev_class->get_tunnel_config(netdev);
430 netdev_unref(struct netdev *dev)
431 OVS_RELEASES(netdev_mutex)
433 ovs_assert(dev->ref_cnt);
434 if (!--dev->ref_cnt) {
435 const struct netdev_class *class = dev->netdev_class;
436 struct netdev_registered_class *rc;
439 dev->netdev_class->destruct(dev);
441 shash_delete(&netdev_shash, dev->node);
443 dev->netdev_class->dealloc(dev);
444 ovs_mutex_unlock(&netdev_mutex);
446 ovs_rwlock_rdlock(&netdev_class_rwlock);
447 rc = netdev_lookup_class(class->type);
448 atomic_sub(&rc->ref_cnt, 1, &old_ref_cnt);
449 ovs_assert(old_ref_cnt > 0);
450 ovs_rwlock_unlock(&netdev_class_rwlock);
452 ovs_mutex_unlock(&netdev_mutex);
456 /* Closes and destroys 'netdev'. */
458 netdev_close(struct netdev *netdev)
459 OVS_EXCLUDED(netdev_mutex)
462 ovs_mutex_lock(&netdev_mutex);
463 netdev_unref(netdev);
467 /* Parses 'netdev_name_', which is of the form [type@]name into its component
468 * pieces. 'name' and 'type' must be freed by the caller. */
470 netdev_parse_name(const char *netdev_name_, char **name, char **type)
472 char *netdev_name = xstrdup(netdev_name_);
475 separator = strchr(netdev_name, '@');
479 *name = xstrdup(separator + 1);
482 *type = xstrdup("system");
487 netdev_rx_open(struct netdev *netdev, struct netdev_rx **rxp)
488 OVS_EXCLUDED(netdev_mutex)
492 if (netdev->netdev_class->rx_alloc) {
493 struct netdev_rx *rx = netdev->netdev_class->rx_alloc();
496 error = netdev->netdev_class->rx_construct(rx);
498 ovs_mutex_lock(&netdev_mutex);
500 ovs_mutex_unlock(&netdev_mutex);
505 netdev->netdev_class->rx_dealloc(rx);
518 netdev_rx_close(struct netdev_rx *rx)
519 OVS_EXCLUDED(netdev_mutex)
522 struct netdev *netdev = rx->netdev;
523 netdev->netdev_class->rx_destruct(rx);
524 netdev->netdev_class->rx_dealloc(rx);
525 netdev_close(netdev);
530 netdev_rx_recv(struct netdev_rx *rx, struct ofpbuf *buffer)
534 ovs_assert(buffer->size == 0);
535 ovs_assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
537 retval = rx->netdev->netdev_class->rx_recv(rx, buffer->data,
538 ofpbuf_tailroom(buffer));
540 COVERAGE_INC(netdev_received);
541 buffer->size += retval;
542 if (buffer->size < ETH_TOTAL_MIN) {
543 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
552 netdev_rx_wait(struct netdev_rx *rx)
554 rx->netdev->netdev_class->rx_wait(rx);
558 netdev_rx_drain(struct netdev_rx *rx)
560 return (rx->netdev->netdev_class->rx_drain
561 ? rx->netdev->netdev_class->rx_drain(rx)
565 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
566 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
567 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
568 * the packet is too big or too small to transmit on the device.
570 * The caller retains ownership of 'buffer' in all cases.
572 * The kernel maintains a packet transmission queue, so the caller is not
573 * expected to do additional queuing of packets.
575 * Some network devices may not implement support for this function. In such
576 * cases this function will always return EOPNOTSUPP. */
578 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
582 error = (netdev->netdev_class->send
583 ? netdev->netdev_class->send(netdev, buffer->data, buffer->size)
586 COVERAGE_INC(netdev_sent);
591 /* Registers with the poll loop to wake up from the next call to poll_block()
592 * when the packet transmission queue has sufficient room to transmit a packet
593 * with netdev_send().
595 * The kernel maintains a packet transmission queue, so the client is not
596 * expected to do additional queuing of packets. Thus, this function is
597 * unlikely to ever be used. It is included for completeness. */
599 netdev_send_wait(struct netdev *netdev)
601 if (netdev->netdev_class->send_wait) {
602 netdev->netdev_class->send_wait(netdev);
606 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
607 * otherwise a positive errno value. */
609 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
611 return netdev->netdev_class->set_etheraddr(netdev, mac);
614 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
615 * the MAC address into 'mac'. On failure, returns a positive errno value and
616 * clears 'mac' to all-zeros. */
618 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
620 return netdev->netdev_class->get_etheraddr(netdev, mac);
623 /* Returns the name of the network device that 'netdev' represents,
624 * e.g. "eth0". The caller must not modify or free the returned string. */
626 netdev_get_name(const struct netdev *netdev)
631 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
632 * (and received) packets, in bytes, not including the hardware header; thus,
633 * this is typically 1500 bytes for Ethernet devices.
635 * If successful, returns 0 and stores the MTU size in '*mtup'. Returns
636 * EOPNOTSUPP if 'netdev' does not have an MTU (as e.g. some tunnels do not).
637 * On other failure, returns a positive errno value. On failure, sets '*mtup'
640 netdev_get_mtu(const struct netdev *netdev, int *mtup)
642 const struct netdev_class *class = netdev->netdev_class;
645 error = class->get_mtu ? class->get_mtu(netdev, mtup) : EOPNOTSUPP;
648 if (error != EOPNOTSUPP) {
649 VLOG_DBG_RL(&rl, "failed to retrieve MTU for network device %s: "
650 "%s", netdev_get_name(netdev), ovs_strerror(error));
656 /* Sets the MTU of 'netdev'. The MTU is the maximum size of transmitted
657 * (and received) packets, in bytes.
659 * If successful, returns 0. Returns EOPNOTSUPP if 'netdev' does not have an
660 * MTU (as e.g. some tunnels do not). On other failure, returns a positive
663 netdev_set_mtu(const struct netdev *netdev, int mtu)
665 const struct netdev_class *class = netdev->netdev_class;
668 error = class->set_mtu ? class->set_mtu(netdev, mtu) : EOPNOTSUPP;
669 if (error && error != EOPNOTSUPP) {
670 VLOG_DBG_RL(&rl, "failed to set MTU for network device %s: %s",
671 netdev_get_name(netdev), ovs_strerror(error));
677 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
678 * failure, returns a negative errno value.
680 * The desired semantics of the ifindex value are a combination of those
681 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
682 * value should be unique within a host and remain stable at least until
683 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
684 * but many systems do not follow this rule anyhow.
686 * Some network devices may not implement support for this function. In such
687 * cases this function will always return -EOPNOTSUPP.
690 netdev_get_ifindex(const struct netdev *netdev)
692 int (*get_ifindex)(const struct netdev *);
694 get_ifindex = netdev->netdev_class->get_ifindex;
696 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
699 /* Stores the features supported by 'netdev' into each of '*current',
700 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
701 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
702 * successful, otherwise a positive errno value. On failure, all of the
703 * passed-in values are set to 0.
705 * Some network devices may not implement support for this function. In such
706 * cases this function will always return EOPNOTSUPP. */
708 netdev_get_features(const struct netdev *netdev,
709 enum netdev_features *current,
710 enum netdev_features *advertised,
711 enum netdev_features *supported,
712 enum netdev_features *peer)
714 int (*get_features)(const struct netdev *netdev,
715 enum netdev_features *current,
716 enum netdev_features *advertised,
717 enum netdev_features *supported,
718 enum netdev_features *peer);
719 enum netdev_features dummy[4];
726 advertised = &dummy[1];
729 supported = &dummy[2];
735 get_features = netdev->netdev_class->get_features;
737 ? get_features(netdev, current, advertised, supported,
741 *current = *advertised = *supported = *peer = 0;
746 /* Returns the maximum speed of a network connection that has the NETDEV_F_*
747 * bits in 'features', in bits per second. If no bits that indicate a speed
748 * are set in 'features', returns 'default_bps'. */
750 netdev_features_to_bps(enum netdev_features features,
751 uint64_t default_bps)
754 F_1000000MB = NETDEV_F_1TB_FD,
755 F_100000MB = NETDEV_F_100GB_FD,
756 F_40000MB = NETDEV_F_40GB_FD,
757 F_10000MB = NETDEV_F_10GB_FD,
758 F_1000MB = NETDEV_F_1GB_HD | NETDEV_F_1GB_FD,
759 F_100MB = NETDEV_F_100MB_HD | NETDEV_F_100MB_FD,
760 F_10MB = NETDEV_F_10MB_HD | NETDEV_F_10MB_FD
763 return ( features & F_1000000MB ? UINT64_C(1000000000000)
764 : features & F_100000MB ? UINT64_C(100000000000)
765 : features & F_40000MB ? UINT64_C(40000000000)
766 : features & F_10000MB ? UINT64_C(10000000000)
767 : features & F_1000MB ? UINT64_C(1000000000)
768 : features & F_100MB ? UINT64_C(100000000)
769 : features & F_10MB ? UINT64_C(10000000)
773 /* Returns true if any of the NETDEV_F_* bits that indicate a full-duplex link
774 * are set in 'features', otherwise false. */
776 netdev_features_is_full_duplex(enum netdev_features features)
778 return (features & (NETDEV_F_10MB_FD | NETDEV_F_100MB_FD | NETDEV_F_1GB_FD
779 | NETDEV_F_10GB_FD | NETDEV_F_40GB_FD
780 | NETDEV_F_100GB_FD | NETDEV_F_1TB_FD)) != 0;
783 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
784 * successful, otherwise a positive errno value. */
786 netdev_set_advertisements(struct netdev *netdev,
787 enum netdev_features advertise)
789 return (netdev->netdev_class->set_advertisements
790 ? netdev->netdev_class->set_advertisements(
795 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
796 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
797 * errno value and sets '*address' to 0 (INADDR_ANY).
799 * The following error values have well-defined meanings:
801 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
803 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
805 * 'address' or 'netmask' or both may be null, in which case the address or
806 * netmask is not reported. */
808 netdev_get_in4(const struct netdev *netdev,
809 struct in_addr *address_, struct in_addr *netmask_)
811 struct in_addr address;
812 struct in_addr netmask;
815 error = (netdev->netdev_class->get_in4
816 ? netdev->netdev_class->get_in4(netdev,
820 address_->s_addr = error ? 0 : address.s_addr;
823 netmask_->s_addr = error ? 0 : netmask.s_addr;
828 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
829 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
830 * positive errno value. */
832 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
834 return (netdev->netdev_class->set_in4
835 ? netdev->netdev_class->set_in4(netdev, addr, mask)
839 /* Obtains ad IPv4 address from device name and save the address in
840 * in4. Returns 0 if successful, otherwise a positive errno value.
843 netdev_get_in4_by_name(const char *device_name, struct in_addr *in4)
845 struct netdev *netdev;
848 error = netdev_open(device_name, "system", &netdev);
850 in4->s_addr = htonl(0);
854 error = netdev_get_in4(netdev, in4, NULL);
855 netdev_close(netdev);
859 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
862 netdev_add_router(struct netdev *netdev, struct in_addr router)
864 COVERAGE_INC(netdev_add_router);
865 return (netdev->netdev_class->add_router
866 ? netdev->netdev_class->add_router(netdev, router)
870 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
871 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
872 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
873 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
874 * a directly connected network) in '*next_hop' and a copy of the name of the
875 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
876 * responsible for freeing '*netdev_name' (by calling free()). */
878 netdev_get_next_hop(const struct netdev *netdev,
879 const struct in_addr *host, struct in_addr *next_hop,
882 int error = (netdev->netdev_class->get_next_hop
883 ? netdev->netdev_class->get_next_hop(
884 host, next_hop, netdev_name)
887 next_hop->s_addr = 0;
893 /* Populates 'smap' with status information.
895 * Populates 'smap' with 'netdev' specific status information. This
896 * information may be used to populate the status column of the Interface table
897 * as defined in ovs-vswitchd.conf.db(5). */
899 netdev_get_status(const struct netdev *netdev, struct smap *smap)
901 return (netdev->netdev_class->get_status
902 ? netdev->netdev_class->get_status(netdev, smap)
906 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
907 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
908 * all-zero-bits (in6addr_any).
910 * The following error values have well-defined meanings:
912 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
914 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
916 * 'in6' may be null, in which case the address itself is not reported. */
918 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
920 struct in6_addr dummy;
923 error = (netdev->netdev_class->get_in6
924 ? netdev->netdev_class->get_in6(netdev,
928 memset(in6, 0, sizeof *in6);
933 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
934 * 'on'. Returns 0 if successful, otherwise a positive errno value. */
936 do_update_flags(struct netdev *netdev, enum netdev_flags off,
937 enum netdev_flags on, enum netdev_flags *old_flagsp,
938 struct netdev_saved_flags **sfp)
939 OVS_EXCLUDED(netdev_mutex)
941 struct netdev_saved_flags *sf = NULL;
942 enum netdev_flags old_flags;
945 error = netdev->netdev_class->update_flags(netdev, off & ~on, on,
948 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
949 off || on ? "set" : "get", netdev_get_name(netdev),
950 ovs_strerror(error));
952 } else if ((off || on) && sfp) {
953 enum netdev_flags new_flags = (old_flags & ~off) | on;
954 enum netdev_flags changed_flags = old_flags ^ new_flags;
956 ovs_mutex_lock(&netdev_mutex);
957 *sfp = sf = xmalloc(sizeof *sf);
959 list_push_front(&netdev->saved_flags_list, &sf->node);
960 sf->saved_flags = changed_flags;
961 sf->saved_values = changed_flags & new_flags;
964 ovs_mutex_unlock(&netdev_mutex);
969 *old_flagsp = old_flags;
978 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
979 * Returns 0 if successful, otherwise a positive errno value. On failure,
980 * stores 0 into '*flagsp'. */
982 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
984 struct netdev *netdev = CONST_CAST(struct netdev *, netdev_);
985 return do_update_flags(netdev, 0, 0, flagsp, NULL);
988 /* Sets the flags for 'netdev' to 'flags'.
989 * Returns 0 if successful, otherwise a positive errno value. */
991 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
992 struct netdev_saved_flags **sfp)
994 return do_update_flags(netdev, -1, flags, NULL, sfp);
997 /* Turns on the specified 'flags' on 'netdev':
999 * - On success, returns 0. If 'sfp' is nonnull, sets '*sfp' to a newly
1000 * allocated 'struct netdev_saved_flags *' that may be passed to
1001 * netdev_restore_flags() to restore the original values of 'flags' on
1002 * 'netdev' (this will happen automatically at program termination if
1003 * netdev_restore_flags() is never called) , or to NULL if no flags were
1006 * - On failure, returns a positive errno value. If 'sfp' is nonnull, sets
1007 * '*sfp' to NULL. */
1009 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
1010 struct netdev_saved_flags **sfp)
1012 return do_update_flags(netdev, 0, flags, NULL, sfp);
1015 /* Turns off the specified 'flags' on 'netdev'. See netdev_turn_flags_on() for
1016 * details of the interface. */
1018 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
1019 struct netdev_saved_flags **sfp)
1021 return do_update_flags(netdev, flags, 0, NULL, sfp);
1024 /* Restores the flags that were saved in 'sf', and destroys 'sf'.
1025 * Does nothing if 'sf' is NULL. */
1027 netdev_restore_flags(struct netdev_saved_flags *sf)
1028 OVS_EXCLUDED(netdev_mutex)
1031 struct netdev *netdev = sf->netdev;
1032 enum netdev_flags old_flags;
1034 netdev->netdev_class->update_flags(netdev,
1035 sf->saved_flags & sf->saved_values,
1036 sf->saved_flags & ~sf->saved_values,
1039 ovs_mutex_lock(&netdev_mutex);
1040 list_remove(&sf->node);
1042 netdev_unref(netdev);
1046 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
1047 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1048 * returns 0. Otherwise, it returns a positive errno value; in particular,
1049 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
1051 netdev_arp_lookup(const struct netdev *netdev,
1052 ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
1054 int error = (netdev->netdev_class->arp_lookup
1055 ? netdev->netdev_class->arp_lookup(netdev, ip, mac)
1058 memset(mac, 0, ETH_ADDR_LEN);
1063 /* Returns true if carrier is active (link light is on) on 'netdev'. */
1065 netdev_get_carrier(const struct netdev *netdev)
1068 enum netdev_flags flags;
1071 netdev_get_flags(netdev, &flags);
1072 if (!(flags & NETDEV_UP)) {
1076 if (!netdev->netdev_class->get_carrier) {
1080 error = netdev->netdev_class->get_carrier(netdev, &carrier);
1082 VLOG_DBG("%s: failed to get network device carrier status, assuming "
1083 "down: %s", netdev_get_name(netdev), ovs_strerror(error));
1090 /* Returns the number of times 'netdev''s carrier has changed. */
1092 netdev_get_carrier_resets(const struct netdev *netdev)
1094 return (netdev->netdev_class->get_carrier_resets
1095 ? netdev->netdev_class->get_carrier_resets(netdev)
1099 /* Attempts to force netdev_get_carrier() to poll 'netdev''s MII registers for
1100 * link status instead of checking 'netdev''s carrier. 'netdev''s MII
1101 * registers will be polled once ever 'interval' milliseconds. If 'netdev'
1102 * does not support MII, another method may be used as a fallback. If
1103 * 'interval' is less than or equal to zero, reverts netdev_get_carrier() to
1104 * its normal behavior.
1106 * Returns 0 if successful, otherwise a positive errno value. */
1108 netdev_set_miimon_interval(struct netdev *netdev, long long int interval)
1110 return (netdev->netdev_class->set_miimon_interval
1111 ? netdev->netdev_class->set_miimon_interval(netdev, interval)
1115 /* Retrieves current device stats for 'netdev'. */
1117 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
1121 COVERAGE_INC(netdev_get_stats);
1122 error = (netdev->netdev_class->get_stats
1123 ? netdev->netdev_class->get_stats(netdev, stats)
1126 memset(stats, 0xff, sizeof *stats);
1131 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
1132 * Returns 0 if successful, otherwise a positive errno value.
1134 * This will probably fail for most network devices. Some devices might only
1135 * allow setting their stats to 0. */
1137 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
1139 return (netdev->netdev_class->set_stats
1140 ? netdev->netdev_class->set_stats(netdev, stats)
1144 /* Attempts to set input rate limiting (policing) policy, such that up to
1145 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
1146 * size of 'kbits' kb. */
1148 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
1149 uint32_t kbits_burst)
1151 return (netdev->netdev_class->set_policing
1152 ? netdev->netdev_class->set_policing(netdev,
1153 kbits_rate, kbits_burst)
1157 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
1158 * empty if 'netdev' does not support QoS. Any names added to 'types' should
1159 * be documented as valid for the "type" column in the "QoS" table in
1160 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1162 * Every network device supports disabling QoS with a type of "", but this type
1163 * will not be added to 'types'.
1165 * The caller must initialize 'types' (e.g. with sset_init()) before calling
1166 * this function. The caller is responsible for destroying 'types' (e.g. with
1167 * sset_destroy()) when it is no longer needed.
1169 * Returns 0 if successful, otherwise a positive errno value. */
1171 netdev_get_qos_types(const struct netdev *netdev, struct sset *types)
1173 const struct netdev_class *class = netdev->netdev_class;
1174 return (class->get_qos_types
1175 ? class->get_qos_types(netdev, types)
1179 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
1180 * which should be "" or one of the types returned by netdev_get_qos_types()
1181 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
1182 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
1183 * 'caps' to all zeros. */
1185 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
1186 struct netdev_qos_capabilities *caps)
1188 const struct netdev_class *class = netdev->netdev_class;
1191 int retval = (class->get_qos_capabilities
1192 ? class->get_qos_capabilities(netdev, type, caps)
1195 memset(caps, 0, sizeof *caps);
1199 /* Every netdev supports turning off QoS. */
1200 memset(caps, 0, sizeof *caps);
1205 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1206 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1207 * the number of queues (zero on failure) in '*n_queuesp'.
1209 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1211 netdev_get_n_queues(const struct netdev *netdev,
1212 const char *type, unsigned int *n_queuesp)
1214 struct netdev_qos_capabilities caps;
1217 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1218 *n_queuesp = caps.n_queues;
1222 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1223 * stores the name of the current form of QoS into '*typep', stores any details
1224 * of configuration as string key-value pairs in 'details', and returns 0. On
1225 * failure, sets '*typep' to NULL and returns a positive errno value.
1227 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1229 * The caller must initialize 'details' as an empty smap (e.g. with
1230 * smap_init()) before calling this function. The caller must free 'details'
1231 * when it is no longer needed (e.g. with smap_destroy()).
1233 * The caller must not modify or free '*typep'.
1235 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1236 * 'netdev'. The contents of 'details' should be documented as valid for
1237 * '*typep' in the "other_config" column in the "QoS" table in
1238 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1240 netdev_get_qos(const struct netdev *netdev,
1241 const char **typep, struct smap *details)
1243 const struct netdev_class *class = netdev->netdev_class;
1246 if (class->get_qos) {
1247 retval = class->get_qos(netdev, typep, details);
1250 smap_clear(details);
1254 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1260 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1261 * with details of configuration from 'details'. Returns 0 if successful,
1262 * otherwise a positive errno value. On error, the previous QoS configuration
1265 * When this function changes the type of QoS (not just 'details'), this also
1266 * resets all queue configuration for 'netdev' to their defaults (which depend
1267 * on the specific type of QoS). Otherwise, the queue configuration for
1268 * 'netdev' is unchanged.
1270 * 'type' should be "" (to disable QoS) or one of the types returned by
1271 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1272 * documented as valid for the given 'type' in the "other_config" column in the
1273 * "QoS" table in vswitchd/vswitch.xml (which is built as
1274 * ovs-vswitchd.conf.db(8)).
1276 * NULL may be specified for 'details' if there are no configuration
1279 netdev_set_qos(struct netdev *netdev,
1280 const char *type, const struct smap *details)
1282 const struct netdev_class *class = netdev->netdev_class;
1288 if (class->set_qos) {
1290 static const struct smap empty = SMAP_INITIALIZER(&empty);
1293 return class->set_qos(netdev, type, details);
1295 return *type ? EOPNOTSUPP : 0;
1299 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1300 * successful, adds that information as string key-value pairs to 'details'.
1301 * Returns 0 if successful, otherwise a positive errno value.
1303 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1304 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1306 * The returned contents of 'details' should be documented as valid for the
1307 * given 'type' in the "other_config" column in the "Queue" table in
1308 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1310 * The caller must initialize 'details' (e.g. with smap_init()) before calling
1311 * this function. The caller must free 'details' when it is no longer needed
1312 * (e.g. with smap_destroy()). */
1314 netdev_get_queue(const struct netdev *netdev,
1315 unsigned int queue_id, struct smap *details)
1317 const struct netdev_class *class = netdev->netdev_class;
1320 retval = (class->get_queue
1321 ? class->get_queue(netdev, queue_id, details)
1324 smap_clear(details);
1329 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1330 * string pairs in 'details'. The contents of 'details' should be documented
1331 * as valid for the given 'type' in the "other_config" column in the "Queue"
1332 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1333 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1334 * given queue's configuration should be unmodified.
1336 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1337 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1339 * This function does not modify 'details', and the caller retains ownership of
1342 netdev_set_queue(struct netdev *netdev,
1343 unsigned int queue_id, const struct smap *details)
1345 const struct netdev_class *class = netdev->netdev_class;
1346 return (class->set_queue
1347 ? class->set_queue(netdev, queue_id, details)
1351 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1352 * of QoS may have a fixed set of queues, in which case attempts to delete them
1353 * will fail with EOPNOTSUPP.
1355 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1356 * given queue will be unmodified.
1358 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1359 * the current form of QoS (e.g. as returned by
1360 * netdev_get_n_queues(netdev)). */
1362 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1364 const struct netdev_class *class = netdev->netdev_class;
1365 return (class->delete_queue
1366 ? class->delete_queue(netdev, queue_id)
1370 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1371 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1372 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1373 * positive errno value and fills 'stats' with values indicating unsupported
1376 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1377 struct netdev_queue_stats *stats)
1379 const struct netdev_class *class = netdev->netdev_class;
1382 retval = (class->get_queue_stats
1383 ? class->get_queue_stats(netdev, queue_id, stats)
1386 stats->tx_bytes = UINT64_MAX;
1387 stats->tx_packets = UINT64_MAX;
1388 stats->tx_errors = UINT64_MAX;
1389 stats->created = LLONG_MIN;
1394 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1395 * its configuration, and the 'aux' specified by the caller. The order of
1396 * iteration is unspecified, but (when successful) each queue is visited
1399 * Calling this function may be more efficient than calling netdev_get_queue()
1402 * 'cb' must not modify or free the 'details' argument passed in. It may
1403 * delete or modify the queue passed in as its 'queue_id' argument. It may
1404 * modify but must not delete any other queue within 'netdev'. 'cb' should not
1405 * add new queues because this may cause some queues to be visited twice or not
1408 * Returns 0 if successful, otherwise a positive errno value. On error, some
1409 * configured queues may not have been included in the iteration. */
1411 netdev_dump_queues(const struct netdev *netdev,
1412 netdev_dump_queues_cb *cb, void *aux)
1414 const struct netdev_class *class = netdev->netdev_class;
1415 return (class->dump_queues
1416 ? class->dump_queues(netdev, cb, aux)
1420 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1421 * its statistics, and the 'aux' specified by the caller. The order of
1422 * iteration is unspecified, but (when successful) each queue is visited
1425 * Calling this function may be more efficient than calling
1426 * netdev_get_queue_stats() for every queue.
1428 * 'cb' must not modify or free the statistics passed in.
1430 * Returns 0 if successful, otherwise a positive errno value. On error, some
1431 * configured queues may not have been included in the iteration. */
1433 netdev_dump_queue_stats(const struct netdev *netdev,
1434 netdev_dump_queue_stats_cb *cb, void *aux)
1436 const struct netdev_class *class = netdev->netdev_class;
1437 return (class->dump_queue_stats
1438 ? class->dump_queue_stats(netdev, cb, aux)
1442 /* Returns a sequence number which indicates changes in one of 'netdev''s
1443 * properties. The returned sequence will be nonzero so that callers have a
1444 * value which they may use as a reset when tracking 'netdev'.
1446 * The returned sequence number will change whenever 'netdev''s flags,
1447 * features, ethernet address, or carrier changes. It may change for other
1448 * reasons as well, or no reason at all. */
1450 netdev_change_seq(const struct netdev *netdev)
1452 return netdev->netdev_class->change_seq(netdev);
1455 /* Returns the class type of 'netdev'.
1457 * The caller must not free the returned value. */
1459 netdev_get_type(const struct netdev *netdev)
1461 return netdev->netdev_class->type;
1464 /* Returns the class associated with 'netdev'. */
1465 const struct netdev_class *
1466 netdev_get_class(const struct netdev *netdev)
1468 return netdev->netdev_class;
1471 /* Returns the netdev with 'name' or NULL if there is none.
1473 * The caller must free the returned netdev with netdev_close(). */
1475 netdev_from_name(const char *name)
1476 OVS_EXCLUDED(netdev_mutex)
1478 struct netdev *netdev;
1480 ovs_mutex_lock(&netdev_mutex);
1481 netdev = shash_find_data(&netdev_shash, name);
1485 ovs_mutex_unlock(&netdev_mutex);
1490 /* Fills 'device_list' with devices that match 'netdev_class'.
1492 * The caller is responsible for initializing and destroying 'device_list' and
1493 * must close each device on the list. */
1495 netdev_get_devices(const struct netdev_class *netdev_class,
1496 struct shash *device_list)
1497 OVS_EXCLUDED(netdev_mutex)
1499 struct shash_node *node;
1501 ovs_mutex_lock(&netdev_mutex);
1502 SHASH_FOR_EACH (node, &netdev_shash) {
1503 struct netdev *dev = node->data;
1505 if (dev->netdev_class == netdev_class) {
1507 shash_add(device_list, node->name, node->data);
1510 ovs_mutex_unlock(&netdev_mutex);
1514 netdev_get_type_from_name(const char *name)
1516 struct netdev *dev = netdev_from_name(name);
1517 const char *type = dev ? netdev_get_type(dev) : NULL;
1523 netdev_rx_get_netdev(const struct netdev_rx *rx)
1525 ovs_assert(rx->netdev->ref_cnt > 0);
1530 netdev_rx_get_name(const struct netdev_rx *rx)
1532 return netdev_get_name(netdev_rx_get_netdev(rx));
1536 restore_all_flags(void *aux OVS_UNUSED)
1538 struct shash_node *node;
1540 SHASH_FOR_EACH (node, &netdev_shash) {
1541 struct netdev *netdev = node->data;
1542 const struct netdev_saved_flags *sf;
1543 enum netdev_flags saved_values;
1544 enum netdev_flags saved_flags;
1546 saved_values = saved_flags = 0;
1547 LIST_FOR_EACH (sf, node, &netdev->saved_flags_list) {
1548 saved_flags |= sf->saved_flags;
1549 saved_values &= ~sf->saved_flags;
1550 saved_values |= sf->saved_flags & sf->saved_values;
1553 enum netdev_flags old_flags;
1555 netdev->netdev_class->update_flags(netdev,
1556 saved_flags & saved_values,
1557 saved_flags & ~saved_values,