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>
28 #include "dynamic-string.h"
29 #include "fatal-signal.h"
32 #include "netdev-provider.h"
33 #include "netdev-vport.h"
35 #include "openflow/openflow.h"
37 #include "poll-loop.h"
44 VLOG_DEFINE_THIS_MODULE(netdev);
46 COVERAGE_DEFINE(netdev_received);
47 COVERAGE_DEFINE(netdev_sent);
48 COVERAGE_DEFINE(netdev_add_router);
49 COVERAGE_DEFINE(netdev_get_stats);
51 static struct shash netdev_classes = SHASH_INITIALIZER(&netdev_classes);
53 /* All created network devices. */
54 static struct shash netdev_dev_shash = SHASH_INITIALIZER(&netdev_dev_shash);
56 /* All open network devices. */
57 static struct list netdev_list = LIST_INITIALIZER(&netdev_list);
59 /* This is set pretty low because we probably won't learn anything from the
60 * additional log messages. */
61 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
63 static void close_all_netdevs(void *aux OVS_UNUSED);
64 static int restore_flags(struct netdev *netdev);
65 void update_device_args(struct netdev_dev *, const struct shash *args);
68 netdev_initialize(void)
75 fatal_signal_add_hook(close_all_netdevs, NULL, NULL, true);
76 netdev_vport_patch_register();
79 netdev_register_provider(&netdev_linux_class);
80 netdev_register_provider(&netdev_internal_class);
81 netdev_register_provider(&netdev_tap_class);
82 netdev_vport_tunnel_register();
85 netdev_register_provider(&netdev_tap_class);
86 netdev_register_provider(&netdev_bsd_class);
91 /* Performs periodic work needed by all the various kinds of netdevs.
93 * If your program opens any netdevs, it must call this function within its
98 struct shash_node *node;
99 SHASH_FOR_EACH(node, &netdev_classes) {
100 const struct netdev_class *netdev_class = node->data;
101 if (netdev_class->run) {
107 /* Arranges for poll_block() to wake up when netdev_run() needs to be called.
109 * If your program opens any netdevs, it must call this function within its
114 struct shash_node *node;
115 SHASH_FOR_EACH(node, &netdev_classes) {
116 const struct netdev_class *netdev_class = node->data;
117 if (netdev_class->wait) {
118 netdev_class->wait();
123 /* Initializes and registers a new netdev provider. After successful
124 * registration, new netdevs of that type can be opened using netdev_open(). */
126 netdev_register_provider(const struct netdev_class *new_class)
128 if (shash_find(&netdev_classes, new_class->type)) {
129 VLOG_WARN("attempted to register duplicate netdev provider: %s",
134 if (new_class->init) {
135 int error = new_class->init();
137 VLOG_ERR("failed to initialize %s network device class: %s",
138 new_class->type, strerror(error));
143 shash_add(&netdev_classes, new_class->type, new_class);
148 /* Unregisters a netdev provider. 'type' must have been previously
149 * registered and not currently be in use by any netdevs. After unregistration
150 * new netdevs of that type cannot be opened using netdev_open(). */
152 netdev_unregister_provider(const char *type)
154 struct shash_node *del_node, *netdev_dev_node;
156 del_node = shash_find(&netdev_classes, type);
158 VLOG_WARN("attempted to unregister a netdev provider that is not "
159 "registered: %s", type);
163 SHASH_FOR_EACH(netdev_dev_node, &netdev_dev_shash) {
164 struct netdev_dev *netdev_dev = netdev_dev_node->data;
165 if (!strcmp(netdev_dev->netdev_class->type, type)) {
166 VLOG_WARN("attempted to unregister in use netdev provider: %s",
172 shash_delete(&netdev_classes, del_node);
177 const struct netdev_class *
178 netdev_lookup_provider(const char *type)
181 return shash_find_data(&netdev_classes, type && type[0] ? type : "system");
184 /* Clears 'types' and enumerates the types of all currently registered netdev
185 * providers into it. The caller must first initialize the sset. */
187 netdev_enumerate_types(struct sset *types)
189 struct shash_node *node;
194 SHASH_FOR_EACH(node, &netdev_classes) {
195 const struct netdev_class *netdev_class = node->data;
196 sset_add(types, netdev_class->type);
200 /* Opens the network device named 'name' (e.g. "eth0") of the specified 'type'
201 * (e.g. "system") and returns zero if successful, otherwise a positive errno
202 * value. On success, sets '*netdevp' to the new network device, otherwise to
205 * Some network devices may need to be configured (with netdev_set_config())
206 * before they can be used. */
208 netdev_open(const char *name, const char *type, struct netdev **netdevp)
210 struct netdev_dev *netdev_dev;
216 netdev_dev = shash_find_data(&netdev_dev_shash, name);
219 const struct netdev_class *class;
221 class = netdev_lookup_provider(type);
223 VLOG_WARN("could not create netdev %s of unknown type %s",
227 error = class->create(class, name, &netdev_dev);
231 ovs_assert(netdev_dev->netdev_class == class);
235 error = netdev_dev->netdev_class->open(netdev_dev, netdevp);
238 netdev_dev->ref_cnt++;
240 if (!netdev_dev->ref_cnt) {
241 netdev_dev_uninit(netdev_dev, true);
248 /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
249 * or NULL if none are needed. */
251 netdev_set_config(struct netdev *netdev, const struct smap *args)
253 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
255 if (netdev_dev->netdev_class->set_config) {
256 struct smap no_args = SMAP_INITIALIZER(&no_args);
257 return netdev_dev->netdev_class->set_config(netdev_dev,
258 args ? args : &no_args);
259 } else if (args && !smap_is_empty(args)) {
260 VLOG_WARN("%s: arguments provided to device that is not configurable",
261 netdev_get_name(netdev));
267 /* Returns the current configuration for 'netdev' in 'args'. The caller must
268 * have already initialized 'args' with smap_init(). Returns 0 on success, in
269 * which case 'args' will be filled with 'netdev''s configuration. On failure
270 * returns a positive errno value, in which case 'args' will be empty.
272 * The caller owns 'args' and its contents and must eventually free them with
275 netdev_get_config(const struct netdev *netdev, struct smap *args)
277 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
281 if (netdev_dev->netdev_class->get_config) {
282 error = netdev_dev->netdev_class->get_config(netdev_dev, args);
293 const struct netdev_tunnel_config *
294 netdev_get_tunnel_config(const struct netdev *netdev)
296 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
298 if (netdev_dev->netdev_class->get_tunnel_config) {
299 return netdev_dev->netdev_class->get_tunnel_config(netdev_dev);
305 /* Closes and destroys 'netdev'. */
307 netdev_close(struct netdev *netdev)
310 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
312 ovs_assert(netdev_dev->ref_cnt);
313 netdev_dev->ref_cnt--;
314 netdev_uninit(netdev, true);
316 /* If the reference count for the netdev device is zero, destroy it. */
317 if (!netdev_dev->ref_cnt) {
318 netdev_dev_uninit(netdev_dev, true);
323 /* Returns true if a network device named 'name' is currently opened,
324 * otherwise false. */
326 netdev_is_open(const char *name)
328 return !!shash_find_data(&netdev_dev_shash, name);
331 /* Parses 'netdev_name_', which is of the form [type@]name into its component
332 * pieces. 'name' and 'type' must be freed by the caller. */
334 netdev_parse_name(const char *netdev_name_, char **name, char **type)
336 char *netdev_name = xstrdup(netdev_name_);
339 separator = strchr(netdev_name, '@');
343 *name = xstrdup(separator + 1);
346 *type = xstrdup("system");
350 /* Attempts to set up 'netdev' for receiving packets with netdev_recv().
351 * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
352 * indicates that the network device does not implement packet reception
353 * through this interface. */
355 netdev_listen(struct netdev *netdev)
357 int (*listen)(struct netdev *);
359 listen = netdev_get_dev(netdev)->netdev_class->listen;
360 return listen ? (listen)(netdev) : EOPNOTSUPP;
363 /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller
364 * must have initialized with sufficient room for the packet. The space
365 * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN
366 * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()).
367 * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN
368 * need not be included.)
370 * This function can only be expected to return a packet if ->listen() has
371 * been called successfully.
373 * If a packet is successfully retrieved, returns 0. In this case 'buffer' is
374 * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a
375 * positive errno value. Returns EAGAIN immediately if no packet is ready to
378 * Some network devices may not implement support for this function. In such
379 * cases this function will always return EOPNOTSUPP. */
381 netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
383 int (*recv)(struct netdev *, void *, size_t);
386 ovs_assert(buffer->size == 0);
387 ovs_assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
389 recv = netdev_get_dev(netdev)->netdev_class->recv;
391 ? (recv)(netdev, buffer->data, ofpbuf_tailroom(buffer))
394 COVERAGE_INC(netdev_received);
395 buffer->size += retval;
396 if (buffer->size < ETH_TOTAL_MIN) {
397 ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
405 /* Registers with the poll loop to wake up from the next call to poll_block()
406 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
408 netdev_recv_wait(struct netdev *netdev)
410 void (*recv_wait)(struct netdev *);
412 recv_wait = netdev_get_dev(netdev)->netdev_class->recv_wait;
418 /* Discards all packets waiting to be received from 'netdev'. */
420 netdev_drain(struct netdev *netdev)
422 int (*drain)(struct netdev *);
424 drain = netdev_get_dev(netdev)->netdev_class->drain;
425 return drain ? drain(netdev) : 0;
428 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
429 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
430 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
431 * the packet is too big or too small to transmit on the device.
433 * The caller retains ownership of 'buffer' in all cases.
435 * The kernel maintains a packet transmission queue, so the caller is not
436 * expected to do additional queuing of packets.
438 * Some network devices may not implement support for this function. In such
439 * cases this function will always return EOPNOTSUPP. */
441 netdev_send(struct netdev *netdev, const struct ofpbuf *buffer)
443 int (*send)(struct netdev *, const void *, size_t);
446 send = netdev_get_dev(netdev)->netdev_class->send;
447 error = send ? (send)(netdev, buffer->data, buffer->size) : EOPNOTSUPP;
449 COVERAGE_INC(netdev_sent);
454 /* Registers with the poll loop to wake up from the next call to poll_block()
455 * when the packet transmission queue has sufficient room to transmit a packet
456 * with netdev_send().
458 * The kernel maintains a packet transmission queue, so the client is not
459 * expected to do additional queuing of packets. Thus, this function is
460 * unlikely to ever be used. It is included for completeness. */
462 netdev_send_wait(struct netdev *netdev)
464 void (*send_wait)(struct netdev *);
466 send_wait = netdev_get_dev(netdev)->netdev_class->send_wait;
472 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
473 * otherwise a positive errno value. */
475 netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN])
477 return netdev_get_dev(netdev)->netdev_class->set_etheraddr(netdev, mac);
480 /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the
481 * the MAC address into 'mac'. On failure, returns a positive errno value and
482 * clears 'mac' to all-zeros. */
484 netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN])
486 return netdev_get_dev(netdev)->netdev_class->get_etheraddr(netdev, mac);
489 /* Returns the name of the network device that 'netdev' represents,
490 * e.g. "eth0". The caller must not modify or free the returned string. */
492 netdev_get_name(const struct netdev *netdev)
494 return netdev_get_dev(netdev)->name;
497 /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted
498 * (and received) packets, in bytes, not including the hardware header; thus,
499 * this is typically 1500 bytes for Ethernet devices.
501 * If successful, returns 0 and stores the MTU size in '*mtup'. Returns
502 * EOPNOTSUPP if 'netdev' does not have an MTU (as e.g. some tunnels do not).
503 * On other failure, returns a positive errno value. On failure, sets '*mtup'
506 netdev_get_mtu(const struct netdev *netdev, int *mtup)
508 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
511 error = class->get_mtu ? class->get_mtu(netdev, mtup) : EOPNOTSUPP;
514 if (error != EOPNOTSUPP) {
515 VLOG_DBG_RL(&rl, "failed to retrieve MTU for network device %s: "
516 "%s", netdev_get_name(netdev), strerror(error));
522 /* Sets the MTU of 'netdev'. The MTU is the maximum size of transmitted
523 * (and received) packets, in bytes.
525 * If successful, returns 0. Returns EOPNOTSUPP if 'netdev' does not have an
526 * MTU (as e.g. some tunnels do not). On other failure, returns a positive
529 netdev_set_mtu(const struct netdev *netdev, int mtu)
531 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
534 error = class->set_mtu ? class->set_mtu(netdev, mtu) : EOPNOTSUPP;
535 if (error && error != EOPNOTSUPP) {
536 VLOG_DBG_RL(&rl, "failed to set MTU for network device %s: %s",
537 netdev_get_name(netdev), strerror(error));
543 /* Returns the ifindex of 'netdev', if successful, as a positive number. On
544 * failure, returns a negative errno value.
546 * The desired semantics of the ifindex value are a combination of those
547 * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex
548 * value should be unique within a host and remain stable at least until
549 * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber"
550 * but many systems do not follow this rule anyhow.
552 * Some network devices may not implement support for this function. In such
553 * cases this function will always return -EOPNOTSUPP.
556 netdev_get_ifindex(const struct netdev *netdev)
558 int (*get_ifindex)(const struct netdev *);
560 get_ifindex = netdev_get_dev(netdev)->netdev_class->get_ifindex;
562 return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP;
565 /* Stores the features supported by 'netdev' into each of '*current',
566 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
567 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
568 * successful, otherwise a positive errno value. On failure, all of the
569 * passed-in values are set to 0.
571 * Some network devices may not implement support for this function. In such
572 * cases this function will always return EOPNOTSUPP. */
574 netdev_get_features(const struct netdev *netdev,
575 enum netdev_features *current,
576 enum netdev_features *advertised,
577 enum netdev_features *supported,
578 enum netdev_features *peer)
580 int (*get_features)(const struct netdev *netdev,
581 enum netdev_features *current,
582 enum netdev_features *advertised,
583 enum netdev_features *supported,
584 enum netdev_features *peer);
585 enum netdev_features dummy[4];
592 advertised = &dummy[1];
595 supported = &dummy[2];
601 get_features = netdev_get_dev(netdev)->netdev_class->get_features;
603 ? get_features(netdev, current, advertised, supported,
607 *current = *advertised = *supported = *peer = 0;
612 /* Returns the maximum speed of a network connection that has the NETDEV_F_*
613 * bits in 'features', in bits per second. If no bits that indicate a speed
614 * are set in 'features', returns 'default_bps'. */
616 netdev_features_to_bps(enum netdev_features features,
617 uint64_t default_bps)
620 F_1000000MB = NETDEV_F_1TB_FD,
621 F_100000MB = NETDEV_F_100GB_FD,
622 F_40000MB = NETDEV_F_40GB_FD,
623 F_10000MB = NETDEV_F_10GB_FD,
624 F_1000MB = NETDEV_F_1GB_HD | NETDEV_F_1GB_FD,
625 F_100MB = NETDEV_F_100MB_HD | NETDEV_F_100MB_FD,
626 F_10MB = NETDEV_F_10MB_HD | NETDEV_F_10MB_FD
629 return ( features & F_1000000MB ? UINT64_C(1000000000000)
630 : features & F_100000MB ? UINT64_C(100000000000)
631 : features & F_40000MB ? UINT64_C(40000000000)
632 : features & F_10000MB ? UINT64_C(10000000000)
633 : features & F_1000MB ? UINT64_C(1000000000)
634 : features & F_100MB ? UINT64_C(100000000)
635 : features & F_10MB ? UINT64_C(10000000)
639 /* Returns true if any of the NETDEV_F_* bits that indicate a full-duplex link
640 * are set in 'features', otherwise false. */
642 netdev_features_is_full_duplex(enum netdev_features features)
644 return (features & (NETDEV_F_10MB_FD | NETDEV_F_100MB_FD | NETDEV_F_1GB_FD
645 | NETDEV_F_10GB_FD | NETDEV_F_40GB_FD
646 | NETDEV_F_100GB_FD | NETDEV_F_1TB_FD)) != 0;
649 /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if
650 * successful, otherwise a positive errno value. */
652 netdev_set_advertisements(struct netdev *netdev,
653 enum netdev_features advertise)
655 return (netdev_get_dev(netdev)->netdev_class->set_advertisements
656 ? netdev_get_dev(netdev)->netdev_class->set_advertisements(
661 /* If 'netdev' has an assigned IPv4 address, sets '*address' to that address
662 * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive
663 * errno value and sets '*address' to 0 (INADDR_ANY).
665 * The following error values have well-defined meanings:
667 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address.
669 * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
671 * 'address' or 'netmask' or both may be null, in which case the address or
672 * netmask is not reported. */
674 netdev_get_in4(const struct netdev *netdev,
675 struct in_addr *address_, struct in_addr *netmask_)
677 struct in_addr address;
678 struct in_addr netmask;
681 error = (netdev_get_dev(netdev)->netdev_class->get_in4
682 ? netdev_get_dev(netdev)->netdev_class->get_in4(netdev,
686 address_->s_addr = error ? 0 : address.s_addr;
689 netmask_->s_addr = error ? 0 : netmask.s_addr;
694 /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If
695 * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a
696 * positive errno value. */
698 netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask)
700 return (netdev_get_dev(netdev)->netdev_class->set_in4
701 ? netdev_get_dev(netdev)->netdev_class->set_in4(netdev, addr, mask)
705 /* Obtains ad IPv4 address from device name and save the address in
706 * in4. Returns 0 if successful, otherwise a positive errno value.
709 netdev_get_in4_by_name(const char *device_name, struct in_addr *in4)
711 struct netdev *netdev;
714 error = netdev_open(device_name, "system", &netdev);
716 in4->s_addr = htonl(0);
720 error = netdev_get_in4(netdev, in4, NULL);
721 netdev_close(netdev);
725 /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds
728 netdev_add_router(struct netdev *netdev, struct in_addr router)
730 COVERAGE_INC(netdev_add_router);
731 return (netdev_get_dev(netdev)->netdev_class->add_router
732 ? netdev_get_dev(netdev)->netdev_class->add_router(netdev, router)
736 /* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to
737 * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0,
738 * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a
739 * next hop is found, stores the next hop gateway's address (0 if 'host' is on
740 * a directly connected network) in '*next_hop' and a copy of the name of the
741 * device to reach 'host' in '*netdev_name', and returns 0. The caller is
742 * responsible for freeing '*netdev_name' (by calling free()). */
744 netdev_get_next_hop(const struct netdev *netdev,
745 const struct in_addr *host, struct in_addr *next_hop,
748 int error = (netdev_get_dev(netdev)->netdev_class->get_next_hop
749 ? netdev_get_dev(netdev)->netdev_class->get_next_hop(
750 host, next_hop, netdev_name)
753 next_hop->s_addr = 0;
759 /* Populates 'smap' with status information.
761 * Populates 'smap' with 'netdev' specific status information. This
762 * information may be used to populate the status column of the Interface table
763 * as defined in ovs-vswitchd.conf.db(5). */
765 netdev_get_status(const struct netdev *netdev, struct smap *smap)
767 struct netdev_dev *dev = netdev_get_dev(netdev);
769 return (dev->netdev_class->get_status
770 ? dev->netdev_class->get_status(netdev, smap)
774 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and
775 * returns 0. Otherwise, returns a positive errno value and sets '*in6' to
776 * all-zero-bits (in6addr_any).
778 * The following error values have well-defined meanings:
780 * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address.
782 * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'.
784 * 'in6' may be null, in which case the address itself is not reported. */
786 netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6)
788 struct in6_addr dummy;
791 error = (netdev_get_dev(netdev)->netdev_class->get_in6
792 ? netdev_get_dev(netdev)->netdev_class->get_in6(netdev,
796 memset(in6, 0, sizeof *in6);
801 /* On 'netdev', turns off the flags in 'off' and then turns on the flags in
802 * 'on'. If 'permanent' is true, the changes will persist; otherwise, they
803 * will be reverted when 'netdev' is closed or the program exits. Returns 0 if
804 * successful, otherwise a positive errno value. */
806 do_update_flags(struct netdev *netdev, enum netdev_flags off,
807 enum netdev_flags on, enum netdev_flags *old_flagsp,
810 enum netdev_flags old_flags;
813 error = netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
814 off & ~on, on, &old_flags);
816 VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s",
817 off || on ? "set" : "get", netdev_get_name(netdev),
820 } else if ((off || on) && !permanent) {
821 enum netdev_flags new_flags = (old_flags & ~off) | on;
822 enum netdev_flags changed_flags = old_flags ^ new_flags;
824 if (!netdev->changed_flags) {
825 netdev->save_flags = old_flags;
827 netdev->changed_flags |= changed_flags;
831 *old_flagsp = old_flags;
836 /* Obtains the current flags for 'netdev' and stores them into '*flagsp'.
837 * Returns 0 if successful, otherwise a positive errno value. On failure,
838 * stores 0 into '*flagsp'. */
840 netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp)
842 struct netdev *netdev = CONST_CAST(struct netdev *, netdev_);
843 return do_update_flags(netdev, 0, 0, flagsp, false);
846 /* Sets the flags for 'netdev' to 'flags'.
847 * If 'permanent' is true, the changes will persist; otherwise, they
848 * will be reverted when 'netdev' is closed or the program exits.
849 * Returns 0 if successful, otherwise a positive errno value. */
851 netdev_set_flags(struct netdev *netdev, enum netdev_flags flags,
854 return do_update_flags(netdev, -1, flags, NULL, permanent);
857 /* Turns on the specified 'flags' on 'netdev'.
858 * If 'permanent' is true, the changes will persist; otherwise, they
859 * will be reverted when 'netdev' is closed or the program exits.
860 * Returns 0 if successful, otherwise a positive errno value. */
862 netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags,
865 return do_update_flags(netdev, 0, flags, NULL, permanent);
868 /* Turns off the specified 'flags' on 'netdev'.
869 * If 'permanent' is true, the changes will persist; otherwise, they
870 * will be reverted when 'netdev' is closed or the program exits.
871 * Returns 0 if successful, otherwise a positive errno value. */
873 netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags,
876 return do_update_flags(netdev, flags, 0, NULL, permanent);
879 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
880 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
881 * returns 0. Otherwise, it returns a positive errno value; in particular,
882 * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */
884 netdev_arp_lookup(const struct netdev *netdev,
885 ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
887 int error = (netdev_get_dev(netdev)->netdev_class->arp_lookup
888 ? netdev_get_dev(netdev)->netdev_class->arp_lookup(netdev,
892 memset(mac, 0, ETH_ADDR_LEN);
897 /* Returns true if carrier is active (link light is on) on 'netdev'. */
899 netdev_get_carrier(const struct netdev *netdev)
902 enum netdev_flags flags;
905 netdev_get_flags(netdev, &flags);
906 if (!(flags & NETDEV_UP)) {
910 if (!netdev_get_dev(netdev)->netdev_class->get_carrier) {
914 error = netdev_get_dev(netdev)->netdev_class->get_carrier(netdev,
917 VLOG_DBG("%s: failed to get network device carrier status, assuming "
918 "down: %s", netdev_get_name(netdev), strerror(error));
925 /* Returns the number of times 'netdev''s carrier has changed. */
927 netdev_get_carrier_resets(const struct netdev *netdev)
929 return (netdev_get_dev(netdev)->netdev_class->get_carrier_resets
930 ? netdev_get_dev(netdev)->netdev_class->get_carrier_resets(netdev)
934 /* Attempts to force netdev_get_carrier() to poll 'netdev''s MII registers for
935 * link status instead of checking 'netdev''s carrier. 'netdev''s MII
936 * registers will be polled once ever 'interval' milliseconds. If 'netdev'
937 * does not support MII, another method may be used as a fallback. If
938 * 'interval' is less than or equal to zero, reverts netdev_get_carrier() to
939 * its normal behavior.
941 * Returns 0 if successful, otherwise a positive errno value. */
943 netdev_set_miimon_interval(struct netdev *netdev, long long int interval)
945 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
946 return (netdev_dev->netdev_class->set_miimon_interval
947 ? netdev_dev->netdev_class->set_miimon_interval(netdev, interval)
951 /* Retrieves current device stats for 'netdev'. */
953 netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
957 COVERAGE_INC(netdev_get_stats);
958 error = (netdev_get_dev(netdev)->netdev_class->get_stats
959 ? netdev_get_dev(netdev)->netdev_class->get_stats(netdev, stats)
962 memset(stats, 0xff, sizeof *stats);
967 /* Attempts to change the stats for 'netdev' to those provided in 'stats'.
968 * Returns 0 if successful, otherwise a positive errno value.
970 * This will probably fail for most network devices. Some devices might only
971 * allow setting their stats to 0. */
973 netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
975 return (netdev_get_dev(netdev)->netdev_class->set_stats
976 ? netdev_get_dev(netdev)->netdev_class->set_stats(netdev, stats)
980 /* Attempts to set input rate limiting (policing) policy, such that up to
981 * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst
982 * size of 'kbits' kb. */
984 netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate,
985 uint32_t kbits_burst)
987 return (netdev_get_dev(netdev)->netdev_class->set_policing
988 ? netdev_get_dev(netdev)->netdev_class->set_policing(netdev,
989 kbits_rate, kbits_burst)
993 /* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it
994 * empty if 'netdev' does not support QoS. Any names added to 'types' should
995 * be documented as valid for the "type" column in the "QoS" table in
996 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
998 * Every network device supports disabling QoS with a type of "", but this type
999 * will not be added to 'types'.
1001 * The caller must initialize 'types' (e.g. with sset_init()) before calling
1002 * this function. The caller is responsible for destroying 'types' (e.g. with
1003 * sset_destroy()) when it is no longer needed.
1005 * Returns 0 if successful, otherwise a positive errno value. */
1007 netdev_get_qos_types(const struct netdev *netdev, struct sset *types)
1009 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1010 return (class->get_qos_types
1011 ? class->get_qos_types(netdev, types)
1015 /* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS,
1016 * which should be "" or one of the types returned by netdev_get_qos_types()
1017 * for 'netdev'. Returns 0 if successful, otherwise a positive errno value.
1018 * On success, initializes 'caps' with the QoS capabilities; on failure, clears
1019 * 'caps' to all zeros. */
1021 netdev_get_qos_capabilities(const struct netdev *netdev, const char *type,
1022 struct netdev_qos_capabilities *caps)
1024 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1027 int retval = (class->get_qos_capabilities
1028 ? class->get_qos_capabilities(netdev, type, caps)
1031 memset(caps, 0, sizeof *caps);
1035 /* Every netdev supports turning off QoS. */
1036 memset(caps, 0, sizeof *caps);
1041 /* Obtains the number of queues supported by 'netdev' for the specified 'type'
1042 * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores
1043 * the number of queues (zero on failure) in '*n_queuesp'.
1045 * This is just a simple wrapper around netdev_get_qos_capabilities(). */
1047 netdev_get_n_queues(const struct netdev *netdev,
1048 const char *type, unsigned int *n_queuesp)
1050 struct netdev_qos_capabilities caps;
1053 retval = netdev_get_qos_capabilities(netdev, type, &caps);
1054 *n_queuesp = caps.n_queues;
1058 /* Queries 'netdev' about its currently configured form of QoS. If successful,
1059 * stores the name of the current form of QoS into '*typep', stores any details
1060 * of configuration as string key-value pairs in 'details', and returns 0. On
1061 * failure, sets '*typep' to NULL and returns a positive errno value.
1063 * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'.
1065 * The caller must initialize 'details' as an empty smap (e.g. with
1066 * smap_init()) before calling this function. The caller must free 'details'
1067 * when it is no longer needed (e.g. with smap_destroy()).
1069 * The caller must not modify or free '*typep'.
1071 * '*typep' will be one of the types returned by netdev_get_qos_types() for
1072 * 'netdev'. The contents of 'details' should be documented as valid for
1073 * '*typep' in the "other_config" column in the "QoS" table in
1074 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */
1076 netdev_get_qos(const struct netdev *netdev,
1077 const char **typep, struct smap *details)
1079 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1082 if (class->get_qos) {
1083 retval = class->get_qos(netdev, typep, details);
1086 smap_clear(details);
1090 /* 'netdev' doesn't support QoS, so report that QoS is disabled. */
1096 /* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type'
1097 * with details of configuration from 'details'. Returns 0 if successful,
1098 * otherwise a positive errno value. On error, the previous QoS configuration
1101 * When this function changes the type of QoS (not just 'details'), this also
1102 * resets all queue configuration for 'netdev' to their defaults (which depend
1103 * on the specific type of QoS). Otherwise, the queue configuration for
1104 * 'netdev' is unchanged.
1106 * 'type' should be "" (to disable QoS) or one of the types returned by
1107 * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be
1108 * documented as valid for the given 'type' in the "other_config" column in the
1109 * "QoS" table in vswitchd/vswitch.xml (which is built as
1110 * ovs-vswitchd.conf.db(8)).
1112 * NULL may be specified for 'details' if there are no configuration
1115 netdev_set_qos(struct netdev *netdev,
1116 const char *type, const struct smap *details)
1118 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1124 if (class->set_qos) {
1126 static struct smap empty = SMAP_INITIALIZER(&empty);
1129 return class->set_qos(netdev, type, details);
1131 return *type ? EOPNOTSUPP : 0;
1135 /* Queries 'netdev' for information about the queue numbered 'queue_id'. If
1136 * successful, adds that information as string key-value pairs to 'details'.
1137 * Returns 0 if successful, otherwise a positive errno value.
1139 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1140 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1142 * The returned contents of 'details' should be documented as valid for the
1143 * given 'type' in the "other_config" column in the "Queue" table in
1144 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1146 * The caller must initialize 'details' (e.g. with smap_init()) before calling
1147 * this function. The caller must free 'details' when it is no longer needed
1148 * (e.g. with smap_destroy()). */
1150 netdev_get_queue(const struct netdev *netdev,
1151 unsigned int queue_id, struct smap *details)
1153 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1156 retval = (class->get_queue
1157 ? class->get_queue(netdev, queue_id, details)
1160 smap_clear(details);
1165 /* Configures the queue numbered 'queue_id' on 'netdev' with the key-value
1166 * string pairs in 'details'. The contents of 'details' should be documented
1167 * as valid for the given 'type' in the "other_config" column in the "Queue"
1168 * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
1169 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1170 * given queue's configuration should be unmodified.
1172 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1173 * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
1175 * This function does not modify 'details', and the caller retains ownership of
1178 netdev_set_queue(struct netdev *netdev,
1179 unsigned int queue_id, const struct smap *details)
1181 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1182 return (class->set_queue
1183 ? class->set_queue(netdev, queue_id, details)
1187 /* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds
1188 * of QoS may have a fixed set of queues, in which case attempts to delete them
1189 * will fail with EOPNOTSUPP.
1191 * Returns 0 if successful, otherwise a positive errno value. On failure, the
1192 * given queue will be unmodified.
1194 * 'queue_id' must be less than the number of queues supported by 'netdev' for
1195 * the current form of QoS (e.g. as returned by
1196 * netdev_get_n_queues(netdev)). */
1198 netdev_delete_queue(struct netdev *netdev, unsigned int queue_id)
1200 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1201 return (class->delete_queue
1202 ? class->delete_queue(netdev, queue_id)
1206 /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and
1207 * fills 'stats' with the queue's statistics; individual members of 'stats' may
1208 * be set to all-1-bits if the statistic is unavailable. On failure, returns a
1209 * positive errno value and fills 'stats' with all-1-bits. */
1211 netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id,
1212 struct netdev_queue_stats *stats)
1214 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1217 retval = (class->get_queue_stats
1218 ? class->get_queue_stats(netdev, queue_id, stats)
1221 memset(stats, 0xff, sizeof *stats);
1226 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1227 * its configuration, and the 'aux' specified by the caller. The order of
1228 * iteration is unspecified, but (when successful) each queue is visited
1231 * Calling this function may be more efficient than calling netdev_get_queue()
1234 * 'cb' must not modify or free the 'details' argument passed in. It may
1235 * delete or modify the queue passed in as its 'queue_id' argument. It may
1236 * modify but must not delete any other queue within 'netdev'. 'cb' should not
1237 * add new queues because this may cause some queues to be visited twice or not
1240 * Returns 0 if successful, otherwise a positive errno value. On error, some
1241 * configured queues may not have been included in the iteration. */
1243 netdev_dump_queues(const struct netdev *netdev,
1244 netdev_dump_queues_cb *cb, void *aux)
1246 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1247 return (class->dump_queues
1248 ? class->dump_queues(netdev, cb, aux)
1252 /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID,
1253 * its statistics, and the 'aux' specified by the caller. The order of
1254 * iteration is unspecified, but (when successful) each queue is visited
1257 * Calling this function may be more efficient than calling
1258 * netdev_get_queue_stats() for every queue.
1260 * 'cb' must not modify or free the statistics passed in.
1262 * Returns 0 if successful, otherwise a positive errno value. On error, some
1263 * configured queues may not have been included in the iteration. */
1265 netdev_dump_queue_stats(const struct netdev *netdev,
1266 netdev_dump_queue_stats_cb *cb, void *aux)
1268 const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class;
1269 return (class->dump_queue_stats
1270 ? class->dump_queue_stats(netdev, cb, aux)
1274 /* Returns a sequence number which indicates changes in one of 'netdev''s
1275 * properties. The returned sequence will be nonzero so that callers have a
1276 * value which they may use as a reset when tracking 'netdev'.
1278 * The returned sequence number will change whenever 'netdev''s flags,
1279 * features, ethernet address, or carrier changes. It may change for other
1280 * reasons as well, or no reason at all. */
1282 netdev_change_seq(const struct netdev *netdev)
1284 return netdev_get_dev(netdev)->netdev_class->change_seq(netdev);
1287 /* Initializes 'netdev_dev' as a netdev device named 'name' of the specified
1288 * 'netdev_class'. This function is ordinarily called from a netdev provider's
1289 * 'create' function.
1291 * This function adds 'netdev_dev' to a netdev-owned shash, so it is
1292 * very important that 'netdev_dev' only be freed after calling
1293 * the refcount drops to zero. */
1295 netdev_dev_init(struct netdev_dev *netdev_dev, const char *name,
1296 const struct netdev_class *netdev_class)
1298 ovs_assert(!shash_find(&netdev_dev_shash, name));
1300 memset(netdev_dev, 0, sizeof *netdev_dev);
1301 netdev_dev->netdev_class = netdev_class;
1302 netdev_dev->name = xstrdup(name);
1303 netdev_dev->node = shash_add(&netdev_dev_shash, name, netdev_dev);
1306 /* Undoes the results of initialization.
1308 * Normally this function does not need to be called as netdev_close has
1309 * the same effect when the refcount drops to zero.
1310 * However, it may be called by providers due to an error on creation
1311 * that occurs after initialization. It this case netdev_close() would
1312 * never be called. */
1314 netdev_dev_uninit(struct netdev_dev *netdev_dev, bool destroy)
1316 char *name = netdev_dev->name;
1318 ovs_assert(!netdev_dev->ref_cnt);
1320 shash_delete(&netdev_dev_shash, netdev_dev->node);
1323 netdev_dev->netdev_class->destroy(netdev_dev);
1328 /* Returns the class type of 'netdev_dev'.
1330 * The caller must not free the returned value. */
1332 netdev_dev_get_type(const struct netdev_dev *netdev_dev)
1334 return netdev_dev->netdev_class->type;
1337 /* Returns the class associated with 'netdev_dev'. */
1338 const struct netdev_class *
1339 netdev_dev_get_class(const struct netdev_dev *netdev_dev)
1341 return netdev_dev->netdev_class;
1344 /* Returns the name of 'netdev_dev'.
1346 * The caller must not free the returned value. */
1348 netdev_dev_get_name(const struct netdev_dev *netdev_dev)
1350 return netdev_dev->name;
1353 /* Returns the netdev_dev with 'name' or NULL if there is none.
1355 * The caller must not free the returned value. */
1357 netdev_dev_from_name(const char *name)
1359 return shash_find_data(&netdev_dev_shash, name);
1362 /* Fills 'device_list' with devices that match 'netdev_class'.
1364 * The caller is responsible for initializing and destroying 'device_list'
1365 * but the contained netdev_devs must not be freed. */
1367 netdev_dev_get_devices(const struct netdev_class *netdev_class,
1368 struct shash *device_list)
1370 struct shash_node *node;
1371 SHASH_FOR_EACH (node, &netdev_dev_shash) {
1372 struct netdev_dev *dev = node->data;
1374 if (dev->netdev_class == netdev_class) {
1375 shash_add(device_list, node->name, node->data);
1380 /* Initializes 'netdev' as a instance of the netdev_dev.
1382 * This function adds 'netdev' to a netdev-owned linked list, so it is very
1383 * important that 'netdev' only be freed after calling netdev_close(). */
1385 netdev_init(struct netdev *netdev, struct netdev_dev *netdev_dev)
1387 memset(netdev, 0, sizeof *netdev);
1388 netdev->netdev_dev = netdev_dev;
1389 list_push_back(&netdev_list, &netdev->node);
1392 /* Undoes the results of initialization.
1394 * Normally this function only needs to be called from netdev_close().
1395 * However, it may be called by providers due to an error on opening
1396 * that occurs after initialization. It this case netdev_close() would
1397 * never be called. */
1399 netdev_uninit(struct netdev *netdev, bool close)
1401 /* Restore flags that we changed, if any. */
1402 int error = restore_flags(netdev);
1403 list_remove(&netdev->node);
1405 VLOG_WARN("failed to restore network device flags on %s: %s",
1406 netdev_get_name(netdev), strerror(error));
1410 netdev_get_dev(netdev)->netdev_class->close(netdev);
1415 /* Returns the class type of 'netdev'.
1417 * The caller must not free the returned value. */
1419 netdev_get_type(const struct netdev *netdev)
1421 return netdev_get_dev(netdev)->netdev_class->type;
1426 netdev_get_type_from_name(const char *name)
1428 const struct netdev_dev *dev = netdev_dev_from_name(name);
1429 return dev ? netdev_dev_get_type(dev) : NULL;
1433 netdev_get_dev(const struct netdev *netdev)
1435 return netdev->netdev_dev;
1438 /* Restore the network device flags on 'netdev' to those that were active
1439 * before we changed them. Returns 0 if successful, otherwise a positive
1442 * To avoid reentry, the caller must ensure that fatal signals are blocked. */
1444 restore_flags(struct netdev *netdev)
1446 if (netdev->changed_flags) {
1447 enum netdev_flags restore = netdev->save_flags & netdev->changed_flags;
1448 enum netdev_flags old_flags;
1449 return netdev_get_dev(netdev)->netdev_class->update_flags(netdev,
1450 netdev->changed_flags & ~restore,
1451 restore, &old_flags);
1456 /* Close all netdevs on shutdown so they can do any needed cleanup such as
1457 * destroying devices, restoring flags, etc. */
1459 close_all_netdevs(void *aux OVS_UNUSED)
1461 struct netdev *netdev, *next;
1462 LIST_FOR_EACH_SAFE(netdev, next, node, &netdev_list) {
1463 netdev_close(netdev);