2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
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.
21 #include <arpa/inet.h>
23 #include <linux/gen_stats.h>
24 #include <linux/if_tun.h>
26 #include <linux/types.h>
27 #include <linux/ethtool.h>
28 #include <linux/mii.h>
29 #include <linux/pkt_sched.h>
30 #include <linux/rtnetlink.h>
31 #include <linux/sockios.h>
32 #include <linux/version.h>
33 #include <sys/types.h>
34 #include <sys/ioctl.h>
35 #include <sys/socket.h>
36 #include <netpacket/packet.h>
37 #include <net/ethernet.h>
39 #include <linux/if_tunnel.h>
40 #include <net/if_arp.h>
41 #include <net/if_packet.h>
42 #include <net/route.h>
43 #include <netinet/in.h>
50 #include "dpif-linux.h"
51 #include "dynamic-string.h"
52 #include "fatal-signal.h"
55 #include "netdev-provider.h"
56 #include "netdev-vport.h"
58 #include "netlink-socket.h"
60 #include "openflow/openflow.h"
62 #include "poll-loop.h"
63 #include "rtnetlink.h"
64 #include "rtnetlink-link.h"
65 #include "socket-util.h"
70 VLOG_DEFINE_THIS_MODULE(netdev_linux);
72 COVERAGE_DEFINE(netdev_get_vlan_vid);
73 COVERAGE_DEFINE(netdev_set_policing);
74 COVERAGE_DEFINE(netdev_arp_lookup);
75 COVERAGE_DEFINE(netdev_get_ifindex);
76 COVERAGE_DEFINE(netdev_get_hwaddr);
77 COVERAGE_DEFINE(netdev_set_hwaddr);
78 COVERAGE_DEFINE(netdev_ethtool);
80 /* These were introduced in Linux 2.6.14, so they might be missing if we have
82 #ifndef ADVERTISED_Pause
83 #define ADVERTISED_Pause (1 << 13)
85 #ifndef ADVERTISED_Asym_Pause
86 #define ADVERTISED_Asym_Pause (1 << 14)
89 /* This was introduced in Linux 2.6.25, so it might be missing if we have old
92 #define TC_RTAB_SIZE 1024
95 static struct rtnetlink_notifier netdev_linux_cache_notifier;
96 static int cache_notifier_refcount;
99 VALID_IFINDEX = 1 << 0,
100 VALID_ETHERADDR = 1 << 1,
104 VALID_CARRIER = 1 << 5,
105 VALID_IS_PSEUDO = 1 << 6, /* Represents is_internal and is_tap. */
106 VALID_POLICING = 1 << 7,
107 VALID_HAVE_VPORT_STATS = 1 << 8
115 /* Traffic control. */
117 /* An instance of a traffic control class. Always associated with a particular
120 * Each TC implementation subclasses this with whatever additional data it
123 const struct tc_ops *ops;
124 struct hmap queues; /* Contains "struct tc_queue"s.
125 * Read by generic TC layer.
126 * Written only by TC implementation. */
129 /* One traffic control queue.
131 * Each TC implementation subclasses this with whatever additional data it
134 struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */
135 unsigned int queue_id; /* OpenFlow queue ID. */
138 /* A particular kind of traffic control. Each implementation generally maps to
139 * one particular Linux qdisc class.
141 * The functions below return 0 if successful or a positive errno value on
142 * failure, except where otherwise noted. All of them must be provided, except
143 * where otherwise noted. */
145 /* Name used by kernel in the TCA_KIND attribute of tcmsg, e.g. "htb".
146 * This is null for tc_ops_default and tc_ops_other, for which there are no
147 * appropriate values. */
148 const char *linux_name;
150 /* Name used in OVS database, e.g. "linux-htb". Must be nonnull. */
151 const char *ovs_name;
153 /* Number of supported OpenFlow queues, 0 for qdiscs that have no
154 * queues. The queues are numbered 0 through n_queues - 1. */
155 unsigned int n_queues;
157 /* Called to install this TC class on 'netdev'. The implementation should
158 * make the Netlink calls required to set up 'netdev' with the right qdisc
159 * and configure it according to 'details'. The implementation may assume
160 * that the current qdisc is the default; that is, there is no need for it
161 * to delete the current qdisc before installing itself.
163 * The contents of 'details' should be documented as valid for 'ovs_name'
164 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
165 * (which is built as ovs-vswitchd.conf.db(8)).
167 * This function must return 0 if and only if it sets 'netdev->tc' to an
168 * initialized 'struct tc'.
170 * (This function is null for tc_ops_other, which cannot be installed. For
171 * other TC classes it should always be nonnull.) */
172 int (*tc_install)(struct netdev *netdev, const struct shash *details);
174 /* Called when the netdev code determines (through a Netlink query) that
175 * this TC class's qdisc is installed on 'netdev', but we didn't install
176 * it ourselves and so don't know any of the details.
178 * 'nlmsg' is the kernel reply to a RTM_GETQDISC Netlink message for
179 * 'netdev'. The TCA_KIND attribute of 'nlmsg' is 'linux_name'. The
180 * implementation should parse the other attributes of 'nlmsg' as
181 * necessary to determine its configuration. If necessary it should also
182 * use Netlink queries to determine the configuration of queues on
185 * This function must return 0 if and only if it sets 'netdev->tc' to an
186 * initialized 'struct tc'. */
187 int (*tc_load)(struct netdev *netdev, struct ofpbuf *nlmsg);
189 /* Destroys the data structures allocated by the implementation as part of
190 * 'tc'. (This includes destroying 'tc->queues' by calling
193 * The implementation should not need to perform any Netlink calls. If
194 * desirable, the caller is responsible for deconfiguring the kernel qdisc.
195 * (But it may not be desirable.)
197 * This function may be null if 'tc' is trivial. */
198 void (*tc_destroy)(struct tc *tc);
200 /* Retrieves details of 'netdev->tc' configuration into 'details'.
202 * The implementation should not need to perform any Netlink calls, because
203 * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have
204 * cached the configuration.
206 * The contents of 'details' should be documented as valid for 'ovs_name'
207 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
208 * (which is built as ovs-vswitchd.conf.db(8)).
210 * This function may be null if 'tc' is not configurable.
212 int (*qdisc_get)(const struct netdev *netdev, struct shash *details);
214 /* Reconfigures 'netdev->tc' according to 'details', performing any
215 * required Netlink calls to complete the reconfiguration.
217 * The contents of 'details' should be documented as valid for 'ovs_name'
218 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
219 * (which is built as ovs-vswitchd.conf.db(8)).
221 * This function may be null if 'tc' is not configurable.
223 int (*qdisc_set)(struct netdev *, const struct shash *details);
225 /* Retrieves details of 'queue' on 'netdev->tc' into 'details'. 'queue' is
226 * one of the 'struct tc_queue's within 'netdev->tc->queues'.
228 * The contents of 'details' should be documented as valid for 'ovs_name'
229 * in the "other_config" column in the "Queue" table in
230 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
232 * The implementation should not need to perform any Netlink calls, because
233 * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have
234 * cached the queue configuration.
236 * This function may be null if 'tc' does not have queues ('n_queues' is
238 int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue,
239 struct shash *details);
241 /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to
242 * 'details', perfoming any required Netlink calls to complete the
243 * reconfiguration. The caller ensures that 'queue_id' is less than
246 * The contents of 'details' should be documented as valid for 'ovs_name'
247 * in the "other_config" column in the "Queue" table in
248 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
250 * This function may be null if 'tc' does not have queues or its queues are
251 * not configurable. */
252 int (*class_set)(struct netdev *, unsigned int queue_id,
253 const struct shash *details);
255 /* Deletes 'queue' from 'netdev->tc'. 'queue' is one of the 'struct
256 * tc_queue's within 'netdev->tc->queues'.
258 * This function may be null if 'tc' does not have queues or its queues
259 * cannot be deleted. */
260 int (*class_delete)(struct netdev *, struct tc_queue *queue);
262 /* Obtains stats for 'queue' from 'netdev->tc'. 'queue' is one of the
263 * 'struct tc_queue's within 'netdev->tc->queues'.
265 * On success, initializes '*stats'.
267 * This function may be null if 'tc' does not have queues or if it cannot
268 * report queue statistics. */
269 int (*class_get_stats)(const struct netdev *netdev,
270 const struct tc_queue *queue,
271 struct netdev_queue_stats *stats);
273 /* Extracts queue stats from 'nlmsg', which is a response to a
274 * RTM_GETTCLASS message, and passes them to 'cb' along with 'aux'.
276 * This function may be null if 'tc' does not have queues or if it cannot
277 * report queue statistics. */
278 int (*class_dump_stats)(const struct netdev *netdev,
279 const struct ofpbuf *nlmsg,
280 netdev_dump_queue_stats_cb *cb, void *aux);
284 tc_init(struct tc *tc, const struct tc_ops *ops)
287 hmap_init(&tc->queues);
291 tc_destroy(struct tc *tc)
293 hmap_destroy(&tc->queues);
296 static const struct tc_ops tc_ops_htb;
297 static const struct tc_ops tc_ops_hfsc;
298 static const struct tc_ops tc_ops_default;
299 static const struct tc_ops tc_ops_other;
301 static const struct tc_ops *tcs[] = {
302 &tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */
303 &tc_ops_hfsc, /* Hierarchical fair service curve. */
304 &tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */
305 &tc_ops_other, /* Some other qdisc. */
309 static unsigned int tc_make_handle(unsigned int major, unsigned int minor);
310 static unsigned int tc_get_major(unsigned int handle);
311 static unsigned int tc_get_minor(unsigned int handle);
313 static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks);
314 static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size);
315 static unsigned int tc_buffer_per_jiffy(unsigned int rate);
317 static struct tcmsg *tc_make_request(const struct netdev *, int type,
318 unsigned int flags, struct ofpbuf *);
319 static int tc_transact(struct ofpbuf *request, struct ofpbuf **replyp);
321 static int tc_parse_qdisc(const struct ofpbuf *, const char **kind,
322 struct nlattr **options);
323 static int tc_parse_class(const struct ofpbuf *, unsigned int *queue_id,
324 struct nlattr **options,
325 struct netdev_queue_stats *);
326 static int tc_query_class(const struct netdev *,
327 unsigned int handle, unsigned int parent,
328 struct ofpbuf **replyp);
329 static int tc_delete_class(const struct netdev *, unsigned int handle);
331 static int tc_del_qdisc(struct netdev *netdev);
332 static int tc_query_qdisc(const struct netdev *netdev);
334 static int tc_calc_cell_log(unsigned int mtu);
335 static void tc_fill_rate(struct tc_ratespec *rate, uint64_t bps, int mtu);
336 static void tc_put_rtab(struct ofpbuf *, uint16_t type,
337 const struct tc_ratespec *rate);
338 static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes);
340 struct netdev_dev_linux {
341 struct netdev_dev netdev_dev;
343 struct shash_node *shash_node;
344 unsigned int cache_valid;
346 /* The following are figured out "on demand" only. They are only valid
347 * when the corresponding VALID_* bit in 'cache_valid' is set. */
349 uint8_t etheraddr[ETH_ADDR_LEN];
350 struct in_addr address, netmask;
354 bool is_internal; /* Is this an openvswitch internal device? */
355 bool is_tap; /* Is this a tuntap device? */
356 uint32_t kbits_rate; /* Policing data. */
357 uint32_t kbits_burst;
358 bool have_vport_stats;
362 struct tap_state tap;
366 struct netdev_linux {
367 struct netdev netdev;
371 /* An AF_INET socket (used for ioctl operations). */
372 static int af_inet_sock = -1;
374 /* A Netlink routing socket that is not subscribed to any multicast groups. */
375 static struct nl_sock *rtnl_sock;
377 struct netdev_linux_notifier {
378 struct netdev_notifier notifier;
382 static struct shash netdev_linux_notifiers =
383 SHASH_INITIALIZER(&netdev_linux_notifiers);
384 static struct rtnetlink_notifier netdev_linux_poll_notifier;
386 /* This is set pretty low because we probably won't learn anything from the
387 * additional log messages. */
388 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
390 static int netdev_linux_init(void);
392 static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
393 int cmd, const char *cmd_name);
394 static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
395 const char *cmd_name);
396 static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
397 int cmd, const char *cmd_name);
398 static int get_flags(const struct netdev *, int *flagsp);
399 static int set_flags(struct netdev *, int flags);
400 static int do_get_ifindex(const char *netdev_name);
401 static int get_ifindex(const struct netdev *, int *ifindexp);
402 static int do_set_addr(struct netdev *netdev,
403 int ioctl_nr, const char *ioctl_name,
404 struct in_addr addr);
405 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
406 static int set_etheraddr(const char *netdev_name, int hwaddr_family,
407 const uint8_t[ETH_ADDR_LEN]);
408 static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
409 static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
412 is_netdev_linux_class(const struct netdev_class *netdev_class)
414 return netdev_class->init == netdev_linux_init;
417 static struct netdev_dev_linux *
418 netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
420 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
421 assert(is_netdev_linux_class(netdev_class));
423 return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
426 static struct netdev_linux *
427 netdev_linux_cast(const struct netdev *netdev)
429 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
430 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
431 assert(is_netdev_linux_class(netdev_class));
433 return CONTAINER_OF(netdev, struct netdev_linux, netdev);
437 netdev_linux_init(void)
439 static int status = -1;
441 /* Create AF_INET socket. */
442 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
443 status = af_inet_sock >= 0 ? 0 : errno;
445 VLOG_ERR("failed to create inet socket: %s", strerror(status));
448 /* Create rtnetlink socket. */
450 status = nl_sock_create(NETLINK_ROUTE, &rtnl_sock);
452 VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",
461 netdev_linux_run(void)
463 rtnetlink_link_notifier_run();
467 netdev_linux_wait(void)
469 rtnetlink_link_notifier_wait();
473 netdev_linux_cache_cb(const struct rtnetlink_link_change *change,
474 void *aux OVS_UNUSED)
476 struct netdev_dev_linux *dev;
478 struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
480 const struct netdev_class *netdev_class =
481 netdev_dev_get_class(base_dev);
483 if (is_netdev_linux_class(netdev_class)) {
484 dev = netdev_dev_linux_cast(base_dev);
485 dev->cache_valid = 0;
489 struct shash device_shash;
490 struct shash_node *node;
492 shash_init(&device_shash);
493 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
494 SHASH_FOR_EACH (node, &device_shash) {
496 dev->cache_valid = 0;
498 shash_destroy(&device_shash);
502 /* Creates system and internal devices. */
504 netdev_linux_create(const struct netdev_class *class,
505 const char *name, const struct shash *args,
506 struct netdev_dev **netdev_devp)
508 struct netdev_dev_linux *netdev_dev;
511 if (!shash_is_empty(args)) {
512 VLOG_WARN("%s: arguments for %s devices should be empty",
516 if (!cache_notifier_refcount) {
517 error = rtnetlink_link_notifier_register(&netdev_linux_cache_notifier,
518 netdev_linux_cache_cb, NULL);
523 cache_notifier_refcount++;
525 netdev_dev = xzalloc(sizeof *netdev_dev);
526 netdev_dev_init(&netdev_dev->netdev_dev, name, args, class);
528 *netdev_devp = &netdev_dev->netdev_dev;
532 /* For most types of netdevs we open the device for each call of
533 * netdev_open(). However, this is not the case with tap devices,
534 * since it is only possible to open the device once. In this
535 * situation we share a single file descriptor, and consequently
536 * buffers, across all readers. Therefore once data is read it will
537 * be unavailable to other reads for tap devices. */
539 netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED,
540 const char *name, const struct shash *args,
541 struct netdev_dev **netdev_devp)
543 struct netdev_dev_linux *netdev_dev;
544 struct tap_state *state;
545 static const char tap_dev[] = "/dev/net/tun";
549 if (!shash_is_empty(args)) {
550 VLOG_WARN("%s: arguments for TAP devices should be empty", name);
553 netdev_dev = xzalloc(sizeof *netdev_dev);
554 state = &netdev_dev->state.tap;
556 /* Open tap device. */
557 state->fd = open(tap_dev, O_RDWR);
560 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
564 /* Create tap device. */
565 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
566 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
567 if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
568 VLOG_WARN("%s: creating tap device failed: %s", name,
574 /* Make non-blocking. */
575 error = set_nonblocking(state->fd);
580 netdev_dev_init(&netdev_dev->netdev_dev, name, args, &netdev_tap_class);
581 *netdev_devp = &netdev_dev->netdev_dev;
590 destroy_tap(struct netdev_dev_linux *netdev_dev)
592 struct tap_state *state = &netdev_dev->state.tap;
594 if (state->fd >= 0) {
599 /* Destroys the netdev device 'netdev_dev_'. */
601 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
603 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
604 const struct netdev_class *class = netdev_dev_get_class(netdev_dev_);
606 if (netdev_dev->tc && netdev_dev->tc->ops->tc_destroy) {
607 netdev_dev->tc->ops->tc_destroy(netdev_dev->tc);
610 if (class == &netdev_linux_class || class == &netdev_internal_class) {
611 cache_notifier_refcount--;
613 if (!cache_notifier_refcount) {
614 rtnetlink_link_notifier_unregister(&netdev_linux_cache_notifier);
616 } else if (class == &netdev_tap_class) {
617 destroy_tap(netdev_dev);
626 netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype,
627 struct netdev **netdevp)
629 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
630 struct netdev_linux *netdev;
631 enum netdev_flags flags;
634 /* Allocate network device. */
635 netdev = xzalloc(sizeof *netdev);
637 netdev_init(&netdev->netdev, netdev_dev_);
639 /* Verify that the device really exists, by attempting to read its flags.
640 * (The flags might be cached, in which case this won't actually do an
643 * Don't do this for "internal" netdevs, though, because those have to be
644 * created as netdev objects before they exist in the kernel, because
645 * creating them in the kernel happens by passing a netdev object to
646 * dpif_port_add(). */
647 if (netdev_dev_get_class(netdev_dev_) != &netdev_internal_class) {
648 error = netdev_get_flags(&netdev->netdev, &flags);
649 if (error == ENODEV) {
654 if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap") &&
655 !netdev_dev->state.tap.opened) {
657 /* We assume that the first user of the tap device is the primary user
658 * and give them the tap FD. Subsequent users probably just expect
659 * this to be a system device so open it normally to avoid send/receive
660 * directions appearing to be reversed. */
661 netdev->fd = netdev_dev->state.tap.fd;
662 netdev_dev->state.tap.opened = true;
663 } else if (ethertype != NETDEV_ETH_TYPE_NONE) {
664 struct sockaddr_ll sll;
668 /* Create file descriptor. */
669 protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
670 : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
672 netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
673 if (netdev->fd < 0) {
678 /* Set non-blocking mode. */
679 error = set_nonblocking(netdev->fd);
684 /* Get ethernet device index. */
685 error = get_ifindex(&netdev->netdev, &ifindex);
690 /* Bind to specific ethernet device. */
691 memset(&sll, 0, sizeof sll);
692 sll.sll_family = AF_PACKET;
693 sll.sll_ifindex = ifindex;
695 (struct sockaddr *) &sll, sizeof sll) < 0) {
697 VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_),
702 /* Between the socket() and bind() calls above, the socket receives all
703 * packets of the requested type on all system interfaces. We do not
704 * want to receive that data, but there is no way to avoid it. So we
705 * must now drain out the receive queue. */
706 error = drain_rcvbuf(netdev->fd);
712 *netdevp = &netdev->netdev;
716 netdev_uninit(&netdev->netdev, true);
720 /* Closes and destroys 'netdev'. */
722 netdev_linux_close(struct netdev *netdev_)
724 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
726 if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
732 /* Initializes 'svec' with a list of the names of all known network devices. */
734 netdev_linux_enumerate(struct svec *svec)
736 struct if_nameindex *names;
738 names = if_nameindex();
742 for (i = 0; names[i].if_name != NULL; i++) {
743 svec_add(svec, names[i].if_name);
745 if_freenameindex(names);
748 VLOG_WARN("could not obtain list of network device names: %s",
755 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
757 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
759 if (netdev->fd < 0) {
760 /* Device was opened with NETDEV_ETH_TYPE_NONE. */
765 ssize_t retval = read(netdev->fd, data, size);
768 } else if (errno != EINTR) {
769 if (errno != EAGAIN) {
770 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
771 strerror(errno), netdev_get_name(netdev_));
778 /* Registers with the poll loop to wake up from the next call to poll_block()
779 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
781 netdev_linux_recv_wait(struct netdev *netdev_)
783 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
784 if (netdev->fd >= 0) {
785 poll_fd_wait(netdev->fd, POLLIN);
789 /* Discards all packets waiting to be received from 'netdev'. */
791 netdev_linux_drain(struct netdev *netdev_)
793 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
794 if (netdev->fd < 0) {
796 } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
798 int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
799 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
803 drain_fd(netdev->fd, ifr.ifr_qlen);
806 return drain_rcvbuf(netdev->fd);
810 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
811 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
812 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
813 * the packet is too big or too small to transmit on the device.
815 * The caller retains ownership of 'buffer' in all cases.
817 * The kernel maintains a packet transmission queue, so the caller is not
818 * expected to do additional queuing of packets. */
820 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
822 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
824 /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
826 if (netdev->fd < 0) {
831 ssize_t retval = write(netdev->fd, data, size);
833 /* The Linux AF_PACKET implementation never blocks waiting for room
834 * for packets, instead returning ENOBUFS. Translate this into
835 * EAGAIN for the caller. */
836 if (errno == ENOBUFS) {
838 } else if (errno == EINTR) {
840 } else if (errno != EAGAIN) {
841 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
842 netdev_get_name(netdev_), strerror(errno));
845 } else if (retval != size) {
846 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
847 "%zu) on %s", retval, size, netdev_get_name(netdev_));
855 /* Registers with the poll loop to wake up from the next call to poll_block()
856 * when the packet transmission queue has sufficient room to transmit a packet
857 * with netdev_send().
859 * The kernel maintains a packet transmission queue, so the client is not
860 * expected to do additional queuing of packets. Thus, this function is
861 * unlikely to ever be used. It is included for completeness. */
863 netdev_linux_send_wait(struct netdev *netdev_)
865 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
866 if (netdev->fd < 0) {
868 } else if (strcmp(netdev_get_type(netdev_), "tap")) {
869 poll_fd_wait(netdev->fd, POLLOUT);
871 /* TAP device always accepts packets.*/
872 poll_immediate_wake();
876 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
877 * otherwise a positive errno value. */
879 netdev_linux_set_etheraddr(struct netdev *netdev_,
880 const uint8_t mac[ETH_ADDR_LEN])
882 struct netdev_dev_linux *netdev_dev =
883 netdev_dev_linux_cast(netdev_get_dev(netdev_));
886 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)
887 || !eth_addr_equals(netdev_dev->etheraddr, mac)) {
888 error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
890 netdev_dev->cache_valid |= VALID_ETHERADDR;
891 memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
899 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
900 * free the returned buffer. */
902 netdev_linux_get_etheraddr(const struct netdev *netdev_,
903 uint8_t mac[ETH_ADDR_LEN])
905 struct netdev_dev_linux *netdev_dev =
906 netdev_dev_linux_cast(netdev_get_dev(netdev_));
907 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
908 int error = get_etheraddr(netdev_get_name(netdev_),
909 netdev_dev->etheraddr);
913 netdev_dev->cache_valid |= VALID_ETHERADDR;
915 memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
919 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
920 * in bytes, not including the hardware header; thus, this is typically 1500
921 * bytes for Ethernet devices. */
923 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
925 struct netdev_dev_linux *netdev_dev =
926 netdev_dev_linux_cast(netdev_get_dev(netdev_));
927 if (!(netdev_dev->cache_valid & VALID_MTU)) {
931 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
932 SIOCGIFMTU, "SIOCGIFMTU");
936 netdev_dev->mtu = ifr.ifr_mtu;
937 netdev_dev->cache_valid |= VALID_MTU;
939 *mtup = netdev_dev->mtu;
943 /* Returns the ifindex of 'netdev', if successful, as a positive number.
944 * On failure, returns a negative errno value. */
946 netdev_linux_get_ifindex(const struct netdev *netdev)
950 error = get_ifindex(netdev, &ifindex);
951 return error ? -error : ifindex;
955 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
957 struct netdev_dev_linux *netdev_dev =
958 netdev_dev_linux_cast(netdev_get_dev(netdev_));
963 if (!(netdev_dev->cache_valid & VALID_CARRIER)) {
967 fn = xasprintf("/sys/class/net/%s/carrier",
968 netdev_get_name(netdev_));
969 fd = open(fn, O_RDONLY);
972 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error));
976 retval = read(fd, line, sizeof line);
979 if (error == EINVAL) {
980 /* This is the normal return value when we try to check carrier
981 * if the network device is not up. */
983 VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error));
986 } else if (retval == 0) {
988 VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn);
992 if (line[0] != '0' && line[0] != '1') {
994 VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)",
998 netdev_dev->carrier = line[0] != '0';
999 netdev_dev->cache_valid |= VALID_CARRIER;
1001 *carrier = netdev_dev->carrier;
1013 netdev_linux_do_miimon(const struct netdev *netdev, int cmd,
1014 const char *cmd_name, struct mii_ioctl_data *data)
1019 memset(&ifr, 0, sizeof ifr);
1020 memcpy(&ifr.ifr_data, data, sizeof *data);
1021 error = netdev_linux_do_ioctl(netdev_get_name(netdev),
1022 &ifr, cmd, cmd_name);
1023 memcpy(data, &ifr.ifr_data, sizeof *data);
1029 netdev_linux_get_miimon(const struct netdev *netdev, bool *miimon)
1031 const char *name = netdev_get_name(netdev);
1032 struct mii_ioctl_data data;
1037 memset(&data, 0, sizeof data);
1038 error = netdev_linux_do_miimon(netdev, SIOCGMIIPHY, "SIOCGMIIPHY", &data);
1040 /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */
1041 data.reg_num = MII_BMSR;
1042 error = netdev_linux_do_miimon(netdev, SIOCGMIIREG, "SIOCGMIIREG",
1046 *miimon = !!(data.val_out & BMSR_LSTATUS);
1048 VLOG_WARN_RL(&rl, "%s: failed to query MII", name);
1051 struct ethtool_cmd ecmd;
1053 VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool",
1056 memset(&ecmd, 0, sizeof ecmd);
1057 error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK,
1060 struct ethtool_value eval;
1062 memcpy(&eval, &ecmd, sizeof eval);
1063 *miimon = !!eval.data;
1065 VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name);
1072 /* Check whether we can we use RTM_GETLINK to get network device statistics.
1073 * In pre-2.6.19 kernels, this was only available if wireless extensions were
1076 check_for_working_netlink_stats(void)
1078 /* Decide on the netdev_get_stats() implementation to use. Netlink is
1079 * preferable, so if that works, we'll use it. */
1080 int ifindex = do_get_ifindex("lo");
1082 VLOG_WARN("failed to get ifindex for lo, "
1083 "obtaining netdev stats from proc");
1086 struct netdev_stats stats;
1087 int error = get_stats_via_netlink(ifindex, &stats);
1089 VLOG_DBG("obtaining netdev stats via rtnetlink");
1092 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
1093 "via proc (you are probably running a pre-2.6.19 "
1094 "kernel)", strerror(error));
1100 /* Brings the 'is_internal' and 'is_tap' members of 'netdev_dev' up-to-date. */
1102 netdev_linux_update_is_pseudo(struct netdev_dev_linux *netdev_dev)
1104 if (!(netdev_dev->cache_valid & VALID_IS_PSEUDO)) {
1105 const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
1106 const char *type = netdev_dev_get_type(&netdev_dev->netdev_dev);
1108 netdev_dev->is_tap = !strcmp(type, "tap");
1109 netdev_dev->is_internal = (!netdev_dev->is_tap
1110 && dpif_linux_is_internal_device(name));
1111 netdev_dev->cache_valid |= VALID_IS_PSEUDO;
1116 swap_uint64(uint64_t *a, uint64_t *b)
1123 /* Retrieves current device stats for 'netdev'. */
1125 netdev_linux_get_stats(const struct netdev *netdev_,
1126 struct netdev_stats *stats)
1128 struct netdev_dev_linux *netdev_dev =
1129 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1130 static int use_netlink_stats = -1;
1133 if (netdev_dev->have_vport_stats ||
1134 !(netdev_dev->cache_valid & VALID_HAVE_VPORT_STATS)) {
1136 error = netdev_vport_get_stats(netdev_, stats);
1137 netdev_dev->have_vport_stats = !error;
1138 netdev_dev->cache_valid |= VALID_HAVE_VPORT_STATS;
1141 if (!netdev_dev->have_vport_stats) {
1142 if (use_netlink_stats < 0) {
1143 use_netlink_stats = check_for_working_netlink_stats();
1145 if (use_netlink_stats) {
1148 error = get_ifindex(netdev_, &ifindex);
1150 error = get_stats_via_netlink(ifindex, stats);
1153 error = get_stats_via_proc(netdev_get_name(netdev_), stats);
1157 /* If this port is an internal port then the transmit and receive stats
1158 * will appear to be swapped relative to the other ports since we are the
1159 * one sending the data, not a remote computer. For consistency, we swap
1160 * them back here. This does not apply if we are getting stats from the
1161 * vport layer because it always tracks stats from the perspective of the
1163 netdev_linux_update_is_pseudo(netdev_dev);
1164 if (!error && !netdev_dev->have_vport_stats &&
1165 (netdev_dev->is_internal || netdev_dev->is_tap)) {
1166 swap_uint64(&stats->rx_packets, &stats->tx_packets);
1167 swap_uint64(&stats->rx_bytes, &stats->tx_bytes);
1168 swap_uint64(&stats->rx_errors, &stats->tx_errors);
1169 swap_uint64(&stats->rx_dropped, &stats->tx_dropped);
1170 stats->rx_length_errors = 0;
1171 stats->rx_over_errors = 0;
1172 stats->rx_crc_errors = 0;
1173 stats->rx_frame_errors = 0;
1174 stats->rx_fifo_errors = 0;
1175 stats->rx_missed_errors = 0;
1176 stats->tx_aborted_errors = 0;
1177 stats->tx_carrier_errors = 0;
1178 stats->tx_fifo_errors = 0;
1179 stats->tx_heartbeat_errors = 0;
1180 stats->tx_window_errors = 0;
1186 /* Stores the features supported by 'netdev' into each of '*current',
1187 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
1188 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
1189 * successful, otherwise a positive errno value. */
1191 netdev_linux_get_features(const struct netdev *netdev,
1192 uint32_t *current, uint32_t *advertised,
1193 uint32_t *supported, uint32_t *peer)
1195 struct ethtool_cmd ecmd;
1198 memset(&ecmd, 0, sizeof ecmd);
1199 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1200 ETHTOOL_GSET, "ETHTOOL_GSET");
1205 /* Supported features. */
1207 if (ecmd.supported & SUPPORTED_10baseT_Half) {
1208 *supported |= OFPPF_10MB_HD;
1210 if (ecmd.supported & SUPPORTED_10baseT_Full) {
1211 *supported |= OFPPF_10MB_FD;
1213 if (ecmd.supported & SUPPORTED_100baseT_Half) {
1214 *supported |= OFPPF_100MB_HD;
1216 if (ecmd.supported & SUPPORTED_100baseT_Full) {
1217 *supported |= OFPPF_100MB_FD;
1219 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
1220 *supported |= OFPPF_1GB_HD;
1222 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
1223 *supported |= OFPPF_1GB_FD;
1225 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
1226 *supported |= OFPPF_10GB_FD;
1228 if (ecmd.supported & SUPPORTED_TP) {
1229 *supported |= OFPPF_COPPER;
1231 if (ecmd.supported & SUPPORTED_FIBRE) {
1232 *supported |= OFPPF_FIBER;
1234 if (ecmd.supported & SUPPORTED_Autoneg) {
1235 *supported |= OFPPF_AUTONEG;
1237 if (ecmd.supported & SUPPORTED_Pause) {
1238 *supported |= OFPPF_PAUSE;
1240 if (ecmd.supported & SUPPORTED_Asym_Pause) {
1241 *supported |= OFPPF_PAUSE_ASYM;
1244 /* Advertised features. */
1246 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
1247 *advertised |= OFPPF_10MB_HD;
1249 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
1250 *advertised |= OFPPF_10MB_FD;
1252 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
1253 *advertised |= OFPPF_100MB_HD;
1255 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
1256 *advertised |= OFPPF_100MB_FD;
1258 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
1259 *advertised |= OFPPF_1GB_HD;
1261 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
1262 *advertised |= OFPPF_1GB_FD;
1264 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
1265 *advertised |= OFPPF_10GB_FD;
1267 if (ecmd.advertising & ADVERTISED_TP) {
1268 *advertised |= OFPPF_COPPER;
1270 if (ecmd.advertising & ADVERTISED_FIBRE) {
1271 *advertised |= OFPPF_FIBER;
1273 if (ecmd.advertising & ADVERTISED_Autoneg) {
1274 *advertised |= OFPPF_AUTONEG;
1276 if (ecmd.advertising & ADVERTISED_Pause) {
1277 *advertised |= OFPPF_PAUSE;
1279 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
1280 *advertised |= OFPPF_PAUSE_ASYM;
1283 /* Current settings. */
1284 if (ecmd.speed == SPEED_10) {
1285 *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD;
1286 } else if (ecmd.speed == SPEED_100) {
1287 *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD;
1288 } else if (ecmd.speed == SPEED_1000) {
1289 *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD;
1290 } else if (ecmd.speed == SPEED_10000) {
1291 *current = OFPPF_10GB_FD;
1296 if (ecmd.port == PORT_TP) {
1297 *current |= OFPPF_COPPER;
1298 } else if (ecmd.port == PORT_FIBRE) {
1299 *current |= OFPPF_FIBER;
1303 *current |= OFPPF_AUTONEG;
1306 /* Peer advertisements. */
1307 *peer = 0; /* XXX */
1312 /* Set the features advertised by 'netdev' to 'advertise'. */
1314 netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
1316 struct ethtool_cmd ecmd;
1319 memset(&ecmd, 0, sizeof ecmd);
1320 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1321 ETHTOOL_GSET, "ETHTOOL_GSET");
1326 ecmd.advertising = 0;
1327 if (advertise & OFPPF_10MB_HD) {
1328 ecmd.advertising |= ADVERTISED_10baseT_Half;
1330 if (advertise & OFPPF_10MB_FD) {
1331 ecmd.advertising |= ADVERTISED_10baseT_Full;
1333 if (advertise & OFPPF_100MB_HD) {
1334 ecmd.advertising |= ADVERTISED_100baseT_Half;
1336 if (advertise & OFPPF_100MB_FD) {
1337 ecmd.advertising |= ADVERTISED_100baseT_Full;
1339 if (advertise & OFPPF_1GB_HD) {
1340 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1342 if (advertise & OFPPF_1GB_FD) {
1343 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1345 if (advertise & OFPPF_10GB_FD) {
1346 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1348 if (advertise & OFPPF_COPPER) {
1349 ecmd.advertising |= ADVERTISED_TP;
1351 if (advertise & OFPPF_FIBER) {
1352 ecmd.advertising |= ADVERTISED_FIBRE;
1354 if (advertise & OFPPF_AUTONEG) {
1355 ecmd.advertising |= ADVERTISED_Autoneg;
1357 if (advertise & OFPPF_PAUSE) {
1358 ecmd.advertising |= ADVERTISED_Pause;
1360 if (advertise & OFPPF_PAUSE_ASYM) {
1361 ecmd.advertising |= ADVERTISED_Asym_Pause;
1363 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1364 ETHTOOL_SSET, "ETHTOOL_SSET");
1367 /* If 'netdev_name' is the name of a VLAN network device (e.g. one created with
1368 * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device
1369 * and returns 0. Otherwise returns a errno value (specifically ENOENT if
1370 * 'netdev_name' is the name of a network device that is not a VLAN device) and
1371 * sets '*vlan_vid' to -1. */
1373 netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1375 const char *netdev_name = netdev_get_name(netdev);
1376 struct ds line = DS_EMPTY_INITIALIZER;
1377 FILE *stream = NULL;
1381 COVERAGE_INC(netdev_get_vlan_vid);
1382 fn = xasprintf("/proc/net/vlan/%s", netdev_name);
1383 stream = fopen(fn, "r");
1389 if (ds_get_line(&line, stream)) {
1390 if (ferror(stream)) {
1392 VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno));
1395 VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn);
1400 if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) {
1402 VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"",
1403 fn, ds_cstr(&line));
1421 #define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress"
1422 #define POLICE_CONFIG_CMD "/sbin/tc filter add dev %s parent ffff: protocol ip prio 50 u32 match ip src 0.0.0.0/0 police rate %dkbit burst %dk mtu 65535 drop flowid :1"
1424 /* Remove ingress policing from 'netdev'. Returns 0 if successful, otherwise a
1425 * positive errno value.
1427 * This function is equivalent to running
1428 * /sbin/tc qdisc del dev %s handle ffff: ingress
1429 * but it is much, much faster.
1432 netdev_linux_remove_policing(struct netdev *netdev)
1434 struct netdev_dev_linux *netdev_dev =
1435 netdev_dev_linux_cast(netdev_get_dev(netdev));
1436 const char *netdev_name = netdev_get_name(netdev);
1438 struct ofpbuf request;
1439 struct tcmsg *tcmsg;
1442 tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request);
1446 tcmsg->tcm_handle = tc_make_handle(0xffff, 0);
1447 tcmsg->tcm_parent = TC_H_INGRESS;
1448 nl_msg_put_string(&request, TCA_KIND, "ingress");
1449 nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0);
1451 error = tc_transact(&request, NULL);
1452 if (error && error != ENOENT && error != EINVAL) {
1453 VLOG_WARN_RL(&rl, "%s: removing policing failed: %s",
1454 netdev_name, strerror(error));
1458 netdev_dev->kbits_rate = 0;
1459 netdev_dev->kbits_burst = 0;
1460 netdev_dev->cache_valid |= VALID_POLICING;
1464 /* Attempts to set input rate limiting (policing) policy. */
1466 netdev_linux_set_policing(struct netdev *netdev,
1467 uint32_t kbits_rate, uint32_t kbits_burst)
1469 struct netdev_dev_linux *netdev_dev =
1470 netdev_dev_linux_cast(netdev_get_dev(netdev));
1471 const char *netdev_name = netdev_get_name(netdev);
1474 COVERAGE_INC(netdev_set_policing);
1476 kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */
1477 : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */
1478 : kbits_burst); /* Stick with user-specified value. */
1480 if (netdev_dev->cache_valid & VALID_POLICING
1481 && netdev_dev->kbits_rate == kbits_rate
1482 && netdev_dev->kbits_burst == kbits_burst) {
1483 /* Assume that settings haven't changed since we last set them. */
1487 netdev_linux_remove_policing(netdev);
1489 snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name);
1490 if (system(command) != 0) {
1491 VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name);
1495 snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name,
1496 kbits_rate, kbits_burst);
1497 if (system(command) != 0) {
1498 VLOG_WARN_RL(&rl, "%s: problem configuring policing",
1503 netdev_dev->kbits_rate = kbits_rate;
1504 netdev_dev->kbits_burst = kbits_burst;
1505 netdev_dev->cache_valid |= VALID_POLICING;
1512 netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED,
1515 const struct tc_ops **opsp;
1517 for (opsp = tcs; *opsp != NULL; opsp++) {
1518 const struct tc_ops *ops = *opsp;
1519 if (ops->tc_install && ops->ovs_name[0] != '\0') {
1520 svec_add(types, ops->ovs_name);
1526 static const struct tc_ops *
1527 tc_lookup_ovs_name(const char *name)
1529 const struct tc_ops **opsp;
1531 for (opsp = tcs; *opsp != NULL; opsp++) {
1532 const struct tc_ops *ops = *opsp;
1533 if (!strcmp(name, ops->ovs_name)) {
1540 static const struct tc_ops *
1541 tc_lookup_linux_name(const char *name)
1543 const struct tc_ops **opsp;
1545 for (opsp = tcs; *opsp != NULL; opsp++) {
1546 const struct tc_ops *ops = *opsp;
1547 if (ops->linux_name && !strcmp(name, ops->linux_name)) {
1554 static struct tc_queue *
1555 tc_find_queue__(const struct netdev *netdev, unsigned int queue_id,
1558 struct netdev_dev_linux *netdev_dev =
1559 netdev_dev_linux_cast(netdev_get_dev(netdev));
1560 struct tc_queue *queue;
1562 HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev_dev->tc->queues) {
1563 if (queue->queue_id == queue_id) {
1570 static struct tc_queue *
1571 tc_find_queue(const struct netdev *netdev, unsigned int queue_id)
1573 return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0));
1577 netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED,
1579 struct netdev_qos_capabilities *caps)
1581 const struct tc_ops *ops = tc_lookup_ovs_name(type);
1585 caps->n_queues = ops->n_queues;
1590 netdev_linux_get_qos(const struct netdev *netdev,
1591 const char **typep, struct shash *details)
1593 struct netdev_dev_linux *netdev_dev =
1594 netdev_dev_linux_cast(netdev_get_dev(netdev));
1597 error = tc_query_qdisc(netdev);
1602 *typep = netdev_dev->tc->ops->ovs_name;
1603 return (netdev_dev->tc->ops->qdisc_get
1604 ? netdev_dev->tc->ops->qdisc_get(netdev, details)
1609 netdev_linux_set_qos(struct netdev *netdev,
1610 const char *type, const struct shash *details)
1612 struct netdev_dev_linux *netdev_dev =
1613 netdev_dev_linux_cast(netdev_get_dev(netdev));
1614 const struct tc_ops *new_ops;
1617 new_ops = tc_lookup_ovs_name(type);
1618 if (!new_ops || !new_ops->tc_install) {
1622 error = tc_query_qdisc(netdev);
1627 if (new_ops == netdev_dev->tc->ops) {
1628 return new_ops->qdisc_set ? new_ops->qdisc_set(netdev, details) : 0;
1630 /* Delete existing qdisc. */
1631 error = tc_del_qdisc(netdev);
1635 assert(netdev_dev->tc == NULL);
1637 /* Install new qdisc. */
1638 error = new_ops->tc_install(netdev, details);
1639 assert((error == 0) == (netdev_dev->tc != NULL));
1646 netdev_linux_get_queue(const struct netdev *netdev,
1647 unsigned int queue_id, struct shash *details)
1649 struct netdev_dev_linux *netdev_dev =
1650 netdev_dev_linux_cast(netdev_get_dev(netdev));
1653 error = tc_query_qdisc(netdev);
1657 struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1659 ? netdev_dev->tc->ops->class_get(netdev, queue, details)
1665 netdev_linux_set_queue(struct netdev *netdev,
1666 unsigned int queue_id, const struct shash *details)
1668 struct netdev_dev_linux *netdev_dev =
1669 netdev_dev_linux_cast(netdev_get_dev(netdev));
1672 error = tc_query_qdisc(netdev);
1675 } else if (queue_id >= netdev_dev->tc->ops->n_queues
1676 || !netdev_dev->tc->ops->class_set) {
1680 return netdev_dev->tc->ops->class_set(netdev, queue_id, details);
1684 netdev_linux_delete_queue(struct netdev *netdev, unsigned int queue_id)
1686 struct netdev_dev_linux *netdev_dev =
1687 netdev_dev_linux_cast(netdev_get_dev(netdev));
1690 error = tc_query_qdisc(netdev);
1693 } else if (!netdev_dev->tc->ops->class_delete) {
1696 struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1698 ? netdev_dev->tc->ops->class_delete(netdev, queue)
1704 netdev_linux_get_queue_stats(const struct netdev *netdev,
1705 unsigned int queue_id,
1706 struct netdev_queue_stats *stats)
1708 struct netdev_dev_linux *netdev_dev =
1709 netdev_dev_linux_cast(netdev_get_dev(netdev));
1712 error = tc_query_qdisc(netdev);
1715 } else if (!netdev_dev->tc->ops->class_get_stats) {
1718 const struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1720 ? netdev_dev->tc->ops->class_get_stats(netdev, queue, stats)
1726 start_queue_dump(const struct netdev *netdev, struct nl_dump *dump)
1728 struct ofpbuf request;
1729 struct tcmsg *tcmsg;
1731 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request);
1735 tcmsg->tcm_parent = 0;
1736 nl_dump_start(dump, rtnl_sock, &request);
1737 ofpbuf_uninit(&request);
1742 netdev_linux_dump_queues(const struct netdev *netdev,
1743 netdev_dump_queues_cb *cb, void *aux)
1745 struct netdev_dev_linux *netdev_dev =
1746 netdev_dev_linux_cast(netdev_get_dev(netdev));
1747 struct tc_queue *queue;
1748 struct shash details;
1752 error = tc_query_qdisc(netdev);
1755 } else if (!netdev_dev->tc->ops->class_get) {
1760 shash_init(&details);
1761 HMAP_FOR_EACH (queue, hmap_node, &netdev_dev->tc->queues) {
1762 shash_clear(&details);
1764 error = netdev_dev->tc->ops->class_get(netdev, queue, &details);
1766 (*cb)(queue->queue_id, &details, aux);
1771 shash_destroy(&details);
1777 netdev_linux_dump_queue_stats(const struct netdev *netdev,
1778 netdev_dump_queue_stats_cb *cb, void *aux)
1780 struct netdev_dev_linux *netdev_dev =
1781 netdev_dev_linux_cast(netdev_get_dev(netdev));
1782 struct nl_dump dump;
1787 error = tc_query_qdisc(netdev);
1790 } else if (!netdev_dev->tc->ops->class_dump_stats) {
1795 if (!start_queue_dump(netdev, &dump)) {
1798 while (nl_dump_next(&dump, &msg)) {
1799 error = netdev_dev->tc->ops->class_dump_stats(netdev, &msg, cb, aux);
1805 error = nl_dump_done(&dump);
1806 return error ? error : last_error;
1810 netdev_linux_get_in4(const struct netdev *netdev_,
1811 struct in_addr *address, struct in_addr *netmask)
1813 struct netdev_dev_linux *netdev_dev =
1814 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1816 if (!(netdev_dev->cache_valid & VALID_IN4)) {
1819 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
1820 SIOCGIFADDR, "SIOCGIFADDR");
1825 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
1826 SIOCGIFNETMASK, "SIOCGIFNETMASK");
1831 netdev_dev->cache_valid |= VALID_IN4;
1833 *address = netdev_dev->address;
1834 *netmask = netdev_dev->netmask;
1835 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
1839 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
1840 struct in_addr netmask)
1842 struct netdev_dev_linux *netdev_dev =
1843 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1846 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
1848 netdev_dev->cache_valid |= VALID_IN4;
1849 netdev_dev->address = address;
1850 netdev_dev->netmask = netmask;
1851 if (address.s_addr != INADDR_ANY) {
1852 error = do_set_addr(netdev_, SIOCSIFNETMASK,
1853 "SIOCSIFNETMASK", netmask);
1860 parse_if_inet6_line(const char *line,
1861 struct in6_addr *in6, char ifname[16 + 1])
1863 uint8_t *s6 = in6->s6_addr;
1864 #define X8 "%2"SCNx8
1866 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
1867 "%*x %*x %*x %*x %16s\n",
1868 &s6[0], &s6[1], &s6[2], &s6[3],
1869 &s6[4], &s6[5], &s6[6], &s6[7],
1870 &s6[8], &s6[9], &s6[10], &s6[11],
1871 &s6[12], &s6[13], &s6[14], &s6[15],
1875 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
1876 * 'in6' is non-null) and returns true. Otherwise, returns false. */
1878 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
1880 struct netdev_dev_linux *netdev_dev =
1881 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1882 if (!(netdev_dev->cache_valid & VALID_IN6)) {
1886 netdev_dev->in6 = in6addr_any;
1888 file = fopen("/proc/net/if_inet6", "r");
1890 const char *name = netdev_get_name(netdev_);
1891 while (fgets(line, sizeof line, file)) {
1892 struct in6_addr in6_tmp;
1893 char ifname[16 + 1];
1894 if (parse_if_inet6_line(line, &in6_tmp, ifname)
1895 && !strcmp(name, ifname))
1897 netdev_dev->in6 = in6_tmp;
1903 netdev_dev->cache_valid |= VALID_IN6;
1905 *in6 = netdev_dev->in6;
1910 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
1912 struct sockaddr_in sin;
1913 memset(&sin, 0, sizeof sin);
1914 sin.sin_family = AF_INET;
1915 sin.sin_addr = addr;
1918 memset(sa, 0, sizeof *sa);
1919 memcpy(sa, &sin, sizeof sin);
1923 do_set_addr(struct netdev *netdev,
1924 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
1927 strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
1928 make_in4_sockaddr(&ifr.ifr_addr, addr);
1930 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
1934 /* Adds 'router' as a default IP gateway. */
1936 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
1938 struct in_addr any = { INADDR_ANY };
1942 memset(&rt, 0, sizeof rt);
1943 make_in4_sockaddr(&rt.rt_dst, any);
1944 make_in4_sockaddr(&rt.rt_gateway, router);
1945 make_in4_sockaddr(&rt.rt_genmask, any);
1946 rt.rt_flags = RTF_UP | RTF_GATEWAY;
1947 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
1949 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
1955 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
1958 static const char fn[] = "/proc/net/route";
1963 *netdev_name = NULL;
1964 stream = fopen(fn, "r");
1965 if (stream == NULL) {
1966 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1971 while (fgets(line, sizeof line, stream)) {
1974 uint32_t dest, gateway, mask;
1975 int refcnt, metric, mtu;
1976 unsigned int flags, use, window, irtt;
1979 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
1981 iface, &dest, &gateway, &flags, &refcnt,
1982 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
1984 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
1988 if (!(flags & RTF_UP)) {
1989 /* Skip routes that aren't up. */
1993 /* The output of 'dest', 'mask', and 'gateway' were given in
1994 * network byte order, so we don't need need any endian
1995 * conversions here. */
1996 if ((dest & mask) == (host->s_addr & mask)) {
1998 /* The host is directly reachable. */
1999 next_hop->s_addr = 0;
2001 /* To reach the host, we must go through a gateway. */
2002 next_hop->s_addr = gateway;
2004 *netdev_name = xstrdup(iface);
2016 netdev_linux_get_status(const struct netdev *netdev, struct shash *sh)
2018 struct ethtool_drvinfo drvinfo;
2021 memset(&drvinfo, 0, sizeof drvinfo);
2022 error = netdev_linux_do_ethtool(netdev_get_name(netdev),
2023 (struct ethtool_cmd *)&drvinfo,
2025 "ETHTOOL_GDRVINFO");
2027 shash_add(sh, "driver_name", xstrdup(drvinfo.driver));
2028 shash_add(sh, "driver_version", xstrdup(drvinfo.version));
2029 shash_add(sh, "firmware_version", xstrdup(drvinfo.fw_version));
2035 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
2036 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
2037 * returns 0. Otherwise, it returns a positive errno value; in particular,
2038 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
2040 netdev_linux_arp_lookup(const struct netdev *netdev,
2041 uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
2044 struct sockaddr_in sin;
2047 memset(&r, 0, sizeof r);
2048 sin.sin_family = AF_INET;
2049 sin.sin_addr.s_addr = ip;
2051 memcpy(&r.arp_pa, &sin, sizeof sin);
2052 r.arp_ha.sa_family = ARPHRD_ETHER;
2054 strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
2055 COVERAGE_INC(netdev_arp_lookup);
2056 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
2058 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
2059 } else if (retval != ENXIO) {
2060 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
2061 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
2067 nd_to_iff_flags(enum netdev_flags nd)
2070 if (nd & NETDEV_UP) {
2073 if (nd & NETDEV_PROMISC) {
2080 iff_to_nd_flags(int iff)
2082 enum netdev_flags nd = 0;
2086 if (iff & IFF_PROMISC) {
2087 nd |= NETDEV_PROMISC;
2093 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
2094 enum netdev_flags on, enum netdev_flags *old_flagsp)
2096 int old_flags, new_flags;
2099 error = get_flags(netdev, &old_flags);
2101 *old_flagsp = iff_to_nd_flags(old_flags);
2102 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
2103 if (new_flags != old_flags) {
2104 error = set_flags(netdev, new_flags);
2111 poll_notify(struct list *list)
2113 struct netdev_linux_notifier *notifier;
2114 LIST_FOR_EACH (notifier, node, list) {
2115 struct netdev_notifier *n = ¬ifier->notifier;
2121 netdev_linux_poll_cb(const struct rtnetlink_link_change *change,
2122 void *aux OVS_UNUSED)
2125 struct list *list = shash_find_data(&netdev_linux_notifiers,
2131 struct shash_node *node;
2132 SHASH_FOR_EACH (node, &netdev_linux_notifiers) {
2133 poll_notify(node->data);
2139 netdev_linux_poll_add(struct netdev *netdev,
2140 void (*cb)(struct netdev_notifier *), void *aux,
2141 struct netdev_notifier **notifierp)
2143 const char *netdev_name = netdev_get_name(netdev);
2144 struct netdev_linux_notifier *notifier;
2147 if (shash_is_empty(&netdev_linux_notifiers)) {
2149 error = rtnetlink_link_notifier_register(&netdev_linux_poll_notifier,
2150 netdev_linux_poll_cb, NULL);
2156 list = shash_find_data(&netdev_linux_notifiers, netdev_name);
2158 list = xmalloc(sizeof *list);
2160 shash_add(&netdev_linux_notifiers, netdev_name, list);
2163 notifier = xmalloc(sizeof *notifier);
2164 netdev_notifier_init(¬ifier->notifier, netdev, cb, aux);
2165 list_push_back(list, ¬ifier->node);
2166 *notifierp = ¬ifier->notifier;
2171 netdev_linux_poll_remove(struct netdev_notifier *notifier_)
2173 struct netdev_linux_notifier *notifier =
2174 CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier);
2177 /* Remove 'notifier' from its list. */
2178 list = list_remove(¬ifier->node);
2179 if (list_is_empty(list)) {
2180 /* The list is now empty. Remove it from the hash and free it. */
2181 const char *netdev_name = netdev_get_name(notifier->notifier.netdev);
2182 shash_delete(&netdev_linux_notifiers,
2183 shash_find(&netdev_linux_notifiers, netdev_name));
2188 /* If that was the last notifier, unregister. */
2189 if (shash_is_empty(&netdev_linux_notifiers)) {
2190 rtnetlink_link_notifier_unregister(&netdev_linux_poll_notifier);
2194 #define NETDEV_LINUX_CLASS(NAME, CREATE, ENUMERATE, SET_STATS) \
2198 netdev_linux_init, \
2200 netdev_linux_wait, \
2203 netdev_linux_destroy, \
2204 NULL, /* set_config */ \
2206 netdev_linux_open, \
2207 netdev_linux_close, \
2211 netdev_linux_recv, \
2212 netdev_linux_recv_wait, \
2213 netdev_linux_drain, \
2215 netdev_linux_send, \
2216 netdev_linux_send_wait, \
2218 netdev_linux_set_etheraddr, \
2219 netdev_linux_get_etheraddr, \
2220 netdev_linux_get_mtu, \
2221 netdev_linux_get_ifindex, \
2222 netdev_linux_get_carrier, \
2223 netdev_linux_get_miimon, \
2224 netdev_linux_get_stats, \
2227 netdev_linux_get_features, \
2228 netdev_linux_set_advertisements, \
2229 netdev_linux_get_vlan_vid, \
2231 netdev_linux_set_policing, \
2232 netdev_linux_get_qos_types, \
2233 netdev_linux_get_qos_capabilities, \
2234 netdev_linux_get_qos, \
2235 netdev_linux_set_qos, \
2236 netdev_linux_get_queue, \
2237 netdev_linux_set_queue, \
2238 netdev_linux_delete_queue, \
2239 netdev_linux_get_queue_stats, \
2240 netdev_linux_dump_queues, \
2241 netdev_linux_dump_queue_stats, \
2243 netdev_linux_get_in4, \
2244 netdev_linux_set_in4, \
2245 netdev_linux_get_in6, \
2246 netdev_linux_add_router, \
2247 netdev_linux_get_next_hop, \
2248 netdev_linux_get_status, \
2249 netdev_linux_arp_lookup, \
2251 netdev_linux_update_flags, \
2253 netdev_linux_poll_add, \
2254 netdev_linux_poll_remove \
2257 const struct netdev_class netdev_linux_class =
2260 netdev_linux_create,
2261 netdev_linux_enumerate,
2262 NULL); /* set_stats */
2264 const struct netdev_class netdev_tap_class =
2267 netdev_linux_create_tap,
2268 NULL, /* enumerate */
2269 NULL); /* set_stats */
2271 const struct netdev_class netdev_internal_class =
2274 netdev_linux_create,
2275 NULL, /* enumerate */
2276 netdev_vport_set_stats);
2278 /* HTB traffic control class. */
2280 #define HTB_N_QUEUES 0xf000
2284 unsigned int max_rate; /* In bytes/s. */
2288 struct tc_queue tc_queue;
2289 unsigned int min_rate; /* In bytes/s. */
2290 unsigned int max_rate; /* In bytes/s. */
2291 unsigned int burst; /* In bytes. */
2292 unsigned int priority; /* Lower values are higher priorities. */
2296 htb_get__(const struct netdev *netdev)
2298 struct netdev_dev_linux *netdev_dev =
2299 netdev_dev_linux_cast(netdev_get_dev(netdev));
2300 return CONTAINER_OF(netdev_dev->tc, struct htb, tc);
2304 htb_install__(struct netdev *netdev, uint64_t max_rate)
2306 struct netdev_dev_linux *netdev_dev =
2307 netdev_dev_linux_cast(netdev_get_dev(netdev));
2310 htb = xmalloc(sizeof *htb);
2311 tc_init(&htb->tc, &tc_ops_htb);
2312 htb->max_rate = max_rate;
2314 netdev_dev->tc = &htb->tc;
2319 /* Create an HTB qdisc.
2321 * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */
2323 htb_setup_qdisc__(struct netdev *netdev)
2326 struct tc_htb_glob opt;
2327 struct ofpbuf request;
2328 struct tcmsg *tcmsg;
2330 tc_del_qdisc(netdev);
2332 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
2333 NLM_F_EXCL | NLM_F_CREATE, &request);
2337 tcmsg->tcm_handle = tc_make_handle(1, 0);
2338 tcmsg->tcm_parent = TC_H_ROOT;
2340 nl_msg_put_string(&request, TCA_KIND, "htb");
2342 memset(&opt, 0, sizeof opt);
2343 opt.rate2quantum = 10;
2347 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2348 nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt);
2349 nl_msg_end_nested(&request, opt_offset);
2351 return tc_transact(&request, NULL);
2354 /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb
2355 * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */
2357 htb_setup_class__(struct netdev *netdev, unsigned int handle,
2358 unsigned int parent, struct htb_class *class)
2361 struct tc_htb_opt opt;
2362 struct ofpbuf request;
2363 struct tcmsg *tcmsg;
2367 netdev_get_mtu(netdev, &mtu);
2369 memset(&opt, 0, sizeof opt);
2370 tc_fill_rate(&opt.rate, class->min_rate, mtu);
2371 tc_fill_rate(&opt.ceil, class->max_rate, mtu);
2372 opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst);
2373 opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst);
2374 opt.prio = class->priority;
2376 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
2380 tcmsg->tcm_handle = handle;
2381 tcmsg->tcm_parent = parent;
2383 nl_msg_put_string(&request, TCA_KIND, "htb");
2384 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2385 nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt);
2386 tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate);
2387 tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil);
2388 nl_msg_end_nested(&request, opt_offset);
2390 error = tc_transact(&request, NULL);
2392 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
2393 "min_rate=%u max_rate=%u burst=%u prio=%u (%s)",
2394 netdev_get_name(netdev),
2395 tc_get_major(handle), tc_get_minor(handle),
2396 tc_get_major(parent), tc_get_minor(parent),
2397 class->min_rate, class->max_rate,
2398 class->burst, class->priority, strerror(error));
2403 /* Parses Netlink attributes in 'options' for HTB parameters and stores a
2404 * description of them into 'details'. The description complies with the
2405 * specification given in the vswitch database documentation for linux-htb
2408 htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class)
2410 static const struct nl_policy tca_htb_policy[] = {
2411 [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false,
2412 .min_len = sizeof(struct tc_htb_opt) },
2415 struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)];
2416 const struct tc_htb_opt *htb;
2418 if (!nl_parse_nested(nl_options, tca_htb_policy,
2419 attrs, ARRAY_SIZE(tca_htb_policy))) {
2420 VLOG_WARN_RL(&rl, "failed to parse HTB class options");
2424 htb = nl_attr_get(attrs[TCA_HTB_PARMS]);
2425 class->min_rate = htb->rate.rate;
2426 class->max_rate = htb->ceil.rate;
2427 class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer);
2428 class->priority = htb->prio;
2433 htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
2434 struct htb_class *options,
2435 struct netdev_queue_stats *stats)
2437 struct nlattr *nl_options;
2438 unsigned int handle;
2441 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
2442 if (!error && queue_id) {
2443 unsigned int major = tc_get_major(handle);
2444 unsigned int minor = tc_get_minor(handle);
2445 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2446 *queue_id = minor - 1;
2451 if (!error && options) {
2452 error = htb_parse_tca_options__(nl_options, options);
2458 htb_parse_qdisc_details__(struct netdev *netdev,
2459 const struct shash *details, struct htb_class *hc)
2461 const char *max_rate_s;
2463 max_rate_s = shash_find_data(details, "max-rate");
2464 hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
2465 if (!hc->max_rate) {
2468 netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
2469 hc->max_rate = netdev_features_to_bps(current) / 8;
2471 hc->min_rate = hc->max_rate;
2477 htb_parse_class_details__(struct netdev *netdev,
2478 const struct shash *details, struct htb_class *hc)
2480 const struct htb *htb = htb_get__(netdev);
2481 const char *min_rate_s = shash_find_data(details, "min-rate");
2482 const char *max_rate_s = shash_find_data(details, "max-rate");
2483 const char *burst_s = shash_find_data(details, "burst");
2484 const char *priority_s = shash_find_data(details, "priority");
2487 /* min-rate. Don't allow a min-rate below 1500 bytes/s. */
2489 /* min-rate is required. */
2492 hc->min_rate = strtoull(min_rate_s, NULL, 10) / 8;
2493 hc->min_rate = MAX(hc->min_rate, 1500);
2494 hc->min_rate = MIN(hc->min_rate, htb->max_rate);
2497 hc->max_rate = (max_rate_s
2498 ? strtoull(max_rate_s, NULL, 10) / 8
2500 hc->max_rate = MAX(hc->max_rate, hc->min_rate);
2501 hc->max_rate = MIN(hc->max_rate, htb->max_rate);
2505 * According to hints in the documentation that I've read, it is important
2506 * that 'burst' be at least as big as the largest frame that might be
2507 * transmitted. Also, making 'burst' a bit bigger than necessary is OK,
2508 * but having it a bit too small is a problem. Since netdev_get_mtu()
2509 * doesn't include the Ethernet header, we need to add at least 14 (18?) to
2510 * the MTU. We actually add 64, instead of 14, as a guard against
2511 * additional headers get tacked on somewhere that we're not aware of. */
2512 netdev_get_mtu(netdev, &mtu);
2513 hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0;
2514 hc->burst = MAX(hc->burst, mtu + 64);
2517 hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0;
2523 htb_query_class__(const struct netdev *netdev, unsigned int handle,
2524 unsigned int parent, struct htb_class *options,
2525 struct netdev_queue_stats *stats)
2527 struct ofpbuf *reply;
2530 error = tc_query_class(netdev, handle, parent, &reply);
2532 error = htb_parse_tcmsg__(reply, NULL, options, stats);
2533 ofpbuf_delete(reply);
2539 htb_tc_install(struct netdev *netdev, const struct shash *details)
2543 error = htb_setup_qdisc__(netdev);
2545 struct htb_class hc;
2547 htb_parse_qdisc_details__(netdev, details, &hc);
2548 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
2549 tc_make_handle(1, 0), &hc);
2551 htb_install__(netdev, hc.max_rate);
2557 static struct htb_class *
2558 htb_class_cast__(const struct tc_queue *queue)
2560 return CONTAINER_OF(queue, struct htb_class, tc_queue);
2564 htb_update_queue__(struct netdev *netdev, unsigned int queue_id,
2565 const struct htb_class *hc)
2567 struct htb *htb = htb_get__(netdev);
2568 size_t hash = hash_int(queue_id, 0);
2569 struct tc_queue *queue;
2570 struct htb_class *hcp;
2572 queue = tc_find_queue__(netdev, queue_id, hash);
2574 hcp = htb_class_cast__(queue);
2576 hcp = xmalloc(sizeof *hcp);
2577 queue = &hcp->tc_queue;
2578 queue->queue_id = queue_id;
2579 hmap_insert(&htb->tc.queues, &queue->hmap_node, hash);
2582 hcp->min_rate = hc->min_rate;
2583 hcp->max_rate = hc->max_rate;
2584 hcp->burst = hc->burst;
2585 hcp->priority = hc->priority;
2589 htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
2592 struct nl_dump dump;
2593 struct htb_class hc;
2596 /* Get qdisc options. */
2598 htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
2599 htb = htb_install__(netdev, hc.max_rate);
2602 if (!start_queue_dump(netdev, &dump)) {
2605 while (nl_dump_next(&dump, &msg)) {
2606 unsigned int queue_id;
2608 if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
2609 htb_update_queue__(netdev, queue_id, &hc);
2612 nl_dump_done(&dump);
2618 htb_tc_destroy(struct tc *tc)
2620 struct htb *htb = CONTAINER_OF(tc, struct htb, tc);
2621 struct htb_class *hc, *next;
2623 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) {
2624 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
2632 htb_qdisc_get(const struct netdev *netdev, struct shash *details)
2634 const struct htb *htb = htb_get__(netdev);
2635 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * htb->max_rate));
2640 htb_qdisc_set(struct netdev *netdev, const struct shash *details)
2642 struct htb_class hc;
2645 htb_parse_qdisc_details__(netdev, details, &hc);
2646 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
2647 tc_make_handle(1, 0), &hc);
2649 htb_get__(netdev)->max_rate = hc.max_rate;
2655 htb_class_get(const struct netdev *netdev OVS_UNUSED,
2656 const struct tc_queue *queue, struct shash *details)
2658 const struct htb_class *hc = htb_class_cast__(queue);
2660 shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate));
2661 if (hc->min_rate != hc->max_rate) {
2662 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate));
2664 shash_add(details, "burst", xasprintf("%llu", 8ULL * hc->burst));
2666 shash_add(details, "priority", xasprintf("%u", hc->priority));
2672 htb_class_set(struct netdev *netdev, unsigned int queue_id,
2673 const struct shash *details)
2675 struct htb_class hc;
2678 error = htb_parse_class_details__(netdev, details, &hc);
2683 error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
2684 tc_make_handle(1, 0xfffe), &hc);
2689 htb_update_queue__(netdev, queue_id, &hc);
2694 htb_class_delete(struct netdev *netdev, struct tc_queue *queue)
2696 struct htb_class *hc = htb_class_cast__(queue);
2697 struct htb *htb = htb_get__(netdev);
2700 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
2702 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
2709 htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
2710 struct netdev_queue_stats *stats)
2712 return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
2713 tc_make_handle(1, 0xfffe), NULL, stats);
2717 htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
2718 const struct ofpbuf *nlmsg,
2719 netdev_dump_queue_stats_cb *cb, void *aux)
2721 struct netdev_queue_stats stats;
2722 unsigned int handle, major, minor;
2725 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
2730 major = tc_get_major(handle);
2731 minor = tc_get_minor(handle);
2732 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2733 (*cb)(minor - 1, &stats, aux);
2738 static const struct tc_ops tc_ops_htb = {
2739 "htb", /* linux_name */
2740 "linux-htb", /* ovs_name */
2741 HTB_N_QUEUES, /* n_queues */
2750 htb_class_get_stats,
2751 htb_class_dump_stats
2754 /* "linux-hfsc" traffic control class. */
2756 #define HFSC_N_QUEUES 0xf000
2764 struct tc_queue tc_queue;
2769 static struct hfsc *
2770 hfsc_get__(const struct netdev *netdev)
2772 struct netdev_dev_linux *netdev_dev;
2773 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
2774 return CONTAINER_OF(netdev_dev->tc, struct hfsc, tc);
2777 static struct hfsc_class *
2778 hfsc_class_cast__(const struct tc_queue *queue)
2780 return CONTAINER_OF(queue, struct hfsc_class, tc_queue);
2783 static struct hfsc *
2784 hfsc_install__(struct netdev *netdev, uint32_t max_rate)
2786 struct netdev_dev_linux * netdev_dev;
2789 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
2790 hfsc = xmalloc(sizeof *hfsc);
2791 tc_init(&hfsc->tc, &tc_ops_hfsc);
2792 hfsc->max_rate = max_rate;
2793 netdev_dev->tc = &hfsc->tc;
2799 hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id,
2800 const struct hfsc_class *hc)
2804 struct hfsc_class *hcp;
2805 struct tc_queue *queue;
2807 hfsc = hfsc_get__(netdev);
2808 hash = hash_int(queue_id, 0);
2810 queue = tc_find_queue__(netdev, queue_id, hash);
2812 hcp = hfsc_class_cast__(queue);
2814 hcp = xmalloc(sizeof *hcp);
2815 queue = &hcp->tc_queue;
2816 queue->queue_id = queue_id;
2817 hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash);
2820 hcp->min_rate = hc->min_rate;
2821 hcp->max_rate = hc->max_rate;
2825 hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class)
2827 const struct tc_service_curve *rsc, *fsc, *usc;
2828 static const struct nl_policy tca_hfsc_policy[] = {
2830 .type = NL_A_UNSPEC,
2832 .min_len = sizeof(struct tc_service_curve),
2835 .type = NL_A_UNSPEC,
2837 .min_len = sizeof(struct tc_service_curve),
2840 .type = NL_A_UNSPEC,
2842 .min_len = sizeof(struct tc_service_curve),
2845 struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)];
2847 if (!nl_parse_nested(nl_options, tca_hfsc_policy,
2848 attrs, ARRAY_SIZE(tca_hfsc_policy))) {
2849 VLOG_WARN_RL(&rl, "failed to parse HFSC class options");
2853 rsc = nl_attr_get(attrs[TCA_HFSC_RSC]);
2854 fsc = nl_attr_get(attrs[TCA_HFSC_FSC]);
2855 usc = nl_attr_get(attrs[TCA_HFSC_USC]);
2857 if (rsc->m1 != 0 || rsc->d != 0 ||
2858 fsc->m1 != 0 || fsc->d != 0 ||
2859 usc->m1 != 0 || usc->d != 0) {
2860 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
2861 "Non-linear service curves are not supported.");
2865 if (rsc->m2 != fsc->m2) {
2866 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
2867 "Real-time service curves are not supported ");
2871 if (rsc->m2 > usc->m2) {
2872 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
2873 "Min-rate service curve is greater than "
2874 "the max-rate service curve.");
2878 class->min_rate = fsc->m2;
2879 class->max_rate = usc->m2;
2884 hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
2885 struct hfsc_class *options,
2886 struct netdev_queue_stats *stats)
2889 unsigned int handle;
2890 struct nlattr *nl_options;
2892 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
2898 unsigned int major, minor;
2900 major = tc_get_major(handle);
2901 minor = tc_get_minor(handle);
2902 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
2903 *queue_id = minor - 1;
2910 error = hfsc_parse_tca_options__(nl_options, options);
2917 hfsc_query_class__(const struct netdev *netdev, unsigned int handle,
2918 unsigned int parent, struct hfsc_class *options,
2919 struct netdev_queue_stats *stats)
2922 struct ofpbuf *reply;
2924 error = tc_query_class(netdev, handle, parent, &reply);
2929 error = hfsc_parse_tcmsg__(reply, NULL, options, stats);
2930 ofpbuf_delete(reply);
2935 hfsc_parse_qdisc_details__(struct netdev *netdev, const struct shash *details,
2936 struct hfsc_class *class)
2939 const char *max_rate_s;
2941 max_rate_s = shash_find_data(details, "max-rate");
2942 max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
2947 netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
2948 max_rate = netdev_features_to_bps(current) / 8;
2951 class->min_rate = max_rate;
2952 class->max_rate = max_rate;
2956 hfsc_parse_class_details__(struct netdev *netdev,
2957 const struct shash *details,
2958 struct hfsc_class * class)
2960 const struct hfsc *hfsc;
2961 uint32_t min_rate, max_rate;
2962 const char *min_rate_s, *max_rate_s;
2964 hfsc = hfsc_get__(netdev);
2965 min_rate_s = shash_find_data(details, "min-rate");
2966 max_rate_s = shash_find_data(details, "max-rate");
2972 min_rate = strtoull(min_rate_s, NULL, 10) / 8;
2973 min_rate = MAX(min_rate, 1500);
2974 min_rate = MIN(min_rate, hfsc->max_rate);
2976 max_rate = (max_rate_s
2977 ? strtoull(max_rate_s, NULL, 10) / 8
2979 max_rate = MAX(max_rate, min_rate);
2980 max_rate = MIN(max_rate, hfsc->max_rate);
2982 class->min_rate = min_rate;
2983 class->max_rate = max_rate;
2988 /* Create an HFSC qdisc.
2990 * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */
2992 hfsc_setup_qdisc__(struct netdev * netdev)
2994 struct tcmsg *tcmsg;
2995 struct ofpbuf request;
2996 struct tc_hfsc_qopt opt;
2998 tc_del_qdisc(netdev);
3000 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
3001 NLM_F_EXCL | NLM_F_CREATE, &request);
3007 tcmsg->tcm_handle = tc_make_handle(1, 0);
3008 tcmsg->tcm_parent = TC_H_ROOT;
3010 memset(&opt, 0, sizeof opt);
3013 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3014 nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt);
3016 return tc_transact(&request, NULL);
3019 /* Create an HFSC class.
3021 * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc
3022 * sc rate <min_rate> ul rate <max_rate>" */
3024 hfsc_setup_class__(struct netdev *netdev, unsigned int handle,
3025 unsigned int parent, struct hfsc_class *class)
3029 struct tcmsg *tcmsg;
3030 struct ofpbuf request;
3031 struct tc_service_curve min, max;
3033 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
3039 tcmsg->tcm_handle = handle;
3040 tcmsg->tcm_parent = parent;
3044 min.m2 = class->min_rate;
3048 max.m2 = class->max_rate;
3050 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3051 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
3052 nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min);
3053 nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min);
3054 nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max);
3055 nl_msg_end_nested(&request, opt_offset);
3057 error = tc_transact(&request, NULL);
3059 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
3060 "min-rate %ubps, max-rate %ubps (%s)",
3061 netdev_get_name(netdev),
3062 tc_get_major(handle), tc_get_minor(handle),
3063 tc_get_major(parent), tc_get_minor(parent),
3064 class->min_rate, class->max_rate, strerror(error));
3071 hfsc_tc_install(struct netdev *netdev, const struct shash *details)
3074 struct hfsc_class class;
3076 error = hfsc_setup_qdisc__(netdev);
3082 hfsc_parse_qdisc_details__(netdev, details, &class);
3083 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3084 tc_make_handle(1, 0), &class);
3090 hfsc_install__(netdev, class.max_rate);
3095 hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3099 struct nl_dump dump;
3100 struct hfsc_class hc;
3103 hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
3104 hfsc = hfsc_install__(netdev, hc.max_rate);
3106 if (!start_queue_dump(netdev, &dump)) {
3110 while (nl_dump_next(&dump, &msg)) {
3111 unsigned int queue_id;
3113 if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
3114 hfsc_update_queue__(netdev, queue_id, &hc);
3118 nl_dump_done(&dump);
3123 hfsc_tc_destroy(struct tc *tc)
3126 struct hfsc_class *hc, *next;
3128 hfsc = CONTAINER_OF(tc, struct hfsc, tc);
3130 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) {
3131 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3140 hfsc_qdisc_get(const struct netdev *netdev, struct shash *details)
3142 const struct hfsc *hfsc;
3143 hfsc = hfsc_get__(netdev);
3144 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hfsc->max_rate));
3149 hfsc_qdisc_set(struct netdev *netdev, const struct shash *details)
3152 struct hfsc_class class;
3154 hfsc_parse_qdisc_details__(netdev, details, &class);
3155 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3156 tc_make_handle(1, 0), &class);
3159 hfsc_get__(netdev)->max_rate = class.max_rate;
3166 hfsc_class_get(const struct netdev *netdev OVS_UNUSED,
3167 const struct tc_queue *queue, struct shash *details)
3169 const struct hfsc_class *hc;
3171 hc = hfsc_class_cast__(queue);
3172 shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate));
3173 if (hc->min_rate != hc->max_rate) {
3174 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate));
3180 hfsc_class_set(struct netdev *netdev, unsigned int queue_id,
3181 const struct shash *details)
3184 struct hfsc_class class;
3186 error = hfsc_parse_class_details__(netdev, details, &class);
3191 error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
3192 tc_make_handle(1, 0xfffe), &class);
3197 hfsc_update_queue__(netdev, queue_id, &class);
3202 hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue)
3206 struct hfsc_class *hc;
3208 hc = hfsc_class_cast__(queue);
3209 hfsc = hfsc_get__(netdev);
3211 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
3213 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3220 hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
3221 struct netdev_queue_stats *stats)
3223 return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
3224 tc_make_handle(1, 0xfffe), NULL, stats);
3228 hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
3229 const struct ofpbuf *nlmsg,
3230 netdev_dump_queue_stats_cb *cb, void *aux)
3232 struct netdev_queue_stats stats;
3233 unsigned int handle, major, minor;
3236 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
3241 major = tc_get_major(handle);
3242 minor = tc_get_minor(handle);
3243 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
3244 (*cb)(minor - 1, &stats, aux);
3249 static const struct tc_ops tc_ops_hfsc = {
3250 "hfsc", /* linux_name */
3251 "linux-hfsc", /* ovs_name */
3252 HFSC_N_QUEUES, /* n_queues */
3253 hfsc_tc_install, /* tc_install */
3254 hfsc_tc_load, /* tc_load */
3255 hfsc_tc_destroy, /* tc_destroy */
3256 hfsc_qdisc_get, /* qdisc_get */
3257 hfsc_qdisc_set, /* qdisc_set */
3258 hfsc_class_get, /* class_get */
3259 hfsc_class_set, /* class_set */
3260 hfsc_class_delete, /* class_delete */
3261 hfsc_class_get_stats, /* class_get_stats */
3262 hfsc_class_dump_stats /* class_dump_stats */
3265 /* "linux-default" traffic control class.
3267 * This class represents the default, unnamed Linux qdisc. It corresponds to
3268 * the "" (empty string) QoS type in the OVS database. */
3271 default_install__(struct netdev *netdev)
3273 struct netdev_dev_linux *netdev_dev =
3274 netdev_dev_linux_cast(netdev_get_dev(netdev));
3275 static struct tc *tc;
3278 tc = xmalloc(sizeof *tc);
3279 tc_init(tc, &tc_ops_default);
3281 netdev_dev->tc = tc;
3285 default_tc_install(struct netdev *netdev,
3286 const struct shash *details OVS_UNUSED)
3288 default_install__(netdev);
3293 default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3295 default_install__(netdev);
3299 static const struct tc_ops tc_ops_default = {
3300 NULL, /* linux_name */
3305 NULL, /* tc_destroy */
3306 NULL, /* qdisc_get */
3307 NULL, /* qdisc_set */
3308 NULL, /* class_get */
3309 NULL, /* class_set */
3310 NULL, /* class_delete */
3311 NULL, /* class_get_stats */
3312 NULL /* class_dump_stats */
3315 /* "linux-other" traffic control class.
3320 other_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3322 struct netdev_dev_linux *netdev_dev =
3323 netdev_dev_linux_cast(netdev_get_dev(netdev));
3324 static struct tc *tc;
3327 tc = xmalloc(sizeof *tc);
3328 tc_init(tc, &tc_ops_other);
3330 netdev_dev->tc = tc;
3334 static const struct tc_ops tc_ops_other = {
3335 NULL, /* linux_name */
3336 "linux-other", /* ovs_name */
3338 NULL, /* tc_install */
3340 NULL, /* tc_destroy */
3341 NULL, /* qdisc_get */
3342 NULL, /* qdisc_set */
3343 NULL, /* class_get */
3344 NULL, /* class_set */
3345 NULL, /* class_delete */
3346 NULL, /* class_get_stats */
3347 NULL /* class_dump_stats */
3350 /* Traffic control. */
3352 /* Number of kernel "tc" ticks per second. */
3353 static double ticks_per_s;
3355 /* Number of kernel "jiffies" per second. This is used for the purpose of
3356 * computing buffer sizes. Generally kernel qdiscs need to be able to buffer
3357 * one jiffy's worth of data.
3359 * There are two possibilities here:
3361 * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the
3362 * approximate range of 100 to 1024. That means that we really need to
3363 * make sure that the qdisc can buffer that much data.
3365 * - 'buffer_hz' is an absurdly large number. That means that the kernel
3366 * has finely granular timers and there's no need to fudge additional room
3367 * for buffers. (There's no extra effort needed to implement that: the
3368 * large 'buffer_hz' is used as a divisor, so practically any number will
3369 * come out as 0 in the division. Small integer results in the case of
3370 * really high dividends won't have any real effect anyhow.)
3372 static unsigned int buffer_hz;
3374 /* Returns tc handle 'major':'minor'. */
3376 tc_make_handle(unsigned int major, unsigned int minor)
3378 return TC_H_MAKE(major << 16, minor);
3381 /* Returns the major number from 'handle'. */
3383 tc_get_major(unsigned int handle)
3385 return TC_H_MAJ(handle) >> 16;
3388 /* Returns the minor number from 'handle'. */
3390 tc_get_minor(unsigned int handle)
3392 return TC_H_MIN(handle);
3395 static struct tcmsg *
3396 tc_make_request(const struct netdev *netdev, int type, unsigned int flags,
3397 struct ofpbuf *request)
3399 struct tcmsg *tcmsg;
3403 error = get_ifindex(netdev, &ifindex);
3408 ofpbuf_init(request, 512);
3409 nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags);
3410 tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg);
3411 tcmsg->tcm_family = AF_UNSPEC;
3412 tcmsg->tcm_ifindex = ifindex;
3413 /* Caller should fill in tcmsg->tcm_handle. */
3414 /* Caller should fill in tcmsg->tcm_parent. */
3420 tc_transact(struct ofpbuf *request, struct ofpbuf **replyp)
3422 int error = nl_sock_transact(rtnl_sock, request, replyp);
3423 ofpbuf_uninit(request);
3430 /* The values in psched are not individually very meaningful, but they are
3431 * important. The tables below show some values seen in the wild.
3435 * - "c" has always been a constant 1000000 since at least Linux 2.4.14.
3436 * (Before that, there are hints that it was 1000000000.)
3438 * - "d" can be unrealistically large, see the comment on 'buffer_hz'
3442 * -----------------------------------
3443 * [1] 000c8000 000f4240 000f4240 00000064
3444 * [2] 000003e8 00000400 000f4240 3b9aca00
3445 * [3] 000003e8 00000400 000f4240 3b9aca00
3446 * [4] 000003e8 00000400 000f4240 00000064
3447 * [5] 000003e8 00000040 000f4240 3b9aca00
3448 * [6] 000003e8 00000040 000f4240 000000f9
3450 * a b c d ticks_per_s buffer_hz
3451 * ------- --------- ---------- ------------- ----------- -------------
3452 * [1] 819,200 1,000,000 1,000,000 100 819,200 100
3453 * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
3454 * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
3455 * [4] 1,000 1,024 1,000,000 100 976,562 100
3456 * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000
3457 * [6] 1,000 64 1,000,000 249 15,625,000 249
3459 * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p
3460 * [2] 2.6.26-1-686-bigmem from Debian lenny
3461 * [3] 2.6.26-2-sparc64 from Debian lenny
3462 * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p
3463 * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion
3464 * [6] 2.6.34 from kernel.org on KVM
3466 static const char fn[] = "/proc/net/psched";
3467 unsigned int a, b, c, d;
3473 stream = fopen(fn, "r");
3475 VLOG_WARN("%s: open failed: %s", fn, strerror(errno));
3479 if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) {
3480 VLOG_WARN("%s: read failed", fn);
3484 VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d);
3488 VLOG_WARN("%s: invalid scheduler parameters", fn);
3492 ticks_per_s = (double) a * c / b;
3496 VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u",
3499 VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz);
3502 /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a
3503 * rate of 'rate' bytes per second. */
3505 tc_ticks_to_bytes(unsigned int rate, unsigned int ticks)
3510 return (rate * ticks) / ticks_per_s;
3513 /* Returns the number of ticks that it would take to transmit 'size' bytes at a
3514 * rate of 'rate' bytes per second. */
3516 tc_bytes_to_ticks(unsigned int rate, unsigned int size)
3521 return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0;
3524 /* Returns the number of bytes that need to be reserved for qdisc buffering at
3525 * a transmission rate of 'rate' bytes per second. */
3527 tc_buffer_per_jiffy(unsigned int rate)
3532 return rate / buffer_hz;
3535 /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc,
3536 * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull,
3537 * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or
3538 * stores NULL into it if it is absent.
3540 * '*kind' and '*options' point into 'msg', so they are owned by whoever owns
3543 * Returns 0 if successful, otherwise a positive errno value. */
3545 tc_parse_qdisc(const struct ofpbuf *msg, const char **kind,
3546 struct nlattr **options)
3548 static const struct nl_policy tca_policy[] = {
3549 [TCA_KIND] = { .type = NL_A_STRING, .optional = false },
3550 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
3552 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
3554 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
3555 tca_policy, ta, ARRAY_SIZE(ta))) {
3556 VLOG_WARN_RL(&rl, "failed to parse qdisc message");
3561 *kind = nl_attr_get_string(ta[TCA_KIND]);
3565 *options = ta[TCA_OPTIONS];
3580 /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the
3581 * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute
3582 * into '*options', and its queue statistics into '*stats'. Any of the output
3583 * arguments may be null.
3585 * Returns 0 if successful, otherwise a positive errno value. */
3587 tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep,
3588 struct nlattr **options, struct netdev_queue_stats *stats)
3590 static const struct nl_policy tca_policy[] = {
3591 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false },
3592 [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false },
3594 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
3596 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
3597 tca_policy, ta, ARRAY_SIZE(ta))) {
3598 VLOG_WARN_RL(&rl, "failed to parse class message");
3603 struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc);
3604 *handlep = tc->tcm_handle;
3608 *options = ta[TCA_OPTIONS];
3612 const struct gnet_stats_queue *gsq;
3613 struct gnet_stats_basic gsb;
3615 static const struct nl_policy stats_policy[] = {
3616 [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false,
3617 .min_len = sizeof gsb },
3618 [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false,
3619 .min_len = sizeof *gsq },
3621 struct nlattr *sa[ARRAY_SIZE(stats_policy)];
3623 if (!nl_parse_nested(ta[TCA_STATS2], stats_policy,
3624 sa, ARRAY_SIZE(sa))) {
3625 VLOG_WARN_RL(&rl, "failed to parse class stats");
3629 /* Alignment issues screw up the length of struct gnet_stats_basic on
3630 * some arch/bitsize combinations. Newer versions of Linux have a
3631 * struct gnet_stats_basic_packed, but we can't depend on that. The
3632 * easiest thing to do is just to make a copy. */
3633 memset(&gsb, 0, sizeof gsb);
3634 memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]),
3635 MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb));
3636 stats->tx_bytes = gsb.bytes;
3637 stats->tx_packets = gsb.packets;
3639 gsq = nl_attr_get(sa[TCA_STATS_QUEUE]);
3640 stats->tx_errors = gsq->drops;
3650 memset(stats, 0, sizeof *stats);
3655 /* Queries the kernel for class with identifier 'handle' and parent 'parent'
3658 tc_query_class(const struct netdev *netdev,
3659 unsigned int handle, unsigned int parent,
3660 struct ofpbuf **replyp)
3662 struct ofpbuf request;
3663 struct tcmsg *tcmsg;
3666 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request);
3670 tcmsg->tcm_handle = handle;
3671 tcmsg->tcm_parent = parent;
3673 error = tc_transact(&request, replyp);
3675 VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)",
3676 netdev_get_name(netdev),
3677 tc_get_major(handle), tc_get_minor(handle),
3678 tc_get_major(parent), tc_get_minor(parent),
3684 /* Equivalent to "tc class del dev <name> handle <handle>". */
3686 tc_delete_class(const struct netdev *netdev, unsigned int handle)
3688 struct ofpbuf request;
3689 struct tcmsg *tcmsg;
3692 tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request);
3696 tcmsg->tcm_handle = handle;
3697 tcmsg->tcm_parent = 0;
3699 error = tc_transact(&request, NULL);
3701 VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)",
3702 netdev_get_name(netdev),
3703 tc_get_major(handle), tc_get_minor(handle),
3709 /* Equivalent to "tc qdisc del dev <name> root". */
3711 tc_del_qdisc(struct netdev *netdev)
3713 struct netdev_dev_linux *netdev_dev =
3714 netdev_dev_linux_cast(netdev_get_dev(netdev));
3715 struct ofpbuf request;
3716 struct tcmsg *tcmsg;
3719 tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request);
3723 tcmsg->tcm_handle = tc_make_handle(1, 0);
3724 tcmsg->tcm_parent = TC_H_ROOT;
3726 error = tc_transact(&request, NULL);
3727 if (error == EINVAL) {
3728 /* EINVAL probably means that the default qdisc was in use, in which
3729 * case we've accomplished our purpose. */
3732 if (!error && netdev_dev->tc) {
3733 if (netdev_dev->tc->ops->tc_destroy) {
3734 netdev_dev->tc->ops->tc_destroy(netdev_dev->tc);
3736 netdev_dev->tc = NULL;
3741 /* If 'netdev''s qdisc type and parameters are not yet known, queries the
3742 * kernel to determine what they are. Returns 0 if successful, otherwise a
3743 * positive errno value. */
3745 tc_query_qdisc(const struct netdev *netdev)
3747 struct netdev_dev_linux *netdev_dev =
3748 netdev_dev_linux_cast(netdev_get_dev(netdev));
3749 struct ofpbuf request, *qdisc;
3750 const struct tc_ops *ops;
3751 struct tcmsg *tcmsg;
3755 if (netdev_dev->tc) {
3759 /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have
3760 * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before
3761 * 2.6.35 without that fix backported to it.
3763 * To avoid the OOPS, we must not make a request that would attempt to dump
3764 * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a
3765 * few others. There are a few ways that I can see to do this, but most of
3766 * them seem to be racy (and if you lose the race the kernel OOPSes). The
3767 * technique chosen here is to assume that any non-default qdisc that we
3768 * create will have a class with handle 1:0. The built-in qdiscs only have
3769 * a class with handle 0:0.
3771 * We could check for Linux 2.6.35+ and use a more straightforward method
3773 tcmsg = tc_make_request(netdev, RTM_GETQDISC, NLM_F_ECHO, &request);
3777 tcmsg->tcm_handle = tc_make_handle(1, 0);
3778 tcmsg->tcm_parent = 0;
3780 /* Figure out what tc class to instantiate. */
3781 error = tc_transact(&request, &qdisc);
3785 error = tc_parse_qdisc(qdisc, &kind, NULL);
3787 ops = &tc_ops_other;
3789 ops = tc_lookup_linux_name(kind);
3791 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1);
3792 VLOG_INFO_RL(&rl2, "unknown qdisc \"%s\"", kind);
3794 ops = &tc_ops_other;
3797 } else if (error == ENOENT) {
3798 /* Either it's a built-in qdisc, or it's a qdisc set up by some
3799 * other entity that doesn't have a handle 1:0. We will assume
3800 * that it's the system default qdisc. */
3801 ops = &tc_ops_default;
3804 /* Who knows? Maybe the device got deleted. */
3805 VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)",
3806 netdev_get_name(netdev), strerror(error));
3807 ops = &tc_ops_other;
3810 /* Instantiate it. */
3811 load_error = ops->tc_load((struct netdev *) netdev, qdisc);
3812 assert((load_error == 0) == (netdev_dev->tc != NULL));
3813 ofpbuf_delete(qdisc);
3815 return error ? error : load_error;
3818 /* Linux traffic control uses tables with 256 entries ("rtab" tables) to
3819 approximate the time to transmit packets of various lengths. For an MTU of
3820 256 or less, each entry is exact; for an MTU of 257 through 512, each entry
3821 represents two possible packet lengths; for a MTU of 513 through 1024, four
3822 possible lengths; and so on.
3824 Returns, for the specified 'mtu', the number of bits that packet lengths
3825 need to be shifted right to fit within such a 256-entry table. */
3827 tc_calc_cell_log(unsigned int mtu)
3832 mtu = ETH_PAYLOAD_MAX;
3834 mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN;
3836 for (cell_log = 0; mtu >= 256; cell_log++) {
3843 /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU
3846 tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu)
3848 memset(rate, 0, sizeof *rate);
3849 rate->cell_log = tc_calc_cell_log(mtu);
3850 /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */
3851 /* rate->cell_align = 0; */ /* distro headers. */
3852 rate->mpu = ETH_TOTAL_MIN;
3856 /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink
3857 * attribute of the specified "type".
3859 * See tc_calc_cell_log() above for a description of "rtab"s. */
3861 tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate)
3866 rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE);
3867 for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) {
3868 unsigned packet_size = (i + 1) << rate->cell_log;
3869 if (packet_size < rate->mpu) {
3870 packet_size = rate->mpu;
3872 rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size);
3876 /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a
3877 * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested
3878 * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of
3881 tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes)
3883 unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu;
3884 return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst));
3888 /* Utility functions. */
3891 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
3893 /* Policy for RTNLGRP_LINK messages.
3895 * There are *many* more fields in these messages, but currently we only
3896 * care about these fields. */
3897 static const struct nl_policy rtnlgrp_link_policy[] = {
3898 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
3899 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
3900 .min_len = sizeof(struct rtnl_link_stats) },
3903 struct ofpbuf request;
3904 struct ofpbuf *reply;
3905 struct ifinfomsg *ifi;
3906 const struct rtnl_link_stats *rtnl_stats;
3907 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
3910 ofpbuf_init(&request, 0);
3911 nl_msg_put_nlmsghdr(&request, sizeof *ifi, RTM_GETLINK, NLM_F_REQUEST);
3912 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
3913 ifi->ifi_family = PF_UNSPEC;
3914 ifi->ifi_index = ifindex;
3915 error = nl_sock_transact(rtnl_sock, &request, &reply);
3916 ofpbuf_uninit(&request);
3921 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
3922 rtnlgrp_link_policy,
3923 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
3924 ofpbuf_delete(reply);
3928 if (!attrs[IFLA_STATS]) {
3929 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
3930 ofpbuf_delete(reply);
3934 rtnl_stats = nl_attr_get(attrs[IFLA_STATS]);
3935 stats->rx_packets = rtnl_stats->rx_packets;
3936 stats->tx_packets = rtnl_stats->tx_packets;
3937 stats->rx_bytes = rtnl_stats->rx_bytes;
3938 stats->tx_bytes = rtnl_stats->tx_bytes;
3939 stats->rx_errors = rtnl_stats->rx_errors;
3940 stats->tx_errors = rtnl_stats->tx_errors;
3941 stats->rx_dropped = rtnl_stats->rx_dropped;
3942 stats->tx_dropped = rtnl_stats->tx_dropped;
3943 stats->multicast = rtnl_stats->multicast;
3944 stats->collisions = rtnl_stats->collisions;
3945 stats->rx_length_errors = rtnl_stats->rx_length_errors;
3946 stats->rx_over_errors = rtnl_stats->rx_over_errors;
3947 stats->rx_crc_errors = rtnl_stats->rx_crc_errors;
3948 stats->rx_frame_errors = rtnl_stats->rx_frame_errors;
3949 stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors;
3950 stats->rx_missed_errors = rtnl_stats->rx_missed_errors;
3951 stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors;
3952 stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors;
3953 stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors;
3954 stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors;
3955 stats->tx_window_errors = rtnl_stats->tx_window_errors;
3957 ofpbuf_delete(reply);
3963 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
3965 static const char fn[] = "/proc/net/dev";
3970 stream = fopen(fn, "r");
3972 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
3977 while (fgets(line, sizeof line, stream)) {
3980 #define X64 "%"SCNu64
3983 X64 X64 X64 X64 X64 X64 X64 "%*u"
3984 X64 X64 X64 X64 X64 X64 X64 "%*u",
3990 &stats->rx_fifo_errors,
3991 &stats->rx_frame_errors,
3997 &stats->tx_fifo_errors,
3999 &stats->tx_carrier_errors) != 15) {
4000 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
4001 } else if (!strcmp(devname, netdev_name)) {
4002 stats->rx_length_errors = UINT64_MAX;
4003 stats->rx_over_errors = UINT64_MAX;
4004 stats->rx_crc_errors = UINT64_MAX;
4005 stats->rx_missed_errors = UINT64_MAX;
4006 stats->tx_aborted_errors = UINT64_MAX;
4007 stats->tx_heartbeat_errors = UINT64_MAX;
4008 stats->tx_window_errors = UINT64_MAX;
4014 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
4020 get_flags(const struct netdev *netdev, int *flags)
4025 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS,
4027 *flags = ifr.ifr_flags;
4032 set_flags(struct netdev *netdev, int flags)
4036 ifr.ifr_flags = flags;
4037 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
4042 do_get_ifindex(const char *netdev_name)
4046 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4047 COVERAGE_INC(netdev_get_ifindex);
4048 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
4049 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
4050 netdev_name, strerror(errno));
4053 return ifr.ifr_ifindex;
4057 get_ifindex(const struct netdev *netdev_, int *ifindexp)
4059 struct netdev_dev_linux *netdev_dev =
4060 netdev_dev_linux_cast(netdev_get_dev(netdev_));
4062 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
4063 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
4067 netdev_dev->cache_valid |= VALID_IFINDEX;
4068 netdev_dev->ifindex = ifindex;
4070 *ifindexp = netdev_dev->ifindex;
4075 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
4080 memset(&ifr, 0, sizeof ifr);
4081 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4082 COVERAGE_INC(netdev_get_hwaddr);
4083 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
4084 VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
4085 netdev_name, strerror(errno));
4088 hwaddr_family = ifr.ifr_hwaddr.sa_family;
4089 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
4090 VLOG_WARN("%s device has unknown hardware address family %d",
4091 netdev_name, hwaddr_family);
4093 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
4098 set_etheraddr(const char *netdev_name, int hwaddr_family,
4099 const uint8_t mac[ETH_ADDR_LEN])
4103 memset(&ifr, 0, sizeof ifr);
4104 strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4105 ifr.ifr_hwaddr.sa_family = hwaddr_family;
4106 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
4107 COVERAGE_INC(netdev_set_hwaddr);
4108 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
4109 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
4110 netdev_name, strerror(errno));
4117 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
4118 int cmd, const char *cmd_name)
4122 memset(&ifr, 0, sizeof ifr);
4123 strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
4124 ifr.ifr_data = (caddr_t) ecmd;
4127 COVERAGE_INC(netdev_ethtool);
4128 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
4131 if (errno != EOPNOTSUPP) {
4132 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
4133 "failed: %s", cmd_name, name, strerror(errno));
4135 /* The device doesn't support this operation. That's pretty
4136 * common, so there's no point in logging anything. */
4143 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
4144 const char *cmd_name)
4146 strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
4147 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
4148 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
4156 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
4157 int cmd, const char *cmd_name)
4162 ifr.ifr_addr.sa_family = AF_INET;
4163 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
4165 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
4166 *ip = sin->sin_addr;