2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "netdev-linux.h"
24 #include <arpa/inet.h>
26 #include <linux/gen_stats.h>
27 #include <linux/if_ether.h>
28 #include <linux/if_tun.h>
29 #include <linux/types.h>
30 #include <linux/ethtool.h>
31 #include <linux/mii.h>
32 #include <linux/pkt_cls.h>
33 #include <linux/pkt_sched.h>
34 #include <linux/rtnetlink.h>
35 #include <linux/sockios.h>
36 #include <linux/version.h>
37 #include <sys/types.h>
38 #include <sys/ioctl.h>
39 #include <sys/socket.h>
40 #include <netpacket/packet.h>
42 #include <net/if_arp.h>
43 #include <net/if_packet.h>
44 #include <net/route.h>
45 #include <netinet/in.h>
52 #include "dpif-linux.h"
53 #include "dynamic-string.h"
54 #include "fatal-signal.h"
57 #include "netdev-provider.h"
58 #include "netdev-vport.h"
60 #include "netlink-notifier.h"
61 #include "netlink-socket.h"
63 #include "openflow/openflow.h"
65 #include "poll-loop.h"
66 #include "rtnetlink-link.h"
67 #include "socket-util.h"
73 VLOG_DEFINE_THIS_MODULE(netdev_linux);
75 COVERAGE_DEFINE(netdev_set_policing);
76 COVERAGE_DEFINE(netdev_arp_lookup);
77 COVERAGE_DEFINE(netdev_get_ifindex);
78 COVERAGE_DEFINE(netdev_get_hwaddr);
79 COVERAGE_DEFINE(netdev_set_hwaddr);
80 COVERAGE_DEFINE(netdev_get_ethtool);
81 COVERAGE_DEFINE(netdev_set_ethtool);
84 /* These were introduced in Linux 2.6.14, so they might be missing if we have
86 #ifndef ADVERTISED_Pause
87 #define ADVERTISED_Pause (1 << 13)
89 #ifndef ADVERTISED_Asym_Pause
90 #define ADVERTISED_Asym_Pause (1 << 14)
93 /* These were introduced in Linux 2.6.24, so they might be missing if we
94 * have old headers. */
95 #ifndef ETHTOOL_GFLAGS
96 #define ETHTOOL_GFLAGS 0x00000025 /* Get flags bitmap(ethtool_value) */
98 #ifndef ETHTOOL_SFLAGS
99 #define ETHTOOL_SFLAGS 0x00000026 /* Set flags bitmap(ethtool_value) */
102 /* This was introduced in Linux 2.6.25, so it might be missing if we have old
105 #define TC_RTAB_SIZE 1024
108 static struct nln_notifier *netdev_linux_cache_notifier = NULL;
109 static int cache_notifier_refcount;
112 VALID_IFINDEX = 1 << 0,
113 VALID_ETHERADDR = 1 << 1,
117 VALID_POLICING = 1 << 5,
118 VALID_VPORT_STAT_ERROR = 1 << 6,
119 VALID_DRVINFO = 1 << 7,
120 VALID_FEATURES = 1 << 8,
128 /* Traffic control. */
130 /* An instance of a traffic control class. Always associated with a particular
133 * Each TC implementation subclasses this with whatever additional data it
136 const struct tc_ops *ops;
137 struct hmap queues; /* Contains "struct tc_queue"s.
138 * Read by generic TC layer.
139 * Written only by TC implementation. */
142 /* One traffic control queue.
144 * Each TC implementation subclasses this with whatever additional data it
147 struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */
148 unsigned int queue_id; /* OpenFlow queue ID. */
151 /* A particular kind of traffic control. Each implementation generally maps to
152 * one particular Linux qdisc class.
154 * The functions below return 0 if successful or a positive errno value on
155 * failure, except where otherwise noted. All of them must be provided, except
156 * where otherwise noted. */
158 /* Name used by kernel in the TCA_KIND attribute of tcmsg, e.g. "htb".
159 * This is null for tc_ops_default and tc_ops_other, for which there are no
160 * appropriate values. */
161 const char *linux_name;
163 /* Name used in OVS database, e.g. "linux-htb". Must be nonnull. */
164 const char *ovs_name;
166 /* Number of supported OpenFlow queues, 0 for qdiscs that have no
167 * queues. The queues are numbered 0 through n_queues - 1. */
168 unsigned int n_queues;
170 /* Called to install this TC class on 'netdev'. The implementation should
171 * make the Netlink calls required to set up 'netdev' with the right qdisc
172 * and configure it according to 'details'. The implementation may assume
173 * that the current qdisc is the default; that is, there is no need for it
174 * to delete the current qdisc before installing itself.
176 * The contents of 'details' should be documented as valid for 'ovs_name'
177 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
178 * (which is built as ovs-vswitchd.conf.db(8)).
180 * This function must return 0 if and only if it sets 'netdev->tc' to an
181 * initialized 'struct tc'.
183 * (This function is null for tc_ops_other, which cannot be installed. For
184 * other TC classes it should always be nonnull.) */
185 int (*tc_install)(struct netdev *netdev, const struct smap *details);
187 /* Called when the netdev code determines (through a Netlink query) that
188 * this TC class's qdisc is installed on 'netdev', but we didn't install
189 * it ourselves and so don't know any of the details.
191 * 'nlmsg' is the kernel reply to a RTM_GETQDISC Netlink message for
192 * 'netdev'. The TCA_KIND attribute of 'nlmsg' is 'linux_name'. The
193 * implementation should parse the other attributes of 'nlmsg' as
194 * necessary to determine its configuration. If necessary it should also
195 * use Netlink queries to determine the configuration of queues on
198 * This function must return 0 if and only if it sets 'netdev->tc' to an
199 * initialized 'struct tc'. */
200 int (*tc_load)(struct netdev *netdev, struct ofpbuf *nlmsg);
202 /* Destroys the data structures allocated by the implementation as part of
203 * 'tc'. (This includes destroying 'tc->queues' by calling
206 * The implementation should not need to perform any Netlink calls. If
207 * desirable, the caller is responsible for deconfiguring the kernel qdisc.
208 * (But it may not be desirable.)
210 * This function may be null if 'tc' is trivial. */
211 void (*tc_destroy)(struct tc *tc);
213 /* Retrieves details of 'netdev->tc' configuration into 'details'.
215 * The implementation should not need to perform any Netlink calls, because
216 * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have
217 * cached the configuration.
219 * The contents of 'details' should be documented as valid for 'ovs_name'
220 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
221 * (which is built as ovs-vswitchd.conf.db(8)).
223 * This function may be null if 'tc' is not configurable.
225 int (*qdisc_get)(const struct netdev *netdev, struct smap *details);
227 /* Reconfigures 'netdev->tc' according to 'details', performing any
228 * required Netlink calls to complete the reconfiguration.
230 * The contents of 'details' should be documented as valid for 'ovs_name'
231 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
232 * (which is built as ovs-vswitchd.conf.db(8)).
234 * This function may be null if 'tc' is not configurable.
236 int (*qdisc_set)(struct netdev *, const struct smap *details);
238 /* Retrieves details of 'queue' on 'netdev->tc' into 'details'. 'queue' is
239 * one of the 'struct tc_queue's within 'netdev->tc->queues'.
241 * The contents of 'details' should be documented as valid for 'ovs_name'
242 * in the "other_config" column in the "Queue" table in
243 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
245 * The implementation should not need to perform any Netlink calls, because
246 * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have
247 * cached the queue configuration.
249 * This function may be null if 'tc' does not have queues ('n_queues' is
251 int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue,
252 struct smap *details);
254 /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to
255 * 'details', perfoming any required Netlink calls to complete the
256 * reconfiguration. The caller ensures that 'queue_id' is less than
259 * The contents of 'details' should be documented as valid for 'ovs_name'
260 * in the "other_config" column in the "Queue" table in
261 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
263 * This function may be null if 'tc' does not have queues or its queues are
264 * not configurable. */
265 int (*class_set)(struct netdev *, unsigned int queue_id,
266 const struct smap *details);
268 /* Deletes 'queue' from 'netdev->tc'. 'queue' is one of the 'struct
269 * tc_queue's within 'netdev->tc->queues'.
271 * This function may be null if 'tc' does not have queues or its queues
272 * cannot be deleted. */
273 int (*class_delete)(struct netdev *, struct tc_queue *queue);
275 /* Obtains stats for 'queue' from 'netdev->tc'. 'queue' is one of the
276 * 'struct tc_queue's within 'netdev->tc->queues'.
278 * On success, initializes '*stats'.
280 * This function may be null if 'tc' does not have queues or if it cannot
281 * report queue statistics. */
282 int (*class_get_stats)(const struct netdev *netdev,
283 const struct tc_queue *queue,
284 struct netdev_queue_stats *stats);
286 /* Extracts queue stats from 'nlmsg', which is a response to a
287 * RTM_GETTCLASS message, and passes them to 'cb' along with 'aux'.
289 * This function may be null if 'tc' does not have queues or if it cannot
290 * report queue statistics. */
291 int (*class_dump_stats)(const struct netdev *netdev,
292 const struct ofpbuf *nlmsg,
293 netdev_dump_queue_stats_cb *cb, void *aux);
297 tc_init(struct tc *tc, const struct tc_ops *ops)
300 hmap_init(&tc->queues);
304 tc_destroy(struct tc *tc)
306 hmap_destroy(&tc->queues);
309 static const struct tc_ops tc_ops_htb;
310 static const struct tc_ops tc_ops_hfsc;
311 static const struct tc_ops tc_ops_default;
312 static const struct tc_ops tc_ops_other;
314 static const struct tc_ops *tcs[] = {
315 &tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */
316 &tc_ops_hfsc, /* Hierarchical fair service curve. */
317 &tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */
318 &tc_ops_other, /* Some other qdisc. */
322 static unsigned int tc_make_handle(unsigned int major, unsigned int minor);
323 static unsigned int tc_get_major(unsigned int handle);
324 static unsigned int tc_get_minor(unsigned int handle);
326 static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks);
327 static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size);
328 static unsigned int tc_buffer_per_jiffy(unsigned int rate);
330 static struct tcmsg *tc_make_request(const struct netdev *, int type,
331 unsigned int flags, struct ofpbuf *);
332 static int tc_transact(struct ofpbuf *request, struct ofpbuf **replyp);
333 static int tc_add_del_ingress_qdisc(struct netdev *netdev, bool add);
334 static int tc_add_policer(struct netdev *netdev, int kbits_rate,
337 static int tc_parse_qdisc(const struct ofpbuf *, const char **kind,
338 struct nlattr **options);
339 static int tc_parse_class(const struct ofpbuf *, unsigned int *queue_id,
340 struct nlattr **options,
341 struct netdev_queue_stats *);
342 static int tc_query_class(const struct netdev *,
343 unsigned int handle, unsigned int parent,
344 struct ofpbuf **replyp);
345 static int tc_delete_class(const struct netdev *, unsigned int handle);
347 static int tc_del_qdisc(struct netdev *netdev);
348 static int tc_query_qdisc(const struct netdev *netdev);
350 static int tc_calc_cell_log(unsigned int mtu);
351 static void tc_fill_rate(struct tc_ratespec *rate, uint64_t bps, int mtu);
352 static void tc_put_rtab(struct ofpbuf *, uint16_t type,
353 const struct tc_ratespec *rate);
354 static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes);
356 struct netdev_dev_linux {
357 struct netdev_dev netdev_dev;
359 struct shash_node *shash_node;
360 unsigned int cache_valid;
361 unsigned int change_seq;
363 bool miimon; /* Link status of last poll. */
364 long long int miimon_interval; /* Miimon Poll rate. Disabled if <= 0. */
365 struct timer miimon_timer;
367 /* The following are figured out "on demand" only. They are only valid
368 * when the corresponding VALID_* bit in 'cache_valid' is set. */
370 uint8_t etheraddr[ETH_ADDR_LEN];
371 struct in_addr address, netmask;
374 unsigned int ifi_flags;
375 long long int carrier_resets;
376 uint32_t kbits_rate; /* Policing data. */
377 uint32_t kbits_burst;
378 int vport_stats_error; /* Cached error code from vport_get_stats().
379 0 or an errno value. */
380 int netdev_mtu_error; /* Cached error code from SIOCGIFMTU or SIOCSIFMTU. */
381 int ether_addr_error; /* Cached error code from set/get etheraddr. */
382 int netdev_policing_error; /* Cached error code from set policing. */
383 int get_features_error; /* Cached error code from ETHTOOL_GSET. */
384 int get_ifindex_error; /* Cached error code from SIOCGIFINDEX. */
386 enum netdev_features current; /* Cached from ETHTOOL_GSET. */
387 enum netdev_features advertised; /* Cached from ETHTOOL_GSET. */
388 enum netdev_features supported; /* Cached from ETHTOOL_GSET. */
389 enum netdev_features peer; /* Cached from ETHTOOL_GSET. */
391 struct ethtool_drvinfo drvinfo; /* Cached from ETHTOOL_GDRVINFO. */
395 struct tap_state tap;
399 struct netdev_linux {
400 struct netdev netdev;
404 /* Sockets used for ioctl operations. */
405 static int af_inet_sock = -1; /* AF_INET, SOCK_DGRAM. */
407 /* A Netlink routing socket that is not subscribed to any multicast groups. */
408 static struct nl_sock *rtnl_sock;
410 /* This is set pretty low because we probably won't learn anything from the
411 * additional log messages. */
412 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
414 static int netdev_linux_init(void);
416 static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
417 int cmd, const char *cmd_name);
418 static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
419 const char *cmd_name);
420 static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
421 int cmd, const char *cmd_name);
422 static int get_flags(const struct netdev_dev *, unsigned int *flags);
423 static int set_flags(struct netdev *, unsigned int flags);
424 static int do_get_ifindex(const char *netdev_name);
425 static int get_ifindex(const struct netdev *, int *ifindexp);
426 static int do_set_addr(struct netdev *netdev,
427 int ioctl_nr, const char *ioctl_name,
428 struct in_addr addr);
429 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
430 static int set_etheraddr(const char *netdev_name, const uint8_t[ETH_ADDR_LEN]);
431 static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
432 static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
433 static int af_packet_sock(void);
434 static void netdev_linux_miimon_run(void);
435 static void netdev_linux_miimon_wait(void);
438 is_netdev_linux_class(const struct netdev_class *netdev_class)
440 return netdev_class->init == netdev_linux_init;
443 static struct netdev_dev_linux *
444 netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
446 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
447 assert(is_netdev_linux_class(netdev_class));
449 return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
452 static struct netdev_linux *
453 netdev_linux_cast(const struct netdev *netdev)
455 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
456 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
457 assert(is_netdev_linux_class(netdev_class));
459 return CONTAINER_OF(netdev, struct netdev_linux, netdev);
463 netdev_linux_init(void)
465 static int status = -1;
467 /* Create AF_INET socket. */
468 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
469 status = af_inet_sock >= 0 ? 0 : errno;
471 VLOG_ERR("failed to create inet socket: %s", strerror(status));
474 /* Create rtnetlink socket. */
476 status = nl_sock_create(NETLINK_ROUTE, &rtnl_sock);
478 VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",
487 netdev_linux_run(void)
489 rtnetlink_link_run();
490 netdev_linux_miimon_run();
494 netdev_linux_wait(void)
496 rtnetlink_link_wait();
497 netdev_linux_miimon_wait();
501 netdev_linux_get_drvinfo(struct netdev_dev_linux *netdev_dev)
506 if (netdev_dev->cache_valid & VALID_DRVINFO) {
510 COVERAGE_INC(netdev_get_ethtool);
511 memset(&netdev_dev->drvinfo, 0, sizeof netdev_dev->drvinfo);
512 error = netdev_linux_do_ethtool(netdev_dev->netdev_dev.name,
513 (struct ethtool_cmd *)&netdev_dev->drvinfo,
517 netdev_dev->cache_valid |= VALID_DRVINFO;
523 netdev_dev_linux_changed(struct netdev_dev_linux *dev,
524 unsigned int ifi_flags,
528 if (!dev->change_seq) {
532 if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) {
533 dev->carrier_resets++;
535 dev->ifi_flags = ifi_flags;
537 dev->cache_valid &= mask;
541 netdev_dev_linux_update(struct netdev_dev_linux *dev,
542 const struct rtnetlink_link_change *change)
544 if (change->nlmsg_type == RTM_NEWLINK) {
546 netdev_dev_linux_changed(dev, change->ifi_flags, VALID_DRVINFO);
548 /* Update netdev from rtnl-change msg. */
550 dev->mtu = change->mtu;
551 dev->cache_valid |= VALID_MTU;
552 dev->netdev_mtu_error = 0;
555 if (!eth_addr_is_zero(change->addr)) {
556 memcpy(dev->etheraddr, change->addr, ETH_ADDR_LEN);
557 dev->cache_valid |= VALID_ETHERADDR;
558 dev->ether_addr_error = 0;
561 dev->ifindex = change->ifi_index;
562 dev->cache_valid |= VALID_IFINDEX;
563 dev->get_ifindex_error = 0;
566 netdev_dev_linux_changed(dev, change->ifi_flags, 0);
571 netdev_linux_cache_cb(const struct rtnetlink_link_change *change,
572 void *aux OVS_UNUSED)
574 struct netdev_dev_linux *dev;
576 struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
578 const struct netdev_class *netdev_class =
579 netdev_dev_get_class(base_dev);
581 if (is_netdev_linux_class(netdev_class)) {
582 dev = netdev_dev_linux_cast(base_dev);
583 netdev_dev_linux_update(dev, change);
587 struct shash device_shash;
588 struct shash_node *node;
590 shash_init(&device_shash);
591 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
592 SHASH_FOR_EACH (node, &device_shash) {
597 get_flags(&dev->netdev_dev, &flags);
598 netdev_dev_linux_changed(dev, flags, 0);
600 shash_destroy(&device_shash);
605 cache_notifier_ref(void)
607 if (!cache_notifier_refcount) {
608 assert(!netdev_linux_cache_notifier);
610 netdev_linux_cache_notifier =
611 rtnetlink_link_notifier_create(netdev_linux_cache_cb, NULL);
613 if (!netdev_linux_cache_notifier) {
617 cache_notifier_refcount++;
623 cache_notifier_unref(void)
625 assert(cache_notifier_refcount > 0);
626 if (!--cache_notifier_refcount) {
627 assert(netdev_linux_cache_notifier);
628 rtnetlink_link_notifier_destroy(netdev_linux_cache_notifier);
629 netdev_linux_cache_notifier = NULL;
633 /* Creates system and internal devices. */
635 netdev_linux_create(const struct netdev_class *class, const char *name,
636 struct netdev_dev **netdev_devp)
638 struct netdev_dev_linux *netdev_dev;
641 error = cache_notifier_ref();
646 netdev_dev = xzalloc(sizeof *netdev_dev);
647 netdev_dev->change_seq = 1;
648 netdev_dev_init(&netdev_dev->netdev_dev, name, class);
649 get_flags(&netdev_dev->netdev_dev, &netdev_dev->ifi_flags);
651 *netdev_devp = &netdev_dev->netdev_dev;
655 /* For most types of netdevs we open the device for each call of
656 * netdev_open(). However, this is not the case with tap devices,
657 * since it is only possible to open the device once. In this
658 * situation we share a single file descriptor, and consequently
659 * buffers, across all readers. Therefore once data is read it will
660 * be unavailable to other reads for tap devices. */
662 netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED,
663 const char *name, struct netdev_dev **netdev_devp)
665 struct netdev_dev_linux *netdev_dev;
666 struct tap_state *state;
667 static const char tap_dev[] = "/dev/net/tun";
671 netdev_dev = xzalloc(sizeof *netdev_dev);
672 state = &netdev_dev->state.tap;
674 error = cache_notifier_ref();
679 /* Open tap device. */
680 state->fd = open(tap_dev, O_RDWR);
683 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
684 goto error_unref_notifier;
687 /* Create tap device. */
688 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
689 ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
690 if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
691 VLOG_WARN("%s: creating tap device failed: %s", name,
694 goto error_unref_notifier;
697 /* Make non-blocking. */
698 error = set_nonblocking(state->fd);
700 goto error_unref_notifier;
703 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
704 *netdev_devp = &netdev_dev->netdev_dev;
707 error_unref_notifier:
708 cache_notifier_unref();
715 destroy_tap(struct netdev_dev_linux *netdev_dev)
717 struct tap_state *state = &netdev_dev->state.tap;
719 if (state->fd >= 0) {
724 /* Destroys the netdev device 'netdev_dev_'. */
726 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
728 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
729 const struct netdev_class *class = netdev_dev_get_class(netdev_dev_);
731 if (netdev_dev->tc && netdev_dev->tc->ops->tc_destroy) {
732 netdev_dev->tc->ops->tc_destroy(netdev_dev->tc);
735 if (class == &netdev_tap_class) {
736 destroy_tap(netdev_dev);
740 cache_notifier_unref();
744 netdev_linux_open(struct netdev_dev *netdev_dev_, struct netdev **netdevp)
746 struct netdev_linux *netdev;
747 enum netdev_flags flags;
750 /* Allocate network device. */
751 netdev = xzalloc(sizeof *netdev);
753 netdev_init(&netdev->netdev, netdev_dev_);
755 /* Verify that the device really exists, by attempting to read its flags.
756 * (The flags might be cached, in which case this won't actually do an
759 * Don't do this for "internal" netdevs, though, because those have to be
760 * created as netdev objects before they exist in the kernel, because
761 * creating them in the kernel happens by passing a netdev object to
762 * dpif_port_add(). */
763 if (netdev_dev_get_class(netdev_dev_) != &netdev_internal_class) {
764 error = netdev_get_flags(&netdev->netdev, &flags);
765 if (error == ENODEV) {
770 *netdevp = &netdev->netdev;
774 netdev_uninit(&netdev->netdev, true);
778 /* Closes and destroys 'netdev'. */
780 netdev_linux_close(struct netdev *netdev_)
782 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
784 if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
791 netdev_linux_listen(struct netdev *netdev_)
793 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
794 struct netdev_dev_linux *netdev_dev =
795 netdev_dev_linux_cast(netdev_get_dev(netdev_));
796 struct sockaddr_ll sll;
801 if (netdev->fd >= 0) {
805 if (!strcmp(netdev_get_type(netdev_), "tap")
806 && !netdev_dev->state.tap.opened) {
807 netdev->fd = netdev_dev->state.tap.fd;
808 netdev_dev->state.tap.opened = true;
812 /* Create file descriptor. */
813 fd = socket(PF_PACKET, SOCK_RAW, 0);
816 VLOG_ERR("failed to create raw socket (%s)", strerror(error));
820 /* Set non-blocking mode. */
821 error = set_nonblocking(fd);
826 /* Get ethernet device index. */
827 error = get_ifindex(&netdev->netdev, &ifindex);
832 /* Bind to specific ethernet device. */
833 memset(&sll, 0, sizeof sll);
834 sll.sll_family = AF_PACKET;
835 sll.sll_ifindex = ifindex;
836 sll.sll_protocol = (OVS_FORCE unsigned short int) htons(ETH_P_ALL);
837 if (bind(fd, (struct sockaddr *) &sll, sizeof sll) < 0) {
839 VLOG_ERR("%s: failed to bind raw socket (%s)",
840 netdev_get_name(netdev_), strerror(error));
855 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
857 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
859 if (netdev->fd < 0) {
860 /* Device is not listening. */
867 retval = (netdev_->netdev_dev->netdev_class == &netdev_tap_class
868 ? read(netdev->fd, data, size)
869 : recv(netdev->fd, data, size, MSG_TRUNC));
871 return retval <= size ? retval : -EMSGSIZE;
872 } else if (errno != EINTR) {
873 if (errno != EAGAIN) {
874 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
875 strerror(errno), netdev_get_name(netdev_));
882 /* Registers with the poll loop to wake up from the next call to poll_block()
883 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
885 netdev_linux_recv_wait(struct netdev *netdev_)
887 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
888 if (netdev->fd >= 0) {
889 poll_fd_wait(netdev->fd, POLLIN);
893 /* Discards all packets waiting to be received from 'netdev'. */
895 netdev_linux_drain(struct netdev *netdev_)
897 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
898 if (netdev->fd < 0) {
900 } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
902 int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
903 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
907 drain_fd(netdev->fd, ifr.ifr_qlen);
910 return drain_rcvbuf(netdev->fd);
914 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
915 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
916 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
917 * the packet is too big or too small to transmit on the device.
919 * The caller retains ownership of 'buffer' in all cases.
921 * The kernel maintains a packet transmission queue, so the caller is not
922 * expected to do additional queuing of packets. */
924 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
926 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
930 if (netdev->fd < 0) {
931 /* Use our AF_PACKET socket to send to this device. */
932 struct sockaddr_ll sll;
939 sock = af_packet_sock();
944 error = get_ifindex(netdev_, &ifindex);
949 /* We don't bother setting most fields in sockaddr_ll because the
950 * kernel ignores them for SOCK_RAW. */
951 memset(&sll, 0, sizeof sll);
952 sll.sll_family = AF_PACKET;
953 sll.sll_ifindex = ifindex;
955 iov.iov_base = CONST_CAST(void *, data);
959 msg.msg_namelen = sizeof sll;
962 msg.msg_control = NULL;
963 msg.msg_controllen = 0;
966 retval = sendmsg(sock, &msg, 0);
968 /* Use the netdev's own fd to send to this device. This is
969 * essential for tap devices, because packets sent to a tap device
970 * with an AF_PACKET socket will loop back to be *received* again
971 * on the tap device. */
972 retval = write(netdev->fd, data, size);
976 /* The Linux AF_PACKET implementation never blocks waiting for room
977 * for packets, instead returning ENOBUFS. Translate this into
978 * EAGAIN for the caller. */
979 if (errno == ENOBUFS) {
981 } else if (errno == EINTR) {
983 } else if (errno != EAGAIN) {
984 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
985 netdev_get_name(netdev_), strerror(errno));
988 } else if (retval != size) {
989 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
990 "%zu) on %s", retval, size, netdev_get_name(netdev_));
998 /* Registers with the poll loop to wake up from the next call to poll_block()
999 * when the packet transmission queue has sufficient room to transmit a packet
1000 * with netdev_send().
1002 * The kernel maintains a packet transmission queue, so the client is not
1003 * expected to do additional queuing of packets. Thus, this function is
1004 * unlikely to ever be used. It is included for completeness. */
1006 netdev_linux_send_wait(struct netdev *netdev_)
1008 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1009 if (netdev->fd < 0) {
1010 /* Nothing to do. */
1011 } else if (strcmp(netdev_get_type(netdev_), "tap")) {
1012 poll_fd_wait(netdev->fd, POLLOUT);
1014 /* TAP device always accepts packets.*/
1015 poll_immediate_wake();
1019 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
1020 * otherwise a positive errno value. */
1022 netdev_linux_set_etheraddr(struct netdev *netdev_,
1023 const uint8_t mac[ETH_ADDR_LEN])
1025 struct netdev_dev_linux *netdev_dev =
1026 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1028 bool up_again = false;
1030 if (netdev_dev->cache_valid & VALID_ETHERADDR) {
1031 if (netdev_dev->ether_addr_error) {
1032 return netdev_dev->ether_addr_error;
1034 if (eth_addr_equals(netdev_dev->etheraddr, mac)) {
1037 netdev_dev->cache_valid &= ~VALID_ETHERADDR;
1040 /* Tap devices must be brought down before setting the address. */
1041 if (!strcmp(netdev_get_type(netdev_), "tap")) {
1042 enum netdev_flags flags;
1044 if (!netdev_get_flags(netdev_, &flags) && (flags & NETDEV_UP)) {
1045 netdev_turn_flags_off(netdev_, NETDEV_UP, false);
1049 error = set_etheraddr(netdev_get_name(netdev_), mac);
1050 if (!error || error == ENODEV) {
1051 netdev_dev->ether_addr_error = error;
1052 netdev_dev->cache_valid |= VALID_ETHERADDR;
1054 memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
1059 netdev_turn_flags_on(netdev_, NETDEV_UP, false);
1065 /* Copies 'netdev''s MAC address to 'mac' which is passed as param. */
1067 netdev_linux_get_etheraddr(const struct netdev *netdev_,
1068 uint8_t mac[ETH_ADDR_LEN])
1070 struct netdev_dev_linux *netdev_dev =
1071 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1073 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
1074 int error = get_etheraddr(netdev_get_name(netdev_),
1075 netdev_dev->etheraddr);
1077 netdev_dev->ether_addr_error = error;
1078 netdev_dev->cache_valid |= VALID_ETHERADDR;
1081 if (!netdev_dev->ether_addr_error) {
1082 memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
1085 return netdev_dev->ether_addr_error;
1088 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
1089 * in bytes, not including the hardware header; thus, this is typically 1500
1090 * bytes for Ethernet devices. */
1092 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
1094 struct netdev_dev_linux *netdev_dev =
1095 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1096 if (!(netdev_dev->cache_valid & VALID_MTU)) {
1100 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1101 SIOCGIFMTU, "SIOCGIFMTU");
1103 netdev_dev->netdev_mtu_error = error;
1104 netdev_dev->mtu = ifr.ifr_mtu;
1105 netdev_dev->cache_valid |= VALID_MTU;
1108 if (!netdev_dev->netdev_mtu_error) {
1109 *mtup = netdev_dev->mtu;
1111 return netdev_dev->netdev_mtu_error;
1114 /* Sets the maximum size of transmitted (MTU) for given device using linux
1115 * networking ioctl interface.
1118 netdev_linux_set_mtu(const struct netdev *netdev_, int mtu)
1120 struct netdev_dev_linux *netdev_dev =
1121 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1125 if (netdev_dev->cache_valid & VALID_MTU) {
1126 if (netdev_dev->netdev_mtu_error) {
1127 return netdev_dev->netdev_mtu_error;
1129 if (netdev_dev->mtu == mtu) {
1132 netdev_dev->cache_valid &= ~VALID_MTU;
1135 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1136 SIOCSIFMTU, "SIOCSIFMTU");
1137 if (!error || error == ENODEV) {
1138 netdev_dev->netdev_mtu_error = error;
1139 netdev_dev->mtu = ifr.ifr_mtu;
1140 netdev_dev->cache_valid |= VALID_MTU;
1145 /* Returns the ifindex of 'netdev', if successful, as a positive number.
1146 * On failure, returns a negative errno value. */
1148 netdev_linux_get_ifindex(const struct netdev *netdev)
1152 error = get_ifindex(netdev, &ifindex);
1153 return error ? -error : ifindex;
1157 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
1159 struct netdev_dev_linux *netdev_dev =
1160 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1162 if (netdev_dev->miimon_interval > 0) {
1163 *carrier = netdev_dev->miimon;
1165 *carrier = (netdev_dev->ifi_flags & IFF_RUNNING) != 0;
1171 static long long int
1172 netdev_linux_get_carrier_resets(const struct netdev *netdev)
1174 return netdev_dev_linux_cast(netdev_get_dev(netdev))->carrier_resets;
1178 netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name,
1179 struct mii_ioctl_data *data)
1184 memset(&ifr, 0, sizeof ifr);
1185 memcpy(&ifr.ifr_data, data, sizeof *data);
1186 error = netdev_linux_do_ioctl(name, &ifr, cmd, cmd_name);
1187 memcpy(data, &ifr.ifr_data, sizeof *data);
1193 netdev_linux_get_miimon(const char *name, bool *miimon)
1195 struct mii_ioctl_data data;
1200 memset(&data, 0, sizeof data);
1201 error = netdev_linux_do_miimon(name, SIOCGMIIPHY, "SIOCGMIIPHY", &data);
1203 /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */
1204 data.reg_num = MII_BMSR;
1205 error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG",
1209 *miimon = !!(data.val_out & BMSR_LSTATUS);
1211 VLOG_WARN_RL(&rl, "%s: failed to query MII", name);
1214 struct ethtool_cmd ecmd;
1216 VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool",
1219 COVERAGE_INC(netdev_get_ethtool);
1220 memset(&ecmd, 0, sizeof ecmd);
1221 error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK,
1224 struct ethtool_value eval;
1226 memcpy(&eval, &ecmd, sizeof eval);
1227 *miimon = !!eval.data;
1229 VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name);
1237 netdev_linux_set_miimon_interval(struct netdev *netdev_,
1238 long long int interval)
1240 struct netdev_dev_linux *netdev_dev;
1242 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev_));
1244 interval = interval > 0 ? MAX(interval, 100) : 0;
1245 if (netdev_dev->miimon_interval != interval) {
1246 netdev_dev->miimon_interval = interval;
1247 timer_set_expired(&netdev_dev->miimon_timer);
1254 netdev_linux_miimon_run(void)
1256 struct shash device_shash;
1257 struct shash_node *node;
1259 shash_init(&device_shash);
1260 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
1261 SHASH_FOR_EACH (node, &device_shash) {
1262 struct netdev_dev_linux *dev = node->data;
1265 if (dev->miimon_interval <= 0 || !timer_expired(&dev->miimon_timer)) {
1269 netdev_linux_get_miimon(dev->netdev_dev.name, &miimon);
1270 if (miimon != dev->miimon) {
1271 dev->miimon = miimon;
1272 netdev_dev_linux_changed(dev, dev->ifi_flags, 0);
1275 timer_set_duration(&dev->miimon_timer, dev->miimon_interval);
1278 shash_destroy(&device_shash);
1282 netdev_linux_miimon_wait(void)
1284 struct shash device_shash;
1285 struct shash_node *node;
1287 shash_init(&device_shash);
1288 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
1289 SHASH_FOR_EACH (node, &device_shash) {
1290 struct netdev_dev_linux *dev = node->data;
1292 if (dev->miimon_interval > 0) {
1293 timer_wait(&dev->miimon_timer);
1296 shash_destroy(&device_shash);
1299 /* Check whether we can we use RTM_GETLINK to get network device statistics.
1300 * In pre-2.6.19 kernels, this was only available if wireless extensions were
1303 check_for_working_netlink_stats(void)
1305 /* Decide on the netdev_get_stats() implementation to use. Netlink is
1306 * preferable, so if that works, we'll use it. */
1307 int ifindex = do_get_ifindex("lo");
1309 VLOG_WARN("failed to get ifindex for lo, "
1310 "obtaining netdev stats from proc");
1313 struct netdev_stats stats;
1314 int error = get_stats_via_netlink(ifindex, &stats);
1316 VLOG_DBG("obtaining netdev stats via rtnetlink");
1319 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
1320 "via proc (you are probably running a pre-2.6.19 "
1321 "kernel)", strerror(error));
1328 swap_uint64(uint64_t *a, uint64_t *b)
1336 get_stats_via_vport(const struct netdev *netdev_,
1337 struct netdev_stats *stats)
1339 struct netdev_dev_linux *netdev_dev =
1340 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1342 if (!netdev_dev->vport_stats_error ||
1343 !(netdev_dev->cache_valid & VALID_VPORT_STAT_ERROR)) {
1346 error = netdev_vport_get_stats(netdev_, stats);
1347 if (error && error != ENOENT) {
1348 VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed "
1349 "(%s)", netdev_get_name(netdev_), strerror(error));
1351 netdev_dev->vport_stats_error = error;
1352 netdev_dev->cache_valid |= VALID_VPORT_STAT_ERROR;
1357 netdev_linux_sys_get_stats(const struct netdev *netdev_,
1358 struct netdev_stats *stats)
1360 static int use_netlink_stats = -1;
1363 if (use_netlink_stats < 0) {
1364 use_netlink_stats = check_for_working_netlink_stats();
1367 if (use_netlink_stats) {
1370 error = get_ifindex(netdev_, &ifindex);
1372 error = get_stats_via_netlink(ifindex, stats);
1375 error = get_stats_via_proc(netdev_get_name(netdev_), stats);
1379 VLOG_WARN_RL(&rl, "%s: linux-sys get stats failed %d",
1380 netdev_get_name(netdev_), error);
1386 /* Retrieves current device stats for 'netdev-linux'. */
1388 netdev_linux_get_stats(const struct netdev *netdev_,
1389 struct netdev_stats *stats)
1391 struct netdev_dev_linux *netdev_dev =
1392 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1393 struct netdev_stats dev_stats;
1396 get_stats_via_vport(netdev_, stats);
1398 error = netdev_linux_sys_get_stats(netdev_, &dev_stats);
1401 if (netdev_dev->vport_stats_error) {
1408 if (netdev_dev->vport_stats_error) {
1409 /* stats not available from OVS then use ioctl stats. */
1412 stats->rx_errors += dev_stats.rx_errors;
1413 stats->tx_errors += dev_stats.tx_errors;
1414 stats->rx_dropped += dev_stats.rx_dropped;
1415 stats->tx_dropped += dev_stats.tx_dropped;
1416 stats->multicast += dev_stats.multicast;
1417 stats->collisions += dev_stats.collisions;
1418 stats->rx_length_errors += dev_stats.rx_length_errors;
1419 stats->rx_over_errors += dev_stats.rx_over_errors;
1420 stats->rx_crc_errors += dev_stats.rx_crc_errors;
1421 stats->rx_frame_errors += dev_stats.rx_frame_errors;
1422 stats->rx_fifo_errors += dev_stats.rx_fifo_errors;
1423 stats->rx_missed_errors += dev_stats.rx_missed_errors;
1424 stats->tx_aborted_errors += dev_stats.tx_aborted_errors;
1425 stats->tx_carrier_errors += dev_stats.tx_carrier_errors;
1426 stats->tx_fifo_errors += dev_stats.tx_fifo_errors;
1427 stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors;
1428 stats->tx_window_errors += dev_stats.tx_window_errors;
1433 /* Retrieves current device stats for 'netdev-tap' netdev or
1434 * netdev-internal. */
1436 netdev_tap_get_stats(const struct netdev *netdev_,
1437 struct netdev_stats *stats)
1439 struct netdev_dev_linux *netdev_dev =
1440 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1441 struct netdev_stats dev_stats;
1444 get_stats_via_vport(netdev_, stats);
1446 error = netdev_linux_sys_get_stats(netdev_, &dev_stats);
1448 if (netdev_dev->vport_stats_error) {
1455 /* If this port is an internal port then the transmit and receive stats
1456 * will appear to be swapped relative to the other ports since we are the
1457 * one sending the data, not a remote computer. For consistency, we swap
1458 * them back here. This does not apply if we are getting stats from the
1459 * vport layer because it always tracks stats from the perspective of the
1461 if (netdev_dev->vport_stats_error) {
1463 swap_uint64(&stats->rx_packets, &stats->tx_packets);
1464 swap_uint64(&stats->rx_bytes, &stats->tx_bytes);
1465 swap_uint64(&stats->rx_errors, &stats->tx_errors);
1466 swap_uint64(&stats->rx_dropped, &stats->tx_dropped);
1467 stats->rx_length_errors = 0;
1468 stats->rx_over_errors = 0;
1469 stats->rx_crc_errors = 0;
1470 stats->rx_frame_errors = 0;
1471 stats->rx_fifo_errors = 0;
1472 stats->rx_missed_errors = 0;
1473 stats->tx_aborted_errors = 0;
1474 stats->tx_carrier_errors = 0;
1475 stats->tx_fifo_errors = 0;
1476 stats->tx_heartbeat_errors = 0;
1477 stats->tx_window_errors = 0;
1479 stats->rx_dropped += dev_stats.tx_dropped;
1480 stats->tx_dropped += dev_stats.rx_dropped;
1482 stats->rx_errors += dev_stats.tx_errors;
1483 stats->tx_errors += dev_stats.rx_errors;
1485 stats->multicast += dev_stats.multicast;
1486 stats->collisions += dev_stats.collisions;
1492 netdev_internal_get_stats(const struct netdev *netdev_,
1493 struct netdev_stats *stats)
1495 struct netdev_dev_linux *netdev_dev =
1496 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1498 get_stats_via_vport(netdev_, stats);
1499 return netdev_dev->vport_stats_error;
1503 netdev_linux_read_features(struct netdev_dev_linux *netdev_dev)
1505 struct ethtool_cmd ecmd;
1509 if (netdev_dev->cache_valid & VALID_FEATURES) {
1513 COVERAGE_INC(netdev_get_ethtool);
1514 memset(&ecmd, 0, sizeof ecmd);
1515 error = netdev_linux_do_ethtool(netdev_dev->netdev_dev.name, &ecmd,
1516 ETHTOOL_GSET, "ETHTOOL_GSET");
1521 /* Supported features. */
1522 netdev_dev->supported = 0;
1523 if (ecmd.supported & SUPPORTED_10baseT_Half) {
1524 netdev_dev->supported |= NETDEV_F_10MB_HD;
1526 if (ecmd.supported & SUPPORTED_10baseT_Full) {
1527 netdev_dev->supported |= NETDEV_F_10MB_FD;
1529 if (ecmd.supported & SUPPORTED_100baseT_Half) {
1530 netdev_dev->supported |= NETDEV_F_100MB_HD;
1532 if (ecmd.supported & SUPPORTED_100baseT_Full) {
1533 netdev_dev->supported |= NETDEV_F_100MB_FD;
1535 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
1536 netdev_dev->supported |= NETDEV_F_1GB_HD;
1538 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
1539 netdev_dev->supported |= NETDEV_F_1GB_FD;
1541 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
1542 netdev_dev->supported |= NETDEV_F_10GB_FD;
1544 if (ecmd.supported & SUPPORTED_TP) {
1545 netdev_dev->supported |= NETDEV_F_COPPER;
1547 if (ecmd.supported & SUPPORTED_FIBRE) {
1548 netdev_dev->supported |= NETDEV_F_FIBER;
1550 if (ecmd.supported & SUPPORTED_Autoneg) {
1551 netdev_dev->supported |= NETDEV_F_AUTONEG;
1553 if (ecmd.supported & SUPPORTED_Pause) {
1554 netdev_dev->supported |= NETDEV_F_PAUSE;
1556 if (ecmd.supported & SUPPORTED_Asym_Pause) {
1557 netdev_dev->supported |= NETDEV_F_PAUSE_ASYM;
1560 /* Advertised features. */
1561 netdev_dev->advertised = 0;
1562 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
1563 netdev_dev->advertised |= NETDEV_F_10MB_HD;
1565 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
1566 netdev_dev->advertised |= NETDEV_F_10MB_FD;
1568 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
1569 netdev_dev->advertised |= NETDEV_F_100MB_HD;
1571 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
1572 netdev_dev->advertised |= NETDEV_F_100MB_FD;
1574 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
1575 netdev_dev->advertised |= NETDEV_F_1GB_HD;
1577 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
1578 netdev_dev->advertised |= NETDEV_F_1GB_FD;
1580 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
1581 netdev_dev->advertised |= NETDEV_F_10GB_FD;
1583 if (ecmd.advertising & ADVERTISED_TP) {
1584 netdev_dev->advertised |= NETDEV_F_COPPER;
1586 if (ecmd.advertising & ADVERTISED_FIBRE) {
1587 netdev_dev->advertised |= NETDEV_F_FIBER;
1589 if (ecmd.advertising & ADVERTISED_Autoneg) {
1590 netdev_dev->advertised |= NETDEV_F_AUTONEG;
1592 if (ecmd.advertising & ADVERTISED_Pause) {
1593 netdev_dev->advertised |= NETDEV_F_PAUSE;
1595 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
1596 netdev_dev->advertised |= NETDEV_F_PAUSE_ASYM;
1599 /* Current settings. */
1601 if (speed == SPEED_10) {
1602 netdev_dev->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD;
1603 } else if (speed == SPEED_100) {
1604 netdev_dev->current = ecmd.duplex ? NETDEV_F_100MB_FD : NETDEV_F_100MB_HD;
1605 } else if (speed == SPEED_1000) {
1606 netdev_dev->current = ecmd.duplex ? NETDEV_F_1GB_FD : NETDEV_F_1GB_HD;
1607 } else if (speed == SPEED_10000) {
1608 netdev_dev->current = NETDEV_F_10GB_FD;
1609 } else if (speed == 40000) {
1610 netdev_dev->current = NETDEV_F_40GB_FD;
1611 } else if (speed == 100000) {
1612 netdev_dev->current = NETDEV_F_100GB_FD;
1613 } else if (speed == 1000000) {
1614 netdev_dev->current = NETDEV_F_1TB_FD;
1616 netdev_dev->current = 0;
1619 if (ecmd.port == PORT_TP) {
1620 netdev_dev->current |= NETDEV_F_COPPER;
1621 } else if (ecmd.port == PORT_FIBRE) {
1622 netdev_dev->current |= NETDEV_F_FIBER;
1626 netdev_dev->current |= NETDEV_F_AUTONEG;
1629 /* Peer advertisements. */
1630 netdev_dev->peer = 0; /* XXX */
1633 netdev_dev->cache_valid |= VALID_FEATURES;
1634 netdev_dev->get_features_error = error;
1637 /* Stores the features supported by 'netdev' into each of '*current',
1638 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
1639 * bitmap of NETDEV_* bits. Returns 0 if successful, otherwise a positive
1642 netdev_linux_get_features(const struct netdev *netdev_,
1643 enum netdev_features *current,
1644 enum netdev_features *advertised,
1645 enum netdev_features *supported,
1646 enum netdev_features *peer)
1648 struct netdev_dev_linux *netdev_dev =
1649 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1651 netdev_linux_read_features(netdev_dev);
1653 if (!netdev_dev->get_features_error) {
1654 *current = netdev_dev->current;
1655 *advertised = netdev_dev->advertised;
1656 *supported = netdev_dev->supported;
1657 *peer = netdev_dev->peer;
1659 return netdev_dev->get_features_error;
1662 /* Set the features advertised by 'netdev' to 'advertise'. */
1664 netdev_linux_set_advertisements(struct netdev *netdev,
1665 enum netdev_features advertise)
1667 struct ethtool_cmd ecmd;
1670 COVERAGE_INC(netdev_get_ethtool);
1671 memset(&ecmd, 0, sizeof ecmd);
1672 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1673 ETHTOOL_GSET, "ETHTOOL_GSET");
1678 ecmd.advertising = 0;
1679 if (advertise & NETDEV_F_10MB_HD) {
1680 ecmd.advertising |= ADVERTISED_10baseT_Half;
1682 if (advertise & NETDEV_F_10MB_FD) {
1683 ecmd.advertising |= ADVERTISED_10baseT_Full;
1685 if (advertise & NETDEV_F_100MB_HD) {
1686 ecmd.advertising |= ADVERTISED_100baseT_Half;
1688 if (advertise & NETDEV_F_100MB_FD) {
1689 ecmd.advertising |= ADVERTISED_100baseT_Full;
1691 if (advertise & NETDEV_F_1GB_HD) {
1692 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1694 if (advertise & NETDEV_F_1GB_FD) {
1695 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1697 if (advertise & NETDEV_F_10GB_FD) {
1698 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1700 if (advertise & NETDEV_F_COPPER) {
1701 ecmd.advertising |= ADVERTISED_TP;
1703 if (advertise & NETDEV_F_FIBER) {
1704 ecmd.advertising |= ADVERTISED_FIBRE;
1706 if (advertise & NETDEV_F_AUTONEG) {
1707 ecmd.advertising |= ADVERTISED_Autoneg;
1709 if (advertise & NETDEV_F_PAUSE) {
1710 ecmd.advertising |= ADVERTISED_Pause;
1712 if (advertise & NETDEV_F_PAUSE_ASYM) {
1713 ecmd.advertising |= ADVERTISED_Asym_Pause;
1715 COVERAGE_INC(netdev_set_ethtool);
1716 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1717 ETHTOOL_SSET, "ETHTOOL_SSET");
1720 /* Attempts to set input rate limiting (policing) policy. Returns 0 if
1721 * successful, otherwise a positive errno value. */
1723 netdev_linux_set_policing(struct netdev *netdev,
1724 uint32_t kbits_rate, uint32_t kbits_burst)
1726 struct netdev_dev_linux *netdev_dev =
1727 netdev_dev_linux_cast(netdev_get_dev(netdev));
1728 const char *netdev_name = netdev_get_name(netdev);
1732 kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */
1733 : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */
1734 : kbits_burst); /* Stick with user-specified value. */
1736 if (netdev_dev->cache_valid & VALID_POLICING) {
1737 if (netdev_dev->netdev_policing_error) {
1738 return netdev_dev->netdev_policing_error;
1741 if (netdev_dev->kbits_rate == kbits_rate &&
1742 netdev_dev->kbits_burst == kbits_burst) {
1743 /* Assume that settings haven't changed since we last set them. */
1746 netdev_dev->cache_valid &= ~VALID_POLICING;
1749 COVERAGE_INC(netdev_set_policing);
1750 /* Remove any existing ingress qdisc. */
1751 error = tc_add_del_ingress_qdisc(netdev, false);
1753 VLOG_WARN_RL(&rl, "%s: removing policing failed: %s",
1754 netdev_name, strerror(error));
1759 error = tc_add_del_ingress_qdisc(netdev, true);
1761 VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s",
1762 netdev_name, strerror(error));
1766 error = tc_add_policer(netdev, kbits_rate, kbits_burst);
1768 VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s",
1769 netdev_name, strerror(error));
1774 netdev_dev->kbits_rate = kbits_rate;
1775 netdev_dev->kbits_burst = kbits_burst;
1778 if (!error || error == ENODEV) {
1779 netdev_dev->netdev_policing_error = error;
1780 netdev_dev->cache_valid |= VALID_POLICING;
1786 netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED,
1789 const struct tc_ops **opsp;
1791 for (opsp = tcs; *opsp != NULL; opsp++) {
1792 const struct tc_ops *ops = *opsp;
1793 if (ops->tc_install && ops->ovs_name[0] != '\0') {
1794 sset_add(types, ops->ovs_name);
1800 static const struct tc_ops *
1801 tc_lookup_ovs_name(const char *name)
1803 const struct tc_ops **opsp;
1805 for (opsp = tcs; *opsp != NULL; opsp++) {
1806 const struct tc_ops *ops = *opsp;
1807 if (!strcmp(name, ops->ovs_name)) {
1814 static const struct tc_ops *
1815 tc_lookup_linux_name(const char *name)
1817 const struct tc_ops **opsp;
1819 for (opsp = tcs; *opsp != NULL; opsp++) {
1820 const struct tc_ops *ops = *opsp;
1821 if (ops->linux_name && !strcmp(name, ops->linux_name)) {
1828 static struct tc_queue *
1829 tc_find_queue__(const struct netdev *netdev, unsigned int queue_id,
1832 struct netdev_dev_linux *netdev_dev =
1833 netdev_dev_linux_cast(netdev_get_dev(netdev));
1834 struct tc_queue *queue;
1836 HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev_dev->tc->queues) {
1837 if (queue->queue_id == queue_id) {
1844 static struct tc_queue *
1845 tc_find_queue(const struct netdev *netdev, unsigned int queue_id)
1847 return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0));
1851 netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED,
1853 struct netdev_qos_capabilities *caps)
1855 const struct tc_ops *ops = tc_lookup_ovs_name(type);
1859 caps->n_queues = ops->n_queues;
1864 netdev_linux_get_qos(const struct netdev *netdev,
1865 const char **typep, struct smap *details)
1867 struct netdev_dev_linux *netdev_dev =
1868 netdev_dev_linux_cast(netdev_get_dev(netdev));
1871 error = tc_query_qdisc(netdev);
1876 *typep = netdev_dev->tc->ops->ovs_name;
1877 return (netdev_dev->tc->ops->qdisc_get
1878 ? netdev_dev->tc->ops->qdisc_get(netdev, details)
1883 netdev_linux_set_qos(struct netdev *netdev,
1884 const char *type, const struct smap *details)
1886 struct netdev_dev_linux *netdev_dev =
1887 netdev_dev_linux_cast(netdev_get_dev(netdev));
1888 const struct tc_ops *new_ops;
1891 new_ops = tc_lookup_ovs_name(type);
1892 if (!new_ops || !new_ops->tc_install) {
1896 error = tc_query_qdisc(netdev);
1901 if (new_ops == netdev_dev->tc->ops) {
1902 return new_ops->qdisc_set ? new_ops->qdisc_set(netdev, details) : 0;
1904 /* Delete existing qdisc. */
1905 error = tc_del_qdisc(netdev);
1909 assert(netdev_dev->tc == NULL);
1911 /* Install new qdisc. */
1912 error = new_ops->tc_install(netdev, details);
1913 assert((error == 0) == (netdev_dev->tc != NULL));
1920 netdev_linux_get_queue(const struct netdev *netdev,
1921 unsigned int queue_id, struct smap *details)
1923 struct netdev_dev_linux *netdev_dev =
1924 netdev_dev_linux_cast(netdev_get_dev(netdev));
1927 error = tc_query_qdisc(netdev);
1931 struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1933 ? netdev_dev->tc->ops->class_get(netdev, queue, details)
1939 netdev_linux_set_queue(struct netdev *netdev,
1940 unsigned int queue_id, const struct smap *details)
1942 struct netdev_dev_linux *netdev_dev =
1943 netdev_dev_linux_cast(netdev_get_dev(netdev));
1946 error = tc_query_qdisc(netdev);
1949 } else if (queue_id >= netdev_dev->tc->ops->n_queues
1950 || !netdev_dev->tc->ops->class_set) {
1954 return netdev_dev->tc->ops->class_set(netdev, queue_id, details);
1958 netdev_linux_delete_queue(struct netdev *netdev, unsigned int queue_id)
1960 struct netdev_dev_linux *netdev_dev =
1961 netdev_dev_linux_cast(netdev_get_dev(netdev));
1964 error = tc_query_qdisc(netdev);
1967 } else if (!netdev_dev->tc->ops->class_delete) {
1970 struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1972 ? netdev_dev->tc->ops->class_delete(netdev, queue)
1978 netdev_linux_get_queue_stats(const struct netdev *netdev,
1979 unsigned int queue_id,
1980 struct netdev_queue_stats *stats)
1982 struct netdev_dev_linux *netdev_dev =
1983 netdev_dev_linux_cast(netdev_get_dev(netdev));
1986 error = tc_query_qdisc(netdev);
1989 } else if (!netdev_dev->tc->ops->class_get_stats) {
1992 const struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1994 ? netdev_dev->tc->ops->class_get_stats(netdev, queue, stats)
2000 start_queue_dump(const struct netdev *netdev, struct nl_dump *dump)
2002 struct ofpbuf request;
2003 struct tcmsg *tcmsg;
2005 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request);
2009 tcmsg->tcm_parent = 0;
2010 nl_dump_start(dump, rtnl_sock, &request);
2011 ofpbuf_uninit(&request);
2016 netdev_linux_dump_queues(const struct netdev *netdev,
2017 netdev_dump_queues_cb *cb, void *aux)
2019 struct netdev_dev_linux *netdev_dev =
2020 netdev_dev_linux_cast(netdev_get_dev(netdev));
2021 struct tc_queue *queue, *next_queue;
2022 struct smap details;
2026 error = tc_query_qdisc(netdev);
2029 } else if (!netdev_dev->tc->ops->class_get) {
2034 smap_init(&details);
2035 HMAP_FOR_EACH_SAFE (queue, next_queue, hmap_node,
2036 &netdev_dev->tc->queues) {
2037 smap_clear(&details);
2039 error = netdev_dev->tc->ops->class_get(netdev, queue, &details);
2041 (*cb)(queue->queue_id, &details, aux);
2046 smap_destroy(&details);
2052 netdev_linux_dump_queue_stats(const struct netdev *netdev,
2053 netdev_dump_queue_stats_cb *cb, void *aux)
2055 struct netdev_dev_linux *netdev_dev =
2056 netdev_dev_linux_cast(netdev_get_dev(netdev));
2057 struct nl_dump dump;
2062 error = tc_query_qdisc(netdev);
2065 } else if (!netdev_dev->tc->ops->class_dump_stats) {
2070 if (!start_queue_dump(netdev, &dump)) {
2073 while (nl_dump_next(&dump, &msg)) {
2074 error = netdev_dev->tc->ops->class_dump_stats(netdev, &msg, cb, aux);
2080 error = nl_dump_done(&dump);
2081 return error ? error : last_error;
2085 netdev_linux_get_in4(const struct netdev *netdev_,
2086 struct in_addr *address, struct in_addr *netmask)
2088 struct netdev_dev_linux *netdev_dev =
2089 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2091 if (!(netdev_dev->cache_valid & VALID_IN4)) {
2094 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
2095 SIOCGIFADDR, "SIOCGIFADDR");
2100 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
2101 SIOCGIFNETMASK, "SIOCGIFNETMASK");
2106 netdev_dev->cache_valid |= VALID_IN4;
2108 *address = netdev_dev->address;
2109 *netmask = netdev_dev->netmask;
2110 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
2114 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
2115 struct in_addr netmask)
2117 struct netdev_dev_linux *netdev_dev =
2118 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2121 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
2123 netdev_dev->cache_valid |= VALID_IN4;
2124 netdev_dev->address = address;
2125 netdev_dev->netmask = netmask;
2126 if (address.s_addr != INADDR_ANY) {
2127 error = do_set_addr(netdev_, SIOCSIFNETMASK,
2128 "SIOCSIFNETMASK", netmask);
2135 parse_if_inet6_line(const char *line,
2136 struct in6_addr *in6, char ifname[16 + 1])
2138 uint8_t *s6 = in6->s6_addr;
2139 #define X8 "%2"SCNx8
2141 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
2142 "%*x %*x %*x %*x %16s\n",
2143 &s6[0], &s6[1], &s6[2], &s6[3],
2144 &s6[4], &s6[5], &s6[6], &s6[7],
2145 &s6[8], &s6[9], &s6[10], &s6[11],
2146 &s6[12], &s6[13], &s6[14], &s6[15],
2150 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
2151 * 'in6' is non-null) and returns true. Otherwise, returns false. */
2153 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
2155 struct netdev_dev_linux *netdev_dev =
2156 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2157 if (!(netdev_dev->cache_valid & VALID_IN6)) {
2161 netdev_dev->in6 = in6addr_any;
2163 file = fopen("/proc/net/if_inet6", "r");
2165 const char *name = netdev_get_name(netdev_);
2166 while (fgets(line, sizeof line, file)) {
2167 struct in6_addr in6_tmp;
2168 char ifname[16 + 1];
2169 if (parse_if_inet6_line(line, &in6_tmp, ifname)
2170 && !strcmp(name, ifname))
2172 netdev_dev->in6 = in6_tmp;
2178 netdev_dev->cache_valid |= VALID_IN6;
2180 *in6 = netdev_dev->in6;
2185 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
2187 struct sockaddr_in sin;
2188 memset(&sin, 0, sizeof sin);
2189 sin.sin_family = AF_INET;
2190 sin.sin_addr = addr;
2193 memset(sa, 0, sizeof *sa);
2194 memcpy(sa, &sin, sizeof sin);
2198 do_set_addr(struct netdev *netdev,
2199 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
2202 ovs_strzcpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
2203 make_in4_sockaddr(&ifr.ifr_addr, addr);
2205 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
2209 /* Adds 'router' as a default IP gateway. */
2211 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
2213 struct in_addr any = { INADDR_ANY };
2217 memset(&rt, 0, sizeof rt);
2218 make_in4_sockaddr(&rt.rt_dst, any);
2219 make_in4_sockaddr(&rt.rt_gateway, router);
2220 make_in4_sockaddr(&rt.rt_genmask, any);
2221 rt.rt_flags = RTF_UP | RTF_GATEWAY;
2222 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
2224 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
2230 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
2233 static const char fn[] = "/proc/net/route";
2238 *netdev_name = NULL;
2239 stream = fopen(fn, "r");
2240 if (stream == NULL) {
2241 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
2246 while (fgets(line, sizeof line, stream)) {
2249 ovs_be32 dest, gateway, mask;
2250 int refcnt, metric, mtu;
2251 unsigned int flags, use, window, irtt;
2254 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
2256 iface, &dest, &gateway, &flags, &refcnt,
2257 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
2259 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
2263 if (!(flags & RTF_UP)) {
2264 /* Skip routes that aren't up. */
2268 /* The output of 'dest', 'mask', and 'gateway' were given in
2269 * network byte order, so we don't need need any endian
2270 * conversions here. */
2271 if ((dest & mask) == (host->s_addr & mask)) {
2273 /* The host is directly reachable. */
2274 next_hop->s_addr = 0;
2276 /* To reach the host, we must go through a gateway. */
2277 next_hop->s_addr = gateway;
2279 *netdev_name = xstrdup(iface);
2291 netdev_linux_get_drv_info(const struct netdev *netdev, struct smap *smap)
2294 struct netdev_dev_linux *netdev_dev =
2295 netdev_dev_linux_cast(netdev_get_dev(netdev));
2297 error = netdev_linux_get_drvinfo(netdev_dev);
2299 smap_add(smap, "driver_name", netdev_dev->drvinfo.driver);
2300 smap_add(smap, "driver_version", netdev_dev->drvinfo.version);
2301 smap_add(smap, "firmware_version", netdev_dev->drvinfo.fw_version);
2307 netdev_internal_get_drv_info(const struct netdev *netdev OVS_UNUSED,
2310 smap_add(smap, "driver_name", "openvswitch");
2314 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
2315 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
2316 * returns 0. Otherwise, it returns a positive errno value; in particular,
2317 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
2319 netdev_linux_arp_lookup(const struct netdev *netdev,
2320 ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
2323 struct sockaddr_in sin;
2326 memset(&r, 0, sizeof r);
2327 memset(&sin, 0, sizeof sin);
2328 sin.sin_family = AF_INET;
2329 sin.sin_addr.s_addr = ip;
2331 memcpy(&r.arp_pa, &sin, sizeof sin);
2332 r.arp_ha.sa_family = ARPHRD_ETHER;
2334 ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
2335 COVERAGE_INC(netdev_arp_lookup);
2336 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
2338 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
2339 } else if (retval != ENXIO) {
2340 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
2341 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
2347 nd_to_iff_flags(enum netdev_flags nd)
2350 if (nd & NETDEV_UP) {
2353 if (nd & NETDEV_PROMISC) {
2360 iff_to_nd_flags(int iff)
2362 enum netdev_flags nd = 0;
2366 if (iff & IFF_PROMISC) {
2367 nd |= NETDEV_PROMISC;
2373 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
2374 enum netdev_flags on, enum netdev_flags *old_flagsp)
2376 struct netdev_dev_linux *netdev_dev;
2377 int old_flags, new_flags;
2380 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
2381 old_flags = netdev_dev->ifi_flags;
2382 *old_flagsp = iff_to_nd_flags(old_flags);
2383 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
2384 if (new_flags != old_flags) {
2385 error = set_flags(netdev, new_flags);
2386 get_flags(&netdev_dev->netdev_dev, &netdev_dev->ifi_flags);
2392 netdev_linux_change_seq(const struct netdev *netdev)
2394 return netdev_dev_linux_cast(netdev_get_dev(netdev))->change_seq;
2397 #define NETDEV_LINUX_CLASS(NAME, CREATE, GET_STATS, SET_STATS, \
2398 GET_FEATURES, GET_STATUS) \
2402 netdev_linux_init, \
2404 netdev_linux_wait, \
2407 netdev_linux_destroy, \
2408 NULL, /* get_config */ \
2409 NULL, /* set_config */ \
2411 netdev_linux_open, \
2412 netdev_linux_close, \
2414 netdev_linux_listen, \
2415 netdev_linux_recv, \
2416 netdev_linux_recv_wait, \
2417 netdev_linux_drain, \
2419 netdev_linux_send, \
2420 netdev_linux_send_wait, \
2422 netdev_linux_set_etheraddr, \
2423 netdev_linux_get_etheraddr, \
2424 netdev_linux_get_mtu, \
2425 netdev_linux_set_mtu, \
2426 netdev_linux_get_ifindex, \
2427 netdev_linux_get_carrier, \
2428 netdev_linux_get_carrier_resets, \
2429 netdev_linux_set_miimon_interval, \
2434 netdev_linux_set_advertisements, \
2436 netdev_linux_set_policing, \
2437 netdev_linux_get_qos_types, \
2438 netdev_linux_get_qos_capabilities, \
2439 netdev_linux_get_qos, \
2440 netdev_linux_set_qos, \
2441 netdev_linux_get_queue, \
2442 netdev_linux_set_queue, \
2443 netdev_linux_delete_queue, \
2444 netdev_linux_get_queue_stats, \
2445 netdev_linux_dump_queues, \
2446 netdev_linux_dump_queue_stats, \
2448 netdev_linux_get_in4, \
2449 netdev_linux_set_in4, \
2450 netdev_linux_get_in6, \
2451 netdev_linux_add_router, \
2452 netdev_linux_get_next_hop, \
2454 netdev_linux_arp_lookup, \
2456 netdev_linux_update_flags, \
2458 netdev_linux_change_seq \
2461 const struct netdev_class netdev_linux_class =
2464 netdev_linux_create,
2465 netdev_linux_get_stats,
2466 NULL, /* set_stats */
2467 netdev_linux_get_features,
2468 netdev_linux_get_drv_info);
2470 const struct netdev_class netdev_tap_class =
2473 netdev_linux_create_tap,
2474 netdev_tap_get_stats,
2475 NULL, /* set_stats */
2476 netdev_linux_get_features,
2477 netdev_linux_get_drv_info);
2479 const struct netdev_class netdev_internal_class =
2482 netdev_linux_create,
2483 netdev_internal_get_stats,
2484 netdev_vport_set_stats,
2485 NULL, /* get_features */
2486 netdev_internal_get_drv_info);
2488 /* HTB traffic control class. */
2490 #define HTB_N_QUEUES 0xf000
2494 unsigned int max_rate; /* In bytes/s. */
2498 struct tc_queue tc_queue;
2499 unsigned int min_rate; /* In bytes/s. */
2500 unsigned int max_rate; /* In bytes/s. */
2501 unsigned int burst; /* In bytes. */
2502 unsigned int priority; /* Lower values are higher priorities. */
2506 htb_get__(const struct netdev *netdev)
2508 struct netdev_dev_linux *netdev_dev =
2509 netdev_dev_linux_cast(netdev_get_dev(netdev));
2510 return CONTAINER_OF(netdev_dev->tc, struct htb, tc);
2514 htb_install__(struct netdev *netdev, uint64_t max_rate)
2516 struct netdev_dev_linux *netdev_dev =
2517 netdev_dev_linux_cast(netdev_get_dev(netdev));
2520 htb = xmalloc(sizeof *htb);
2521 tc_init(&htb->tc, &tc_ops_htb);
2522 htb->max_rate = max_rate;
2524 netdev_dev->tc = &htb->tc;
2527 /* Create an HTB qdisc.
2529 * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */
2531 htb_setup_qdisc__(struct netdev *netdev)
2534 struct tc_htb_glob opt;
2535 struct ofpbuf request;
2536 struct tcmsg *tcmsg;
2538 tc_del_qdisc(netdev);
2540 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
2541 NLM_F_EXCL | NLM_F_CREATE, &request);
2545 tcmsg->tcm_handle = tc_make_handle(1, 0);
2546 tcmsg->tcm_parent = TC_H_ROOT;
2548 nl_msg_put_string(&request, TCA_KIND, "htb");
2550 memset(&opt, 0, sizeof opt);
2551 opt.rate2quantum = 10;
2555 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2556 nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt);
2557 nl_msg_end_nested(&request, opt_offset);
2559 return tc_transact(&request, NULL);
2562 /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb
2563 * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */
2565 htb_setup_class__(struct netdev *netdev, unsigned int handle,
2566 unsigned int parent, struct htb_class *class)
2569 struct tc_htb_opt opt;
2570 struct ofpbuf request;
2571 struct tcmsg *tcmsg;
2575 error = netdev_get_mtu(netdev, &mtu);
2577 VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU",
2578 netdev_get_name(netdev));
2582 memset(&opt, 0, sizeof opt);
2583 tc_fill_rate(&opt.rate, class->min_rate, mtu);
2584 tc_fill_rate(&opt.ceil, class->max_rate, mtu);
2585 opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst);
2586 opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst);
2587 opt.prio = class->priority;
2589 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
2593 tcmsg->tcm_handle = handle;
2594 tcmsg->tcm_parent = parent;
2596 nl_msg_put_string(&request, TCA_KIND, "htb");
2597 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2598 nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt);
2599 tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate);
2600 tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil);
2601 nl_msg_end_nested(&request, opt_offset);
2603 error = tc_transact(&request, NULL);
2605 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
2606 "min_rate=%u max_rate=%u burst=%u prio=%u (%s)",
2607 netdev_get_name(netdev),
2608 tc_get_major(handle), tc_get_minor(handle),
2609 tc_get_major(parent), tc_get_minor(parent),
2610 class->min_rate, class->max_rate,
2611 class->burst, class->priority, strerror(error));
2616 /* Parses Netlink attributes in 'options' for HTB parameters and stores a
2617 * description of them into 'details'. The description complies with the
2618 * specification given in the vswitch database documentation for linux-htb
2621 htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class)
2623 static const struct nl_policy tca_htb_policy[] = {
2624 [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false,
2625 .min_len = sizeof(struct tc_htb_opt) },
2628 struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)];
2629 const struct tc_htb_opt *htb;
2631 if (!nl_parse_nested(nl_options, tca_htb_policy,
2632 attrs, ARRAY_SIZE(tca_htb_policy))) {
2633 VLOG_WARN_RL(&rl, "failed to parse HTB class options");
2637 htb = nl_attr_get(attrs[TCA_HTB_PARMS]);
2638 class->min_rate = htb->rate.rate;
2639 class->max_rate = htb->ceil.rate;
2640 class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer);
2641 class->priority = htb->prio;
2646 htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
2647 struct htb_class *options,
2648 struct netdev_queue_stats *stats)
2650 struct nlattr *nl_options;
2651 unsigned int handle;
2654 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
2655 if (!error && queue_id) {
2656 unsigned int major = tc_get_major(handle);
2657 unsigned int minor = tc_get_minor(handle);
2658 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2659 *queue_id = minor - 1;
2664 if (!error && options) {
2665 error = htb_parse_tca_options__(nl_options, options);
2671 htb_parse_qdisc_details__(struct netdev *netdev,
2672 const struct smap *details, struct htb_class *hc)
2674 const char *max_rate_s;
2676 max_rate_s = smap_get(details, "max-rate");
2677 hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
2678 if (!hc->max_rate) {
2679 enum netdev_features current;
2681 netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
2682 hc->max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8;
2684 hc->min_rate = hc->max_rate;
2690 htb_parse_class_details__(struct netdev *netdev,
2691 const struct smap *details, struct htb_class *hc)
2693 const struct htb *htb = htb_get__(netdev);
2694 const char *min_rate_s = smap_get(details, "min-rate");
2695 const char *max_rate_s = smap_get(details, "max-rate");
2696 const char *burst_s = smap_get(details, "burst");
2697 const char *priority_s = smap_get(details, "priority");
2700 error = netdev_get_mtu(netdev, &mtu);
2702 VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU",
2703 netdev_get_name(netdev));
2707 /* HTB requires at least an mtu sized min-rate to send any traffic even
2708 * on uncongested links. */
2709 hc->min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
2710 hc->min_rate = MAX(hc->min_rate, mtu);
2711 hc->min_rate = MIN(hc->min_rate, htb->max_rate);
2714 hc->max_rate = (max_rate_s
2715 ? strtoull(max_rate_s, NULL, 10) / 8
2717 hc->max_rate = MAX(hc->max_rate, hc->min_rate);
2718 hc->max_rate = MIN(hc->max_rate, htb->max_rate);
2722 * According to hints in the documentation that I've read, it is important
2723 * that 'burst' be at least as big as the largest frame that might be
2724 * transmitted. Also, making 'burst' a bit bigger than necessary is OK,
2725 * but having it a bit too small is a problem. Since netdev_get_mtu()
2726 * doesn't include the Ethernet header, we need to add at least 14 (18?) to
2727 * the MTU. We actually add 64, instead of 14, as a guard against
2728 * additional headers get tacked on somewhere that we're not aware of. */
2729 hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0;
2730 hc->burst = MAX(hc->burst, mtu + 64);
2733 hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0;
2739 htb_query_class__(const struct netdev *netdev, unsigned int handle,
2740 unsigned int parent, struct htb_class *options,
2741 struct netdev_queue_stats *stats)
2743 struct ofpbuf *reply;
2746 error = tc_query_class(netdev, handle, parent, &reply);
2748 error = htb_parse_tcmsg__(reply, NULL, options, stats);
2749 ofpbuf_delete(reply);
2755 htb_tc_install(struct netdev *netdev, const struct smap *details)
2759 error = htb_setup_qdisc__(netdev);
2761 struct htb_class hc;
2763 htb_parse_qdisc_details__(netdev, details, &hc);
2764 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
2765 tc_make_handle(1, 0), &hc);
2767 htb_install__(netdev, hc.max_rate);
2773 static struct htb_class *
2774 htb_class_cast__(const struct tc_queue *queue)
2776 return CONTAINER_OF(queue, struct htb_class, tc_queue);
2780 htb_update_queue__(struct netdev *netdev, unsigned int queue_id,
2781 const struct htb_class *hc)
2783 struct htb *htb = htb_get__(netdev);
2784 size_t hash = hash_int(queue_id, 0);
2785 struct tc_queue *queue;
2786 struct htb_class *hcp;
2788 queue = tc_find_queue__(netdev, queue_id, hash);
2790 hcp = htb_class_cast__(queue);
2792 hcp = xmalloc(sizeof *hcp);
2793 queue = &hcp->tc_queue;
2794 queue->queue_id = queue_id;
2795 hmap_insert(&htb->tc.queues, &queue->hmap_node, hash);
2798 hcp->min_rate = hc->min_rate;
2799 hcp->max_rate = hc->max_rate;
2800 hcp->burst = hc->burst;
2801 hcp->priority = hc->priority;
2805 htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
2808 struct nl_dump dump;
2809 struct htb_class hc;
2811 /* Get qdisc options. */
2813 htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
2814 htb_install__(netdev, hc.max_rate);
2817 if (!start_queue_dump(netdev, &dump)) {
2820 while (nl_dump_next(&dump, &msg)) {
2821 unsigned int queue_id;
2823 if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
2824 htb_update_queue__(netdev, queue_id, &hc);
2827 nl_dump_done(&dump);
2833 htb_tc_destroy(struct tc *tc)
2835 struct htb *htb = CONTAINER_OF(tc, struct htb, tc);
2836 struct htb_class *hc, *next;
2838 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) {
2839 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
2847 htb_qdisc_get(const struct netdev *netdev, struct smap *details)
2849 const struct htb *htb = htb_get__(netdev);
2850 smap_add_format(details, "max-rate", "%llu", 8ULL * htb->max_rate);
2855 htb_qdisc_set(struct netdev *netdev, const struct smap *details)
2857 struct htb_class hc;
2860 htb_parse_qdisc_details__(netdev, details, &hc);
2861 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
2862 tc_make_handle(1, 0), &hc);
2864 htb_get__(netdev)->max_rate = hc.max_rate;
2870 htb_class_get(const struct netdev *netdev OVS_UNUSED,
2871 const struct tc_queue *queue, struct smap *details)
2873 const struct htb_class *hc = htb_class_cast__(queue);
2875 smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate);
2876 if (hc->min_rate != hc->max_rate) {
2877 smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate);
2879 smap_add_format(details, "burst", "%llu", 8ULL * hc->burst);
2881 smap_add_format(details, "priority", "%u", hc->priority);
2887 htb_class_set(struct netdev *netdev, unsigned int queue_id,
2888 const struct smap *details)
2890 struct htb_class hc;
2893 error = htb_parse_class_details__(netdev, details, &hc);
2898 error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
2899 tc_make_handle(1, 0xfffe), &hc);
2904 htb_update_queue__(netdev, queue_id, &hc);
2909 htb_class_delete(struct netdev *netdev, struct tc_queue *queue)
2911 struct htb_class *hc = htb_class_cast__(queue);
2912 struct htb *htb = htb_get__(netdev);
2915 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
2917 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
2924 htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
2925 struct netdev_queue_stats *stats)
2927 return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
2928 tc_make_handle(1, 0xfffe), NULL, stats);
2932 htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
2933 const struct ofpbuf *nlmsg,
2934 netdev_dump_queue_stats_cb *cb, void *aux)
2936 struct netdev_queue_stats stats;
2937 unsigned int handle, major, minor;
2940 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
2945 major = tc_get_major(handle);
2946 minor = tc_get_minor(handle);
2947 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2948 (*cb)(minor - 1, &stats, aux);
2953 static const struct tc_ops tc_ops_htb = {
2954 "htb", /* linux_name */
2955 "linux-htb", /* ovs_name */
2956 HTB_N_QUEUES, /* n_queues */
2965 htb_class_get_stats,
2966 htb_class_dump_stats
2969 /* "linux-hfsc" traffic control class. */
2971 #define HFSC_N_QUEUES 0xf000
2979 struct tc_queue tc_queue;
2984 static struct hfsc *
2985 hfsc_get__(const struct netdev *netdev)
2987 struct netdev_dev_linux *netdev_dev;
2988 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
2989 return CONTAINER_OF(netdev_dev->tc, struct hfsc, tc);
2992 static struct hfsc_class *
2993 hfsc_class_cast__(const struct tc_queue *queue)
2995 return CONTAINER_OF(queue, struct hfsc_class, tc_queue);
2999 hfsc_install__(struct netdev *netdev, uint32_t max_rate)
3001 struct netdev_dev_linux * netdev_dev;
3004 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
3005 hfsc = xmalloc(sizeof *hfsc);
3006 tc_init(&hfsc->tc, &tc_ops_hfsc);
3007 hfsc->max_rate = max_rate;
3008 netdev_dev->tc = &hfsc->tc;
3012 hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id,
3013 const struct hfsc_class *hc)
3017 struct hfsc_class *hcp;
3018 struct tc_queue *queue;
3020 hfsc = hfsc_get__(netdev);
3021 hash = hash_int(queue_id, 0);
3023 queue = tc_find_queue__(netdev, queue_id, hash);
3025 hcp = hfsc_class_cast__(queue);
3027 hcp = xmalloc(sizeof *hcp);
3028 queue = &hcp->tc_queue;
3029 queue->queue_id = queue_id;
3030 hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash);
3033 hcp->min_rate = hc->min_rate;
3034 hcp->max_rate = hc->max_rate;
3038 hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class)
3040 const struct tc_service_curve *rsc, *fsc, *usc;
3041 static const struct nl_policy tca_hfsc_policy[] = {
3043 .type = NL_A_UNSPEC,
3045 .min_len = sizeof(struct tc_service_curve),
3048 .type = NL_A_UNSPEC,
3050 .min_len = sizeof(struct tc_service_curve),
3053 .type = NL_A_UNSPEC,
3055 .min_len = sizeof(struct tc_service_curve),
3058 struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)];
3060 if (!nl_parse_nested(nl_options, tca_hfsc_policy,
3061 attrs, ARRAY_SIZE(tca_hfsc_policy))) {
3062 VLOG_WARN_RL(&rl, "failed to parse HFSC class options");
3066 rsc = nl_attr_get(attrs[TCA_HFSC_RSC]);
3067 fsc = nl_attr_get(attrs[TCA_HFSC_FSC]);
3068 usc = nl_attr_get(attrs[TCA_HFSC_USC]);
3070 if (rsc->m1 != 0 || rsc->d != 0 ||
3071 fsc->m1 != 0 || fsc->d != 0 ||
3072 usc->m1 != 0 || usc->d != 0) {
3073 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3074 "Non-linear service curves are not supported.");
3078 if (rsc->m2 != fsc->m2) {
3079 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3080 "Real-time service curves are not supported ");
3084 if (rsc->m2 > usc->m2) {
3085 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3086 "Min-rate service curve is greater than "
3087 "the max-rate service curve.");
3091 class->min_rate = fsc->m2;
3092 class->max_rate = usc->m2;
3097 hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
3098 struct hfsc_class *options,
3099 struct netdev_queue_stats *stats)
3102 unsigned int handle;
3103 struct nlattr *nl_options;
3105 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
3111 unsigned int major, minor;
3113 major = tc_get_major(handle);
3114 minor = tc_get_minor(handle);
3115 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
3116 *queue_id = minor - 1;
3123 error = hfsc_parse_tca_options__(nl_options, options);
3130 hfsc_query_class__(const struct netdev *netdev, unsigned int handle,
3131 unsigned int parent, struct hfsc_class *options,
3132 struct netdev_queue_stats *stats)
3135 struct ofpbuf *reply;
3137 error = tc_query_class(netdev, handle, parent, &reply);
3142 error = hfsc_parse_tcmsg__(reply, NULL, options, stats);
3143 ofpbuf_delete(reply);
3148 hfsc_parse_qdisc_details__(struct netdev *netdev, const struct smap *details,
3149 struct hfsc_class *class)
3152 const char *max_rate_s;
3154 max_rate_s = smap_get(details, "max-rate");
3155 max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
3158 enum netdev_features current;
3160 netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
3161 max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8;
3164 class->min_rate = max_rate;
3165 class->max_rate = max_rate;
3169 hfsc_parse_class_details__(struct netdev *netdev,
3170 const struct smap *details,
3171 struct hfsc_class * class)
3173 const struct hfsc *hfsc;
3174 uint32_t min_rate, max_rate;
3175 const char *min_rate_s, *max_rate_s;
3177 hfsc = hfsc_get__(netdev);
3178 min_rate_s = smap_get(details, "min-rate");
3179 max_rate_s = smap_get(details, "max-rate");
3181 min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
3182 min_rate = MAX(min_rate, 1);
3183 min_rate = MIN(min_rate, hfsc->max_rate);
3185 max_rate = (max_rate_s
3186 ? strtoull(max_rate_s, NULL, 10) / 8
3188 max_rate = MAX(max_rate, min_rate);
3189 max_rate = MIN(max_rate, hfsc->max_rate);
3191 class->min_rate = min_rate;
3192 class->max_rate = max_rate;
3197 /* Create an HFSC qdisc.
3199 * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */
3201 hfsc_setup_qdisc__(struct netdev * netdev)
3203 struct tcmsg *tcmsg;
3204 struct ofpbuf request;
3205 struct tc_hfsc_qopt opt;
3207 tc_del_qdisc(netdev);
3209 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
3210 NLM_F_EXCL | NLM_F_CREATE, &request);
3216 tcmsg->tcm_handle = tc_make_handle(1, 0);
3217 tcmsg->tcm_parent = TC_H_ROOT;
3219 memset(&opt, 0, sizeof opt);
3222 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3223 nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt);
3225 return tc_transact(&request, NULL);
3228 /* Create an HFSC class.
3230 * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc
3231 * sc rate <min_rate> ul rate <max_rate>" */
3233 hfsc_setup_class__(struct netdev *netdev, unsigned int handle,
3234 unsigned int parent, struct hfsc_class *class)
3238 struct tcmsg *tcmsg;
3239 struct ofpbuf request;
3240 struct tc_service_curve min, max;
3242 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
3248 tcmsg->tcm_handle = handle;
3249 tcmsg->tcm_parent = parent;
3253 min.m2 = class->min_rate;
3257 max.m2 = class->max_rate;
3259 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3260 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
3261 nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min);
3262 nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min);
3263 nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max);
3264 nl_msg_end_nested(&request, opt_offset);
3266 error = tc_transact(&request, NULL);
3268 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
3269 "min-rate %ubps, max-rate %ubps (%s)",
3270 netdev_get_name(netdev),
3271 tc_get_major(handle), tc_get_minor(handle),
3272 tc_get_major(parent), tc_get_minor(parent),
3273 class->min_rate, class->max_rate, strerror(error));
3280 hfsc_tc_install(struct netdev *netdev, const struct smap *details)
3283 struct hfsc_class class;
3285 error = hfsc_setup_qdisc__(netdev);
3291 hfsc_parse_qdisc_details__(netdev, details, &class);
3292 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3293 tc_make_handle(1, 0), &class);
3299 hfsc_install__(netdev, class.max_rate);
3304 hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3307 struct nl_dump dump;
3308 struct hfsc_class hc;
3311 hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
3312 hfsc_install__(netdev, hc.max_rate);
3314 if (!start_queue_dump(netdev, &dump)) {
3318 while (nl_dump_next(&dump, &msg)) {
3319 unsigned int queue_id;
3321 if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
3322 hfsc_update_queue__(netdev, queue_id, &hc);
3326 nl_dump_done(&dump);
3331 hfsc_tc_destroy(struct tc *tc)
3334 struct hfsc_class *hc, *next;
3336 hfsc = CONTAINER_OF(tc, struct hfsc, tc);
3338 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) {
3339 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3348 hfsc_qdisc_get(const struct netdev *netdev, struct smap *details)
3350 const struct hfsc *hfsc;
3351 hfsc = hfsc_get__(netdev);
3352 smap_add_format(details, "max-rate", "%llu", 8ULL * hfsc->max_rate);
3357 hfsc_qdisc_set(struct netdev *netdev, const struct smap *details)
3360 struct hfsc_class class;
3362 hfsc_parse_qdisc_details__(netdev, details, &class);
3363 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3364 tc_make_handle(1, 0), &class);
3367 hfsc_get__(netdev)->max_rate = class.max_rate;
3374 hfsc_class_get(const struct netdev *netdev OVS_UNUSED,
3375 const struct tc_queue *queue, struct smap *details)
3377 const struct hfsc_class *hc;
3379 hc = hfsc_class_cast__(queue);
3380 smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate);
3381 if (hc->min_rate != hc->max_rate) {
3382 smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate);
3388 hfsc_class_set(struct netdev *netdev, unsigned int queue_id,
3389 const struct smap *details)
3392 struct hfsc_class class;
3394 error = hfsc_parse_class_details__(netdev, details, &class);
3399 error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
3400 tc_make_handle(1, 0xfffe), &class);
3405 hfsc_update_queue__(netdev, queue_id, &class);
3410 hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue)
3414 struct hfsc_class *hc;
3416 hc = hfsc_class_cast__(queue);
3417 hfsc = hfsc_get__(netdev);
3419 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
3421 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3428 hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
3429 struct netdev_queue_stats *stats)
3431 return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
3432 tc_make_handle(1, 0xfffe), NULL, stats);
3436 hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
3437 const struct ofpbuf *nlmsg,
3438 netdev_dump_queue_stats_cb *cb, void *aux)
3440 struct netdev_queue_stats stats;
3441 unsigned int handle, major, minor;
3444 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
3449 major = tc_get_major(handle);
3450 minor = tc_get_minor(handle);
3451 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
3452 (*cb)(minor - 1, &stats, aux);
3457 static const struct tc_ops tc_ops_hfsc = {
3458 "hfsc", /* linux_name */
3459 "linux-hfsc", /* ovs_name */
3460 HFSC_N_QUEUES, /* n_queues */
3461 hfsc_tc_install, /* tc_install */
3462 hfsc_tc_load, /* tc_load */
3463 hfsc_tc_destroy, /* tc_destroy */
3464 hfsc_qdisc_get, /* qdisc_get */
3465 hfsc_qdisc_set, /* qdisc_set */
3466 hfsc_class_get, /* class_get */
3467 hfsc_class_set, /* class_set */
3468 hfsc_class_delete, /* class_delete */
3469 hfsc_class_get_stats, /* class_get_stats */
3470 hfsc_class_dump_stats /* class_dump_stats */
3473 /* "linux-default" traffic control class.
3475 * This class represents the default, unnamed Linux qdisc. It corresponds to
3476 * the "" (empty string) QoS type in the OVS database. */
3479 default_install__(struct netdev *netdev)
3481 struct netdev_dev_linux *netdev_dev =
3482 netdev_dev_linux_cast(netdev_get_dev(netdev));
3483 static struct tc *tc;
3486 tc = xmalloc(sizeof *tc);
3487 tc_init(tc, &tc_ops_default);
3489 netdev_dev->tc = tc;
3493 default_tc_install(struct netdev *netdev,
3494 const struct smap *details OVS_UNUSED)
3496 default_install__(netdev);
3501 default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3503 default_install__(netdev);
3507 static const struct tc_ops tc_ops_default = {
3508 NULL, /* linux_name */
3513 NULL, /* tc_destroy */
3514 NULL, /* qdisc_get */
3515 NULL, /* qdisc_set */
3516 NULL, /* class_get */
3517 NULL, /* class_set */
3518 NULL, /* class_delete */
3519 NULL, /* class_get_stats */
3520 NULL /* class_dump_stats */
3523 /* "linux-other" traffic control class.
3528 other_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3530 struct netdev_dev_linux *netdev_dev =
3531 netdev_dev_linux_cast(netdev_get_dev(netdev));
3532 static struct tc *tc;
3535 tc = xmalloc(sizeof *tc);
3536 tc_init(tc, &tc_ops_other);
3538 netdev_dev->tc = tc;
3542 static const struct tc_ops tc_ops_other = {
3543 NULL, /* linux_name */
3544 "linux-other", /* ovs_name */
3546 NULL, /* tc_install */
3548 NULL, /* tc_destroy */
3549 NULL, /* qdisc_get */
3550 NULL, /* qdisc_set */
3551 NULL, /* class_get */
3552 NULL, /* class_set */
3553 NULL, /* class_delete */
3554 NULL, /* class_get_stats */
3555 NULL /* class_dump_stats */
3558 /* Traffic control. */
3560 /* Number of kernel "tc" ticks per second. */
3561 static double ticks_per_s;
3563 /* Number of kernel "jiffies" per second. This is used for the purpose of
3564 * computing buffer sizes. Generally kernel qdiscs need to be able to buffer
3565 * one jiffy's worth of data.
3567 * There are two possibilities here:
3569 * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the
3570 * approximate range of 100 to 1024. That means that we really need to
3571 * make sure that the qdisc can buffer that much data.
3573 * - 'buffer_hz' is an absurdly large number. That means that the kernel
3574 * has finely granular timers and there's no need to fudge additional room
3575 * for buffers. (There's no extra effort needed to implement that: the
3576 * large 'buffer_hz' is used as a divisor, so practically any number will
3577 * come out as 0 in the division. Small integer results in the case of
3578 * really high dividends won't have any real effect anyhow.)
3580 static unsigned int buffer_hz;
3582 /* Returns tc handle 'major':'minor'. */
3584 tc_make_handle(unsigned int major, unsigned int minor)
3586 return TC_H_MAKE(major << 16, minor);
3589 /* Returns the major number from 'handle'. */
3591 tc_get_major(unsigned int handle)
3593 return TC_H_MAJ(handle) >> 16;
3596 /* Returns the minor number from 'handle'. */
3598 tc_get_minor(unsigned int handle)
3600 return TC_H_MIN(handle);
3603 static struct tcmsg *
3604 tc_make_request(const struct netdev *netdev, int type, unsigned int flags,
3605 struct ofpbuf *request)
3607 struct tcmsg *tcmsg;
3611 error = get_ifindex(netdev, &ifindex);
3616 ofpbuf_init(request, 512);
3617 nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags);
3618 tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg);
3619 tcmsg->tcm_family = AF_UNSPEC;
3620 tcmsg->tcm_ifindex = ifindex;
3621 /* Caller should fill in tcmsg->tcm_handle. */
3622 /* Caller should fill in tcmsg->tcm_parent. */
3628 tc_transact(struct ofpbuf *request, struct ofpbuf **replyp)
3630 int error = nl_sock_transact(rtnl_sock, request, replyp);
3631 ofpbuf_uninit(request);
3635 /* Adds or deletes a root ingress qdisc on 'netdev'. We use this for
3636 * policing configuration.
3638 * This function is equivalent to running the following when 'add' is true:
3639 * /sbin/tc qdisc add dev <devname> handle ffff: ingress
3641 * This function is equivalent to running the following when 'add' is false:
3642 * /sbin/tc qdisc del dev <devname> handle ffff: ingress
3644 * The configuration and stats may be seen with the following command:
3645 * /sbin/tc -s qdisc show dev <devname>
3647 * Returns 0 if successful, otherwise a positive errno value.
3650 tc_add_del_ingress_qdisc(struct netdev *netdev, bool add)
3652 struct ofpbuf request;
3653 struct tcmsg *tcmsg;
3655 int type = add ? RTM_NEWQDISC : RTM_DELQDISC;
3656 int flags = add ? NLM_F_EXCL | NLM_F_CREATE : 0;
3658 tcmsg = tc_make_request(netdev, type, flags, &request);
3662 tcmsg->tcm_handle = tc_make_handle(0xffff, 0);
3663 tcmsg->tcm_parent = TC_H_INGRESS;
3664 nl_msg_put_string(&request, TCA_KIND, "ingress");
3665 nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0);
3667 error = tc_transact(&request, NULL);
3669 /* If we're deleting the qdisc, don't worry about some of the
3670 * error conditions. */
3671 if (!add && (error == ENOENT || error == EINVAL)) {
3680 /* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size
3683 * This function is equivalent to running:
3684 * /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49
3685 * basic police rate <kbits_rate>kbit burst <kbits_burst>k
3688 * The configuration and stats may be seen with the following command:
3689 * /sbin/tc -s filter show <devname> eth0 parent ffff:
3691 * Returns 0 if successful, otherwise a positive errno value.
3694 tc_add_policer(struct netdev *netdev, int kbits_rate, int kbits_burst)
3696 struct tc_police tc_police;
3697 struct ofpbuf request;
3698 struct tcmsg *tcmsg;
3699 size_t basic_offset;
3700 size_t police_offset;
3704 memset(&tc_police, 0, sizeof tc_police);
3705 tc_police.action = TC_POLICE_SHOT;
3706 tc_police.mtu = mtu;
3707 tc_fill_rate(&tc_police.rate, kbits_rate/8 * 1000, mtu);
3708 tc_police.burst = tc_bytes_to_ticks(tc_police.rate.rate,
3709 kbits_burst * 1024);
3711 tcmsg = tc_make_request(netdev, RTM_NEWTFILTER,
3712 NLM_F_EXCL | NLM_F_CREATE, &request);
3716 tcmsg->tcm_parent = tc_make_handle(0xffff, 0);
3717 tcmsg->tcm_info = tc_make_handle(49,
3718 (OVS_FORCE uint16_t) htons(ETH_P_ALL));
3720 nl_msg_put_string(&request, TCA_KIND, "basic");
3721 basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
3722 police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE);
3723 nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police);
3724 tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate);
3725 nl_msg_end_nested(&request, police_offset);
3726 nl_msg_end_nested(&request, basic_offset);
3728 error = tc_transact(&request, NULL);
3739 /* The values in psched are not individually very meaningful, but they are
3740 * important. The tables below show some values seen in the wild.
3744 * - "c" has always been a constant 1000000 since at least Linux 2.4.14.
3745 * (Before that, there are hints that it was 1000000000.)
3747 * - "d" can be unrealistically large, see the comment on 'buffer_hz'
3751 * -----------------------------------
3752 * [1] 000c8000 000f4240 000f4240 00000064
3753 * [2] 000003e8 00000400 000f4240 3b9aca00
3754 * [3] 000003e8 00000400 000f4240 3b9aca00
3755 * [4] 000003e8 00000400 000f4240 00000064
3756 * [5] 000003e8 00000040 000f4240 3b9aca00
3757 * [6] 000003e8 00000040 000f4240 000000f9
3759 * a b c d ticks_per_s buffer_hz
3760 * ------- --------- ---------- ------------- ----------- -------------
3761 * [1] 819,200 1,000,000 1,000,000 100 819,200 100
3762 * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
3763 * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
3764 * [4] 1,000 1,024 1,000,000 100 976,562 100
3765 * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000
3766 * [6] 1,000 64 1,000,000 249 15,625,000 249
3768 * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p
3769 * [2] 2.6.26-1-686-bigmem from Debian lenny
3770 * [3] 2.6.26-2-sparc64 from Debian lenny
3771 * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p
3772 * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion
3773 * [6] 2.6.34 from kernel.org on KVM
3775 static const char fn[] = "/proc/net/psched";
3776 unsigned int a, b, c, d;
3782 stream = fopen(fn, "r");
3784 VLOG_WARN("%s: open failed: %s", fn, strerror(errno));
3788 if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) {
3789 VLOG_WARN("%s: read failed", fn);
3793 VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d);
3797 VLOG_WARN("%s: invalid scheduler parameters", fn);
3801 ticks_per_s = (double) a * c / b;
3805 VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u",
3808 VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz);
3811 /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a
3812 * rate of 'rate' bytes per second. */
3814 tc_ticks_to_bytes(unsigned int rate, unsigned int ticks)
3819 return (rate * ticks) / ticks_per_s;
3822 /* Returns the number of ticks that it would take to transmit 'size' bytes at a
3823 * rate of 'rate' bytes per second. */
3825 tc_bytes_to_ticks(unsigned int rate, unsigned int size)
3830 return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0;
3833 /* Returns the number of bytes that need to be reserved for qdisc buffering at
3834 * a transmission rate of 'rate' bytes per second. */
3836 tc_buffer_per_jiffy(unsigned int rate)
3841 return rate / buffer_hz;
3844 /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc,
3845 * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull,
3846 * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or
3847 * stores NULL into it if it is absent.
3849 * '*kind' and '*options' point into 'msg', so they are owned by whoever owns
3852 * Returns 0 if successful, otherwise a positive errno value. */
3854 tc_parse_qdisc(const struct ofpbuf *msg, const char **kind,
3855 struct nlattr **options)
3857 static const struct nl_policy tca_policy[] = {
3858 [TCA_KIND] = { .type = NL_A_STRING, .optional = false },
3859 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
3861 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
3863 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
3864 tca_policy, ta, ARRAY_SIZE(ta))) {
3865 VLOG_WARN_RL(&rl, "failed to parse qdisc message");
3870 *kind = nl_attr_get_string(ta[TCA_KIND]);
3874 *options = ta[TCA_OPTIONS];
3889 /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the
3890 * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute
3891 * into '*options', and its queue statistics into '*stats'. Any of the output
3892 * arguments may be null.
3894 * Returns 0 if successful, otherwise a positive errno value. */
3896 tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep,
3897 struct nlattr **options, struct netdev_queue_stats *stats)
3899 static const struct nl_policy tca_policy[] = {
3900 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false },
3901 [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false },
3903 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
3905 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
3906 tca_policy, ta, ARRAY_SIZE(ta))) {
3907 VLOG_WARN_RL(&rl, "failed to parse class message");
3912 struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc);
3913 *handlep = tc->tcm_handle;
3917 *options = ta[TCA_OPTIONS];
3921 const struct gnet_stats_queue *gsq;
3922 struct gnet_stats_basic gsb;
3924 static const struct nl_policy stats_policy[] = {
3925 [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false,
3926 .min_len = sizeof gsb },
3927 [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false,
3928 .min_len = sizeof *gsq },
3930 struct nlattr *sa[ARRAY_SIZE(stats_policy)];
3932 if (!nl_parse_nested(ta[TCA_STATS2], stats_policy,
3933 sa, ARRAY_SIZE(sa))) {
3934 VLOG_WARN_RL(&rl, "failed to parse class stats");
3938 /* Alignment issues screw up the length of struct gnet_stats_basic on
3939 * some arch/bitsize combinations. Newer versions of Linux have a
3940 * struct gnet_stats_basic_packed, but we can't depend on that. The
3941 * easiest thing to do is just to make a copy. */
3942 memset(&gsb, 0, sizeof gsb);
3943 memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]),
3944 MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb));
3945 stats->tx_bytes = gsb.bytes;
3946 stats->tx_packets = gsb.packets;
3948 gsq = nl_attr_get(sa[TCA_STATS_QUEUE]);
3949 stats->tx_errors = gsq->drops;
3959 memset(stats, 0, sizeof *stats);
3964 /* Queries the kernel for class with identifier 'handle' and parent 'parent'
3967 tc_query_class(const struct netdev *netdev,
3968 unsigned int handle, unsigned int parent,
3969 struct ofpbuf **replyp)
3971 struct ofpbuf request;
3972 struct tcmsg *tcmsg;
3975 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request);
3979 tcmsg->tcm_handle = handle;
3980 tcmsg->tcm_parent = parent;
3982 error = tc_transact(&request, replyp);
3984 VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)",
3985 netdev_get_name(netdev),
3986 tc_get_major(handle), tc_get_minor(handle),
3987 tc_get_major(parent), tc_get_minor(parent),
3993 /* Equivalent to "tc class del dev <name> handle <handle>". */
3995 tc_delete_class(const struct netdev *netdev, unsigned int handle)
3997 struct ofpbuf request;
3998 struct tcmsg *tcmsg;
4001 tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request);
4005 tcmsg->tcm_handle = handle;
4006 tcmsg->tcm_parent = 0;
4008 error = tc_transact(&request, NULL);
4010 VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)",
4011 netdev_get_name(netdev),
4012 tc_get_major(handle), tc_get_minor(handle),
4018 /* Equivalent to "tc qdisc del dev <name> root". */
4020 tc_del_qdisc(struct netdev *netdev)
4022 struct netdev_dev_linux *netdev_dev =
4023 netdev_dev_linux_cast(netdev_get_dev(netdev));
4024 struct ofpbuf request;
4025 struct tcmsg *tcmsg;
4028 tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request);
4032 tcmsg->tcm_handle = tc_make_handle(1, 0);
4033 tcmsg->tcm_parent = TC_H_ROOT;
4035 error = tc_transact(&request, NULL);
4036 if (error == EINVAL) {
4037 /* EINVAL probably means that the default qdisc was in use, in which
4038 * case we've accomplished our purpose. */
4041 if (!error && netdev_dev->tc) {
4042 if (netdev_dev->tc->ops->tc_destroy) {
4043 netdev_dev->tc->ops->tc_destroy(netdev_dev->tc);
4045 netdev_dev->tc = NULL;
4050 /* If 'netdev''s qdisc type and parameters are not yet known, queries the
4051 * kernel to determine what they are. Returns 0 if successful, otherwise a
4052 * positive errno value. */
4054 tc_query_qdisc(const struct netdev *netdev)
4056 struct netdev_dev_linux *netdev_dev =
4057 netdev_dev_linux_cast(netdev_get_dev(netdev));
4058 struct ofpbuf request, *qdisc;
4059 const struct tc_ops *ops;
4060 struct tcmsg *tcmsg;
4064 if (netdev_dev->tc) {
4068 /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have
4069 * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before
4070 * 2.6.35 without that fix backported to it.
4072 * To avoid the OOPS, we must not make a request that would attempt to dump
4073 * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a
4074 * few others. There are a few ways that I can see to do this, but most of
4075 * them seem to be racy (and if you lose the race the kernel OOPSes). The
4076 * technique chosen here is to assume that any non-default qdisc that we
4077 * create will have a class with handle 1:0. The built-in qdiscs only have
4078 * a class with handle 0:0.
4080 * We could check for Linux 2.6.35+ and use a more straightforward method
4082 tcmsg = tc_make_request(netdev, RTM_GETQDISC, NLM_F_ECHO, &request);
4086 tcmsg->tcm_handle = tc_make_handle(1, 0);
4087 tcmsg->tcm_parent = 0;
4089 /* Figure out what tc class to instantiate. */
4090 error = tc_transact(&request, &qdisc);
4094 error = tc_parse_qdisc(qdisc, &kind, NULL);
4096 ops = &tc_ops_other;
4098 ops = tc_lookup_linux_name(kind);
4100 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1);
4101 VLOG_INFO_RL(&rl2, "unknown qdisc \"%s\"", kind);
4103 ops = &tc_ops_other;
4106 } else if (error == ENOENT) {
4107 /* Either it's a built-in qdisc, or it's a qdisc set up by some
4108 * other entity that doesn't have a handle 1:0. We will assume
4109 * that it's the system default qdisc. */
4110 ops = &tc_ops_default;
4113 /* Who knows? Maybe the device got deleted. */
4114 VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)",
4115 netdev_get_name(netdev), strerror(error));
4116 ops = &tc_ops_other;
4119 /* Instantiate it. */
4120 load_error = ops->tc_load(CONST_CAST(struct netdev *, netdev), qdisc);
4121 assert((load_error == 0) == (netdev_dev->tc != NULL));
4122 ofpbuf_delete(qdisc);
4124 return error ? error : load_error;
4127 /* Linux traffic control uses tables with 256 entries ("rtab" tables) to
4128 approximate the time to transmit packets of various lengths. For an MTU of
4129 256 or less, each entry is exact; for an MTU of 257 through 512, each entry
4130 represents two possible packet lengths; for a MTU of 513 through 1024, four
4131 possible lengths; and so on.
4133 Returns, for the specified 'mtu', the number of bits that packet lengths
4134 need to be shifted right to fit within such a 256-entry table. */
4136 tc_calc_cell_log(unsigned int mtu)
4141 mtu = ETH_PAYLOAD_MAX;
4143 mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN;
4145 for (cell_log = 0; mtu >= 256; cell_log++) {
4152 /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU
4155 tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu)
4157 memset(rate, 0, sizeof *rate);
4158 rate->cell_log = tc_calc_cell_log(mtu);
4159 /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */
4160 /* rate->cell_align = 0; */ /* distro headers. */
4161 rate->mpu = ETH_TOTAL_MIN;
4165 /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink
4166 * attribute of the specified "type".
4168 * See tc_calc_cell_log() above for a description of "rtab"s. */
4170 tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate)
4175 rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE);
4176 for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) {
4177 unsigned packet_size = (i + 1) << rate->cell_log;
4178 if (packet_size < rate->mpu) {
4179 packet_size = rate->mpu;
4181 rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size);
4185 /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a
4186 * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested
4187 * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of
4190 tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes)
4192 unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu;
4193 return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst));
4196 /* Linux-only functions declared in netdev-linux.h */
4198 /* Returns a fd for an AF_INET socket or a negative errno value. */
4200 netdev_linux_get_af_inet_sock(void)
4202 int error = netdev_linux_init();
4203 return error ? -error : af_inet_sock;
4206 /* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If
4207 * 'enable' is true, the bit is set. Otherwise, it is cleared. */
4209 netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag,
4210 const char *flag_name, bool enable)
4212 const char *netdev_name = netdev_get_name(netdev);
4213 struct ethtool_value evalue;
4217 COVERAGE_INC(netdev_get_ethtool);
4218 memset(&evalue, 0, sizeof evalue);
4219 error = netdev_linux_do_ethtool(netdev_name,
4220 (struct ethtool_cmd *)&evalue,
4221 ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4226 COVERAGE_INC(netdev_set_ethtool);
4227 evalue.data = new_flags = (evalue.data & ~flag) | (enable ? flag : 0);
4228 error = netdev_linux_do_ethtool(netdev_name,
4229 (struct ethtool_cmd *)&evalue,
4230 ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS");
4235 COVERAGE_INC(netdev_get_ethtool);
4236 memset(&evalue, 0, sizeof evalue);
4237 error = netdev_linux_do_ethtool(netdev_name,
4238 (struct ethtool_cmd *)&evalue,
4239 ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4244 if (new_flags != evalue.data) {
4245 VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network "
4246 "device %s failed", enable ? "enable" : "disable",
4247 flag_name, netdev_name);
4254 /* Utility functions. */
4256 /* Copies 'src' into 'dst', performing format conversion in the process. */
4258 netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst,
4259 const struct rtnl_link_stats *src)
4261 dst->rx_packets = src->rx_packets;
4262 dst->tx_packets = src->tx_packets;
4263 dst->rx_bytes = src->rx_bytes;
4264 dst->tx_bytes = src->tx_bytes;
4265 dst->rx_errors = src->rx_errors;
4266 dst->tx_errors = src->tx_errors;
4267 dst->rx_dropped = src->rx_dropped;
4268 dst->tx_dropped = src->tx_dropped;
4269 dst->multicast = src->multicast;
4270 dst->collisions = src->collisions;
4271 dst->rx_length_errors = src->rx_length_errors;
4272 dst->rx_over_errors = src->rx_over_errors;
4273 dst->rx_crc_errors = src->rx_crc_errors;
4274 dst->rx_frame_errors = src->rx_frame_errors;
4275 dst->rx_fifo_errors = src->rx_fifo_errors;
4276 dst->rx_missed_errors = src->rx_missed_errors;
4277 dst->tx_aborted_errors = src->tx_aborted_errors;
4278 dst->tx_carrier_errors = src->tx_carrier_errors;
4279 dst->tx_fifo_errors = src->tx_fifo_errors;
4280 dst->tx_heartbeat_errors = src->tx_heartbeat_errors;
4281 dst->tx_window_errors = src->tx_window_errors;
4285 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
4287 /* Policy for RTNLGRP_LINK messages.
4289 * There are *many* more fields in these messages, but currently we only
4290 * care about these fields. */
4291 static const struct nl_policy rtnlgrp_link_policy[] = {
4292 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
4293 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
4294 .min_len = sizeof(struct rtnl_link_stats) },
4297 struct ofpbuf request;
4298 struct ofpbuf *reply;
4299 struct ifinfomsg *ifi;
4300 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
4303 ofpbuf_init(&request, 0);
4304 nl_msg_put_nlmsghdr(&request, sizeof *ifi, RTM_GETLINK, NLM_F_REQUEST);
4305 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
4306 ifi->ifi_family = PF_UNSPEC;
4307 ifi->ifi_index = ifindex;
4308 error = nl_sock_transact(rtnl_sock, &request, &reply);
4309 ofpbuf_uninit(&request);
4314 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
4315 rtnlgrp_link_policy,
4316 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
4317 ofpbuf_delete(reply);
4321 if (!attrs[IFLA_STATS]) {
4322 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
4323 ofpbuf_delete(reply);
4327 netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(attrs[IFLA_STATS]));
4329 ofpbuf_delete(reply);
4335 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
4337 static const char fn[] = "/proc/net/dev";
4342 stream = fopen(fn, "r");
4344 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
4349 while (fgets(line, sizeof line, stream)) {
4352 #define X64 "%"SCNu64
4355 X64 X64 X64 X64 X64 X64 X64 "%*u"
4356 X64 X64 X64 X64 X64 X64 X64 "%*u",
4362 &stats->rx_fifo_errors,
4363 &stats->rx_frame_errors,
4369 &stats->tx_fifo_errors,
4371 &stats->tx_carrier_errors) != 15) {
4372 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
4373 } else if (!strcmp(devname, netdev_name)) {
4374 stats->rx_length_errors = UINT64_MAX;
4375 stats->rx_over_errors = UINT64_MAX;
4376 stats->rx_crc_errors = UINT64_MAX;
4377 stats->rx_missed_errors = UINT64_MAX;
4378 stats->tx_aborted_errors = UINT64_MAX;
4379 stats->tx_heartbeat_errors = UINT64_MAX;
4380 stats->tx_window_errors = UINT64_MAX;
4386 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
4392 get_flags(const struct netdev_dev *dev, unsigned int *flags)
4398 error = netdev_linux_do_ioctl(dev->name, &ifr, SIOCGIFFLAGS,
4401 *flags = ifr.ifr_flags;
4407 set_flags(struct netdev *netdev, unsigned int flags)
4411 ifr.ifr_flags = flags;
4412 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
4417 do_get_ifindex(const char *netdev_name)
4421 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4422 COVERAGE_INC(netdev_get_ifindex);
4423 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
4424 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
4425 netdev_name, strerror(errno));
4428 return ifr.ifr_ifindex;
4432 get_ifindex(const struct netdev *netdev_, int *ifindexp)
4434 struct netdev_dev_linux *netdev_dev =
4435 netdev_dev_linux_cast(netdev_get_dev(netdev_));
4437 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
4438 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
4441 netdev_dev->get_ifindex_error = -ifindex;
4442 netdev_dev->ifindex = 0;
4444 netdev_dev->get_ifindex_error = 0;
4445 netdev_dev->ifindex = ifindex;
4447 netdev_dev->cache_valid |= VALID_IFINDEX;
4450 *ifindexp = netdev_dev->ifindex;
4451 return netdev_dev->get_ifindex_error;
4455 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
4460 memset(&ifr, 0, sizeof ifr);
4461 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4462 COVERAGE_INC(netdev_get_hwaddr);
4463 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
4464 /* ENODEV probably means that a vif disappeared asynchronously and
4465 * hasn't been removed from the database yet, so reduce the log level
4466 * to INFO for that case. */
4467 VLOG(errno == ENODEV ? VLL_INFO : VLL_ERR,
4468 "ioctl(SIOCGIFHWADDR) on %s device failed: %s",
4469 netdev_name, strerror(errno));
4472 hwaddr_family = ifr.ifr_hwaddr.sa_family;
4473 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
4474 VLOG_WARN("%s device has unknown hardware address family %d",
4475 netdev_name, hwaddr_family);
4477 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
4482 set_etheraddr(const char *netdev_name,
4483 const uint8_t mac[ETH_ADDR_LEN])
4487 memset(&ifr, 0, sizeof ifr);
4488 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4489 ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
4490 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
4491 COVERAGE_INC(netdev_set_hwaddr);
4492 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
4493 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
4494 netdev_name, strerror(errno));
4501 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
4502 int cmd, const char *cmd_name)
4506 memset(&ifr, 0, sizeof ifr);
4507 ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
4508 ifr.ifr_data = (caddr_t) ecmd;
4511 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
4514 if (errno != EOPNOTSUPP) {
4515 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
4516 "failed: %s", cmd_name, name, strerror(errno));
4518 /* The device doesn't support this operation. That's pretty
4519 * common, so there's no point in logging anything. */
4526 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
4527 const char *cmd_name)
4529 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
4530 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
4531 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
4539 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
4540 int cmd, const char *cmd_name)
4545 ifr.ifr_addr.sa_family = AF_INET;
4546 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
4548 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
4549 *ip = sin->sin_addr;
4554 /* Returns an AF_PACKET raw socket or a negative errno value. */
4556 af_packet_sock(void)
4558 static int sock = INT_MIN;
4560 if (sock == INT_MIN) {
4561 sock = socket(AF_PACKET, SOCK_RAW, 0);
4563 set_nonblocking(sock);
4566 VLOG_ERR("failed to create packet socket: %s", strerror(errno));