2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
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"
74 VLOG_DEFINE_THIS_MODULE(netdev_linux);
76 COVERAGE_DEFINE(netdev_set_policing);
77 COVERAGE_DEFINE(netdev_arp_lookup);
78 COVERAGE_DEFINE(netdev_get_ifindex);
79 COVERAGE_DEFINE(netdev_get_hwaddr);
80 COVERAGE_DEFINE(netdev_set_hwaddr);
81 COVERAGE_DEFINE(netdev_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 shash *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 shash *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 shash *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 shash *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 shash *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 memset(&netdev_dev->drvinfo, 0, sizeof netdev_dev->drvinfo);
511 error = netdev_linux_do_ethtool(netdev_dev->netdev_dev.name,
512 (struct ethtool_cmd *)&netdev_dev->drvinfo,
516 netdev_dev->cache_valid |= VALID_DRVINFO;
522 netdev_dev_linux_changed(struct netdev_dev_linux *dev,
523 unsigned int ifi_flags,
527 if (!dev->change_seq) {
531 if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) {
532 dev->carrier_resets++;
534 dev->ifi_flags = ifi_flags;
536 dev->cache_valid &= mask;
540 netdev_dev_linux_update(struct netdev_dev_linux *dev,
541 const struct rtnetlink_link_change *change)
543 if (change->nlmsg_type == RTM_NEWLINK) {
545 netdev_dev_linux_changed(dev, change->ifi_flags, VALID_DRVINFO);
547 /* Update netdev from rtnl-change msg. */
549 dev->mtu = change->mtu;
550 dev->cache_valid |= VALID_MTU;
551 dev->netdev_mtu_error = 0;
554 if (!eth_addr_is_zero(change->addr)) {
555 memcpy(dev->etheraddr, change->addr, ETH_ADDR_LEN);
556 dev->cache_valid |= VALID_ETHERADDR;
557 dev->ether_addr_error = 0;
560 dev->ifindex = change->ifi_index;
561 dev->cache_valid |= VALID_IFINDEX;
562 dev->get_ifindex_error = 0;
565 netdev_dev_linux_changed(dev, change->ifi_flags, 0);
570 netdev_linux_cache_cb(const struct rtnetlink_link_change *change,
571 void *aux OVS_UNUSED)
573 struct netdev_dev_linux *dev;
575 struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
577 const struct netdev_class *netdev_class =
578 netdev_dev_get_class(base_dev);
580 if (is_netdev_linux_class(netdev_class)) {
581 dev = netdev_dev_linux_cast(base_dev);
582 netdev_dev_linux_update(dev, change);
586 struct shash device_shash;
587 struct shash_node *node;
589 shash_init(&device_shash);
590 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
591 SHASH_FOR_EACH (node, &device_shash) {
596 get_flags(&dev->netdev_dev, &flags);
597 netdev_dev_linux_changed(dev, flags, 0);
599 shash_destroy(&device_shash);
604 cache_notifier_ref(void)
606 if (!cache_notifier_refcount) {
607 assert(!netdev_linux_cache_notifier);
609 netdev_linux_cache_notifier =
610 rtnetlink_link_notifier_create(netdev_linux_cache_cb, NULL);
612 if (!netdev_linux_cache_notifier) {
616 cache_notifier_refcount++;
622 cache_notifier_unref(void)
624 assert(cache_notifier_refcount > 0);
625 if (!--cache_notifier_refcount) {
626 assert(netdev_linux_cache_notifier);
627 rtnetlink_link_notifier_destroy(netdev_linux_cache_notifier);
628 netdev_linux_cache_notifier = NULL;
632 /* Creates system and internal devices. */
634 netdev_linux_create(const struct netdev_class *class, const char *name,
635 struct netdev_dev **netdev_devp)
637 struct netdev_dev_linux *netdev_dev;
640 error = cache_notifier_ref();
645 netdev_dev = xzalloc(sizeof *netdev_dev);
646 netdev_dev->change_seq = 1;
647 netdev_dev_init(&netdev_dev->netdev_dev, name, class);
648 get_flags(&netdev_dev->netdev_dev, &netdev_dev->ifi_flags);
650 *netdev_devp = &netdev_dev->netdev_dev;
654 /* For most types of netdevs we open the device for each call of
655 * netdev_open(). However, this is not the case with tap devices,
656 * since it is only possible to open the device once. In this
657 * situation we share a single file descriptor, and consequently
658 * buffers, across all readers. Therefore once data is read it will
659 * be unavailable to other reads for tap devices. */
661 netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED,
662 const char *name, struct netdev_dev **netdev_devp)
664 struct netdev_dev_linux *netdev_dev;
665 struct tap_state *state;
666 static const char tap_dev[] = "/dev/net/tun";
670 netdev_dev = xzalloc(sizeof *netdev_dev);
671 state = &netdev_dev->state.tap;
673 error = cache_notifier_ref();
678 /* Open tap device. */
679 state->fd = open(tap_dev, O_RDWR);
682 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
683 goto error_unref_notifier;
686 /* Create tap device. */
687 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
688 ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
689 if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
690 VLOG_WARN("%s: creating tap device failed: %s", name,
693 goto error_unref_notifier;
696 /* Make non-blocking. */
697 error = set_nonblocking(state->fd);
699 goto error_unref_notifier;
702 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
703 *netdev_devp = &netdev_dev->netdev_dev;
706 error_unref_notifier:
707 cache_notifier_unref();
714 destroy_tap(struct netdev_dev_linux *netdev_dev)
716 struct tap_state *state = &netdev_dev->state.tap;
718 if (state->fd >= 0) {
723 /* Destroys the netdev device 'netdev_dev_'. */
725 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
727 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
728 const struct netdev_class *class = netdev_dev_get_class(netdev_dev_);
730 if (netdev_dev->tc && netdev_dev->tc->ops->tc_destroy) {
731 netdev_dev->tc->ops->tc_destroy(netdev_dev->tc);
734 if (class == &netdev_tap_class) {
735 destroy_tap(netdev_dev);
739 cache_notifier_unref();
743 netdev_linux_open(struct netdev_dev *netdev_dev_, struct netdev **netdevp)
745 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
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 if (!strncmp(netdev_dev_get_type(netdev_dev_), "tap", 3) &&
771 !netdev_dev->state.tap.opened) {
773 /* We assume that the first user of the tap device is the primary user
774 * and give them the tap FD. Subsequent users probably just expect
775 * this to be a system device so open it normally to avoid send/receive
776 * directions appearing to be reversed. */
777 netdev->fd = netdev_dev->state.tap.fd;
778 netdev_dev->state.tap.opened = true;
781 *netdevp = &netdev->netdev;
785 netdev_uninit(&netdev->netdev, true);
789 /* Closes and destroys 'netdev'. */
791 netdev_linux_close(struct netdev *netdev_)
793 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
795 if (netdev->fd > 0 && strncmp(netdev_get_type(netdev_), "tap", 3)) {
802 netdev_linux_listen(struct netdev *netdev_)
804 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
805 struct sockaddr_ll sll;
810 if (netdev->fd >= 0) {
814 /* Create file descriptor. */
815 fd = socket(PF_PACKET, SOCK_RAW, 0);
818 VLOG_ERR("failed to create raw socket (%s)", strerror(error));
822 /* Set non-blocking mode. */
823 error = set_nonblocking(fd);
828 /* Get ethernet device index. */
829 error = get_ifindex(&netdev->netdev, &ifindex);
834 /* Bind to specific ethernet device. */
835 memset(&sll, 0, sizeof sll);
836 sll.sll_family = AF_PACKET;
837 sll.sll_ifindex = ifindex;
838 sll.sll_protocol = (OVS_FORCE unsigned short int) htons(ETH_P_ALL);
839 if (bind(fd, (struct sockaddr *) &sll, sizeof sll) < 0) {
841 VLOG_ERR("%s: failed to bind raw socket (%s)",
842 netdev_get_name(netdev_), strerror(error));
857 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
859 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
861 if (netdev->fd < 0) {
862 /* Device is not listening. */
869 retval = (netdev_->netdev_dev->netdev_class == &netdev_tap_class
870 ? read(netdev->fd, data, size)
871 : recv(netdev->fd, data, size, MSG_TRUNC));
873 return retval <= size ? retval : -EMSGSIZE;
874 } else if (errno != EINTR) {
875 if (errno != EAGAIN) {
876 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
877 strerror(errno), netdev_get_name(netdev_));
884 /* Registers with the poll loop to wake up from the next call to poll_block()
885 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
887 netdev_linux_recv_wait(struct netdev *netdev_)
889 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
890 if (netdev->fd >= 0) {
891 poll_fd_wait(netdev->fd, POLLIN);
895 /* Discards all packets waiting to be received from 'netdev'. */
897 netdev_linux_drain(struct netdev *netdev_)
899 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
900 if (netdev->fd < 0) {
902 } else if (!strncmp(netdev_get_type(netdev_), "tap", 3)) {
904 int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
905 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
909 drain_fd(netdev->fd, ifr.ifr_qlen);
912 return drain_rcvbuf(netdev->fd);
916 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
917 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
918 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
919 * the packet is too big or too small to transmit on the device.
921 * The caller retains ownership of 'buffer' in all cases.
923 * The kernel maintains a packet transmission queue, so the caller is not
924 * expected to do additional queuing of packets. */
926 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
928 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
932 if (netdev->fd < 0) {
933 /* Use our AF_PACKET socket to send to this device. */
934 struct sockaddr_ll sll;
941 sock = af_packet_sock();
946 error = get_ifindex(netdev_, &ifindex);
951 /* We don't bother setting most fields in sockaddr_ll because the
952 * kernel ignores them for SOCK_RAW. */
953 memset(&sll, 0, sizeof sll);
954 sll.sll_family = AF_PACKET;
955 sll.sll_ifindex = ifindex;
957 iov.iov_base = (void *) data;
961 msg.msg_namelen = sizeof sll;
964 msg.msg_control = NULL;
965 msg.msg_controllen = 0;
968 retval = sendmsg(sock, &msg, 0);
970 /* Use the netdev's own fd to send to this device. This is
971 * essential for tap devices, because packets sent to a tap device
972 * with an AF_PACKET socket will loop back to be *received* again
973 * on the tap device. */
974 retval = write(netdev->fd, data, size);
978 /* The Linux AF_PACKET implementation never blocks waiting for room
979 * for packets, instead returning ENOBUFS. Translate this into
980 * EAGAIN for the caller. */
981 if (errno == ENOBUFS) {
983 } else if (errno == EINTR) {
985 } else if (errno != EAGAIN) {
986 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
987 netdev_get_name(netdev_), strerror(errno));
990 } else if (retval != size) {
991 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
992 "%zu) on %s", retval, size, netdev_get_name(netdev_));
1000 /* Registers with the poll loop to wake up from the next call to poll_block()
1001 * when the packet transmission queue has sufficient room to transmit a packet
1002 * with netdev_send().
1004 * The kernel maintains a packet transmission queue, so the client is not
1005 * expected to do additional queuing of packets. Thus, this function is
1006 * unlikely to ever be used. It is included for completeness. */
1008 netdev_linux_send_wait(struct netdev *netdev_)
1010 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1011 if (netdev->fd < 0) {
1012 /* Nothing to do. */
1013 } else if (strncmp(netdev_get_type(netdev_), "tap", 3)) {
1014 poll_fd_wait(netdev->fd, POLLOUT);
1016 /* TAP device always accepts packets.*/
1017 poll_immediate_wake();
1021 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
1022 * otherwise a positive errno value. */
1024 netdev_linux_set_etheraddr(struct netdev *netdev_,
1025 const uint8_t mac[ETH_ADDR_LEN])
1027 struct netdev_dev_linux *netdev_dev =
1028 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1031 if (netdev_dev->cache_valid & VALID_ETHERADDR) {
1032 if (netdev_dev->ether_addr_error) {
1033 return netdev_dev->ether_addr_error;
1035 if (eth_addr_equals(netdev_dev->etheraddr, mac)) {
1038 netdev_dev->cache_valid &= ~VALID_ETHERADDR;
1041 error = set_etheraddr(netdev_get_name(netdev_), mac);
1042 if (!error || error == ENODEV) {
1043 netdev_dev->ether_addr_error = error;
1044 netdev_dev->cache_valid |= VALID_ETHERADDR;
1046 memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
1053 /* Copies 'netdev''s MAC address to 'mac' which is passed as param. */
1055 netdev_linux_get_etheraddr(const struct netdev *netdev_,
1056 uint8_t mac[ETH_ADDR_LEN])
1058 struct netdev_dev_linux *netdev_dev =
1059 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1061 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
1062 int error = get_etheraddr(netdev_get_name(netdev_),
1063 netdev_dev->etheraddr);
1065 netdev_dev->ether_addr_error = error;
1066 netdev_dev->cache_valid |= VALID_ETHERADDR;
1069 if (!netdev_dev->ether_addr_error) {
1070 memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
1073 return netdev_dev->ether_addr_error;
1076 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
1077 * in bytes, not including the hardware header; thus, this is typically 1500
1078 * bytes for Ethernet devices. */
1080 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
1082 struct netdev_dev_linux *netdev_dev =
1083 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1084 if (!(netdev_dev->cache_valid & VALID_MTU)) {
1088 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1089 SIOCGIFMTU, "SIOCGIFMTU");
1091 netdev_dev->netdev_mtu_error = error;
1092 netdev_dev->mtu = ifr.ifr_mtu;
1093 netdev_dev->cache_valid |= VALID_MTU;
1096 if (!netdev_dev->netdev_mtu_error) {
1097 *mtup = netdev_dev->mtu;
1099 return netdev_dev->netdev_mtu_error;
1102 /* Sets the maximum size of transmitted (MTU) for given device using linux
1103 * networking ioctl interface.
1106 netdev_linux_set_mtu(const struct netdev *netdev_, int mtu)
1108 struct netdev_dev_linux *netdev_dev =
1109 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1113 if (netdev_dev->cache_valid & VALID_MTU) {
1114 if (netdev_dev->netdev_mtu_error) {
1115 return netdev_dev->netdev_mtu_error;
1117 if (netdev_dev->mtu == mtu) {
1120 netdev_dev->cache_valid &= ~VALID_MTU;
1123 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1124 SIOCSIFMTU, "SIOCSIFMTU");
1125 if (!error || error == ENODEV) {
1126 netdev_dev->netdev_mtu_error = error;
1127 netdev_dev->mtu = ifr.ifr_mtu;
1128 netdev_dev->cache_valid |= VALID_MTU;
1133 /* Returns the ifindex of 'netdev', if successful, as a positive number.
1134 * On failure, returns a negative errno value. */
1136 netdev_linux_get_ifindex(const struct netdev *netdev)
1140 error = get_ifindex(netdev, &ifindex);
1141 return error ? -error : ifindex;
1145 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
1147 struct netdev_dev_linux *netdev_dev =
1148 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1150 if (netdev_dev->miimon_interval > 0) {
1151 *carrier = netdev_dev->miimon;
1153 *carrier = (netdev_dev->ifi_flags & IFF_RUNNING) != 0;
1159 static long long int
1160 netdev_linux_get_carrier_resets(const struct netdev *netdev)
1162 return netdev_dev_linux_cast(netdev_get_dev(netdev))->carrier_resets;
1166 netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name,
1167 struct mii_ioctl_data *data)
1172 memset(&ifr, 0, sizeof ifr);
1173 memcpy(&ifr.ifr_data, data, sizeof *data);
1174 error = netdev_linux_do_ioctl(name, &ifr, cmd, cmd_name);
1175 memcpy(data, &ifr.ifr_data, sizeof *data);
1181 netdev_linux_get_miimon(const char *name, bool *miimon)
1183 struct mii_ioctl_data data;
1188 memset(&data, 0, sizeof data);
1189 error = netdev_linux_do_miimon(name, SIOCGMIIPHY, "SIOCGMIIPHY", &data);
1191 /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */
1192 data.reg_num = MII_BMSR;
1193 error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG",
1197 *miimon = !!(data.val_out & BMSR_LSTATUS);
1199 VLOG_WARN_RL(&rl, "%s: failed to query MII", name);
1202 struct ethtool_cmd ecmd;
1204 VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool",
1207 memset(&ecmd, 0, sizeof ecmd);
1208 error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK,
1211 struct ethtool_value eval;
1213 memcpy(&eval, &ecmd, sizeof eval);
1214 *miimon = !!eval.data;
1216 VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name);
1224 netdev_linux_set_miimon_interval(struct netdev *netdev_,
1225 long long int interval)
1227 struct netdev_dev_linux *netdev_dev;
1229 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev_));
1231 interval = interval > 0 ? MAX(interval, 100) : 0;
1232 if (netdev_dev->miimon_interval != interval) {
1233 netdev_dev->miimon_interval = interval;
1234 timer_set_expired(&netdev_dev->miimon_timer);
1241 netdev_linux_miimon_run(void)
1243 struct shash device_shash;
1244 struct shash_node *node;
1246 shash_init(&device_shash);
1247 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
1248 SHASH_FOR_EACH (node, &device_shash) {
1249 struct netdev_dev_linux *dev = node->data;
1252 if (dev->miimon_interval <= 0 || !timer_expired(&dev->miimon_timer)) {
1256 netdev_linux_get_miimon(dev->netdev_dev.name, &miimon);
1257 if (miimon != dev->miimon) {
1258 dev->miimon = miimon;
1259 netdev_dev_linux_changed(dev, dev->ifi_flags, 0);
1262 timer_set_duration(&dev->miimon_timer, dev->miimon_interval);
1265 shash_destroy(&device_shash);
1269 netdev_linux_miimon_wait(void)
1271 struct shash device_shash;
1272 struct shash_node *node;
1274 shash_init(&device_shash);
1275 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
1276 SHASH_FOR_EACH (node, &device_shash) {
1277 struct netdev_dev_linux *dev = node->data;
1279 if (dev->miimon_interval > 0) {
1280 timer_wait(&dev->miimon_timer);
1283 shash_destroy(&device_shash);
1286 /* Check whether we can we use RTM_GETLINK to get network device statistics.
1287 * In pre-2.6.19 kernels, this was only available if wireless extensions were
1290 check_for_working_netlink_stats(void)
1292 /* Decide on the netdev_get_stats() implementation to use. Netlink is
1293 * preferable, so if that works, we'll use it. */
1294 int ifindex = do_get_ifindex("lo");
1296 VLOG_WARN("failed to get ifindex for lo, "
1297 "obtaining netdev stats from proc");
1300 struct netdev_stats stats;
1301 int error = get_stats_via_netlink(ifindex, &stats);
1303 VLOG_DBG("obtaining netdev stats via rtnetlink");
1306 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
1307 "via proc (you are probably running a pre-2.6.19 "
1308 "kernel)", strerror(error));
1315 swap_uint64(uint64_t *a, uint64_t *b)
1323 get_stats_via_vport(const struct netdev *netdev_,
1324 struct netdev_stats *stats)
1326 struct netdev_dev_linux *netdev_dev =
1327 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1329 if (!netdev_dev->vport_stats_error ||
1330 !(netdev_dev->cache_valid & VALID_VPORT_STAT_ERROR)) {
1333 error = netdev_vport_get_stats(netdev_, stats);
1335 VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed "
1336 "(%s)", netdev_get_name(netdev_), strerror(error));
1338 netdev_dev->vport_stats_error = error;
1339 netdev_dev->cache_valid |= VALID_VPORT_STAT_ERROR;
1344 netdev_linux_sys_get_stats(const struct netdev *netdev_,
1345 struct netdev_stats *stats)
1347 static int use_netlink_stats = -1;
1350 if (use_netlink_stats < 0) {
1351 use_netlink_stats = check_for_working_netlink_stats();
1354 if (use_netlink_stats) {
1357 error = get_ifindex(netdev_, &ifindex);
1359 error = get_stats_via_netlink(ifindex, stats);
1362 error = get_stats_via_proc(netdev_get_name(netdev_), stats);
1366 VLOG_WARN_RL(&rl, "%s: linux-sys get stats failed %d",
1367 netdev_get_name(netdev_), error);
1373 /* Retrieves current device stats for 'netdev-linux'. */
1375 netdev_linux_get_stats(const struct netdev *netdev_,
1376 struct netdev_stats *stats)
1378 struct netdev_dev_linux *netdev_dev =
1379 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1380 struct netdev_stats dev_stats;
1383 get_stats_via_vport(netdev_, stats);
1385 error = netdev_linux_sys_get_stats(netdev_, &dev_stats);
1388 if (netdev_dev->vport_stats_error) {
1395 if (netdev_dev->vport_stats_error) {
1396 /* stats not available from OVS then use ioctl stats. */
1399 stats->rx_errors += dev_stats.rx_errors;
1400 stats->tx_errors += dev_stats.tx_errors;
1401 stats->rx_dropped += dev_stats.rx_dropped;
1402 stats->tx_dropped += dev_stats.tx_dropped;
1403 stats->multicast += dev_stats.multicast;
1404 stats->collisions += dev_stats.collisions;
1405 stats->rx_length_errors += dev_stats.rx_length_errors;
1406 stats->rx_over_errors += dev_stats.rx_over_errors;
1407 stats->rx_crc_errors += dev_stats.rx_crc_errors;
1408 stats->rx_frame_errors += dev_stats.rx_frame_errors;
1409 stats->rx_fifo_errors += dev_stats.rx_fifo_errors;
1410 stats->rx_missed_errors += dev_stats.rx_missed_errors;
1411 stats->tx_aborted_errors += dev_stats.tx_aborted_errors;
1412 stats->tx_carrier_errors += dev_stats.tx_carrier_errors;
1413 stats->tx_fifo_errors += dev_stats.tx_fifo_errors;
1414 stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors;
1415 stats->tx_window_errors += dev_stats.tx_window_errors;
1420 /* Retrieves current device stats for 'netdev-tap' netdev or
1421 * netdev-internal. */
1423 netdev_tap_get_stats(const struct netdev *netdev_,
1424 struct netdev_stats *stats)
1426 struct netdev_dev_linux *netdev_dev =
1427 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1428 struct netdev_stats dev_stats;
1431 get_stats_via_vport(netdev_, stats);
1433 error = netdev_linux_sys_get_stats(netdev_, &dev_stats);
1435 if (netdev_dev->vport_stats_error) {
1442 /* If this port is an internal port then the transmit and receive stats
1443 * will appear to be swapped relative to the other ports since we are the
1444 * one sending the data, not a remote computer. For consistency, we swap
1445 * them back here. This does not apply if we are getting stats from the
1446 * vport layer because it always tracks stats from the perspective of the
1448 if (netdev_dev->vport_stats_error) {
1450 swap_uint64(&stats->rx_packets, &stats->tx_packets);
1451 swap_uint64(&stats->rx_bytes, &stats->tx_bytes);
1452 swap_uint64(&stats->rx_errors, &stats->tx_errors);
1453 swap_uint64(&stats->rx_dropped, &stats->tx_dropped);
1454 stats->rx_length_errors = 0;
1455 stats->rx_over_errors = 0;
1456 stats->rx_crc_errors = 0;
1457 stats->rx_frame_errors = 0;
1458 stats->rx_fifo_errors = 0;
1459 stats->rx_missed_errors = 0;
1460 stats->tx_aborted_errors = 0;
1461 stats->tx_carrier_errors = 0;
1462 stats->tx_fifo_errors = 0;
1463 stats->tx_heartbeat_errors = 0;
1464 stats->tx_window_errors = 0;
1466 stats->rx_dropped += dev_stats.tx_dropped;
1467 stats->tx_dropped += dev_stats.rx_dropped;
1469 stats->rx_errors += dev_stats.tx_errors;
1470 stats->tx_errors += dev_stats.rx_errors;
1472 stats->multicast += dev_stats.multicast;
1473 stats->collisions += dev_stats.collisions;
1479 netdev_internal_get_stats(const struct netdev *netdev_,
1480 struct netdev_stats *stats)
1482 struct netdev_dev_linux *netdev_dev =
1483 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1485 get_stats_via_vport(netdev_, stats);
1486 return netdev_dev->vport_stats_error;
1490 netdev_linux_read_features(struct netdev_dev_linux *netdev_dev)
1492 struct ethtool_cmd ecmd;
1496 if (netdev_dev->cache_valid & VALID_FEATURES) {
1500 memset(&ecmd, 0, sizeof ecmd);
1501 error = netdev_linux_do_ethtool(netdev_dev->netdev_dev.name, &ecmd,
1502 ETHTOOL_GSET, "ETHTOOL_GSET");
1507 /* Supported features. */
1508 netdev_dev->supported = 0;
1509 if (ecmd.supported & SUPPORTED_10baseT_Half) {
1510 netdev_dev->supported |= NETDEV_F_10MB_HD;
1512 if (ecmd.supported & SUPPORTED_10baseT_Full) {
1513 netdev_dev->supported |= NETDEV_F_10MB_FD;
1515 if (ecmd.supported & SUPPORTED_100baseT_Half) {
1516 netdev_dev->supported |= NETDEV_F_100MB_HD;
1518 if (ecmd.supported & SUPPORTED_100baseT_Full) {
1519 netdev_dev->supported |= NETDEV_F_100MB_FD;
1521 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
1522 netdev_dev->supported |= NETDEV_F_1GB_HD;
1524 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
1525 netdev_dev->supported |= NETDEV_F_1GB_FD;
1527 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
1528 netdev_dev->supported |= NETDEV_F_10GB_FD;
1530 if (ecmd.supported & SUPPORTED_TP) {
1531 netdev_dev->supported |= NETDEV_F_COPPER;
1533 if (ecmd.supported & SUPPORTED_FIBRE) {
1534 netdev_dev->supported |= NETDEV_F_FIBER;
1536 if (ecmd.supported & SUPPORTED_Autoneg) {
1537 netdev_dev->supported |= NETDEV_F_AUTONEG;
1539 if (ecmd.supported & SUPPORTED_Pause) {
1540 netdev_dev->supported |= NETDEV_F_PAUSE;
1542 if (ecmd.supported & SUPPORTED_Asym_Pause) {
1543 netdev_dev->supported |= NETDEV_F_PAUSE_ASYM;
1546 /* Advertised features. */
1547 netdev_dev->advertised = 0;
1548 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
1549 netdev_dev->advertised |= NETDEV_F_10MB_HD;
1551 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
1552 netdev_dev->advertised |= NETDEV_F_10MB_FD;
1554 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
1555 netdev_dev->advertised |= NETDEV_F_100MB_HD;
1557 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
1558 netdev_dev->advertised |= NETDEV_F_100MB_FD;
1560 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
1561 netdev_dev->advertised |= NETDEV_F_1GB_HD;
1563 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
1564 netdev_dev->advertised |= NETDEV_F_1GB_FD;
1566 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
1567 netdev_dev->advertised |= NETDEV_F_10GB_FD;
1569 if (ecmd.advertising & ADVERTISED_TP) {
1570 netdev_dev->advertised |= NETDEV_F_COPPER;
1572 if (ecmd.advertising & ADVERTISED_FIBRE) {
1573 netdev_dev->advertised |= NETDEV_F_FIBER;
1575 if (ecmd.advertising & ADVERTISED_Autoneg) {
1576 netdev_dev->advertised |= NETDEV_F_AUTONEG;
1578 if (ecmd.advertising & ADVERTISED_Pause) {
1579 netdev_dev->advertised |= NETDEV_F_PAUSE;
1581 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
1582 netdev_dev->advertised |= NETDEV_F_PAUSE_ASYM;
1585 /* Current settings. */
1587 if (speed == SPEED_10) {
1588 netdev_dev->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD;
1589 } else if (speed == SPEED_100) {
1590 netdev_dev->current = ecmd.duplex ? NETDEV_F_100MB_FD : NETDEV_F_100MB_HD;
1591 } else if (speed == SPEED_1000) {
1592 netdev_dev->current = ecmd.duplex ? NETDEV_F_1GB_FD : NETDEV_F_1GB_HD;
1593 } else if (speed == SPEED_10000) {
1594 netdev_dev->current = NETDEV_F_10GB_FD;
1595 } else if (speed == 40000) {
1596 netdev_dev->current = NETDEV_F_40GB_FD;
1597 } else if (speed == 100000) {
1598 netdev_dev->current = NETDEV_F_100GB_FD;
1599 } else if (speed == 1000000) {
1600 netdev_dev->current = NETDEV_F_1TB_FD;
1602 netdev_dev->current = 0;
1605 if (ecmd.port == PORT_TP) {
1606 netdev_dev->current |= NETDEV_F_COPPER;
1607 } else if (ecmd.port == PORT_FIBRE) {
1608 netdev_dev->current |= NETDEV_F_FIBER;
1612 netdev_dev->current |= NETDEV_F_AUTONEG;
1615 /* Peer advertisements. */
1616 netdev_dev->peer = 0; /* XXX */
1619 netdev_dev->cache_valid |= VALID_FEATURES;
1620 netdev_dev->get_features_error = error;
1623 /* Stores the features supported by 'netdev' into each of '*current',
1624 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
1625 * bitmap of NETDEV_* bits. Returns 0 if successful, otherwise a positive
1628 netdev_linux_get_features(const struct netdev *netdev_,
1629 enum netdev_features *current,
1630 enum netdev_features *advertised,
1631 enum netdev_features *supported,
1632 enum netdev_features *peer)
1634 struct netdev_dev_linux *netdev_dev =
1635 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1637 netdev_linux_read_features(netdev_dev);
1639 if (!netdev_dev->get_features_error) {
1640 *current = netdev_dev->current;
1641 *advertised = netdev_dev->advertised;
1642 *supported = netdev_dev->supported;
1643 *peer = netdev_dev->peer;
1645 return netdev_dev->get_features_error;
1648 /* Set the features advertised by 'netdev' to 'advertise'. */
1650 netdev_linux_set_advertisements(struct netdev *netdev,
1651 enum netdev_features advertise)
1653 struct ethtool_cmd ecmd;
1656 memset(&ecmd, 0, sizeof ecmd);
1657 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1658 ETHTOOL_GSET, "ETHTOOL_GSET");
1663 ecmd.advertising = 0;
1664 if (advertise & NETDEV_F_10MB_HD) {
1665 ecmd.advertising |= ADVERTISED_10baseT_Half;
1667 if (advertise & NETDEV_F_10MB_FD) {
1668 ecmd.advertising |= ADVERTISED_10baseT_Full;
1670 if (advertise & NETDEV_F_100MB_HD) {
1671 ecmd.advertising |= ADVERTISED_100baseT_Half;
1673 if (advertise & NETDEV_F_100MB_FD) {
1674 ecmd.advertising |= ADVERTISED_100baseT_Full;
1676 if (advertise & NETDEV_F_1GB_HD) {
1677 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1679 if (advertise & NETDEV_F_1GB_FD) {
1680 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1682 if (advertise & NETDEV_F_10GB_FD) {
1683 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1685 if (advertise & NETDEV_F_COPPER) {
1686 ecmd.advertising |= ADVERTISED_TP;
1688 if (advertise & NETDEV_F_FIBER) {
1689 ecmd.advertising |= ADVERTISED_FIBRE;
1691 if (advertise & NETDEV_F_AUTONEG) {
1692 ecmd.advertising |= ADVERTISED_Autoneg;
1694 if (advertise & NETDEV_F_PAUSE) {
1695 ecmd.advertising |= ADVERTISED_Pause;
1697 if (advertise & NETDEV_F_PAUSE_ASYM) {
1698 ecmd.advertising |= ADVERTISED_Asym_Pause;
1700 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1701 ETHTOOL_SSET, "ETHTOOL_SSET");
1704 /* Attempts to set input rate limiting (policing) policy. Returns 0 if
1705 * successful, otherwise a positive errno value. */
1707 netdev_linux_set_policing(struct netdev *netdev,
1708 uint32_t kbits_rate, uint32_t kbits_burst)
1710 struct netdev_dev_linux *netdev_dev =
1711 netdev_dev_linux_cast(netdev_get_dev(netdev));
1712 const char *netdev_name = netdev_get_name(netdev);
1716 kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */
1717 : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */
1718 : kbits_burst); /* Stick with user-specified value. */
1720 if (netdev_dev->cache_valid & VALID_POLICING) {
1721 if (netdev_dev->netdev_policing_error) {
1722 return netdev_dev->netdev_policing_error;
1725 if (netdev_dev->kbits_rate == kbits_rate &&
1726 netdev_dev->kbits_burst == kbits_burst) {
1727 /* Assume that settings haven't changed since we last set them. */
1730 netdev_dev->cache_valid &= ~VALID_POLICING;
1733 COVERAGE_INC(netdev_set_policing);
1734 /* Remove any existing ingress qdisc. */
1735 error = tc_add_del_ingress_qdisc(netdev, false);
1737 VLOG_WARN_RL(&rl, "%s: removing policing failed: %s",
1738 netdev_name, strerror(error));
1743 error = tc_add_del_ingress_qdisc(netdev, true);
1745 VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s",
1746 netdev_name, strerror(error));
1750 error = tc_add_policer(netdev, kbits_rate, kbits_burst);
1752 VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s",
1753 netdev_name, strerror(error));
1758 netdev_dev->kbits_rate = kbits_rate;
1759 netdev_dev->kbits_burst = kbits_burst;
1762 if (!error || error == ENODEV) {
1763 netdev_dev->netdev_policing_error = error;
1764 netdev_dev->cache_valid |= VALID_POLICING;
1770 netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED,
1773 const struct tc_ops **opsp;
1775 for (opsp = tcs; *opsp != NULL; opsp++) {
1776 const struct tc_ops *ops = *opsp;
1777 if (ops->tc_install && ops->ovs_name[0] != '\0') {
1778 sset_add(types, ops->ovs_name);
1785 netdev_linux_create_tap_pl(const struct netdev_class *class OVS_UNUSED,
1786 const char *name, struct netdev_dev **netdev_devp)
1788 struct netdev_dev_linux *netdev_dev;
1789 struct tap_state *state;
1790 char real_name[IFNAMSIZ];
1793 netdev_dev = xzalloc(sizeof *netdev_dev);
1794 state = &netdev_dev->state.tap;
1796 error = cache_notifier_ref();
1801 /* Open tap device. */
1802 state->fd = tun_alloc(IFF_TAP, real_name);
1803 if (state->fd < 0) {
1805 VLOG_WARN("tun_alloc(IFF_TAP, %s) failed: %s", name, strerror(error));
1806 goto error_unref_notifier;
1808 if (strcmp(name, real_name)) {
1809 VLOG_WARN("tap_pl: requested %s, created %s", name, real_name);
1812 /* Make non-blocking. */
1813 error = set_nonblocking(state->fd);
1815 goto error_unref_notifier;
1818 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
1819 *netdev_devp = &netdev_dev->netdev_dev;
1822 error_unref_notifier:
1823 cache_notifier_unref();
1829 static const struct tc_ops *
1830 tc_lookup_ovs_name(const char *name)
1832 const struct tc_ops **opsp;
1834 for (opsp = tcs; *opsp != NULL; opsp++) {
1835 const struct tc_ops *ops = *opsp;
1836 if (!strcmp(name, ops->ovs_name)) {
1843 static const struct tc_ops *
1844 tc_lookup_linux_name(const char *name)
1846 const struct tc_ops **opsp;
1848 for (opsp = tcs; *opsp != NULL; opsp++) {
1849 const struct tc_ops *ops = *opsp;
1850 if (ops->linux_name && !strcmp(name, ops->linux_name)) {
1857 static struct tc_queue *
1858 tc_find_queue__(const struct netdev *netdev, unsigned int queue_id,
1861 struct netdev_dev_linux *netdev_dev =
1862 netdev_dev_linux_cast(netdev_get_dev(netdev));
1863 struct tc_queue *queue;
1865 HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev_dev->tc->queues) {
1866 if (queue->queue_id == queue_id) {
1873 static struct tc_queue *
1874 tc_find_queue(const struct netdev *netdev, unsigned int queue_id)
1876 return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0));
1880 netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED,
1882 struct netdev_qos_capabilities *caps)
1884 const struct tc_ops *ops = tc_lookup_ovs_name(type);
1888 caps->n_queues = ops->n_queues;
1893 netdev_linux_get_qos(const struct netdev *netdev,
1894 const char **typep, struct shash *details)
1896 struct netdev_dev_linux *netdev_dev =
1897 netdev_dev_linux_cast(netdev_get_dev(netdev));
1900 error = tc_query_qdisc(netdev);
1905 *typep = netdev_dev->tc->ops->ovs_name;
1906 return (netdev_dev->tc->ops->qdisc_get
1907 ? netdev_dev->tc->ops->qdisc_get(netdev, details)
1912 netdev_linux_set_qos(struct netdev *netdev,
1913 const char *type, const struct shash *details)
1915 struct netdev_dev_linux *netdev_dev =
1916 netdev_dev_linux_cast(netdev_get_dev(netdev));
1917 const struct tc_ops *new_ops;
1920 new_ops = tc_lookup_ovs_name(type);
1921 if (!new_ops || !new_ops->tc_install) {
1925 error = tc_query_qdisc(netdev);
1930 if (new_ops == netdev_dev->tc->ops) {
1931 return new_ops->qdisc_set ? new_ops->qdisc_set(netdev, details) : 0;
1933 /* Delete existing qdisc. */
1934 error = tc_del_qdisc(netdev);
1938 assert(netdev_dev->tc == NULL);
1940 /* Install new qdisc. */
1941 error = new_ops->tc_install(netdev, details);
1942 assert((error == 0) == (netdev_dev->tc != NULL));
1949 netdev_linux_get_queue(const struct netdev *netdev,
1950 unsigned int queue_id, struct shash *details)
1952 struct netdev_dev_linux *netdev_dev =
1953 netdev_dev_linux_cast(netdev_get_dev(netdev));
1956 error = tc_query_qdisc(netdev);
1960 struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1962 ? netdev_dev->tc->ops->class_get(netdev, queue, details)
1968 netdev_linux_set_queue(struct netdev *netdev,
1969 unsigned int queue_id, const struct shash *details)
1971 struct netdev_dev_linux *netdev_dev =
1972 netdev_dev_linux_cast(netdev_get_dev(netdev));
1975 error = tc_query_qdisc(netdev);
1978 } else if (queue_id >= netdev_dev->tc->ops->n_queues
1979 || !netdev_dev->tc->ops->class_set) {
1983 return netdev_dev->tc->ops->class_set(netdev, queue_id, details);
1987 netdev_linux_delete_queue(struct netdev *netdev, unsigned int queue_id)
1989 struct netdev_dev_linux *netdev_dev =
1990 netdev_dev_linux_cast(netdev_get_dev(netdev));
1993 error = tc_query_qdisc(netdev);
1996 } else if (!netdev_dev->tc->ops->class_delete) {
1999 struct tc_queue *queue = tc_find_queue(netdev, queue_id);
2001 ? netdev_dev->tc->ops->class_delete(netdev, queue)
2007 netdev_linux_get_queue_stats(const struct netdev *netdev,
2008 unsigned int queue_id,
2009 struct netdev_queue_stats *stats)
2011 struct netdev_dev_linux *netdev_dev =
2012 netdev_dev_linux_cast(netdev_get_dev(netdev));
2015 error = tc_query_qdisc(netdev);
2018 } else if (!netdev_dev->tc->ops->class_get_stats) {
2021 const struct tc_queue *queue = tc_find_queue(netdev, queue_id);
2023 ? netdev_dev->tc->ops->class_get_stats(netdev, queue, stats)
2029 start_queue_dump(const struct netdev *netdev, struct nl_dump *dump)
2031 struct ofpbuf request;
2032 struct tcmsg *tcmsg;
2034 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request);
2038 tcmsg->tcm_parent = 0;
2039 nl_dump_start(dump, rtnl_sock, &request);
2040 ofpbuf_uninit(&request);
2045 netdev_linux_dump_queues(const struct netdev *netdev,
2046 netdev_dump_queues_cb *cb, void *aux)
2048 struct netdev_dev_linux *netdev_dev =
2049 netdev_dev_linux_cast(netdev_get_dev(netdev));
2050 struct tc_queue *queue, *next_queue;
2051 struct shash details;
2055 error = tc_query_qdisc(netdev);
2058 } else if (!netdev_dev->tc->ops->class_get) {
2063 shash_init(&details);
2064 HMAP_FOR_EACH_SAFE (queue, next_queue, hmap_node,
2065 &netdev_dev->tc->queues) {
2066 shash_clear(&details);
2068 error = netdev_dev->tc->ops->class_get(netdev, queue, &details);
2070 (*cb)(queue->queue_id, &details, aux);
2075 shash_destroy(&details);
2081 netdev_linux_dump_queue_stats(const struct netdev *netdev,
2082 netdev_dump_queue_stats_cb *cb, void *aux)
2084 struct netdev_dev_linux *netdev_dev =
2085 netdev_dev_linux_cast(netdev_get_dev(netdev));
2086 struct nl_dump dump;
2091 error = tc_query_qdisc(netdev);
2094 } else if (!netdev_dev->tc->ops->class_dump_stats) {
2099 if (!start_queue_dump(netdev, &dump)) {
2102 while (nl_dump_next(&dump, &msg)) {
2103 error = netdev_dev->tc->ops->class_dump_stats(netdev, &msg, cb, aux);
2109 error = nl_dump_done(&dump);
2110 return error ? error : last_error;
2114 netdev_linux_get_in4(const struct netdev *netdev_,
2115 struct in_addr *address, struct in_addr *netmask)
2117 struct netdev_dev_linux *netdev_dev =
2118 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2120 if (!(netdev_dev->cache_valid & VALID_IN4)) {
2123 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
2124 SIOCGIFADDR, "SIOCGIFADDR");
2129 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
2130 SIOCGIFNETMASK, "SIOCGIFNETMASK");
2135 netdev_dev->cache_valid |= VALID_IN4;
2137 *address = netdev_dev->address;
2138 *netmask = netdev_dev->netmask;
2139 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
2143 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
2144 struct in_addr netmask)
2146 struct netdev_dev_linux *netdev_dev =
2147 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2150 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
2152 netdev_dev->cache_valid |= VALID_IN4;
2153 netdev_dev->address = address;
2154 netdev_dev->netmask = netmask;
2155 if (address.s_addr != INADDR_ANY) {
2156 error = do_set_addr(netdev_, SIOCSIFNETMASK,
2157 "SIOCSIFNETMASK", netmask);
2164 parse_if_inet6_line(const char *line,
2165 struct in6_addr *in6, char ifname[16 + 1])
2167 uint8_t *s6 = in6->s6_addr;
2168 #define X8 "%2"SCNx8
2170 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
2171 "%*x %*x %*x %*x %16s\n",
2172 &s6[0], &s6[1], &s6[2], &s6[3],
2173 &s6[4], &s6[5], &s6[6], &s6[7],
2174 &s6[8], &s6[9], &s6[10], &s6[11],
2175 &s6[12], &s6[13], &s6[14], &s6[15],
2179 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
2180 * 'in6' is non-null) and returns true. Otherwise, returns false. */
2182 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
2184 struct netdev_dev_linux *netdev_dev =
2185 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2186 if (!(netdev_dev->cache_valid & VALID_IN6)) {
2190 netdev_dev->in6 = in6addr_any;
2192 file = fopen("/proc/net/if_inet6", "r");
2194 const char *name = netdev_get_name(netdev_);
2195 while (fgets(line, sizeof line, file)) {
2196 struct in6_addr in6_tmp;
2197 char ifname[16 + 1];
2198 if (parse_if_inet6_line(line, &in6_tmp, ifname)
2199 && !strcmp(name, ifname))
2201 netdev_dev->in6 = in6_tmp;
2207 netdev_dev->cache_valid |= VALID_IN6;
2209 *in6 = netdev_dev->in6;
2214 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
2216 struct sockaddr_in sin;
2217 memset(&sin, 0, sizeof sin);
2218 sin.sin_family = AF_INET;
2219 sin.sin_addr = addr;
2222 memset(sa, 0, sizeof *sa);
2223 memcpy(sa, &sin, sizeof sin);
2227 do_set_addr(struct netdev *netdev,
2228 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
2231 ovs_strzcpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
2232 make_in4_sockaddr(&ifr.ifr_addr, addr);
2234 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
2238 /* Adds 'router' as a default IP gateway. */
2240 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
2242 struct in_addr any = { INADDR_ANY };
2246 memset(&rt, 0, sizeof rt);
2247 make_in4_sockaddr(&rt.rt_dst, any);
2248 make_in4_sockaddr(&rt.rt_gateway, router);
2249 make_in4_sockaddr(&rt.rt_genmask, any);
2250 rt.rt_flags = RTF_UP | RTF_GATEWAY;
2251 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
2253 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
2259 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
2262 static const char fn[] = "/proc/net/route";
2267 *netdev_name = NULL;
2268 stream = fopen(fn, "r");
2269 if (stream == NULL) {
2270 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
2275 while (fgets(line, sizeof line, stream)) {
2278 ovs_be32 dest, gateway, mask;
2279 int refcnt, metric, mtu;
2280 unsigned int flags, use, window, irtt;
2283 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
2285 iface, &dest, &gateway, &flags, &refcnt,
2286 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
2288 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
2292 if (!(flags & RTF_UP)) {
2293 /* Skip routes that aren't up. */
2297 /* The output of 'dest', 'mask', and 'gateway' were given in
2298 * network byte order, so we don't need need any endian
2299 * conversions here. */
2300 if ((dest & mask) == (host->s_addr & mask)) {
2302 /* The host is directly reachable. */
2303 next_hop->s_addr = 0;
2305 /* To reach the host, we must go through a gateway. */
2306 next_hop->s_addr = gateway;
2308 *netdev_name = xstrdup(iface);
2320 netdev_linux_get_drv_info(const struct netdev *netdev, struct shash *sh)
2323 struct netdev_dev_linux *netdev_dev =
2324 netdev_dev_linux_cast(netdev_get_dev(netdev));
2326 error = netdev_linux_get_drvinfo(netdev_dev);
2328 shash_add(sh, "driver_name", xstrdup(netdev_dev->drvinfo.driver));
2329 shash_add(sh, "driver_version", xstrdup(netdev_dev->drvinfo.version));
2330 shash_add(sh, "firmware_version", xstrdup(netdev_dev->drvinfo.fw_version));
2336 netdev_internal_get_drv_info(const struct netdev *netdev OVS_UNUSED, struct shash *sh)
2338 shash_add(sh, "driver_name", xstrdup("openvswitch"));
2342 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
2343 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
2344 * returns 0. Otherwise, it returns a positive errno value; in particular,
2345 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
2347 netdev_linux_arp_lookup(const struct netdev *netdev,
2348 ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
2351 struct sockaddr_in sin;
2354 memset(&r, 0, sizeof r);
2355 memset(&sin, 0, sizeof sin);
2356 sin.sin_family = AF_INET;
2357 sin.sin_addr.s_addr = ip;
2359 memcpy(&r.arp_pa, &sin, sizeof sin);
2360 r.arp_ha.sa_family = ARPHRD_ETHER;
2362 ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
2363 COVERAGE_INC(netdev_arp_lookup);
2364 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
2366 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
2367 } else if (retval != ENXIO) {
2368 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
2369 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
2375 nd_to_iff_flags(enum netdev_flags nd)
2378 if (nd & NETDEV_UP) {
2381 if (nd & NETDEV_PROMISC) {
2388 iff_to_nd_flags(int iff)
2390 enum netdev_flags nd = 0;
2394 if (iff & IFF_PROMISC) {
2395 nd |= NETDEV_PROMISC;
2401 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
2402 enum netdev_flags on, enum netdev_flags *old_flagsp)
2404 struct netdev_dev_linux *netdev_dev;
2405 int old_flags, new_flags;
2408 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
2409 old_flags = netdev_dev->ifi_flags;
2410 *old_flagsp = iff_to_nd_flags(old_flags);
2411 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
2412 if (new_flags != old_flags) {
2413 error = set_flags(netdev, new_flags);
2414 get_flags(&netdev_dev->netdev_dev, &netdev_dev->ifi_flags);
2420 netdev_linux_change_seq(const struct netdev *netdev)
2422 return netdev_dev_linux_cast(netdev_get_dev(netdev))->change_seq;
2425 #define NETDEV_LINUX_CLASS(NAME, CREATE, GET_STATS, SET_STATS, \
2426 GET_FEATURES, GET_STATUS) \
2430 netdev_linux_init, \
2432 netdev_linux_wait, \
2435 netdev_linux_destroy, \
2436 NULL, /* get_config */ \
2437 NULL, /* set_config */ \
2439 netdev_linux_open, \
2440 netdev_linux_close, \
2442 netdev_linux_listen, \
2443 netdev_linux_recv, \
2444 netdev_linux_recv_wait, \
2445 netdev_linux_drain, \
2447 netdev_linux_send, \
2448 netdev_linux_send_wait, \
2450 netdev_linux_set_etheraddr, \
2451 netdev_linux_get_etheraddr, \
2452 netdev_linux_get_mtu, \
2453 netdev_linux_set_mtu, \
2454 netdev_linux_get_ifindex, \
2455 netdev_linux_get_carrier, \
2456 netdev_linux_get_carrier_resets, \
2457 netdev_linux_set_miimon_interval, \
2462 netdev_linux_set_advertisements, \
2464 netdev_linux_set_policing, \
2465 netdev_linux_get_qos_types, \
2466 netdev_linux_get_qos_capabilities, \
2467 netdev_linux_get_qos, \
2468 netdev_linux_set_qos, \
2469 netdev_linux_get_queue, \
2470 netdev_linux_set_queue, \
2471 netdev_linux_delete_queue, \
2472 netdev_linux_get_queue_stats, \
2473 netdev_linux_dump_queues, \
2474 netdev_linux_dump_queue_stats, \
2476 netdev_linux_get_in4, \
2477 netdev_linux_set_in4, \
2478 netdev_linux_get_in6, \
2479 netdev_linux_add_router, \
2480 netdev_linux_get_next_hop, \
2482 netdev_linux_arp_lookup, \
2484 netdev_linux_update_flags, \
2486 netdev_linux_change_seq \
2489 const struct netdev_class netdev_linux_class =
2492 netdev_linux_create,
2493 netdev_linux_get_stats,
2494 NULL, /* set_stats */
2495 netdev_linux_get_features,
2496 netdev_linux_get_drv_info);
2498 const struct netdev_class netdev_tap_class =
2501 netdev_linux_create_tap,
2502 netdev_tap_get_stats,
2503 NULL, /* set_stats */
2504 netdev_linux_get_features,
2505 netdev_linux_get_drv_info);
2507 const struct netdev_class netdev_internal_class =
2510 netdev_linux_create,
2511 netdev_internal_get_stats,
2512 netdev_vport_set_stats,
2513 NULL, /* get_features */
2514 netdev_internal_get_drv_info);
2516 const struct netdev_class netdev_tap_pl_class =
2519 netdev_linux_create_tap_pl,
2520 netdev_tap_get_stats,
2521 NULL, /* set_stats */
2522 netdev_linux_get_features,
2523 netdev_linux_get_drv_info);
2525 /* HTB traffic control class. */
2527 #define HTB_N_QUEUES 0xf000
2531 unsigned int max_rate; /* In bytes/s. */
2535 struct tc_queue tc_queue;
2536 unsigned int min_rate; /* In bytes/s. */
2537 unsigned int max_rate; /* In bytes/s. */
2538 unsigned int burst; /* In bytes. */
2539 unsigned int priority; /* Lower values are higher priorities. */
2543 htb_get__(const struct netdev *netdev)
2545 struct netdev_dev_linux *netdev_dev =
2546 netdev_dev_linux_cast(netdev_get_dev(netdev));
2547 return CONTAINER_OF(netdev_dev->tc, struct htb, tc);
2551 htb_install__(struct netdev *netdev, uint64_t max_rate)
2553 struct netdev_dev_linux *netdev_dev =
2554 netdev_dev_linux_cast(netdev_get_dev(netdev));
2557 htb = xmalloc(sizeof *htb);
2558 tc_init(&htb->tc, &tc_ops_htb);
2559 htb->max_rate = max_rate;
2561 netdev_dev->tc = &htb->tc;
2564 /* Create an HTB qdisc.
2566 * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */
2568 htb_setup_qdisc__(struct netdev *netdev)
2571 struct tc_htb_glob opt;
2572 struct ofpbuf request;
2573 struct tcmsg *tcmsg;
2575 tc_del_qdisc(netdev);
2577 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
2578 NLM_F_EXCL | NLM_F_CREATE, &request);
2582 tcmsg->tcm_handle = tc_make_handle(1, 0);
2583 tcmsg->tcm_parent = TC_H_ROOT;
2585 nl_msg_put_string(&request, TCA_KIND, "htb");
2587 memset(&opt, 0, sizeof opt);
2588 opt.rate2quantum = 10;
2592 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2593 nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt);
2594 nl_msg_end_nested(&request, opt_offset);
2596 return tc_transact(&request, NULL);
2599 /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb
2600 * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */
2602 htb_setup_class__(struct netdev *netdev, unsigned int handle,
2603 unsigned int parent, struct htb_class *class)
2606 struct tc_htb_opt opt;
2607 struct ofpbuf request;
2608 struct tcmsg *tcmsg;
2612 error = netdev_get_mtu(netdev, &mtu);
2614 VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU",
2615 netdev_get_name(netdev));
2619 memset(&opt, 0, sizeof opt);
2620 tc_fill_rate(&opt.rate, class->min_rate, mtu);
2621 tc_fill_rate(&opt.ceil, class->max_rate, mtu);
2622 opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst);
2623 opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst);
2624 opt.prio = class->priority;
2626 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
2630 tcmsg->tcm_handle = handle;
2631 tcmsg->tcm_parent = parent;
2633 nl_msg_put_string(&request, TCA_KIND, "htb");
2634 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2635 nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt);
2636 tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate);
2637 tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil);
2638 nl_msg_end_nested(&request, opt_offset);
2640 error = tc_transact(&request, NULL);
2642 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
2643 "min_rate=%u max_rate=%u burst=%u prio=%u (%s)",
2644 netdev_get_name(netdev),
2645 tc_get_major(handle), tc_get_minor(handle),
2646 tc_get_major(parent), tc_get_minor(parent),
2647 class->min_rate, class->max_rate,
2648 class->burst, class->priority, strerror(error));
2653 /* Parses Netlink attributes in 'options' for HTB parameters and stores a
2654 * description of them into 'details'. The description complies with the
2655 * specification given in the vswitch database documentation for linux-htb
2658 htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class)
2660 static const struct nl_policy tca_htb_policy[] = {
2661 [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false,
2662 .min_len = sizeof(struct tc_htb_opt) },
2665 struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)];
2666 const struct tc_htb_opt *htb;
2668 if (!nl_parse_nested(nl_options, tca_htb_policy,
2669 attrs, ARRAY_SIZE(tca_htb_policy))) {
2670 VLOG_WARN_RL(&rl, "failed to parse HTB class options");
2674 htb = nl_attr_get(attrs[TCA_HTB_PARMS]);
2675 class->min_rate = htb->rate.rate;
2676 class->max_rate = htb->ceil.rate;
2677 class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer);
2678 class->priority = htb->prio;
2683 htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
2684 struct htb_class *options,
2685 struct netdev_queue_stats *stats)
2687 struct nlattr *nl_options;
2688 unsigned int handle;
2691 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
2692 if (!error && queue_id) {
2693 unsigned int major = tc_get_major(handle);
2694 unsigned int minor = tc_get_minor(handle);
2695 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2696 *queue_id = minor - 1;
2701 if (!error && options) {
2702 error = htb_parse_tca_options__(nl_options, options);
2708 htb_parse_qdisc_details__(struct netdev *netdev,
2709 const struct shash *details, struct htb_class *hc)
2711 const char *max_rate_s;
2713 max_rate_s = shash_find_data(details, "max-rate");
2714 hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
2715 if (!hc->max_rate) {
2716 enum netdev_features current;
2718 netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
2719 hc->max_rate = netdev_features_to_bps(current) / 8;
2721 hc->min_rate = hc->max_rate;
2727 htb_parse_class_details__(struct netdev *netdev,
2728 const struct shash *details, struct htb_class *hc)
2730 const struct htb *htb = htb_get__(netdev);
2731 const char *min_rate_s = shash_find_data(details, "min-rate");
2732 const char *max_rate_s = shash_find_data(details, "max-rate");
2733 const char *burst_s = shash_find_data(details, "burst");
2734 const char *priority_s = shash_find_data(details, "priority");
2737 error = netdev_get_mtu(netdev, &mtu);
2739 VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU",
2740 netdev_get_name(netdev));
2744 /* HTB requires at least an mtu sized min-rate to send any traffic even
2745 * on uncongested links. */
2746 hc->min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
2747 hc->min_rate = MAX(hc->min_rate, mtu);
2748 hc->min_rate = MIN(hc->min_rate, htb->max_rate);
2751 hc->max_rate = (max_rate_s
2752 ? strtoull(max_rate_s, NULL, 10) / 8
2754 hc->max_rate = MAX(hc->max_rate, hc->min_rate);
2755 hc->max_rate = MIN(hc->max_rate, htb->max_rate);
2759 * According to hints in the documentation that I've read, it is important
2760 * that 'burst' be at least as big as the largest frame that might be
2761 * transmitted. Also, making 'burst' a bit bigger than necessary is OK,
2762 * but having it a bit too small is a problem. Since netdev_get_mtu()
2763 * doesn't include the Ethernet header, we need to add at least 14 (18?) to
2764 * the MTU. We actually add 64, instead of 14, as a guard against
2765 * additional headers get tacked on somewhere that we're not aware of. */
2766 hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0;
2767 hc->burst = MAX(hc->burst, mtu + 64);
2770 hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0;
2776 htb_query_class__(const struct netdev *netdev, unsigned int handle,
2777 unsigned int parent, struct htb_class *options,
2778 struct netdev_queue_stats *stats)
2780 struct ofpbuf *reply;
2783 error = tc_query_class(netdev, handle, parent, &reply);
2785 error = htb_parse_tcmsg__(reply, NULL, options, stats);
2786 ofpbuf_delete(reply);
2792 htb_tc_install(struct netdev *netdev, const struct shash *details)
2796 error = htb_setup_qdisc__(netdev);
2798 struct htb_class hc;
2800 htb_parse_qdisc_details__(netdev, details, &hc);
2801 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
2802 tc_make_handle(1, 0), &hc);
2804 htb_install__(netdev, hc.max_rate);
2810 static struct htb_class *
2811 htb_class_cast__(const struct tc_queue *queue)
2813 return CONTAINER_OF(queue, struct htb_class, tc_queue);
2817 htb_update_queue__(struct netdev *netdev, unsigned int queue_id,
2818 const struct htb_class *hc)
2820 struct htb *htb = htb_get__(netdev);
2821 size_t hash = hash_int(queue_id, 0);
2822 struct tc_queue *queue;
2823 struct htb_class *hcp;
2825 queue = tc_find_queue__(netdev, queue_id, hash);
2827 hcp = htb_class_cast__(queue);
2829 hcp = xmalloc(sizeof *hcp);
2830 queue = &hcp->tc_queue;
2831 queue->queue_id = queue_id;
2832 hmap_insert(&htb->tc.queues, &queue->hmap_node, hash);
2835 hcp->min_rate = hc->min_rate;
2836 hcp->max_rate = hc->max_rate;
2837 hcp->burst = hc->burst;
2838 hcp->priority = hc->priority;
2842 htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
2845 struct nl_dump dump;
2846 struct htb_class hc;
2848 /* Get qdisc options. */
2850 htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
2851 htb_install__(netdev, hc.max_rate);
2854 if (!start_queue_dump(netdev, &dump)) {
2857 while (nl_dump_next(&dump, &msg)) {
2858 unsigned int queue_id;
2860 if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
2861 htb_update_queue__(netdev, queue_id, &hc);
2864 nl_dump_done(&dump);
2870 htb_tc_destroy(struct tc *tc)
2872 struct htb *htb = CONTAINER_OF(tc, struct htb, tc);
2873 struct htb_class *hc, *next;
2875 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) {
2876 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
2884 htb_qdisc_get(const struct netdev *netdev, struct shash *details)
2886 const struct htb *htb = htb_get__(netdev);
2887 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * htb->max_rate));
2892 htb_qdisc_set(struct netdev *netdev, const struct shash *details)
2894 struct htb_class hc;
2897 htb_parse_qdisc_details__(netdev, details, &hc);
2898 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
2899 tc_make_handle(1, 0), &hc);
2901 htb_get__(netdev)->max_rate = hc.max_rate;
2907 htb_class_get(const struct netdev *netdev OVS_UNUSED,
2908 const struct tc_queue *queue, struct shash *details)
2910 const struct htb_class *hc = htb_class_cast__(queue);
2912 shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate));
2913 if (hc->min_rate != hc->max_rate) {
2914 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate));
2916 shash_add(details, "burst", xasprintf("%llu", 8ULL * hc->burst));
2918 shash_add(details, "priority", xasprintf("%u", hc->priority));
2924 htb_class_set(struct netdev *netdev, unsigned int queue_id,
2925 const struct shash *details)
2927 struct htb_class hc;
2930 error = htb_parse_class_details__(netdev, details, &hc);
2935 error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
2936 tc_make_handle(1, 0xfffe), &hc);
2941 htb_update_queue__(netdev, queue_id, &hc);
2946 htb_class_delete(struct netdev *netdev, struct tc_queue *queue)
2948 struct htb_class *hc = htb_class_cast__(queue);
2949 struct htb *htb = htb_get__(netdev);
2952 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
2954 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
2961 htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
2962 struct netdev_queue_stats *stats)
2964 return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
2965 tc_make_handle(1, 0xfffe), NULL, stats);
2969 htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
2970 const struct ofpbuf *nlmsg,
2971 netdev_dump_queue_stats_cb *cb, void *aux)
2973 struct netdev_queue_stats stats;
2974 unsigned int handle, major, minor;
2977 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
2982 major = tc_get_major(handle);
2983 minor = tc_get_minor(handle);
2984 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2985 (*cb)(minor - 1, &stats, aux);
2990 static const struct tc_ops tc_ops_htb = {
2991 "htb", /* linux_name */
2992 "linux-htb", /* ovs_name */
2993 HTB_N_QUEUES, /* n_queues */
3002 htb_class_get_stats,
3003 htb_class_dump_stats
3006 /* "linux-hfsc" traffic control class. */
3008 #define HFSC_N_QUEUES 0xf000
3016 struct tc_queue tc_queue;
3021 static struct hfsc *
3022 hfsc_get__(const struct netdev *netdev)
3024 struct netdev_dev_linux *netdev_dev;
3025 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
3026 return CONTAINER_OF(netdev_dev->tc, struct hfsc, tc);
3029 static struct hfsc_class *
3030 hfsc_class_cast__(const struct tc_queue *queue)
3032 return CONTAINER_OF(queue, struct hfsc_class, tc_queue);
3036 hfsc_install__(struct netdev *netdev, uint32_t max_rate)
3038 struct netdev_dev_linux * netdev_dev;
3041 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
3042 hfsc = xmalloc(sizeof *hfsc);
3043 tc_init(&hfsc->tc, &tc_ops_hfsc);
3044 hfsc->max_rate = max_rate;
3045 netdev_dev->tc = &hfsc->tc;
3049 hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id,
3050 const struct hfsc_class *hc)
3054 struct hfsc_class *hcp;
3055 struct tc_queue *queue;
3057 hfsc = hfsc_get__(netdev);
3058 hash = hash_int(queue_id, 0);
3060 queue = tc_find_queue__(netdev, queue_id, hash);
3062 hcp = hfsc_class_cast__(queue);
3064 hcp = xmalloc(sizeof *hcp);
3065 queue = &hcp->tc_queue;
3066 queue->queue_id = queue_id;
3067 hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash);
3070 hcp->min_rate = hc->min_rate;
3071 hcp->max_rate = hc->max_rate;
3075 hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class)
3077 const struct tc_service_curve *rsc, *fsc, *usc;
3078 static const struct nl_policy tca_hfsc_policy[] = {
3080 .type = NL_A_UNSPEC,
3082 .min_len = sizeof(struct tc_service_curve),
3085 .type = NL_A_UNSPEC,
3087 .min_len = sizeof(struct tc_service_curve),
3090 .type = NL_A_UNSPEC,
3092 .min_len = sizeof(struct tc_service_curve),
3095 struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)];
3097 if (!nl_parse_nested(nl_options, tca_hfsc_policy,
3098 attrs, ARRAY_SIZE(tca_hfsc_policy))) {
3099 VLOG_WARN_RL(&rl, "failed to parse HFSC class options");
3103 rsc = nl_attr_get(attrs[TCA_HFSC_RSC]);
3104 fsc = nl_attr_get(attrs[TCA_HFSC_FSC]);
3105 usc = nl_attr_get(attrs[TCA_HFSC_USC]);
3107 if (rsc->m1 != 0 || rsc->d != 0 ||
3108 fsc->m1 != 0 || fsc->d != 0 ||
3109 usc->m1 != 0 || usc->d != 0) {
3110 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3111 "Non-linear service curves are not supported.");
3115 if (rsc->m2 != fsc->m2) {
3116 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3117 "Real-time service curves are not supported ");
3121 if (rsc->m2 > usc->m2) {
3122 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3123 "Min-rate service curve is greater than "
3124 "the max-rate service curve.");
3128 class->min_rate = fsc->m2;
3129 class->max_rate = usc->m2;
3134 hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
3135 struct hfsc_class *options,
3136 struct netdev_queue_stats *stats)
3139 unsigned int handle;
3140 struct nlattr *nl_options;
3142 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
3148 unsigned int major, minor;
3150 major = tc_get_major(handle);
3151 minor = tc_get_minor(handle);
3152 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
3153 *queue_id = minor - 1;
3160 error = hfsc_parse_tca_options__(nl_options, options);
3167 hfsc_query_class__(const struct netdev *netdev, unsigned int handle,
3168 unsigned int parent, struct hfsc_class *options,
3169 struct netdev_queue_stats *stats)
3172 struct ofpbuf *reply;
3174 error = tc_query_class(netdev, handle, parent, &reply);
3179 error = hfsc_parse_tcmsg__(reply, NULL, options, stats);
3180 ofpbuf_delete(reply);
3185 hfsc_parse_qdisc_details__(struct netdev *netdev, const struct shash *details,
3186 struct hfsc_class *class)
3189 const char *max_rate_s;
3191 max_rate_s = shash_find_data(details, "max-rate");
3192 max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
3195 enum netdev_features current;
3197 netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
3198 max_rate = netdev_features_to_bps(current) / 8;
3201 class->min_rate = max_rate;
3202 class->max_rate = max_rate;
3206 hfsc_parse_class_details__(struct netdev *netdev,
3207 const struct shash *details,
3208 struct hfsc_class * class)
3210 const struct hfsc *hfsc;
3211 uint32_t min_rate, max_rate;
3212 const char *min_rate_s, *max_rate_s;
3214 hfsc = hfsc_get__(netdev);
3215 min_rate_s = shash_find_data(details, "min-rate");
3216 max_rate_s = shash_find_data(details, "max-rate");
3218 min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
3219 min_rate = MAX(min_rate, 1);
3220 min_rate = MIN(min_rate, hfsc->max_rate);
3222 max_rate = (max_rate_s
3223 ? strtoull(max_rate_s, NULL, 10) / 8
3225 max_rate = MAX(max_rate, min_rate);
3226 max_rate = MIN(max_rate, hfsc->max_rate);
3228 class->min_rate = min_rate;
3229 class->max_rate = max_rate;
3234 /* Create an HFSC qdisc.
3236 * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */
3238 hfsc_setup_qdisc__(struct netdev * netdev)
3240 struct tcmsg *tcmsg;
3241 struct ofpbuf request;
3242 struct tc_hfsc_qopt opt;
3244 tc_del_qdisc(netdev);
3246 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
3247 NLM_F_EXCL | NLM_F_CREATE, &request);
3253 tcmsg->tcm_handle = tc_make_handle(1, 0);
3254 tcmsg->tcm_parent = TC_H_ROOT;
3256 memset(&opt, 0, sizeof opt);
3259 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3260 nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt);
3262 return tc_transact(&request, NULL);
3265 /* Create an HFSC class.
3267 * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc
3268 * sc rate <min_rate> ul rate <max_rate>" */
3270 hfsc_setup_class__(struct netdev *netdev, unsigned int handle,
3271 unsigned int parent, struct hfsc_class *class)
3275 struct tcmsg *tcmsg;
3276 struct ofpbuf request;
3277 struct tc_service_curve min, max;
3279 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
3285 tcmsg->tcm_handle = handle;
3286 tcmsg->tcm_parent = parent;
3290 min.m2 = class->min_rate;
3294 max.m2 = class->max_rate;
3296 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3297 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
3298 nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min);
3299 nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min);
3300 nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max);
3301 nl_msg_end_nested(&request, opt_offset);
3303 error = tc_transact(&request, NULL);
3305 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
3306 "min-rate %ubps, max-rate %ubps (%s)",
3307 netdev_get_name(netdev),
3308 tc_get_major(handle), tc_get_minor(handle),
3309 tc_get_major(parent), tc_get_minor(parent),
3310 class->min_rate, class->max_rate, strerror(error));
3317 hfsc_tc_install(struct netdev *netdev, const struct shash *details)
3320 struct hfsc_class class;
3322 error = hfsc_setup_qdisc__(netdev);
3328 hfsc_parse_qdisc_details__(netdev, details, &class);
3329 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3330 tc_make_handle(1, 0), &class);
3336 hfsc_install__(netdev, class.max_rate);
3341 hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3344 struct nl_dump dump;
3345 struct hfsc_class hc;
3348 hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
3349 hfsc_install__(netdev, hc.max_rate);
3351 if (!start_queue_dump(netdev, &dump)) {
3355 while (nl_dump_next(&dump, &msg)) {
3356 unsigned int queue_id;
3358 if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
3359 hfsc_update_queue__(netdev, queue_id, &hc);
3363 nl_dump_done(&dump);
3368 hfsc_tc_destroy(struct tc *tc)
3371 struct hfsc_class *hc, *next;
3373 hfsc = CONTAINER_OF(tc, struct hfsc, tc);
3375 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) {
3376 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3385 hfsc_qdisc_get(const struct netdev *netdev, struct shash *details)
3387 const struct hfsc *hfsc;
3388 hfsc = hfsc_get__(netdev);
3389 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hfsc->max_rate));
3394 hfsc_qdisc_set(struct netdev *netdev, const struct shash *details)
3397 struct hfsc_class class;
3399 hfsc_parse_qdisc_details__(netdev, details, &class);
3400 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3401 tc_make_handle(1, 0), &class);
3404 hfsc_get__(netdev)->max_rate = class.max_rate;
3411 hfsc_class_get(const struct netdev *netdev OVS_UNUSED,
3412 const struct tc_queue *queue, struct shash *details)
3414 const struct hfsc_class *hc;
3416 hc = hfsc_class_cast__(queue);
3417 shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate));
3418 if (hc->min_rate != hc->max_rate) {
3419 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate));
3425 hfsc_class_set(struct netdev *netdev, unsigned int queue_id,
3426 const struct shash *details)
3429 struct hfsc_class class;
3431 error = hfsc_parse_class_details__(netdev, details, &class);
3436 error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
3437 tc_make_handle(1, 0xfffe), &class);
3442 hfsc_update_queue__(netdev, queue_id, &class);
3447 hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue)
3451 struct hfsc_class *hc;
3453 hc = hfsc_class_cast__(queue);
3454 hfsc = hfsc_get__(netdev);
3456 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
3458 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3465 hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
3466 struct netdev_queue_stats *stats)
3468 return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
3469 tc_make_handle(1, 0xfffe), NULL, stats);
3473 hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
3474 const struct ofpbuf *nlmsg,
3475 netdev_dump_queue_stats_cb *cb, void *aux)
3477 struct netdev_queue_stats stats;
3478 unsigned int handle, major, minor;
3481 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
3486 major = tc_get_major(handle);
3487 minor = tc_get_minor(handle);
3488 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
3489 (*cb)(minor - 1, &stats, aux);
3494 static const struct tc_ops tc_ops_hfsc = {
3495 "hfsc", /* linux_name */
3496 "linux-hfsc", /* ovs_name */
3497 HFSC_N_QUEUES, /* n_queues */
3498 hfsc_tc_install, /* tc_install */
3499 hfsc_tc_load, /* tc_load */
3500 hfsc_tc_destroy, /* tc_destroy */
3501 hfsc_qdisc_get, /* qdisc_get */
3502 hfsc_qdisc_set, /* qdisc_set */
3503 hfsc_class_get, /* class_get */
3504 hfsc_class_set, /* class_set */
3505 hfsc_class_delete, /* class_delete */
3506 hfsc_class_get_stats, /* class_get_stats */
3507 hfsc_class_dump_stats /* class_dump_stats */
3510 /* "linux-default" traffic control class.
3512 * This class represents the default, unnamed Linux qdisc. It corresponds to
3513 * the "" (empty string) QoS type in the OVS database. */
3516 default_install__(struct netdev *netdev)
3518 struct netdev_dev_linux *netdev_dev =
3519 netdev_dev_linux_cast(netdev_get_dev(netdev));
3520 static struct tc *tc;
3523 tc = xmalloc(sizeof *tc);
3524 tc_init(tc, &tc_ops_default);
3526 netdev_dev->tc = tc;
3530 default_tc_install(struct netdev *netdev,
3531 const struct shash *details OVS_UNUSED)
3533 default_install__(netdev);
3538 default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3540 default_install__(netdev);
3544 static const struct tc_ops tc_ops_default = {
3545 NULL, /* linux_name */
3550 NULL, /* tc_destroy */
3551 NULL, /* qdisc_get */
3552 NULL, /* qdisc_set */
3553 NULL, /* class_get */
3554 NULL, /* class_set */
3555 NULL, /* class_delete */
3556 NULL, /* class_get_stats */
3557 NULL /* class_dump_stats */
3560 /* "linux-other" traffic control class.
3565 other_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3567 struct netdev_dev_linux *netdev_dev =
3568 netdev_dev_linux_cast(netdev_get_dev(netdev));
3569 static struct tc *tc;
3572 tc = xmalloc(sizeof *tc);
3573 tc_init(tc, &tc_ops_other);
3575 netdev_dev->tc = tc;
3579 static const struct tc_ops tc_ops_other = {
3580 NULL, /* linux_name */
3581 "linux-other", /* ovs_name */
3583 NULL, /* tc_install */
3585 NULL, /* tc_destroy */
3586 NULL, /* qdisc_get */
3587 NULL, /* qdisc_set */
3588 NULL, /* class_get */
3589 NULL, /* class_set */
3590 NULL, /* class_delete */
3591 NULL, /* class_get_stats */
3592 NULL /* class_dump_stats */
3595 /* Traffic control. */
3597 /* Number of kernel "tc" ticks per second. */
3598 static double ticks_per_s;
3600 /* Number of kernel "jiffies" per second. This is used for the purpose of
3601 * computing buffer sizes. Generally kernel qdiscs need to be able to buffer
3602 * one jiffy's worth of data.
3604 * There are two possibilities here:
3606 * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the
3607 * approximate range of 100 to 1024. That means that we really need to
3608 * make sure that the qdisc can buffer that much data.
3610 * - 'buffer_hz' is an absurdly large number. That means that the kernel
3611 * has finely granular timers and there's no need to fudge additional room
3612 * for buffers. (There's no extra effort needed to implement that: the
3613 * large 'buffer_hz' is used as a divisor, so practically any number will
3614 * come out as 0 in the division. Small integer results in the case of
3615 * really high dividends won't have any real effect anyhow.)
3617 static unsigned int buffer_hz;
3619 /* Returns tc handle 'major':'minor'. */
3621 tc_make_handle(unsigned int major, unsigned int minor)
3623 return TC_H_MAKE(major << 16, minor);
3626 /* Returns the major number from 'handle'. */
3628 tc_get_major(unsigned int handle)
3630 return TC_H_MAJ(handle) >> 16;
3633 /* Returns the minor number from 'handle'. */
3635 tc_get_minor(unsigned int handle)
3637 return TC_H_MIN(handle);
3640 static struct tcmsg *
3641 tc_make_request(const struct netdev *netdev, int type, unsigned int flags,
3642 struct ofpbuf *request)
3644 struct tcmsg *tcmsg;
3648 error = get_ifindex(netdev, &ifindex);
3653 ofpbuf_init(request, 512);
3654 nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags);
3655 tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg);
3656 tcmsg->tcm_family = AF_UNSPEC;
3657 tcmsg->tcm_ifindex = ifindex;
3658 /* Caller should fill in tcmsg->tcm_handle. */
3659 /* Caller should fill in tcmsg->tcm_parent. */
3665 tc_transact(struct ofpbuf *request, struct ofpbuf **replyp)
3667 int error = nl_sock_transact(rtnl_sock, request, replyp);
3668 ofpbuf_uninit(request);
3672 /* Adds or deletes a root ingress qdisc on 'netdev'. We use this for
3673 * policing configuration.
3675 * This function is equivalent to running the following when 'add' is true:
3676 * /sbin/tc qdisc add dev <devname> handle ffff: ingress
3678 * This function is equivalent to running the following when 'add' is false:
3679 * /sbin/tc qdisc del dev <devname> handle ffff: ingress
3681 * The configuration and stats may be seen with the following command:
3682 * /sbin/tc -s qdisc show dev <devname>
3684 * Returns 0 if successful, otherwise a positive errno value.
3687 tc_add_del_ingress_qdisc(struct netdev *netdev, bool add)
3689 struct ofpbuf request;
3690 struct tcmsg *tcmsg;
3692 int type = add ? RTM_NEWQDISC : RTM_DELQDISC;
3693 int flags = add ? NLM_F_EXCL | NLM_F_CREATE : 0;
3695 tcmsg = tc_make_request(netdev, type, flags, &request);
3699 tcmsg->tcm_handle = tc_make_handle(0xffff, 0);
3700 tcmsg->tcm_parent = TC_H_INGRESS;
3701 nl_msg_put_string(&request, TCA_KIND, "ingress");
3702 nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0);
3704 error = tc_transact(&request, NULL);
3706 /* If we're deleting the qdisc, don't worry about some of the
3707 * error conditions. */
3708 if (!add && (error == ENOENT || error == EINVAL)) {
3717 /* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size
3720 * This function is equivalent to running:
3721 * /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49
3722 * basic police rate <kbits_rate>kbit burst <kbits_burst>k
3725 * The configuration and stats may be seen with the following command:
3726 * /sbin/tc -s filter show <devname> eth0 parent ffff:
3728 * Returns 0 if successful, otherwise a positive errno value.
3731 tc_add_policer(struct netdev *netdev, int kbits_rate, int kbits_burst)
3733 struct tc_police tc_police;
3734 struct ofpbuf request;
3735 struct tcmsg *tcmsg;
3736 size_t basic_offset;
3737 size_t police_offset;
3741 memset(&tc_police, 0, sizeof tc_police);
3742 tc_police.action = TC_POLICE_SHOT;
3743 tc_police.mtu = mtu;
3744 tc_fill_rate(&tc_police.rate, kbits_rate/8 * 1000, mtu);
3745 tc_police.burst = tc_bytes_to_ticks(tc_police.rate.rate,
3746 kbits_burst * 1024);
3748 tcmsg = tc_make_request(netdev, RTM_NEWTFILTER,
3749 NLM_F_EXCL | NLM_F_CREATE, &request);
3753 tcmsg->tcm_parent = tc_make_handle(0xffff, 0);
3754 tcmsg->tcm_info = tc_make_handle(49,
3755 (OVS_FORCE uint16_t) htons(ETH_P_ALL));
3757 nl_msg_put_string(&request, TCA_KIND, "basic");
3758 basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
3759 police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE);
3760 nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police);
3761 tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate);
3762 nl_msg_end_nested(&request, police_offset);
3763 nl_msg_end_nested(&request, basic_offset);
3765 error = tc_transact(&request, NULL);
3776 /* The values in psched are not individually very meaningful, but they are
3777 * important. The tables below show some values seen in the wild.
3781 * - "c" has always been a constant 1000000 since at least Linux 2.4.14.
3782 * (Before that, there are hints that it was 1000000000.)
3784 * - "d" can be unrealistically large, see the comment on 'buffer_hz'
3788 * -----------------------------------
3789 * [1] 000c8000 000f4240 000f4240 00000064
3790 * [2] 000003e8 00000400 000f4240 3b9aca00
3791 * [3] 000003e8 00000400 000f4240 3b9aca00
3792 * [4] 000003e8 00000400 000f4240 00000064
3793 * [5] 000003e8 00000040 000f4240 3b9aca00
3794 * [6] 000003e8 00000040 000f4240 000000f9
3796 * a b c d ticks_per_s buffer_hz
3797 * ------- --------- ---------- ------------- ----------- -------------
3798 * [1] 819,200 1,000,000 1,000,000 100 819,200 100
3799 * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
3800 * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
3801 * [4] 1,000 1,024 1,000,000 100 976,562 100
3802 * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000
3803 * [6] 1,000 64 1,000,000 249 15,625,000 249
3805 * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p
3806 * [2] 2.6.26-1-686-bigmem from Debian lenny
3807 * [3] 2.6.26-2-sparc64 from Debian lenny
3808 * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p
3809 * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion
3810 * [6] 2.6.34 from kernel.org on KVM
3812 static const char fn[] = "/proc/net/psched";
3813 unsigned int a, b, c, d;
3819 stream = fopen(fn, "r");
3821 VLOG_WARN("%s: open failed: %s", fn, strerror(errno));
3825 if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) {
3826 VLOG_WARN("%s: read failed", fn);
3830 VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d);
3834 VLOG_WARN("%s: invalid scheduler parameters", fn);
3838 ticks_per_s = (double) a * c / b;
3842 VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u",
3845 VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz);
3848 /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a
3849 * rate of 'rate' bytes per second. */
3851 tc_ticks_to_bytes(unsigned int rate, unsigned int ticks)
3856 return (rate * ticks) / ticks_per_s;
3859 /* Returns the number of ticks that it would take to transmit 'size' bytes at a
3860 * rate of 'rate' bytes per second. */
3862 tc_bytes_to_ticks(unsigned int rate, unsigned int size)
3867 return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0;
3870 /* Returns the number of bytes that need to be reserved for qdisc buffering at
3871 * a transmission rate of 'rate' bytes per second. */
3873 tc_buffer_per_jiffy(unsigned int rate)
3878 return rate / buffer_hz;
3881 /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc,
3882 * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull,
3883 * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or
3884 * stores NULL into it if it is absent.
3886 * '*kind' and '*options' point into 'msg', so they are owned by whoever owns
3889 * Returns 0 if successful, otherwise a positive errno value. */
3891 tc_parse_qdisc(const struct ofpbuf *msg, const char **kind,
3892 struct nlattr **options)
3894 static const struct nl_policy tca_policy[] = {
3895 [TCA_KIND] = { .type = NL_A_STRING, .optional = false },
3896 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
3898 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
3900 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
3901 tca_policy, ta, ARRAY_SIZE(ta))) {
3902 VLOG_WARN_RL(&rl, "failed to parse qdisc message");
3907 *kind = nl_attr_get_string(ta[TCA_KIND]);
3911 *options = ta[TCA_OPTIONS];
3926 /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the
3927 * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute
3928 * into '*options', and its queue statistics into '*stats'. Any of the output
3929 * arguments may be null.
3931 * Returns 0 if successful, otherwise a positive errno value. */
3933 tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep,
3934 struct nlattr **options, struct netdev_queue_stats *stats)
3936 static const struct nl_policy tca_policy[] = {
3937 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false },
3938 [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false },
3940 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
3942 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
3943 tca_policy, ta, ARRAY_SIZE(ta))) {
3944 VLOG_WARN_RL(&rl, "failed to parse class message");
3949 struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc);
3950 *handlep = tc->tcm_handle;
3954 *options = ta[TCA_OPTIONS];
3958 const struct gnet_stats_queue *gsq;
3959 struct gnet_stats_basic gsb;
3961 static const struct nl_policy stats_policy[] = {
3962 [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false,
3963 .min_len = sizeof gsb },
3964 [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false,
3965 .min_len = sizeof *gsq },
3967 struct nlattr *sa[ARRAY_SIZE(stats_policy)];
3969 if (!nl_parse_nested(ta[TCA_STATS2], stats_policy,
3970 sa, ARRAY_SIZE(sa))) {
3971 VLOG_WARN_RL(&rl, "failed to parse class stats");
3975 /* Alignment issues screw up the length of struct gnet_stats_basic on
3976 * some arch/bitsize combinations. Newer versions of Linux have a
3977 * struct gnet_stats_basic_packed, but we can't depend on that. The
3978 * easiest thing to do is just to make a copy. */
3979 memset(&gsb, 0, sizeof gsb);
3980 memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]),
3981 MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb));
3982 stats->tx_bytes = gsb.bytes;
3983 stats->tx_packets = gsb.packets;
3985 gsq = nl_attr_get(sa[TCA_STATS_QUEUE]);
3986 stats->tx_errors = gsq->drops;
3996 memset(stats, 0, sizeof *stats);
4001 /* Queries the kernel for class with identifier 'handle' and parent 'parent'
4004 tc_query_class(const struct netdev *netdev,
4005 unsigned int handle, unsigned int parent,
4006 struct ofpbuf **replyp)
4008 struct ofpbuf request;
4009 struct tcmsg *tcmsg;
4012 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request);
4016 tcmsg->tcm_handle = handle;
4017 tcmsg->tcm_parent = parent;
4019 error = tc_transact(&request, replyp);
4021 VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)",
4022 netdev_get_name(netdev),
4023 tc_get_major(handle), tc_get_minor(handle),
4024 tc_get_major(parent), tc_get_minor(parent),
4030 /* Equivalent to "tc class del dev <name> handle <handle>". */
4032 tc_delete_class(const struct netdev *netdev, unsigned int handle)
4034 struct ofpbuf request;
4035 struct tcmsg *tcmsg;
4038 tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request);
4042 tcmsg->tcm_handle = handle;
4043 tcmsg->tcm_parent = 0;
4045 error = tc_transact(&request, NULL);
4047 VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)",
4048 netdev_get_name(netdev),
4049 tc_get_major(handle), tc_get_minor(handle),
4055 /* Equivalent to "tc qdisc del dev <name> root". */
4057 tc_del_qdisc(struct netdev *netdev)
4059 struct netdev_dev_linux *netdev_dev =
4060 netdev_dev_linux_cast(netdev_get_dev(netdev));
4061 struct ofpbuf request;
4062 struct tcmsg *tcmsg;
4065 tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request);
4069 tcmsg->tcm_handle = tc_make_handle(1, 0);
4070 tcmsg->tcm_parent = TC_H_ROOT;
4072 error = tc_transact(&request, NULL);
4073 if (error == EINVAL) {
4074 /* EINVAL probably means that the default qdisc was in use, in which
4075 * case we've accomplished our purpose. */
4078 if (!error && netdev_dev->tc) {
4079 if (netdev_dev->tc->ops->tc_destroy) {
4080 netdev_dev->tc->ops->tc_destroy(netdev_dev->tc);
4082 netdev_dev->tc = NULL;
4087 /* If 'netdev''s qdisc type and parameters are not yet known, queries the
4088 * kernel to determine what they are. Returns 0 if successful, otherwise a
4089 * positive errno value. */
4091 tc_query_qdisc(const struct netdev *netdev)
4093 struct netdev_dev_linux *netdev_dev =
4094 netdev_dev_linux_cast(netdev_get_dev(netdev));
4095 struct ofpbuf request, *qdisc;
4096 const struct tc_ops *ops;
4097 struct tcmsg *tcmsg;
4101 if (netdev_dev->tc) {
4105 /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have
4106 * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before
4107 * 2.6.35 without that fix backported to it.
4109 * To avoid the OOPS, we must not make a request that would attempt to dump
4110 * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a
4111 * few others. There are a few ways that I can see to do this, but most of
4112 * them seem to be racy (and if you lose the race the kernel OOPSes). The
4113 * technique chosen here is to assume that any non-default qdisc that we
4114 * create will have a class with handle 1:0. The built-in qdiscs only have
4115 * a class with handle 0:0.
4117 * We could check for Linux 2.6.35+ and use a more straightforward method
4119 tcmsg = tc_make_request(netdev, RTM_GETQDISC, NLM_F_ECHO, &request);
4123 tcmsg->tcm_handle = tc_make_handle(1, 0);
4124 tcmsg->tcm_parent = 0;
4126 /* Figure out what tc class to instantiate. */
4127 error = tc_transact(&request, &qdisc);
4131 error = tc_parse_qdisc(qdisc, &kind, NULL);
4133 ops = &tc_ops_other;
4135 ops = tc_lookup_linux_name(kind);
4137 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1);
4138 VLOG_INFO_RL(&rl2, "unknown qdisc \"%s\"", kind);
4140 ops = &tc_ops_other;
4143 } else if (error == ENOENT) {
4144 /* Either it's a built-in qdisc, or it's a qdisc set up by some
4145 * other entity that doesn't have a handle 1:0. We will assume
4146 * that it's the system default qdisc. */
4147 ops = &tc_ops_default;
4150 /* Who knows? Maybe the device got deleted. */
4151 VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)",
4152 netdev_get_name(netdev), strerror(error));
4153 ops = &tc_ops_other;
4156 /* Instantiate it. */
4157 load_error = ops->tc_load((struct netdev *) netdev, qdisc);
4158 assert((load_error == 0) == (netdev_dev->tc != NULL));
4159 ofpbuf_delete(qdisc);
4161 return error ? error : load_error;
4164 /* Linux traffic control uses tables with 256 entries ("rtab" tables) to
4165 approximate the time to transmit packets of various lengths. For an MTU of
4166 256 or less, each entry is exact; for an MTU of 257 through 512, each entry
4167 represents two possible packet lengths; for a MTU of 513 through 1024, four
4168 possible lengths; and so on.
4170 Returns, for the specified 'mtu', the number of bits that packet lengths
4171 need to be shifted right to fit within such a 256-entry table. */
4173 tc_calc_cell_log(unsigned int mtu)
4178 mtu = ETH_PAYLOAD_MAX;
4180 mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN;
4182 for (cell_log = 0; mtu >= 256; cell_log++) {
4189 /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU
4192 tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu)
4194 memset(rate, 0, sizeof *rate);
4195 rate->cell_log = tc_calc_cell_log(mtu);
4196 /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */
4197 /* rate->cell_align = 0; */ /* distro headers. */
4198 rate->mpu = ETH_TOTAL_MIN;
4202 /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink
4203 * attribute of the specified "type".
4205 * See tc_calc_cell_log() above for a description of "rtab"s. */
4207 tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate)
4212 rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE);
4213 for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) {
4214 unsigned packet_size = (i + 1) << rate->cell_log;
4215 if (packet_size < rate->mpu) {
4216 packet_size = rate->mpu;
4218 rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size);
4222 /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a
4223 * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested
4224 * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of
4227 tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes)
4229 unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu;
4230 return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst));
4233 /* Linux-only functions declared in netdev-linux.h */
4235 /* Returns a fd for an AF_INET socket or a negative errno value. */
4237 netdev_linux_get_af_inet_sock(void)
4239 int error = netdev_linux_init();
4240 return error ? -error : af_inet_sock;
4243 /* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If
4244 * 'enable' is true, the bit is set. Otherwise, it is cleared. */
4246 netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag,
4247 const char *flag_name, bool enable)
4249 const char *netdev_name = netdev_get_name(netdev);
4250 struct ethtool_value evalue;
4254 memset(&evalue, 0, sizeof evalue);
4255 error = netdev_linux_do_ethtool(netdev_name,
4256 (struct ethtool_cmd *)&evalue,
4257 ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4262 evalue.data = new_flags = (evalue.data & ~flag) | (enable ? flag : 0);
4263 error = netdev_linux_do_ethtool(netdev_name,
4264 (struct ethtool_cmd *)&evalue,
4265 ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS");
4270 memset(&evalue, 0, sizeof evalue);
4271 error = netdev_linux_do_ethtool(netdev_name,
4272 (struct ethtool_cmd *)&evalue,
4273 ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4278 if (new_flags != evalue.data) {
4279 VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network "
4280 "device %s failed", enable ? "enable" : "disable",
4281 flag_name, netdev_name);
4288 /* Utility functions. */
4290 /* Copies 'src' into 'dst', performing format conversion in the process. */
4292 netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst,
4293 const struct rtnl_link_stats *src)
4295 dst->rx_packets = src->rx_packets;
4296 dst->tx_packets = src->tx_packets;
4297 dst->rx_bytes = src->rx_bytes;
4298 dst->tx_bytes = src->tx_bytes;
4299 dst->rx_errors = src->rx_errors;
4300 dst->tx_errors = src->tx_errors;
4301 dst->rx_dropped = src->rx_dropped;
4302 dst->tx_dropped = src->tx_dropped;
4303 dst->multicast = src->multicast;
4304 dst->collisions = src->collisions;
4305 dst->rx_length_errors = src->rx_length_errors;
4306 dst->rx_over_errors = src->rx_over_errors;
4307 dst->rx_crc_errors = src->rx_crc_errors;
4308 dst->rx_frame_errors = src->rx_frame_errors;
4309 dst->rx_fifo_errors = src->rx_fifo_errors;
4310 dst->rx_missed_errors = src->rx_missed_errors;
4311 dst->tx_aborted_errors = src->tx_aborted_errors;
4312 dst->tx_carrier_errors = src->tx_carrier_errors;
4313 dst->tx_fifo_errors = src->tx_fifo_errors;
4314 dst->tx_heartbeat_errors = src->tx_heartbeat_errors;
4315 dst->tx_window_errors = src->tx_window_errors;
4319 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
4321 /* Policy for RTNLGRP_LINK messages.
4323 * There are *many* more fields in these messages, but currently we only
4324 * care about these fields. */
4325 static const struct nl_policy rtnlgrp_link_policy[] = {
4326 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
4327 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
4328 .min_len = sizeof(struct rtnl_link_stats) },
4331 struct ofpbuf request;
4332 struct ofpbuf *reply;
4333 struct ifinfomsg *ifi;
4334 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
4337 ofpbuf_init(&request, 0);
4338 nl_msg_put_nlmsghdr(&request, sizeof *ifi, RTM_GETLINK, NLM_F_REQUEST);
4339 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
4340 ifi->ifi_family = PF_UNSPEC;
4341 ifi->ifi_index = ifindex;
4342 error = nl_sock_transact(rtnl_sock, &request, &reply);
4343 ofpbuf_uninit(&request);
4348 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
4349 rtnlgrp_link_policy,
4350 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
4351 ofpbuf_delete(reply);
4355 if (!attrs[IFLA_STATS]) {
4356 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
4357 ofpbuf_delete(reply);
4361 netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(attrs[IFLA_STATS]));
4363 ofpbuf_delete(reply);
4369 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
4371 static const char fn[] = "/proc/net/dev";
4376 stream = fopen(fn, "r");
4378 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
4383 while (fgets(line, sizeof line, stream)) {
4386 #define X64 "%"SCNu64
4389 X64 X64 X64 X64 X64 X64 X64 "%*u"
4390 X64 X64 X64 X64 X64 X64 X64 "%*u",
4396 &stats->rx_fifo_errors,
4397 &stats->rx_frame_errors,
4403 &stats->tx_fifo_errors,
4405 &stats->tx_carrier_errors) != 15) {
4406 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
4407 } else if (!strcmp(devname, netdev_name)) {
4408 stats->rx_length_errors = UINT64_MAX;
4409 stats->rx_over_errors = UINT64_MAX;
4410 stats->rx_crc_errors = UINT64_MAX;
4411 stats->rx_missed_errors = UINT64_MAX;
4412 stats->tx_aborted_errors = UINT64_MAX;
4413 stats->tx_heartbeat_errors = UINT64_MAX;
4414 stats->tx_window_errors = UINT64_MAX;
4420 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
4426 get_flags(const struct netdev_dev *dev, unsigned int *flags)
4432 error = netdev_linux_do_ioctl(dev->name, &ifr, SIOCGIFFLAGS,
4435 *flags = ifr.ifr_flags;
4441 set_flags(struct netdev *netdev, unsigned int flags)
4445 ifr.ifr_flags = flags;
4446 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
4451 do_get_ifindex(const char *netdev_name)
4455 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4456 COVERAGE_INC(netdev_get_ifindex);
4457 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
4458 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
4459 netdev_name, strerror(errno));
4462 return ifr.ifr_ifindex;
4466 get_ifindex(const struct netdev *netdev_, int *ifindexp)
4468 struct netdev_dev_linux *netdev_dev =
4469 netdev_dev_linux_cast(netdev_get_dev(netdev_));
4471 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
4472 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
4475 netdev_dev->get_ifindex_error = -ifindex;
4476 netdev_dev->ifindex = 0;
4478 netdev_dev->get_ifindex_error = 0;
4479 netdev_dev->ifindex = ifindex;
4481 netdev_dev->cache_valid |= VALID_IFINDEX;
4484 *ifindexp = netdev_dev->ifindex;
4485 return netdev_dev->get_ifindex_error;
4489 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
4494 memset(&ifr, 0, sizeof ifr);
4495 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4496 COVERAGE_INC(netdev_get_hwaddr);
4497 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
4498 /* ENODEV probably means that a vif disappeared asynchronously and
4499 * hasn't been removed from the database yet, so reduce the log level
4500 * to INFO for that case. */
4501 VLOG(errno == ENODEV ? VLL_INFO : VLL_ERR,
4502 "ioctl(SIOCGIFHWADDR) on %s device failed: %s",
4503 netdev_name, strerror(errno));
4506 hwaddr_family = ifr.ifr_hwaddr.sa_family;
4507 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
4508 VLOG_WARN("%s device has unknown hardware address family %d",
4509 netdev_name, hwaddr_family);
4511 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
4516 set_etheraddr(const char *netdev_name,
4517 const uint8_t mac[ETH_ADDR_LEN])
4521 memset(&ifr, 0, sizeof ifr);
4522 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4523 ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
4524 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
4525 COVERAGE_INC(netdev_set_hwaddr);
4526 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
4527 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
4528 netdev_name, strerror(errno));
4535 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
4536 int cmd, const char *cmd_name)
4540 memset(&ifr, 0, sizeof ifr);
4541 ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
4542 ifr.ifr_data = (caddr_t) ecmd;
4545 COVERAGE_INC(netdev_ethtool);
4546 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
4549 if (errno != EOPNOTSUPP) {
4550 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
4551 "failed: %s", cmd_name, name, strerror(errno));
4553 /* The device doesn't support this operation. That's pretty
4554 * common, so there's no point in logging anything. */
4561 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
4562 const char *cmd_name)
4564 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
4565 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
4566 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
4574 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
4575 int cmd, const char *cmd_name)
4580 ifr.ifr_addr.sa_family = AF_INET;
4581 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
4583 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
4584 *ip = sin->sin_addr;
4589 /* Returns an AF_PACKET raw socket or a negative errno value. */
4591 af_packet_sock(void)
4593 static int sock = INT_MIN;
4595 if (sock == INT_MIN) {
4596 sock = socket(AF_PACKET, SOCK_RAW, 0);
4598 set_nonblocking(sock);
4601 VLOG_ERR("failed to create packet socket: %s", strerror(errno));