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>
30 #include <linux/types.h>
31 #include <linux/ethtool.h>
32 #include <linux/mii.h>
33 #include <linux/pkt_cls.h>
34 #include <linux/pkt_sched.h>
35 #include <linux/rtnetlink.h>
36 #include <linux/sockios.h>
37 #include <linux/version.h>
38 #include <sys/types.h>
39 #include <sys/ioctl.h>
40 #include <sys/socket.h>
41 #include <netpacket/packet.h>
43 #include <net/if_arp.h>
44 #include <net/if_packet.h>
45 #include <net/route.h>
46 #include <netinet/in.h>
53 #include "dpif-linux.h"
54 #include "dynamic-string.h"
55 #include "fatal-signal.h"
58 #include "netdev-provider.h"
59 #include "netdev-vport.h"
61 #include "netlink-notifier.h"
62 #include "netlink-socket.h"
64 #include "openflow/openflow.h"
66 #include "poll-loop.h"
67 #include "rtnetlink-link.h"
68 #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 (!strcmp(netdev_dev_get_type(netdev_dev_), "tap") &&
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 && strcmp(netdev_get_type(netdev_), "tap")) {
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 (!strcmp(netdev_get_type(netdev_), "tap")) {
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 (strcmp(netdev_get_type(netdev_), "tap")) {
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);
1784 static const struct tc_ops *
1785 tc_lookup_ovs_name(const char *name)
1787 const struct tc_ops **opsp;
1789 for (opsp = tcs; *opsp != NULL; opsp++) {
1790 const struct tc_ops *ops = *opsp;
1791 if (!strcmp(name, ops->ovs_name)) {
1798 static const struct tc_ops *
1799 tc_lookup_linux_name(const char *name)
1801 const struct tc_ops **opsp;
1803 for (opsp = tcs; *opsp != NULL; opsp++) {
1804 const struct tc_ops *ops = *opsp;
1805 if (ops->linux_name && !strcmp(name, ops->linux_name)) {
1812 static struct tc_queue *
1813 tc_find_queue__(const struct netdev *netdev, unsigned int queue_id,
1816 struct netdev_dev_linux *netdev_dev =
1817 netdev_dev_linux_cast(netdev_get_dev(netdev));
1818 struct tc_queue *queue;
1820 HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev_dev->tc->queues) {
1821 if (queue->queue_id == queue_id) {
1828 static struct tc_queue *
1829 tc_find_queue(const struct netdev *netdev, unsigned int queue_id)
1831 return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0));
1835 netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED,
1837 struct netdev_qos_capabilities *caps)
1839 const struct tc_ops *ops = tc_lookup_ovs_name(type);
1843 caps->n_queues = ops->n_queues;
1848 netdev_linux_get_qos(const struct netdev *netdev,
1849 const char **typep, struct shash *details)
1851 struct netdev_dev_linux *netdev_dev =
1852 netdev_dev_linux_cast(netdev_get_dev(netdev));
1855 error = tc_query_qdisc(netdev);
1860 *typep = netdev_dev->tc->ops->ovs_name;
1861 return (netdev_dev->tc->ops->qdisc_get
1862 ? netdev_dev->tc->ops->qdisc_get(netdev, details)
1867 netdev_linux_set_qos(struct netdev *netdev,
1868 const char *type, const struct shash *details)
1870 struct netdev_dev_linux *netdev_dev =
1871 netdev_dev_linux_cast(netdev_get_dev(netdev));
1872 const struct tc_ops *new_ops;
1875 new_ops = tc_lookup_ovs_name(type);
1876 if (!new_ops || !new_ops->tc_install) {
1880 error = tc_query_qdisc(netdev);
1885 if (new_ops == netdev_dev->tc->ops) {
1886 return new_ops->qdisc_set ? new_ops->qdisc_set(netdev, details) : 0;
1888 /* Delete existing qdisc. */
1889 error = tc_del_qdisc(netdev);
1893 assert(netdev_dev->tc == NULL);
1895 /* Install new qdisc. */
1896 error = new_ops->tc_install(netdev, details);
1897 assert((error == 0) == (netdev_dev->tc != NULL));
1904 netdev_linux_get_queue(const struct netdev *netdev,
1905 unsigned int queue_id, struct shash *details)
1907 struct netdev_dev_linux *netdev_dev =
1908 netdev_dev_linux_cast(netdev_get_dev(netdev));
1911 error = tc_query_qdisc(netdev);
1915 struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1917 ? netdev_dev->tc->ops->class_get(netdev, queue, details)
1923 netdev_linux_set_queue(struct netdev *netdev,
1924 unsigned int queue_id, const struct shash *details)
1926 struct netdev_dev_linux *netdev_dev =
1927 netdev_dev_linux_cast(netdev_get_dev(netdev));
1930 error = tc_query_qdisc(netdev);
1933 } else if (queue_id >= netdev_dev->tc->ops->n_queues
1934 || !netdev_dev->tc->ops->class_set) {
1938 return netdev_dev->tc->ops->class_set(netdev, queue_id, details);
1942 netdev_linux_delete_queue(struct netdev *netdev, unsigned int queue_id)
1944 struct netdev_dev_linux *netdev_dev =
1945 netdev_dev_linux_cast(netdev_get_dev(netdev));
1948 error = tc_query_qdisc(netdev);
1951 } else if (!netdev_dev->tc->ops->class_delete) {
1954 struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1956 ? netdev_dev->tc->ops->class_delete(netdev, queue)
1962 netdev_linux_get_queue_stats(const struct netdev *netdev,
1963 unsigned int queue_id,
1964 struct netdev_queue_stats *stats)
1966 struct netdev_dev_linux *netdev_dev =
1967 netdev_dev_linux_cast(netdev_get_dev(netdev));
1970 error = tc_query_qdisc(netdev);
1973 } else if (!netdev_dev->tc->ops->class_get_stats) {
1976 const struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1978 ? netdev_dev->tc->ops->class_get_stats(netdev, queue, stats)
1984 start_queue_dump(const struct netdev *netdev, struct nl_dump *dump)
1986 struct ofpbuf request;
1987 struct tcmsg *tcmsg;
1989 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request);
1993 tcmsg->tcm_parent = 0;
1994 nl_dump_start(dump, rtnl_sock, &request);
1995 ofpbuf_uninit(&request);
2000 netdev_linux_dump_queues(const struct netdev *netdev,
2001 netdev_dump_queues_cb *cb, void *aux)
2003 struct netdev_dev_linux *netdev_dev =
2004 netdev_dev_linux_cast(netdev_get_dev(netdev));
2005 struct tc_queue *queue, *next_queue;
2006 struct shash details;
2010 error = tc_query_qdisc(netdev);
2013 } else if (!netdev_dev->tc->ops->class_get) {
2018 shash_init(&details);
2019 HMAP_FOR_EACH_SAFE (queue, next_queue, hmap_node,
2020 &netdev_dev->tc->queues) {
2021 shash_clear(&details);
2023 error = netdev_dev->tc->ops->class_get(netdev, queue, &details);
2025 (*cb)(queue->queue_id, &details, aux);
2030 shash_destroy(&details);
2036 netdev_linux_dump_queue_stats(const struct netdev *netdev,
2037 netdev_dump_queue_stats_cb *cb, void *aux)
2039 struct netdev_dev_linux *netdev_dev =
2040 netdev_dev_linux_cast(netdev_get_dev(netdev));
2041 struct nl_dump dump;
2046 error = tc_query_qdisc(netdev);
2049 } else if (!netdev_dev->tc->ops->class_dump_stats) {
2054 if (!start_queue_dump(netdev, &dump)) {
2057 while (nl_dump_next(&dump, &msg)) {
2058 error = netdev_dev->tc->ops->class_dump_stats(netdev, &msg, cb, aux);
2064 error = nl_dump_done(&dump);
2065 return error ? error : last_error;
2069 netdev_linux_get_in4(const struct netdev *netdev_,
2070 struct in_addr *address, struct in_addr *netmask)
2072 struct netdev_dev_linux *netdev_dev =
2073 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2075 if (!(netdev_dev->cache_valid & VALID_IN4)) {
2078 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
2079 SIOCGIFADDR, "SIOCGIFADDR");
2084 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
2085 SIOCGIFNETMASK, "SIOCGIFNETMASK");
2090 netdev_dev->cache_valid |= VALID_IN4;
2092 *address = netdev_dev->address;
2093 *netmask = netdev_dev->netmask;
2094 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
2098 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
2099 struct in_addr netmask)
2101 struct netdev_dev_linux *netdev_dev =
2102 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2105 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
2107 netdev_dev->cache_valid |= VALID_IN4;
2108 netdev_dev->address = address;
2109 netdev_dev->netmask = netmask;
2110 if (address.s_addr != INADDR_ANY) {
2111 error = do_set_addr(netdev_, SIOCSIFNETMASK,
2112 "SIOCSIFNETMASK", netmask);
2119 parse_if_inet6_line(const char *line,
2120 struct in6_addr *in6, char ifname[16 + 1])
2122 uint8_t *s6 = in6->s6_addr;
2123 #define X8 "%2"SCNx8
2125 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
2126 "%*x %*x %*x %*x %16s\n",
2127 &s6[0], &s6[1], &s6[2], &s6[3],
2128 &s6[4], &s6[5], &s6[6], &s6[7],
2129 &s6[8], &s6[9], &s6[10], &s6[11],
2130 &s6[12], &s6[13], &s6[14], &s6[15],
2134 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
2135 * 'in6' is non-null) and returns true. Otherwise, returns false. */
2137 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
2139 struct netdev_dev_linux *netdev_dev =
2140 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2141 if (!(netdev_dev->cache_valid & VALID_IN6)) {
2145 netdev_dev->in6 = in6addr_any;
2147 file = fopen("/proc/net/if_inet6", "r");
2149 const char *name = netdev_get_name(netdev_);
2150 while (fgets(line, sizeof line, file)) {
2151 struct in6_addr in6_tmp;
2152 char ifname[16 + 1];
2153 if (parse_if_inet6_line(line, &in6_tmp, ifname)
2154 && !strcmp(name, ifname))
2156 netdev_dev->in6 = in6_tmp;
2162 netdev_dev->cache_valid |= VALID_IN6;
2164 *in6 = netdev_dev->in6;
2169 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
2171 struct sockaddr_in sin;
2172 memset(&sin, 0, sizeof sin);
2173 sin.sin_family = AF_INET;
2174 sin.sin_addr = addr;
2177 memset(sa, 0, sizeof *sa);
2178 memcpy(sa, &sin, sizeof sin);
2182 do_set_addr(struct netdev *netdev,
2183 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
2186 ovs_strzcpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
2187 make_in4_sockaddr(&ifr.ifr_addr, addr);
2189 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
2193 /* Adds 'router' as a default IP gateway. */
2195 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
2197 struct in_addr any = { INADDR_ANY };
2201 memset(&rt, 0, sizeof rt);
2202 make_in4_sockaddr(&rt.rt_dst, any);
2203 make_in4_sockaddr(&rt.rt_gateway, router);
2204 make_in4_sockaddr(&rt.rt_genmask, any);
2205 rt.rt_flags = RTF_UP | RTF_GATEWAY;
2206 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
2208 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
2214 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
2217 static const char fn[] = "/proc/net/route";
2222 *netdev_name = NULL;
2223 stream = fopen(fn, "r");
2224 if (stream == NULL) {
2225 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
2230 while (fgets(line, sizeof line, stream)) {
2233 ovs_be32 dest, gateway, mask;
2234 int refcnt, metric, mtu;
2235 unsigned int flags, use, window, irtt;
2238 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
2240 iface, &dest, &gateway, &flags, &refcnt,
2241 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
2243 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
2247 if (!(flags & RTF_UP)) {
2248 /* Skip routes that aren't up. */
2252 /* The output of 'dest', 'mask', and 'gateway' were given in
2253 * network byte order, so we don't need need any endian
2254 * conversions here. */
2255 if ((dest & mask) == (host->s_addr & mask)) {
2257 /* The host is directly reachable. */
2258 next_hop->s_addr = 0;
2260 /* To reach the host, we must go through a gateway. */
2261 next_hop->s_addr = gateway;
2263 *netdev_name = xstrdup(iface);
2275 netdev_linux_get_status(const struct netdev *netdev, struct shash *sh)
2278 struct netdev_dev_linux *netdev_dev =
2279 netdev_dev_linux_cast(netdev_get_dev(netdev));
2281 error = netdev_linux_get_drvinfo(netdev_dev);
2283 shash_add(sh, "driver_name", xstrdup(netdev_dev->drvinfo.driver));
2284 shash_add(sh, "driver_version", xstrdup(netdev_dev->drvinfo.version));
2285 shash_add(sh, "firmware_version", xstrdup(netdev_dev->drvinfo.fw_version));
2291 netdev_internal_get_status(const struct netdev *netdev OVS_UNUSED, struct shash *sh)
2293 shash_add(sh, "driver_name", xstrdup("openvswitch"));
2297 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
2298 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
2299 * returns 0. Otherwise, it returns a positive errno value; in particular,
2300 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
2302 netdev_linux_arp_lookup(const struct netdev *netdev,
2303 ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
2306 struct sockaddr_in sin;
2309 memset(&r, 0, sizeof r);
2310 memset(&sin, 0, sizeof sin);
2311 sin.sin_family = AF_INET;
2312 sin.sin_addr.s_addr = ip;
2314 memcpy(&r.arp_pa, &sin, sizeof sin);
2315 r.arp_ha.sa_family = ARPHRD_ETHER;
2317 ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
2318 COVERAGE_INC(netdev_arp_lookup);
2319 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
2321 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
2322 } else if (retval != ENXIO) {
2323 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
2324 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
2330 nd_to_iff_flags(enum netdev_flags nd)
2333 if (nd & NETDEV_UP) {
2336 if (nd & NETDEV_PROMISC) {
2343 iff_to_nd_flags(int iff)
2345 enum netdev_flags nd = 0;
2349 if (iff & IFF_PROMISC) {
2350 nd |= NETDEV_PROMISC;
2356 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
2357 enum netdev_flags on, enum netdev_flags *old_flagsp)
2359 struct netdev_dev_linux *netdev_dev;
2360 int old_flags, new_flags;
2363 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
2364 old_flags = netdev_dev->ifi_flags;
2365 *old_flagsp = iff_to_nd_flags(old_flags);
2366 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
2367 if (new_flags != old_flags) {
2368 error = set_flags(netdev, new_flags);
2369 get_flags(&netdev_dev->netdev_dev, &netdev_dev->ifi_flags);
2375 netdev_linux_change_seq(const struct netdev *netdev)
2377 return netdev_dev_linux_cast(netdev_get_dev(netdev))->change_seq;
2380 #define NETDEV_LINUX_CLASS(NAME, CREATE, GET_STATS, SET_STATS, \
2381 GET_FEATURES, GET_STATUS) \
2385 netdev_linux_init, \
2387 netdev_linux_wait, \
2390 netdev_linux_destroy, \
2391 NULL, /* get_config */ \
2392 NULL, /* set_config */ \
2394 netdev_linux_open, \
2395 netdev_linux_close, \
2397 netdev_linux_listen, \
2398 netdev_linux_recv, \
2399 netdev_linux_recv_wait, \
2400 netdev_linux_drain, \
2402 netdev_linux_send, \
2403 netdev_linux_send_wait, \
2405 netdev_linux_set_etheraddr, \
2406 netdev_linux_get_etheraddr, \
2407 netdev_linux_get_mtu, \
2408 netdev_linux_set_mtu, \
2409 netdev_linux_get_ifindex, \
2410 netdev_linux_get_carrier, \
2411 netdev_linux_get_carrier_resets, \
2412 netdev_linux_set_miimon_interval, \
2417 netdev_linux_set_advertisements, \
2419 netdev_linux_set_policing, \
2420 netdev_linux_get_qos_types, \
2421 netdev_linux_get_qos_capabilities, \
2422 netdev_linux_get_qos, \
2423 netdev_linux_set_qos, \
2424 netdev_linux_get_queue, \
2425 netdev_linux_set_queue, \
2426 netdev_linux_delete_queue, \
2427 netdev_linux_get_queue_stats, \
2428 netdev_linux_dump_queues, \
2429 netdev_linux_dump_queue_stats, \
2431 netdev_linux_get_in4, \
2432 netdev_linux_set_in4, \
2433 netdev_linux_get_in6, \
2434 netdev_linux_add_router, \
2435 netdev_linux_get_next_hop, \
2437 netdev_linux_arp_lookup, \
2439 netdev_linux_update_flags, \
2441 netdev_linux_change_seq \
2444 const struct netdev_class netdev_linux_class =
2447 netdev_linux_create,
2448 netdev_linux_get_stats,
2449 NULL, /* set_stats */
2450 netdev_linux_get_features,
2451 netdev_linux_get_status);
2453 const struct netdev_class netdev_tap_class =
2456 netdev_linux_create_tap,
2457 netdev_tap_get_stats,
2458 NULL, /* set_stats */
2459 netdev_linux_get_features,
2460 netdev_linux_get_status);
2462 const struct netdev_class netdev_internal_class =
2465 netdev_linux_create,
2466 netdev_internal_get_stats,
2467 netdev_vport_set_stats,
2468 NULL, /* get_features */
2469 netdev_internal_get_status);
2471 /* HTB traffic control class. */
2473 #define HTB_N_QUEUES 0xf000
2477 unsigned int max_rate; /* In bytes/s. */
2481 struct tc_queue tc_queue;
2482 unsigned int min_rate; /* In bytes/s. */
2483 unsigned int max_rate; /* In bytes/s. */
2484 unsigned int burst; /* In bytes. */
2485 unsigned int priority; /* Lower values are higher priorities. */
2489 htb_get__(const struct netdev *netdev)
2491 struct netdev_dev_linux *netdev_dev =
2492 netdev_dev_linux_cast(netdev_get_dev(netdev));
2493 return CONTAINER_OF(netdev_dev->tc, struct htb, tc);
2497 htb_install__(struct netdev *netdev, uint64_t max_rate)
2499 struct netdev_dev_linux *netdev_dev =
2500 netdev_dev_linux_cast(netdev_get_dev(netdev));
2503 htb = xmalloc(sizeof *htb);
2504 tc_init(&htb->tc, &tc_ops_htb);
2505 htb->max_rate = max_rate;
2507 netdev_dev->tc = &htb->tc;
2510 /* Create an HTB qdisc.
2512 * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */
2514 htb_setup_qdisc__(struct netdev *netdev)
2517 struct tc_htb_glob opt;
2518 struct ofpbuf request;
2519 struct tcmsg *tcmsg;
2521 tc_del_qdisc(netdev);
2523 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
2524 NLM_F_EXCL | NLM_F_CREATE, &request);
2528 tcmsg->tcm_handle = tc_make_handle(1, 0);
2529 tcmsg->tcm_parent = TC_H_ROOT;
2531 nl_msg_put_string(&request, TCA_KIND, "htb");
2533 memset(&opt, 0, sizeof opt);
2534 opt.rate2quantum = 10;
2538 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2539 nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt);
2540 nl_msg_end_nested(&request, opt_offset);
2542 return tc_transact(&request, NULL);
2545 /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb
2546 * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */
2548 htb_setup_class__(struct netdev *netdev, unsigned int handle,
2549 unsigned int parent, struct htb_class *class)
2552 struct tc_htb_opt opt;
2553 struct ofpbuf request;
2554 struct tcmsg *tcmsg;
2558 error = netdev_get_mtu(netdev, &mtu);
2560 VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU",
2561 netdev_get_name(netdev));
2565 memset(&opt, 0, sizeof opt);
2566 tc_fill_rate(&opt.rate, class->min_rate, mtu);
2567 tc_fill_rate(&opt.ceil, class->max_rate, mtu);
2568 opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst);
2569 opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst);
2570 opt.prio = class->priority;
2572 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
2576 tcmsg->tcm_handle = handle;
2577 tcmsg->tcm_parent = parent;
2579 nl_msg_put_string(&request, TCA_KIND, "htb");
2580 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2581 nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt);
2582 tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate);
2583 tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil);
2584 nl_msg_end_nested(&request, opt_offset);
2586 error = tc_transact(&request, NULL);
2588 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
2589 "min_rate=%u max_rate=%u burst=%u prio=%u (%s)",
2590 netdev_get_name(netdev),
2591 tc_get_major(handle), tc_get_minor(handle),
2592 tc_get_major(parent), tc_get_minor(parent),
2593 class->min_rate, class->max_rate,
2594 class->burst, class->priority, strerror(error));
2599 /* Parses Netlink attributes in 'options' for HTB parameters and stores a
2600 * description of them into 'details'. The description complies with the
2601 * specification given in the vswitch database documentation for linux-htb
2604 htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class)
2606 static const struct nl_policy tca_htb_policy[] = {
2607 [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false,
2608 .min_len = sizeof(struct tc_htb_opt) },
2611 struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)];
2612 const struct tc_htb_opt *htb;
2614 if (!nl_parse_nested(nl_options, tca_htb_policy,
2615 attrs, ARRAY_SIZE(tca_htb_policy))) {
2616 VLOG_WARN_RL(&rl, "failed to parse HTB class options");
2620 htb = nl_attr_get(attrs[TCA_HTB_PARMS]);
2621 class->min_rate = htb->rate.rate;
2622 class->max_rate = htb->ceil.rate;
2623 class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer);
2624 class->priority = htb->prio;
2629 htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
2630 struct htb_class *options,
2631 struct netdev_queue_stats *stats)
2633 struct nlattr *nl_options;
2634 unsigned int handle;
2637 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
2638 if (!error && queue_id) {
2639 unsigned int major = tc_get_major(handle);
2640 unsigned int minor = tc_get_minor(handle);
2641 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2642 *queue_id = minor - 1;
2647 if (!error && options) {
2648 error = htb_parse_tca_options__(nl_options, options);
2654 htb_parse_qdisc_details__(struct netdev *netdev,
2655 const struct shash *details, struct htb_class *hc)
2657 const char *max_rate_s;
2659 max_rate_s = shash_find_data(details, "max-rate");
2660 hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
2661 if (!hc->max_rate) {
2662 enum netdev_features current;
2664 netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
2665 hc->max_rate = netdev_features_to_bps(current) / 8;
2667 hc->min_rate = hc->max_rate;
2673 htb_parse_class_details__(struct netdev *netdev,
2674 const struct shash *details, struct htb_class *hc)
2676 const struct htb *htb = htb_get__(netdev);
2677 const char *min_rate_s = shash_find_data(details, "min-rate");
2678 const char *max_rate_s = shash_find_data(details, "max-rate");
2679 const char *burst_s = shash_find_data(details, "burst");
2680 const char *priority_s = shash_find_data(details, "priority");
2683 error = netdev_get_mtu(netdev, &mtu);
2685 VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU",
2686 netdev_get_name(netdev));
2690 /* HTB requires at least an mtu sized min-rate to send any traffic even
2691 * on uncongested links. */
2692 hc->min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
2693 hc->min_rate = MAX(hc->min_rate, mtu);
2694 hc->min_rate = MIN(hc->min_rate, htb->max_rate);
2697 hc->max_rate = (max_rate_s
2698 ? strtoull(max_rate_s, NULL, 10) / 8
2700 hc->max_rate = MAX(hc->max_rate, hc->min_rate);
2701 hc->max_rate = MIN(hc->max_rate, htb->max_rate);
2705 * According to hints in the documentation that I've read, it is important
2706 * that 'burst' be at least as big as the largest frame that might be
2707 * transmitted. Also, making 'burst' a bit bigger than necessary is OK,
2708 * but having it a bit too small is a problem. Since netdev_get_mtu()
2709 * doesn't include the Ethernet header, we need to add at least 14 (18?) to
2710 * the MTU. We actually add 64, instead of 14, as a guard against
2711 * additional headers get tacked on somewhere that we're not aware of. */
2712 hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0;
2713 hc->burst = MAX(hc->burst, mtu + 64);
2716 hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0;
2722 htb_query_class__(const struct netdev *netdev, unsigned int handle,
2723 unsigned int parent, struct htb_class *options,
2724 struct netdev_queue_stats *stats)
2726 struct ofpbuf *reply;
2729 error = tc_query_class(netdev, handle, parent, &reply);
2731 error = htb_parse_tcmsg__(reply, NULL, options, stats);
2732 ofpbuf_delete(reply);
2738 htb_tc_install(struct netdev *netdev, const struct shash *details)
2742 error = htb_setup_qdisc__(netdev);
2744 struct htb_class hc;
2746 htb_parse_qdisc_details__(netdev, details, &hc);
2747 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
2748 tc_make_handle(1, 0), &hc);
2750 htb_install__(netdev, hc.max_rate);
2756 static struct htb_class *
2757 htb_class_cast__(const struct tc_queue *queue)
2759 return CONTAINER_OF(queue, struct htb_class, tc_queue);
2763 htb_update_queue__(struct netdev *netdev, unsigned int queue_id,
2764 const struct htb_class *hc)
2766 struct htb *htb = htb_get__(netdev);
2767 size_t hash = hash_int(queue_id, 0);
2768 struct tc_queue *queue;
2769 struct htb_class *hcp;
2771 queue = tc_find_queue__(netdev, queue_id, hash);
2773 hcp = htb_class_cast__(queue);
2775 hcp = xmalloc(sizeof *hcp);
2776 queue = &hcp->tc_queue;
2777 queue->queue_id = queue_id;
2778 hmap_insert(&htb->tc.queues, &queue->hmap_node, hash);
2781 hcp->min_rate = hc->min_rate;
2782 hcp->max_rate = hc->max_rate;
2783 hcp->burst = hc->burst;
2784 hcp->priority = hc->priority;
2788 htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
2791 struct nl_dump dump;
2792 struct htb_class hc;
2794 /* Get qdisc options. */
2796 htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
2797 htb_install__(netdev, hc.max_rate);
2800 if (!start_queue_dump(netdev, &dump)) {
2803 while (nl_dump_next(&dump, &msg)) {
2804 unsigned int queue_id;
2806 if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
2807 htb_update_queue__(netdev, queue_id, &hc);
2810 nl_dump_done(&dump);
2816 htb_tc_destroy(struct tc *tc)
2818 struct htb *htb = CONTAINER_OF(tc, struct htb, tc);
2819 struct htb_class *hc, *next;
2821 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) {
2822 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
2830 htb_qdisc_get(const struct netdev *netdev, struct shash *details)
2832 const struct htb *htb = htb_get__(netdev);
2833 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * htb->max_rate));
2838 htb_qdisc_set(struct netdev *netdev, const struct shash *details)
2840 struct htb_class hc;
2843 htb_parse_qdisc_details__(netdev, details, &hc);
2844 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
2845 tc_make_handle(1, 0), &hc);
2847 htb_get__(netdev)->max_rate = hc.max_rate;
2853 htb_class_get(const struct netdev *netdev OVS_UNUSED,
2854 const struct tc_queue *queue, struct shash *details)
2856 const struct htb_class *hc = htb_class_cast__(queue);
2858 shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate));
2859 if (hc->min_rate != hc->max_rate) {
2860 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate));
2862 shash_add(details, "burst", xasprintf("%llu", 8ULL * hc->burst));
2864 shash_add(details, "priority", xasprintf("%u", hc->priority));
2870 htb_class_set(struct netdev *netdev, unsigned int queue_id,
2871 const struct shash *details)
2873 struct htb_class hc;
2876 error = htb_parse_class_details__(netdev, details, &hc);
2881 error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
2882 tc_make_handle(1, 0xfffe), &hc);
2887 htb_update_queue__(netdev, queue_id, &hc);
2892 htb_class_delete(struct netdev *netdev, struct tc_queue *queue)
2894 struct htb_class *hc = htb_class_cast__(queue);
2895 struct htb *htb = htb_get__(netdev);
2898 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
2900 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
2907 htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
2908 struct netdev_queue_stats *stats)
2910 return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
2911 tc_make_handle(1, 0xfffe), NULL, stats);
2915 htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
2916 const struct ofpbuf *nlmsg,
2917 netdev_dump_queue_stats_cb *cb, void *aux)
2919 struct netdev_queue_stats stats;
2920 unsigned int handle, major, minor;
2923 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
2928 major = tc_get_major(handle);
2929 minor = tc_get_minor(handle);
2930 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2931 (*cb)(minor - 1, &stats, aux);
2936 static const struct tc_ops tc_ops_htb = {
2937 "htb", /* linux_name */
2938 "linux-htb", /* ovs_name */
2939 HTB_N_QUEUES, /* n_queues */
2948 htb_class_get_stats,
2949 htb_class_dump_stats
2952 /* "linux-hfsc" traffic control class. */
2954 #define HFSC_N_QUEUES 0xf000
2962 struct tc_queue tc_queue;
2967 static struct hfsc *
2968 hfsc_get__(const struct netdev *netdev)
2970 struct netdev_dev_linux *netdev_dev;
2971 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
2972 return CONTAINER_OF(netdev_dev->tc, struct hfsc, tc);
2975 static struct hfsc_class *
2976 hfsc_class_cast__(const struct tc_queue *queue)
2978 return CONTAINER_OF(queue, struct hfsc_class, tc_queue);
2982 hfsc_install__(struct netdev *netdev, uint32_t max_rate)
2984 struct netdev_dev_linux * netdev_dev;
2987 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
2988 hfsc = xmalloc(sizeof *hfsc);
2989 tc_init(&hfsc->tc, &tc_ops_hfsc);
2990 hfsc->max_rate = max_rate;
2991 netdev_dev->tc = &hfsc->tc;
2995 hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id,
2996 const struct hfsc_class *hc)
3000 struct hfsc_class *hcp;
3001 struct tc_queue *queue;
3003 hfsc = hfsc_get__(netdev);
3004 hash = hash_int(queue_id, 0);
3006 queue = tc_find_queue__(netdev, queue_id, hash);
3008 hcp = hfsc_class_cast__(queue);
3010 hcp = xmalloc(sizeof *hcp);
3011 queue = &hcp->tc_queue;
3012 queue->queue_id = queue_id;
3013 hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash);
3016 hcp->min_rate = hc->min_rate;
3017 hcp->max_rate = hc->max_rate;
3021 hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class)
3023 const struct tc_service_curve *rsc, *fsc, *usc;
3024 static const struct nl_policy tca_hfsc_policy[] = {
3026 .type = NL_A_UNSPEC,
3028 .min_len = sizeof(struct tc_service_curve),
3031 .type = NL_A_UNSPEC,
3033 .min_len = sizeof(struct tc_service_curve),
3036 .type = NL_A_UNSPEC,
3038 .min_len = sizeof(struct tc_service_curve),
3041 struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)];
3043 if (!nl_parse_nested(nl_options, tca_hfsc_policy,
3044 attrs, ARRAY_SIZE(tca_hfsc_policy))) {
3045 VLOG_WARN_RL(&rl, "failed to parse HFSC class options");
3049 rsc = nl_attr_get(attrs[TCA_HFSC_RSC]);
3050 fsc = nl_attr_get(attrs[TCA_HFSC_FSC]);
3051 usc = nl_attr_get(attrs[TCA_HFSC_USC]);
3053 if (rsc->m1 != 0 || rsc->d != 0 ||
3054 fsc->m1 != 0 || fsc->d != 0 ||
3055 usc->m1 != 0 || usc->d != 0) {
3056 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3057 "Non-linear service curves are not supported.");
3061 if (rsc->m2 != fsc->m2) {
3062 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3063 "Real-time service curves are not supported ");
3067 if (rsc->m2 > usc->m2) {
3068 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3069 "Min-rate service curve is greater than "
3070 "the max-rate service curve.");
3074 class->min_rate = fsc->m2;
3075 class->max_rate = usc->m2;
3080 hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
3081 struct hfsc_class *options,
3082 struct netdev_queue_stats *stats)
3085 unsigned int handle;
3086 struct nlattr *nl_options;
3088 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
3094 unsigned int major, minor;
3096 major = tc_get_major(handle);
3097 minor = tc_get_minor(handle);
3098 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
3099 *queue_id = minor - 1;
3106 error = hfsc_parse_tca_options__(nl_options, options);
3113 hfsc_query_class__(const struct netdev *netdev, unsigned int handle,
3114 unsigned int parent, struct hfsc_class *options,
3115 struct netdev_queue_stats *stats)
3118 struct ofpbuf *reply;
3120 error = tc_query_class(netdev, handle, parent, &reply);
3125 error = hfsc_parse_tcmsg__(reply, NULL, options, stats);
3126 ofpbuf_delete(reply);
3131 hfsc_parse_qdisc_details__(struct netdev *netdev, const struct shash *details,
3132 struct hfsc_class *class)
3135 const char *max_rate_s;
3137 max_rate_s = shash_find_data(details, "max-rate");
3138 max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
3141 enum netdev_features current;
3143 netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
3144 max_rate = netdev_features_to_bps(current) / 8;
3147 class->min_rate = max_rate;
3148 class->max_rate = max_rate;
3152 hfsc_parse_class_details__(struct netdev *netdev,
3153 const struct shash *details,
3154 struct hfsc_class * class)
3156 const struct hfsc *hfsc;
3157 uint32_t min_rate, max_rate;
3158 const char *min_rate_s, *max_rate_s;
3160 hfsc = hfsc_get__(netdev);
3161 min_rate_s = shash_find_data(details, "min-rate");
3162 max_rate_s = shash_find_data(details, "max-rate");
3164 min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
3165 min_rate = MAX(min_rate, 1);
3166 min_rate = MIN(min_rate, hfsc->max_rate);
3168 max_rate = (max_rate_s
3169 ? strtoull(max_rate_s, NULL, 10) / 8
3171 max_rate = MAX(max_rate, min_rate);
3172 max_rate = MIN(max_rate, hfsc->max_rate);
3174 class->min_rate = min_rate;
3175 class->max_rate = max_rate;
3180 /* Create an HFSC qdisc.
3182 * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */
3184 hfsc_setup_qdisc__(struct netdev * netdev)
3186 struct tcmsg *tcmsg;
3187 struct ofpbuf request;
3188 struct tc_hfsc_qopt opt;
3190 tc_del_qdisc(netdev);
3192 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
3193 NLM_F_EXCL | NLM_F_CREATE, &request);
3199 tcmsg->tcm_handle = tc_make_handle(1, 0);
3200 tcmsg->tcm_parent = TC_H_ROOT;
3202 memset(&opt, 0, sizeof opt);
3205 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3206 nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt);
3208 return tc_transact(&request, NULL);
3211 /* Create an HFSC class.
3213 * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc
3214 * sc rate <min_rate> ul rate <max_rate>" */
3216 hfsc_setup_class__(struct netdev *netdev, unsigned int handle,
3217 unsigned int parent, struct hfsc_class *class)
3221 struct tcmsg *tcmsg;
3222 struct ofpbuf request;
3223 struct tc_service_curve min, max;
3225 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
3231 tcmsg->tcm_handle = handle;
3232 tcmsg->tcm_parent = parent;
3236 min.m2 = class->min_rate;
3240 max.m2 = class->max_rate;
3242 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3243 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
3244 nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min);
3245 nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min);
3246 nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max);
3247 nl_msg_end_nested(&request, opt_offset);
3249 error = tc_transact(&request, NULL);
3251 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
3252 "min-rate %ubps, max-rate %ubps (%s)",
3253 netdev_get_name(netdev),
3254 tc_get_major(handle), tc_get_minor(handle),
3255 tc_get_major(parent), tc_get_minor(parent),
3256 class->min_rate, class->max_rate, strerror(error));
3263 hfsc_tc_install(struct netdev *netdev, const struct shash *details)
3266 struct hfsc_class class;
3268 error = hfsc_setup_qdisc__(netdev);
3274 hfsc_parse_qdisc_details__(netdev, details, &class);
3275 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3276 tc_make_handle(1, 0), &class);
3282 hfsc_install__(netdev, class.max_rate);
3287 hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3290 struct nl_dump dump;
3291 struct hfsc_class hc;
3294 hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
3295 hfsc_install__(netdev, hc.max_rate);
3297 if (!start_queue_dump(netdev, &dump)) {
3301 while (nl_dump_next(&dump, &msg)) {
3302 unsigned int queue_id;
3304 if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
3305 hfsc_update_queue__(netdev, queue_id, &hc);
3309 nl_dump_done(&dump);
3314 hfsc_tc_destroy(struct tc *tc)
3317 struct hfsc_class *hc, *next;
3319 hfsc = CONTAINER_OF(tc, struct hfsc, tc);
3321 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) {
3322 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3331 hfsc_qdisc_get(const struct netdev *netdev, struct shash *details)
3333 const struct hfsc *hfsc;
3334 hfsc = hfsc_get__(netdev);
3335 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hfsc->max_rate));
3340 hfsc_qdisc_set(struct netdev *netdev, const struct shash *details)
3343 struct hfsc_class class;
3345 hfsc_parse_qdisc_details__(netdev, details, &class);
3346 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3347 tc_make_handle(1, 0), &class);
3350 hfsc_get__(netdev)->max_rate = class.max_rate;
3357 hfsc_class_get(const struct netdev *netdev OVS_UNUSED,
3358 const struct tc_queue *queue, struct shash *details)
3360 const struct hfsc_class *hc;
3362 hc = hfsc_class_cast__(queue);
3363 shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate));
3364 if (hc->min_rate != hc->max_rate) {
3365 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate));
3371 hfsc_class_set(struct netdev *netdev, unsigned int queue_id,
3372 const struct shash *details)
3375 struct hfsc_class class;
3377 error = hfsc_parse_class_details__(netdev, details, &class);
3382 error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
3383 tc_make_handle(1, 0xfffe), &class);
3388 hfsc_update_queue__(netdev, queue_id, &class);
3393 hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue)
3397 struct hfsc_class *hc;
3399 hc = hfsc_class_cast__(queue);
3400 hfsc = hfsc_get__(netdev);
3402 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
3404 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3411 hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
3412 struct netdev_queue_stats *stats)
3414 return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
3415 tc_make_handle(1, 0xfffe), NULL, stats);
3419 hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
3420 const struct ofpbuf *nlmsg,
3421 netdev_dump_queue_stats_cb *cb, void *aux)
3423 struct netdev_queue_stats stats;
3424 unsigned int handle, major, minor;
3427 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
3432 major = tc_get_major(handle);
3433 minor = tc_get_minor(handle);
3434 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
3435 (*cb)(minor - 1, &stats, aux);
3440 static const struct tc_ops tc_ops_hfsc = {
3441 "hfsc", /* linux_name */
3442 "linux-hfsc", /* ovs_name */
3443 HFSC_N_QUEUES, /* n_queues */
3444 hfsc_tc_install, /* tc_install */
3445 hfsc_tc_load, /* tc_load */
3446 hfsc_tc_destroy, /* tc_destroy */
3447 hfsc_qdisc_get, /* qdisc_get */
3448 hfsc_qdisc_set, /* qdisc_set */
3449 hfsc_class_get, /* class_get */
3450 hfsc_class_set, /* class_set */
3451 hfsc_class_delete, /* class_delete */
3452 hfsc_class_get_stats, /* class_get_stats */
3453 hfsc_class_dump_stats /* class_dump_stats */
3456 /* "linux-default" traffic control class.
3458 * This class represents the default, unnamed Linux qdisc. It corresponds to
3459 * the "" (empty string) QoS type in the OVS database. */
3462 default_install__(struct netdev *netdev)
3464 struct netdev_dev_linux *netdev_dev =
3465 netdev_dev_linux_cast(netdev_get_dev(netdev));
3466 static struct tc *tc;
3469 tc = xmalloc(sizeof *tc);
3470 tc_init(tc, &tc_ops_default);
3472 netdev_dev->tc = tc;
3476 default_tc_install(struct netdev *netdev,
3477 const struct shash *details OVS_UNUSED)
3479 default_install__(netdev);
3484 default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3486 default_install__(netdev);
3490 static const struct tc_ops tc_ops_default = {
3491 NULL, /* linux_name */
3496 NULL, /* tc_destroy */
3497 NULL, /* qdisc_get */
3498 NULL, /* qdisc_set */
3499 NULL, /* class_get */
3500 NULL, /* class_set */
3501 NULL, /* class_delete */
3502 NULL, /* class_get_stats */
3503 NULL /* class_dump_stats */
3506 /* "linux-other" traffic control class.
3511 other_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3513 struct netdev_dev_linux *netdev_dev =
3514 netdev_dev_linux_cast(netdev_get_dev(netdev));
3515 static struct tc *tc;
3518 tc = xmalloc(sizeof *tc);
3519 tc_init(tc, &tc_ops_other);
3521 netdev_dev->tc = tc;
3525 static const struct tc_ops tc_ops_other = {
3526 NULL, /* linux_name */
3527 "linux-other", /* ovs_name */
3529 NULL, /* tc_install */
3531 NULL, /* tc_destroy */
3532 NULL, /* qdisc_get */
3533 NULL, /* qdisc_set */
3534 NULL, /* class_get */
3535 NULL, /* class_set */
3536 NULL, /* class_delete */
3537 NULL, /* class_get_stats */
3538 NULL /* class_dump_stats */
3541 /* Traffic control. */
3543 /* Number of kernel "tc" ticks per second. */
3544 static double ticks_per_s;
3546 /* Number of kernel "jiffies" per second. This is used for the purpose of
3547 * computing buffer sizes. Generally kernel qdiscs need to be able to buffer
3548 * one jiffy's worth of data.
3550 * There are two possibilities here:
3552 * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the
3553 * approximate range of 100 to 1024. That means that we really need to
3554 * make sure that the qdisc can buffer that much data.
3556 * - 'buffer_hz' is an absurdly large number. That means that the kernel
3557 * has finely granular timers and there's no need to fudge additional room
3558 * for buffers. (There's no extra effort needed to implement that: the
3559 * large 'buffer_hz' is used as a divisor, so practically any number will
3560 * come out as 0 in the division. Small integer results in the case of
3561 * really high dividends won't have any real effect anyhow.)
3563 static unsigned int buffer_hz;
3565 /* Returns tc handle 'major':'minor'. */
3567 tc_make_handle(unsigned int major, unsigned int minor)
3569 return TC_H_MAKE(major << 16, minor);
3572 /* Returns the major number from 'handle'. */
3574 tc_get_major(unsigned int handle)
3576 return TC_H_MAJ(handle) >> 16;
3579 /* Returns the minor number from 'handle'. */
3581 tc_get_minor(unsigned int handle)
3583 return TC_H_MIN(handle);
3586 static struct tcmsg *
3587 tc_make_request(const struct netdev *netdev, int type, unsigned int flags,
3588 struct ofpbuf *request)
3590 struct tcmsg *tcmsg;
3594 error = get_ifindex(netdev, &ifindex);
3599 ofpbuf_init(request, 512);
3600 nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags);
3601 tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg);
3602 tcmsg->tcm_family = AF_UNSPEC;
3603 tcmsg->tcm_ifindex = ifindex;
3604 /* Caller should fill in tcmsg->tcm_handle. */
3605 /* Caller should fill in tcmsg->tcm_parent. */
3611 tc_transact(struct ofpbuf *request, struct ofpbuf **replyp)
3613 int error = nl_sock_transact(rtnl_sock, request, replyp);
3614 ofpbuf_uninit(request);
3618 /* Adds or deletes a root ingress qdisc on 'netdev'. We use this for
3619 * policing configuration.
3621 * This function is equivalent to running the following when 'add' is true:
3622 * /sbin/tc qdisc add dev <devname> handle ffff: ingress
3624 * This function is equivalent to running the following when 'add' is false:
3625 * /sbin/tc qdisc del dev <devname> handle ffff: ingress
3627 * The configuration and stats may be seen with the following command:
3628 * /sbin/tc -s qdisc show dev <devname>
3630 * Returns 0 if successful, otherwise a positive errno value.
3633 tc_add_del_ingress_qdisc(struct netdev *netdev, bool add)
3635 struct ofpbuf request;
3636 struct tcmsg *tcmsg;
3638 int type = add ? RTM_NEWQDISC : RTM_DELQDISC;
3639 int flags = add ? NLM_F_EXCL | NLM_F_CREATE : 0;
3641 tcmsg = tc_make_request(netdev, type, flags, &request);
3645 tcmsg->tcm_handle = tc_make_handle(0xffff, 0);
3646 tcmsg->tcm_parent = TC_H_INGRESS;
3647 nl_msg_put_string(&request, TCA_KIND, "ingress");
3648 nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0);
3650 error = tc_transact(&request, NULL);
3652 /* If we're deleting the qdisc, don't worry about some of the
3653 * error conditions. */
3654 if (!add && (error == ENOENT || error == EINVAL)) {
3663 /* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size
3666 * This function is equivalent to running:
3667 * /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49
3668 * basic police rate <kbits_rate>kbit burst <kbits_burst>k
3671 * The configuration and stats may be seen with the following command:
3672 * /sbin/tc -s filter show <devname> eth0 parent ffff:
3674 * Returns 0 if successful, otherwise a positive errno value.
3677 tc_add_policer(struct netdev *netdev, int kbits_rate, int kbits_burst)
3679 struct tc_police tc_police;
3680 struct ofpbuf request;
3681 struct tcmsg *tcmsg;
3682 size_t basic_offset;
3683 size_t police_offset;
3687 memset(&tc_police, 0, sizeof tc_police);
3688 tc_police.action = TC_POLICE_SHOT;
3689 tc_police.mtu = mtu;
3690 tc_fill_rate(&tc_police.rate, kbits_rate/8 * 1000, mtu);
3691 tc_police.burst = tc_bytes_to_ticks(tc_police.rate.rate,
3692 kbits_burst * 1024);
3694 tcmsg = tc_make_request(netdev, RTM_NEWTFILTER,
3695 NLM_F_EXCL | NLM_F_CREATE, &request);
3699 tcmsg->tcm_parent = tc_make_handle(0xffff, 0);
3700 tcmsg->tcm_info = tc_make_handle(49,
3701 (OVS_FORCE uint16_t) htons(ETH_P_ALL));
3703 nl_msg_put_string(&request, TCA_KIND, "basic");
3704 basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
3705 police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE);
3706 nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police);
3707 tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate);
3708 nl_msg_end_nested(&request, police_offset);
3709 nl_msg_end_nested(&request, basic_offset);
3711 error = tc_transact(&request, NULL);
3722 /* The values in psched are not individually very meaningful, but they are
3723 * important. The tables below show some values seen in the wild.
3727 * - "c" has always been a constant 1000000 since at least Linux 2.4.14.
3728 * (Before that, there are hints that it was 1000000000.)
3730 * - "d" can be unrealistically large, see the comment on 'buffer_hz'
3734 * -----------------------------------
3735 * [1] 000c8000 000f4240 000f4240 00000064
3736 * [2] 000003e8 00000400 000f4240 3b9aca00
3737 * [3] 000003e8 00000400 000f4240 3b9aca00
3738 * [4] 000003e8 00000400 000f4240 00000064
3739 * [5] 000003e8 00000040 000f4240 3b9aca00
3740 * [6] 000003e8 00000040 000f4240 000000f9
3742 * a b c d ticks_per_s buffer_hz
3743 * ------- --------- ---------- ------------- ----------- -------------
3744 * [1] 819,200 1,000,000 1,000,000 100 819,200 100
3745 * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
3746 * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
3747 * [4] 1,000 1,024 1,000,000 100 976,562 100
3748 * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000
3749 * [6] 1,000 64 1,000,000 249 15,625,000 249
3751 * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p
3752 * [2] 2.6.26-1-686-bigmem from Debian lenny
3753 * [3] 2.6.26-2-sparc64 from Debian lenny
3754 * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p
3755 * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion
3756 * [6] 2.6.34 from kernel.org on KVM
3758 static const char fn[] = "/proc/net/psched";
3759 unsigned int a, b, c, d;
3765 stream = fopen(fn, "r");
3767 VLOG_WARN("%s: open failed: %s", fn, strerror(errno));
3771 if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) {
3772 VLOG_WARN("%s: read failed", fn);
3776 VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d);
3780 VLOG_WARN("%s: invalid scheduler parameters", fn);
3784 ticks_per_s = (double) a * c / b;
3788 VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u",
3791 VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz);
3794 /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a
3795 * rate of 'rate' bytes per second. */
3797 tc_ticks_to_bytes(unsigned int rate, unsigned int ticks)
3802 return (rate * ticks) / ticks_per_s;
3805 /* Returns the number of ticks that it would take to transmit 'size' bytes at a
3806 * rate of 'rate' bytes per second. */
3808 tc_bytes_to_ticks(unsigned int rate, unsigned int size)
3813 return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0;
3816 /* Returns the number of bytes that need to be reserved for qdisc buffering at
3817 * a transmission rate of 'rate' bytes per second. */
3819 tc_buffer_per_jiffy(unsigned int rate)
3824 return rate / buffer_hz;
3827 /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc,
3828 * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull,
3829 * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or
3830 * stores NULL into it if it is absent.
3832 * '*kind' and '*options' point into 'msg', so they are owned by whoever owns
3835 * Returns 0 if successful, otherwise a positive errno value. */
3837 tc_parse_qdisc(const struct ofpbuf *msg, const char **kind,
3838 struct nlattr **options)
3840 static const struct nl_policy tca_policy[] = {
3841 [TCA_KIND] = { .type = NL_A_STRING, .optional = false },
3842 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
3844 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
3846 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
3847 tca_policy, ta, ARRAY_SIZE(ta))) {
3848 VLOG_WARN_RL(&rl, "failed to parse qdisc message");
3853 *kind = nl_attr_get_string(ta[TCA_KIND]);
3857 *options = ta[TCA_OPTIONS];
3872 /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the
3873 * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute
3874 * into '*options', and its queue statistics into '*stats'. Any of the output
3875 * arguments may be null.
3877 * Returns 0 if successful, otherwise a positive errno value. */
3879 tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep,
3880 struct nlattr **options, struct netdev_queue_stats *stats)
3882 static const struct nl_policy tca_policy[] = {
3883 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false },
3884 [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false },
3886 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
3888 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
3889 tca_policy, ta, ARRAY_SIZE(ta))) {
3890 VLOG_WARN_RL(&rl, "failed to parse class message");
3895 struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc);
3896 *handlep = tc->tcm_handle;
3900 *options = ta[TCA_OPTIONS];
3904 const struct gnet_stats_queue *gsq;
3905 struct gnet_stats_basic gsb;
3907 static const struct nl_policy stats_policy[] = {
3908 [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false,
3909 .min_len = sizeof gsb },
3910 [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false,
3911 .min_len = sizeof *gsq },
3913 struct nlattr *sa[ARRAY_SIZE(stats_policy)];
3915 if (!nl_parse_nested(ta[TCA_STATS2], stats_policy,
3916 sa, ARRAY_SIZE(sa))) {
3917 VLOG_WARN_RL(&rl, "failed to parse class stats");
3921 /* Alignment issues screw up the length of struct gnet_stats_basic on
3922 * some arch/bitsize combinations. Newer versions of Linux have a
3923 * struct gnet_stats_basic_packed, but we can't depend on that. The
3924 * easiest thing to do is just to make a copy. */
3925 memset(&gsb, 0, sizeof gsb);
3926 memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]),
3927 MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb));
3928 stats->tx_bytes = gsb.bytes;
3929 stats->tx_packets = gsb.packets;
3931 gsq = nl_attr_get(sa[TCA_STATS_QUEUE]);
3932 stats->tx_errors = gsq->drops;
3942 memset(stats, 0, sizeof *stats);
3947 /* Queries the kernel for class with identifier 'handle' and parent 'parent'
3950 tc_query_class(const struct netdev *netdev,
3951 unsigned int handle, unsigned int parent,
3952 struct ofpbuf **replyp)
3954 struct ofpbuf request;
3955 struct tcmsg *tcmsg;
3958 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request);
3962 tcmsg->tcm_handle = handle;
3963 tcmsg->tcm_parent = parent;
3965 error = tc_transact(&request, replyp);
3967 VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)",
3968 netdev_get_name(netdev),
3969 tc_get_major(handle), tc_get_minor(handle),
3970 tc_get_major(parent), tc_get_minor(parent),
3976 /* Equivalent to "tc class del dev <name> handle <handle>". */
3978 tc_delete_class(const struct netdev *netdev, unsigned int handle)
3980 struct ofpbuf request;
3981 struct tcmsg *tcmsg;
3984 tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request);
3988 tcmsg->tcm_handle = handle;
3989 tcmsg->tcm_parent = 0;
3991 error = tc_transact(&request, NULL);
3993 VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)",
3994 netdev_get_name(netdev),
3995 tc_get_major(handle), tc_get_minor(handle),
4001 /* Equivalent to "tc qdisc del dev <name> root". */
4003 tc_del_qdisc(struct netdev *netdev)
4005 struct netdev_dev_linux *netdev_dev =
4006 netdev_dev_linux_cast(netdev_get_dev(netdev));
4007 struct ofpbuf request;
4008 struct tcmsg *tcmsg;
4011 tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request);
4015 tcmsg->tcm_handle = tc_make_handle(1, 0);
4016 tcmsg->tcm_parent = TC_H_ROOT;
4018 error = tc_transact(&request, NULL);
4019 if (error == EINVAL) {
4020 /* EINVAL probably means that the default qdisc was in use, in which
4021 * case we've accomplished our purpose. */
4024 if (!error && netdev_dev->tc) {
4025 if (netdev_dev->tc->ops->tc_destroy) {
4026 netdev_dev->tc->ops->tc_destroy(netdev_dev->tc);
4028 netdev_dev->tc = NULL;
4033 /* If 'netdev''s qdisc type and parameters are not yet known, queries the
4034 * kernel to determine what they are. Returns 0 if successful, otherwise a
4035 * positive errno value. */
4037 tc_query_qdisc(const struct netdev *netdev)
4039 struct netdev_dev_linux *netdev_dev =
4040 netdev_dev_linux_cast(netdev_get_dev(netdev));
4041 struct ofpbuf request, *qdisc;
4042 const struct tc_ops *ops;
4043 struct tcmsg *tcmsg;
4047 if (netdev_dev->tc) {
4051 /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have
4052 * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before
4053 * 2.6.35 without that fix backported to it.
4055 * To avoid the OOPS, we must not make a request that would attempt to dump
4056 * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a
4057 * few others. There are a few ways that I can see to do this, but most of
4058 * them seem to be racy (and if you lose the race the kernel OOPSes). The
4059 * technique chosen here is to assume that any non-default qdisc that we
4060 * create will have a class with handle 1:0. The built-in qdiscs only have
4061 * a class with handle 0:0.
4063 * We could check for Linux 2.6.35+ and use a more straightforward method
4065 tcmsg = tc_make_request(netdev, RTM_GETQDISC, NLM_F_ECHO, &request);
4069 tcmsg->tcm_handle = tc_make_handle(1, 0);
4070 tcmsg->tcm_parent = 0;
4072 /* Figure out what tc class to instantiate. */
4073 error = tc_transact(&request, &qdisc);
4077 error = tc_parse_qdisc(qdisc, &kind, NULL);
4079 ops = &tc_ops_other;
4081 ops = tc_lookup_linux_name(kind);
4083 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1);
4084 VLOG_INFO_RL(&rl2, "unknown qdisc \"%s\"", kind);
4086 ops = &tc_ops_other;
4089 } else if (error == ENOENT) {
4090 /* Either it's a built-in qdisc, or it's a qdisc set up by some
4091 * other entity that doesn't have a handle 1:0. We will assume
4092 * that it's the system default qdisc. */
4093 ops = &tc_ops_default;
4096 /* Who knows? Maybe the device got deleted. */
4097 VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)",
4098 netdev_get_name(netdev), strerror(error));
4099 ops = &tc_ops_other;
4102 /* Instantiate it. */
4103 load_error = ops->tc_load((struct netdev *) netdev, qdisc);
4104 assert((load_error == 0) == (netdev_dev->tc != NULL));
4105 ofpbuf_delete(qdisc);
4107 return error ? error : load_error;
4110 /* Linux traffic control uses tables with 256 entries ("rtab" tables) to
4111 approximate the time to transmit packets of various lengths. For an MTU of
4112 256 or less, each entry is exact; for an MTU of 257 through 512, each entry
4113 represents two possible packet lengths; for a MTU of 513 through 1024, four
4114 possible lengths; and so on.
4116 Returns, for the specified 'mtu', the number of bits that packet lengths
4117 need to be shifted right to fit within such a 256-entry table. */
4119 tc_calc_cell_log(unsigned int mtu)
4124 mtu = ETH_PAYLOAD_MAX;
4126 mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN;
4128 for (cell_log = 0; mtu >= 256; cell_log++) {
4135 /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU
4138 tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu)
4140 memset(rate, 0, sizeof *rate);
4141 rate->cell_log = tc_calc_cell_log(mtu);
4142 /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */
4143 /* rate->cell_align = 0; */ /* distro headers. */
4144 rate->mpu = ETH_TOTAL_MIN;
4148 /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink
4149 * attribute of the specified "type".
4151 * See tc_calc_cell_log() above for a description of "rtab"s. */
4153 tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate)
4158 rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE);
4159 for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) {
4160 unsigned packet_size = (i + 1) << rate->cell_log;
4161 if (packet_size < rate->mpu) {
4162 packet_size = rate->mpu;
4164 rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size);
4168 /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a
4169 * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested
4170 * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of
4173 tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes)
4175 unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu;
4176 return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst));
4179 /* Linux-only functions declared in netdev-linux.h */
4181 /* Returns a fd for an AF_INET socket or a negative errno value. */
4183 netdev_linux_get_af_inet_sock(void)
4185 int error = netdev_linux_init();
4186 return error ? -error : af_inet_sock;
4189 /* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If
4190 * 'enable' is true, the bit is set. Otherwise, it is cleared. */
4192 netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag,
4193 const char *flag_name, bool enable)
4195 const char *netdev_name = netdev_get_name(netdev);
4196 struct ethtool_value evalue;
4200 memset(&evalue, 0, sizeof evalue);
4201 error = netdev_linux_do_ethtool(netdev_name,
4202 (struct ethtool_cmd *)&evalue,
4203 ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4208 evalue.data = new_flags = (evalue.data & ~flag) | (enable ? flag : 0);
4209 error = netdev_linux_do_ethtool(netdev_name,
4210 (struct ethtool_cmd *)&evalue,
4211 ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS");
4216 memset(&evalue, 0, sizeof evalue);
4217 error = netdev_linux_do_ethtool(netdev_name,
4218 (struct ethtool_cmd *)&evalue,
4219 ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4224 if (new_flags != evalue.data) {
4225 VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network "
4226 "device %s failed", enable ? "enable" : "disable",
4227 flag_name, netdev_name);
4234 /* Utility functions. */
4236 /* Copies 'src' into 'dst', performing format conversion in the process. */
4238 netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst,
4239 const struct rtnl_link_stats *src)
4241 dst->rx_packets = src->rx_packets;
4242 dst->tx_packets = src->tx_packets;
4243 dst->rx_bytes = src->rx_bytes;
4244 dst->tx_bytes = src->tx_bytes;
4245 dst->rx_errors = src->rx_errors;
4246 dst->tx_errors = src->tx_errors;
4247 dst->rx_dropped = src->rx_dropped;
4248 dst->tx_dropped = src->tx_dropped;
4249 dst->multicast = src->multicast;
4250 dst->collisions = src->collisions;
4251 dst->rx_length_errors = src->rx_length_errors;
4252 dst->rx_over_errors = src->rx_over_errors;
4253 dst->rx_crc_errors = src->rx_crc_errors;
4254 dst->rx_frame_errors = src->rx_frame_errors;
4255 dst->rx_fifo_errors = src->rx_fifo_errors;
4256 dst->rx_missed_errors = src->rx_missed_errors;
4257 dst->tx_aborted_errors = src->tx_aborted_errors;
4258 dst->tx_carrier_errors = src->tx_carrier_errors;
4259 dst->tx_fifo_errors = src->tx_fifo_errors;
4260 dst->tx_heartbeat_errors = src->tx_heartbeat_errors;
4261 dst->tx_window_errors = src->tx_window_errors;
4265 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
4267 /* Policy for RTNLGRP_LINK messages.
4269 * There are *many* more fields in these messages, but currently we only
4270 * care about these fields. */
4271 static const struct nl_policy rtnlgrp_link_policy[] = {
4272 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
4273 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
4274 .min_len = sizeof(struct rtnl_link_stats) },
4277 struct ofpbuf request;
4278 struct ofpbuf *reply;
4279 struct ifinfomsg *ifi;
4280 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
4283 ofpbuf_init(&request, 0);
4284 nl_msg_put_nlmsghdr(&request, sizeof *ifi, RTM_GETLINK, NLM_F_REQUEST);
4285 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
4286 ifi->ifi_family = PF_UNSPEC;
4287 ifi->ifi_index = ifindex;
4288 error = nl_sock_transact(rtnl_sock, &request, &reply);
4289 ofpbuf_uninit(&request);
4294 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
4295 rtnlgrp_link_policy,
4296 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
4297 ofpbuf_delete(reply);
4301 if (!attrs[IFLA_STATS]) {
4302 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
4303 ofpbuf_delete(reply);
4307 netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(attrs[IFLA_STATS]));
4309 ofpbuf_delete(reply);
4315 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
4317 static const char fn[] = "/proc/net/dev";
4322 stream = fopen(fn, "r");
4324 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
4329 while (fgets(line, sizeof line, stream)) {
4332 #define X64 "%"SCNu64
4335 X64 X64 X64 X64 X64 X64 X64 "%*u"
4336 X64 X64 X64 X64 X64 X64 X64 "%*u",
4342 &stats->rx_fifo_errors,
4343 &stats->rx_frame_errors,
4349 &stats->tx_fifo_errors,
4351 &stats->tx_carrier_errors) != 15) {
4352 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
4353 } else if (!strcmp(devname, netdev_name)) {
4354 stats->rx_length_errors = UINT64_MAX;
4355 stats->rx_over_errors = UINT64_MAX;
4356 stats->rx_crc_errors = UINT64_MAX;
4357 stats->rx_missed_errors = UINT64_MAX;
4358 stats->tx_aborted_errors = UINT64_MAX;
4359 stats->tx_heartbeat_errors = UINT64_MAX;
4360 stats->tx_window_errors = UINT64_MAX;
4366 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
4372 get_flags(const struct netdev_dev *dev, unsigned int *flags)
4378 error = netdev_linux_do_ioctl(dev->name, &ifr, SIOCGIFFLAGS,
4381 *flags = ifr.ifr_flags;
4387 set_flags(struct netdev *netdev, unsigned int flags)
4391 ifr.ifr_flags = flags;
4392 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
4397 do_get_ifindex(const char *netdev_name)
4401 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4402 COVERAGE_INC(netdev_get_ifindex);
4403 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
4404 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
4405 netdev_name, strerror(errno));
4408 return ifr.ifr_ifindex;
4412 get_ifindex(const struct netdev *netdev_, int *ifindexp)
4414 struct netdev_dev_linux *netdev_dev =
4415 netdev_dev_linux_cast(netdev_get_dev(netdev_));
4417 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
4418 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
4421 netdev_dev->get_ifindex_error = -ifindex;
4422 netdev_dev->ifindex = 0;
4424 netdev_dev->get_ifindex_error = 0;
4425 netdev_dev->ifindex = ifindex;
4427 netdev_dev->cache_valid |= VALID_IFINDEX;
4430 *ifindexp = netdev_dev->ifindex;
4431 return netdev_dev->get_ifindex_error;
4435 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
4440 memset(&ifr, 0, sizeof ifr);
4441 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4442 COVERAGE_INC(netdev_get_hwaddr);
4443 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
4444 /* ENODEV probably means that a vif disappeared asynchronously and
4445 * hasn't been removed from the database yet, so reduce the log level
4446 * to INFO for that case. */
4447 VLOG(errno == ENODEV ? VLL_INFO : VLL_ERR,
4448 "ioctl(SIOCGIFHWADDR) on %s device failed: %s",
4449 netdev_name, strerror(errno));
4452 hwaddr_family = ifr.ifr_hwaddr.sa_family;
4453 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
4454 VLOG_WARN("%s device has unknown hardware address family %d",
4455 netdev_name, hwaddr_family);
4457 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
4462 set_etheraddr(const char *netdev_name,
4463 const uint8_t mac[ETH_ADDR_LEN])
4467 memset(&ifr, 0, sizeof ifr);
4468 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4469 ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
4470 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
4471 COVERAGE_INC(netdev_set_hwaddr);
4472 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
4473 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
4474 netdev_name, strerror(errno));
4481 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
4482 int cmd, const char *cmd_name)
4486 memset(&ifr, 0, sizeof ifr);
4487 ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
4488 ifr.ifr_data = (caddr_t) ecmd;
4491 COVERAGE_INC(netdev_ethtool);
4492 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
4495 if (errno != EOPNOTSUPP) {
4496 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
4497 "failed: %s", cmd_name, name, strerror(errno));
4499 /* The device doesn't support this operation. That's pretty
4500 * common, so there's no point in logging anything. */
4507 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
4508 const char *cmd_name)
4510 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
4511 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
4512 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
4520 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
4521 int cmd, const char *cmd_name)
4526 ifr.ifr_addr.sa_family = AF_INET;
4527 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
4529 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
4530 *ip = sin->sin_addr;
4535 /* Returns an AF_PACKET raw socket or a negative errno value. */
4537 af_packet_sock(void)
4539 static int sock = INT_MIN;
4541 if (sock == INT_MIN) {
4542 sock = socket(AF_PACKET, SOCK_RAW, 0);
4544 set_nonblocking(sock);
4547 VLOG_ERR("failed to create packet socket: %s", strerror(errno));