2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
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);
83 /* These were introduced in Linux 2.6.14, so they might be missing if we have
85 #ifndef ADVERTISED_Pause
86 #define ADVERTISED_Pause (1 << 13)
88 #ifndef ADVERTISED_Asym_Pause
89 #define ADVERTISED_Asym_Pause (1 << 14)
92 /* These were introduced in Linux 2.6.24, so they might be missing if we
93 * have old headers. */
94 #ifndef ETHTOOL_GFLAGS
95 #define ETHTOOL_GFLAGS 0x00000025 /* Get flags bitmap(ethtool_value) */
97 #ifndef ETHTOOL_SFLAGS
98 #define ETHTOOL_SFLAGS 0x00000026 /* Set flags bitmap(ethtool_value) */
101 /* This was introduced in Linux 2.6.25, so it might be missing if we have old
104 #define TC_RTAB_SIZE 1024
107 static struct nln_notifier *netdev_linux_cache_notifier = NULL;
108 static int cache_notifier_refcount;
111 VALID_IFINDEX = 1 << 0,
112 VALID_ETHERADDR = 1 << 1,
116 VALID_POLICING = 1 << 5,
117 VALID_HAVE_VPORT_STATS = 1 << 6
125 /* Traffic control. */
127 /* An instance of a traffic control class. Always associated with a particular
130 * Each TC implementation subclasses this with whatever additional data it
133 const struct tc_ops *ops;
134 struct hmap queues; /* Contains "struct tc_queue"s.
135 * Read by generic TC layer.
136 * Written only by TC implementation. */
139 /* One traffic control queue.
141 * Each TC implementation subclasses this with whatever additional data it
144 struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */
145 unsigned int queue_id; /* OpenFlow queue ID. */
148 /* A particular kind of traffic control. Each implementation generally maps to
149 * one particular Linux qdisc class.
151 * The functions below return 0 if successful or a positive errno value on
152 * failure, except where otherwise noted. All of them must be provided, except
153 * where otherwise noted. */
155 /* Name used by kernel in the TCA_KIND attribute of tcmsg, e.g. "htb".
156 * This is null for tc_ops_default and tc_ops_other, for which there are no
157 * appropriate values. */
158 const char *linux_name;
160 /* Name used in OVS database, e.g. "linux-htb". Must be nonnull. */
161 const char *ovs_name;
163 /* Number of supported OpenFlow queues, 0 for qdiscs that have no
164 * queues. The queues are numbered 0 through n_queues - 1. */
165 unsigned int n_queues;
167 /* Called to install this TC class on 'netdev'. The implementation should
168 * make the Netlink calls required to set up 'netdev' with the right qdisc
169 * and configure it according to 'details'. The implementation may assume
170 * that the current qdisc is the default; that is, there is no need for it
171 * to delete the current qdisc before installing itself.
173 * The contents of 'details' should be documented as valid for 'ovs_name'
174 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
175 * (which is built as ovs-vswitchd.conf.db(8)).
177 * This function must return 0 if and only if it sets 'netdev->tc' to an
178 * initialized 'struct tc'.
180 * (This function is null for tc_ops_other, which cannot be installed. For
181 * other TC classes it should always be nonnull.) */
182 int (*tc_install)(struct netdev *netdev, const struct shash *details);
184 /* Called when the netdev code determines (through a Netlink query) that
185 * this TC class's qdisc is installed on 'netdev', but we didn't install
186 * it ourselves and so don't know any of the details.
188 * 'nlmsg' is the kernel reply to a RTM_GETQDISC Netlink message for
189 * 'netdev'. The TCA_KIND attribute of 'nlmsg' is 'linux_name'. The
190 * implementation should parse the other attributes of 'nlmsg' as
191 * necessary to determine its configuration. If necessary it should also
192 * use Netlink queries to determine the configuration of queues on
195 * This function must return 0 if and only if it sets 'netdev->tc' to an
196 * initialized 'struct tc'. */
197 int (*tc_load)(struct netdev *netdev, struct ofpbuf *nlmsg);
199 /* Destroys the data structures allocated by the implementation as part of
200 * 'tc'. (This includes destroying 'tc->queues' by calling
203 * The implementation should not need to perform any Netlink calls. If
204 * desirable, the caller is responsible for deconfiguring the kernel qdisc.
205 * (But it may not be desirable.)
207 * This function may be null if 'tc' is trivial. */
208 void (*tc_destroy)(struct tc *tc);
210 /* Retrieves details of 'netdev->tc' configuration into 'details'.
212 * The implementation should not need to perform any Netlink calls, because
213 * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have
214 * cached the configuration.
216 * The contents of 'details' should be documented as valid for 'ovs_name'
217 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
218 * (which is built as ovs-vswitchd.conf.db(8)).
220 * This function may be null if 'tc' is not configurable.
222 int (*qdisc_get)(const struct netdev *netdev, struct shash *details);
224 /* Reconfigures 'netdev->tc' according to 'details', performing any
225 * required Netlink calls to complete the reconfiguration.
227 * The contents of 'details' should be documented as valid for 'ovs_name'
228 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
229 * (which is built as ovs-vswitchd.conf.db(8)).
231 * This function may be null if 'tc' is not configurable.
233 int (*qdisc_set)(struct netdev *, const struct shash *details);
235 /* Retrieves details of 'queue' on 'netdev->tc' into 'details'. 'queue' is
236 * one of the 'struct tc_queue's within 'netdev->tc->queues'.
238 * The contents of 'details' should be documented as valid for 'ovs_name'
239 * in the "other_config" column in the "Queue" table in
240 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
242 * The implementation should not need to perform any Netlink calls, because
243 * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have
244 * cached the queue configuration.
246 * This function may be null if 'tc' does not have queues ('n_queues' is
248 int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue,
249 struct shash *details);
251 /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to
252 * 'details', perfoming any required Netlink calls to complete the
253 * reconfiguration. The caller ensures that 'queue_id' is less than
256 * The contents of 'details' should be documented as valid for 'ovs_name'
257 * in the "other_config" column in the "Queue" table in
258 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
260 * This function may be null if 'tc' does not have queues or its queues are
261 * not configurable. */
262 int (*class_set)(struct netdev *, unsigned int queue_id,
263 const struct shash *details);
265 /* Deletes 'queue' from 'netdev->tc'. 'queue' is one of the 'struct
266 * tc_queue's within 'netdev->tc->queues'.
268 * This function may be null if 'tc' does not have queues or its queues
269 * cannot be deleted. */
270 int (*class_delete)(struct netdev *, struct tc_queue *queue);
272 /* Obtains stats for 'queue' from 'netdev->tc'. 'queue' is one of the
273 * 'struct tc_queue's within 'netdev->tc->queues'.
275 * On success, initializes '*stats'.
277 * This function may be null if 'tc' does not have queues or if it cannot
278 * report queue statistics. */
279 int (*class_get_stats)(const struct netdev *netdev,
280 const struct tc_queue *queue,
281 struct netdev_queue_stats *stats);
283 /* Extracts queue stats from 'nlmsg', which is a response to a
284 * RTM_GETTCLASS message, and passes them to 'cb' along with 'aux'.
286 * This function may be null if 'tc' does not have queues or if it cannot
287 * report queue statistics. */
288 int (*class_dump_stats)(const struct netdev *netdev,
289 const struct ofpbuf *nlmsg,
290 netdev_dump_queue_stats_cb *cb, void *aux);
294 tc_init(struct tc *tc, const struct tc_ops *ops)
297 hmap_init(&tc->queues);
301 tc_destroy(struct tc *tc)
303 hmap_destroy(&tc->queues);
306 static const struct tc_ops tc_ops_htb;
307 static const struct tc_ops tc_ops_hfsc;
308 static const struct tc_ops tc_ops_default;
309 static const struct tc_ops tc_ops_other;
311 static const struct tc_ops *tcs[] = {
312 &tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */
313 &tc_ops_hfsc, /* Hierarchical fair service curve. */
314 &tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */
315 &tc_ops_other, /* Some other qdisc. */
319 static unsigned int tc_make_handle(unsigned int major, unsigned int minor);
320 static unsigned int tc_get_major(unsigned int handle);
321 static unsigned int tc_get_minor(unsigned int handle);
323 static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks);
324 static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size);
325 static unsigned int tc_buffer_per_jiffy(unsigned int rate);
327 static struct tcmsg *tc_make_request(const struct netdev *, int type,
328 unsigned int flags, struct ofpbuf *);
329 static int tc_transact(struct ofpbuf *request, struct ofpbuf **replyp);
330 static int tc_add_del_ingress_qdisc(struct netdev *netdev, bool add);
331 static int tc_add_policer(struct netdev *netdev, int kbits_rate,
334 static int tc_parse_qdisc(const struct ofpbuf *, const char **kind,
335 struct nlattr **options);
336 static int tc_parse_class(const struct ofpbuf *, unsigned int *queue_id,
337 struct nlattr **options,
338 struct netdev_queue_stats *);
339 static int tc_query_class(const struct netdev *,
340 unsigned int handle, unsigned int parent,
341 struct ofpbuf **replyp);
342 static int tc_delete_class(const struct netdev *, unsigned int handle);
344 static int tc_del_qdisc(struct netdev *netdev);
345 static int tc_query_qdisc(const struct netdev *netdev);
347 static int tc_calc_cell_log(unsigned int mtu);
348 static void tc_fill_rate(struct tc_ratespec *rate, uint64_t bps, int mtu);
349 static void tc_put_rtab(struct ofpbuf *, uint16_t type,
350 const struct tc_ratespec *rate);
351 static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes);
353 struct netdev_dev_linux {
354 struct netdev_dev netdev_dev;
356 struct shash_node *shash_node;
357 unsigned int cache_valid;
358 unsigned int change_seq;
360 bool miimon; /* Link status of last poll. */
361 long long int miimon_interval; /* Miimon Poll rate. Disabled if <= 0. */
362 struct timer miimon_timer;
364 /* The following are figured out "on demand" only. They are only valid
365 * when the corresponding VALID_* bit in 'cache_valid' is set. */
367 uint8_t etheraddr[ETH_ADDR_LEN];
368 struct in_addr address, netmask;
371 unsigned int ifi_flags;
372 long long int carrier_resets;
373 uint32_t kbits_rate; /* Policing data. */
374 uint32_t kbits_burst;
375 bool have_vport_stats;
379 struct tap_state tap;
383 struct netdev_linux {
384 struct netdev netdev;
388 /* Sockets used for ioctl operations. */
389 static int af_inet_sock = -1; /* AF_INET, SOCK_DGRAM. */
391 /* A Netlink routing socket that is not subscribed to any multicast groups. */
392 static struct nl_sock *rtnl_sock;
394 /* This is set pretty low because we probably won't learn anything from the
395 * additional log messages. */
396 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
398 static int netdev_linux_init(void);
400 static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
401 int cmd, const char *cmd_name);
402 static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
403 const char *cmd_name);
404 static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
405 int cmd, const char *cmd_name);
406 static int get_flags(const struct netdev_dev *, unsigned int *flags);
407 static int set_flags(struct netdev *, unsigned int flags);
408 static int do_get_ifindex(const char *netdev_name);
409 static int get_ifindex(const struct netdev *, int *ifindexp);
410 static int do_set_addr(struct netdev *netdev,
411 int ioctl_nr, const char *ioctl_name,
412 struct in_addr addr);
413 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
414 static int set_etheraddr(const char *netdev_name, int hwaddr_family,
415 const uint8_t[ETH_ADDR_LEN]);
416 static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
417 static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
418 static int af_packet_sock(void);
419 static void netdev_linux_miimon_run(void);
420 static void netdev_linux_miimon_wait(void);
423 is_netdev_linux_class(const struct netdev_class *netdev_class)
425 return netdev_class->init == netdev_linux_init;
428 static struct netdev_dev_linux *
429 netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
431 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
432 assert(is_netdev_linux_class(netdev_class));
434 return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
437 static struct netdev_linux *
438 netdev_linux_cast(const struct netdev *netdev)
440 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
441 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
442 assert(is_netdev_linux_class(netdev_class));
444 return CONTAINER_OF(netdev, struct netdev_linux, netdev);
448 netdev_linux_init(void)
450 static int status = -1;
452 /* Create AF_INET socket. */
453 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
454 status = af_inet_sock >= 0 ? 0 : errno;
456 VLOG_ERR("failed to create inet socket: %s", strerror(status));
459 /* Create rtnetlink socket. */
461 status = nl_sock_create(NETLINK_ROUTE, &rtnl_sock);
463 VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",
472 netdev_linux_run(void)
474 rtnetlink_link_run();
475 netdev_linux_miimon_run();
479 netdev_linux_wait(void)
481 rtnetlink_link_wait();
482 netdev_linux_miimon_wait();
486 netdev_dev_linux_changed(struct netdev_dev_linux *dev)
489 if (!dev->change_seq) {
492 dev->cache_valid = 0;
496 netdev_linux_cache_cb(const struct rtnetlink_link_change *change,
497 void *aux OVS_UNUSED)
499 struct netdev_dev_linux *dev;
501 struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
503 const struct netdev_class *netdev_class =
504 netdev_dev_get_class(base_dev);
506 if (is_netdev_linux_class(netdev_class)) {
507 dev = netdev_dev_linux_cast(base_dev);
509 if ((dev->ifi_flags ^ change->ifi_flags) & IFF_RUNNING) {
510 dev->carrier_resets++;
512 dev->ifi_flags = change->ifi_flags;
514 netdev_dev_linux_changed(dev);
518 struct shash device_shash;
519 struct shash_node *node;
521 shash_init(&device_shash);
522 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
523 SHASH_FOR_EACH (node, &device_shash) {
528 get_flags(&dev->netdev_dev, &flags);
529 if ((dev->ifi_flags ^ flags) & IFF_RUNNING) {
530 dev->carrier_resets++;
532 dev->ifi_flags = flags;
534 netdev_dev_linux_changed(dev);
536 shash_destroy(&device_shash);
541 cache_notifier_ref(void)
543 if (!cache_notifier_refcount) {
544 assert(!netdev_linux_cache_notifier);
546 netdev_linux_cache_notifier =
547 rtnetlink_link_notifier_create(netdev_linux_cache_cb, NULL);
549 if (!netdev_linux_cache_notifier) {
553 cache_notifier_refcount++;
559 cache_notifier_unref(void)
561 assert(cache_notifier_refcount > 0);
562 if (!--cache_notifier_refcount) {
563 assert(netdev_linux_cache_notifier);
564 rtnetlink_link_notifier_destroy(netdev_linux_cache_notifier);
565 netdev_linux_cache_notifier = NULL;
569 /* Creates system and internal devices. */
571 netdev_linux_create(const struct netdev_class *class, const char *name,
572 struct netdev_dev **netdev_devp)
574 struct netdev_dev_linux *netdev_dev;
577 error = cache_notifier_ref();
582 netdev_dev = xzalloc(sizeof *netdev_dev);
583 netdev_dev->change_seq = 1;
584 netdev_dev_init(&netdev_dev->netdev_dev, name, class);
585 get_flags(&netdev_dev->netdev_dev, &netdev_dev->ifi_flags);
587 *netdev_devp = &netdev_dev->netdev_dev;
591 /* For most types of netdevs we open the device for each call of
592 * netdev_open(). However, this is not the case with tap devices,
593 * since it is only possible to open the device once. In this
594 * situation we share a single file descriptor, and consequently
595 * buffers, across all readers. Therefore once data is read it will
596 * be unavailable to other reads for tap devices. */
598 netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED,
599 const char *name, struct netdev_dev **netdev_devp)
601 struct netdev_dev_linux *netdev_dev;
602 struct tap_state *state;
603 static const char tap_dev[] = "/dev/net/tun";
607 netdev_dev = xzalloc(sizeof *netdev_dev);
608 state = &netdev_dev->state.tap;
610 error = cache_notifier_ref();
615 /* Open tap device. */
616 state->fd = open(tap_dev, O_RDWR);
619 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
620 goto error_unref_notifier;
623 /* Create tap device. */
624 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
625 ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
626 if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
627 VLOG_WARN("%s: creating tap device failed: %s", name,
630 goto error_unref_notifier;
633 /* Make non-blocking. */
634 error = set_nonblocking(state->fd);
636 goto error_unref_notifier;
639 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
640 *netdev_devp = &netdev_dev->netdev_dev;
643 error_unref_notifier:
644 cache_notifier_unref();
651 destroy_tap(struct netdev_dev_linux *netdev_dev)
653 struct tap_state *state = &netdev_dev->state.tap;
655 if (state->fd >= 0) {
660 /* Destroys the netdev device 'netdev_dev_'. */
662 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
664 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
665 const struct netdev_class *class = netdev_dev_get_class(netdev_dev_);
667 if (netdev_dev->tc && netdev_dev->tc->ops->tc_destroy) {
668 netdev_dev->tc->ops->tc_destroy(netdev_dev->tc);
671 if (class == &netdev_tap_class) {
672 destroy_tap(netdev_dev);
676 cache_notifier_unref();
680 netdev_linux_open(struct netdev_dev *netdev_dev_, struct netdev **netdevp)
682 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
683 struct netdev_linux *netdev;
684 enum netdev_flags flags;
687 /* Allocate network device. */
688 netdev = xzalloc(sizeof *netdev);
690 netdev_init(&netdev->netdev, netdev_dev_);
692 /* Verify that the device really exists, by attempting to read its flags.
693 * (The flags might be cached, in which case this won't actually do an
696 * Don't do this for "internal" netdevs, though, because those have to be
697 * created as netdev objects before they exist in the kernel, because
698 * creating them in the kernel happens by passing a netdev object to
699 * dpif_port_add(). */
700 if (netdev_dev_get_class(netdev_dev_) != &netdev_internal_class) {
701 error = netdev_get_flags(&netdev->netdev, &flags);
702 if (error == ENODEV) {
707 if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap") &&
708 !netdev_dev->state.tap.opened) {
710 /* We assume that the first user of the tap device is the primary user
711 * and give them the tap FD. Subsequent users probably just expect
712 * this to be a system device so open it normally to avoid send/receive
713 * directions appearing to be reversed. */
714 netdev->fd = netdev_dev->state.tap.fd;
715 netdev_dev->state.tap.opened = true;
718 *netdevp = &netdev->netdev;
722 netdev_uninit(&netdev->netdev, true);
726 /* Closes and destroys 'netdev'. */
728 netdev_linux_close(struct netdev *netdev_)
730 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
732 if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
739 netdev_linux_listen(struct netdev *netdev_)
741 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
742 struct sockaddr_ll sll;
747 if (netdev->fd >= 0) {
751 /* Create file descriptor. */
752 fd = socket(PF_PACKET, SOCK_RAW, 0);
755 VLOG_ERR("failed to create raw socket (%s)", strerror(error));
759 /* Set non-blocking mode. */
760 error = set_nonblocking(fd);
765 /* Get ethernet device index. */
766 error = get_ifindex(&netdev->netdev, &ifindex);
771 /* Bind to specific ethernet device. */
772 memset(&sll, 0, sizeof sll);
773 sll.sll_family = AF_PACKET;
774 sll.sll_ifindex = ifindex;
775 sll.sll_protocol = (OVS_FORCE unsigned short int) htons(ETH_P_ALL);
776 if (bind(fd, (struct sockaddr *) &sll, sizeof sll) < 0) {
778 VLOG_ERR("%s: failed to bind raw socket (%s)",
779 netdev_get_name(netdev_), strerror(error));
794 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
796 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
798 if (netdev->fd < 0) {
799 /* Device is not listening. */
804 ssize_t retval = recv(netdev->fd, data, size, MSG_TRUNC);
806 return retval <= size ? retval : -EMSGSIZE;
807 } else if (errno != EINTR) {
808 if (errno != EAGAIN) {
809 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
810 strerror(errno), netdev_get_name(netdev_));
817 /* Registers with the poll loop to wake up from the next call to poll_block()
818 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
820 netdev_linux_recv_wait(struct netdev *netdev_)
822 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
823 if (netdev->fd >= 0) {
824 poll_fd_wait(netdev->fd, POLLIN);
828 /* Discards all packets waiting to be received from 'netdev'. */
830 netdev_linux_drain(struct netdev *netdev_)
832 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
833 if (netdev->fd < 0) {
835 } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
837 int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
838 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
842 drain_fd(netdev->fd, ifr.ifr_qlen);
845 return drain_rcvbuf(netdev->fd);
849 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
850 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
851 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
852 * the packet is too big or too small to transmit on the device.
854 * The caller retains ownership of 'buffer' in all cases.
856 * The kernel maintains a packet transmission queue, so the caller is not
857 * expected to do additional queuing of packets. */
859 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
861 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
865 if (netdev->fd < 0) {
866 /* Use our AF_PACKET socket to send to this device. */
867 struct sockaddr_ll sll;
874 sock = af_packet_sock();
879 error = get_ifindex(netdev_, &ifindex);
884 /* We don't bother setting most fields in sockaddr_ll because the
885 * kernel ignores them for SOCK_RAW. */
886 memset(&sll, 0, sizeof sll);
887 sll.sll_family = AF_PACKET;
888 sll.sll_ifindex = ifindex;
890 iov.iov_base = (void *) data;
894 msg.msg_namelen = sizeof sll;
897 msg.msg_control = NULL;
898 msg.msg_controllen = 0;
901 retval = sendmsg(sock, &msg, 0);
903 /* Use the netdev's own fd to send to this device. This is
904 * essential for tap devices, because packets sent to a tap device
905 * with an AF_PACKET socket will loop back to be *received* again
906 * on the tap device. */
907 retval = write(netdev->fd, data, size);
911 /* The Linux AF_PACKET implementation never blocks waiting for room
912 * for packets, instead returning ENOBUFS. Translate this into
913 * EAGAIN for the caller. */
914 if (errno == ENOBUFS) {
916 } else if (errno == EINTR) {
918 } else if (errno != EAGAIN) {
919 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
920 netdev_get_name(netdev_), strerror(errno));
923 } else if (retval != size) {
924 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
925 "%zu) on %s", retval, size, netdev_get_name(netdev_));
933 /* Registers with the poll loop to wake up from the next call to poll_block()
934 * when the packet transmission queue has sufficient room to transmit a packet
935 * with netdev_send().
937 * The kernel maintains a packet transmission queue, so the client is not
938 * expected to do additional queuing of packets. Thus, this function is
939 * unlikely to ever be used. It is included for completeness. */
941 netdev_linux_send_wait(struct netdev *netdev_)
943 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
944 if (netdev->fd < 0) {
946 } else if (strcmp(netdev_get_type(netdev_), "tap")) {
947 poll_fd_wait(netdev->fd, POLLOUT);
949 /* TAP device always accepts packets.*/
950 poll_immediate_wake();
954 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
955 * otherwise a positive errno value. */
957 netdev_linux_set_etheraddr(struct netdev *netdev_,
958 const uint8_t mac[ETH_ADDR_LEN])
960 struct netdev_dev_linux *netdev_dev =
961 netdev_dev_linux_cast(netdev_get_dev(netdev_));
964 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)
965 || !eth_addr_equals(netdev_dev->etheraddr, mac)) {
966 error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
968 netdev_dev->cache_valid |= VALID_ETHERADDR;
969 memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
977 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
978 * free the returned buffer. */
980 netdev_linux_get_etheraddr(const struct netdev *netdev_,
981 uint8_t mac[ETH_ADDR_LEN])
983 struct netdev_dev_linux *netdev_dev =
984 netdev_dev_linux_cast(netdev_get_dev(netdev_));
985 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
986 int error = get_etheraddr(netdev_get_name(netdev_),
987 netdev_dev->etheraddr);
991 netdev_dev->cache_valid |= VALID_ETHERADDR;
993 memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
997 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
998 * in bytes, not including the hardware header; thus, this is typically 1500
999 * bytes for Ethernet devices. */
1001 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
1003 struct netdev_dev_linux *netdev_dev =
1004 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1005 if (!(netdev_dev->cache_valid & VALID_MTU)) {
1009 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1010 SIOCGIFMTU, "SIOCGIFMTU");
1014 netdev_dev->mtu = ifr.ifr_mtu;
1015 netdev_dev->cache_valid |= VALID_MTU;
1017 *mtup = netdev_dev->mtu;
1021 /* Sets the maximum size of transmitted (MTU) for given device using linux
1022 * networking ioctl interface.
1025 netdev_linux_set_mtu(const struct netdev *netdev_, int mtu)
1027 struct netdev_dev_linux *netdev_dev =
1028 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1032 if (netdev_dev->cache_valid & VALID_MTU &&
1033 netdev_dev->mtu == mtu) {
1037 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1038 SIOCSIFMTU, "SIOCSIFMTU");
1043 netdev_dev->mtu = ifr.ifr_mtu;
1044 netdev_dev->cache_valid |= VALID_MTU;
1048 /* Returns the ifindex of 'netdev', if successful, as a positive number.
1049 * On failure, returns a negative errno value. */
1051 netdev_linux_get_ifindex(const struct netdev *netdev)
1055 error = get_ifindex(netdev, &ifindex);
1056 return error ? -error : ifindex;
1060 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
1062 struct netdev_dev_linux *netdev_dev =
1063 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1065 if (netdev_dev->miimon_interval > 0) {
1066 *carrier = netdev_dev->miimon;
1068 *carrier = (netdev_dev->ifi_flags & IFF_RUNNING) != 0;
1074 static long long int
1075 netdev_linux_get_carrier_resets(const struct netdev *netdev)
1077 return netdev_dev_linux_cast(netdev_get_dev(netdev))->carrier_resets;
1081 netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name,
1082 struct mii_ioctl_data *data)
1087 memset(&ifr, 0, sizeof ifr);
1088 memcpy(&ifr.ifr_data, data, sizeof *data);
1089 error = netdev_linux_do_ioctl(name, &ifr, cmd, cmd_name);
1090 memcpy(data, &ifr.ifr_data, sizeof *data);
1096 netdev_linux_get_miimon(const char *name, bool *miimon)
1098 struct mii_ioctl_data data;
1103 memset(&data, 0, sizeof data);
1104 error = netdev_linux_do_miimon(name, SIOCGMIIPHY, "SIOCGMIIPHY", &data);
1106 /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */
1107 data.reg_num = MII_BMSR;
1108 error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG",
1112 *miimon = !!(data.val_out & BMSR_LSTATUS);
1114 VLOG_WARN_RL(&rl, "%s: failed to query MII", name);
1117 struct ethtool_cmd ecmd;
1119 VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool",
1122 memset(&ecmd, 0, sizeof ecmd);
1123 error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK,
1126 struct ethtool_value eval;
1128 memcpy(&eval, &ecmd, sizeof eval);
1129 *miimon = !!eval.data;
1131 VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name);
1139 netdev_linux_set_miimon_interval(struct netdev *netdev_,
1140 long long int interval)
1142 struct netdev_dev_linux *netdev_dev;
1144 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev_));
1146 interval = interval > 0 ? MAX(interval, 100) : 0;
1147 if (netdev_dev->miimon_interval != interval) {
1148 netdev_dev->miimon_interval = interval;
1149 timer_set_expired(&netdev_dev->miimon_timer);
1156 netdev_linux_miimon_run(void)
1158 struct shash device_shash;
1159 struct shash_node *node;
1161 shash_init(&device_shash);
1162 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
1163 SHASH_FOR_EACH (node, &device_shash) {
1164 struct netdev_dev_linux *dev = node->data;
1167 if (dev->miimon_interval <= 0 || !timer_expired(&dev->miimon_timer)) {
1171 netdev_linux_get_miimon(dev->netdev_dev.name, &miimon);
1172 if (miimon != dev->miimon) {
1173 dev->miimon = miimon;
1174 netdev_dev_linux_changed(dev);
1177 timer_set_duration(&dev->miimon_timer, dev->miimon_interval);
1180 shash_destroy(&device_shash);
1184 netdev_linux_miimon_wait(void)
1186 struct shash device_shash;
1187 struct shash_node *node;
1189 shash_init(&device_shash);
1190 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
1191 SHASH_FOR_EACH (node, &device_shash) {
1192 struct netdev_dev_linux *dev = node->data;
1194 if (dev->miimon_interval > 0) {
1195 timer_wait(&dev->miimon_timer);
1198 shash_destroy(&device_shash);
1201 /* Check whether we can we use RTM_GETLINK to get network device statistics.
1202 * In pre-2.6.19 kernels, this was only available if wireless extensions were
1205 check_for_working_netlink_stats(void)
1207 /* Decide on the netdev_get_stats() implementation to use. Netlink is
1208 * preferable, so if that works, we'll use it. */
1209 int ifindex = do_get_ifindex("lo");
1211 VLOG_WARN("failed to get ifindex for lo, "
1212 "obtaining netdev stats from proc");
1215 struct netdev_stats stats;
1216 int error = get_stats_via_netlink(ifindex, &stats);
1218 VLOG_DBG("obtaining netdev stats via rtnetlink");
1221 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
1222 "via proc (you are probably running a pre-2.6.19 "
1223 "kernel)", strerror(error));
1230 swap_uint64(uint64_t *a, uint64_t *b)
1238 get_stats_via_vport(const struct netdev *netdev_,
1239 struct netdev_stats *stats)
1241 struct netdev_dev_linux *netdev_dev =
1242 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1244 if (netdev_dev->have_vport_stats ||
1245 !(netdev_dev->cache_valid & VALID_HAVE_VPORT_STATS)) {
1248 error = netdev_vport_get_stats(netdev_, stats);
1250 VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed "
1251 "(%s)", netdev_get_name(netdev_), strerror(error));
1253 netdev_dev->have_vport_stats = !error;
1254 netdev_dev->cache_valid |= VALID_HAVE_VPORT_STATS;
1259 netdev_linux_sys_get_stats(const struct netdev *netdev_,
1260 struct netdev_stats *stats)
1262 static int use_netlink_stats = -1;
1265 if (use_netlink_stats < 0) {
1266 use_netlink_stats = check_for_working_netlink_stats();
1269 if (use_netlink_stats) {
1272 error = get_ifindex(netdev_, &ifindex);
1274 error = get_stats_via_netlink(ifindex, stats);
1277 error = get_stats_via_proc(netdev_get_name(netdev_), stats);
1281 VLOG_WARN_RL(&rl, "%s: linux-sys get stats failed %d",
1282 netdev_get_name(netdev_), error);
1288 /* Retrieves current device stats for 'netdev-linux'. */
1290 netdev_linux_get_stats(const struct netdev *netdev_,
1291 struct netdev_stats *stats)
1293 struct netdev_dev_linux *netdev_dev =
1294 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1295 struct netdev_stats dev_stats;
1298 get_stats_via_vport(netdev_, stats);
1300 error = netdev_linux_sys_get_stats(netdev_, &dev_stats);
1303 if (!netdev_dev->have_vport_stats) {
1310 if (!netdev_dev->have_vport_stats) {
1311 /* stats not available from OVS then use ioctl stats. */
1314 stats->rx_errors += dev_stats.rx_errors;
1315 stats->tx_errors += dev_stats.tx_errors;
1316 stats->rx_dropped += dev_stats.rx_dropped;
1317 stats->tx_dropped += dev_stats.tx_dropped;
1318 stats->multicast += dev_stats.multicast;
1319 stats->collisions += dev_stats.collisions;
1320 stats->rx_length_errors += dev_stats.rx_length_errors;
1321 stats->rx_over_errors += dev_stats.rx_over_errors;
1322 stats->rx_crc_errors += dev_stats.rx_crc_errors;
1323 stats->rx_frame_errors += dev_stats.rx_frame_errors;
1324 stats->rx_fifo_errors += dev_stats.rx_fifo_errors;
1325 stats->rx_missed_errors += dev_stats.rx_missed_errors;
1326 stats->tx_aborted_errors += dev_stats.tx_aborted_errors;
1327 stats->tx_carrier_errors += dev_stats.tx_carrier_errors;
1328 stats->tx_fifo_errors += dev_stats.tx_fifo_errors;
1329 stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors;
1330 stats->tx_window_errors += dev_stats.tx_window_errors;
1335 /* Retrieves current device stats for 'netdev-tap' netdev or
1336 * netdev-internal. */
1338 netdev_pseudo_get_stats(const struct netdev *netdev_,
1339 struct netdev_stats *stats)
1341 struct netdev_dev_linux *netdev_dev =
1342 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1343 struct netdev_stats dev_stats;
1346 get_stats_via_vport(netdev_, stats);
1348 error = netdev_linux_sys_get_stats(netdev_, &dev_stats);
1350 if (!netdev_dev->have_vport_stats) {
1357 /* If this port is an internal port then the transmit and receive stats
1358 * will appear to be swapped relative to the other ports since we are the
1359 * one sending the data, not a remote computer. For consistency, we swap
1360 * them back here. This does not apply if we are getting stats from the
1361 * vport layer because it always tracks stats from the perspective of the
1363 if (!netdev_dev->have_vport_stats) {
1365 swap_uint64(&stats->rx_packets, &stats->tx_packets);
1366 swap_uint64(&stats->rx_bytes, &stats->tx_bytes);
1367 swap_uint64(&stats->rx_errors, &stats->tx_errors);
1368 swap_uint64(&stats->rx_dropped, &stats->tx_dropped);
1369 stats->rx_length_errors = 0;
1370 stats->rx_over_errors = 0;
1371 stats->rx_crc_errors = 0;
1372 stats->rx_frame_errors = 0;
1373 stats->rx_fifo_errors = 0;
1374 stats->rx_missed_errors = 0;
1375 stats->tx_aborted_errors = 0;
1376 stats->tx_carrier_errors = 0;
1377 stats->tx_fifo_errors = 0;
1378 stats->tx_heartbeat_errors = 0;
1379 stats->tx_window_errors = 0;
1381 stats->rx_dropped += dev_stats.tx_dropped;
1382 stats->tx_dropped += dev_stats.rx_dropped;
1384 stats->rx_errors += dev_stats.tx_errors;
1385 stats->tx_errors += dev_stats.rx_errors;
1387 stats->multicast += dev_stats.multicast;
1388 stats->collisions += dev_stats.collisions;
1393 /* Stores the features supported by 'netdev' into each of '*current',
1394 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
1395 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
1396 * successful, otherwise a positive errno value. */
1398 netdev_linux_get_features(const struct netdev *netdev,
1399 uint32_t *current, uint32_t *advertised,
1400 uint32_t *supported, uint32_t *peer)
1402 struct ethtool_cmd ecmd;
1405 memset(&ecmd, 0, sizeof ecmd);
1406 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1407 ETHTOOL_GSET, "ETHTOOL_GSET");
1412 /* Supported features. */
1414 if (ecmd.supported & SUPPORTED_10baseT_Half) {
1415 *supported |= OFPPF_10MB_HD;
1417 if (ecmd.supported & SUPPORTED_10baseT_Full) {
1418 *supported |= OFPPF_10MB_FD;
1420 if (ecmd.supported & SUPPORTED_100baseT_Half) {
1421 *supported |= OFPPF_100MB_HD;
1423 if (ecmd.supported & SUPPORTED_100baseT_Full) {
1424 *supported |= OFPPF_100MB_FD;
1426 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
1427 *supported |= OFPPF_1GB_HD;
1429 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
1430 *supported |= OFPPF_1GB_FD;
1432 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
1433 *supported |= OFPPF_10GB_FD;
1435 if (ecmd.supported & SUPPORTED_TP) {
1436 *supported |= OFPPF_COPPER;
1438 if (ecmd.supported & SUPPORTED_FIBRE) {
1439 *supported |= OFPPF_FIBER;
1441 if (ecmd.supported & SUPPORTED_Autoneg) {
1442 *supported |= OFPPF_AUTONEG;
1444 if (ecmd.supported & SUPPORTED_Pause) {
1445 *supported |= OFPPF_PAUSE;
1447 if (ecmd.supported & SUPPORTED_Asym_Pause) {
1448 *supported |= OFPPF_PAUSE_ASYM;
1451 /* Advertised features. */
1453 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
1454 *advertised |= OFPPF_10MB_HD;
1456 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
1457 *advertised |= OFPPF_10MB_FD;
1459 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
1460 *advertised |= OFPPF_100MB_HD;
1462 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
1463 *advertised |= OFPPF_100MB_FD;
1465 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
1466 *advertised |= OFPPF_1GB_HD;
1468 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
1469 *advertised |= OFPPF_1GB_FD;
1471 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
1472 *advertised |= OFPPF_10GB_FD;
1474 if (ecmd.advertising & ADVERTISED_TP) {
1475 *advertised |= OFPPF_COPPER;
1477 if (ecmd.advertising & ADVERTISED_FIBRE) {
1478 *advertised |= OFPPF_FIBER;
1480 if (ecmd.advertising & ADVERTISED_Autoneg) {
1481 *advertised |= OFPPF_AUTONEG;
1483 if (ecmd.advertising & ADVERTISED_Pause) {
1484 *advertised |= OFPPF_PAUSE;
1486 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
1487 *advertised |= OFPPF_PAUSE_ASYM;
1490 /* Current settings. */
1491 if (ecmd.speed == SPEED_10) {
1492 *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD;
1493 } else if (ecmd.speed == SPEED_100) {
1494 *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD;
1495 } else if (ecmd.speed == SPEED_1000) {
1496 *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD;
1497 } else if (ecmd.speed == SPEED_10000) {
1498 *current = OFPPF_10GB_FD;
1503 if (ecmd.port == PORT_TP) {
1504 *current |= OFPPF_COPPER;
1505 } else if (ecmd.port == PORT_FIBRE) {
1506 *current |= OFPPF_FIBER;
1510 *current |= OFPPF_AUTONEG;
1513 /* Peer advertisements. */
1514 *peer = 0; /* XXX */
1519 /* Set the features advertised by 'netdev' to 'advertise'. */
1521 netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
1523 struct ethtool_cmd ecmd;
1526 memset(&ecmd, 0, sizeof ecmd);
1527 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1528 ETHTOOL_GSET, "ETHTOOL_GSET");
1533 ecmd.advertising = 0;
1534 if (advertise & OFPPF_10MB_HD) {
1535 ecmd.advertising |= ADVERTISED_10baseT_Half;
1537 if (advertise & OFPPF_10MB_FD) {
1538 ecmd.advertising |= ADVERTISED_10baseT_Full;
1540 if (advertise & OFPPF_100MB_HD) {
1541 ecmd.advertising |= ADVERTISED_100baseT_Half;
1543 if (advertise & OFPPF_100MB_FD) {
1544 ecmd.advertising |= ADVERTISED_100baseT_Full;
1546 if (advertise & OFPPF_1GB_HD) {
1547 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1549 if (advertise & OFPPF_1GB_FD) {
1550 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1552 if (advertise & OFPPF_10GB_FD) {
1553 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1555 if (advertise & OFPPF_COPPER) {
1556 ecmd.advertising |= ADVERTISED_TP;
1558 if (advertise & OFPPF_FIBER) {
1559 ecmd.advertising |= ADVERTISED_FIBRE;
1561 if (advertise & OFPPF_AUTONEG) {
1562 ecmd.advertising |= ADVERTISED_Autoneg;
1564 if (advertise & OFPPF_PAUSE) {
1565 ecmd.advertising |= ADVERTISED_Pause;
1567 if (advertise & OFPPF_PAUSE_ASYM) {
1568 ecmd.advertising |= ADVERTISED_Asym_Pause;
1570 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1571 ETHTOOL_SSET, "ETHTOOL_SSET");
1574 /* Attempts to set input rate limiting (policing) policy. Returns 0 if
1575 * successful, otherwise a positive errno value. */
1577 netdev_linux_set_policing(struct netdev *netdev,
1578 uint32_t kbits_rate, uint32_t kbits_burst)
1580 struct netdev_dev_linux *netdev_dev =
1581 netdev_dev_linux_cast(netdev_get_dev(netdev));
1582 const char *netdev_name = netdev_get_name(netdev);
1585 COVERAGE_INC(netdev_set_policing);
1587 kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */
1588 : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */
1589 : kbits_burst); /* Stick with user-specified value. */
1591 if (netdev_dev->cache_valid & VALID_POLICING
1592 && netdev_dev->kbits_rate == kbits_rate
1593 && netdev_dev->kbits_burst == kbits_burst) {
1594 /* Assume that settings haven't changed since we last set them. */
1598 /* Remove any existing ingress qdisc. */
1599 error = tc_add_del_ingress_qdisc(netdev, false);
1601 VLOG_WARN_RL(&rl, "%s: removing policing failed: %s",
1602 netdev_name, strerror(error));
1607 error = tc_add_del_ingress_qdisc(netdev, true);
1609 VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s",
1610 netdev_name, strerror(error));
1614 error = tc_add_policer(netdev, kbits_rate, kbits_burst);
1616 VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s",
1617 netdev_name, strerror(error));
1622 netdev_dev->kbits_rate = kbits_rate;
1623 netdev_dev->kbits_burst = kbits_burst;
1624 netdev_dev->cache_valid |= VALID_POLICING;
1630 netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED,
1633 const struct tc_ops **opsp;
1635 for (opsp = tcs; *opsp != NULL; opsp++) {
1636 const struct tc_ops *ops = *opsp;
1637 if (ops->tc_install && ops->ovs_name[0] != '\0') {
1638 sset_add(types, ops->ovs_name);
1644 static const struct tc_ops *
1645 tc_lookup_ovs_name(const char *name)
1647 const struct tc_ops **opsp;
1649 for (opsp = tcs; *opsp != NULL; opsp++) {
1650 const struct tc_ops *ops = *opsp;
1651 if (!strcmp(name, ops->ovs_name)) {
1658 static const struct tc_ops *
1659 tc_lookup_linux_name(const char *name)
1661 const struct tc_ops **opsp;
1663 for (opsp = tcs; *opsp != NULL; opsp++) {
1664 const struct tc_ops *ops = *opsp;
1665 if (ops->linux_name && !strcmp(name, ops->linux_name)) {
1672 static struct tc_queue *
1673 tc_find_queue__(const struct netdev *netdev, unsigned int queue_id,
1676 struct netdev_dev_linux *netdev_dev =
1677 netdev_dev_linux_cast(netdev_get_dev(netdev));
1678 struct tc_queue *queue;
1680 HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev_dev->tc->queues) {
1681 if (queue->queue_id == queue_id) {
1688 static struct tc_queue *
1689 tc_find_queue(const struct netdev *netdev, unsigned int queue_id)
1691 return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0));
1695 netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED,
1697 struct netdev_qos_capabilities *caps)
1699 const struct tc_ops *ops = tc_lookup_ovs_name(type);
1703 caps->n_queues = ops->n_queues;
1708 netdev_linux_get_qos(const struct netdev *netdev,
1709 const char **typep, struct shash *details)
1711 struct netdev_dev_linux *netdev_dev =
1712 netdev_dev_linux_cast(netdev_get_dev(netdev));
1715 error = tc_query_qdisc(netdev);
1720 *typep = netdev_dev->tc->ops->ovs_name;
1721 return (netdev_dev->tc->ops->qdisc_get
1722 ? netdev_dev->tc->ops->qdisc_get(netdev, details)
1727 netdev_linux_set_qos(struct netdev *netdev,
1728 const char *type, const struct shash *details)
1730 struct netdev_dev_linux *netdev_dev =
1731 netdev_dev_linux_cast(netdev_get_dev(netdev));
1732 const struct tc_ops *new_ops;
1735 new_ops = tc_lookup_ovs_name(type);
1736 if (!new_ops || !new_ops->tc_install) {
1740 error = tc_query_qdisc(netdev);
1745 if (new_ops == netdev_dev->tc->ops) {
1746 return new_ops->qdisc_set ? new_ops->qdisc_set(netdev, details) : 0;
1748 /* Delete existing qdisc. */
1749 error = tc_del_qdisc(netdev);
1753 assert(netdev_dev->tc == NULL);
1755 /* Install new qdisc. */
1756 error = new_ops->tc_install(netdev, details);
1757 assert((error == 0) == (netdev_dev->tc != NULL));
1764 netdev_linux_get_queue(const struct netdev *netdev,
1765 unsigned int queue_id, struct shash *details)
1767 struct netdev_dev_linux *netdev_dev =
1768 netdev_dev_linux_cast(netdev_get_dev(netdev));
1771 error = tc_query_qdisc(netdev);
1775 struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1777 ? netdev_dev->tc->ops->class_get(netdev, queue, details)
1783 netdev_linux_set_queue(struct netdev *netdev,
1784 unsigned int queue_id, const struct shash *details)
1786 struct netdev_dev_linux *netdev_dev =
1787 netdev_dev_linux_cast(netdev_get_dev(netdev));
1790 error = tc_query_qdisc(netdev);
1793 } else if (queue_id >= netdev_dev->tc->ops->n_queues
1794 || !netdev_dev->tc->ops->class_set) {
1798 return netdev_dev->tc->ops->class_set(netdev, queue_id, details);
1802 netdev_linux_delete_queue(struct netdev *netdev, unsigned int queue_id)
1804 struct netdev_dev_linux *netdev_dev =
1805 netdev_dev_linux_cast(netdev_get_dev(netdev));
1808 error = tc_query_qdisc(netdev);
1811 } else if (!netdev_dev->tc->ops->class_delete) {
1814 struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1816 ? netdev_dev->tc->ops->class_delete(netdev, queue)
1822 netdev_linux_get_queue_stats(const struct netdev *netdev,
1823 unsigned int queue_id,
1824 struct netdev_queue_stats *stats)
1826 struct netdev_dev_linux *netdev_dev =
1827 netdev_dev_linux_cast(netdev_get_dev(netdev));
1830 error = tc_query_qdisc(netdev);
1833 } else if (!netdev_dev->tc->ops->class_get_stats) {
1836 const struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1838 ? netdev_dev->tc->ops->class_get_stats(netdev, queue, stats)
1844 start_queue_dump(const struct netdev *netdev, struct nl_dump *dump)
1846 struct ofpbuf request;
1847 struct tcmsg *tcmsg;
1849 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request);
1853 tcmsg->tcm_parent = 0;
1854 nl_dump_start(dump, rtnl_sock, &request);
1855 ofpbuf_uninit(&request);
1860 netdev_linux_dump_queues(const struct netdev *netdev,
1861 netdev_dump_queues_cb *cb, void *aux)
1863 struct netdev_dev_linux *netdev_dev =
1864 netdev_dev_linux_cast(netdev_get_dev(netdev));
1865 struct tc_queue *queue;
1866 struct shash details;
1870 error = tc_query_qdisc(netdev);
1873 } else if (!netdev_dev->tc->ops->class_get) {
1878 shash_init(&details);
1879 HMAP_FOR_EACH (queue, hmap_node, &netdev_dev->tc->queues) {
1880 shash_clear(&details);
1882 error = netdev_dev->tc->ops->class_get(netdev, queue, &details);
1884 (*cb)(queue->queue_id, &details, aux);
1889 shash_destroy(&details);
1895 netdev_linux_dump_queue_stats(const struct netdev *netdev,
1896 netdev_dump_queue_stats_cb *cb, void *aux)
1898 struct netdev_dev_linux *netdev_dev =
1899 netdev_dev_linux_cast(netdev_get_dev(netdev));
1900 struct nl_dump dump;
1905 error = tc_query_qdisc(netdev);
1908 } else if (!netdev_dev->tc->ops->class_dump_stats) {
1913 if (!start_queue_dump(netdev, &dump)) {
1916 while (nl_dump_next(&dump, &msg)) {
1917 error = netdev_dev->tc->ops->class_dump_stats(netdev, &msg, cb, aux);
1923 error = nl_dump_done(&dump);
1924 return error ? error : last_error;
1928 netdev_linux_get_in4(const struct netdev *netdev_,
1929 struct in_addr *address, struct in_addr *netmask)
1931 struct netdev_dev_linux *netdev_dev =
1932 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1934 if (!(netdev_dev->cache_valid & VALID_IN4)) {
1937 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
1938 SIOCGIFADDR, "SIOCGIFADDR");
1943 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
1944 SIOCGIFNETMASK, "SIOCGIFNETMASK");
1949 netdev_dev->cache_valid |= VALID_IN4;
1951 *address = netdev_dev->address;
1952 *netmask = netdev_dev->netmask;
1953 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
1957 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
1958 struct in_addr netmask)
1960 struct netdev_dev_linux *netdev_dev =
1961 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1964 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
1966 netdev_dev->cache_valid |= VALID_IN4;
1967 netdev_dev->address = address;
1968 netdev_dev->netmask = netmask;
1969 if (address.s_addr != INADDR_ANY) {
1970 error = do_set_addr(netdev_, SIOCSIFNETMASK,
1971 "SIOCSIFNETMASK", netmask);
1978 parse_if_inet6_line(const char *line,
1979 struct in6_addr *in6, char ifname[16 + 1])
1981 uint8_t *s6 = in6->s6_addr;
1982 #define X8 "%2"SCNx8
1984 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
1985 "%*x %*x %*x %*x %16s\n",
1986 &s6[0], &s6[1], &s6[2], &s6[3],
1987 &s6[4], &s6[5], &s6[6], &s6[7],
1988 &s6[8], &s6[9], &s6[10], &s6[11],
1989 &s6[12], &s6[13], &s6[14], &s6[15],
1993 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
1994 * 'in6' is non-null) and returns true. Otherwise, returns false. */
1996 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
1998 struct netdev_dev_linux *netdev_dev =
1999 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2000 if (!(netdev_dev->cache_valid & VALID_IN6)) {
2004 netdev_dev->in6 = in6addr_any;
2006 file = fopen("/proc/net/if_inet6", "r");
2008 const char *name = netdev_get_name(netdev_);
2009 while (fgets(line, sizeof line, file)) {
2010 struct in6_addr in6_tmp;
2011 char ifname[16 + 1];
2012 if (parse_if_inet6_line(line, &in6_tmp, ifname)
2013 && !strcmp(name, ifname))
2015 netdev_dev->in6 = in6_tmp;
2021 netdev_dev->cache_valid |= VALID_IN6;
2023 *in6 = netdev_dev->in6;
2028 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
2030 struct sockaddr_in sin;
2031 memset(&sin, 0, sizeof sin);
2032 sin.sin_family = AF_INET;
2033 sin.sin_addr = addr;
2036 memset(sa, 0, sizeof *sa);
2037 memcpy(sa, &sin, sizeof sin);
2041 do_set_addr(struct netdev *netdev,
2042 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
2045 ovs_strzcpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
2046 make_in4_sockaddr(&ifr.ifr_addr, addr);
2048 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
2052 /* Adds 'router' as a default IP gateway. */
2054 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
2056 struct in_addr any = { INADDR_ANY };
2060 memset(&rt, 0, sizeof rt);
2061 make_in4_sockaddr(&rt.rt_dst, any);
2062 make_in4_sockaddr(&rt.rt_gateway, router);
2063 make_in4_sockaddr(&rt.rt_genmask, any);
2064 rt.rt_flags = RTF_UP | RTF_GATEWAY;
2065 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
2067 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
2073 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
2076 static const char fn[] = "/proc/net/route";
2081 *netdev_name = NULL;
2082 stream = fopen(fn, "r");
2083 if (stream == NULL) {
2084 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
2089 while (fgets(line, sizeof line, stream)) {
2092 ovs_be32 dest, gateway, mask;
2093 int refcnt, metric, mtu;
2094 unsigned int flags, use, window, irtt;
2097 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
2099 iface, &dest, &gateway, &flags, &refcnt,
2100 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
2102 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
2106 if (!(flags & RTF_UP)) {
2107 /* Skip routes that aren't up. */
2111 /* The output of 'dest', 'mask', and 'gateway' were given in
2112 * network byte order, so we don't need need any endian
2113 * conversions here. */
2114 if ((dest & mask) == (host->s_addr & mask)) {
2116 /* The host is directly reachable. */
2117 next_hop->s_addr = 0;
2119 /* To reach the host, we must go through a gateway. */
2120 next_hop->s_addr = gateway;
2122 *netdev_name = xstrdup(iface);
2134 netdev_linux_get_status(const struct netdev *netdev, struct shash *sh)
2136 struct ethtool_drvinfo drvinfo;
2139 memset(&drvinfo, 0, sizeof drvinfo);
2140 error = netdev_linux_do_ethtool(netdev_get_name(netdev),
2141 (struct ethtool_cmd *)&drvinfo,
2143 "ETHTOOL_GDRVINFO");
2145 shash_add(sh, "driver_name", xstrdup(drvinfo.driver));
2146 shash_add(sh, "driver_version", xstrdup(drvinfo.version));
2147 shash_add(sh, "firmware_version", xstrdup(drvinfo.fw_version));
2153 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
2154 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
2155 * returns 0. Otherwise, it returns a positive errno value; in particular,
2156 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
2158 netdev_linux_arp_lookup(const struct netdev *netdev,
2159 ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
2162 struct sockaddr_in sin;
2165 memset(&r, 0, sizeof r);
2166 memset(&sin, 0, sizeof sin);
2167 sin.sin_family = AF_INET;
2168 sin.sin_addr.s_addr = ip;
2170 memcpy(&r.arp_pa, &sin, sizeof sin);
2171 r.arp_ha.sa_family = ARPHRD_ETHER;
2173 ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
2174 COVERAGE_INC(netdev_arp_lookup);
2175 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
2177 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
2178 } else if (retval != ENXIO) {
2179 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
2180 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
2186 nd_to_iff_flags(enum netdev_flags nd)
2189 if (nd & NETDEV_UP) {
2192 if (nd & NETDEV_PROMISC) {
2199 iff_to_nd_flags(int iff)
2201 enum netdev_flags nd = 0;
2205 if (iff & IFF_PROMISC) {
2206 nd |= NETDEV_PROMISC;
2212 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
2213 enum netdev_flags on, enum netdev_flags *old_flagsp)
2215 struct netdev_dev_linux *netdev_dev;
2216 int old_flags, new_flags;
2219 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
2220 old_flags = netdev_dev->ifi_flags;
2221 *old_flagsp = iff_to_nd_flags(old_flags);
2222 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
2223 if (new_flags != old_flags) {
2224 error = set_flags(netdev, new_flags);
2225 get_flags(&netdev_dev->netdev_dev, &netdev_dev->ifi_flags);
2231 netdev_linux_change_seq(const struct netdev *netdev)
2233 return netdev_dev_linux_cast(netdev_get_dev(netdev))->change_seq;
2236 #define NETDEV_LINUX_CLASS(NAME, CREATE, GET_STATS, SET_STATS) \
2240 netdev_linux_init, \
2242 netdev_linux_wait, \
2245 netdev_linux_destroy, \
2246 NULL, /* get_config */ \
2247 NULL, /* set_config */ \
2249 netdev_linux_open, \
2250 netdev_linux_close, \
2252 netdev_linux_listen, \
2253 netdev_linux_recv, \
2254 netdev_linux_recv_wait, \
2255 netdev_linux_drain, \
2257 netdev_linux_send, \
2258 netdev_linux_send_wait, \
2260 netdev_linux_set_etheraddr, \
2261 netdev_linux_get_etheraddr, \
2262 netdev_linux_get_mtu, \
2263 netdev_linux_set_mtu, \
2264 netdev_linux_get_ifindex, \
2265 netdev_linux_get_carrier, \
2266 netdev_linux_get_carrier_resets, \
2267 netdev_linux_set_miimon_interval, \
2271 netdev_linux_get_features, \
2272 netdev_linux_set_advertisements, \
2274 netdev_linux_set_policing, \
2275 netdev_linux_get_qos_types, \
2276 netdev_linux_get_qos_capabilities, \
2277 netdev_linux_get_qos, \
2278 netdev_linux_set_qos, \
2279 netdev_linux_get_queue, \
2280 netdev_linux_set_queue, \
2281 netdev_linux_delete_queue, \
2282 netdev_linux_get_queue_stats, \
2283 netdev_linux_dump_queues, \
2284 netdev_linux_dump_queue_stats, \
2286 netdev_linux_get_in4, \
2287 netdev_linux_set_in4, \
2288 netdev_linux_get_in6, \
2289 netdev_linux_add_router, \
2290 netdev_linux_get_next_hop, \
2291 netdev_linux_get_status, \
2292 netdev_linux_arp_lookup, \
2294 netdev_linux_update_flags, \
2296 netdev_linux_change_seq \
2299 const struct netdev_class netdev_linux_class =
2302 netdev_linux_create,
2303 netdev_linux_get_stats,
2304 NULL); /* set_stats */
2306 const struct netdev_class netdev_tap_class =
2309 netdev_linux_create_tap,
2310 netdev_pseudo_get_stats,
2311 NULL); /* set_stats */
2313 const struct netdev_class netdev_internal_class =
2316 netdev_linux_create,
2317 netdev_pseudo_get_stats,
2318 netdev_vport_set_stats);
2320 /* HTB traffic control class. */
2322 #define HTB_N_QUEUES 0xf000
2326 unsigned int max_rate; /* In bytes/s. */
2330 struct tc_queue tc_queue;
2331 unsigned int min_rate; /* In bytes/s. */
2332 unsigned int max_rate; /* In bytes/s. */
2333 unsigned int burst; /* In bytes. */
2334 unsigned int priority; /* Lower values are higher priorities. */
2338 htb_get__(const struct netdev *netdev)
2340 struct netdev_dev_linux *netdev_dev =
2341 netdev_dev_linux_cast(netdev_get_dev(netdev));
2342 return CONTAINER_OF(netdev_dev->tc, struct htb, tc);
2346 htb_install__(struct netdev *netdev, uint64_t max_rate)
2348 struct netdev_dev_linux *netdev_dev =
2349 netdev_dev_linux_cast(netdev_get_dev(netdev));
2352 htb = xmalloc(sizeof *htb);
2353 tc_init(&htb->tc, &tc_ops_htb);
2354 htb->max_rate = max_rate;
2356 netdev_dev->tc = &htb->tc;
2359 /* Create an HTB qdisc.
2361 * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */
2363 htb_setup_qdisc__(struct netdev *netdev)
2366 struct tc_htb_glob opt;
2367 struct ofpbuf request;
2368 struct tcmsg *tcmsg;
2370 tc_del_qdisc(netdev);
2372 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
2373 NLM_F_EXCL | NLM_F_CREATE, &request);
2377 tcmsg->tcm_handle = tc_make_handle(1, 0);
2378 tcmsg->tcm_parent = TC_H_ROOT;
2380 nl_msg_put_string(&request, TCA_KIND, "htb");
2382 memset(&opt, 0, sizeof opt);
2383 opt.rate2quantum = 10;
2387 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2388 nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt);
2389 nl_msg_end_nested(&request, opt_offset);
2391 return tc_transact(&request, NULL);
2394 /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb
2395 * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */
2397 htb_setup_class__(struct netdev *netdev, unsigned int handle,
2398 unsigned int parent, struct htb_class *class)
2401 struct tc_htb_opt opt;
2402 struct ofpbuf request;
2403 struct tcmsg *tcmsg;
2407 error = netdev_get_mtu(netdev, &mtu);
2409 VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU",
2410 netdev_get_name(netdev));
2414 memset(&opt, 0, sizeof opt);
2415 tc_fill_rate(&opt.rate, class->min_rate, mtu);
2416 tc_fill_rate(&opt.ceil, class->max_rate, mtu);
2417 opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst);
2418 opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst);
2419 opt.prio = class->priority;
2421 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
2425 tcmsg->tcm_handle = handle;
2426 tcmsg->tcm_parent = parent;
2428 nl_msg_put_string(&request, TCA_KIND, "htb");
2429 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2430 nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt);
2431 tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate);
2432 tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil);
2433 nl_msg_end_nested(&request, opt_offset);
2435 error = tc_transact(&request, NULL);
2437 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
2438 "min_rate=%u max_rate=%u burst=%u prio=%u (%s)",
2439 netdev_get_name(netdev),
2440 tc_get_major(handle), tc_get_minor(handle),
2441 tc_get_major(parent), tc_get_minor(parent),
2442 class->min_rate, class->max_rate,
2443 class->burst, class->priority, strerror(error));
2448 /* Parses Netlink attributes in 'options' for HTB parameters and stores a
2449 * description of them into 'details'. The description complies with the
2450 * specification given in the vswitch database documentation for linux-htb
2453 htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class)
2455 static const struct nl_policy tca_htb_policy[] = {
2456 [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false,
2457 .min_len = sizeof(struct tc_htb_opt) },
2460 struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)];
2461 const struct tc_htb_opt *htb;
2463 if (!nl_parse_nested(nl_options, tca_htb_policy,
2464 attrs, ARRAY_SIZE(tca_htb_policy))) {
2465 VLOG_WARN_RL(&rl, "failed to parse HTB class options");
2469 htb = nl_attr_get(attrs[TCA_HTB_PARMS]);
2470 class->min_rate = htb->rate.rate;
2471 class->max_rate = htb->ceil.rate;
2472 class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer);
2473 class->priority = htb->prio;
2478 htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
2479 struct htb_class *options,
2480 struct netdev_queue_stats *stats)
2482 struct nlattr *nl_options;
2483 unsigned int handle;
2486 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
2487 if (!error && queue_id) {
2488 unsigned int major = tc_get_major(handle);
2489 unsigned int minor = tc_get_minor(handle);
2490 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2491 *queue_id = minor - 1;
2496 if (!error && options) {
2497 error = htb_parse_tca_options__(nl_options, options);
2503 htb_parse_qdisc_details__(struct netdev *netdev,
2504 const struct shash *details, struct htb_class *hc)
2506 const char *max_rate_s;
2508 max_rate_s = shash_find_data(details, "max-rate");
2509 hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
2510 if (!hc->max_rate) {
2513 netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
2514 hc->max_rate = netdev_features_to_bps(current) / 8;
2516 hc->min_rate = hc->max_rate;
2522 htb_parse_class_details__(struct netdev *netdev,
2523 const struct shash *details, struct htb_class *hc)
2525 const struct htb *htb = htb_get__(netdev);
2526 const char *min_rate_s = shash_find_data(details, "min-rate");
2527 const char *max_rate_s = shash_find_data(details, "max-rate");
2528 const char *burst_s = shash_find_data(details, "burst");
2529 const char *priority_s = shash_find_data(details, "priority");
2532 error = netdev_get_mtu(netdev, &mtu);
2534 VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU",
2535 netdev_get_name(netdev));
2539 /* HTB requires at least an mtu sized min-rate to send any traffic even
2540 * on uncongested links. */
2541 hc->min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
2542 hc->min_rate = MAX(hc->min_rate, mtu);
2543 hc->min_rate = MIN(hc->min_rate, htb->max_rate);
2546 hc->max_rate = (max_rate_s
2547 ? strtoull(max_rate_s, NULL, 10) / 8
2549 hc->max_rate = MAX(hc->max_rate, hc->min_rate);
2550 hc->max_rate = MIN(hc->max_rate, htb->max_rate);
2554 * According to hints in the documentation that I've read, it is important
2555 * that 'burst' be at least as big as the largest frame that might be
2556 * transmitted. Also, making 'burst' a bit bigger than necessary is OK,
2557 * but having it a bit too small is a problem. Since netdev_get_mtu()
2558 * doesn't include the Ethernet header, we need to add at least 14 (18?) to
2559 * the MTU. We actually add 64, instead of 14, as a guard against
2560 * additional headers get tacked on somewhere that we're not aware of. */
2561 hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0;
2562 hc->burst = MAX(hc->burst, mtu + 64);
2565 hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0;
2571 htb_query_class__(const struct netdev *netdev, unsigned int handle,
2572 unsigned int parent, struct htb_class *options,
2573 struct netdev_queue_stats *stats)
2575 struct ofpbuf *reply;
2578 error = tc_query_class(netdev, handle, parent, &reply);
2580 error = htb_parse_tcmsg__(reply, NULL, options, stats);
2581 ofpbuf_delete(reply);
2587 htb_tc_install(struct netdev *netdev, const struct shash *details)
2591 error = htb_setup_qdisc__(netdev);
2593 struct htb_class hc;
2595 htb_parse_qdisc_details__(netdev, details, &hc);
2596 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
2597 tc_make_handle(1, 0), &hc);
2599 htb_install__(netdev, hc.max_rate);
2605 static struct htb_class *
2606 htb_class_cast__(const struct tc_queue *queue)
2608 return CONTAINER_OF(queue, struct htb_class, tc_queue);
2612 htb_update_queue__(struct netdev *netdev, unsigned int queue_id,
2613 const struct htb_class *hc)
2615 struct htb *htb = htb_get__(netdev);
2616 size_t hash = hash_int(queue_id, 0);
2617 struct tc_queue *queue;
2618 struct htb_class *hcp;
2620 queue = tc_find_queue__(netdev, queue_id, hash);
2622 hcp = htb_class_cast__(queue);
2624 hcp = xmalloc(sizeof *hcp);
2625 queue = &hcp->tc_queue;
2626 queue->queue_id = queue_id;
2627 hmap_insert(&htb->tc.queues, &queue->hmap_node, hash);
2630 hcp->min_rate = hc->min_rate;
2631 hcp->max_rate = hc->max_rate;
2632 hcp->burst = hc->burst;
2633 hcp->priority = hc->priority;
2637 htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
2640 struct nl_dump dump;
2641 struct htb_class hc;
2643 /* Get qdisc options. */
2645 htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
2646 htb_install__(netdev, hc.max_rate);
2649 if (!start_queue_dump(netdev, &dump)) {
2652 while (nl_dump_next(&dump, &msg)) {
2653 unsigned int queue_id;
2655 if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
2656 htb_update_queue__(netdev, queue_id, &hc);
2659 nl_dump_done(&dump);
2665 htb_tc_destroy(struct tc *tc)
2667 struct htb *htb = CONTAINER_OF(tc, struct htb, tc);
2668 struct htb_class *hc, *next;
2670 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) {
2671 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
2679 htb_qdisc_get(const struct netdev *netdev, struct shash *details)
2681 const struct htb *htb = htb_get__(netdev);
2682 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * htb->max_rate));
2687 htb_qdisc_set(struct netdev *netdev, const struct shash *details)
2689 struct htb_class hc;
2692 htb_parse_qdisc_details__(netdev, details, &hc);
2693 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
2694 tc_make_handle(1, 0), &hc);
2696 htb_get__(netdev)->max_rate = hc.max_rate;
2702 htb_class_get(const struct netdev *netdev OVS_UNUSED,
2703 const struct tc_queue *queue, struct shash *details)
2705 const struct htb_class *hc = htb_class_cast__(queue);
2707 shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate));
2708 if (hc->min_rate != hc->max_rate) {
2709 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate));
2711 shash_add(details, "burst", xasprintf("%llu", 8ULL * hc->burst));
2713 shash_add(details, "priority", xasprintf("%u", hc->priority));
2719 htb_class_set(struct netdev *netdev, unsigned int queue_id,
2720 const struct shash *details)
2722 struct htb_class hc;
2725 error = htb_parse_class_details__(netdev, details, &hc);
2730 error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
2731 tc_make_handle(1, 0xfffe), &hc);
2736 htb_update_queue__(netdev, queue_id, &hc);
2741 htb_class_delete(struct netdev *netdev, struct tc_queue *queue)
2743 struct htb_class *hc = htb_class_cast__(queue);
2744 struct htb *htb = htb_get__(netdev);
2747 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
2749 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
2756 htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
2757 struct netdev_queue_stats *stats)
2759 return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
2760 tc_make_handle(1, 0xfffe), NULL, stats);
2764 htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
2765 const struct ofpbuf *nlmsg,
2766 netdev_dump_queue_stats_cb *cb, void *aux)
2768 struct netdev_queue_stats stats;
2769 unsigned int handle, major, minor;
2772 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
2777 major = tc_get_major(handle);
2778 minor = tc_get_minor(handle);
2779 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2780 (*cb)(minor - 1, &stats, aux);
2785 static const struct tc_ops tc_ops_htb = {
2786 "htb", /* linux_name */
2787 "linux-htb", /* ovs_name */
2788 HTB_N_QUEUES, /* n_queues */
2797 htb_class_get_stats,
2798 htb_class_dump_stats
2801 /* "linux-hfsc" traffic control class. */
2803 #define HFSC_N_QUEUES 0xf000
2811 struct tc_queue tc_queue;
2816 static struct hfsc *
2817 hfsc_get__(const struct netdev *netdev)
2819 struct netdev_dev_linux *netdev_dev;
2820 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
2821 return CONTAINER_OF(netdev_dev->tc, struct hfsc, tc);
2824 static struct hfsc_class *
2825 hfsc_class_cast__(const struct tc_queue *queue)
2827 return CONTAINER_OF(queue, struct hfsc_class, tc_queue);
2831 hfsc_install__(struct netdev *netdev, uint32_t max_rate)
2833 struct netdev_dev_linux * netdev_dev;
2836 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
2837 hfsc = xmalloc(sizeof *hfsc);
2838 tc_init(&hfsc->tc, &tc_ops_hfsc);
2839 hfsc->max_rate = max_rate;
2840 netdev_dev->tc = &hfsc->tc;
2844 hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id,
2845 const struct hfsc_class *hc)
2849 struct hfsc_class *hcp;
2850 struct tc_queue *queue;
2852 hfsc = hfsc_get__(netdev);
2853 hash = hash_int(queue_id, 0);
2855 queue = tc_find_queue__(netdev, queue_id, hash);
2857 hcp = hfsc_class_cast__(queue);
2859 hcp = xmalloc(sizeof *hcp);
2860 queue = &hcp->tc_queue;
2861 queue->queue_id = queue_id;
2862 hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash);
2865 hcp->min_rate = hc->min_rate;
2866 hcp->max_rate = hc->max_rate;
2870 hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class)
2872 const struct tc_service_curve *rsc, *fsc, *usc;
2873 static const struct nl_policy tca_hfsc_policy[] = {
2875 .type = NL_A_UNSPEC,
2877 .min_len = sizeof(struct tc_service_curve),
2880 .type = NL_A_UNSPEC,
2882 .min_len = sizeof(struct tc_service_curve),
2885 .type = NL_A_UNSPEC,
2887 .min_len = sizeof(struct tc_service_curve),
2890 struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)];
2892 if (!nl_parse_nested(nl_options, tca_hfsc_policy,
2893 attrs, ARRAY_SIZE(tca_hfsc_policy))) {
2894 VLOG_WARN_RL(&rl, "failed to parse HFSC class options");
2898 rsc = nl_attr_get(attrs[TCA_HFSC_RSC]);
2899 fsc = nl_attr_get(attrs[TCA_HFSC_FSC]);
2900 usc = nl_attr_get(attrs[TCA_HFSC_USC]);
2902 if (rsc->m1 != 0 || rsc->d != 0 ||
2903 fsc->m1 != 0 || fsc->d != 0 ||
2904 usc->m1 != 0 || usc->d != 0) {
2905 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
2906 "Non-linear service curves are not supported.");
2910 if (rsc->m2 != fsc->m2) {
2911 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
2912 "Real-time service curves are not supported ");
2916 if (rsc->m2 > usc->m2) {
2917 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
2918 "Min-rate service curve is greater than "
2919 "the max-rate service curve.");
2923 class->min_rate = fsc->m2;
2924 class->max_rate = usc->m2;
2929 hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
2930 struct hfsc_class *options,
2931 struct netdev_queue_stats *stats)
2934 unsigned int handle;
2935 struct nlattr *nl_options;
2937 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
2943 unsigned int major, minor;
2945 major = tc_get_major(handle);
2946 minor = tc_get_minor(handle);
2947 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
2948 *queue_id = minor - 1;
2955 error = hfsc_parse_tca_options__(nl_options, options);
2962 hfsc_query_class__(const struct netdev *netdev, unsigned int handle,
2963 unsigned int parent, struct hfsc_class *options,
2964 struct netdev_queue_stats *stats)
2967 struct ofpbuf *reply;
2969 error = tc_query_class(netdev, handle, parent, &reply);
2974 error = hfsc_parse_tcmsg__(reply, NULL, options, stats);
2975 ofpbuf_delete(reply);
2980 hfsc_parse_qdisc_details__(struct netdev *netdev, const struct shash *details,
2981 struct hfsc_class *class)
2984 const char *max_rate_s;
2986 max_rate_s = shash_find_data(details, "max-rate");
2987 max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
2992 netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
2993 max_rate = netdev_features_to_bps(current) / 8;
2996 class->min_rate = max_rate;
2997 class->max_rate = max_rate;
3001 hfsc_parse_class_details__(struct netdev *netdev,
3002 const struct shash *details,
3003 struct hfsc_class * class)
3005 const struct hfsc *hfsc;
3006 uint32_t min_rate, max_rate;
3007 const char *min_rate_s, *max_rate_s;
3009 hfsc = hfsc_get__(netdev);
3010 min_rate_s = shash_find_data(details, "min-rate");
3011 max_rate_s = shash_find_data(details, "max-rate");
3013 min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
3014 min_rate = MAX(min_rate, 1);
3015 min_rate = MIN(min_rate, hfsc->max_rate);
3017 max_rate = (max_rate_s
3018 ? strtoull(max_rate_s, NULL, 10) / 8
3020 max_rate = MAX(max_rate, min_rate);
3021 max_rate = MIN(max_rate, hfsc->max_rate);
3023 class->min_rate = min_rate;
3024 class->max_rate = max_rate;
3029 /* Create an HFSC qdisc.
3031 * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */
3033 hfsc_setup_qdisc__(struct netdev * netdev)
3035 struct tcmsg *tcmsg;
3036 struct ofpbuf request;
3037 struct tc_hfsc_qopt opt;
3039 tc_del_qdisc(netdev);
3041 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
3042 NLM_F_EXCL | NLM_F_CREATE, &request);
3048 tcmsg->tcm_handle = tc_make_handle(1, 0);
3049 tcmsg->tcm_parent = TC_H_ROOT;
3051 memset(&opt, 0, sizeof opt);
3054 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3055 nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt);
3057 return tc_transact(&request, NULL);
3060 /* Create an HFSC class.
3062 * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc
3063 * sc rate <min_rate> ul rate <max_rate>" */
3065 hfsc_setup_class__(struct netdev *netdev, unsigned int handle,
3066 unsigned int parent, struct hfsc_class *class)
3070 struct tcmsg *tcmsg;
3071 struct ofpbuf request;
3072 struct tc_service_curve min, max;
3074 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
3080 tcmsg->tcm_handle = handle;
3081 tcmsg->tcm_parent = parent;
3085 min.m2 = class->min_rate;
3089 max.m2 = class->max_rate;
3091 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3092 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
3093 nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min);
3094 nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min);
3095 nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max);
3096 nl_msg_end_nested(&request, opt_offset);
3098 error = tc_transact(&request, NULL);
3100 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
3101 "min-rate %ubps, max-rate %ubps (%s)",
3102 netdev_get_name(netdev),
3103 tc_get_major(handle), tc_get_minor(handle),
3104 tc_get_major(parent), tc_get_minor(parent),
3105 class->min_rate, class->max_rate, strerror(error));
3112 hfsc_tc_install(struct netdev *netdev, const struct shash *details)
3115 struct hfsc_class class;
3117 error = hfsc_setup_qdisc__(netdev);
3123 hfsc_parse_qdisc_details__(netdev, details, &class);
3124 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3125 tc_make_handle(1, 0), &class);
3131 hfsc_install__(netdev, class.max_rate);
3136 hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3139 struct nl_dump dump;
3140 struct hfsc_class hc;
3143 hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
3144 hfsc_install__(netdev, hc.max_rate);
3146 if (!start_queue_dump(netdev, &dump)) {
3150 while (nl_dump_next(&dump, &msg)) {
3151 unsigned int queue_id;
3153 if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
3154 hfsc_update_queue__(netdev, queue_id, &hc);
3158 nl_dump_done(&dump);
3163 hfsc_tc_destroy(struct tc *tc)
3166 struct hfsc_class *hc, *next;
3168 hfsc = CONTAINER_OF(tc, struct hfsc, tc);
3170 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) {
3171 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3180 hfsc_qdisc_get(const struct netdev *netdev, struct shash *details)
3182 const struct hfsc *hfsc;
3183 hfsc = hfsc_get__(netdev);
3184 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hfsc->max_rate));
3189 hfsc_qdisc_set(struct netdev *netdev, const struct shash *details)
3192 struct hfsc_class class;
3194 hfsc_parse_qdisc_details__(netdev, details, &class);
3195 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3196 tc_make_handle(1, 0), &class);
3199 hfsc_get__(netdev)->max_rate = class.max_rate;
3206 hfsc_class_get(const struct netdev *netdev OVS_UNUSED,
3207 const struct tc_queue *queue, struct shash *details)
3209 const struct hfsc_class *hc;
3211 hc = hfsc_class_cast__(queue);
3212 shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate));
3213 if (hc->min_rate != hc->max_rate) {
3214 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate));
3220 hfsc_class_set(struct netdev *netdev, unsigned int queue_id,
3221 const struct shash *details)
3224 struct hfsc_class class;
3226 error = hfsc_parse_class_details__(netdev, details, &class);
3231 error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
3232 tc_make_handle(1, 0xfffe), &class);
3237 hfsc_update_queue__(netdev, queue_id, &class);
3242 hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue)
3246 struct hfsc_class *hc;
3248 hc = hfsc_class_cast__(queue);
3249 hfsc = hfsc_get__(netdev);
3251 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
3253 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3260 hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
3261 struct netdev_queue_stats *stats)
3263 return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
3264 tc_make_handle(1, 0xfffe), NULL, stats);
3268 hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
3269 const struct ofpbuf *nlmsg,
3270 netdev_dump_queue_stats_cb *cb, void *aux)
3272 struct netdev_queue_stats stats;
3273 unsigned int handle, major, minor;
3276 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
3281 major = tc_get_major(handle);
3282 minor = tc_get_minor(handle);
3283 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
3284 (*cb)(minor - 1, &stats, aux);
3289 static const struct tc_ops tc_ops_hfsc = {
3290 "hfsc", /* linux_name */
3291 "linux-hfsc", /* ovs_name */
3292 HFSC_N_QUEUES, /* n_queues */
3293 hfsc_tc_install, /* tc_install */
3294 hfsc_tc_load, /* tc_load */
3295 hfsc_tc_destroy, /* tc_destroy */
3296 hfsc_qdisc_get, /* qdisc_get */
3297 hfsc_qdisc_set, /* qdisc_set */
3298 hfsc_class_get, /* class_get */
3299 hfsc_class_set, /* class_set */
3300 hfsc_class_delete, /* class_delete */
3301 hfsc_class_get_stats, /* class_get_stats */
3302 hfsc_class_dump_stats /* class_dump_stats */
3305 /* "linux-default" traffic control class.
3307 * This class represents the default, unnamed Linux qdisc. It corresponds to
3308 * the "" (empty string) QoS type in the OVS database. */
3311 default_install__(struct netdev *netdev)
3313 struct netdev_dev_linux *netdev_dev =
3314 netdev_dev_linux_cast(netdev_get_dev(netdev));
3315 static struct tc *tc;
3318 tc = xmalloc(sizeof *tc);
3319 tc_init(tc, &tc_ops_default);
3321 netdev_dev->tc = tc;
3325 default_tc_install(struct netdev *netdev,
3326 const struct shash *details OVS_UNUSED)
3328 default_install__(netdev);
3333 default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3335 default_install__(netdev);
3339 static const struct tc_ops tc_ops_default = {
3340 NULL, /* linux_name */
3345 NULL, /* tc_destroy */
3346 NULL, /* qdisc_get */
3347 NULL, /* qdisc_set */
3348 NULL, /* class_get */
3349 NULL, /* class_set */
3350 NULL, /* class_delete */
3351 NULL, /* class_get_stats */
3352 NULL /* class_dump_stats */
3355 /* "linux-other" traffic control class.
3360 other_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3362 struct netdev_dev_linux *netdev_dev =
3363 netdev_dev_linux_cast(netdev_get_dev(netdev));
3364 static struct tc *tc;
3367 tc = xmalloc(sizeof *tc);
3368 tc_init(tc, &tc_ops_other);
3370 netdev_dev->tc = tc;
3374 static const struct tc_ops tc_ops_other = {
3375 NULL, /* linux_name */
3376 "linux-other", /* ovs_name */
3378 NULL, /* tc_install */
3380 NULL, /* tc_destroy */
3381 NULL, /* qdisc_get */
3382 NULL, /* qdisc_set */
3383 NULL, /* class_get */
3384 NULL, /* class_set */
3385 NULL, /* class_delete */
3386 NULL, /* class_get_stats */
3387 NULL /* class_dump_stats */
3390 /* Traffic control. */
3392 /* Number of kernel "tc" ticks per second. */
3393 static double ticks_per_s;
3395 /* Number of kernel "jiffies" per second. This is used for the purpose of
3396 * computing buffer sizes. Generally kernel qdiscs need to be able to buffer
3397 * one jiffy's worth of data.
3399 * There are two possibilities here:
3401 * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the
3402 * approximate range of 100 to 1024. That means that we really need to
3403 * make sure that the qdisc can buffer that much data.
3405 * - 'buffer_hz' is an absurdly large number. That means that the kernel
3406 * has finely granular timers and there's no need to fudge additional room
3407 * for buffers. (There's no extra effort needed to implement that: the
3408 * large 'buffer_hz' is used as a divisor, so practically any number will
3409 * come out as 0 in the division. Small integer results in the case of
3410 * really high dividends won't have any real effect anyhow.)
3412 static unsigned int buffer_hz;
3414 /* Returns tc handle 'major':'minor'. */
3416 tc_make_handle(unsigned int major, unsigned int minor)
3418 return TC_H_MAKE(major << 16, minor);
3421 /* Returns the major number from 'handle'. */
3423 tc_get_major(unsigned int handle)
3425 return TC_H_MAJ(handle) >> 16;
3428 /* Returns the minor number from 'handle'. */
3430 tc_get_minor(unsigned int handle)
3432 return TC_H_MIN(handle);
3435 static struct tcmsg *
3436 tc_make_request(const struct netdev *netdev, int type, unsigned int flags,
3437 struct ofpbuf *request)
3439 struct tcmsg *tcmsg;
3443 error = get_ifindex(netdev, &ifindex);
3448 ofpbuf_init(request, 512);
3449 nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags);
3450 tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg);
3451 tcmsg->tcm_family = AF_UNSPEC;
3452 tcmsg->tcm_ifindex = ifindex;
3453 /* Caller should fill in tcmsg->tcm_handle. */
3454 /* Caller should fill in tcmsg->tcm_parent. */
3460 tc_transact(struct ofpbuf *request, struct ofpbuf **replyp)
3462 int error = nl_sock_transact(rtnl_sock, request, replyp);
3463 ofpbuf_uninit(request);
3467 /* Adds or deletes a root ingress qdisc on 'netdev'. We use this for
3468 * policing configuration.
3470 * This function is equivalent to running the following when 'add' is true:
3471 * /sbin/tc qdisc add dev <devname> handle ffff: ingress
3473 * This function is equivalent to running the following when 'add' is false:
3474 * /sbin/tc qdisc del dev <devname> handle ffff: ingress
3476 * The configuration and stats may be seen with the following command:
3477 * /sbin/tc -s qdisc show dev <devname>
3479 * Returns 0 if successful, otherwise a positive errno value.
3482 tc_add_del_ingress_qdisc(struct netdev *netdev, bool add)
3484 struct ofpbuf request;
3485 struct tcmsg *tcmsg;
3487 int type = add ? RTM_NEWQDISC : RTM_DELQDISC;
3488 int flags = add ? NLM_F_EXCL | NLM_F_CREATE : 0;
3490 tcmsg = tc_make_request(netdev, type, flags, &request);
3494 tcmsg->tcm_handle = tc_make_handle(0xffff, 0);
3495 tcmsg->tcm_parent = TC_H_INGRESS;
3496 nl_msg_put_string(&request, TCA_KIND, "ingress");
3497 nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0);
3499 error = tc_transact(&request, NULL);
3501 /* If we're deleting the qdisc, don't worry about some of the
3502 * error conditions. */
3503 if (!add && (error == ENOENT || error == EINVAL)) {
3512 /* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size
3515 * This function is equivalent to running:
3516 * /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49
3517 * basic police rate <kbits_rate>kbit burst <kbits_burst>k
3520 * The configuration and stats may be seen with the following command:
3521 * /sbin/tc -s filter show <devname> eth0 parent ffff:
3523 * Returns 0 if successful, otherwise a positive errno value.
3526 tc_add_policer(struct netdev *netdev, int kbits_rate, int kbits_burst)
3528 struct tc_police tc_police;
3529 struct ofpbuf request;
3530 struct tcmsg *tcmsg;
3531 size_t basic_offset;
3532 size_t police_offset;
3536 memset(&tc_police, 0, sizeof tc_police);
3537 tc_police.action = TC_POLICE_SHOT;
3538 tc_police.mtu = mtu;
3539 tc_fill_rate(&tc_police.rate, kbits_rate/8 * 1000, mtu);
3540 tc_police.burst = tc_bytes_to_ticks(tc_police.rate.rate,
3541 kbits_burst * 1024);
3543 tcmsg = tc_make_request(netdev, RTM_NEWTFILTER,
3544 NLM_F_EXCL | NLM_F_CREATE, &request);
3548 tcmsg->tcm_parent = tc_make_handle(0xffff, 0);
3549 tcmsg->tcm_info = tc_make_handle(49,
3550 (OVS_FORCE uint16_t) htons(ETH_P_ALL));
3552 nl_msg_put_string(&request, TCA_KIND, "basic");
3553 basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
3554 police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE);
3555 nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police);
3556 tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate);
3557 nl_msg_end_nested(&request, police_offset);
3558 nl_msg_end_nested(&request, basic_offset);
3560 error = tc_transact(&request, NULL);
3571 /* The values in psched are not individually very meaningful, but they are
3572 * important. The tables below show some values seen in the wild.
3576 * - "c" has always been a constant 1000000 since at least Linux 2.4.14.
3577 * (Before that, there are hints that it was 1000000000.)
3579 * - "d" can be unrealistically large, see the comment on 'buffer_hz'
3583 * -----------------------------------
3584 * [1] 000c8000 000f4240 000f4240 00000064
3585 * [2] 000003e8 00000400 000f4240 3b9aca00
3586 * [3] 000003e8 00000400 000f4240 3b9aca00
3587 * [4] 000003e8 00000400 000f4240 00000064
3588 * [5] 000003e8 00000040 000f4240 3b9aca00
3589 * [6] 000003e8 00000040 000f4240 000000f9
3591 * a b c d ticks_per_s buffer_hz
3592 * ------- --------- ---------- ------------- ----------- -------------
3593 * [1] 819,200 1,000,000 1,000,000 100 819,200 100
3594 * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
3595 * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
3596 * [4] 1,000 1,024 1,000,000 100 976,562 100
3597 * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000
3598 * [6] 1,000 64 1,000,000 249 15,625,000 249
3600 * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p
3601 * [2] 2.6.26-1-686-bigmem from Debian lenny
3602 * [3] 2.6.26-2-sparc64 from Debian lenny
3603 * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p
3604 * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion
3605 * [6] 2.6.34 from kernel.org on KVM
3607 static const char fn[] = "/proc/net/psched";
3608 unsigned int a, b, c, d;
3614 stream = fopen(fn, "r");
3616 VLOG_WARN("%s: open failed: %s", fn, strerror(errno));
3620 if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) {
3621 VLOG_WARN("%s: read failed", fn);
3625 VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d);
3629 VLOG_WARN("%s: invalid scheduler parameters", fn);
3633 ticks_per_s = (double) a * c / b;
3637 VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u",
3640 VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz);
3643 /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a
3644 * rate of 'rate' bytes per second. */
3646 tc_ticks_to_bytes(unsigned int rate, unsigned int ticks)
3651 return (rate * ticks) / ticks_per_s;
3654 /* Returns the number of ticks that it would take to transmit 'size' bytes at a
3655 * rate of 'rate' bytes per second. */
3657 tc_bytes_to_ticks(unsigned int rate, unsigned int size)
3662 return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0;
3665 /* Returns the number of bytes that need to be reserved for qdisc buffering at
3666 * a transmission rate of 'rate' bytes per second. */
3668 tc_buffer_per_jiffy(unsigned int rate)
3673 return rate / buffer_hz;
3676 /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc,
3677 * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull,
3678 * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or
3679 * stores NULL into it if it is absent.
3681 * '*kind' and '*options' point into 'msg', so they are owned by whoever owns
3684 * Returns 0 if successful, otherwise a positive errno value. */
3686 tc_parse_qdisc(const struct ofpbuf *msg, const char **kind,
3687 struct nlattr **options)
3689 static const struct nl_policy tca_policy[] = {
3690 [TCA_KIND] = { .type = NL_A_STRING, .optional = false },
3691 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
3693 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
3695 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
3696 tca_policy, ta, ARRAY_SIZE(ta))) {
3697 VLOG_WARN_RL(&rl, "failed to parse qdisc message");
3702 *kind = nl_attr_get_string(ta[TCA_KIND]);
3706 *options = ta[TCA_OPTIONS];
3721 /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the
3722 * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute
3723 * into '*options', and its queue statistics into '*stats'. Any of the output
3724 * arguments may be null.
3726 * Returns 0 if successful, otherwise a positive errno value. */
3728 tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep,
3729 struct nlattr **options, struct netdev_queue_stats *stats)
3731 static const struct nl_policy tca_policy[] = {
3732 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false },
3733 [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false },
3735 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
3737 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
3738 tca_policy, ta, ARRAY_SIZE(ta))) {
3739 VLOG_WARN_RL(&rl, "failed to parse class message");
3744 struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc);
3745 *handlep = tc->tcm_handle;
3749 *options = ta[TCA_OPTIONS];
3753 const struct gnet_stats_queue *gsq;
3754 struct gnet_stats_basic gsb;
3756 static const struct nl_policy stats_policy[] = {
3757 [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false,
3758 .min_len = sizeof gsb },
3759 [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false,
3760 .min_len = sizeof *gsq },
3762 struct nlattr *sa[ARRAY_SIZE(stats_policy)];
3764 if (!nl_parse_nested(ta[TCA_STATS2], stats_policy,
3765 sa, ARRAY_SIZE(sa))) {
3766 VLOG_WARN_RL(&rl, "failed to parse class stats");
3770 /* Alignment issues screw up the length of struct gnet_stats_basic on
3771 * some arch/bitsize combinations. Newer versions of Linux have a
3772 * struct gnet_stats_basic_packed, but we can't depend on that. The
3773 * easiest thing to do is just to make a copy. */
3774 memset(&gsb, 0, sizeof gsb);
3775 memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]),
3776 MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb));
3777 stats->tx_bytes = gsb.bytes;
3778 stats->tx_packets = gsb.packets;
3780 gsq = nl_attr_get(sa[TCA_STATS_QUEUE]);
3781 stats->tx_errors = gsq->drops;
3791 memset(stats, 0, sizeof *stats);
3796 /* Queries the kernel for class with identifier 'handle' and parent 'parent'
3799 tc_query_class(const struct netdev *netdev,
3800 unsigned int handle, unsigned int parent,
3801 struct ofpbuf **replyp)
3803 struct ofpbuf request;
3804 struct tcmsg *tcmsg;
3807 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request);
3811 tcmsg->tcm_handle = handle;
3812 tcmsg->tcm_parent = parent;
3814 error = tc_transact(&request, replyp);
3816 VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)",
3817 netdev_get_name(netdev),
3818 tc_get_major(handle), tc_get_minor(handle),
3819 tc_get_major(parent), tc_get_minor(parent),
3825 /* Equivalent to "tc class del dev <name> handle <handle>". */
3827 tc_delete_class(const struct netdev *netdev, unsigned int handle)
3829 struct ofpbuf request;
3830 struct tcmsg *tcmsg;
3833 tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request);
3837 tcmsg->tcm_handle = handle;
3838 tcmsg->tcm_parent = 0;
3840 error = tc_transact(&request, NULL);
3842 VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)",
3843 netdev_get_name(netdev),
3844 tc_get_major(handle), tc_get_minor(handle),
3850 /* Equivalent to "tc qdisc del dev <name> root". */
3852 tc_del_qdisc(struct netdev *netdev)
3854 struct netdev_dev_linux *netdev_dev =
3855 netdev_dev_linux_cast(netdev_get_dev(netdev));
3856 struct ofpbuf request;
3857 struct tcmsg *tcmsg;
3860 tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request);
3864 tcmsg->tcm_handle = tc_make_handle(1, 0);
3865 tcmsg->tcm_parent = TC_H_ROOT;
3867 error = tc_transact(&request, NULL);
3868 if (error == EINVAL) {
3869 /* EINVAL probably means that the default qdisc was in use, in which
3870 * case we've accomplished our purpose. */
3873 if (!error && netdev_dev->tc) {
3874 if (netdev_dev->tc->ops->tc_destroy) {
3875 netdev_dev->tc->ops->tc_destroy(netdev_dev->tc);
3877 netdev_dev->tc = NULL;
3882 /* If 'netdev''s qdisc type and parameters are not yet known, queries the
3883 * kernel to determine what they are. Returns 0 if successful, otherwise a
3884 * positive errno value. */
3886 tc_query_qdisc(const struct netdev *netdev)
3888 struct netdev_dev_linux *netdev_dev =
3889 netdev_dev_linux_cast(netdev_get_dev(netdev));
3890 struct ofpbuf request, *qdisc;
3891 const struct tc_ops *ops;
3892 struct tcmsg *tcmsg;
3896 if (netdev_dev->tc) {
3900 /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have
3901 * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before
3902 * 2.6.35 without that fix backported to it.
3904 * To avoid the OOPS, we must not make a request that would attempt to dump
3905 * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a
3906 * few others. There are a few ways that I can see to do this, but most of
3907 * them seem to be racy (and if you lose the race the kernel OOPSes). The
3908 * technique chosen here is to assume that any non-default qdisc that we
3909 * create will have a class with handle 1:0. The built-in qdiscs only have
3910 * a class with handle 0:0.
3912 * We could check for Linux 2.6.35+ and use a more straightforward method
3914 tcmsg = tc_make_request(netdev, RTM_GETQDISC, NLM_F_ECHO, &request);
3918 tcmsg->tcm_handle = tc_make_handle(1, 0);
3919 tcmsg->tcm_parent = 0;
3921 /* Figure out what tc class to instantiate. */
3922 error = tc_transact(&request, &qdisc);
3926 error = tc_parse_qdisc(qdisc, &kind, NULL);
3928 ops = &tc_ops_other;
3930 ops = tc_lookup_linux_name(kind);
3932 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1);
3933 VLOG_INFO_RL(&rl2, "unknown qdisc \"%s\"", kind);
3935 ops = &tc_ops_other;
3938 } else if (error == ENOENT) {
3939 /* Either it's a built-in qdisc, or it's a qdisc set up by some
3940 * other entity that doesn't have a handle 1:0. We will assume
3941 * that it's the system default qdisc. */
3942 ops = &tc_ops_default;
3945 /* Who knows? Maybe the device got deleted. */
3946 VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)",
3947 netdev_get_name(netdev), strerror(error));
3948 ops = &tc_ops_other;
3951 /* Instantiate it. */
3952 load_error = ops->tc_load((struct netdev *) netdev, qdisc);
3953 assert((load_error == 0) == (netdev_dev->tc != NULL));
3954 ofpbuf_delete(qdisc);
3956 return error ? error : load_error;
3959 /* Linux traffic control uses tables with 256 entries ("rtab" tables) to
3960 approximate the time to transmit packets of various lengths. For an MTU of
3961 256 or less, each entry is exact; for an MTU of 257 through 512, each entry
3962 represents two possible packet lengths; for a MTU of 513 through 1024, four
3963 possible lengths; and so on.
3965 Returns, for the specified 'mtu', the number of bits that packet lengths
3966 need to be shifted right to fit within such a 256-entry table. */
3968 tc_calc_cell_log(unsigned int mtu)
3973 mtu = ETH_PAYLOAD_MAX;
3975 mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN;
3977 for (cell_log = 0; mtu >= 256; cell_log++) {
3984 /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU
3987 tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu)
3989 memset(rate, 0, sizeof *rate);
3990 rate->cell_log = tc_calc_cell_log(mtu);
3991 /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */
3992 /* rate->cell_align = 0; */ /* distro headers. */
3993 rate->mpu = ETH_TOTAL_MIN;
3997 /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink
3998 * attribute of the specified "type".
4000 * See tc_calc_cell_log() above for a description of "rtab"s. */
4002 tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate)
4007 rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE);
4008 for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) {
4009 unsigned packet_size = (i + 1) << rate->cell_log;
4010 if (packet_size < rate->mpu) {
4011 packet_size = rate->mpu;
4013 rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size);
4017 /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a
4018 * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested
4019 * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of
4022 tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes)
4024 unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu;
4025 return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst));
4028 /* Linux-only functions declared in netdev-linux.h */
4030 /* Returns a fd for an AF_INET socket or a negative errno value. */
4032 netdev_linux_get_af_inet_sock(void)
4034 int error = netdev_linux_init();
4035 return error ? -error : af_inet_sock;
4038 /* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If
4039 * 'enable' is true, the bit is set. Otherwise, it is cleared. */
4041 netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag,
4042 const char *flag_name, bool enable)
4044 const char *netdev_name = netdev_get_name(netdev);
4045 struct ethtool_value evalue;
4049 memset(&evalue, 0, sizeof evalue);
4050 error = netdev_linux_do_ethtool(netdev_name,
4051 (struct ethtool_cmd *)&evalue,
4052 ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4057 evalue.data = new_flags = (evalue.data & ~flag) | (enable ? flag : 0);
4058 error = netdev_linux_do_ethtool(netdev_name,
4059 (struct ethtool_cmd *)&evalue,
4060 ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS");
4065 memset(&evalue, 0, sizeof evalue);
4066 error = netdev_linux_do_ethtool(netdev_name,
4067 (struct ethtool_cmd *)&evalue,
4068 ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4073 if (new_flags != evalue.data) {
4074 VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network "
4075 "device %s failed", enable ? "enable" : "disable",
4076 flag_name, netdev_name);
4083 /* Utility functions. */
4085 /* Copies 'src' into 'dst', performing format conversion in the process. */
4087 netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst,
4088 const struct rtnl_link_stats *src)
4090 dst->rx_packets = src->rx_packets;
4091 dst->tx_packets = src->tx_packets;
4092 dst->rx_bytes = src->rx_bytes;
4093 dst->tx_bytes = src->tx_bytes;
4094 dst->rx_errors = src->rx_errors;
4095 dst->tx_errors = src->tx_errors;
4096 dst->rx_dropped = src->rx_dropped;
4097 dst->tx_dropped = src->tx_dropped;
4098 dst->multicast = src->multicast;
4099 dst->collisions = src->collisions;
4100 dst->rx_length_errors = src->rx_length_errors;
4101 dst->rx_over_errors = src->rx_over_errors;
4102 dst->rx_crc_errors = src->rx_crc_errors;
4103 dst->rx_frame_errors = src->rx_frame_errors;
4104 dst->rx_fifo_errors = src->rx_fifo_errors;
4105 dst->rx_missed_errors = src->rx_missed_errors;
4106 dst->tx_aborted_errors = src->tx_aborted_errors;
4107 dst->tx_carrier_errors = src->tx_carrier_errors;
4108 dst->tx_fifo_errors = src->tx_fifo_errors;
4109 dst->tx_heartbeat_errors = src->tx_heartbeat_errors;
4110 dst->tx_window_errors = src->tx_window_errors;
4114 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
4116 /* Policy for RTNLGRP_LINK messages.
4118 * There are *many* more fields in these messages, but currently we only
4119 * care about these fields. */
4120 static const struct nl_policy rtnlgrp_link_policy[] = {
4121 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
4122 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
4123 .min_len = sizeof(struct rtnl_link_stats) },
4126 struct ofpbuf request;
4127 struct ofpbuf *reply;
4128 struct ifinfomsg *ifi;
4129 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
4132 ofpbuf_init(&request, 0);
4133 nl_msg_put_nlmsghdr(&request, sizeof *ifi, RTM_GETLINK, NLM_F_REQUEST);
4134 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
4135 ifi->ifi_family = PF_UNSPEC;
4136 ifi->ifi_index = ifindex;
4137 error = nl_sock_transact(rtnl_sock, &request, &reply);
4138 ofpbuf_uninit(&request);
4143 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
4144 rtnlgrp_link_policy,
4145 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
4146 ofpbuf_delete(reply);
4150 if (!attrs[IFLA_STATS]) {
4151 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
4152 ofpbuf_delete(reply);
4156 netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(attrs[IFLA_STATS]));
4158 ofpbuf_delete(reply);
4164 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
4166 static const char fn[] = "/proc/net/dev";
4171 stream = fopen(fn, "r");
4173 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
4178 while (fgets(line, sizeof line, stream)) {
4181 #define X64 "%"SCNu64
4184 X64 X64 X64 X64 X64 X64 X64 "%*u"
4185 X64 X64 X64 X64 X64 X64 X64 "%*u",
4191 &stats->rx_fifo_errors,
4192 &stats->rx_frame_errors,
4198 &stats->tx_fifo_errors,
4200 &stats->tx_carrier_errors) != 15) {
4201 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
4202 } else if (!strcmp(devname, netdev_name)) {
4203 stats->rx_length_errors = UINT64_MAX;
4204 stats->rx_over_errors = UINT64_MAX;
4205 stats->rx_crc_errors = UINT64_MAX;
4206 stats->rx_missed_errors = UINT64_MAX;
4207 stats->tx_aborted_errors = UINT64_MAX;
4208 stats->tx_heartbeat_errors = UINT64_MAX;
4209 stats->tx_window_errors = UINT64_MAX;
4215 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
4221 get_flags(const struct netdev_dev *dev, unsigned int *flags)
4227 error = netdev_linux_do_ioctl(dev->name, &ifr, SIOCGIFFLAGS,
4230 *flags = ifr.ifr_flags;
4236 set_flags(struct netdev *netdev, unsigned int flags)
4240 ifr.ifr_flags = flags;
4241 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
4246 do_get_ifindex(const char *netdev_name)
4250 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4251 COVERAGE_INC(netdev_get_ifindex);
4252 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
4253 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
4254 netdev_name, strerror(errno));
4257 return ifr.ifr_ifindex;
4261 get_ifindex(const struct netdev *netdev_, int *ifindexp)
4263 struct netdev_dev_linux *netdev_dev =
4264 netdev_dev_linux_cast(netdev_get_dev(netdev_));
4266 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
4267 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
4271 netdev_dev->cache_valid |= VALID_IFINDEX;
4272 netdev_dev->ifindex = ifindex;
4274 *ifindexp = netdev_dev->ifindex;
4279 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
4284 memset(&ifr, 0, sizeof ifr);
4285 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4286 COVERAGE_INC(netdev_get_hwaddr);
4287 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
4288 /* ENODEV probably means that a vif disappeared asynchronously and
4289 * hasn't been removed from the database yet, so reduce the log level
4290 * to INFO for that case. */
4291 VLOG(errno == ENODEV ? VLL_INFO : VLL_ERR,
4292 "ioctl(SIOCGIFHWADDR) on %s device failed: %s",
4293 netdev_name, strerror(errno));
4296 hwaddr_family = ifr.ifr_hwaddr.sa_family;
4297 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
4298 VLOG_WARN("%s device has unknown hardware address family %d",
4299 netdev_name, hwaddr_family);
4301 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
4306 set_etheraddr(const char *netdev_name, int hwaddr_family,
4307 const uint8_t mac[ETH_ADDR_LEN])
4311 memset(&ifr, 0, sizeof ifr);
4312 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4313 ifr.ifr_hwaddr.sa_family = hwaddr_family;
4314 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
4315 COVERAGE_INC(netdev_set_hwaddr);
4316 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
4317 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
4318 netdev_name, strerror(errno));
4325 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
4326 int cmd, const char *cmd_name)
4330 memset(&ifr, 0, sizeof ifr);
4331 ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
4332 ifr.ifr_data = (caddr_t) ecmd;
4335 COVERAGE_INC(netdev_ethtool);
4336 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
4339 if (errno != EOPNOTSUPP) {
4340 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
4341 "failed: %s", cmd_name, name, strerror(errno));
4343 /* The device doesn't support this operation. That's pretty
4344 * common, so there's no point in logging anything. */
4351 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
4352 const char *cmd_name)
4354 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
4355 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
4356 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
4364 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
4365 int cmd, const char *cmd_name)
4370 ifr.ifr_addr.sa_family = AF_INET;
4371 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
4373 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
4374 *ip = sin->sin_addr;
4379 /* Returns an AF_PACKET raw socket or a negative errno value. */
4381 af_packet_sock(void)
4383 static int sock = INT_MIN;
4385 if (sock == INT_MIN) {
4386 sock = socket(AF_PACKET, SOCK_RAW, 0);
4388 set_nonblocking(sock);
4391 VLOG_ERR("failed to create packet socket: %s", strerror(errno));