2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "netdev-linux.h"
24 #include <arpa/inet.h>
26 #include <linux/gen_stats.h>
27 #include <linux/if_ether.h>
28 #include <linux/if_tun.h>
30 #include <linux/types.h>
31 #include <linux/ethtool.h>
32 #include <linux/mii.h>
33 #include <linux/pkt_cls.h>
34 #include <linux/pkt_sched.h>
35 #include <linux/rtnetlink.h>
36 #include <linux/sockios.h>
37 #include <linux/version.h>
38 #include <sys/types.h>
39 #include <sys/ioctl.h>
40 #include <sys/socket.h>
41 #include <netpacket/packet.h>
43 #include <net/if_arp.h>
44 #include <net/if_packet.h>
45 #include <net/route.h>
46 #include <netinet/in.h>
53 #include "dpif-linux.h"
54 #include "dynamic-string.h"
55 #include "fatal-signal.h"
58 #include "netdev-provider.h"
59 #include "netdev-vport.h"
61 #include "netlink-notifier.h"
62 #include "netlink-socket.h"
64 #include "openflow/openflow.h"
66 #include "poll-loop.h"
67 #include "rtnetlink-link.h"
68 #include "socket-util.h"
74 VLOG_DEFINE_THIS_MODULE(netdev_linux);
76 COVERAGE_DEFINE(netdev_set_policing);
77 COVERAGE_DEFINE(netdev_arp_lookup);
78 COVERAGE_DEFINE(netdev_get_ifindex);
79 COVERAGE_DEFINE(netdev_get_hwaddr);
80 COVERAGE_DEFINE(netdev_set_hwaddr);
81 COVERAGE_DEFINE(netdev_ethtool);
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_VPORT_STAT_ERROR = 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 int vport_stats_error; /* Cached error code from vport_get_stats().
376 0 or an errno value. */
380 struct tap_state tap;
384 struct netdev_linux {
385 struct netdev netdev;
389 /* Sockets used for ioctl operations. */
390 static int af_inet_sock = -1; /* AF_INET, SOCK_DGRAM. */
392 /* A Netlink routing socket that is not subscribed to any multicast groups. */
393 static struct nl_sock *rtnl_sock;
395 /* This is set pretty low because we probably won't learn anything from the
396 * additional log messages. */
397 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
399 static int netdev_linux_init(void);
401 static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
402 int cmd, const char *cmd_name);
403 static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
404 const char *cmd_name);
405 static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
406 int cmd, const char *cmd_name);
407 static int get_flags(const struct netdev_dev *, unsigned int *flags);
408 static int set_flags(struct netdev *, unsigned int flags);
409 static int do_get_ifindex(const char *netdev_name);
410 static int get_ifindex(const struct netdev *, int *ifindexp);
411 static int do_set_addr(struct netdev *netdev,
412 int ioctl_nr, const char *ioctl_name,
413 struct in_addr addr);
414 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
415 static int set_etheraddr(const char *netdev_name, int hwaddr_family,
416 const uint8_t[ETH_ADDR_LEN]);
417 static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
418 static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
419 static int af_packet_sock(void);
420 static void netdev_linux_miimon_run(void);
421 static void netdev_linux_miimon_wait(void);
424 is_netdev_linux_class(const struct netdev_class *netdev_class)
426 return netdev_class->init == netdev_linux_init;
429 static struct netdev_dev_linux *
430 netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
432 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
433 assert(is_netdev_linux_class(netdev_class));
435 return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
438 static struct netdev_linux *
439 netdev_linux_cast(const struct netdev *netdev)
441 struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
442 const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
443 assert(is_netdev_linux_class(netdev_class));
445 return CONTAINER_OF(netdev, struct netdev_linux, netdev);
449 netdev_linux_init(void)
451 static int status = -1;
453 /* Create AF_INET socket. */
454 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
455 status = af_inet_sock >= 0 ? 0 : errno;
457 VLOG_ERR("failed to create inet socket: %s", strerror(status));
460 /* Create rtnetlink socket. */
462 status = nl_sock_create(NETLINK_ROUTE, &rtnl_sock);
464 VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",
473 netdev_linux_run(void)
475 rtnetlink_link_run();
476 netdev_linux_miimon_run();
480 netdev_linux_wait(void)
482 rtnetlink_link_wait();
483 netdev_linux_miimon_wait();
487 netdev_dev_linux_changed(struct netdev_dev_linux *dev, unsigned int ifi_flags)
490 if (!dev->change_seq) {
494 if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) {
495 dev->carrier_resets++;
497 dev->ifi_flags = ifi_flags;
499 dev->cache_valid = 0;
503 netdev_linux_cache_cb(const struct rtnetlink_link_change *change,
504 void *aux OVS_UNUSED)
506 struct netdev_dev_linux *dev;
508 struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
510 const struct netdev_class *netdev_class =
511 netdev_dev_get_class(base_dev);
513 if (is_netdev_linux_class(netdev_class)) {
514 dev = netdev_dev_linux_cast(base_dev);
515 netdev_dev_linux_changed(dev, change->ifi_flags);
519 struct shash device_shash;
520 struct shash_node *node;
522 shash_init(&device_shash);
523 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
524 SHASH_FOR_EACH (node, &device_shash) {
529 get_flags(&dev->netdev_dev, &flags);
530 netdev_dev_linux_changed(dev, flags);
532 shash_destroy(&device_shash);
537 cache_notifier_ref(void)
539 if (!cache_notifier_refcount) {
540 assert(!netdev_linux_cache_notifier);
542 netdev_linux_cache_notifier =
543 rtnetlink_link_notifier_create(netdev_linux_cache_cb, NULL);
545 if (!netdev_linux_cache_notifier) {
549 cache_notifier_refcount++;
555 cache_notifier_unref(void)
557 assert(cache_notifier_refcount > 0);
558 if (!--cache_notifier_refcount) {
559 assert(netdev_linux_cache_notifier);
560 rtnetlink_link_notifier_destroy(netdev_linux_cache_notifier);
561 netdev_linux_cache_notifier = NULL;
565 /* Creates system and internal devices. */
567 netdev_linux_create(const struct netdev_class *class, const char *name,
568 struct netdev_dev **netdev_devp)
570 struct netdev_dev_linux *netdev_dev;
573 error = cache_notifier_ref();
578 netdev_dev = xzalloc(sizeof *netdev_dev);
579 netdev_dev->change_seq = 1;
580 netdev_dev_init(&netdev_dev->netdev_dev, name, class);
581 get_flags(&netdev_dev->netdev_dev, &netdev_dev->ifi_flags);
583 *netdev_devp = &netdev_dev->netdev_dev;
587 /* For most types of netdevs we open the device for each call of
588 * netdev_open(). However, this is not the case with tap devices,
589 * since it is only possible to open the device once. In this
590 * situation we share a single file descriptor, and consequently
591 * buffers, across all readers. Therefore once data is read it will
592 * be unavailable to other reads for tap devices. */
594 netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED,
595 const char *name, struct netdev_dev **netdev_devp)
597 struct netdev_dev_linux *netdev_dev;
598 struct tap_state *state;
599 static const char tap_dev[] = "/dev/net/tun";
603 netdev_dev = xzalloc(sizeof *netdev_dev);
604 state = &netdev_dev->state.tap;
606 error = cache_notifier_ref();
611 /* Open tap device. */
612 state->fd = open(tap_dev, O_RDWR);
615 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
616 goto error_unref_notifier;
619 /* Create tap device. */
620 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
621 ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
622 if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
623 VLOG_WARN("%s: creating tap device failed: %s", name,
626 goto error_unref_notifier;
629 /* Make non-blocking. */
630 error = set_nonblocking(state->fd);
632 goto error_unref_notifier;
635 netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
636 *netdev_devp = &netdev_dev->netdev_dev;
639 error_unref_notifier:
640 cache_notifier_unref();
647 destroy_tap(struct netdev_dev_linux *netdev_dev)
649 struct tap_state *state = &netdev_dev->state.tap;
651 if (state->fd >= 0) {
656 /* Destroys the netdev device 'netdev_dev_'. */
658 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
660 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
661 const struct netdev_class *class = netdev_dev_get_class(netdev_dev_);
663 if (netdev_dev->tc && netdev_dev->tc->ops->tc_destroy) {
664 netdev_dev->tc->ops->tc_destroy(netdev_dev->tc);
667 if (class == &netdev_tap_class) {
668 destroy_tap(netdev_dev);
672 cache_notifier_unref();
676 netdev_linux_open(struct netdev_dev *netdev_dev_, struct netdev **netdevp)
678 struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
679 struct netdev_linux *netdev;
680 enum netdev_flags flags;
683 /* Allocate network device. */
684 netdev = xzalloc(sizeof *netdev);
686 netdev_init(&netdev->netdev, netdev_dev_);
688 /* Verify that the device really exists, by attempting to read its flags.
689 * (The flags might be cached, in which case this won't actually do an
692 * Don't do this for "internal" netdevs, though, because those have to be
693 * created as netdev objects before they exist in the kernel, because
694 * creating them in the kernel happens by passing a netdev object to
695 * dpif_port_add(). */
696 if (netdev_dev_get_class(netdev_dev_) != &netdev_internal_class) {
697 error = netdev_get_flags(&netdev->netdev, &flags);
698 if (error == ENODEV) {
703 if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap") &&
704 !netdev_dev->state.tap.opened) {
706 /* We assume that the first user of the tap device is the primary user
707 * and give them the tap FD. Subsequent users probably just expect
708 * this to be a system device so open it normally to avoid send/receive
709 * directions appearing to be reversed. */
710 netdev->fd = netdev_dev->state.tap.fd;
711 netdev_dev->state.tap.opened = true;
714 *netdevp = &netdev->netdev;
718 netdev_uninit(&netdev->netdev, true);
722 /* Closes and destroys 'netdev'. */
724 netdev_linux_close(struct netdev *netdev_)
726 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
728 if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
735 netdev_linux_listen(struct netdev *netdev_)
737 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
738 struct sockaddr_ll sll;
743 if (netdev->fd >= 0) {
747 /* Create file descriptor. */
748 fd = socket(PF_PACKET, SOCK_RAW, 0);
751 VLOG_ERR("failed to create raw socket (%s)", strerror(error));
755 /* Set non-blocking mode. */
756 error = set_nonblocking(fd);
761 /* Get ethernet device index. */
762 error = get_ifindex(&netdev->netdev, &ifindex);
767 /* Bind to specific ethernet device. */
768 memset(&sll, 0, sizeof sll);
769 sll.sll_family = AF_PACKET;
770 sll.sll_ifindex = ifindex;
771 sll.sll_protocol = (OVS_FORCE unsigned short int) htons(ETH_P_ALL);
772 if (bind(fd, (struct sockaddr *) &sll, sizeof sll) < 0) {
774 VLOG_ERR("%s: failed to bind raw socket (%s)",
775 netdev_get_name(netdev_), strerror(error));
790 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
792 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
794 if (netdev->fd < 0) {
795 /* Device is not listening. */
800 ssize_t retval = recv(netdev->fd, data, size, MSG_TRUNC);
802 return retval <= size ? retval : -EMSGSIZE;
803 } else if (errno != EINTR) {
804 if (errno != EAGAIN) {
805 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
806 strerror(errno), netdev_get_name(netdev_));
813 /* Registers with the poll loop to wake up from the next call to poll_block()
814 * when a packet is ready to be received with netdev_recv() on 'netdev'. */
816 netdev_linux_recv_wait(struct netdev *netdev_)
818 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
819 if (netdev->fd >= 0) {
820 poll_fd_wait(netdev->fd, POLLIN);
824 /* Discards all packets waiting to be received from 'netdev'. */
826 netdev_linux_drain(struct netdev *netdev_)
828 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
829 if (netdev->fd < 0) {
831 } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
833 int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
834 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
838 drain_fd(netdev->fd, ifr.ifr_qlen);
841 return drain_rcvbuf(netdev->fd);
845 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
846 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
847 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
848 * the packet is too big or too small to transmit on the device.
850 * The caller retains ownership of 'buffer' in all cases.
852 * The kernel maintains a packet transmission queue, so the caller is not
853 * expected to do additional queuing of packets. */
855 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
857 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
861 if (netdev->fd < 0) {
862 /* Use our AF_PACKET socket to send to this device. */
863 struct sockaddr_ll sll;
870 sock = af_packet_sock();
875 error = get_ifindex(netdev_, &ifindex);
880 /* We don't bother setting most fields in sockaddr_ll because the
881 * kernel ignores them for SOCK_RAW. */
882 memset(&sll, 0, sizeof sll);
883 sll.sll_family = AF_PACKET;
884 sll.sll_ifindex = ifindex;
886 iov.iov_base = (void *) data;
890 msg.msg_namelen = sizeof sll;
893 msg.msg_control = NULL;
894 msg.msg_controllen = 0;
897 retval = sendmsg(sock, &msg, 0);
899 /* Use the netdev's own fd to send to this device. This is
900 * essential for tap devices, because packets sent to a tap device
901 * with an AF_PACKET socket will loop back to be *received* again
902 * on the tap device. */
903 retval = write(netdev->fd, data, size);
907 /* The Linux AF_PACKET implementation never blocks waiting for room
908 * for packets, instead returning ENOBUFS. Translate this into
909 * EAGAIN for the caller. */
910 if (errno == ENOBUFS) {
912 } else if (errno == EINTR) {
914 } else if (errno != EAGAIN) {
915 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
916 netdev_get_name(netdev_), strerror(errno));
919 } else if (retval != size) {
920 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
921 "%zu) on %s", retval, size, netdev_get_name(netdev_));
929 /* Registers with the poll loop to wake up from the next call to poll_block()
930 * when the packet transmission queue has sufficient room to transmit a packet
931 * with netdev_send().
933 * The kernel maintains a packet transmission queue, so the client is not
934 * expected to do additional queuing of packets. Thus, this function is
935 * unlikely to ever be used. It is included for completeness. */
937 netdev_linux_send_wait(struct netdev *netdev_)
939 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
940 if (netdev->fd < 0) {
942 } else if (strcmp(netdev_get_type(netdev_), "tap")) {
943 poll_fd_wait(netdev->fd, POLLOUT);
945 /* TAP device always accepts packets.*/
946 poll_immediate_wake();
950 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
951 * otherwise a positive errno value. */
953 netdev_linux_set_etheraddr(struct netdev *netdev_,
954 const uint8_t mac[ETH_ADDR_LEN])
956 struct netdev_dev_linux *netdev_dev =
957 netdev_dev_linux_cast(netdev_get_dev(netdev_));
960 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)
961 || !eth_addr_equals(netdev_dev->etheraddr, mac)) {
962 error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
964 netdev_dev->cache_valid |= VALID_ETHERADDR;
965 memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
973 /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or
974 * free the returned buffer. */
976 netdev_linux_get_etheraddr(const struct netdev *netdev_,
977 uint8_t mac[ETH_ADDR_LEN])
979 struct netdev_dev_linux *netdev_dev =
980 netdev_dev_linux_cast(netdev_get_dev(netdev_));
981 if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
982 int error = get_etheraddr(netdev_get_name(netdev_),
983 netdev_dev->etheraddr);
987 netdev_dev->cache_valid |= VALID_ETHERADDR;
989 memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
993 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
994 * in bytes, not including the hardware header; thus, this is typically 1500
995 * bytes for Ethernet devices. */
997 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
999 struct netdev_dev_linux *netdev_dev =
1000 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1001 if (!(netdev_dev->cache_valid & VALID_MTU)) {
1005 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1006 SIOCGIFMTU, "SIOCGIFMTU");
1010 netdev_dev->mtu = ifr.ifr_mtu;
1011 netdev_dev->cache_valid |= VALID_MTU;
1013 *mtup = netdev_dev->mtu;
1017 /* Sets the maximum size of transmitted (MTU) for given device using linux
1018 * networking ioctl interface.
1021 netdev_linux_set_mtu(const struct netdev *netdev_, int mtu)
1023 struct netdev_dev_linux *netdev_dev =
1024 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1028 if (netdev_dev->cache_valid & VALID_MTU &&
1029 netdev_dev->mtu == mtu) {
1033 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1034 SIOCSIFMTU, "SIOCSIFMTU");
1039 netdev_dev->mtu = ifr.ifr_mtu;
1040 netdev_dev->cache_valid |= VALID_MTU;
1044 /* Returns the ifindex of 'netdev', if successful, as a positive number.
1045 * On failure, returns a negative errno value. */
1047 netdev_linux_get_ifindex(const struct netdev *netdev)
1051 error = get_ifindex(netdev, &ifindex);
1052 return error ? -error : ifindex;
1056 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
1058 struct netdev_dev_linux *netdev_dev =
1059 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1061 if (netdev_dev->miimon_interval > 0) {
1062 *carrier = netdev_dev->miimon;
1064 *carrier = (netdev_dev->ifi_flags & IFF_RUNNING) != 0;
1070 static long long int
1071 netdev_linux_get_carrier_resets(const struct netdev *netdev)
1073 return netdev_dev_linux_cast(netdev_get_dev(netdev))->carrier_resets;
1077 netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name,
1078 struct mii_ioctl_data *data)
1083 memset(&ifr, 0, sizeof ifr);
1084 memcpy(&ifr.ifr_data, data, sizeof *data);
1085 error = netdev_linux_do_ioctl(name, &ifr, cmd, cmd_name);
1086 memcpy(data, &ifr.ifr_data, sizeof *data);
1092 netdev_linux_get_miimon(const char *name, bool *miimon)
1094 struct mii_ioctl_data data;
1099 memset(&data, 0, sizeof data);
1100 error = netdev_linux_do_miimon(name, SIOCGMIIPHY, "SIOCGMIIPHY", &data);
1102 /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */
1103 data.reg_num = MII_BMSR;
1104 error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG",
1108 *miimon = !!(data.val_out & BMSR_LSTATUS);
1110 VLOG_WARN_RL(&rl, "%s: failed to query MII", name);
1113 struct ethtool_cmd ecmd;
1115 VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool",
1118 memset(&ecmd, 0, sizeof ecmd);
1119 error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK,
1122 struct ethtool_value eval;
1124 memcpy(&eval, &ecmd, sizeof eval);
1125 *miimon = !!eval.data;
1127 VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name);
1135 netdev_linux_set_miimon_interval(struct netdev *netdev_,
1136 long long int interval)
1138 struct netdev_dev_linux *netdev_dev;
1140 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev_));
1142 interval = interval > 0 ? MAX(interval, 100) : 0;
1143 if (netdev_dev->miimon_interval != interval) {
1144 netdev_dev->miimon_interval = interval;
1145 timer_set_expired(&netdev_dev->miimon_timer);
1152 netdev_linux_miimon_run(void)
1154 struct shash device_shash;
1155 struct shash_node *node;
1157 shash_init(&device_shash);
1158 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
1159 SHASH_FOR_EACH (node, &device_shash) {
1160 struct netdev_dev_linux *dev = node->data;
1163 if (dev->miimon_interval <= 0 || !timer_expired(&dev->miimon_timer)) {
1167 netdev_linux_get_miimon(dev->netdev_dev.name, &miimon);
1168 if (miimon != dev->miimon) {
1169 dev->miimon = miimon;
1170 netdev_dev_linux_changed(dev, dev->ifi_flags);
1173 timer_set_duration(&dev->miimon_timer, dev->miimon_interval);
1176 shash_destroy(&device_shash);
1180 netdev_linux_miimon_wait(void)
1182 struct shash device_shash;
1183 struct shash_node *node;
1185 shash_init(&device_shash);
1186 netdev_dev_get_devices(&netdev_linux_class, &device_shash);
1187 SHASH_FOR_EACH (node, &device_shash) {
1188 struct netdev_dev_linux *dev = node->data;
1190 if (dev->miimon_interval > 0) {
1191 timer_wait(&dev->miimon_timer);
1194 shash_destroy(&device_shash);
1197 /* Check whether we can we use RTM_GETLINK to get network device statistics.
1198 * In pre-2.6.19 kernels, this was only available if wireless extensions were
1201 check_for_working_netlink_stats(void)
1203 /* Decide on the netdev_get_stats() implementation to use. Netlink is
1204 * preferable, so if that works, we'll use it. */
1205 int ifindex = do_get_ifindex("lo");
1207 VLOG_WARN("failed to get ifindex for lo, "
1208 "obtaining netdev stats from proc");
1211 struct netdev_stats stats;
1212 int error = get_stats_via_netlink(ifindex, &stats);
1214 VLOG_DBG("obtaining netdev stats via rtnetlink");
1217 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
1218 "via proc (you are probably running a pre-2.6.19 "
1219 "kernel)", strerror(error));
1226 swap_uint64(uint64_t *a, uint64_t *b)
1234 get_stats_via_vport(const struct netdev *netdev_,
1235 struct netdev_stats *stats)
1237 struct netdev_dev_linux *netdev_dev =
1238 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1240 if (!netdev_dev->vport_stats_error ||
1241 !(netdev_dev->cache_valid & VALID_VPORT_STAT_ERROR)) {
1244 error = netdev_vport_get_stats(netdev_, stats);
1246 VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed "
1247 "(%s)", netdev_get_name(netdev_), strerror(error));
1249 netdev_dev->vport_stats_error = error;
1250 netdev_dev->cache_valid |= VALID_VPORT_STAT_ERROR;
1255 netdev_linux_sys_get_stats(const struct netdev *netdev_,
1256 struct netdev_stats *stats)
1258 static int use_netlink_stats = -1;
1261 if (use_netlink_stats < 0) {
1262 use_netlink_stats = check_for_working_netlink_stats();
1265 if (use_netlink_stats) {
1268 error = get_ifindex(netdev_, &ifindex);
1270 error = get_stats_via_netlink(ifindex, stats);
1273 error = get_stats_via_proc(netdev_get_name(netdev_), stats);
1277 VLOG_WARN_RL(&rl, "%s: linux-sys get stats failed %d",
1278 netdev_get_name(netdev_), error);
1284 /* Retrieves current device stats for 'netdev-linux'. */
1286 netdev_linux_get_stats(const struct netdev *netdev_,
1287 struct netdev_stats *stats)
1289 struct netdev_dev_linux *netdev_dev =
1290 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1291 struct netdev_stats dev_stats;
1294 get_stats_via_vport(netdev_, stats);
1296 error = netdev_linux_sys_get_stats(netdev_, &dev_stats);
1299 if (netdev_dev->vport_stats_error) {
1306 if (netdev_dev->vport_stats_error) {
1307 /* stats not available from OVS then use ioctl stats. */
1310 stats->rx_errors += dev_stats.rx_errors;
1311 stats->tx_errors += dev_stats.tx_errors;
1312 stats->rx_dropped += dev_stats.rx_dropped;
1313 stats->tx_dropped += dev_stats.tx_dropped;
1314 stats->multicast += dev_stats.multicast;
1315 stats->collisions += dev_stats.collisions;
1316 stats->rx_length_errors += dev_stats.rx_length_errors;
1317 stats->rx_over_errors += dev_stats.rx_over_errors;
1318 stats->rx_crc_errors += dev_stats.rx_crc_errors;
1319 stats->rx_frame_errors += dev_stats.rx_frame_errors;
1320 stats->rx_fifo_errors += dev_stats.rx_fifo_errors;
1321 stats->rx_missed_errors += dev_stats.rx_missed_errors;
1322 stats->tx_aborted_errors += dev_stats.tx_aborted_errors;
1323 stats->tx_carrier_errors += dev_stats.tx_carrier_errors;
1324 stats->tx_fifo_errors += dev_stats.tx_fifo_errors;
1325 stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors;
1326 stats->tx_window_errors += dev_stats.tx_window_errors;
1331 /* Retrieves current device stats for 'netdev-tap' netdev or
1332 * netdev-internal. */
1334 netdev_tap_get_stats(const struct netdev *netdev_,
1335 struct netdev_stats *stats)
1337 struct netdev_dev_linux *netdev_dev =
1338 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1339 struct netdev_stats dev_stats;
1342 get_stats_via_vport(netdev_, stats);
1344 error = netdev_linux_sys_get_stats(netdev_, &dev_stats);
1346 if (netdev_dev->vport_stats_error) {
1353 /* If this port is an internal port then the transmit and receive stats
1354 * will appear to be swapped relative to the other ports since we are the
1355 * one sending the data, not a remote computer. For consistency, we swap
1356 * them back here. This does not apply if we are getting stats from the
1357 * vport layer because it always tracks stats from the perspective of the
1359 if (netdev_dev->vport_stats_error) {
1361 swap_uint64(&stats->rx_packets, &stats->tx_packets);
1362 swap_uint64(&stats->rx_bytes, &stats->tx_bytes);
1363 swap_uint64(&stats->rx_errors, &stats->tx_errors);
1364 swap_uint64(&stats->rx_dropped, &stats->tx_dropped);
1365 stats->rx_length_errors = 0;
1366 stats->rx_over_errors = 0;
1367 stats->rx_crc_errors = 0;
1368 stats->rx_frame_errors = 0;
1369 stats->rx_fifo_errors = 0;
1370 stats->rx_missed_errors = 0;
1371 stats->tx_aborted_errors = 0;
1372 stats->tx_carrier_errors = 0;
1373 stats->tx_fifo_errors = 0;
1374 stats->tx_heartbeat_errors = 0;
1375 stats->tx_window_errors = 0;
1377 stats->rx_dropped += dev_stats.tx_dropped;
1378 stats->tx_dropped += dev_stats.rx_dropped;
1380 stats->rx_errors += dev_stats.tx_errors;
1381 stats->tx_errors += dev_stats.rx_errors;
1383 stats->multicast += dev_stats.multicast;
1384 stats->collisions += dev_stats.collisions;
1390 netdev_internal_get_stats(const struct netdev *netdev_,
1391 struct netdev_stats *stats)
1393 struct netdev_dev_linux *netdev_dev =
1394 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1396 get_stats_via_vport(netdev_, stats);
1397 return netdev_dev->vport_stats_error;
1400 /* Stores the features supported by 'netdev' into each of '*current',
1401 * '*advertised', '*supported', and '*peer' that are non-null. Each value is a
1402 * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
1403 * successful, otherwise a positive errno value. */
1405 netdev_linux_get_features(const struct netdev *netdev,
1406 uint32_t *current, uint32_t *advertised,
1407 uint32_t *supported, uint32_t *peer)
1409 struct ethtool_cmd ecmd;
1412 memset(&ecmd, 0, sizeof ecmd);
1413 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1414 ETHTOOL_GSET, "ETHTOOL_GSET");
1419 /* Supported features. */
1421 if (ecmd.supported & SUPPORTED_10baseT_Half) {
1422 *supported |= OFPPF_10MB_HD;
1424 if (ecmd.supported & SUPPORTED_10baseT_Full) {
1425 *supported |= OFPPF_10MB_FD;
1427 if (ecmd.supported & SUPPORTED_100baseT_Half) {
1428 *supported |= OFPPF_100MB_HD;
1430 if (ecmd.supported & SUPPORTED_100baseT_Full) {
1431 *supported |= OFPPF_100MB_FD;
1433 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
1434 *supported |= OFPPF_1GB_HD;
1436 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
1437 *supported |= OFPPF_1GB_FD;
1439 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
1440 *supported |= OFPPF_10GB_FD;
1442 if (ecmd.supported & SUPPORTED_TP) {
1443 *supported |= OFPPF_COPPER;
1445 if (ecmd.supported & SUPPORTED_FIBRE) {
1446 *supported |= OFPPF_FIBER;
1448 if (ecmd.supported & SUPPORTED_Autoneg) {
1449 *supported |= OFPPF_AUTONEG;
1451 if (ecmd.supported & SUPPORTED_Pause) {
1452 *supported |= OFPPF_PAUSE;
1454 if (ecmd.supported & SUPPORTED_Asym_Pause) {
1455 *supported |= OFPPF_PAUSE_ASYM;
1458 /* Advertised features. */
1460 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
1461 *advertised |= OFPPF_10MB_HD;
1463 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
1464 *advertised |= OFPPF_10MB_FD;
1466 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
1467 *advertised |= OFPPF_100MB_HD;
1469 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
1470 *advertised |= OFPPF_100MB_FD;
1472 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
1473 *advertised |= OFPPF_1GB_HD;
1475 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
1476 *advertised |= OFPPF_1GB_FD;
1478 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
1479 *advertised |= OFPPF_10GB_FD;
1481 if (ecmd.advertising & ADVERTISED_TP) {
1482 *advertised |= OFPPF_COPPER;
1484 if (ecmd.advertising & ADVERTISED_FIBRE) {
1485 *advertised |= OFPPF_FIBER;
1487 if (ecmd.advertising & ADVERTISED_Autoneg) {
1488 *advertised |= OFPPF_AUTONEG;
1490 if (ecmd.advertising & ADVERTISED_Pause) {
1491 *advertised |= OFPPF_PAUSE;
1493 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
1494 *advertised |= OFPPF_PAUSE_ASYM;
1497 /* Current settings. */
1498 if (ecmd.speed == SPEED_10) {
1499 *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD;
1500 } else if (ecmd.speed == SPEED_100) {
1501 *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD;
1502 } else if (ecmd.speed == SPEED_1000) {
1503 *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD;
1504 } else if (ecmd.speed == SPEED_10000) {
1505 *current = OFPPF_10GB_FD;
1510 if (ecmd.port == PORT_TP) {
1511 *current |= OFPPF_COPPER;
1512 } else if (ecmd.port == PORT_FIBRE) {
1513 *current |= OFPPF_FIBER;
1517 *current |= OFPPF_AUTONEG;
1520 /* Peer advertisements. */
1521 *peer = 0; /* XXX */
1526 /* Set the features advertised by 'netdev' to 'advertise'. */
1528 netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
1530 struct ethtool_cmd ecmd;
1533 memset(&ecmd, 0, sizeof ecmd);
1534 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1535 ETHTOOL_GSET, "ETHTOOL_GSET");
1540 ecmd.advertising = 0;
1541 if (advertise & OFPPF_10MB_HD) {
1542 ecmd.advertising |= ADVERTISED_10baseT_Half;
1544 if (advertise & OFPPF_10MB_FD) {
1545 ecmd.advertising |= ADVERTISED_10baseT_Full;
1547 if (advertise & OFPPF_100MB_HD) {
1548 ecmd.advertising |= ADVERTISED_100baseT_Half;
1550 if (advertise & OFPPF_100MB_FD) {
1551 ecmd.advertising |= ADVERTISED_100baseT_Full;
1553 if (advertise & OFPPF_1GB_HD) {
1554 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1556 if (advertise & OFPPF_1GB_FD) {
1557 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1559 if (advertise & OFPPF_10GB_FD) {
1560 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1562 if (advertise & OFPPF_COPPER) {
1563 ecmd.advertising |= ADVERTISED_TP;
1565 if (advertise & OFPPF_FIBER) {
1566 ecmd.advertising |= ADVERTISED_FIBRE;
1568 if (advertise & OFPPF_AUTONEG) {
1569 ecmd.advertising |= ADVERTISED_Autoneg;
1571 if (advertise & OFPPF_PAUSE) {
1572 ecmd.advertising |= ADVERTISED_Pause;
1574 if (advertise & OFPPF_PAUSE_ASYM) {
1575 ecmd.advertising |= ADVERTISED_Asym_Pause;
1577 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1578 ETHTOOL_SSET, "ETHTOOL_SSET");
1581 /* Attempts to set input rate limiting (policing) policy. Returns 0 if
1582 * successful, otherwise a positive errno value. */
1584 netdev_linux_set_policing(struct netdev *netdev,
1585 uint32_t kbits_rate, uint32_t kbits_burst)
1587 struct netdev_dev_linux *netdev_dev =
1588 netdev_dev_linux_cast(netdev_get_dev(netdev));
1589 const char *netdev_name = netdev_get_name(netdev);
1593 kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */
1594 : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */
1595 : kbits_burst); /* Stick with user-specified value. */
1597 if (netdev_dev->cache_valid & VALID_POLICING
1598 && netdev_dev->kbits_rate == kbits_rate
1599 && netdev_dev->kbits_burst == kbits_burst) {
1600 /* Assume that settings haven't changed since we last set them. */
1604 COVERAGE_INC(netdev_set_policing);
1605 /* Remove any existing ingress qdisc. */
1606 error = tc_add_del_ingress_qdisc(netdev, false);
1608 VLOG_WARN_RL(&rl, "%s: removing policing failed: %s",
1609 netdev_name, strerror(error));
1614 error = tc_add_del_ingress_qdisc(netdev, true);
1616 VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s",
1617 netdev_name, strerror(error));
1621 error = tc_add_policer(netdev, kbits_rate, kbits_burst);
1623 VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s",
1624 netdev_name, strerror(error));
1629 netdev_dev->kbits_rate = kbits_rate;
1630 netdev_dev->kbits_burst = kbits_burst;
1631 netdev_dev->cache_valid |= VALID_POLICING;
1637 netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED,
1640 const struct tc_ops **opsp;
1642 for (opsp = tcs; *opsp != NULL; opsp++) {
1643 const struct tc_ops *ops = *opsp;
1644 if (ops->tc_install && ops->ovs_name[0] != '\0') {
1645 sset_add(types, ops->ovs_name);
1651 static const struct tc_ops *
1652 tc_lookup_ovs_name(const char *name)
1654 const struct tc_ops **opsp;
1656 for (opsp = tcs; *opsp != NULL; opsp++) {
1657 const struct tc_ops *ops = *opsp;
1658 if (!strcmp(name, ops->ovs_name)) {
1665 static const struct tc_ops *
1666 tc_lookup_linux_name(const char *name)
1668 const struct tc_ops **opsp;
1670 for (opsp = tcs; *opsp != NULL; opsp++) {
1671 const struct tc_ops *ops = *opsp;
1672 if (ops->linux_name && !strcmp(name, ops->linux_name)) {
1679 static struct tc_queue *
1680 tc_find_queue__(const struct netdev *netdev, unsigned int queue_id,
1683 struct netdev_dev_linux *netdev_dev =
1684 netdev_dev_linux_cast(netdev_get_dev(netdev));
1685 struct tc_queue *queue;
1687 HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev_dev->tc->queues) {
1688 if (queue->queue_id == queue_id) {
1695 static struct tc_queue *
1696 tc_find_queue(const struct netdev *netdev, unsigned int queue_id)
1698 return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0));
1702 netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED,
1704 struct netdev_qos_capabilities *caps)
1706 const struct tc_ops *ops = tc_lookup_ovs_name(type);
1710 caps->n_queues = ops->n_queues;
1715 netdev_linux_get_qos(const struct netdev *netdev,
1716 const char **typep, struct shash *details)
1718 struct netdev_dev_linux *netdev_dev =
1719 netdev_dev_linux_cast(netdev_get_dev(netdev));
1722 error = tc_query_qdisc(netdev);
1727 *typep = netdev_dev->tc->ops->ovs_name;
1728 return (netdev_dev->tc->ops->qdisc_get
1729 ? netdev_dev->tc->ops->qdisc_get(netdev, details)
1734 netdev_linux_set_qos(struct netdev *netdev,
1735 const char *type, const struct shash *details)
1737 struct netdev_dev_linux *netdev_dev =
1738 netdev_dev_linux_cast(netdev_get_dev(netdev));
1739 const struct tc_ops *new_ops;
1742 new_ops = tc_lookup_ovs_name(type);
1743 if (!new_ops || !new_ops->tc_install) {
1747 error = tc_query_qdisc(netdev);
1752 if (new_ops == netdev_dev->tc->ops) {
1753 return new_ops->qdisc_set ? new_ops->qdisc_set(netdev, details) : 0;
1755 /* Delete existing qdisc. */
1756 error = tc_del_qdisc(netdev);
1760 assert(netdev_dev->tc == NULL);
1762 /* Install new qdisc. */
1763 error = new_ops->tc_install(netdev, details);
1764 assert((error == 0) == (netdev_dev->tc != NULL));
1771 netdev_linux_get_queue(const struct netdev *netdev,
1772 unsigned int queue_id, struct shash *details)
1774 struct netdev_dev_linux *netdev_dev =
1775 netdev_dev_linux_cast(netdev_get_dev(netdev));
1778 error = tc_query_qdisc(netdev);
1782 struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1784 ? netdev_dev->tc->ops->class_get(netdev, queue, details)
1790 netdev_linux_set_queue(struct netdev *netdev,
1791 unsigned int queue_id, const struct shash *details)
1793 struct netdev_dev_linux *netdev_dev =
1794 netdev_dev_linux_cast(netdev_get_dev(netdev));
1797 error = tc_query_qdisc(netdev);
1800 } else if (queue_id >= netdev_dev->tc->ops->n_queues
1801 || !netdev_dev->tc->ops->class_set) {
1805 return netdev_dev->tc->ops->class_set(netdev, queue_id, details);
1809 netdev_linux_delete_queue(struct netdev *netdev, unsigned int queue_id)
1811 struct netdev_dev_linux *netdev_dev =
1812 netdev_dev_linux_cast(netdev_get_dev(netdev));
1815 error = tc_query_qdisc(netdev);
1818 } else if (!netdev_dev->tc->ops->class_delete) {
1821 struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1823 ? netdev_dev->tc->ops->class_delete(netdev, queue)
1829 netdev_linux_get_queue_stats(const struct netdev *netdev,
1830 unsigned int queue_id,
1831 struct netdev_queue_stats *stats)
1833 struct netdev_dev_linux *netdev_dev =
1834 netdev_dev_linux_cast(netdev_get_dev(netdev));
1837 error = tc_query_qdisc(netdev);
1840 } else if (!netdev_dev->tc->ops->class_get_stats) {
1843 const struct tc_queue *queue = tc_find_queue(netdev, queue_id);
1845 ? netdev_dev->tc->ops->class_get_stats(netdev, queue, stats)
1851 start_queue_dump(const struct netdev *netdev, struct nl_dump *dump)
1853 struct ofpbuf request;
1854 struct tcmsg *tcmsg;
1856 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request);
1860 tcmsg->tcm_parent = 0;
1861 nl_dump_start(dump, rtnl_sock, &request);
1862 ofpbuf_uninit(&request);
1867 netdev_linux_dump_queues(const struct netdev *netdev,
1868 netdev_dump_queues_cb *cb, void *aux)
1870 struct netdev_dev_linux *netdev_dev =
1871 netdev_dev_linux_cast(netdev_get_dev(netdev));
1872 struct tc_queue *queue;
1873 struct shash details;
1877 error = tc_query_qdisc(netdev);
1880 } else if (!netdev_dev->tc->ops->class_get) {
1885 shash_init(&details);
1886 HMAP_FOR_EACH (queue, hmap_node, &netdev_dev->tc->queues) {
1887 shash_clear(&details);
1889 error = netdev_dev->tc->ops->class_get(netdev, queue, &details);
1891 (*cb)(queue->queue_id, &details, aux);
1896 shash_destroy(&details);
1902 netdev_linux_dump_queue_stats(const struct netdev *netdev,
1903 netdev_dump_queue_stats_cb *cb, void *aux)
1905 struct netdev_dev_linux *netdev_dev =
1906 netdev_dev_linux_cast(netdev_get_dev(netdev));
1907 struct nl_dump dump;
1912 error = tc_query_qdisc(netdev);
1915 } else if (!netdev_dev->tc->ops->class_dump_stats) {
1920 if (!start_queue_dump(netdev, &dump)) {
1923 while (nl_dump_next(&dump, &msg)) {
1924 error = netdev_dev->tc->ops->class_dump_stats(netdev, &msg, cb, aux);
1930 error = nl_dump_done(&dump);
1931 return error ? error : last_error;
1935 netdev_linux_get_in4(const struct netdev *netdev_,
1936 struct in_addr *address, struct in_addr *netmask)
1938 struct netdev_dev_linux *netdev_dev =
1939 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1941 if (!(netdev_dev->cache_valid & VALID_IN4)) {
1944 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
1945 SIOCGIFADDR, "SIOCGIFADDR");
1950 error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
1951 SIOCGIFNETMASK, "SIOCGIFNETMASK");
1956 netdev_dev->cache_valid |= VALID_IN4;
1958 *address = netdev_dev->address;
1959 *netmask = netdev_dev->netmask;
1960 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
1964 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
1965 struct in_addr netmask)
1967 struct netdev_dev_linux *netdev_dev =
1968 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1971 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
1973 netdev_dev->cache_valid |= VALID_IN4;
1974 netdev_dev->address = address;
1975 netdev_dev->netmask = netmask;
1976 if (address.s_addr != INADDR_ANY) {
1977 error = do_set_addr(netdev_, SIOCSIFNETMASK,
1978 "SIOCSIFNETMASK", netmask);
1985 parse_if_inet6_line(const char *line,
1986 struct in6_addr *in6, char ifname[16 + 1])
1988 uint8_t *s6 = in6->s6_addr;
1989 #define X8 "%2"SCNx8
1991 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
1992 "%*x %*x %*x %*x %16s\n",
1993 &s6[0], &s6[1], &s6[2], &s6[3],
1994 &s6[4], &s6[5], &s6[6], &s6[7],
1995 &s6[8], &s6[9], &s6[10], &s6[11],
1996 &s6[12], &s6[13], &s6[14], &s6[15],
2000 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
2001 * 'in6' is non-null) and returns true. Otherwise, returns false. */
2003 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
2005 struct netdev_dev_linux *netdev_dev =
2006 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2007 if (!(netdev_dev->cache_valid & VALID_IN6)) {
2011 netdev_dev->in6 = in6addr_any;
2013 file = fopen("/proc/net/if_inet6", "r");
2015 const char *name = netdev_get_name(netdev_);
2016 while (fgets(line, sizeof line, file)) {
2017 struct in6_addr in6_tmp;
2018 char ifname[16 + 1];
2019 if (parse_if_inet6_line(line, &in6_tmp, ifname)
2020 && !strcmp(name, ifname))
2022 netdev_dev->in6 = in6_tmp;
2028 netdev_dev->cache_valid |= VALID_IN6;
2030 *in6 = netdev_dev->in6;
2035 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
2037 struct sockaddr_in sin;
2038 memset(&sin, 0, sizeof sin);
2039 sin.sin_family = AF_INET;
2040 sin.sin_addr = addr;
2043 memset(sa, 0, sizeof *sa);
2044 memcpy(sa, &sin, sizeof sin);
2048 do_set_addr(struct netdev *netdev,
2049 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
2052 ovs_strzcpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
2053 make_in4_sockaddr(&ifr.ifr_addr, addr);
2055 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
2059 /* Adds 'router' as a default IP gateway. */
2061 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
2063 struct in_addr any = { INADDR_ANY };
2067 memset(&rt, 0, sizeof rt);
2068 make_in4_sockaddr(&rt.rt_dst, any);
2069 make_in4_sockaddr(&rt.rt_gateway, router);
2070 make_in4_sockaddr(&rt.rt_genmask, any);
2071 rt.rt_flags = RTF_UP | RTF_GATEWAY;
2072 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
2074 VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
2080 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
2083 static const char fn[] = "/proc/net/route";
2088 *netdev_name = NULL;
2089 stream = fopen(fn, "r");
2090 if (stream == NULL) {
2091 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
2096 while (fgets(line, sizeof line, stream)) {
2099 ovs_be32 dest, gateway, mask;
2100 int refcnt, metric, mtu;
2101 unsigned int flags, use, window, irtt;
2104 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
2106 iface, &dest, &gateway, &flags, &refcnt,
2107 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
2109 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
2113 if (!(flags & RTF_UP)) {
2114 /* Skip routes that aren't up. */
2118 /* The output of 'dest', 'mask', and 'gateway' were given in
2119 * network byte order, so we don't need need any endian
2120 * conversions here. */
2121 if ((dest & mask) == (host->s_addr & mask)) {
2123 /* The host is directly reachable. */
2124 next_hop->s_addr = 0;
2126 /* To reach the host, we must go through a gateway. */
2127 next_hop->s_addr = gateway;
2129 *netdev_name = xstrdup(iface);
2141 netdev_linux_get_status(const struct netdev *netdev, struct shash *sh)
2143 struct ethtool_drvinfo drvinfo;
2146 memset(&drvinfo, 0, sizeof drvinfo);
2147 error = netdev_linux_do_ethtool(netdev_get_name(netdev),
2148 (struct ethtool_cmd *)&drvinfo,
2150 "ETHTOOL_GDRVINFO");
2152 shash_add(sh, "driver_name", xstrdup(drvinfo.driver));
2153 shash_add(sh, "driver_version", xstrdup(drvinfo.version));
2154 shash_add(sh, "firmware_version", xstrdup(drvinfo.fw_version));
2160 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
2161 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
2162 * returns 0. Otherwise, it returns a positive errno value; in particular,
2163 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
2165 netdev_linux_arp_lookup(const struct netdev *netdev,
2166 ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
2169 struct sockaddr_in sin;
2172 memset(&r, 0, sizeof r);
2173 memset(&sin, 0, sizeof sin);
2174 sin.sin_family = AF_INET;
2175 sin.sin_addr.s_addr = ip;
2177 memcpy(&r.arp_pa, &sin, sizeof sin);
2178 r.arp_ha.sa_family = ARPHRD_ETHER;
2180 ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
2181 COVERAGE_INC(netdev_arp_lookup);
2182 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
2184 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
2185 } else if (retval != ENXIO) {
2186 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
2187 netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
2193 nd_to_iff_flags(enum netdev_flags nd)
2196 if (nd & NETDEV_UP) {
2199 if (nd & NETDEV_PROMISC) {
2206 iff_to_nd_flags(int iff)
2208 enum netdev_flags nd = 0;
2212 if (iff & IFF_PROMISC) {
2213 nd |= NETDEV_PROMISC;
2219 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
2220 enum netdev_flags on, enum netdev_flags *old_flagsp)
2222 struct netdev_dev_linux *netdev_dev;
2223 int old_flags, new_flags;
2226 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
2227 old_flags = netdev_dev->ifi_flags;
2228 *old_flagsp = iff_to_nd_flags(old_flags);
2229 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
2230 if (new_flags != old_flags) {
2231 error = set_flags(netdev, new_flags);
2232 get_flags(&netdev_dev->netdev_dev, &netdev_dev->ifi_flags);
2238 netdev_linux_change_seq(const struct netdev *netdev)
2240 return netdev_dev_linux_cast(netdev_get_dev(netdev))->change_seq;
2243 #define NETDEV_LINUX_CLASS(NAME, CREATE, GET_STATS, SET_STATS) \
2247 netdev_linux_init, \
2249 netdev_linux_wait, \
2252 netdev_linux_destroy, \
2253 NULL, /* get_config */ \
2254 NULL, /* set_config */ \
2256 netdev_linux_open, \
2257 netdev_linux_close, \
2259 netdev_linux_listen, \
2260 netdev_linux_recv, \
2261 netdev_linux_recv_wait, \
2262 netdev_linux_drain, \
2264 netdev_linux_send, \
2265 netdev_linux_send_wait, \
2267 netdev_linux_set_etheraddr, \
2268 netdev_linux_get_etheraddr, \
2269 netdev_linux_get_mtu, \
2270 netdev_linux_set_mtu, \
2271 netdev_linux_get_ifindex, \
2272 netdev_linux_get_carrier, \
2273 netdev_linux_get_carrier_resets, \
2274 netdev_linux_set_miimon_interval, \
2278 netdev_linux_get_features, \
2279 netdev_linux_set_advertisements, \
2281 netdev_linux_set_policing, \
2282 netdev_linux_get_qos_types, \
2283 netdev_linux_get_qos_capabilities, \
2284 netdev_linux_get_qos, \
2285 netdev_linux_set_qos, \
2286 netdev_linux_get_queue, \
2287 netdev_linux_set_queue, \
2288 netdev_linux_delete_queue, \
2289 netdev_linux_get_queue_stats, \
2290 netdev_linux_dump_queues, \
2291 netdev_linux_dump_queue_stats, \
2293 netdev_linux_get_in4, \
2294 netdev_linux_set_in4, \
2295 netdev_linux_get_in6, \
2296 netdev_linux_add_router, \
2297 netdev_linux_get_next_hop, \
2298 netdev_linux_get_status, \
2299 netdev_linux_arp_lookup, \
2301 netdev_linux_update_flags, \
2303 netdev_linux_change_seq \
2306 const struct netdev_class netdev_linux_class =
2309 netdev_linux_create,
2310 netdev_linux_get_stats,
2311 NULL); /* set_stats */
2313 const struct netdev_class netdev_tap_class =
2316 netdev_linux_create_tap,
2317 netdev_tap_get_stats,
2318 NULL); /* set_stats */
2320 const struct netdev_class netdev_internal_class =
2323 netdev_linux_create,
2324 netdev_internal_get_stats,
2325 netdev_vport_set_stats);
2327 /* HTB traffic control class. */
2329 #define HTB_N_QUEUES 0xf000
2333 unsigned int max_rate; /* In bytes/s. */
2337 struct tc_queue tc_queue;
2338 unsigned int min_rate; /* In bytes/s. */
2339 unsigned int max_rate; /* In bytes/s. */
2340 unsigned int burst; /* In bytes. */
2341 unsigned int priority; /* Lower values are higher priorities. */
2345 htb_get__(const struct netdev *netdev)
2347 struct netdev_dev_linux *netdev_dev =
2348 netdev_dev_linux_cast(netdev_get_dev(netdev));
2349 return CONTAINER_OF(netdev_dev->tc, struct htb, tc);
2353 htb_install__(struct netdev *netdev, uint64_t max_rate)
2355 struct netdev_dev_linux *netdev_dev =
2356 netdev_dev_linux_cast(netdev_get_dev(netdev));
2359 htb = xmalloc(sizeof *htb);
2360 tc_init(&htb->tc, &tc_ops_htb);
2361 htb->max_rate = max_rate;
2363 netdev_dev->tc = &htb->tc;
2366 /* Create an HTB qdisc.
2368 * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */
2370 htb_setup_qdisc__(struct netdev *netdev)
2373 struct tc_htb_glob opt;
2374 struct ofpbuf request;
2375 struct tcmsg *tcmsg;
2377 tc_del_qdisc(netdev);
2379 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
2380 NLM_F_EXCL | NLM_F_CREATE, &request);
2384 tcmsg->tcm_handle = tc_make_handle(1, 0);
2385 tcmsg->tcm_parent = TC_H_ROOT;
2387 nl_msg_put_string(&request, TCA_KIND, "htb");
2389 memset(&opt, 0, sizeof opt);
2390 opt.rate2quantum = 10;
2394 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2395 nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt);
2396 nl_msg_end_nested(&request, opt_offset);
2398 return tc_transact(&request, NULL);
2401 /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb
2402 * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */
2404 htb_setup_class__(struct netdev *netdev, unsigned int handle,
2405 unsigned int parent, struct htb_class *class)
2408 struct tc_htb_opt opt;
2409 struct ofpbuf request;
2410 struct tcmsg *tcmsg;
2414 error = netdev_get_mtu(netdev, &mtu);
2416 VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU",
2417 netdev_get_name(netdev));
2421 memset(&opt, 0, sizeof opt);
2422 tc_fill_rate(&opt.rate, class->min_rate, mtu);
2423 tc_fill_rate(&opt.ceil, class->max_rate, mtu);
2424 opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst);
2425 opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst);
2426 opt.prio = class->priority;
2428 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
2432 tcmsg->tcm_handle = handle;
2433 tcmsg->tcm_parent = parent;
2435 nl_msg_put_string(&request, TCA_KIND, "htb");
2436 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2437 nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt);
2438 tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate);
2439 tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil);
2440 nl_msg_end_nested(&request, opt_offset);
2442 error = tc_transact(&request, NULL);
2444 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
2445 "min_rate=%u max_rate=%u burst=%u prio=%u (%s)",
2446 netdev_get_name(netdev),
2447 tc_get_major(handle), tc_get_minor(handle),
2448 tc_get_major(parent), tc_get_minor(parent),
2449 class->min_rate, class->max_rate,
2450 class->burst, class->priority, strerror(error));
2455 /* Parses Netlink attributes in 'options' for HTB parameters and stores a
2456 * description of them into 'details'. The description complies with the
2457 * specification given in the vswitch database documentation for linux-htb
2460 htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class)
2462 static const struct nl_policy tca_htb_policy[] = {
2463 [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false,
2464 .min_len = sizeof(struct tc_htb_opt) },
2467 struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)];
2468 const struct tc_htb_opt *htb;
2470 if (!nl_parse_nested(nl_options, tca_htb_policy,
2471 attrs, ARRAY_SIZE(tca_htb_policy))) {
2472 VLOG_WARN_RL(&rl, "failed to parse HTB class options");
2476 htb = nl_attr_get(attrs[TCA_HTB_PARMS]);
2477 class->min_rate = htb->rate.rate;
2478 class->max_rate = htb->ceil.rate;
2479 class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer);
2480 class->priority = htb->prio;
2485 htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
2486 struct htb_class *options,
2487 struct netdev_queue_stats *stats)
2489 struct nlattr *nl_options;
2490 unsigned int handle;
2493 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
2494 if (!error && queue_id) {
2495 unsigned int major = tc_get_major(handle);
2496 unsigned int minor = tc_get_minor(handle);
2497 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2498 *queue_id = minor - 1;
2503 if (!error && options) {
2504 error = htb_parse_tca_options__(nl_options, options);
2510 htb_parse_qdisc_details__(struct netdev *netdev,
2511 const struct shash *details, struct htb_class *hc)
2513 const char *max_rate_s;
2515 max_rate_s = shash_find_data(details, "max-rate");
2516 hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
2517 if (!hc->max_rate) {
2520 netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
2521 hc->max_rate = netdev_features_to_bps(current) / 8;
2523 hc->min_rate = hc->max_rate;
2529 htb_parse_class_details__(struct netdev *netdev,
2530 const struct shash *details, struct htb_class *hc)
2532 const struct htb *htb = htb_get__(netdev);
2533 const char *min_rate_s = shash_find_data(details, "min-rate");
2534 const char *max_rate_s = shash_find_data(details, "max-rate");
2535 const char *burst_s = shash_find_data(details, "burst");
2536 const char *priority_s = shash_find_data(details, "priority");
2539 error = netdev_get_mtu(netdev, &mtu);
2541 VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU",
2542 netdev_get_name(netdev));
2546 /* HTB requires at least an mtu sized min-rate to send any traffic even
2547 * on uncongested links. */
2548 hc->min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
2549 hc->min_rate = MAX(hc->min_rate, mtu);
2550 hc->min_rate = MIN(hc->min_rate, htb->max_rate);
2553 hc->max_rate = (max_rate_s
2554 ? strtoull(max_rate_s, NULL, 10) / 8
2556 hc->max_rate = MAX(hc->max_rate, hc->min_rate);
2557 hc->max_rate = MIN(hc->max_rate, htb->max_rate);
2561 * According to hints in the documentation that I've read, it is important
2562 * that 'burst' be at least as big as the largest frame that might be
2563 * transmitted. Also, making 'burst' a bit bigger than necessary is OK,
2564 * but having it a bit too small is a problem. Since netdev_get_mtu()
2565 * doesn't include the Ethernet header, we need to add at least 14 (18?) to
2566 * the MTU. We actually add 64, instead of 14, as a guard against
2567 * additional headers get tacked on somewhere that we're not aware of. */
2568 hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0;
2569 hc->burst = MAX(hc->burst, mtu + 64);
2572 hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0;
2578 htb_query_class__(const struct netdev *netdev, unsigned int handle,
2579 unsigned int parent, struct htb_class *options,
2580 struct netdev_queue_stats *stats)
2582 struct ofpbuf *reply;
2585 error = tc_query_class(netdev, handle, parent, &reply);
2587 error = htb_parse_tcmsg__(reply, NULL, options, stats);
2588 ofpbuf_delete(reply);
2594 htb_tc_install(struct netdev *netdev, const struct shash *details)
2598 error = htb_setup_qdisc__(netdev);
2600 struct htb_class hc;
2602 htb_parse_qdisc_details__(netdev, details, &hc);
2603 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
2604 tc_make_handle(1, 0), &hc);
2606 htb_install__(netdev, hc.max_rate);
2612 static struct htb_class *
2613 htb_class_cast__(const struct tc_queue *queue)
2615 return CONTAINER_OF(queue, struct htb_class, tc_queue);
2619 htb_update_queue__(struct netdev *netdev, unsigned int queue_id,
2620 const struct htb_class *hc)
2622 struct htb *htb = htb_get__(netdev);
2623 size_t hash = hash_int(queue_id, 0);
2624 struct tc_queue *queue;
2625 struct htb_class *hcp;
2627 queue = tc_find_queue__(netdev, queue_id, hash);
2629 hcp = htb_class_cast__(queue);
2631 hcp = xmalloc(sizeof *hcp);
2632 queue = &hcp->tc_queue;
2633 queue->queue_id = queue_id;
2634 hmap_insert(&htb->tc.queues, &queue->hmap_node, hash);
2637 hcp->min_rate = hc->min_rate;
2638 hcp->max_rate = hc->max_rate;
2639 hcp->burst = hc->burst;
2640 hcp->priority = hc->priority;
2644 htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
2647 struct nl_dump dump;
2648 struct htb_class hc;
2650 /* Get qdisc options. */
2652 htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
2653 htb_install__(netdev, hc.max_rate);
2656 if (!start_queue_dump(netdev, &dump)) {
2659 while (nl_dump_next(&dump, &msg)) {
2660 unsigned int queue_id;
2662 if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
2663 htb_update_queue__(netdev, queue_id, &hc);
2666 nl_dump_done(&dump);
2672 htb_tc_destroy(struct tc *tc)
2674 struct htb *htb = CONTAINER_OF(tc, struct htb, tc);
2675 struct htb_class *hc, *next;
2677 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) {
2678 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
2686 htb_qdisc_get(const struct netdev *netdev, struct shash *details)
2688 const struct htb *htb = htb_get__(netdev);
2689 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * htb->max_rate));
2694 htb_qdisc_set(struct netdev *netdev, const struct shash *details)
2696 struct htb_class hc;
2699 htb_parse_qdisc_details__(netdev, details, &hc);
2700 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
2701 tc_make_handle(1, 0), &hc);
2703 htb_get__(netdev)->max_rate = hc.max_rate;
2709 htb_class_get(const struct netdev *netdev OVS_UNUSED,
2710 const struct tc_queue *queue, struct shash *details)
2712 const struct htb_class *hc = htb_class_cast__(queue);
2714 shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate));
2715 if (hc->min_rate != hc->max_rate) {
2716 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate));
2718 shash_add(details, "burst", xasprintf("%llu", 8ULL * hc->burst));
2720 shash_add(details, "priority", xasprintf("%u", hc->priority));
2726 htb_class_set(struct netdev *netdev, unsigned int queue_id,
2727 const struct shash *details)
2729 struct htb_class hc;
2732 error = htb_parse_class_details__(netdev, details, &hc);
2737 error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
2738 tc_make_handle(1, 0xfffe), &hc);
2743 htb_update_queue__(netdev, queue_id, &hc);
2748 htb_class_delete(struct netdev *netdev, struct tc_queue *queue)
2750 struct htb_class *hc = htb_class_cast__(queue);
2751 struct htb *htb = htb_get__(netdev);
2754 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
2756 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
2763 htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
2764 struct netdev_queue_stats *stats)
2766 return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
2767 tc_make_handle(1, 0xfffe), NULL, stats);
2771 htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
2772 const struct ofpbuf *nlmsg,
2773 netdev_dump_queue_stats_cb *cb, void *aux)
2775 struct netdev_queue_stats stats;
2776 unsigned int handle, major, minor;
2779 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
2784 major = tc_get_major(handle);
2785 minor = tc_get_minor(handle);
2786 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2787 (*cb)(minor - 1, &stats, aux);
2792 static const struct tc_ops tc_ops_htb = {
2793 "htb", /* linux_name */
2794 "linux-htb", /* ovs_name */
2795 HTB_N_QUEUES, /* n_queues */
2804 htb_class_get_stats,
2805 htb_class_dump_stats
2808 /* "linux-hfsc" traffic control class. */
2810 #define HFSC_N_QUEUES 0xf000
2818 struct tc_queue tc_queue;
2823 static struct hfsc *
2824 hfsc_get__(const struct netdev *netdev)
2826 struct netdev_dev_linux *netdev_dev;
2827 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
2828 return CONTAINER_OF(netdev_dev->tc, struct hfsc, tc);
2831 static struct hfsc_class *
2832 hfsc_class_cast__(const struct tc_queue *queue)
2834 return CONTAINER_OF(queue, struct hfsc_class, tc_queue);
2838 hfsc_install__(struct netdev *netdev, uint32_t max_rate)
2840 struct netdev_dev_linux * netdev_dev;
2843 netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
2844 hfsc = xmalloc(sizeof *hfsc);
2845 tc_init(&hfsc->tc, &tc_ops_hfsc);
2846 hfsc->max_rate = max_rate;
2847 netdev_dev->tc = &hfsc->tc;
2851 hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id,
2852 const struct hfsc_class *hc)
2856 struct hfsc_class *hcp;
2857 struct tc_queue *queue;
2859 hfsc = hfsc_get__(netdev);
2860 hash = hash_int(queue_id, 0);
2862 queue = tc_find_queue__(netdev, queue_id, hash);
2864 hcp = hfsc_class_cast__(queue);
2866 hcp = xmalloc(sizeof *hcp);
2867 queue = &hcp->tc_queue;
2868 queue->queue_id = queue_id;
2869 hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash);
2872 hcp->min_rate = hc->min_rate;
2873 hcp->max_rate = hc->max_rate;
2877 hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class)
2879 const struct tc_service_curve *rsc, *fsc, *usc;
2880 static const struct nl_policy tca_hfsc_policy[] = {
2882 .type = NL_A_UNSPEC,
2884 .min_len = sizeof(struct tc_service_curve),
2887 .type = NL_A_UNSPEC,
2889 .min_len = sizeof(struct tc_service_curve),
2892 .type = NL_A_UNSPEC,
2894 .min_len = sizeof(struct tc_service_curve),
2897 struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)];
2899 if (!nl_parse_nested(nl_options, tca_hfsc_policy,
2900 attrs, ARRAY_SIZE(tca_hfsc_policy))) {
2901 VLOG_WARN_RL(&rl, "failed to parse HFSC class options");
2905 rsc = nl_attr_get(attrs[TCA_HFSC_RSC]);
2906 fsc = nl_attr_get(attrs[TCA_HFSC_FSC]);
2907 usc = nl_attr_get(attrs[TCA_HFSC_USC]);
2909 if (rsc->m1 != 0 || rsc->d != 0 ||
2910 fsc->m1 != 0 || fsc->d != 0 ||
2911 usc->m1 != 0 || usc->d != 0) {
2912 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
2913 "Non-linear service curves are not supported.");
2917 if (rsc->m2 != fsc->m2) {
2918 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
2919 "Real-time service curves are not supported ");
2923 if (rsc->m2 > usc->m2) {
2924 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
2925 "Min-rate service curve is greater than "
2926 "the max-rate service curve.");
2930 class->min_rate = fsc->m2;
2931 class->max_rate = usc->m2;
2936 hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
2937 struct hfsc_class *options,
2938 struct netdev_queue_stats *stats)
2941 unsigned int handle;
2942 struct nlattr *nl_options;
2944 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
2950 unsigned int major, minor;
2952 major = tc_get_major(handle);
2953 minor = tc_get_minor(handle);
2954 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
2955 *queue_id = minor - 1;
2962 error = hfsc_parse_tca_options__(nl_options, options);
2969 hfsc_query_class__(const struct netdev *netdev, unsigned int handle,
2970 unsigned int parent, struct hfsc_class *options,
2971 struct netdev_queue_stats *stats)
2974 struct ofpbuf *reply;
2976 error = tc_query_class(netdev, handle, parent, &reply);
2981 error = hfsc_parse_tcmsg__(reply, NULL, options, stats);
2982 ofpbuf_delete(reply);
2987 hfsc_parse_qdisc_details__(struct netdev *netdev, const struct shash *details,
2988 struct hfsc_class *class)
2991 const char *max_rate_s;
2993 max_rate_s = shash_find_data(details, "max-rate");
2994 max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
2999 netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
3000 max_rate = netdev_features_to_bps(current) / 8;
3003 class->min_rate = max_rate;
3004 class->max_rate = max_rate;
3008 hfsc_parse_class_details__(struct netdev *netdev,
3009 const struct shash *details,
3010 struct hfsc_class * class)
3012 const struct hfsc *hfsc;
3013 uint32_t min_rate, max_rate;
3014 const char *min_rate_s, *max_rate_s;
3016 hfsc = hfsc_get__(netdev);
3017 min_rate_s = shash_find_data(details, "min-rate");
3018 max_rate_s = shash_find_data(details, "max-rate");
3020 min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
3021 min_rate = MAX(min_rate, 1);
3022 min_rate = MIN(min_rate, hfsc->max_rate);
3024 max_rate = (max_rate_s
3025 ? strtoull(max_rate_s, NULL, 10) / 8
3027 max_rate = MAX(max_rate, min_rate);
3028 max_rate = MIN(max_rate, hfsc->max_rate);
3030 class->min_rate = min_rate;
3031 class->max_rate = max_rate;
3036 /* Create an HFSC qdisc.
3038 * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */
3040 hfsc_setup_qdisc__(struct netdev * netdev)
3042 struct tcmsg *tcmsg;
3043 struct ofpbuf request;
3044 struct tc_hfsc_qopt opt;
3046 tc_del_qdisc(netdev);
3048 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
3049 NLM_F_EXCL | NLM_F_CREATE, &request);
3055 tcmsg->tcm_handle = tc_make_handle(1, 0);
3056 tcmsg->tcm_parent = TC_H_ROOT;
3058 memset(&opt, 0, sizeof opt);
3061 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3062 nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt);
3064 return tc_transact(&request, NULL);
3067 /* Create an HFSC class.
3069 * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc
3070 * sc rate <min_rate> ul rate <max_rate>" */
3072 hfsc_setup_class__(struct netdev *netdev, unsigned int handle,
3073 unsigned int parent, struct hfsc_class *class)
3077 struct tcmsg *tcmsg;
3078 struct ofpbuf request;
3079 struct tc_service_curve min, max;
3081 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
3087 tcmsg->tcm_handle = handle;
3088 tcmsg->tcm_parent = parent;
3092 min.m2 = class->min_rate;
3096 max.m2 = class->max_rate;
3098 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3099 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
3100 nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min);
3101 nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min);
3102 nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max);
3103 nl_msg_end_nested(&request, opt_offset);
3105 error = tc_transact(&request, NULL);
3107 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
3108 "min-rate %ubps, max-rate %ubps (%s)",
3109 netdev_get_name(netdev),
3110 tc_get_major(handle), tc_get_minor(handle),
3111 tc_get_major(parent), tc_get_minor(parent),
3112 class->min_rate, class->max_rate, strerror(error));
3119 hfsc_tc_install(struct netdev *netdev, const struct shash *details)
3122 struct hfsc_class class;
3124 error = hfsc_setup_qdisc__(netdev);
3130 hfsc_parse_qdisc_details__(netdev, details, &class);
3131 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3132 tc_make_handle(1, 0), &class);
3138 hfsc_install__(netdev, class.max_rate);
3143 hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3146 struct nl_dump dump;
3147 struct hfsc_class hc;
3150 hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
3151 hfsc_install__(netdev, hc.max_rate);
3153 if (!start_queue_dump(netdev, &dump)) {
3157 while (nl_dump_next(&dump, &msg)) {
3158 unsigned int queue_id;
3160 if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
3161 hfsc_update_queue__(netdev, queue_id, &hc);
3165 nl_dump_done(&dump);
3170 hfsc_tc_destroy(struct tc *tc)
3173 struct hfsc_class *hc, *next;
3175 hfsc = CONTAINER_OF(tc, struct hfsc, tc);
3177 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) {
3178 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3187 hfsc_qdisc_get(const struct netdev *netdev, struct shash *details)
3189 const struct hfsc *hfsc;
3190 hfsc = hfsc_get__(netdev);
3191 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hfsc->max_rate));
3196 hfsc_qdisc_set(struct netdev *netdev, const struct shash *details)
3199 struct hfsc_class class;
3201 hfsc_parse_qdisc_details__(netdev, details, &class);
3202 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3203 tc_make_handle(1, 0), &class);
3206 hfsc_get__(netdev)->max_rate = class.max_rate;
3213 hfsc_class_get(const struct netdev *netdev OVS_UNUSED,
3214 const struct tc_queue *queue, struct shash *details)
3216 const struct hfsc_class *hc;
3218 hc = hfsc_class_cast__(queue);
3219 shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate));
3220 if (hc->min_rate != hc->max_rate) {
3221 shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate));
3227 hfsc_class_set(struct netdev *netdev, unsigned int queue_id,
3228 const struct shash *details)
3231 struct hfsc_class class;
3233 error = hfsc_parse_class_details__(netdev, details, &class);
3238 error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
3239 tc_make_handle(1, 0xfffe), &class);
3244 hfsc_update_queue__(netdev, queue_id, &class);
3249 hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue)
3253 struct hfsc_class *hc;
3255 hc = hfsc_class_cast__(queue);
3256 hfsc = hfsc_get__(netdev);
3258 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
3260 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3267 hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
3268 struct netdev_queue_stats *stats)
3270 return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
3271 tc_make_handle(1, 0xfffe), NULL, stats);
3275 hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
3276 const struct ofpbuf *nlmsg,
3277 netdev_dump_queue_stats_cb *cb, void *aux)
3279 struct netdev_queue_stats stats;
3280 unsigned int handle, major, minor;
3283 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
3288 major = tc_get_major(handle);
3289 minor = tc_get_minor(handle);
3290 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
3291 (*cb)(minor - 1, &stats, aux);
3296 static const struct tc_ops tc_ops_hfsc = {
3297 "hfsc", /* linux_name */
3298 "linux-hfsc", /* ovs_name */
3299 HFSC_N_QUEUES, /* n_queues */
3300 hfsc_tc_install, /* tc_install */
3301 hfsc_tc_load, /* tc_load */
3302 hfsc_tc_destroy, /* tc_destroy */
3303 hfsc_qdisc_get, /* qdisc_get */
3304 hfsc_qdisc_set, /* qdisc_set */
3305 hfsc_class_get, /* class_get */
3306 hfsc_class_set, /* class_set */
3307 hfsc_class_delete, /* class_delete */
3308 hfsc_class_get_stats, /* class_get_stats */
3309 hfsc_class_dump_stats /* class_dump_stats */
3312 /* "linux-default" traffic control class.
3314 * This class represents the default, unnamed Linux qdisc. It corresponds to
3315 * the "" (empty string) QoS type in the OVS database. */
3318 default_install__(struct netdev *netdev)
3320 struct netdev_dev_linux *netdev_dev =
3321 netdev_dev_linux_cast(netdev_get_dev(netdev));
3322 static struct tc *tc;
3325 tc = xmalloc(sizeof *tc);
3326 tc_init(tc, &tc_ops_default);
3328 netdev_dev->tc = tc;
3332 default_tc_install(struct netdev *netdev,
3333 const struct shash *details OVS_UNUSED)
3335 default_install__(netdev);
3340 default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3342 default_install__(netdev);
3346 static const struct tc_ops tc_ops_default = {
3347 NULL, /* linux_name */
3352 NULL, /* tc_destroy */
3353 NULL, /* qdisc_get */
3354 NULL, /* qdisc_set */
3355 NULL, /* class_get */
3356 NULL, /* class_set */
3357 NULL, /* class_delete */
3358 NULL, /* class_get_stats */
3359 NULL /* class_dump_stats */
3362 /* "linux-other" traffic control class.
3367 other_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3369 struct netdev_dev_linux *netdev_dev =
3370 netdev_dev_linux_cast(netdev_get_dev(netdev));
3371 static struct tc *tc;
3374 tc = xmalloc(sizeof *tc);
3375 tc_init(tc, &tc_ops_other);
3377 netdev_dev->tc = tc;
3381 static const struct tc_ops tc_ops_other = {
3382 NULL, /* linux_name */
3383 "linux-other", /* ovs_name */
3385 NULL, /* tc_install */
3387 NULL, /* tc_destroy */
3388 NULL, /* qdisc_get */
3389 NULL, /* qdisc_set */
3390 NULL, /* class_get */
3391 NULL, /* class_set */
3392 NULL, /* class_delete */
3393 NULL, /* class_get_stats */
3394 NULL /* class_dump_stats */
3397 /* Traffic control. */
3399 /* Number of kernel "tc" ticks per second. */
3400 static double ticks_per_s;
3402 /* Number of kernel "jiffies" per second. This is used for the purpose of
3403 * computing buffer sizes. Generally kernel qdiscs need to be able to buffer
3404 * one jiffy's worth of data.
3406 * There are two possibilities here:
3408 * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the
3409 * approximate range of 100 to 1024. That means that we really need to
3410 * make sure that the qdisc can buffer that much data.
3412 * - 'buffer_hz' is an absurdly large number. That means that the kernel
3413 * has finely granular timers and there's no need to fudge additional room
3414 * for buffers. (There's no extra effort needed to implement that: the
3415 * large 'buffer_hz' is used as a divisor, so practically any number will
3416 * come out as 0 in the division. Small integer results in the case of
3417 * really high dividends won't have any real effect anyhow.)
3419 static unsigned int buffer_hz;
3421 /* Returns tc handle 'major':'minor'. */
3423 tc_make_handle(unsigned int major, unsigned int minor)
3425 return TC_H_MAKE(major << 16, minor);
3428 /* Returns the major number from 'handle'. */
3430 tc_get_major(unsigned int handle)
3432 return TC_H_MAJ(handle) >> 16;
3435 /* Returns the minor number from 'handle'. */
3437 tc_get_minor(unsigned int handle)
3439 return TC_H_MIN(handle);
3442 static struct tcmsg *
3443 tc_make_request(const struct netdev *netdev, int type, unsigned int flags,
3444 struct ofpbuf *request)
3446 struct tcmsg *tcmsg;
3450 error = get_ifindex(netdev, &ifindex);
3455 ofpbuf_init(request, 512);
3456 nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags);
3457 tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg);
3458 tcmsg->tcm_family = AF_UNSPEC;
3459 tcmsg->tcm_ifindex = ifindex;
3460 /* Caller should fill in tcmsg->tcm_handle. */
3461 /* Caller should fill in tcmsg->tcm_parent. */
3467 tc_transact(struct ofpbuf *request, struct ofpbuf **replyp)
3469 int error = nl_sock_transact(rtnl_sock, request, replyp);
3470 ofpbuf_uninit(request);
3474 /* Adds or deletes a root ingress qdisc on 'netdev'. We use this for
3475 * policing configuration.
3477 * This function is equivalent to running the following when 'add' is true:
3478 * /sbin/tc qdisc add dev <devname> handle ffff: ingress
3480 * This function is equivalent to running the following when 'add' is false:
3481 * /sbin/tc qdisc del dev <devname> handle ffff: ingress
3483 * The configuration and stats may be seen with the following command:
3484 * /sbin/tc -s qdisc show dev <devname>
3486 * Returns 0 if successful, otherwise a positive errno value.
3489 tc_add_del_ingress_qdisc(struct netdev *netdev, bool add)
3491 struct ofpbuf request;
3492 struct tcmsg *tcmsg;
3494 int type = add ? RTM_NEWQDISC : RTM_DELQDISC;
3495 int flags = add ? NLM_F_EXCL | NLM_F_CREATE : 0;
3497 tcmsg = tc_make_request(netdev, type, flags, &request);
3501 tcmsg->tcm_handle = tc_make_handle(0xffff, 0);
3502 tcmsg->tcm_parent = TC_H_INGRESS;
3503 nl_msg_put_string(&request, TCA_KIND, "ingress");
3504 nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0);
3506 error = tc_transact(&request, NULL);
3508 /* If we're deleting the qdisc, don't worry about some of the
3509 * error conditions. */
3510 if (!add && (error == ENOENT || error == EINVAL)) {
3519 /* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size
3522 * This function is equivalent to running:
3523 * /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49
3524 * basic police rate <kbits_rate>kbit burst <kbits_burst>k
3527 * The configuration and stats may be seen with the following command:
3528 * /sbin/tc -s filter show <devname> eth0 parent ffff:
3530 * Returns 0 if successful, otherwise a positive errno value.
3533 tc_add_policer(struct netdev *netdev, int kbits_rate, int kbits_burst)
3535 struct tc_police tc_police;
3536 struct ofpbuf request;
3537 struct tcmsg *tcmsg;
3538 size_t basic_offset;
3539 size_t police_offset;
3543 memset(&tc_police, 0, sizeof tc_police);
3544 tc_police.action = TC_POLICE_SHOT;
3545 tc_police.mtu = mtu;
3546 tc_fill_rate(&tc_police.rate, kbits_rate/8 * 1000, mtu);
3547 tc_police.burst = tc_bytes_to_ticks(tc_police.rate.rate,
3548 kbits_burst * 1024);
3550 tcmsg = tc_make_request(netdev, RTM_NEWTFILTER,
3551 NLM_F_EXCL | NLM_F_CREATE, &request);
3555 tcmsg->tcm_parent = tc_make_handle(0xffff, 0);
3556 tcmsg->tcm_info = tc_make_handle(49,
3557 (OVS_FORCE uint16_t) htons(ETH_P_ALL));
3559 nl_msg_put_string(&request, TCA_KIND, "basic");
3560 basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
3561 police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE);
3562 nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police);
3563 tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate);
3564 nl_msg_end_nested(&request, police_offset);
3565 nl_msg_end_nested(&request, basic_offset);
3567 error = tc_transact(&request, NULL);
3578 /* The values in psched are not individually very meaningful, but they are
3579 * important. The tables below show some values seen in the wild.
3583 * - "c" has always been a constant 1000000 since at least Linux 2.4.14.
3584 * (Before that, there are hints that it was 1000000000.)
3586 * - "d" can be unrealistically large, see the comment on 'buffer_hz'
3590 * -----------------------------------
3591 * [1] 000c8000 000f4240 000f4240 00000064
3592 * [2] 000003e8 00000400 000f4240 3b9aca00
3593 * [3] 000003e8 00000400 000f4240 3b9aca00
3594 * [4] 000003e8 00000400 000f4240 00000064
3595 * [5] 000003e8 00000040 000f4240 3b9aca00
3596 * [6] 000003e8 00000040 000f4240 000000f9
3598 * a b c d ticks_per_s buffer_hz
3599 * ------- --------- ---------- ------------- ----------- -------------
3600 * [1] 819,200 1,000,000 1,000,000 100 819,200 100
3601 * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
3602 * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
3603 * [4] 1,000 1,024 1,000,000 100 976,562 100
3604 * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000
3605 * [6] 1,000 64 1,000,000 249 15,625,000 249
3607 * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p
3608 * [2] 2.6.26-1-686-bigmem from Debian lenny
3609 * [3] 2.6.26-2-sparc64 from Debian lenny
3610 * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p
3611 * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion
3612 * [6] 2.6.34 from kernel.org on KVM
3614 static const char fn[] = "/proc/net/psched";
3615 unsigned int a, b, c, d;
3621 stream = fopen(fn, "r");
3623 VLOG_WARN("%s: open failed: %s", fn, strerror(errno));
3627 if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) {
3628 VLOG_WARN("%s: read failed", fn);
3632 VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d);
3636 VLOG_WARN("%s: invalid scheduler parameters", fn);
3640 ticks_per_s = (double) a * c / b;
3644 VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u",
3647 VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz);
3650 /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a
3651 * rate of 'rate' bytes per second. */
3653 tc_ticks_to_bytes(unsigned int rate, unsigned int ticks)
3658 return (rate * ticks) / ticks_per_s;
3661 /* Returns the number of ticks that it would take to transmit 'size' bytes at a
3662 * rate of 'rate' bytes per second. */
3664 tc_bytes_to_ticks(unsigned int rate, unsigned int size)
3669 return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0;
3672 /* Returns the number of bytes that need to be reserved for qdisc buffering at
3673 * a transmission rate of 'rate' bytes per second. */
3675 tc_buffer_per_jiffy(unsigned int rate)
3680 return rate / buffer_hz;
3683 /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc,
3684 * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull,
3685 * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or
3686 * stores NULL into it if it is absent.
3688 * '*kind' and '*options' point into 'msg', so they are owned by whoever owns
3691 * Returns 0 if successful, otherwise a positive errno value. */
3693 tc_parse_qdisc(const struct ofpbuf *msg, const char **kind,
3694 struct nlattr **options)
3696 static const struct nl_policy tca_policy[] = {
3697 [TCA_KIND] = { .type = NL_A_STRING, .optional = false },
3698 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
3700 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
3702 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
3703 tca_policy, ta, ARRAY_SIZE(ta))) {
3704 VLOG_WARN_RL(&rl, "failed to parse qdisc message");
3709 *kind = nl_attr_get_string(ta[TCA_KIND]);
3713 *options = ta[TCA_OPTIONS];
3728 /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the
3729 * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute
3730 * into '*options', and its queue statistics into '*stats'. Any of the output
3731 * arguments may be null.
3733 * Returns 0 if successful, otherwise a positive errno value. */
3735 tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep,
3736 struct nlattr **options, struct netdev_queue_stats *stats)
3738 static const struct nl_policy tca_policy[] = {
3739 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false },
3740 [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false },
3742 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
3744 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
3745 tca_policy, ta, ARRAY_SIZE(ta))) {
3746 VLOG_WARN_RL(&rl, "failed to parse class message");
3751 struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc);
3752 *handlep = tc->tcm_handle;
3756 *options = ta[TCA_OPTIONS];
3760 const struct gnet_stats_queue *gsq;
3761 struct gnet_stats_basic gsb;
3763 static const struct nl_policy stats_policy[] = {
3764 [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false,
3765 .min_len = sizeof gsb },
3766 [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false,
3767 .min_len = sizeof *gsq },
3769 struct nlattr *sa[ARRAY_SIZE(stats_policy)];
3771 if (!nl_parse_nested(ta[TCA_STATS2], stats_policy,
3772 sa, ARRAY_SIZE(sa))) {
3773 VLOG_WARN_RL(&rl, "failed to parse class stats");
3777 /* Alignment issues screw up the length of struct gnet_stats_basic on
3778 * some arch/bitsize combinations. Newer versions of Linux have a
3779 * struct gnet_stats_basic_packed, but we can't depend on that. The
3780 * easiest thing to do is just to make a copy. */
3781 memset(&gsb, 0, sizeof gsb);
3782 memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]),
3783 MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb));
3784 stats->tx_bytes = gsb.bytes;
3785 stats->tx_packets = gsb.packets;
3787 gsq = nl_attr_get(sa[TCA_STATS_QUEUE]);
3788 stats->tx_errors = gsq->drops;
3798 memset(stats, 0, sizeof *stats);
3803 /* Queries the kernel for class with identifier 'handle' and parent 'parent'
3806 tc_query_class(const struct netdev *netdev,
3807 unsigned int handle, unsigned int parent,
3808 struct ofpbuf **replyp)
3810 struct ofpbuf request;
3811 struct tcmsg *tcmsg;
3814 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request);
3818 tcmsg->tcm_handle = handle;
3819 tcmsg->tcm_parent = parent;
3821 error = tc_transact(&request, replyp);
3823 VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)",
3824 netdev_get_name(netdev),
3825 tc_get_major(handle), tc_get_minor(handle),
3826 tc_get_major(parent), tc_get_minor(parent),
3832 /* Equivalent to "tc class del dev <name> handle <handle>". */
3834 tc_delete_class(const struct netdev *netdev, unsigned int handle)
3836 struct ofpbuf request;
3837 struct tcmsg *tcmsg;
3840 tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request);
3844 tcmsg->tcm_handle = handle;
3845 tcmsg->tcm_parent = 0;
3847 error = tc_transact(&request, NULL);
3849 VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)",
3850 netdev_get_name(netdev),
3851 tc_get_major(handle), tc_get_minor(handle),
3857 /* Equivalent to "tc qdisc del dev <name> root". */
3859 tc_del_qdisc(struct netdev *netdev)
3861 struct netdev_dev_linux *netdev_dev =
3862 netdev_dev_linux_cast(netdev_get_dev(netdev));
3863 struct ofpbuf request;
3864 struct tcmsg *tcmsg;
3867 tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request);
3871 tcmsg->tcm_handle = tc_make_handle(1, 0);
3872 tcmsg->tcm_parent = TC_H_ROOT;
3874 error = tc_transact(&request, NULL);
3875 if (error == EINVAL) {
3876 /* EINVAL probably means that the default qdisc was in use, in which
3877 * case we've accomplished our purpose. */
3880 if (!error && netdev_dev->tc) {
3881 if (netdev_dev->tc->ops->tc_destroy) {
3882 netdev_dev->tc->ops->tc_destroy(netdev_dev->tc);
3884 netdev_dev->tc = NULL;
3889 /* If 'netdev''s qdisc type and parameters are not yet known, queries the
3890 * kernel to determine what they are. Returns 0 if successful, otherwise a
3891 * positive errno value. */
3893 tc_query_qdisc(const struct netdev *netdev)
3895 struct netdev_dev_linux *netdev_dev =
3896 netdev_dev_linux_cast(netdev_get_dev(netdev));
3897 struct ofpbuf request, *qdisc;
3898 const struct tc_ops *ops;
3899 struct tcmsg *tcmsg;
3903 if (netdev_dev->tc) {
3907 /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have
3908 * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before
3909 * 2.6.35 without that fix backported to it.
3911 * To avoid the OOPS, we must not make a request that would attempt to dump
3912 * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a
3913 * few others. There are a few ways that I can see to do this, but most of
3914 * them seem to be racy (and if you lose the race the kernel OOPSes). The
3915 * technique chosen here is to assume that any non-default qdisc that we
3916 * create will have a class with handle 1:0. The built-in qdiscs only have
3917 * a class with handle 0:0.
3919 * We could check for Linux 2.6.35+ and use a more straightforward method
3921 tcmsg = tc_make_request(netdev, RTM_GETQDISC, NLM_F_ECHO, &request);
3925 tcmsg->tcm_handle = tc_make_handle(1, 0);
3926 tcmsg->tcm_parent = 0;
3928 /* Figure out what tc class to instantiate. */
3929 error = tc_transact(&request, &qdisc);
3933 error = tc_parse_qdisc(qdisc, &kind, NULL);
3935 ops = &tc_ops_other;
3937 ops = tc_lookup_linux_name(kind);
3939 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1);
3940 VLOG_INFO_RL(&rl2, "unknown qdisc \"%s\"", kind);
3942 ops = &tc_ops_other;
3945 } else if (error == ENOENT) {
3946 /* Either it's a built-in qdisc, or it's a qdisc set up by some
3947 * other entity that doesn't have a handle 1:0. We will assume
3948 * that it's the system default qdisc. */
3949 ops = &tc_ops_default;
3952 /* Who knows? Maybe the device got deleted. */
3953 VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)",
3954 netdev_get_name(netdev), strerror(error));
3955 ops = &tc_ops_other;
3958 /* Instantiate it. */
3959 load_error = ops->tc_load((struct netdev *) netdev, qdisc);
3960 assert((load_error == 0) == (netdev_dev->tc != NULL));
3961 ofpbuf_delete(qdisc);
3963 return error ? error : load_error;
3966 /* Linux traffic control uses tables with 256 entries ("rtab" tables) to
3967 approximate the time to transmit packets of various lengths. For an MTU of
3968 256 or less, each entry is exact; for an MTU of 257 through 512, each entry
3969 represents two possible packet lengths; for a MTU of 513 through 1024, four
3970 possible lengths; and so on.
3972 Returns, for the specified 'mtu', the number of bits that packet lengths
3973 need to be shifted right to fit within such a 256-entry table. */
3975 tc_calc_cell_log(unsigned int mtu)
3980 mtu = ETH_PAYLOAD_MAX;
3982 mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN;
3984 for (cell_log = 0; mtu >= 256; cell_log++) {
3991 /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU
3994 tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu)
3996 memset(rate, 0, sizeof *rate);
3997 rate->cell_log = tc_calc_cell_log(mtu);
3998 /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */
3999 /* rate->cell_align = 0; */ /* distro headers. */
4000 rate->mpu = ETH_TOTAL_MIN;
4004 /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink
4005 * attribute of the specified "type".
4007 * See tc_calc_cell_log() above for a description of "rtab"s. */
4009 tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate)
4014 rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE);
4015 for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) {
4016 unsigned packet_size = (i + 1) << rate->cell_log;
4017 if (packet_size < rate->mpu) {
4018 packet_size = rate->mpu;
4020 rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size);
4024 /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a
4025 * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested
4026 * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of
4029 tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes)
4031 unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu;
4032 return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst));
4035 /* Linux-only functions declared in netdev-linux.h */
4037 /* Returns a fd for an AF_INET socket or a negative errno value. */
4039 netdev_linux_get_af_inet_sock(void)
4041 int error = netdev_linux_init();
4042 return error ? -error : af_inet_sock;
4045 /* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If
4046 * 'enable' is true, the bit is set. Otherwise, it is cleared. */
4048 netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag,
4049 const char *flag_name, bool enable)
4051 const char *netdev_name = netdev_get_name(netdev);
4052 struct ethtool_value evalue;
4056 memset(&evalue, 0, sizeof evalue);
4057 error = netdev_linux_do_ethtool(netdev_name,
4058 (struct ethtool_cmd *)&evalue,
4059 ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4064 evalue.data = new_flags = (evalue.data & ~flag) | (enable ? flag : 0);
4065 error = netdev_linux_do_ethtool(netdev_name,
4066 (struct ethtool_cmd *)&evalue,
4067 ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS");
4072 memset(&evalue, 0, sizeof evalue);
4073 error = netdev_linux_do_ethtool(netdev_name,
4074 (struct ethtool_cmd *)&evalue,
4075 ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4080 if (new_flags != evalue.data) {
4081 VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network "
4082 "device %s failed", enable ? "enable" : "disable",
4083 flag_name, netdev_name);
4090 /* Utility functions. */
4092 /* Copies 'src' into 'dst', performing format conversion in the process. */
4094 netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst,
4095 const struct rtnl_link_stats *src)
4097 dst->rx_packets = src->rx_packets;
4098 dst->tx_packets = src->tx_packets;
4099 dst->rx_bytes = src->rx_bytes;
4100 dst->tx_bytes = src->tx_bytes;
4101 dst->rx_errors = src->rx_errors;
4102 dst->tx_errors = src->tx_errors;
4103 dst->rx_dropped = src->rx_dropped;
4104 dst->tx_dropped = src->tx_dropped;
4105 dst->multicast = src->multicast;
4106 dst->collisions = src->collisions;
4107 dst->rx_length_errors = src->rx_length_errors;
4108 dst->rx_over_errors = src->rx_over_errors;
4109 dst->rx_crc_errors = src->rx_crc_errors;
4110 dst->rx_frame_errors = src->rx_frame_errors;
4111 dst->rx_fifo_errors = src->rx_fifo_errors;
4112 dst->rx_missed_errors = src->rx_missed_errors;
4113 dst->tx_aborted_errors = src->tx_aborted_errors;
4114 dst->tx_carrier_errors = src->tx_carrier_errors;
4115 dst->tx_fifo_errors = src->tx_fifo_errors;
4116 dst->tx_heartbeat_errors = src->tx_heartbeat_errors;
4117 dst->tx_window_errors = src->tx_window_errors;
4121 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
4123 /* Policy for RTNLGRP_LINK messages.
4125 * There are *many* more fields in these messages, but currently we only
4126 * care about these fields. */
4127 static const struct nl_policy rtnlgrp_link_policy[] = {
4128 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
4129 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
4130 .min_len = sizeof(struct rtnl_link_stats) },
4133 struct ofpbuf request;
4134 struct ofpbuf *reply;
4135 struct ifinfomsg *ifi;
4136 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
4139 ofpbuf_init(&request, 0);
4140 nl_msg_put_nlmsghdr(&request, sizeof *ifi, RTM_GETLINK, NLM_F_REQUEST);
4141 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
4142 ifi->ifi_family = PF_UNSPEC;
4143 ifi->ifi_index = ifindex;
4144 error = nl_sock_transact(rtnl_sock, &request, &reply);
4145 ofpbuf_uninit(&request);
4150 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
4151 rtnlgrp_link_policy,
4152 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
4153 ofpbuf_delete(reply);
4157 if (!attrs[IFLA_STATS]) {
4158 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
4159 ofpbuf_delete(reply);
4163 netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(attrs[IFLA_STATS]));
4165 ofpbuf_delete(reply);
4171 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
4173 static const char fn[] = "/proc/net/dev";
4178 stream = fopen(fn, "r");
4180 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
4185 while (fgets(line, sizeof line, stream)) {
4188 #define X64 "%"SCNu64
4191 X64 X64 X64 X64 X64 X64 X64 "%*u"
4192 X64 X64 X64 X64 X64 X64 X64 "%*u",
4198 &stats->rx_fifo_errors,
4199 &stats->rx_frame_errors,
4205 &stats->tx_fifo_errors,
4207 &stats->tx_carrier_errors) != 15) {
4208 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
4209 } else if (!strcmp(devname, netdev_name)) {
4210 stats->rx_length_errors = UINT64_MAX;
4211 stats->rx_over_errors = UINT64_MAX;
4212 stats->rx_crc_errors = UINT64_MAX;
4213 stats->rx_missed_errors = UINT64_MAX;
4214 stats->tx_aborted_errors = UINT64_MAX;
4215 stats->tx_heartbeat_errors = UINT64_MAX;
4216 stats->tx_window_errors = UINT64_MAX;
4222 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
4228 get_flags(const struct netdev_dev *dev, unsigned int *flags)
4234 error = netdev_linux_do_ioctl(dev->name, &ifr, SIOCGIFFLAGS,
4237 *flags = ifr.ifr_flags;
4243 set_flags(struct netdev *netdev, unsigned int flags)
4247 ifr.ifr_flags = flags;
4248 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
4253 do_get_ifindex(const char *netdev_name)
4257 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4258 COVERAGE_INC(netdev_get_ifindex);
4259 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
4260 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
4261 netdev_name, strerror(errno));
4264 return ifr.ifr_ifindex;
4268 get_ifindex(const struct netdev *netdev_, int *ifindexp)
4270 struct netdev_dev_linux *netdev_dev =
4271 netdev_dev_linux_cast(netdev_get_dev(netdev_));
4273 if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
4274 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
4278 netdev_dev->cache_valid |= VALID_IFINDEX;
4279 netdev_dev->ifindex = ifindex;
4281 *ifindexp = netdev_dev->ifindex;
4286 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
4291 memset(&ifr, 0, sizeof ifr);
4292 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4293 COVERAGE_INC(netdev_get_hwaddr);
4294 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
4295 /* ENODEV probably means that a vif disappeared asynchronously and
4296 * hasn't been removed from the database yet, so reduce the log level
4297 * to INFO for that case. */
4298 VLOG(errno == ENODEV ? VLL_INFO : VLL_ERR,
4299 "ioctl(SIOCGIFHWADDR) on %s device failed: %s",
4300 netdev_name, strerror(errno));
4303 hwaddr_family = ifr.ifr_hwaddr.sa_family;
4304 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
4305 VLOG_WARN("%s device has unknown hardware address family %d",
4306 netdev_name, hwaddr_family);
4308 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
4313 set_etheraddr(const char *netdev_name, int hwaddr_family,
4314 const uint8_t mac[ETH_ADDR_LEN])
4318 memset(&ifr, 0, sizeof ifr);
4319 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4320 ifr.ifr_hwaddr.sa_family = hwaddr_family;
4321 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
4322 COVERAGE_INC(netdev_set_hwaddr);
4323 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
4324 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
4325 netdev_name, strerror(errno));
4332 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
4333 int cmd, const char *cmd_name)
4337 memset(&ifr, 0, sizeof ifr);
4338 ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
4339 ifr.ifr_data = (caddr_t) ecmd;
4342 COVERAGE_INC(netdev_ethtool);
4343 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
4346 if (errno != EOPNOTSUPP) {
4347 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
4348 "failed: %s", cmd_name, name, strerror(errno));
4350 /* The device doesn't support this operation. That's pretty
4351 * common, so there's no point in logging anything. */
4358 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
4359 const char *cmd_name)
4361 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
4362 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
4363 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
4371 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
4372 int cmd, const char *cmd_name)
4377 ifr.ifr_addr.sa_family = AF_INET;
4378 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
4380 const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
4381 *ip = sin->sin_addr;
4386 /* Returns an AF_PACKET raw socket or a negative errno value. */
4388 af_packet_sock(void)
4390 static int sock = INT_MIN;
4392 if (sock == INT_MIN) {
4393 sock = socket(AF_PACKET, SOCK_RAW, 0);
4395 set_nonblocking(sock);
4398 VLOG_ERR("failed to create packet socket: %s", strerror(errno));