2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "netdev-linux.h"
23 #include <arpa/inet.h>
25 #include <linux/filter.h>
26 #include <linux/gen_stats.h>
27 #include <linux/if_ether.h>
28 #include <linux/if_tun.h>
29 #include <linux/types.h>
30 #include <linux/ethtool.h>
31 #include <linux/mii.h>
32 #include <linux/pkt_cls.h>
33 #include <linux/pkt_sched.h>
34 #include <linux/rtnetlink.h>
35 #include <linux/sockios.h>
36 #include <sys/types.h>
37 #include <sys/ioctl.h>
38 #include <sys/socket.h>
39 #include <netpacket/packet.h>
41 #include <net/if_arp.h>
42 #include <net/if_packet.h>
43 #include <net/route.h>
44 #include <netinet/in.h>
51 #include "dpif-linux.h"
52 #include "dpif-netdev.h"
53 #include "dynamic-string.h"
54 #include "fatal-signal.h"
57 #include "netdev-provider.h"
58 #include "netdev-vport.h"
59 #include "netlink-notifier.h"
60 #include "netlink-socket.h"
63 #include "openflow/openflow.h"
64 #include "ovs-atomic.h"
66 #include "poll-loop.h"
67 #include "rtnetlink-link.h"
69 #include "socket-util.h"
72 #include "unaligned.h"
75 VLOG_DEFINE_THIS_MODULE(netdev_linux);
77 COVERAGE_DEFINE(netdev_set_policing);
78 COVERAGE_DEFINE(netdev_arp_lookup);
79 COVERAGE_DEFINE(netdev_get_ifindex);
80 COVERAGE_DEFINE(netdev_get_hwaddr);
81 COVERAGE_DEFINE(netdev_set_hwaddr);
82 COVERAGE_DEFINE(netdev_get_ethtool);
83 COVERAGE_DEFINE(netdev_set_ethtool);
86 /* These were introduced in Linux 2.6.14, so they might be missing if we have
88 #ifndef ADVERTISED_Pause
89 #define ADVERTISED_Pause (1 << 13)
91 #ifndef ADVERTISED_Asym_Pause
92 #define ADVERTISED_Asym_Pause (1 << 14)
95 /* These were introduced in Linux 2.6.24, so they might be missing if we
96 * have old headers. */
97 #ifndef ETHTOOL_GFLAGS
98 #define ETHTOOL_GFLAGS 0x00000025 /* Get flags bitmap(ethtool_value) */
100 #ifndef ETHTOOL_SFLAGS
101 #define ETHTOOL_SFLAGS 0x00000026 /* Set flags bitmap(ethtool_value) */
104 /* This was introduced in Linux 2.6.25, so it might be missing if we have old
107 #define TC_RTAB_SIZE 1024
110 /* Linux 2.6.21 introduced struct tpacket_auxdata.
111 * Linux 2.6.27 added the tp_vlan_tci member.
112 * Linux 3.0 defined TP_STATUS_VLAN_VALID.
113 * Linux 3.13 repurposed a padding member for tp_vlan_tpid and defined
114 * TP_STATUS_VLAN_TPID_VALID.
116 * With all this churn it's easiest to unconditionally define a replacement
117 * structure that has everything we want.
119 #ifndef PACKET_AUXDATA
120 #define PACKET_AUXDATA 8
122 #ifndef TP_STATUS_VLAN_VALID
123 #define TP_STATUS_VLAN_VALID (1 << 4)
125 #ifndef TP_STATUS_VLAN_TPID_VALID
126 #define TP_STATUS_VLAN_TPID_VALID (1 << 6)
128 #undef tpacket_auxdata
129 #define tpacket_auxdata rpl_tpacket_auxdata
130 struct tpacket_auxdata {
136 uint16_t tp_vlan_tci;
137 uint16_t tp_vlan_tpid;
141 VALID_IFINDEX = 1 << 0,
142 VALID_ETHERADDR = 1 << 1,
146 VALID_POLICING = 1 << 5,
147 VALID_VPORT_STAT_ERROR = 1 << 6,
148 VALID_DRVINFO = 1 << 7,
149 VALID_FEATURES = 1 << 8,
152 /* Traffic control. */
154 /* An instance of a traffic control class. Always associated with a particular
157 * Each TC implementation subclasses this with whatever additional data it
160 const struct tc_ops *ops;
161 struct hmap queues; /* Contains "struct tc_queue"s.
162 * Read by generic TC layer.
163 * Written only by TC implementation. */
166 #define TC_INITIALIZER(TC, OPS) { OPS, HMAP_INITIALIZER(&(TC)->queues) }
168 /* One traffic control queue.
170 * Each TC implementation subclasses this with whatever additional data it
173 struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */
174 unsigned int queue_id; /* OpenFlow queue ID. */
175 long long int created; /* Time queue was created, in msecs. */
178 /* A particular kind of traffic control. Each implementation generally maps to
179 * one particular Linux qdisc class.
181 * The functions below return 0 if successful or a positive errno value on
182 * failure, except where otherwise noted. All of them must be provided, except
183 * where otherwise noted. */
185 /* Name used by kernel in the TCA_KIND attribute of tcmsg, e.g. "htb".
186 * This is null for tc_ops_default and tc_ops_other, for which there are no
187 * appropriate values. */
188 const char *linux_name;
190 /* Name used in OVS database, e.g. "linux-htb". Must be nonnull. */
191 const char *ovs_name;
193 /* Number of supported OpenFlow queues, 0 for qdiscs that have no
194 * queues. The queues are numbered 0 through n_queues - 1. */
195 unsigned int n_queues;
197 /* Called to install this TC class on 'netdev'. The implementation should
198 * make the Netlink calls required to set up 'netdev' with the right qdisc
199 * and configure it according to 'details'. The implementation may assume
200 * that the current qdisc is the default; that is, there is no need for it
201 * to delete the current qdisc before installing itself.
203 * The contents of 'details' should be documented as valid for 'ovs_name'
204 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
205 * (which is built as ovs-vswitchd.conf.db(8)).
207 * This function must return 0 if and only if it sets 'netdev->tc' to an
208 * initialized 'struct tc'.
210 * (This function is null for tc_ops_other, which cannot be installed. For
211 * other TC classes it should always be nonnull.) */
212 int (*tc_install)(struct netdev *netdev, const struct smap *details);
214 /* Called when the netdev code determines (through a Netlink query) that
215 * this TC class's qdisc is installed on 'netdev', but we didn't install
216 * it ourselves and so don't know any of the details.
218 * 'nlmsg' is the kernel reply to a RTM_GETQDISC Netlink message for
219 * 'netdev'. The TCA_KIND attribute of 'nlmsg' is 'linux_name'. The
220 * implementation should parse the other attributes of 'nlmsg' as
221 * necessary to determine its configuration. If necessary it should also
222 * use Netlink queries to determine the configuration of queues on
225 * This function must return 0 if and only if it sets 'netdev->tc' to an
226 * initialized 'struct tc'. */
227 int (*tc_load)(struct netdev *netdev, struct ofpbuf *nlmsg);
229 /* Destroys the data structures allocated by the implementation as part of
230 * 'tc'. (This includes destroying 'tc->queues' by calling
233 * The implementation should not need to perform any Netlink calls. If
234 * desirable, the caller is responsible for deconfiguring the kernel qdisc.
235 * (But it may not be desirable.)
237 * This function may be null if 'tc' is trivial. */
238 void (*tc_destroy)(struct tc *tc);
240 /* Retrieves details of 'netdev->tc' configuration into 'details'.
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 configuration.
246 * The contents of 'details' should be documented as valid for 'ovs_name'
247 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
248 * (which is built as ovs-vswitchd.conf.db(8)).
250 * This function may be null if 'tc' is not configurable.
252 int (*qdisc_get)(const struct netdev *netdev, struct smap *details);
254 /* Reconfigures 'netdev->tc' according to 'details', performing any
255 * required Netlink calls to complete the reconfiguration.
257 * The contents of 'details' should be documented as valid for 'ovs_name'
258 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
259 * (which is built as ovs-vswitchd.conf.db(8)).
261 * This function may be null if 'tc' is not configurable.
263 int (*qdisc_set)(struct netdev *, const struct smap *details);
265 /* Retrieves details of 'queue' on 'netdev->tc' into 'details'. 'queue' is
266 * one of the 'struct tc_queue's within 'netdev->tc->queues'.
268 * The contents of 'details' should be documented as valid for 'ovs_name'
269 * in the "other_config" column in the "Queue" table in
270 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
272 * The implementation should not need to perform any Netlink calls, because
273 * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have
274 * cached the queue configuration.
276 * This function may be null if 'tc' does not have queues ('n_queues' is
278 int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue,
279 struct smap *details);
281 /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to
282 * 'details', perfoming any required Netlink calls to complete the
283 * reconfiguration. The caller ensures that 'queue_id' is less than
286 * The contents of 'details' should be documented as valid for 'ovs_name'
287 * in the "other_config" column in the "Queue" table in
288 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
290 * This function may be null if 'tc' does not have queues or its queues are
291 * not configurable. */
292 int (*class_set)(struct netdev *, unsigned int queue_id,
293 const struct smap *details);
295 /* Deletes 'queue' from 'netdev->tc'. 'queue' is one of the 'struct
296 * tc_queue's within 'netdev->tc->queues'.
298 * This function may be null if 'tc' does not have queues or its queues
299 * cannot be deleted. */
300 int (*class_delete)(struct netdev *, struct tc_queue *queue);
302 /* Obtains stats for 'queue' from 'netdev->tc'. 'queue' is one of the
303 * 'struct tc_queue's within 'netdev->tc->queues'.
305 * On success, initializes '*stats'.
307 * This function may be null if 'tc' does not have queues or if it cannot
308 * report queue statistics. */
309 int (*class_get_stats)(const struct netdev *netdev,
310 const struct tc_queue *queue,
311 struct netdev_queue_stats *stats);
313 /* Extracts queue stats from 'nlmsg', which is a response to a
314 * RTM_GETTCLASS message, and passes them to 'cb' along with 'aux'.
316 * This function may be null if 'tc' does not have queues or if it cannot
317 * report queue statistics. */
318 int (*class_dump_stats)(const struct netdev *netdev,
319 const struct ofpbuf *nlmsg,
320 netdev_dump_queue_stats_cb *cb, void *aux);
324 tc_init(struct tc *tc, const struct tc_ops *ops)
327 hmap_init(&tc->queues);
331 tc_destroy(struct tc *tc)
333 hmap_destroy(&tc->queues);
336 static const struct tc_ops tc_ops_htb;
337 static const struct tc_ops tc_ops_hfsc;
338 static const struct tc_ops tc_ops_default;
339 static const struct tc_ops tc_ops_other;
341 static const struct tc_ops *const tcs[] = {
342 &tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */
343 &tc_ops_hfsc, /* Hierarchical fair service curve. */
344 &tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */
345 &tc_ops_other, /* Some other qdisc. */
349 static unsigned int tc_make_handle(unsigned int major, unsigned int minor);
350 static unsigned int tc_get_major(unsigned int handle);
351 static unsigned int tc_get_minor(unsigned int handle);
353 static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks);
354 static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size);
355 static unsigned int tc_buffer_per_jiffy(unsigned int rate);
357 static struct tcmsg *tc_make_request(const struct netdev *, int type,
358 unsigned int flags, struct ofpbuf *);
359 static int tc_transact(struct ofpbuf *request, struct ofpbuf **replyp);
360 static int tc_add_del_ingress_qdisc(struct netdev *netdev, bool add);
361 static int tc_add_policer(struct netdev *netdev, int kbits_rate,
364 static int tc_parse_qdisc(const struct ofpbuf *, const char **kind,
365 struct nlattr **options);
366 static int tc_parse_class(const struct ofpbuf *, unsigned int *queue_id,
367 struct nlattr **options,
368 struct netdev_queue_stats *);
369 static int tc_query_class(const struct netdev *,
370 unsigned int handle, unsigned int parent,
371 struct ofpbuf **replyp);
372 static int tc_delete_class(const struct netdev *, unsigned int handle);
374 static int tc_del_qdisc(struct netdev *netdev);
375 static int tc_query_qdisc(const struct netdev *netdev);
377 static int tc_calc_cell_log(unsigned int mtu);
378 static void tc_fill_rate(struct tc_ratespec *rate, uint64_t bps, int mtu);
379 static void tc_put_rtab(struct ofpbuf *, uint16_t type,
380 const struct tc_ratespec *rate);
381 static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes);
383 struct netdev_linux {
386 /* Protects all members below. */
387 struct ovs_mutex mutex;
389 unsigned int cache_valid;
391 bool miimon; /* Link status of last poll. */
392 long long int miimon_interval; /* Miimon Poll rate. Disabled if <= 0. */
393 struct timer miimon_timer;
395 /* The following are figured out "on demand" only. They are only valid
396 * when the corresponding VALID_* bit in 'cache_valid' is set. */
398 uint8_t etheraddr[ETH_ADDR_LEN];
399 struct in_addr address, netmask;
402 unsigned int ifi_flags;
403 long long int carrier_resets;
404 uint32_t kbits_rate; /* Policing data. */
405 uint32_t kbits_burst;
406 int vport_stats_error; /* Cached error code from vport_get_stats().
407 0 or an errno value. */
408 int netdev_mtu_error; /* Cached error code from SIOCGIFMTU or SIOCSIFMTU. */
409 int ether_addr_error; /* Cached error code from set/get etheraddr. */
410 int netdev_policing_error; /* Cached error code from set policing. */
411 int get_features_error; /* Cached error code from ETHTOOL_GSET. */
412 int get_ifindex_error; /* Cached error code from SIOCGIFINDEX. */
414 enum netdev_features current; /* Cached from ETHTOOL_GSET. */
415 enum netdev_features advertised; /* Cached from ETHTOOL_GSET. */
416 enum netdev_features supported; /* Cached from ETHTOOL_GSET. */
418 struct ethtool_drvinfo drvinfo; /* Cached from ETHTOOL_GDRVINFO. */
421 /* For devices of class netdev_tap_class only. */
425 struct netdev_rxq_linux {
426 struct netdev_rxq up;
431 /* This is set pretty low because we probably won't learn anything from the
432 * additional log messages. */
433 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
435 /* Polling miimon status for all ports causes performance degradation when
436 * handling a large number of ports. If there are no devices using miimon, then
437 * we skip netdev_linux_miimon_run() and netdev_linux_miimon_wait(). */
438 static atomic_int miimon_cnt = ATOMIC_VAR_INIT(0);
440 static void netdev_linux_run(void);
442 static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
443 int cmd, const char *cmd_name);
444 static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
445 int cmd, const char *cmd_name);
446 static int get_flags(const struct netdev *, unsigned int *flags);
447 static int set_flags(const char *, unsigned int flags);
448 static int update_flags(struct netdev_linux *netdev, enum netdev_flags off,
449 enum netdev_flags on, enum netdev_flags *old_flagsp)
450 OVS_REQUIRES(netdev->mutex);
451 static int do_get_ifindex(const char *netdev_name);
452 static int get_ifindex(const struct netdev *, int *ifindexp);
453 static int do_set_addr(struct netdev *netdev,
454 int ioctl_nr, const char *ioctl_name,
455 struct in_addr addr);
456 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
457 static int set_etheraddr(const char *netdev_name, const uint8_t[ETH_ADDR_LEN]);
458 static int get_stats_via_netlink(const struct netdev *, struct netdev_stats *);
459 static int af_packet_sock(void);
460 static bool netdev_linux_miimon_enabled(void);
461 static void netdev_linux_miimon_run(void);
462 static void netdev_linux_miimon_wait(void);
463 static int netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup);
466 is_netdev_linux_class(const struct netdev_class *netdev_class)
468 return netdev_class->run == netdev_linux_run;
472 is_tap_netdev(const struct netdev *netdev)
474 return netdev_get_class(netdev) == &netdev_tap_class;
477 static struct netdev_linux *
478 netdev_linux_cast(const struct netdev *netdev)
480 ovs_assert(is_netdev_linux_class(netdev_get_class(netdev)));
482 return CONTAINER_OF(netdev, struct netdev_linux, up);
485 static struct netdev_rxq_linux *
486 netdev_rxq_linux_cast(const struct netdev_rxq *rx)
488 ovs_assert(is_netdev_linux_class(netdev_get_class(rx->netdev)));
489 return CONTAINER_OF(rx, struct netdev_rxq_linux, up);
492 static void netdev_linux_update(struct netdev_linux *netdev,
493 const struct rtnetlink_link_change *)
494 OVS_REQUIRES(netdev->mutex);
495 static void netdev_linux_changed(struct netdev_linux *netdev,
496 unsigned int ifi_flags, unsigned int mask)
497 OVS_REQUIRES(netdev->mutex);
499 /* Returns a NETLINK_ROUTE socket listening for RTNLGRP_LINK changes, or NULL
500 * if no such socket could be created. */
501 static struct nl_sock *
502 netdev_linux_notify_sock(void)
504 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
505 static struct nl_sock *sock;
507 if (ovsthread_once_start(&once)) {
510 error = nl_sock_create(NETLINK_ROUTE, &sock);
512 error = nl_sock_join_mcgroup(sock, RTNLGRP_LINK);
514 nl_sock_destroy(sock);
518 ovsthread_once_done(&once);
525 netdev_linux_miimon_enabled(void)
529 atomic_read(&miimon_cnt, &miimon);
534 netdev_linux_run(void)
536 struct nl_sock *sock;
539 if (netdev_linux_miimon_enabled()) {
540 netdev_linux_miimon_run();
543 sock = netdev_linux_notify_sock();
549 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
550 uint64_t buf_stub[4096 / 8];
553 ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub);
554 error = nl_sock_recv(sock, &buf, false);
556 struct rtnetlink_link_change change;
558 if (rtnetlink_link_parse(&buf, &change)) {
559 struct netdev *netdev_ = netdev_from_name(change.ifname);
560 if (netdev_ && is_netdev_linux_class(netdev_->netdev_class)) {
561 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
563 ovs_mutex_lock(&netdev->mutex);
564 netdev_linux_update(netdev, &change);
565 ovs_mutex_unlock(&netdev->mutex);
567 netdev_close(netdev_);
569 } else if (error == ENOBUFS) {
570 struct shash device_shash;
571 struct shash_node *node;
575 shash_init(&device_shash);
576 netdev_get_devices(&netdev_linux_class, &device_shash);
577 SHASH_FOR_EACH (node, &device_shash) {
578 struct netdev *netdev_ = node->data;
579 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
582 ovs_mutex_lock(&netdev->mutex);
583 get_flags(netdev_, &flags);
584 netdev_linux_changed(netdev, flags, 0);
585 ovs_mutex_unlock(&netdev->mutex);
587 netdev_close(netdev_);
589 shash_destroy(&device_shash);
590 } else if (error != EAGAIN) {
591 VLOG_WARN_RL(&rl, "error reading or parsing netlink (%s)",
592 ovs_strerror(error));
599 netdev_linux_wait(void)
601 struct nl_sock *sock;
603 if (netdev_linux_miimon_enabled()) {
604 netdev_linux_miimon_wait();
606 sock = netdev_linux_notify_sock();
608 nl_sock_wait(sock, POLLIN);
613 netdev_linux_changed(struct netdev_linux *dev,
614 unsigned int ifi_flags, unsigned int mask)
615 OVS_REQUIRES(dev->mutex)
617 netdev_change_seq_changed(&dev->up);
619 if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) {
620 dev->carrier_resets++;
622 dev->ifi_flags = ifi_flags;
624 dev->cache_valid &= mask;
628 netdev_linux_update(struct netdev_linux *dev,
629 const struct rtnetlink_link_change *change)
630 OVS_REQUIRES(dev->mutex)
632 if (change->nlmsg_type == RTM_NEWLINK) {
634 netdev_linux_changed(dev, change->ifi_flags, VALID_DRVINFO);
636 /* Update netdev from rtnl-change msg. */
638 dev->mtu = change->mtu;
639 dev->cache_valid |= VALID_MTU;
640 dev->netdev_mtu_error = 0;
643 if (!eth_addr_is_zero(change->addr)) {
644 memcpy(dev->etheraddr, change->addr, ETH_ADDR_LEN);
645 dev->cache_valid |= VALID_ETHERADDR;
646 dev->ether_addr_error = 0;
649 dev->ifindex = change->ifi_index;
650 dev->cache_valid |= VALID_IFINDEX;
651 dev->get_ifindex_error = 0;
654 netdev_linux_changed(dev, change->ifi_flags, 0);
658 static struct netdev *
659 netdev_linux_alloc(void)
661 struct netdev_linux *netdev = xzalloc(sizeof *netdev);
666 netdev_linux_common_construct(struct netdev_linux *netdev)
668 ovs_mutex_init(&netdev->mutex);
671 /* Creates system and internal devices. */
673 netdev_linux_construct(struct netdev *netdev_)
675 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
678 netdev_linux_common_construct(netdev);
680 error = get_flags(&netdev->up, &netdev->ifi_flags);
681 if (error == ENODEV) {
682 if (netdev->up.netdev_class != &netdev_internal_class) {
683 /* The device does not exist, so don't allow it to be opened. */
686 /* "Internal" netdevs have to be created as netdev objects before
687 * they exist in the kernel, because creating them in the kernel
688 * happens by passing a netdev object to dpif_port_add().
689 * Therefore, ignore the error. */
696 /* For most types of netdevs we open the device for each call of
697 * netdev_open(). However, this is not the case with tap devices,
698 * since it is only possible to open the device once. In this
699 * situation we share a single file descriptor, and consequently
700 * buffers, across all readers. Therefore once data is read it will
701 * be unavailable to other reads for tap devices. */
703 netdev_linux_construct_tap(struct netdev *netdev_)
705 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
706 static const char tap_dev[] = "/dev/net/tun";
707 const char *name = netdev_->name;
711 netdev_linux_common_construct(netdev);
713 /* Open tap device. */
714 netdev->tap_fd = open(tap_dev, O_RDWR);
715 if (netdev->tap_fd < 0) {
717 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, ovs_strerror(error));
721 /* Create tap device. */
722 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
723 ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
724 if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) {
725 VLOG_WARN("%s: creating tap device failed: %s", name,
726 ovs_strerror(errno));
731 /* Make non-blocking. */
732 error = set_nonblocking(netdev->tap_fd);
740 close(netdev->tap_fd);
745 netdev_linux_destruct(struct netdev *netdev_)
747 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
749 if (netdev->tc && netdev->tc->ops->tc_destroy) {
750 netdev->tc->ops->tc_destroy(netdev->tc);
753 if (netdev_get_class(netdev_) == &netdev_tap_class
754 && netdev->tap_fd >= 0)
756 close(netdev->tap_fd);
759 if (netdev->miimon_interval > 0) {
761 atomic_sub(&miimon_cnt, 1, &junk);
764 ovs_mutex_destroy(&netdev->mutex);
768 netdev_linux_dealloc(struct netdev *netdev_)
770 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
774 static struct netdev_rxq *
775 netdev_linux_rxq_alloc(void)
777 struct netdev_rxq_linux *rx = xzalloc(sizeof *rx);
782 netdev_linux_rxq_construct(struct netdev_rxq *rxq_)
784 struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
785 struct netdev *netdev_ = rx->up.netdev;
786 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
789 ovs_mutex_lock(&netdev->mutex);
790 rx->is_tap = is_tap_netdev(netdev_);
792 rx->fd = netdev->tap_fd;
794 struct sockaddr_ll sll;
796 /* Result of tcpdump -dd inbound */
797 static const struct sock_filter filt[] = {
798 { 0x28, 0, 0, 0xfffff004 }, /* ldh [0] */
799 { 0x15, 0, 1, 0x00000004 }, /* jeq #4 jt 2 jf 3 */
800 { 0x6, 0, 0, 0x00000000 }, /* ret #0 */
801 { 0x6, 0, 0, 0x0000ffff } /* ret #65535 */
803 static const struct sock_fprog fprog = {
804 ARRAY_SIZE(filt), (struct sock_filter *) filt
807 /* Create file descriptor. */
808 rx->fd = socket(PF_PACKET, SOCK_RAW, 0);
811 VLOG_ERR("failed to create raw socket (%s)", ovs_strerror(error));
816 if (setsockopt(rx->fd, SOL_PACKET, PACKET_AUXDATA, &val, sizeof val)) {
818 VLOG_ERR("%s: failed to mark socket for auxdata (%s)",
819 netdev_get_name(netdev_), ovs_strerror(error));
823 /* Set non-blocking mode. */
824 error = set_nonblocking(rx->fd);
829 /* Get ethernet device index. */
830 error = get_ifindex(&netdev->up, &ifindex);
835 /* Bind to specific ethernet device. */
836 memset(&sll, 0, sizeof sll);
837 sll.sll_family = AF_PACKET;
838 sll.sll_ifindex = ifindex;
839 sll.sll_protocol = htons(ETH_P_ALL);
840 if (bind(rx->fd, (struct sockaddr *) &sll, sizeof sll) < 0) {
842 VLOG_ERR("%s: failed to bind raw socket (%s)",
843 netdev_get_name(netdev_), ovs_strerror(error));
847 /* Filter for only inbound packets. */
848 error = setsockopt(rx->fd, SOL_SOCKET, SO_ATTACH_FILTER, &fprog,
852 VLOG_ERR("%s: failed to attach filter (%s)",
853 netdev_get_name(netdev_), ovs_strerror(error));
857 ovs_mutex_unlock(&netdev->mutex);
865 ovs_mutex_unlock(&netdev->mutex);
870 netdev_linux_rxq_destruct(struct netdev_rxq *rxq_)
872 struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
880 netdev_linux_rxq_dealloc(struct netdev_rxq *rxq_)
882 struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
888 auxdata_to_vlan_tpid(const struct tpacket_auxdata *aux)
890 if (aux->tp_status & TP_STATUS_VLAN_TPID_VALID) {
891 return htons(aux->tp_vlan_tpid);
893 return htons(ETH_TYPE_VLAN);
898 auxdata_has_vlan_tci(const struct tpacket_auxdata *aux)
900 return aux->tp_vlan_tci || aux->tp_status & TP_STATUS_VLAN_VALID;
904 netdev_linux_rxq_recv_sock(int fd, struct ofpbuf *buffer)
909 struct cmsghdr *cmsg;
912 char buffer[CMSG_SPACE(sizeof(struct tpacket_auxdata))];
916 /* Reserve headroom for a single VLAN tag */
917 ofpbuf_reserve(buffer, VLAN_HEADER_LEN);
918 size = ofpbuf_tailroom(buffer);
920 iov.iov_base = ofpbuf_data(buffer);
922 msgh.msg_name = NULL;
923 msgh.msg_namelen = 0;
926 msgh.msg_control = &cmsg_buffer;
927 msgh.msg_controllen = sizeof cmsg_buffer;
931 retval = recvmsg(fd, &msgh, MSG_TRUNC);
932 } while (retval < 0 && errno == EINTR);
936 } else if (retval > size) {
940 ofpbuf_set_size(buffer, ofpbuf_size(buffer) + retval);
942 for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg; cmsg = CMSG_NXTHDR(&msgh, cmsg)) {
943 const struct tpacket_auxdata *aux;
945 if (cmsg->cmsg_level != SOL_PACKET
946 || cmsg->cmsg_type != PACKET_AUXDATA
947 || cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata))) {
951 aux = ALIGNED_CAST(struct tpacket_auxdata *, CMSG_DATA(cmsg));
952 if (auxdata_has_vlan_tci(aux)) {
953 if (retval < ETH_HEADER_LEN) {
957 eth_push_vlan(buffer, auxdata_to_vlan_tpid(aux),
958 htons(aux->tp_vlan_tci));
967 netdev_linux_rxq_recv_tap(int fd, struct ofpbuf *buffer)
970 size_t size = ofpbuf_tailroom(buffer);
973 retval = read(fd, ofpbuf_data(buffer), size);
974 } while (retval < 0 && errno == EINTR);
978 } else if (retval > size) {
982 ofpbuf_set_size(buffer, ofpbuf_size(buffer) + retval);
987 netdev_linux_rxq_recv(struct netdev_rxq *rxq_, struct ofpbuf **packet, int *c)
989 struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
990 struct netdev *netdev = rx->up.netdev;
991 struct ofpbuf *buffer;
995 if (netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu)) {
996 mtu = ETH_PAYLOAD_MAX;
999 buffer = ofpbuf_new_with_headroom(VLAN_ETH_HEADER_LEN + mtu, DP_NETDEV_HEADROOM);
1001 retval = (rx->is_tap
1002 ? netdev_linux_rxq_recv_tap(rx->fd, buffer)
1003 : netdev_linux_rxq_recv_sock(rx->fd, buffer));
1006 if (retval != EAGAIN && retval != EMSGSIZE) {
1007 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
1008 ovs_strerror(errno), netdev_rxq_get_name(rxq_));
1010 ofpbuf_delete(buffer);
1012 dp_packet_pad(buffer);
1021 netdev_linux_rxq_wait(struct netdev_rxq *rxq_)
1023 struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
1024 poll_fd_wait(rx->fd, POLLIN);
1028 netdev_linux_rxq_drain(struct netdev_rxq *rxq_)
1030 struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
1033 int error = af_inet_ifreq_ioctl(netdev_rxq_get_name(rxq_), &ifr,
1034 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
1038 drain_fd(rx->fd, ifr.ifr_qlen);
1041 return drain_rcvbuf(rx->fd);
1045 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
1046 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
1047 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
1048 * the packet is too big or too small to transmit on the device.
1050 * The caller retains ownership of 'buffer' in all cases.
1052 * The kernel maintains a packet transmission queue, so the caller is not
1053 * expected to do additional queuing of packets. */
1055 netdev_linux_send(struct netdev *netdev_, struct ofpbuf *pkt, bool may_steal)
1057 const void *data = ofpbuf_data(pkt);
1058 size_t size = ofpbuf_size(pkt);
1063 if (!is_tap_netdev(netdev_)) {
1064 /* Use our AF_PACKET socket to send to this device. */
1065 struct sockaddr_ll sll;
1071 sock = af_packet_sock();
1076 ifindex = netdev_get_ifindex(netdev_);
1081 /* We don't bother setting most fields in sockaddr_ll because the
1082 * kernel ignores them for SOCK_RAW. */
1083 memset(&sll, 0, sizeof sll);
1084 sll.sll_family = AF_PACKET;
1085 sll.sll_ifindex = ifindex;
1087 iov.iov_base = CONST_CAST(void *, data);
1090 msg.msg_name = &sll;
1091 msg.msg_namelen = sizeof sll;
1094 msg.msg_control = NULL;
1095 msg.msg_controllen = 0;
1098 retval = sendmsg(sock, &msg, 0);
1100 /* Use the tap fd to send to this device. This is essential for
1101 * tap devices, because packets sent to a tap device with an
1102 * AF_PACKET socket will loop back to be *received* again on the
1103 * tap device. This doesn't occur on other interface types
1104 * because we attach a socket filter to the rx socket. */
1105 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1107 retval = write(netdev->tap_fd, data, size);
1115 /* The Linux AF_PACKET implementation never blocks waiting for room
1116 * for packets, instead returning ENOBUFS. Translate this into
1117 * EAGAIN for the caller. */
1118 if (errno == ENOBUFS) {
1120 } else if (errno == EINTR) {
1122 } else if (errno != EAGAIN) {
1123 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
1124 netdev_get_name(netdev_), ovs_strerror(errno));
1127 } else if (retval != size) {
1128 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%"PRIuSIZE" bytes of "
1129 "%"PRIuSIZE") on %s", retval, size, netdev_get_name(netdev_));
1137 /* Registers with the poll loop to wake up from the next call to poll_block()
1138 * when the packet transmission queue has sufficient room to transmit a packet
1139 * with netdev_send().
1141 * The kernel maintains a packet transmission queue, so the client is not
1142 * expected to do additional queuing of packets. Thus, this function is
1143 * unlikely to ever be used. It is included for completeness. */
1145 netdev_linux_send_wait(struct netdev *netdev)
1147 if (is_tap_netdev(netdev)) {
1148 /* TAP device always accepts packets.*/
1149 poll_immediate_wake();
1153 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
1154 * otherwise a positive errno value. */
1156 netdev_linux_set_etheraddr(struct netdev *netdev_,
1157 const uint8_t mac[ETH_ADDR_LEN])
1159 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1160 enum netdev_flags old_flags = 0;
1163 ovs_mutex_lock(&netdev->mutex);
1165 if (netdev->cache_valid & VALID_ETHERADDR) {
1166 error = netdev->ether_addr_error;
1167 if (error || eth_addr_equals(netdev->etheraddr, mac)) {
1170 netdev->cache_valid &= ~VALID_ETHERADDR;
1173 /* Tap devices must be brought down before setting the address. */
1174 if (is_tap_netdev(netdev_)) {
1175 update_flags(netdev, NETDEV_UP, 0, &old_flags);
1177 error = set_etheraddr(netdev_get_name(netdev_), mac);
1178 if (!error || error == ENODEV) {
1179 netdev->ether_addr_error = error;
1180 netdev->cache_valid |= VALID_ETHERADDR;
1182 memcpy(netdev->etheraddr, mac, ETH_ADDR_LEN);
1186 if (is_tap_netdev(netdev_) && old_flags & NETDEV_UP) {
1187 update_flags(netdev, 0, NETDEV_UP, &old_flags);
1191 ovs_mutex_unlock(&netdev->mutex);
1195 /* Copies 'netdev''s MAC address to 'mac' which is passed as param. */
1197 netdev_linux_get_etheraddr(const struct netdev *netdev_,
1198 uint8_t mac[ETH_ADDR_LEN])
1200 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1203 ovs_mutex_lock(&netdev->mutex);
1204 if (!(netdev->cache_valid & VALID_ETHERADDR)) {
1205 netdev->ether_addr_error = get_etheraddr(netdev_get_name(netdev_),
1207 netdev->cache_valid |= VALID_ETHERADDR;
1210 error = netdev->ether_addr_error;
1212 memcpy(mac, netdev->etheraddr, ETH_ADDR_LEN);
1214 ovs_mutex_unlock(&netdev->mutex);
1220 netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup)
1224 if (!(netdev->cache_valid & VALID_MTU)) {
1227 netdev->netdev_mtu_error = af_inet_ifreq_ioctl(
1228 netdev_get_name(&netdev->up), &ifr, SIOCGIFMTU, "SIOCGIFMTU");
1229 netdev->mtu = ifr.ifr_mtu;
1230 netdev->cache_valid |= VALID_MTU;
1233 error = netdev->netdev_mtu_error;
1235 *mtup = netdev->mtu;
1241 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
1242 * in bytes, not including the hardware header; thus, this is typically 1500
1243 * bytes for Ethernet devices. */
1245 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
1247 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1250 ovs_mutex_lock(&netdev->mutex);
1251 error = netdev_linux_get_mtu__(netdev, mtup);
1252 ovs_mutex_unlock(&netdev->mutex);
1257 /* Sets the maximum size of transmitted (MTU) for given device using linux
1258 * networking ioctl interface.
1261 netdev_linux_set_mtu(const struct netdev *netdev_, int mtu)
1263 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1267 ovs_mutex_lock(&netdev->mutex);
1268 if (netdev->cache_valid & VALID_MTU) {
1269 error = netdev->netdev_mtu_error;
1270 if (error || netdev->mtu == mtu) {
1273 netdev->cache_valid &= ~VALID_MTU;
1276 error = af_inet_ifreq_ioctl(netdev_get_name(netdev_), &ifr,
1277 SIOCSIFMTU, "SIOCSIFMTU");
1278 if (!error || error == ENODEV) {
1279 netdev->netdev_mtu_error = error;
1280 netdev->mtu = ifr.ifr_mtu;
1281 netdev->cache_valid |= VALID_MTU;
1284 ovs_mutex_unlock(&netdev->mutex);
1288 /* Returns the ifindex of 'netdev', if successful, as a positive number.
1289 * On failure, returns a negative errno value. */
1291 netdev_linux_get_ifindex(const struct netdev *netdev_)
1293 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1296 ovs_mutex_lock(&netdev->mutex);
1297 error = get_ifindex(netdev_, &ifindex);
1298 ovs_mutex_unlock(&netdev->mutex);
1300 return error ? -error : ifindex;
1304 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
1306 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1308 ovs_mutex_lock(&netdev->mutex);
1309 if (netdev->miimon_interval > 0) {
1310 *carrier = netdev->miimon;
1312 *carrier = (netdev->ifi_flags & IFF_RUNNING) != 0;
1314 ovs_mutex_unlock(&netdev->mutex);
1319 static long long int
1320 netdev_linux_get_carrier_resets(const struct netdev *netdev_)
1322 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1323 long long int carrier_resets;
1325 ovs_mutex_lock(&netdev->mutex);
1326 carrier_resets = netdev->carrier_resets;
1327 ovs_mutex_unlock(&netdev->mutex);
1329 return carrier_resets;
1333 netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name,
1334 struct mii_ioctl_data *data)
1339 memset(&ifr, 0, sizeof ifr);
1340 memcpy(&ifr.ifr_data, data, sizeof *data);
1341 error = af_inet_ifreq_ioctl(name, &ifr, cmd, cmd_name);
1342 memcpy(data, &ifr.ifr_data, sizeof *data);
1348 netdev_linux_get_miimon(const char *name, bool *miimon)
1350 struct mii_ioctl_data data;
1355 memset(&data, 0, sizeof data);
1356 error = netdev_linux_do_miimon(name, SIOCGMIIPHY, "SIOCGMIIPHY", &data);
1358 /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */
1359 data.reg_num = MII_BMSR;
1360 error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG",
1364 *miimon = !!(data.val_out & BMSR_LSTATUS);
1366 VLOG_WARN_RL(&rl, "%s: failed to query MII", name);
1369 struct ethtool_cmd ecmd;
1371 VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool",
1374 COVERAGE_INC(netdev_get_ethtool);
1375 memset(&ecmd, 0, sizeof ecmd);
1376 error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK,
1379 struct ethtool_value eval;
1381 memcpy(&eval, &ecmd, sizeof eval);
1382 *miimon = !!eval.data;
1384 VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name);
1392 netdev_linux_set_miimon_interval(struct netdev *netdev_,
1393 long long int interval)
1395 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1397 ovs_mutex_lock(&netdev->mutex);
1398 interval = interval > 0 ? MAX(interval, 100) : 0;
1399 if (netdev->miimon_interval != interval) {
1402 if (interval && !netdev->miimon_interval) {
1403 atomic_add(&miimon_cnt, 1, &junk);
1404 } else if (!interval && netdev->miimon_interval) {
1405 atomic_sub(&miimon_cnt, 1, &junk);
1408 netdev->miimon_interval = interval;
1409 timer_set_expired(&netdev->miimon_timer);
1411 ovs_mutex_unlock(&netdev->mutex);
1417 netdev_linux_miimon_run(void)
1419 struct shash device_shash;
1420 struct shash_node *node;
1422 shash_init(&device_shash);
1423 netdev_get_devices(&netdev_linux_class, &device_shash);
1424 SHASH_FOR_EACH (node, &device_shash) {
1425 struct netdev *netdev = node->data;
1426 struct netdev_linux *dev = netdev_linux_cast(netdev);
1429 ovs_mutex_lock(&dev->mutex);
1430 if (dev->miimon_interval > 0 && timer_expired(&dev->miimon_timer)) {
1431 netdev_linux_get_miimon(dev->up.name, &miimon);
1432 if (miimon != dev->miimon) {
1433 dev->miimon = miimon;
1434 netdev_linux_changed(dev, dev->ifi_flags, 0);
1437 timer_set_duration(&dev->miimon_timer, dev->miimon_interval);
1439 ovs_mutex_unlock(&dev->mutex);
1440 netdev_close(netdev);
1443 shash_destroy(&device_shash);
1447 netdev_linux_miimon_wait(void)
1449 struct shash device_shash;
1450 struct shash_node *node;
1452 shash_init(&device_shash);
1453 netdev_get_devices(&netdev_linux_class, &device_shash);
1454 SHASH_FOR_EACH (node, &device_shash) {
1455 struct netdev *netdev = node->data;
1456 struct netdev_linux *dev = netdev_linux_cast(netdev);
1458 ovs_mutex_lock(&dev->mutex);
1459 if (dev->miimon_interval > 0) {
1460 timer_wait(&dev->miimon_timer);
1462 ovs_mutex_unlock(&dev->mutex);
1463 netdev_close(netdev);
1465 shash_destroy(&device_shash);
1469 swap_uint64(uint64_t *a, uint64_t *b)
1476 /* Copies 'src' into 'dst', performing format conversion in the process.
1478 * 'src' is allowed to be misaligned. */
1480 netdev_stats_from_ovs_vport_stats(struct netdev_stats *dst,
1481 const struct ovs_vport_stats *src)
1483 dst->rx_packets = get_unaligned_u64(&src->rx_packets);
1484 dst->tx_packets = get_unaligned_u64(&src->tx_packets);
1485 dst->rx_bytes = get_unaligned_u64(&src->rx_bytes);
1486 dst->tx_bytes = get_unaligned_u64(&src->tx_bytes);
1487 dst->rx_errors = get_unaligned_u64(&src->rx_errors);
1488 dst->tx_errors = get_unaligned_u64(&src->tx_errors);
1489 dst->rx_dropped = get_unaligned_u64(&src->rx_dropped);
1490 dst->tx_dropped = get_unaligned_u64(&src->tx_dropped);
1492 dst->collisions = 0;
1493 dst->rx_length_errors = 0;
1494 dst->rx_over_errors = 0;
1495 dst->rx_crc_errors = 0;
1496 dst->rx_frame_errors = 0;
1497 dst->rx_fifo_errors = 0;
1498 dst->rx_missed_errors = 0;
1499 dst->tx_aborted_errors = 0;
1500 dst->tx_carrier_errors = 0;
1501 dst->tx_fifo_errors = 0;
1502 dst->tx_heartbeat_errors = 0;
1503 dst->tx_window_errors = 0;
1507 get_stats_via_vport__(const struct netdev *netdev, struct netdev_stats *stats)
1509 struct dpif_linux_vport reply;
1513 error = dpif_linux_vport_get(netdev_get_name(netdev), &reply, &buf);
1516 } else if (!reply.stats) {
1521 netdev_stats_from_ovs_vport_stats(stats, reply.stats);
1529 get_stats_via_vport(const struct netdev *netdev_,
1530 struct netdev_stats *stats)
1532 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1534 if (!netdev->vport_stats_error ||
1535 !(netdev->cache_valid & VALID_VPORT_STAT_ERROR)) {
1538 error = get_stats_via_vport__(netdev_, stats);
1539 if (error && error != ENOENT) {
1540 VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed "
1542 netdev_get_name(netdev_), ovs_strerror(error));
1544 netdev->vport_stats_error = error;
1545 netdev->cache_valid |= VALID_VPORT_STAT_ERROR;
1549 /* Retrieves current device stats for 'netdev-linux'. */
1551 netdev_linux_get_stats(const struct netdev *netdev_,
1552 struct netdev_stats *stats)
1554 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1555 struct netdev_stats dev_stats;
1558 ovs_mutex_lock(&netdev->mutex);
1559 get_stats_via_vport(netdev_, stats);
1560 error = get_stats_via_netlink(netdev_, &dev_stats);
1562 if (!netdev->vport_stats_error) {
1565 } else if (netdev->vport_stats_error) {
1566 /* stats not available from OVS then use netdev stats. */
1569 /* Use kernel netdev's packet and byte counts since vport's counters
1570 * do not reflect packet counts on the wire when GSO, TSO or GRO are
1572 stats->rx_packets = dev_stats.rx_packets;
1573 stats->rx_bytes = dev_stats.rx_bytes;
1574 stats->tx_packets = dev_stats.tx_packets;
1575 stats->tx_bytes = dev_stats.tx_bytes;
1577 stats->rx_errors += dev_stats.rx_errors;
1578 stats->tx_errors += dev_stats.tx_errors;
1579 stats->rx_dropped += dev_stats.rx_dropped;
1580 stats->tx_dropped += dev_stats.tx_dropped;
1581 stats->multicast += dev_stats.multicast;
1582 stats->collisions += dev_stats.collisions;
1583 stats->rx_length_errors += dev_stats.rx_length_errors;
1584 stats->rx_over_errors += dev_stats.rx_over_errors;
1585 stats->rx_crc_errors += dev_stats.rx_crc_errors;
1586 stats->rx_frame_errors += dev_stats.rx_frame_errors;
1587 stats->rx_fifo_errors += dev_stats.rx_fifo_errors;
1588 stats->rx_missed_errors += dev_stats.rx_missed_errors;
1589 stats->tx_aborted_errors += dev_stats.tx_aborted_errors;
1590 stats->tx_carrier_errors += dev_stats.tx_carrier_errors;
1591 stats->tx_fifo_errors += dev_stats.tx_fifo_errors;
1592 stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors;
1593 stats->tx_window_errors += dev_stats.tx_window_errors;
1595 ovs_mutex_unlock(&netdev->mutex);
1600 /* Retrieves current device stats for 'netdev-tap' netdev or
1601 * netdev-internal. */
1603 netdev_tap_get_stats(const struct netdev *netdev_, struct netdev_stats *stats)
1605 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1606 struct netdev_stats dev_stats;
1609 ovs_mutex_lock(&netdev->mutex);
1610 get_stats_via_vport(netdev_, stats);
1611 error = get_stats_via_netlink(netdev_, &dev_stats);
1613 if (!netdev->vport_stats_error) {
1616 } else if (netdev->vport_stats_error) {
1617 /* Transmit and receive stats will appear to be swapped relative to the
1618 * other ports since we are the one sending the data, not a remote
1619 * computer. For consistency, we swap them back here. This does not
1620 * apply if we are getting stats from the vport layer because it always
1621 * tracks stats from the perspective of the switch. */
1624 swap_uint64(&stats->rx_packets, &stats->tx_packets);
1625 swap_uint64(&stats->rx_bytes, &stats->tx_bytes);
1626 swap_uint64(&stats->rx_errors, &stats->tx_errors);
1627 swap_uint64(&stats->rx_dropped, &stats->tx_dropped);
1628 stats->rx_length_errors = 0;
1629 stats->rx_over_errors = 0;
1630 stats->rx_crc_errors = 0;
1631 stats->rx_frame_errors = 0;
1632 stats->rx_fifo_errors = 0;
1633 stats->rx_missed_errors = 0;
1634 stats->tx_aborted_errors = 0;
1635 stats->tx_carrier_errors = 0;
1636 stats->tx_fifo_errors = 0;
1637 stats->tx_heartbeat_errors = 0;
1638 stats->tx_window_errors = 0;
1640 /* Use kernel netdev's packet and byte counts since vport counters
1641 * do not reflect packet counts on the wire when GSO, TSO or GRO
1643 stats->rx_packets = dev_stats.tx_packets;
1644 stats->rx_bytes = dev_stats.tx_bytes;
1645 stats->tx_packets = dev_stats.rx_packets;
1646 stats->tx_bytes = dev_stats.rx_bytes;
1648 stats->rx_dropped += dev_stats.tx_dropped;
1649 stats->tx_dropped += dev_stats.rx_dropped;
1651 stats->rx_errors += dev_stats.tx_errors;
1652 stats->tx_errors += dev_stats.rx_errors;
1654 stats->multicast += dev_stats.multicast;
1655 stats->collisions += dev_stats.collisions;
1657 ovs_mutex_unlock(&netdev->mutex);
1663 netdev_internal_get_stats(const struct netdev *netdev_,
1664 struct netdev_stats *stats)
1666 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1669 ovs_mutex_lock(&netdev->mutex);
1670 get_stats_via_vport(netdev_, stats);
1671 error = netdev->vport_stats_error;
1672 ovs_mutex_unlock(&netdev->mutex);
1678 netdev_internal_set_stats(struct netdev *netdev,
1679 const struct netdev_stats *stats)
1681 struct ovs_vport_stats vport_stats;
1682 struct dpif_linux_vport vport;
1685 vport_stats.rx_packets = stats->rx_packets;
1686 vport_stats.tx_packets = stats->tx_packets;
1687 vport_stats.rx_bytes = stats->rx_bytes;
1688 vport_stats.tx_bytes = stats->tx_bytes;
1689 vport_stats.rx_errors = stats->rx_errors;
1690 vport_stats.tx_errors = stats->tx_errors;
1691 vport_stats.rx_dropped = stats->rx_dropped;
1692 vport_stats.tx_dropped = stats->tx_dropped;
1694 dpif_linux_vport_init(&vport);
1695 vport.cmd = OVS_VPORT_CMD_SET;
1696 vport.name = netdev_get_name(netdev);
1697 vport.stats = &vport_stats;
1699 err = dpif_linux_vport_transact(&vport, NULL, NULL);
1701 /* If the vport layer doesn't know about the device, that doesn't mean it
1702 * doesn't exist (after all were able to open it when netdev_open() was
1703 * called), it just means that it isn't attached and we'll be getting
1704 * stats a different way. */
1705 if (err == ENODEV) {
1713 netdev_linux_read_features(struct netdev_linux *netdev)
1715 struct ethtool_cmd ecmd;
1719 if (netdev->cache_valid & VALID_FEATURES) {
1723 COVERAGE_INC(netdev_get_ethtool);
1724 memset(&ecmd, 0, sizeof ecmd);
1725 error = netdev_linux_do_ethtool(netdev->up.name, &ecmd,
1726 ETHTOOL_GSET, "ETHTOOL_GSET");
1731 /* Supported features. */
1732 netdev->supported = 0;
1733 if (ecmd.supported & SUPPORTED_10baseT_Half) {
1734 netdev->supported |= NETDEV_F_10MB_HD;
1736 if (ecmd.supported & SUPPORTED_10baseT_Full) {
1737 netdev->supported |= NETDEV_F_10MB_FD;
1739 if (ecmd.supported & SUPPORTED_100baseT_Half) {
1740 netdev->supported |= NETDEV_F_100MB_HD;
1742 if (ecmd.supported & SUPPORTED_100baseT_Full) {
1743 netdev->supported |= NETDEV_F_100MB_FD;
1745 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
1746 netdev->supported |= NETDEV_F_1GB_HD;
1748 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
1749 netdev->supported |= NETDEV_F_1GB_FD;
1751 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
1752 netdev->supported |= NETDEV_F_10GB_FD;
1754 if (ecmd.supported & SUPPORTED_TP) {
1755 netdev->supported |= NETDEV_F_COPPER;
1757 if (ecmd.supported & SUPPORTED_FIBRE) {
1758 netdev->supported |= NETDEV_F_FIBER;
1760 if (ecmd.supported & SUPPORTED_Autoneg) {
1761 netdev->supported |= NETDEV_F_AUTONEG;
1763 if (ecmd.supported & SUPPORTED_Pause) {
1764 netdev->supported |= NETDEV_F_PAUSE;
1766 if (ecmd.supported & SUPPORTED_Asym_Pause) {
1767 netdev->supported |= NETDEV_F_PAUSE_ASYM;
1770 /* Advertised features. */
1771 netdev->advertised = 0;
1772 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
1773 netdev->advertised |= NETDEV_F_10MB_HD;
1775 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
1776 netdev->advertised |= NETDEV_F_10MB_FD;
1778 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
1779 netdev->advertised |= NETDEV_F_100MB_HD;
1781 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
1782 netdev->advertised |= NETDEV_F_100MB_FD;
1784 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
1785 netdev->advertised |= NETDEV_F_1GB_HD;
1787 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
1788 netdev->advertised |= NETDEV_F_1GB_FD;
1790 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
1791 netdev->advertised |= NETDEV_F_10GB_FD;
1793 if (ecmd.advertising & ADVERTISED_TP) {
1794 netdev->advertised |= NETDEV_F_COPPER;
1796 if (ecmd.advertising & ADVERTISED_FIBRE) {
1797 netdev->advertised |= NETDEV_F_FIBER;
1799 if (ecmd.advertising & ADVERTISED_Autoneg) {
1800 netdev->advertised |= NETDEV_F_AUTONEG;
1802 if (ecmd.advertising & ADVERTISED_Pause) {
1803 netdev->advertised |= NETDEV_F_PAUSE;
1805 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
1806 netdev->advertised |= NETDEV_F_PAUSE_ASYM;
1809 /* Current settings. */
1811 if (speed == SPEED_10) {
1812 netdev->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD;
1813 } else if (speed == SPEED_100) {
1814 netdev->current = ecmd.duplex ? NETDEV_F_100MB_FD : NETDEV_F_100MB_HD;
1815 } else if (speed == SPEED_1000) {
1816 netdev->current = ecmd.duplex ? NETDEV_F_1GB_FD : NETDEV_F_1GB_HD;
1817 } else if (speed == SPEED_10000) {
1818 netdev->current = NETDEV_F_10GB_FD;
1819 } else if (speed == 40000) {
1820 netdev->current = NETDEV_F_40GB_FD;
1821 } else if (speed == 100000) {
1822 netdev->current = NETDEV_F_100GB_FD;
1823 } else if (speed == 1000000) {
1824 netdev->current = NETDEV_F_1TB_FD;
1826 netdev->current = 0;
1829 if (ecmd.port == PORT_TP) {
1830 netdev->current |= NETDEV_F_COPPER;
1831 } else if (ecmd.port == PORT_FIBRE) {
1832 netdev->current |= NETDEV_F_FIBER;
1836 netdev->current |= NETDEV_F_AUTONEG;
1840 netdev->cache_valid |= VALID_FEATURES;
1841 netdev->get_features_error = error;
1844 /* Stores the features supported by 'netdev' into of '*current', '*advertised',
1845 * '*supported', and '*peer'. Each value is a bitmap of NETDEV_* bits.
1846 * Returns 0 if successful, otherwise a positive errno value. */
1848 netdev_linux_get_features(const struct netdev *netdev_,
1849 enum netdev_features *current,
1850 enum netdev_features *advertised,
1851 enum netdev_features *supported,
1852 enum netdev_features *peer)
1854 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1857 ovs_mutex_lock(&netdev->mutex);
1858 netdev_linux_read_features(netdev);
1859 if (!netdev->get_features_error) {
1860 *current = netdev->current;
1861 *advertised = netdev->advertised;
1862 *supported = netdev->supported;
1863 *peer = 0; /* XXX */
1865 error = netdev->get_features_error;
1866 ovs_mutex_unlock(&netdev->mutex);
1871 /* Set the features advertised by 'netdev' to 'advertise'. */
1873 netdev_linux_set_advertisements(struct netdev *netdev_,
1874 enum netdev_features advertise)
1876 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1877 struct ethtool_cmd ecmd;
1880 ovs_mutex_lock(&netdev->mutex);
1882 COVERAGE_INC(netdev_get_ethtool);
1883 memset(&ecmd, 0, sizeof ecmd);
1884 error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd,
1885 ETHTOOL_GSET, "ETHTOOL_GSET");
1890 ecmd.advertising = 0;
1891 if (advertise & NETDEV_F_10MB_HD) {
1892 ecmd.advertising |= ADVERTISED_10baseT_Half;
1894 if (advertise & NETDEV_F_10MB_FD) {
1895 ecmd.advertising |= ADVERTISED_10baseT_Full;
1897 if (advertise & NETDEV_F_100MB_HD) {
1898 ecmd.advertising |= ADVERTISED_100baseT_Half;
1900 if (advertise & NETDEV_F_100MB_FD) {
1901 ecmd.advertising |= ADVERTISED_100baseT_Full;
1903 if (advertise & NETDEV_F_1GB_HD) {
1904 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1906 if (advertise & NETDEV_F_1GB_FD) {
1907 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1909 if (advertise & NETDEV_F_10GB_FD) {
1910 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1912 if (advertise & NETDEV_F_COPPER) {
1913 ecmd.advertising |= ADVERTISED_TP;
1915 if (advertise & NETDEV_F_FIBER) {
1916 ecmd.advertising |= ADVERTISED_FIBRE;
1918 if (advertise & NETDEV_F_AUTONEG) {
1919 ecmd.advertising |= ADVERTISED_Autoneg;
1921 if (advertise & NETDEV_F_PAUSE) {
1922 ecmd.advertising |= ADVERTISED_Pause;
1924 if (advertise & NETDEV_F_PAUSE_ASYM) {
1925 ecmd.advertising |= ADVERTISED_Asym_Pause;
1927 COVERAGE_INC(netdev_set_ethtool);
1928 error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd,
1929 ETHTOOL_SSET, "ETHTOOL_SSET");
1932 ovs_mutex_unlock(&netdev->mutex);
1936 /* Attempts to set input rate limiting (policing) policy. Returns 0 if
1937 * successful, otherwise a positive errno value. */
1939 netdev_linux_set_policing(struct netdev *netdev_,
1940 uint32_t kbits_rate, uint32_t kbits_burst)
1942 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1943 const char *netdev_name = netdev_get_name(netdev_);
1946 kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */
1947 : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */
1948 : kbits_burst); /* Stick with user-specified value. */
1950 ovs_mutex_lock(&netdev->mutex);
1951 if (netdev->cache_valid & VALID_POLICING) {
1952 error = netdev->netdev_policing_error;
1953 if (error || (netdev->kbits_rate == kbits_rate &&
1954 netdev->kbits_burst == kbits_burst)) {
1955 /* Assume that settings haven't changed since we last set them. */
1958 netdev->cache_valid &= ~VALID_POLICING;
1961 COVERAGE_INC(netdev_set_policing);
1962 /* Remove any existing ingress qdisc. */
1963 error = tc_add_del_ingress_qdisc(netdev_, false);
1965 VLOG_WARN_RL(&rl, "%s: removing policing failed: %s",
1966 netdev_name, ovs_strerror(error));
1971 error = tc_add_del_ingress_qdisc(netdev_, true);
1973 VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s",
1974 netdev_name, ovs_strerror(error));
1978 error = tc_add_policer(netdev_, kbits_rate, kbits_burst);
1980 VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s",
1981 netdev_name, ovs_strerror(error));
1986 netdev->kbits_rate = kbits_rate;
1987 netdev->kbits_burst = kbits_burst;
1990 if (!error || error == ENODEV) {
1991 netdev->netdev_policing_error = error;
1992 netdev->cache_valid |= VALID_POLICING;
1994 ovs_mutex_unlock(&netdev->mutex);
1999 netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED,
2002 const struct tc_ops *const *opsp;
2004 for (opsp = tcs; *opsp != NULL; opsp++) {
2005 const struct tc_ops *ops = *opsp;
2006 if (ops->tc_install && ops->ovs_name[0] != '\0') {
2007 sset_add(types, ops->ovs_name);
2013 static const struct tc_ops *
2014 tc_lookup_ovs_name(const char *name)
2016 const struct tc_ops *const *opsp;
2018 for (opsp = tcs; *opsp != NULL; opsp++) {
2019 const struct tc_ops *ops = *opsp;
2020 if (!strcmp(name, ops->ovs_name)) {
2027 static const struct tc_ops *
2028 tc_lookup_linux_name(const char *name)
2030 const struct tc_ops *const *opsp;
2032 for (opsp = tcs; *opsp != NULL; opsp++) {
2033 const struct tc_ops *ops = *opsp;
2034 if (ops->linux_name && !strcmp(name, ops->linux_name)) {
2041 static struct tc_queue *
2042 tc_find_queue__(const struct netdev *netdev_, unsigned int queue_id,
2045 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2046 struct tc_queue *queue;
2048 HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev->tc->queues) {
2049 if (queue->queue_id == queue_id) {
2056 static struct tc_queue *
2057 tc_find_queue(const struct netdev *netdev, unsigned int queue_id)
2059 return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0));
2063 netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED,
2065 struct netdev_qos_capabilities *caps)
2067 const struct tc_ops *ops = tc_lookup_ovs_name(type);
2071 caps->n_queues = ops->n_queues;
2076 netdev_linux_get_qos(const struct netdev *netdev_,
2077 const char **typep, struct smap *details)
2079 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2082 ovs_mutex_lock(&netdev->mutex);
2083 error = tc_query_qdisc(netdev_);
2085 *typep = netdev->tc->ops->ovs_name;
2086 error = (netdev->tc->ops->qdisc_get
2087 ? netdev->tc->ops->qdisc_get(netdev_, details)
2090 ovs_mutex_unlock(&netdev->mutex);
2096 netdev_linux_set_qos(struct netdev *netdev_,
2097 const char *type, const struct smap *details)
2099 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2100 const struct tc_ops *new_ops;
2103 new_ops = tc_lookup_ovs_name(type);
2104 if (!new_ops || !new_ops->tc_install) {
2108 ovs_mutex_lock(&netdev->mutex);
2109 error = tc_query_qdisc(netdev_);
2114 if (new_ops == netdev->tc->ops) {
2115 error = new_ops->qdisc_set ? new_ops->qdisc_set(netdev_, details) : 0;
2117 /* Delete existing qdisc. */
2118 error = tc_del_qdisc(netdev_);
2122 ovs_assert(netdev->tc == NULL);
2124 /* Install new qdisc. */
2125 error = new_ops->tc_install(netdev_, details);
2126 ovs_assert((error == 0) == (netdev->tc != NULL));
2130 ovs_mutex_unlock(&netdev->mutex);
2135 netdev_linux_get_queue(const struct netdev *netdev_,
2136 unsigned int queue_id, struct smap *details)
2138 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2141 ovs_mutex_lock(&netdev->mutex);
2142 error = tc_query_qdisc(netdev_);
2144 struct tc_queue *queue = tc_find_queue(netdev_, queue_id);
2146 ? netdev->tc->ops->class_get(netdev_, queue, details)
2149 ovs_mutex_unlock(&netdev->mutex);
2155 netdev_linux_set_queue(struct netdev *netdev_,
2156 unsigned int queue_id, const struct smap *details)
2158 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2161 ovs_mutex_lock(&netdev->mutex);
2162 error = tc_query_qdisc(netdev_);
2164 error = (queue_id < netdev->tc->ops->n_queues
2165 && netdev->tc->ops->class_set
2166 ? netdev->tc->ops->class_set(netdev_, queue_id, details)
2169 ovs_mutex_unlock(&netdev->mutex);
2175 netdev_linux_delete_queue(struct netdev *netdev_, unsigned int queue_id)
2177 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2180 ovs_mutex_lock(&netdev->mutex);
2181 error = tc_query_qdisc(netdev_);
2183 if (netdev->tc->ops->class_delete) {
2184 struct tc_queue *queue = tc_find_queue(netdev_, queue_id);
2186 ? netdev->tc->ops->class_delete(netdev_, queue)
2192 ovs_mutex_unlock(&netdev->mutex);
2198 netdev_linux_get_queue_stats(const struct netdev *netdev_,
2199 unsigned int queue_id,
2200 struct netdev_queue_stats *stats)
2202 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2205 ovs_mutex_lock(&netdev->mutex);
2206 error = tc_query_qdisc(netdev_);
2208 if (netdev->tc->ops->class_get_stats) {
2209 const struct tc_queue *queue = tc_find_queue(netdev_, queue_id);
2211 stats->created = queue->created;
2212 error = netdev->tc->ops->class_get_stats(netdev_, queue,
2221 ovs_mutex_unlock(&netdev->mutex);
2226 struct queue_dump_state {
2227 struct nl_dump dump;
2232 start_queue_dump(const struct netdev *netdev, struct queue_dump_state *state)
2234 struct ofpbuf request;
2235 struct tcmsg *tcmsg;
2237 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request);
2241 tcmsg->tcm_parent = 0;
2242 nl_dump_start(&state->dump, NETLINK_ROUTE, &request);
2243 ofpbuf_uninit(&request);
2245 ofpbuf_init(&state->buf, NL_DUMP_BUFSIZE);
2250 finish_queue_dump(struct queue_dump_state *state)
2252 ofpbuf_uninit(&state->buf);
2253 return nl_dump_done(&state->dump);
2256 struct netdev_linux_queue_state {
2257 unsigned int *queues;
2263 netdev_linux_queue_dump_start(const struct netdev *netdev_, void **statep)
2265 const struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2268 ovs_mutex_lock(&netdev->mutex);
2269 error = tc_query_qdisc(netdev_);
2271 if (netdev->tc->ops->class_get) {
2272 struct netdev_linux_queue_state *state;
2273 struct tc_queue *queue;
2276 *statep = state = xmalloc(sizeof *state);
2277 state->n_queues = hmap_count(&netdev->tc->queues);
2278 state->cur_queue = 0;
2279 state->queues = xmalloc(state->n_queues * sizeof *state->queues);
2282 HMAP_FOR_EACH (queue, hmap_node, &netdev->tc->queues) {
2283 state->queues[i++] = queue->queue_id;
2289 ovs_mutex_unlock(&netdev->mutex);
2295 netdev_linux_queue_dump_next(const struct netdev *netdev_, void *state_,
2296 unsigned int *queue_idp, struct smap *details)
2298 const struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2299 struct netdev_linux_queue_state *state = state_;
2302 ovs_mutex_lock(&netdev->mutex);
2303 while (state->cur_queue < state->n_queues) {
2304 unsigned int queue_id = state->queues[state->cur_queue++];
2305 struct tc_queue *queue = tc_find_queue(netdev_, queue_id);
2308 *queue_idp = queue_id;
2309 error = netdev->tc->ops->class_get(netdev_, queue, details);
2313 ovs_mutex_unlock(&netdev->mutex);
2319 netdev_linux_queue_dump_done(const struct netdev *netdev OVS_UNUSED,
2322 struct netdev_linux_queue_state *state = state_;
2324 free(state->queues);
2330 netdev_linux_dump_queue_stats(const struct netdev *netdev_,
2331 netdev_dump_queue_stats_cb *cb, void *aux)
2333 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2336 ovs_mutex_lock(&netdev->mutex);
2337 error = tc_query_qdisc(netdev_);
2339 struct queue_dump_state state;
2341 if (!netdev->tc->ops->class_dump_stats) {
2343 } else if (!start_queue_dump(netdev_, &state)) {
2349 while (nl_dump_next(&state.dump, &msg, &state.buf)) {
2350 retval = netdev->tc->ops->class_dump_stats(netdev_, &msg,
2357 retval = finish_queue_dump(&state);
2363 ovs_mutex_unlock(&netdev->mutex);
2369 netdev_linux_get_in4(const struct netdev *netdev_,
2370 struct in_addr *address, struct in_addr *netmask)
2372 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2375 ovs_mutex_lock(&netdev->mutex);
2376 if (!(netdev->cache_valid & VALID_IN4)) {
2377 error = netdev_linux_get_ipv4(netdev_, &netdev->address,
2378 SIOCGIFADDR, "SIOCGIFADDR");
2380 error = netdev_linux_get_ipv4(netdev_, &netdev->netmask,
2381 SIOCGIFNETMASK, "SIOCGIFNETMASK");
2383 netdev->cache_valid |= VALID_IN4;
2391 if (netdev->address.s_addr != INADDR_ANY) {
2392 *address = netdev->address;
2393 *netmask = netdev->netmask;
2395 error = EADDRNOTAVAIL;
2398 ovs_mutex_unlock(&netdev->mutex);
2404 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
2405 struct in_addr netmask)
2407 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2410 ovs_mutex_lock(&netdev->mutex);
2411 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
2413 netdev->cache_valid |= VALID_IN4;
2414 netdev->address = address;
2415 netdev->netmask = netmask;
2416 if (address.s_addr != INADDR_ANY) {
2417 error = do_set_addr(netdev_, SIOCSIFNETMASK,
2418 "SIOCSIFNETMASK", netmask);
2421 ovs_mutex_unlock(&netdev->mutex);
2427 parse_if_inet6_line(const char *line,
2428 struct in6_addr *in6, char ifname[16 + 1])
2430 uint8_t *s6 = in6->s6_addr;
2431 #define X8 "%2"SCNx8
2432 return ovs_scan(line,
2433 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
2434 "%*x %*x %*x %*x %16s\n",
2435 &s6[0], &s6[1], &s6[2], &s6[3],
2436 &s6[4], &s6[5], &s6[6], &s6[7],
2437 &s6[8], &s6[9], &s6[10], &s6[11],
2438 &s6[12], &s6[13], &s6[14], &s6[15],
2442 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
2443 * 'in6' is non-null) and returns true. Otherwise, returns false. */
2445 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
2447 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2449 ovs_mutex_lock(&netdev->mutex);
2450 if (!(netdev->cache_valid & VALID_IN6)) {
2454 netdev->in6 = in6addr_any;
2456 file = fopen("/proc/net/if_inet6", "r");
2458 const char *name = netdev_get_name(netdev_);
2459 while (fgets(line, sizeof line, file)) {
2460 struct in6_addr in6_tmp;
2461 char ifname[16 + 1];
2462 if (parse_if_inet6_line(line, &in6_tmp, ifname)
2463 && !strcmp(name, ifname))
2465 netdev->in6 = in6_tmp;
2471 netdev->cache_valid |= VALID_IN6;
2474 ovs_mutex_unlock(&netdev->mutex);
2480 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
2482 struct sockaddr_in sin;
2483 memset(&sin, 0, sizeof sin);
2484 sin.sin_family = AF_INET;
2485 sin.sin_addr = addr;
2488 memset(sa, 0, sizeof *sa);
2489 memcpy(sa, &sin, sizeof sin);
2493 do_set_addr(struct netdev *netdev,
2494 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
2498 make_in4_sockaddr(&ifr.ifr_addr, addr);
2499 return af_inet_ifreq_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
2503 /* Adds 'router' as a default IP gateway. */
2505 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
2507 struct in_addr any = { INADDR_ANY };
2511 memset(&rt, 0, sizeof rt);
2512 make_in4_sockaddr(&rt.rt_dst, any);
2513 make_in4_sockaddr(&rt.rt_gateway, router);
2514 make_in4_sockaddr(&rt.rt_genmask, any);
2515 rt.rt_flags = RTF_UP | RTF_GATEWAY;
2516 error = af_inet_ioctl(SIOCADDRT, &rt);
2518 VLOG_WARN("ioctl(SIOCADDRT): %s", ovs_strerror(error));
2524 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
2527 static const char fn[] = "/proc/net/route";
2532 *netdev_name = NULL;
2533 stream = fopen(fn, "r");
2534 if (stream == NULL) {
2535 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno));
2540 while (fgets(line, sizeof line, stream)) {
2543 ovs_be32 dest, gateway, mask;
2544 int refcnt, metric, mtu;
2545 unsigned int flags, use, window, irtt;
2548 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
2550 iface, &dest, &gateway, &flags, &refcnt,
2551 &use, &metric, &mask, &mtu, &window, &irtt)) {
2552 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
2556 if (!(flags & RTF_UP)) {
2557 /* Skip routes that aren't up. */
2561 /* The output of 'dest', 'mask', and 'gateway' were given in
2562 * network byte order, so we don't need need any endian
2563 * conversions here. */
2564 if ((dest & mask) == (host->s_addr & mask)) {
2566 /* The host is directly reachable. */
2567 next_hop->s_addr = 0;
2569 /* To reach the host, we must go through a gateway. */
2570 next_hop->s_addr = gateway;
2572 *netdev_name = xstrdup(iface);
2584 netdev_linux_get_status(const struct netdev *netdev_, struct smap *smap)
2586 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2589 ovs_mutex_lock(&netdev->mutex);
2590 if (!(netdev->cache_valid & VALID_DRVINFO)) {
2591 struct ethtool_cmd *cmd = (struct ethtool_cmd *) &netdev->drvinfo;
2593 COVERAGE_INC(netdev_get_ethtool);
2594 memset(&netdev->drvinfo, 0, sizeof netdev->drvinfo);
2595 error = netdev_linux_do_ethtool(netdev->up.name,
2598 "ETHTOOL_GDRVINFO");
2600 netdev->cache_valid |= VALID_DRVINFO;
2605 smap_add(smap, "driver_name", netdev->drvinfo.driver);
2606 smap_add(smap, "driver_version", netdev->drvinfo.version);
2607 smap_add(smap, "firmware_version", netdev->drvinfo.fw_version);
2609 ovs_mutex_unlock(&netdev->mutex);
2615 netdev_internal_get_status(const struct netdev *netdev OVS_UNUSED,
2618 smap_add(smap, "driver_name", "openvswitch");
2622 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
2623 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
2624 * returns 0. Otherwise, it returns a positive errno value; in particular,
2625 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
2627 netdev_linux_arp_lookup(const struct netdev *netdev,
2628 ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
2631 struct sockaddr_in sin;
2634 memset(&r, 0, sizeof r);
2635 memset(&sin, 0, sizeof sin);
2636 sin.sin_family = AF_INET;
2637 sin.sin_addr.s_addr = ip;
2639 memcpy(&r.arp_pa, &sin, sizeof sin);
2640 r.arp_ha.sa_family = ARPHRD_ETHER;
2642 ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
2643 COVERAGE_INC(netdev_arp_lookup);
2644 retval = af_inet_ioctl(SIOCGARP, &r);
2646 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
2647 } else if (retval != ENXIO) {
2648 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
2649 netdev_get_name(netdev), IP_ARGS(ip),
2650 ovs_strerror(retval));
2656 nd_to_iff_flags(enum netdev_flags nd)
2659 if (nd & NETDEV_UP) {
2662 if (nd & NETDEV_PROMISC) {
2665 if (nd & NETDEV_LOOPBACK) {
2666 iff |= IFF_LOOPBACK;
2672 iff_to_nd_flags(int iff)
2674 enum netdev_flags nd = 0;
2678 if (iff & IFF_PROMISC) {
2679 nd |= NETDEV_PROMISC;
2681 if (iff & IFF_LOOPBACK) {
2682 nd |= NETDEV_LOOPBACK;
2688 update_flags(struct netdev_linux *netdev, enum netdev_flags off,
2689 enum netdev_flags on, enum netdev_flags *old_flagsp)
2690 OVS_REQUIRES(netdev->mutex)
2692 int old_flags, new_flags;
2695 old_flags = netdev->ifi_flags;
2696 *old_flagsp = iff_to_nd_flags(old_flags);
2697 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
2698 if (new_flags != old_flags) {
2699 error = set_flags(netdev_get_name(&netdev->up), new_flags);
2700 get_flags(&netdev->up, &netdev->ifi_flags);
2707 netdev_linux_update_flags(struct netdev *netdev_, enum netdev_flags off,
2708 enum netdev_flags on, enum netdev_flags *old_flagsp)
2710 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2713 ovs_mutex_lock(&netdev->mutex);
2714 error = update_flags(netdev, off, on, old_flagsp);
2715 ovs_mutex_unlock(&netdev->mutex);
2720 #define NETDEV_LINUX_CLASS(NAME, CONSTRUCT, GET_STATS, SET_STATS, \
2721 GET_FEATURES, GET_STATUS) \
2727 netdev_linux_wait, \
2729 netdev_linux_alloc, \
2731 netdev_linux_destruct, \
2732 netdev_linux_dealloc, \
2733 NULL, /* get_config */ \
2734 NULL, /* set_config */ \
2735 NULL, /* get_tunnel_config */ \
2737 netdev_linux_send, \
2738 netdev_linux_send_wait, \
2740 netdev_linux_set_etheraddr, \
2741 netdev_linux_get_etheraddr, \
2742 netdev_linux_get_mtu, \
2743 netdev_linux_set_mtu, \
2744 netdev_linux_get_ifindex, \
2745 netdev_linux_get_carrier, \
2746 netdev_linux_get_carrier_resets, \
2747 netdev_linux_set_miimon_interval, \
2752 netdev_linux_set_advertisements, \
2754 netdev_linux_set_policing, \
2755 netdev_linux_get_qos_types, \
2756 netdev_linux_get_qos_capabilities, \
2757 netdev_linux_get_qos, \
2758 netdev_linux_set_qos, \
2759 netdev_linux_get_queue, \
2760 netdev_linux_set_queue, \
2761 netdev_linux_delete_queue, \
2762 netdev_linux_get_queue_stats, \
2763 netdev_linux_queue_dump_start, \
2764 netdev_linux_queue_dump_next, \
2765 netdev_linux_queue_dump_done, \
2766 netdev_linux_dump_queue_stats, \
2768 netdev_linux_get_in4, \
2769 netdev_linux_set_in4, \
2770 netdev_linux_get_in6, \
2771 netdev_linux_add_router, \
2772 netdev_linux_get_next_hop, \
2774 netdev_linux_arp_lookup, \
2776 netdev_linux_update_flags, \
2778 netdev_linux_rxq_alloc, \
2779 netdev_linux_rxq_construct, \
2780 netdev_linux_rxq_destruct, \
2781 netdev_linux_rxq_dealloc, \
2782 netdev_linux_rxq_recv, \
2783 netdev_linux_rxq_wait, \
2784 netdev_linux_rxq_drain, \
2787 const struct netdev_class netdev_linux_class =
2790 netdev_linux_construct,
2791 netdev_linux_get_stats,
2792 NULL, /* set_stats */
2793 netdev_linux_get_features,
2794 netdev_linux_get_status);
2796 const struct netdev_class netdev_tap_class =
2799 netdev_linux_construct_tap,
2800 netdev_tap_get_stats,
2801 NULL, /* set_stats */
2802 netdev_linux_get_features,
2803 netdev_linux_get_status);
2805 const struct netdev_class netdev_internal_class =
2808 netdev_linux_construct,
2809 netdev_internal_get_stats,
2810 netdev_internal_set_stats,
2811 NULL, /* get_features */
2812 netdev_internal_get_status);
2814 /* HTB traffic control class. */
2816 #define HTB_N_QUEUES 0xf000
2817 #define HTB_RATE2QUANTUM 10
2821 unsigned int max_rate; /* In bytes/s. */
2825 struct tc_queue tc_queue;
2826 unsigned int min_rate; /* In bytes/s. */
2827 unsigned int max_rate; /* In bytes/s. */
2828 unsigned int burst; /* In bytes. */
2829 unsigned int priority; /* Lower values are higher priorities. */
2833 htb_get__(const struct netdev *netdev_)
2835 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2836 return CONTAINER_OF(netdev->tc, struct htb, tc);
2840 htb_install__(struct netdev *netdev_, uint64_t max_rate)
2842 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2845 htb = xmalloc(sizeof *htb);
2846 tc_init(&htb->tc, &tc_ops_htb);
2847 htb->max_rate = max_rate;
2849 netdev->tc = &htb->tc;
2852 /* Create an HTB qdisc.
2854 * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */
2856 htb_setup_qdisc__(struct netdev *netdev)
2859 struct tc_htb_glob opt;
2860 struct ofpbuf request;
2861 struct tcmsg *tcmsg;
2863 tc_del_qdisc(netdev);
2865 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
2866 NLM_F_EXCL | NLM_F_CREATE, &request);
2870 tcmsg->tcm_handle = tc_make_handle(1, 0);
2871 tcmsg->tcm_parent = TC_H_ROOT;
2873 nl_msg_put_string(&request, TCA_KIND, "htb");
2875 memset(&opt, 0, sizeof opt);
2876 opt.rate2quantum = HTB_RATE2QUANTUM;
2880 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2881 nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt);
2882 nl_msg_end_nested(&request, opt_offset);
2884 return tc_transact(&request, NULL);
2887 /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb
2888 * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */
2890 htb_setup_class__(struct netdev *netdev, unsigned int handle,
2891 unsigned int parent, struct htb_class *class)
2894 struct tc_htb_opt opt;
2895 struct ofpbuf request;
2896 struct tcmsg *tcmsg;
2900 error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu);
2902 VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU",
2903 netdev_get_name(netdev));
2907 memset(&opt, 0, sizeof opt);
2908 tc_fill_rate(&opt.rate, class->min_rate, mtu);
2909 tc_fill_rate(&opt.ceil, class->max_rate, mtu);
2910 /* Makes sure the quantum is at least MTU. Setting quantum will
2911 * make htb ignore the r2q for this class. */
2912 if ((class->min_rate / HTB_RATE2QUANTUM) < mtu) {
2915 opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst);
2916 opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst);
2917 opt.prio = class->priority;
2919 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
2923 tcmsg->tcm_handle = handle;
2924 tcmsg->tcm_parent = parent;
2926 nl_msg_put_string(&request, TCA_KIND, "htb");
2927 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2928 nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt);
2929 tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate);
2930 tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil);
2931 nl_msg_end_nested(&request, opt_offset);
2933 error = tc_transact(&request, NULL);
2935 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
2936 "min_rate=%u max_rate=%u burst=%u prio=%u (%s)",
2937 netdev_get_name(netdev),
2938 tc_get_major(handle), tc_get_minor(handle),
2939 tc_get_major(parent), tc_get_minor(parent),
2940 class->min_rate, class->max_rate,
2941 class->burst, class->priority, ovs_strerror(error));
2946 /* Parses Netlink attributes in 'options' for HTB parameters and stores a
2947 * description of them into 'details'. The description complies with the
2948 * specification given in the vswitch database documentation for linux-htb
2951 htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class)
2953 static const struct nl_policy tca_htb_policy[] = {
2954 [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false,
2955 .min_len = sizeof(struct tc_htb_opt) },
2958 struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)];
2959 const struct tc_htb_opt *htb;
2961 if (!nl_parse_nested(nl_options, tca_htb_policy,
2962 attrs, ARRAY_SIZE(tca_htb_policy))) {
2963 VLOG_WARN_RL(&rl, "failed to parse HTB class options");
2967 htb = nl_attr_get(attrs[TCA_HTB_PARMS]);
2968 class->min_rate = htb->rate.rate;
2969 class->max_rate = htb->ceil.rate;
2970 class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer);
2971 class->priority = htb->prio;
2976 htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
2977 struct htb_class *options,
2978 struct netdev_queue_stats *stats)
2980 struct nlattr *nl_options;
2981 unsigned int handle;
2984 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
2985 if (!error && queue_id) {
2986 unsigned int major = tc_get_major(handle);
2987 unsigned int minor = tc_get_minor(handle);
2988 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2989 *queue_id = minor - 1;
2994 if (!error && options) {
2995 error = htb_parse_tca_options__(nl_options, options);
3001 htb_parse_qdisc_details__(struct netdev *netdev_,
3002 const struct smap *details, struct htb_class *hc)
3004 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
3005 const char *max_rate_s;
3007 max_rate_s = smap_get(details, "max-rate");
3008 hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
3009 if (!hc->max_rate) {
3010 enum netdev_features current;
3012 netdev_linux_read_features(netdev);
3013 current = !netdev->get_features_error ? netdev->current : 0;
3014 hc->max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8;
3016 hc->min_rate = hc->max_rate;
3022 htb_parse_class_details__(struct netdev *netdev,
3023 const struct smap *details, struct htb_class *hc)
3025 const struct htb *htb = htb_get__(netdev);
3026 const char *min_rate_s = smap_get(details, "min-rate");
3027 const char *max_rate_s = smap_get(details, "max-rate");
3028 const char *burst_s = smap_get(details, "burst");
3029 const char *priority_s = smap_get(details, "priority");
3032 error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu);
3034 VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU",
3035 netdev_get_name(netdev));
3039 /* HTB requires at least an mtu sized min-rate to send any traffic even
3040 * on uncongested links. */
3041 hc->min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
3042 hc->min_rate = MAX(hc->min_rate, mtu);
3043 hc->min_rate = MIN(hc->min_rate, htb->max_rate);
3046 hc->max_rate = (max_rate_s
3047 ? strtoull(max_rate_s, NULL, 10) / 8
3049 hc->max_rate = MAX(hc->max_rate, hc->min_rate);
3050 hc->max_rate = MIN(hc->max_rate, htb->max_rate);
3054 * According to hints in the documentation that I've read, it is important
3055 * that 'burst' be at least as big as the largest frame that might be
3056 * transmitted. Also, making 'burst' a bit bigger than necessary is OK,
3057 * but having it a bit too small is a problem. Since netdev_get_mtu()
3058 * doesn't include the Ethernet header, we need to add at least 14 (18?) to
3059 * the MTU. We actually add 64, instead of 14, as a guard against
3060 * additional headers get tacked on somewhere that we're not aware of. */
3061 hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0;
3062 hc->burst = MAX(hc->burst, mtu + 64);
3065 hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0;
3071 htb_query_class__(const struct netdev *netdev, unsigned int handle,
3072 unsigned int parent, struct htb_class *options,
3073 struct netdev_queue_stats *stats)
3075 struct ofpbuf *reply;
3078 error = tc_query_class(netdev, handle, parent, &reply);
3080 error = htb_parse_tcmsg__(reply, NULL, options, stats);
3081 ofpbuf_delete(reply);
3087 htb_tc_install(struct netdev *netdev, const struct smap *details)
3091 error = htb_setup_qdisc__(netdev);
3093 struct htb_class hc;
3095 htb_parse_qdisc_details__(netdev, details, &hc);
3096 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3097 tc_make_handle(1, 0), &hc);
3099 htb_install__(netdev, hc.max_rate);
3105 static struct htb_class *
3106 htb_class_cast__(const struct tc_queue *queue)
3108 return CONTAINER_OF(queue, struct htb_class, tc_queue);
3112 htb_update_queue__(struct netdev *netdev, unsigned int queue_id,
3113 const struct htb_class *hc)
3115 struct htb *htb = htb_get__(netdev);
3116 size_t hash = hash_int(queue_id, 0);
3117 struct tc_queue *queue;
3118 struct htb_class *hcp;
3120 queue = tc_find_queue__(netdev, queue_id, hash);
3122 hcp = htb_class_cast__(queue);
3124 hcp = xmalloc(sizeof *hcp);
3125 queue = &hcp->tc_queue;
3126 queue->queue_id = queue_id;
3127 queue->created = time_msec();
3128 hmap_insert(&htb->tc.queues, &queue->hmap_node, hash);
3131 hcp->min_rate = hc->min_rate;
3132 hcp->max_rate = hc->max_rate;
3133 hcp->burst = hc->burst;
3134 hcp->priority = hc->priority;
3138 htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3141 struct queue_dump_state state;
3142 struct htb_class hc;
3144 /* Get qdisc options. */
3146 htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
3147 htb_install__(netdev, hc.max_rate);
3150 if (!start_queue_dump(netdev, &state)) {
3153 while (nl_dump_next(&state.dump, &msg, &state.buf)) {
3154 unsigned int queue_id;
3156 if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
3157 htb_update_queue__(netdev, queue_id, &hc);
3160 finish_queue_dump(&state);
3166 htb_tc_destroy(struct tc *tc)
3168 struct htb *htb = CONTAINER_OF(tc, struct htb, tc);
3169 struct htb_class *hc, *next;
3171 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) {
3172 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
3180 htb_qdisc_get(const struct netdev *netdev, struct smap *details)
3182 const struct htb *htb = htb_get__(netdev);
3183 smap_add_format(details, "max-rate", "%llu", 8ULL * htb->max_rate);
3188 htb_qdisc_set(struct netdev *netdev, const struct smap *details)
3190 struct htb_class hc;
3193 htb_parse_qdisc_details__(netdev, details, &hc);
3194 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3195 tc_make_handle(1, 0), &hc);
3197 htb_get__(netdev)->max_rate = hc.max_rate;
3203 htb_class_get(const struct netdev *netdev OVS_UNUSED,
3204 const struct tc_queue *queue, struct smap *details)
3206 const struct htb_class *hc = htb_class_cast__(queue);
3208 smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate);
3209 if (hc->min_rate != hc->max_rate) {
3210 smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate);
3212 smap_add_format(details, "burst", "%llu", 8ULL * hc->burst);
3214 smap_add_format(details, "priority", "%u", hc->priority);
3220 htb_class_set(struct netdev *netdev, unsigned int queue_id,
3221 const struct smap *details)
3223 struct htb_class hc;
3226 error = htb_parse_class_details__(netdev, details, &hc);
3231 error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
3232 tc_make_handle(1, 0xfffe), &hc);
3237 htb_update_queue__(netdev, queue_id, &hc);
3242 htb_class_delete(struct netdev *netdev, struct tc_queue *queue)
3244 struct htb_class *hc = htb_class_cast__(queue);
3245 struct htb *htb = htb_get__(netdev);
3248 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
3250 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
3257 htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
3258 struct netdev_queue_stats *stats)
3260 return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
3261 tc_make_handle(1, 0xfffe), NULL, stats);
3265 htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
3266 const struct ofpbuf *nlmsg,
3267 netdev_dump_queue_stats_cb *cb, void *aux)
3269 struct netdev_queue_stats stats;
3270 unsigned int handle, major, minor;
3273 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
3278 major = tc_get_major(handle);
3279 minor = tc_get_minor(handle);
3280 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
3281 (*cb)(minor - 1, &stats, aux);
3286 static const struct tc_ops tc_ops_htb = {
3287 "htb", /* linux_name */
3288 "linux-htb", /* ovs_name */
3289 HTB_N_QUEUES, /* n_queues */
3298 htb_class_get_stats,
3299 htb_class_dump_stats
3302 /* "linux-hfsc" traffic control class. */
3304 #define HFSC_N_QUEUES 0xf000
3312 struct tc_queue tc_queue;
3317 static struct hfsc *
3318 hfsc_get__(const struct netdev *netdev_)
3320 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
3321 return CONTAINER_OF(netdev->tc, struct hfsc, tc);
3324 static struct hfsc_class *
3325 hfsc_class_cast__(const struct tc_queue *queue)
3327 return CONTAINER_OF(queue, struct hfsc_class, tc_queue);
3331 hfsc_install__(struct netdev *netdev_, uint32_t max_rate)
3333 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
3336 hfsc = xmalloc(sizeof *hfsc);
3337 tc_init(&hfsc->tc, &tc_ops_hfsc);
3338 hfsc->max_rate = max_rate;
3339 netdev->tc = &hfsc->tc;
3343 hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id,
3344 const struct hfsc_class *hc)
3348 struct hfsc_class *hcp;
3349 struct tc_queue *queue;
3351 hfsc = hfsc_get__(netdev);
3352 hash = hash_int(queue_id, 0);
3354 queue = tc_find_queue__(netdev, queue_id, hash);
3356 hcp = hfsc_class_cast__(queue);
3358 hcp = xmalloc(sizeof *hcp);
3359 queue = &hcp->tc_queue;
3360 queue->queue_id = queue_id;
3361 queue->created = time_msec();
3362 hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash);
3365 hcp->min_rate = hc->min_rate;
3366 hcp->max_rate = hc->max_rate;
3370 hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class)
3372 const struct tc_service_curve *rsc, *fsc, *usc;
3373 static const struct nl_policy tca_hfsc_policy[] = {
3375 .type = NL_A_UNSPEC,
3377 .min_len = sizeof(struct tc_service_curve),
3380 .type = NL_A_UNSPEC,
3382 .min_len = sizeof(struct tc_service_curve),
3385 .type = NL_A_UNSPEC,
3387 .min_len = sizeof(struct tc_service_curve),
3390 struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)];
3392 if (!nl_parse_nested(nl_options, tca_hfsc_policy,
3393 attrs, ARRAY_SIZE(tca_hfsc_policy))) {
3394 VLOG_WARN_RL(&rl, "failed to parse HFSC class options");
3398 rsc = nl_attr_get(attrs[TCA_HFSC_RSC]);
3399 fsc = nl_attr_get(attrs[TCA_HFSC_FSC]);
3400 usc = nl_attr_get(attrs[TCA_HFSC_USC]);
3402 if (rsc->m1 != 0 || rsc->d != 0 ||
3403 fsc->m1 != 0 || fsc->d != 0 ||
3404 usc->m1 != 0 || usc->d != 0) {
3405 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3406 "Non-linear service curves are not supported.");
3410 if (rsc->m2 != fsc->m2) {
3411 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3412 "Real-time service curves are not supported ");
3416 if (rsc->m2 > usc->m2) {
3417 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3418 "Min-rate service curve is greater than "
3419 "the max-rate service curve.");
3423 class->min_rate = fsc->m2;
3424 class->max_rate = usc->m2;
3429 hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
3430 struct hfsc_class *options,
3431 struct netdev_queue_stats *stats)
3434 unsigned int handle;
3435 struct nlattr *nl_options;
3437 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
3443 unsigned int major, minor;
3445 major = tc_get_major(handle);
3446 minor = tc_get_minor(handle);
3447 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
3448 *queue_id = minor - 1;
3455 error = hfsc_parse_tca_options__(nl_options, options);
3462 hfsc_query_class__(const struct netdev *netdev, unsigned int handle,
3463 unsigned int parent, struct hfsc_class *options,
3464 struct netdev_queue_stats *stats)
3467 struct ofpbuf *reply;
3469 error = tc_query_class(netdev, handle, parent, &reply);
3474 error = hfsc_parse_tcmsg__(reply, NULL, options, stats);
3475 ofpbuf_delete(reply);
3480 hfsc_parse_qdisc_details__(struct netdev *netdev_, const struct smap *details,
3481 struct hfsc_class *class)
3483 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
3485 const char *max_rate_s;
3487 max_rate_s = smap_get(details, "max-rate");
3488 max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
3491 enum netdev_features current;
3493 netdev_linux_read_features(netdev);
3494 current = !netdev->get_features_error ? netdev->current : 0;
3495 max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8;
3498 class->min_rate = max_rate;
3499 class->max_rate = max_rate;
3503 hfsc_parse_class_details__(struct netdev *netdev,
3504 const struct smap *details,
3505 struct hfsc_class * class)
3507 const struct hfsc *hfsc;
3508 uint32_t min_rate, max_rate;
3509 const char *min_rate_s, *max_rate_s;
3511 hfsc = hfsc_get__(netdev);
3512 min_rate_s = smap_get(details, "min-rate");
3513 max_rate_s = smap_get(details, "max-rate");
3515 min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
3516 min_rate = MAX(min_rate, 1);
3517 min_rate = MIN(min_rate, hfsc->max_rate);
3519 max_rate = (max_rate_s
3520 ? strtoull(max_rate_s, NULL, 10) / 8
3522 max_rate = MAX(max_rate, min_rate);
3523 max_rate = MIN(max_rate, hfsc->max_rate);
3525 class->min_rate = min_rate;
3526 class->max_rate = max_rate;
3531 /* Create an HFSC qdisc.
3533 * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */
3535 hfsc_setup_qdisc__(struct netdev * netdev)
3537 struct tcmsg *tcmsg;
3538 struct ofpbuf request;
3539 struct tc_hfsc_qopt opt;
3541 tc_del_qdisc(netdev);
3543 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
3544 NLM_F_EXCL | NLM_F_CREATE, &request);
3550 tcmsg->tcm_handle = tc_make_handle(1, 0);
3551 tcmsg->tcm_parent = TC_H_ROOT;
3553 memset(&opt, 0, sizeof opt);
3556 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3557 nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt);
3559 return tc_transact(&request, NULL);
3562 /* Create an HFSC class.
3564 * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc
3565 * sc rate <min_rate> ul rate <max_rate>" */
3567 hfsc_setup_class__(struct netdev *netdev, unsigned int handle,
3568 unsigned int parent, struct hfsc_class *class)
3572 struct tcmsg *tcmsg;
3573 struct ofpbuf request;
3574 struct tc_service_curve min, max;
3576 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
3582 tcmsg->tcm_handle = handle;
3583 tcmsg->tcm_parent = parent;
3587 min.m2 = class->min_rate;
3591 max.m2 = class->max_rate;
3593 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3594 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
3595 nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min);
3596 nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min);
3597 nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max);
3598 nl_msg_end_nested(&request, opt_offset);
3600 error = tc_transact(&request, NULL);
3602 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
3603 "min-rate %ubps, max-rate %ubps (%s)",
3604 netdev_get_name(netdev),
3605 tc_get_major(handle), tc_get_minor(handle),
3606 tc_get_major(parent), tc_get_minor(parent),
3607 class->min_rate, class->max_rate, ovs_strerror(error));
3614 hfsc_tc_install(struct netdev *netdev, const struct smap *details)
3617 struct hfsc_class class;
3619 error = hfsc_setup_qdisc__(netdev);
3625 hfsc_parse_qdisc_details__(netdev, details, &class);
3626 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3627 tc_make_handle(1, 0), &class);
3633 hfsc_install__(netdev, class.max_rate);
3638 hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3641 struct queue_dump_state state;
3642 struct hfsc_class hc;
3645 hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
3646 hfsc_install__(netdev, hc.max_rate);
3648 if (!start_queue_dump(netdev, &state)) {
3652 while (nl_dump_next(&state.dump, &msg, &state.buf)) {
3653 unsigned int queue_id;
3655 if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
3656 hfsc_update_queue__(netdev, queue_id, &hc);
3660 finish_queue_dump(&state);
3665 hfsc_tc_destroy(struct tc *tc)
3668 struct hfsc_class *hc, *next;
3670 hfsc = CONTAINER_OF(tc, struct hfsc, tc);
3672 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) {
3673 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3682 hfsc_qdisc_get(const struct netdev *netdev, struct smap *details)
3684 const struct hfsc *hfsc;
3685 hfsc = hfsc_get__(netdev);
3686 smap_add_format(details, "max-rate", "%llu", 8ULL * hfsc->max_rate);
3691 hfsc_qdisc_set(struct netdev *netdev, const struct smap *details)
3694 struct hfsc_class class;
3696 hfsc_parse_qdisc_details__(netdev, details, &class);
3697 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3698 tc_make_handle(1, 0), &class);
3701 hfsc_get__(netdev)->max_rate = class.max_rate;
3708 hfsc_class_get(const struct netdev *netdev OVS_UNUSED,
3709 const struct tc_queue *queue, struct smap *details)
3711 const struct hfsc_class *hc;
3713 hc = hfsc_class_cast__(queue);
3714 smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate);
3715 if (hc->min_rate != hc->max_rate) {
3716 smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate);
3722 hfsc_class_set(struct netdev *netdev, unsigned int queue_id,
3723 const struct smap *details)
3726 struct hfsc_class class;
3728 error = hfsc_parse_class_details__(netdev, details, &class);
3733 error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
3734 tc_make_handle(1, 0xfffe), &class);
3739 hfsc_update_queue__(netdev, queue_id, &class);
3744 hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue)
3748 struct hfsc_class *hc;
3750 hc = hfsc_class_cast__(queue);
3751 hfsc = hfsc_get__(netdev);
3753 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
3755 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3762 hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
3763 struct netdev_queue_stats *stats)
3765 return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
3766 tc_make_handle(1, 0xfffe), NULL, stats);
3770 hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
3771 const struct ofpbuf *nlmsg,
3772 netdev_dump_queue_stats_cb *cb, void *aux)
3774 struct netdev_queue_stats stats;
3775 unsigned int handle, major, minor;
3778 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
3783 major = tc_get_major(handle);
3784 minor = tc_get_minor(handle);
3785 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
3786 (*cb)(minor - 1, &stats, aux);
3791 static const struct tc_ops tc_ops_hfsc = {
3792 "hfsc", /* linux_name */
3793 "linux-hfsc", /* ovs_name */
3794 HFSC_N_QUEUES, /* n_queues */
3795 hfsc_tc_install, /* tc_install */
3796 hfsc_tc_load, /* tc_load */
3797 hfsc_tc_destroy, /* tc_destroy */
3798 hfsc_qdisc_get, /* qdisc_get */
3799 hfsc_qdisc_set, /* qdisc_set */
3800 hfsc_class_get, /* class_get */
3801 hfsc_class_set, /* class_set */
3802 hfsc_class_delete, /* class_delete */
3803 hfsc_class_get_stats, /* class_get_stats */
3804 hfsc_class_dump_stats /* class_dump_stats */
3807 /* "linux-default" traffic control class.
3809 * This class represents the default, unnamed Linux qdisc. It corresponds to
3810 * the "" (empty string) QoS type in the OVS database. */
3813 default_install__(struct netdev *netdev_)
3815 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
3816 static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_default);
3818 /* Nothing but a tc class implementation is allowed to write to a tc. This
3819 * class never does that, so we can legitimately use a const tc object. */
3820 netdev->tc = CONST_CAST(struct tc *, &tc);
3824 default_tc_install(struct netdev *netdev,
3825 const struct smap *details OVS_UNUSED)
3827 default_install__(netdev);
3832 default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3834 default_install__(netdev);
3838 static const struct tc_ops tc_ops_default = {
3839 NULL, /* linux_name */
3844 NULL, /* tc_destroy */
3845 NULL, /* qdisc_get */
3846 NULL, /* qdisc_set */
3847 NULL, /* class_get */
3848 NULL, /* class_set */
3849 NULL, /* class_delete */
3850 NULL, /* class_get_stats */
3851 NULL /* class_dump_stats */
3854 /* "linux-other" traffic control class.
3859 other_tc_load(struct netdev *netdev_, struct ofpbuf *nlmsg OVS_UNUSED)
3861 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
3862 static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_other);
3864 /* Nothing but a tc class implementation is allowed to write to a tc. This
3865 * class never does that, so we can legitimately use a const tc object. */
3866 netdev->tc = CONST_CAST(struct tc *, &tc);
3870 static const struct tc_ops tc_ops_other = {
3871 NULL, /* linux_name */
3872 "linux-other", /* ovs_name */
3874 NULL, /* tc_install */
3876 NULL, /* tc_destroy */
3877 NULL, /* qdisc_get */
3878 NULL, /* qdisc_set */
3879 NULL, /* class_get */
3880 NULL, /* class_set */
3881 NULL, /* class_delete */
3882 NULL, /* class_get_stats */
3883 NULL /* class_dump_stats */
3886 /* Traffic control. */
3888 /* Number of kernel "tc" ticks per second. */
3889 static double ticks_per_s;
3891 /* Number of kernel "jiffies" per second. This is used for the purpose of
3892 * computing buffer sizes. Generally kernel qdiscs need to be able to buffer
3893 * one jiffy's worth of data.
3895 * There are two possibilities here:
3897 * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the
3898 * approximate range of 100 to 1024. That means that we really need to
3899 * make sure that the qdisc can buffer that much data.
3901 * - 'buffer_hz' is an absurdly large number. That means that the kernel
3902 * has finely granular timers and there's no need to fudge additional room
3903 * for buffers. (There's no extra effort needed to implement that: the
3904 * large 'buffer_hz' is used as a divisor, so practically any number will
3905 * come out as 0 in the division. Small integer results in the case of
3906 * really high dividends won't have any real effect anyhow.)
3908 static unsigned int buffer_hz;
3910 /* Returns tc handle 'major':'minor'. */
3912 tc_make_handle(unsigned int major, unsigned int minor)
3914 return TC_H_MAKE(major << 16, minor);
3917 /* Returns the major number from 'handle'. */
3919 tc_get_major(unsigned int handle)
3921 return TC_H_MAJ(handle) >> 16;
3924 /* Returns the minor number from 'handle'. */
3926 tc_get_minor(unsigned int handle)
3928 return TC_H_MIN(handle);
3931 static struct tcmsg *
3932 tc_make_request(const struct netdev *netdev, int type, unsigned int flags,
3933 struct ofpbuf *request)
3935 struct tcmsg *tcmsg;
3939 error = get_ifindex(netdev, &ifindex);
3944 ofpbuf_init(request, 512);
3945 nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags);
3946 tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg);
3947 tcmsg->tcm_family = AF_UNSPEC;
3948 tcmsg->tcm_ifindex = ifindex;
3949 /* Caller should fill in tcmsg->tcm_handle. */
3950 /* Caller should fill in tcmsg->tcm_parent. */
3956 tc_transact(struct ofpbuf *request, struct ofpbuf **replyp)
3958 int error = nl_transact(NETLINK_ROUTE, request, replyp);
3959 ofpbuf_uninit(request);
3963 /* Adds or deletes a root ingress qdisc on 'netdev'. We use this for
3964 * policing configuration.
3966 * This function is equivalent to running the following when 'add' is true:
3967 * /sbin/tc qdisc add dev <devname> handle ffff: ingress
3969 * This function is equivalent to running the following when 'add' is false:
3970 * /sbin/tc qdisc del dev <devname> handle ffff: ingress
3972 * The configuration and stats may be seen with the following command:
3973 * /sbin/tc -s qdisc show dev <devname>
3975 * Returns 0 if successful, otherwise a positive errno value.
3978 tc_add_del_ingress_qdisc(struct netdev *netdev, bool add)
3980 struct ofpbuf request;
3981 struct tcmsg *tcmsg;
3983 int type = add ? RTM_NEWQDISC : RTM_DELQDISC;
3984 int flags = add ? NLM_F_EXCL | NLM_F_CREATE : 0;
3986 tcmsg = tc_make_request(netdev, type, flags, &request);
3990 tcmsg->tcm_handle = tc_make_handle(0xffff, 0);
3991 tcmsg->tcm_parent = TC_H_INGRESS;
3992 nl_msg_put_string(&request, TCA_KIND, "ingress");
3993 nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0);
3995 error = tc_transact(&request, NULL);
3997 /* If we're deleting the qdisc, don't worry about some of the
3998 * error conditions. */
3999 if (!add && (error == ENOENT || error == EINVAL)) {
4008 /* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size
4011 * This function is equivalent to running:
4012 * /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49
4013 * basic police rate <kbits_rate>kbit burst <kbits_burst>k
4016 * The configuration and stats may be seen with the following command:
4017 * /sbin/tc -s filter show <devname> eth0 parent ffff:
4019 * Returns 0 if successful, otherwise a positive errno value.
4022 tc_add_policer(struct netdev *netdev, int kbits_rate, int kbits_burst)
4024 struct tc_police tc_police;
4025 struct ofpbuf request;
4026 struct tcmsg *tcmsg;
4027 size_t basic_offset;
4028 size_t police_offset;
4032 memset(&tc_police, 0, sizeof tc_police);
4033 tc_police.action = TC_POLICE_SHOT;
4034 tc_police.mtu = mtu;
4035 tc_fill_rate(&tc_police.rate, ((uint64_t) kbits_rate * 1000)/8, mtu);
4036 tc_police.burst = tc_bytes_to_ticks(tc_police.rate.rate,
4037 kbits_burst * 1024);
4039 tcmsg = tc_make_request(netdev, RTM_NEWTFILTER,
4040 NLM_F_EXCL | NLM_F_CREATE, &request);
4044 tcmsg->tcm_parent = tc_make_handle(0xffff, 0);
4045 tcmsg->tcm_info = tc_make_handle(49,
4046 (OVS_FORCE uint16_t) htons(ETH_P_ALL));
4048 nl_msg_put_string(&request, TCA_KIND, "basic");
4049 basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
4050 police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE);
4051 nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police);
4052 tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate);
4053 nl_msg_end_nested(&request, police_offset);
4054 nl_msg_end_nested(&request, basic_offset);
4056 error = tc_transact(&request, NULL);
4067 /* The values in psched are not individually very meaningful, but they are
4068 * important. The tables below show some values seen in the wild.
4072 * - "c" has always been a constant 1000000 since at least Linux 2.4.14.
4073 * (Before that, there are hints that it was 1000000000.)
4075 * - "d" can be unrealistically large, see the comment on 'buffer_hz'
4079 * -----------------------------------
4080 * [1] 000c8000 000f4240 000f4240 00000064
4081 * [2] 000003e8 00000400 000f4240 3b9aca00
4082 * [3] 000003e8 00000400 000f4240 3b9aca00
4083 * [4] 000003e8 00000400 000f4240 00000064
4084 * [5] 000003e8 00000040 000f4240 3b9aca00
4085 * [6] 000003e8 00000040 000f4240 000000f9
4087 * a b c d ticks_per_s buffer_hz
4088 * ------- --------- ---------- ------------- ----------- -------------
4089 * [1] 819,200 1,000,000 1,000,000 100 819,200 100
4090 * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
4091 * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
4092 * [4] 1,000 1,024 1,000,000 100 976,562 100
4093 * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000
4094 * [6] 1,000 64 1,000,000 249 15,625,000 249
4096 * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p
4097 * [2] 2.6.26-1-686-bigmem from Debian lenny
4098 * [3] 2.6.26-2-sparc64 from Debian lenny
4099 * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p
4100 * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion
4101 * [6] 2.6.34 from kernel.org on KVM
4103 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
4104 static const char fn[] = "/proc/net/psched";
4105 unsigned int a, b, c, d;
4108 if (!ovsthread_once_start(&once)) {
4115 stream = fopen(fn, "r");
4117 VLOG_WARN("%s: open failed: %s", fn, ovs_strerror(errno));
4121 if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) {
4122 VLOG_WARN("%s: read failed", fn);
4126 VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d);
4130 VLOG_WARN("%s: invalid scheduler parameters", fn);
4134 ticks_per_s = (double) a * c / b;
4138 VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u",
4141 VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz);
4144 ovsthread_once_done(&once);
4147 /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a
4148 * rate of 'rate' bytes per second. */
4150 tc_ticks_to_bytes(unsigned int rate, unsigned int ticks)
4153 return (rate * ticks) / ticks_per_s;
4156 /* Returns the number of ticks that it would take to transmit 'size' bytes at a
4157 * rate of 'rate' bytes per second. */
4159 tc_bytes_to_ticks(unsigned int rate, unsigned int size)
4162 return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0;
4165 /* Returns the number of bytes that need to be reserved for qdisc buffering at
4166 * a transmission rate of 'rate' bytes per second. */
4168 tc_buffer_per_jiffy(unsigned int rate)
4171 return rate / buffer_hz;
4174 /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc,
4175 * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull,
4176 * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or
4177 * stores NULL into it if it is absent.
4179 * '*kind' and '*options' point into 'msg', so they are owned by whoever owns
4182 * Returns 0 if successful, otherwise a positive errno value. */
4184 tc_parse_qdisc(const struct ofpbuf *msg, const char **kind,
4185 struct nlattr **options)
4187 static const struct nl_policy tca_policy[] = {
4188 [TCA_KIND] = { .type = NL_A_STRING, .optional = false },
4189 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
4191 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
4193 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
4194 tca_policy, ta, ARRAY_SIZE(ta))) {
4195 VLOG_WARN_RL(&rl, "failed to parse qdisc message");
4200 *kind = nl_attr_get_string(ta[TCA_KIND]);
4204 *options = ta[TCA_OPTIONS];
4219 /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the
4220 * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute
4221 * into '*options', and its queue statistics into '*stats'. Any of the output
4222 * arguments may be null.
4224 * Returns 0 if successful, otherwise a positive errno value. */
4226 tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep,
4227 struct nlattr **options, struct netdev_queue_stats *stats)
4229 static const struct nl_policy tca_policy[] = {
4230 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false },
4231 [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false },
4233 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
4235 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
4236 tca_policy, ta, ARRAY_SIZE(ta))) {
4237 VLOG_WARN_RL(&rl, "failed to parse class message");
4242 struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc);
4243 *handlep = tc->tcm_handle;
4247 *options = ta[TCA_OPTIONS];
4251 const struct gnet_stats_queue *gsq;
4252 struct gnet_stats_basic gsb;
4254 static const struct nl_policy stats_policy[] = {
4255 [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false,
4256 .min_len = sizeof gsb },
4257 [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false,
4258 .min_len = sizeof *gsq },
4260 struct nlattr *sa[ARRAY_SIZE(stats_policy)];
4262 if (!nl_parse_nested(ta[TCA_STATS2], stats_policy,
4263 sa, ARRAY_SIZE(sa))) {
4264 VLOG_WARN_RL(&rl, "failed to parse class stats");
4268 /* Alignment issues screw up the length of struct gnet_stats_basic on
4269 * some arch/bitsize combinations. Newer versions of Linux have a
4270 * struct gnet_stats_basic_packed, but we can't depend on that. The
4271 * easiest thing to do is just to make a copy. */
4272 memset(&gsb, 0, sizeof gsb);
4273 memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]),
4274 MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb));
4275 stats->tx_bytes = gsb.bytes;
4276 stats->tx_packets = gsb.packets;
4278 gsq = nl_attr_get(sa[TCA_STATS_QUEUE]);
4279 stats->tx_errors = gsq->drops;
4289 memset(stats, 0, sizeof *stats);
4294 /* Queries the kernel for class with identifier 'handle' and parent 'parent'
4297 tc_query_class(const struct netdev *netdev,
4298 unsigned int handle, unsigned int parent,
4299 struct ofpbuf **replyp)
4301 struct ofpbuf request;
4302 struct tcmsg *tcmsg;
4305 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request);
4309 tcmsg->tcm_handle = handle;
4310 tcmsg->tcm_parent = parent;
4312 error = tc_transact(&request, replyp);
4314 VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)",
4315 netdev_get_name(netdev),
4316 tc_get_major(handle), tc_get_minor(handle),
4317 tc_get_major(parent), tc_get_minor(parent),
4318 ovs_strerror(error));
4323 /* Equivalent to "tc class del dev <name> handle <handle>". */
4325 tc_delete_class(const struct netdev *netdev, unsigned int handle)
4327 struct ofpbuf request;
4328 struct tcmsg *tcmsg;
4331 tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request);
4335 tcmsg->tcm_handle = handle;
4336 tcmsg->tcm_parent = 0;
4338 error = tc_transact(&request, NULL);
4340 VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)",
4341 netdev_get_name(netdev),
4342 tc_get_major(handle), tc_get_minor(handle),
4343 ovs_strerror(error));
4348 /* Equivalent to "tc qdisc del dev <name> root". */
4350 tc_del_qdisc(struct netdev *netdev_)
4352 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
4353 struct ofpbuf request;
4354 struct tcmsg *tcmsg;
4357 tcmsg = tc_make_request(netdev_, RTM_DELQDISC, 0, &request);
4361 tcmsg->tcm_handle = tc_make_handle(1, 0);
4362 tcmsg->tcm_parent = TC_H_ROOT;
4364 error = tc_transact(&request, NULL);
4365 if (error == EINVAL) {
4366 /* EINVAL probably means that the default qdisc was in use, in which
4367 * case we've accomplished our purpose. */
4370 if (!error && netdev->tc) {
4371 if (netdev->tc->ops->tc_destroy) {
4372 netdev->tc->ops->tc_destroy(netdev->tc);
4379 /* If 'netdev''s qdisc type and parameters are not yet known, queries the
4380 * kernel to determine what they are. Returns 0 if successful, otherwise a
4381 * positive errno value. */
4383 tc_query_qdisc(const struct netdev *netdev_)
4385 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
4386 struct ofpbuf request, *qdisc;
4387 const struct tc_ops *ops;
4388 struct tcmsg *tcmsg;
4396 /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have
4397 * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before
4398 * 2.6.35 without that fix backported to it.
4400 * To avoid the OOPS, we must not make a request that would attempt to dump
4401 * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a
4402 * few others. There are a few ways that I can see to do this, but most of
4403 * them seem to be racy (and if you lose the race the kernel OOPSes). The
4404 * technique chosen here is to assume that any non-default qdisc that we
4405 * create will have a class with handle 1:0. The built-in qdiscs only have
4406 * a class with handle 0:0.
4408 * We could check for Linux 2.6.35+ and use a more straightforward method
4410 tcmsg = tc_make_request(netdev_, RTM_GETQDISC, NLM_F_ECHO, &request);
4414 tcmsg->tcm_handle = tc_make_handle(1, 0);
4415 tcmsg->tcm_parent = 0;
4417 /* Figure out what tc class to instantiate. */
4418 error = tc_transact(&request, &qdisc);
4422 error = tc_parse_qdisc(qdisc, &kind, NULL);
4424 ops = &tc_ops_other;
4426 ops = tc_lookup_linux_name(kind);
4428 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1);
4429 VLOG_INFO_RL(&rl2, "unknown qdisc \"%s\"", kind);
4431 ops = &tc_ops_other;
4434 } else if (error == ENOENT) {
4435 /* Either it's a built-in qdisc, or it's a qdisc set up by some
4436 * other entity that doesn't have a handle 1:0. We will assume
4437 * that it's the system default qdisc. */
4438 ops = &tc_ops_default;
4441 /* Who knows? Maybe the device got deleted. */
4442 VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)",
4443 netdev_get_name(netdev_), ovs_strerror(error));
4444 ops = &tc_ops_other;
4447 /* Instantiate it. */
4448 load_error = ops->tc_load(CONST_CAST(struct netdev *, netdev_), qdisc);
4449 ovs_assert((load_error == 0) == (netdev->tc != NULL));
4450 ofpbuf_delete(qdisc);
4452 return error ? error : load_error;
4455 /* Linux traffic control uses tables with 256 entries ("rtab" tables) to
4456 approximate the time to transmit packets of various lengths. For an MTU of
4457 256 or less, each entry is exact; for an MTU of 257 through 512, each entry
4458 represents two possible packet lengths; for a MTU of 513 through 1024, four
4459 possible lengths; and so on.
4461 Returns, for the specified 'mtu', the number of bits that packet lengths
4462 need to be shifted right to fit within such a 256-entry table. */
4464 tc_calc_cell_log(unsigned int mtu)
4469 mtu = ETH_PAYLOAD_MAX;
4471 mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN;
4473 for (cell_log = 0; mtu >= 256; cell_log++) {
4480 /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU
4483 tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu)
4485 memset(rate, 0, sizeof *rate);
4486 rate->cell_log = tc_calc_cell_log(mtu);
4487 /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */
4488 /* rate->cell_align = 0; */ /* distro headers. */
4489 rate->mpu = ETH_TOTAL_MIN;
4493 /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink
4494 * attribute of the specified "type".
4496 * See tc_calc_cell_log() above for a description of "rtab"s. */
4498 tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate)
4503 rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE);
4504 for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) {
4505 unsigned packet_size = (i + 1) << rate->cell_log;
4506 if (packet_size < rate->mpu) {
4507 packet_size = rate->mpu;
4509 rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size);
4513 /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a
4514 * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested
4515 * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of
4518 tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes)
4520 unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu;
4521 return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst));
4524 /* Linux-only functions declared in netdev-linux.h */
4526 /* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If
4527 * 'enable' is true, the bit is set. Otherwise, it is cleared. */
4529 netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag,
4530 const char *flag_name, bool enable)
4532 const char *netdev_name = netdev_get_name(netdev);
4533 struct ethtool_value evalue;
4537 COVERAGE_INC(netdev_get_ethtool);
4538 memset(&evalue, 0, sizeof evalue);
4539 error = netdev_linux_do_ethtool(netdev_name,
4540 (struct ethtool_cmd *)&evalue,
4541 ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4546 COVERAGE_INC(netdev_set_ethtool);
4547 evalue.data = new_flags = (evalue.data & ~flag) | (enable ? flag : 0);
4548 error = netdev_linux_do_ethtool(netdev_name,
4549 (struct ethtool_cmd *)&evalue,
4550 ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS");
4555 COVERAGE_INC(netdev_get_ethtool);
4556 memset(&evalue, 0, sizeof evalue);
4557 error = netdev_linux_do_ethtool(netdev_name,
4558 (struct ethtool_cmd *)&evalue,
4559 ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4564 if (new_flags != evalue.data) {
4565 VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network "
4566 "device %s failed", enable ? "enable" : "disable",
4567 flag_name, netdev_name);
4574 /* Utility functions. */
4576 /* Copies 'src' into 'dst', performing format conversion in the process. */
4578 netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst,
4579 const struct rtnl_link_stats *src)
4581 dst->rx_packets = src->rx_packets;
4582 dst->tx_packets = src->tx_packets;
4583 dst->rx_bytes = src->rx_bytes;
4584 dst->tx_bytes = src->tx_bytes;
4585 dst->rx_errors = src->rx_errors;
4586 dst->tx_errors = src->tx_errors;
4587 dst->rx_dropped = src->rx_dropped;
4588 dst->tx_dropped = src->tx_dropped;
4589 dst->multicast = src->multicast;
4590 dst->collisions = src->collisions;
4591 dst->rx_length_errors = src->rx_length_errors;
4592 dst->rx_over_errors = src->rx_over_errors;
4593 dst->rx_crc_errors = src->rx_crc_errors;
4594 dst->rx_frame_errors = src->rx_frame_errors;
4595 dst->rx_fifo_errors = src->rx_fifo_errors;
4596 dst->rx_missed_errors = src->rx_missed_errors;
4597 dst->tx_aborted_errors = src->tx_aborted_errors;
4598 dst->tx_carrier_errors = src->tx_carrier_errors;
4599 dst->tx_fifo_errors = src->tx_fifo_errors;
4600 dst->tx_heartbeat_errors = src->tx_heartbeat_errors;
4601 dst->tx_window_errors = src->tx_window_errors;
4605 get_stats_via_netlink(const struct netdev *netdev_, struct netdev_stats *stats)
4607 struct ofpbuf request;
4608 struct ofpbuf *reply;
4611 ofpbuf_init(&request, 0);
4612 nl_msg_put_nlmsghdr(&request,
4613 sizeof(struct ifinfomsg) + NL_ATTR_SIZE(IFNAMSIZ),
4614 RTM_GETLINK, NLM_F_REQUEST);
4615 ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg));
4616 nl_msg_put_string(&request, IFLA_IFNAME, netdev_get_name(netdev_));
4617 error = nl_transact(NETLINK_ROUTE, &request, &reply);
4618 ofpbuf_uninit(&request);
4623 if (ofpbuf_try_pull(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg))) {
4624 const struct nlattr *a = nl_attr_find(reply, 0, IFLA_STATS);
4625 if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats)) {
4626 netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(a));
4629 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
4633 VLOG_WARN_RL(&rl, "short RTM_GETLINK reply");
4638 ofpbuf_delete(reply);
4643 get_flags(const struct netdev *dev, unsigned int *flags)
4649 error = af_inet_ifreq_ioctl(dev->name, &ifr, SIOCGIFFLAGS, "SIOCGIFFLAGS");
4651 *flags = ifr.ifr_flags;
4657 set_flags(const char *name, unsigned int flags)
4661 ifr.ifr_flags = flags;
4662 return af_inet_ifreq_ioctl(name, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS");
4666 do_get_ifindex(const char *netdev_name)
4671 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4672 COVERAGE_INC(netdev_get_ifindex);
4674 error = af_inet_ioctl(SIOCGIFINDEX, &ifr);
4676 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
4677 netdev_name, ovs_strerror(error));
4680 return ifr.ifr_ifindex;
4684 get_ifindex(const struct netdev *netdev_, int *ifindexp)
4686 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
4688 if (!(netdev->cache_valid & VALID_IFINDEX)) {
4689 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
4692 netdev->get_ifindex_error = -ifindex;
4693 netdev->ifindex = 0;
4695 netdev->get_ifindex_error = 0;
4696 netdev->ifindex = ifindex;
4698 netdev->cache_valid |= VALID_IFINDEX;
4701 *ifindexp = netdev->ifindex;
4702 return netdev->get_ifindex_error;
4706 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
4712 memset(&ifr, 0, sizeof ifr);
4713 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4714 COVERAGE_INC(netdev_get_hwaddr);
4715 error = af_inet_ioctl(SIOCGIFHWADDR, &ifr);
4717 /* ENODEV probably means that a vif disappeared asynchronously and
4718 * hasn't been removed from the database yet, so reduce the log level
4719 * to INFO for that case. */
4720 VLOG(error == ENODEV ? VLL_INFO : VLL_ERR,
4721 "ioctl(SIOCGIFHWADDR) on %s device failed: %s",
4722 netdev_name, ovs_strerror(error));
4725 hwaddr_family = ifr.ifr_hwaddr.sa_family;
4726 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
4727 VLOG_WARN("%s device has unknown hardware address family %d",
4728 netdev_name, hwaddr_family);
4730 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
4735 set_etheraddr(const char *netdev_name,
4736 const uint8_t mac[ETH_ADDR_LEN])
4741 memset(&ifr, 0, sizeof ifr);
4742 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4743 ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
4744 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
4745 COVERAGE_INC(netdev_set_hwaddr);
4746 error = af_inet_ioctl(SIOCSIFHWADDR, &ifr);
4748 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
4749 netdev_name, ovs_strerror(error));
4755 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
4756 int cmd, const char *cmd_name)
4761 memset(&ifr, 0, sizeof ifr);
4762 ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
4763 ifr.ifr_data = (caddr_t) ecmd;
4766 error = af_inet_ioctl(SIOCETHTOOL, &ifr);
4768 if (error != EOPNOTSUPP) {
4769 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
4770 "failed: %s", cmd_name, name, ovs_strerror(error));
4772 /* The device doesn't support this operation. That's pretty
4773 * common, so there's no point in logging anything. */
4780 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
4781 int cmd, const char *cmd_name)
4786 ifr.ifr_addr.sa_family = AF_INET;
4787 error = af_inet_ifreq_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
4789 const struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *,
4791 *ip = sin->sin_addr;
4796 /* Returns an AF_PACKET raw socket or a negative errno value. */
4798 af_packet_sock(void)
4800 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
4803 if (ovsthread_once_start(&once)) {
4804 sock = socket(AF_PACKET, SOCK_RAW, 0);
4806 int error = set_nonblocking(sock);
4813 VLOG_ERR("failed to create packet socket: %s",
4814 ovs_strerror(errno));
4816 ovsthread_once_done(&once);