2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 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 <linux/version.h>
37 #include <sys/types.h>
38 #include <sys/ioctl.h>
39 #include <sys/socket.h>
40 #include <netpacket/packet.h>
42 #include <net/if_arp.h>
43 #include <net/if_packet.h>
44 #include <net/route.h>
45 #include <netinet/in.h>
52 #include "dpif-linux.h"
53 #include "dynamic-string.h"
54 #include "fatal-signal.h"
57 #include "netdev-provider.h"
58 #include "netdev-vport.h"
59 #include "netlink-notifier.h"
60 #include "netlink-socket.h"
63 #include "openflow/openflow.h"
65 #include "poll-loop.h"
66 #include "rtnetlink-link.h"
68 #include "socket-util.h"
71 #include "unaligned.h"
74 VLOG_DEFINE_THIS_MODULE(netdev_linux);
76 COVERAGE_DEFINE(netdev_set_policing);
77 COVERAGE_DEFINE(netdev_arp_lookup);
78 COVERAGE_DEFINE(netdev_get_ifindex);
79 COVERAGE_DEFINE(netdev_get_hwaddr);
80 COVERAGE_DEFINE(netdev_set_hwaddr);
81 COVERAGE_DEFINE(netdev_get_ethtool);
82 COVERAGE_DEFINE(netdev_set_ethtool);
85 /* These were introduced in Linux 2.6.14, so they might be missing if we have
87 #ifndef ADVERTISED_Pause
88 #define ADVERTISED_Pause (1 << 13)
90 #ifndef ADVERTISED_Asym_Pause
91 #define ADVERTISED_Asym_Pause (1 << 14)
94 /* These were introduced in Linux 2.6.24, so they might be missing if we
95 * have old headers. */
96 #ifndef ETHTOOL_GFLAGS
97 #define ETHTOOL_GFLAGS 0x00000025 /* Get flags bitmap(ethtool_value) */
99 #ifndef ETHTOOL_SFLAGS
100 #define ETHTOOL_SFLAGS 0x00000026 /* Set flags bitmap(ethtool_value) */
103 /* This was introduced in Linux 2.6.25, so it might be missing if we have old
106 #define TC_RTAB_SIZE 1024
109 static struct nln_notifier *netdev_linux_cache_notifier = NULL;
110 static int cache_notifier_refcount;
113 VALID_IFINDEX = 1 << 0,
114 VALID_ETHERADDR = 1 << 1,
118 VALID_POLICING = 1 << 5,
119 VALID_VPORT_STAT_ERROR = 1 << 6,
120 VALID_DRVINFO = 1 << 7,
121 VALID_FEATURES = 1 << 8,
124 /* Traffic control. */
126 /* An instance of a traffic control class. Always associated with a particular
129 * Each TC implementation subclasses this with whatever additional data it
132 const struct tc_ops *ops;
133 struct hmap queues; /* Contains "struct tc_queue"s.
134 * Read by generic TC layer.
135 * Written only by TC implementation. */
138 #define TC_INITIALIZER(TC, OPS) { OPS, HMAP_INITIALIZER(&(TC)->queues) }
140 /* One traffic control queue.
142 * Each TC implementation subclasses this with whatever additional data it
145 struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */
146 unsigned int queue_id; /* OpenFlow queue ID. */
147 long long int created; /* Time queue was created, in msecs. */
150 /* A particular kind of traffic control. Each implementation generally maps to
151 * one particular Linux qdisc class.
153 * The functions below return 0 if successful or a positive errno value on
154 * failure, except where otherwise noted. All of them must be provided, except
155 * where otherwise noted. */
157 /* Name used by kernel in the TCA_KIND attribute of tcmsg, e.g. "htb".
158 * This is null for tc_ops_default and tc_ops_other, for which there are no
159 * appropriate values. */
160 const char *linux_name;
162 /* Name used in OVS database, e.g. "linux-htb". Must be nonnull. */
163 const char *ovs_name;
165 /* Number of supported OpenFlow queues, 0 for qdiscs that have no
166 * queues. The queues are numbered 0 through n_queues - 1. */
167 unsigned int n_queues;
169 /* Called to install this TC class on 'netdev'. The implementation should
170 * make the Netlink calls required to set up 'netdev' with the right qdisc
171 * and configure it according to 'details'. The implementation may assume
172 * that the current qdisc is the default; that is, there is no need for it
173 * to delete the current qdisc before installing itself.
175 * The contents of 'details' should be documented as valid for 'ovs_name'
176 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
177 * (which is built as ovs-vswitchd.conf.db(8)).
179 * This function must return 0 if and only if it sets 'netdev->tc' to an
180 * initialized 'struct tc'.
182 * (This function is null for tc_ops_other, which cannot be installed. For
183 * other TC classes it should always be nonnull.) */
184 int (*tc_install)(struct netdev *netdev, const struct smap *details);
186 /* Called when the netdev code determines (through a Netlink query) that
187 * this TC class's qdisc is installed on 'netdev', but we didn't install
188 * it ourselves and so don't know any of the details.
190 * 'nlmsg' is the kernel reply to a RTM_GETQDISC Netlink message for
191 * 'netdev'. The TCA_KIND attribute of 'nlmsg' is 'linux_name'. The
192 * implementation should parse the other attributes of 'nlmsg' as
193 * necessary to determine its configuration. If necessary it should also
194 * use Netlink queries to determine the configuration of queues on
197 * This function must return 0 if and only if it sets 'netdev->tc' to an
198 * initialized 'struct tc'. */
199 int (*tc_load)(struct netdev *netdev, struct ofpbuf *nlmsg);
201 /* Destroys the data structures allocated by the implementation as part of
202 * 'tc'. (This includes destroying 'tc->queues' by calling
205 * The implementation should not need to perform any Netlink calls. If
206 * desirable, the caller is responsible for deconfiguring the kernel qdisc.
207 * (But it may not be desirable.)
209 * This function may be null if 'tc' is trivial. */
210 void (*tc_destroy)(struct tc *tc);
212 /* Retrieves details of 'netdev->tc' configuration into 'details'.
214 * The implementation should not need to perform any Netlink calls, because
215 * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have
216 * cached the configuration.
218 * The contents of 'details' should be documented as valid for 'ovs_name'
219 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
220 * (which is built as ovs-vswitchd.conf.db(8)).
222 * This function may be null if 'tc' is not configurable.
224 int (*qdisc_get)(const struct netdev *netdev, struct smap *details);
226 /* Reconfigures 'netdev->tc' according to 'details', performing any
227 * required Netlink calls to complete the reconfiguration.
229 * The contents of 'details' should be documented as valid for 'ovs_name'
230 * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml
231 * (which is built as ovs-vswitchd.conf.db(8)).
233 * This function may be null if 'tc' is not configurable.
235 int (*qdisc_set)(struct netdev *, const struct smap *details);
237 /* Retrieves details of 'queue' on 'netdev->tc' into 'details'. 'queue' is
238 * one of the 'struct tc_queue's within 'netdev->tc->queues'.
240 * The contents of 'details' should be documented as valid for 'ovs_name'
241 * in the "other_config" column in the "Queue" table in
242 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
244 * The implementation should not need to perform any Netlink calls, because
245 * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have
246 * cached the queue configuration.
248 * This function may be null if 'tc' does not have queues ('n_queues' is
250 int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue,
251 struct smap *details);
253 /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to
254 * 'details', perfoming any required Netlink calls to complete the
255 * reconfiguration. The caller ensures that 'queue_id' is less than
258 * The contents of 'details' should be documented as valid for 'ovs_name'
259 * in the "other_config" column in the "Queue" table in
260 * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)).
262 * This function may be null if 'tc' does not have queues or its queues are
263 * not configurable. */
264 int (*class_set)(struct netdev *, unsigned int queue_id,
265 const struct smap *details);
267 /* Deletes 'queue' from 'netdev->tc'. 'queue' is one of the 'struct
268 * tc_queue's within 'netdev->tc->queues'.
270 * This function may be null if 'tc' does not have queues or its queues
271 * cannot be deleted. */
272 int (*class_delete)(struct netdev *, struct tc_queue *queue);
274 /* Obtains stats for 'queue' from 'netdev->tc'. 'queue' is one of the
275 * 'struct tc_queue's within 'netdev->tc->queues'.
277 * On success, initializes '*stats'.
279 * This function may be null if 'tc' does not have queues or if it cannot
280 * report queue statistics. */
281 int (*class_get_stats)(const struct netdev *netdev,
282 const struct tc_queue *queue,
283 struct netdev_queue_stats *stats);
285 /* Extracts queue stats from 'nlmsg', which is a response to a
286 * RTM_GETTCLASS message, and passes them to 'cb' along with 'aux'.
288 * This function may be null if 'tc' does not have queues or if it cannot
289 * report queue statistics. */
290 int (*class_dump_stats)(const struct netdev *netdev,
291 const struct ofpbuf *nlmsg,
292 netdev_dump_queue_stats_cb *cb, void *aux);
296 tc_init(struct tc *tc, const struct tc_ops *ops)
299 hmap_init(&tc->queues);
303 tc_destroy(struct tc *tc)
305 hmap_destroy(&tc->queues);
308 static const struct tc_ops tc_ops_htb;
309 static const struct tc_ops tc_ops_hfsc;
310 static const struct tc_ops tc_ops_default;
311 static const struct tc_ops tc_ops_other;
313 static const struct tc_ops *const tcs[] = {
314 &tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */
315 &tc_ops_hfsc, /* Hierarchical fair service curve. */
316 &tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */
317 &tc_ops_other, /* Some other qdisc. */
321 static unsigned int tc_make_handle(unsigned int major, unsigned int minor);
322 static unsigned int tc_get_major(unsigned int handle);
323 static unsigned int tc_get_minor(unsigned int handle);
325 static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks);
326 static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size);
327 static unsigned int tc_buffer_per_jiffy(unsigned int rate);
329 static struct tcmsg *tc_make_request(const struct netdev *, int type,
330 unsigned int flags, struct ofpbuf *);
331 static int tc_transact(struct ofpbuf *request, struct ofpbuf **replyp);
332 static int tc_add_del_ingress_qdisc(struct netdev *netdev, bool add);
333 static int tc_add_policer(struct netdev *netdev, int kbits_rate,
336 static int tc_parse_qdisc(const struct ofpbuf *, const char **kind,
337 struct nlattr **options);
338 static int tc_parse_class(const struct ofpbuf *, unsigned int *queue_id,
339 struct nlattr **options,
340 struct netdev_queue_stats *);
341 static int tc_query_class(const struct netdev *,
342 unsigned int handle, unsigned int parent,
343 struct ofpbuf **replyp);
344 static int tc_delete_class(const struct netdev *, unsigned int handle);
346 static int tc_del_qdisc(struct netdev *netdev);
347 static int tc_query_qdisc(const struct netdev *netdev);
349 static int tc_calc_cell_log(unsigned int mtu);
350 static void tc_fill_rate(struct tc_ratespec *rate, uint64_t bps, int mtu);
351 static void tc_put_rtab(struct ofpbuf *, uint16_t type,
352 const struct tc_ratespec *rate);
353 static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes);
355 struct netdev_linux {
358 unsigned int cache_valid;
359 unsigned int change_seq;
361 bool miimon; /* Link status of last poll. */
362 long long int miimon_interval; /* Miimon Poll rate. Disabled if <= 0. */
363 struct timer miimon_timer;
365 /* The following are figured out "on demand" only. They are only valid
366 * when the corresponding VALID_* bit in 'cache_valid' is set. */
368 uint8_t etheraddr[ETH_ADDR_LEN];
369 struct in_addr address, netmask;
372 unsigned int ifi_flags;
373 long long int carrier_resets;
374 uint32_t kbits_rate; /* Policing data. */
375 uint32_t kbits_burst;
376 int vport_stats_error; /* Cached error code from vport_get_stats().
377 0 or an errno value. */
378 int netdev_mtu_error; /* Cached error code from SIOCGIFMTU or SIOCSIFMTU. */
379 int ether_addr_error; /* Cached error code from set/get etheraddr. */
380 int netdev_policing_error; /* Cached error code from set policing. */
381 int get_features_error; /* Cached error code from ETHTOOL_GSET. */
382 int get_ifindex_error; /* Cached error code from SIOCGIFINDEX. */
384 enum netdev_features current; /* Cached from ETHTOOL_GSET. */
385 enum netdev_features advertised; /* Cached from ETHTOOL_GSET. */
386 enum netdev_features supported; /* Cached from ETHTOOL_GSET. */
388 struct ethtool_drvinfo drvinfo; /* Cached from ETHTOOL_GDRVINFO. */
391 /* For devices of class netdev_tap_class only. */
395 struct netdev_rx_linux {
401 static const struct netdev_rx_class netdev_rx_linux_class;
403 /* Sockets used for ioctl operations. */
404 static int af_inet_sock = -1; /* AF_INET, SOCK_DGRAM. */
406 /* This is set pretty low because we probably won't learn anything from the
407 * additional log messages. */
408 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
410 static int netdev_linux_init(void);
412 static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
413 int cmd, const char *cmd_name);
414 static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
415 const char *cmd_name);
416 static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
417 int cmd, const char *cmd_name);
418 static int get_flags(const struct netdev *, unsigned int *flags);
419 static int set_flags(const char *, unsigned int flags);
420 static int do_get_ifindex(const char *netdev_name);
421 static int get_ifindex(const struct netdev *, int *ifindexp);
422 static int do_set_addr(struct netdev *netdev,
423 int ioctl_nr, const char *ioctl_name,
424 struct in_addr addr);
425 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
426 static int set_etheraddr(const char *netdev_name, const uint8_t[ETH_ADDR_LEN]);
427 static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
428 static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
429 static int af_packet_sock(void);
430 static void netdev_linux_miimon_run(void);
431 static void netdev_linux_miimon_wait(void);
434 is_netdev_linux_class(const struct netdev_class *netdev_class)
436 return netdev_class->init == netdev_linux_init;
440 is_tap_netdev(const struct netdev *netdev)
442 return netdev_get_class(netdev) == &netdev_tap_class;
445 static struct netdev_linux *
446 netdev_linux_cast(const struct netdev *netdev)
448 ovs_assert(is_netdev_linux_class(netdev_get_class(netdev)));
450 return CONTAINER_OF(netdev, struct netdev_linux, up);
453 static struct netdev_rx_linux *
454 netdev_rx_linux_cast(const struct netdev_rx *rx)
456 netdev_rx_assert_class(rx, &netdev_rx_linux_class);
457 return CONTAINER_OF(rx, struct netdev_rx_linux, up);
461 netdev_linux_init(void)
463 static int status = -1;
465 /* Create AF_INET socket. */
466 af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
467 status = af_inet_sock >= 0 ? 0 : errno;
469 VLOG_ERR("failed to create inet socket: %s", ovs_strerror(status));
476 netdev_linux_run(void)
478 rtnetlink_link_run();
479 netdev_linux_miimon_run();
483 netdev_linux_wait(void)
485 rtnetlink_link_wait();
486 netdev_linux_miimon_wait();
490 netdev_linux_changed(struct netdev_linux *dev,
491 unsigned int ifi_flags, unsigned int mask)
494 if (!dev->change_seq) {
498 if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) {
499 dev->carrier_resets++;
501 dev->ifi_flags = ifi_flags;
503 dev->cache_valid &= mask;
507 netdev_linux_update(struct netdev_linux *dev,
508 const struct rtnetlink_link_change *change)
510 if (change->nlmsg_type == RTM_NEWLINK) {
512 netdev_linux_changed(dev, change->ifi_flags, VALID_DRVINFO);
514 /* Update netdev from rtnl-change msg. */
516 dev->mtu = change->mtu;
517 dev->cache_valid |= VALID_MTU;
518 dev->netdev_mtu_error = 0;
521 if (!eth_addr_is_zero(change->addr)) {
522 memcpy(dev->etheraddr, change->addr, ETH_ADDR_LEN);
523 dev->cache_valid |= VALID_ETHERADDR;
524 dev->ether_addr_error = 0;
527 dev->ifindex = change->ifi_index;
528 dev->cache_valid |= VALID_IFINDEX;
529 dev->get_ifindex_error = 0;
532 netdev_linux_changed(dev, change->ifi_flags, 0);
537 netdev_linux_cache_cb(const struct rtnetlink_link_change *change,
538 void *aux OVS_UNUSED)
540 struct netdev_linux *dev;
542 struct netdev *base_dev = netdev_from_name(change->ifname);
543 if (base_dev && is_netdev_linux_class(netdev_get_class(base_dev))) {
544 netdev_linux_update(netdev_linux_cast(base_dev), change);
547 struct shash device_shash;
548 struct shash_node *node;
550 shash_init(&device_shash);
551 netdev_get_devices(&netdev_linux_class, &device_shash);
552 SHASH_FOR_EACH (node, &device_shash) {
553 struct netdev *netdev = node->data;
556 dev = netdev_linux_cast(netdev);
558 get_flags(&dev->up, &flags);
559 netdev_linux_changed(dev, flags, 0);
561 shash_destroy(&device_shash);
566 cache_notifier_ref(void)
568 if (!cache_notifier_refcount) {
569 ovs_assert(!netdev_linux_cache_notifier);
571 netdev_linux_cache_notifier =
572 rtnetlink_link_notifier_create(netdev_linux_cache_cb, NULL);
574 if (!netdev_linux_cache_notifier) {
578 cache_notifier_refcount++;
584 cache_notifier_unref(void)
586 ovs_assert(cache_notifier_refcount > 0);
587 if (!--cache_notifier_refcount) {
588 ovs_assert(netdev_linux_cache_notifier);
589 rtnetlink_link_notifier_destroy(netdev_linux_cache_notifier);
590 netdev_linux_cache_notifier = NULL;
594 /* Creates system and internal devices. */
596 netdev_linux_create(const struct netdev_class *class, const char *name,
597 struct netdev **netdevp)
599 struct netdev_linux *netdev;
602 error = cache_notifier_ref();
607 netdev = xzalloc(sizeof *netdev);
608 netdev->change_seq = 1;
609 netdev_init(&netdev->up, name, class);
610 error = get_flags(&netdev->up, &netdev->ifi_flags);
611 if (error == ENODEV) {
612 if (class != &netdev_internal_class) {
613 /* The device does not exist, so don't allow it to be opened. */
614 netdev_uninit(&netdev->up, false);
615 cache_notifier_unref();
619 /* "Internal" netdevs have to be created as netdev objects before
620 * they exist in the kernel, because creating them in the kernel
621 * happens by passing a netdev object to dpif_port_add().
622 * Therefore, ignore the error. */
626 *netdevp = &netdev->up;
630 /* For most types of netdevs we open the device for each call of
631 * netdev_open(). However, this is not the case with tap devices,
632 * since it is only possible to open the device once. In this
633 * situation we share a single file descriptor, and consequently
634 * buffers, across all readers. Therefore once data is read it will
635 * be unavailable to other reads for tap devices. */
637 netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED,
638 const char *name, struct netdev **netdevp)
640 struct netdev_linux *netdev;
641 static const char tap_dev[] = "/dev/net/tun";
645 netdev = xzalloc(sizeof *netdev);
646 netdev->change_seq = 1;
648 error = cache_notifier_ref();
653 /* Open tap device. */
654 netdev->tap_fd = open(tap_dev, O_RDWR);
655 if (netdev->tap_fd < 0) {
657 VLOG_WARN("opening \"%s\" failed: %s", tap_dev, ovs_strerror(error));
658 goto error_unref_notifier;
661 /* Create tap device. */
662 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
663 ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
664 if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) {
665 VLOG_WARN("%s: creating tap device failed: %s", name,
666 ovs_strerror(errno));
671 /* Make non-blocking. */
672 error = set_nonblocking(netdev->tap_fd);
677 netdev_init(&netdev->up, name, &netdev_tap_class);
678 *netdevp = &netdev->up;
682 close(netdev->tap_fd);
683 error_unref_notifier:
684 cache_notifier_unref();
691 netdev_linux_destroy(struct netdev *netdev_)
693 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
695 if (netdev->tc && netdev->tc->ops->tc_destroy) {
696 netdev->tc->ops->tc_destroy(netdev->tc);
699 if (netdev_get_class(netdev_) == &netdev_tap_class
700 && netdev->tap_fd >= 0)
702 close(netdev->tap_fd);
706 cache_notifier_unref();
710 netdev_linux_rx_open(struct netdev *netdev_, struct netdev_rx **rxp)
712 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
713 bool is_tap = is_tap_netdev(netdev_);
714 struct netdev_rx_linux *rx;
721 struct sockaddr_ll sll;
723 /* Result of tcpdump -dd inbound */
724 static struct sock_filter filt[] = {
725 { 0x28, 0, 0, 0xfffff004 }, /* ldh [0] */
726 { 0x15, 0, 1, 0x00000004 }, /* jeq #4 jt 2 jf 3 */
727 { 0x6, 0, 0, 0x00000000 }, /* ret #0 */
728 { 0x6, 0, 0, 0x0000ffff } /* ret #65535 */
730 static struct sock_fprog fprog = { ARRAY_SIZE(filt), filt };
732 /* Create file descriptor. */
733 fd = socket(PF_PACKET, SOCK_RAW, 0);
736 VLOG_ERR("failed to create raw socket (%s)", ovs_strerror(error));
740 /* Set non-blocking mode. */
741 error = set_nonblocking(fd);
746 /* Get ethernet device index. */
747 error = get_ifindex(&netdev->up, &ifindex);
752 /* Bind to specific ethernet device. */
753 memset(&sll, 0, sizeof sll);
754 sll.sll_family = AF_PACKET;
755 sll.sll_ifindex = ifindex;
756 sll.sll_protocol = (OVS_FORCE unsigned short int) htons(ETH_P_ALL);
757 if (bind(fd, (struct sockaddr *) &sll, sizeof sll) < 0) {
759 VLOG_ERR("%s: failed to bind raw socket (%s)",
760 netdev_get_name(netdev_), ovs_strerror(error));
764 /* Filter for only inbound packets. */
765 error = setsockopt(fd, SOL_SOCKET, SO_ATTACH_FILTER, &fprog,
769 VLOG_ERR("%s: failed attach filter (%s)",
770 netdev_get_name(netdev_), ovs_strerror(error));
775 rx = xmalloc(sizeof *rx);
776 netdev_rx_init(&rx->up, netdev_, &netdev_rx_linux_class);
791 netdev_rx_linux_destroy(struct netdev_rx *rx_)
793 struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_);
802 netdev_rx_linux_recv(struct netdev_rx *rx_, void *data, size_t size)
804 struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_);
809 ? read(rx->fd, data, size)
810 : recv(rx->fd, data, size, MSG_TRUNC));
811 } while (retval < 0 && errno == EINTR);
814 return retval > size ? -EMSGSIZE : retval;
816 if (errno != EAGAIN) {
817 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
818 ovs_strerror(errno), netdev_rx_get_name(rx_));
825 netdev_rx_linux_wait(struct netdev_rx *rx_)
827 struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_);
828 poll_fd_wait(rx->fd, POLLIN);
832 netdev_rx_linux_drain(struct netdev_rx *rx_)
834 struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_);
837 int error = netdev_linux_do_ioctl(netdev_rx_get_name(rx_), &ifr,
838 SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
842 drain_fd(rx->fd, ifr.ifr_qlen);
845 return drain_rcvbuf(rx->fd);
849 /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive
850 * errno value. Returns EAGAIN without blocking if the packet cannot be queued
851 * immediately. Returns EMSGSIZE if a partial packet was transmitted or if
852 * the packet is too big or too small to transmit on the device.
854 * The caller retains ownership of 'buffer' in all cases.
856 * The kernel maintains a packet transmission queue, so the caller is not
857 * expected to do additional queuing of packets. */
859 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
864 if (!is_tap_netdev(netdev_)) {
865 /* Use our AF_PACKET socket to send to this device. */
866 struct sockaddr_ll sll;
873 sock = af_packet_sock();
878 error = get_ifindex(netdev_, &ifindex);
883 /* We don't bother setting most fields in sockaddr_ll because the
884 * kernel ignores them for SOCK_RAW. */
885 memset(&sll, 0, sizeof sll);
886 sll.sll_family = AF_PACKET;
887 sll.sll_ifindex = ifindex;
889 iov.iov_base = CONST_CAST(void *, data);
893 msg.msg_namelen = sizeof sll;
896 msg.msg_control = NULL;
897 msg.msg_controllen = 0;
900 retval = sendmsg(sock, &msg, 0);
902 /* Use the tap fd to send to this device. This is essential for
903 * tap devices, because packets sent to a tap device with an
904 * AF_PACKET socket will loop back to be *received* again on the
905 * tap device. This doesn't occur on other interface types
906 * because we attach a socket filter to the rx socket. */
907 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
909 retval = write(netdev->tap_fd, data, size);
913 /* The Linux AF_PACKET implementation never blocks waiting for room
914 * for packets, instead returning ENOBUFS. Translate this into
915 * EAGAIN for the caller. */
916 if (errno == ENOBUFS) {
918 } else if (errno == EINTR) {
920 } else if (errno != EAGAIN) {
921 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
922 netdev_get_name(netdev_), ovs_strerror(errno));
925 } else if (retval != size) {
926 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
927 "%zu) on %s", retval, size, netdev_get_name(netdev_));
935 /* Registers with the poll loop to wake up from the next call to poll_block()
936 * when the packet transmission queue has sufficient room to transmit a packet
937 * with netdev_send().
939 * The kernel maintains a packet transmission queue, so the client is not
940 * expected to do additional queuing of packets. Thus, this function is
941 * unlikely to ever be used. It is included for completeness. */
943 netdev_linux_send_wait(struct netdev *netdev)
945 if (is_tap_netdev(netdev)) {
946 /* TAP device always accepts packets.*/
947 poll_immediate_wake();
951 /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
952 * otherwise a positive errno value. */
954 netdev_linux_set_etheraddr(struct netdev *netdev_,
955 const uint8_t mac[ETH_ADDR_LEN])
957 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
958 struct netdev_saved_flags *sf = NULL;
961 if (netdev->cache_valid & VALID_ETHERADDR) {
962 if (netdev->ether_addr_error) {
963 return netdev->ether_addr_error;
965 if (eth_addr_equals(netdev->etheraddr, mac)) {
968 netdev->cache_valid &= ~VALID_ETHERADDR;
971 /* Tap devices must be brought down before setting the address. */
972 if (is_tap_netdev(netdev_)) {
973 netdev_turn_flags_off(netdev_, NETDEV_UP, &sf);
975 error = set_etheraddr(netdev_get_name(netdev_), mac);
976 if (!error || error == ENODEV) {
977 netdev->ether_addr_error = error;
978 netdev->cache_valid |= VALID_ETHERADDR;
980 memcpy(netdev->etheraddr, mac, ETH_ADDR_LEN);
984 netdev_restore_flags(sf);
989 /* Copies 'netdev''s MAC address to 'mac' which is passed as param. */
991 netdev_linux_get_etheraddr(const struct netdev *netdev_,
992 uint8_t mac[ETH_ADDR_LEN])
994 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
996 if (!(netdev->cache_valid & VALID_ETHERADDR)) {
997 int error = get_etheraddr(netdev_get_name(netdev_),
1000 netdev->ether_addr_error = error;
1001 netdev->cache_valid |= VALID_ETHERADDR;
1004 if (!netdev->ether_addr_error) {
1005 memcpy(mac, netdev->etheraddr, ETH_ADDR_LEN);
1008 return netdev->ether_addr_error;
1011 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
1012 * in bytes, not including the hardware header; thus, this is typically 1500
1013 * bytes for Ethernet devices. */
1015 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
1017 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1018 if (!(netdev->cache_valid & VALID_MTU)) {
1022 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1023 SIOCGIFMTU, "SIOCGIFMTU");
1025 netdev->netdev_mtu_error = error;
1026 netdev->mtu = ifr.ifr_mtu;
1027 netdev->cache_valid |= VALID_MTU;
1030 if (!netdev->netdev_mtu_error) {
1031 *mtup = netdev->mtu;
1033 return netdev->netdev_mtu_error;
1036 /* Sets the maximum size of transmitted (MTU) for given device using linux
1037 * networking ioctl interface.
1040 netdev_linux_set_mtu(const struct netdev *netdev_, int mtu)
1042 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1046 if (netdev->cache_valid & VALID_MTU) {
1047 if (netdev->netdev_mtu_error) {
1048 return netdev->netdev_mtu_error;
1050 if (netdev->mtu == mtu) {
1053 netdev->cache_valid &= ~VALID_MTU;
1056 error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1057 SIOCSIFMTU, "SIOCSIFMTU");
1058 if (!error || error == ENODEV) {
1059 netdev->netdev_mtu_error = error;
1060 netdev->mtu = ifr.ifr_mtu;
1061 netdev->cache_valid |= VALID_MTU;
1066 /* Returns the ifindex of 'netdev', if successful, as a positive number.
1067 * On failure, returns a negative errno value. */
1069 netdev_linux_get_ifindex(const struct netdev *netdev)
1073 error = get_ifindex(netdev, &ifindex);
1074 return error ? -error : ifindex;
1078 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
1080 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1082 if (netdev->miimon_interval > 0) {
1083 *carrier = netdev->miimon;
1085 *carrier = (netdev->ifi_flags & IFF_RUNNING) != 0;
1091 static long long int
1092 netdev_linux_get_carrier_resets(const struct netdev *netdev)
1094 return netdev_linux_cast(netdev)->carrier_resets;
1098 netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name,
1099 struct mii_ioctl_data *data)
1104 memset(&ifr, 0, sizeof ifr);
1105 memcpy(&ifr.ifr_data, data, sizeof *data);
1106 error = netdev_linux_do_ioctl(name, &ifr, cmd, cmd_name);
1107 memcpy(data, &ifr.ifr_data, sizeof *data);
1113 netdev_linux_get_miimon(const char *name, bool *miimon)
1115 struct mii_ioctl_data data;
1120 memset(&data, 0, sizeof data);
1121 error = netdev_linux_do_miimon(name, SIOCGMIIPHY, "SIOCGMIIPHY", &data);
1123 /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */
1124 data.reg_num = MII_BMSR;
1125 error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG",
1129 *miimon = !!(data.val_out & BMSR_LSTATUS);
1131 VLOG_WARN_RL(&rl, "%s: failed to query MII", name);
1134 struct ethtool_cmd ecmd;
1136 VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool",
1139 COVERAGE_INC(netdev_get_ethtool);
1140 memset(&ecmd, 0, sizeof ecmd);
1141 error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK,
1144 struct ethtool_value eval;
1146 memcpy(&eval, &ecmd, sizeof eval);
1147 *miimon = !!eval.data;
1149 VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name);
1157 netdev_linux_set_miimon_interval(struct netdev *netdev_,
1158 long long int interval)
1160 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1162 interval = interval > 0 ? MAX(interval, 100) : 0;
1163 if (netdev->miimon_interval != interval) {
1164 netdev->miimon_interval = interval;
1165 timer_set_expired(&netdev->miimon_timer);
1172 netdev_linux_miimon_run(void)
1174 struct shash device_shash;
1175 struct shash_node *node;
1177 shash_init(&device_shash);
1178 netdev_get_devices(&netdev_linux_class, &device_shash);
1179 SHASH_FOR_EACH (node, &device_shash) {
1180 struct netdev *netdev = node->data;
1181 struct netdev_linux *dev = netdev_linux_cast(netdev);
1184 if (dev->miimon_interval <= 0 || !timer_expired(&dev->miimon_timer)) {
1188 netdev_linux_get_miimon(dev->up.name, &miimon);
1189 if (miimon != dev->miimon) {
1190 dev->miimon = miimon;
1191 netdev_linux_changed(dev, dev->ifi_flags, 0);
1194 timer_set_duration(&dev->miimon_timer, dev->miimon_interval);
1197 shash_destroy(&device_shash);
1201 netdev_linux_miimon_wait(void)
1203 struct shash device_shash;
1204 struct shash_node *node;
1206 shash_init(&device_shash);
1207 netdev_get_devices(&netdev_linux_class, &device_shash);
1208 SHASH_FOR_EACH (node, &device_shash) {
1209 struct netdev *netdev = node->data;
1210 struct netdev_linux *dev = netdev_linux_cast(netdev);
1212 if (dev->miimon_interval > 0) {
1213 timer_wait(&dev->miimon_timer);
1216 shash_destroy(&device_shash);
1219 /* Check whether we can we use RTM_GETLINK to get network device statistics.
1220 * In pre-2.6.19 kernels, this was only available if wireless extensions were
1223 check_for_working_netlink_stats(void)
1225 /* Decide on the netdev_get_stats() implementation to use. Netlink is
1226 * preferable, so if that works, we'll use it. */
1227 int ifindex = do_get_ifindex("lo");
1229 VLOG_WARN("failed to get ifindex for lo, "
1230 "obtaining netdev stats from proc");
1233 struct netdev_stats stats;
1234 int error = get_stats_via_netlink(ifindex, &stats);
1236 VLOG_DBG("obtaining netdev stats via rtnetlink");
1239 VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
1240 "via proc (you are probably running a pre-2.6.19 "
1241 "kernel)", ovs_strerror(error));
1248 swap_uint64(uint64_t *a, uint64_t *b)
1255 /* Copies 'src' into 'dst', performing format conversion in the process.
1257 * 'src' is allowed to be misaligned. */
1259 netdev_stats_from_ovs_vport_stats(struct netdev_stats *dst,
1260 const struct ovs_vport_stats *src)
1262 dst->rx_packets = get_unaligned_u64(&src->rx_packets);
1263 dst->tx_packets = get_unaligned_u64(&src->tx_packets);
1264 dst->rx_bytes = get_unaligned_u64(&src->rx_bytes);
1265 dst->tx_bytes = get_unaligned_u64(&src->tx_bytes);
1266 dst->rx_errors = get_unaligned_u64(&src->rx_errors);
1267 dst->tx_errors = get_unaligned_u64(&src->tx_errors);
1268 dst->rx_dropped = get_unaligned_u64(&src->rx_dropped);
1269 dst->tx_dropped = get_unaligned_u64(&src->tx_dropped);
1271 dst->collisions = 0;
1272 dst->rx_length_errors = 0;
1273 dst->rx_over_errors = 0;
1274 dst->rx_crc_errors = 0;
1275 dst->rx_frame_errors = 0;
1276 dst->rx_fifo_errors = 0;
1277 dst->rx_missed_errors = 0;
1278 dst->tx_aborted_errors = 0;
1279 dst->tx_carrier_errors = 0;
1280 dst->tx_fifo_errors = 0;
1281 dst->tx_heartbeat_errors = 0;
1282 dst->tx_window_errors = 0;
1286 get_stats_via_vport__(const struct netdev *netdev, struct netdev_stats *stats)
1288 struct dpif_linux_vport reply;
1292 error = dpif_linux_vport_get(netdev_get_name(netdev), &reply, &buf);
1295 } else if (!reply.stats) {
1300 netdev_stats_from_ovs_vport_stats(stats, reply.stats);
1308 get_stats_via_vport(const struct netdev *netdev_,
1309 struct netdev_stats *stats)
1311 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1313 if (!netdev->vport_stats_error ||
1314 !(netdev->cache_valid & VALID_VPORT_STAT_ERROR)) {
1317 error = get_stats_via_vport__(netdev_, stats);
1318 if (error && error != ENOENT) {
1319 VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed "
1321 netdev_get_name(netdev_), ovs_strerror(error));
1323 netdev->vport_stats_error = error;
1324 netdev->cache_valid |= VALID_VPORT_STAT_ERROR;
1329 netdev_linux_sys_get_stats(const struct netdev *netdev_,
1330 struct netdev_stats *stats)
1332 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
1333 static int use_netlink_stats;
1336 if (ovsthread_once_start(&once)) {
1337 use_netlink_stats = check_for_working_netlink_stats();
1338 ovsthread_once_done(&once);
1341 if (use_netlink_stats) {
1344 error = get_ifindex(netdev_, &ifindex);
1346 error = get_stats_via_netlink(ifindex, stats);
1349 error = get_stats_via_proc(netdev_get_name(netdev_), stats);
1353 VLOG_WARN_RL(&rl, "%s: linux-sys get stats failed %d",
1354 netdev_get_name(netdev_), error);
1360 /* Retrieves current device stats for 'netdev-linux'. */
1362 netdev_linux_get_stats(const struct netdev *netdev_,
1363 struct netdev_stats *stats)
1365 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1366 struct netdev_stats dev_stats;
1369 get_stats_via_vport(netdev_, stats);
1371 error = netdev_linux_sys_get_stats(netdev_, &dev_stats);
1374 if (netdev->vport_stats_error) {
1381 if (netdev->vport_stats_error) {
1382 /* stats not available from OVS then use ioctl stats. */
1385 stats->rx_errors += dev_stats.rx_errors;
1386 stats->tx_errors += dev_stats.tx_errors;
1387 stats->rx_dropped += dev_stats.rx_dropped;
1388 stats->tx_dropped += dev_stats.tx_dropped;
1389 stats->multicast += dev_stats.multicast;
1390 stats->collisions += dev_stats.collisions;
1391 stats->rx_length_errors += dev_stats.rx_length_errors;
1392 stats->rx_over_errors += dev_stats.rx_over_errors;
1393 stats->rx_crc_errors += dev_stats.rx_crc_errors;
1394 stats->rx_frame_errors += dev_stats.rx_frame_errors;
1395 stats->rx_fifo_errors += dev_stats.rx_fifo_errors;
1396 stats->rx_missed_errors += dev_stats.rx_missed_errors;
1397 stats->tx_aborted_errors += dev_stats.tx_aborted_errors;
1398 stats->tx_carrier_errors += dev_stats.tx_carrier_errors;
1399 stats->tx_fifo_errors += dev_stats.tx_fifo_errors;
1400 stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors;
1401 stats->tx_window_errors += dev_stats.tx_window_errors;
1406 /* Retrieves current device stats for 'netdev-tap' netdev or
1407 * netdev-internal. */
1409 netdev_tap_get_stats(const struct netdev *netdev_, struct netdev_stats *stats)
1411 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1412 struct netdev_stats dev_stats;
1415 get_stats_via_vport(netdev_, stats);
1417 error = netdev_linux_sys_get_stats(netdev_, &dev_stats);
1419 if (netdev->vport_stats_error) {
1426 /* If this port is an internal port then the transmit and receive stats
1427 * will appear to be swapped relative to the other ports since we are the
1428 * one sending the data, not a remote computer. For consistency, we swap
1429 * them back here. This does not apply if we are getting stats from the
1430 * vport layer because it always tracks stats from the perspective of the
1432 if (netdev->vport_stats_error) {
1434 swap_uint64(&stats->rx_packets, &stats->tx_packets);
1435 swap_uint64(&stats->rx_bytes, &stats->tx_bytes);
1436 swap_uint64(&stats->rx_errors, &stats->tx_errors);
1437 swap_uint64(&stats->rx_dropped, &stats->tx_dropped);
1438 stats->rx_length_errors = 0;
1439 stats->rx_over_errors = 0;
1440 stats->rx_crc_errors = 0;
1441 stats->rx_frame_errors = 0;
1442 stats->rx_fifo_errors = 0;
1443 stats->rx_missed_errors = 0;
1444 stats->tx_aborted_errors = 0;
1445 stats->tx_carrier_errors = 0;
1446 stats->tx_fifo_errors = 0;
1447 stats->tx_heartbeat_errors = 0;
1448 stats->tx_window_errors = 0;
1450 stats->rx_dropped += dev_stats.tx_dropped;
1451 stats->tx_dropped += dev_stats.rx_dropped;
1453 stats->rx_errors += dev_stats.tx_errors;
1454 stats->tx_errors += dev_stats.rx_errors;
1456 stats->multicast += dev_stats.multicast;
1457 stats->collisions += dev_stats.collisions;
1463 netdev_internal_get_stats(const struct netdev *netdev_,
1464 struct netdev_stats *stats)
1466 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1468 get_stats_via_vport(netdev_, stats);
1469 return netdev->vport_stats_error;
1473 netdev_internal_set_stats(struct netdev *netdev,
1474 const struct netdev_stats *stats)
1476 struct ovs_vport_stats vport_stats;
1477 struct dpif_linux_vport vport;
1480 vport_stats.rx_packets = stats->rx_packets;
1481 vport_stats.tx_packets = stats->tx_packets;
1482 vport_stats.rx_bytes = stats->rx_bytes;
1483 vport_stats.tx_bytes = stats->tx_bytes;
1484 vport_stats.rx_errors = stats->rx_errors;
1485 vport_stats.tx_errors = stats->tx_errors;
1486 vport_stats.rx_dropped = stats->rx_dropped;
1487 vport_stats.tx_dropped = stats->tx_dropped;
1489 dpif_linux_vport_init(&vport);
1490 vport.cmd = OVS_VPORT_CMD_SET;
1491 vport.name = netdev_get_name(netdev);
1492 vport.stats = &vport_stats;
1494 err = dpif_linux_vport_transact(&vport, NULL, NULL);
1496 /* If the vport layer doesn't know about the device, that doesn't mean it
1497 * doesn't exist (after all were able to open it when netdev_open() was
1498 * called), it just means that it isn't attached and we'll be getting
1499 * stats a different way. */
1500 if (err == ENODEV) {
1508 netdev_linux_read_features(struct netdev_linux *netdev)
1510 struct ethtool_cmd ecmd;
1514 if (netdev->cache_valid & VALID_FEATURES) {
1518 COVERAGE_INC(netdev_get_ethtool);
1519 memset(&ecmd, 0, sizeof ecmd);
1520 error = netdev_linux_do_ethtool(netdev->up.name, &ecmd,
1521 ETHTOOL_GSET, "ETHTOOL_GSET");
1526 /* Supported features. */
1527 netdev->supported = 0;
1528 if (ecmd.supported & SUPPORTED_10baseT_Half) {
1529 netdev->supported |= NETDEV_F_10MB_HD;
1531 if (ecmd.supported & SUPPORTED_10baseT_Full) {
1532 netdev->supported |= NETDEV_F_10MB_FD;
1534 if (ecmd.supported & SUPPORTED_100baseT_Half) {
1535 netdev->supported |= NETDEV_F_100MB_HD;
1537 if (ecmd.supported & SUPPORTED_100baseT_Full) {
1538 netdev->supported |= NETDEV_F_100MB_FD;
1540 if (ecmd.supported & SUPPORTED_1000baseT_Half) {
1541 netdev->supported |= NETDEV_F_1GB_HD;
1543 if (ecmd.supported & SUPPORTED_1000baseT_Full) {
1544 netdev->supported |= NETDEV_F_1GB_FD;
1546 if (ecmd.supported & SUPPORTED_10000baseT_Full) {
1547 netdev->supported |= NETDEV_F_10GB_FD;
1549 if (ecmd.supported & SUPPORTED_TP) {
1550 netdev->supported |= NETDEV_F_COPPER;
1552 if (ecmd.supported & SUPPORTED_FIBRE) {
1553 netdev->supported |= NETDEV_F_FIBER;
1555 if (ecmd.supported & SUPPORTED_Autoneg) {
1556 netdev->supported |= NETDEV_F_AUTONEG;
1558 if (ecmd.supported & SUPPORTED_Pause) {
1559 netdev->supported |= NETDEV_F_PAUSE;
1561 if (ecmd.supported & SUPPORTED_Asym_Pause) {
1562 netdev->supported |= NETDEV_F_PAUSE_ASYM;
1565 /* Advertised features. */
1566 netdev->advertised = 0;
1567 if (ecmd.advertising & ADVERTISED_10baseT_Half) {
1568 netdev->advertised |= NETDEV_F_10MB_HD;
1570 if (ecmd.advertising & ADVERTISED_10baseT_Full) {
1571 netdev->advertised |= NETDEV_F_10MB_FD;
1573 if (ecmd.advertising & ADVERTISED_100baseT_Half) {
1574 netdev->advertised |= NETDEV_F_100MB_HD;
1576 if (ecmd.advertising & ADVERTISED_100baseT_Full) {
1577 netdev->advertised |= NETDEV_F_100MB_FD;
1579 if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
1580 netdev->advertised |= NETDEV_F_1GB_HD;
1582 if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
1583 netdev->advertised |= NETDEV_F_1GB_FD;
1585 if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
1586 netdev->advertised |= NETDEV_F_10GB_FD;
1588 if (ecmd.advertising & ADVERTISED_TP) {
1589 netdev->advertised |= NETDEV_F_COPPER;
1591 if (ecmd.advertising & ADVERTISED_FIBRE) {
1592 netdev->advertised |= NETDEV_F_FIBER;
1594 if (ecmd.advertising & ADVERTISED_Autoneg) {
1595 netdev->advertised |= NETDEV_F_AUTONEG;
1597 if (ecmd.advertising & ADVERTISED_Pause) {
1598 netdev->advertised |= NETDEV_F_PAUSE;
1600 if (ecmd.advertising & ADVERTISED_Asym_Pause) {
1601 netdev->advertised |= NETDEV_F_PAUSE_ASYM;
1604 /* Current settings. */
1606 if (speed == SPEED_10) {
1607 netdev->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD;
1608 } else if (speed == SPEED_100) {
1609 netdev->current = ecmd.duplex ? NETDEV_F_100MB_FD : NETDEV_F_100MB_HD;
1610 } else if (speed == SPEED_1000) {
1611 netdev->current = ecmd.duplex ? NETDEV_F_1GB_FD : NETDEV_F_1GB_HD;
1612 } else if (speed == SPEED_10000) {
1613 netdev->current = NETDEV_F_10GB_FD;
1614 } else if (speed == 40000) {
1615 netdev->current = NETDEV_F_40GB_FD;
1616 } else if (speed == 100000) {
1617 netdev->current = NETDEV_F_100GB_FD;
1618 } else if (speed == 1000000) {
1619 netdev->current = NETDEV_F_1TB_FD;
1621 netdev->current = 0;
1624 if (ecmd.port == PORT_TP) {
1625 netdev->current |= NETDEV_F_COPPER;
1626 } else if (ecmd.port == PORT_FIBRE) {
1627 netdev->current |= NETDEV_F_FIBER;
1631 netdev->current |= NETDEV_F_AUTONEG;
1635 netdev->cache_valid |= VALID_FEATURES;
1636 netdev->get_features_error = error;
1639 /* Stores the features supported by 'netdev' into of '*current', '*advertised',
1640 * '*supported', and '*peer'. Each value is a bitmap of NETDEV_* bits.
1641 * Returns 0 if successful, otherwise a positive errno value. */
1643 netdev_linux_get_features(const struct netdev *netdev_,
1644 enum netdev_features *current,
1645 enum netdev_features *advertised,
1646 enum netdev_features *supported,
1647 enum netdev_features *peer)
1649 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1651 netdev_linux_read_features(netdev);
1653 if (!netdev->get_features_error) {
1654 *current = netdev->current;
1655 *advertised = netdev->advertised;
1656 *supported = netdev->supported;
1657 *peer = 0; /* XXX */
1659 return netdev->get_features_error;
1662 /* Set the features advertised by 'netdev' to 'advertise'. */
1664 netdev_linux_set_advertisements(struct netdev *netdev,
1665 enum netdev_features advertise)
1667 struct ethtool_cmd ecmd;
1670 COVERAGE_INC(netdev_get_ethtool);
1671 memset(&ecmd, 0, sizeof ecmd);
1672 error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1673 ETHTOOL_GSET, "ETHTOOL_GSET");
1678 ecmd.advertising = 0;
1679 if (advertise & NETDEV_F_10MB_HD) {
1680 ecmd.advertising |= ADVERTISED_10baseT_Half;
1682 if (advertise & NETDEV_F_10MB_FD) {
1683 ecmd.advertising |= ADVERTISED_10baseT_Full;
1685 if (advertise & NETDEV_F_100MB_HD) {
1686 ecmd.advertising |= ADVERTISED_100baseT_Half;
1688 if (advertise & NETDEV_F_100MB_FD) {
1689 ecmd.advertising |= ADVERTISED_100baseT_Full;
1691 if (advertise & NETDEV_F_1GB_HD) {
1692 ecmd.advertising |= ADVERTISED_1000baseT_Half;
1694 if (advertise & NETDEV_F_1GB_FD) {
1695 ecmd.advertising |= ADVERTISED_1000baseT_Full;
1697 if (advertise & NETDEV_F_10GB_FD) {
1698 ecmd.advertising |= ADVERTISED_10000baseT_Full;
1700 if (advertise & NETDEV_F_COPPER) {
1701 ecmd.advertising |= ADVERTISED_TP;
1703 if (advertise & NETDEV_F_FIBER) {
1704 ecmd.advertising |= ADVERTISED_FIBRE;
1706 if (advertise & NETDEV_F_AUTONEG) {
1707 ecmd.advertising |= ADVERTISED_Autoneg;
1709 if (advertise & NETDEV_F_PAUSE) {
1710 ecmd.advertising |= ADVERTISED_Pause;
1712 if (advertise & NETDEV_F_PAUSE_ASYM) {
1713 ecmd.advertising |= ADVERTISED_Asym_Pause;
1715 COVERAGE_INC(netdev_set_ethtool);
1716 return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1717 ETHTOOL_SSET, "ETHTOOL_SSET");
1720 /* Attempts to set input rate limiting (policing) policy. Returns 0 if
1721 * successful, otherwise a positive errno value. */
1723 netdev_linux_set_policing(struct netdev *netdev_,
1724 uint32_t kbits_rate, uint32_t kbits_burst)
1726 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1727 const char *netdev_name = netdev_get_name(netdev_);
1731 kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */
1732 : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */
1733 : kbits_burst); /* Stick with user-specified value. */
1735 if (netdev->cache_valid & VALID_POLICING) {
1736 if (netdev->netdev_policing_error) {
1737 return netdev->netdev_policing_error;
1740 if (netdev->kbits_rate == kbits_rate &&
1741 netdev->kbits_burst == kbits_burst) {
1742 /* Assume that settings haven't changed since we last set them. */
1745 netdev->cache_valid &= ~VALID_POLICING;
1748 COVERAGE_INC(netdev_set_policing);
1749 /* Remove any existing ingress qdisc. */
1750 error = tc_add_del_ingress_qdisc(netdev_, false);
1752 VLOG_WARN_RL(&rl, "%s: removing policing failed: %s",
1753 netdev_name, ovs_strerror(error));
1758 error = tc_add_del_ingress_qdisc(netdev_, true);
1760 VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s",
1761 netdev_name, ovs_strerror(error));
1765 error = tc_add_policer(netdev_, kbits_rate, kbits_burst);
1767 VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s",
1768 netdev_name, ovs_strerror(error));
1773 netdev->kbits_rate = kbits_rate;
1774 netdev->kbits_burst = kbits_burst;
1777 if (!error || error == ENODEV) {
1778 netdev->netdev_policing_error = error;
1779 netdev->cache_valid |= VALID_POLICING;
1785 netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED,
1788 const struct tc_ops *const *opsp;
1790 for (opsp = tcs; *opsp != NULL; opsp++) {
1791 const struct tc_ops *ops = *opsp;
1792 if (ops->tc_install && ops->ovs_name[0] != '\0') {
1793 sset_add(types, ops->ovs_name);
1799 static const struct tc_ops *
1800 tc_lookup_ovs_name(const char *name)
1802 const struct tc_ops *const *opsp;
1804 for (opsp = tcs; *opsp != NULL; opsp++) {
1805 const struct tc_ops *ops = *opsp;
1806 if (!strcmp(name, ops->ovs_name)) {
1813 static const struct tc_ops *
1814 tc_lookup_linux_name(const char *name)
1816 const struct tc_ops *const *opsp;
1818 for (opsp = tcs; *opsp != NULL; opsp++) {
1819 const struct tc_ops *ops = *opsp;
1820 if (ops->linux_name && !strcmp(name, ops->linux_name)) {
1827 static struct tc_queue *
1828 tc_find_queue__(const struct netdev *netdev_, unsigned int queue_id,
1831 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1832 struct tc_queue *queue;
1834 HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev->tc->queues) {
1835 if (queue->queue_id == queue_id) {
1842 static struct tc_queue *
1843 tc_find_queue(const struct netdev *netdev, unsigned int queue_id)
1845 return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0));
1849 netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED,
1851 struct netdev_qos_capabilities *caps)
1853 const struct tc_ops *ops = tc_lookup_ovs_name(type);
1857 caps->n_queues = ops->n_queues;
1862 netdev_linux_get_qos(const struct netdev *netdev_,
1863 const char **typep, struct smap *details)
1865 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1868 error = tc_query_qdisc(netdev_);
1873 *typep = netdev->tc->ops->ovs_name;
1874 return (netdev->tc->ops->qdisc_get
1875 ? netdev->tc->ops->qdisc_get(netdev_, details)
1880 netdev_linux_set_qos(struct netdev *netdev_,
1881 const char *type, const struct smap *details)
1883 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1884 const struct tc_ops *new_ops;
1887 new_ops = tc_lookup_ovs_name(type);
1888 if (!new_ops || !new_ops->tc_install) {
1892 error = tc_query_qdisc(netdev_);
1897 if (new_ops == netdev->tc->ops) {
1898 return new_ops->qdisc_set ? new_ops->qdisc_set(netdev_, details) : 0;
1900 /* Delete existing qdisc. */
1901 error = tc_del_qdisc(netdev_);
1905 ovs_assert(netdev->tc == NULL);
1907 /* Install new qdisc. */
1908 error = new_ops->tc_install(netdev_, details);
1909 ovs_assert((error == 0) == (netdev->tc != NULL));
1916 netdev_linux_get_queue(const struct netdev *netdev_,
1917 unsigned int queue_id, struct smap *details)
1919 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1922 error = tc_query_qdisc(netdev_);
1926 struct tc_queue *queue = tc_find_queue(netdev_, queue_id);
1928 ? netdev->tc->ops->class_get(netdev_, queue, details)
1934 netdev_linux_set_queue(struct netdev *netdev_,
1935 unsigned int queue_id, const struct smap *details)
1937 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1940 error = tc_query_qdisc(netdev_);
1943 } else if (queue_id >= netdev->tc->ops->n_queues
1944 || !netdev->tc->ops->class_set) {
1948 return netdev->tc->ops->class_set(netdev_, queue_id, details);
1952 netdev_linux_delete_queue(struct netdev *netdev_, unsigned int queue_id)
1954 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1957 error = tc_query_qdisc(netdev_);
1960 } else if (!netdev->tc->ops->class_delete) {
1963 struct tc_queue *queue = tc_find_queue(netdev_, queue_id);
1965 ? netdev->tc->ops->class_delete(netdev_, queue)
1971 netdev_linux_get_queue_stats(const struct netdev *netdev_,
1972 unsigned int queue_id,
1973 struct netdev_queue_stats *stats)
1975 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1978 error = tc_query_qdisc(netdev_);
1981 } else if (!netdev->tc->ops->class_get_stats) {
1984 const struct tc_queue *queue = tc_find_queue(netdev_, queue_id);
1988 stats->created = queue->created;
1989 return netdev->tc->ops->class_get_stats(netdev_, queue, stats);
1994 start_queue_dump(const struct netdev *netdev, struct nl_dump *dump)
1996 struct ofpbuf request;
1997 struct tcmsg *tcmsg;
1999 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request);
2003 tcmsg->tcm_parent = 0;
2004 nl_dump_start(dump, NETLINK_ROUTE, &request);
2005 ofpbuf_uninit(&request);
2010 netdev_linux_dump_queues(const struct netdev *netdev_,
2011 netdev_dump_queues_cb *cb, void *aux)
2013 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2014 struct tc_queue *queue, *next_queue;
2015 struct smap details;
2019 error = tc_query_qdisc(netdev_);
2022 } else if (!netdev->tc->ops->class_get) {
2027 smap_init(&details);
2028 HMAP_FOR_EACH_SAFE (queue, next_queue, hmap_node,
2029 &netdev->tc->queues) {
2030 smap_clear(&details);
2032 error = netdev->tc->ops->class_get(netdev_, queue, &details);
2034 (*cb)(queue->queue_id, &details, aux);
2039 smap_destroy(&details);
2045 netdev_linux_dump_queue_stats(const struct netdev *netdev_,
2046 netdev_dump_queue_stats_cb *cb, void *aux)
2048 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2049 struct nl_dump dump;
2054 error = tc_query_qdisc(netdev_);
2057 } else if (!netdev->tc->ops->class_dump_stats) {
2062 if (!start_queue_dump(netdev_, &dump)) {
2065 while (nl_dump_next(&dump, &msg)) {
2066 error = netdev->tc->ops->class_dump_stats(netdev_, &msg, cb, aux);
2072 error = nl_dump_done(&dump);
2073 return error ? error : last_error;
2077 netdev_linux_get_in4(const struct netdev *netdev_,
2078 struct in_addr *address, struct in_addr *netmask)
2080 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2082 if (!(netdev->cache_valid & VALID_IN4)) {
2085 error = netdev_linux_get_ipv4(netdev_, &netdev->address,
2086 SIOCGIFADDR, "SIOCGIFADDR");
2091 error = netdev_linux_get_ipv4(netdev_, &netdev->netmask,
2092 SIOCGIFNETMASK, "SIOCGIFNETMASK");
2097 netdev->cache_valid |= VALID_IN4;
2099 *address = netdev->address;
2100 *netmask = netdev->netmask;
2101 return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
2105 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
2106 struct in_addr netmask)
2108 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2111 error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
2113 netdev->cache_valid |= VALID_IN4;
2114 netdev->address = address;
2115 netdev->netmask = netmask;
2116 if (address.s_addr != INADDR_ANY) {
2117 error = do_set_addr(netdev_, SIOCSIFNETMASK,
2118 "SIOCSIFNETMASK", netmask);
2125 parse_if_inet6_line(const char *line,
2126 struct in6_addr *in6, char ifname[16 + 1])
2128 uint8_t *s6 = in6->s6_addr;
2129 #define X8 "%2"SCNx8
2131 " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
2132 "%*x %*x %*x %*x %16s\n",
2133 &s6[0], &s6[1], &s6[2], &s6[3],
2134 &s6[4], &s6[5], &s6[6], &s6[7],
2135 &s6[8], &s6[9], &s6[10], &s6[11],
2136 &s6[12], &s6[13], &s6[14], &s6[15],
2140 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
2141 * 'in6' is non-null) and returns true. Otherwise, returns false. */
2143 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
2145 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2146 if (!(netdev->cache_valid & VALID_IN6)) {
2150 netdev->in6 = in6addr_any;
2152 file = fopen("/proc/net/if_inet6", "r");
2154 const char *name = netdev_get_name(netdev_);
2155 while (fgets(line, sizeof line, file)) {
2156 struct in6_addr in6_tmp;
2157 char ifname[16 + 1];
2158 if (parse_if_inet6_line(line, &in6_tmp, ifname)
2159 && !strcmp(name, ifname))
2161 netdev->in6 = in6_tmp;
2167 netdev->cache_valid |= VALID_IN6;
2174 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
2176 struct sockaddr_in sin;
2177 memset(&sin, 0, sizeof sin);
2178 sin.sin_family = AF_INET;
2179 sin.sin_addr = addr;
2182 memset(sa, 0, sizeof *sa);
2183 memcpy(sa, &sin, sizeof sin);
2187 do_set_addr(struct netdev *netdev,
2188 int ioctl_nr, const char *ioctl_name, struct in_addr addr)
2191 ovs_strzcpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
2192 make_in4_sockaddr(&ifr.ifr_addr, addr);
2194 return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
2198 /* Adds 'router' as a default IP gateway. */
2200 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
2202 struct in_addr any = { INADDR_ANY };
2206 memset(&rt, 0, sizeof rt);
2207 make_in4_sockaddr(&rt.rt_dst, any);
2208 make_in4_sockaddr(&rt.rt_gateway, router);
2209 make_in4_sockaddr(&rt.rt_genmask, any);
2210 rt.rt_flags = RTF_UP | RTF_GATEWAY;
2211 error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
2213 VLOG_WARN("ioctl(SIOCADDRT): %s", ovs_strerror(error));
2219 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
2222 static const char fn[] = "/proc/net/route";
2227 *netdev_name = NULL;
2228 stream = fopen(fn, "r");
2229 if (stream == NULL) {
2230 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno));
2235 while (fgets(line, sizeof line, stream)) {
2238 ovs_be32 dest, gateway, mask;
2239 int refcnt, metric, mtu;
2240 unsigned int flags, use, window, irtt;
2243 "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
2245 iface, &dest, &gateway, &flags, &refcnt,
2246 &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
2248 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
2252 if (!(flags & RTF_UP)) {
2253 /* Skip routes that aren't up. */
2257 /* The output of 'dest', 'mask', and 'gateway' were given in
2258 * network byte order, so we don't need need any endian
2259 * conversions here. */
2260 if ((dest & mask) == (host->s_addr & mask)) {
2262 /* The host is directly reachable. */
2263 next_hop->s_addr = 0;
2265 /* To reach the host, we must go through a gateway. */
2266 next_hop->s_addr = gateway;
2268 *netdev_name = xstrdup(iface);
2280 netdev_linux_get_status(const struct netdev *netdev_, struct smap *smap)
2282 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2285 if (!(netdev->cache_valid & VALID_DRVINFO)) {
2286 struct ethtool_cmd *cmd = (struct ethtool_cmd *) &netdev->drvinfo;
2288 COVERAGE_INC(netdev_get_ethtool);
2289 memset(&netdev->drvinfo, 0, sizeof netdev->drvinfo);
2290 error = netdev_linux_do_ethtool(netdev->up.name,
2293 "ETHTOOL_GDRVINFO");
2295 netdev->cache_valid |= VALID_DRVINFO;
2300 smap_add(smap, "driver_name", netdev->drvinfo.driver);
2301 smap_add(smap, "driver_version", netdev->drvinfo.version);
2302 smap_add(smap, "firmware_version", netdev->drvinfo.fw_version);
2308 netdev_internal_get_status(const struct netdev *netdev OVS_UNUSED,
2311 smap_add(smap, "driver_name", "openvswitch");
2315 /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
2316 * successfully retrieved, it stores the corresponding MAC address in 'mac' and
2317 * returns 0. Otherwise, it returns a positive errno value; in particular,
2318 * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
2320 netdev_linux_arp_lookup(const struct netdev *netdev,
2321 ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
2324 struct sockaddr_in sin;
2327 memset(&r, 0, sizeof r);
2328 memset(&sin, 0, sizeof sin);
2329 sin.sin_family = AF_INET;
2330 sin.sin_addr.s_addr = ip;
2332 memcpy(&r.arp_pa, &sin, sizeof sin);
2333 r.arp_ha.sa_family = ARPHRD_ETHER;
2335 ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
2336 COVERAGE_INC(netdev_arp_lookup);
2337 retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
2339 memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
2340 } else if (retval != ENXIO) {
2341 VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
2342 netdev_get_name(netdev), IP_ARGS(ip),
2343 ovs_strerror(retval));
2349 nd_to_iff_flags(enum netdev_flags nd)
2352 if (nd & NETDEV_UP) {
2355 if (nd & NETDEV_PROMISC) {
2362 iff_to_nd_flags(int iff)
2364 enum netdev_flags nd = 0;
2368 if (iff & IFF_PROMISC) {
2369 nd |= NETDEV_PROMISC;
2375 netdev_linux_update_flags(struct netdev *netdev_, enum netdev_flags off,
2376 enum netdev_flags on, enum netdev_flags *old_flagsp)
2378 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2379 int old_flags, new_flags;
2382 old_flags = netdev->ifi_flags;
2383 *old_flagsp = iff_to_nd_flags(old_flags);
2384 new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
2385 if (new_flags != old_flags) {
2386 error = set_flags(netdev_get_name(netdev_), new_flags);
2387 get_flags(netdev_, &netdev->ifi_flags);
2393 netdev_linux_change_seq(const struct netdev *netdev)
2395 return netdev_linux_cast(netdev)->change_seq;
2398 #define NETDEV_LINUX_CLASS(NAME, CREATE, GET_STATS, SET_STATS, \
2399 GET_FEATURES, GET_STATUS) \
2403 netdev_linux_init, \
2405 netdev_linux_wait, \
2408 netdev_linux_destroy, \
2409 NULL, /* get_config */ \
2410 NULL, /* set_config */ \
2411 NULL, /* get_tunnel_config */ \
2413 netdev_linux_rx_open, \
2415 netdev_linux_send, \
2416 netdev_linux_send_wait, \
2418 netdev_linux_set_etheraddr, \
2419 netdev_linux_get_etheraddr, \
2420 netdev_linux_get_mtu, \
2421 netdev_linux_set_mtu, \
2422 netdev_linux_get_ifindex, \
2423 netdev_linux_get_carrier, \
2424 netdev_linux_get_carrier_resets, \
2425 netdev_linux_set_miimon_interval, \
2430 netdev_linux_set_advertisements, \
2432 netdev_linux_set_policing, \
2433 netdev_linux_get_qos_types, \
2434 netdev_linux_get_qos_capabilities, \
2435 netdev_linux_get_qos, \
2436 netdev_linux_set_qos, \
2437 netdev_linux_get_queue, \
2438 netdev_linux_set_queue, \
2439 netdev_linux_delete_queue, \
2440 netdev_linux_get_queue_stats, \
2441 netdev_linux_dump_queues, \
2442 netdev_linux_dump_queue_stats, \
2444 netdev_linux_get_in4, \
2445 netdev_linux_set_in4, \
2446 netdev_linux_get_in6, \
2447 netdev_linux_add_router, \
2448 netdev_linux_get_next_hop, \
2450 netdev_linux_arp_lookup, \
2452 netdev_linux_update_flags, \
2454 netdev_linux_change_seq \
2457 const struct netdev_class netdev_linux_class =
2460 netdev_linux_create,
2461 netdev_linux_get_stats,
2462 NULL, /* set_stats */
2463 netdev_linux_get_features,
2464 netdev_linux_get_status);
2466 const struct netdev_class netdev_tap_class =
2469 netdev_linux_create_tap,
2470 netdev_tap_get_stats,
2471 NULL, /* set_stats */
2472 netdev_linux_get_features,
2473 netdev_linux_get_status);
2475 const struct netdev_class netdev_internal_class =
2478 netdev_linux_create,
2479 netdev_internal_get_stats,
2480 netdev_internal_set_stats,
2481 NULL, /* get_features */
2482 netdev_internal_get_status);
2484 static const struct netdev_rx_class netdev_rx_linux_class = {
2485 netdev_rx_linux_destroy,
2486 netdev_rx_linux_recv,
2487 netdev_rx_linux_wait,
2488 netdev_rx_linux_drain,
2491 /* HTB traffic control class. */
2493 #define HTB_N_QUEUES 0xf000
2497 unsigned int max_rate; /* In bytes/s. */
2501 struct tc_queue tc_queue;
2502 unsigned int min_rate; /* In bytes/s. */
2503 unsigned int max_rate; /* In bytes/s. */
2504 unsigned int burst; /* In bytes. */
2505 unsigned int priority; /* Lower values are higher priorities. */
2509 htb_get__(const struct netdev *netdev_)
2511 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2512 return CONTAINER_OF(netdev->tc, struct htb, tc);
2516 htb_install__(struct netdev *netdev_, uint64_t max_rate)
2518 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2521 htb = xmalloc(sizeof *htb);
2522 tc_init(&htb->tc, &tc_ops_htb);
2523 htb->max_rate = max_rate;
2525 netdev->tc = &htb->tc;
2528 /* Create an HTB qdisc.
2530 * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */
2532 htb_setup_qdisc__(struct netdev *netdev)
2535 struct tc_htb_glob opt;
2536 struct ofpbuf request;
2537 struct tcmsg *tcmsg;
2539 tc_del_qdisc(netdev);
2541 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
2542 NLM_F_EXCL | NLM_F_CREATE, &request);
2546 tcmsg->tcm_handle = tc_make_handle(1, 0);
2547 tcmsg->tcm_parent = TC_H_ROOT;
2549 nl_msg_put_string(&request, TCA_KIND, "htb");
2551 memset(&opt, 0, sizeof opt);
2552 opt.rate2quantum = 10;
2556 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2557 nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt);
2558 nl_msg_end_nested(&request, opt_offset);
2560 return tc_transact(&request, NULL);
2563 /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb
2564 * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */
2566 htb_setup_class__(struct netdev *netdev, unsigned int handle,
2567 unsigned int parent, struct htb_class *class)
2570 struct tc_htb_opt opt;
2571 struct ofpbuf request;
2572 struct tcmsg *tcmsg;
2576 error = netdev_get_mtu(netdev, &mtu);
2578 VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU",
2579 netdev_get_name(netdev));
2583 memset(&opt, 0, sizeof opt);
2584 tc_fill_rate(&opt.rate, class->min_rate, mtu);
2585 tc_fill_rate(&opt.ceil, class->max_rate, mtu);
2586 opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst);
2587 opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst);
2588 opt.prio = class->priority;
2590 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
2594 tcmsg->tcm_handle = handle;
2595 tcmsg->tcm_parent = parent;
2597 nl_msg_put_string(&request, TCA_KIND, "htb");
2598 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
2599 nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt);
2600 tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate);
2601 tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil);
2602 nl_msg_end_nested(&request, opt_offset);
2604 error = tc_transact(&request, NULL);
2606 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
2607 "min_rate=%u max_rate=%u burst=%u prio=%u (%s)",
2608 netdev_get_name(netdev),
2609 tc_get_major(handle), tc_get_minor(handle),
2610 tc_get_major(parent), tc_get_minor(parent),
2611 class->min_rate, class->max_rate,
2612 class->burst, class->priority, ovs_strerror(error));
2617 /* Parses Netlink attributes in 'options' for HTB parameters and stores a
2618 * description of them into 'details'. The description complies with the
2619 * specification given in the vswitch database documentation for linux-htb
2622 htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class)
2624 static const struct nl_policy tca_htb_policy[] = {
2625 [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false,
2626 .min_len = sizeof(struct tc_htb_opt) },
2629 struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)];
2630 const struct tc_htb_opt *htb;
2632 if (!nl_parse_nested(nl_options, tca_htb_policy,
2633 attrs, ARRAY_SIZE(tca_htb_policy))) {
2634 VLOG_WARN_RL(&rl, "failed to parse HTB class options");
2638 htb = nl_attr_get(attrs[TCA_HTB_PARMS]);
2639 class->min_rate = htb->rate.rate;
2640 class->max_rate = htb->ceil.rate;
2641 class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer);
2642 class->priority = htb->prio;
2647 htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
2648 struct htb_class *options,
2649 struct netdev_queue_stats *stats)
2651 struct nlattr *nl_options;
2652 unsigned int handle;
2655 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
2656 if (!error && queue_id) {
2657 unsigned int major = tc_get_major(handle);
2658 unsigned int minor = tc_get_minor(handle);
2659 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2660 *queue_id = minor - 1;
2665 if (!error && options) {
2666 error = htb_parse_tca_options__(nl_options, options);
2672 htb_parse_qdisc_details__(struct netdev *netdev,
2673 const struct smap *details, struct htb_class *hc)
2675 const char *max_rate_s;
2677 max_rate_s = smap_get(details, "max-rate");
2678 hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
2679 if (!hc->max_rate) {
2680 enum netdev_features current;
2682 netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
2683 hc->max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8;
2685 hc->min_rate = hc->max_rate;
2691 htb_parse_class_details__(struct netdev *netdev,
2692 const struct smap *details, struct htb_class *hc)
2694 const struct htb *htb = htb_get__(netdev);
2695 const char *min_rate_s = smap_get(details, "min-rate");
2696 const char *max_rate_s = smap_get(details, "max-rate");
2697 const char *burst_s = smap_get(details, "burst");
2698 const char *priority_s = smap_get(details, "priority");
2701 error = netdev_get_mtu(netdev, &mtu);
2703 VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU",
2704 netdev_get_name(netdev));
2708 /* HTB requires at least an mtu sized min-rate to send any traffic even
2709 * on uncongested links. */
2710 hc->min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
2711 hc->min_rate = MAX(hc->min_rate, mtu);
2712 hc->min_rate = MIN(hc->min_rate, htb->max_rate);
2715 hc->max_rate = (max_rate_s
2716 ? strtoull(max_rate_s, NULL, 10) / 8
2718 hc->max_rate = MAX(hc->max_rate, hc->min_rate);
2719 hc->max_rate = MIN(hc->max_rate, htb->max_rate);
2723 * According to hints in the documentation that I've read, it is important
2724 * that 'burst' be at least as big as the largest frame that might be
2725 * transmitted. Also, making 'burst' a bit bigger than necessary is OK,
2726 * but having it a bit too small is a problem. Since netdev_get_mtu()
2727 * doesn't include the Ethernet header, we need to add at least 14 (18?) to
2728 * the MTU. We actually add 64, instead of 14, as a guard against
2729 * additional headers get tacked on somewhere that we're not aware of. */
2730 hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0;
2731 hc->burst = MAX(hc->burst, mtu + 64);
2734 hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0;
2740 htb_query_class__(const struct netdev *netdev, unsigned int handle,
2741 unsigned int parent, struct htb_class *options,
2742 struct netdev_queue_stats *stats)
2744 struct ofpbuf *reply;
2747 error = tc_query_class(netdev, handle, parent, &reply);
2749 error = htb_parse_tcmsg__(reply, NULL, options, stats);
2750 ofpbuf_delete(reply);
2756 htb_tc_install(struct netdev *netdev, const struct smap *details)
2760 error = htb_setup_qdisc__(netdev);
2762 struct htb_class hc;
2764 htb_parse_qdisc_details__(netdev, details, &hc);
2765 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
2766 tc_make_handle(1, 0), &hc);
2768 htb_install__(netdev, hc.max_rate);
2774 static struct htb_class *
2775 htb_class_cast__(const struct tc_queue *queue)
2777 return CONTAINER_OF(queue, struct htb_class, tc_queue);
2781 htb_update_queue__(struct netdev *netdev, unsigned int queue_id,
2782 const struct htb_class *hc)
2784 struct htb *htb = htb_get__(netdev);
2785 size_t hash = hash_int(queue_id, 0);
2786 struct tc_queue *queue;
2787 struct htb_class *hcp;
2789 queue = tc_find_queue__(netdev, queue_id, hash);
2791 hcp = htb_class_cast__(queue);
2793 hcp = xmalloc(sizeof *hcp);
2794 queue = &hcp->tc_queue;
2795 queue->queue_id = queue_id;
2796 queue->created = time_msec();
2797 hmap_insert(&htb->tc.queues, &queue->hmap_node, hash);
2800 hcp->min_rate = hc->min_rate;
2801 hcp->max_rate = hc->max_rate;
2802 hcp->burst = hc->burst;
2803 hcp->priority = hc->priority;
2807 htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
2810 struct nl_dump dump;
2811 struct htb_class hc;
2813 /* Get qdisc options. */
2815 htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
2816 htb_install__(netdev, hc.max_rate);
2819 if (!start_queue_dump(netdev, &dump)) {
2822 while (nl_dump_next(&dump, &msg)) {
2823 unsigned int queue_id;
2825 if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
2826 htb_update_queue__(netdev, queue_id, &hc);
2829 nl_dump_done(&dump);
2835 htb_tc_destroy(struct tc *tc)
2837 struct htb *htb = CONTAINER_OF(tc, struct htb, tc);
2838 struct htb_class *hc, *next;
2840 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) {
2841 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
2849 htb_qdisc_get(const struct netdev *netdev, struct smap *details)
2851 const struct htb *htb = htb_get__(netdev);
2852 smap_add_format(details, "max-rate", "%llu", 8ULL * htb->max_rate);
2857 htb_qdisc_set(struct netdev *netdev, const struct smap *details)
2859 struct htb_class hc;
2862 htb_parse_qdisc_details__(netdev, details, &hc);
2863 error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
2864 tc_make_handle(1, 0), &hc);
2866 htb_get__(netdev)->max_rate = hc.max_rate;
2872 htb_class_get(const struct netdev *netdev OVS_UNUSED,
2873 const struct tc_queue *queue, struct smap *details)
2875 const struct htb_class *hc = htb_class_cast__(queue);
2877 smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate);
2878 if (hc->min_rate != hc->max_rate) {
2879 smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate);
2881 smap_add_format(details, "burst", "%llu", 8ULL * hc->burst);
2883 smap_add_format(details, "priority", "%u", hc->priority);
2889 htb_class_set(struct netdev *netdev, unsigned int queue_id,
2890 const struct smap *details)
2892 struct htb_class hc;
2895 error = htb_parse_class_details__(netdev, details, &hc);
2900 error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
2901 tc_make_handle(1, 0xfffe), &hc);
2906 htb_update_queue__(netdev, queue_id, &hc);
2911 htb_class_delete(struct netdev *netdev, struct tc_queue *queue)
2913 struct htb_class *hc = htb_class_cast__(queue);
2914 struct htb *htb = htb_get__(netdev);
2917 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
2919 hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
2926 htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
2927 struct netdev_queue_stats *stats)
2929 return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
2930 tc_make_handle(1, 0xfffe), NULL, stats);
2934 htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
2935 const struct ofpbuf *nlmsg,
2936 netdev_dump_queue_stats_cb *cb, void *aux)
2938 struct netdev_queue_stats stats;
2939 unsigned int handle, major, minor;
2942 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
2947 major = tc_get_major(handle);
2948 minor = tc_get_minor(handle);
2949 if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
2950 (*cb)(minor - 1, &stats, aux);
2955 static const struct tc_ops tc_ops_htb = {
2956 "htb", /* linux_name */
2957 "linux-htb", /* ovs_name */
2958 HTB_N_QUEUES, /* n_queues */
2967 htb_class_get_stats,
2968 htb_class_dump_stats
2971 /* "linux-hfsc" traffic control class. */
2973 #define HFSC_N_QUEUES 0xf000
2981 struct tc_queue tc_queue;
2986 static struct hfsc *
2987 hfsc_get__(const struct netdev *netdev_)
2989 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
2990 return CONTAINER_OF(netdev->tc, struct hfsc, tc);
2993 static struct hfsc_class *
2994 hfsc_class_cast__(const struct tc_queue *queue)
2996 return CONTAINER_OF(queue, struct hfsc_class, tc_queue);
3000 hfsc_install__(struct netdev *netdev_, uint32_t max_rate)
3002 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
3005 hfsc = xmalloc(sizeof *hfsc);
3006 tc_init(&hfsc->tc, &tc_ops_hfsc);
3007 hfsc->max_rate = max_rate;
3008 netdev->tc = &hfsc->tc;
3012 hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id,
3013 const struct hfsc_class *hc)
3017 struct hfsc_class *hcp;
3018 struct tc_queue *queue;
3020 hfsc = hfsc_get__(netdev);
3021 hash = hash_int(queue_id, 0);
3023 queue = tc_find_queue__(netdev, queue_id, hash);
3025 hcp = hfsc_class_cast__(queue);
3027 hcp = xmalloc(sizeof *hcp);
3028 queue = &hcp->tc_queue;
3029 queue->queue_id = queue_id;
3030 queue->created = time_msec();
3031 hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash);
3034 hcp->min_rate = hc->min_rate;
3035 hcp->max_rate = hc->max_rate;
3039 hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class)
3041 const struct tc_service_curve *rsc, *fsc, *usc;
3042 static const struct nl_policy tca_hfsc_policy[] = {
3044 .type = NL_A_UNSPEC,
3046 .min_len = sizeof(struct tc_service_curve),
3049 .type = NL_A_UNSPEC,
3051 .min_len = sizeof(struct tc_service_curve),
3054 .type = NL_A_UNSPEC,
3056 .min_len = sizeof(struct tc_service_curve),
3059 struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)];
3061 if (!nl_parse_nested(nl_options, tca_hfsc_policy,
3062 attrs, ARRAY_SIZE(tca_hfsc_policy))) {
3063 VLOG_WARN_RL(&rl, "failed to parse HFSC class options");
3067 rsc = nl_attr_get(attrs[TCA_HFSC_RSC]);
3068 fsc = nl_attr_get(attrs[TCA_HFSC_FSC]);
3069 usc = nl_attr_get(attrs[TCA_HFSC_USC]);
3071 if (rsc->m1 != 0 || rsc->d != 0 ||
3072 fsc->m1 != 0 || fsc->d != 0 ||
3073 usc->m1 != 0 || usc->d != 0) {
3074 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3075 "Non-linear service curves are not supported.");
3079 if (rsc->m2 != fsc->m2) {
3080 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3081 "Real-time service curves are not supported ");
3085 if (rsc->m2 > usc->m2) {
3086 VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
3087 "Min-rate service curve is greater than "
3088 "the max-rate service curve.");
3092 class->min_rate = fsc->m2;
3093 class->max_rate = usc->m2;
3098 hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
3099 struct hfsc_class *options,
3100 struct netdev_queue_stats *stats)
3103 unsigned int handle;
3104 struct nlattr *nl_options;
3106 error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
3112 unsigned int major, minor;
3114 major = tc_get_major(handle);
3115 minor = tc_get_minor(handle);
3116 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
3117 *queue_id = minor - 1;
3124 error = hfsc_parse_tca_options__(nl_options, options);
3131 hfsc_query_class__(const struct netdev *netdev, unsigned int handle,
3132 unsigned int parent, struct hfsc_class *options,
3133 struct netdev_queue_stats *stats)
3136 struct ofpbuf *reply;
3138 error = tc_query_class(netdev, handle, parent, &reply);
3143 error = hfsc_parse_tcmsg__(reply, NULL, options, stats);
3144 ofpbuf_delete(reply);
3149 hfsc_parse_qdisc_details__(struct netdev *netdev, const struct smap *details,
3150 struct hfsc_class *class)
3153 const char *max_rate_s;
3155 max_rate_s = smap_get(details, "max-rate");
3156 max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
3159 enum netdev_features current;
3161 netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
3162 max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8;
3165 class->min_rate = max_rate;
3166 class->max_rate = max_rate;
3170 hfsc_parse_class_details__(struct netdev *netdev,
3171 const struct smap *details,
3172 struct hfsc_class * class)
3174 const struct hfsc *hfsc;
3175 uint32_t min_rate, max_rate;
3176 const char *min_rate_s, *max_rate_s;
3178 hfsc = hfsc_get__(netdev);
3179 min_rate_s = smap_get(details, "min-rate");
3180 max_rate_s = smap_get(details, "max-rate");
3182 min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
3183 min_rate = MAX(min_rate, 1);
3184 min_rate = MIN(min_rate, hfsc->max_rate);
3186 max_rate = (max_rate_s
3187 ? strtoull(max_rate_s, NULL, 10) / 8
3189 max_rate = MAX(max_rate, min_rate);
3190 max_rate = MIN(max_rate, hfsc->max_rate);
3192 class->min_rate = min_rate;
3193 class->max_rate = max_rate;
3198 /* Create an HFSC qdisc.
3200 * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */
3202 hfsc_setup_qdisc__(struct netdev * netdev)
3204 struct tcmsg *tcmsg;
3205 struct ofpbuf request;
3206 struct tc_hfsc_qopt opt;
3208 tc_del_qdisc(netdev);
3210 tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
3211 NLM_F_EXCL | NLM_F_CREATE, &request);
3217 tcmsg->tcm_handle = tc_make_handle(1, 0);
3218 tcmsg->tcm_parent = TC_H_ROOT;
3220 memset(&opt, 0, sizeof opt);
3223 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3224 nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt);
3226 return tc_transact(&request, NULL);
3229 /* Create an HFSC class.
3231 * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc
3232 * sc rate <min_rate> ul rate <max_rate>" */
3234 hfsc_setup_class__(struct netdev *netdev, unsigned int handle,
3235 unsigned int parent, struct hfsc_class *class)
3239 struct tcmsg *tcmsg;
3240 struct ofpbuf request;
3241 struct tc_service_curve min, max;
3243 tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
3249 tcmsg->tcm_handle = handle;
3250 tcmsg->tcm_parent = parent;
3254 min.m2 = class->min_rate;
3258 max.m2 = class->max_rate;
3260 nl_msg_put_string(&request, TCA_KIND, "hfsc");
3261 opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
3262 nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min);
3263 nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min);
3264 nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max);
3265 nl_msg_end_nested(&request, opt_offset);
3267 error = tc_transact(&request, NULL);
3269 VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
3270 "min-rate %ubps, max-rate %ubps (%s)",
3271 netdev_get_name(netdev),
3272 tc_get_major(handle), tc_get_minor(handle),
3273 tc_get_major(parent), tc_get_minor(parent),
3274 class->min_rate, class->max_rate, ovs_strerror(error));
3281 hfsc_tc_install(struct netdev *netdev, const struct smap *details)
3284 struct hfsc_class class;
3286 error = hfsc_setup_qdisc__(netdev);
3292 hfsc_parse_qdisc_details__(netdev, details, &class);
3293 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3294 tc_make_handle(1, 0), &class);
3300 hfsc_install__(netdev, class.max_rate);
3305 hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3308 struct nl_dump dump;
3309 struct hfsc_class hc;
3312 hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
3313 hfsc_install__(netdev, hc.max_rate);
3315 if (!start_queue_dump(netdev, &dump)) {
3319 while (nl_dump_next(&dump, &msg)) {
3320 unsigned int queue_id;
3322 if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
3323 hfsc_update_queue__(netdev, queue_id, &hc);
3327 nl_dump_done(&dump);
3332 hfsc_tc_destroy(struct tc *tc)
3335 struct hfsc_class *hc, *next;
3337 hfsc = CONTAINER_OF(tc, struct hfsc, tc);
3339 HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) {
3340 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3349 hfsc_qdisc_get(const struct netdev *netdev, struct smap *details)
3351 const struct hfsc *hfsc;
3352 hfsc = hfsc_get__(netdev);
3353 smap_add_format(details, "max-rate", "%llu", 8ULL * hfsc->max_rate);
3358 hfsc_qdisc_set(struct netdev *netdev, const struct smap *details)
3361 struct hfsc_class class;
3363 hfsc_parse_qdisc_details__(netdev, details, &class);
3364 error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
3365 tc_make_handle(1, 0), &class);
3368 hfsc_get__(netdev)->max_rate = class.max_rate;
3375 hfsc_class_get(const struct netdev *netdev OVS_UNUSED,
3376 const struct tc_queue *queue, struct smap *details)
3378 const struct hfsc_class *hc;
3380 hc = hfsc_class_cast__(queue);
3381 smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate);
3382 if (hc->min_rate != hc->max_rate) {
3383 smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate);
3389 hfsc_class_set(struct netdev *netdev, unsigned int queue_id,
3390 const struct smap *details)
3393 struct hfsc_class class;
3395 error = hfsc_parse_class_details__(netdev, details, &class);
3400 error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
3401 tc_make_handle(1, 0xfffe), &class);
3406 hfsc_update_queue__(netdev, queue_id, &class);
3411 hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue)
3415 struct hfsc_class *hc;
3417 hc = hfsc_class_cast__(queue);
3418 hfsc = hfsc_get__(netdev);
3420 error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
3422 hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
3429 hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
3430 struct netdev_queue_stats *stats)
3432 return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
3433 tc_make_handle(1, 0xfffe), NULL, stats);
3437 hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
3438 const struct ofpbuf *nlmsg,
3439 netdev_dump_queue_stats_cb *cb, void *aux)
3441 struct netdev_queue_stats stats;
3442 unsigned int handle, major, minor;
3445 error = tc_parse_class(nlmsg, &handle, NULL, &stats);
3450 major = tc_get_major(handle);
3451 minor = tc_get_minor(handle);
3452 if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
3453 (*cb)(minor - 1, &stats, aux);
3458 static const struct tc_ops tc_ops_hfsc = {
3459 "hfsc", /* linux_name */
3460 "linux-hfsc", /* ovs_name */
3461 HFSC_N_QUEUES, /* n_queues */
3462 hfsc_tc_install, /* tc_install */
3463 hfsc_tc_load, /* tc_load */
3464 hfsc_tc_destroy, /* tc_destroy */
3465 hfsc_qdisc_get, /* qdisc_get */
3466 hfsc_qdisc_set, /* qdisc_set */
3467 hfsc_class_get, /* class_get */
3468 hfsc_class_set, /* class_set */
3469 hfsc_class_delete, /* class_delete */
3470 hfsc_class_get_stats, /* class_get_stats */
3471 hfsc_class_dump_stats /* class_dump_stats */
3474 /* "linux-default" traffic control class.
3476 * This class represents the default, unnamed Linux qdisc. It corresponds to
3477 * the "" (empty string) QoS type in the OVS database. */
3480 default_install__(struct netdev *netdev_)
3482 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
3483 static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_default);
3485 /* Nothing but a tc class implementation is allowed to write to a tc. This
3486 * class never does that, so we can legitimately use a const tc object. */
3487 netdev->tc = CONST_CAST(struct tc *, &tc);
3491 default_tc_install(struct netdev *netdev,
3492 const struct smap *details OVS_UNUSED)
3494 default_install__(netdev);
3499 default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
3501 default_install__(netdev);
3505 static const struct tc_ops tc_ops_default = {
3506 NULL, /* linux_name */
3511 NULL, /* tc_destroy */
3512 NULL, /* qdisc_get */
3513 NULL, /* qdisc_set */
3514 NULL, /* class_get */
3515 NULL, /* class_set */
3516 NULL, /* class_delete */
3517 NULL, /* class_get_stats */
3518 NULL /* class_dump_stats */
3521 /* "linux-other" traffic control class.
3526 other_tc_load(struct netdev *netdev_, struct ofpbuf *nlmsg OVS_UNUSED)
3528 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
3529 static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_other);
3531 /* Nothing but a tc class implementation is allowed to write to a tc. This
3532 * class never does that, so we can legitimately use a const tc object. */
3533 netdev->tc = CONST_CAST(struct tc *, &tc);
3537 static const struct tc_ops tc_ops_other = {
3538 NULL, /* linux_name */
3539 "linux-other", /* ovs_name */
3541 NULL, /* tc_install */
3543 NULL, /* tc_destroy */
3544 NULL, /* qdisc_get */
3545 NULL, /* qdisc_set */
3546 NULL, /* class_get */
3547 NULL, /* class_set */
3548 NULL, /* class_delete */
3549 NULL, /* class_get_stats */
3550 NULL /* class_dump_stats */
3553 /* Traffic control. */
3555 /* Number of kernel "tc" ticks per second. */
3556 static double ticks_per_s;
3558 /* Number of kernel "jiffies" per second. This is used for the purpose of
3559 * computing buffer sizes. Generally kernel qdiscs need to be able to buffer
3560 * one jiffy's worth of data.
3562 * There are two possibilities here:
3564 * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the
3565 * approximate range of 100 to 1024. That means that we really need to
3566 * make sure that the qdisc can buffer that much data.
3568 * - 'buffer_hz' is an absurdly large number. That means that the kernel
3569 * has finely granular timers and there's no need to fudge additional room
3570 * for buffers. (There's no extra effort needed to implement that: the
3571 * large 'buffer_hz' is used as a divisor, so practically any number will
3572 * come out as 0 in the division. Small integer results in the case of
3573 * really high dividends won't have any real effect anyhow.)
3575 static unsigned int buffer_hz;
3577 /* Returns tc handle 'major':'minor'. */
3579 tc_make_handle(unsigned int major, unsigned int minor)
3581 return TC_H_MAKE(major << 16, minor);
3584 /* Returns the major number from 'handle'. */
3586 tc_get_major(unsigned int handle)
3588 return TC_H_MAJ(handle) >> 16;
3591 /* Returns the minor number from 'handle'. */
3593 tc_get_minor(unsigned int handle)
3595 return TC_H_MIN(handle);
3598 static struct tcmsg *
3599 tc_make_request(const struct netdev *netdev, int type, unsigned int flags,
3600 struct ofpbuf *request)
3602 struct tcmsg *tcmsg;
3606 error = get_ifindex(netdev, &ifindex);
3611 ofpbuf_init(request, 512);
3612 nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags);
3613 tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg);
3614 tcmsg->tcm_family = AF_UNSPEC;
3615 tcmsg->tcm_ifindex = ifindex;
3616 /* Caller should fill in tcmsg->tcm_handle. */
3617 /* Caller should fill in tcmsg->tcm_parent. */
3623 tc_transact(struct ofpbuf *request, struct ofpbuf **replyp)
3625 int error = nl_transact(NETLINK_ROUTE, request, replyp);
3626 ofpbuf_uninit(request);
3630 /* Adds or deletes a root ingress qdisc on 'netdev'. We use this for
3631 * policing configuration.
3633 * This function is equivalent to running the following when 'add' is true:
3634 * /sbin/tc qdisc add dev <devname> handle ffff: ingress
3636 * This function is equivalent to running the following when 'add' is false:
3637 * /sbin/tc qdisc del dev <devname> handle ffff: ingress
3639 * The configuration and stats may be seen with the following command:
3640 * /sbin/tc -s qdisc show dev <devname>
3642 * Returns 0 if successful, otherwise a positive errno value.
3645 tc_add_del_ingress_qdisc(struct netdev *netdev, bool add)
3647 struct ofpbuf request;
3648 struct tcmsg *tcmsg;
3650 int type = add ? RTM_NEWQDISC : RTM_DELQDISC;
3651 int flags = add ? NLM_F_EXCL | NLM_F_CREATE : 0;
3653 tcmsg = tc_make_request(netdev, type, flags, &request);
3657 tcmsg->tcm_handle = tc_make_handle(0xffff, 0);
3658 tcmsg->tcm_parent = TC_H_INGRESS;
3659 nl_msg_put_string(&request, TCA_KIND, "ingress");
3660 nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0);
3662 error = tc_transact(&request, NULL);
3664 /* If we're deleting the qdisc, don't worry about some of the
3665 * error conditions. */
3666 if (!add && (error == ENOENT || error == EINVAL)) {
3675 /* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size
3678 * This function is equivalent to running:
3679 * /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49
3680 * basic police rate <kbits_rate>kbit burst <kbits_burst>k
3683 * The configuration and stats may be seen with the following command:
3684 * /sbin/tc -s filter show <devname> eth0 parent ffff:
3686 * Returns 0 if successful, otherwise a positive errno value.
3689 tc_add_policer(struct netdev *netdev, int kbits_rate, int kbits_burst)
3691 struct tc_police tc_police;
3692 struct ofpbuf request;
3693 struct tcmsg *tcmsg;
3694 size_t basic_offset;
3695 size_t police_offset;
3699 memset(&tc_police, 0, sizeof tc_police);
3700 tc_police.action = TC_POLICE_SHOT;
3701 tc_police.mtu = mtu;
3702 tc_fill_rate(&tc_police.rate, (kbits_rate * 1000)/8, mtu);
3703 tc_police.burst = tc_bytes_to_ticks(tc_police.rate.rate,
3704 kbits_burst * 1024);
3706 tcmsg = tc_make_request(netdev, RTM_NEWTFILTER,
3707 NLM_F_EXCL | NLM_F_CREATE, &request);
3711 tcmsg->tcm_parent = tc_make_handle(0xffff, 0);
3712 tcmsg->tcm_info = tc_make_handle(49,
3713 (OVS_FORCE uint16_t) htons(ETH_P_ALL));
3715 nl_msg_put_string(&request, TCA_KIND, "basic");
3716 basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
3717 police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE);
3718 nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police);
3719 tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate);
3720 nl_msg_end_nested(&request, police_offset);
3721 nl_msg_end_nested(&request, basic_offset);
3723 error = tc_transact(&request, NULL);
3734 /* The values in psched are not individually very meaningful, but they are
3735 * important. The tables below show some values seen in the wild.
3739 * - "c" has always been a constant 1000000 since at least Linux 2.4.14.
3740 * (Before that, there are hints that it was 1000000000.)
3742 * - "d" can be unrealistically large, see the comment on 'buffer_hz'
3746 * -----------------------------------
3747 * [1] 000c8000 000f4240 000f4240 00000064
3748 * [2] 000003e8 00000400 000f4240 3b9aca00
3749 * [3] 000003e8 00000400 000f4240 3b9aca00
3750 * [4] 000003e8 00000400 000f4240 00000064
3751 * [5] 000003e8 00000040 000f4240 3b9aca00
3752 * [6] 000003e8 00000040 000f4240 000000f9
3754 * a b c d ticks_per_s buffer_hz
3755 * ------- --------- ---------- ------------- ----------- -------------
3756 * [1] 819,200 1,000,000 1,000,000 100 819,200 100
3757 * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
3758 * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
3759 * [4] 1,000 1,024 1,000,000 100 976,562 100
3760 * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000
3761 * [6] 1,000 64 1,000,000 249 15,625,000 249
3763 * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p
3764 * [2] 2.6.26-1-686-bigmem from Debian lenny
3765 * [3] 2.6.26-2-sparc64 from Debian lenny
3766 * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p
3767 * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion
3768 * [6] 2.6.34 from kernel.org on KVM
3770 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
3771 static const char fn[] = "/proc/net/psched";
3772 unsigned int a, b, c, d;
3775 if (!ovsthread_once_start(&once)) {
3782 stream = fopen(fn, "r");
3784 VLOG_WARN("%s: open failed: %s", fn, ovs_strerror(errno));
3788 if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) {
3789 VLOG_WARN("%s: read failed", fn);
3793 VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d);
3797 VLOG_WARN("%s: invalid scheduler parameters", fn);
3801 ticks_per_s = (double) a * c / b;
3805 VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u",
3808 VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz);
3811 ovsthread_once_done(&once);
3814 /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a
3815 * rate of 'rate' bytes per second. */
3817 tc_ticks_to_bytes(unsigned int rate, unsigned int ticks)
3820 return (rate * ticks) / ticks_per_s;
3823 /* Returns the number of ticks that it would take to transmit 'size' bytes at a
3824 * rate of 'rate' bytes per second. */
3826 tc_bytes_to_ticks(unsigned int rate, unsigned int size)
3829 return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0;
3832 /* Returns the number of bytes that need to be reserved for qdisc buffering at
3833 * a transmission rate of 'rate' bytes per second. */
3835 tc_buffer_per_jiffy(unsigned int rate)
3838 return rate / buffer_hz;
3841 /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc,
3842 * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull,
3843 * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or
3844 * stores NULL into it if it is absent.
3846 * '*kind' and '*options' point into 'msg', so they are owned by whoever owns
3849 * Returns 0 if successful, otherwise a positive errno value. */
3851 tc_parse_qdisc(const struct ofpbuf *msg, const char **kind,
3852 struct nlattr **options)
3854 static const struct nl_policy tca_policy[] = {
3855 [TCA_KIND] = { .type = NL_A_STRING, .optional = false },
3856 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
3858 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
3860 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
3861 tca_policy, ta, ARRAY_SIZE(ta))) {
3862 VLOG_WARN_RL(&rl, "failed to parse qdisc message");
3867 *kind = nl_attr_get_string(ta[TCA_KIND]);
3871 *options = ta[TCA_OPTIONS];
3886 /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the
3887 * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute
3888 * into '*options', and its queue statistics into '*stats'. Any of the output
3889 * arguments may be null.
3891 * Returns 0 if successful, otherwise a positive errno value. */
3893 tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep,
3894 struct nlattr **options, struct netdev_queue_stats *stats)
3896 static const struct nl_policy tca_policy[] = {
3897 [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false },
3898 [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false },
3900 struct nlattr *ta[ARRAY_SIZE(tca_policy)];
3902 if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg),
3903 tca_policy, ta, ARRAY_SIZE(ta))) {
3904 VLOG_WARN_RL(&rl, "failed to parse class message");
3909 struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc);
3910 *handlep = tc->tcm_handle;
3914 *options = ta[TCA_OPTIONS];
3918 const struct gnet_stats_queue *gsq;
3919 struct gnet_stats_basic gsb;
3921 static const struct nl_policy stats_policy[] = {
3922 [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false,
3923 .min_len = sizeof gsb },
3924 [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false,
3925 .min_len = sizeof *gsq },
3927 struct nlattr *sa[ARRAY_SIZE(stats_policy)];
3929 if (!nl_parse_nested(ta[TCA_STATS2], stats_policy,
3930 sa, ARRAY_SIZE(sa))) {
3931 VLOG_WARN_RL(&rl, "failed to parse class stats");
3935 /* Alignment issues screw up the length of struct gnet_stats_basic on
3936 * some arch/bitsize combinations. Newer versions of Linux have a
3937 * struct gnet_stats_basic_packed, but we can't depend on that. The
3938 * easiest thing to do is just to make a copy. */
3939 memset(&gsb, 0, sizeof gsb);
3940 memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]),
3941 MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb));
3942 stats->tx_bytes = gsb.bytes;
3943 stats->tx_packets = gsb.packets;
3945 gsq = nl_attr_get(sa[TCA_STATS_QUEUE]);
3946 stats->tx_errors = gsq->drops;
3956 memset(stats, 0, sizeof *stats);
3961 /* Queries the kernel for class with identifier 'handle' and parent 'parent'
3964 tc_query_class(const struct netdev *netdev,
3965 unsigned int handle, unsigned int parent,
3966 struct ofpbuf **replyp)
3968 struct ofpbuf request;
3969 struct tcmsg *tcmsg;
3972 tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request);
3976 tcmsg->tcm_handle = handle;
3977 tcmsg->tcm_parent = parent;
3979 error = tc_transact(&request, replyp);
3981 VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)",
3982 netdev_get_name(netdev),
3983 tc_get_major(handle), tc_get_minor(handle),
3984 tc_get_major(parent), tc_get_minor(parent),
3985 ovs_strerror(error));
3990 /* Equivalent to "tc class del dev <name> handle <handle>". */
3992 tc_delete_class(const struct netdev *netdev, unsigned int handle)
3994 struct ofpbuf request;
3995 struct tcmsg *tcmsg;
3998 tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request);
4002 tcmsg->tcm_handle = handle;
4003 tcmsg->tcm_parent = 0;
4005 error = tc_transact(&request, NULL);
4007 VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)",
4008 netdev_get_name(netdev),
4009 tc_get_major(handle), tc_get_minor(handle),
4010 ovs_strerror(error));
4015 /* Equivalent to "tc qdisc del dev <name> root". */
4017 tc_del_qdisc(struct netdev *netdev_)
4019 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
4020 struct ofpbuf request;
4021 struct tcmsg *tcmsg;
4024 tcmsg = tc_make_request(netdev_, RTM_DELQDISC, 0, &request);
4028 tcmsg->tcm_handle = tc_make_handle(1, 0);
4029 tcmsg->tcm_parent = TC_H_ROOT;
4031 error = tc_transact(&request, NULL);
4032 if (error == EINVAL) {
4033 /* EINVAL probably means that the default qdisc was in use, in which
4034 * case we've accomplished our purpose. */
4037 if (!error && netdev->tc) {
4038 if (netdev->tc->ops->tc_destroy) {
4039 netdev->tc->ops->tc_destroy(netdev->tc);
4046 /* If 'netdev''s qdisc type and parameters are not yet known, queries the
4047 * kernel to determine what they are. Returns 0 if successful, otherwise a
4048 * positive errno value. */
4050 tc_query_qdisc(const struct netdev *netdev_)
4052 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
4053 struct ofpbuf request, *qdisc;
4054 const struct tc_ops *ops;
4055 struct tcmsg *tcmsg;
4063 /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have
4064 * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before
4065 * 2.6.35 without that fix backported to it.
4067 * To avoid the OOPS, we must not make a request that would attempt to dump
4068 * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a
4069 * few others. There are a few ways that I can see to do this, but most of
4070 * them seem to be racy (and if you lose the race the kernel OOPSes). The
4071 * technique chosen here is to assume that any non-default qdisc that we
4072 * create will have a class with handle 1:0. The built-in qdiscs only have
4073 * a class with handle 0:0.
4075 * We could check for Linux 2.6.35+ and use a more straightforward method
4077 tcmsg = tc_make_request(netdev_, RTM_GETQDISC, NLM_F_ECHO, &request);
4081 tcmsg->tcm_handle = tc_make_handle(1, 0);
4082 tcmsg->tcm_parent = 0;
4084 /* Figure out what tc class to instantiate. */
4085 error = tc_transact(&request, &qdisc);
4089 error = tc_parse_qdisc(qdisc, &kind, NULL);
4091 ops = &tc_ops_other;
4093 ops = tc_lookup_linux_name(kind);
4095 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1);
4096 VLOG_INFO_RL(&rl2, "unknown qdisc \"%s\"", kind);
4098 ops = &tc_ops_other;
4101 } else if (error == ENOENT) {
4102 /* Either it's a built-in qdisc, or it's a qdisc set up by some
4103 * other entity that doesn't have a handle 1:0. We will assume
4104 * that it's the system default qdisc. */
4105 ops = &tc_ops_default;
4108 /* Who knows? Maybe the device got deleted. */
4109 VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)",
4110 netdev_get_name(netdev_), ovs_strerror(error));
4111 ops = &tc_ops_other;
4114 /* Instantiate it. */
4115 load_error = ops->tc_load(CONST_CAST(struct netdev *, netdev_), qdisc);
4116 ovs_assert((load_error == 0) == (netdev->tc != NULL));
4117 ofpbuf_delete(qdisc);
4119 return error ? error : load_error;
4122 /* Linux traffic control uses tables with 256 entries ("rtab" tables) to
4123 approximate the time to transmit packets of various lengths. For an MTU of
4124 256 or less, each entry is exact; for an MTU of 257 through 512, each entry
4125 represents two possible packet lengths; for a MTU of 513 through 1024, four
4126 possible lengths; and so on.
4128 Returns, for the specified 'mtu', the number of bits that packet lengths
4129 need to be shifted right to fit within such a 256-entry table. */
4131 tc_calc_cell_log(unsigned int mtu)
4136 mtu = ETH_PAYLOAD_MAX;
4138 mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN;
4140 for (cell_log = 0; mtu >= 256; cell_log++) {
4147 /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU
4150 tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu)
4152 memset(rate, 0, sizeof *rate);
4153 rate->cell_log = tc_calc_cell_log(mtu);
4154 /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */
4155 /* rate->cell_align = 0; */ /* distro headers. */
4156 rate->mpu = ETH_TOTAL_MIN;
4160 /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink
4161 * attribute of the specified "type".
4163 * See tc_calc_cell_log() above for a description of "rtab"s. */
4165 tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate)
4170 rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE);
4171 for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) {
4172 unsigned packet_size = (i + 1) << rate->cell_log;
4173 if (packet_size < rate->mpu) {
4174 packet_size = rate->mpu;
4176 rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size);
4180 /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a
4181 * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested
4182 * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of
4185 tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes)
4187 unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu;
4188 return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst));
4191 /* Linux-only functions declared in netdev-linux.h */
4193 /* Returns a fd for an AF_INET socket or a negative errno value. */
4195 netdev_linux_get_af_inet_sock(void)
4197 int error = netdev_linux_init();
4198 return error ? -error : af_inet_sock;
4201 /* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If
4202 * 'enable' is true, the bit is set. Otherwise, it is cleared. */
4204 netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag,
4205 const char *flag_name, bool enable)
4207 const char *netdev_name = netdev_get_name(netdev);
4208 struct ethtool_value evalue;
4212 COVERAGE_INC(netdev_get_ethtool);
4213 memset(&evalue, 0, sizeof evalue);
4214 error = netdev_linux_do_ethtool(netdev_name,
4215 (struct ethtool_cmd *)&evalue,
4216 ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4221 COVERAGE_INC(netdev_set_ethtool);
4222 evalue.data = new_flags = (evalue.data & ~flag) | (enable ? flag : 0);
4223 error = netdev_linux_do_ethtool(netdev_name,
4224 (struct ethtool_cmd *)&evalue,
4225 ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS");
4230 COVERAGE_INC(netdev_get_ethtool);
4231 memset(&evalue, 0, sizeof evalue);
4232 error = netdev_linux_do_ethtool(netdev_name,
4233 (struct ethtool_cmd *)&evalue,
4234 ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4239 if (new_flags != evalue.data) {
4240 VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network "
4241 "device %s failed", enable ? "enable" : "disable",
4242 flag_name, netdev_name);
4249 /* Utility functions. */
4251 /* Copies 'src' into 'dst', performing format conversion in the process. */
4253 netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst,
4254 const struct rtnl_link_stats *src)
4256 dst->rx_packets = src->rx_packets;
4257 dst->tx_packets = src->tx_packets;
4258 dst->rx_bytes = src->rx_bytes;
4259 dst->tx_bytes = src->tx_bytes;
4260 dst->rx_errors = src->rx_errors;
4261 dst->tx_errors = src->tx_errors;
4262 dst->rx_dropped = src->rx_dropped;
4263 dst->tx_dropped = src->tx_dropped;
4264 dst->multicast = src->multicast;
4265 dst->collisions = src->collisions;
4266 dst->rx_length_errors = src->rx_length_errors;
4267 dst->rx_over_errors = src->rx_over_errors;
4268 dst->rx_crc_errors = src->rx_crc_errors;
4269 dst->rx_frame_errors = src->rx_frame_errors;
4270 dst->rx_fifo_errors = src->rx_fifo_errors;
4271 dst->rx_missed_errors = src->rx_missed_errors;
4272 dst->tx_aborted_errors = src->tx_aborted_errors;
4273 dst->tx_carrier_errors = src->tx_carrier_errors;
4274 dst->tx_fifo_errors = src->tx_fifo_errors;
4275 dst->tx_heartbeat_errors = src->tx_heartbeat_errors;
4276 dst->tx_window_errors = src->tx_window_errors;
4280 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
4282 /* Policy for RTNLGRP_LINK messages.
4284 * There are *many* more fields in these messages, but currently we only
4285 * care about these fields. */
4286 static const struct nl_policy rtnlgrp_link_policy[] = {
4287 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
4288 [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
4289 .min_len = sizeof(struct rtnl_link_stats) },
4292 struct ofpbuf request;
4293 struct ofpbuf *reply;
4294 struct ifinfomsg *ifi;
4295 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
4298 ofpbuf_init(&request, 0);
4299 nl_msg_put_nlmsghdr(&request, sizeof *ifi, RTM_GETLINK, NLM_F_REQUEST);
4300 ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
4301 ifi->ifi_family = PF_UNSPEC;
4302 ifi->ifi_index = ifindex;
4303 error = nl_transact(NETLINK_ROUTE, &request, &reply);
4304 ofpbuf_uninit(&request);
4309 if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
4310 rtnlgrp_link_policy,
4311 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
4312 ofpbuf_delete(reply);
4316 if (!attrs[IFLA_STATS]) {
4317 VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
4318 ofpbuf_delete(reply);
4322 netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(attrs[IFLA_STATS]));
4324 ofpbuf_delete(reply);
4330 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
4332 static const char fn[] = "/proc/net/dev";
4337 stream = fopen(fn, "r");
4339 VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno));
4344 while (fgets(line, sizeof line, stream)) {
4347 #define X64 "%"SCNu64
4350 X64 X64 X64 X64 X64 X64 X64 "%*u"
4351 X64 X64 X64 X64 X64 X64 X64 "%*u",
4357 &stats->rx_fifo_errors,
4358 &stats->rx_frame_errors,
4364 &stats->tx_fifo_errors,
4366 &stats->tx_carrier_errors) != 15) {
4367 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
4368 } else if (!strcmp(devname, netdev_name)) {
4369 stats->rx_length_errors = UINT64_MAX;
4370 stats->rx_over_errors = UINT64_MAX;
4371 stats->rx_crc_errors = UINT64_MAX;
4372 stats->rx_missed_errors = UINT64_MAX;
4373 stats->tx_aborted_errors = UINT64_MAX;
4374 stats->tx_heartbeat_errors = UINT64_MAX;
4375 stats->tx_window_errors = UINT64_MAX;
4381 VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
4387 get_flags(const struct netdev *dev, unsigned int *flags)
4393 error = netdev_linux_do_ioctl(dev->name, &ifr, SIOCGIFFLAGS,
4396 *flags = ifr.ifr_flags;
4402 set_flags(const char *name, unsigned int flags)
4406 ifr.ifr_flags = flags;
4407 return netdev_linux_do_ioctl(name, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS");
4411 do_get_ifindex(const char *netdev_name)
4415 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4416 COVERAGE_INC(netdev_get_ifindex);
4417 if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
4418 VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
4419 netdev_name, ovs_strerror(errno));
4422 return ifr.ifr_ifindex;
4426 get_ifindex(const struct netdev *netdev_, int *ifindexp)
4428 struct netdev_linux *netdev = netdev_linux_cast(netdev_);
4430 if (!(netdev->cache_valid & VALID_IFINDEX)) {
4431 int ifindex = do_get_ifindex(netdev_get_name(netdev_));
4434 netdev->get_ifindex_error = -ifindex;
4435 netdev->ifindex = 0;
4437 netdev->get_ifindex_error = 0;
4438 netdev->ifindex = ifindex;
4440 netdev->cache_valid |= VALID_IFINDEX;
4443 *ifindexp = netdev->ifindex;
4444 return netdev->get_ifindex_error;
4448 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
4453 memset(&ifr, 0, sizeof ifr);
4454 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4455 COVERAGE_INC(netdev_get_hwaddr);
4456 if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
4457 /* ENODEV probably means that a vif disappeared asynchronously and
4458 * hasn't been removed from the database yet, so reduce the log level
4459 * to INFO for that case. */
4460 VLOG(errno == ENODEV ? VLL_INFO : VLL_ERR,
4461 "ioctl(SIOCGIFHWADDR) on %s device failed: %s",
4462 netdev_name, ovs_strerror(errno));
4465 hwaddr_family = ifr.ifr_hwaddr.sa_family;
4466 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
4467 VLOG_WARN("%s device has unknown hardware address family %d",
4468 netdev_name, hwaddr_family);
4470 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
4475 set_etheraddr(const char *netdev_name,
4476 const uint8_t mac[ETH_ADDR_LEN])
4480 memset(&ifr, 0, sizeof ifr);
4481 ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
4482 ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
4483 memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
4484 COVERAGE_INC(netdev_set_hwaddr);
4485 if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
4486 VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
4487 netdev_name, ovs_strerror(errno));
4494 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
4495 int cmd, const char *cmd_name)
4499 memset(&ifr, 0, sizeof ifr);
4500 ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
4501 ifr.ifr_data = (caddr_t) ecmd;
4504 if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
4507 if (errno != EOPNOTSUPP) {
4508 VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
4509 "failed: %s", cmd_name, name, ovs_strerror(errno));
4511 /* The device doesn't support this operation. That's pretty
4512 * common, so there's no point in logging anything. */
4519 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
4520 const char *cmd_name)
4522 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
4523 if (ioctl(af_inet_sock, cmd, ifr) == -1) {
4524 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
4525 ovs_strerror(errno));
4532 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
4533 int cmd, const char *cmd_name)
4538 ifr.ifr_addr.sa_family = AF_INET;
4539 error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
4541 const struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *,
4543 *ip = sin->sin_addr;
4548 /* Returns an AF_PACKET raw socket or a negative errno value. */
4550 af_packet_sock(void)
4552 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
4555 if (ovsthread_once_start(&once)) {
4556 sock = socket(AF_PACKET, SOCK_RAW, 0);
4558 int error = set_nonblocking(sock);
4565 VLOG_ERR("failed to create packet socket: %s",
4566 ovs_strerror(errno));
4568 ovsthread_once_done(&once);