+netdev_linux_poll_add(struct netdev *netdev,
+ void (*cb)(struct netdev_notifier *), void *aux,
+ struct netdev_notifier **notifierp)
+{
+ const char *netdev_name = netdev_get_name(netdev);
+ struct netdev_linux_notifier *notifier;
+ struct list *list;
+
+ if (shash_is_empty(&netdev_linux_notifiers)) {
+ int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier,
+ netdev_linux_poll_cb, NULL);
+ if (error) {
+ return error;
+ }
+ }
+
+ list = shash_find_data(&netdev_linux_notifiers, netdev_name);
+ if (!list) {
+ list = xmalloc(sizeof *list);
+ list_init(list);
+ shash_add(&netdev_linux_notifiers, netdev_name, list);
+ }
+
+ notifier = xmalloc(sizeof *notifier);
+ netdev_notifier_init(¬ifier->notifier, netdev, cb, aux);
+ list_push_back(list, ¬ifier->node);
+ *notifierp = ¬ifier->notifier;
+ return 0;
+}
+
+static void
+netdev_linux_poll_remove(struct netdev_notifier *notifier_)
+{
+ struct netdev_linux_notifier *notifier =
+ CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier);
+ struct list *list;
+
+ /* Remove 'notifier' from its list. */
+ list = list_remove(¬ifier->node);
+ if (list_is_empty(list)) {
+ /* The list is now empty. Remove it from the hash and free it. */
+ const char *netdev_name = netdev_get_name(notifier->notifier.netdev);
+ shash_delete(&netdev_linux_notifiers,
+ shash_find(&netdev_linux_notifiers, netdev_name));
+ free(list);
+ }
+ free(notifier);
+
+ /* If that was the last notifier, unregister. */
+ if (shash_is_empty(&netdev_linux_notifiers)) {
+ rtnetlink_notifier_unregister(&netdev_linux_poll_notifier);
+ }
+}
+
+const struct netdev_class netdev_linux_class = {
+ "system",
+
+ netdev_linux_init,
+ netdev_linux_run,
+ netdev_linux_wait,
+
+ netdev_linux_create_system,
+ netdev_linux_destroy,
+ NULL, /* reconfigure */
+
+ netdev_linux_open,
+ netdev_linux_close,
+
+ netdev_linux_enumerate,
+
+ netdev_linux_recv,
+ netdev_linux_recv_wait,
+ netdev_linux_drain,
+
+ netdev_linux_send,
+ netdev_linux_send_wait,
+
+ netdev_linux_set_etheraddr,
+ netdev_linux_get_etheraddr,
+ netdev_linux_get_mtu,
+ netdev_linux_get_ifindex,
+ netdev_linux_get_carrier,
+ netdev_linux_get_stats,
+ netdev_vport_set_stats,
+
+ netdev_linux_get_features,
+ netdev_linux_set_advertisements,
+ netdev_linux_get_vlan_vid,
+
+ netdev_linux_set_policing,
+ netdev_linux_get_qos_types,
+ netdev_linux_get_qos_capabilities,
+ netdev_linux_get_qos,
+ netdev_linux_set_qos,
+ netdev_linux_get_queue,
+ netdev_linux_set_queue,
+ netdev_linux_delete_queue,
+ netdev_linux_get_queue_stats,
+ netdev_linux_dump_queues,
+ netdev_linux_dump_queue_stats,
+
+ netdev_linux_get_in4,
+ netdev_linux_set_in4,
+ netdev_linux_get_in6,
+ netdev_linux_add_router,
+ netdev_linux_get_next_hop,
+ netdev_linux_arp_lookup,
+
+ netdev_linux_update_flags,
+
+ netdev_linux_poll_add,
+ netdev_linux_poll_remove,
+};
+
+const struct netdev_class netdev_tap_class = {
+ "tap",
+
+ netdev_linux_init,
+ netdev_linux_run,
+ netdev_linux_wait,
+
+ netdev_linux_create_tap,
+ netdev_linux_destroy,
+ NULL, /* reconfigure */
+
+ netdev_linux_open,
+ netdev_linux_close,
+
+ NULL, /* enumerate */
+
+ netdev_linux_recv,
+ netdev_linux_recv_wait,
+ netdev_linux_drain,
+
+ netdev_linux_send,
+ netdev_linux_send_wait,
+
+ netdev_linux_set_etheraddr,
+ netdev_linux_get_etheraddr,
+ netdev_linux_get_mtu,
+ netdev_linux_get_ifindex,
+ netdev_linux_get_carrier,
+ netdev_linux_get_stats,
+ NULL, /* set_stats */
+
+ netdev_linux_get_features,
+ netdev_linux_set_advertisements,
+ netdev_linux_get_vlan_vid,
+
+ netdev_linux_set_policing,
+ netdev_linux_get_qos_types,
+ netdev_linux_get_qos_capabilities,
+ netdev_linux_get_qos,
+ netdev_linux_set_qos,
+ netdev_linux_get_queue,
+ netdev_linux_set_queue,
+ netdev_linux_delete_queue,
+ netdev_linux_get_queue_stats,
+ netdev_linux_dump_queues,
+ netdev_linux_dump_queue_stats,
+
+ netdev_linux_get_in4,
+ netdev_linux_set_in4,
+ netdev_linux_get_in6,
+ netdev_linux_add_router,
+ netdev_linux_get_next_hop,
+ netdev_linux_arp_lookup,
+
+ netdev_linux_update_flags,
+
+ netdev_linux_poll_add,
+ netdev_linux_poll_remove,
+};
+\f
+/* HTB traffic control class. */
+
+#define HTB_N_QUEUES 0xf000
+
+struct htb {
+ struct tc tc;
+ unsigned int max_rate; /* In bytes/s. */
+};
+
+struct htb_class {
+ unsigned int min_rate; /* In bytes/s. */
+ unsigned int max_rate; /* In bytes/s. */
+ unsigned int burst; /* In bytes. */
+ unsigned int priority; /* Lower values are higher priorities. */
+};
+
+static struct htb *
+htb_get__(const struct netdev *netdev)
+{
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev));
+ return CONTAINER_OF(netdev_dev->tc, struct htb, tc);
+}
+
+static struct htb *
+htb_install__(struct netdev *netdev, uint64_t max_rate)
+{
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev));
+ struct htb *htb;
+
+ htb = xmalloc(sizeof *htb);
+ tc_init(&htb->tc, &tc_ops_htb);
+ htb->max_rate = max_rate;
+
+ netdev_dev->tc = &htb->tc;
+
+ return htb;
+}
+
+/* Create an HTB qdisc.
+ *
+ * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default
+ * 0". */
+static int
+htb_setup_qdisc__(struct netdev *netdev)
+{
+ size_t opt_offset;
+ struct tc_htb_glob opt;
+ struct ofpbuf request;
+ struct tcmsg *tcmsg;
+
+ tc_del_qdisc(netdev);
+
+ tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
+ NLM_F_EXCL | NLM_F_CREATE, &request);
+ tcmsg->tcm_handle = tc_make_handle(1, 0);
+ tcmsg->tcm_parent = TC_H_ROOT;
+
+ nl_msg_put_string(&request, TCA_KIND, "htb");
+
+ memset(&opt, 0, sizeof opt);
+ opt.rate2quantum = 10;
+ opt.version = 3;
+ opt.defcls = 0;
+
+ opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
+ nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt);
+ nl_msg_end_nested(&request, opt_offset);
+
+ return tc_transact(&request, NULL);
+}
+
+/* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb
+ * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */
+static int
+htb_setup_class__(struct netdev *netdev, unsigned int handle,
+ unsigned int parent, struct htb_class *class)
+{
+ size_t opt_offset;
+ struct tc_htb_opt opt;
+ struct ofpbuf request;
+ struct tcmsg *tcmsg;
+ int error;
+ int mtu;
+
+ netdev_get_mtu(netdev, &mtu);
+
+ memset(&opt, 0, sizeof opt);
+ tc_fill_rate(&opt.rate, class->min_rate, mtu);
+ tc_fill_rate(&opt.ceil, class->max_rate, mtu);
+ opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst);
+ opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst);
+ opt.prio = class->priority;
+
+ tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
+ tcmsg->tcm_handle = handle;
+ tcmsg->tcm_parent = parent;
+
+ nl_msg_put_string(&request, TCA_KIND, "htb");
+ opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
+ nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt);
+ tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate);
+ tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil);
+ nl_msg_end_nested(&request, opt_offset);
+
+ error = tc_transact(&request, NULL);
+ if (error) {
+ VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
+ "min_rate=%u max_rate=%u burst=%u prio=%u (%s)",
+ netdev_get_name(netdev),
+ tc_get_major(handle), tc_get_minor(handle),
+ tc_get_major(parent), tc_get_minor(parent),
+ class->min_rate, class->max_rate,
+ class->burst, class->priority, strerror(error));
+ }
+ return error;
+}
+
+/* Parses Netlink attributes in 'options' for HTB parameters and stores a
+ * description of them into 'details'. The description complies with the
+ * specification given in the vswitch database documentation for linux-htb
+ * queue details. */
+static int
+htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class)
+{
+ static const struct nl_policy tca_htb_policy[] = {
+ [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false,
+ .min_len = sizeof(struct tc_htb_opt) },
+ };
+
+ struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)];
+ const struct tc_htb_opt *htb;
+
+ if (!nl_parse_nested(nl_options, tca_htb_policy,
+ attrs, ARRAY_SIZE(tca_htb_policy))) {
+ VLOG_WARN_RL(&rl, "failed to parse HTB class options");
+ return EPROTO;
+ }
+
+ htb = nl_attr_get(attrs[TCA_HTB_PARMS]);
+ class->min_rate = htb->rate.rate;
+ class->max_rate = htb->ceil.rate;
+ class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer);
+ class->priority = htb->prio;
+ return 0;
+}
+
+static int
+htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
+ struct htb_class *options,
+ struct netdev_queue_stats *stats)
+{
+ struct nlattr *nl_options;
+ unsigned int handle;
+ int error;
+
+ error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
+ if (!error && queue_id) {
+ unsigned int major = tc_get_major(handle);
+ unsigned int minor = tc_get_minor(handle);
+ if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
+ *queue_id = minor - 1;
+ } else {
+ error = EPROTO;
+ }
+ }
+ if (!error && options) {
+ error = htb_parse_tca_options__(nl_options, options);
+ }
+ return error;
+}
+
+static void
+htb_parse_qdisc_details__(struct netdev *netdev,
+ const struct shash *details, struct htb_class *hc)
+{
+ const char *max_rate_s;
+
+ max_rate_s = shash_find_data(details, "max-rate");
+ hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
+ if (!hc->max_rate) {
+ uint32_t current;
+
+ netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
+ hc->max_rate = netdev_features_to_bps(current) / 8;
+ }
+ hc->min_rate = hc->max_rate;
+ hc->burst = 0;
+ hc->priority = 0;
+}
+
+static int
+htb_parse_class_details__(struct netdev *netdev,
+ const struct shash *details, struct htb_class *hc)
+{
+ const struct htb *htb = htb_get__(netdev);
+ const char *min_rate_s = shash_find_data(details, "min-rate");
+ const char *max_rate_s = shash_find_data(details, "max-rate");
+ const char *burst_s = shash_find_data(details, "burst");
+ const char *priority_s = shash_find_data(details, "priority");
+ int mtu;
+
+ /* min-rate */
+ if (!min_rate_s) {
+ /* min-rate is required. */
+ return EINVAL;
+ }
+ hc->min_rate = strtoull(min_rate_s, NULL, 10) / 8;
+ hc->min_rate = MAX(hc->min_rate, 0);
+ hc->min_rate = MIN(hc->min_rate, htb->max_rate);
+
+ /* max-rate */
+ hc->max_rate = (max_rate_s
+ ? strtoull(max_rate_s, NULL, 10) / 8
+ : htb->max_rate);
+ hc->max_rate = MAX(hc->max_rate, hc->min_rate);
+ hc->max_rate = MIN(hc->max_rate, htb->max_rate);
+
+ /* burst
+ *
+ * According to hints in the documentation that I've read, it is important
+ * that 'burst' be at least as big as the largest frame that might be
+ * transmitted. Also, making 'burst' a bit bigger than necessary is OK,
+ * but having it a bit too small is a problem. Since netdev_get_mtu()
+ * doesn't include the Ethernet header, we need to add at least 14 (18?) to
+ * the MTU. We actually add 64, instead of 14, as a guard against
+ * additional headers get tacked on somewhere that we're not aware of. */
+ netdev_get_mtu(netdev, &mtu);
+ hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0;
+ hc->burst = MAX(hc->burst, mtu + 64);
+
+ /* priority */
+ hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0;
+
+ return 0;
+}
+
+static int
+htb_query_class__(const struct netdev *netdev, unsigned int handle,
+ unsigned int parent, struct htb_class *options,
+ struct netdev_queue_stats *stats)
+{
+ struct ofpbuf *reply;
+ int error;
+
+ error = tc_query_class(netdev, handle, parent, &reply);
+ if (!error) {
+ error = htb_parse_tcmsg__(reply, NULL, options, stats);
+ ofpbuf_delete(reply);
+ }
+ return error;
+}
+
+static int
+htb_tc_install(struct netdev *netdev, const struct shash *details)
+{
+ int error;
+
+ error = htb_setup_qdisc__(netdev);
+ if (!error) {
+ struct htb_class hc;
+
+ htb_parse_qdisc_details__(netdev, details, &hc);
+ error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
+ tc_make_handle(1, 0), &hc);
+ if (!error) {
+ htb_install__(netdev, hc.max_rate);
+ }
+ }
+ return error;
+}
+
+static void
+htb_update_queue__(struct netdev *netdev, unsigned int queue_id,
+ const struct htb_class *hc)
+{
+ struct htb *htb = htb_get__(netdev);
+ struct htb_class *hcp;
+
+ hcp = port_array_get(&htb->tc.queues, queue_id);
+ if (!hcp) {
+ hcp = xmalloc(sizeof *hcp);
+ port_array_set(&htb->tc.queues, queue_id, hcp);
+ }
+ *hcp = *hc;
+}
+
+static int
+htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
+{
+ struct shash details = SHASH_INITIALIZER(&details);
+ struct ofpbuf msg;
+ struct nl_dump dump;
+ struct htb_class hc;
+ struct htb *htb;
+
+ /* Get qdisc options. */
+ hc.max_rate = 0;
+ htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
+ htb = htb_install__(netdev, hc.max_rate);
+
+ /* Get queues. */
+ start_queue_dump(netdev, &dump);
+ shash_init(&details);
+ while (nl_dump_next(&dump, &msg)) {
+ unsigned int queue_id;
+
+ if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
+ htb_update_queue__(netdev, queue_id, &hc);
+ }
+ }
+ nl_dump_done(&dump);
+
+ return 0;
+}
+
+static void
+htb_tc_destroy(struct tc *tc)
+{
+ struct htb *htb = CONTAINER_OF(tc, struct htb, tc);
+ unsigned int queue_id;
+ struct htb_class *hc;
+
+ PORT_ARRAY_FOR_EACH (hc, &htb->tc.queues, queue_id) {
+ free(hc);
+ }
+ tc_destroy(tc);
+ free(htb);
+}
+
+static int
+htb_qdisc_get(const struct netdev *netdev, struct shash *details)
+{
+ const struct htb *htb = htb_get__(netdev);
+ shash_add(details, "max-rate", xasprintf("%llu", 8ULL * htb->max_rate));
+ return 0;
+}
+
+static int
+htb_qdisc_set(struct netdev *netdev, const struct shash *details)
+{
+ struct htb_class hc;
+ int error;
+
+ htb_parse_qdisc_details__(netdev, details, &hc);
+ error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe),
+ tc_make_handle(1, 0), &hc);
+ if (!error) {
+ htb_get__(netdev)->max_rate = hc.max_rate;
+ }
+ return error;
+}
+
+static int
+htb_class_get(const struct netdev *netdev, unsigned int queue_id,
+ struct shash *details)
+{
+ const struct htb *htb = htb_get__(netdev);
+ const struct htb_class *hc;
+
+ hc = port_array_get(&htb->tc.queues, queue_id);
+ assert(hc != NULL);
+
+ shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate));
+ if (hc->min_rate != hc->max_rate) {
+ shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate));
+ }
+ shash_add(details, "burst", xasprintf("%llu", 8ULL * hc->burst));
+ if (hc->priority) {
+ shash_add(details, "priority", xasprintf("%u", hc->priority));
+ }
+ return 0;
+}
+
+static int
+htb_class_set(struct netdev *netdev, unsigned int queue_id,
+ const struct shash *details)
+{
+ struct htb_class hc;
+ int error;
+
+ error = htb_parse_class_details__(netdev, details, &hc);
+ if (error) {
+ return error;
+ }
+
+ error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
+ tc_make_handle(1, 0xfffe), &hc);
+ if (error) {
+ return error;
+ }
+
+ htb_update_queue__(netdev, queue_id, &hc);
+ return 0;
+}
+
+static int
+htb_class_delete(struct netdev *netdev, unsigned int queue_id)
+{
+ struct htb *htb = htb_get__(netdev);
+ struct htb_class *hc;
+ int error;
+
+ hc = port_array_get(&htb->tc.queues, queue_id);
+ assert(hc != NULL);
+
+ error = tc_delete_class(netdev, tc_make_handle(1, queue_id + 1));
+ if (!error) {
+ free(hc);
+ port_array_delete(&htb->tc.queues, queue_id);
+ }
+ return error;
+}
+
+static int
+htb_class_get_stats(const struct netdev *netdev, unsigned int queue_id,
+ struct netdev_queue_stats *stats)
+{
+ return htb_query_class__(netdev, tc_make_handle(1, queue_id + 1),
+ tc_make_handle(1, 0xfffe), NULL, stats);
+}
+
+static int
+htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
+ const struct ofpbuf *nlmsg,
+ netdev_dump_queue_stats_cb *cb, void *aux)
+{
+ struct netdev_queue_stats stats;
+ unsigned int handle, major, minor;
+ int error;
+
+ error = tc_parse_class(nlmsg, &handle, NULL, &stats);
+ if (error) {
+ return error;
+ }
+
+ major = tc_get_major(handle);
+ minor = tc_get_minor(handle);
+ if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
+ (*cb)(tc_get_minor(handle), &stats, aux);
+ }
+ return 0;
+}
+
+static const struct tc_ops tc_ops_htb = {
+ "htb", /* linux_name */
+ "linux-htb", /* ovs_name */
+ HTB_N_QUEUES, /* n_queues */
+ htb_tc_install,
+ htb_tc_load,
+ htb_tc_destroy,
+ htb_qdisc_get,
+ htb_qdisc_set,
+ htb_class_get,
+ htb_class_set,
+ htb_class_delete,
+ htb_class_get_stats,
+ htb_class_dump_stats
+};
+\f
+/* "linux-default" traffic control class.
+ *
+ * This class represents the default, unnamed Linux qdisc. It corresponds to
+ * the "" (empty string) QoS type in the OVS database. */
+
+static void
+default_install__(struct netdev *netdev)
+{
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev));
+ static struct tc *tc;
+
+ if (!tc) {
+ tc = xmalloc(sizeof *tc);
+ tc_init(tc, &tc_ops_default);
+ }
+ netdev_dev->tc = tc;
+}
+
+static int
+default_tc_install(struct netdev *netdev,
+ const struct shash *details OVS_UNUSED)
+{
+ default_install__(netdev);
+ return 0;
+}
+
+static int
+default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
+{
+ default_install__(netdev);
+ return 0;
+}
+
+static const struct tc_ops tc_ops_default = {
+ NULL, /* linux_name */
+ "", /* ovs_name */
+ 0, /* n_queues */
+ default_tc_install,
+ default_tc_load,
+ NULL, /* tc_destroy */
+ NULL, /* qdisc_get */
+ NULL, /* qdisc_set */
+ NULL, /* class_get */
+ NULL, /* class_set */
+ NULL, /* class_delete */
+ NULL, /* class_get_stats */
+ NULL /* class_dump_stats */
+};
+\f
+/* "linux-other" traffic control class.
+ *
+ * */
+
+static int
+other_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
+{
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev));
+ static struct tc *tc;
+
+ if (!tc) {
+ tc = xmalloc(sizeof *tc);
+ tc_init(tc, &tc_ops_other);
+ }
+ netdev_dev->tc = tc;
+ return 0;
+}
+
+static const struct tc_ops tc_ops_other = {
+ NULL, /* linux_name */
+ "linux-other", /* ovs_name */
+ 0, /* n_queues */
+ NULL, /* tc_install */
+ other_tc_load,
+ NULL, /* tc_destroy */
+ NULL, /* qdisc_get */
+ NULL, /* qdisc_set */
+ NULL, /* class_get */
+ NULL, /* class_set */
+ NULL, /* class_delete */
+ NULL, /* class_get_stats */
+ NULL /* class_dump_stats */
+};
+\f
+/* Traffic control. */
+
+/* Number of kernel "tc" ticks per second. */
+static double ticks_per_s;
+
+/* Number of kernel "jiffies" per second. This is used for the purpose of
+ * computing buffer sizes. Generally kernel qdiscs need to be able to buffer
+ * one jiffy's worth of data.
+ *
+ * There are two possibilities here:
+ *
+ * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the
+ * approximate range of 100 to 1024. That means that we really need to
+ * make sure that the qdisc can buffer that much data.
+ *
+ * - 'buffer_hz' is an absurdly large number. That means that the kernel
+ * has finely granular timers and there's no need to fudge additional room
+ * for buffers. (There's no extra effort needed to implement that: the
+ * large 'buffer_hz' is used as a divisor, so practically any number will
+ * come out as 0 in the division. Small integer results in the case of
+ * really high dividends won't have any real effect anyhow.)
+ */
+static unsigned int buffer_hz;
+
+/* Returns tc handle 'major':'minor'. */
+static unsigned int
+tc_make_handle(unsigned int major, unsigned int minor)
+{
+ return TC_H_MAKE(major << 16, minor);
+}
+
+/* Returns the major number from 'handle'. */
+static unsigned int
+tc_get_major(unsigned int handle)
+{
+ return TC_H_MAJ(handle) >> 16;
+}
+
+/* Returns the minor number from 'handle'. */
+static unsigned int
+tc_get_minor(unsigned int handle)
+{
+ return TC_H_MIN(handle);
+}
+
+static struct tcmsg *
+tc_make_request(const struct netdev *netdev, int type, unsigned int flags,
+ struct ofpbuf *request)
+{
+ struct tcmsg *tcmsg;
+ int ifindex;
+ int error;
+
+ error = get_ifindex(netdev, &ifindex);
+ if (error) {
+ return NULL;
+ }
+
+ ofpbuf_init(request, 512);
+ nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags);
+ tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg);
+ tcmsg->tcm_family = AF_UNSPEC;
+ tcmsg->tcm_ifindex = ifindex;
+ /* Caller should fill in tcmsg->tcm_handle. */
+ /* Caller should fill in tcmsg->tcm_parent. */
+
+ return tcmsg;
+}
+
+static int
+tc_transact(struct ofpbuf *request, struct ofpbuf **replyp)
+{
+ int error = nl_sock_transact(rtnl_sock, request, replyp);
+ ofpbuf_uninit(request);
+ return error;
+}
+
+static void
+read_psched(void)
+{
+ /* The values in psched are not individually very meaningful, but they are
+ * important. The tables below show some values seen in the wild.
+ *
+ * Some notes:
+ *
+ * - "c" has always been a constant 1000000 since at least Linux 2.4.14.
+ * (Before that, there are hints that it was 1000000000.)
+ *
+ * - "d" can be unrealistically large, see the comment on 'buffer_hz'
+ * above.
+ *
+ * /proc/net/psched
+ * -----------------------------------
+ * [1] 000c8000 000f4240 000f4240 00000064
+ * [2] 000003e8 00000400 000f4240 3b9aca00
+ * [3] 000003e8 00000400 000f4240 3b9aca00
+ * [4] 000003e8 00000400 000f4240 00000064
+ * [5] 000003e8 00000040 000f4240 3b9aca00
+ * [6] 000003e8 00000040 000f4240 000000f9
+ *
+ * a b c d ticks_per_s buffer_hz
+ * ------- --------- ---------- ------------- ----------- -------------
+ * [1] 819,200 1,000,000 1,000,000 100 819,200 100
+ * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
+ * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000
+ * [4] 1,000 1,024 1,000,000 100 976,562 100
+ * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000
+ * [6] 1,000 64 1,000,000 249 15,625,000 249
+ *
+ * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p
+ * [2] 2.6.26-1-686-bigmem from Debian lenny
+ * [3] 2.6.26-2-sparc64 from Debian lenny
+ * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p
+ * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion
+ * [6] 2.6.34 from kernel.org on KVM
+ */
+ static const char fn[] = "/proc/net/psched";
+ unsigned int a, b, c, d;
+ FILE *stream;
+
+ ticks_per_s = 1.0;
+ buffer_hz = 100;
+
+ stream = fopen(fn, "r");
+ if (!stream) {
+ VLOG_WARN("%s: open failed: %s", fn, strerror(errno));
+ return;
+ }
+
+ if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) {
+ VLOG_WARN("%s: read failed", fn);
+ fclose(stream);
+ return;
+ }
+ VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d);
+ fclose(stream);
+
+ if (!a || !c) {
+ VLOG_WARN("%s: invalid scheduler parameters", fn);
+ return;
+ }
+
+ ticks_per_s = (double) a * c / b;
+ if (c == 1000000) {
+ buffer_hz = d;
+ } else {
+ VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u",
+ fn, a, b, c, d);
+ }
+ VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz);
+}
+
+/* Returns the number of bytes that can be transmitted in 'ticks' ticks at a
+ * rate of 'rate' bytes per second. */
+static unsigned int
+tc_ticks_to_bytes(unsigned int rate, unsigned int ticks)
+{
+ if (!buffer_hz) {
+ read_psched();
+ }
+ return (rate * ticks) / ticks_per_s;
+}
+
+/* Returns the number of ticks that it would take to transmit 'size' bytes at a
+ * rate of 'rate' bytes per second. */
+static unsigned int
+tc_bytes_to_ticks(unsigned int rate, unsigned int size)
+{
+ if (!buffer_hz) {
+ read_psched();
+ }
+ return ((unsigned long long int) ticks_per_s * size) / rate;
+}
+
+/* Returns the number of bytes that need to be reserved for qdisc buffering at
+ * a transmission rate of 'rate' bytes per second. */
+static unsigned int
+tc_buffer_per_jiffy(unsigned int rate)
+{
+ if (!buffer_hz) {
+ read_psched();
+ }
+ return rate / buffer_hz;
+}
+
+/* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc,
+ * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull,
+ * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or
+ * stores NULL into it if it is absent.
+ *
+ * '*kind' and '*options' point into 'msg', so they are owned by whoever owns
+ * 'msg'.
+ *
+ * Returns 0 if successful, otherwise a positive errno value. */
+static int
+tc_parse_qdisc(const struct ofpbuf *msg, const char **kind,
+ struct nlattr **options)