2 * net/sched/sch_tbf.c Token Bucket Filter queue.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 * Dmitry Torokhov <dtor@mail.ru> - allow attaching inner qdiscs -
11 * original idea by Martin Devera
15 #include <linux/module.h>
16 #include <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/string.h>
19 #include <linux/errno.h>
20 #include <linux/skbuff.h>
21 #include <net/netlink.h>
22 #include <net/sch_generic.h>
23 #include <net/pkt_sched.h>
27 /* Simple Token Bucket Filter.
28 =======================================
38 A data flow obeys TBF with rate R and depth B, if for any
39 time interval t_i...t_f the number of transmitted bits
40 does not exceed B + R*(t_f-t_i).
42 Packetized version of this definition:
43 The sequence of packets of sizes s_i served at moments t_i
44 obeys TBF, if for any i<=k:
46 s_i+....+s_k <= B + R*(t_k - t_i)
51 Let N(t_i) be B/R initially and N(t) grow continuously with time as:
53 N(t+delta) = min{B/R, N(t) + delta}
55 If the first packet in queue has length S, it may be
56 transmitted only at the time t_* when S/R <= N(t_*),
57 and in this case N(t) jumps:
59 N(t_* + 0) = N(t_* - 0) - S/R.
63 Actually, QoS requires two TBF to be applied to a data stream.
64 One of them controls steady state burst size, another
65 one with rate P (peak rate) and depth M (equal to link MTU)
66 limits bursts at a smaller time scale.
68 It is easy to see that P>R, and B>M. If P is infinity, this double
69 TBF is equivalent to a single one.
71 When TBF works in reshaping mode, latency is estimated as:
73 lat = max ((L-B)/R, (L-M)/P)
79 If TBF throttles, it starts a watchdog timer, which will wake it up
80 when it is ready to transmit.
81 Note that the minimal timer resolution is 1/HZ.
82 If no new packets arrive during this period,
83 or if the device is not awaken by EOI for some previous packet,
84 TBF can stop its activity for 1/HZ.
87 This means, that with depth B, the maximal rate is
91 F.e. for 10Mbit ethernet and HZ=100 the minimal allowed B is ~10Kbytes.
93 Note that the peak rate TBF is much more tough: with MTU 1500
94 P_crit = 150Kbytes/sec. So, if you need greater peak
95 rates, use alpha with HZ=1000 :-)
97 With classful TBF, limit is just kept for backwards compatibility.
98 It is passed to the default bfifo qdisc - if the inner qdisc is
99 changed the limit is not effective anymore.
102 struct tbf_sched_data {
104 u32 limit; /* Maximal length of backlog: bytes */
105 s64 buffer; /* Token bucket depth/rate: MUST BE >= MTU/B */
108 struct psched_ratecfg rate;
109 struct psched_ratecfg peak;
113 s64 tokens; /* Current number of B tokens */
114 s64 ptokens; /* Current number of P tokens */
115 s64 t_c; /* Time check-point */
116 struct Qdisc *qdisc; /* Inner qdisc, default - bfifo queue */
117 struct qdisc_watchdog watchdog; /* Watchdog timer */
121 /* Time to Length, convert time in ns to length in bytes
122 * to determinate how many bytes can be sent in given time.
124 static u64 psched_ns_t2l(const struct psched_ratecfg *r,
128 * len = (time_in_ns * r->rate_bytes_ps) / NSEC_PER_SEC
130 u64 len = time_in_ns * r->rate_bytes_ps;
132 do_div(len, NSEC_PER_SEC);
134 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) {
139 if (len > r->overhead)
148 * Return length of individual segments of a gso packet,
149 * including all headers (MAC, IP, TCP/UDP)
151 static unsigned int skb_gso_seglen(const struct sk_buff *skb)
153 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
154 const struct skb_shared_info *shinfo = skb_shinfo(skb);
156 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
157 hdr_len += tcp_hdrlen(skb);
159 hdr_len += sizeof(struct udphdr);
160 return hdr_len + shinfo->gso_size;
163 /* GSO packet is too big, segment it so that tbf can transmit
164 * each segment in time
166 static int tbf_segment(struct sk_buff *skb, struct Qdisc *sch)
168 struct tbf_sched_data *q = qdisc_priv(sch);
169 struct sk_buff *segs, *nskb;
170 netdev_features_t features = netif_skb_features(skb);
173 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
175 if (IS_ERR_OR_NULL(segs))
176 return qdisc_reshape_fail(skb, sch);
182 qdisc_skb_cb(segs)->pkt_len = segs->len;
183 ret = qdisc_enqueue(segs, q->qdisc);
184 if (ret != NET_XMIT_SUCCESS) {
185 if (net_xmit_drop_count(ret))
194 qdisc_tree_decrease_qlen(sch, 1 - nb);
196 return nb > 0 ? NET_XMIT_SUCCESS : NET_XMIT_DROP;
199 static int tbf_enqueue(struct sk_buff *skb, struct Qdisc *sch)
201 struct tbf_sched_data *q = qdisc_priv(sch);
204 if (qdisc_pkt_len(skb) > q->max_size) {
205 if (skb_is_gso(skb) && skb_gso_seglen(skb) <= q->max_size)
206 return tbf_segment(skb, sch);
207 return qdisc_reshape_fail(skb, sch);
209 ret = qdisc_enqueue(skb, q->qdisc);
210 if (ret != NET_XMIT_SUCCESS) {
211 if (net_xmit_drop_count(ret))
217 return NET_XMIT_SUCCESS;
220 static unsigned int tbf_drop(struct Qdisc *sch)
222 struct tbf_sched_data *q = qdisc_priv(sch);
223 unsigned int len = 0;
225 if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) {
232 static struct sk_buff *tbf_dequeue(struct Qdisc *sch)
234 struct tbf_sched_data *q = qdisc_priv(sch);
237 skb = q->qdisc->ops->peek(q->qdisc);
243 unsigned int len = qdisc_pkt_len(skb);
245 now = ktime_to_ns(ktime_get());
246 toks = min_t(s64, now - q->t_c, q->buffer);
248 if (q->peak_present) {
249 ptoks = toks + q->ptokens;
252 ptoks -= (s64) psched_l2t_ns(&q->peak, len);
255 if (toks > q->buffer)
257 toks -= (s64) psched_l2t_ns(&q->rate, len);
259 if ((toks|ptoks) >= 0) {
260 skb = qdisc_dequeue_peeked(q->qdisc);
268 qdisc_unthrottled(sch);
269 qdisc_bstats_update(sch, skb);
273 qdisc_watchdog_schedule_ns(&q->watchdog,
274 now + max_t(long, -toks, -ptoks));
276 /* Maybe we have a shorter packet in the queue,
277 which can be sent now. It sounds cool,
278 but, however, this is wrong in principle.
279 We MUST NOT reorder packets under these circumstances.
281 Really, if we split the flow into independent
282 subflows, it would be a very good solution.
283 This is the main idea of all FQ algorithms
284 (cf. CSZ, HPFQ, HFSC)
287 sch->qstats.overlimits++;
292 static void tbf_reset(struct Qdisc *sch)
294 struct tbf_sched_data *q = qdisc_priv(sch);
296 qdisc_reset(q->qdisc);
298 q->t_c = ktime_to_ns(ktime_get());
299 q->tokens = q->buffer;
301 qdisc_watchdog_cancel(&q->watchdog);
304 static const struct nla_policy tbf_policy[TCA_TBF_MAX + 1] = {
305 [TCA_TBF_PARMS] = { .len = sizeof(struct tc_tbf_qopt) },
306 [TCA_TBF_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
307 [TCA_TBF_PTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
308 [TCA_TBF_RATE64] = { .type = NLA_U64 },
309 [TCA_TBF_PRATE64] = { .type = NLA_U64 },
312 static int tbf_change(struct Qdisc *sch, struct nlattr *opt)
315 struct tbf_sched_data *q = qdisc_priv(sch);
316 struct nlattr *tb[TCA_TBF_MAX + 1];
317 struct tc_tbf_qopt *qopt;
318 struct Qdisc *child = NULL;
319 struct psched_ratecfg rate;
320 struct psched_ratecfg peak;
323 u64 rate64 = 0, prate64 = 0;
325 err = nla_parse_nested(tb, TCA_TBF_MAX, opt, tbf_policy);
330 if (tb[TCA_TBF_PARMS] == NULL)
333 qopt = nla_data(tb[TCA_TBF_PARMS]);
334 if (qopt->rate.linklayer == TC_LINKLAYER_UNAWARE)
335 qdisc_put_rtab(qdisc_get_rtab(&qopt->rate,
338 if (qopt->peakrate.linklayer == TC_LINKLAYER_UNAWARE)
339 qdisc_put_rtab(qdisc_get_rtab(&qopt->peakrate,
342 if (q->qdisc != &noop_qdisc) {
343 err = fifo_set_limit(q->qdisc, qopt->limit);
346 } else if (qopt->limit > 0) {
347 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, qopt->limit);
349 err = PTR_ERR(child);
354 buffer = min_t(u64, PSCHED_TICKS2NS(qopt->buffer), ~0U);
355 mtu = min_t(u64, PSCHED_TICKS2NS(qopt->mtu), ~0U);
357 if (tb[TCA_TBF_RATE64])
358 rate64 = nla_get_u64(tb[TCA_TBF_RATE64]);
359 psched_ratecfg_precompute(&rate, &qopt->rate, rate64);
361 max_size = min_t(u64, psched_ns_t2l(&rate, buffer), ~0U);
363 if (qopt->peakrate.rate) {
364 if (tb[TCA_TBF_PRATE64])
365 prate64 = nla_get_u64(tb[TCA_TBF_PRATE64]);
366 psched_ratecfg_precompute(&peak, &qopt->peakrate, prate64);
367 if (peak.rate_bytes_ps <= rate.rate_bytes_ps) {
368 pr_warn_ratelimited("sch_tbf: peakrate %llu is lower than or equals to rate %llu !\n",
369 peak.rate_bytes_ps, rate.rate_bytes_ps);
374 max_size = min_t(u64, max_size, psched_ns_t2l(&peak, mtu));
377 if (max_size < psched_mtu(qdisc_dev(sch)))
378 pr_warn_ratelimited("sch_tbf: burst %llu is lower than device %s mtu (%u) !\n",
379 max_size, qdisc_dev(sch)->name,
380 psched_mtu(qdisc_dev(sch)));
389 qdisc_tree_decrease_qlen(q->qdisc, q->qdisc->q.qlen);
390 qdisc_destroy(q->qdisc);
393 q->limit = qopt->limit;
394 q->mtu = PSCHED_TICKS2NS(qopt->mtu);
395 q->max_size = max_size;
396 q->buffer = PSCHED_TICKS2NS(qopt->buffer);
397 q->tokens = q->buffer;
400 memcpy(&q->rate, &rate, sizeof(struct psched_ratecfg));
401 if (qopt->peakrate.rate) {
402 memcpy(&q->peak, &peak, sizeof(struct psched_ratecfg));
403 q->peak_present = true;
405 q->peak_present = false;
408 sch_tree_unlock(sch);
414 static int tbf_init(struct Qdisc *sch, struct nlattr *opt)
416 struct tbf_sched_data *q = qdisc_priv(sch);
421 q->t_c = ktime_to_ns(ktime_get());
422 qdisc_watchdog_init(&q->watchdog, sch);
423 q->qdisc = &noop_qdisc;
425 return tbf_change(sch, opt);
428 static void tbf_destroy(struct Qdisc *sch)
430 struct tbf_sched_data *q = qdisc_priv(sch);
432 qdisc_watchdog_cancel(&q->watchdog);
433 qdisc_destroy(q->qdisc);
436 static int tbf_dump(struct Qdisc *sch, struct sk_buff *skb)
438 struct tbf_sched_data *q = qdisc_priv(sch);
440 struct tc_tbf_qopt opt;
442 sch->qstats.backlog = q->qdisc->qstats.backlog;
443 nest = nla_nest_start(skb, TCA_OPTIONS);
445 goto nla_put_failure;
447 opt.limit = q->limit;
448 psched_ratecfg_getrate(&opt.rate, &q->rate);
450 psched_ratecfg_getrate(&opt.peakrate, &q->peak);
452 memset(&opt.peakrate, 0, sizeof(opt.peakrate));
453 opt.mtu = PSCHED_NS2TICKS(q->mtu);
454 opt.buffer = PSCHED_NS2TICKS(q->buffer);
455 if (nla_put(skb, TCA_TBF_PARMS, sizeof(opt), &opt))
456 goto nla_put_failure;
457 if (q->rate.rate_bytes_ps >= (1ULL << 32) &&
458 nla_put_u64(skb, TCA_TBF_RATE64, q->rate.rate_bytes_ps))
459 goto nla_put_failure;
460 if (q->peak_present &&
461 q->peak.rate_bytes_ps >= (1ULL << 32) &&
462 nla_put_u64(skb, TCA_TBF_PRATE64, q->peak.rate_bytes_ps))
463 goto nla_put_failure;
465 nla_nest_end(skb, nest);
469 nla_nest_cancel(skb, nest);
473 static int tbf_dump_class(struct Qdisc *sch, unsigned long cl,
474 struct sk_buff *skb, struct tcmsg *tcm)
476 struct tbf_sched_data *q = qdisc_priv(sch);
478 tcm->tcm_handle |= TC_H_MIN(1);
479 tcm->tcm_info = q->qdisc->handle;
484 static int tbf_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
487 struct tbf_sched_data *q = qdisc_priv(sch);
495 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
497 sch_tree_unlock(sch);
502 static struct Qdisc *tbf_leaf(struct Qdisc *sch, unsigned long arg)
504 struct tbf_sched_data *q = qdisc_priv(sch);
508 static unsigned long tbf_get(struct Qdisc *sch, u32 classid)
513 static void tbf_put(struct Qdisc *sch, unsigned long arg)
517 static void tbf_walk(struct Qdisc *sch, struct qdisc_walker *walker)
520 if (walker->count >= walker->skip)
521 if (walker->fn(sch, 1, walker) < 0) {
529 static const struct Qdisc_class_ops tbf_class_ops = {
535 .dump = tbf_dump_class,
538 static struct Qdisc_ops tbf_qdisc_ops __read_mostly = {
540 .cl_ops = &tbf_class_ops,
542 .priv_size = sizeof(struct tbf_sched_data),
543 .enqueue = tbf_enqueue,
544 .dequeue = tbf_dequeue,
545 .peek = qdisc_peek_dequeued,
549 .destroy = tbf_destroy,
550 .change = tbf_change,
552 .owner = THIS_MODULE,
555 static int __init tbf_module_init(void)
557 return register_qdisc(&tbf_qdisc_ops);
560 static void __exit tbf_module_exit(void)
562 unregister_qdisc(&tbf_qdisc_ops);
564 module_init(tbf_module_init)
565 module_exit(tbf_module_exit)
566 MODULE_LICENSE("GPL");