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/pkt_sched.h>
25 /* Simple Token Bucket Filter.
26 =======================================
36 A data flow obeys TBF with rate R and depth B, if for any
37 time interval t_i...t_f the number of transmitted bits
38 does not exceed B + R*(t_f-t_i).
40 Packetized version of this definition:
41 The sequence of packets of sizes s_i served at moments t_i
42 obeys TBF, if for any i<=k:
44 s_i+....+s_k <= B + R*(t_k - t_i)
49 Let N(t_i) be B/R initially and N(t) grow continuously with time as:
51 N(t+delta) = min{B/R, N(t) + delta}
53 If the first packet in queue has length S, it may be
54 transmitted only at the time t_* when S/R <= N(t_*),
55 and in this case N(t) jumps:
57 N(t_* + 0) = N(t_* - 0) - S/R.
61 Actually, QoS requires two TBF to be applied to a data stream.
62 One of them controls steady state burst size, another
63 one with rate P (peak rate) and depth M (equal to link MTU)
64 limits bursts at a smaller time scale.
66 It is easy to see that P>R, and B>M. If P is infinity, this double
67 TBF is equivalent to a single one.
69 When TBF works in reshaping mode, latency is estimated as:
71 lat = max ((L-B)/R, (L-M)/P)
77 If TBF throttles, it starts a watchdog timer, which will wake it up
78 when it is ready to transmit.
79 Note that the minimal timer resolution is 1/HZ.
80 If no new packets arrive during this period,
81 or if the device is not awaken by EOI for some previous packet,
82 TBF can stop its activity for 1/HZ.
85 This means, that with depth B, the maximal rate is
89 F.e. for 10Mbit ethernet and HZ=100 the minimal allowed B is ~10Kbytes.
91 Note that the peak rate TBF is much more tough: with MTU 1500
92 P_crit = 150Kbytes/sec. So, if you need greater peak
93 rates, use alpha with HZ=1000 :-)
95 With classful TBF, limit is just kept for backwards compatibility.
96 It is passed to the default bfifo qdisc - if the inner qdisc is
97 changed the limit is not effective anymore.
100 struct tbf_sched_data
103 u32 limit; /* Maximal length of backlog: bytes */
104 u32 buffer; /* Token bucket depth/rate: MUST BE >= MTU/B */
107 struct qdisc_rate_table *R_tab;
108 struct qdisc_rate_table *P_tab;
111 long tokens; /* Current number of B tokens */
112 long ptokens; /* Current number of P tokens */
113 psched_time_t t_c; /* Time check-point */
114 struct Qdisc *qdisc; /* Inner qdisc, default - bfifo queue */
115 struct qdisc_watchdog watchdog; /* Watchdog timer */
118 #define L2T(q,L) qdisc_l2t((q)->R_tab,L)
119 #define L2T_P(q,L) qdisc_l2t((q)->P_tab,L)
121 static int tbf_enqueue(struct sk_buff *skb, struct Qdisc* sch)
123 struct tbf_sched_data *q = qdisc_priv(sch);
126 if (skb->len > q->max_size)
127 return qdisc_reshape_fail(skb, sch);
129 if ((ret = q->qdisc->enqueue(skb, q->qdisc)) != 0) {
135 sch->bstats.bytes += skb->len;
136 sch->bstats.packets++;
140 static int tbf_requeue(struct sk_buff *skb, struct Qdisc* sch)
142 struct tbf_sched_data *q = qdisc_priv(sch);
145 if ((ret = q->qdisc->ops->requeue(skb, q->qdisc)) == 0) {
147 sch->qstats.requeues++;
153 static unsigned int tbf_drop(struct Qdisc* sch)
155 struct tbf_sched_data *q = qdisc_priv(sch);
156 unsigned int len = 0;
158 if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) {
165 static struct sk_buff *tbf_dequeue(struct Qdisc* sch)
167 struct tbf_sched_data *q = qdisc_priv(sch);
170 skb = q->qdisc->dequeue(q->qdisc);
176 unsigned int len = skb->len;
178 now = psched_get_time();
179 toks = psched_tdiff_bounded(now, q->t_c, q->buffer);
182 ptoks = toks + q->ptokens;
183 if (ptoks > (long)q->mtu)
185 ptoks -= L2T_P(q, len);
188 if (toks > (long)q->buffer)
192 if ((toks|ptoks) >= 0) {
197 sch->flags &= ~TCQ_F_THROTTLED;
201 qdisc_watchdog_schedule(&q->watchdog,
202 now + max_t(long, -toks, -ptoks));
204 /* Maybe we have a shorter packet in the queue,
205 which can be sent now. It sounds cool,
206 but, however, this is wrong in principle.
207 We MUST NOT reorder packets under these circumstances.
209 Really, if we split the flow into independent
210 subflows, it would be a very good solution.
211 This is the main idea of all FQ algorithms
212 (cf. CSZ, HPFQ, HFSC)
215 if (q->qdisc->ops->requeue(skb, q->qdisc) != NET_XMIT_SUCCESS) {
216 /* When requeue fails skb is dropped */
217 qdisc_tree_decrease_qlen(q->qdisc, 1);
221 sch->qstats.overlimits++;
226 static void tbf_reset(struct Qdisc* sch)
228 struct tbf_sched_data *q = qdisc_priv(sch);
230 qdisc_reset(q->qdisc);
232 q->t_c = psched_get_time();
233 q->tokens = q->buffer;
235 qdisc_watchdog_cancel(&q->watchdog);
238 static struct Qdisc *tbf_create_dflt_qdisc(struct Qdisc *sch, u32 limit)
244 q = qdisc_create_dflt(sch->dev, &bfifo_qdisc_ops,
245 TC_H_MAKE(sch->handle, 1));
247 nla = kmalloc(nla_attr_size(sizeof(struct tc_fifo_qopt)),
250 nla->nla_type = RTM_NEWQDISC;
251 nla->nla_len = nla_attr_size(sizeof(struct tc_fifo_qopt));
252 ((struct tc_fifo_qopt *)nla_data(nla))->limit = limit;
254 ret = q->ops->change(q, nla);
266 static const struct nla_policy tbf_policy[TCA_TBF_MAX + 1] = {
267 [TCA_TBF_PARMS] = { .len = sizeof(struct tc_tbf_qopt) },
268 [TCA_TBF_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
269 [TCA_TBF_PTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
272 static int tbf_change(struct Qdisc* sch, struct nlattr *opt)
275 struct tbf_sched_data *q = qdisc_priv(sch);
276 struct nlattr *tb[TCA_TBF_PTAB + 1];
277 struct tc_tbf_qopt *qopt;
278 struct qdisc_rate_table *rtab = NULL;
279 struct qdisc_rate_table *ptab = NULL;
280 struct Qdisc *child = NULL;
283 err = nla_parse_nested(tb, TCA_TBF_PTAB, opt, tbf_policy);
288 if (tb[TCA_TBF_PARMS] == NULL)
291 qopt = nla_data(tb[TCA_TBF_PARMS]);
292 rtab = qdisc_get_rtab(&qopt->rate, tb[TCA_TBF_RTAB]);
296 if (qopt->peakrate.rate) {
297 if (qopt->peakrate.rate > qopt->rate.rate)
298 ptab = qdisc_get_rtab(&qopt->peakrate, tb[TCA_TBF_PTAB]);
303 for (n = 0; n < 256; n++)
304 if (rtab->data[n] > qopt->buffer) break;
305 max_size = (n << qopt->rate.cell_log)-1;
309 for (n = 0; n < 256; n++)
310 if (ptab->data[n] > qopt->mtu) break;
311 size = (n << qopt->peakrate.cell_log)-1;
312 if (size < max_size) max_size = size;
317 if (qopt->limit > 0) {
318 if ((child = tbf_create_dflt_qdisc(sch, qopt->limit)) == NULL)
324 qdisc_tree_decrease_qlen(q->qdisc, q->qdisc->q.qlen);
325 qdisc_destroy(xchg(&q->qdisc, child));
327 q->limit = qopt->limit;
329 q->max_size = max_size;
330 q->buffer = qopt->buffer;
331 q->tokens = q->buffer;
333 rtab = xchg(&q->R_tab, rtab);
334 ptab = xchg(&q->P_tab, ptab);
335 sch_tree_unlock(sch);
339 qdisc_put_rtab(rtab);
341 qdisc_put_rtab(ptab);
345 static int tbf_init(struct Qdisc* sch, struct nlattr *opt)
347 struct tbf_sched_data *q = qdisc_priv(sch);
352 q->t_c = psched_get_time();
353 qdisc_watchdog_init(&q->watchdog, sch);
354 q->qdisc = &noop_qdisc;
356 return tbf_change(sch, opt);
359 static void tbf_destroy(struct Qdisc *sch)
361 struct tbf_sched_data *q = qdisc_priv(sch);
363 qdisc_watchdog_cancel(&q->watchdog);
366 qdisc_put_rtab(q->P_tab);
368 qdisc_put_rtab(q->R_tab);
370 qdisc_destroy(q->qdisc);
373 static int tbf_dump(struct Qdisc *sch, struct sk_buff *skb)
375 struct tbf_sched_data *q = qdisc_priv(sch);
377 struct tc_tbf_qopt opt;
379 nest = nla_nest_start(skb, TCA_OPTIONS);
381 goto nla_put_failure;
383 opt.limit = q->limit;
384 opt.rate = q->R_tab->rate;
386 opt.peakrate = q->P_tab->rate;
388 memset(&opt.peakrate, 0, sizeof(opt.peakrate));
390 opt.buffer = q->buffer;
391 NLA_PUT(skb, TCA_TBF_PARMS, sizeof(opt), &opt);
393 nla_nest_end(skb, nest);
397 nla_nest_cancel(skb, nest);
401 static int tbf_dump_class(struct Qdisc *sch, unsigned long cl,
402 struct sk_buff *skb, struct tcmsg *tcm)
404 struct tbf_sched_data *q = qdisc_priv(sch);
406 if (cl != 1) /* only one class */
409 tcm->tcm_handle |= TC_H_MIN(1);
410 tcm->tcm_info = q->qdisc->handle;
415 static int tbf_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
418 struct tbf_sched_data *q = qdisc_priv(sch);
424 *old = xchg(&q->qdisc, new);
425 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
427 sch_tree_unlock(sch);
432 static struct Qdisc *tbf_leaf(struct Qdisc *sch, unsigned long arg)
434 struct tbf_sched_data *q = qdisc_priv(sch);
438 static unsigned long tbf_get(struct Qdisc *sch, u32 classid)
443 static void tbf_put(struct Qdisc *sch, unsigned long arg)
447 static int tbf_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
448 struct nlattr **tca, unsigned long *arg)
453 static int tbf_delete(struct Qdisc *sch, unsigned long arg)
458 static void tbf_walk(struct Qdisc *sch, struct qdisc_walker *walker)
461 if (walker->count >= walker->skip)
462 if (walker->fn(sch, 1, walker) < 0) {
470 static struct tcf_proto **tbf_find_tcf(struct Qdisc *sch, unsigned long cl)
475 static const struct Qdisc_class_ops tbf_class_ops =
481 .change = tbf_change_class,
482 .delete = tbf_delete,
484 .tcf_chain = tbf_find_tcf,
485 .dump = tbf_dump_class,
488 static struct Qdisc_ops tbf_qdisc_ops __read_mostly = {
490 .cl_ops = &tbf_class_ops,
492 .priv_size = sizeof(struct tbf_sched_data),
493 .enqueue = tbf_enqueue,
494 .dequeue = tbf_dequeue,
495 .requeue = tbf_requeue,
499 .destroy = tbf_destroy,
500 .change = tbf_change,
502 .owner = THIS_MODULE,
505 static int __init tbf_module_init(void)
507 return register_qdisc(&tbf_qdisc_ops);
510 static void __exit tbf_module_exit(void)
512 unregister_qdisc(&tbf_qdisc_ops);
514 module_init(tbf_module_init)
515 module_exit(tbf_module_exit)
516 MODULE_LICENSE("GPL");