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 <asm/uaccess.h>
17 #include <asm/system.h>
18 #include <linux/bitops.h>
19 #include <linux/types.h>
20 #include <linux/kernel.h>
21 #include <linux/jiffies.h>
22 #include <linux/string.h>
24 #include <linux/socket.h>
25 #include <linux/sockios.h>
27 #include <linux/errno.h>
28 #include <linux/interrupt.h>
29 #include <linux/if_ether.h>
30 #include <linux/inet.h>
31 #include <linux/netdevice.h>
32 #include <linux/etherdevice.h>
33 #include <linux/notifier.h>
35 #include <net/route.h>
36 #include <linux/skbuff.h>
38 #include <net/pkt_sched.h>
41 /* Simple Token Bucket Filter.
42 =======================================
52 A data flow obeys TBF with rate R and depth B, if for any
53 time interval t_i...t_f the number of transmitted bits
54 does not exceed B + R*(t_f-t_i).
56 Packetized version of this definition:
57 The sequence of packets of sizes s_i served at moments t_i
58 obeys TBF, if for any i<=k:
60 s_i+....+s_k <= B + R*(t_k - t_i)
65 Let N(t_i) be B/R initially and N(t) grow continuously with time as:
67 N(t+delta) = min{B/R, N(t) + delta}
69 If the first packet in queue has length S, it may be
70 transmitted only at the time t_* when S/R <= N(t_*),
71 and in this case N(t) jumps:
73 N(t_* + 0) = N(t_* - 0) - S/R.
77 Actually, QoS requires two TBF to be applied to a data stream.
78 One of them controls steady state burst size, another
79 one with rate P (peak rate) and depth M (equal to link MTU)
80 limits bursts at a smaller time scale.
82 It is easy to see that P>R, and B>M. If P is infinity, this double
83 TBF is equivalent to a single one.
85 When TBF works in reshaping mode, latency is estimated as:
87 lat = max ((L-B)/R, (L-M)/P)
93 If TBF throttles, it starts a watchdog timer, which will wake it up
94 when it is ready to transmit.
95 Note that the minimal timer resolution is 1/HZ.
96 If no new packets arrive during this period,
97 or if the device is not awaken by EOI for some previous packet,
98 TBF can stop its activity for 1/HZ.
101 This means, that with depth B, the maximal rate is
105 F.e. for 10Mbit ethernet and HZ=100 the minimal allowed B is ~10Kbytes.
107 Note that the peak rate TBF is much more tough: with MTU 1500
108 P_crit = 150Kbytes/sec. So, if you need greater peak
109 rates, use alpha with HZ=1000 :-)
111 With classful TBF, limit is just kept for backwards compatibility.
112 It is passed to the default bfifo qdisc - if the inner qdisc is
113 changed the limit is not effective anymore.
116 struct tbf_sched_data
119 u32 limit; /* Maximal length of backlog: bytes */
120 u32 buffer; /* Token bucket depth/rate: MUST BE >= MTU/B */
123 struct qdisc_rate_table *R_tab;
124 struct qdisc_rate_table *P_tab;
127 long tokens; /* Current number of B tokens */
128 long ptokens; /* Current number of P tokens */
129 psched_time_t t_c; /* Time check-point */
130 struct timer_list wd_timer; /* Watchdog timer */
131 struct Qdisc *qdisc; /* Inner qdisc, default - bfifo queue */
134 #define L2T(q,L) ((q)->R_tab->data[(L)>>(q)->R_tab->rate.cell_log])
135 #define L2T_P(q,L) ((q)->P_tab->data[(L)>>(q)->P_tab->rate.cell_log])
137 static int tbf_enqueue(struct sk_buff *skb, struct Qdisc* sch)
139 struct tbf_sched_data *q = qdisc_priv(sch);
142 if (skb->len > q->max_size) {
144 #ifdef CONFIG_NET_CLS_POLICE
145 if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
149 return NET_XMIT_DROP;
152 if ((ret = q->qdisc->enqueue(skb, q->qdisc)) != 0) {
158 sch->bstats.bytes += skb->len;
159 sch->bstats.packets++;
163 static int tbf_requeue(struct sk_buff *skb, struct Qdisc* sch)
165 struct tbf_sched_data *q = qdisc_priv(sch);
168 if ((ret = q->qdisc->ops->requeue(skb, q->qdisc)) == 0) {
170 sch->qstats.requeues++;
176 static unsigned int tbf_drop(struct Qdisc* sch)
178 struct tbf_sched_data *q = qdisc_priv(sch);
179 unsigned int len = 0;
181 if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) {
188 static void tbf_watchdog(unsigned long arg)
190 struct Qdisc *sch = (struct Qdisc*)arg;
192 sch->flags &= ~TCQ_F_THROTTLED;
193 netif_schedule(sch->dev);
196 static struct sk_buff *tbf_dequeue(struct Qdisc* sch)
198 struct tbf_sched_data *q = qdisc_priv(sch);
201 skb = q->qdisc->dequeue(q->qdisc);
207 unsigned int len = skb->len;
209 PSCHED_GET_TIME(now);
211 toks = PSCHED_TDIFF_SAFE(now, q->t_c, q->buffer);
214 ptoks = toks + q->ptokens;
215 if (ptoks > (long)q->mtu)
217 ptoks -= L2T_P(q, len);
220 if (toks > (long)q->buffer)
224 if ((toks|ptoks) >= 0) {
229 sch->flags &= ~TCQ_F_THROTTLED;
233 delay = PSCHED_US2JIFFIE(max_t(long, -toks, -ptoks));
238 mod_timer(&q->wd_timer, jiffies+delay);
240 /* Maybe we have a shorter packet in the queue,
241 which can be sent now. It sounds cool,
242 but, however, this is wrong in principle.
243 We MUST NOT reorder packets under these circumstances.
245 Really, if we split the flow into independent
246 subflows, it would be a very good solution.
247 This is the main idea of all FQ algorithms
248 (cf. CSZ, HPFQ, HFSC)
251 if (q->qdisc->ops->requeue(skb, q->qdisc) != NET_XMIT_SUCCESS) {
252 /* When requeue fails skb is dropped */
253 qdisc_tree_decrease_qlen(q->qdisc, 1);
257 sch->flags |= TCQ_F_THROTTLED;
258 sch->qstats.overlimits++;
263 static void tbf_reset(struct Qdisc* sch)
265 struct tbf_sched_data *q = qdisc_priv(sch);
267 qdisc_reset(q->qdisc);
269 PSCHED_GET_TIME(q->t_c);
270 q->tokens = q->buffer;
272 sch->flags &= ~TCQ_F_THROTTLED;
273 del_timer(&q->wd_timer);
276 static struct Qdisc *tbf_create_dflt_qdisc(struct Qdisc *sch, u32 limit)
282 q = qdisc_create_dflt(sch->dev, &bfifo_qdisc_ops,
283 TC_H_MAKE(sch->handle, 1));
285 rta = kmalloc(RTA_LENGTH(sizeof(struct tc_fifo_qopt)), GFP_KERNEL);
287 rta->rta_type = RTM_NEWQDISC;
288 rta->rta_len = RTA_LENGTH(sizeof(struct tc_fifo_qopt));
289 ((struct tc_fifo_qopt *)RTA_DATA(rta))->limit = limit;
291 ret = q->ops->change(q, rta);
303 static int tbf_change(struct Qdisc* sch, struct rtattr *opt)
306 struct tbf_sched_data *q = qdisc_priv(sch);
307 struct rtattr *tb[TCA_TBF_PTAB];
308 struct tc_tbf_qopt *qopt;
309 struct qdisc_rate_table *rtab = NULL;
310 struct qdisc_rate_table *ptab = NULL;
311 struct Qdisc *child = NULL;
314 if (rtattr_parse_nested(tb, TCA_TBF_PTAB, opt) ||
315 tb[TCA_TBF_PARMS-1] == NULL ||
316 RTA_PAYLOAD(tb[TCA_TBF_PARMS-1]) < sizeof(*qopt))
319 qopt = RTA_DATA(tb[TCA_TBF_PARMS-1]);
320 rtab = qdisc_get_rtab(&qopt->rate, tb[TCA_TBF_RTAB-1]);
324 if (qopt->peakrate.rate) {
325 if (qopt->peakrate.rate > qopt->rate.rate)
326 ptab = qdisc_get_rtab(&qopt->peakrate, tb[TCA_TBF_PTAB-1]);
331 for (n = 0; n < 256; n++)
332 if (rtab->data[n] > qopt->buffer) break;
333 max_size = (n << qopt->rate.cell_log)-1;
337 for (n = 0; n < 256; n++)
338 if (ptab->data[n] > qopt->mtu) break;
339 size = (n << qopt->peakrate.cell_log)-1;
340 if (size < max_size) max_size = size;
345 if (qopt->limit > 0) {
346 if ((child = tbf_create_dflt_qdisc(sch, qopt->limit)) == NULL)
352 qdisc_tree_decrease_qlen(q->qdisc, q->qdisc->q.qlen);
353 qdisc_destroy(xchg(&q->qdisc, child));
355 q->limit = qopt->limit;
357 q->max_size = max_size;
358 q->buffer = qopt->buffer;
359 q->tokens = q->buffer;
361 rtab = xchg(&q->R_tab, rtab);
362 ptab = xchg(&q->P_tab, ptab);
363 sch_tree_unlock(sch);
367 qdisc_put_rtab(rtab);
369 qdisc_put_rtab(ptab);
373 static int tbf_init(struct Qdisc* sch, struct rtattr *opt)
375 struct tbf_sched_data *q = qdisc_priv(sch);
380 PSCHED_GET_TIME(q->t_c);
381 init_timer(&q->wd_timer);
382 q->wd_timer.function = tbf_watchdog;
383 q->wd_timer.data = (unsigned long)sch;
385 q->qdisc = &noop_qdisc;
387 return tbf_change(sch, opt);
390 static void tbf_destroy(struct Qdisc *sch)
392 struct tbf_sched_data *q = qdisc_priv(sch);
394 del_timer(&q->wd_timer);
397 qdisc_put_rtab(q->P_tab);
399 qdisc_put_rtab(q->R_tab);
401 qdisc_destroy(q->qdisc);
404 static int tbf_dump(struct Qdisc *sch, struct sk_buff *skb)
406 struct tbf_sched_data *q = qdisc_priv(sch);
407 unsigned char *b = skb->tail;
409 struct tc_tbf_qopt opt;
411 rta = (struct rtattr*)b;
412 RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
414 opt.limit = q->limit;
415 opt.rate = q->R_tab->rate;
417 opt.peakrate = q->P_tab->rate;
419 memset(&opt.peakrate, 0, sizeof(opt.peakrate));
421 opt.buffer = q->buffer;
422 RTA_PUT(skb, TCA_TBF_PARMS, sizeof(opt), &opt);
423 rta->rta_len = skb->tail - b;
428 skb_trim(skb, b - skb->data);
432 static int tbf_dump_class(struct Qdisc *sch, unsigned long cl,
433 struct sk_buff *skb, struct tcmsg *tcm)
435 struct tbf_sched_data *q = qdisc_priv(sch);
437 if (cl != 1) /* only one class */
440 tcm->tcm_handle |= TC_H_MIN(1);
441 tcm->tcm_info = q->qdisc->handle;
446 static int tbf_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
449 struct tbf_sched_data *q = qdisc_priv(sch);
455 *old = xchg(&q->qdisc, new);
456 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
458 sch_tree_unlock(sch);
463 static struct Qdisc *tbf_leaf(struct Qdisc *sch, unsigned long arg)
465 struct tbf_sched_data *q = qdisc_priv(sch);
469 static unsigned long tbf_get(struct Qdisc *sch, u32 classid)
474 static void tbf_put(struct Qdisc *sch, unsigned long arg)
478 static int tbf_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
479 struct rtattr **tca, unsigned long *arg)
484 static int tbf_delete(struct Qdisc *sch, unsigned long arg)
489 static void tbf_walk(struct Qdisc *sch, struct qdisc_walker *walker)
492 if (walker->count >= walker->skip)
493 if (walker->fn(sch, 1, walker) < 0) {
501 static struct tcf_proto **tbf_find_tcf(struct Qdisc *sch, unsigned long cl)
506 static struct Qdisc_class_ops tbf_class_ops =
512 .change = tbf_change_class,
513 .delete = tbf_delete,
515 .tcf_chain = tbf_find_tcf,
516 .dump = tbf_dump_class,
519 static struct Qdisc_ops tbf_qdisc_ops = {
521 .cl_ops = &tbf_class_ops,
523 .priv_size = sizeof(struct tbf_sched_data),
524 .enqueue = tbf_enqueue,
525 .dequeue = tbf_dequeue,
526 .requeue = tbf_requeue,
530 .destroy = tbf_destroy,
531 .change = tbf_change,
533 .owner = THIS_MODULE,
536 static int __init tbf_module_init(void)
538 return register_qdisc(&tbf_qdisc_ops);
541 static void __exit tbf_module_exit(void)
543 unregister_qdisc(&tbf_qdisc_ops);
545 module_init(tbf_module_init)
546 module_exit(tbf_module_exit)
547 MODULE_LICENSE("GPL");