blob: b8e156319d7bc48c261ccad7110bbeb84ba44856 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * net/sched/sch_tbf.c Token Bucket Filter queue.
3 *
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.
8 *
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
12 *
13 */
14
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/types.h>
17#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/errno.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/skbuff.h>
Patrick McHardy0ba48052007-07-02 22:49:07 -070021#include <net/netlink.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <net/pkt_sched.h>
23
24
25/* Simple Token Bucket Filter.
26 =======================================
27
28 SOURCE.
29 -------
30
31 None.
32
33 Description.
34 ------------
35
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).
39
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:
43
44 s_i+....+s_k <= B + R*(t_k - t_i)
45
46 Algorithm.
47 ----------
48
49 Let N(t_i) be B/R initially and N(t) grow continuously with time as:
50
51 N(t+delta) = min{B/R, N(t) + delta}
52
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:
56
57 N(t_* + 0) = N(t_* - 0) - S/R.
58
59
60
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.
65
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.
68
69 When TBF works in reshaping mode, latency is estimated as:
70
71 lat = max ((L-B)/R, (L-M)/P)
72
73
74 NOTES.
75 ------
76
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.
83
84
85 This means, that with depth B, the maximal rate is
86
87 R_crit = B*HZ
88
89 F.e. for 10Mbit ethernet and HZ=100 the minimal allowed B is ~10Kbytes.
90
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 :-)
94
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.
98*/
99
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000100struct tbf_sched_data {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101/* Parameters */
102 u32 limit; /* Maximal length of backlog: bytes */
103 u32 buffer; /* Token bucket depth/rate: MUST BE >= MTU/B */
104 u32 mtu;
105 u32 max_size;
106 struct qdisc_rate_table *R_tab;
107 struct qdisc_rate_table *P_tab;
108
109/* Variables */
110 long tokens; /* Current number of B tokens */
111 long ptokens; /* Current number of P tokens */
112 psched_time_t t_c; /* Time check-point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 struct Qdisc *qdisc; /* Inner qdisc, default - bfifo queue */
Patrick McHardyf7f593e2007-03-16 01:20:07 -0700114 struct qdisc_watchdog watchdog; /* Watchdog timer */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115};
116
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000117#define L2T(q, L) qdisc_l2t((q)->R_tab, L)
118#define L2T_P(q, L) qdisc_l2t((q)->P_tab, L)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000120static int tbf_enqueue(struct sk_buff *skb, struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
122 struct tbf_sched_data *q = qdisc_priv(sch);
123 int ret;
124
David S. Miller69747652008-08-17 23:55:36 -0700125 if (qdisc_pkt_len(skb) > q->max_size)
126 return qdisc_reshape_fail(skb, sch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Jussi Kivilinna5f861732008-07-20 00:08:04 -0700128 ret = qdisc_enqueue(skb, q->qdisc);
Ben Greear9871e502010-08-10 01:45:40 -0700129 if (ret != NET_XMIT_SUCCESS) {
Jarek Poplawski378a2f02008-08-04 22:31:03 -0700130 if (net_xmit_drop_count(ret))
131 sch->qstats.drops++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 return ret;
133 }
134
135 sch->q.qlen++;
Ben Greear9871e502010-08-10 01:45:40 -0700136 return NET_XMIT_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137}
138
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000139static unsigned int tbf_drop(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140{
141 struct tbf_sched_data *q = qdisc_priv(sch);
Patrick McHardy6d037a22006-03-20 19:00:49 -0800142 unsigned int len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
Patrick McHardy6d037a22006-03-20 19:00:49 -0800144 if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 sch->q.qlen--;
146 sch->qstats.drops++;
147 }
148 return len;
149}
150
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000151static struct sk_buff *tbf_dequeue(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152{
153 struct tbf_sched_data *q = qdisc_priv(sch);
154 struct sk_buff *skb;
155
Jarek Poplawski03c05f02008-10-31 00:46:19 -0700156 skb = q->qdisc->ops->peek(q->qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
158 if (skb) {
159 psched_time_t now;
Patrick McHardyf7f593e2007-03-16 01:20:07 -0700160 long toks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 long ptoks = 0;
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700162 unsigned int len = qdisc_pkt_len(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
Patrick McHardy3bebcda2007-03-23 11:29:25 -0700164 now = psched_get_time();
Patrick McHardy03cc45c2007-03-23 11:29:11 -0700165 toks = psched_tdiff_bounded(now, q->t_c, q->buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
167 if (q->P_tab) {
168 ptoks = toks + q->ptokens;
169 if (ptoks > (long)q->mtu)
170 ptoks = q->mtu;
171 ptoks -= L2T_P(q, len);
172 }
173 toks += q->tokens;
174 if (toks > (long)q->buffer)
175 toks = q->buffer;
176 toks -= L2T(q, len);
177
178 if ((toks|ptoks) >= 0) {
Jarek Poplawski77be1552008-10-31 00:47:01 -0700179 skb = qdisc_dequeue_peeked(q->qdisc);
Jarek Poplawski03c05f02008-10-31 00:46:19 -0700180 if (unlikely(!skb))
181 return NULL;
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 q->t_c = now;
184 q->tokens = toks;
185 q->ptokens = ptoks;
186 sch->q.qlen--;
Eric Dumazetfd245a42011-01-20 05:27:16 +0000187 qdisc_unthrottled(sch);
Eric Dumazet9190b3b2011-01-20 23:31:33 -0800188 qdisc_bstats_update(sch, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 return skb;
190 }
191
Patrick McHardyf7f593e2007-03-16 01:20:07 -0700192 qdisc_watchdog_schedule(&q->watchdog,
193 now + max_t(long, -toks, -ptoks));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
195 /* Maybe we have a shorter packet in the queue,
196 which can be sent now. It sounds cool,
197 but, however, this is wrong in principle.
198 We MUST NOT reorder packets under these circumstances.
199
200 Really, if we split the flow into independent
201 subflows, it would be a very good solution.
202 This is the main idea of all FQ algorithms
203 (cf. CSZ, HPFQ, HFSC)
204 */
205
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 sch->qstats.overlimits++;
207 }
208 return NULL;
209}
210
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000211static void tbf_reset(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212{
213 struct tbf_sched_data *q = qdisc_priv(sch);
214
215 qdisc_reset(q->qdisc);
216 sch->q.qlen = 0;
Patrick McHardy3bebcda2007-03-23 11:29:25 -0700217 q->t_c = psched_get_time();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 q->tokens = q->buffer;
219 q->ptokens = q->mtu;
Patrick McHardyf7f593e2007-03-16 01:20:07 -0700220 qdisc_watchdog_cancel(&q->watchdog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221}
222
Patrick McHardy27a34212008-01-23 20:35:39 -0800223static const struct nla_policy tbf_policy[TCA_TBF_MAX + 1] = {
224 [TCA_TBF_PARMS] = { .len = sizeof(struct tc_tbf_qopt) },
225 [TCA_TBF_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
226 [TCA_TBF_PTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
227};
228
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000229static int tbf_change(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230{
Patrick McHardycee63722008-01-23 20:33:32 -0800231 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 struct tbf_sched_data *q = qdisc_priv(sch);
Patrick McHardy1e904742008-01-22 22:11:17 -0800233 struct nlattr *tb[TCA_TBF_PTAB + 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 struct tc_tbf_qopt *qopt;
235 struct qdisc_rate_table *rtab = NULL;
236 struct qdisc_rate_table *ptab = NULL;
237 struct Qdisc *child = NULL;
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000238 int max_size, n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Patrick McHardy27a34212008-01-23 20:35:39 -0800240 err = nla_parse_nested(tb, TCA_TBF_PTAB, opt, tbf_policy);
Patrick McHardycee63722008-01-23 20:33:32 -0800241 if (err < 0)
242 return err;
243
244 err = -EINVAL;
Patrick McHardy27a34212008-01-23 20:35:39 -0800245 if (tb[TCA_TBF_PARMS] == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 goto done;
247
Patrick McHardy1e904742008-01-22 22:11:17 -0800248 qopt = nla_data(tb[TCA_TBF_PARMS]);
249 rtab = qdisc_get_rtab(&qopt->rate, tb[TCA_TBF_RTAB]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 if (rtab == NULL)
251 goto done;
252
253 if (qopt->peakrate.rate) {
254 if (qopt->peakrate.rate > qopt->rate.rate)
Patrick McHardy1e904742008-01-22 22:11:17 -0800255 ptab = qdisc_get_rtab(&qopt->peakrate, tb[TCA_TBF_PTAB]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 if (ptab == NULL)
257 goto done;
258 }
259
260 for (n = 0; n < 256; n++)
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000261 if (rtab->data[n] > qopt->buffer)
262 break;
263 max_size = (n << qopt->rate.cell_log) - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 if (ptab) {
265 int size;
266
267 for (n = 0; n < 256; n++)
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000268 if (ptab->data[n] > qopt->mtu)
269 break;
270 size = (n << qopt->peakrate.cell_log) - 1;
271 if (size < max_size)
272 max_size = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 }
274 if (max_size < 0)
275 goto done;
276
stephen hemmingerf0cd1502010-05-14 14:38:59 +0000277 if (q->qdisc != &noop_qdisc) {
278 err = fifo_set_limit(q->qdisc, qopt->limit);
279 if (err)
280 goto done;
281 } else if (qopt->limit > 0) {
Patrick McHardyfb0305c2008-07-05 23:40:21 -0700282 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, qopt->limit);
283 if (IS_ERR(child)) {
284 err = PTR_ERR(child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 goto done;
Patrick McHardyfb0305c2008-07-05 23:40:21 -0700286 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 }
288
289 sch_tree_lock(sch);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800290 if (child) {
291 qdisc_tree_decrease_qlen(q->qdisc, q->qdisc->q.qlen);
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800292 qdisc_destroy(q->qdisc);
293 q->qdisc = child;
Patrick McHardy5e50da02006-11-29 17:36:20 -0800294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 q->limit = qopt->limit;
296 q->mtu = qopt->mtu;
297 q->max_size = max_size;
298 q->buffer = qopt->buffer;
299 q->tokens = q->buffer;
300 q->ptokens = q->mtu;
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800301
Ilpo Järvinena0bffff2009-03-21 13:36:17 -0700302 swap(q->R_tab, rtab);
303 swap(q->P_tab, ptab);
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800304
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 sch_tree_unlock(sch);
306 err = 0;
307done:
308 if (rtab)
309 qdisc_put_rtab(rtab);
310 if (ptab)
311 qdisc_put_rtab(ptab);
312 return err;
313}
314
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000315static int tbf_init(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316{
317 struct tbf_sched_data *q = qdisc_priv(sch);
318
319 if (opt == NULL)
320 return -EINVAL;
321
Patrick McHardy3bebcda2007-03-23 11:29:25 -0700322 q->t_c = psched_get_time();
Patrick McHardyf7f593e2007-03-16 01:20:07 -0700323 qdisc_watchdog_init(&q->watchdog, sch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 q->qdisc = &noop_qdisc;
325
326 return tbf_change(sch, opt);
327}
328
329static void tbf_destroy(struct Qdisc *sch)
330{
331 struct tbf_sched_data *q = qdisc_priv(sch);
332
Patrick McHardyf7f593e2007-03-16 01:20:07 -0700333 qdisc_watchdog_cancel(&q->watchdog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
335 if (q->P_tab)
336 qdisc_put_rtab(q->P_tab);
337 if (q->R_tab)
338 qdisc_put_rtab(q->R_tab);
339
340 qdisc_destroy(q->qdisc);
341}
342
343static int tbf_dump(struct Qdisc *sch, struct sk_buff *skb)
344{
345 struct tbf_sched_data *q = qdisc_priv(sch);
Patrick McHardy4b3550ef2008-01-23 20:34:11 -0800346 struct nlattr *nest;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 struct tc_tbf_qopt opt;
348
Eric Dumazetb0460e42011-12-28 23:27:44 +0000349 sch->qstats.backlog = q->qdisc->qstats.backlog;
Patrick McHardy4b3550ef2008-01-23 20:34:11 -0800350 nest = nla_nest_start(skb, TCA_OPTIONS);
351 if (nest == NULL)
352 goto nla_put_failure;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
354 opt.limit = q->limit;
355 opt.rate = q->R_tab->rate;
356 if (q->P_tab)
357 opt.peakrate = q->P_tab->rate;
358 else
359 memset(&opt.peakrate, 0, sizeof(opt.peakrate));
360 opt.mtu = q->mtu;
361 opt.buffer = q->buffer;
Patrick McHardy1e904742008-01-22 22:11:17 -0800362 NLA_PUT(skb, TCA_TBF_PARMS, sizeof(opt), &opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363
Patrick McHardy4b3550ef2008-01-23 20:34:11 -0800364 nla_nest_end(skb, nest);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 return skb->len;
366
Patrick McHardy1e904742008-01-22 22:11:17 -0800367nla_put_failure:
Patrick McHardy4b3550ef2008-01-23 20:34:11 -0800368 nla_nest_cancel(skb, nest);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 return -1;
370}
371
372static int tbf_dump_class(struct Qdisc *sch, unsigned long cl,
373 struct sk_buff *skb, struct tcmsg *tcm)
374{
375 struct tbf_sched_data *q = qdisc_priv(sch);
376
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 tcm->tcm_handle |= TC_H_MIN(1);
378 tcm->tcm_info = q->qdisc->handle;
379
380 return 0;
381}
382
383static int tbf_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
384 struct Qdisc **old)
385{
386 struct tbf_sched_data *q = qdisc_priv(sch);
387
388 if (new == NULL)
389 new = &noop_qdisc;
390
391 sch_tree_lock(sch);
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800392 *old = q->qdisc;
393 q->qdisc = new;
Patrick McHardy5e50da02006-11-29 17:36:20 -0800394 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 qdisc_reset(*old);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 sch_tree_unlock(sch);
397
398 return 0;
399}
400
401static struct Qdisc *tbf_leaf(struct Qdisc *sch, unsigned long arg)
402{
403 struct tbf_sched_data *q = qdisc_priv(sch);
404 return q->qdisc;
405}
406
407static unsigned long tbf_get(struct Qdisc *sch, u32 classid)
408{
409 return 1;
410}
411
412static void tbf_put(struct Qdisc *sch, unsigned long arg)
413{
414}
415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416static void tbf_walk(struct Qdisc *sch, struct qdisc_walker *walker)
417{
418 if (!walker->stop) {
419 if (walker->count >= walker->skip)
420 if (walker->fn(sch, 1, walker) < 0) {
421 walker->stop = 1;
422 return;
423 }
424 walker->count++;
425 }
426}
427
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000428static const struct Qdisc_class_ops tbf_class_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 .graft = tbf_graft,
430 .leaf = tbf_leaf,
431 .get = tbf_get,
432 .put = tbf_put,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 .walk = tbf_walk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 .dump = tbf_dump_class,
435};
436
Eric Dumazet20fea082007-11-14 01:44:41 -0800437static struct Qdisc_ops tbf_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 .next = NULL,
439 .cl_ops = &tbf_class_ops,
440 .id = "tbf",
441 .priv_size = sizeof(struct tbf_sched_data),
442 .enqueue = tbf_enqueue,
443 .dequeue = tbf_dequeue,
Jarek Poplawski77be1552008-10-31 00:47:01 -0700444 .peek = qdisc_peek_dequeued,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 .drop = tbf_drop,
446 .init = tbf_init,
447 .reset = tbf_reset,
448 .destroy = tbf_destroy,
449 .change = tbf_change,
450 .dump = tbf_dump,
451 .owner = THIS_MODULE,
452};
453
454static int __init tbf_module_init(void)
455{
456 return register_qdisc(&tbf_qdisc_ops);
457}
458
459static void __exit tbf_module_exit(void)
460{
461 unregister_qdisc(&tbf_qdisc_ops);
462}
463module_init(tbf_module_init)
464module_exit(tbf_module_exit)
465MODULE_LICENSE("GPL");