]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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 | ||
1da177e4 | 15 | #include <linux/module.h> |
1da177e4 LT |
16 | #include <linux/types.h> |
17 | #include <linux/kernel.h> | |
1da177e4 | 18 | #include <linux/string.h> |
1da177e4 | 19 | #include <linux/errno.h> |
1da177e4 | 20 | #include <linux/skbuff.h> |
0ba48053 | 21 | #include <net/netlink.h> |
1da177e4 LT |
22 | #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 | ||
100 | struct tbf_sched_data | |
101 | { | |
102 | /* Parameters */ | |
103 | u32 limit; /* Maximal length of backlog: bytes */ | |
104 | u32 buffer; /* Token bucket depth/rate: MUST BE >= MTU/B */ | |
105 | u32 mtu; | |
106 | u32 max_size; | |
107 | struct qdisc_rate_table *R_tab; | |
108 | struct qdisc_rate_table *P_tab; | |
109 | ||
110 | /* Variables */ | |
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 */ | |
1da177e4 | 114 | struct Qdisc *qdisc; /* Inner qdisc, default - bfifo queue */ |
f7f593e3 | 115 | struct qdisc_watchdog watchdog; /* Watchdog timer */ |
1da177e4 LT |
116 | }; |
117 | ||
e9bef55d JDB |
118 | #define L2T(q,L) qdisc_l2t((q)->R_tab,L) |
119 | #define L2T_P(q,L) qdisc_l2t((q)->P_tab,L) | |
1da177e4 LT |
120 | |
121 | static int tbf_enqueue(struct sk_buff *skb, struct Qdisc* sch) | |
122 | { | |
123 | struct tbf_sched_data *q = qdisc_priv(sch); | |
124 | int ret; | |
125 | ||
69747650 DM |
126 | if (qdisc_pkt_len(skb) > q->max_size) |
127 | return qdisc_reshape_fail(skb, sch); | |
1da177e4 | 128 | |
5f86173b JK |
129 | ret = qdisc_enqueue(skb, q->qdisc); |
130 | if (ret != 0) { | |
378a2f09 JP |
131 | if (net_xmit_drop_count(ret)) |
132 | sch->qstats.drops++; | |
1da177e4 LT |
133 | return ret; |
134 | } | |
135 | ||
136 | sch->q.qlen++; | |
0abf77e5 | 137 | sch->bstats.bytes += qdisc_pkt_len(skb); |
1da177e4 LT |
138 | sch->bstats.packets++; |
139 | return 0; | |
140 | } | |
141 | ||
1da177e4 LT |
142 | static unsigned int tbf_drop(struct Qdisc* sch) |
143 | { | |
144 | struct tbf_sched_data *q = qdisc_priv(sch); | |
6d037a26 | 145 | unsigned int len = 0; |
1da177e4 | 146 | |
6d037a26 | 147 | if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) { |
1da177e4 LT |
148 | sch->q.qlen--; |
149 | sch->qstats.drops++; | |
150 | } | |
151 | return len; | |
152 | } | |
153 | ||
1da177e4 LT |
154 | static struct sk_buff *tbf_dequeue(struct Qdisc* sch) |
155 | { | |
156 | struct tbf_sched_data *q = qdisc_priv(sch); | |
157 | struct sk_buff *skb; | |
158 | ||
03c05f0d | 159 | skb = q->qdisc->ops->peek(q->qdisc); |
1da177e4 LT |
160 | |
161 | if (skb) { | |
162 | psched_time_t now; | |
f7f593e3 | 163 | long toks; |
1da177e4 | 164 | long ptoks = 0; |
0abf77e5 | 165 | unsigned int len = qdisc_pkt_len(skb); |
1da177e4 | 166 | |
3bebcda2 | 167 | now = psched_get_time(); |
03cc45c0 | 168 | toks = psched_tdiff_bounded(now, q->t_c, q->buffer); |
1da177e4 LT |
169 | |
170 | if (q->P_tab) { | |
171 | ptoks = toks + q->ptokens; | |
172 | if (ptoks > (long)q->mtu) | |
173 | ptoks = q->mtu; | |
174 | ptoks -= L2T_P(q, len); | |
175 | } | |
176 | toks += q->tokens; | |
177 | if (toks > (long)q->buffer) | |
178 | toks = q->buffer; | |
179 | toks -= L2T(q, len); | |
180 | ||
181 | if ((toks|ptoks) >= 0) { | |
77be155c | 182 | skb = qdisc_dequeue_peeked(q->qdisc); |
03c05f0d JP |
183 | if (unlikely(!skb)) |
184 | return NULL; | |
185 | ||
1da177e4 LT |
186 | q->t_c = now; |
187 | q->tokens = toks; | |
188 | q->ptokens = ptoks; | |
189 | sch->q.qlen--; | |
190 | sch->flags &= ~TCQ_F_THROTTLED; | |
191 | return skb; | |
192 | } | |
193 | ||
f7f593e3 PM |
194 | qdisc_watchdog_schedule(&q->watchdog, |
195 | now + max_t(long, -toks, -ptoks)); | |
1da177e4 LT |
196 | |
197 | /* Maybe we have a shorter packet in the queue, | |
198 | which can be sent now. It sounds cool, | |
199 | but, however, this is wrong in principle. | |
200 | We MUST NOT reorder packets under these circumstances. | |
201 | ||
202 | Really, if we split the flow into independent | |
203 | subflows, it would be a very good solution. | |
204 | This is the main idea of all FQ algorithms | |
205 | (cf. CSZ, HPFQ, HFSC) | |
206 | */ | |
207 | ||
1da177e4 LT |
208 | sch->qstats.overlimits++; |
209 | } | |
210 | return NULL; | |
211 | } | |
212 | ||
213 | static void tbf_reset(struct Qdisc* sch) | |
214 | { | |
215 | struct tbf_sched_data *q = qdisc_priv(sch); | |
216 | ||
217 | qdisc_reset(q->qdisc); | |
218 | sch->q.qlen = 0; | |
3bebcda2 | 219 | q->t_c = psched_get_time(); |
1da177e4 LT |
220 | q->tokens = q->buffer; |
221 | q->ptokens = q->mtu; | |
f7f593e3 | 222 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
223 | } |
224 | ||
27a3421e PM |
225 | static const struct nla_policy tbf_policy[TCA_TBF_MAX + 1] = { |
226 | [TCA_TBF_PARMS] = { .len = sizeof(struct tc_tbf_qopt) }, | |
227 | [TCA_TBF_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
228 | [TCA_TBF_PTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
229 | }; | |
230 | ||
1e90474c | 231 | static int tbf_change(struct Qdisc* sch, struct nlattr *opt) |
1da177e4 | 232 | { |
cee63723 | 233 | int err; |
1da177e4 | 234 | struct tbf_sched_data *q = qdisc_priv(sch); |
1e90474c | 235 | struct nlattr *tb[TCA_TBF_PTAB + 1]; |
1da177e4 LT |
236 | struct tc_tbf_qopt *qopt; |
237 | struct qdisc_rate_table *rtab = NULL; | |
238 | struct qdisc_rate_table *ptab = NULL; | |
239 | struct Qdisc *child = NULL; | |
240 | int max_size,n; | |
241 | ||
27a3421e | 242 | err = nla_parse_nested(tb, TCA_TBF_PTAB, opt, tbf_policy); |
cee63723 PM |
243 | if (err < 0) |
244 | return err; | |
245 | ||
246 | err = -EINVAL; | |
27a3421e | 247 | if (tb[TCA_TBF_PARMS] == NULL) |
1da177e4 LT |
248 | goto done; |
249 | ||
1e90474c PM |
250 | qopt = nla_data(tb[TCA_TBF_PARMS]); |
251 | rtab = qdisc_get_rtab(&qopt->rate, tb[TCA_TBF_RTAB]); | |
1da177e4 LT |
252 | if (rtab == NULL) |
253 | goto done; | |
254 | ||
255 | if (qopt->peakrate.rate) { | |
256 | if (qopt->peakrate.rate > qopt->rate.rate) | |
1e90474c | 257 | ptab = qdisc_get_rtab(&qopt->peakrate, tb[TCA_TBF_PTAB]); |
1da177e4 LT |
258 | if (ptab == NULL) |
259 | goto done; | |
260 | } | |
261 | ||
262 | for (n = 0; n < 256; n++) | |
263 | if (rtab->data[n] > qopt->buffer) break; | |
264 | max_size = (n << qopt->rate.cell_log)-1; | |
265 | if (ptab) { | |
266 | int size; | |
267 | ||
268 | for (n = 0; n < 256; n++) | |
269 | if (ptab->data[n] > qopt->mtu) break; | |
270 | size = (n << qopt->peakrate.cell_log)-1; | |
271 | if (size < max_size) max_size = size; | |
272 | } | |
273 | if (max_size < 0) | |
274 | goto done; | |
275 | ||
053cfed7 | 276 | if (qopt->limit > 0) { |
fb0305ce PM |
277 | child = fifo_create_dflt(sch, &bfifo_qdisc_ops, qopt->limit); |
278 | if (IS_ERR(child)) { | |
279 | err = PTR_ERR(child); | |
1da177e4 | 280 | goto done; |
fb0305ce | 281 | } |
1da177e4 LT |
282 | } |
283 | ||
284 | sch_tree_lock(sch); | |
5e50da01 PM |
285 | if (child) { |
286 | qdisc_tree_decrease_qlen(q->qdisc, q->qdisc->q.qlen); | |
053cfed7 | 287 | qdisc_destroy(xchg(&q->qdisc, child)); |
5e50da01 | 288 | } |
1da177e4 LT |
289 | q->limit = qopt->limit; |
290 | q->mtu = qopt->mtu; | |
291 | q->max_size = max_size; | |
292 | q->buffer = qopt->buffer; | |
293 | q->tokens = q->buffer; | |
294 | q->ptokens = q->mtu; | |
295 | rtab = xchg(&q->R_tab, rtab); | |
296 | ptab = xchg(&q->P_tab, ptab); | |
297 | sch_tree_unlock(sch); | |
298 | err = 0; | |
299 | done: | |
300 | if (rtab) | |
301 | qdisc_put_rtab(rtab); | |
302 | if (ptab) | |
303 | qdisc_put_rtab(ptab); | |
304 | return err; | |
305 | } | |
306 | ||
1e90474c | 307 | static int tbf_init(struct Qdisc* sch, struct nlattr *opt) |
1da177e4 LT |
308 | { |
309 | struct tbf_sched_data *q = qdisc_priv(sch); | |
310 | ||
311 | if (opt == NULL) | |
312 | return -EINVAL; | |
313 | ||
3bebcda2 | 314 | q->t_c = psched_get_time(); |
f7f593e3 | 315 | qdisc_watchdog_init(&q->watchdog, sch); |
1da177e4 LT |
316 | q->qdisc = &noop_qdisc; |
317 | ||
318 | return tbf_change(sch, opt); | |
319 | } | |
320 | ||
321 | static void tbf_destroy(struct Qdisc *sch) | |
322 | { | |
323 | struct tbf_sched_data *q = qdisc_priv(sch); | |
324 | ||
f7f593e3 | 325 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
326 | |
327 | if (q->P_tab) | |
328 | qdisc_put_rtab(q->P_tab); | |
329 | if (q->R_tab) | |
330 | qdisc_put_rtab(q->R_tab); | |
331 | ||
332 | qdisc_destroy(q->qdisc); | |
333 | } | |
334 | ||
335 | static int tbf_dump(struct Qdisc *sch, struct sk_buff *skb) | |
336 | { | |
337 | struct tbf_sched_data *q = qdisc_priv(sch); | |
4b3550ef | 338 | struct nlattr *nest; |
1da177e4 LT |
339 | struct tc_tbf_qopt opt; |
340 | ||
4b3550ef PM |
341 | nest = nla_nest_start(skb, TCA_OPTIONS); |
342 | if (nest == NULL) | |
343 | goto nla_put_failure; | |
1da177e4 LT |
344 | |
345 | opt.limit = q->limit; | |
346 | opt.rate = q->R_tab->rate; | |
347 | if (q->P_tab) | |
348 | opt.peakrate = q->P_tab->rate; | |
349 | else | |
350 | memset(&opt.peakrate, 0, sizeof(opt.peakrate)); | |
351 | opt.mtu = q->mtu; | |
352 | opt.buffer = q->buffer; | |
1e90474c | 353 | NLA_PUT(skb, TCA_TBF_PARMS, sizeof(opt), &opt); |
1da177e4 | 354 | |
4b3550ef | 355 | nla_nest_end(skb, nest); |
1da177e4 LT |
356 | return skb->len; |
357 | ||
1e90474c | 358 | nla_put_failure: |
4b3550ef | 359 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
360 | return -1; |
361 | } | |
362 | ||
363 | static int tbf_dump_class(struct Qdisc *sch, unsigned long cl, | |
364 | struct sk_buff *skb, struct tcmsg *tcm) | |
365 | { | |
366 | struct tbf_sched_data *q = qdisc_priv(sch); | |
367 | ||
368 | if (cl != 1) /* only one class */ | |
369 | return -ENOENT; | |
370 | ||
371 | tcm->tcm_handle |= TC_H_MIN(1); | |
372 | tcm->tcm_info = q->qdisc->handle; | |
373 | ||
374 | return 0; | |
375 | } | |
376 | ||
377 | static int tbf_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
378 | struct Qdisc **old) | |
379 | { | |
380 | struct tbf_sched_data *q = qdisc_priv(sch); | |
381 | ||
382 | if (new == NULL) | |
383 | new = &noop_qdisc; | |
384 | ||
385 | sch_tree_lock(sch); | |
386 | *old = xchg(&q->qdisc, new); | |
5e50da01 | 387 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); |
1da177e4 | 388 | qdisc_reset(*old); |
1da177e4 LT |
389 | sch_tree_unlock(sch); |
390 | ||
391 | return 0; | |
392 | } | |
393 | ||
394 | static struct Qdisc *tbf_leaf(struct Qdisc *sch, unsigned long arg) | |
395 | { | |
396 | struct tbf_sched_data *q = qdisc_priv(sch); | |
397 | return q->qdisc; | |
398 | } | |
399 | ||
400 | static unsigned long tbf_get(struct Qdisc *sch, u32 classid) | |
401 | { | |
402 | return 1; | |
403 | } | |
404 | ||
405 | static void tbf_put(struct Qdisc *sch, unsigned long arg) | |
406 | { | |
407 | } | |
408 | ||
10297b99 | 409 | static int tbf_change_class(struct Qdisc *sch, u32 classid, u32 parentid, |
1e90474c | 410 | struct nlattr **tca, unsigned long *arg) |
1da177e4 LT |
411 | { |
412 | return -ENOSYS; | |
413 | } | |
414 | ||
415 | static int tbf_delete(struct Qdisc *sch, unsigned long arg) | |
416 | { | |
417 | return -ENOSYS; | |
418 | } | |
419 | ||
420 | static void tbf_walk(struct Qdisc *sch, struct qdisc_walker *walker) | |
421 | { | |
422 | if (!walker->stop) { | |
423 | if (walker->count >= walker->skip) | |
424 | if (walker->fn(sch, 1, walker) < 0) { | |
425 | walker->stop = 1; | |
426 | return; | |
427 | } | |
428 | walker->count++; | |
429 | } | |
430 | } | |
431 | ||
432 | static struct tcf_proto **tbf_find_tcf(struct Qdisc *sch, unsigned long cl) | |
433 | { | |
434 | return NULL; | |
435 | } | |
436 | ||
20fea08b | 437 | static const struct Qdisc_class_ops tbf_class_ops = |
1da177e4 LT |
438 | { |
439 | .graft = tbf_graft, | |
440 | .leaf = tbf_leaf, | |
441 | .get = tbf_get, | |
442 | .put = tbf_put, | |
443 | .change = tbf_change_class, | |
444 | .delete = tbf_delete, | |
445 | .walk = tbf_walk, | |
446 | .tcf_chain = tbf_find_tcf, | |
447 | .dump = tbf_dump_class, | |
448 | }; | |
449 | ||
20fea08b | 450 | static struct Qdisc_ops tbf_qdisc_ops __read_mostly = { |
1da177e4 LT |
451 | .next = NULL, |
452 | .cl_ops = &tbf_class_ops, | |
453 | .id = "tbf", | |
454 | .priv_size = sizeof(struct tbf_sched_data), | |
455 | .enqueue = tbf_enqueue, | |
456 | .dequeue = tbf_dequeue, | |
77be155c | 457 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
458 | .drop = tbf_drop, |
459 | .init = tbf_init, | |
460 | .reset = tbf_reset, | |
461 | .destroy = tbf_destroy, | |
462 | .change = tbf_change, | |
463 | .dump = tbf_dump, | |
464 | .owner = THIS_MODULE, | |
465 | }; | |
466 | ||
467 | static int __init tbf_module_init(void) | |
468 | { | |
469 | return register_qdisc(&tbf_qdisc_ops); | |
470 | } | |
471 | ||
472 | static void __exit tbf_module_exit(void) | |
473 | { | |
474 | unregister_qdisc(&tbf_qdisc_ops); | |
475 | } | |
476 | module_init(tbf_module_init) | |
477 | module_exit(tbf_module_exit) | |
478 | MODULE_LICENSE("GPL"); |