]>
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> |
b757c933 | 22 | #include <net/sch_generic.h> |
1da177e4 LT |
23 | #include <net/pkt_sched.h> |
24 | ||
25 | ||
26 | /* Simple Token Bucket Filter. | |
27 | ======================================= | |
28 | ||
29 | SOURCE. | |
30 | ------- | |
31 | ||
32 | None. | |
33 | ||
34 | Description. | |
35 | ------------ | |
36 | ||
37 | A data flow obeys TBF with rate R and depth B, if for any | |
38 | time interval t_i...t_f the number of transmitted bits | |
39 | does not exceed B + R*(t_f-t_i). | |
40 | ||
41 | Packetized version of this definition: | |
42 | The sequence of packets of sizes s_i served at moments t_i | |
43 | obeys TBF, if for any i<=k: | |
44 | ||
45 | s_i+....+s_k <= B + R*(t_k - t_i) | |
46 | ||
47 | Algorithm. | |
48 | ---------- | |
49 | ||
50 | Let N(t_i) be B/R initially and N(t) grow continuously with time as: | |
51 | ||
52 | N(t+delta) = min{B/R, N(t) + delta} | |
53 | ||
54 | If the first packet in queue has length S, it may be | |
55 | transmitted only at the time t_* when S/R <= N(t_*), | |
56 | and in this case N(t) jumps: | |
57 | ||
58 | N(t_* + 0) = N(t_* - 0) - S/R. | |
59 | ||
60 | ||
61 | ||
62 | Actually, QoS requires two TBF to be applied to a data stream. | |
63 | One of them controls steady state burst size, another | |
64 | one with rate P (peak rate) and depth M (equal to link MTU) | |
65 | limits bursts at a smaller time scale. | |
66 | ||
67 | It is easy to see that P>R, and B>M. If P is infinity, this double | |
68 | TBF is equivalent to a single one. | |
69 | ||
70 | When TBF works in reshaping mode, latency is estimated as: | |
71 | ||
72 | lat = max ((L-B)/R, (L-M)/P) | |
73 | ||
74 | ||
75 | NOTES. | |
76 | ------ | |
77 | ||
78 | If TBF throttles, it starts a watchdog timer, which will wake it up | |
79 | when it is ready to transmit. | |
80 | Note that the minimal timer resolution is 1/HZ. | |
81 | If no new packets arrive during this period, | |
82 | or if the device is not awaken by EOI for some previous packet, | |
83 | TBF can stop its activity for 1/HZ. | |
84 | ||
85 | ||
86 | This means, that with depth B, the maximal rate is | |
87 | ||
88 | R_crit = B*HZ | |
89 | ||
90 | F.e. for 10Mbit ethernet and HZ=100 the minimal allowed B is ~10Kbytes. | |
91 | ||
92 | Note that the peak rate TBF is much more tough: with MTU 1500 | |
93 | P_crit = 150Kbytes/sec. So, if you need greater peak | |
94 | rates, use alpha with HZ=1000 :-) | |
95 | ||
96 | With classful TBF, limit is just kept for backwards compatibility. | |
97 | It is passed to the default bfifo qdisc - if the inner qdisc is | |
98 | changed the limit is not effective anymore. | |
99 | */ | |
100 | ||
cc7ec456 | 101 | struct tbf_sched_data { |
1da177e4 LT |
102 | /* Parameters */ |
103 | u32 limit; /* Maximal length of backlog: bytes */ | |
b757c933 JP |
104 | s64 buffer; /* Token bucket depth/rate: MUST BE >= MTU/B */ |
105 | s64 mtu; | |
1da177e4 | 106 | u32 max_size; |
b757c933 JP |
107 | struct psched_ratecfg rate; |
108 | struct psched_ratecfg peak; | |
109 | bool peak_present; | |
1da177e4 LT |
110 | |
111 | /* Variables */ | |
b757c933 JP |
112 | s64 tokens; /* Current number of B tokens */ |
113 | s64 ptokens; /* Current number of P tokens */ | |
114 | s64 t_c; /* Time check-point */ | |
1da177e4 | 115 | struct Qdisc *qdisc; /* Inner qdisc, default - bfifo queue */ |
f7f593e3 | 116 | struct qdisc_watchdog watchdog; /* Watchdog timer */ |
1da177e4 LT |
117 | }; |
118 | ||
cc7ec456 | 119 | static int tbf_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
1da177e4 LT |
120 | { |
121 | struct tbf_sched_data *q = qdisc_priv(sch); | |
122 | int ret; | |
123 | ||
69747650 DM |
124 | if (qdisc_pkt_len(skb) > q->max_size) |
125 | return qdisc_reshape_fail(skb, sch); | |
1da177e4 | 126 | |
5f86173b | 127 | ret = qdisc_enqueue(skb, q->qdisc); |
9871e50e | 128 | if (ret != NET_XMIT_SUCCESS) { |
378a2f09 JP |
129 | if (net_xmit_drop_count(ret)) |
130 | sch->qstats.drops++; | |
1da177e4 LT |
131 | return ret; |
132 | } | |
133 | ||
134 | sch->q.qlen++; | |
9871e50e | 135 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
136 | } |
137 | ||
cc7ec456 | 138 | static unsigned int tbf_drop(struct Qdisc *sch) |
1da177e4 LT |
139 | { |
140 | struct tbf_sched_data *q = qdisc_priv(sch); | |
6d037a26 | 141 | unsigned int len = 0; |
1da177e4 | 142 | |
6d037a26 | 143 | if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) { |
1da177e4 LT |
144 | sch->q.qlen--; |
145 | sch->qstats.drops++; | |
146 | } | |
147 | return len; | |
148 | } | |
149 | ||
cc7ec456 | 150 | static struct sk_buff *tbf_dequeue(struct Qdisc *sch) |
1da177e4 LT |
151 | { |
152 | struct tbf_sched_data *q = qdisc_priv(sch); | |
153 | struct sk_buff *skb; | |
154 | ||
03c05f0d | 155 | skb = q->qdisc->ops->peek(q->qdisc); |
1da177e4 LT |
156 | |
157 | if (skb) { | |
b757c933 JP |
158 | s64 now; |
159 | s64 toks; | |
160 | s64 ptoks = 0; | |
0abf77e5 | 161 | unsigned int len = qdisc_pkt_len(skb); |
1da177e4 | 162 | |
b757c933 JP |
163 | now = ktime_to_ns(ktime_get()); |
164 | toks = min_t(s64, now - q->t_c, q->buffer); | |
1da177e4 | 165 | |
b757c933 | 166 | if (q->peak_present) { |
1da177e4 | 167 | ptoks = toks + q->ptokens; |
b757c933 | 168 | if (ptoks > q->mtu) |
1da177e4 | 169 | ptoks = q->mtu; |
b757c933 | 170 | ptoks -= (s64) psched_l2t_ns(&q->peak, len); |
1da177e4 LT |
171 | } |
172 | toks += q->tokens; | |
b757c933 | 173 | if (toks > q->buffer) |
1da177e4 | 174 | toks = q->buffer; |
b757c933 | 175 | toks -= (s64) psched_l2t_ns(&q->rate, len); |
1da177e4 LT |
176 | |
177 | if ((toks|ptoks) >= 0) { | |
77be155c | 178 | skb = qdisc_dequeue_peeked(q->qdisc); |
03c05f0d JP |
179 | if (unlikely(!skb)) |
180 | return NULL; | |
181 | ||
1da177e4 LT |
182 | q->t_c = now; |
183 | q->tokens = toks; | |
184 | q->ptokens = ptoks; | |
185 | sch->q.qlen--; | |
fd245a4a | 186 | qdisc_unthrottled(sch); |
9190b3b3 | 187 | qdisc_bstats_update(sch, skb); |
1da177e4 LT |
188 | return skb; |
189 | } | |
190 | ||
b757c933 JP |
191 | qdisc_watchdog_schedule_ns(&q->watchdog, |
192 | now + max_t(long, -toks, -ptoks)); | |
1da177e4 LT |
193 | |
194 | /* Maybe we have a shorter packet in the queue, | |
195 | which can be sent now. It sounds cool, | |
196 | but, however, this is wrong in principle. | |
197 | We MUST NOT reorder packets under these circumstances. | |
198 | ||
199 | Really, if we split the flow into independent | |
200 | subflows, it would be a very good solution. | |
201 | This is the main idea of all FQ algorithms | |
202 | (cf. CSZ, HPFQ, HFSC) | |
203 | */ | |
204 | ||
1da177e4 LT |
205 | sch->qstats.overlimits++; |
206 | } | |
207 | return NULL; | |
208 | } | |
209 | ||
cc7ec456 | 210 | static void tbf_reset(struct Qdisc *sch) |
1da177e4 LT |
211 | { |
212 | struct tbf_sched_data *q = qdisc_priv(sch); | |
213 | ||
214 | qdisc_reset(q->qdisc); | |
215 | sch->q.qlen = 0; | |
b757c933 | 216 | q->t_c = ktime_to_ns(ktime_get()); |
1da177e4 LT |
217 | q->tokens = q->buffer; |
218 | q->ptokens = q->mtu; | |
f7f593e3 | 219 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
220 | } |
221 | ||
27a3421e PM |
222 | static const struct nla_policy tbf_policy[TCA_TBF_MAX + 1] = { |
223 | [TCA_TBF_PARMS] = { .len = sizeof(struct tc_tbf_qopt) }, | |
224 | [TCA_TBF_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
225 | [TCA_TBF_PTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, | |
226 | }; | |
227 | ||
cc7ec456 | 228 | static int tbf_change(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 | 229 | { |
cee63723 | 230 | int err; |
1da177e4 | 231 | struct tbf_sched_data *q = qdisc_priv(sch); |
1e90474c | 232 | struct nlattr *tb[TCA_TBF_PTAB + 1]; |
1da177e4 LT |
233 | struct tc_tbf_qopt *qopt; |
234 | struct qdisc_rate_table *rtab = NULL; | |
235 | struct qdisc_rate_table *ptab = NULL; | |
236 | struct Qdisc *child = NULL; | |
cc7ec456 | 237 | int max_size, n; |
1da177e4 | 238 | |
27a3421e | 239 | err = nla_parse_nested(tb, TCA_TBF_PTAB, opt, tbf_policy); |
cee63723 PM |
240 | if (err < 0) |
241 | return err; | |
242 | ||
243 | err = -EINVAL; | |
27a3421e | 244 | if (tb[TCA_TBF_PARMS] == NULL) |
1da177e4 LT |
245 | goto done; |
246 | ||
1e90474c PM |
247 | qopt = nla_data(tb[TCA_TBF_PARMS]); |
248 | rtab = qdisc_get_rtab(&qopt->rate, tb[TCA_TBF_RTAB]); | |
1da177e4 LT |
249 | if (rtab == NULL) |
250 | goto done; | |
251 | ||
252 | if (qopt->peakrate.rate) { | |
253 | if (qopt->peakrate.rate > qopt->rate.rate) | |
1e90474c | 254 | ptab = qdisc_get_rtab(&qopt->peakrate, tb[TCA_TBF_PTAB]); |
1da177e4 LT |
255 | if (ptab == NULL) |
256 | goto done; | |
257 | } | |
258 | ||
259 | for (n = 0; n < 256; n++) | |
cc7ec456 ED |
260 | if (rtab->data[n] > qopt->buffer) |
261 | break; | |
262 | max_size = (n << qopt->rate.cell_log) - 1; | |
1da177e4 LT |
263 | if (ptab) { |
264 | int size; | |
265 | ||
266 | for (n = 0; n < 256; n++) | |
cc7ec456 ED |
267 | if (ptab->data[n] > qopt->mtu) |
268 | break; | |
269 | size = (n << qopt->peakrate.cell_log) - 1; | |
270 | if (size < max_size) | |
271 | max_size = size; | |
1da177e4 LT |
272 | } |
273 | if (max_size < 0) | |
274 | goto done; | |
275 | ||
f0cd1508 | 276 | if (q->qdisc != &noop_qdisc) { |
277 | err = fifo_set_limit(q->qdisc, qopt->limit); | |
278 | if (err) | |
279 | goto done; | |
280 | } else if (qopt->limit > 0) { | |
fb0305ce PM |
281 | child = fifo_create_dflt(sch, &bfifo_qdisc_ops, qopt->limit); |
282 | if (IS_ERR(child)) { | |
283 | err = PTR_ERR(child); | |
1da177e4 | 284 | goto done; |
fb0305ce | 285 | } |
1da177e4 LT |
286 | } |
287 | ||
288 | sch_tree_lock(sch); | |
5e50da01 PM |
289 | if (child) { |
290 | qdisc_tree_decrease_qlen(q->qdisc, q->qdisc->q.qlen); | |
b94c8afc PM |
291 | qdisc_destroy(q->qdisc); |
292 | q->qdisc = child; | |
5e50da01 | 293 | } |
1da177e4 | 294 | q->limit = qopt->limit; |
b757c933 | 295 | q->mtu = PSCHED_TICKS2NS(qopt->mtu); |
1da177e4 | 296 | q->max_size = max_size; |
b757c933 | 297 | q->buffer = PSCHED_TICKS2NS(qopt->buffer); |
1da177e4 LT |
298 | q->tokens = q->buffer; |
299 | q->ptokens = q->mtu; | |
b94c8afc | 300 | |
01cb71d2 | 301 | psched_ratecfg_precompute(&q->rate, &rtab->rate); |
b757c933 | 302 | if (ptab) { |
01cb71d2 | 303 | psched_ratecfg_precompute(&q->peak, &ptab->rate); |
b757c933 JP |
304 | q->peak_present = true; |
305 | } else { | |
306 | q->peak_present = false; | |
307 | } | |
b94c8afc | 308 | |
1da177e4 LT |
309 | sch_tree_unlock(sch); |
310 | err = 0; | |
311 | done: | |
312 | if (rtab) | |
313 | qdisc_put_rtab(rtab); | |
314 | if (ptab) | |
315 | qdisc_put_rtab(ptab); | |
316 | return err; | |
317 | } | |
318 | ||
cc7ec456 | 319 | static int tbf_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
320 | { |
321 | struct tbf_sched_data *q = qdisc_priv(sch); | |
322 | ||
323 | if (opt == NULL) | |
324 | return -EINVAL; | |
325 | ||
b757c933 | 326 | q->t_c = ktime_to_ns(ktime_get()); |
f7f593e3 | 327 | qdisc_watchdog_init(&q->watchdog, sch); |
1da177e4 LT |
328 | q->qdisc = &noop_qdisc; |
329 | ||
330 | return tbf_change(sch, opt); | |
331 | } | |
332 | ||
333 | static void tbf_destroy(struct Qdisc *sch) | |
334 | { | |
335 | struct tbf_sched_data *q = qdisc_priv(sch); | |
336 | ||
f7f593e3 | 337 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
338 | qdisc_destroy(q->qdisc); |
339 | } | |
340 | ||
341 | static int tbf_dump(struct Qdisc *sch, struct sk_buff *skb) | |
342 | { | |
343 | struct tbf_sched_data *q = qdisc_priv(sch); | |
4b3550ef | 344 | struct nlattr *nest; |
1da177e4 LT |
345 | struct tc_tbf_qopt opt; |
346 | ||
b0460e44 | 347 | sch->qstats.backlog = q->qdisc->qstats.backlog; |
4b3550ef PM |
348 | nest = nla_nest_start(skb, TCA_OPTIONS); |
349 | if (nest == NULL) | |
350 | goto nla_put_failure; | |
1da177e4 LT |
351 | |
352 | opt.limit = q->limit; | |
01cb71d2 | 353 | psched_ratecfg_getrate(&opt.rate, &q->rate); |
b757c933 | 354 | if (q->peak_present) |
01cb71d2 | 355 | psched_ratecfg_getrate(&opt.peakrate, &q->peak); |
1da177e4 LT |
356 | else |
357 | memset(&opt.peakrate, 0, sizeof(opt.peakrate)); | |
b757c933 JP |
358 | opt.mtu = PSCHED_NS2TICKS(q->mtu); |
359 | opt.buffer = PSCHED_NS2TICKS(q->buffer); | |
1b34ec43 DM |
360 | if (nla_put(skb, TCA_TBF_PARMS, sizeof(opt), &opt)) |
361 | goto nla_put_failure; | |
1da177e4 | 362 | |
4b3550ef | 363 | nla_nest_end(skb, nest); |
1da177e4 LT |
364 | return skb->len; |
365 | ||
1e90474c | 366 | nla_put_failure: |
4b3550ef | 367 | nla_nest_cancel(skb, nest); |
1da177e4 LT |
368 | return -1; |
369 | } | |
370 | ||
371 | static int tbf_dump_class(struct Qdisc *sch, unsigned long cl, | |
372 | struct sk_buff *skb, struct tcmsg *tcm) | |
373 | { | |
374 | struct tbf_sched_data *q = qdisc_priv(sch); | |
375 | ||
1da177e4 LT |
376 | tcm->tcm_handle |= TC_H_MIN(1); |
377 | tcm->tcm_info = q->qdisc->handle; | |
378 | ||
379 | return 0; | |
380 | } | |
381 | ||
382 | static int tbf_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
383 | struct Qdisc **old) | |
384 | { | |
385 | struct tbf_sched_data *q = qdisc_priv(sch); | |
386 | ||
387 | if (new == NULL) | |
388 | new = &noop_qdisc; | |
389 | ||
390 | sch_tree_lock(sch); | |
b94c8afc PM |
391 | *old = q->qdisc; |
392 | q->qdisc = new; | |
5e50da01 | 393 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); |
1da177e4 | 394 | qdisc_reset(*old); |
1da177e4 LT |
395 | sch_tree_unlock(sch); |
396 | ||
397 | return 0; | |
398 | } | |
399 | ||
400 | static struct Qdisc *tbf_leaf(struct Qdisc *sch, unsigned long arg) | |
401 | { | |
402 | struct tbf_sched_data *q = qdisc_priv(sch); | |
403 | return q->qdisc; | |
404 | } | |
405 | ||
406 | static unsigned long tbf_get(struct Qdisc *sch, u32 classid) | |
407 | { | |
408 | return 1; | |
409 | } | |
410 | ||
411 | static void tbf_put(struct Qdisc *sch, unsigned long arg) | |
412 | { | |
413 | } | |
414 | ||
1da177e4 LT |
415 | static void tbf_walk(struct Qdisc *sch, struct qdisc_walker *walker) |
416 | { | |
417 | if (!walker->stop) { | |
418 | if (walker->count >= walker->skip) | |
419 | if (walker->fn(sch, 1, walker) < 0) { | |
420 | walker->stop = 1; | |
421 | return; | |
422 | } | |
423 | walker->count++; | |
424 | } | |
425 | } | |
426 | ||
cc7ec456 | 427 | static const struct Qdisc_class_ops tbf_class_ops = { |
1da177e4 LT |
428 | .graft = tbf_graft, |
429 | .leaf = tbf_leaf, | |
430 | .get = tbf_get, | |
431 | .put = tbf_put, | |
1da177e4 | 432 | .walk = tbf_walk, |
1da177e4 LT |
433 | .dump = tbf_dump_class, |
434 | }; | |
435 | ||
20fea08b | 436 | static struct Qdisc_ops tbf_qdisc_ops __read_mostly = { |
1da177e4 LT |
437 | .next = NULL, |
438 | .cl_ops = &tbf_class_ops, | |
439 | .id = "tbf", | |
440 | .priv_size = sizeof(struct tbf_sched_data), | |
441 | .enqueue = tbf_enqueue, | |
442 | .dequeue = tbf_dequeue, | |
77be155c | 443 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
444 | .drop = tbf_drop, |
445 | .init = tbf_init, | |
446 | .reset = tbf_reset, | |
447 | .destroy = tbf_destroy, | |
448 | .change = tbf_change, | |
449 | .dump = tbf_dump, | |
450 | .owner = THIS_MODULE, | |
451 | }; | |
452 | ||
453 | static int __init tbf_module_init(void) | |
454 | { | |
455 | return register_qdisc(&tbf_qdisc_ops); | |
456 | } | |
457 | ||
458 | static void __exit tbf_module_exit(void) | |
459 | { | |
460 | unregister_qdisc(&tbf_qdisc_ops); | |
461 | } | |
462 | module_init(tbf_module_init) | |
463 | module_exit(tbf_module_exit) | |
464 | MODULE_LICENSE("GPL"); |