blob: 94cecef7014517b5784bdc297c36a8fad644ca28 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * net/sched/sch_prio.c Simple 3-band priority "scheduler".
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>
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +090010 * Fixes: 19990609: J Hadi Salim <hadi@nortelnetworks.com>:
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Init -- EINVAL when opt undefined
12 */
13
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/types.h>
16#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/errno.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/skbuff.h>
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -070020#include <net/netlink.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <net/pkt_sched.h>
22
23
24struct prio_sched_data
25{
26 int bands;
27 struct tcf_proto *filter_list;
28 u8 prio2band[TC_PRIO_MAX+1];
29 struct Qdisc *queues[TCQ_PRIO_BANDS];
30};
31
32
33static struct Qdisc *
34prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
35{
36 struct prio_sched_data *q = qdisc_priv(sch);
37 u32 band = skb->priority;
38 struct tcf_result res;
Patrick McHardybdba91e2007-07-30 17:07:14 -070039 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
Jarek Poplawskic27f3392008-08-04 22:39:11 -070041 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 if (TC_H_MAJ(skb->priority) != sch->handle) {
Patrick McHardybdba91e2007-07-30 17:07:14 -070043 err = tc_classify(skb, q->filter_list, &res);
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#ifdef CONFIG_NET_CLS_ACT
Lucas Nussbaumdbaaa072007-08-30 22:35:46 -070045 switch (err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 case TC_ACT_STOLEN:
47 case TC_ACT_QUEUED:
Jarek Poplawski378a2f02008-08-04 22:31:03 -070048 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 case TC_ACT_SHOT:
50 return NULL;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -070051 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#endif
Patrick McHardybdba91e2007-07-30 17:07:14 -070053 if (!q->filter_list || err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 if (TC_H_MAJ(band))
55 band = 0;
David S. Miller1d8ae3f2008-07-15 02:52:19 -070056 return q->queues[q->prio2band[band&TC_PRIO_MAX]];
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 }
58 band = res.classid;
59 }
60 band = TC_H_MIN(band) - 1;
Jamal Hadi Salim3e5c2d32007-05-14 02:57:19 -070061 if (band >= q->bands)
David S. Miller1d8ae3f2008-07-15 02:52:19 -070062 return q->queues[q->prio2band[0]];
63
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 return q->queues[band];
65}
66
67static int
68prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
69{
70 struct Qdisc *qdisc;
71 int ret;
72
73 qdisc = prio_classify(skb, sch, &ret);
74#ifdef CONFIG_NET_CLS_ACT
75 if (qdisc == NULL) {
Jamal Hadi Salim29f1df62006-01-08 22:35:55 -080076
Jarek Poplawskic27f3392008-08-04 22:39:11 -070077 if (ret & __NET_XMIT_BYPASS)
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 sch->qstats.drops++;
79 kfree_skb(skb);
80 return ret;
81 }
82#endif
83
Jussi Kivilinna5f861732008-07-20 00:08:04 -070084 ret = qdisc_enqueue(skb, qdisc);
85 if (ret == NET_XMIT_SUCCESS) {
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -070086 sch->bstats.bytes += qdisc_pkt_len(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 sch->bstats.packets++;
88 sch->q.qlen++;
89 return NET_XMIT_SUCCESS;
90 }
Jarek Poplawski378a2f02008-08-04 22:31:03 -070091 if (net_xmit_drop_count(ret))
92 sch->qstats.drops++;
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +090093 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -070094}
95
Patrick McHardy48a8f512008-10-31 00:44:18 -070096static struct sk_buff *prio_peek(struct Qdisc *sch)
97{
98 struct prio_sched_data *q = qdisc_priv(sch);
99 int prio;
100
101 for (prio = 0; prio < q->bands; prio++) {
102 struct Qdisc *qdisc = q->queues[prio];
103 struct sk_buff *skb = qdisc->ops->peek(qdisc);
104 if (skb)
105 return skb;
106 }
107 return NULL;
108}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700110static struct sk_buff *prio_dequeue(struct Qdisc* sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 struct prio_sched_data *q = qdisc_priv(sch);
113 int prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114
115 for (prio = 0; prio < q->bands; prio++) {
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700116 struct Qdisc *qdisc = q->queues[prio];
117 struct sk_buff *skb = qdisc->dequeue(qdisc);
118 if (skb) {
119 sch->q.qlen--;
120 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 }
122 }
123 return NULL;
124
125}
126
127static unsigned int prio_drop(struct Qdisc* sch)
128{
129 struct prio_sched_data *q = qdisc_priv(sch);
130 int prio;
131 unsigned int len;
132 struct Qdisc *qdisc;
133
134 for (prio = q->bands-1; prio >= 0; prio--) {
135 qdisc = q->queues[prio];
Patrick McHardy6d037a22006-03-20 19:00:49 -0800136 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 sch->q.qlen--;
138 return len;
139 }
140 }
141 return 0;
142}
143
144
145static void
146prio_reset(struct Qdisc* sch)
147{
148 int prio;
149 struct prio_sched_data *q = qdisc_priv(sch);
150
151 for (prio=0; prio<q->bands; prio++)
152 qdisc_reset(q->queues[prio]);
153 sch->q.qlen = 0;
154}
155
156static void
157prio_destroy(struct Qdisc* sch)
158{
159 int prio;
160 struct prio_sched_data *q = qdisc_priv(sch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
Patrick McHardyff31ab52008-07-01 19:52:38 -0700162 tcf_destroy_chain(&q->filter_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 for (prio=0; prio<q->bands; prio++)
164 qdisc_destroy(q->queues[prio]);
165}
166
Patrick McHardy1e904742008-01-22 22:11:17 -0800167static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168{
169 struct prio_sched_data *q = qdisc_priv(sch);
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700170 struct tc_prio_qopt *qopt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 int i;
172
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700173 if (nla_len(opt) < sizeof(*qopt))
174 return -EINVAL;
175 qopt = nla_data(opt);
Patrick McHardycee63722008-01-23 20:33:32 -0800176
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700177 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 return -EINVAL;
179
180 for (i=0; i<=TC_PRIO_MAX; i++) {
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700181 if (qopt->priomap[i] >= qopt->bands)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 return -EINVAL;
183 }
184
185 sch_tree_lock(sch);
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700186 q->bands = qopt->bands;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
188
189 for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800190 struct Qdisc *child = q->queues[i];
191 q->queues[i] = &noop_qdisc;
Patrick McHardy5e50da02006-11-29 17:36:20 -0800192 if (child != &noop_qdisc) {
193 qdisc_tree_decrease_qlen(child, child->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 qdisc_destroy(child);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800195 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 }
197 sch_tree_unlock(sch);
198
Amnon Aaronsohndd914b42006-01-17 02:24:26 -0800199 for (i=0; i<q->bands; i++) {
200 if (q->queues[i] == &noop_qdisc) {
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800201 struct Qdisc *child, *old;
David S. Miller5ce2d482008-07-08 17:06:30 -0700202 child = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
David S. Millerbb949fb2008-07-08 16:55:56 -0700203 &pfifo_qdisc_ops,
Patrick McHardy9f9afec2006-11-29 17:35:18 -0800204 TC_H_MAKE(sch->handle, i + 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 if (child) {
206 sch_tree_lock(sch);
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800207 old = q->queues[i];
208 q->queues[i] = child;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800210 if (old != &noop_qdisc) {
211 qdisc_tree_decrease_qlen(old,
212 old->q.qlen);
213 qdisc_destroy(old);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800214 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 sch_tree_unlock(sch);
216 }
217 }
218 }
219 return 0;
220}
221
Patrick McHardy1e904742008-01-22 22:11:17 -0800222static int prio_init(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223{
224 struct prio_sched_data *q = qdisc_priv(sch);
225 int i;
226
227 for (i=0; i<TCQ_PRIO_BANDS; i++)
228 q->queues[i] = &noop_qdisc;
229
230 if (opt == NULL) {
231 return -EINVAL;
232 } else {
233 int err;
234
235 if ((err= prio_tune(sch, opt)) != 0)
236 return err;
237 }
238 return 0;
239}
240
241static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
242{
243 struct prio_sched_data *q = qdisc_priv(sch);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700244 unsigned char *b = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 struct tc_prio_qopt opt;
246
247 opt.bands = q->bands;
248 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700249
Thomas Graf2c10b322008-09-02 17:30:27 -0700250 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700251
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 return skb->len;
253
Patrick McHardy1e904742008-01-22 22:11:17 -0800254nla_put_failure:
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -0700255 nlmsg_trim(skb, b);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 return -1;
257}
258
259static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
260 struct Qdisc **old)
261{
262 struct prio_sched_data *q = qdisc_priv(sch);
263 unsigned long band = arg - 1;
264
265 if (band >= q->bands)
266 return -EINVAL;
267
268 if (new == NULL)
269 new = &noop_qdisc;
270
271 sch_tree_lock(sch);
272 *old = q->queues[band];
273 q->queues[band] = new;
Patrick McHardy5e50da02006-11-29 17:36:20 -0800274 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 qdisc_reset(*old);
276 sch_tree_unlock(sch);
277
278 return 0;
279}
280
281static struct Qdisc *
282prio_leaf(struct Qdisc *sch, unsigned long arg)
283{
284 struct prio_sched_data *q = qdisc_priv(sch);
285 unsigned long band = arg - 1;
286
287 if (band >= q->bands)
288 return NULL;
289
290 return q->queues[band];
291}
292
293static unsigned long prio_get(struct Qdisc *sch, u32 classid)
294{
295 struct prio_sched_data *q = qdisc_priv(sch);
296 unsigned long band = TC_H_MIN(classid);
297
298 if (band - 1 >= q->bands)
299 return 0;
300 return band;
301}
302
303static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
304{
305 return prio_get(sch, classid);
306}
307
308
309static void prio_put(struct Qdisc *q, unsigned long cl)
310{
311 return;
312}
313
Patrick McHardy1e904742008-01-22 22:11:17 -0800314static int prio_change(struct Qdisc *sch, u32 handle, u32 parent, struct nlattr **tca, unsigned long *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
316 unsigned long cl = *arg;
317 struct prio_sched_data *q = qdisc_priv(sch);
318
319 if (cl - 1 > q->bands)
320 return -ENOENT;
321 return 0;
322}
323
324static int prio_delete(struct Qdisc *sch, unsigned long cl)
325{
326 struct prio_sched_data *q = qdisc_priv(sch);
327 if (cl - 1 > q->bands)
328 return -ENOENT;
329 return 0;
330}
331
332
333static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
334 struct tcmsg *tcm)
335{
336 struct prio_sched_data *q = qdisc_priv(sch);
337
338 if (cl - 1 > q->bands)
339 return -ENOENT;
340 tcm->tcm_handle |= TC_H_MIN(cl);
341 if (q->queues[cl-1])
342 tcm->tcm_info = q->queues[cl-1]->handle;
343 return 0;
344}
345
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800346static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
347 struct gnet_dump *d)
348{
349 struct prio_sched_data *q = qdisc_priv(sch);
350 struct Qdisc *cl_q;
351
352 cl_q = q->queues[cl - 1];
353 if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 ||
354 gnet_stats_copy_queue(d, &cl_q->qstats) < 0)
355 return -1;
356
357 return 0;
358}
359
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
361{
362 struct prio_sched_data *q = qdisc_priv(sch);
363 int prio;
364
365 if (arg->stop)
366 return;
367
368 for (prio = 0; prio < q->bands; prio++) {
369 if (arg->count < arg->skip) {
370 arg->count++;
371 continue;
372 }
373 if (arg->fn(sch, prio+1, arg) < 0) {
374 arg->stop = 1;
375 break;
376 }
377 arg->count++;
378 }
379}
380
381static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
382{
383 struct prio_sched_data *q = qdisc_priv(sch);
384
385 if (cl)
386 return NULL;
387 return &q->filter_list;
388}
389
Eric Dumazet20fea082007-11-14 01:44:41 -0800390static const struct Qdisc_class_ops prio_class_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 .graft = prio_graft,
392 .leaf = prio_leaf,
393 .get = prio_get,
394 .put = prio_put,
395 .change = prio_change,
396 .delete = prio_delete,
397 .walk = prio_walk,
398 .tcf_chain = prio_find_tcf,
399 .bind_tcf = prio_bind,
400 .unbind_tcf = prio_put,
401 .dump = prio_dump_class,
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800402 .dump_stats = prio_dump_class_stats,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403};
404
Eric Dumazet20fea082007-11-14 01:44:41 -0800405static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 .next = NULL,
407 .cl_ops = &prio_class_ops,
408 .id = "prio",
409 .priv_size = sizeof(struct prio_sched_data),
410 .enqueue = prio_enqueue,
411 .dequeue = prio_dequeue,
Patrick McHardy48a8f512008-10-31 00:44:18 -0700412 .peek = prio_peek,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 .drop = prio_drop,
414 .init = prio_init,
415 .reset = prio_reset,
416 .destroy = prio_destroy,
417 .change = prio_tune,
418 .dump = prio_dump,
419 .owner = THIS_MODULE,
420};
421
422static int __init prio_module_init(void)
423{
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700424 return register_qdisc(&prio_qdisc_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425}
426
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900427static void __exit prio_module_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428{
429 unregister_qdisc(&prio_qdisc_ops);
430}
431
432module_init(prio_module_init)
433module_exit(prio_module_exit)
434
435MODULE_LICENSE("GPL");