blob: b1c95bce33ce6f2f8a360e93ab234f8db1ab42c9 [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090015#include <linux/slab.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>
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -070021#include <net/netlink.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <net/pkt_sched.h>
23
24
25struct prio_sched_data
26{
27 int bands;
28 struct tcf_proto *filter_list;
29 u8 prio2band[TC_PRIO_MAX+1];
30 struct Qdisc *queues[TCQ_PRIO_BANDS];
31};
32
33
34static struct Qdisc *
35prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
36{
37 struct prio_sched_data *q = qdisc_priv(sch);
38 u32 band = skb->priority;
39 struct tcf_result res;
Patrick McHardybdba91e2007-07-30 17:07:14 -070040 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Jarek Poplawskic27f3392008-08-04 22:39:11 -070042 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 if (TC_H_MAJ(skb->priority) != sch->handle) {
Patrick McHardybdba91e2007-07-30 17:07:14 -070044 err = tc_classify(skb, q->filter_list, &res);
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#ifdef CONFIG_NET_CLS_ACT
Lucas Nussbaumdbaaa072007-08-30 22:35:46 -070046 switch (err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 case TC_ACT_STOLEN:
48 case TC_ACT_QUEUED:
Jarek Poplawski378a2f02008-08-04 22:31:03 -070049 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 case TC_ACT_SHOT:
51 return NULL;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -070052 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#endif
Patrick McHardybdba91e2007-07-30 17:07:14 -070054 if (!q->filter_list || err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 if (TC_H_MAJ(band))
56 band = 0;
David S. Miller1d8ae3f2008-07-15 02:52:19 -070057 return q->queues[q->prio2band[band&TC_PRIO_MAX]];
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 }
59 band = res.classid;
60 }
61 band = TC_H_MIN(band) - 1;
Jamal Hadi Salim3e5c2d32007-05-14 02:57:19 -070062 if (band >= q->bands)
David S. Miller1d8ae3f2008-07-15 02:52:19 -070063 return q->queues[q->prio2band[0]];
64
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 return q->queues[band];
66}
67
68static int
69prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
70{
71 struct Qdisc *qdisc;
72 int ret;
73
74 qdisc = prio_classify(skb, sch, &ret);
75#ifdef CONFIG_NET_CLS_ACT
76 if (qdisc == NULL) {
Jamal Hadi Salim29f1df62006-01-08 22:35:55 -080077
Jarek Poplawskic27f3392008-08-04 22:39:11 -070078 if (ret & __NET_XMIT_BYPASS)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 sch->qstats.drops++;
80 kfree_skb(skb);
81 return ret;
82 }
83#endif
84
Jussi Kivilinna5f861732008-07-20 00:08:04 -070085 ret = qdisc_enqueue(skb, qdisc);
86 if (ret == NET_XMIT_SUCCESS) {
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -070087 sch->bstats.bytes += qdisc_pkt_len(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 sch->bstats.packets++;
89 sch->q.qlen++;
90 return NET_XMIT_SUCCESS;
91 }
Jarek Poplawski378a2f02008-08-04 22:31:03 -070092 if (net_xmit_drop_count(ret))
93 sch->qstats.drops++;
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +090094 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095}
96
Patrick McHardy48a8f512008-10-31 00:44:18 -070097static struct sk_buff *prio_peek(struct Qdisc *sch)
98{
99 struct prio_sched_data *q = qdisc_priv(sch);
100 int prio;
101
102 for (prio = 0; prio < q->bands; prio++) {
103 struct Qdisc *qdisc = q->queues[prio];
104 struct sk_buff *skb = qdisc->ops->peek(qdisc);
105 if (skb)
106 return skb;
107 }
108 return NULL;
109}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700111static struct sk_buff *prio_dequeue(struct Qdisc* sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 struct prio_sched_data *q = qdisc_priv(sch);
114 int prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
116 for (prio = 0; prio < q->bands; prio++) {
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700117 struct Qdisc *qdisc = q->queues[prio];
118 struct sk_buff *skb = qdisc->dequeue(qdisc);
119 if (skb) {
120 sch->q.qlen--;
121 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 }
123 }
124 return NULL;
125
126}
127
128static unsigned int prio_drop(struct Qdisc* sch)
129{
130 struct prio_sched_data *q = qdisc_priv(sch);
131 int prio;
132 unsigned int len;
133 struct Qdisc *qdisc;
134
135 for (prio = q->bands-1; prio >= 0; prio--) {
136 qdisc = q->queues[prio];
Patrick McHardy6d037a22006-03-20 19:00:49 -0800137 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 sch->q.qlen--;
139 return len;
140 }
141 }
142 return 0;
143}
144
145
146static void
147prio_reset(struct Qdisc* sch)
148{
149 int prio;
150 struct prio_sched_data *q = qdisc_priv(sch);
151
152 for (prio=0; prio<q->bands; prio++)
153 qdisc_reset(q->queues[prio]);
154 sch->q.qlen = 0;
155}
156
157static void
158prio_destroy(struct Qdisc* sch)
159{
160 int prio;
161 struct prio_sched_data *q = qdisc_priv(sch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Patrick McHardyff31ab52008-07-01 19:52:38 -0700163 tcf_destroy_chain(&q->filter_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 for (prio=0; prio<q->bands; prio++)
165 qdisc_destroy(q->queues[prio]);
166}
167
Patrick McHardy1e904742008-01-22 22:11:17 -0800168static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169{
170 struct prio_sched_data *q = qdisc_priv(sch);
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700171 struct tc_prio_qopt *qopt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 int i;
173
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700174 if (nla_len(opt) < sizeof(*qopt))
175 return -EINVAL;
176 qopt = nla_data(opt);
Patrick McHardycee63722008-01-23 20:33:32 -0800177
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700178 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 return -EINVAL;
180
181 for (i=0; i<=TC_PRIO_MAX; i++) {
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700182 if (qopt->priomap[i] >= qopt->bands)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 return -EINVAL;
184 }
185
186 sch_tree_lock(sch);
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700187 q->bands = qopt->bands;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
189
190 for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800191 struct Qdisc *child = q->queues[i];
192 q->queues[i] = &noop_qdisc;
Patrick McHardy5e50da02006-11-29 17:36:20 -0800193 if (child != &noop_qdisc) {
194 qdisc_tree_decrease_qlen(child, child->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 qdisc_destroy(child);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800196 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 }
198 sch_tree_unlock(sch);
199
Amnon Aaronsohndd914b42006-01-17 02:24:26 -0800200 for (i=0; i<q->bands; i++) {
201 if (q->queues[i] == &noop_qdisc) {
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800202 struct Qdisc *child, *old;
Changli Gao3511c912010-10-16 13:04:08 +0000203 child = qdisc_create_dflt(sch->dev_queue,
David S. Millerbb949fb2008-07-08 16:55:56 -0700204 &pfifo_qdisc_ops,
Patrick McHardy9f9afec2006-11-29 17:35:18 -0800205 TC_H_MAKE(sch->handle, i + 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 if (child) {
207 sch_tree_lock(sch);
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800208 old = q->queues[i];
209 q->queues[i] = child;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800211 if (old != &noop_qdisc) {
212 qdisc_tree_decrease_qlen(old,
213 old->q.qlen);
214 qdisc_destroy(old);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800215 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 sch_tree_unlock(sch);
217 }
218 }
219 }
220 return 0;
221}
222
Patrick McHardy1e904742008-01-22 22:11:17 -0800223static int prio_init(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224{
225 struct prio_sched_data *q = qdisc_priv(sch);
226 int i;
227
228 for (i=0; i<TCQ_PRIO_BANDS; i++)
229 q->queues[i] = &noop_qdisc;
230
231 if (opt == NULL) {
232 return -EINVAL;
233 } else {
234 int err;
235
236 if ((err= prio_tune(sch, opt)) != 0)
237 return err;
238 }
239 return 0;
240}
241
242static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
243{
244 struct prio_sched_data *q = qdisc_priv(sch);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700245 unsigned char *b = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 struct tc_prio_qopt opt;
247
248 opt.bands = q->bands;
249 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700250
Thomas Graf2c10b322008-09-02 17:30:27 -0700251 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700252
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 return skb->len;
254
Patrick McHardy1e904742008-01-22 22:11:17 -0800255nla_put_failure:
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -0700256 nlmsg_trim(skb, b);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 return -1;
258}
259
260static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
261 struct Qdisc **old)
262{
263 struct prio_sched_data *q = qdisc_priv(sch);
264 unsigned long band = arg - 1;
265
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 if (new == NULL)
267 new = &noop_qdisc;
268
269 sch_tree_lock(sch);
270 *old = q->queues[band];
271 q->queues[band] = new;
Patrick McHardy5e50da02006-11-29 17:36:20 -0800272 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 qdisc_reset(*old);
274 sch_tree_unlock(sch);
275
276 return 0;
277}
278
279static struct Qdisc *
280prio_leaf(struct Qdisc *sch, unsigned long arg)
281{
282 struct prio_sched_data *q = qdisc_priv(sch);
283 unsigned long band = arg - 1;
284
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 return q->queues[band];
286}
287
288static unsigned long prio_get(struct Qdisc *sch, u32 classid)
289{
290 struct prio_sched_data *q = qdisc_priv(sch);
291 unsigned long band = TC_H_MIN(classid);
292
293 if (band - 1 >= q->bands)
294 return 0;
295 return band;
296}
297
298static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
299{
300 return prio_get(sch, classid);
301}
302
303
304static void prio_put(struct Qdisc *q, unsigned long cl)
305{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306}
307
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
309 struct tcmsg *tcm)
310{
311 struct prio_sched_data *q = qdisc_priv(sch);
312
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 tcm->tcm_handle |= TC_H_MIN(cl);
Patrick McHardy5b9a9cc2009-09-04 06:41:17 +0000314 tcm->tcm_info = q->queues[cl-1]->handle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 return 0;
316}
317
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800318static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
319 struct gnet_dump *d)
320{
321 struct prio_sched_data *q = qdisc_priv(sch);
322 struct Qdisc *cl_q;
323
324 cl_q = q->queues[cl - 1];
Jarek Poplawskia19d2152009-09-17 10:26:07 -0700325 cl_q->qstats.qlen = cl_q->q.qlen;
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800326 if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 ||
327 gnet_stats_copy_queue(d, &cl_q->qstats) < 0)
328 return -1;
329
330 return 0;
331}
332
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
334{
335 struct prio_sched_data *q = qdisc_priv(sch);
336 int prio;
337
338 if (arg->stop)
339 return;
340
341 for (prio = 0; prio < q->bands; prio++) {
342 if (arg->count < arg->skip) {
343 arg->count++;
344 continue;
345 }
346 if (arg->fn(sch, prio+1, arg) < 0) {
347 arg->stop = 1;
348 break;
349 }
350 arg->count++;
351 }
352}
353
354static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
355{
356 struct prio_sched_data *q = qdisc_priv(sch);
357
358 if (cl)
359 return NULL;
360 return &q->filter_list;
361}
362
Eric Dumazet20fea082007-11-14 01:44:41 -0800363static const struct Qdisc_class_ops prio_class_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 .graft = prio_graft,
365 .leaf = prio_leaf,
366 .get = prio_get,
367 .put = prio_put,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 .walk = prio_walk,
369 .tcf_chain = prio_find_tcf,
370 .bind_tcf = prio_bind,
371 .unbind_tcf = prio_put,
372 .dump = prio_dump_class,
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800373 .dump_stats = prio_dump_class_stats,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374};
375
Eric Dumazet20fea082007-11-14 01:44:41 -0800376static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 .next = NULL,
378 .cl_ops = &prio_class_ops,
379 .id = "prio",
380 .priv_size = sizeof(struct prio_sched_data),
381 .enqueue = prio_enqueue,
382 .dequeue = prio_dequeue,
Patrick McHardy48a8f512008-10-31 00:44:18 -0700383 .peek = prio_peek,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 .drop = prio_drop,
385 .init = prio_init,
386 .reset = prio_reset,
387 .destroy = prio_destroy,
388 .change = prio_tune,
389 .dump = prio_dump,
390 .owner = THIS_MODULE,
391};
392
393static int __init prio_module_init(void)
394{
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700395 return register_qdisc(&prio_qdisc_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396}
397
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900398static void __exit prio_module_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399{
400 unregister_qdisc(&prio_qdisc_ops);
401}
402
403module_init(prio_module_init)
404module_exit(prio_module_exit)
405
406MODULE_LICENSE("GPL");