blob: 79359b69ad8d6fcf35b6937673d4ae332bbf5158 [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
Eric Dumazetcc7ec452011-01-19 19:26:56 +000025struct prio_sched_data {
Linus Torvalds1da177e2005-04-16 15:20:36 -070026 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;
Eric Dumazetcc7ec452011-01-19 19:26:56 +000056 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) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 sch->q.qlen++;
87 return NET_XMIT_SUCCESS;
88 }
Jarek Poplawski378a2f02008-08-04 22:31:03 -070089 if (net_xmit_drop_count(ret))
90 sch->qstats.drops++;
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +090091 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092}
93
Patrick McHardy48a8f512008-10-31 00:44:18 -070094static struct sk_buff *prio_peek(struct Qdisc *sch)
95{
96 struct prio_sched_data *q = qdisc_priv(sch);
97 int prio;
98
99 for (prio = 0; prio < q->bands; prio++) {
100 struct Qdisc *qdisc = q->queues[prio];
101 struct sk_buff *skb = qdisc->ops->peek(qdisc);
102 if (skb)
103 return skb;
104 }
105 return NULL;
106}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000108static struct sk_buff *prio_dequeue(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 struct prio_sched_data *q = qdisc_priv(sch);
111 int prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
113 for (prio = 0; prio < q->bands; prio++) {
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700114 struct Qdisc *qdisc = q->queues[prio];
Florian Westphal35576192011-08-09 02:04:43 +0000115 struct sk_buff *skb = qdisc_dequeue_peeked(qdisc);
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700116 if (skb) {
Eric Dumazet9190b3b2011-01-20 23:31:33 -0800117 qdisc_bstats_update(sch, skb);
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700118 sch->q.qlen--;
119 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 }
121 }
122 return NULL;
123
124}
125
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000126static unsigned int prio_drop(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127{
128 struct prio_sched_data *q = qdisc_priv(sch);
129 int prio;
130 unsigned int len;
131 struct Qdisc *qdisc;
132
133 for (prio = q->bands-1; prio >= 0; prio--) {
134 qdisc = q->queues[prio];
Patrick McHardy6d037a22006-03-20 19:00:49 -0800135 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 sch->q.qlen--;
137 return len;
138 }
139 }
140 return 0;
141}
142
143
144static void
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000145prio_reset(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
147 int prio;
148 struct prio_sched_data *q = qdisc_priv(sch);
149
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000150 for (prio = 0; prio < q->bands; prio++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 qdisc_reset(q->queues[prio]);
152 sch->q.qlen = 0;
153}
154
155static void
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000156prio_destroy(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157{
158 int prio;
159 struct prio_sched_data *q = qdisc_priv(sch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160
Patrick McHardyff31ab52008-07-01 19:52:38 -0700161 tcf_destroy_chain(&q->filter_list);
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000162 for (prio = 0; prio < q->bands; prio++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 qdisc_destroy(q->queues[prio]);
164}
165
Patrick McHardy1e904742008-01-22 22:11:17 -0800166static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167{
168 struct prio_sched_data *q = qdisc_priv(sch);
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700169 struct tc_prio_qopt *qopt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 int i;
171
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700172 if (nla_len(opt) < sizeof(*qopt))
173 return -EINVAL;
174 qopt = nla_data(opt);
Patrick McHardycee63722008-01-23 20:33:32 -0800175
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700176 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 return -EINVAL;
178
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000179 for (i = 0; i <= TC_PRIO_MAX; i++) {
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700180 if (qopt->priomap[i] >= qopt->bands)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 return -EINVAL;
182 }
183
184 sch_tree_lock(sch);
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700185 q->bands = qopt->bands;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
187
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000188 for (i = q->bands; i < TCQ_PRIO_BANDS; i++) {
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800189 struct Qdisc *child = q->queues[i];
190 q->queues[i] = &noop_qdisc;
Patrick McHardy5e50da02006-11-29 17:36:20 -0800191 if (child != &noop_qdisc) {
192 qdisc_tree_decrease_qlen(child, child->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 qdisc_destroy(child);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800194 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 }
196 sch_tree_unlock(sch);
197
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000198 for (i = 0; i < q->bands; i++) {
Amnon Aaronsohndd914b42006-01-17 02:24:26 -0800199 if (q->queues[i] == &noop_qdisc) {
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800200 struct Qdisc *child, *old;
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000201
Changli Gao3511c912010-10-16 13:04:08 +0000202 child = qdisc_create_dflt(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
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000227 for (i = 0; i < TCQ_PRIO_BANDS; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 q->queues[i] = &noop_qdisc;
229
230 if (opt == NULL) {
231 return -EINVAL;
232 } else {
233 int err;
234
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000235 if ((err = prio_tune(sch, opt)) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 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;
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000248 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1);
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700249
David S. Miller1b34ec42012-03-29 05:11:39 -0400250 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
251 goto nla_put_failure;
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 }
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000346 if (arg->fn(sch, prio + 1, arg) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 arg->stop = 1;
348 break;
349 }
350 arg->count++;
351 }
352}
353
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000354static struct tcf_proto **prio_find_tcf(struct Qdisc *sch, unsigned long cl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355{
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");