blob: ba6487f2741f9fb20e55d39eeab4e7448b76cdac [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;
John Fastabend25d8c0d2014-09-12 20:05:27 -070027 struct tcf_proto __rcu *filter_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -070028 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;
John Fastabend25d8c0d2014-09-12 20:05:27 -070039 struct tcf_proto *fl;
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) {
John Fastabend25d8c0d2014-09-12 20:05:27 -070044 fl = rcu_dereference_bh(q->filter_list);
Daniel Borkmann3b3ae882015-08-26 23:00:06 +020045 err = tc_classify(skb, fl, &res, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#ifdef CONFIG_NET_CLS_ACT
Lucas Nussbaumdbaaa072007-08-30 22:35:46 -070047 switch (err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070048 case TC_ACT_STOLEN:
49 case TC_ACT_QUEUED:
Jarek Poplawski378a2f02008-08-04 22:31:03 -070050 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051 case TC_ACT_SHOT:
52 return NULL;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -070053 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#endif
John Fastabend25d8c0d2014-09-12 20:05:27 -070055 if (!fl || err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 if (TC_H_MAJ(band))
57 band = 0;
Eric Dumazetcc7ec452011-01-19 19:26:56 +000058 return q->queues[q->prio2band[band & TC_PRIO_MAX]];
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 }
60 band = res.classid;
61 }
62 band = TC_H_MIN(band) - 1;
Jamal Hadi Salim3e5c2d32007-05-14 02:57:19 -070063 if (band >= q->bands)
David S. Miller1d8ae3f2008-07-15 02:52:19 -070064 return q->queues[q->prio2band[0]];
65
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 return q->queues[band];
67}
68
69static int
70prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
71{
72 struct Qdisc *qdisc;
73 int ret;
74
75 qdisc = prio_classify(skb, sch, &ret);
76#ifdef CONFIG_NET_CLS_ACT
77 if (qdisc == NULL) {
Jamal Hadi Salim29f1df62006-01-08 22:35:55 -080078
Jarek Poplawskic27f3392008-08-04 22:39:11 -070079 if (ret & __NET_XMIT_BYPASS)
John Fastabend25331d62014-09-28 11:53:29 -070080 qdisc_qstats_drop(sch);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 kfree_skb(skb);
82 return ret;
83 }
84#endif
85
Jussi Kivilinna5f861732008-07-20 00:08:04 -070086 ret = qdisc_enqueue(skb, qdisc);
87 if (ret == NET_XMIT_SUCCESS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 sch->q.qlen++;
89 return NET_XMIT_SUCCESS;
90 }
Jarek Poplawski378a2f02008-08-04 22:31:03 -070091 if (net_xmit_drop_count(ret))
John Fastabend25331d62014-09-28 11:53:29 -070092 qdisc_qstats_drop(sch);
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
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000110static 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];
Florian Westphal35576192011-08-09 02:04:43 +0000117 struct sk_buff *skb = qdisc_dequeue_peeked(qdisc);
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700118 if (skb) {
Eric Dumazet9190b3b2011-01-20 23:31:33 -0800119 qdisc_bstats_update(sch, skb);
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700120 sch->q.qlen--;
121 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 }
123 }
124 return NULL;
125
126}
127
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000128static unsigned int prio_drop(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129{
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
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000147prio_reset(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148{
149 int prio;
150 struct prio_sched_data *q = qdisc_priv(sch);
151
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000152 for (prio = 0; prio < q->bands; prio++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 qdisc_reset(q->queues[prio]);
154 sch->q.qlen = 0;
155}
156
157static void
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000158prio_destroy(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159{
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);
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000164 for (prio = 0; prio < q->bands; prio++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 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
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000181 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
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000190 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
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000200 for (i = 0; i < q->bands; i++) {
Amnon Aaronsohndd914b42006-01-17 02:24:26 -0800201 if (q->queues[i] == &noop_qdisc) {
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800202 struct Qdisc *child, *old;
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000203
Changli Gao3511c912010-10-16 13:04:08 +0000204 child = qdisc_create_dflt(sch->dev_queue,
David S. Millerbb949fb2008-07-08 16:55:56 -0700205 &pfifo_qdisc_ops,
Patrick McHardy9f9afec2006-11-29 17:35:18 -0800206 TC_H_MAKE(sch->handle, i + 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 if (child) {
208 sch_tree_lock(sch);
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800209 old = q->queues[i];
210 q->queues[i] = child;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800212 if (old != &noop_qdisc) {
213 qdisc_tree_decrease_qlen(old,
214 old->q.qlen);
215 qdisc_destroy(old);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800216 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 sch_tree_unlock(sch);
218 }
219 }
220 }
221 return 0;
222}
223
Patrick McHardy1e904742008-01-22 22:11:17 -0800224static int prio_init(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
226 struct prio_sched_data *q = qdisc_priv(sch);
227 int i;
228
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000229 for (i = 0; i < TCQ_PRIO_BANDS; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 q->queues[i] = &noop_qdisc;
231
232 if (opt == NULL) {
233 return -EINVAL;
234 } else {
235 int err;
236
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000237 if ((err = prio_tune(sch, opt)) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 return err;
239 }
240 return 0;
241}
242
243static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
244{
245 struct prio_sched_data *q = qdisc_priv(sch);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700246 unsigned char *b = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 struct tc_prio_qopt opt;
248
249 opt.bands = q->bands;
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000250 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1);
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700251
David S. Miller1b34ec42012-03-29 05:11:39 -0400252 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
253 goto nla_put_failure;
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700254
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 return skb->len;
256
Patrick McHardy1e904742008-01-22 22:11:17 -0800257nla_put_failure:
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -0700258 nlmsg_trim(skb, b);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 return -1;
260}
261
262static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
263 struct Qdisc **old)
264{
265 struct prio_sched_data *q = qdisc_priv(sch);
266 unsigned long band = arg - 1;
267
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return q->queues[band];
288}
289
290static unsigned long prio_get(struct Qdisc *sch, u32 classid)
291{
292 struct prio_sched_data *q = qdisc_priv(sch);
293 unsigned long band = TC_H_MIN(classid);
294
295 if (band - 1 >= q->bands)
296 return 0;
297 return band;
298}
299
300static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
301{
302 return prio_get(sch, classid);
303}
304
305
306static void prio_put(struct Qdisc *q, unsigned long cl)
307{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308}
309
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
311 struct tcmsg *tcm)
312{
313 struct prio_sched_data *q = qdisc_priv(sch);
314
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 tcm->tcm_handle |= TC_H_MIN(cl);
Patrick McHardy5b9a9cc2009-09-04 06:41:17 +0000316 tcm->tcm_info = q->queues[cl-1]->handle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 return 0;
318}
319
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800320static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
321 struct gnet_dump *d)
322{
323 struct prio_sched_data *q = qdisc_priv(sch);
324 struct Qdisc *cl_q;
325
326 cl_q = q->queues[cl - 1];
John Fastabend22e0f8b2014-09-28 11:52:56 -0700327 if (gnet_stats_copy_basic(d, NULL, &cl_q->bstats) < 0 ||
John Fastabendb0ab6f92014-09-28 11:54:24 -0700328 gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0)
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800329 return -1;
330
331 return 0;
332}
333
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
335{
336 struct prio_sched_data *q = qdisc_priv(sch);
337 int prio;
338
339 if (arg->stop)
340 return;
341
342 for (prio = 0; prio < q->bands; prio++) {
343 if (arg->count < arg->skip) {
344 arg->count++;
345 continue;
346 }
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000347 if (arg->fn(sch, prio + 1, arg) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 arg->stop = 1;
349 break;
350 }
351 arg->count++;
352 }
353}
354
John Fastabend25d8c0d2014-09-12 20:05:27 -0700355static struct tcf_proto __rcu **prio_find_tcf(struct Qdisc *sch,
356 unsigned long cl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357{
358 struct prio_sched_data *q = qdisc_priv(sch);
359
360 if (cl)
361 return NULL;
362 return &q->filter_list;
363}
364
Eric Dumazet20fea082007-11-14 01:44:41 -0800365static const struct Qdisc_class_ops prio_class_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 .graft = prio_graft,
367 .leaf = prio_leaf,
368 .get = prio_get,
369 .put = prio_put,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 .walk = prio_walk,
371 .tcf_chain = prio_find_tcf,
372 .bind_tcf = prio_bind,
373 .unbind_tcf = prio_put,
374 .dump = prio_dump_class,
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800375 .dump_stats = prio_dump_class_stats,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376};
377
Eric Dumazet20fea082007-11-14 01:44:41 -0800378static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 .next = NULL,
380 .cl_ops = &prio_class_ops,
381 .id = "prio",
382 .priv_size = sizeof(struct prio_sched_data),
383 .enqueue = prio_enqueue,
384 .dequeue = prio_dequeue,
Patrick McHardy48a8f512008-10-31 00:44:18 -0700385 .peek = prio_peek,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 .drop = prio_drop,
387 .init = prio_init,
388 .reset = prio_reset,
389 .destroy = prio_destroy,
390 .change = prio_tune,
391 .dump = prio_dump,
392 .owner = THIS_MODULE,
393};
394
395static int __init prio_module_init(void)
396{
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700397 return register_qdisc(&prio_qdisc_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398}
399
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900400static void __exit prio_module_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401{
402 unregister_qdisc(&prio_qdisc_ops);
403}
404
405module_init(prio_module_init)
406module_exit(prio_module_exit)
407
408MODULE_LICENSE("GPL");