blob: 3651da3e280211fddfa167dbe595283af399ac29 [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
96
97static int
98prio_requeue(struct sk_buff *skb, struct Qdisc* sch)
99{
100 struct Qdisc *qdisc;
101 int ret;
102
103 qdisc = prio_classify(skb, sch, &ret);
104#ifdef CONFIG_NET_CLS_ACT
105 if (qdisc == NULL) {
Jarek Poplawskic27f3392008-08-04 22:39:11 -0700106 if (ret & __NET_XMIT_BYPASS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 sch->qstats.drops++;
108 kfree_skb(skb);
109 return ret;
110 }
111#endif
112
113 if ((ret = qdisc->ops->requeue(skb, qdisc)) == NET_XMIT_SUCCESS) {
114 sch->q.qlen++;
115 sch->qstats.requeues++;
David S. Miller4cf7cb22008-08-17 22:45:17 -0700116 return NET_XMIT_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 }
Jarek Poplawski378a2f02008-08-04 22:31:03 -0700118 if (net_xmit_drop_count(ret))
119 sch->qstats.drops++;
Jussi Kivilinna0d40b6e2008-08-17 22:43:56 -0700120 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121}
122
Patrick McHardy48a8f512008-10-31 00:44:18 -0700123static struct sk_buff *prio_peek(struct Qdisc *sch)
124{
125 struct prio_sched_data *q = qdisc_priv(sch);
126 int prio;
127
128 for (prio = 0; prio < q->bands; prio++) {
129 struct Qdisc *qdisc = q->queues[prio];
130 struct sk_buff *skb = qdisc->ops->peek(qdisc);
131 if (skb)
132 return skb;
133 }
134 return NULL;
135}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700137static struct sk_buff *prio_dequeue(struct Qdisc* sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 struct prio_sched_data *q = qdisc_priv(sch);
140 int prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
142 for (prio = 0; prio < q->bands; prio++) {
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700143 struct Qdisc *qdisc = q->queues[prio];
144 struct sk_buff *skb = qdisc->dequeue(qdisc);
145 if (skb) {
146 sch->q.qlen--;
147 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 }
149 }
150 return NULL;
151
152}
153
154static unsigned int prio_drop(struct Qdisc* sch)
155{
156 struct prio_sched_data *q = qdisc_priv(sch);
157 int prio;
158 unsigned int len;
159 struct Qdisc *qdisc;
160
161 for (prio = q->bands-1; prio >= 0; prio--) {
162 qdisc = q->queues[prio];
Patrick McHardy6d037a22006-03-20 19:00:49 -0800163 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 sch->q.qlen--;
165 return len;
166 }
167 }
168 return 0;
169}
170
171
172static void
173prio_reset(struct Qdisc* sch)
174{
175 int prio;
176 struct prio_sched_data *q = qdisc_priv(sch);
177
178 for (prio=0; prio<q->bands; prio++)
179 qdisc_reset(q->queues[prio]);
180 sch->q.qlen = 0;
181}
182
183static void
184prio_destroy(struct Qdisc* sch)
185{
186 int prio;
187 struct prio_sched_data *q = qdisc_priv(sch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Patrick McHardyff31ab52008-07-01 19:52:38 -0700189 tcf_destroy_chain(&q->filter_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 for (prio=0; prio<q->bands; prio++)
191 qdisc_destroy(q->queues[prio]);
192}
193
Patrick McHardy1e904742008-01-22 22:11:17 -0800194static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195{
196 struct prio_sched_data *q = qdisc_priv(sch);
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700197 struct tc_prio_qopt *qopt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 int i;
199
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700200 if (nla_len(opt) < sizeof(*qopt))
201 return -EINVAL;
202 qopt = nla_data(opt);
Patrick McHardycee63722008-01-23 20:33:32 -0800203
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700204 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 return -EINVAL;
206
207 for (i=0; i<=TC_PRIO_MAX; i++) {
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700208 if (qopt->priomap[i] >= qopt->bands)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 return -EINVAL;
210 }
211
212 sch_tree_lock(sch);
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700213 q->bands = qopt->bands;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
215
216 for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
217 struct Qdisc *child = xchg(&q->queues[i], &noop_qdisc);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800218 if (child != &noop_qdisc) {
219 qdisc_tree_decrease_qlen(child, child->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 qdisc_destroy(child);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800221 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 }
223 sch_tree_unlock(sch);
224
Amnon Aaronsohndd914b42006-01-17 02:24:26 -0800225 for (i=0; i<q->bands; i++) {
226 if (q->queues[i] == &noop_qdisc) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 struct Qdisc *child;
David S. Miller5ce2d482008-07-08 17:06:30 -0700228 child = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
David S. Millerbb949fb2008-07-08 16:55:56 -0700229 &pfifo_qdisc_ops,
Patrick McHardy9f9afec2006-11-29 17:35:18 -0800230 TC_H_MAKE(sch->handle, i + 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 if (child) {
232 sch_tree_lock(sch);
Amnon Aaronsohndd914b42006-01-17 02:24:26 -0800233 child = xchg(&q->queues[i], child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
Patrick McHardy5e50da02006-11-29 17:36:20 -0800235 if (child != &noop_qdisc) {
236 qdisc_tree_decrease_qlen(child,
237 child->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 qdisc_destroy(child);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800239 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 sch_tree_unlock(sch);
241 }
242 }
243 }
244 return 0;
245}
246
Patrick McHardy1e904742008-01-22 22:11:17 -0800247static int prio_init(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248{
249 struct prio_sched_data *q = qdisc_priv(sch);
250 int i;
251
252 for (i=0; i<TCQ_PRIO_BANDS; i++)
253 q->queues[i] = &noop_qdisc;
254
255 if (opt == NULL) {
256 return -EINVAL;
257 } else {
258 int err;
259
260 if ((err= prio_tune(sch, opt)) != 0)
261 return err;
262 }
263 return 0;
264}
265
266static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
267{
268 struct prio_sched_data *q = qdisc_priv(sch);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700269 unsigned char *b = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 struct tc_prio_qopt opt;
271
272 opt.bands = q->bands;
273 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700274
Thomas Graf2c10b322008-09-02 17:30:27 -0700275 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700276
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 return skb->len;
278
Patrick McHardy1e904742008-01-22 22:11:17 -0800279nla_put_failure:
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -0700280 nlmsg_trim(skb, b);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 return -1;
282}
283
284static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
285 struct Qdisc **old)
286{
287 struct prio_sched_data *q = qdisc_priv(sch);
288 unsigned long band = arg - 1;
289
290 if (band >= q->bands)
291 return -EINVAL;
292
293 if (new == NULL)
294 new = &noop_qdisc;
295
296 sch_tree_lock(sch);
297 *old = q->queues[band];
298 q->queues[band] = new;
Patrick McHardy5e50da02006-11-29 17:36:20 -0800299 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 qdisc_reset(*old);
301 sch_tree_unlock(sch);
302
303 return 0;
304}
305
306static struct Qdisc *
307prio_leaf(struct Qdisc *sch, unsigned long arg)
308{
309 struct prio_sched_data *q = qdisc_priv(sch);
310 unsigned long band = arg - 1;
311
312 if (band >= q->bands)
313 return NULL;
314
315 return q->queues[band];
316}
317
318static unsigned long prio_get(struct Qdisc *sch, u32 classid)
319{
320 struct prio_sched_data *q = qdisc_priv(sch);
321 unsigned long band = TC_H_MIN(classid);
322
323 if (band - 1 >= q->bands)
324 return 0;
325 return band;
326}
327
328static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
329{
330 return prio_get(sch, classid);
331}
332
333
334static void prio_put(struct Qdisc *q, unsigned long cl)
335{
336 return;
337}
338
Patrick McHardy1e904742008-01-22 22:11:17 -0800339static int prio_change(struct Qdisc *sch, u32 handle, u32 parent, struct nlattr **tca, unsigned long *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340{
341 unsigned long cl = *arg;
342 struct prio_sched_data *q = qdisc_priv(sch);
343
344 if (cl - 1 > q->bands)
345 return -ENOENT;
346 return 0;
347}
348
349static int prio_delete(struct Qdisc *sch, unsigned long cl)
350{
351 struct prio_sched_data *q = qdisc_priv(sch);
352 if (cl - 1 > q->bands)
353 return -ENOENT;
354 return 0;
355}
356
357
358static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
359 struct tcmsg *tcm)
360{
361 struct prio_sched_data *q = qdisc_priv(sch);
362
363 if (cl - 1 > q->bands)
364 return -ENOENT;
365 tcm->tcm_handle |= TC_H_MIN(cl);
366 if (q->queues[cl-1])
367 tcm->tcm_info = q->queues[cl-1]->handle;
368 return 0;
369}
370
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800371static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
372 struct gnet_dump *d)
373{
374 struct prio_sched_data *q = qdisc_priv(sch);
375 struct Qdisc *cl_q;
376
377 cl_q = q->queues[cl - 1];
378 if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 ||
379 gnet_stats_copy_queue(d, &cl_q->qstats) < 0)
380 return -1;
381
382 return 0;
383}
384
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
386{
387 struct prio_sched_data *q = qdisc_priv(sch);
388 int prio;
389
390 if (arg->stop)
391 return;
392
393 for (prio = 0; prio < q->bands; prio++) {
394 if (arg->count < arg->skip) {
395 arg->count++;
396 continue;
397 }
398 if (arg->fn(sch, prio+1, arg) < 0) {
399 arg->stop = 1;
400 break;
401 }
402 arg->count++;
403 }
404}
405
406static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
407{
408 struct prio_sched_data *q = qdisc_priv(sch);
409
410 if (cl)
411 return NULL;
412 return &q->filter_list;
413}
414
Eric Dumazet20fea082007-11-14 01:44:41 -0800415static const struct Qdisc_class_ops prio_class_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 .graft = prio_graft,
417 .leaf = prio_leaf,
418 .get = prio_get,
419 .put = prio_put,
420 .change = prio_change,
421 .delete = prio_delete,
422 .walk = prio_walk,
423 .tcf_chain = prio_find_tcf,
424 .bind_tcf = prio_bind,
425 .unbind_tcf = prio_put,
426 .dump = prio_dump_class,
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800427 .dump_stats = prio_dump_class_stats,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428};
429
Eric Dumazet20fea082007-11-14 01:44:41 -0800430static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 .next = NULL,
432 .cl_ops = &prio_class_ops,
433 .id = "prio",
434 .priv_size = sizeof(struct prio_sched_data),
435 .enqueue = prio_enqueue,
436 .dequeue = prio_dequeue,
Patrick McHardy48a8f512008-10-31 00:44:18 -0700437 .peek = prio_peek,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 .requeue = prio_requeue,
439 .drop = prio_drop,
440 .init = prio_init,
441 .reset = prio_reset,
442 .destroy = prio_destroy,
443 .change = prio_tune,
444 .dump = prio_dump,
445 .owner = THIS_MODULE,
446};
447
448static int __init prio_module_init(void)
449{
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700450 return register_qdisc(&prio_qdisc_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451}
452
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900453static void __exit prio_module_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454{
455 unregister_qdisc(&prio_qdisc_ops);
456}
457
458module_init(prio_module_init)
459module_exit(prio_module_exit)
460
461MODULE_LICENSE("GPL");