blob: a356450b747ba57552d6c1e1e35255935580f02a [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) {
WANG Cong6529d752016-06-01 16:15:16 -070088 qdisc_qstats_backlog_inc(sch, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 sch->q.qlen++;
90 return NET_XMIT_SUCCESS;
91 }
Jarek Poplawski378a2f02008-08-04 22:31:03 -070092 if (net_xmit_drop_count(ret))
John Fastabend25331d62014-09-28 11:53:29 -070093 qdisc_qstats_drop(sch);
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
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000111static 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];
Florian Westphal35576192011-08-09 02:04:43 +0000118 struct sk_buff *skb = qdisc_dequeue_peeked(qdisc);
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700119 if (skb) {
Eric Dumazet9190b3b2011-01-20 23:31:33 -0800120 qdisc_bstats_update(sch, skb);
WANG Cong6529d752016-06-01 16:15:16 -0700121 qdisc_qstats_backlog_dec(sch, skb);
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700122 sch->q.qlen--;
123 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 }
125 }
126 return NULL;
127
128}
129
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000130static unsigned int prio_drop(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131{
132 struct prio_sched_data *q = qdisc_priv(sch);
133 int prio;
134 unsigned int len;
135 struct Qdisc *qdisc;
136
137 for (prio = q->bands-1; prio >= 0; prio--) {
138 qdisc = q->queues[prio];
Patrick McHardy6d037a22006-03-20 19:00:49 -0800139 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
WANG Cong6529d752016-06-01 16:15:16 -0700140 sch->qstats.backlog -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 sch->q.qlen--;
142 return len;
143 }
144 }
145 return 0;
146}
147
148
149static void
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000150prio_reset(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151{
152 int prio;
153 struct prio_sched_data *q = qdisc_priv(sch);
154
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000155 for (prio = 0; prio < q->bands; prio++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 qdisc_reset(q->queues[prio]);
WANG Cong6529d752016-06-01 16:15:16 -0700157 sch->qstats.backlog = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 sch->q.qlen = 0;
159}
160
161static void
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000162prio_destroy(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163{
164 int prio;
165 struct prio_sched_data *q = qdisc_priv(sch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
Patrick McHardyff31ab52008-07-01 19:52:38 -0700167 tcf_destroy_chain(&q->filter_list);
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000168 for (prio = 0; prio < q->bands; prio++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 qdisc_destroy(q->queues[prio]);
170}
171
Patrick McHardy1e904742008-01-22 22:11:17 -0800172static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173{
174 struct prio_sched_data *q = qdisc_priv(sch);
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700175 struct Qdisc *queues[TCQ_PRIO_BANDS];
176 int oldbands = q->bands, i;
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700177 struct tc_prio_qopt *qopt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700179 if (nla_len(opt) < sizeof(*qopt))
180 return -EINVAL;
181 qopt = nla_data(opt);
Patrick McHardycee63722008-01-23 20:33:32 -0800182
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700183 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 return -EINVAL;
185
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000186 for (i = 0; i <= TC_PRIO_MAX; i++) {
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700187 if (qopt->priomap[i] >= qopt->bands)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 return -EINVAL;
189 }
190
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700191 /* Before commit, make sure we can allocate all new qdiscs */
192 for (i = oldbands; i < qopt->bands; i++) {
193 queues[i] = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
194 TC_H_MAKE(sch->handle, i + 1));
195 if (!queues[i]) {
196 while (i > oldbands)
197 qdisc_destroy(queues[--i]);
198 return -ENOMEM;
199 }
200 }
201
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 sch_tree_lock(sch);
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700203 q->bands = qopt->bands;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
205
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700206 for (i = q->bands; i < oldbands; i++) {
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800207 struct Qdisc *child = q->queues[i];
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700208
209 qdisc_tree_reduce_backlog(child, child->q.qlen,
210 child->qstats.backlog);
211 qdisc_destroy(child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 }
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700213
214 for (i = oldbands; i < q->bands; i++)
215 q->queues[i] = queues[i];
216
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 sch_tree_unlock(sch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 return 0;
219}
220
Patrick McHardy1e904742008-01-22 22:11:17 -0800221static int prio_init(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222{
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700223 if (!opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700226 return prio_tune(sch, opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227}
228
229static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
230{
231 struct prio_sched_data *q = qdisc_priv(sch);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700232 unsigned char *b = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 struct tc_prio_qopt opt;
234
235 opt.bands = q->bands;
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000236 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1);
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700237
David S. Miller1b34ec42012-03-29 05:11:39 -0400238 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
239 goto nla_put_failure;
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700240
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 return skb->len;
242
Patrick McHardy1e904742008-01-22 22:11:17 -0800243nla_put_failure:
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -0700244 nlmsg_trim(skb, b);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 return -1;
246}
247
248static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
249 struct Qdisc **old)
250{
251 struct prio_sched_data *q = qdisc_priv(sch);
252 unsigned long band = arg - 1;
253
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 if (new == NULL)
255 new = &noop_qdisc;
256
WANG Cong86a79962016-02-25 14:55:00 -0800257 *old = qdisc_replace(sch, new, &q->queues[band]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 return 0;
259}
260
261static struct Qdisc *
262prio_leaf(struct Qdisc *sch, unsigned long arg)
263{
264 struct prio_sched_data *q = qdisc_priv(sch);
265 unsigned long band = arg - 1;
266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 return q->queues[band];
268}
269
270static unsigned long prio_get(struct Qdisc *sch, u32 classid)
271{
272 struct prio_sched_data *q = qdisc_priv(sch);
273 unsigned long band = TC_H_MIN(classid);
274
275 if (band - 1 >= q->bands)
276 return 0;
277 return band;
278}
279
280static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
281{
282 return prio_get(sch, classid);
283}
284
285
286static void prio_put(struct Qdisc *q, unsigned long cl)
287{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288}
289
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
291 struct tcmsg *tcm)
292{
293 struct prio_sched_data *q = qdisc_priv(sch);
294
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 tcm->tcm_handle |= TC_H_MIN(cl);
Patrick McHardy5b9a9cc2009-09-04 06:41:17 +0000296 tcm->tcm_info = q->queues[cl-1]->handle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 return 0;
298}
299
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800300static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
301 struct gnet_dump *d)
302{
303 struct prio_sched_data *q = qdisc_priv(sch);
304 struct Qdisc *cl_q;
305
306 cl_q = q->queues[cl - 1];
John Fastabend22e0f8b2014-09-28 11:52:56 -0700307 if (gnet_stats_copy_basic(d, NULL, &cl_q->bstats) < 0 ||
John Fastabendb0ab6f92014-09-28 11:54:24 -0700308 gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0)
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800309 return -1;
310
311 return 0;
312}
313
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
315{
316 struct prio_sched_data *q = qdisc_priv(sch);
317 int prio;
318
319 if (arg->stop)
320 return;
321
322 for (prio = 0; prio < q->bands; prio++) {
323 if (arg->count < arg->skip) {
324 arg->count++;
325 continue;
326 }
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000327 if (arg->fn(sch, prio + 1, arg) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 arg->stop = 1;
329 break;
330 }
331 arg->count++;
332 }
333}
334
John Fastabend25d8c0d2014-09-12 20:05:27 -0700335static struct tcf_proto __rcu **prio_find_tcf(struct Qdisc *sch,
336 unsigned long cl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337{
338 struct prio_sched_data *q = qdisc_priv(sch);
339
340 if (cl)
341 return NULL;
342 return &q->filter_list;
343}
344
Eric Dumazet20fea082007-11-14 01:44:41 -0800345static const struct Qdisc_class_ops prio_class_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 .graft = prio_graft,
347 .leaf = prio_leaf,
348 .get = prio_get,
349 .put = prio_put,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 .walk = prio_walk,
351 .tcf_chain = prio_find_tcf,
352 .bind_tcf = prio_bind,
353 .unbind_tcf = prio_put,
354 .dump = prio_dump_class,
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800355 .dump_stats = prio_dump_class_stats,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356};
357
Eric Dumazet20fea082007-11-14 01:44:41 -0800358static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 .next = NULL,
360 .cl_ops = &prio_class_ops,
361 .id = "prio",
362 .priv_size = sizeof(struct prio_sched_data),
363 .enqueue = prio_enqueue,
364 .dequeue = prio_dequeue,
Patrick McHardy48a8f512008-10-31 00:44:18 -0700365 .peek = prio_peek,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 .drop = prio_drop,
367 .init = prio_init,
368 .reset = prio_reset,
369 .destroy = prio_destroy,
370 .change = prio_tune,
371 .dump = prio_dump,
372 .owner = THIS_MODULE,
373};
374
375static int __init prio_module_init(void)
376{
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700377 return register_qdisc(&prio_qdisc_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378}
379
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900380static void __exit prio_module_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381{
382 unregister_qdisc(&prio_qdisc_ops);
383}
384
385module_init(prio_module_init)
386module_exit(prio_module_exit)
387
388MODULE_LICENSE("GPL");