blob: 92c2e6d448d7984af35d6beb2cb3aea717b76511 [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>
Jiri Pirkocf1facd2017-02-09 14:38:56 +010023#include <net/pkt_cls.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
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
Eric Dumazet520ac302016-06-21 23:16:49 -070070prio_enqueue(struct sk_buff *skb, struct Qdisc *sch, struct sk_buff **to_free)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071{
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
Eric Dumazet520ac302016-06-21 23:16:49 -070086 ret = qdisc_enqueue(skb, qdisc, to_free);
Jussi Kivilinna5f861732008-07-20 00:08:04 -070087 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130static void
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000131prio_reset(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132{
133 int prio;
134 struct prio_sched_data *q = qdisc_priv(sch);
135
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000136 for (prio = 0; prio < q->bands; prio++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 qdisc_reset(q->queues[prio]);
WANG Cong6529d752016-06-01 16:15:16 -0700138 sch->qstats.backlog = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 sch->q.qlen = 0;
140}
141
142static void
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000143prio_destroy(struct Qdisc *sch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144{
145 int prio;
146 struct prio_sched_data *q = qdisc_priv(sch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
Patrick McHardyff31ab52008-07-01 19:52:38 -0700148 tcf_destroy_chain(&q->filter_list);
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000149 for (prio = 0; prio < q->bands; prio++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 qdisc_destroy(q->queues[prio]);
151}
152
Patrick McHardy1e904742008-01-22 22:11:17 -0800153static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154{
155 struct prio_sched_data *q = qdisc_priv(sch);
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700156 struct Qdisc *queues[TCQ_PRIO_BANDS];
157 int oldbands = q->bands, i;
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700158 struct tc_prio_qopt *qopt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700160 if (nla_len(opt) < sizeof(*qopt))
161 return -EINVAL;
162 qopt = nla_data(opt);
Patrick McHardycee63722008-01-23 20:33:32 -0800163
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700164 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 return -EINVAL;
166
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000167 for (i = 0; i <= TC_PRIO_MAX; i++) {
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700168 if (qopt->priomap[i] >= qopt->bands)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 return -EINVAL;
170 }
171
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700172 /* Before commit, make sure we can allocate all new qdiscs */
173 for (i = oldbands; i < qopt->bands; i++) {
174 queues[i] = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
175 TC_H_MAKE(sch->handle, i + 1));
176 if (!queues[i]) {
177 while (i > oldbands)
178 qdisc_destroy(queues[--i]);
179 return -ENOMEM;
180 }
181 }
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 sch_tree_lock(sch);
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700184 q->bands = qopt->bands;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
186
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700187 for (i = q->bands; i < oldbands; i++) {
Patrick McHardyb94c8af2008-11-20 04:11:36 -0800188 struct Qdisc *child = q->queues[i];
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700189
190 qdisc_tree_reduce_backlog(child, child->q.qlen,
191 child->qstats.backlog);
192 qdisc_destroy(child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 }
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700194
Jiri Kosina49b49972017-03-08 16:03:32 +0100195 for (i = oldbands; i < q->bands; i++) {
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700196 q->queues[i] = queues[i];
Jiri Kosina49b49972017-03-08 16:03:32 +0100197 if (q->queues[i] != &noop_qdisc)
198 qdisc_hash_add(q->queues[i], true);
199 }
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700200
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 sch_tree_unlock(sch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 return 0;
203}
204
Patrick McHardy1e904742008-01-22 22:11:17 -0800205static int prio_init(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206{
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700207 if (!opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
Eric Dumazet3d7c8252016-06-13 11:33:32 -0700210 return prio_tune(sch, opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211}
212
213static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
214{
215 struct prio_sched_data *q = qdisc_priv(sch);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700216 unsigned char *b = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 struct tc_prio_qopt opt;
218
219 opt.bands = q->bands;
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000220 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1);
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700221
David S. Miller1b34ec42012-03-29 05:11:39 -0400222 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
223 goto nla_put_failure;
Peter P Waskiewicz Jrd62733c2007-06-28 21:04:31 -0700224
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 return skb->len;
226
Patrick McHardy1e904742008-01-22 22:11:17 -0800227nla_put_failure:
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -0700228 nlmsg_trim(skb, b);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 return -1;
230}
231
232static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
233 struct Qdisc **old)
234{
235 struct prio_sched_data *q = qdisc_priv(sch);
236 unsigned long band = arg - 1;
237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 if (new == NULL)
239 new = &noop_qdisc;
240
WANG Cong86a79962016-02-25 14:55:00 -0800241 *old = qdisc_replace(sch, new, &q->queues[band]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 return 0;
243}
244
245static struct Qdisc *
246prio_leaf(struct Qdisc *sch, unsigned long arg)
247{
248 struct prio_sched_data *q = qdisc_priv(sch);
249 unsigned long band = arg - 1;
250
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 return q->queues[band];
252}
253
254static unsigned long prio_get(struct Qdisc *sch, u32 classid)
255{
256 struct prio_sched_data *q = qdisc_priv(sch);
257 unsigned long band = TC_H_MIN(classid);
258
259 if (band - 1 >= q->bands)
260 return 0;
261 return band;
262}
263
264static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
265{
266 return prio_get(sch, classid);
267}
268
269
270static void prio_put(struct Qdisc *q, unsigned long cl)
271{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272}
273
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
275 struct tcmsg *tcm)
276{
277 struct prio_sched_data *q = qdisc_priv(sch);
278
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 tcm->tcm_handle |= TC_H_MIN(cl);
Patrick McHardy5b9a9cc2009-09-04 06:41:17 +0000280 tcm->tcm_info = q->queues[cl-1]->handle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 return 0;
282}
283
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800284static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
285 struct gnet_dump *d)
286{
287 struct prio_sched_data *q = qdisc_priv(sch);
288 struct Qdisc *cl_q;
289
290 cl_q = q->queues[cl - 1];
Eric Dumazetedb09eb2016-06-06 09:37:16 -0700291 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
292 d, NULL, &cl_q->bstats) < 0 ||
John Fastabendb0ab6f92014-09-28 11:54:24 -0700293 gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0)
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800294 return -1;
295
296 return 0;
297}
298
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
300{
301 struct prio_sched_data *q = qdisc_priv(sch);
302 int prio;
303
304 if (arg->stop)
305 return;
306
307 for (prio = 0; prio < q->bands; prio++) {
308 if (arg->count < arg->skip) {
309 arg->count++;
310 continue;
311 }
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000312 if (arg->fn(sch, prio + 1, arg) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 arg->stop = 1;
314 break;
315 }
316 arg->count++;
317 }
318}
319
John Fastabend25d8c0d2014-09-12 20:05:27 -0700320static struct tcf_proto __rcu **prio_find_tcf(struct Qdisc *sch,
321 unsigned long cl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322{
323 struct prio_sched_data *q = qdisc_priv(sch);
324
325 if (cl)
326 return NULL;
327 return &q->filter_list;
328}
329
Eric Dumazet20fea082007-11-14 01:44:41 -0800330static const struct Qdisc_class_ops prio_class_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 .graft = prio_graft,
332 .leaf = prio_leaf,
333 .get = prio_get,
334 .put = prio_put,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 .walk = prio_walk,
336 .tcf_chain = prio_find_tcf,
337 .bind_tcf = prio_bind,
338 .unbind_tcf = prio_put,
339 .dump = prio_dump_class,
Jarek Poplawski2cf6c362007-01-31 12:21:24 -0800340 .dump_stats = prio_dump_class_stats,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341};
342
Eric Dumazet20fea082007-11-14 01:44:41 -0800343static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 .next = NULL,
345 .cl_ops = &prio_class_ops,
346 .id = "prio",
347 .priv_size = sizeof(struct prio_sched_data),
348 .enqueue = prio_enqueue,
349 .dequeue = prio_dequeue,
Patrick McHardy48a8f512008-10-31 00:44:18 -0700350 .peek = prio_peek,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 .init = prio_init,
352 .reset = prio_reset,
353 .destroy = prio_destroy,
354 .change = prio_tune,
355 .dump = prio_dump,
356 .owner = THIS_MODULE,
357};
358
359static int __init prio_module_init(void)
360{
David S. Miller1d8ae3f2008-07-15 02:52:19 -0700361 return register_qdisc(&prio_qdisc_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362}
363
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900364static void __exit prio_module_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365{
366 unregister_qdisc(&prio_qdisc_ops);
367}
368
369module_init(prio_module_init)
370module_exit(prio_module_exit)
371
372MODULE_LICENSE("GPL");