xref: /openbmc/linux/net/sched/sch_prio.c (revision 9dae47aba0a055f761176d9297371d5bb24289ec)
1 /*
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>
10  * Fixes:       19990609: J Hadi Salim <hadi@nortelnetworks.com>:
11  *              Init --  EINVAL when opt undefined
12  */
13 
14 #include <linux/module.h>
15 #include <linux/slab.h>
16 #include <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/string.h>
19 #include <linux/errno.h>
20 #include <linux/skbuff.h>
21 #include <net/netlink.h>
22 #include <net/pkt_sched.h>
23 #include <net/pkt_cls.h>
24 
25 struct prio_sched_data {
26 	int bands;
27 	struct tcf_proto __rcu *filter_list;
28 	struct tcf_block *block;
29 	u8  prio2band[TC_PRIO_MAX+1];
30 	struct Qdisc *queues[TCQ_PRIO_BANDS];
31 };
32 
33 
34 static struct Qdisc *
35 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
36 {
37 	struct prio_sched_data *q = qdisc_priv(sch);
38 	u32 band = skb->priority;
39 	struct tcf_result res;
40 	struct tcf_proto *fl;
41 	int err;
42 
43 	*qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
44 	if (TC_H_MAJ(skb->priority) != sch->handle) {
45 		fl = rcu_dereference_bh(q->filter_list);
46 		err = tcf_classify(skb, fl, &res, false);
47 #ifdef CONFIG_NET_CLS_ACT
48 		switch (err) {
49 		case TC_ACT_STOLEN:
50 		case TC_ACT_QUEUED:
51 		case TC_ACT_TRAP:
52 			*qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
53 			/* fall through */
54 		case TC_ACT_SHOT:
55 			return NULL;
56 		}
57 #endif
58 		if (!fl || err < 0) {
59 			if (TC_H_MAJ(band))
60 				band = 0;
61 			return q->queues[q->prio2band[band & TC_PRIO_MAX]];
62 		}
63 		band = res.classid;
64 	}
65 	band = TC_H_MIN(band) - 1;
66 	if (band >= q->bands)
67 		return q->queues[q->prio2band[0]];
68 
69 	return q->queues[band];
70 }
71 
72 static int
73 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch, struct sk_buff **to_free)
74 {
75 	struct Qdisc *qdisc;
76 	int ret;
77 
78 	qdisc = prio_classify(skb, sch, &ret);
79 #ifdef CONFIG_NET_CLS_ACT
80 	if (qdisc == NULL) {
81 
82 		if (ret & __NET_XMIT_BYPASS)
83 			qdisc_qstats_drop(sch);
84 		__qdisc_drop(skb, to_free);
85 		return ret;
86 	}
87 #endif
88 
89 	ret = qdisc_enqueue(skb, qdisc, to_free);
90 	if (ret == NET_XMIT_SUCCESS) {
91 		qdisc_qstats_backlog_inc(sch, skb);
92 		sch->q.qlen++;
93 		return NET_XMIT_SUCCESS;
94 	}
95 	if (net_xmit_drop_count(ret))
96 		qdisc_qstats_drop(sch);
97 	return ret;
98 }
99 
100 static struct sk_buff *prio_peek(struct Qdisc *sch)
101 {
102 	struct prio_sched_data *q = qdisc_priv(sch);
103 	int prio;
104 
105 	for (prio = 0; prio < q->bands; prio++) {
106 		struct Qdisc *qdisc = q->queues[prio];
107 		struct sk_buff *skb = qdisc->ops->peek(qdisc);
108 		if (skb)
109 			return skb;
110 	}
111 	return NULL;
112 }
113 
114 static struct sk_buff *prio_dequeue(struct Qdisc *sch)
115 {
116 	struct prio_sched_data *q = qdisc_priv(sch);
117 	int prio;
118 
119 	for (prio = 0; prio < q->bands; prio++) {
120 		struct Qdisc *qdisc = q->queues[prio];
121 		struct sk_buff *skb = qdisc_dequeue_peeked(qdisc);
122 		if (skb) {
123 			qdisc_bstats_update(sch, skb);
124 			qdisc_qstats_backlog_dec(sch, skb);
125 			sch->q.qlen--;
126 			return skb;
127 		}
128 	}
129 	return NULL;
130 
131 }
132 
133 static void
134 prio_reset(struct Qdisc *sch)
135 {
136 	int prio;
137 	struct prio_sched_data *q = qdisc_priv(sch);
138 
139 	for (prio = 0; prio < q->bands; prio++)
140 		qdisc_reset(q->queues[prio]);
141 	sch->qstats.backlog = 0;
142 	sch->q.qlen = 0;
143 }
144 
145 static void
146 prio_destroy(struct Qdisc *sch)
147 {
148 	int prio;
149 	struct prio_sched_data *q = qdisc_priv(sch);
150 
151 	tcf_block_put(q->block);
152 	for (prio = 0; prio < q->bands; prio++)
153 		qdisc_destroy(q->queues[prio]);
154 }
155 
156 static int prio_tune(struct Qdisc *sch, struct nlattr *opt,
157 		     struct netlink_ext_ack *extack)
158 {
159 	struct prio_sched_data *q = qdisc_priv(sch);
160 	struct Qdisc *queues[TCQ_PRIO_BANDS];
161 	int oldbands = q->bands, i;
162 	struct tc_prio_qopt *qopt;
163 
164 	if (nla_len(opt) < sizeof(*qopt))
165 		return -EINVAL;
166 	qopt = nla_data(opt);
167 
168 	if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
169 		return -EINVAL;
170 
171 	for (i = 0; i <= TC_PRIO_MAX; i++) {
172 		if (qopt->priomap[i] >= qopt->bands)
173 			return -EINVAL;
174 	}
175 
176 	/* Before commit, make sure we can allocate all new qdiscs */
177 	for (i = oldbands; i < qopt->bands; i++) {
178 		queues[i] = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
179 					      TC_H_MAKE(sch->handle, i + 1),
180 					      extack);
181 		if (!queues[i]) {
182 			while (i > oldbands)
183 				qdisc_destroy(queues[--i]);
184 			return -ENOMEM;
185 		}
186 	}
187 
188 	sch_tree_lock(sch);
189 	q->bands = qopt->bands;
190 	memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
191 
192 	for (i = q->bands; i < oldbands; i++) {
193 		struct Qdisc *child = q->queues[i];
194 
195 		qdisc_tree_reduce_backlog(child, child->q.qlen,
196 					  child->qstats.backlog);
197 		qdisc_destroy(child);
198 	}
199 
200 	for (i = oldbands; i < q->bands; i++) {
201 		q->queues[i] = queues[i];
202 		if (q->queues[i] != &noop_qdisc)
203 			qdisc_hash_add(q->queues[i], true);
204 	}
205 
206 	sch_tree_unlock(sch);
207 	return 0;
208 }
209 
210 static int prio_init(struct Qdisc *sch, struct nlattr *opt,
211 		     struct netlink_ext_ack *extack)
212 {
213 	struct prio_sched_data *q = qdisc_priv(sch);
214 	int err;
215 
216 	if (!opt)
217 		return -EINVAL;
218 
219 	err = tcf_block_get(&q->block, &q->filter_list, sch, extack);
220 	if (err)
221 		return err;
222 
223 	return prio_tune(sch, opt, extack);
224 }
225 
226 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
227 {
228 	struct prio_sched_data *q = qdisc_priv(sch);
229 	unsigned char *b = skb_tail_pointer(skb);
230 	struct tc_prio_qopt opt;
231 
232 	opt.bands = q->bands;
233 	memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1);
234 
235 	if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
236 		goto nla_put_failure;
237 
238 	return skb->len;
239 
240 nla_put_failure:
241 	nlmsg_trim(skb, b);
242 	return -1;
243 }
244 
245 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
246 		      struct Qdisc **old, struct netlink_ext_ack *extack)
247 {
248 	struct prio_sched_data *q = qdisc_priv(sch);
249 	unsigned long band = arg - 1;
250 
251 	if (new == NULL)
252 		new = &noop_qdisc;
253 
254 	*old = qdisc_replace(sch, new, &q->queues[band]);
255 	return 0;
256 }
257 
258 static struct Qdisc *
259 prio_leaf(struct Qdisc *sch, unsigned long arg)
260 {
261 	struct prio_sched_data *q = qdisc_priv(sch);
262 	unsigned long band = arg - 1;
263 
264 	return q->queues[band];
265 }
266 
267 static unsigned long prio_find(struct Qdisc *sch, u32 classid)
268 {
269 	struct prio_sched_data *q = qdisc_priv(sch);
270 	unsigned long band = TC_H_MIN(classid);
271 
272 	if (band - 1 >= q->bands)
273 		return 0;
274 	return band;
275 }
276 
277 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
278 {
279 	return prio_find(sch, classid);
280 }
281 
282 
283 static void prio_unbind(struct Qdisc *q, unsigned long cl)
284 {
285 }
286 
287 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
288 			   struct tcmsg *tcm)
289 {
290 	struct prio_sched_data *q = qdisc_priv(sch);
291 
292 	tcm->tcm_handle |= TC_H_MIN(cl);
293 	tcm->tcm_info = q->queues[cl-1]->handle;
294 	return 0;
295 }
296 
297 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
298 				 struct gnet_dump *d)
299 {
300 	struct prio_sched_data *q = qdisc_priv(sch);
301 	struct Qdisc *cl_q;
302 
303 	cl_q = q->queues[cl - 1];
304 	if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
305 				  d, NULL, &cl_q->bstats) < 0 ||
306 	    gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0)
307 		return -1;
308 
309 	return 0;
310 }
311 
312 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
313 {
314 	struct prio_sched_data *q = qdisc_priv(sch);
315 	int prio;
316 
317 	if (arg->stop)
318 		return;
319 
320 	for (prio = 0; prio < q->bands; prio++) {
321 		if (arg->count < arg->skip) {
322 			arg->count++;
323 			continue;
324 		}
325 		if (arg->fn(sch, prio + 1, arg) < 0) {
326 			arg->stop = 1;
327 			break;
328 		}
329 		arg->count++;
330 	}
331 }
332 
333 static struct tcf_block *prio_tcf_block(struct Qdisc *sch, unsigned long cl,
334 					struct netlink_ext_ack *extack)
335 {
336 	struct prio_sched_data *q = qdisc_priv(sch);
337 
338 	if (cl)
339 		return NULL;
340 	return q->block;
341 }
342 
343 static const struct Qdisc_class_ops prio_class_ops = {
344 	.graft		=	prio_graft,
345 	.leaf		=	prio_leaf,
346 	.find		=	prio_find,
347 	.walk		=	prio_walk,
348 	.tcf_block	=	prio_tcf_block,
349 	.bind_tcf	=	prio_bind,
350 	.unbind_tcf	=	prio_unbind,
351 	.dump		=	prio_dump_class,
352 	.dump_stats	=	prio_dump_class_stats,
353 };
354 
355 static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
356 	.next		=	NULL,
357 	.cl_ops		=	&prio_class_ops,
358 	.id		=	"prio",
359 	.priv_size	=	sizeof(struct prio_sched_data),
360 	.enqueue	=	prio_enqueue,
361 	.dequeue	=	prio_dequeue,
362 	.peek		=	prio_peek,
363 	.init		=	prio_init,
364 	.reset		=	prio_reset,
365 	.destroy	=	prio_destroy,
366 	.change		=	prio_tune,
367 	.dump		=	prio_dump,
368 	.owner		=	THIS_MODULE,
369 };
370 
371 static int __init prio_module_init(void)
372 {
373 	return register_qdisc(&prio_qdisc_ops);
374 }
375 
376 static void __exit prio_module_exit(void)
377 {
378 	unregister_qdisc(&prio_qdisc_ops);
379 }
380 
381 module_init(prio_module_init)
382 module_exit(prio_module_exit)
383 
384 MODULE_LICENSE("GPL");
385