xref: /openbmc/linux/net/sched/sch_prio.c (revision 7aacf86b)
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 		case TC_ACT_SHOT:
54 			return NULL;
55 		}
56 #endif
57 		if (!fl || err < 0) {
58 			if (TC_H_MAJ(band))
59 				band = 0;
60 			return q->queues[q->prio2band[band & TC_PRIO_MAX]];
61 		}
62 		band = res.classid;
63 	}
64 	band = TC_H_MIN(band) - 1;
65 	if (band >= q->bands)
66 		return q->queues[q->prio2band[0]];
67 
68 	return q->queues[band];
69 }
70 
71 static int
72 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch, struct sk_buff **to_free)
73 {
74 	struct Qdisc *qdisc;
75 	int ret;
76 
77 	qdisc = prio_classify(skb, sch, &ret);
78 #ifdef CONFIG_NET_CLS_ACT
79 	if (qdisc == NULL) {
80 
81 		if (ret & __NET_XMIT_BYPASS)
82 			qdisc_qstats_drop(sch);
83 		kfree_skb(skb);
84 		return ret;
85 	}
86 #endif
87 
88 	ret = qdisc_enqueue(skb, qdisc, to_free);
89 	if (ret == NET_XMIT_SUCCESS) {
90 		qdisc_qstats_backlog_inc(sch, skb);
91 		sch->q.qlen++;
92 		return NET_XMIT_SUCCESS;
93 	}
94 	if (net_xmit_drop_count(ret))
95 		qdisc_qstats_drop(sch);
96 	return ret;
97 }
98 
99 static struct sk_buff *prio_peek(struct Qdisc *sch)
100 {
101 	struct prio_sched_data *q = qdisc_priv(sch);
102 	int prio;
103 
104 	for (prio = 0; prio < q->bands; prio++) {
105 		struct Qdisc *qdisc = q->queues[prio];
106 		struct sk_buff *skb = qdisc->ops->peek(qdisc);
107 		if (skb)
108 			return skb;
109 	}
110 	return NULL;
111 }
112 
113 static struct sk_buff *prio_dequeue(struct Qdisc *sch)
114 {
115 	struct prio_sched_data *q = qdisc_priv(sch);
116 	int prio;
117 
118 	for (prio = 0; prio < q->bands; prio++) {
119 		struct Qdisc *qdisc = q->queues[prio];
120 		struct sk_buff *skb = qdisc_dequeue_peeked(qdisc);
121 		if (skb) {
122 			qdisc_bstats_update(sch, skb);
123 			qdisc_qstats_backlog_dec(sch, skb);
124 			sch->q.qlen--;
125 			return skb;
126 		}
127 	}
128 	return NULL;
129 
130 }
131 
132 static void
133 prio_reset(struct Qdisc *sch)
134 {
135 	int prio;
136 	struct prio_sched_data *q = qdisc_priv(sch);
137 
138 	for (prio = 0; prio < q->bands; prio++)
139 		qdisc_reset(q->queues[prio]);
140 	sch->qstats.backlog = 0;
141 	sch->q.qlen = 0;
142 }
143 
144 static void
145 prio_destroy(struct Qdisc *sch)
146 {
147 	int prio;
148 	struct prio_sched_data *q = qdisc_priv(sch);
149 
150 	tcf_block_put(q->block);
151 	for (prio = 0; prio < q->bands; prio++)
152 		qdisc_destroy(q->queues[prio]);
153 }
154 
155 static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
156 {
157 	struct prio_sched_data *q = qdisc_priv(sch);
158 	struct Qdisc *queues[TCQ_PRIO_BANDS];
159 	int oldbands = q->bands, i;
160 	struct tc_prio_qopt *qopt;
161 
162 	if (nla_len(opt) < sizeof(*qopt))
163 		return -EINVAL;
164 	qopt = nla_data(opt);
165 
166 	if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
167 		return -EINVAL;
168 
169 	for (i = 0; i <= TC_PRIO_MAX; i++) {
170 		if (qopt->priomap[i] >= qopt->bands)
171 			return -EINVAL;
172 	}
173 
174 	/* Before commit, make sure we can allocate all new qdiscs */
175 	for (i = oldbands; i < qopt->bands; i++) {
176 		queues[i] = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
177 					      TC_H_MAKE(sch->handle, i + 1));
178 		if (!queues[i]) {
179 			while (i > oldbands)
180 				qdisc_destroy(queues[--i]);
181 			return -ENOMEM;
182 		}
183 	}
184 
185 	sch_tree_lock(sch);
186 	q->bands = qopt->bands;
187 	memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
188 
189 	for (i = q->bands; i < oldbands; i++) {
190 		struct Qdisc *child = q->queues[i];
191 
192 		qdisc_tree_reduce_backlog(child, child->q.qlen,
193 					  child->qstats.backlog);
194 		qdisc_destroy(child);
195 	}
196 
197 	for (i = oldbands; i < q->bands; i++) {
198 		q->queues[i] = queues[i];
199 		if (q->queues[i] != &noop_qdisc)
200 			qdisc_hash_add(q->queues[i], true);
201 	}
202 
203 	sch_tree_unlock(sch);
204 	return 0;
205 }
206 
207 static int prio_init(struct Qdisc *sch, struct nlattr *opt)
208 {
209 	struct prio_sched_data *q = qdisc_priv(sch);
210 	int err;
211 
212 	if (!opt)
213 		return -EINVAL;
214 
215 	err = tcf_block_get(&q->block, &q->filter_list);
216 	if (err)
217 		return err;
218 
219 	return prio_tune(sch, opt);
220 }
221 
222 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
223 {
224 	struct prio_sched_data *q = qdisc_priv(sch);
225 	unsigned char *b = skb_tail_pointer(skb);
226 	struct tc_prio_qopt opt;
227 
228 	opt.bands = q->bands;
229 	memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1);
230 
231 	if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
232 		goto nla_put_failure;
233 
234 	return skb->len;
235 
236 nla_put_failure:
237 	nlmsg_trim(skb, b);
238 	return -1;
239 }
240 
241 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
242 		      struct Qdisc **old)
243 {
244 	struct prio_sched_data *q = qdisc_priv(sch);
245 	unsigned long band = arg - 1;
246 
247 	if (new == NULL)
248 		new = &noop_qdisc;
249 
250 	*old = qdisc_replace(sch, new, &q->queues[band]);
251 	return 0;
252 }
253 
254 static struct Qdisc *
255 prio_leaf(struct Qdisc *sch, unsigned long arg)
256 {
257 	struct prio_sched_data *q = qdisc_priv(sch);
258 	unsigned long band = arg - 1;
259 
260 	return q->queues[band];
261 }
262 
263 static unsigned long prio_get(struct Qdisc *sch, u32 classid)
264 {
265 	struct prio_sched_data *q = qdisc_priv(sch);
266 	unsigned long band = TC_H_MIN(classid);
267 
268 	if (band - 1 >= q->bands)
269 		return 0;
270 	return band;
271 }
272 
273 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
274 {
275 	return prio_get(sch, classid);
276 }
277 
278 
279 static void prio_put(struct Qdisc *q, unsigned long cl)
280 {
281 }
282 
283 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
284 			   struct tcmsg *tcm)
285 {
286 	struct prio_sched_data *q = qdisc_priv(sch);
287 
288 	tcm->tcm_handle |= TC_H_MIN(cl);
289 	tcm->tcm_info = q->queues[cl-1]->handle;
290 	return 0;
291 }
292 
293 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
294 				 struct gnet_dump *d)
295 {
296 	struct prio_sched_data *q = qdisc_priv(sch);
297 	struct Qdisc *cl_q;
298 
299 	cl_q = q->queues[cl - 1];
300 	if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
301 				  d, NULL, &cl_q->bstats) < 0 ||
302 	    gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0)
303 		return -1;
304 
305 	return 0;
306 }
307 
308 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
309 {
310 	struct prio_sched_data *q = qdisc_priv(sch);
311 	int prio;
312 
313 	if (arg->stop)
314 		return;
315 
316 	for (prio = 0; prio < q->bands; prio++) {
317 		if (arg->count < arg->skip) {
318 			arg->count++;
319 			continue;
320 		}
321 		if (arg->fn(sch, prio + 1, arg) < 0) {
322 			arg->stop = 1;
323 			break;
324 		}
325 		arg->count++;
326 	}
327 }
328 
329 static struct tcf_block *prio_tcf_block(struct Qdisc *sch, unsigned long cl)
330 {
331 	struct prio_sched_data *q = qdisc_priv(sch);
332 
333 	if (cl)
334 		return NULL;
335 	return q->block;
336 }
337 
338 static const struct Qdisc_class_ops prio_class_ops = {
339 	.graft		=	prio_graft,
340 	.leaf		=	prio_leaf,
341 	.get		=	prio_get,
342 	.put		=	prio_put,
343 	.walk		=	prio_walk,
344 	.tcf_block	=	prio_tcf_block,
345 	.bind_tcf	=	prio_bind,
346 	.unbind_tcf	=	prio_put,
347 	.dump		=	prio_dump_class,
348 	.dump_stats	=	prio_dump_class_stats,
349 };
350 
351 static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
352 	.next		=	NULL,
353 	.cl_ops		=	&prio_class_ops,
354 	.id		=	"prio",
355 	.priv_size	=	sizeof(struct prio_sched_data),
356 	.enqueue	=	prio_enqueue,
357 	.dequeue	=	prio_dequeue,
358 	.peek		=	prio_peek,
359 	.init		=	prio_init,
360 	.reset		=	prio_reset,
361 	.destroy	=	prio_destroy,
362 	.change		=	prio_tune,
363 	.dump		=	prio_dump,
364 	.owner		=	THIS_MODULE,
365 };
366 
367 static int __init prio_module_init(void)
368 {
369 	return register_qdisc(&prio_qdisc_ops);
370 }
371 
372 static void __exit prio_module_exit(void)
373 {
374 	unregister_qdisc(&prio_qdisc_ops);
375 }
376 
377 module_init(prio_module_init)
378 module_exit(prio_module_exit)
379 
380 MODULE_LICENSE("GPL");
381