xref: /openbmc/linux/net/sched/sch_drr.c (revision bf070bb0)
1 /*
2  * net/sched/sch_drr.c         Deficit Round Robin scheduler
3  *
4  * Copyright (c) 2008 Patrick McHardy <kaber@trash.net>
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * version 2 as published by the Free Software Foundation.
9  */
10 
11 #include <linux/module.h>
12 #include <linux/slab.h>
13 #include <linux/init.h>
14 #include <linux/errno.h>
15 #include <linux/netdevice.h>
16 #include <linux/pkt_sched.h>
17 #include <net/sch_generic.h>
18 #include <net/pkt_sched.h>
19 #include <net/pkt_cls.h>
20 
21 struct drr_class {
22 	struct Qdisc_class_common	common;
23 	unsigned int			filter_cnt;
24 
25 	struct gnet_stats_basic_packed		bstats;
26 	struct gnet_stats_queue		qstats;
27 	struct net_rate_estimator __rcu *rate_est;
28 	struct list_head		alist;
29 	struct Qdisc			*qdisc;
30 
31 	u32				quantum;
32 	u32				deficit;
33 };
34 
35 struct drr_sched {
36 	struct list_head		active;
37 	struct tcf_proto __rcu		*filter_list;
38 	struct tcf_block		*block;
39 	struct Qdisc_class_hash		clhash;
40 };
41 
42 static struct drr_class *drr_find_class(struct Qdisc *sch, u32 classid)
43 {
44 	struct drr_sched *q = qdisc_priv(sch);
45 	struct Qdisc_class_common *clc;
46 
47 	clc = qdisc_class_find(&q->clhash, classid);
48 	if (clc == NULL)
49 		return NULL;
50 	return container_of(clc, struct drr_class, common);
51 }
52 
53 static void drr_purge_queue(struct drr_class *cl)
54 {
55 	unsigned int len = cl->qdisc->q.qlen;
56 	unsigned int backlog = cl->qdisc->qstats.backlog;
57 
58 	qdisc_reset(cl->qdisc);
59 	qdisc_tree_reduce_backlog(cl->qdisc, len, backlog);
60 }
61 
62 static const struct nla_policy drr_policy[TCA_DRR_MAX + 1] = {
63 	[TCA_DRR_QUANTUM]	= { .type = NLA_U32 },
64 };
65 
66 static int drr_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
67 			    struct nlattr **tca, unsigned long *arg)
68 {
69 	struct drr_sched *q = qdisc_priv(sch);
70 	struct drr_class *cl = (struct drr_class *)*arg;
71 	struct nlattr *opt = tca[TCA_OPTIONS];
72 	struct nlattr *tb[TCA_DRR_MAX + 1];
73 	u32 quantum;
74 	int err;
75 
76 	if (!opt)
77 		return -EINVAL;
78 
79 	err = nla_parse_nested(tb, TCA_DRR_MAX, opt, drr_policy, NULL);
80 	if (err < 0)
81 		return err;
82 
83 	if (tb[TCA_DRR_QUANTUM]) {
84 		quantum = nla_get_u32(tb[TCA_DRR_QUANTUM]);
85 		if (quantum == 0)
86 			return -EINVAL;
87 	} else
88 		quantum = psched_mtu(qdisc_dev(sch));
89 
90 	if (cl != NULL) {
91 		if (tca[TCA_RATE]) {
92 			err = gen_replace_estimator(&cl->bstats, NULL,
93 						    &cl->rate_est,
94 						    NULL,
95 						    qdisc_root_sleeping_running(sch),
96 						    tca[TCA_RATE]);
97 			if (err)
98 				return err;
99 		}
100 
101 		sch_tree_lock(sch);
102 		if (tb[TCA_DRR_QUANTUM])
103 			cl->quantum = quantum;
104 		sch_tree_unlock(sch);
105 
106 		return 0;
107 	}
108 
109 	cl = kzalloc(sizeof(struct drr_class), GFP_KERNEL);
110 	if (cl == NULL)
111 		return -ENOBUFS;
112 
113 	cl->common.classid = classid;
114 	cl->quantum	   = quantum;
115 	cl->qdisc	   = qdisc_create_dflt(sch->dev_queue,
116 					       &pfifo_qdisc_ops, classid);
117 	if (cl->qdisc == NULL)
118 		cl->qdisc = &noop_qdisc;
119 	else
120 		qdisc_hash_add(cl->qdisc, true);
121 
122 	if (tca[TCA_RATE]) {
123 		err = gen_replace_estimator(&cl->bstats, NULL, &cl->rate_est,
124 					    NULL,
125 					    qdisc_root_sleeping_running(sch),
126 					    tca[TCA_RATE]);
127 		if (err) {
128 			qdisc_destroy(cl->qdisc);
129 			kfree(cl);
130 			return err;
131 		}
132 	}
133 
134 	sch_tree_lock(sch);
135 	qdisc_class_hash_insert(&q->clhash, &cl->common);
136 	sch_tree_unlock(sch);
137 
138 	qdisc_class_hash_grow(sch, &q->clhash);
139 
140 	*arg = (unsigned long)cl;
141 	return 0;
142 }
143 
144 static void drr_destroy_class(struct Qdisc *sch, struct drr_class *cl)
145 {
146 	gen_kill_estimator(&cl->rate_est);
147 	qdisc_destroy(cl->qdisc);
148 	kfree(cl);
149 }
150 
151 static int drr_delete_class(struct Qdisc *sch, unsigned long arg)
152 {
153 	struct drr_sched *q = qdisc_priv(sch);
154 	struct drr_class *cl = (struct drr_class *)arg;
155 
156 	if (cl->filter_cnt > 0)
157 		return -EBUSY;
158 
159 	sch_tree_lock(sch);
160 
161 	drr_purge_queue(cl);
162 	qdisc_class_hash_remove(&q->clhash, &cl->common);
163 
164 	sch_tree_unlock(sch);
165 
166 	drr_destroy_class(sch, cl);
167 	return 0;
168 }
169 
170 static unsigned long drr_search_class(struct Qdisc *sch, u32 classid)
171 {
172 	return (unsigned long)drr_find_class(sch, classid);
173 }
174 
175 static struct tcf_block *drr_tcf_block(struct Qdisc *sch, unsigned long cl)
176 {
177 	struct drr_sched *q = qdisc_priv(sch);
178 
179 	if (cl)
180 		return NULL;
181 
182 	return q->block;
183 }
184 
185 static unsigned long drr_bind_tcf(struct Qdisc *sch, unsigned long parent,
186 				  u32 classid)
187 {
188 	struct drr_class *cl = drr_find_class(sch, classid);
189 
190 	if (cl != NULL)
191 		cl->filter_cnt++;
192 
193 	return (unsigned long)cl;
194 }
195 
196 static void drr_unbind_tcf(struct Qdisc *sch, unsigned long arg)
197 {
198 	struct drr_class *cl = (struct drr_class *)arg;
199 
200 	cl->filter_cnt--;
201 }
202 
203 static int drr_graft_class(struct Qdisc *sch, unsigned long arg,
204 			   struct Qdisc *new, struct Qdisc **old)
205 {
206 	struct drr_class *cl = (struct drr_class *)arg;
207 
208 	if (new == NULL) {
209 		new = qdisc_create_dflt(sch->dev_queue,
210 					&pfifo_qdisc_ops, cl->common.classid);
211 		if (new == NULL)
212 			new = &noop_qdisc;
213 	}
214 
215 	*old = qdisc_replace(sch, new, &cl->qdisc);
216 	return 0;
217 }
218 
219 static struct Qdisc *drr_class_leaf(struct Qdisc *sch, unsigned long arg)
220 {
221 	struct drr_class *cl = (struct drr_class *)arg;
222 
223 	return cl->qdisc;
224 }
225 
226 static void drr_qlen_notify(struct Qdisc *csh, unsigned long arg)
227 {
228 	struct drr_class *cl = (struct drr_class *)arg;
229 
230 	list_del(&cl->alist);
231 }
232 
233 static int drr_dump_class(struct Qdisc *sch, unsigned long arg,
234 			  struct sk_buff *skb, struct tcmsg *tcm)
235 {
236 	struct drr_class *cl = (struct drr_class *)arg;
237 	struct nlattr *nest;
238 
239 	tcm->tcm_parent	= TC_H_ROOT;
240 	tcm->tcm_handle	= cl->common.classid;
241 	tcm->tcm_info	= cl->qdisc->handle;
242 
243 	nest = nla_nest_start(skb, TCA_OPTIONS);
244 	if (nest == NULL)
245 		goto nla_put_failure;
246 	if (nla_put_u32(skb, TCA_DRR_QUANTUM, cl->quantum))
247 		goto nla_put_failure;
248 	return nla_nest_end(skb, nest);
249 
250 nla_put_failure:
251 	nla_nest_cancel(skb, nest);
252 	return -EMSGSIZE;
253 }
254 
255 static int drr_dump_class_stats(struct Qdisc *sch, unsigned long arg,
256 				struct gnet_dump *d)
257 {
258 	struct drr_class *cl = (struct drr_class *)arg;
259 	__u32 qlen = cl->qdisc->q.qlen;
260 	struct tc_drr_stats xstats;
261 
262 	memset(&xstats, 0, sizeof(xstats));
263 	if (qlen)
264 		xstats.deficit = cl->deficit;
265 
266 	if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
267 				  d, NULL, &cl->bstats) < 0 ||
268 	    gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
269 	    gnet_stats_copy_queue(d, NULL, &cl->qdisc->qstats, qlen) < 0)
270 		return -1;
271 
272 	return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
273 }
274 
275 static void drr_walk(struct Qdisc *sch, struct qdisc_walker *arg)
276 {
277 	struct drr_sched *q = qdisc_priv(sch);
278 	struct drr_class *cl;
279 	unsigned int i;
280 
281 	if (arg->stop)
282 		return;
283 
284 	for (i = 0; i < q->clhash.hashsize; i++) {
285 		hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
286 			if (arg->count < arg->skip) {
287 				arg->count++;
288 				continue;
289 			}
290 			if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
291 				arg->stop = 1;
292 				return;
293 			}
294 			arg->count++;
295 		}
296 	}
297 }
298 
299 static struct drr_class *drr_classify(struct sk_buff *skb, struct Qdisc *sch,
300 				      int *qerr)
301 {
302 	struct drr_sched *q = qdisc_priv(sch);
303 	struct drr_class *cl;
304 	struct tcf_result res;
305 	struct tcf_proto *fl;
306 	int result;
307 
308 	if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) {
309 		cl = drr_find_class(sch, skb->priority);
310 		if (cl != NULL)
311 			return cl;
312 	}
313 
314 	*qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
315 	fl = rcu_dereference_bh(q->filter_list);
316 	result = tcf_classify(skb, fl, &res, false);
317 	if (result >= 0) {
318 #ifdef CONFIG_NET_CLS_ACT
319 		switch (result) {
320 		case TC_ACT_QUEUED:
321 		case TC_ACT_STOLEN:
322 		case TC_ACT_TRAP:
323 			*qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
324 			/* fall through */
325 		case TC_ACT_SHOT:
326 			return NULL;
327 		}
328 #endif
329 		cl = (struct drr_class *)res.class;
330 		if (cl == NULL)
331 			cl = drr_find_class(sch, res.classid);
332 		return cl;
333 	}
334 	return NULL;
335 }
336 
337 static int drr_enqueue(struct sk_buff *skb, struct Qdisc *sch,
338 		       struct sk_buff **to_free)
339 {
340 	struct drr_sched *q = qdisc_priv(sch);
341 	struct drr_class *cl;
342 	int err = 0;
343 
344 	cl = drr_classify(skb, sch, &err);
345 	if (cl == NULL) {
346 		if (err & __NET_XMIT_BYPASS)
347 			qdisc_qstats_drop(sch);
348 		__qdisc_drop(skb, to_free);
349 		return err;
350 	}
351 
352 	err = qdisc_enqueue(skb, cl->qdisc, to_free);
353 	if (unlikely(err != NET_XMIT_SUCCESS)) {
354 		if (net_xmit_drop_count(err)) {
355 			cl->qstats.drops++;
356 			qdisc_qstats_drop(sch);
357 		}
358 		return err;
359 	}
360 
361 	if (cl->qdisc->q.qlen == 1) {
362 		list_add_tail(&cl->alist, &q->active);
363 		cl->deficit = cl->quantum;
364 	}
365 
366 	qdisc_qstats_backlog_inc(sch, skb);
367 	sch->q.qlen++;
368 	return err;
369 }
370 
371 static struct sk_buff *drr_dequeue(struct Qdisc *sch)
372 {
373 	struct drr_sched *q = qdisc_priv(sch);
374 	struct drr_class *cl;
375 	struct sk_buff *skb;
376 	unsigned int len;
377 
378 	if (list_empty(&q->active))
379 		goto out;
380 	while (1) {
381 		cl = list_first_entry(&q->active, struct drr_class, alist);
382 		skb = cl->qdisc->ops->peek(cl->qdisc);
383 		if (skb == NULL) {
384 			qdisc_warn_nonwc(__func__, cl->qdisc);
385 			goto out;
386 		}
387 
388 		len = qdisc_pkt_len(skb);
389 		if (len <= cl->deficit) {
390 			cl->deficit -= len;
391 			skb = qdisc_dequeue_peeked(cl->qdisc);
392 			if (unlikely(skb == NULL))
393 				goto out;
394 			if (cl->qdisc->q.qlen == 0)
395 				list_del(&cl->alist);
396 
397 			bstats_update(&cl->bstats, skb);
398 			qdisc_bstats_update(sch, skb);
399 			qdisc_qstats_backlog_dec(sch, skb);
400 			sch->q.qlen--;
401 			return skb;
402 		}
403 
404 		cl->deficit += cl->quantum;
405 		list_move_tail(&cl->alist, &q->active);
406 	}
407 out:
408 	return NULL;
409 }
410 
411 static int drr_init_qdisc(struct Qdisc *sch, struct nlattr *opt)
412 {
413 	struct drr_sched *q = qdisc_priv(sch);
414 	int err;
415 
416 	err = tcf_block_get(&q->block, &q->filter_list, sch);
417 	if (err)
418 		return err;
419 	err = qdisc_class_hash_init(&q->clhash);
420 	if (err < 0)
421 		return err;
422 	INIT_LIST_HEAD(&q->active);
423 	return 0;
424 }
425 
426 static void drr_reset_qdisc(struct Qdisc *sch)
427 {
428 	struct drr_sched *q = qdisc_priv(sch);
429 	struct drr_class *cl;
430 	unsigned int i;
431 
432 	for (i = 0; i < q->clhash.hashsize; i++) {
433 		hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
434 			if (cl->qdisc->q.qlen)
435 				list_del(&cl->alist);
436 			qdisc_reset(cl->qdisc);
437 		}
438 	}
439 	sch->qstats.backlog = 0;
440 	sch->q.qlen = 0;
441 }
442 
443 static void drr_destroy_qdisc(struct Qdisc *sch)
444 {
445 	struct drr_sched *q = qdisc_priv(sch);
446 	struct drr_class *cl;
447 	struct hlist_node *next;
448 	unsigned int i;
449 
450 	tcf_block_put(q->block);
451 
452 	for (i = 0; i < q->clhash.hashsize; i++) {
453 		hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i],
454 					  common.hnode)
455 			drr_destroy_class(sch, cl);
456 	}
457 	qdisc_class_hash_destroy(&q->clhash);
458 }
459 
460 static const struct Qdisc_class_ops drr_class_ops = {
461 	.change		= drr_change_class,
462 	.delete		= drr_delete_class,
463 	.find		= drr_search_class,
464 	.tcf_block	= drr_tcf_block,
465 	.bind_tcf	= drr_bind_tcf,
466 	.unbind_tcf	= drr_unbind_tcf,
467 	.graft		= drr_graft_class,
468 	.leaf		= drr_class_leaf,
469 	.qlen_notify	= drr_qlen_notify,
470 	.dump		= drr_dump_class,
471 	.dump_stats	= drr_dump_class_stats,
472 	.walk		= drr_walk,
473 };
474 
475 static struct Qdisc_ops drr_qdisc_ops __read_mostly = {
476 	.cl_ops		= &drr_class_ops,
477 	.id		= "drr",
478 	.priv_size	= sizeof(struct drr_sched),
479 	.enqueue	= drr_enqueue,
480 	.dequeue	= drr_dequeue,
481 	.peek		= qdisc_peek_dequeued,
482 	.init		= drr_init_qdisc,
483 	.reset		= drr_reset_qdisc,
484 	.destroy	= drr_destroy_qdisc,
485 	.owner		= THIS_MODULE,
486 };
487 
488 static int __init drr_init(void)
489 {
490 	return register_qdisc(&drr_qdisc_ops);
491 }
492 
493 static void __exit drr_exit(void)
494 {
495 	unregister_qdisc(&drr_qdisc_ops);
496 }
497 
498 module_init(drr_init);
499 module_exit(drr_exit);
500 MODULE_LICENSE("GPL");
501