1 /*
2  * net/sched/sch_cbq.c	Class-Based Queueing discipline.
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  *
11  */
12 
13 #include <linux/module.h>
14 #include <linux/slab.h>
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/string.h>
18 #include <linux/errno.h>
19 #include <linux/skbuff.h>
20 #include <net/netlink.h>
21 #include <net/pkt_sched.h>
22 
23 
24 /*	Class-Based Queueing (CBQ) algorithm.
25 	=======================================
26 
27 	Sources: [1] Sally Floyd and Van Jacobson, "Link-sharing and Resource
28 		 Management Models for Packet Networks",
29 		 IEEE/ACM Transactions on Networking, Vol.3, No.4, 1995
30 
31 		 [2] Sally Floyd, "Notes on CBQ and Guaranteed Service", 1995
32 
33 		 [3] Sally Floyd, "Notes on Class-Based Queueing: Setting
34 		 Parameters", 1996
35 
36 		 [4] Sally Floyd and Michael Speer, "Experimental Results
37 		 for Class-Based Queueing", 1998, not published.
38 
39 	-----------------------------------------------------------------------
40 
41 	Algorithm skeleton was taken from NS simulator cbq.cc.
42 	If someone wants to check this code against the LBL version,
43 	he should take into account that ONLY the skeleton was borrowed,
44 	the implementation is different. Particularly:
45 
46 	--- The WRR algorithm is different. Our version looks more
47 	reasonable (I hope) and works when quanta are allowed to be
48 	less than MTU, which is always the case when real time classes
49 	have small rates. Note, that the statement of [3] is
50 	incomplete, delay may actually be estimated even if class
51 	per-round allotment is less than MTU. Namely, if per-round
52 	allotment is W*r_i, and r_1+...+r_k = r < 1
53 
54 	delay_i <= ([MTU/(W*r_i)]*W*r + W*r + k*MTU)/B
55 
56 	In the worst case we have IntServ estimate with D = W*r+k*MTU
57 	and C = MTU*r. The proof (if correct at all) is trivial.
58 
59 
60 	--- It seems that cbq-2.0 is not very accurate. At least, I cannot
61 	interpret some places, which look like wrong translations
62 	from NS. Anyone is advised to find these differences
63 	and explain to me, why I am wrong 8).
64 
65 	--- Linux has no EOI event, so that we cannot estimate true class
66 	idle time. Workaround is to consider the next dequeue event
67 	as sign that previous packet is finished. This is wrong because of
68 	internal device queueing, but on a permanently loaded link it is true.
69 	Moreover, combined with clock integrator, this scheme looks
70 	very close to an ideal solution.  */
71 
72 struct cbq_sched_data;
73 
74 
75 struct cbq_class {
76 	struct Qdisc_class_common common;
77 	struct cbq_class	*next_alive;	/* next class with backlog in this priority band */
78 
79 /* Parameters */
80 	unsigned char		priority;	/* class priority */
81 	unsigned char		priority2;	/* priority to be used after overlimit */
82 	unsigned char		ewma_log;	/* time constant for idle time calculation */
83 	unsigned char		ovl_strategy;
84 #ifdef CONFIG_NET_CLS_ACT
85 	unsigned char		police;
86 #endif
87 
88 	u32			defmap;
89 
90 	/* Link-sharing scheduler parameters */
91 	long			maxidle;	/* Class parameters: see below. */
92 	long			offtime;
93 	long			minidle;
94 	u32			avpkt;
95 	struct qdisc_rate_table	*R_tab;
96 
97 	/* Overlimit strategy parameters */
98 	void			(*overlimit)(struct cbq_class *cl);
99 	psched_tdiff_t		penalty;
100 
101 	/* General scheduler (WRR) parameters */
102 	long			allot;
103 	long			quantum;	/* Allotment per WRR round */
104 	long			weight;		/* Relative allotment: see below */
105 
106 	struct Qdisc		*qdisc;		/* Ptr to CBQ discipline */
107 	struct cbq_class	*split;		/* Ptr to split node */
108 	struct cbq_class	*share;		/* Ptr to LS parent in the class tree */
109 	struct cbq_class	*tparent;	/* Ptr to tree parent in the class tree */
110 	struct cbq_class	*borrow;	/* NULL if class is bandwidth limited;
111 						   parent otherwise */
112 	struct cbq_class	*sibling;	/* Sibling chain */
113 	struct cbq_class	*children;	/* Pointer to children chain */
114 
115 	struct Qdisc		*q;		/* Elementary queueing discipline */
116 
117 
118 /* Variables */
119 	unsigned char		cpriority;	/* Effective priority */
120 	unsigned char		delayed;
121 	unsigned char		level;		/* level of the class in hierarchy:
122 						   0 for leaf classes, and maximal
123 						   level of children + 1 for nodes.
124 						 */
125 
126 	psched_time_t		last;		/* Last end of service */
127 	psched_time_t		undertime;
128 	long			avgidle;
129 	long			deficit;	/* Saved deficit for WRR */
130 	psched_time_t		penalized;
131 	struct gnet_stats_basic_packed bstats;
132 	struct gnet_stats_queue qstats;
133 	struct gnet_stats_rate_est rate_est;
134 	struct tc_cbq_xstats	xstats;
135 
136 	struct tcf_proto	*filter_list;
137 
138 	int			refcnt;
139 	int			filters;
140 
141 	struct cbq_class	*defaults[TC_PRIO_MAX + 1];
142 };
143 
144 struct cbq_sched_data {
145 	struct Qdisc_class_hash	clhash;			/* Hash table of all classes */
146 	int			nclasses[TC_CBQ_MAXPRIO + 1];
147 	unsigned int		quanta[TC_CBQ_MAXPRIO + 1];
148 
149 	struct cbq_class	link;
150 
151 	unsigned int		activemask;
152 	struct cbq_class	*active[TC_CBQ_MAXPRIO + 1];	/* List of all classes
153 								   with backlog */
154 
155 #ifdef CONFIG_NET_CLS_ACT
156 	struct cbq_class	*rx_class;
157 #endif
158 	struct cbq_class	*tx_class;
159 	struct cbq_class	*tx_borrowed;
160 	int			tx_len;
161 	psched_time_t		now;		/* Cached timestamp */
162 	psched_time_t		now_rt;		/* Cached real time */
163 	unsigned int		pmask;
164 
165 	struct hrtimer		delay_timer;
166 	struct qdisc_watchdog	watchdog;	/* Watchdog timer,
167 						   started when CBQ has
168 						   backlog, but cannot
169 						   transmit just now */
170 	psched_tdiff_t		wd_expires;
171 	int			toplevel;
172 	u32			hgenerator;
173 };
174 
175 
176 #define L2T(cl, len)	qdisc_l2t((cl)->R_tab, len)
177 
178 static inline struct cbq_class *
cbq_class_lookup(struct cbq_sched_data * q,u32 classid)179 cbq_class_lookup(struct cbq_sched_data *q, u32 classid)
180 {
181 	struct Qdisc_class_common *clc;
182 
183 	clc = qdisc_class_find(&q->clhash, classid);
184 	if (clc == NULL)
185 		return NULL;
186 	return container_of(clc, struct cbq_class, common);
187 }
188 
189 #ifdef CONFIG_NET_CLS_ACT
190 
191 static struct cbq_class *
cbq_reclassify(struct sk_buff * skb,struct cbq_class * this)192 cbq_reclassify(struct sk_buff *skb, struct cbq_class *this)
193 {
194 	struct cbq_class *cl;
195 
196 	for (cl = this->tparent; cl; cl = cl->tparent) {
197 		struct cbq_class *new = cl->defaults[TC_PRIO_BESTEFFORT];
198 
199 		if (new != NULL && new != this)
200 			return new;
201 	}
202 	return NULL;
203 }
204 
205 #endif
206 
207 /* Classify packet. The procedure is pretty complicated, but
208  * it allows us to combine link sharing and priority scheduling
209  * transparently.
210  *
211  * Namely, you can put link sharing rules (f.e. route based) at root of CBQ,
212  * so that it resolves to split nodes. Then packets are classified
213  * by logical priority, or a more specific classifier may be attached
214  * to the split node.
215  */
216 
217 static struct cbq_class *
cbq_classify(struct sk_buff * skb,struct Qdisc * sch,int * qerr)218 cbq_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
219 {
220 	struct cbq_sched_data *q = qdisc_priv(sch);
221 	struct cbq_class *head = &q->link;
222 	struct cbq_class **defmap;
223 	struct cbq_class *cl = NULL;
224 	u32 prio = skb->priority;
225 	struct tcf_result res;
226 
227 	/*
228 	 *  Step 1. If skb->priority points to one of our classes, use it.
229 	 */
230 	if (TC_H_MAJ(prio ^ sch->handle) == 0 &&
231 	    (cl = cbq_class_lookup(q, prio)) != NULL)
232 		return cl;
233 
234 	*qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
235 	for (;;) {
236 		int result = 0;
237 		defmap = head->defaults;
238 
239 		/*
240 		 * Step 2+n. Apply classifier.
241 		 */
242 		if (!head->filter_list ||
243 		    (result = tc_classify_compat(skb, head->filter_list, &res)) < 0)
244 			goto fallback;
245 
246 		cl = (void *)res.class;
247 		if (!cl) {
248 			if (TC_H_MAJ(res.classid))
249 				cl = cbq_class_lookup(q, res.classid);
250 			else if ((cl = defmap[res.classid & TC_PRIO_MAX]) == NULL)
251 				cl = defmap[TC_PRIO_BESTEFFORT];
252 
253 			if (cl == NULL || cl->level >= head->level)
254 				goto fallback;
255 		}
256 
257 #ifdef CONFIG_NET_CLS_ACT
258 		switch (result) {
259 		case TC_ACT_QUEUED:
260 		case TC_ACT_STOLEN:
261 			*qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
262 		case TC_ACT_SHOT:
263 			return NULL;
264 		case TC_ACT_RECLASSIFY:
265 			return cbq_reclassify(skb, cl);
266 		}
267 #endif
268 		if (cl->level == 0)
269 			return cl;
270 
271 		/*
272 		 * Step 3+n. If classifier selected a link sharing class,
273 		 *	   apply agency specific classifier.
274 		 *	   Repeat this procdure until we hit a leaf node.
275 		 */
276 		head = cl;
277 	}
278 
279 fallback:
280 	cl = head;
281 
282 	/*
283 	 * Step 4. No success...
284 	 */
285 	if (TC_H_MAJ(prio) == 0 &&
286 	    !(cl = head->defaults[prio & TC_PRIO_MAX]) &&
287 	    !(cl = head->defaults[TC_PRIO_BESTEFFORT]))
288 		return head;
289 
290 	return cl;
291 }
292 
293 /*
294  * A packet has just been enqueued on the empty class.
295  * cbq_activate_class adds it to the tail of active class list
296  * of its priority band.
297  */
298 
cbq_activate_class(struct cbq_class * cl)299 static inline void cbq_activate_class(struct cbq_class *cl)
300 {
301 	struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
302 	int prio = cl->cpriority;
303 	struct cbq_class *cl_tail;
304 
305 	cl_tail = q->active[prio];
306 	q->active[prio] = cl;
307 
308 	if (cl_tail != NULL) {
309 		cl->next_alive = cl_tail->next_alive;
310 		cl_tail->next_alive = cl;
311 	} else {
312 		cl->next_alive = cl;
313 		q->activemask |= (1<<prio);
314 	}
315 }
316 
317 /*
318  * Unlink class from active chain.
319  * Note that this same procedure is done directly in cbq_dequeue*
320  * during round-robin procedure.
321  */
322 
cbq_deactivate_class(struct cbq_class * this)323 static void cbq_deactivate_class(struct cbq_class *this)
324 {
325 	struct cbq_sched_data *q = qdisc_priv(this->qdisc);
326 	int prio = this->cpriority;
327 	struct cbq_class *cl;
328 	struct cbq_class *cl_prev = q->active[prio];
329 
330 	do {
331 		cl = cl_prev->next_alive;
332 		if (cl == this) {
333 			cl_prev->next_alive = cl->next_alive;
334 			cl->next_alive = NULL;
335 
336 			if (cl == q->active[prio]) {
337 				q->active[prio] = cl_prev;
338 				if (cl == q->active[prio]) {
339 					q->active[prio] = NULL;
340 					q->activemask &= ~(1<<prio);
341 					return;
342 				}
343 			}
344 			return;
345 		}
346 	} while ((cl_prev = cl) != q->active[prio]);
347 }
348 
349 static void
cbq_mark_toplevel(struct cbq_sched_data * q,struct cbq_class * cl)350 cbq_mark_toplevel(struct cbq_sched_data *q, struct cbq_class *cl)
351 {
352 	int toplevel = q->toplevel;
353 
354 	if (toplevel > cl->level && !(qdisc_is_throttled(cl->q))) {
355 		psched_time_t now;
356 		psched_tdiff_t incr;
357 
358 		now = psched_get_time();
359 		incr = now - q->now_rt;
360 		now = q->now + incr;
361 
362 		do {
363 			if (cl->undertime < now) {
364 				q->toplevel = cl->level;
365 				return;
366 			}
367 		} while ((cl = cl->borrow) != NULL && toplevel > cl->level);
368 	}
369 }
370 
371 static int
cbq_enqueue(struct sk_buff * skb,struct Qdisc * sch)372 cbq_enqueue(struct sk_buff *skb, struct Qdisc *sch)
373 {
374 	struct cbq_sched_data *q = qdisc_priv(sch);
375 	int uninitialized_var(ret);
376 	struct cbq_class *cl = cbq_classify(skb, sch, &ret);
377 
378 #ifdef CONFIG_NET_CLS_ACT
379 	q->rx_class = cl;
380 #endif
381 	if (cl == NULL) {
382 		if (ret & __NET_XMIT_BYPASS)
383 			sch->qstats.drops++;
384 		kfree_skb(skb);
385 		return ret;
386 	}
387 
388 #ifdef CONFIG_NET_CLS_ACT
389 	cl->q->__parent = sch;
390 #endif
391 	ret = qdisc_enqueue(skb, cl->q);
392 	if (ret == NET_XMIT_SUCCESS) {
393 		sch->q.qlen++;
394 		cbq_mark_toplevel(q, cl);
395 		if (!cl->next_alive)
396 			cbq_activate_class(cl);
397 		return ret;
398 	}
399 
400 	if (net_xmit_drop_count(ret)) {
401 		sch->qstats.drops++;
402 		cbq_mark_toplevel(q, cl);
403 		cl->qstats.drops++;
404 	}
405 	return ret;
406 }
407 
408 /* Overlimit actions */
409 
410 /* TC_CBQ_OVL_CLASSIC: (default) penalize leaf class by adding offtime */
411 
cbq_ovl_classic(struct cbq_class * cl)412 static void cbq_ovl_classic(struct cbq_class *cl)
413 {
414 	struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
415 	psched_tdiff_t delay = cl->undertime - q->now;
416 
417 	if (!cl->delayed) {
418 		delay += cl->offtime;
419 
420 		/*
421 		 * Class goes to sleep, so that it will have no
422 		 * chance to work avgidle. Let's forgive it 8)
423 		 *
424 		 * BTW cbq-2.0 has a crap in this
425 		 * place, apparently they forgot to shift it by cl->ewma_log.
426 		 */
427 		if (cl->avgidle < 0)
428 			delay -= (-cl->avgidle) - ((-cl->avgidle) >> cl->ewma_log);
429 		if (cl->avgidle < cl->minidle)
430 			cl->avgidle = cl->minidle;
431 		if (delay <= 0)
432 			delay = 1;
433 		cl->undertime = q->now + delay;
434 
435 		cl->xstats.overactions++;
436 		cl->delayed = 1;
437 	}
438 	if (q->wd_expires == 0 || q->wd_expires > delay)
439 		q->wd_expires = delay;
440 
441 	/* Dirty work! We must schedule wakeups based on
442 	 * real available rate, rather than leaf rate,
443 	 * which may be tiny (even zero).
444 	 */
445 	if (q->toplevel == TC_CBQ_MAXLEVEL) {
446 		struct cbq_class *b;
447 		psched_tdiff_t base_delay = q->wd_expires;
448 
449 		for (b = cl->borrow; b; b = b->borrow) {
450 			delay = b->undertime - q->now;
451 			if (delay < base_delay) {
452 				if (delay <= 0)
453 					delay = 1;
454 				base_delay = delay;
455 			}
456 		}
457 
458 		q->wd_expires = base_delay;
459 	}
460 }
461 
462 /* TC_CBQ_OVL_RCLASSIC: penalize by offtime classes in hierarchy, when
463  * they go overlimit
464  */
465 
cbq_ovl_rclassic(struct cbq_class * cl)466 static void cbq_ovl_rclassic(struct cbq_class *cl)
467 {
468 	struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
469 	struct cbq_class *this = cl;
470 
471 	do {
472 		if (cl->level > q->toplevel) {
473 			cl = NULL;
474 			break;
475 		}
476 	} while ((cl = cl->borrow) != NULL);
477 
478 	if (cl == NULL)
479 		cl = this;
480 	cbq_ovl_classic(cl);
481 }
482 
483 /* TC_CBQ_OVL_DELAY: delay until it will go to underlimit */
484 
cbq_ovl_delay(struct cbq_class * cl)485 static void cbq_ovl_delay(struct cbq_class *cl)
486 {
487 	struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
488 	psched_tdiff_t delay = cl->undertime - q->now;
489 
490 	if (test_bit(__QDISC_STATE_DEACTIVATED,
491 		     &qdisc_root_sleeping(cl->qdisc)->state))
492 		return;
493 
494 	if (!cl->delayed) {
495 		psched_time_t sched = q->now;
496 		ktime_t expires;
497 
498 		delay += cl->offtime;
499 		if (cl->avgidle < 0)
500 			delay -= (-cl->avgidle) - ((-cl->avgidle) >> cl->ewma_log);
501 		if (cl->avgidle < cl->minidle)
502 			cl->avgidle = cl->minidle;
503 		cl->undertime = q->now + delay;
504 
505 		if (delay > 0) {
506 			sched += delay + cl->penalty;
507 			cl->penalized = sched;
508 			cl->cpriority = TC_CBQ_MAXPRIO;
509 			q->pmask |= (1<<TC_CBQ_MAXPRIO);
510 
511 			expires = ktime_set(0, 0);
512 			expires = ktime_add_ns(expires, PSCHED_TICKS2NS(sched));
513 			if (hrtimer_try_to_cancel(&q->delay_timer) &&
514 			    ktime_to_ns(ktime_sub(
515 					hrtimer_get_expires(&q->delay_timer),
516 					expires)) > 0)
517 				hrtimer_set_expires(&q->delay_timer, expires);
518 			hrtimer_restart(&q->delay_timer);
519 			cl->delayed = 1;
520 			cl->xstats.overactions++;
521 			return;
522 		}
523 		delay = 1;
524 	}
525 	if (q->wd_expires == 0 || q->wd_expires > delay)
526 		q->wd_expires = delay;
527 }
528 
529 /* TC_CBQ_OVL_LOWPRIO: penalize class by lowering its priority band */
530 
cbq_ovl_lowprio(struct cbq_class * cl)531 static void cbq_ovl_lowprio(struct cbq_class *cl)
532 {
533 	struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
534 
535 	cl->penalized = q->now + cl->penalty;
536 
537 	if (cl->cpriority != cl->priority2) {
538 		cl->cpriority = cl->priority2;
539 		q->pmask |= (1<<cl->cpriority);
540 		cl->xstats.overactions++;
541 	}
542 	cbq_ovl_classic(cl);
543 }
544 
545 /* TC_CBQ_OVL_DROP: penalize class by dropping */
546 
cbq_ovl_drop(struct cbq_class * cl)547 static void cbq_ovl_drop(struct cbq_class *cl)
548 {
549 	if (cl->q->ops->drop)
550 		if (cl->q->ops->drop(cl->q))
551 			cl->qdisc->q.qlen--;
552 	cl->xstats.overactions++;
553 	cbq_ovl_classic(cl);
554 }
555 
cbq_undelay_prio(struct cbq_sched_data * q,int prio,psched_time_t now)556 static psched_tdiff_t cbq_undelay_prio(struct cbq_sched_data *q, int prio,
557 				       psched_time_t now)
558 {
559 	struct cbq_class *cl;
560 	struct cbq_class *cl_prev = q->active[prio];
561 	psched_time_t sched = now;
562 
563 	if (cl_prev == NULL)
564 		return 0;
565 
566 	do {
567 		cl = cl_prev->next_alive;
568 		if (now - cl->penalized > 0) {
569 			cl_prev->next_alive = cl->next_alive;
570 			cl->next_alive = NULL;
571 			cl->cpriority = cl->priority;
572 			cl->delayed = 0;
573 			cbq_activate_class(cl);
574 
575 			if (cl == q->active[prio]) {
576 				q->active[prio] = cl_prev;
577 				if (cl == q->active[prio]) {
578 					q->active[prio] = NULL;
579 					return 0;
580 				}
581 			}
582 
583 			cl = cl_prev->next_alive;
584 		} else if (sched - cl->penalized > 0)
585 			sched = cl->penalized;
586 	} while ((cl_prev = cl) != q->active[prio]);
587 
588 	return sched - now;
589 }
590 
cbq_undelay(struct hrtimer * timer)591 static enum hrtimer_restart cbq_undelay(struct hrtimer *timer)
592 {
593 	struct cbq_sched_data *q = container_of(timer, struct cbq_sched_data,
594 						delay_timer);
595 	struct Qdisc *sch = q->watchdog.qdisc;
596 	psched_time_t now;
597 	psched_tdiff_t delay = 0;
598 	unsigned int pmask;
599 
600 	now = psched_get_time();
601 
602 	pmask = q->pmask;
603 	q->pmask = 0;
604 
605 	while (pmask) {
606 		int prio = ffz(~pmask);
607 		psched_tdiff_t tmp;
608 
609 		pmask &= ~(1<<prio);
610 
611 		tmp = cbq_undelay_prio(q, prio, now);
612 		if (tmp > 0) {
613 			q->pmask |= 1<<prio;
614 			if (tmp < delay || delay == 0)
615 				delay = tmp;
616 		}
617 	}
618 
619 	if (delay) {
620 		ktime_t time;
621 
622 		time = ktime_set(0, 0);
623 		time = ktime_add_ns(time, PSCHED_TICKS2NS(now + delay));
624 		hrtimer_start(&q->delay_timer, time, HRTIMER_MODE_ABS);
625 	}
626 
627 	qdisc_unthrottled(sch);
628 	__netif_schedule(qdisc_root(sch));
629 	return HRTIMER_NORESTART;
630 }
631 
632 #ifdef CONFIG_NET_CLS_ACT
cbq_reshape_fail(struct sk_buff * skb,struct Qdisc * child)633 static int cbq_reshape_fail(struct sk_buff *skb, struct Qdisc *child)
634 {
635 	struct Qdisc *sch = child->__parent;
636 	struct cbq_sched_data *q = qdisc_priv(sch);
637 	struct cbq_class *cl = q->rx_class;
638 
639 	q->rx_class = NULL;
640 
641 	if (cl && (cl = cbq_reclassify(skb, cl)) != NULL) {
642 		int ret;
643 
644 		cbq_mark_toplevel(q, cl);
645 
646 		q->rx_class = cl;
647 		cl->q->__parent = sch;
648 
649 		ret = qdisc_enqueue(skb, cl->q);
650 		if (ret == NET_XMIT_SUCCESS) {
651 			sch->q.qlen++;
652 			if (!cl->next_alive)
653 				cbq_activate_class(cl);
654 			return 0;
655 		}
656 		if (net_xmit_drop_count(ret))
657 			sch->qstats.drops++;
658 		return 0;
659 	}
660 
661 	sch->qstats.drops++;
662 	return -1;
663 }
664 #endif
665 
666 /*
667  * It is mission critical procedure.
668  *
669  * We "regenerate" toplevel cutoff, if transmitting class
670  * has backlog and it is not regulated. It is not part of
671  * original CBQ description, but looks more reasonable.
672  * Probably, it is wrong. This question needs further investigation.
673  */
674 
675 static inline void
cbq_update_toplevel(struct cbq_sched_data * q,struct cbq_class * cl,struct cbq_class * borrowed)676 cbq_update_toplevel(struct cbq_sched_data *q, struct cbq_class *cl,
677 		    struct cbq_class *borrowed)
678 {
679 	if (cl && q->toplevel >= borrowed->level) {
680 		if (cl->q->q.qlen > 1) {
681 			do {
682 				if (borrowed->undertime == PSCHED_PASTPERFECT) {
683 					q->toplevel = borrowed->level;
684 					return;
685 				}
686 			} while ((borrowed = borrowed->borrow) != NULL);
687 		}
688 #if 0
689 	/* It is not necessary now. Uncommenting it
690 	   will save CPU cycles, but decrease fairness.
691 	 */
692 		q->toplevel = TC_CBQ_MAXLEVEL;
693 #endif
694 	}
695 }
696 
697 static void
cbq_update(struct cbq_sched_data * q)698 cbq_update(struct cbq_sched_data *q)
699 {
700 	struct cbq_class *this = q->tx_class;
701 	struct cbq_class *cl = this;
702 	int len = q->tx_len;
703 
704 	q->tx_class = NULL;
705 
706 	for ( ; cl; cl = cl->share) {
707 		long avgidle = cl->avgidle;
708 		long idle;
709 
710 		cl->bstats.packets++;
711 		cl->bstats.bytes += len;
712 
713 		/*
714 		 * (now - last) is total time between packet right edges.
715 		 * (last_pktlen/rate) is "virtual" busy time, so that
716 		 *
717 		 *	idle = (now - last) - last_pktlen/rate
718 		 */
719 
720 		idle = q->now - cl->last;
721 		if ((unsigned long)idle > 128*1024*1024) {
722 			avgidle = cl->maxidle;
723 		} else {
724 			idle -= L2T(cl, len);
725 
726 		/* true_avgidle := (1-W)*true_avgidle + W*idle,
727 		 * where W=2^{-ewma_log}. But cl->avgidle is scaled:
728 		 * cl->avgidle == true_avgidle/W,
729 		 * hence:
730 		 */
731 			avgidle += idle - (avgidle>>cl->ewma_log);
732 		}
733 
734 		if (avgidle <= 0) {
735 			/* Overlimit or at-limit */
736 
737 			if (avgidle < cl->minidle)
738 				avgidle = cl->minidle;
739 
740 			cl->avgidle = avgidle;
741 
742 			/* Calculate expected time, when this class
743 			 * will be allowed to send.
744 			 * It will occur, when:
745 			 * (1-W)*true_avgidle + W*delay = 0, i.e.
746 			 * idle = (1/W - 1)*(-true_avgidle)
747 			 * or
748 			 * idle = (1 - W)*(-cl->avgidle);
749 			 */
750 			idle = (-avgidle) - ((-avgidle) >> cl->ewma_log);
751 
752 			/*
753 			 * That is not all.
754 			 * To maintain the rate allocated to the class,
755 			 * we add to undertime virtual clock,
756 			 * necessary to complete transmitted packet.
757 			 * (len/phys_bandwidth has been already passed
758 			 * to the moment of cbq_update)
759 			 */
760 
761 			idle -= L2T(&q->link, len);
762 			idle += L2T(cl, len);
763 
764 			cl->undertime = q->now + idle;
765 		} else {
766 			/* Underlimit */
767 
768 			cl->undertime = PSCHED_PASTPERFECT;
769 			if (avgidle > cl->maxidle)
770 				cl->avgidle = cl->maxidle;
771 			else
772 				cl->avgidle = avgidle;
773 		}
774 		cl->last = q->now;
775 	}
776 
777 	cbq_update_toplevel(q, this, q->tx_borrowed);
778 }
779 
780 static inline struct cbq_class *
cbq_under_limit(struct cbq_class * cl)781 cbq_under_limit(struct cbq_class *cl)
782 {
783 	struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
784 	struct cbq_class *this_cl = cl;
785 
786 	if (cl->tparent == NULL)
787 		return cl;
788 
789 	if (cl->undertime == PSCHED_PASTPERFECT || q->now >= cl->undertime) {
790 		cl->delayed = 0;
791 		return cl;
792 	}
793 
794 	do {
795 		/* It is very suspicious place. Now overlimit
796 		 * action is generated for not bounded classes
797 		 * only if link is completely congested.
798 		 * Though it is in agree with ancestor-only paradigm,
799 		 * it looks very stupid. Particularly,
800 		 * it means that this chunk of code will either
801 		 * never be called or result in strong amplification
802 		 * of burstiness. Dangerous, silly, and, however,
803 		 * no another solution exists.
804 		 */
805 		cl = cl->borrow;
806 		if (!cl) {
807 			this_cl->qstats.overlimits++;
808 			this_cl->overlimit(this_cl);
809 			return NULL;
810 		}
811 		if (cl->level > q->toplevel)
812 			return NULL;
813 	} while (cl->undertime != PSCHED_PASTPERFECT && q->now < cl->undertime);
814 
815 	cl->delayed = 0;
816 	return cl;
817 }
818 
819 static inline struct sk_buff *
cbq_dequeue_prio(struct Qdisc * sch,int prio)820 cbq_dequeue_prio(struct Qdisc *sch, int prio)
821 {
822 	struct cbq_sched_data *q = qdisc_priv(sch);
823 	struct cbq_class *cl_tail, *cl_prev, *cl;
824 	struct sk_buff *skb;
825 	int deficit;
826 
827 	cl_tail = cl_prev = q->active[prio];
828 	cl = cl_prev->next_alive;
829 
830 	do {
831 		deficit = 0;
832 
833 		/* Start round */
834 		do {
835 			struct cbq_class *borrow = cl;
836 
837 			if (cl->q->q.qlen &&
838 			    (borrow = cbq_under_limit(cl)) == NULL)
839 				goto skip_class;
840 
841 			if (cl->deficit <= 0) {
842 				/* Class exhausted its allotment per
843 				 * this round. Switch to the next one.
844 				 */
845 				deficit = 1;
846 				cl->deficit += cl->quantum;
847 				goto next_class;
848 			}
849 
850 			skb = cl->q->dequeue(cl->q);
851 
852 			/* Class did not give us any skb :-(
853 			 * It could occur even if cl->q->q.qlen != 0
854 			 * f.e. if cl->q == "tbf"
855 			 */
856 			if (skb == NULL)
857 				goto skip_class;
858 
859 			cl->deficit -= qdisc_pkt_len(skb);
860 			q->tx_class = cl;
861 			q->tx_borrowed = borrow;
862 			if (borrow != cl) {
863 #ifndef CBQ_XSTATS_BORROWS_BYTES
864 				borrow->xstats.borrows++;
865 				cl->xstats.borrows++;
866 #else
867 				borrow->xstats.borrows += qdisc_pkt_len(skb);
868 				cl->xstats.borrows += qdisc_pkt_len(skb);
869 #endif
870 			}
871 			q->tx_len = qdisc_pkt_len(skb);
872 
873 			if (cl->deficit <= 0) {
874 				q->active[prio] = cl;
875 				cl = cl->next_alive;
876 				cl->deficit += cl->quantum;
877 			}
878 			return skb;
879 
880 skip_class:
881 			if (cl->q->q.qlen == 0 || prio != cl->cpriority) {
882 				/* Class is empty or penalized.
883 				 * Unlink it from active chain.
884 				 */
885 				cl_prev->next_alive = cl->next_alive;
886 				cl->next_alive = NULL;
887 
888 				/* Did cl_tail point to it? */
889 				if (cl == cl_tail) {
890 					/* Repair it! */
891 					cl_tail = cl_prev;
892 
893 					/* Was it the last class in this band? */
894 					if (cl == cl_tail) {
895 						/* Kill the band! */
896 						q->active[prio] = NULL;
897 						q->activemask &= ~(1<<prio);
898 						if (cl->q->q.qlen)
899 							cbq_activate_class(cl);
900 						return NULL;
901 					}
902 
903 					q->active[prio] = cl_tail;
904 				}
905 				if (cl->q->q.qlen)
906 					cbq_activate_class(cl);
907 
908 				cl = cl_prev;
909 			}
910 
911 next_class:
912 			cl_prev = cl;
913 			cl = cl->next_alive;
914 		} while (cl_prev != cl_tail);
915 	} while (deficit);
916 
917 	q->active[prio] = cl_prev;
918 
919 	return NULL;
920 }
921 
922 static inline struct sk_buff *
cbq_dequeue_1(struct Qdisc * sch)923 cbq_dequeue_1(struct Qdisc *sch)
924 {
925 	struct cbq_sched_data *q = qdisc_priv(sch);
926 	struct sk_buff *skb;
927 	unsigned int activemask;
928 
929 	activemask = q->activemask & 0xFF;
930 	while (activemask) {
931 		int prio = ffz(~activemask);
932 		activemask &= ~(1<<prio);
933 		skb = cbq_dequeue_prio(sch, prio);
934 		if (skb)
935 			return skb;
936 	}
937 	return NULL;
938 }
939 
940 static struct sk_buff *
cbq_dequeue(struct Qdisc * sch)941 cbq_dequeue(struct Qdisc *sch)
942 {
943 	struct sk_buff *skb;
944 	struct cbq_sched_data *q = qdisc_priv(sch);
945 	psched_time_t now;
946 	psched_tdiff_t incr;
947 
948 	now = psched_get_time();
949 	incr = now - q->now_rt;
950 
951 	if (q->tx_class) {
952 		psched_tdiff_t incr2;
953 		/* Time integrator. We calculate EOS time
954 		 * by adding expected packet transmission time.
955 		 * If real time is greater, we warp artificial clock,
956 		 * so that:
957 		 *
958 		 * cbq_time = max(real_time, work);
959 		 */
960 		incr2 = L2T(&q->link, q->tx_len);
961 		q->now += incr2;
962 		cbq_update(q);
963 		if ((incr -= incr2) < 0)
964 			incr = 0;
965 	}
966 	q->now += incr;
967 	q->now_rt = now;
968 
969 	for (;;) {
970 		q->wd_expires = 0;
971 
972 		skb = cbq_dequeue_1(sch);
973 		if (skb) {
974 			qdisc_bstats_update(sch, skb);
975 			sch->q.qlen--;
976 			qdisc_unthrottled(sch);
977 			return skb;
978 		}
979 
980 		/* All the classes are overlimit.
981 		 *
982 		 * It is possible, if:
983 		 *
984 		 * 1. Scheduler is empty.
985 		 * 2. Toplevel cutoff inhibited borrowing.
986 		 * 3. Root class is overlimit.
987 		 *
988 		 * Reset 2d and 3d conditions and retry.
989 		 *
990 		 * Note, that NS and cbq-2.0 are buggy, peeking
991 		 * an arbitrary class is appropriate for ancestor-only
992 		 * sharing, but not for toplevel algorithm.
993 		 *
994 		 * Our version is better, but slower, because it requires
995 		 * two passes, but it is unavoidable with top-level sharing.
996 		 */
997 
998 		if (q->toplevel == TC_CBQ_MAXLEVEL &&
999 		    q->link.undertime == PSCHED_PASTPERFECT)
1000 			break;
1001 
1002 		q->toplevel = TC_CBQ_MAXLEVEL;
1003 		q->link.undertime = PSCHED_PASTPERFECT;
1004 	}
1005 
1006 	/* No packets in scheduler or nobody wants to give them to us :-(
1007 	 * Sigh... start watchdog timer in the last case.
1008 	 */
1009 
1010 	if (sch->q.qlen) {
1011 		sch->qstats.overlimits++;
1012 		if (q->wd_expires)
1013 			qdisc_watchdog_schedule(&q->watchdog,
1014 						now + q->wd_expires);
1015 	}
1016 	return NULL;
1017 }
1018 
1019 /* CBQ class maintanance routines */
1020 
cbq_adjust_levels(struct cbq_class * this)1021 static void cbq_adjust_levels(struct cbq_class *this)
1022 {
1023 	if (this == NULL)
1024 		return;
1025 
1026 	do {
1027 		int level = 0;
1028 		struct cbq_class *cl;
1029 
1030 		cl = this->children;
1031 		if (cl) {
1032 			do {
1033 				if (cl->level > level)
1034 					level = cl->level;
1035 			} while ((cl = cl->sibling) != this->children);
1036 		}
1037 		this->level = level + 1;
1038 	} while ((this = this->tparent) != NULL);
1039 }
1040 
cbq_normalize_quanta(struct cbq_sched_data * q,int prio)1041 static void cbq_normalize_quanta(struct cbq_sched_data *q, int prio)
1042 {
1043 	struct cbq_class *cl;
1044 	struct hlist_node *n;
1045 	unsigned int h;
1046 
1047 	if (q->quanta[prio] == 0)
1048 		return;
1049 
1050 	for (h = 0; h < q->clhash.hashsize; h++) {
1051 		hlist_for_each_entry(cl, n, &q->clhash.hash[h], common.hnode) {
1052 			/* BUGGGG... Beware! This expression suffer of
1053 			 * arithmetic overflows!
1054 			 */
1055 			if (cl->priority == prio) {
1056 				cl->quantum = (cl->weight*cl->allot*q->nclasses[prio])/
1057 					q->quanta[prio];
1058 			}
1059 			if (cl->quantum <= 0 || cl->quantum>32*qdisc_dev(cl->qdisc)->mtu) {
1060 				pr_warning("CBQ: class %08x has bad quantum==%ld, repaired.\n",
1061 					   cl->common.classid, cl->quantum);
1062 				cl->quantum = qdisc_dev(cl->qdisc)->mtu/2 + 1;
1063 			}
1064 		}
1065 	}
1066 }
1067 
cbq_sync_defmap(struct cbq_class * cl)1068 static void cbq_sync_defmap(struct cbq_class *cl)
1069 {
1070 	struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
1071 	struct cbq_class *split = cl->split;
1072 	unsigned int h;
1073 	int i;
1074 
1075 	if (split == NULL)
1076 		return;
1077 
1078 	for (i = 0; i <= TC_PRIO_MAX; i++) {
1079 		if (split->defaults[i] == cl && !(cl->defmap & (1<<i)))
1080 			split->defaults[i] = NULL;
1081 	}
1082 
1083 	for (i = 0; i <= TC_PRIO_MAX; i++) {
1084 		int level = split->level;
1085 
1086 		if (split->defaults[i])
1087 			continue;
1088 
1089 		for (h = 0; h < q->clhash.hashsize; h++) {
1090 			struct hlist_node *n;
1091 			struct cbq_class *c;
1092 
1093 			hlist_for_each_entry(c, n, &q->clhash.hash[h],
1094 					     common.hnode) {
1095 				if (c->split == split && c->level < level &&
1096 				    c->defmap & (1<<i)) {
1097 					split->defaults[i] = c;
1098 					level = c->level;
1099 				}
1100 			}
1101 		}
1102 	}
1103 }
1104 
cbq_change_defmap(struct cbq_class * cl,u32 splitid,u32 def,u32 mask)1105 static void cbq_change_defmap(struct cbq_class *cl, u32 splitid, u32 def, u32 mask)
1106 {
1107 	struct cbq_class *split = NULL;
1108 
1109 	if (splitid == 0) {
1110 		split = cl->split;
1111 		if (!split)
1112 			return;
1113 		splitid = split->common.classid;
1114 	}
1115 
1116 	if (split == NULL || split->common.classid != splitid) {
1117 		for (split = cl->tparent; split; split = split->tparent)
1118 			if (split->common.classid == splitid)
1119 				break;
1120 	}
1121 
1122 	if (split == NULL)
1123 		return;
1124 
1125 	if (cl->split != split) {
1126 		cl->defmap = 0;
1127 		cbq_sync_defmap(cl);
1128 		cl->split = split;
1129 		cl->defmap = def & mask;
1130 	} else
1131 		cl->defmap = (cl->defmap & ~mask) | (def & mask);
1132 
1133 	cbq_sync_defmap(cl);
1134 }
1135 
cbq_unlink_class(struct cbq_class * this)1136 static void cbq_unlink_class(struct cbq_class *this)
1137 {
1138 	struct cbq_class *cl, **clp;
1139 	struct cbq_sched_data *q = qdisc_priv(this->qdisc);
1140 
1141 	qdisc_class_hash_remove(&q->clhash, &this->common);
1142 
1143 	if (this->tparent) {
1144 		clp = &this->sibling;
1145 		cl = *clp;
1146 		do {
1147 			if (cl == this) {
1148 				*clp = cl->sibling;
1149 				break;
1150 			}
1151 			clp = &cl->sibling;
1152 		} while ((cl = *clp) != this->sibling);
1153 
1154 		if (this->tparent->children == this) {
1155 			this->tparent->children = this->sibling;
1156 			if (this->sibling == this)
1157 				this->tparent->children = NULL;
1158 		}
1159 	} else {
1160 		WARN_ON(this->sibling != this);
1161 	}
1162 }
1163 
cbq_link_class(struct cbq_class * this)1164 static void cbq_link_class(struct cbq_class *this)
1165 {
1166 	struct cbq_sched_data *q = qdisc_priv(this->qdisc);
1167 	struct cbq_class *parent = this->tparent;
1168 
1169 	this->sibling = this;
1170 	qdisc_class_hash_insert(&q->clhash, &this->common);
1171 
1172 	if (parent == NULL)
1173 		return;
1174 
1175 	if (parent->children == NULL) {
1176 		parent->children = this;
1177 	} else {
1178 		this->sibling = parent->children->sibling;
1179 		parent->children->sibling = this;
1180 	}
1181 }
1182 
cbq_drop(struct Qdisc * sch)1183 static unsigned int cbq_drop(struct Qdisc *sch)
1184 {
1185 	struct cbq_sched_data *q = qdisc_priv(sch);
1186 	struct cbq_class *cl, *cl_head;
1187 	int prio;
1188 	unsigned int len;
1189 
1190 	for (prio = TC_CBQ_MAXPRIO; prio >= 0; prio--) {
1191 		cl_head = q->active[prio];
1192 		if (!cl_head)
1193 			continue;
1194 
1195 		cl = cl_head;
1196 		do {
1197 			if (cl->q->ops->drop && (len = cl->q->ops->drop(cl->q))) {
1198 				sch->q.qlen--;
1199 				if (!cl->q->q.qlen)
1200 					cbq_deactivate_class(cl);
1201 				return len;
1202 			}
1203 		} while ((cl = cl->next_alive) != cl_head);
1204 	}
1205 	return 0;
1206 }
1207 
1208 static void
cbq_reset(struct Qdisc * sch)1209 cbq_reset(struct Qdisc *sch)
1210 {
1211 	struct cbq_sched_data *q = qdisc_priv(sch);
1212 	struct cbq_class *cl;
1213 	struct hlist_node *n;
1214 	int prio;
1215 	unsigned int h;
1216 
1217 	q->activemask = 0;
1218 	q->pmask = 0;
1219 	q->tx_class = NULL;
1220 	q->tx_borrowed = NULL;
1221 	qdisc_watchdog_cancel(&q->watchdog);
1222 	hrtimer_cancel(&q->delay_timer);
1223 	q->toplevel = TC_CBQ_MAXLEVEL;
1224 	q->now = psched_get_time();
1225 	q->now_rt = q->now;
1226 
1227 	for (prio = 0; prio <= TC_CBQ_MAXPRIO; prio++)
1228 		q->active[prio] = NULL;
1229 
1230 	for (h = 0; h < q->clhash.hashsize; h++) {
1231 		hlist_for_each_entry(cl, n, &q->clhash.hash[h], common.hnode) {
1232 			qdisc_reset(cl->q);
1233 
1234 			cl->next_alive = NULL;
1235 			cl->undertime = PSCHED_PASTPERFECT;
1236 			cl->avgidle = cl->maxidle;
1237 			cl->deficit = cl->quantum;
1238 			cl->cpriority = cl->priority;
1239 		}
1240 	}
1241 	sch->q.qlen = 0;
1242 }
1243 
1244 
cbq_set_lss(struct cbq_class * cl,struct tc_cbq_lssopt * lss)1245 static int cbq_set_lss(struct cbq_class *cl, struct tc_cbq_lssopt *lss)
1246 {
1247 	if (lss->change & TCF_CBQ_LSS_FLAGS) {
1248 		cl->share = (lss->flags & TCF_CBQ_LSS_ISOLATED) ? NULL : cl->tparent;
1249 		cl->borrow = (lss->flags & TCF_CBQ_LSS_BOUNDED) ? NULL : cl->tparent;
1250 	}
1251 	if (lss->change & TCF_CBQ_LSS_EWMA)
1252 		cl->ewma_log = lss->ewma_log;
1253 	if (lss->change & TCF_CBQ_LSS_AVPKT)
1254 		cl->avpkt = lss->avpkt;
1255 	if (lss->change & TCF_CBQ_LSS_MINIDLE)
1256 		cl->minidle = -(long)lss->minidle;
1257 	if (lss->change & TCF_CBQ_LSS_MAXIDLE) {
1258 		cl->maxidle = lss->maxidle;
1259 		cl->avgidle = lss->maxidle;
1260 	}
1261 	if (lss->change & TCF_CBQ_LSS_OFFTIME)
1262 		cl->offtime = lss->offtime;
1263 	return 0;
1264 }
1265 
cbq_rmprio(struct cbq_sched_data * q,struct cbq_class * cl)1266 static void cbq_rmprio(struct cbq_sched_data *q, struct cbq_class *cl)
1267 {
1268 	q->nclasses[cl->priority]--;
1269 	q->quanta[cl->priority] -= cl->weight;
1270 	cbq_normalize_quanta(q, cl->priority);
1271 }
1272 
cbq_addprio(struct cbq_sched_data * q,struct cbq_class * cl)1273 static void cbq_addprio(struct cbq_sched_data *q, struct cbq_class *cl)
1274 {
1275 	q->nclasses[cl->priority]++;
1276 	q->quanta[cl->priority] += cl->weight;
1277 	cbq_normalize_quanta(q, cl->priority);
1278 }
1279 
cbq_set_wrr(struct cbq_class * cl,struct tc_cbq_wrropt * wrr)1280 static int cbq_set_wrr(struct cbq_class *cl, struct tc_cbq_wrropt *wrr)
1281 {
1282 	struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
1283 
1284 	if (wrr->allot)
1285 		cl->allot = wrr->allot;
1286 	if (wrr->weight)
1287 		cl->weight = wrr->weight;
1288 	if (wrr->priority) {
1289 		cl->priority = wrr->priority - 1;
1290 		cl->cpriority = cl->priority;
1291 		if (cl->priority >= cl->priority2)
1292 			cl->priority2 = TC_CBQ_MAXPRIO - 1;
1293 	}
1294 
1295 	cbq_addprio(q, cl);
1296 	return 0;
1297 }
1298 
cbq_set_overlimit(struct cbq_class * cl,struct tc_cbq_ovl * ovl)1299 static int cbq_set_overlimit(struct cbq_class *cl, struct tc_cbq_ovl *ovl)
1300 {
1301 	switch (ovl->strategy) {
1302 	case TC_CBQ_OVL_CLASSIC:
1303 		cl->overlimit = cbq_ovl_classic;
1304 		break;
1305 	case TC_CBQ_OVL_DELAY:
1306 		cl->overlimit = cbq_ovl_delay;
1307 		break;
1308 	case TC_CBQ_OVL_LOWPRIO:
1309 		if (ovl->priority2 - 1 >= TC_CBQ_MAXPRIO ||
1310 		    ovl->priority2 - 1 <= cl->priority)
1311 			return -EINVAL;
1312 		cl->priority2 = ovl->priority2 - 1;
1313 		cl->overlimit = cbq_ovl_lowprio;
1314 		break;
1315 	case TC_CBQ_OVL_DROP:
1316 		cl->overlimit = cbq_ovl_drop;
1317 		break;
1318 	case TC_CBQ_OVL_RCLASSIC:
1319 		cl->overlimit = cbq_ovl_rclassic;
1320 		break;
1321 	default:
1322 		return -EINVAL;
1323 	}
1324 	cl->penalty = ovl->penalty;
1325 	return 0;
1326 }
1327 
1328 #ifdef CONFIG_NET_CLS_ACT
cbq_set_police(struct cbq_class * cl,struct tc_cbq_police * p)1329 static int cbq_set_police(struct cbq_class *cl, struct tc_cbq_police *p)
1330 {
1331 	cl->police = p->police;
1332 
1333 	if (cl->q->handle) {
1334 		if (p->police == TC_POLICE_RECLASSIFY)
1335 			cl->q->reshape_fail = cbq_reshape_fail;
1336 		else
1337 			cl->q->reshape_fail = NULL;
1338 	}
1339 	return 0;
1340 }
1341 #endif
1342 
cbq_set_fopt(struct cbq_class * cl,struct tc_cbq_fopt * fopt)1343 static int cbq_set_fopt(struct cbq_class *cl, struct tc_cbq_fopt *fopt)
1344 {
1345 	cbq_change_defmap(cl, fopt->split, fopt->defmap, fopt->defchange);
1346 	return 0;
1347 }
1348 
1349 static const struct nla_policy cbq_policy[TCA_CBQ_MAX + 1] = {
1350 	[TCA_CBQ_LSSOPT]	= { .len = sizeof(struct tc_cbq_lssopt) },
1351 	[TCA_CBQ_WRROPT]	= { .len = sizeof(struct tc_cbq_wrropt) },
1352 	[TCA_CBQ_FOPT]		= { .len = sizeof(struct tc_cbq_fopt) },
1353 	[TCA_CBQ_OVL_STRATEGY]	= { .len = sizeof(struct tc_cbq_ovl) },
1354 	[TCA_CBQ_RATE]		= { .len = sizeof(struct tc_ratespec) },
1355 	[TCA_CBQ_RTAB]		= { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
1356 	[TCA_CBQ_POLICE]	= { .len = sizeof(struct tc_cbq_police) },
1357 };
1358 
cbq_init(struct Qdisc * sch,struct nlattr * opt)1359 static int cbq_init(struct Qdisc *sch, struct nlattr *opt)
1360 {
1361 	struct cbq_sched_data *q = qdisc_priv(sch);
1362 	struct nlattr *tb[TCA_CBQ_MAX + 1];
1363 	struct tc_ratespec *r;
1364 	int err;
1365 
1366 	err = nla_parse_nested(tb, TCA_CBQ_MAX, opt, cbq_policy);
1367 	if (err < 0)
1368 		return err;
1369 
1370 	if (tb[TCA_CBQ_RTAB] == NULL || tb[TCA_CBQ_RATE] == NULL)
1371 		return -EINVAL;
1372 
1373 	r = nla_data(tb[TCA_CBQ_RATE]);
1374 
1375 	if ((q->link.R_tab = qdisc_get_rtab(r, tb[TCA_CBQ_RTAB])) == NULL)
1376 		return -EINVAL;
1377 
1378 	err = qdisc_class_hash_init(&q->clhash);
1379 	if (err < 0)
1380 		goto put_rtab;
1381 
1382 	q->link.refcnt = 1;
1383 	q->link.sibling = &q->link;
1384 	q->link.common.classid = sch->handle;
1385 	q->link.qdisc = sch;
1386 	q->link.q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
1387 				      sch->handle);
1388 	if (!q->link.q)
1389 		q->link.q = &noop_qdisc;
1390 
1391 	q->link.priority = TC_CBQ_MAXPRIO - 1;
1392 	q->link.priority2 = TC_CBQ_MAXPRIO - 1;
1393 	q->link.cpriority = TC_CBQ_MAXPRIO - 1;
1394 	q->link.ovl_strategy = TC_CBQ_OVL_CLASSIC;
1395 	q->link.overlimit = cbq_ovl_classic;
1396 	q->link.allot = psched_mtu(qdisc_dev(sch));
1397 	q->link.quantum = q->link.allot;
1398 	q->link.weight = q->link.R_tab->rate.rate;
1399 
1400 	q->link.ewma_log = TC_CBQ_DEF_EWMA;
1401 	q->link.avpkt = q->link.allot/2;
1402 	q->link.minidle = -0x7FFFFFFF;
1403 
1404 	qdisc_watchdog_init(&q->watchdog, sch);
1405 	hrtimer_init(&q->delay_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1406 	q->delay_timer.function = cbq_undelay;
1407 	q->toplevel = TC_CBQ_MAXLEVEL;
1408 	q->now = psched_get_time();
1409 	q->now_rt = q->now;
1410 
1411 	cbq_link_class(&q->link);
1412 
1413 	if (tb[TCA_CBQ_LSSOPT])
1414 		cbq_set_lss(&q->link, nla_data(tb[TCA_CBQ_LSSOPT]));
1415 
1416 	cbq_addprio(q, &q->link);
1417 	return 0;
1418 
1419 put_rtab:
1420 	qdisc_put_rtab(q->link.R_tab);
1421 	return err;
1422 }
1423 
cbq_dump_rate(struct sk_buff * skb,struct cbq_class * cl)1424 static int cbq_dump_rate(struct sk_buff *skb, struct cbq_class *cl)
1425 {
1426 	unsigned char *b = skb_tail_pointer(skb);
1427 
1428 	NLA_PUT(skb, TCA_CBQ_RATE, sizeof(cl->R_tab->rate), &cl->R_tab->rate);
1429 	return skb->len;
1430 
1431 nla_put_failure:
1432 	nlmsg_trim(skb, b);
1433 	return -1;
1434 }
1435 
cbq_dump_lss(struct sk_buff * skb,struct cbq_class * cl)1436 static int cbq_dump_lss(struct sk_buff *skb, struct cbq_class *cl)
1437 {
1438 	unsigned char *b = skb_tail_pointer(skb);
1439 	struct tc_cbq_lssopt opt;
1440 
1441 	opt.flags = 0;
1442 	if (cl->borrow == NULL)
1443 		opt.flags |= TCF_CBQ_LSS_BOUNDED;
1444 	if (cl->share == NULL)
1445 		opt.flags |= TCF_CBQ_LSS_ISOLATED;
1446 	opt.ewma_log = cl->ewma_log;
1447 	opt.level = cl->level;
1448 	opt.avpkt = cl->avpkt;
1449 	opt.maxidle = cl->maxidle;
1450 	opt.minidle = (u32)(-cl->minidle);
1451 	opt.offtime = cl->offtime;
1452 	opt.change = ~0;
1453 	NLA_PUT(skb, TCA_CBQ_LSSOPT, sizeof(opt), &opt);
1454 	return skb->len;
1455 
1456 nla_put_failure:
1457 	nlmsg_trim(skb, b);
1458 	return -1;
1459 }
1460 
cbq_dump_wrr(struct sk_buff * skb,struct cbq_class * cl)1461 static int cbq_dump_wrr(struct sk_buff *skb, struct cbq_class *cl)
1462 {
1463 	unsigned char *b = skb_tail_pointer(skb);
1464 	struct tc_cbq_wrropt opt;
1465 
1466 	opt.flags = 0;
1467 	opt.allot = cl->allot;
1468 	opt.priority = cl->priority + 1;
1469 	opt.cpriority = cl->cpriority + 1;
1470 	opt.weight = cl->weight;
1471 	NLA_PUT(skb, TCA_CBQ_WRROPT, sizeof(opt), &opt);
1472 	return skb->len;
1473 
1474 nla_put_failure:
1475 	nlmsg_trim(skb, b);
1476 	return -1;
1477 }
1478 
cbq_dump_ovl(struct sk_buff * skb,struct cbq_class * cl)1479 static int cbq_dump_ovl(struct sk_buff *skb, struct cbq_class *cl)
1480 {
1481 	unsigned char *b = skb_tail_pointer(skb);
1482 	struct tc_cbq_ovl opt;
1483 
1484 	opt.strategy = cl->ovl_strategy;
1485 	opt.priority2 = cl->priority2 + 1;
1486 	opt.pad = 0;
1487 	opt.penalty = cl->penalty;
1488 	NLA_PUT(skb, TCA_CBQ_OVL_STRATEGY, sizeof(opt), &opt);
1489 	return skb->len;
1490 
1491 nla_put_failure:
1492 	nlmsg_trim(skb, b);
1493 	return -1;
1494 }
1495 
cbq_dump_fopt(struct sk_buff * skb,struct cbq_class * cl)1496 static int cbq_dump_fopt(struct sk_buff *skb, struct cbq_class *cl)
1497 {
1498 	unsigned char *b = skb_tail_pointer(skb);
1499 	struct tc_cbq_fopt opt;
1500 
1501 	if (cl->split || cl->defmap) {
1502 		opt.split = cl->split ? cl->split->common.classid : 0;
1503 		opt.defmap = cl->defmap;
1504 		opt.defchange = ~0;
1505 		NLA_PUT(skb, TCA_CBQ_FOPT, sizeof(opt), &opt);
1506 	}
1507 	return skb->len;
1508 
1509 nla_put_failure:
1510 	nlmsg_trim(skb, b);
1511 	return -1;
1512 }
1513 
1514 #ifdef CONFIG_NET_CLS_ACT
cbq_dump_police(struct sk_buff * skb,struct cbq_class * cl)1515 static int cbq_dump_police(struct sk_buff *skb, struct cbq_class *cl)
1516 {
1517 	unsigned char *b = skb_tail_pointer(skb);
1518 	struct tc_cbq_police opt;
1519 
1520 	if (cl->police) {
1521 		opt.police = cl->police;
1522 		opt.__res1 = 0;
1523 		opt.__res2 = 0;
1524 		NLA_PUT(skb, TCA_CBQ_POLICE, sizeof(opt), &opt);
1525 	}
1526 	return skb->len;
1527 
1528 nla_put_failure:
1529 	nlmsg_trim(skb, b);
1530 	return -1;
1531 }
1532 #endif
1533 
cbq_dump_attr(struct sk_buff * skb,struct cbq_class * cl)1534 static int cbq_dump_attr(struct sk_buff *skb, struct cbq_class *cl)
1535 {
1536 	if (cbq_dump_lss(skb, cl) < 0 ||
1537 	    cbq_dump_rate(skb, cl) < 0 ||
1538 	    cbq_dump_wrr(skb, cl) < 0 ||
1539 	    cbq_dump_ovl(skb, cl) < 0 ||
1540 #ifdef CONFIG_NET_CLS_ACT
1541 	    cbq_dump_police(skb, cl) < 0 ||
1542 #endif
1543 	    cbq_dump_fopt(skb, cl) < 0)
1544 		return -1;
1545 	return 0;
1546 }
1547 
cbq_dump(struct Qdisc * sch,struct sk_buff * skb)1548 static int cbq_dump(struct Qdisc *sch, struct sk_buff *skb)
1549 {
1550 	struct cbq_sched_data *q = qdisc_priv(sch);
1551 	struct nlattr *nest;
1552 
1553 	nest = nla_nest_start(skb, TCA_OPTIONS);
1554 	if (nest == NULL)
1555 		goto nla_put_failure;
1556 	if (cbq_dump_attr(skb, &q->link) < 0)
1557 		goto nla_put_failure;
1558 	nla_nest_end(skb, nest);
1559 	return skb->len;
1560 
1561 nla_put_failure:
1562 	nla_nest_cancel(skb, nest);
1563 	return -1;
1564 }
1565 
1566 static int
cbq_dump_stats(struct Qdisc * sch,struct gnet_dump * d)1567 cbq_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
1568 {
1569 	struct cbq_sched_data *q = qdisc_priv(sch);
1570 
1571 	q->link.xstats.avgidle = q->link.avgidle;
1572 	return gnet_stats_copy_app(d, &q->link.xstats, sizeof(q->link.xstats));
1573 }
1574 
1575 static int
cbq_dump_class(struct Qdisc * sch,unsigned long arg,struct sk_buff * skb,struct tcmsg * tcm)1576 cbq_dump_class(struct Qdisc *sch, unsigned long arg,
1577 	       struct sk_buff *skb, struct tcmsg *tcm)
1578 {
1579 	struct cbq_class *cl = (struct cbq_class *)arg;
1580 	struct nlattr *nest;
1581 
1582 	if (cl->tparent)
1583 		tcm->tcm_parent = cl->tparent->common.classid;
1584 	else
1585 		tcm->tcm_parent = TC_H_ROOT;
1586 	tcm->tcm_handle = cl->common.classid;
1587 	tcm->tcm_info = cl->q->handle;
1588 
1589 	nest = nla_nest_start(skb, TCA_OPTIONS);
1590 	if (nest == NULL)
1591 		goto nla_put_failure;
1592 	if (cbq_dump_attr(skb, cl) < 0)
1593 		goto nla_put_failure;
1594 	nla_nest_end(skb, nest);
1595 	return skb->len;
1596 
1597 nla_put_failure:
1598 	nla_nest_cancel(skb, nest);
1599 	return -1;
1600 }
1601 
1602 static int
cbq_dump_class_stats(struct Qdisc * sch,unsigned long arg,struct gnet_dump * d)1603 cbq_dump_class_stats(struct Qdisc *sch, unsigned long arg,
1604 	struct gnet_dump *d)
1605 {
1606 	struct cbq_sched_data *q = qdisc_priv(sch);
1607 	struct cbq_class *cl = (struct cbq_class *)arg;
1608 
1609 	cl->qstats.qlen = cl->q->q.qlen;
1610 	cl->xstats.avgidle = cl->avgidle;
1611 	cl->xstats.undertime = 0;
1612 
1613 	if (cl->undertime != PSCHED_PASTPERFECT)
1614 		cl->xstats.undertime = cl->undertime - q->now;
1615 
1616 	if (gnet_stats_copy_basic(d, &cl->bstats) < 0 ||
1617 	    gnet_stats_copy_rate_est(d, &cl->bstats, &cl->rate_est) < 0 ||
1618 	    gnet_stats_copy_queue(d, &cl->qstats) < 0)
1619 		return -1;
1620 
1621 	return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats));
1622 }
1623 
cbq_graft(struct Qdisc * sch,unsigned long arg,struct Qdisc * new,struct Qdisc ** old)1624 static int cbq_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1625 		     struct Qdisc **old)
1626 {
1627 	struct cbq_class *cl = (struct cbq_class *)arg;
1628 
1629 	if (new == NULL) {
1630 		new = qdisc_create_dflt(sch->dev_queue,
1631 					&pfifo_qdisc_ops, cl->common.classid);
1632 		if (new == NULL)
1633 			return -ENOBUFS;
1634 	} else {
1635 #ifdef CONFIG_NET_CLS_ACT
1636 		if (cl->police == TC_POLICE_RECLASSIFY)
1637 			new->reshape_fail = cbq_reshape_fail;
1638 #endif
1639 	}
1640 	sch_tree_lock(sch);
1641 	*old = cl->q;
1642 	cl->q = new;
1643 	qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
1644 	qdisc_reset(*old);
1645 	sch_tree_unlock(sch);
1646 
1647 	return 0;
1648 }
1649 
cbq_leaf(struct Qdisc * sch,unsigned long arg)1650 static struct Qdisc *cbq_leaf(struct Qdisc *sch, unsigned long arg)
1651 {
1652 	struct cbq_class *cl = (struct cbq_class *)arg;
1653 
1654 	return cl->q;
1655 }
1656 
cbq_qlen_notify(struct Qdisc * sch,unsigned long arg)1657 static void cbq_qlen_notify(struct Qdisc *sch, unsigned long arg)
1658 {
1659 	struct cbq_class *cl = (struct cbq_class *)arg;
1660 
1661 	if (cl->q->q.qlen == 0)
1662 		cbq_deactivate_class(cl);
1663 }
1664 
cbq_get(struct Qdisc * sch,u32 classid)1665 static unsigned long cbq_get(struct Qdisc *sch, u32 classid)
1666 {
1667 	struct cbq_sched_data *q = qdisc_priv(sch);
1668 	struct cbq_class *cl = cbq_class_lookup(q, classid);
1669 
1670 	if (cl) {
1671 		cl->refcnt++;
1672 		return (unsigned long)cl;
1673 	}
1674 	return 0;
1675 }
1676 
cbq_destroy_class(struct Qdisc * sch,struct cbq_class * cl)1677 static void cbq_destroy_class(struct Qdisc *sch, struct cbq_class *cl)
1678 {
1679 	struct cbq_sched_data *q = qdisc_priv(sch);
1680 
1681 	WARN_ON(cl->filters);
1682 
1683 	tcf_destroy_chain(&cl->filter_list);
1684 	qdisc_destroy(cl->q);
1685 	qdisc_put_rtab(cl->R_tab);
1686 	gen_kill_estimator(&cl->bstats, &cl->rate_est);
1687 	if (cl != &q->link)
1688 		kfree(cl);
1689 }
1690 
cbq_destroy(struct Qdisc * sch)1691 static void cbq_destroy(struct Qdisc *sch)
1692 {
1693 	struct cbq_sched_data *q = qdisc_priv(sch);
1694 	struct hlist_node *n, *next;
1695 	struct cbq_class *cl;
1696 	unsigned int h;
1697 
1698 #ifdef CONFIG_NET_CLS_ACT
1699 	q->rx_class = NULL;
1700 #endif
1701 	/*
1702 	 * Filters must be destroyed first because we don't destroy the
1703 	 * classes from root to leafs which means that filters can still
1704 	 * be bound to classes which have been destroyed already. --TGR '04
1705 	 */
1706 	for (h = 0; h < q->clhash.hashsize; h++) {
1707 		hlist_for_each_entry(cl, n, &q->clhash.hash[h], common.hnode)
1708 			tcf_destroy_chain(&cl->filter_list);
1709 	}
1710 	for (h = 0; h < q->clhash.hashsize; h++) {
1711 		hlist_for_each_entry_safe(cl, n, next, &q->clhash.hash[h],
1712 					  common.hnode)
1713 			cbq_destroy_class(sch, cl);
1714 	}
1715 	qdisc_class_hash_destroy(&q->clhash);
1716 }
1717 
cbq_put(struct Qdisc * sch,unsigned long arg)1718 static void cbq_put(struct Qdisc *sch, unsigned long arg)
1719 {
1720 	struct cbq_class *cl = (struct cbq_class *)arg;
1721 
1722 	if (--cl->refcnt == 0) {
1723 #ifdef CONFIG_NET_CLS_ACT
1724 		spinlock_t *root_lock = qdisc_root_sleeping_lock(sch);
1725 		struct cbq_sched_data *q = qdisc_priv(sch);
1726 
1727 		spin_lock_bh(root_lock);
1728 		if (q->rx_class == cl)
1729 			q->rx_class = NULL;
1730 		spin_unlock_bh(root_lock);
1731 #endif
1732 
1733 		cbq_destroy_class(sch, cl);
1734 	}
1735 }
1736 
1737 static int
cbq_change_class(struct Qdisc * sch,u32 classid,u32 parentid,struct nlattr ** tca,unsigned long * arg)1738 cbq_change_class(struct Qdisc *sch, u32 classid, u32 parentid, struct nlattr **tca,
1739 		 unsigned long *arg)
1740 {
1741 	int err;
1742 	struct cbq_sched_data *q = qdisc_priv(sch);
1743 	struct cbq_class *cl = (struct cbq_class *)*arg;
1744 	struct nlattr *opt = tca[TCA_OPTIONS];
1745 	struct nlattr *tb[TCA_CBQ_MAX + 1];
1746 	struct cbq_class *parent;
1747 	struct qdisc_rate_table *rtab = NULL;
1748 
1749 	if (opt == NULL)
1750 		return -EINVAL;
1751 
1752 	err = nla_parse_nested(tb, TCA_CBQ_MAX, opt, cbq_policy);
1753 	if (err < 0)
1754 		return err;
1755 
1756 	if (cl) {
1757 		/* Check parent */
1758 		if (parentid) {
1759 			if (cl->tparent &&
1760 			    cl->tparent->common.classid != parentid)
1761 				return -EINVAL;
1762 			if (!cl->tparent && parentid != TC_H_ROOT)
1763 				return -EINVAL;
1764 		}
1765 
1766 		if (tb[TCA_CBQ_RATE]) {
1767 			rtab = qdisc_get_rtab(nla_data(tb[TCA_CBQ_RATE]),
1768 					      tb[TCA_CBQ_RTAB]);
1769 			if (rtab == NULL)
1770 				return -EINVAL;
1771 		}
1772 
1773 		if (tca[TCA_RATE]) {
1774 			err = gen_replace_estimator(&cl->bstats, &cl->rate_est,
1775 						    qdisc_root_sleeping_lock(sch),
1776 						    tca[TCA_RATE]);
1777 			if (err) {
1778 				if (rtab)
1779 					qdisc_put_rtab(rtab);
1780 				return err;
1781 			}
1782 		}
1783 
1784 		/* Change class parameters */
1785 		sch_tree_lock(sch);
1786 
1787 		if (cl->next_alive != NULL)
1788 			cbq_deactivate_class(cl);
1789 
1790 		if (rtab) {
1791 			qdisc_put_rtab(cl->R_tab);
1792 			cl->R_tab = rtab;
1793 		}
1794 
1795 		if (tb[TCA_CBQ_LSSOPT])
1796 			cbq_set_lss(cl, nla_data(tb[TCA_CBQ_LSSOPT]));
1797 
1798 		if (tb[TCA_CBQ_WRROPT]) {
1799 			cbq_rmprio(q, cl);
1800 			cbq_set_wrr(cl, nla_data(tb[TCA_CBQ_WRROPT]));
1801 		}
1802 
1803 		if (tb[TCA_CBQ_OVL_STRATEGY])
1804 			cbq_set_overlimit(cl, nla_data(tb[TCA_CBQ_OVL_STRATEGY]));
1805 
1806 #ifdef CONFIG_NET_CLS_ACT
1807 		if (tb[TCA_CBQ_POLICE])
1808 			cbq_set_police(cl, nla_data(tb[TCA_CBQ_POLICE]));
1809 #endif
1810 
1811 		if (tb[TCA_CBQ_FOPT])
1812 			cbq_set_fopt(cl, nla_data(tb[TCA_CBQ_FOPT]));
1813 
1814 		if (cl->q->q.qlen)
1815 			cbq_activate_class(cl);
1816 
1817 		sch_tree_unlock(sch);
1818 
1819 		return 0;
1820 	}
1821 
1822 	if (parentid == TC_H_ROOT)
1823 		return -EINVAL;
1824 
1825 	if (tb[TCA_CBQ_WRROPT] == NULL || tb[TCA_CBQ_RATE] == NULL ||
1826 	    tb[TCA_CBQ_LSSOPT] == NULL)
1827 		return -EINVAL;
1828 
1829 	rtab = qdisc_get_rtab(nla_data(tb[TCA_CBQ_RATE]), tb[TCA_CBQ_RTAB]);
1830 	if (rtab == NULL)
1831 		return -EINVAL;
1832 
1833 	if (classid) {
1834 		err = -EINVAL;
1835 		if (TC_H_MAJ(classid ^ sch->handle) ||
1836 		    cbq_class_lookup(q, classid))
1837 			goto failure;
1838 	} else {
1839 		int i;
1840 		classid = TC_H_MAKE(sch->handle, 0x8000);
1841 
1842 		for (i = 0; i < 0x8000; i++) {
1843 			if (++q->hgenerator >= 0x8000)
1844 				q->hgenerator = 1;
1845 			if (cbq_class_lookup(q, classid|q->hgenerator) == NULL)
1846 				break;
1847 		}
1848 		err = -ENOSR;
1849 		if (i >= 0x8000)
1850 			goto failure;
1851 		classid = classid|q->hgenerator;
1852 	}
1853 
1854 	parent = &q->link;
1855 	if (parentid) {
1856 		parent = cbq_class_lookup(q, parentid);
1857 		err = -EINVAL;
1858 		if (parent == NULL)
1859 			goto failure;
1860 	}
1861 
1862 	err = -ENOBUFS;
1863 	cl = kzalloc(sizeof(*cl), GFP_KERNEL);
1864 	if (cl == NULL)
1865 		goto failure;
1866 
1867 	if (tca[TCA_RATE]) {
1868 		err = gen_new_estimator(&cl->bstats, &cl->rate_est,
1869 					qdisc_root_sleeping_lock(sch),
1870 					tca[TCA_RATE]);
1871 		if (err) {
1872 			kfree(cl);
1873 			goto failure;
1874 		}
1875 	}
1876 
1877 	cl->R_tab = rtab;
1878 	rtab = NULL;
1879 	cl->refcnt = 1;
1880 	cl->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, classid);
1881 	if (!cl->q)
1882 		cl->q = &noop_qdisc;
1883 	cl->common.classid = classid;
1884 	cl->tparent = parent;
1885 	cl->qdisc = sch;
1886 	cl->allot = parent->allot;
1887 	cl->quantum = cl->allot;
1888 	cl->weight = cl->R_tab->rate.rate;
1889 
1890 	sch_tree_lock(sch);
1891 	cbq_link_class(cl);
1892 	cl->borrow = cl->tparent;
1893 	if (cl->tparent != &q->link)
1894 		cl->share = cl->tparent;
1895 	cbq_adjust_levels(parent);
1896 	cl->minidle = -0x7FFFFFFF;
1897 	cbq_set_lss(cl, nla_data(tb[TCA_CBQ_LSSOPT]));
1898 	cbq_set_wrr(cl, nla_data(tb[TCA_CBQ_WRROPT]));
1899 	if (cl->ewma_log == 0)
1900 		cl->ewma_log = q->link.ewma_log;
1901 	if (cl->maxidle == 0)
1902 		cl->maxidle = q->link.maxidle;
1903 	if (cl->avpkt == 0)
1904 		cl->avpkt = q->link.avpkt;
1905 	cl->overlimit = cbq_ovl_classic;
1906 	if (tb[TCA_CBQ_OVL_STRATEGY])
1907 		cbq_set_overlimit(cl, nla_data(tb[TCA_CBQ_OVL_STRATEGY]));
1908 #ifdef CONFIG_NET_CLS_ACT
1909 	if (tb[TCA_CBQ_POLICE])
1910 		cbq_set_police(cl, nla_data(tb[TCA_CBQ_POLICE]));
1911 #endif
1912 	if (tb[TCA_CBQ_FOPT])
1913 		cbq_set_fopt(cl, nla_data(tb[TCA_CBQ_FOPT]));
1914 	sch_tree_unlock(sch);
1915 
1916 	qdisc_class_hash_grow(sch, &q->clhash);
1917 
1918 	*arg = (unsigned long)cl;
1919 	return 0;
1920 
1921 failure:
1922 	qdisc_put_rtab(rtab);
1923 	return err;
1924 }
1925 
cbq_delete(struct Qdisc * sch,unsigned long arg)1926 static int cbq_delete(struct Qdisc *sch, unsigned long arg)
1927 {
1928 	struct cbq_sched_data *q = qdisc_priv(sch);
1929 	struct cbq_class *cl = (struct cbq_class *)arg;
1930 	unsigned int qlen;
1931 
1932 	if (cl->filters || cl->children || cl == &q->link)
1933 		return -EBUSY;
1934 
1935 	sch_tree_lock(sch);
1936 
1937 	qlen = cl->q->q.qlen;
1938 	qdisc_reset(cl->q);
1939 	qdisc_tree_decrease_qlen(cl->q, qlen);
1940 
1941 	if (cl->next_alive)
1942 		cbq_deactivate_class(cl);
1943 
1944 	if (q->tx_borrowed == cl)
1945 		q->tx_borrowed = q->tx_class;
1946 	if (q->tx_class == cl) {
1947 		q->tx_class = NULL;
1948 		q->tx_borrowed = NULL;
1949 	}
1950 #ifdef CONFIG_NET_CLS_ACT
1951 	if (q->rx_class == cl)
1952 		q->rx_class = NULL;
1953 #endif
1954 
1955 	cbq_unlink_class(cl);
1956 	cbq_adjust_levels(cl->tparent);
1957 	cl->defmap = 0;
1958 	cbq_sync_defmap(cl);
1959 
1960 	cbq_rmprio(q, cl);
1961 	sch_tree_unlock(sch);
1962 
1963 	BUG_ON(--cl->refcnt == 0);
1964 	/*
1965 	 * This shouldn't happen: we "hold" one cops->get() when called
1966 	 * from tc_ctl_tclass; the destroy method is done from cops->put().
1967 	 */
1968 
1969 	return 0;
1970 }
1971 
cbq_find_tcf(struct Qdisc * sch,unsigned long arg)1972 static struct tcf_proto **cbq_find_tcf(struct Qdisc *sch, unsigned long arg)
1973 {
1974 	struct cbq_sched_data *q = qdisc_priv(sch);
1975 	struct cbq_class *cl = (struct cbq_class *)arg;
1976 
1977 	if (cl == NULL)
1978 		cl = &q->link;
1979 
1980 	return &cl->filter_list;
1981 }
1982 
cbq_bind_filter(struct Qdisc * sch,unsigned long parent,u32 classid)1983 static unsigned long cbq_bind_filter(struct Qdisc *sch, unsigned long parent,
1984 				     u32 classid)
1985 {
1986 	struct cbq_sched_data *q = qdisc_priv(sch);
1987 	struct cbq_class *p = (struct cbq_class *)parent;
1988 	struct cbq_class *cl = cbq_class_lookup(q, classid);
1989 
1990 	if (cl) {
1991 		if (p && p->level <= cl->level)
1992 			return 0;
1993 		cl->filters++;
1994 		return (unsigned long)cl;
1995 	}
1996 	return 0;
1997 }
1998 
cbq_unbind_filter(struct Qdisc * sch,unsigned long arg)1999 static void cbq_unbind_filter(struct Qdisc *sch, unsigned long arg)
2000 {
2001 	struct cbq_class *cl = (struct cbq_class *)arg;
2002 
2003 	cl->filters--;
2004 }
2005 
cbq_walk(struct Qdisc * sch,struct qdisc_walker * arg)2006 static void cbq_walk(struct Qdisc *sch, struct qdisc_walker *arg)
2007 {
2008 	struct cbq_sched_data *q = qdisc_priv(sch);
2009 	struct cbq_class *cl;
2010 	struct hlist_node *n;
2011 	unsigned int h;
2012 
2013 	if (arg->stop)
2014 		return;
2015 
2016 	for (h = 0; h < q->clhash.hashsize; h++) {
2017 		hlist_for_each_entry(cl, n, &q->clhash.hash[h], common.hnode) {
2018 			if (arg->count < arg->skip) {
2019 				arg->count++;
2020 				continue;
2021 			}
2022 			if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
2023 				arg->stop = 1;
2024 				return;
2025 			}
2026 			arg->count++;
2027 		}
2028 	}
2029 }
2030 
2031 static const struct Qdisc_class_ops cbq_class_ops = {
2032 	.graft		=	cbq_graft,
2033 	.leaf		=	cbq_leaf,
2034 	.qlen_notify	=	cbq_qlen_notify,
2035 	.get		=	cbq_get,
2036 	.put		=	cbq_put,
2037 	.change		=	cbq_change_class,
2038 	.delete		=	cbq_delete,
2039 	.walk		=	cbq_walk,
2040 	.tcf_chain	=	cbq_find_tcf,
2041 	.bind_tcf	=	cbq_bind_filter,
2042 	.unbind_tcf	=	cbq_unbind_filter,
2043 	.dump		=	cbq_dump_class,
2044 	.dump_stats	=	cbq_dump_class_stats,
2045 };
2046 
2047 static struct Qdisc_ops cbq_qdisc_ops __read_mostly = {
2048 	.next		=	NULL,
2049 	.cl_ops		=	&cbq_class_ops,
2050 	.id		=	"cbq",
2051 	.priv_size	=	sizeof(struct cbq_sched_data),
2052 	.enqueue	=	cbq_enqueue,
2053 	.dequeue	=	cbq_dequeue,
2054 	.peek		=	qdisc_peek_dequeued,
2055 	.drop		=	cbq_drop,
2056 	.init		=	cbq_init,
2057 	.reset		=	cbq_reset,
2058 	.destroy	=	cbq_destroy,
2059 	.change		=	NULL,
2060 	.dump		=	cbq_dump,
2061 	.dump_stats	=	cbq_dump_stats,
2062 	.owner		=	THIS_MODULE,
2063 };
2064 
cbq_module_init(void)2065 static int __init cbq_module_init(void)
2066 {
2067 	return register_qdisc(&cbq_qdisc_ops);
2068 }
cbq_module_exit(void)2069 static void __exit cbq_module_exit(void)
2070 {
2071 	unregister_qdisc(&cbq_qdisc_ops);
2072 }
2073 module_init(cbq_module_init)
2074 module_exit(cbq_module_exit)
2075 MODULE_LICENSE("GPL");
2076