xref: /openbmc/linux/net/sched/sch_htb.c (revision 90099433)
1 /*
2  * net/sched/sch_htb.c	Hierarchical token bucket, feed tree version
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:	Martin Devera, <devik@cdi.cz>
10  *
11  * Credits (in time order) for older HTB versions:
12  *              Stef Coene <stef.coene@docum.org>
13  *			HTB support at LARTC mailing list
14  *		Ondrej Kraus, <krauso@barr.cz>
15  *			found missing INIT_QDISC(htb)
16  *		Vladimir Smelhaus, Aamer Akhter, Bert Hubert
17  *			helped a lot to locate nasty class stall bug
18  *		Andi Kleen, Jamal Hadi, Bert Hubert
19  *			code review and helpful comments on shaping
20  *		Tomasz Wrona, <tw@eter.tym.pl>
21  *			created test case so that I was able to fix nasty bug
22  *		Wilfried Weissmann
23  *			spotted bug in dequeue code and helped with fix
24  *		Jiri Fojtasek
25  *			fixed requeue routine
26  *		and many others. thanks.
27  */
28 #include <linux/module.h>
29 #include <linux/moduleparam.h>
30 #include <linux/types.h>
31 #include <linux/kernel.h>
32 #include <linux/string.h>
33 #include <linux/errno.h>
34 #include <linux/skbuff.h>
35 #include <linux/list.h>
36 #include <linux/compiler.h>
37 #include <linux/rbtree.h>
38 #include <linux/workqueue.h>
39 #include <linux/slab.h>
40 #include <net/netlink.h>
41 #include <net/pkt_sched.h>
42 
43 /* HTB algorithm.
44     Author: devik@cdi.cz
45     ========================================================================
46     HTB is like TBF with multiple classes. It is also similar to CBQ because
47     it allows to assign priority to each class in hierarchy.
48     In fact it is another implementation of Floyd's formal sharing.
49 
50     Levels:
51     Each class is assigned level. Leaf has ALWAYS level 0 and root
52     classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level
53     one less than their parent.
54 */
55 
56 static int htb_hysteresis __read_mostly = 0; /* whether to use mode hysteresis for speedup */
57 #define HTB_VER 0x30011		/* major must be matched with number suplied by TC as version */
58 
59 #if HTB_VER >> 16 != TC_HTB_PROTOVER
60 #error "Mismatched sch_htb.c and pkt_sch.h"
61 #endif
62 
63 /* Module parameter and sysfs export */
64 module_param    (htb_hysteresis, int, 0640);
65 MODULE_PARM_DESC(htb_hysteresis, "Hysteresis mode, less CPU load, less accurate");
66 
67 /* used internaly to keep status of single class */
68 enum htb_cmode {
69 	HTB_CANT_SEND,		/* class can't send and can't borrow */
70 	HTB_MAY_BORROW,		/* class can't send but may borrow */
71 	HTB_CAN_SEND		/* class can send */
72 };
73 
74 /* interior & leaf nodes; props specific to leaves are marked L: */
75 struct htb_class {
76 	struct Qdisc_class_common common;
77 	/* general class parameters */
78 	struct gnet_stats_basic_packed bstats;
79 	struct gnet_stats_queue qstats;
80 	struct gnet_stats_rate_est rate_est;
81 	struct tc_htb_xstats xstats;	/* our special stats */
82 	int refcnt;		/* usage count of this class */
83 
84 	/* topology */
85 	int level;		/* our level (see above) */
86 	unsigned int children;
87 	struct htb_class *parent;	/* parent class */
88 
89 	int prio;		/* these two are used only by leaves... */
90 	int quantum;		/* but stored for parent-to-leaf return */
91 
92 	union {
93 		struct htb_class_leaf {
94 			struct Qdisc *q;
95 			int deficit[TC_HTB_MAXDEPTH];
96 			struct list_head drop_list;
97 		} leaf;
98 		struct htb_class_inner {
99 			struct rb_root feed[TC_HTB_NUMPRIO];	/* feed trees */
100 			struct rb_node *ptr[TC_HTB_NUMPRIO];	/* current class ptr */
101 			/* When class changes from state 1->2 and disconnects from
102 			 * parent's feed then we lost ptr value and start from the
103 			 * first child again. Here we store classid of the
104 			 * last valid ptr (used when ptr is NULL).
105 			 */
106 			u32 last_ptr_id[TC_HTB_NUMPRIO];
107 		} inner;
108 	} un;
109 	struct rb_node node[TC_HTB_NUMPRIO];	/* node for self or feed tree */
110 	struct rb_node pq_node;	/* node for event queue */
111 	psched_time_t pq_key;
112 
113 	int prio_activity;	/* for which prios are we active */
114 	enum htb_cmode cmode;	/* current mode of the class */
115 
116 	/* class attached filters */
117 	struct tcf_proto *filter_list;
118 	int filter_cnt;
119 
120 	/* token bucket parameters */
121 	struct qdisc_rate_table *rate;	/* rate table of the class itself */
122 	struct qdisc_rate_table *ceil;	/* ceiling rate (limits borrows too) */
123 	long buffer, cbuffer;	/* token bucket depth/rate */
124 	psched_tdiff_t mbuffer;	/* max wait time */
125 	long tokens, ctokens;	/* current number of tokens */
126 	psched_time_t t_c;	/* checkpoint time */
127 };
128 
129 struct htb_sched {
130 	struct Qdisc_class_hash clhash;
131 	struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */
132 
133 	/* self list - roots of self generating tree */
134 	struct rb_root row[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO];
135 	int row_mask[TC_HTB_MAXDEPTH];
136 	struct rb_node *ptr[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO];
137 	u32 last_ptr_id[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO];
138 
139 	/* self wait list - roots of wait PQs per row */
140 	struct rb_root wait_pq[TC_HTB_MAXDEPTH];
141 
142 	/* time of nearest event per level (row) */
143 	psched_time_t near_ev_cache[TC_HTB_MAXDEPTH];
144 
145 	int defcls;		/* class where unclassified flows go to */
146 
147 	/* filters for qdisc itself */
148 	struct tcf_proto *filter_list;
149 
150 	int rate2quantum;	/* quant = rate / rate2quantum */
151 	psched_time_t now;	/* cached dequeue time */
152 	struct qdisc_watchdog watchdog;
153 
154 	/* non shaped skbs; let them go directly thru */
155 	struct sk_buff_head direct_queue;
156 	int direct_qlen;	/* max qlen of above */
157 
158 	long direct_pkts;
159 
160 #define HTB_WARN_TOOMANYEVENTS	0x1
161 	unsigned int warned;	/* only one warning */
162 	struct work_struct work;
163 };
164 
165 /* find class in global hash table using given handle */
166 static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch)
167 {
168 	struct htb_sched *q = qdisc_priv(sch);
169 	struct Qdisc_class_common *clc;
170 
171 	clc = qdisc_class_find(&q->clhash, handle);
172 	if (clc == NULL)
173 		return NULL;
174 	return container_of(clc, struct htb_class, common);
175 }
176 
177 /**
178  * htb_classify - classify a packet into class
179  *
180  * It returns NULL if the packet should be dropped or -1 if the packet
181  * should be passed directly thru. In all other cases leaf class is returned.
182  * We allow direct class selection by classid in priority. The we examine
183  * filters in qdisc and in inner nodes (if higher filter points to the inner
184  * node). If we end up with classid MAJOR:0 we enqueue the skb into special
185  * internal fifo (direct). These packets then go directly thru. If we still
186  * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessful
187  * then finish and return direct queue.
188  */
189 #define HTB_DIRECT ((struct htb_class *)-1L)
190 
191 static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch,
192 				      int *qerr)
193 {
194 	struct htb_sched *q = qdisc_priv(sch);
195 	struct htb_class *cl;
196 	struct tcf_result res;
197 	struct tcf_proto *tcf;
198 	int result;
199 
200 	/* allow to select class by setting skb->priority to valid classid;
201 	 * note that nfmark can be used too by attaching filter fw with no
202 	 * rules in it
203 	 */
204 	if (skb->priority == sch->handle)
205 		return HTB_DIRECT;	/* X:0 (direct flow) selected */
206 	cl = htb_find(skb->priority, sch);
207 	if (cl && cl->level == 0)
208 		return cl;
209 
210 	*qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
211 	tcf = q->filter_list;
212 	while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) {
213 #ifdef CONFIG_NET_CLS_ACT
214 		switch (result) {
215 		case TC_ACT_QUEUED:
216 		case TC_ACT_STOLEN:
217 			*qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
218 		case TC_ACT_SHOT:
219 			return NULL;
220 		}
221 #endif
222 		cl = (void *)res.class;
223 		if (!cl) {
224 			if (res.classid == sch->handle)
225 				return HTB_DIRECT;	/* X:0 (direct flow) */
226 			cl = htb_find(res.classid, sch);
227 			if (!cl)
228 				break;	/* filter selected invalid classid */
229 		}
230 		if (!cl->level)
231 			return cl;	/* we hit leaf; return it */
232 
233 		/* we have got inner class; apply inner filter chain */
234 		tcf = cl->filter_list;
235 	}
236 	/* classification failed; try to use default class */
237 	cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch);
238 	if (!cl || cl->level)
239 		return HTB_DIRECT;	/* bad default .. this is safe bet */
240 	return cl;
241 }
242 
243 /**
244  * htb_add_to_id_tree - adds class to the round robin list
245  *
246  * Routine adds class to the list (actually tree) sorted by classid.
247  * Make sure that class is not already on such list for given prio.
248  */
249 static void htb_add_to_id_tree(struct rb_root *root,
250 			       struct htb_class *cl, int prio)
251 {
252 	struct rb_node **p = &root->rb_node, *parent = NULL;
253 
254 	while (*p) {
255 		struct htb_class *c;
256 		parent = *p;
257 		c = rb_entry(parent, struct htb_class, node[prio]);
258 
259 		if (cl->common.classid > c->common.classid)
260 			p = &parent->rb_right;
261 		else
262 			p = &parent->rb_left;
263 	}
264 	rb_link_node(&cl->node[prio], parent, p);
265 	rb_insert_color(&cl->node[prio], root);
266 }
267 
268 /**
269  * htb_add_to_wait_tree - adds class to the event queue with delay
270  *
271  * The class is added to priority event queue to indicate that class will
272  * change its mode in cl->pq_key microseconds. Make sure that class is not
273  * already in the queue.
274  */
275 static void htb_add_to_wait_tree(struct htb_sched *q,
276 				 struct htb_class *cl, long delay)
277 {
278 	struct rb_node **p = &q->wait_pq[cl->level].rb_node, *parent = NULL;
279 
280 	cl->pq_key = q->now + delay;
281 	if (cl->pq_key == q->now)
282 		cl->pq_key++;
283 
284 	/* update the nearest event cache */
285 	if (q->near_ev_cache[cl->level] > cl->pq_key)
286 		q->near_ev_cache[cl->level] = cl->pq_key;
287 
288 	while (*p) {
289 		struct htb_class *c;
290 		parent = *p;
291 		c = rb_entry(parent, struct htb_class, pq_node);
292 		if (cl->pq_key >= c->pq_key)
293 			p = &parent->rb_right;
294 		else
295 			p = &parent->rb_left;
296 	}
297 	rb_link_node(&cl->pq_node, parent, p);
298 	rb_insert_color(&cl->pq_node, &q->wait_pq[cl->level]);
299 }
300 
301 /**
302  * htb_next_rb_node - finds next node in binary tree
303  *
304  * When we are past last key we return NULL.
305  * Average complexity is 2 steps per call.
306  */
307 static inline void htb_next_rb_node(struct rb_node **n)
308 {
309 	*n = rb_next(*n);
310 }
311 
312 /**
313  * htb_add_class_to_row - add class to its row
314  *
315  * The class is added to row at priorities marked in mask.
316  * It does nothing if mask == 0.
317  */
318 static inline void htb_add_class_to_row(struct htb_sched *q,
319 					struct htb_class *cl, int mask)
320 {
321 	q->row_mask[cl->level] |= mask;
322 	while (mask) {
323 		int prio = ffz(~mask);
324 		mask &= ~(1 << prio);
325 		htb_add_to_id_tree(q->row[cl->level] + prio, cl, prio);
326 	}
327 }
328 
329 /* If this triggers, it is a bug in this code, but it need not be fatal */
330 static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root)
331 {
332 	if (RB_EMPTY_NODE(rb)) {
333 		WARN_ON(1);
334 	} else {
335 		rb_erase(rb, root);
336 		RB_CLEAR_NODE(rb);
337 	}
338 }
339 
340 
341 /**
342  * htb_remove_class_from_row - removes class from its row
343  *
344  * The class is removed from row at priorities marked in mask.
345  * It does nothing if mask == 0.
346  */
347 static inline void htb_remove_class_from_row(struct htb_sched *q,
348 						 struct htb_class *cl, int mask)
349 {
350 	int m = 0;
351 
352 	while (mask) {
353 		int prio = ffz(~mask);
354 
355 		mask &= ~(1 << prio);
356 		if (q->ptr[cl->level][prio] == cl->node + prio)
357 			htb_next_rb_node(q->ptr[cl->level] + prio);
358 
359 		htb_safe_rb_erase(cl->node + prio, q->row[cl->level] + prio);
360 		if (!q->row[cl->level][prio].rb_node)
361 			m |= 1 << prio;
362 	}
363 	q->row_mask[cl->level] &= ~m;
364 }
365 
366 /**
367  * htb_activate_prios - creates active classe's feed chain
368  *
369  * The class is connected to ancestors and/or appropriate rows
370  * for priorities it is participating on. cl->cmode must be new
371  * (activated) mode. It does nothing if cl->prio_activity == 0.
372  */
373 static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl)
374 {
375 	struct htb_class *p = cl->parent;
376 	long m, mask = cl->prio_activity;
377 
378 	while (cl->cmode == HTB_MAY_BORROW && p && mask) {
379 		m = mask;
380 		while (m) {
381 			int prio = ffz(~m);
382 			m &= ~(1 << prio);
383 
384 			if (p->un.inner.feed[prio].rb_node)
385 				/* parent already has its feed in use so that
386 				 * reset bit in mask as parent is already ok
387 				 */
388 				mask &= ~(1 << prio);
389 
390 			htb_add_to_id_tree(p->un.inner.feed + prio, cl, prio);
391 		}
392 		p->prio_activity |= mask;
393 		cl = p;
394 		p = cl->parent;
395 
396 	}
397 	if (cl->cmode == HTB_CAN_SEND && mask)
398 		htb_add_class_to_row(q, cl, mask);
399 }
400 
401 /**
402  * htb_deactivate_prios - remove class from feed chain
403  *
404  * cl->cmode must represent old mode (before deactivation). It does
405  * nothing if cl->prio_activity == 0. Class is removed from all feed
406  * chains and rows.
407  */
408 static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl)
409 {
410 	struct htb_class *p = cl->parent;
411 	long m, mask = cl->prio_activity;
412 
413 	while (cl->cmode == HTB_MAY_BORROW && p && mask) {
414 		m = mask;
415 		mask = 0;
416 		while (m) {
417 			int prio = ffz(~m);
418 			m &= ~(1 << prio);
419 
420 			if (p->un.inner.ptr[prio] == cl->node + prio) {
421 				/* we are removing child which is pointed to from
422 				 * parent feed - forget the pointer but remember
423 				 * classid
424 				 */
425 				p->un.inner.last_ptr_id[prio] = cl->common.classid;
426 				p->un.inner.ptr[prio] = NULL;
427 			}
428 
429 			htb_safe_rb_erase(cl->node + prio, p->un.inner.feed + prio);
430 
431 			if (!p->un.inner.feed[prio].rb_node)
432 				mask |= 1 << prio;
433 		}
434 
435 		p->prio_activity &= ~mask;
436 		cl = p;
437 		p = cl->parent;
438 
439 	}
440 	if (cl->cmode == HTB_CAN_SEND && mask)
441 		htb_remove_class_from_row(q, cl, mask);
442 }
443 
444 static inline long htb_lowater(const struct htb_class *cl)
445 {
446 	if (htb_hysteresis)
447 		return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0;
448 	else
449 		return 0;
450 }
451 static inline long htb_hiwater(const struct htb_class *cl)
452 {
453 	if (htb_hysteresis)
454 		return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0;
455 	else
456 		return 0;
457 }
458 
459 
460 /**
461  * htb_class_mode - computes and returns current class mode
462  *
463  * It computes cl's mode at time cl->t_c+diff and returns it. If mode
464  * is not HTB_CAN_SEND then cl->pq_key is updated to time difference
465  * from now to time when cl will change its state.
466  * Also it is worth to note that class mode doesn't change simply
467  * at cl->{c,}tokens == 0 but there can rather be hysteresis of
468  * 0 .. -cl->{c,}buffer range. It is meant to limit number of
469  * mode transitions per time unit. The speed gain is about 1/6.
470  */
471 static inline enum htb_cmode
472 htb_class_mode(struct htb_class *cl, long *diff)
473 {
474 	long toks;
475 
476 	if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) {
477 		*diff = -toks;
478 		return HTB_CANT_SEND;
479 	}
480 
481 	if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl))
482 		return HTB_CAN_SEND;
483 
484 	*diff = -toks;
485 	return HTB_MAY_BORROW;
486 }
487 
488 /**
489  * htb_change_class_mode - changes classe's mode
490  *
491  * This should be the only way how to change classe's mode under normal
492  * cirsumstances. Routine will update feed lists linkage, change mode
493  * and add class to the wait event queue if appropriate. New mode should
494  * be different from old one and cl->pq_key has to be valid if changing
495  * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree).
496  */
497 static void
498 htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, long *diff)
499 {
500 	enum htb_cmode new_mode = htb_class_mode(cl, diff);
501 
502 	if (new_mode == cl->cmode)
503 		return;
504 
505 	if (cl->prio_activity) {	/* not necessary: speed optimization */
506 		if (cl->cmode != HTB_CANT_SEND)
507 			htb_deactivate_prios(q, cl);
508 		cl->cmode = new_mode;
509 		if (new_mode != HTB_CANT_SEND)
510 			htb_activate_prios(q, cl);
511 	} else
512 		cl->cmode = new_mode;
513 }
514 
515 /**
516  * htb_activate - inserts leaf cl into appropriate active feeds
517  *
518  * Routine learns (new) priority of leaf and activates feed chain
519  * for the prio. It can be called on already active leaf safely.
520  * It also adds leaf into droplist.
521  */
522 static inline void htb_activate(struct htb_sched *q, struct htb_class *cl)
523 {
524 	WARN_ON(cl->level || !cl->un.leaf.q || !cl->un.leaf.q->q.qlen);
525 
526 	if (!cl->prio_activity) {
527 		cl->prio_activity = 1 << cl->prio;
528 		htb_activate_prios(q, cl);
529 		list_add_tail(&cl->un.leaf.drop_list,
530 			      q->drops + cl->prio);
531 	}
532 }
533 
534 /**
535  * htb_deactivate - remove leaf cl from active feeds
536  *
537  * Make sure that leaf is active. In the other words it can't be called
538  * with non-active leaf. It also removes class from the drop list.
539  */
540 static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl)
541 {
542 	WARN_ON(!cl->prio_activity);
543 
544 	htb_deactivate_prios(q, cl);
545 	cl->prio_activity = 0;
546 	list_del_init(&cl->un.leaf.drop_list);
547 }
548 
549 static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch)
550 {
551 	int uninitialized_var(ret);
552 	struct htb_sched *q = qdisc_priv(sch);
553 	struct htb_class *cl = htb_classify(skb, sch, &ret);
554 
555 	if (cl == HTB_DIRECT) {
556 		/* enqueue to helper queue */
557 		if (q->direct_queue.qlen < q->direct_qlen) {
558 			__skb_queue_tail(&q->direct_queue, skb);
559 			q->direct_pkts++;
560 		} else {
561 			return qdisc_drop(skb, sch);
562 		}
563 #ifdef CONFIG_NET_CLS_ACT
564 	} else if (!cl) {
565 		if (ret & __NET_XMIT_BYPASS)
566 			sch->qstats.drops++;
567 		kfree_skb(skb);
568 		return ret;
569 #endif
570 	} else if ((ret = qdisc_enqueue(skb, cl->un.leaf.q)) != NET_XMIT_SUCCESS) {
571 		if (net_xmit_drop_count(ret)) {
572 			sch->qstats.drops++;
573 			cl->qstats.drops++;
574 		}
575 		return ret;
576 	} else {
577 		bstats_update(&cl->bstats, skb);
578 		htb_activate(q, cl);
579 	}
580 
581 	sch->q.qlen++;
582 	return NET_XMIT_SUCCESS;
583 }
584 
585 static inline void htb_accnt_tokens(struct htb_class *cl, int bytes, long diff)
586 {
587 	long toks = diff + cl->tokens;
588 
589 	if (toks > cl->buffer)
590 		toks = cl->buffer;
591 	toks -= (long) qdisc_l2t(cl->rate, bytes);
592 	if (toks <= -cl->mbuffer)
593 		toks = 1 - cl->mbuffer;
594 
595 	cl->tokens = toks;
596 }
597 
598 static inline void htb_accnt_ctokens(struct htb_class *cl, int bytes, long diff)
599 {
600 	long toks = diff + cl->ctokens;
601 
602 	if (toks > cl->cbuffer)
603 		toks = cl->cbuffer;
604 	toks -= (long) qdisc_l2t(cl->ceil, bytes);
605 	if (toks <= -cl->mbuffer)
606 		toks = 1 - cl->mbuffer;
607 
608 	cl->ctokens = toks;
609 }
610 
611 /**
612  * htb_charge_class - charges amount "bytes" to leaf and ancestors
613  *
614  * Routine assumes that packet "bytes" long was dequeued from leaf cl
615  * borrowing from "level". It accounts bytes to ceil leaky bucket for
616  * leaf and all ancestors and to rate bucket for ancestors at levels
617  * "level" and higher. It also handles possible change of mode resulting
618  * from the update. Note that mode can also increase here (MAY_BORROW to
619  * CAN_SEND) because we can use more precise clock that event queue here.
620  * In such case we remove class from event queue first.
621  */
622 static void htb_charge_class(struct htb_sched *q, struct htb_class *cl,
623 			     int level, struct sk_buff *skb)
624 {
625 	int bytes = qdisc_pkt_len(skb);
626 	enum htb_cmode old_mode;
627 	long diff;
628 
629 	while (cl) {
630 		diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer);
631 		if (cl->level >= level) {
632 			if (cl->level == level)
633 				cl->xstats.lends++;
634 			htb_accnt_tokens(cl, bytes, diff);
635 		} else {
636 			cl->xstats.borrows++;
637 			cl->tokens += diff;	/* we moved t_c; update tokens */
638 		}
639 		htb_accnt_ctokens(cl, bytes, diff);
640 		cl->t_c = q->now;
641 
642 		old_mode = cl->cmode;
643 		diff = 0;
644 		htb_change_class_mode(q, cl, &diff);
645 		if (old_mode != cl->cmode) {
646 			if (old_mode != HTB_CAN_SEND)
647 				htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level);
648 			if (cl->cmode != HTB_CAN_SEND)
649 				htb_add_to_wait_tree(q, cl, diff);
650 		}
651 
652 		/* update basic stats except for leaves which are already updated */
653 		if (cl->level)
654 			bstats_update(&cl->bstats, skb);
655 
656 		cl = cl->parent;
657 	}
658 }
659 
660 /**
661  * htb_do_events - make mode changes to classes at the level
662  *
663  * Scans event queue for pending events and applies them. Returns time of
664  * next pending event (0 for no event in pq, q->now for too many events).
665  * Note: Applied are events whose have cl->pq_key <= q->now.
666  */
667 static psched_time_t htb_do_events(struct htb_sched *q, int level,
668 				   unsigned long start)
669 {
670 	/* don't run for longer than 2 jiffies; 2 is used instead of
671 	 * 1 to simplify things when jiffy is going to be incremented
672 	 * too soon
673 	 */
674 	unsigned long stop_at = start + 2;
675 	while (time_before(jiffies, stop_at)) {
676 		struct htb_class *cl;
677 		long diff;
678 		struct rb_node *p = rb_first(&q->wait_pq[level]);
679 
680 		if (!p)
681 			return 0;
682 
683 		cl = rb_entry(p, struct htb_class, pq_node);
684 		if (cl->pq_key > q->now)
685 			return cl->pq_key;
686 
687 		htb_safe_rb_erase(p, q->wait_pq + level);
688 		diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer);
689 		htb_change_class_mode(q, cl, &diff);
690 		if (cl->cmode != HTB_CAN_SEND)
691 			htb_add_to_wait_tree(q, cl, diff);
692 	}
693 
694 	/* too much load - let's continue after a break for scheduling */
695 	if (!(q->warned & HTB_WARN_TOOMANYEVENTS)) {
696 		pr_warning("htb: too many events!\n");
697 		q->warned |= HTB_WARN_TOOMANYEVENTS;
698 	}
699 
700 	return q->now;
701 }
702 
703 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL
704  * is no such one exists.
705  */
706 static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n,
707 					      u32 id)
708 {
709 	struct rb_node *r = NULL;
710 	while (n) {
711 		struct htb_class *cl =
712 		    rb_entry(n, struct htb_class, node[prio]);
713 
714 		if (id > cl->common.classid) {
715 			n = n->rb_right;
716 		} else if (id < cl->common.classid) {
717 			r = n;
718 			n = n->rb_left;
719 		} else {
720 			return n;
721 		}
722 	}
723 	return r;
724 }
725 
726 /**
727  * htb_lookup_leaf - returns next leaf class in DRR order
728  *
729  * Find leaf where current feed pointers points to.
730  */
731 static struct htb_class *htb_lookup_leaf(struct rb_root *tree, int prio,
732 					 struct rb_node **pptr, u32 * pid)
733 {
734 	int i;
735 	struct {
736 		struct rb_node *root;
737 		struct rb_node **pptr;
738 		u32 *pid;
739 	} stk[TC_HTB_MAXDEPTH], *sp = stk;
740 
741 	BUG_ON(!tree->rb_node);
742 	sp->root = tree->rb_node;
743 	sp->pptr = pptr;
744 	sp->pid = pid;
745 
746 	for (i = 0; i < 65535; i++) {
747 		if (!*sp->pptr && *sp->pid) {
748 			/* ptr was invalidated but id is valid - try to recover
749 			 * the original or next ptr
750 			 */
751 			*sp->pptr =
752 			    htb_id_find_next_upper(prio, sp->root, *sp->pid);
753 		}
754 		*sp->pid = 0;	/* ptr is valid now so that remove this hint as it
755 				 * can become out of date quickly
756 				 */
757 		if (!*sp->pptr) {	/* we are at right end; rewind & go up */
758 			*sp->pptr = sp->root;
759 			while ((*sp->pptr)->rb_left)
760 				*sp->pptr = (*sp->pptr)->rb_left;
761 			if (sp > stk) {
762 				sp--;
763 				if (!*sp->pptr) {
764 					WARN_ON(1);
765 					return NULL;
766 				}
767 				htb_next_rb_node(sp->pptr);
768 			}
769 		} else {
770 			struct htb_class *cl;
771 			cl = rb_entry(*sp->pptr, struct htb_class, node[prio]);
772 			if (!cl->level)
773 				return cl;
774 			(++sp)->root = cl->un.inner.feed[prio].rb_node;
775 			sp->pptr = cl->un.inner.ptr + prio;
776 			sp->pid = cl->un.inner.last_ptr_id + prio;
777 		}
778 	}
779 	WARN_ON(1);
780 	return NULL;
781 }
782 
783 /* dequeues packet at given priority and level; call only if
784  * you are sure that there is active class at prio/level
785  */
786 static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, int prio,
787 					int level)
788 {
789 	struct sk_buff *skb = NULL;
790 	struct htb_class *cl, *start;
791 	/* look initial class up in the row */
792 	start = cl = htb_lookup_leaf(q->row[level] + prio, prio,
793 				     q->ptr[level] + prio,
794 				     q->last_ptr_id[level] + prio);
795 
796 	do {
797 next:
798 		if (unlikely(!cl))
799 			return NULL;
800 
801 		/* class can be empty - it is unlikely but can be true if leaf
802 		 * qdisc drops packets in enqueue routine or if someone used
803 		 * graft operation on the leaf since last dequeue;
804 		 * simply deactivate and skip such class
805 		 */
806 		if (unlikely(cl->un.leaf.q->q.qlen == 0)) {
807 			struct htb_class *next;
808 			htb_deactivate(q, cl);
809 
810 			/* row/level might become empty */
811 			if ((q->row_mask[level] & (1 << prio)) == 0)
812 				return NULL;
813 
814 			next = htb_lookup_leaf(q->row[level] + prio,
815 					       prio, q->ptr[level] + prio,
816 					       q->last_ptr_id[level] + prio);
817 
818 			if (cl == start)	/* fix start if we just deleted it */
819 				start = next;
820 			cl = next;
821 			goto next;
822 		}
823 
824 		skb = cl->un.leaf.q->dequeue(cl->un.leaf.q);
825 		if (likely(skb != NULL))
826 			break;
827 
828 		qdisc_warn_nonwc("htb", cl->un.leaf.q);
829 		htb_next_rb_node((level ? cl->parent->un.inner.ptr : q->
830 				  ptr[0]) + prio);
831 		cl = htb_lookup_leaf(q->row[level] + prio, prio,
832 				     q->ptr[level] + prio,
833 				     q->last_ptr_id[level] + prio);
834 
835 	} while (cl != start);
836 
837 	if (likely(skb != NULL)) {
838 		cl->un.leaf.deficit[level] -= qdisc_pkt_len(skb);
839 		if (cl->un.leaf.deficit[level] < 0) {
840 			cl->un.leaf.deficit[level] += cl->quantum;
841 			htb_next_rb_node((level ? cl->parent->un.inner.ptr : q->
842 					  ptr[0]) + prio);
843 		}
844 		/* this used to be after charge_class but this constelation
845 		 * gives us slightly better performance
846 		 */
847 		if (!cl->un.leaf.q->q.qlen)
848 			htb_deactivate(q, cl);
849 		htb_charge_class(q, cl, level, skb);
850 	}
851 	return skb;
852 }
853 
854 static struct sk_buff *htb_dequeue(struct Qdisc *sch)
855 {
856 	struct sk_buff *skb;
857 	struct htb_sched *q = qdisc_priv(sch);
858 	int level;
859 	psched_time_t next_event;
860 	unsigned long start_at;
861 
862 	/* try to dequeue direct packets as high prio (!) to minimize cpu work */
863 	skb = __skb_dequeue(&q->direct_queue);
864 	if (skb != NULL) {
865 ok:
866 		qdisc_bstats_update(sch, skb);
867 		qdisc_unthrottled(sch);
868 		sch->q.qlen--;
869 		return skb;
870 	}
871 
872 	if (!sch->q.qlen)
873 		goto fin;
874 	q->now = psched_get_time();
875 	start_at = jiffies;
876 
877 	next_event = q->now + 5 * PSCHED_TICKS_PER_SEC;
878 
879 	for (level = 0; level < TC_HTB_MAXDEPTH; level++) {
880 		/* common case optimization - skip event handler quickly */
881 		int m;
882 		psched_time_t event;
883 
884 		if (q->now >= q->near_ev_cache[level]) {
885 			event = htb_do_events(q, level, start_at);
886 			if (!event)
887 				event = q->now + PSCHED_TICKS_PER_SEC;
888 			q->near_ev_cache[level] = event;
889 		} else
890 			event = q->near_ev_cache[level];
891 
892 		if (next_event > event)
893 			next_event = event;
894 
895 		m = ~q->row_mask[level];
896 		while (m != (int)(-1)) {
897 			int prio = ffz(m);
898 
899 			m |= 1 << prio;
900 			skb = htb_dequeue_tree(q, prio, level);
901 			if (likely(skb != NULL))
902 				goto ok;
903 		}
904 	}
905 	sch->qstats.overlimits++;
906 	if (likely(next_event > q->now))
907 		qdisc_watchdog_schedule(&q->watchdog, next_event);
908 	else
909 		schedule_work(&q->work);
910 fin:
911 	return skb;
912 }
913 
914 /* try to drop from each class (by prio) until one succeed */
915 static unsigned int htb_drop(struct Qdisc *sch)
916 {
917 	struct htb_sched *q = qdisc_priv(sch);
918 	int prio;
919 
920 	for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) {
921 		struct list_head *p;
922 		list_for_each(p, q->drops + prio) {
923 			struct htb_class *cl = list_entry(p, struct htb_class,
924 							  un.leaf.drop_list);
925 			unsigned int len;
926 			if (cl->un.leaf.q->ops->drop &&
927 			    (len = cl->un.leaf.q->ops->drop(cl->un.leaf.q))) {
928 				sch->q.qlen--;
929 				if (!cl->un.leaf.q->q.qlen)
930 					htb_deactivate(q, cl);
931 				return len;
932 			}
933 		}
934 	}
935 	return 0;
936 }
937 
938 /* reset all classes */
939 /* always caled under BH & queue lock */
940 static void htb_reset(struct Qdisc *sch)
941 {
942 	struct htb_sched *q = qdisc_priv(sch);
943 	struct htb_class *cl;
944 	struct hlist_node *n;
945 	unsigned int i;
946 
947 	for (i = 0; i < q->clhash.hashsize; i++) {
948 		hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) {
949 			if (cl->level)
950 				memset(&cl->un.inner, 0, sizeof(cl->un.inner));
951 			else {
952 				if (cl->un.leaf.q)
953 					qdisc_reset(cl->un.leaf.q);
954 				INIT_LIST_HEAD(&cl->un.leaf.drop_list);
955 			}
956 			cl->prio_activity = 0;
957 			cl->cmode = HTB_CAN_SEND;
958 
959 		}
960 	}
961 	qdisc_watchdog_cancel(&q->watchdog);
962 	__skb_queue_purge(&q->direct_queue);
963 	sch->q.qlen = 0;
964 	memset(q->row, 0, sizeof(q->row));
965 	memset(q->row_mask, 0, sizeof(q->row_mask));
966 	memset(q->wait_pq, 0, sizeof(q->wait_pq));
967 	memset(q->ptr, 0, sizeof(q->ptr));
968 	for (i = 0; i < TC_HTB_NUMPRIO; i++)
969 		INIT_LIST_HEAD(q->drops + i);
970 }
971 
972 static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = {
973 	[TCA_HTB_PARMS]	= { .len = sizeof(struct tc_htb_opt) },
974 	[TCA_HTB_INIT]	= { .len = sizeof(struct tc_htb_glob) },
975 	[TCA_HTB_CTAB]	= { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
976 	[TCA_HTB_RTAB]	= { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
977 };
978 
979 static void htb_work_func(struct work_struct *work)
980 {
981 	struct htb_sched *q = container_of(work, struct htb_sched, work);
982 	struct Qdisc *sch = q->watchdog.qdisc;
983 
984 	__netif_schedule(qdisc_root(sch));
985 }
986 
987 static int htb_init(struct Qdisc *sch, struct nlattr *opt)
988 {
989 	struct htb_sched *q = qdisc_priv(sch);
990 	struct nlattr *tb[TCA_HTB_INIT + 1];
991 	struct tc_htb_glob *gopt;
992 	int err;
993 	int i;
994 
995 	if (!opt)
996 		return -EINVAL;
997 
998 	err = nla_parse_nested(tb, TCA_HTB_INIT, opt, htb_policy);
999 	if (err < 0)
1000 		return err;
1001 
1002 	if (tb[TCA_HTB_INIT] == NULL) {
1003 		pr_err("HTB: hey probably you have bad tc tool ?\n");
1004 		return -EINVAL;
1005 	}
1006 	gopt = nla_data(tb[TCA_HTB_INIT]);
1007 	if (gopt->version != HTB_VER >> 16) {
1008 		pr_err("HTB: need tc/htb version %d (minor is %d), you have %d\n",
1009 		       HTB_VER >> 16, HTB_VER & 0xffff, gopt->version);
1010 		return -EINVAL;
1011 	}
1012 
1013 	err = qdisc_class_hash_init(&q->clhash);
1014 	if (err < 0)
1015 		return err;
1016 	for (i = 0; i < TC_HTB_NUMPRIO; i++)
1017 		INIT_LIST_HEAD(q->drops + i);
1018 
1019 	qdisc_watchdog_init(&q->watchdog, sch);
1020 	INIT_WORK(&q->work, htb_work_func);
1021 	skb_queue_head_init(&q->direct_queue);
1022 
1023 	q->direct_qlen = qdisc_dev(sch)->tx_queue_len;
1024 	if (q->direct_qlen < 2)	/* some devices have zero tx_queue_len */
1025 		q->direct_qlen = 2;
1026 
1027 	if ((q->rate2quantum = gopt->rate2quantum) < 1)
1028 		q->rate2quantum = 1;
1029 	q->defcls = gopt->defcls;
1030 
1031 	return 0;
1032 }
1033 
1034 static int htb_dump(struct Qdisc *sch, struct sk_buff *skb)
1035 {
1036 	spinlock_t *root_lock = qdisc_root_sleeping_lock(sch);
1037 	struct htb_sched *q = qdisc_priv(sch);
1038 	struct nlattr *nest;
1039 	struct tc_htb_glob gopt;
1040 
1041 	spin_lock_bh(root_lock);
1042 
1043 	gopt.direct_pkts = q->direct_pkts;
1044 	gopt.version = HTB_VER;
1045 	gopt.rate2quantum = q->rate2quantum;
1046 	gopt.defcls = q->defcls;
1047 	gopt.debug = 0;
1048 
1049 	nest = nla_nest_start(skb, TCA_OPTIONS);
1050 	if (nest == NULL)
1051 		goto nla_put_failure;
1052 	if (nla_put(skb, TCA_HTB_INIT, sizeof(gopt), &gopt))
1053 		goto nla_put_failure;
1054 	nla_nest_end(skb, nest);
1055 
1056 	spin_unlock_bh(root_lock);
1057 	return skb->len;
1058 
1059 nla_put_failure:
1060 	spin_unlock_bh(root_lock);
1061 	nla_nest_cancel(skb, nest);
1062 	return -1;
1063 }
1064 
1065 static int htb_dump_class(struct Qdisc *sch, unsigned long arg,
1066 			  struct sk_buff *skb, struct tcmsg *tcm)
1067 {
1068 	struct htb_class *cl = (struct htb_class *)arg;
1069 	spinlock_t *root_lock = qdisc_root_sleeping_lock(sch);
1070 	struct nlattr *nest;
1071 	struct tc_htb_opt opt;
1072 
1073 	spin_lock_bh(root_lock);
1074 	tcm->tcm_parent = cl->parent ? cl->parent->common.classid : TC_H_ROOT;
1075 	tcm->tcm_handle = cl->common.classid;
1076 	if (!cl->level && cl->un.leaf.q)
1077 		tcm->tcm_info = cl->un.leaf.q->handle;
1078 
1079 	nest = nla_nest_start(skb, TCA_OPTIONS);
1080 	if (nest == NULL)
1081 		goto nla_put_failure;
1082 
1083 	memset(&opt, 0, sizeof(opt));
1084 
1085 	opt.rate = cl->rate->rate;
1086 	opt.buffer = cl->buffer;
1087 	opt.ceil = cl->ceil->rate;
1088 	opt.cbuffer = cl->cbuffer;
1089 	opt.quantum = cl->quantum;
1090 	opt.prio = cl->prio;
1091 	opt.level = cl->level;
1092 	if (nla_put(skb, TCA_HTB_PARMS, sizeof(opt), &opt))
1093 		goto nla_put_failure;
1094 
1095 	nla_nest_end(skb, nest);
1096 	spin_unlock_bh(root_lock);
1097 	return skb->len;
1098 
1099 nla_put_failure:
1100 	spin_unlock_bh(root_lock);
1101 	nla_nest_cancel(skb, nest);
1102 	return -1;
1103 }
1104 
1105 static int
1106 htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d)
1107 {
1108 	struct htb_class *cl = (struct htb_class *)arg;
1109 
1110 	if (!cl->level && cl->un.leaf.q)
1111 		cl->qstats.qlen = cl->un.leaf.q->q.qlen;
1112 	cl->xstats.tokens = cl->tokens;
1113 	cl->xstats.ctokens = cl->ctokens;
1114 
1115 	if (gnet_stats_copy_basic(d, &cl->bstats) < 0 ||
1116 	    gnet_stats_copy_rate_est(d, NULL, &cl->rate_est) < 0 ||
1117 	    gnet_stats_copy_queue(d, &cl->qstats) < 0)
1118 		return -1;
1119 
1120 	return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats));
1121 }
1122 
1123 static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1124 		     struct Qdisc **old)
1125 {
1126 	struct htb_class *cl = (struct htb_class *)arg;
1127 
1128 	if (cl->level)
1129 		return -EINVAL;
1130 	if (new == NULL &&
1131 	    (new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
1132 				     cl->common.classid)) == NULL)
1133 		return -ENOBUFS;
1134 
1135 	sch_tree_lock(sch);
1136 	*old = cl->un.leaf.q;
1137 	cl->un.leaf.q = new;
1138 	if (*old != NULL) {
1139 		qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
1140 		qdisc_reset(*old);
1141 	}
1142 	sch_tree_unlock(sch);
1143 	return 0;
1144 }
1145 
1146 static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg)
1147 {
1148 	struct htb_class *cl = (struct htb_class *)arg;
1149 	return !cl->level ? cl->un.leaf.q : NULL;
1150 }
1151 
1152 static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg)
1153 {
1154 	struct htb_class *cl = (struct htb_class *)arg;
1155 
1156 	if (cl->un.leaf.q->q.qlen == 0)
1157 		htb_deactivate(qdisc_priv(sch), cl);
1158 }
1159 
1160 static unsigned long htb_get(struct Qdisc *sch, u32 classid)
1161 {
1162 	struct htb_class *cl = htb_find(classid, sch);
1163 	if (cl)
1164 		cl->refcnt++;
1165 	return (unsigned long)cl;
1166 }
1167 
1168 static inline int htb_parent_last_child(struct htb_class *cl)
1169 {
1170 	if (!cl->parent)
1171 		/* the root class */
1172 		return 0;
1173 	if (cl->parent->children > 1)
1174 		/* not the last child */
1175 		return 0;
1176 	return 1;
1177 }
1178 
1179 static void htb_parent_to_leaf(struct htb_sched *q, struct htb_class *cl,
1180 			       struct Qdisc *new_q)
1181 {
1182 	struct htb_class *parent = cl->parent;
1183 
1184 	WARN_ON(cl->level || !cl->un.leaf.q || cl->prio_activity);
1185 
1186 	if (parent->cmode != HTB_CAN_SEND)
1187 		htb_safe_rb_erase(&parent->pq_node, q->wait_pq + parent->level);
1188 
1189 	parent->level = 0;
1190 	memset(&parent->un.inner, 0, sizeof(parent->un.inner));
1191 	INIT_LIST_HEAD(&parent->un.leaf.drop_list);
1192 	parent->un.leaf.q = new_q ? new_q : &noop_qdisc;
1193 	parent->tokens = parent->buffer;
1194 	parent->ctokens = parent->cbuffer;
1195 	parent->t_c = psched_get_time();
1196 	parent->cmode = HTB_CAN_SEND;
1197 }
1198 
1199 static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl)
1200 {
1201 	if (!cl->level) {
1202 		WARN_ON(!cl->un.leaf.q);
1203 		qdisc_destroy(cl->un.leaf.q);
1204 	}
1205 	gen_kill_estimator(&cl->bstats, &cl->rate_est);
1206 	qdisc_put_rtab(cl->rate);
1207 	qdisc_put_rtab(cl->ceil);
1208 
1209 	tcf_destroy_chain(&cl->filter_list);
1210 	kfree(cl);
1211 }
1212 
1213 static void htb_destroy(struct Qdisc *sch)
1214 {
1215 	struct htb_sched *q = qdisc_priv(sch);
1216 	struct hlist_node *n, *next;
1217 	struct htb_class *cl;
1218 	unsigned int i;
1219 
1220 	cancel_work_sync(&q->work);
1221 	qdisc_watchdog_cancel(&q->watchdog);
1222 	/* This line used to be after htb_destroy_class call below
1223 	 * and surprisingly it worked in 2.4. But it must precede it
1224 	 * because filter need its target class alive to be able to call
1225 	 * unbind_filter on it (without Oops).
1226 	 */
1227 	tcf_destroy_chain(&q->filter_list);
1228 
1229 	for (i = 0; i < q->clhash.hashsize; i++) {
1230 		hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode)
1231 			tcf_destroy_chain(&cl->filter_list);
1232 	}
1233 	for (i = 0; i < q->clhash.hashsize; i++) {
1234 		hlist_for_each_entry_safe(cl, n, next, &q->clhash.hash[i],
1235 					  common.hnode)
1236 			htb_destroy_class(sch, cl);
1237 	}
1238 	qdisc_class_hash_destroy(&q->clhash);
1239 	__skb_queue_purge(&q->direct_queue);
1240 }
1241 
1242 static int htb_delete(struct Qdisc *sch, unsigned long arg)
1243 {
1244 	struct htb_sched *q = qdisc_priv(sch);
1245 	struct htb_class *cl = (struct htb_class *)arg;
1246 	unsigned int qlen;
1247 	struct Qdisc *new_q = NULL;
1248 	int last_child = 0;
1249 
1250 	// TODO: why don't allow to delete subtree ? references ? does
1251 	// tc subsys quarantee us that in htb_destroy it holds no class
1252 	// refs so that we can remove children safely there ?
1253 	if (cl->children || cl->filter_cnt)
1254 		return -EBUSY;
1255 
1256 	if (!cl->level && htb_parent_last_child(cl)) {
1257 		new_q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
1258 					  cl->parent->common.classid);
1259 		last_child = 1;
1260 	}
1261 
1262 	sch_tree_lock(sch);
1263 
1264 	if (!cl->level) {
1265 		qlen = cl->un.leaf.q->q.qlen;
1266 		qdisc_reset(cl->un.leaf.q);
1267 		qdisc_tree_decrease_qlen(cl->un.leaf.q, qlen);
1268 	}
1269 
1270 	/* delete from hash and active; remainder in destroy_class */
1271 	qdisc_class_hash_remove(&q->clhash, &cl->common);
1272 	if (cl->parent)
1273 		cl->parent->children--;
1274 
1275 	if (cl->prio_activity)
1276 		htb_deactivate(q, cl);
1277 
1278 	if (cl->cmode != HTB_CAN_SEND)
1279 		htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level);
1280 
1281 	if (last_child)
1282 		htb_parent_to_leaf(q, cl, new_q);
1283 
1284 	BUG_ON(--cl->refcnt == 0);
1285 	/*
1286 	 * This shouldn't happen: we "hold" one cops->get() when called
1287 	 * from tc_ctl_tclass; the destroy method is done from cops->put().
1288 	 */
1289 
1290 	sch_tree_unlock(sch);
1291 	return 0;
1292 }
1293 
1294 static void htb_put(struct Qdisc *sch, unsigned long arg)
1295 {
1296 	struct htb_class *cl = (struct htb_class *)arg;
1297 
1298 	if (--cl->refcnt == 0)
1299 		htb_destroy_class(sch, cl);
1300 }
1301 
1302 static int htb_change_class(struct Qdisc *sch, u32 classid,
1303 			    u32 parentid, struct nlattr **tca,
1304 			    unsigned long *arg)
1305 {
1306 	int err = -EINVAL;
1307 	struct htb_sched *q = qdisc_priv(sch);
1308 	struct htb_class *cl = (struct htb_class *)*arg, *parent;
1309 	struct nlattr *opt = tca[TCA_OPTIONS];
1310 	struct qdisc_rate_table *rtab = NULL, *ctab = NULL;
1311 	struct nlattr *tb[__TCA_HTB_MAX];
1312 	struct tc_htb_opt *hopt;
1313 
1314 	/* extract all subattrs from opt attr */
1315 	if (!opt)
1316 		goto failure;
1317 
1318 	err = nla_parse_nested(tb, TCA_HTB_MAX, opt, htb_policy);
1319 	if (err < 0)
1320 		goto failure;
1321 
1322 	err = -EINVAL;
1323 	if (tb[TCA_HTB_PARMS] == NULL)
1324 		goto failure;
1325 
1326 	parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch);
1327 
1328 	hopt = nla_data(tb[TCA_HTB_PARMS]);
1329 
1330 	rtab = qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB]);
1331 	ctab = qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB]);
1332 	if (!rtab || !ctab)
1333 		goto failure;
1334 
1335 	if (!cl) {		/* new class */
1336 		struct Qdisc *new_q;
1337 		int prio;
1338 		struct {
1339 			struct nlattr		nla;
1340 			struct gnet_estimator	opt;
1341 		} est = {
1342 			.nla = {
1343 				.nla_len	= nla_attr_size(sizeof(est.opt)),
1344 				.nla_type	= TCA_RATE,
1345 			},
1346 			.opt = {
1347 				/* 4s interval, 16s averaging constant */
1348 				.interval	= 2,
1349 				.ewma_log	= 2,
1350 			},
1351 		};
1352 
1353 		/* check for valid classid */
1354 		if (!classid || TC_H_MAJ(classid ^ sch->handle) ||
1355 		    htb_find(classid, sch))
1356 			goto failure;
1357 
1358 		/* check maximal depth */
1359 		if (parent && parent->parent && parent->parent->level < 2) {
1360 			pr_err("htb: tree is too deep\n");
1361 			goto failure;
1362 		}
1363 		err = -ENOBUFS;
1364 		cl = kzalloc(sizeof(*cl), GFP_KERNEL);
1365 		if (!cl)
1366 			goto failure;
1367 
1368 		err = gen_new_estimator(&cl->bstats, &cl->rate_est,
1369 					qdisc_root_sleeping_lock(sch),
1370 					tca[TCA_RATE] ? : &est.nla);
1371 		if (err) {
1372 			kfree(cl);
1373 			goto failure;
1374 		}
1375 
1376 		cl->refcnt = 1;
1377 		cl->children = 0;
1378 		INIT_LIST_HEAD(&cl->un.leaf.drop_list);
1379 		RB_CLEAR_NODE(&cl->pq_node);
1380 
1381 		for (prio = 0; prio < TC_HTB_NUMPRIO; prio++)
1382 			RB_CLEAR_NODE(&cl->node[prio]);
1383 
1384 		/* create leaf qdisc early because it uses kmalloc(GFP_KERNEL)
1385 		 * so that can't be used inside of sch_tree_lock
1386 		 * -- thanks to Karlis Peisenieks
1387 		 */
1388 		new_q = qdisc_create_dflt(sch->dev_queue,
1389 					  &pfifo_qdisc_ops, classid);
1390 		sch_tree_lock(sch);
1391 		if (parent && !parent->level) {
1392 			unsigned int qlen = parent->un.leaf.q->q.qlen;
1393 
1394 			/* turn parent into inner node */
1395 			qdisc_reset(parent->un.leaf.q);
1396 			qdisc_tree_decrease_qlen(parent->un.leaf.q, qlen);
1397 			qdisc_destroy(parent->un.leaf.q);
1398 			if (parent->prio_activity)
1399 				htb_deactivate(q, parent);
1400 
1401 			/* remove from evt list because of level change */
1402 			if (parent->cmode != HTB_CAN_SEND) {
1403 				htb_safe_rb_erase(&parent->pq_node, q->wait_pq);
1404 				parent->cmode = HTB_CAN_SEND;
1405 			}
1406 			parent->level = (parent->parent ? parent->parent->level
1407 					 : TC_HTB_MAXDEPTH) - 1;
1408 			memset(&parent->un.inner, 0, sizeof(parent->un.inner));
1409 		}
1410 		/* leaf (we) needs elementary qdisc */
1411 		cl->un.leaf.q = new_q ? new_q : &noop_qdisc;
1412 
1413 		cl->common.classid = classid;
1414 		cl->parent = parent;
1415 
1416 		/* set class to be in HTB_CAN_SEND state */
1417 		cl->tokens = hopt->buffer;
1418 		cl->ctokens = hopt->cbuffer;
1419 		cl->mbuffer = 60 * PSCHED_TICKS_PER_SEC;	/* 1min */
1420 		cl->t_c = psched_get_time();
1421 		cl->cmode = HTB_CAN_SEND;
1422 
1423 		/* attach to the hash list and parent's family */
1424 		qdisc_class_hash_insert(&q->clhash, &cl->common);
1425 		if (parent)
1426 			parent->children++;
1427 	} else {
1428 		if (tca[TCA_RATE]) {
1429 			err = gen_replace_estimator(&cl->bstats, &cl->rate_est,
1430 						    qdisc_root_sleeping_lock(sch),
1431 						    tca[TCA_RATE]);
1432 			if (err)
1433 				return err;
1434 		}
1435 		sch_tree_lock(sch);
1436 	}
1437 
1438 	/* it used to be a nasty bug here, we have to check that node
1439 	 * is really leaf before changing cl->un.leaf !
1440 	 */
1441 	if (!cl->level) {
1442 		cl->quantum = rtab->rate.rate / q->rate2quantum;
1443 		if (!hopt->quantum && cl->quantum < 1000) {
1444 			pr_warning(
1445 			       "HTB: quantum of class %X is small. Consider r2q change.\n",
1446 			       cl->common.classid);
1447 			cl->quantum = 1000;
1448 		}
1449 		if (!hopt->quantum && cl->quantum > 200000) {
1450 			pr_warning(
1451 			       "HTB: quantum of class %X is big. Consider r2q change.\n",
1452 			       cl->common.classid);
1453 			cl->quantum = 200000;
1454 		}
1455 		if (hopt->quantum)
1456 			cl->quantum = hopt->quantum;
1457 		if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO)
1458 			cl->prio = TC_HTB_NUMPRIO - 1;
1459 	}
1460 
1461 	cl->buffer = hopt->buffer;
1462 	cl->cbuffer = hopt->cbuffer;
1463 	if (cl->rate)
1464 		qdisc_put_rtab(cl->rate);
1465 	cl->rate = rtab;
1466 	if (cl->ceil)
1467 		qdisc_put_rtab(cl->ceil);
1468 	cl->ceil = ctab;
1469 	sch_tree_unlock(sch);
1470 
1471 	qdisc_class_hash_grow(sch, &q->clhash);
1472 
1473 	*arg = (unsigned long)cl;
1474 	return 0;
1475 
1476 failure:
1477 	if (rtab)
1478 		qdisc_put_rtab(rtab);
1479 	if (ctab)
1480 		qdisc_put_rtab(ctab);
1481 	return err;
1482 }
1483 
1484 static struct tcf_proto **htb_find_tcf(struct Qdisc *sch, unsigned long arg)
1485 {
1486 	struct htb_sched *q = qdisc_priv(sch);
1487 	struct htb_class *cl = (struct htb_class *)arg;
1488 	struct tcf_proto **fl = cl ? &cl->filter_list : &q->filter_list;
1489 
1490 	return fl;
1491 }
1492 
1493 static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent,
1494 				     u32 classid)
1495 {
1496 	struct htb_class *cl = htb_find(classid, sch);
1497 
1498 	/*if (cl && !cl->level) return 0;
1499 	 * The line above used to be there to prevent attaching filters to
1500 	 * leaves. But at least tc_index filter uses this just to get class
1501 	 * for other reasons so that we have to allow for it.
1502 	 * ----
1503 	 * 19.6.2002 As Werner explained it is ok - bind filter is just
1504 	 * another way to "lock" the class - unlike "get" this lock can
1505 	 * be broken by class during destroy IIUC.
1506 	 */
1507 	if (cl)
1508 		cl->filter_cnt++;
1509 	return (unsigned long)cl;
1510 }
1511 
1512 static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg)
1513 {
1514 	struct htb_class *cl = (struct htb_class *)arg;
1515 
1516 	if (cl)
1517 		cl->filter_cnt--;
1518 }
1519 
1520 static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1521 {
1522 	struct htb_sched *q = qdisc_priv(sch);
1523 	struct htb_class *cl;
1524 	struct hlist_node *n;
1525 	unsigned int i;
1526 
1527 	if (arg->stop)
1528 		return;
1529 
1530 	for (i = 0; i < q->clhash.hashsize; i++) {
1531 		hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) {
1532 			if (arg->count < arg->skip) {
1533 				arg->count++;
1534 				continue;
1535 			}
1536 			if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
1537 				arg->stop = 1;
1538 				return;
1539 			}
1540 			arg->count++;
1541 		}
1542 	}
1543 }
1544 
1545 static const struct Qdisc_class_ops htb_class_ops = {
1546 	.graft		=	htb_graft,
1547 	.leaf		=	htb_leaf,
1548 	.qlen_notify	=	htb_qlen_notify,
1549 	.get		=	htb_get,
1550 	.put		=	htb_put,
1551 	.change		=	htb_change_class,
1552 	.delete		=	htb_delete,
1553 	.walk		=	htb_walk,
1554 	.tcf_chain	=	htb_find_tcf,
1555 	.bind_tcf	=	htb_bind_filter,
1556 	.unbind_tcf	=	htb_unbind_filter,
1557 	.dump		=	htb_dump_class,
1558 	.dump_stats	=	htb_dump_class_stats,
1559 };
1560 
1561 static struct Qdisc_ops htb_qdisc_ops __read_mostly = {
1562 	.cl_ops		=	&htb_class_ops,
1563 	.id		=	"htb",
1564 	.priv_size	=	sizeof(struct htb_sched),
1565 	.enqueue	=	htb_enqueue,
1566 	.dequeue	=	htb_dequeue,
1567 	.peek		=	qdisc_peek_dequeued,
1568 	.drop		=	htb_drop,
1569 	.init		=	htb_init,
1570 	.reset		=	htb_reset,
1571 	.destroy	=	htb_destroy,
1572 	.dump		=	htb_dump,
1573 	.owner		=	THIS_MODULE,
1574 };
1575 
1576 static int __init htb_module_init(void)
1577 {
1578 	return register_qdisc(&htb_qdisc_ops);
1579 }
1580 static void __exit htb_module_exit(void)
1581 {
1582 	unregister_qdisc(&htb_qdisc_ops);
1583 }
1584 
1585 module_init(htb_module_init)
1586 module_exit(htb_module_exit)
1587 MODULE_LICENSE("GPL");
1588