Initialise scatter/gather list in sg driver
[linux-2.6/libata-dev.git] / net / sched / sch_htb.c
blob5e608a64935af6b644c6ecf3596f7dd84c32d35a
1 /*
2 * net/sched/sch_htb.c Hierarchical token bucket, feed tree version
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.
9 * Authors: Martin Devera, <devik@cdi.cz>
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.
28 * $Id: sch_htb.c,v 1.25 2003/12/07 11:08:25 devik Exp devik $
30 #include <linux/module.h>
31 #include <linux/types.h>
32 #include <linux/kernel.h>
33 #include <linux/string.h>
34 #include <linux/errno.h>
35 #include <linux/skbuff.h>
36 #include <linux/list.h>
37 #include <linux/compiler.h>
38 #include <linux/rbtree.h>
39 #include <net/netlink.h>
40 #include <net/pkt_sched.h>
42 /* HTB algorithm.
43 Author: devik@cdi.cz
44 ========================================================================
45 HTB is like TBF with multiple classes. It is also similar to CBQ because
46 it allows to assign priority to each class in hierarchy.
47 In fact it is another implementation of Floyd's formal sharing.
49 Levels:
50 Each class is assigned level. Leaf has ALWAYS level 0 and root
51 classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level
52 one less than their parent.
55 #define HTB_HSIZE 16 /* classid hash size */
56 #define HTB_HYSTERESIS 1 /* whether to use mode hysteresis for speedup */
57 #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */
59 #if HTB_VER >> 16 != TC_HTB_PROTOVER
60 #error "Mismatched sch_htb.c and pkt_sch.h"
61 #endif
63 /* used internaly to keep status of single class */
64 enum htb_cmode {
65 HTB_CANT_SEND, /* class can't send and can't borrow */
66 HTB_MAY_BORROW, /* class can't send but may borrow */
67 HTB_CAN_SEND /* class can send */
70 /* interior & leaf nodes; props specific to leaves are marked L: */
71 struct htb_class {
72 /* general class parameters */
73 u32 classid;
74 struct gnet_stats_basic bstats;
75 struct gnet_stats_queue qstats;
76 struct gnet_stats_rate_est rate_est;
77 struct tc_htb_xstats xstats; /* our special stats */
78 int refcnt; /* usage count of this class */
80 /* topology */
81 int level; /* our level (see above) */
82 struct htb_class *parent; /* parent class */
83 struct hlist_node hlist; /* classid hash list item */
84 struct list_head sibling; /* sibling list item */
85 struct list_head children; /* children list */
87 union {
88 struct htb_class_leaf {
89 struct Qdisc *q;
90 int prio;
91 int aprio;
92 int quantum;
93 int deficit[TC_HTB_MAXDEPTH];
94 struct list_head drop_list;
95 } leaf;
96 struct htb_class_inner {
97 struct rb_root feed[TC_HTB_NUMPRIO]; /* feed trees */
98 struct rb_node *ptr[TC_HTB_NUMPRIO]; /* current class ptr */
99 /* When class changes from state 1->2 and disconnects from
100 parent's feed then we lost ptr value and start from the
101 first child again. Here we store classid of the
102 last valid ptr (used when ptr is NULL). */
103 u32 last_ptr_id[TC_HTB_NUMPRIO];
104 } inner;
105 } un;
106 struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */
107 struct rb_node pq_node; /* node for event queue */
108 psched_time_t pq_key;
110 int prio_activity; /* for which prios are we active */
111 enum htb_cmode cmode; /* current mode of the class */
113 /* class attached filters */
114 struct tcf_proto *filter_list;
115 int filter_cnt;
117 int warned; /* only one warning about non work conserving .. */
119 /* token bucket parameters */
120 struct qdisc_rate_table *rate; /* rate table of the class itself */
121 struct qdisc_rate_table *ceil; /* ceiling rate (limits borrows too) */
122 long buffer, cbuffer; /* token bucket depth/rate */
123 psched_tdiff_t mbuffer; /* max wait time */
124 long tokens, ctokens; /* current number of tokens */
125 psched_time_t t_c; /* checkpoint time */
127 int prio; /* For parent to leaf return possible here */
128 int quantum; /* we do backup. Finally full replacement */
129 /* of un.leaf originals should be done. */
132 static inline long L2T(struct htb_class *cl, struct qdisc_rate_table *rate,
133 int size)
135 long result = qdisc_l2t(rate, size);
136 return result;
139 struct htb_sched {
140 struct list_head root; /* root classes list */
141 struct hlist_head hash[HTB_HSIZE]; /* hashed by classid */
142 struct list_head drops[TC_HTB_NUMPRIO];/* active leaves (for drops) */
144 /* self list - roots of self generating tree */
145 struct rb_root row[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO];
146 int row_mask[TC_HTB_MAXDEPTH];
147 struct rb_node *ptr[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO];
148 u32 last_ptr_id[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO];
150 /* self wait list - roots of wait PQs per row */
151 struct rb_root wait_pq[TC_HTB_MAXDEPTH];
153 /* time of nearest event per level (row) */
154 psched_time_t near_ev_cache[TC_HTB_MAXDEPTH];
156 /* whether we hit non-work conserving class during this dequeue; we use */
157 int nwc_hit; /* this to disable mindelay complaint in dequeue */
159 int defcls; /* class where unclassified flows go to */
161 /* filters for qdisc itself */
162 struct tcf_proto *filter_list;
163 int filter_cnt;
165 int rate2quantum; /* quant = rate / rate2quantum */
166 psched_time_t now; /* cached dequeue time */
167 struct qdisc_watchdog watchdog;
169 /* non shaped skbs; let them go directly thru */
170 struct sk_buff_head direct_queue;
171 int direct_qlen; /* max qlen of above */
173 long direct_pkts;
176 /* compute hash of size HTB_HSIZE for given handle */
177 static inline int htb_hash(u32 h)
179 #if HTB_HSIZE != 16
180 #error "Declare new hash for your HTB_HSIZE"
181 #endif
182 h ^= h >> 8; /* stolen from cbq_hash */
183 h ^= h >> 4;
184 return h & 0xf;
187 /* find class in global hash table using given handle */
188 static inline struct htb_class *htb_find(u32 handle, struct Qdisc *sch)
190 struct htb_sched *q = qdisc_priv(sch);
191 struct hlist_node *p;
192 struct htb_class *cl;
194 if (TC_H_MAJ(handle) != sch->handle)
195 return NULL;
197 hlist_for_each_entry(cl, p, q->hash + htb_hash(handle), hlist) {
198 if (cl->classid == handle)
199 return cl;
201 return NULL;
205 * htb_classify - classify a packet into class
207 * It returns NULL if the packet should be dropped or -1 if the packet
208 * should be passed directly thru. In all other cases leaf class is returned.
209 * We allow direct class selection by classid in priority. The we examine
210 * filters in qdisc and in inner nodes (if higher filter points to the inner
211 * node). If we end up with classid MAJOR:0 we enqueue the skb into special
212 * internal fifo (direct). These packets then go directly thru. If we still
213 * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessfull
214 * then finish and return direct queue.
216 #define HTB_DIRECT (struct htb_class*)-1
217 static inline u32 htb_classid(struct htb_class *cl)
219 return (cl && cl != HTB_DIRECT) ? cl->classid : TC_H_UNSPEC;
222 static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch,
223 int *qerr)
225 struct htb_sched *q = qdisc_priv(sch);
226 struct htb_class *cl;
227 struct tcf_result res;
228 struct tcf_proto *tcf;
229 int result;
231 /* allow to select class by setting skb->priority to valid classid;
232 note that nfmark can be used too by attaching filter fw with no
233 rules in it */
234 if (skb->priority == sch->handle)
235 return HTB_DIRECT; /* X:0 (direct flow) selected */
236 if ((cl = htb_find(skb->priority, sch)) != NULL && cl->level == 0)
237 return cl;
239 *qerr = NET_XMIT_BYPASS;
240 tcf = q->filter_list;
241 while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) {
242 #ifdef CONFIG_NET_CLS_ACT
243 switch (result) {
244 case TC_ACT_QUEUED:
245 case TC_ACT_STOLEN:
246 *qerr = NET_XMIT_SUCCESS;
247 case TC_ACT_SHOT:
248 return NULL;
250 #endif
251 if ((cl = (void *)res.class) == NULL) {
252 if (res.classid == sch->handle)
253 return HTB_DIRECT; /* X:0 (direct flow) */
254 if ((cl = htb_find(res.classid, sch)) == NULL)
255 break; /* filter selected invalid classid */
257 if (!cl->level)
258 return cl; /* we hit leaf; return it */
260 /* we have got inner class; apply inner filter chain */
261 tcf = cl->filter_list;
263 /* classification failed; try to use default class */
264 cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch);
265 if (!cl || cl->level)
266 return HTB_DIRECT; /* bad default .. this is safe bet */
267 return cl;
271 * htb_add_to_id_tree - adds class to the round robin list
273 * Routine adds class to the list (actually tree) sorted by classid.
274 * Make sure that class is not already on such list for given prio.
276 static void htb_add_to_id_tree(struct rb_root *root,
277 struct htb_class *cl, int prio)
279 struct rb_node **p = &root->rb_node, *parent = NULL;
281 while (*p) {
282 struct htb_class *c;
283 parent = *p;
284 c = rb_entry(parent, struct htb_class, node[prio]);
286 if (cl->classid > c->classid)
287 p = &parent->rb_right;
288 else
289 p = &parent->rb_left;
291 rb_link_node(&cl->node[prio], parent, p);
292 rb_insert_color(&cl->node[prio], root);
296 * htb_add_to_wait_tree - adds class to the event queue with delay
298 * The class is added to priority event queue to indicate that class will
299 * change its mode in cl->pq_key microseconds. Make sure that class is not
300 * already in the queue.
302 static void htb_add_to_wait_tree(struct htb_sched *q,
303 struct htb_class *cl, long delay)
305 struct rb_node **p = &q->wait_pq[cl->level].rb_node, *parent = NULL;
307 cl->pq_key = q->now + delay;
308 if (cl->pq_key == q->now)
309 cl->pq_key++;
311 /* update the nearest event cache */
312 if (q->near_ev_cache[cl->level] > cl->pq_key)
313 q->near_ev_cache[cl->level] = cl->pq_key;
315 while (*p) {
316 struct htb_class *c;
317 parent = *p;
318 c = rb_entry(parent, struct htb_class, pq_node);
319 if (cl->pq_key >= c->pq_key)
320 p = &parent->rb_right;
321 else
322 p = &parent->rb_left;
324 rb_link_node(&cl->pq_node, parent, p);
325 rb_insert_color(&cl->pq_node, &q->wait_pq[cl->level]);
329 * htb_next_rb_node - finds next node in binary tree
331 * When we are past last key we return NULL.
332 * Average complexity is 2 steps per call.
334 static inline void htb_next_rb_node(struct rb_node **n)
336 *n = rb_next(*n);
340 * htb_add_class_to_row - add class to its row
342 * The class is added to row at priorities marked in mask.
343 * It does nothing if mask == 0.
345 static inline void htb_add_class_to_row(struct htb_sched *q,
346 struct htb_class *cl, int mask)
348 q->row_mask[cl->level] |= mask;
349 while (mask) {
350 int prio = ffz(~mask);
351 mask &= ~(1 << prio);
352 htb_add_to_id_tree(q->row[cl->level] + prio, cl, prio);
356 /* If this triggers, it is a bug in this code, but it need not be fatal */
357 static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root)
359 if (RB_EMPTY_NODE(rb)) {
360 WARN_ON(1);
361 } else {
362 rb_erase(rb, root);
363 RB_CLEAR_NODE(rb);
369 * htb_remove_class_from_row - removes class from its row
371 * The class is removed from row at priorities marked in mask.
372 * It does nothing if mask == 0.
374 static inline void htb_remove_class_from_row(struct htb_sched *q,
375 struct htb_class *cl, int mask)
377 int m = 0;
379 while (mask) {
380 int prio = ffz(~mask);
382 mask &= ~(1 << prio);
383 if (q->ptr[cl->level][prio] == cl->node + prio)
384 htb_next_rb_node(q->ptr[cl->level] + prio);
386 htb_safe_rb_erase(cl->node + prio, q->row[cl->level] + prio);
387 if (!q->row[cl->level][prio].rb_node)
388 m |= 1 << prio;
390 q->row_mask[cl->level] &= ~m;
394 * htb_activate_prios - creates active classe's feed chain
396 * The class is connected to ancestors and/or appropriate rows
397 * for priorities it is participating on. cl->cmode must be new
398 * (activated) mode. It does nothing if cl->prio_activity == 0.
400 static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl)
402 struct htb_class *p = cl->parent;
403 long m, mask = cl->prio_activity;
405 while (cl->cmode == HTB_MAY_BORROW && p && mask) {
406 m = mask;
407 while (m) {
408 int prio = ffz(~m);
409 m &= ~(1 << prio);
411 if (p->un.inner.feed[prio].rb_node)
412 /* parent already has its feed in use so that
413 reset bit in mask as parent is already ok */
414 mask &= ~(1 << prio);
416 htb_add_to_id_tree(p->un.inner.feed + prio, cl, prio);
418 p->prio_activity |= mask;
419 cl = p;
420 p = cl->parent;
423 if (cl->cmode == HTB_CAN_SEND && mask)
424 htb_add_class_to_row(q, cl, mask);
428 * htb_deactivate_prios - remove class from feed chain
430 * cl->cmode must represent old mode (before deactivation). It does
431 * nothing if cl->prio_activity == 0. Class is removed from all feed
432 * chains and rows.
434 static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl)
436 struct htb_class *p = cl->parent;
437 long m, mask = cl->prio_activity;
439 while (cl->cmode == HTB_MAY_BORROW && p && mask) {
440 m = mask;
441 mask = 0;
442 while (m) {
443 int prio = ffz(~m);
444 m &= ~(1 << prio);
446 if (p->un.inner.ptr[prio] == cl->node + prio) {
447 /* we are removing child which is pointed to from
448 parent feed - forget the pointer but remember
449 classid */
450 p->un.inner.last_ptr_id[prio] = cl->classid;
451 p->un.inner.ptr[prio] = NULL;
454 htb_safe_rb_erase(cl->node + prio, p->un.inner.feed + prio);
456 if (!p->un.inner.feed[prio].rb_node)
457 mask |= 1 << prio;
460 p->prio_activity &= ~mask;
461 cl = p;
462 p = cl->parent;
465 if (cl->cmode == HTB_CAN_SEND && mask)
466 htb_remove_class_from_row(q, cl, mask);
469 #if HTB_HYSTERESIS
470 static inline long htb_lowater(const struct htb_class *cl)
472 return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0;
474 static inline long htb_hiwater(const struct htb_class *cl)
476 return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0;
478 #else
479 #define htb_lowater(cl) (0)
480 #define htb_hiwater(cl) (0)
481 #endif
484 * htb_class_mode - computes and returns current class mode
486 * It computes cl's mode at time cl->t_c+diff and returns it. If mode
487 * is not HTB_CAN_SEND then cl->pq_key is updated to time difference
488 * from now to time when cl will change its state.
489 * Also it is worth to note that class mode doesn't change simply
490 * at cl->{c,}tokens == 0 but there can rather be hysteresis of
491 * 0 .. -cl->{c,}buffer range. It is meant to limit number of
492 * mode transitions per time unit. The speed gain is about 1/6.
494 static inline enum htb_cmode
495 htb_class_mode(struct htb_class *cl, long *diff)
497 long toks;
499 if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) {
500 *diff = -toks;
501 return HTB_CANT_SEND;
504 if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl))
505 return HTB_CAN_SEND;
507 *diff = -toks;
508 return HTB_MAY_BORROW;
512 * htb_change_class_mode - changes classe's mode
514 * This should be the only way how to change classe's mode under normal
515 * cirsumstances. Routine will update feed lists linkage, change mode
516 * and add class to the wait event queue if appropriate. New mode should
517 * be different from old one and cl->pq_key has to be valid if changing
518 * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree).
520 static void
521 htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, long *diff)
523 enum htb_cmode new_mode = htb_class_mode(cl, diff);
525 if (new_mode == cl->cmode)
526 return;
528 if (cl->prio_activity) { /* not necessary: speed optimization */
529 if (cl->cmode != HTB_CANT_SEND)
530 htb_deactivate_prios(q, cl);
531 cl->cmode = new_mode;
532 if (new_mode != HTB_CANT_SEND)
533 htb_activate_prios(q, cl);
534 } else
535 cl->cmode = new_mode;
539 * htb_activate - inserts leaf cl into appropriate active feeds
541 * Routine learns (new) priority of leaf and activates feed chain
542 * for the prio. It can be called on already active leaf safely.
543 * It also adds leaf into droplist.
545 static inline void htb_activate(struct htb_sched *q, struct htb_class *cl)
547 BUG_TRAP(!cl->level && cl->un.leaf.q && cl->un.leaf.q->q.qlen);
549 if (!cl->prio_activity) {
550 cl->prio_activity = 1 << (cl->un.leaf.aprio = cl->un.leaf.prio);
551 htb_activate_prios(q, cl);
552 list_add_tail(&cl->un.leaf.drop_list,
553 q->drops + cl->un.leaf.aprio);
558 * htb_deactivate - remove leaf cl from active feeds
560 * Make sure that leaf is active. In the other words it can't be called
561 * with non-active leaf. It also removes class from the drop list.
563 static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl)
565 BUG_TRAP(cl->prio_activity);
567 htb_deactivate_prios(q, cl);
568 cl->prio_activity = 0;
569 list_del_init(&cl->un.leaf.drop_list);
572 static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch)
574 int ret;
575 struct htb_sched *q = qdisc_priv(sch);
576 struct htb_class *cl = htb_classify(skb, sch, &ret);
578 if (cl == HTB_DIRECT) {
579 /* enqueue to helper queue */
580 if (q->direct_queue.qlen < q->direct_qlen) {
581 __skb_queue_tail(&q->direct_queue, skb);
582 q->direct_pkts++;
583 } else {
584 kfree_skb(skb);
585 sch->qstats.drops++;
586 return NET_XMIT_DROP;
588 #ifdef CONFIG_NET_CLS_ACT
589 } else if (!cl) {
590 if (ret == NET_XMIT_BYPASS)
591 sch->qstats.drops++;
592 kfree_skb(skb);
593 return ret;
594 #endif
595 } else if (cl->un.leaf.q->enqueue(skb, cl->un.leaf.q) !=
596 NET_XMIT_SUCCESS) {
597 sch->qstats.drops++;
598 cl->qstats.drops++;
599 return NET_XMIT_DROP;
600 } else {
601 cl->bstats.packets +=
602 skb_is_gso(skb)?skb_shinfo(skb)->gso_segs:1;
603 cl->bstats.bytes += skb->len;
604 htb_activate(q, cl);
607 sch->q.qlen++;
608 sch->bstats.packets += skb_is_gso(skb)?skb_shinfo(skb)->gso_segs:1;
609 sch->bstats.bytes += skb->len;
610 return NET_XMIT_SUCCESS;
613 /* TODO: requeuing packet charges it to policers again !! */
614 static int htb_requeue(struct sk_buff *skb, struct Qdisc *sch)
616 struct htb_sched *q = qdisc_priv(sch);
617 int ret = NET_XMIT_SUCCESS;
618 struct htb_class *cl = htb_classify(skb, sch, &ret);
619 struct sk_buff *tskb;
621 if (cl == HTB_DIRECT || !cl) {
622 /* enqueue to helper queue */
623 if (q->direct_queue.qlen < q->direct_qlen && cl) {
624 __skb_queue_head(&q->direct_queue, skb);
625 } else {
626 __skb_queue_head(&q->direct_queue, skb);
627 tskb = __skb_dequeue_tail(&q->direct_queue);
628 kfree_skb(tskb);
629 sch->qstats.drops++;
630 return NET_XMIT_CN;
632 } else if (cl->un.leaf.q->ops->requeue(skb, cl->un.leaf.q) !=
633 NET_XMIT_SUCCESS) {
634 sch->qstats.drops++;
635 cl->qstats.drops++;
636 return NET_XMIT_DROP;
637 } else
638 htb_activate(q, cl);
640 sch->q.qlen++;
641 sch->qstats.requeues++;
642 return NET_XMIT_SUCCESS;
646 * htb_charge_class - charges amount "bytes" to leaf and ancestors
648 * Routine assumes that packet "bytes" long was dequeued from leaf cl
649 * borrowing from "level". It accounts bytes to ceil leaky bucket for
650 * leaf and all ancestors and to rate bucket for ancestors at levels
651 * "level" and higher. It also handles possible change of mode resulting
652 * from the update. Note that mode can also increase here (MAY_BORROW to
653 * CAN_SEND) because we can use more precise clock that event queue here.
654 * In such case we remove class from event queue first.
656 static void htb_charge_class(struct htb_sched *q, struct htb_class *cl,
657 int level, struct sk_buff *skb)
659 int bytes = skb->len;
660 long toks, diff;
661 enum htb_cmode old_mode;
663 #define HTB_ACCNT(T,B,R) toks = diff + cl->T; \
664 if (toks > cl->B) toks = cl->B; \
665 toks -= L2T(cl, cl->R, bytes); \
666 if (toks <= -cl->mbuffer) toks = 1-cl->mbuffer; \
667 cl->T = toks
669 while (cl) {
670 diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer);
671 if (cl->level >= level) {
672 if (cl->level == level)
673 cl->xstats.lends++;
674 HTB_ACCNT(tokens, buffer, rate);
675 } else {
676 cl->xstats.borrows++;
677 cl->tokens += diff; /* we moved t_c; update tokens */
679 HTB_ACCNT(ctokens, cbuffer, ceil);
680 cl->t_c = q->now;
682 old_mode = cl->cmode;
683 diff = 0;
684 htb_change_class_mode(q, cl, &diff);
685 if (old_mode != cl->cmode) {
686 if (old_mode != HTB_CAN_SEND)
687 htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level);
688 if (cl->cmode != HTB_CAN_SEND)
689 htb_add_to_wait_tree(q, cl, diff);
692 /* update byte stats except for leaves which are already updated */
693 if (cl->level) {
694 cl->bstats.bytes += bytes;
695 cl->bstats.packets += skb_is_gso(skb)?
696 skb_shinfo(skb)->gso_segs:1;
698 cl = cl->parent;
703 * htb_do_events - make mode changes to classes at the level
705 * Scans event queue for pending events and applies them. Returns time of
706 * next pending event (0 for no event in pq).
707 * Note: Applied are events whose have cl->pq_key <= q->now.
709 static psched_time_t htb_do_events(struct htb_sched *q, int level)
711 int i;
713 for (i = 0; i < 500; i++) {
714 struct htb_class *cl;
715 long diff;
716 struct rb_node *p = rb_first(&q->wait_pq[level]);
718 if (!p)
719 return 0;
721 cl = rb_entry(p, struct htb_class, pq_node);
722 if (cl->pq_key > q->now)
723 return cl->pq_key;
725 htb_safe_rb_erase(p, q->wait_pq + level);
726 diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer);
727 htb_change_class_mode(q, cl, &diff);
728 if (cl->cmode != HTB_CAN_SEND)
729 htb_add_to_wait_tree(q, cl, diff);
731 if (net_ratelimit())
732 printk(KERN_WARNING "htb: too many events !\n");
733 return q->now + PSCHED_TICKS_PER_SEC / 10;
736 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL
737 is no such one exists. */
738 static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n,
739 u32 id)
741 struct rb_node *r = NULL;
742 while (n) {
743 struct htb_class *cl =
744 rb_entry(n, struct htb_class, node[prio]);
745 if (id == cl->classid)
746 return n;
748 if (id > cl->classid) {
749 n = n->rb_right;
750 } else {
751 r = n;
752 n = n->rb_left;
755 return r;
759 * htb_lookup_leaf - returns next leaf class in DRR order
761 * Find leaf where current feed pointers points to.
763 static struct htb_class *htb_lookup_leaf(struct rb_root *tree, int prio,
764 struct rb_node **pptr, u32 * pid)
766 int i;
767 struct {
768 struct rb_node *root;
769 struct rb_node **pptr;
770 u32 *pid;
771 } stk[TC_HTB_MAXDEPTH], *sp = stk;
773 BUG_TRAP(tree->rb_node);
774 sp->root = tree->rb_node;
775 sp->pptr = pptr;
776 sp->pid = pid;
778 for (i = 0; i < 65535; i++) {
779 if (!*sp->pptr && *sp->pid) {
780 /* ptr was invalidated but id is valid - try to recover
781 the original or next ptr */
782 *sp->pptr =
783 htb_id_find_next_upper(prio, sp->root, *sp->pid);
785 *sp->pid = 0; /* ptr is valid now so that remove this hint as it
786 can become out of date quickly */
787 if (!*sp->pptr) { /* we are at right end; rewind & go up */
788 *sp->pptr = sp->root;
789 while ((*sp->pptr)->rb_left)
790 *sp->pptr = (*sp->pptr)->rb_left;
791 if (sp > stk) {
792 sp--;
793 BUG_TRAP(*sp->pptr);
794 if (!*sp->pptr)
795 return NULL;
796 htb_next_rb_node(sp->pptr);
798 } else {
799 struct htb_class *cl;
800 cl = rb_entry(*sp->pptr, struct htb_class, node[prio]);
801 if (!cl->level)
802 return cl;
803 (++sp)->root = cl->un.inner.feed[prio].rb_node;
804 sp->pptr = cl->un.inner.ptr + prio;
805 sp->pid = cl->un.inner.last_ptr_id + prio;
808 BUG_TRAP(0);
809 return NULL;
812 /* dequeues packet at given priority and level; call only if
813 you are sure that there is active class at prio/level */
814 static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, int prio,
815 int level)
817 struct sk_buff *skb = NULL;
818 struct htb_class *cl, *start;
819 /* look initial class up in the row */
820 start = cl = htb_lookup_leaf(q->row[level] + prio, prio,
821 q->ptr[level] + prio,
822 q->last_ptr_id[level] + prio);
824 do {
825 next:
826 BUG_TRAP(cl);
827 if (!cl)
828 return NULL;
830 /* class can be empty - it is unlikely but can be true if leaf
831 qdisc drops packets in enqueue routine or if someone used
832 graft operation on the leaf since last dequeue;
833 simply deactivate and skip such class */
834 if (unlikely(cl->un.leaf.q->q.qlen == 0)) {
835 struct htb_class *next;
836 htb_deactivate(q, cl);
838 /* row/level might become empty */
839 if ((q->row_mask[level] & (1 << prio)) == 0)
840 return NULL;
842 next = htb_lookup_leaf(q->row[level] + prio,
843 prio, q->ptr[level] + prio,
844 q->last_ptr_id[level] + prio);
846 if (cl == start) /* fix start if we just deleted it */
847 start = next;
848 cl = next;
849 goto next;
852 skb = cl->un.leaf.q->dequeue(cl->un.leaf.q);
853 if (likely(skb != NULL))
854 break;
855 if (!cl->warned) {
856 printk(KERN_WARNING
857 "htb: class %X isn't work conserving ?!\n",
858 cl->classid);
859 cl->warned = 1;
861 q->nwc_hit++;
862 htb_next_rb_node((level ? cl->parent->un.inner.ptr : q->
863 ptr[0]) + prio);
864 cl = htb_lookup_leaf(q->row[level] + prio, prio,
865 q->ptr[level] + prio,
866 q->last_ptr_id[level] + prio);
868 } while (cl != start);
870 if (likely(skb != NULL)) {
871 if ((cl->un.leaf.deficit[level] -= skb->len) < 0) {
872 cl->un.leaf.deficit[level] += cl->un.leaf.quantum;
873 htb_next_rb_node((level ? cl->parent->un.inner.ptr : q->
874 ptr[0]) + prio);
876 /* this used to be after charge_class but this constelation
877 gives us slightly better performance */
878 if (!cl->un.leaf.q->q.qlen)
879 htb_deactivate(q, cl);
880 htb_charge_class(q, cl, level, skb);
882 return skb;
885 static struct sk_buff *htb_dequeue(struct Qdisc *sch)
887 struct sk_buff *skb = NULL;
888 struct htb_sched *q = qdisc_priv(sch);
889 int level;
890 psched_time_t next_event;
892 /* try to dequeue direct packets as high prio (!) to minimize cpu work */
893 skb = __skb_dequeue(&q->direct_queue);
894 if (skb != NULL) {
895 sch->flags &= ~TCQ_F_THROTTLED;
896 sch->q.qlen--;
897 return skb;
900 if (!sch->q.qlen)
901 goto fin;
902 q->now = psched_get_time();
904 next_event = q->now + 5 * PSCHED_TICKS_PER_SEC;
905 q->nwc_hit = 0;
906 for (level = 0; level < TC_HTB_MAXDEPTH; level++) {
907 /* common case optimization - skip event handler quickly */
908 int m;
909 psched_time_t event;
911 if (q->now >= q->near_ev_cache[level]) {
912 event = htb_do_events(q, level);
913 if (!event)
914 event = q->now + PSCHED_TICKS_PER_SEC;
915 q->near_ev_cache[level] = event;
916 } else
917 event = q->near_ev_cache[level];
919 if (event && next_event > event)
920 next_event = event;
922 m = ~q->row_mask[level];
923 while (m != (int)(-1)) {
924 int prio = ffz(m);
925 m |= 1 << prio;
926 skb = htb_dequeue_tree(q, prio, level);
927 if (likely(skb != NULL)) {
928 sch->q.qlen--;
929 sch->flags &= ~TCQ_F_THROTTLED;
930 goto fin;
934 sch->qstats.overlimits++;
935 qdisc_watchdog_schedule(&q->watchdog, next_event);
936 fin:
937 return skb;
940 /* try to drop from each class (by prio) until one succeed */
941 static unsigned int htb_drop(struct Qdisc *sch)
943 struct htb_sched *q = qdisc_priv(sch);
944 int prio;
946 for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) {
947 struct list_head *p;
948 list_for_each(p, q->drops + prio) {
949 struct htb_class *cl = list_entry(p, struct htb_class,
950 un.leaf.drop_list);
951 unsigned int len;
952 if (cl->un.leaf.q->ops->drop &&
953 (len = cl->un.leaf.q->ops->drop(cl->un.leaf.q))) {
954 sch->q.qlen--;
955 if (!cl->un.leaf.q->q.qlen)
956 htb_deactivate(q, cl);
957 return len;
961 return 0;
964 /* reset all classes */
965 /* always caled under BH & queue lock */
966 static void htb_reset(struct Qdisc *sch)
968 struct htb_sched *q = qdisc_priv(sch);
969 int i;
971 for (i = 0; i < HTB_HSIZE; i++) {
972 struct hlist_node *p;
973 struct htb_class *cl;
975 hlist_for_each_entry(cl, p, q->hash + i, hlist) {
976 if (cl->level)
977 memset(&cl->un.inner, 0, sizeof(cl->un.inner));
978 else {
979 if (cl->un.leaf.q)
980 qdisc_reset(cl->un.leaf.q);
981 INIT_LIST_HEAD(&cl->un.leaf.drop_list);
983 cl->prio_activity = 0;
984 cl->cmode = HTB_CAN_SEND;
988 qdisc_watchdog_cancel(&q->watchdog);
989 __skb_queue_purge(&q->direct_queue);
990 sch->q.qlen = 0;
991 memset(q->row, 0, sizeof(q->row));
992 memset(q->row_mask, 0, sizeof(q->row_mask));
993 memset(q->wait_pq, 0, sizeof(q->wait_pq));
994 memset(q->ptr, 0, sizeof(q->ptr));
995 for (i = 0; i < TC_HTB_NUMPRIO; i++)
996 INIT_LIST_HEAD(q->drops + i);
999 static int htb_init(struct Qdisc *sch, struct rtattr *opt)
1001 struct htb_sched *q = qdisc_priv(sch);
1002 struct rtattr *tb[TCA_HTB_INIT];
1003 struct tc_htb_glob *gopt;
1004 int i;
1005 if (!opt || rtattr_parse_nested(tb, TCA_HTB_INIT, opt) ||
1006 tb[TCA_HTB_INIT - 1] == NULL ||
1007 RTA_PAYLOAD(tb[TCA_HTB_INIT - 1]) < sizeof(*gopt)) {
1008 printk(KERN_ERR "HTB: hey probably you have bad tc tool ?\n");
1009 return -EINVAL;
1011 gopt = RTA_DATA(tb[TCA_HTB_INIT - 1]);
1012 if (gopt->version != HTB_VER >> 16) {
1013 printk(KERN_ERR
1014 "HTB: need tc/htb version %d (minor is %d), you have %d\n",
1015 HTB_VER >> 16, HTB_VER & 0xffff, gopt->version);
1016 return -EINVAL;
1019 INIT_LIST_HEAD(&q->root);
1020 for (i = 0; i < HTB_HSIZE; i++)
1021 INIT_HLIST_HEAD(q->hash + i);
1022 for (i = 0; i < TC_HTB_NUMPRIO; i++)
1023 INIT_LIST_HEAD(q->drops + i);
1025 qdisc_watchdog_init(&q->watchdog, sch);
1026 skb_queue_head_init(&q->direct_queue);
1028 q->direct_qlen = sch->dev->tx_queue_len;
1029 if (q->direct_qlen < 2) /* some devices have zero tx_queue_len */
1030 q->direct_qlen = 2;
1032 if ((q->rate2quantum = gopt->rate2quantum) < 1)
1033 q->rate2quantum = 1;
1034 q->defcls = gopt->defcls;
1036 return 0;
1039 static int htb_dump(struct Qdisc *sch, struct sk_buff *skb)
1041 struct htb_sched *q = qdisc_priv(sch);
1042 unsigned char *b = skb_tail_pointer(skb);
1043 struct rtattr *rta;
1044 struct tc_htb_glob gopt;
1045 spin_lock_bh(&sch->dev->queue_lock);
1046 gopt.direct_pkts = q->direct_pkts;
1048 gopt.version = HTB_VER;
1049 gopt.rate2quantum = q->rate2quantum;
1050 gopt.defcls = q->defcls;
1051 gopt.debug = 0;
1052 rta = (struct rtattr *)b;
1053 RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
1054 RTA_PUT(skb, TCA_HTB_INIT, sizeof(gopt), &gopt);
1055 rta->rta_len = skb_tail_pointer(skb) - b;
1056 spin_unlock_bh(&sch->dev->queue_lock);
1057 return skb->len;
1058 rtattr_failure:
1059 spin_unlock_bh(&sch->dev->queue_lock);
1060 nlmsg_trim(skb, skb_tail_pointer(skb));
1061 return -1;
1064 static int htb_dump_class(struct Qdisc *sch, unsigned long arg,
1065 struct sk_buff *skb, struct tcmsg *tcm)
1067 struct htb_class *cl = (struct htb_class *)arg;
1068 unsigned char *b = skb_tail_pointer(skb);
1069 struct rtattr *rta;
1070 struct tc_htb_opt opt;
1072 spin_lock_bh(&sch->dev->queue_lock);
1073 tcm->tcm_parent = cl->parent ? cl->parent->classid : TC_H_ROOT;
1074 tcm->tcm_handle = cl->classid;
1075 if (!cl->level && cl->un.leaf.q)
1076 tcm->tcm_info = cl->un.leaf.q->handle;
1078 rta = (struct rtattr *)b;
1079 RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
1081 memset(&opt, 0, sizeof(opt));
1083 opt.rate = cl->rate->rate;
1084 opt.buffer = cl->buffer;
1085 opt.ceil = cl->ceil->rate;
1086 opt.cbuffer = cl->cbuffer;
1087 opt.quantum = cl->un.leaf.quantum;
1088 opt.prio = cl->un.leaf.prio;
1089 opt.level = cl->level;
1090 RTA_PUT(skb, TCA_HTB_PARMS, sizeof(opt), &opt);
1091 rta->rta_len = skb_tail_pointer(skb) - b;
1092 spin_unlock_bh(&sch->dev->queue_lock);
1093 return skb->len;
1094 rtattr_failure:
1095 spin_unlock_bh(&sch->dev->queue_lock);
1096 nlmsg_trim(skb, b);
1097 return -1;
1100 static int
1101 htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d)
1103 struct htb_class *cl = (struct htb_class *)arg;
1105 if (!cl->level && cl->un.leaf.q)
1106 cl->qstats.qlen = cl->un.leaf.q->q.qlen;
1107 cl->xstats.tokens = cl->tokens;
1108 cl->xstats.ctokens = cl->ctokens;
1110 if (gnet_stats_copy_basic(d, &cl->bstats) < 0 ||
1111 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
1112 gnet_stats_copy_queue(d, &cl->qstats) < 0)
1113 return -1;
1115 return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats));
1118 static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1119 struct Qdisc **old)
1121 struct htb_class *cl = (struct htb_class *)arg;
1123 if (cl && !cl->level) {
1124 if (new == NULL &&
1125 (new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
1126 cl->classid))
1127 == NULL)
1128 return -ENOBUFS;
1129 sch_tree_lock(sch);
1130 if ((*old = xchg(&cl->un.leaf.q, new)) != NULL) {
1131 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
1132 qdisc_reset(*old);
1134 sch_tree_unlock(sch);
1135 return 0;
1137 return -ENOENT;
1140 static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg)
1142 struct htb_class *cl = (struct htb_class *)arg;
1143 return (cl && !cl->level) ? cl->un.leaf.q : NULL;
1146 static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg)
1148 struct htb_class *cl = (struct htb_class *)arg;
1150 if (cl->un.leaf.q->q.qlen == 0)
1151 htb_deactivate(qdisc_priv(sch), cl);
1154 static unsigned long htb_get(struct Qdisc *sch, u32 classid)
1156 struct htb_class *cl = htb_find(classid, sch);
1157 if (cl)
1158 cl->refcnt++;
1159 return (unsigned long)cl;
1162 static inline int htb_parent_last_child(struct htb_class *cl)
1164 if (!cl->parent)
1165 /* the root class */
1166 return 0;
1168 if (!(cl->parent->children.next == &cl->sibling &&
1169 cl->parent->children.prev == &cl->sibling))
1170 /* not the last child */
1171 return 0;
1173 return 1;
1176 static void htb_parent_to_leaf(struct htb_class *cl, struct Qdisc *new_q)
1178 struct htb_class *parent = cl->parent;
1180 BUG_TRAP(!cl->level && cl->un.leaf.q && !cl->prio_activity);
1182 parent->level = 0;
1183 memset(&parent->un.inner, 0, sizeof(parent->un.inner));
1184 INIT_LIST_HEAD(&parent->un.leaf.drop_list);
1185 parent->un.leaf.q = new_q ? new_q : &noop_qdisc;
1186 parent->un.leaf.quantum = parent->quantum;
1187 parent->un.leaf.prio = parent->prio;
1188 parent->tokens = parent->buffer;
1189 parent->ctokens = parent->cbuffer;
1190 parent->t_c = psched_get_time();
1191 parent->cmode = HTB_CAN_SEND;
1194 static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl)
1196 struct htb_sched *q = qdisc_priv(sch);
1198 if (!cl->level) {
1199 BUG_TRAP(cl->un.leaf.q);
1200 qdisc_destroy(cl->un.leaf.q);
1202 gen_kill_estimator(&cl->bstats, &cl->rate_est);
1203 qdisc_put_rtab(cl->rate);
1204 qdisc_put_rtab(cl->ceil);
1206 tcf_destroy_chain(cl->filter_list);
1208 while (!list_empty(&cl->children))
1209 htb_destroy_class(sch, list_entry(cl->children.next,
1210 struct htb_class, sibling));
1212 /* note: this delete may happen twice (see htb_delete) */
1213 hlist_del_init(&cl->hlist);
1214 list_del(&cl->sibling);
1216 if (cl->prio_activity)
1217 htb_deactivate(q, cl);
1219 if (cl->cmode != HTB_CAN_SEND)
1220 htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level);
1222 kfree(cl);
1225 /* always caled under BH & queue lock */
1226 static void htb_destroy(struct Qdisc *sch)
1228 struct htb_sched *q = qdisc_priv(sch);
1230 qdisc_watchdog_cancel(&q->watchdog);
1231 /* This line used to be after htb_destroy_class call below
1232 and surprisingly it worked in 2.4. But it must precede it
1233 because filter need its target class alive to be able to call
1234 unbind_filter on it (without Oops). */
1235 tcf_destroy_chain(q->filter_list);
1237 while (!list_empty(&q->root))
1238 htb_destroy_class(sch, list_entry(q->root.next,
1239 struct htb_class, sibling));
1241 __skb_queue_purge(&q->direct_queue);
1244 static int htb_delete(struct Qdisc *sch, unsigned long arg)
1246 struct htb_sched *q = qdisc_priv(sch);
1247 struct htb_class *cl = (struct htb_class *)arg;
1248 unsigned int qlen;
1249 struct Qdisc *new_q = NULL;
1250 int last_child = 0;
1252 // TODO: why don't allow to delete subtree ? references ? does
1253 // tc subsys quarantee us that in htb_destroy it holds no class
1254 // refs so that we can remove children safely there ?
1255 if (!list_empty(&cl->children) || cl->filter_cnt)
1256 return -EBUSY;
1258 if (!cl->level && htb_parent_last_child(cl)) {
1259 new_q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
1260 cl->parent->classid);
1261 last_child = 1;
1264 sch_tree_lock(sch);
1266 if (!cl->level) {
1267 qlen = cl->un.leaf.q->q.qlen;
1268 qdisc_reset(cl->un.leaf.q);
1269 qdisc_tree_decrease_qlen(cl->un.leaf.q, qlen);
1272 /* delete from hash and active; remainder in destroy_class */
1273 hlist_del_init(&cl->hlist);
1275 if (cl->prio_activity)
1276 htb_deactivate(q, cl);
1278 if (last_child)
1279 htb_parent_to_leaf(cl, new_q);
1281 if (--cl->refcnt == 0)
1282 htb_destroy_class(sch, cl);
1284 sch_tree_unlock(sch);
1285 return 0;
1288 static void htb_put(struct Qdisc *sch, unsigned long arg)
1290 struct htb_class *cl = (struct htb_class *)arg;
1292 if (--cl->refcnt == 0)
1293 htb_destroy_class(sch, cl);
1296 static int htb_change_class(struct Qdisc *sch, u32 classid,
1297 u32 parentid, struct rtattr **tca,
1298 unsigned long *arg)
1300 int err = -EINVAL;
1301 struct htb_sched *q = qdisc_priv(sch);
1302 struct htb_class *cl = (struct htb_class *)*arg, *parent;
1303 struct rtattr *opt = tca[TCA_OPTIONS - 1];
1304 struct qdisc_rate_table *rtab = NULL, *ctab = NULL;
1305 struct rtattr *tb[TCA_HTB_RTAB];
1306 struct tc_htb_opt *hopt;
1308 /* extract all subattrs from opt attr */
1309 if (!opt || rtattr_parse_nested(tb, TCA_HTB_RTAB, opt) ||
1310 tb[TCA_HTB_PARMS - 1] == NULL ||
1311 RTA_PAYLOAD(tb[TCA_HTB_PARMS - 1]) < sizeof(*hopt))
1312 goto failure;
1314 parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch);
1316 hopt = RTA_DATA(tb[TCA_HTB_PARMS - 1]);
1318 rtab = qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB - 1]);
1319 ctab = qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB - 1]);
1320 if (!rtab || !ctab)
1321 goto failure;
1323 if (!cl) { /* new class */
1324 struct Qdisc *new_q;
1325 int prio;
1326 struct {
1327 struct rtattr rta;
1328 struct gnet_estimator opt;
1329 } est = {
1330 .rta = {
1331 .rta_len = RTA_LENGTH(sizeof(est.opt)),
1332 .rta_type = TCA_RATE,
1334 .opt = {
1335 /* 4s interval, 16s averaging constant */
1336 .interval = 2,
1337 .ewma_log = 2,
1341 /* check for valid classid */
1342 if (!classid || TC_H_MAJ(classid ^ sch->handle)
1343 || htb_find(classid, sch))
1344 goto failure;
1346 /* check maximal depth */
1347 if (parent && parent->parent && parent->parent->level < 2) {
1348 printk(KERN_ERR "htb: tree is too deep\n");
1349 goto failure;
1351 err = -ENOBUFS;
1352 if ((cl = kzalloc(sizeof(*cl), GFP_KERNEL)) == NULL)
1353 goto failure;
1355 gen_new_estimator(&cl->bstats, &cl->rate_est,
1356 &sch->dev->queue_lock,
1357 tca[TCA_RATE-1] ? : &est.rta);
1358 cl->refcnt = 1;
1359 INIT_LIST_HEAD(&cl->sibling);
1360 INIT_HLIST_NODE(&cl->hlist);
1361 INIT_LIST_HEAD(&cl->children);
1362 INIT_LIST_HEAD(&cl->un.leaf.drop_list);
1363 RB_CLEAR_NODE(&cl->pq_node);
1365 for (prio = 0; prio < TC_HTB_NUMPRIO; prio++)
1366 RB_CLEAR_NODE(&cl->node[prio]);
1368 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL)
1369 so that can't be used inside of sch_tree_lock
1370 -- thanks to Karlis Peisenieks */
1371 new_q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, classid);
1372 sch_tree_lock(sch);
1373 if (parent && !parent->level) {
1374 unsigned int qlen = parent->un.leaf.q->q.qlen;
1376 /* turn parent into inner node */
1377 qdisc_reset(parent->un.leaf.q);
1378 qdisc_tree_decrease_qlen(parent->un.leaf.q, qlen);
1379 qdisc_destroy(parent->un.leaf.q);
1380 if (parent->prio_activity)
1381 htb_deactivate(q, parent);
1383 /* remove from evt list because of level change */
1384 if (parent->cmode != HTB_CAN_SEND) {
1385 htb_safe_rb_erase(&parent->pq_node, q->wait_pq);
1386 parent->cmode = HTB_CAN_SEND;
1388 parent->level = (parent->parent ? parent->parent->level
1389 : TC_HTB_MAXDEPTH) - 1;
1390 memset(&parent->un.inner, 0, sizeof(parent->un.inner));
1392 /* leaf (we) needs elementary qdisc */
1393 cl->un.leaf.q = new_q ? new_q : &noop_qdisc;
1395 cl->classid = classid;
1396 cl->parent = parent;
1398 /* set class to be in HTB_CAN_SEND state */
1399 cl->tokens = hopt->buffer;
1400 cl->ctokens = hopt->cbuffer;
1401 cl->mbuffer = 60 * PSCHED_TICKS_PER_SEC; /* 1min */
1402 cl->t_c = psched_get_time();
1403 cl->cmode = HTB_CAN_SEND;
1405 /* attach to the hash list and parent's family */
1406 hlist_add_head(&cl->hlist, q->hash + htb_hash(classid));
1407 list_add_tail(&cl->sibling,
1408 parent ? &parent->children : &q->root);
1409 } else {
1410 if (tca[TCA_RATE-1])
1411 gen_replace_estimator(&cl->bstats, &cl->rate_est,
1412 &sch->dev->queue_lock,
1413 tca[TCA_RATE-1]);
1414 sch_tree_lock(sch);
1417 /* it used to be a nasty bug here, we have to check that node
1418 is really leaf before changing cl->un.leaf ! */
1419 if (!cl->level) {
1420 cl->un.leaf.quantum = rtab->rate.rate / q->rate2quantum;
1421 if (!hopt->quantum && cl->un.leaf.quantum < 1000) {
1422 printk(KERN_WARNING
1423 "HTB: quantum of class %X is small. Consider r2q change.\n",
1424 cl->classid);
1425 cl->un.leaf.quantum = 1000;
1427 if (!hopt->quantum && cl->un.leaf.quantum > 200000) {
1428 printk(KERN_WARNING
1429 "HTB: quantum of class %X is big. Consider r2q change.\n",
1430 cl->classid);
1431 cl->un.leaf.quantum = 200000;
1433 if (hopt->quantum)
1434 cl->un.leaf.quantum = hopt->quantum;
1435 if ((cl->un.leaf.prio = hopt->prio) >= TC_HTB_NUMPRIO)
1436 cl->un.leaf.prio = TC_HTB_NUMPRIO - 1;
1438 /* backup for htb_parent_to_leaf */
1439 cl->quantum = cl->un.leaf.quantum;
1440 cl->prio = cl->un.leaf.prio;
1443 cl->buffer = hopt->buffer;
1444 cl->cbuffer = hopt->cbuffer;
1445 if (cl->rate)
1446 qdisc_put_rtab(cl->rate);
1447 cl->rate = rtab;
1448 if (cl->ceil)
1449 qdisc_put_rtab(cl->ceil);
1450 cl->ceil = ctab;
1451 sch_tree_unlock(sch);
1453 *arg = (unsigned long)cl;
1454 return 0;
1456 failure:
1457 if (rtab)
1458 qdisc_put_rtab(rtab);
1459 if (ctab)
1460 qdisc_put_rtab(ctab);
1461 return err;
1464 static struct tcf_proto **htb_find_tcf(struct Qdisc *sch, unsigned long arg)
1466 struct htb_sched *q = qdisc_priv(sch);
1467 struct htb_class *cl = (struct htb_class *)arg;
1468 struct tcf_proto **fl = cl ? &cl->filter_list : &q->filter_list;
1470 return fl;
1473 static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent,
1474 u32 classid)
1476 struct htb_sched *q = qdisc_priv(sch);
1477 struct htb_class *cl = htb_find(classid, sch);
1479 /*if (cl && !cl->level) return 0;
1480 The line above used to be there to prevent attaching filters to
1481 leaves. But at least tc_index filter uses this just to get class
1482 for other reasons so that we have to allow for it.
1483 ----
1484 19.6.2002 As Werner explained it is ok - bind filter is just
1485 another way to "lock" the class - unlike "get" this lock can
1486 be broken by class during destroy IIUC.
1488 if (cl)
1489 cl->filter_cnt++;
1490 else
1491 q->filter_cnt++;
1492 return (unsigned long)cl;
1495 static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg)
1497 struct htb_sched *q = qdisc_priv(sch);
1498 struct htb_class *cl = (struct htb_class *)arg;
1500 if (cl)
1501 cl->filter_cnt--;
1502 else
1503 q->filter_cnt--;
1506 static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1508 struct htb_sched *q = qdisc_priv(sch);
1509 int i;
1511 if (arg->stop)
1512 return;
1514 for (i = 0; i < HTB_HSIZE; i++) {
1515 struct hlist_node *p;
1516 struct htb_class *cl;
1518 hlist_for_each_entry(cl, p, q->hash + i, hlist) {
1519 if (arg->count < arg->skip) {
1520 arg->count++;
1521 continue;
1523 if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
1524 arg->stop = 1;
1525 return;
1527 arg->count++;
1532 static struct Qdisc_class_ops htb_class_ops = {
1533 .graft = htb_graft,
1534 .leaf = htb_leaf,
1535 .qlen_notify = htb_qlen_notify,
1536 .get = htb_get,
1537 .put = htb_put,
1538 .change = htb_change_class,
1539 .delete = htb_delete,
1540 .walk = htb_walk,
1541 .tcf_chain = htb_find_tcf,
1542 .bind_tcf = htb_bind_filter,
1543 .unbind_tcf = htb_unbind_filter,
1544 .dump = htb_dump_class,
1545 .dump_stats = htb_dump_class_stats,
1548 static struct Qdisc_ops htb_qdisc_ops = {
1549 .next = NULL,
1550 .cl_ops = &htb_class_ops,
1551 .id = "htb",
1552 .priv_size = sizeof(struct htb_sched),
1553 .enqueue = htb_enqueue,
1554 .dequeue = htb_dequeue,
1555 .requeue = htb_requeue,
1556 .drop = htb_drop,
1557 .init = htb_init,
1558 .reset = htb_reset,
1559 .destroy = htb_destroy,
1560 .change = NULL /* htb_change */,
1561 .dump = htb_dump,
1562 .owner = THIS_MODULE,
1565 static int __init htb_module_init(void)
1567 return register_qdisc(&htb_qdisc_ops);
1569 static void __exit htb_module_exit(void)
1571 unregister_qdisc(&htb_qdisc_ops);
1574 module_init(htb_module_init)
1575 module_exit(htb_module_exit)
1576 MODULE_LICENSE("GPL");