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
23 * spotted bug in dequeue code and helped with fix
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>
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.
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"
63 /* used internaly to keep status of single class */
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: */
72 /* general class parameters */
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 */
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 */
88 struct htb_class_leaf {
93 int deficit[TC_HTB_MAXDEPTH];
94 struct list_head drop_list;
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];
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;
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,
135 long result = qdisc_l2t(rate, size);
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;
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 */
176 /* compute hash of size HTB_HSIZE for given handle */
177 static inline int htb_hash(u32 h)
180 #error "Declare new hash for your HTB_HSIZE"
182 h ^= h >> 8; /* stolen from cbq_hash */
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)
197 hlist_for_each_entry(cl, p, q->hash + htb_hash(handle), hlist) {
198 if (cl->classid == handle)
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
218 static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch,
221 struct htb_sched *q = qdisc_priv(sch);
222 struct htb_class *cl;
223 struct tcf_result res;
224 struct tcf_proto *tcf;
227 /* allow to select class by setting skb->priority to valid classid;
228 note that nfmark can be used too by attaching filter fw with no
230 if (skb->priority == sch->handle)
231 return HTB_DIRECT; /* X:0 (direct flow) selected */
232 if ((cl = htb_find(skb->priority, sch)) != NULL && cl->level == 0)
235 *qerr = NET_XMIT_BYPASS;
236 tcf = q->filter_list;
237 while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) {
238 #ifdef CONFIG_NET_CLS_ACT
242 *qerr = NET_XMIT_SUCCESS;
247 if ((cl = (void *)res.class) == NULL) {
248 if (res.classid == sch->handle)
249 return HTB_DIRECT; /* X:0 (direct flow) */
250 if ((cl = htb_find(res.classid, sch)) == NULL)
251 break; /* filter selected invalid classid */
254 return cl; /* we hit leaf; return it */
256 /* we have got inner class; apply inner filter chain */
257 tcf = cl->filter_list;
259 /* classification failed; try to use default class */
260 cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch);
261 if (!cl || cl->level)
262 return HTB_DIRECT; /* bad default .. this is safe bet */
267 * htb_add_to_id_tree - adds class to the round robin list
269 * Routine adds class to the list (actually tree) sorted by classid.
270 * Make sure that class is not already on such list for given prio.
272 static void htb_add_to_id_tree(struct rb_root *root,
273 struct htb_class *cl, int prio)
275 struct rb_node **p = &root->rb_node, *parent = NULL;
280 c = rb_entry(parent, struct htb_class, node[prio]);
282 if (cl->classid > c->classid)
283 p = &parent->rb_right;
285 p = &parent->rb_left;
287 rb_link_node(&cl->node[prio], parent, p);
288 rb_insert_color(&cl->node[prio], root);
292 * htb_add_to_wait_tree - adds class to the event queue with delay
294 * The class is added to priority event queue to indicate that class will
295 * change its mode in cl->pq_key microseconds. Make sure that class is not
296 * already in the queue.
298 static void htb_add_to_wait_tree(struct htb_sched *q,
299 struct htb_class *cl, long delay)
301 struct rb_node **p = &q->wait_pq[cl->level].rb_node, *parent = NULL;
303 cl->pq_key = q->now + delay;
304 if (cl->pq_key == q->now)
307 /* update the nearest event cache */
308 if (q->near_ev_cache[cl->level] > cl->pq_key)
309 q->near_ev_cache[cl->level] = cl->pq_key;
314 c = rb_entry(parent, struct htb_class, pq_node);
315 if (cl->pq_key >= c->pq_key)
316 p = &parent->rb_right;
318 p = &parent->rb_left;
320 rb_link_node(&cl->pq_node, parent, p);
321 rb_insert_color(&cl->pq_node, &q->wait_pq[cl->level]);
325 * htb_next_rb_node - finds next node in binary tree
327 * When we are past last key we return NULL.
328 * Average complexity is 2 steps per call.
330 static inline void htb_next_rb_node(struct rb_node **n)
336 * htb_add_class_to_row - add class to its row
338 * The class is added to row at priorities marked in mask.
339 * It does nothing if mask == 0.
341 static inline void htb_add_class_to_row(struct htb_sched *q,
342 struct htb_class *cl, int mask)
344 q->row_mask[cl->level] |= mask;
346 int prio = ffz(~mask);
347 mask &= ~(1 << prio);
348 htb_add_to_id_tree(q->row[cl->level] + prio, cl, prio);
352 /* If this triggers, it is a bug in this code, but it need not be fatal */
353 static void htb_safe_rb_erase(struct rb_node *rb, struct rb_root *root)
355 if (RB_EMPTY_NODE(rb)) {
365 * htb_remove_class_from_row - removes class from its row
367 * The class is removed from row at priorities marked in mask.
368 * It does nothing if mask == 0.
370 static inline void htb_remove_class_from_row(struct htb_sched *q,
371 struct htb_class *cl, int mask)
376 int prio = ffz(~mask);
378 mask &= ~(1 << prio);
379 if (q->ptr[cl->level][prio] == cl->node + prio)
380 htb_next_rb_node(q->ptr[cl->level] + prio);
382 htb_safe_rb_erase(cl->node + prio, q->row[cl->level] + prio);
383 if (!q->row[cl->level][prio].rb_node)
386 q->row_mask[cl->level] &= ~m;
390 * htb_activate_prios - creates active classe's feed chain
392 * The class is connected to ancestors and/or appropriate rows
393 * for priorities it is participating on. cl->cmode must be new
394 * (activated) mode. It does nothing if cl->prio_activity == 0.
396 static void htb_activate_prios(struct htb_sched *q, struct htb_class *cl)
398 struct htb_class *p = cl->parent;
399 long m, mask = cl->prio_activity;
401 while (cl->cmode == HTB_MAY_BORROW && p && mask) {
407 if (p->un.inner.feed[prio].rb_node)
408 /* parent already has its feed in use so that
409 reset bit in mask as parent is already ok */
410 mask &= ~(1 << prio);
412 htb_add_to_id_tree(p->un.inner.feed + prio, cl, prio);
414 p->prio_activity |= mask;
419 if (cl->cmode == HTB_CAN_SEND && mask)
420 htb_add_class_to_row(q, cl, mask);
424 * htb_deactivate_prios - remove class from feed chain
426 * cl->cmode must represent old mode (before deactivation). It does
427 * nothing if cl->prio_activity == 0. Class is removed from all feed
430 static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl)
432 struct htb_class *p = cl->parent;
433 long m, mask = cl->prio_activity;
435 while (cl->cmode == HTB_MAY_BORROW && p && mask) {
442 if (p->un.inner.ptr[prio] == cl->node + prio) {
443 /* we are removing child which is pointed to from
444 parent feed - forget the pointer but remember
446 p->un.inner.last_ptr_id[prio] = cl->classid;
447 p->un.inner.ptr[prio] = NULL;
450 htb_safe_rb_erase(cl->node + prio, p->un.inner.feed + prio);
452 if (!p->un.inner.feed[prio].rb_node)
456 p->prio_activity &= ~mask;
461 if (cl->cmode == HTB_CAN_SEND && mask)
462 htb_remove_class_from_row(q, cl, mask);
466 static inline long htb_lowater(const struct htb_class *cl)
468 return cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : 0;
470 static inline long htb_hiwater(const struct htb_class *cl)
472 return cl->cmode == HTB_CAN_SEND ? -cl->buffer : 0;
475 #define htb_lowater(cl) (0)
476 #define htb_hiwater(cl) (0)
480 * htb_class_mode - computes and returns current class mode
482 * It computes cl's mode at time cl->t_c+diff and returns it. If mode
483 * is not HTB_CAN_SEND then cl->pq_key is updated to time difference
484 * from now to time when cl will change its state.
485 * Also it is worth to note that class mode doesn't change simply
486 * at cl->{c,}tokens == 0 but there can rather be hysteresis of
487 * 0 .. -cl->{c,}buffer range. It is meant to limit number of
488 * mode transitions per time unit. The speed gain is about 1/6.
490 static inline enum htb_cmode
491 htb_class_mode(struct htb_class *cl, long *diff)
495 if ((toks = (cl->ctokens + *diff)) < htb_lowater(cl)) {
497 return HTB_CANT_SEND;
500 if ((toks = (cl->tokens + *diff)) >= htb_hiwater(cl))
504 return HTB_MAY_BORROW;
508 * htb_change_class_mode - changes classe's mode
510 * This should be the only way how to change classe's mode under normal
511 * cirsumstances. Routine will update feed lists linkage, change mode
512 * and add class to the wait event queue if appropriate. New mode should
513 * be different from old one and cl->pq_key has to be valid if changing
514 * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree).
517 htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, long *diff)
519 enum htb_cmode new_mode = htb_class_mode(cl, diff);
521 if (new_mode == cl->cmode)
524 if (cl->prio_activity) { /* not necessary: speed optimization */
525 if (cl->cmode != HTB_CANT_SEND)
526 htb_deactivate_prios(q, cl);
527 cl->cmode = new_mode;
528 if (new_mode != HTB_CANT_SEND)
529 htb_activate_prios(q, cl);
531 cl->cmode = new_mode;
535 * htb_activate - inserts leaf cl into appropriate active feeds
537 * Routine learns (new) priority of leaf and activates feed chain
538 * for the prio. It can be called on already active leaf safely.
539 * It also adds leaf into droplist.
541 static inline void htb_activate(struct htb_sched *q, struct htb_class *cl)
543 BUG_TRAP(!cl->level && cl->un.leaf.q && cl->un.leaf.q->q.qlen);
545 if (!cl->prio_activity) {
546 cl->prio_activity = 1 << (cl->un.leaf.aprio = cl->un.leaf.prio);
547 htb_activate_prios(q, cl);
548 list_add_tail(&cl->un.leaf.drop_list,
549 q->drops + cl->un.leaf.aprio);
554 * htb_deactivate - remove leaf cl from active feeds
556 * Make sure that leaf is active. In the other words it can't be called
557 * with non-active leaf. It also removes class from the drop list.
559 static inline void htb_deactivate(struct htb_sched *q, struct htb_class *cl)
561 BUG_TRAP(cl->prio_activity);
563 htb_deactivate_prios(q, cl);
564 cl->prio_activity = 0;
565 list_del_init(&cl->un.leaf.drop_list);
568 static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch)
571 struct htb_sched *q = qdisc_priv(sch);
572 struct htb_class *cl = htb_classify(skb, sch, &ret);
574 if (cl == HTB_DIRECT) {
575 /* enqueue to helper queue */
576 if (q->direct_queue.qlen < q->direct_qlen) {
577 __skb_queue_tail(&q->direct_queue, skb);
582 return NET_XMIT_DROP;
584 #ifdef CONFIG_NET_CLS_ACT
586 if (ret == NET_XMIT_BYPASS)
591 } else if (cl->un.leaf.q->enqueue(skb, cl->un.leaf.q) !=
595 return NET_XMIT_DROP;
597 cl->bstats.packets +=
598 skb_is_gso(skb)?skb_shinfo(skb)->gso_segs:1;
599 cl->bstats.bytes += skb->len;
604 sch->bstats.packets += skb_is_gso(skb)?skb_shinfo(skb)->gso_segs:1;
605 sch->bstats.bytes += skb->len;
606 return NET_XMIT_SUCCESS;
609 /* TODO: requeuing packet charges it to policers again !! */
610 static int htb_requeue(struct sk_buff *skb, struct Qdisc *sch)
613 struct htb_sched *q = qdisc_priv(sch);
614 struct htb_class *cl = htb_classify(skb, sch, &ret);
615 struct sk_buff *tskb;
617 if (cl == HTB_DIRECT) {
618 /* enqueue to helper queue */
619 if (q->direct_queue.qlen < q->direct_qlen) {
620 __skb_queue_head(&q->direct_queue, skb);
622 __skb_queue_head(&q->direct_queue, skb);
623 tskb = __skb_dequeue_tail(&q->direct_queue);
628 #ifdef CONFIG_NET_CLS_ACT
630 if (ret == NET_XMIT_BYPASS)
635 } else if (cl->un.leaf.q->ops->requeue(skb, cl->un.leaf.q) !=
639 return NET_XMIT_DROP;
644 sch->qstats.requeues++;
645 return NET_XMIT_SUCCESS;
649 * htb_charge_class - charges amount "bytes" to leaf and ancestors
651 * Routine assumes that packet "bytes" long was dequeued from leaf cl
652 * borrowing from "level". It accounts bytes to ceil leaky bucket for
653 * leaf and all ancestors and to rate bucket for ancestors at levels
654 * "level" and higher. It also handles possible change of mode resulting
655 * from the update. Note that mode can also increase here (MAY_BORROW to
656 * CAN_SEND) because we can use more precise clock that event queue here.
657 * In such case we remove class from event queue first.
659 static void htb_charge_class(struct htb_sched *q, struct htb_class *cl,
660 int level, struct sk_buff *skb)
662 int bytes = skb->len;
664 enum htb_cmode old_mode;
666 #define HTB_ACCNT(T,B,R) toks = diff + cl->T; \
667 if (toks > cl->B) toks = cl->B; \
668 toks -= L2T(cl, cl->R, bytes); \
669 if (toks <= -cl->mbuffer) toks = 1-cl->mbuffer; \
673 diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer);
674 if (cl->level >= level) {
675 if (cl->level == level)
677 HTB_ACCNT(tokens, buffer, rate);
679 cl->xstats.borrows++;
680 cl->tokens += diff; /* we moved t_c; update tokens */
682 HTB_ACCNT(ctokens, cbuffer, ceil);
685 old_mode = cl->cmode;
687 htb_change_class_mode(q, cl, &diff);
688 if (old_mode != cl->cmode) {
689 if (old_mode != HTB_CAN_SEND)
690 htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level);
691 if (cl->cmode != HTB_CAN_SEND)
692 htb_add_to_wait_tree(q, cl, diff);
695 /* update byte stats except for leaves which are already updated */
697 cl->bstats.bytes += bytes;
698 cl->bstats.packets += skb_is_gso(skb)?
699 skb_shinfo(skb)->gso_segs:1;
706 * htb_do_events - make mode changes to classes at the level
708 * Scans event queue for pending events and applies them. Returns time of
709 * next pending event (0 for no event in pq).
710 * Note: Applied are events whose have cl->pq_key <= q->now.
712 static psched_time_t htb_do_events(struct htb_sched *q, int level)
716 for (i = 0; i < 500; i++) {
717 struct htb_class *cl;
719 struct rb_node *p = rb_first(&q->wait_pq[level]);
724 cl = rb_entry(p, struct htb_class, pq_node);
725 if (cl->pq_key > q->now)
728 htb_safe_rb_erase(p, q->wait_pq + level);
729 diff = psched_tdiff_bounded(q->now, cl->t_c, cl->mbuffer);
730 htb_change_class_mode(q, cl, &diff);
731 if (cl->cmode != HTB_CAN_SEND)
732 htb_add_to_wait_tree(q, cl, diff);
735 printk(KERN_WARNING "htb: too many events !\n");
736 return q->now + PSCHED_TICKS_PER_SEC / 10;
739 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL
740 is no such one exists. */
741 static struct rb_node *htb_id_find_next_upper(int prio, struct rb_node *n,
744 struct rb_node *r = NULL;
746 struct htb_class *cl =
747 rb_entry(n, struct htb_class, node[prio]);
748 if (id == cl->classid)
751 if (id > cl->classid) {
762 * htb_lookup_leaf - returns next leaf class in DRR order
764 * Find leaf where current feed pointers points to.
766 static struct htb_class *htb_lookup_leaf(struct rb_root *tree, int prio,
767 struct rb_node **pptr, u32 * pid)
771 struct rb_node *root;
772 struct rb_node **pptr;
774 } stk[TC_HTB_MAXDEPTH], *sp = stk;
776 BUG_TRAP(tree->rb_node);
777 sp->root = tree->rb_node;
781 for (i = 0; i < 65535; i++) {
782 if (!*sp->pptr && *sp->pid) {
783 /* ptr was invalidated but id is valid - try to recover
784 the original or next ptr */
786 htb_id_find_next_upper(prio, sp->root, *sp->pid);
788 *sp->pid = 0; /* ptr is valid now so that remove this hint as it
789 can become out of date quickly */
790 if (!*sp->pptr) { /* we are at right end; rewind & go up */
791 *sp->pptr = sp->root;
792 while ((*sp->pptr)->rb_left)
793 *sp->pptr = (*sp->pptr)->rb_left;
799 htb_next_rb_node(sp->pptr);
802 struct htb_class *cl;
803 cl = rb_entry(*sp->pptr, struct htb_class, node[prio]);
806 (++sp)->root = cl->un.inner.feed[prio].rb_node;
807 sp->pptr = cl->un.inner.ptr + prio;
808 sp->pid = cl->un.inner.last_ptr_id + prio;
815 /* dequeues packet at given priority and level; call only if
816 you are sure that there is active class at prio/level */
817 static struct sk_buff *htb_dequeue_tree(struct htb_sched *q, int prio,
820 struct sk_buff *skb = NULL;
821 struct htb_class *cl, *start;
822 /* look initial class up in the row */
823 start = cl = htb_lookup_leaf(q->row[level] + prio, prio,
824 q->ptr[level] + prio,
825 q->last_ptr_id[level] + prio);
833 /* class can be empty - it is unlikely but can be true if leaf
834 qdisc drops packets in enqueue routine or if someone used
835 graft operation on the leaf since last dequeue;
836 simply deactivate and skip such class */
837 if (unlikely(cl->un.leaf.q->q.qlen == 0)) {
838 struct htb_class *next;
839 htb_deactivate(q, cl);
841 /* row/level might become empty */
842 if ((q->row_mask[level] & (1 << prio)) == 0)
845 next = htb_lookup_leaf(q->row[level] + prio,
846 prio, q->ptr[level] + prio,
847 q->last_ptr_id[level] + prio);
849 if (cl == start) /* fix start if we just deleted it */
855 skb = cl->un.leaf.q->dequeue(cl->un.leaf.q);
856 if (likely(skb != NULL))
860 "htb: class %X isn't work conserving ?!\n",
865 htb_next_rb_node((level ? cl->parent->un.inner.ptr : q->
867 cl = htb_lookup_leaf(q->row[level] + prio, prio,
868 q->ptr[level] + prio,
869 q->last_ptr_id[level] + prio);
871 } while (cl != start);
873 if (likely(skb != NULL)) {
874 if ((cl->un.leaf.deficit[level] -= skb->len) < 0) {
875 cl->un.leaf.deficit[level] += cl->un.leaf.quantum;
876 htb_next_rb_node((level ? cl->parent->un.inner.ptr : q->
879 /* this used to be after charge_class but this constelation
880 gives us slightly better performance */
881 if (!cl->un.leaf.q->q.qlen)
882 htb_deactivate(q, cl);
883 htb_charge_class(q, cl, level, skb);
888 static struct sk_buff *htb_dequeue(struct Qdisc *sch)
890 struct sk_buff *skb = NULL;
891 struct htb_sched *q = qdisc_priv(sch);
893 psched_time_t next_event;
895 /* try to dequeue direct packets as high prio (!) to minimize cpu work */
896 skb = __skb_dequeue(&q->direct_queue);
898 sch->flags &= ~TCQ_F_THROTTLED;
905 q->now = psched_get_time();
907 next_event = q->now + 5 * PSCHED_TICKS_PER_SEC;
909 for (level = 0; level < TC_HTB_MAXDEPTH; level++) {
910 /* common case optimization - skip event handler quickly */
914 if (q->now >= q->near_ev_cache[level]) {
915 event = htb_do_events(q, level);
917 event = q->now + PSCHED_TICKS_PER_SEC;
918 q->near_ev_cache[level] = event;
920 event = q->near_ev_cache[level];
922 if (event && next_event > event)
925 m = ~q->row_mask[level];
926 while (m != (int)(-1)) {
929 skb = htb_dequeue_tree(q, prio, level);
930 if (likely(skb != NULL)) {
932 sch->flags &= ~TCQ_F_THROTTLED;
937 sch->qstats.overlimits++;
938 qdisc_watchdog_schedule(&q->watchdog, next_event);
943 /* try to drop from each class (by prio) until one succeed */
944 static unsigned int htb_drop(struct Qdisc *sch)
946 struct htb_sched *q = qdisc_priv(sch);
949 for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) {
951 list_for_each(p, q->drops + prio) {
952 struct htb_class *cl = list_entry(p, struct htb_class,
955 if (cl->un.leaf.q->ops->drop &&
956 (len = cl->un.leaf.q->ops->drop(cl->un.leaf.q))) {
958 if (!cl->un.leaf.q->q.qlen)
959 htb_deactivate(q, cl);
967 /* reset all classes */
968 /* always caled under BH & queue lock */
969 static void htb_reset(struct Qdisc *sch)
971 struct htb_sched *q = qdisc_priv(sch);
974 for (i = 0; i < HTB_HSIZE; i++) {
975 struct hlist_node *p;
976 struct htb_class *cl;
978 hlist_for_each_entry(cl, p, q->hash + i, hlist) {
980 memset(&cl->un.inner, 0, sizeof(cl->un.inner));
983 qdisc_reset(cl->un.leaf.q);
984 INIT_LIST_HEAD(&cl->un.leaf.drop_list);
986 cl->prio_activity = 0;
987 cl->cmode = HTB_CAN_SEND;
991 qdisc_watchdog_cancel(&q->watchdog);
992 __skb_queue_purge(&q->direct_queue);
994 memset(q->row, 0, sizeof(q->row));
995 memset(q->row_mask, 0, sizeof(q->row_mask));
996 memset(q->wait_pq, 0, sizeof(q->wait_pq));
997 memset(q->ptr, 0, sizeof(q->ptr));
998 for (i = 0; i < TC_HTB_NUMPRIO; i++)
999 INIT_LIST_HEAD(q->drops + i);
1002 static const struct nla_policy htb_policy[TCA_HTB_MAX + 1] = {
1003 [TCA_HTB_PARMS] = { .len = sizeof(struct tc_htb_opt) },
1004 [TCA_HTB_INIT] = { .len = sizeof(struct tc_htb_glob) },
1005 [TCA_HTB_CTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
1006 [TCA_HTB_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
1009 static int htb_init(struct Qdisc *sch, struct nlattr *opt)
1011 struct htb_sched *q = qdisc_priv(sch);
1012 struct nlattr *tb[TCA_HTB_INIT + 1];
1013 struct tc_htb_glob *gopt;
1020 err = nla_parse_nested(tb, TCA_HTB_INIT, opt, htb_policy);
1024 if (tb[TCA_HTB_INIT] == NULL) {
1025 printk(KERN_ERR "HTB: hey probably you have bad tc tool ?\n");
1028 gopt = nla_data(tb[TCA_HTB_INIT]);
1029 if (gopt->version != HTB_VER >> 16) {
1031 "HTB: need tc/htb version %d (minor is %d), you have %d\n",
1032 HTB_VER >> 16, HTB_VER & 0xffff, gopt->version);
1036 INIT_LIST_HEAD(&q->root);
1037 for (i = 0; i < HTB_HSIZE; i++)
1038 INIT_HLIST_HEAD(q->hash + i);
1039 for (i = 0; i < TC_HTB_NUMPRIO; i++)
1040 INIT_LIST_HEAD(q->drops + i);
1042 qdisc_watchdog_init(&q->watchdog, sch);
1043 skb_queue_head_init(&q->direct_queue);
1045 q->direct_qlen = sch->dev->tx_queue_len;
1046 if (q->direct_qlen < 2) /* some devices have zero tx_queue_len */
1049 if ((q->rate2quantum = gopt->rate2quantum) < 1)
1050 q->rate2quantum = 1;
1051 q->defcls = gopt->defcls;
1056 static int htb_dump(struct Qdisc *sch, struct sk_buff *skb)
1058 struct htb_sched *q = qdisc_priv(sch);
1059 struct nlattr *nest;
1060 struct tc_htb_glob gopt;
1062 spin_lock_bh(&sch->dev->queue_lock);
1064 gopt.direct_pkts = q->direct_pkts;
1065 gopt.version = HTB_VER;
1066 gopt.rate2quantum = q->rate2quantum;
1067 gopt.defcls = q->defcls;
1070 nest = nla_nest_start(skb, TCA_OPTIONS);
1072 goto nla_put_failure;
1073 NLA_PUT(skb, TCA_HTB_INIT, sizeof(gopt), &gopt);
1074 nla_nest_end(skb, nest);
1076 spin_unlock_bh(&sch->dev->queue_lock);
1080 spin_unlock_bh(&sch->dev->queue_lock);
1081 nla_nest_cancel(skb, nest);
1085 static int htb_dump_class(struct Qdisc *sch, unsigned long arg,
1086 struct sk_buff *skb, struct tcmsg *tcm)
1088 struct htb_class *cl = (struct htb_class *)arg;
1089 struct nlattr *nest;
1090 struct tc_htb_opt opt;
1092 spin_lock_bh(&sch->dev->queue_lock);
1093 tcm->tcm_parent = cl->parent ? cl->parent->classid : TC_H_ROOT;
1094 tcm->tcm_handle = cl->classid;
1095 if (!cl->level && cl->un.leaf.q)
1096 tcm->tcm_info = cl->un.leaf.q->handle;
1098 nest = nla_nest_start(skb, TCA_OPTIONS);
1100 goto nla_put_failure;
1102 memset(&opt, 0, sizeof(opt));
1104 opt.rate = cl->rate->rate;
1105 opt.buffer = cl->buffer;
1106 opt.ceil = cl->ceil->rate;
1107 opt.cbuffer = cl->cbuffer;
1108 opt.quantum = cl->un.leaf.quantum;
1109 opt.prio = cl->un.leaf.prio;
1110 opt.level = cl->level;
1111 NLA_PUT(skb, TCA_HTB_PARMS, sizeof(opt), &opt);
1113 nla_nest_end(skb, nest);
1114 spin_unlock_bh(&sch->dev->queue_lock);
1118 spin_unlock_bh(&sch->dev->queue_lock);
1119 nla_nest_cancel(skb, nest);
1124 htb_dump_class_stats(struct Qdisc *sch, unsigned long arg, struct gnet_dump *d)
1126 struct htb_class *cl = (struct htb_class *)arg;
1128 if (!cl->level && cl->un.leaf.q)
1129 cl->qstats.qlen = cl->un.leaf.q->q.qlen;
1130 cl->xstats.tokens = cl->tokens;
1131 cl->xstats.ctokens = cl->ctokens;
1133 if (gnet_stats_copy_basic(d, &cl->bstats) < 0 ||
1134 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
1135 gnet_stats_copy_queue(d, &cl->qstats) < 0)
1138 return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats));
1141 static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1144 struct htb_class *cl = (struct htb_class *)arg;
1146 if (cl && !cl->level) {
1148 (new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
1153 if ((*old = xchg(&cl->un.leaf.q, new)) != NULL) {
1154 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
1157 sch_tree_unlock(sch);
1163 static struct Qdisc *htb_leaf(struct Qdisc *sch, unsigned long arg)
1165 struct htb_class *cl = (struct htb_class *)arg;
1166 return (cl && !cl->level) ? cl->un.leaf.q : NULL;
1169 static void htb_qlen_notify(struct Qdisc *sch, unsigned long arg)
1171 struct htb_class *cl = (struct htb_class *)arg;
1173 if (cl->un.leaf.q->q.qlen == 0)
1174 htb_deactivate(qdisc_priv(sch), cl);
1177 static unsigned long htb_get(struct Qdisc *sch, u32 classid)
1179 struct htb_class *cl = htb_find(classid, sch);
1182 return (unsigned long)cl;
1185 static inline int htb_parent_last_child(struct htb_class *cl)
1188 /* the root class */
1191 if (!(cl->parent->children.next == &cl->sibling &&
1192 cl->parent->children.prev == &cl->sibling))
1193 /* not the last child */
1199 static void htb_parent_to_leaf(struct htb_class *cl, struct Qdisc *new_q)
1201 struct htb_class *parent = cl->parent;
1203 BUG_TRAP(!cl->level && cl->un.leaf.q && !cl->prio_activity);
1206 memset(&parent->un.inner, 0, sizeof(parent->un.inner));
1207 INIT_LIST_HEAD(&parent->un.leaf.drop_list);
1208 parent->un.leaf.q = new_q ? new_q : &noop_qdisc;
1209 parent->un.leaf.quantum = parent->quantum;
1210 parent->un.leaf.prio = parent->prio;
1211 parent->tokens = parent->buffer;
1212 parent->ctokens = parent->cbuffer;
1213 parent->t_c = psched_get_time();
1214 parent->cmode = HTB_CAN_SEND;
1217 static void htb_destroy_class(struct Qdisc *sch, struct htb_class *cl)
1219 struct htb_sched *q = qdisc_priv(sch);
1222 BUG_TRAP(cl->un.leaf.q);
1223 qdisc_destroy(cl->un.leaf.q);
1225 gen_kill_estimator(&cl->bstats, &cl->rate_est);
1226 qdisc_put_rtab(cl->rate);
1227 qdisc_put_rtab(cl->ceil);
1229 tcf_destroy_chain(cl->filter_list);
1231 while (!list_empty(&cl->children))
1232 htb_destroy_class(sch, list_entry(cl->children.next,
1233 struct htb_class, sibling));
1235 /* note: this delete may happen twice (see htb_delete) */
1236 hlist_del_init(&cl->hlist);
1237 list_del(&cl->sibling);
1239 if (cl->prio_activity)
1240 htb_deactivate(q, cl);
1242 if (cl->cmode != HTB_CAN_SEND)
1243 htb_safe_rb_erase(&cl->pq_node, q->wait_pq + cl->level);
1248 /* always caled under BH & queue lock */
1249 static void htb_destroy(struct Qdisc *sch)
1251 struct htb_sched *q = qdisc_priv(sch);
1253 qdisc_watchdog_cancel(&q->watchdog);
1254 /* This line used to be after htb_destroy_class call below
1255 and surprisingly it worked in 2.4. But it must precede it
1256 because filter need its target class alive to be able to call
1257 unbind_filter on it (without Oops). */
1258 tcf_destroy_chain(q->filter_list);
1260 while (!list_empty(&q->root))
1261 htb_destroy_class(sch, list_entry(q->root.next,
1262 struct htb_class, sibling));
1264 __skb_queue_purge(&q->direct_queue);
1267 static int htb_delete(struct Qdisc *sch, unsigned long arg)
1269 struct htb_sched *q = qdisc_priv(sch);
1270 struct htb_class *cl = (struct htb_class *)arg;
1272 struct Qdisc *new_q = NULL;
1275 // TODO: why don't allow to delete subtree ? references ? does
1276 // tc subsys quarantee us that in htb_destroy it holds no class
1277 // refs so that we can remove children safely there ?
1278 if (!list_empty(&cl->children) || cl->filter_cnt)
1281 if (!cl->level && htb_parent_last_child(cl)) {
1282 new_q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
1283 cl->parent->classid);
1290 qlen = cl->un.leaf.q->q.qlen;
1291 qdisc_reset(cl->un.leaf.q);
1292 qdisc_tree_decrease_qlen(cl->un.leaf.q, qlen);
1295 /* delete from hash and active; remainder in destroy_class */
1296 hlist_del_init(&cl->hlist);
1298 if (cl->prio_activity)
1299 htb_deactivate(q, cl);
1302 htb_parent_to_leaf(cl, new_q);
1304 if (--cl->refcnt == 0)
1305 htb_destroy_class(sch, cl);
1307 sch_tree_unlock(sch);
1311 static void htb_put(struct Qdisc *sch, unsigned long arg)
1313 struct htb_class *cl = (struct htb_class *)arg;
1315 if (--cl->refcnt == 0)
1316 htb_destroy_class(sch, cl);
1319 static int htb_change_class(struct Qdisc *sch, u32 classid,
1320 u32 parentid, struct nlattr **tca,
1324 struct htb_sched *q = qdisc_priv(sch);
1325 struct htb_class *cl = (struct htb_class *)*arg, *parent;
1326 struct nlattr *opt = tca[TCA_OPTIONS];
1327 struct qdisc_rate_table *rtab = NULL, *ctab = NULL;
1328 struct nlattr *tb[TCA_HTB_RTAB + 1];
1329 struct tc_htb_opt *hopt;
1331 /* extract all subattrs from opt attr */
1335 err = nla_parse_nested(tb, TCA_HTB_RTAB, opt, htb_policy);
1340 if (tb[TCA_HTB_PARMS] == NULL)
1343 parent = parentid == TC_H_ROOT ? NULL : htb_find(parentid, sch);
1345 hopt = nla_data(tb[TCA_HTB_PARMS]);
1347 rtab = qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB]);
1348 ctab = qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB]);
1352 if (!cl) { /* new class */
1353 struct Qdisc *new_q;
1357 struct gnet_estimator opt;
1360 .nla_len = nla_attr_size(sizeof(est.opt)),
1361 .nla_type = TCA_RATE,
1364 /* 4s interval, 16s averaging constant */
1370 /* check for valid classid */
1371 if (!classid || TC_H_MAJ(classid ^ sch->handle)
1372 || htb_find(classid, sch))
1375 /* check maximal depth */
1376 if (parent && parent->parent && parent->parent->level < 2) {
1377 printk(KERN_ERR "htb: tree is too deep\n");
1381 if ((cl = kzalloc(sizeof(*cl), GFP_KERNEL)) == NULL)
1384 gen_new_estimator(&cl->bstats, &cl->rate_est,
1385 &sch->dev->queue_lock,
1386 tca[TCA_RATE] ? : &est.nla);
1388 INIT_LIST_HEAD(&cl->sibling);
1389 INIT_HLIST_NODE(&cl->hlist);
1390 INIT_LIST_HEAD(&cl->children);
1391 INIT_LIST_HEAD(&cl->un.leaf.drop_list);
1392 RB_CLEAR_NODE(&cl->pq_node);
1394 for (prio = 0; prio < TC_HTB_NUMPRIO; prio++)
1395 RB_CLEAR_NODE(&cl->node[prio]);
1397 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL)
1398 so that can't be used inside of sch_tree_lock
1399 -- thanks to Karlis Peisenieks */
1400 new_q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, classid);
1402 if (parent && !parent->level) {
1403 unsigned int qlen = parent->un.leaf.q->q.qlen;
1405 /* turn parent into inner node */
1406 qdisc_reset(parent->un.leaf.q);
1407 qdisc_tree_decrease_qlen(parent->un.leaf.q, qlen);
1408 qdisc_destroy(parent->un.leaf.q);
1409 if (parent->prio_activity)
1410 htb_deactivate(q, parent);
1412 /* remove from evt list because of level change */
1413 if (parent->cmode != HTB_CAN_SEND) {
1414 htb_safe_rb_erase(&parent->pq_node, q->wait_pq);
1415 parent->cmode = HTB_CAN_SEND;
1417 parent->level = (parent->parent ? parent->parent->level
1418 : TC_HTB_MAXDEPTH) - 1;
1419 memset(&parent->un.inner, 0, sizeof(parent->un.inner));
1421 /* leaf (we) needs elementary qdisc */
1422 cl->un.leaf.q = new_q ? new_q : &noop_qdisc;
1424 cl->classid = classid;
1425 cl->parent = parent;
1427 /* set class to be in HTB_CAN_SEND state */
1428 cl->tokens = hopt->buffer;
1429 cl->ctokens = hopt->cbuffer;
1430 cl->mbuffer = 60 * PSCHED_TICKS_PER_SEC; /* 1min */
1431 cl->t_c = psched_get_time();
1432 cl->cmode = HTB_CAN_SEND;
1434 /* attach to the hash list and parent's family */
1435 hlist_add_head(&cl->hlist, q->hash + htb_hash(classid));
1436 list_add_tail(&cl->sibling,
1437 parent ? &parent->children : &q->root);
1440 gen_replace_estimator(&cl->bstats, &cl->rate_est,
1441 &sch->dev->queue_lock,
1446 /* it used to be a nasty bug here, we have to check that node
1447 is really leaf before changing cl->un.leaf ! */
1449 cl->un.leaf.quantum = rtab->rate.rate / q->rate2quantum;
1450 if (!hopt->quantum && cl->un.leaf.quantum < 1000) {
1452 "HTB: quantum of class %X is small. Consider r2q change.\n",
1454 cl->un.leaf.quantum = 1000;
1456 if (!hopt->quantum && cl->un.leaf.quantum > 200000) {
1458 "HTB: quantum of class %X is big. Consider r2q change.\n",
1460 cl->un.leaf.quantum = 200000;
1463 cl->un.leaf.quantum = hopt->quantum;
1464 if ((cl->un.leaf.prio = hopt->prio) >= TC_HTB_NUMPRIO)
1465 cl->un.leaf.prio = TC_HTB_NUMPRIO - 1;
1467 /* backup for htb_parent_to_leaf */
1468 cl->quantum = cl->un.leaf.quantum;
1469 cl->prio = cl->un.leaf.prio;
1472 cl->buffer = hopt->buffer;
1473 cl->cbuffer = hopt->cbuffer;
1475 qdisc_put_rtab(cl->rate);
1478 qdisc_put_rtab(cl->ceil);
1480 sch_tree_unlock(sch);
1482 *arg = (unsigned long)cl;
1487 qdisc_put_rtab(rtab);
1489 qdisc_put_rtab(ctab);
1493 static struct tcf_proto **htb_find_tcf(struct Qdisc *sch, unsigned long arg)
1495 struct htb_sched *q = qdisc_priv(sch);
1496 struct htb_class *cl = (struct htb_class *)arg;
1497 struct tcf_proto **fl = cl ? &cl->filter_list : &q->filter_list;
1502 static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent,
1505 struct htb_sched *q = qdisc_priv(sch);
1506 struct htb_class *cl = htb_find(classid, sch);
1508 /*if (cl && !cl->level) return 0;
1509 The line above used to be there to prevent attaching filters to
1510 leaves. But at least tc_index filter uses this just to get class
1511 for other reasons so that we have to allow for it.
1513 19.6.2002 As Werner explained it is ok - bind filter is just
1514 another way to "lock" the class - unlike "get" this lock can
1515 be broken by class during destroy IIUC.
1521 return (unsigned long)cl;
1524 static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg)
1526 struct htb_sched *q = qdisc_priv(sch);
1527 struct htb_class *cl = (struct htb_class *)arg;
1535 static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1537 struct htb_sched *q = qdisc_priv(sch);
1543 for (i = 0; i < HTB_HSIZE; i++) {
1544 struct hlist_node *p;
1545 struct htb_class *cl;
1547 hlist_for_each_entry(cl, p, q->hash + i, hlist) {
1548 if (arg->count < arg->skip) {
1552 if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
1561 static const struct Qdisc_class_ops htb_class_ops = {
1564 .qlen_notify = htb_qlen_notify,
1567 .change = htb_change_class,
1568 .delete = htb_delete,
1570 .tcf_chain = htb_find_tcf,
1571 .bind_tcf = htb_bind_filter,
1572 .unbind_tcf = htb_unbind_filter,
1573 .dump = htb_dump_class,
1574 .dump_stats = htb_dump_class_stats,
1577 static struct Qdisc_ops htb_qdisc_ops __read_mostly = {
1579 .cl_ops = &htb_class_ops,
1581 .priv_size = sizeof(struct htb_sched),
1582 .enqueue = htb_enqueue,
1583 .dequeue = htb_dequeue,
1584 .requeue = htb_requeue,
1588 .destroy = htb_destroy,
1589 .change = NULL /* htb_change */,
1591 .owner = THIS_MODULE,
1594 static int __init htb_module_init(void)
1596 return register_qdisc(&htb_qdisc_ops);
1598 static void __exit htb_module_exit(void)
1600 unregister_qdisc(&htb_qdisc_ops);
1603 module_init(htb_module_init)
1604 module_exit(htb_module_exit)
1605 MODULE_LICENSE("GPL");