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/config.h>
31 #include <linux/module.h>
32 #include <asm/uaccess.h>
33 #include <asm/system.h>
34 #include <asm/bitops.h>
35 #include <linux/types.h>
36 #include <linux/kernel.h>
37 #include <linux/sched.h>
38 #include <linux/string.h>
40 #include <linux/socket.h>
41 #include <linux/sockios.h>
43 #include <linux/errno.h>
44 #include <linux/interrupt.h>
45 #include <linux/if_ether.h>
46 #include <linux/inet.h>
47 #include <linux/netdevice.h>
48 #include <linux/etherdevice.h>
49 #include <linux/notifier.h>
51 #include <net/route.h>
52 #include <linux/skbuff.h>
53 #include <linux/list.h>
54 #include <linux/compiler.h>
56 #include <net/pkt_sched.h>
57 #include <linux/rbtree.h>
61 ========================================================================
62 HTB is like TBF with multiple classes. It is also similar to CBQ because
63 it allows to assign priority to each class in hierarchy.
64 In fact it is another implementation of Floyd's formal sharing.
67 Each class is assigned level. Leaf has ALWAYS level 0 and root
68 classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level
69 one less than their parent.
72 #define HTB_HSIZE 16 /* classid hash size */
73 #define HTB_EWMAC 2 /* rate average over HTB_EWMAC*HTB_HSIZE sec */
74 #define HTB_DEBUG 1 /* compile debugging support (activated by tc tool) */
75 #define HTB_RATECM 1 /* whether to use rate computer */
76 #define HTB_HYSTERESIS 1/* whether to use mode hysteresis for speedup */
77 #define HTB_QLOCK(S) spin_lock_bh(&(S)->dev->queue_lock)
78 #define HTB_QUNLOCK(S) spin_unlock_bh(&(S)->dev->queue_lock)
79 #define HTB_VER 0x30010 /* major must be matched with number suplied by TC as version */
81 #if HTB_VER >> 16 != TC_HTB_PROTOVER
82 #error "Mismatched sch_htb.c and pkt_sch.h"
85 /* debugging support; S is subsystem, these are defined:
90 4 - dequeue one prio DRR part
91 5 - dequeue class accounting
92 6 - class overlimit status computation
97 12 - fast dequeue cache
99 L is level; 0 = none, 1 = basic info, 2 = detailed, 3 = full
100 q->debug uint32 contains 16 2-bit fields one for subsystem starting
104 #define HTB_DBG_COND(S,L) (((q->debug>>(2*S))&3) >= L)
105 #define HTB_DBG(S,L,FMT,ARG...) if (HTB_DBG_COND(S,L)) \
106 printk(KERN_DEBUG FMT,##ARG)
107 #define HTB_CHCL(cl) BUG_TRAP((cl)->magic == HTB_CMAGIC)
109 #define HTB_ARGQ struct htb_sched *q,
115 #define HTB_CMAGIC 0xFEFAFEF1
116 #define htb_safe_rb_erase(N,R) do { BUG_TRAP((N)->rb_color != -1); \
117 if ((N)->rb_color == -1) break; \
119 (N)->rb_color = -1; } while (0)
121 #define HTB_DBG_COND(S,L) (0)
122 #define HTB_DBG(S,L,FMT,ARG...)
126 #define htb_safe_rb_erase(N,R) rb_erase(N,R)
130 /* used internaly to keep status of single class */
132 HTB_CANT_SEND, /* class can't send and can't borrow */
133 HTB_MAY_BORROW, /* class can't send but may borrow */
134 HTB_CAN_SEND /* class can send */
137 /* interior & leaf nodes; props specific to leaves are marked L: */
143 /* general class parameters */
145 struct tc_stats stats; /* generic stats */
146 spinlock_t *stats_lock;
147 struct tc_htb_xstats xstats;/* our special stats */
148 int refcnt; /* usage count of this class */
151 /* rate measurement counters */
152 unsigned long rate_bytes,sum_bytes;
153 unsigned long rate_packets,sum_packets;
157 int level; /* our level (see above) */
158 struct htb_class *parent; /* parent class */
159 struct list_head hlist; /* classid hash list item */
160 struct list_head sibling; /* sibling list item */
161 struct list_head children; /* children list */
164 struct htb_class_leaf {
169 int deficit[TC_HTB_MAXDEPTH];
170 struct list_head drop_list;
172 struct htb_class_inner {
173 struct rb_root feed[TC_HTB_NUMPRIO]; /* feed trees */
174 struct rb_node *ptr[TC_HTB_NUMPRIO]; /* current class ptr */
177 struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */
178 struct rb_node pq_node; /* node for event queue */
179 unsigned long pq_key; /* the same type as jiffies global */
181 int prio_activity; /* for which prios are we active */
182 enum htb_cmode cmode; /* current mode of the class */
184 /* class attached filters */
185 struct tcf_proto *filter_list;
188 int warned; /* only one warning about non work conserving .. */
190 /* token bucket parameters */
191 struct qdisc_rate_table *rate; /* rate table of the class itself */
192 struct qdisc_rate_table *ceil; /* ceiling rate (limits borrows too) */
193 long buffer,cbuffer; /* token bucket depth/rate */
194 long mbuffer; /* max wait time */
195 long tokens,ctokens; /* current number of tokens */
196 psched_time_t t_c; /* checkpoint time */
199 /* TODO: maybe compute rate when size is too large .. or drop ? */
200 static __inline__ long L2T(struct htb_class *cl,struct qdisc_rate_table *rate,
203 int slot = size >> rate->rate.cell_log;
208 return rate->data[slot];
213 struct list_head root; /* root classes list */
214 struct list_head hash[HTB_HSIZE]; /* hashed by classid */
215 struct list_head drops[TC_HTB_NUMPRIO]; /* active leaves (for drops) */
217 /* self list - roots of self generating tree */
218 struct rb_root row[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO];
219 int row_mask[TC_HTB_MAXDEPTH];
220 struct rb_node *ptr[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO];
222 /* self wait list - roots of wait PQs per row */
223 struct rb_root wait_pq[TC_HTB_MAXDEPTH];
225 /* time of nearest event per level (row) */
226 unsigned long near_ev_cache[TC_HTB_MAXDEPTH];
228 /* cached value of jiffies in dequeue */
229 unsigned long jiffies;
231 /* whether we hit non-work conserving class during this dequeue; we use */
232 int nwc_hit; /* this to disable mindelay complaint in dequeue */
234 int defcls; /* class where unclassified flows go to */
235 u32 debug; /* subsystem debug levels */
237 /* filters for qdisc itself */
238 struct tcf_proto *filter_list;
241 int rate2quantum; /* quant = rate / rate2quantum */
242 psched_time_t now; /* cached dequeue time */
243 struct timer_list timer; /* send delay timer */
245 struct timer_list rttim; /* rate computer timer */
246 int recmp_bucket; /* which hash bucket to recompute next */
249 /* non shaped skbs; let them go directly thru */
250 struct sk_buff_head direct_queue;
251 int direct_qlen; /* max qlen of above */
256 /* compute hash of size HTB_HSIZE for given handle */
257 static __inline__ int htb_hash(u32 h)
260 #error "Declare new hash for your HTB_HSIZE"
262 h ^= h>>8; /* stolen from cbq_hash */
267 /* find class in global hash table using given handle */
268 static __inline__ struct htb_class *htb_find(u32 handle, struct Qdisc *sch)
270 struct htb_sched *q = (struct htb_sched *)sch->data;
272 if (TC_H_MAJ(handle) != sch->handle)
275 list_for_each (p,q->hash+htb_hash(handle)) {
276 struct htb_class *cl = list_entry(p,struct htb_class,hlist);
277 if (cl->classid == handle)
284 * htb_classify - classify a packet into class
286 * It returns NULL if the packet should be dropped or -1 if the packet
287 * should be passed directly thru. In all other cases leaf class is returned.
288 * We allow direct class selection by classid in priority. The we examine
289 * filters in qdisc and in inner nodes (if higher filter points to the inner
290 * node). If we end up with classid MAJOR:0 we enqueue the skb into special
291 * internal fifo (direct). These packets then go directly thru. If we still
292 * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessfull
293 * then finish and return direct queue.
295 #define HTB_DIRECT (struct htb_class*)-1
296 static inline u32 htb_classid(struct htb_class *cl)
298 return (cl && cl != HTB_DIRECT) ? cl->classid : TC_H_UNSPEC;
301 static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch, int *qres)
303 struct htb_sched *q = (struct htb_sched *)sch->data;
304 struct htb_class *cl;
305 struct tcf_result res;
306 struct tcf_proto *tcf;
309 /* allow to select class by setting skb->priority to valid classid;
310 note that nfmark can be used too by attaching filter fw with no
312 if (skb->priority == sch->handle)
313 return HTB_DIRECT; /* X:0 (direct flow) selected */
314 if ((cl = htb_find(skb->priority,sch)) != NULL && cl->level == 0)
317 tcf = q->filter_list;
318 while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) {
319 #ifdef CONFIG_NET_CLS_ACT
322 case TC_ACT_SHOT: /* Stop and kfree */
323 *qres = NET_XMIT_DROP;
330 case TC_ACT_RECLASSIFY: /* Things look good */
342 #ifdef CONFIG_NET_CLS_POLICE
343 if (result == TC_POLICE_SHOT)
347 if ((cl = (void*)res.class) == NULL) {
348 if (res.classid == sch->handle)
349 return HTB_DIRECT; /* X:0 (direct flow) */
350 if ((cl = htb_find(res.classid,sch)) == NULL)
351 break; /* filter selected invalid classid */
354 return cl; /* we hit leaf; return it */
356 /* we have got inner class; apply inner filter chain */
357 tcf = cl->filter_list;
359 /* classification failed; try to use default class */
360 cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle),q->defcls),sch);
361 if (!cl || cl->level)
362 return HTB_DIRECT; /* bad default .. this is safe bet */
367 static void htb_next_rb_node(struct rb_node **n);
368 #define HTB_DUMTREE(root,memb) if(root) { \
369 struct rb_node *n = (root)->rb_node; \
370 while (n->rb_left) n = n->rb_left; \
372 struct htb_class *cl = rb_entry(n, struct htb_class, memb); \
373 printk(" %x",cl->classid); htb_next_rb_node (&n); \
376 static void htb_debug_dump (struct htb_sched *q)
379 printk(KERN_DEBUG "htb*g j=%lu lj=%lu\n",jiffies,q->jiffies);
381 for (i=TC_HTB_MAXDEPTH-1;i>=0;i--) {
382 printk(KERN_DEBUG "htb*r%d m=%x",i,q->row_mask[i]);
383 for (p=0;p<TC_HTB_NUMPRIO;p++) {
384 if (!q->row[i][p].rb_node) continue;
386 HTB_DUMTREE(q->row[i]+p,node[p]);
391 for (i = 0; i < HTB_HSIZE; i++) {
393 list_for_each (l,q->hash+i) {
394 struct htb_class *cl = list_entry(l,struct htb_class,hlist);
395 long diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32)cl->mbuffer);
396 printk(KERN_DEBUG "htb*c%x m=%d t=%ld c=%ld pq=%lu df=%ld ql=%d "
398 cl->classid,cl->cmode,cl->tokens,cl->ctokens,
399 cl->pq_node.rb_color==-1?0:cl->pq_key,diff,
400 cl->level?0:cl->un.leaf.q->q.qlen,cl->prio_activity);
402 for (p=0;p<TC_HTB_NUMPRIO;p++) {
403 if (!cl->un.inner.feed[p].rb_node) continue;
404 printk(" p%d a=%x:",p,cl->un.inner.ptr[p]?rb_entry(cl->un.inner.ptr[p], struct htb_class,node[p])->classid:0);
405 HTB_DUMTREE(cl->un.inner.feed+p,node[p]);
413 * htb_add_to_id_tree - adds class to the round robin list
415 * Routine adds class to the list (actually tree) sorted by classid.
416 * Make sure that class is not already on such list for given prio.
418 static void htb_add_to_id_tree (HTB_ARGQ struct rb_root *root,
419 struct htb_class *cl,int prio)
421 struct rb_node **p = &root->rb_node, *parent = NULL;
422 HTB_DBG(7,3,"htb_add_id_tree cl=%X prio=%d\n",cl->classid,prio);
424 if (cl->node[prio].rb_color != -1) { BUG_TRAP(0); return; }
427 struct htb_class *x = rb_entry(*p,struct htb_class,node[prio]);
432 struct htb_class *c; parent = *p;
433 c = rb_entry(parent, struct htb_class, node[prio]);
435 if (cl->classid > c->classid)
436 p = &parent->rb_right;
438 p = &parent->rb_left;
440 rb_link_node(&cl->node[prio], parent, p);
441 rb_insert_color(&cl->node[prio], root);
445 * htb_add_to_wait_tree - adds class to the event queue with delay
447 * The class is added to priority event queue to indicate that class will
448 * change its mode in cl->pq_key microseconds. Make sure that class is not
449 * already in the queue.
451 static void htb_add_to_wait_tree (struct htb_sched *q,
452 struct htb_class *cl,long delay,int debug_hint)
454 struct rb_node **p = &q->wait_pq[cl->level].rb_node, *parent = NULL;
455 HTB_DBG(7,3,"htb_add_wt cl=%X key=%lu\n",cl->classid,cl->pq_key);
457 if (cl->pq_node.rb_color != -1) { BUG_TRAP(0); return; }
459 if ((delay <= 0 || delay > cl->mbuffer) && net_ratelimit())
460 printk(KERN_ERR "HTB: suspicious delay in wait_tree d=%ld cl=%X h=%d\n",delay,cl->classid,debug_hint);
462 cl->pq_key = q->jiffies + PSCHED_US2JIFFIE(delay);
463 if (cl->pq_key == q->jiffies)
466 /* update the nearest event cache */
467 if (time_after(q->near_ev_cache[cl->level], cl->pq_key))
468 q->near_ev_cache[cl->level] = cl->pq_key;
471 struct htb_class *c; parent = *p;
472 c = rb_entry(parent, struct htb_class, pq_node);
473 if (time_after_eq(cl->pq_key, c->pq_key))
474 p = &parent->rb_right;
476 p = &parent->rb_left;
478 rb_link_node(&cl->pq_node, parent, p);
479 rb_insert_color(&cl->pq_node, &q->wait_pq[cl->level]);
483 * htb_next_rb_node - finds next node in binary tree
485 * When we are past last key we return NULL.
486 * Average complexity is 2 steps per call.
488 static void htb_next_rb_node(struct rb_node **n)
494 * htb_add_class_to_row - add class to its row
496 * The class is added to row at priorities marked in mask.
497 * It does nothing if mask == 0.
499 static inline void htb_add_class_to_row(struct htb_sched *q,
500 struct htb_class *cl,int mask)
502 HTB_DBG(7,2,"htb_addrow cl=%X mask=%X rmask=%X\n",
503 cl->classid,mask,q->row_mask[cl->level]);
505 q->row_mask[cl->level] |= mask;
507 int prio = ffz(~mask);
508 mask &= ~(1 << prio);
509 htb_add_to_id_tree(HTB_PASSQ q->row[cl->level]+prio,cl,prio);
514 * htb_remove_class_from_row - removes class from its row
516 * The class is removed from row at priorities marked in mask.
517 * It does nothing if mask == 0.
519 static __inline__ void htb_remove_class_from_row(struct htb_sched *q,
520 struct htb_class *cl,int mask)
525 int prio = ffz(~mask);
526 mask &= ~(1 << prio);
527 if (q->ptr[cl->level][prio] == cl->node+prio)
528 htb_next_rb_node(q->ptr[cl->level]+prio);
529 htb_safe_rb_erase(cl->node + prio,q->row[cl->level]+prio);
530 if (!q->row[cl->level][prio].rb_node)
533 HTB_DBG(7,2,"htb_delrow cl=%X mask=%X rmask=%X maskdel=%X\n",
534 cl->classid,mask,q->row_mask[cl->level],m);
535 q->row_mask[cl->level] &= ~m;
539 * htb_activate_prios - creates active classe's feed chain
541 * The class is connected to ancestors and/or appropriate rows
542 * for priorities it is participating on. cl->cmode must be new
543 * (activated) mode. It does nothing if cl->prio_activity == 0.
545 static void htb_activate_prios(struct htb_sched *q,struct htb_class *cl)
547 struct htb_class *p = cl->parent;
548 long m,mask = cl->prio_activity;
549 HTB_DBG(7,2,"htb_act_prios cl=%X mask=%lX cmode=%d\n",cl->classid,mask,cl->cmode);
552 while (cl->cmode == HTB_MAY_BORROW && p && mask) {
554 m = mask; while (m) {
558 if (p->un.inner.feed[prio].rb_node)
559 /* parent already has its feed in use so that
560 reset bit in mask as parent is already ok */
561 mask &= ~(1 << prio);
563 htb_add_to_id_tree(HTB_PASSQ p->un.inner.feed+prio,cl,prio);
565 HTB_DBG(7,3,"htb_act_pr_aft p=%X pact=%X mask=%lX pmode=%d\n",
566 p->classid,p->prio_activity,mask,p->cmode);
567 p->prio_activity |= mask;
568 cl = p; p = cl->parent;
571 if (cl->cmode == HTB_CAN_SEND && mask)
572 htb_add_class_to_row(q,cl,mask);
576 * htb_deactivate_prios - remove class from feed chain
578 * cl->cmode must represent old mode (before deactivation). It does
579 * nothing if cl->prio_activity == 0. Class is removed from all feed
582 static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl)
584 struct htb_class *p = cl->parent;
585 long m,mask = cl->prio_activity;
586 HTB_DBG(7,2,"htb_deact_prios cl=%X mask=%lX cmode=%d\n",cl->classid,mask,cl->cmode);
589 while (cl->cmode == HTB_MAY_BORROW && p && mask) {
595 if (p->un.inner.ptr[prio] == cl->node+prio)
596 htb_next_rb_node(p->un.inner.ptr + prio);
598 htb_safe_rb_erase(cl->node + prio,p->un.inner.feed + prio);
600 if (!p->un.inner.feed[prio].rb_node)
603 HTB_DBG(7,3,"htb_deact_pr_aft p=%X pact=%X mask=%lX pmode=%d\n",
604 p->classid,p->prio_activity,mask,p->cmode);
605 p->prio_activity &= ~mask;
606 cl = p; p = cl->parent;
609 if (cl->cmode == HTB_CAN_SEND && mask)
610 htb_remove_class_from_row(q,cl,mask);
614 * htb_class_mode - computes and returns current class mode
616 * It computes cl's mode at time cl->t_c+diff and returns it. If mode
617 * is not HTB_CAN_SEND then cl->pq_key is updated to time difference
618 * from now to time when cl will change its state.
619 * Also it is worth to note that class mode doesn't change simply
620 * at cl->{c,}tokens == 0 but there can rather be hysteresis of
621 * 0 .. -cl->{c,}buffer range. It is meant to limit number of
622 * mode transitions per time unit. The speed gain is about 1/6.
624 static __inline__ enum htb_cmode
625 htb_class_mode(struct htb_class *cl,long *diff)
629 if ((toks = (cl->ctokens + *diff)) < (
631 cl->cmode != HTB_CANT_SEND ? -cl->cbuffer :
635 return HTB_CANT_SEND;
637 if ((toks = (cl->tokens + *diff)) >= (
639 cl->cmode == HTB_CAN_SEND ? -cl->buffer :
645 return HTB_MAY_BORROW;
649 * htb_change_class_mode - changes classe's mode
651 * This should be the only way how to change classe's mode under normal
652 * cirsumstances. Routine will update feed lists linkage, change mode
653 * and add class to the wait event queue if appropriate. New mode should
654 * be different from old one and cl->pq_key has to be valid if changing
655 * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree).
658 htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, long *diff)
660 enum htb_cmode new_mode = htb_class_mode(cl,diff);
663 HTB_DBG(7,1,"htb_chging_clmode %d->%d cl=%X\n",cl->cmode,new_mode,cl->classid);
665 if (new_mode == cl->cmode)
668 if (cl->prio_activity) { /* not necessary: speed optimization */
669 if (cl->cmode != HTB_CANT_SEND)
670 htb_deactivate_prios(q,cl);
671 cl->cmode = new_mode;
672 if (new_mode != HTB_CANT_SEND)
673 htb_activate_prios(q,cl);
675 cl->cmode = new_mode;
679 * htb_activate - inserts leaf cl into appropriate active feeds
681 * Routine learns (new) priority of leaf and activates feed chain
682 * for the prio. It can be called on already active leaf safely.
683 * It also adds leaf into droplist.
685 static __inline__ void htb_activate(struct htb_sched *q,struct htb_class *cl)
687 BUG_TRAP(!cl->level && cl->un.leaf.q && cl->un.leaf.q->q.qlen);
689 if (!cl->prio_activity) {
690 cl->prio_activity = 1 << (cl->un.leaf.aprio = cl->un.leaf.prio);
691 htb_activate_prios(q,cl);
692 list_add_tail(&cl->un.leaf.drop_list,q->drops+cl->un.leaf.aprio);
697 * htb_deactivate - remove leaf cl from active feeds
699 * Make sure that leaf is active. In the other words it can't be called
700 * with non-active leaf. It also removes class from the drop list.
702 static __inline__ void
703 htb_deactivate(struct htb_sched *q,struct htb_class *cl)
705 BUG_TRAP(cl->prio_activity);
707 htb_deactivate_prios(q,cl);
708 cl->prio_activity = 0;
709 list_del_init(&cl->un.leaf.drop_list);
712 static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch)
714 int ret = NET_XMIT_SUCCESS;
715 struct htb_sched *q = (struct htb_sched *)sch->data;
716 struct htb_class *cl = htb_classify(skb,sch,&ret);
719 #ifdef CONFIG_NET_CLS_ACT
720 if (cl == HTB_DIRECT ) {
721 if (q->direct_queue.qlen < q->direct_qlen ) {
722 __skb_queue_tail(&q->direct_queue, skb);
726 if (NET_XMIT_DROP == ret) {
732 if (cl == HTB_DIRECT || !cl) {
733 /* enqueue to helper queue */
734 if (q->direct_queue.qlen < q->direct_qlen && cl) {
735 __skb_queue_tail(&q->direct_queue, skb);
740 return NET_XMIT_DROP;
744 else if (cl->un.leaf.q->enqueue(skb, cl->un.leaf.q) != NET_XMIT_SUCCESS) {
747 return NET_XMIT_DROP;
749 cl->stats.packets++; cl->stats.bytes += skb->len;
754 sch->stats.packets++; sch->stats.bytes += skb->len;
755 HTB_DBG(1,1,"htb_enq_ok cl=%X skb=%p\n",(cl && cl != HTB_DIRECT)?cl->classid:0,skb);
756 return NET_XMIT_SUCCESS;
759 /* TODO: requeuing packet charges it to policers again !! */
760 static int htb_requeue(struct sk_buff *skb, struct Qdisc *sch)
762 struct htb_sched *q = (struct htb_sched *)sch->data;
763 int ret = NET_XMIT_SUCCESS;
764 struct htb_class *cl = htb_classify(skb,sch, &ret);
765 struct sk_buff *tskb;
767 if (cl == HTB_DIRECT || !cl) {
768 /* enqueue to helper queue */
769 if (q->direct_queue.qlen < q->direct_qlen && cl) {
770 __skb_queue_head(&q->direct_queue, skb);
772 __skb_queue_head(&q->direct_queue, skb);
773 tskb = __skb_dequeue_tail(&q->direct_queue);
778 } else if (cl->un.leaf.q->ops->requeue(skb, cl->un.leaf.q) != NET_XMIT_SUCCESS) {
781 return NET_XMIT_DROP;
786 HTB_DBG(1,1,"htb_req_ok cl=%X skb=%p\n",(cl && cl != HTB_DIRECT)?cl->classid:0,skb);
787 return NET_XMIT_SUCCESS;
790 static void htb_timer(unsigned long arg)
792 struct Qdisc *sch = (struct Qdisc*)arg;
793 sch->flags &= ~TCQ_F_THROTTLED;
795 netif_schedule(sch->dev);
799 #define RT_GEN(D,R) R+=D-(R/HTB_EWMAC);D=0
800 static void htb_rate_timer(unsigned long arg)
802 struct Qdisc *sch = (struct Qdisc*)arg;
803 struct htb_sched *q = (struct htb_sched *)sch->data;
806 /* lock queue so that we can muck with it */
808 HTB_DBG(10,1,"htb_rttmr j=%ld\n",jiffies);
810 q->rttim.expires = jiffies + HZ;
811 add_timer(&q->rttim);
813 /* scan and recompute one bucket at time */
814 if (++q->recmp_bucket >= HTB_HSIZE)
816 list_for_each (p,q->hash+q->recmp_bucket) {
817 struct htb_class *cl = list_entry(p,struct htb_class,hlist);
818 HTB_DBG(10,2,"htb_rttmr_cl cl=%X sbyte=%lu spkt=%lu\n",
819 cl->classid,cl->sum_bytes,cl->sum_packets);
820 RT_GEN (cl->sum_bytes,cl->rate_bytes);
821 RT_GEN (cl->sum_packets,cl->rate_packets);
828 * htb_charge_class - charges amount "bytes" to leaf and ancestors
830 * Routine assumes that packet "bytes" long was dequeued from leaf cl
831 * borrowing from "level". It accounts bytes to ceil leaky bucket for
832 * leaf and all ancestors and to rate bucket for ancestors at levels
833 * "level" and higher. It also handles possible change of mode resulting
834 * from the update. Note that mode can also increase here (MAY_BORROW to
835 * CAN_SEND) because we can use more precise clock that event queue here.
836 * In such case we remove class from event queue first.
838 static void htb_charge_class(struct htb_sched *q,struct htb_class *cl,
842 enum htb_cmode old_mode;
843 HTB_DBG(5,1,"htb_chrg_cl cl=%X lev=%d len=%d\n",cl->classid,level,bytes);
845 #define HTB_ACCNT(T,B,R) toks = diff + cl->T; \
846 if (toks > cl->B) toks = cl->B; \
847 toks -= L2T(cl, cl->R, bytes); \
848 if (toks <= -cl->mbuffer) toks = 1-cl->mbuffer; \
853 diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32)cl->mbuffer);
855 if (diff > cl->mbuffer || diff < 0 || PSCHED_TLESS(q->now, cl->t_c)) {
857 printk(KERN_ERR "HTB: bad diff in charge, cl=%X diff=%lX now=%Lu then=%Lu j=%lu\n",
859 (unsigned long long) q->now,
860 (unsigned long long) cl->t_c,
865 if (cl->level >= level) {
866 if (cl->level == level) cl->xstats.lends++;
867 HTB_ACCNT (tokens,buffer,rate);
869 cl->xstats.borrows++;
870 cl->tokens += diff; /* we moved t_c; update tokens */
872 HTB_ACCNT (ctokens,cbuffer,ceil);
874 HTB_DBG(5,2,"htb_chrg_clp cl=%X diff=%ld tok=%ld ctok=%ld\n",cl->classid,diff,cl->tokens,cl->ctokens);
876 old_mode = cl->cmode; diff = 0;
877 htb_change_class_mode(q,cl,&diff);
878 if (old_mode != cl->cmode) {
879 if (old_mode != HTB_CAN_SEND)
880 htb_safe_rb_erase(&cl->pq_node,q->wait_pq+cl->level);
881 if (cl->cmode != HTB_CAN_SEND)
882 htb_add_to_wait_tree (q,cl,diff,1);
886 /* update rate counters */
887 cl->sum_bytes += bytes; cl->sum_packets++;
890 /* update byte stats except for leaves which are already updated */
892 cl->stats.bytes += bytes;
900 * htb_do_events - make mode changes to classes at the level
902 * Scans event queue for pending events and applies them. Returns jiffies to
903 * next pending event (0 for no event in pq).
904 * Note: Aplied are events whose have cl->pq_key <= jiffies.
906 static long htb_do_events(struct htb_sched *q,int level)
909 HTB_DBG(8,1,"htb_do_events l=%d root=%p rmask=%X\n",
910 level,q->wait_pq[level].rb_node,q->row_mask[level]);
911 for (i = 0; i < 500; i++) {
912 struct htb_class *cl;
914 struct rb_node *p = q->wait_pq[level].rb_node;
916 while (p->rb_left) p = p->rb_left;
918 cl = rb_entry(p, struct htb_class, pq_node);
919 if (time_after(cl->pq_key, q->jiffies)) {
920 HTB_DBG(8,3,"htb_do_ev_ret delay=%ld\n",cl->pq_key - q->jiffies);
921 return cl->pq_key - q->jiffies;
923 htb_safe_rb_erase(p,q->wait_pq+level);
924 diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32)cl->mbuffer);
926 if (diff > cl->mbuffer || diff < 0 || PSCHED_TLESS(q->now, cl->t_c)) {
928 printk(KERN_ERR "HTB: bad diff in events, cl=%X diff=%lX now=%Lu then=%Lu j=%lu\n",
930 (unsigned long long) q->now,
931 (unsigned long long) cl->t_c,
936 htb_change_class_mode(q,cl,&diff);
937 if (cl->cmode != HTB_CAN_SEND)
938 htb_add_to_wait_tree (q,cl,diff,2);
941 printk(KERN_WARNING "htb: too many events !\n");
946 * htb_lookup_leaf - returns next leaf class in DRR order
948 * Find leaf where current feed pointers points to.
950 static struct htb_class *
951 htb_lookup_leaf(struct rb_root *tree,int prio,struct rb_node **pptr)
955 struct rb_node *root;
956 struct rb_node **pptr;
957 } stk[TC_HTB_MAXDEPTH],*sp = stk;
959 BUG_TRAP(tree->rb_node);
960 sp->root = tree->rb_node;
963 for (i = 0; i < 65535; i++) {
964 if (!*sp->pptr) { /* we are at right end; rewind & go up */
965 *sp->pptr = sp->root;
966 while ((*sp->pptr)->rb_left)
967 *sp->pptr = (*sp->pptr)->rb_left;
970 BUG_TRAP(*sp->pptr); if(!*sp->pptr) return NULL;
971 htb_next_rb_node (sp->pptr);
974 struct htb_class *cl;
975 cl = rb_entry(*sp->pptr,struct htb_class,node[prio]);
979 (++sp)->root = cl->un.inner.feed[prio].rb_node;
980 sp->pptr = cl->un.inner.ptr+prio;
987 /* dequeues packet at given priority and level; call only if
988 you are sure that there is active class at prio/level */
989 static struct sk_buff *
990 htb_dequeue_tree(struct htb_sched *q,int prio,int level)
992 struct sk_buff *skb = NULL;
993 struct htb_class *cl,*start;
994 /* look initial class up in the row */
995 start = cl = htb_lookup_leaf (q->row[level]+prio,prio,q->ptr[level]+prio);
1000 if (!cl) return NULL;
1001 HTB_DBG(4,1,"htb_deq_tr prio=%d lev=%d cl=%X defic=%d\n",
1002 prio,level,cl->classid,cl->un.leaf.deficit[level]);
1004 /* class can be empty - it is unlikely but can be true if leaf
1005 qdisc drops packets in enqueue routine or if someone used
1006 graft operation on the leaf since last dequeue;
1007 simply deactivate and skip such class */
1008 if (unlikely(cl->un.leaf.q->q.qlen == 0)) {
1009 struct htb_class *next;
1010 htb_deactivate(q,cl);
1012 /* row/level might become empty */
1013 if ((q->row_mask[level] & (1 << prio)) == 0)
1016 next = htb_lookup_leaf (q->row[level]+prio,
1017 prio,q->ptr[level]+prio);
1018 if (cl == start) /* fix start if we just deleted it */
1024 if (likely((skb = cl->un.leaf.q->dequeue(cl->un.leaf.q)) != NULL))
1027 printk(KERN_WARNING "htb: class %X isn't work conserving ?!\n",cl->classid);
1031 htb_next_rb_node((level?cl->parent->un.inner.ptr:q->ptr[0])+prio);
1032 cl = htb_lookup_leaf (q->row[level]+prio,prio,q->ptr[level]+prio);
1033 } while (cl != start);
1035 if (likely(skb != NULL)) {
1036 if ((cl->un.leaf.deficit[level] -= skb->len) < 0) {
1037 HTB_DBG(4,2,"htb_next_cl oldptr=%p quant_add=%d\n",
1038 level?cl->parent->un.inner.ptr[prio]:q->ptr[0][prio],cl->un.leaf.quantum);
1039 cl->un.leaf.deficit[level] += cl->un.leaf.quantum;
1040 htb_next_rb_node((level?cl->parent->un.inner.ptr:q->ptr[0])+prio);
1042 /* this used to be after charge_class but this constelation
1043 gives us slightly better performance */
1044 if (!cl->un.leaf.q->q.qlen)
1045 htb_deactivate (q,cl);
1046 htb_charge_class (q,cl,level,skb->len);
1051 static void htb_delay_by(struct Qdisc *sch,long delay)
1053 struct htb_sched *q = (struct htb_sched *)sch->data;
1054 if (delay <= 0) delay = 1;
1055 if (unlikely(delay > 5*HZ)) {
1056 if (net_ratelimit())
1057 printk(KERN_INFO "HTB delay %ld > 5sec\n", delay);
1060 /* why don't use jiffies here ? because expires can be in past */
1061 mod_timer(&q->timer, q->jiffies + delay);
1062 sch->flags |= TCQ_F_THROTTLED;
1063 sch->stats.overlimits++;
1064 HTB_DBG(3,1,"htb_deq t_delay=%ld\n",delay);
1067 static struct sk_buff *htb_dequeue(struct Qdisc *sch)
1069 struct sk_buff *skb = NULL;
1070 struct htb_sched *q = (struct htb_sched *)sch->data;
1077 q->jiffies = jiffies;
1078 HTB_DBG(3,1,"htb_deq dircnt=%d qlen=%d\n",skb_queue_len(&q->direct_queue),
1081 /* try to dequeue direct packets as high prio (!) to minimize cpu work */
1082 if ((skb = __skb_dequeue(&q->direct_queue)) != NULL) {
1083 sch->flags &= ~TCQ_F_THROTTLED;
1088 if (!sch->q.qlen) goto fin;
1089 PSCHED_GET_TIME(q->now);
1091 min_delay = LONG_MAX;
1093 for (level = 0; level < TC_HTB_MAXDEPTH; level++) {
1094 /* common case optimization - skip event handler quickly */
1097 if (time_after_eq(q->jiffies, q->near_ev_cache[level])) {
1098 delay = htb_do_events(q,level);
1099 q->near_ev_cache[level] = q->jiffies + (delay ? delay : HZ);
1104 delay = q->near_ev_cache[level] - q->jiffies;
1106 if (delay && min_delay > delay)
1108 m = ~q->row_mask[level];
1109 while (m != (int)(-1)) {
1112 skb = htb_dequeue_tree(q,prio,level);
1113 if (likely(skb != NULL)) {
1115 sch->flags &= ~TCQ_F_THROTTLED;
1121 if (!q->nwc_hit && min_delay >= 10*HZ && net_ratelimit()) {
1122 if (min_delay == LONG_MAX) {
1123 printk(KERN_ERR "HTB: dequeue bug (%d,%lu,%lu), report it please !\n",
1124 evs_used,q->jiffies,jiffies);
1127 printk(KERN_WARNING "HTB: mindelay=%ld, some class has "
1128 "too small rate\n",min_delay);
1131 htb_delay_by (sch,min_delay > 5*HZ ? 5*HZ : min_delay);
1133 HTB_DBG(3,1,"htb_deq_end %s j=%lu skb=%p\n",sch->dev->name,q->jiffies,skb);
1137 /* try to drop from each class (by prio) until one succeed */
1138 static unsigned int htb_drop(struct Qdisc* sch)
1140 struct htb_sched *q = (struct htb_sched *)sch->data;
1143 for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) {
1144 struct list_head *p;
1145 list_for_each (p,q->drops+prio) {
1146 struct htb_class *cl = list_entry(p, struct htb_class,
1149 if (cl->un.leaf.q->ops->drop &&
1150 (len = cl->un.leaf.q->ops->drop(cl->un.leaf.q))) {
1152 if (!cl->un.leaf.q->q.qlen)
1153 htb_deactivate (q,cl);
1161 /* reset all classes */
1162 /* always caled under BH & queue lock */
1163 static void htb_reset(struct Qdisc* sch)
1165 struct htb_sched *q = (struct htb_sched *)sch->data;
1167 HTB_DBG(0,1,"htb_reset sch=%p, handle=%X\n",sch,sch->handle);
1169 for (i = 0; i < HTB_HSIZE; i++) {
1170 struct list_head *p;
1171 list_for_each (p,q->hash+i) {
1172 struct htb_class *cl = list_entry(p,struct htb_class,hlist);
1174 memset(&cl->un.inner,0,sizeof(cl->un.inner));
1177 qdisc_reset(cl->un.leaf.q);
1178 INIT_LIST_HEAD(&cl->un.leaf.drop_list);
1180 cl->prio_activity = 0;
1181 cl->cmode = HTB_CAN_SEND;
1183 cl->pq_node.rb_color = -1;
1184 memset(cl->node,255,sizeof(cl->node));
1189 sch->flags &= ~TCQ_F_THROTTLED;
1190 del_timer(&q->timer);
1191 __skb_queue_purge(&q->direct_queue);
1193 memset(q->row,0,sizeof(q->row));
1194 memset(q->row_mask,0,sizeof(q->row_mask));
1195 memset(q->wait_pq,0,sizeof(q->wait_pq));
1196 memset(q->ptr,0,sizeof(q->ptr));
1197 for (i = 0; i < TC_HTB_NUMPRIO; i++)
1198 INIT_LIST_HEAD(q->drops+i);
1201 static int htb_init(struct Qdisc *sch, struct rtattr *opt)
1203 struct htb_sched *q = (struct htb_sched*)sch->data;
1204 struct rtattr *tb[TCA_HTB_INIT];
1205 struct tc_htb_glob *gopt;
1208 printk(KERN_INFO "HTB init, kernel part version %d.%d\n",
1209 HTB_VER >> 16,HTB_VER & 0xffff);
1211 if (!opt || rtattr_parse(tb, TCA_HTB_INIT, RTA_DATA(opt), RTA_PAYLOAD(opt)) ||
1212 tb[TCA_HTB_INIT-1] == NULL ||
1213 RTA_PAYLOAD(tb[TCA_HTB_INIT-1]) < sizeof(*gopt)) {
1214 printk(KERN_ERR "HTB: hey probably you have bad tc tool ?\n");
1217 gopt = RTA_DATA(tb[TCA_HTB_INIT-1]);
1218 if (gopt->version != HTB_VER >> 16) {
1219 printk(KERN_ERR "HTB: need tc/htb version %d (minor is %d), you have %d\n",
1220 HTB_VER >> 16,HTB_VER & 0xffff,gopt->version);
1223 memset(q,0,sizeof(*q));
1224 q->debug = gopt->debug;
1225 HTB_DBG(0,1,"htb_init sch=%p handle=%X r2q=%d\n",sch,sch->handle,gopt->rate2quantum);
1227 INIT_LIST_HEAD(&q->root);
1228 for (i = 0; i < HTB_HSIZE; i++)
1229 INIT_LIST_HEAD(q->hash+i);
1230 for (i = 0; i < TC_HTB_NUMPRIO; i++)
1231 INIT_LIST_HEAD(q->drops+i);
1233 init_timer(&q->timer);
1234 skb_queue_head_init(&q->direct_queue);
1236 q->direct_qlen = sch->dev->tx_queue_len;
1237 if (q->direct_qlen < 2) /* some devices have zero tx_queue_len */
1239 q->timer.function = htb_timer;
1240 q->timer.data = (unsigned long)sch;
1243 init_timer(&q->rttim);
1244 q->rttim.function = htb_rate_timer;
1245 q->rttim.data = (unsigned long)sch;
1246 q->rttim.expires = jiffies + HZ;
1247 add_timer(&q->rttim);
1249 if ((q->rate2quantum = gopt->rate2quantum) < 1)
1250 q->rate2quantum = 1;
1251 q->defcls = gopt->defcls;
1256 static int htb_dump(struct Qdisc *sch, struct sk_buff *skb)
1258 struct htb_sched *q = (struct htb_sched*)sch->data;
1259 unsigned char *b = skb->tail;
1261 struct tc_htb_glob gopt;
1262 HTB_DBG(0,1,"htb_dump sch=%p, handle=%X\n",sch,sch->handle);
1265 gopt.direct_pkts = q->direct_pkts;
1268 if (HTB_DBG_COND(0,2))
1271 gopt.version = HTB_VER;
1272 gopt.rate2quantum = q->rate2quantum;
1273 gopt.defcls = q->defcls;
1274 gopt.debug = q->debug;
1275 rta = (struct rtattr*)b;
1276 RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
1277 RTA_PUT(skb, TCA_HTB_INIT, sizeof(gopt), &gopt);
1278 rta->rta_len = skb->tail - b;
1279 sch->stats.qlen = sch->q.qlen;
1280 RTA_PUT(skb, TCA_STATS, sizeof(sch->stats), &sch->stats);
1285 skb_trim(skb, skb->tail - skb->data);
1289 static int htb_dump_class(struct Qdisc *sch, unsigned long arg,
1290 struct sk_buff *skb, struct tcmsg *tcm)
1293 struct htb_sched *q = (struct htb_sched*)sch->data;
1295 struct htb_class *cl = (struct htb_class*)arg;
1296 unsigned char *b = skb->tail;
1298 struct tc_htb_opt opt;
1300 HTB_DBG(0,1,"htb_dump_class handle=%X clid=%X\n",sch->handle,cl->classid);
1303 tcm->tcm_parent = cl->parent ? cl->parent->classid : TC_H_ROOT;
1304 tcm->tcm_handle = cl->classid;
1305 if (!cl->level && cl->un.leaf.q) {
1306 tcm->tcm_info = cl->un.leaf.q->handle;
1307 cl->stats.qlen = cl->un.leaf.q->q.qlen;
1310 rta = (struct rtattr*)b;
1311 RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
1313 memset (&opt,0,sizeof(opt));
1315 opt.rate = cl->rate->rate; opt.buffer = cl->buffer;
1316 opt.ceil = cl->ceil->rate; opt.cbuffer = cl->cbuffer;
1317 opt.quantum = cl->un.leaf.quantum; opt.prio = cl->un.leaf.prio;
1318 opt.level = cl->level;
1319 RTA_PUT(skb, TCA_HTB_PARMS, sizeof(opt), &opt);
1320 rta->rta_len = skb->tail - b;
1323 cl->stats.bps = cl->rate_bytes/(HTB_EWMAC*HTB_HSIZE);
1324 cl->stats.pps = cl->rate_packets/(HTB_EWMAC*HTB_HSIZE);
1327 cl->xstats.tokens = cl->tokens;
1328 cl->xstats.ctokens = cl->ctokens;
1329 RTA_PUT(skb, TCA_STATS, sizeof(cl->stats), &cl->stats);
1330 RTA_PUT(skb, TCA_XSTATS, sizeof(cl->xstats), &cl->xstats);
1335 skb_trim(skb, b - skb->data);
1339 static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1342 struct htb_class *cl = (struct htb_class*)arg;
1344 if (cl && !cl->level) {
1345 if (new == NULL && (new = qdisc_create_dflt(sch->dev,
1346 &pfifo_qdisc_ops)) == NULL)
1349 if ((*old = xchg(&cl->un.leaf.q, new)) != NULL) {
1350 if (cl->prio_activity)
1351 htb_deactivate ((struct htb_sched*)sch->data,cl);
1353 /* TODO: is it correct ? Why CBQ doesn't do it ? */
1354 sch->q.qlen -= (*old)->q.qlen;
1357 sch_tree_unlock(sch);
1363 static struct Qdisc * htb_leaf(struct Qdisc *sch, unsigned long arg)
1365 struct htb_class *cl = (struct htb_class*)arg;
1366 return (cl && !cl->level) ? cl->un.leaf.q : NULL;
1369 static unsigned long htb_get(struct Qdisc *sch, u32 classid)
1372 struct htb_sched *q = (struct htb_sched *)sch->data;
1374 struct htb_class *cl = htb_find(classid,sch);
1375 HTB_DBG(0,1,"htb_get clid=%X q=%p cl=%p ref=%d\n",classid,q,cl,cl?cl->refcnt:0);
1378 return (unsigned long)cl;
1381 static void htb_destroy_filters(struct tcf_proto **fl)
1383 struct tcf_proto *tp;
1385 while ((tp = *fl) != NULL) {
1391 static void htb_destroy_class(struct Qdisc* sch,struct htb_class *cl)
1393 struct htb_sched *q = (struct htb_sched *)sch->data;
1394 HTB_DBG(0,1,"htb_destrycls clid=%X ref=%d\n", cl?cl->classid:0,cl?cl->refcnt:0);
1396 BUG_TRAP(cl->un.leaf.q);
1397 sch->q.qlen -= cl->un.leaf.q->q.qlen;
1398 qdisc_destroy(cl->un.leaf.q);
1400 qdisc_put_rtab(cl->rate);
1401 qdisc_put_rtab(cl->ceil);
1403 #ifdef CONFIG_NET_ESTIMATOR
1404 qdisc_kill_estimator(&cl->stats);
1406 htb_destroy_filters (&cl->filter_list);
1408 while (!list_empty(&cl->children))
1409 htb_destroy_class (sch,list_entry(cl->children.next,
1410 struct htb_class,sibling));
1412 /* note: this delete may happen twice (see htb_delete) */
1413 list_del(&cl->hlist);
1414 list_del(&cl->sibling);
1416 if (cl->prio_activity)
1417 htb_deactivate (q,cl);
1419 if (cl->cmode != HTB_CAN_SEND)
1420 htb_safe_rb_erase(&cl->pq_node,q->wait_pq+cl->level);
1425 /* always caled under BH & queue lock */
1426 static void htb_destroy(struct Qdisc* sch)
1428 struct htb_sched *q = (struct htb_sched *)sch->data;
1429 HTB_DBG(0,1,"htb_destroy q=%p\n",q);
1431 del_timer_sync (&q->timer);
1433 del_timer_sync (&q->rttim);
1435 /* This line used to be after htb_destroy_class call below
1436 and surprisingly it worked in 2.4. But it must precede it
1437 because filter need its target class alive to be able to call
1438 unbind_filter on it (without Oops). */
1439 htb_destroy_filters(&q->filter_list);
1441 while (!list_empty(&q->root))
1442 htb_destroy_class (sch,list_entry(q->root.next,
1443 struct htb_class,sibling));
1445 __skb_queue_purge(&q->direct_queue);
1448 static int htb_delete(struct Qdisc *sch, unsigned long arg)
1450 struct htb_sched *q = (struct htb_sched *)sch->data;
1451 struct htb_class *cl = (struct htb_class*)arg;
1452 HTB_DBG(0,1,"htb_delete q=%p cl=%X ref=%d\n",q,cl?cl->classid:0,cl?cl->refcnt:0);
1454 // TODO: why don't allow to delete subtree ? references ? does
1455 // tc subsys quarantee us that in htb_destroy it holds no class
1456 // refs so that we can remove children safely there ?
1457 if (!list_empty(&cl->children) || cl->filter_cnt)
1462 /* delete from hash and active; remainder in destroy_class */
1463 list_del_init(&cl->hlist);
1464 if (cl->prio_activity)
1465 htb_deactivate (q,cl);
1467 if (--cl->refcnt == 0)
1468 htb_destroy_class(sch,cl);
1470 sch_tree_unlock(sch);
1474 static void htb_put(struct Qdisc *sch, unsigned long arg)
1477 struct htb_sched *q = (struct htb_sched *)sch->data;
1479 struct htb_class *cl = (struct htb_class*)arg;
1480 HTB_DBG(0,1,"htb_put q=%p cl=%X ref=%d\n",q,cl?cl->classid:0,cl?cl->refcnt:0);
1482 if (--cl->refcnt == 0)
1483 htb_destroy_class(sch,cl);
1486 static int htb_change_class(struct Qdisc *sch, u32 classid,
1487 u32 parentid, struct rtattr **tca, unsigned long *arg)
1490 struct htb_sched *q = (struct htb_sched *)sch->data;
1491 struct htb_class *cl = (struct htb_class*)*arg,*parent;
1492 struct rtattr *opt = tca[TCA_OPTIONS-1];
1493 struct qdisc_rate_table *rtab = NULL, *ctab = NULL;
1494 struct rtattr *tb[TCA_HTB_RTAB];
1495 struct tc_htb_opt *hopt;
1497 /* extract all subattrs from opt attr */
1498 if (!opt || rtattr_parse(tb, TCA_HTB_RTAB, RTA_DATA(opt), RTA_PAYLOAD(opt)) ||
1499 tb[TCA_HTB_PARMS-1] == NULL ||
1500 RTA_PAYLOAD(tb[TCA_HTB_PARMS-1]) < sizeof(*hopt))
1503 parent = parentid == TC_H_ROOT ? NULL : htb_find (parentid,sch);
1505 hopt = RTA_DATA(tb[TCA_HTB_PARMS-1]);
1506 HTB_DBG(0,1,"htb_chg cl=%p(%X), clid=%X, parid=%X, opt/prio=%d, rate=%u, buff=%d, quant=%d\n", cl,cl?cl->classid:0,classid,parentid,(int)hopt->prio,hopt->rate.rate,hopt->buffer,hopt->quantum);
1507 rtab = qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB-1]);
1508 ctab = qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB-1]);
1509 if (!rtab || !ctab) goto failure;
1511 if (!cl) { /* new class */
1512 struct Qdisc *new_q;
1513 /* check for valid classid */
1514 if (!classid || TC_H_MAJ(classid^sch->handle) || htb_find(classid,sch))
1517 /* check maximal depth */
1518 if (parent && parent->parent && parent->parent->level < 2) {
1519 printk(KERN_ERR "htb: tree is too deep\n");
1523 if ((cl = kmalloc(sizeof(*cl), GFP_KERNEL)) == NULL)
1526 memset(cl, 0, sizeof(*cl));
1528 INIT_LIST_HEAD(&cl->sibling);
1529 INIT_LIST_HEAD(&cl->hlist);
1530 INIT_LIST_HEAD(&cl->children);
1531 INIT_LIST_HEAD(&cl->un.leaf.drop_list);
1533 cl->magic = HTB_CMAGIC;
1536 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL)
1537 so that can't be used inside of sch_tree_lock
1538 -- thanks to Karlis Peisenieks */
1539 new_q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops);
1541 if (parent && !parent->level) {
1542 /* turn parent into inner node */
1543 sch->q.qlen -= parent->un.leaf.q->q.qlen;
1544 qdisc_destroy (parent->un.leaf.q);
1545 if (parent->prio_activity)
1546 htb_deactivate (q,parent);
1548 /* remove from evt list because of level change */
1549 if (parent->cmode != HTB_CAN_SEND) {
1550 htb_safe_rb_erase(&parent->pq_node,q->wait_pq /*+0*/);
1551 parent->cmode = HTB_CAN_SEND;
1553 parent->level = (parent->parent ? parent->parent->level
1554 : TC_HTB_MAXDEPTH) - 1;
1555 memset (&parent->un.inner,0,sizeof(parent->un.inner));
1557 /* leaf (we) needs elementary qdisc */
1558 cl->un.leaf.q = new_q ? new_q : &noop_qdisc;
1560 cl->classid = classid; cl->parent = parent;
1562 /* set class to be in HTB_CAN_SEND state */
1563 cl->tokens = hopt->buffer;
1564 cl->ctokens = hopt->cbuffer;
1565 cl->mbuffer = 60000000; /* 1min */
1566 PSCHED_GET_TIME(cl->t_c);
1567 cl->cmode = HTB_CAN_SEND;
1569 /* attach to the hash list and parent's family */
1570 list_add_tail(&cl->hlist, q->hash+htb_hash(classid));
1571 list_add_tail(&cl->sibling, parent ? &parent->children : &q->root);
1575 for (i = 0; i < TC_HTB_NUMPRIO; i++) cl->node[i].rb_color = -1;
1576 cl->pq_node.rb_color = -1;
1579 } else sch_tree_lock(sch);
1581 /* it used to be a nasty bug here, we have to check that node
1582 is really leaf before changing cl->un.leaf ! */
1584 cl->un.leaf.quantum = rtab->rate.rate / q->rate2quantum;
1585 if (!hopt->quantum && cl->un.leaf.quantum < 1000) {
1586 printk(KERN_WARNING "HTB: quantum of class %X is small. Consider r2q change.\n", cl->classid);
1587 cl->un.leaf.quantum = 1000;
1589 if (!hopt->quantum && cl->un.leaf.quantum > 200000) {
1590 printk(KERN_WARNING "HTB: quantum of class %X is big. Consider r2q change.\n", cl->classid);
1591 cl->un.leaf.quantum = 200000;
1594 cl->un.leaf.quantum = hopt->quantum;
1595 if ((cl->un.leaf.prio = hopt->prio) >= TC_HTB_NUMPRIO)
1596 cl->un.leaf.prio = TC_HTB_NUMPRIO - 1;
1599 cl->buffer = hopt->buffer;
1600 cl->cbuffer = hopt->cbuffer;
1601 if (cl->rate) qdisc_put_rtab(cl->rate); cl->rate = rtab;
1602 if (cl->ceil) qdisc_put_rtab(cl->ceil); cl->ceil = ctab;
1603 sch_tree_unlock(sch);
1605 *arg = (unsigned long)cl;
1609 if (rtab) qdisc_put_rtab(rtab);
1610 if (ctab) qdisc_put_rtab(ctab);
1614 static struct tcf_proto **htb_find_tcf(struct Qdisc *sch, unsigned long arg)
1616 struct htb_sched *q = (struct htb_sched *)sch->data;
1617 struct htb_class *cl = (struct htb_class *)arg;
1618 struct tcf_proto **fl = cl ? &cl->filter_list : &q->filter_list;
1619 HTB_DBG(0,2,"htb_tcf q=%p clid=%X fref=%d fl=%p\n",q,cl?cl->classid:0,cl?cl->filter_cnt:q->filter_cnt,*fl);
1623 static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent,
1626 struct htb_sched *q = (struct htb_sched *)sch->data;
1627 struct htb_class *cl = htb_find (classid,sch);
1628 HTB_DBG(0,2,"htb_bind q=%p clid=%X cl=%p fref=%d\n",q,classid,cl,cl?cl->filter_cnt:q->filter_cnt);
1629 /*if (cl && !cl->level) return 0;
1630 The line above used to be there to prevent attaching filters to
1631 leaves. But at least tc_index filter uses this just to get class
1632 for other reasons so that we have to allow for it.
1634 19.6.2002 As Werner explained it is ok - bind filter is just
1635 another way to "lock" the class - unlike "get" this lock can
1636 be broken by class during destroy IIUC.
1642 return (unsigned long)cl;
1645 static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg)
1647 struct htb_sched *q = (struct htb_sched *)sch->data;
1648 struct htb_class *cl = (struct htb_class *)arg;
1649 HTB_DBG(0,2,"htb_unbind q=%p cl=%p fref=%d\n",q,cl,cl?cl->filter_cnt:q->filter_cnt);
1656 static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1658 struct htb_sched *q = (struct htb_sched *)sch->data;
1664 for (i = 0; i < HTB_HSIZE; i++) {
1665 struct list_head *p;
1666 list_for_each (p,q->hash+i) {
1667 struct htb_class *cl = list_entry(p,struct htb_class,hlist);
1668 if (arg->count < arg->skip) {
1672 if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
1681 static struct Qdisc_class_ops htb_class_ops = {
1686 .change = htb_change_class,
1687 .delete = htb_delete,
1689 .tcf_chain = htb_find_tcf,
1690 .bind_tcf = htb_bind_filter,
1691 .unbind_tcf = htb_unbind_filter,
1692 .dump = htb_dump_class,
1695 static struct Qdisc_ops htb_qdisc_ops = {
1697 .cl_ops = &htb_class_ops,
1699 .priv_size = sizeof(struct htb_sched),
1700 .enqueue = htb_enqueue,
1701 .dequeue = htb_dequeue,
1702 .requeue = htb_requeue,
1706 .destroy = htb_destroy,
1707 .change = NULL /* htb_change */,
1709 .owner = THIS_MODULE,
1712 static int __init htb_module_init(void)
1714 return register_qdisc(&htb_qdisc_ops);
1716 static void __exit htb_module_exit(void)
1718 unregister_qdisc(&htb_qdisc_ops);
1720 module_init(htb_module_init)
1721 module_exit(htb_module_exit)
1722 MODULE_LICENSE("GPL");