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 <asm/uaccess.h>
32 #include <asm/system.h>
33 #include <linux/bitops.h>
34 #include <linux/types.h>
35 #include <linux/kernel.h>
36 #include <linux/string.h>
38 #include <linux/socket.h>
39 #include <linux/sockios.h>
41 #include <linux/errno.h>
42 #include <linux/interrupt.h>
43 #include <linux/if_ether.h>
44 #include <linux/inet.h>
45 #include <linux/netdevice.h>
46 #include <linux/etherdevice.h>
47 #include <linux/notifier.h>
49 #include <net/route.h>
50 #include <linux/skbuff.h>
51 #include <linux/list.h>
52 #include <linux/compiler.h>
53 #include <net/netlink.h>
55 #include <net/pkt_sched.h>
56 #include <linux/rbtree.h>
60 ========================================================================
61 HTB is like TBF with multiple classes. It is also similar to CBQ because
62 it allows to assign priority to each class in hierarchy.
63 In fact it is another implementation of Floyd's formal sharing.
66 Each class is assigned level. Leaf has ALWAYS level 0 and root
67 classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level
68 one less than their parent.
71 #define HTB_HSIZE 16 /* classid hash size */
72 #define HTB_EWMAC 2 /* rate average over HTB_EWMAC*HTB_HSIZE sec */
73 #define HTB_RATECM 1 /* whether to use rate computer */
74 //#define HTB_HYSTERESIS 0 /* whether to use mode hysteresis for speedup */
75 #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */
77 #if HTB_VER >> 16 != TC_HTB_PROTOVER
78 #error "Mismatched sch_htb.c and pkt_sch.h"
81 /* used internaly to keep status of single class */
83 HTB_CANT_SEND
, /* class can't send and can't borrow */
84 HTB_MAY_BORROW
, /* class can't send but may borrow */
85 HTB_CAN_SEND
/* class can send */
88 /* interior & leaf nodes; props specific to leaves are marked L: */
90 /* general class parameters */
92 struct gnet_stats_basic bstats
;
93 struct gnet_stats_queue qstats
;
94 struct gnet_stats_rate_est rate_est
;
95 struct tc_htb_xstats xstats
; /* our special stats */
96 int refcnt
; /* usage count of this class */
99 /* rate measurement counters */
100 unsigned long rate_bytes
, sum_bytes
;
101 unsigned long rate_packets
, sum_packets
;
105 int level
; /* our level (see above) */
106 struct htb_class
*parent
; /* parent class */
107 struct hlist_node hlist
; /* classid hash list item */
108 struct list_head sibling
; /* sibling list item */
109 struct list_head children
; /* children list */
112 struct htb_class_leaf
{
117 int deficit
[TC_HTB_MAXDEPTH
];
118 struct list_head drop_list
;
120 struct htb_class_inner
{
121 struct rb_root feed
[TC_HTB_NUMPRIO
]; /* feed trees */
122 struct rb_node
*ptr
[TC_HTB_NUMPRIO
]; /* current class ptr */
123 /* When class changes from state 1->2 and disconnects from
124 parent's feed then we lost ptr value and start from the
125 first child again. Here we store classid of the
126 last valid ptr (used when ptr is NULL). */
127 u32 last_ptr_id
[TC_HTB_NUMPRIO
];
130 struct rb_node node
[TC_HTB_NUMPRIO
]; /* node for self or feed tree */
131 struct rb_node pq_node
; /* node for event queue */
132 psched_time_t pq_key
;
134 int prio_activity
; /* for which prios are we active */
135 enum htb_cmode cmode
; /* current mode of the class */
137 /* class attached filters */
138 struct tcf_proto
*filter_list
;
141 int warned
; /* only one warning about non work conserving .. */
143 /* token bucket parameters */
144 struct qdisc_rate_table
*rate
; /* rate table of the class itself */
145 struct qdisc_rate_table
*ceil
; /* ceiling rate (limits borrows too) */
146 long buffer
, cbuffer
; /* token bucket depth/rate */
147 psched_tdiff_t mbuffer
; /* max wait time */
148 long tokens
, ctokens
; /* current number of tokens */
149 psched_time_t t_c
; /* checkpoint time */
151 int prio
; /* For parent to leaf return possible here */
152 int quantum
; /* we do backup. Finally full replacement */
153 /* of un.leaf originals should be done. */
156 /* TODO: maybe compute rate when size is too large .. or drop ? */
157 static inline long L2T(struct htb_class
*cl
, struct qdisc_rate_table
*rate
,
160 long result
= qdisc_l2t(rate
, size
);
161 if (result
> rate
->data
[255])
167 struct list_head root
; /* root classes list */
168 struct hlist_head hash
[HTB_HSIZE
]; /* hashed by classid */
169 struct list_head drops
[TC_HTB_NUMPRIO
];/* active leaves (for drops) */
171 /* self list - roots of self generating tree */
172 struct rb_root row
[TC_HTB_MAXDEPTH
][TC_HTB_NUMPRIO
];
173 int row_mask
[TC_HTB_MAXDEPTH
];
174 struct rb_node
*ptr
[TC_HTB_MAXDEPTH
][TC_HTB_NUMPRIO
];
175 u32 last_ptr_id
[TC_HTB_MAXDEPTH
][TC_HTB_NUMPRIO
];
177 /* self wait list - roots of wait PQs per row */
178 struct rb_root wait_pq
[TC_HTB_MAXDEPTH
];
180 /* time of nearest event per level (row) */
181 psched_time_t near_ev_cache
[TC_HTB_MAXDEPTH
];
183 /* whether we hit non-work conserving class during this dequeue; we use */
184 int nwc_hit
; /* this to disable mindelay complaint in dequeue */
186 int defcls
; /* class where unclassified flows go to */
188 /* filters for qdisc itself */
189 struct tcf_proto
*filter_list
;
192 int rate2quantum
; /* quant = rate / rate2quantum */
193 psched_time_t now
; /* cached dequeue time */
194 struct qdisc_watchdog watchdog
;
196 struct timer_list rttim
; /* rate computer timer */
197 int recmp_bucket
; /* which hash bucket to recompute next */
200 /* non shaped skbs; let them go directly thru */
201 struct sk_buff_head direct_queue
;
202 int direct_qlen
; /* max qlen of above */
207 /* compute hash of size HTB_HSIZE for given handle */
208 static inline int htb_hash(u32 h
)
211 #error "Declare new hash for your HTB_HSIZE"
213 h
^= h
>> 8; /* stolen from cbq_hash */
218 /* find class in global hash table using given handle */
219 static inline struct htb_class
*htb_find(u32 handle
, struct Qdisc
*sch
)
221 struct htb_sched
*q
= qdisc_priv(sch
);
222 struct hlist_node
*p
;
223 struct htb_class
*cl
;
225 if (TC_H_MAJ(handle
) != sch
->handle
)
228 hlist_for_each_entry(cl
, p
, q
->hash
+ htb_hash(handle
), hlist
) {
229 if (cl
->classid
== handle
)
236 * htb_classify - classify a packet into class
238 * It returns NULL if the packet should be dropped or -1 if the packet
239 * should be passed directly thru. In all other cases leaf class is returned.
240 * We allow direct class selection by classid in priority. The we examine
241 * filters in qdisc and in inner nodes (if higher filter points to the inner
242 * node). If we end up with classid MAJOR:0 we enqueue the skb into special
243 * internal fifo (direct). These packets then go directly thru. If we still
244 * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessfull
245 * then finish and return direct queue.
247 #define HTB_DIRECT (struct htb_class*)-1
248 static inline u32
htb_classid(struct htb_class
*cl
)
250 return (cl
&& cl
!= HTB_DIRECT
) ? cl
->classid
: TC_H_UNSPEC
;
253 static struct htb_class
*htb_classify(struct sk_buff
*skb
, struct Qdisc
*sch
,
256 struct htb_sched
*q
= qdisc_priv(sch
);
257 struct htb_class
*cl
;
258 struct tcf_result res
;
259 struct tcf_proto
*tcf
;
262 /* allow to select class by setting skb->priority to valid classid;
263 note that nfmark can be used too by attaching filter fw with no
265 if (skb
->priority
== sch
->handle
)
266 return HTB_DIRECT
; /* X:0 (direct flow) selected */
267 if ((cl
= htb_find(skb
->priority
, sch
)) != NULL
&& cl
->level
== 0)
270 *qerr
= NET_XMIT_BYPASS
;
271 tcf
= q
->filter_list
;
272 while (tcf
&& (result
= tc_classify(skb
, tcf
, &res
)) >= 0) {
273 #ifdef CONFIG_NET_CLS_ACT
277 *qerr
= NET_XMIT_SUCCESS
;
281 #elif defined(CONFIG_NET_CLS_POLICE)
282 if (result
== TC_POLICE_SHOT
)
285 if ((cl
= (void *)res
.class) == NULL
) {
286 if (res
.classid
== sch
->handle
)
287 return HTB_DIRECT
; /* X:0 (direct flow) */
288 if ((cl
= htb_find(res
.classid
, sch
)) == NULL
)
289 break; /* filter selected invalid classid */
292 return cl
; /* we hit leaf; return it */
294 /* we have got inner class; apply inner filter chain */
295 tcf
= cl
->filter_list
;
297 /* classification failed; try to use default class */
298 cl
= htb_find(TC_H_MAKE(TC_H_MAJ(sch
->handle
), q
->defcls
), sch
);
299 if (!cl
|| cl
->level
)
300 return HTB_DIRECT
; /* bad default .. this is safe bet */
305 * htb_add_to_id_tree - adds class to the round robin list
307 * Routine adds class to the list (actually tree) sorted by classid.
308 * Make sure that class is not already on such list for given prio.
310 static void htb_add_to_id_tree(struct rb_root
*root
,
311 struct htb_class
*cl
, int prio
)
313 struct rb_node
**p
= &root
->rb_node
, *parent
= NULL
;
318 c
= rb_entry(parent
, struct htb_class
, node
[prio
]);
320 if (cl
->classid
> c
->classid
)
321 p
= &parent
->rb_right
;
323 p
= &parent
->rb_left
;
325 rb_link_node(&cl
->node
[prio
], parent
, p
);
326 rb_insert_color(&cl
->node
[prio
], root
);
330 * htb_add_to_wait_tree - adds class to the event queue with delay
332 * The class is added to priority event queue to indicate that class will
333 * change its mode in cl->pq_key microseconds. Make sure that class is not
334 * already in the queue.
336 static void htb_add_to_wait_tree(struct htb_sched
*q
,
337 struct htb_class
*cl
, long delay
)
339 struct rb_node
**p
= &q
->wait_pq
[cl
->level
].rb_node
, *parent
= NULL
;
341 cl
->pq_key
= q
->now
+ delay
;
342 if (cl
->pq_key
== q
->now
)
345 /* update the nearest event cache */
346 if (q
->near_ev_cache
[cl
->level
] > cl
->pq_key
)
347 q
->near_ev_cache
[cl
->level
] = cl
->pq_key
;
352 c
= rb_entry(parent
, struct htb_class
, pq_node
);
353 if (cl
->pq_key
>= c
->pq_key
)
354 p
= &parent
->rb_right
;
356 p
= &parent
->rb_left
;
358 rb_link_node(&cl
->pq_node
, parent
, p
);
359 rb_insert_color(&cl
->pq_node
, &q
->wait_pq
[cl
->level
]);
363 * htb_next_rb_node - finds next node in binary tree
365 * When we are past last key we return NULL.
366 * Average complexity is 2 steps per call.
368 static inline void htb_next_rb_node(struct rb_node
**n
)
374 * htb_add_class_to_row - add class to its row
376 * The class is added to row at priorities marked in mask.
377 * It does nothing if mask == 0.
379 static inline void htb_add_class_to_row(struct htb_sched
*q
,
380 struct htb_class
*cl
, int mask
)
382 q
->row_mask
[cl
->level
] |= mask
;
384 int prio
= ffz(~mask
);
385 mask
&= ~(1 << prio
);
386 htb_add_to_id_tree(q
->row
[cl
->level
] + prio
, cl
, prio
);
390 /* If this triggers, it is a bug in this code, but it need not be fatal */
391 static void htb_safe_rb_erase(struct rb_node
*rb
, struct rb_root
*root
)
393 if (RB_EMPTY_NODE(rb
)) {
403 * htb_remove_class_from_row - removes class from its row
405 * The class is removed from row at priorities marked in mask.
406 * It does nothing if mask == 0.
408 static inline void htb_remove_class_from_row(struct htb_sched
*q
,
409 struct htb_class
*cl
, int mask
)
414 int prio
= ffz(~mask
);
416 mask
&= ~(1 << prio
);
417 if (q
->ptr
[cl
->level
][prio
] == cl
->node
+ prio
)
418 htb_next_rb_node(q
->ptr
[cl
->level
] + prio
);
420 htb_safe_rb_erase(cl
->node
+ prio
, q
->row
[cl
->level
] + prio
);
421 if (!q
->row
[cl
->level
][prio
].rb_node
)
424 q
->row_mask
[cl
->level
] &= ~m
;
428 * htb_activate_prios - creates active classe's feed chain
430 * The class is connected to ancestors and/or appropriate rows
431 * for priorities it is participating on. cl->cmode must be new
432 * (activated) mode. It does nothing if cl->prio_activity == 0.
434 static void htb_activate_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
) {
445 if (p
->un
.inner
.feed
[prio
].rb_node
)
446 /* parent already has its feed in use so that
447 reset bit in mask as parent is already ok */
448 mask
&= ~(1 << prio
);
450 htb_add_to_id_tree(p
->un
.inner
.feed
+ prio
, cl
, prio
);
452 p
->prio_activity
|= mask
;
457 if (cl
->cmode
== HTB_CAN_SEND
&& mask
)
458 htb_add_class_to_row(q
, cl
, mask
);
462 * htb_deactivate_prios - remove class from feed chain
464 * cl->cmode must represent old mode (before deactivation). It does
465 * nothing if cl->prio_activity == 0. Class is removed from all feed
468 static void htb_deactivate_prios(struct htb_sched
*q
, struct htb_class
*cl
)
470 struct htb_class
*p
= cl
->parent
;
471 long m
, mask
= cl
->prio_activity
;
473 while (cl
->cmode
== HTB_MAY_BORROW
&& p
&& mask
) {
480 if (p
->un
.inner
.ptr
[prio
] == cl
->node
+ prio
) {
481 /* we are removing child which is pointed to from
482 parent feed - forget the pointer but remember
484 p
->un
.inner
.last_ptr_id
[prio
] = cl
->classid
;
485 p
->un
.inner
.ptr
[prio
] = NULL
;
488 htb_safe_rb_erase(cl
->node
+ prio
, p
->un
.inner
.feed
+ prio
);
490 if (!p
->un
.inner
.feed
[prio
].rb_node
)
494 p
->prio_activity
&= ~mask
;
499 if (cl
->cmode
== HTB_CAN_SEND
&& mask
)
500 htb_remove_class_from_row(q
, cl
, mask
);
504 static inline long htb_lowater(const struct htb_class
*cl
)
506 return cl
->cmode
!= HTB_CANT_SEND
? -cl
->cbuffer
: 0;
508 static inline long htb_hiwater(const struct htb_class
*cl
)
510 return cl
->cmode
== HTB_CAN_SEND
? -cl
->buffer
: 0;
513 #define htb_lowater(cl) (0)
514 #define htb_hiwater(cl) (0)
518 * htb_class_mode - computes and returns current class mode
520 * It computes cl's mode at time cl->t_c+diff and returns it. If mode
521 * is not HTB_CAN_SEND then cl->pq_key is updated to time difference
522 * from now to time when cl will change its state.
523 * Also it is worth to note that class mode doesn't change simply
524 * at cl->{c,}tokens == 0 but there can rather be hysteresis of
525 * 0 .. -cl->{c,}buffer range. It is meant to limit number of
526 * mode transitions per time unit. The speed gain is about 1/6.
528 static inline enum htb_cmode
529 htb_class_mode(struct htb_class
*cl
, long *diff
)
533 if ((toks
= (cl
->ctokens
+ *diff
)) < htb_lowater(cl
)) {
535 return HTB_CANT_SEND
;
538 if ((toks
= (cl
->tokens
+ *diff
)) >= htb_hiwater(cl
))
542 return HTB_MAY_BORROW
;
546 * htb_change_class_mode - changes classe's mode
548 * This should be the only way how to change classe's mode under normal
549 * cirsumstances. Routine will update feed lists linkage, change mode
550 * and add class to the wait event queue if appropriate. New mode should
551 * be different from old one and cl->pq_key has to be valid if changing
552 * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree).
555 htb_change_class_mode(struct htb_sched
*q
, struct htb_class
*cl
, long *diff
)
557 enum htb_cmode new_mode
= htb_class_mode(cl
, diff
);
559 if (new_mode
== cl
->cmode
)
562 if (cl
->prio_activity
) { /* not necessary: speed optimization */
563 if (cl
->cmode
!= HTB_CANT_SEND
)
564 htb_deactivate_prios(q
, cl
);
565 cl
->cmode
= new_mode
;
566 if (new_mode
!= HTB_CANT_SEND
)
567 htb_activate_prios(q
, cl
);
569 cl
->cmode
= new_mode
;
573 * htb_activate - inserts leaf cl into appropriate active feeds
575 * Routine learns (new) priority of leaf and activates feed chain
576 * for the prio. It can be called on already active leaf safely.
577 * It also adds leaf into droplist.
579 static inline void htb_activate(struct htb_sched
*q
, struct htb_class
*cl
)
581 BUG_TRAP(!cl
->level
&& cl
->un
.leaf
.q
&& cl
->un
.leaf
.q
->q
.qlen
);
583 if (!cl
->prio_activity
) {
584 cl
->prio_activity
= 1 << (cl
->un
.leaf
.aprio
= cl
->un
.leaf
.prio
);
585 htb_activate_prios(q
, cl
);
586 list_add_tail(&cl
->un
.leaf
.drop_list
,
587 q
->drops
+ cl
->un
.leaf
.aprio
);
592 * htb_deactivate - remove leaf cl from active feeds
594 * Make sure that leaf is active. In the other words it can't be called
595 * with non-active leaf. It also removes class from the drop list.
597 static inline void htb_deactivate(struct htb_sched
*q
, struct htb_class
*cl
)
599 BUG_TRAP(cl
->prio_activity
);
601 htb_deactivate_prios(q
, cl
);
602 cl
->prio_activity
= 0;
603 list_del_init(&cl
->un
.leaf
.drop_list
);
606 static int htb_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
)
609 struct htb_sched
*q
= qdisc_priv(sch
);
610 struct htb_class
*cl
= htb_classify(skb
, sch
, &ret
);
612 if (cl
== HTB_DIRECT
) {
613 /* enqueue to helper queue */
614 if (q
->direct_queue
.qlen
< q
->direct_qlen
) {
615 __skb_queue_tail(&q
->direct_queue
, skb
);
620 return NET_XMIT_DROP
;
622 #ifdef CONFIG_NET_CLS_ACT
624 if (ret
== NET_XMIT_BYPASS
)
629 } else if (cl
->un
.leaf
.q
->enqueue(skb
, cl
->un
.leaf
.q
) !=
635 cl
->bstats
.packets
++;
636 cl
->bstats
.bytes
+= skb
->len
;
641 sch
->bstats
.packets
++;
642 sch
->bstats
.bytes
+= skb
->len
;
643 return NET_XMIT_SUCCESS
;
646 /* TODO: requeuing packet charges it to policers again !! */
647 static int htb_requeue(struct sk_buff
*skb
, struct Qdisc
*sch
)
650 struct htb_sched
*q
= qdisc_priv(sch
);
651 struct htb_class
*cl
= htb_classify(skb
, sch
, &ret
);
652 struct sk_buff
*tskb
;
654 if (cl
== HTB_DIRECT
) {
655 /* enqueue to helper queue */
656 if (q
->direct_queue
.qlen
< q
->direct_qlen
) {
657 __skb_queue_head(&q
->direct_queue
, skb
);
659 __skb_queue_head(&q
->direct_queue
, skb
);
660 tskb
= __skb_dequeue_tail(&q
->direct_queue
);
665 #ifdef CONFIG_NET_CLS_ACT
667 if (ret
== NET_XMIT_BYPASS
)
672 } else if (cl
->un
.leaf
.q
->ops
->requeue(skb
, cl
->un
.leaf
.q
) !=
681 sch
->qstats
.requeues
++;
682 return NET_XMIT_SUCCESS
;
686 #define RT_GEN(D,R) R+=D-(R/HTB_EWMAC);D=0
687 static void htb_rate_timer(unsigned long arg
)
689 struct Qdisc
*sch
= (struct Qdisc
*)arg
;
690 struct htb_sched
*q
= qdisc_priv(sch
);
691 struct hlist_node
*p
;
692 struct htb_class
*cl
;
695 /* lock queue so that we can muck with it */
696 spin_lock_bh(&sch
->dev
->queue_lock
);
698 q
->rttim
.expires
= jiffies
+ HZ
;
699 add_timer(&q
->rttim
);
701 /* scan and recompute one bucket at time */
702 if (++q
->recmp_bucket
>= HTB_HSIZE
)
705 hlist_for_each_entry(cl
,p
, q
->hash
+ q
->recmp_bucket
, hlist
) {
706 RT_GEN(cl
->sum_bytes
, cl
->rate_bytes
);
707 RT_GEN(cl
->sum_packets
, cl
->rate_packets
);
709 spin_unlock_bh(&sch
->dev
->queue_lock
);
714 * htb_charge_class - charges amount "bytes" to leaf and ancestors
716 * Routine assumes that packet "bytes" long was dequeued from leaf cl
717 * borrowing from "level". It accounts bytes to ceil leaky bucket for
718 * leaf and all ancestors and to rate bucket for ancestors at levels
719 * "level" and higher. It also handles possible change of mode resulting
720 * from the update. Note that mode can also increase here (MAY_BORROW to
721 * CAN_SEND) because we can use more precise clock that event queue here.
722 * In such case we remove class from event queue first.
724 static void htb_charge_class(struct htb_sched
*q
, struct htb_class
*cl
,
725 int level
, int bytes
)
728 enum htb_cmode old_mode
;
730 #define HTB_ACCNT(T,B,R) toks = diff + cl->T; \
731 if (toks > cl->B) toks = cl->B; \
732 toks -= L2T(cl, cl->R, bytes); \
733 if (toks <= -cl->mbuffer) toks = 1-cl->mbuffer; \
737 diff
= psched_tdiff_bounded(q
->now
, cl
->t_c
, cl
->mbuffer
);
738 if (cl
->level
>= level
) {
739 if (cl
->level
== level
)
741 HTB_ACCNT(tokens
, buffer
, rate
);
743 cl
->xstats
.borrows
++;
744 cl
->tokens
+= diff
; /* we moved t_c; update tokens */
746 HTB_ACCNT(ctokens
, cbuffer
, ceil
);
749 old_mode
= cl
->cmode
;
751 htb_change_class_mode(q
, cl
, &diff
);
752 if (old_mode
!= cl
->cmode
) {
753 if (old_mode
!= HTB_CAN_SEND
)
754 htb_safe_rb_erase(&cl
->pq_node
, q
->wait_pq
+ cl
->level
);
755 if (cl
->cmode
!= HTB_CAN_SEND
)
756 htb_add_to_wait_tree(q
, cl
, diff
);
759 /* update rate counters */
760 cl
->sum_bytes
+= bytes
;
764 /* update byte stats except for leaves which are already updated */
766 cl
->bstats
.bytes
+= bytes
;
767 cl
->bstats
.packets
++;
774 * htb_do_events - make mode changes to classes at the level
776 * Scans event queue for pending events and applies them. Returns time of
777 * next pending event (0 for no event in pq).
778 * Note: Applied are events whose have cl->pq_key <= q->now.
780 static psched_time_t
htb_do_events(struct htb_sched
*q
, int level
)
784 for (i
= 0; i
< 500; i
++) {
785 struct htb_class
*cl
;
787 struct rb_node
*p
= rb_first(&q
->wait_pq
[level
]);
792 cl
= rb_entry(p
, struct htb_class
, pq_node
);
793 if (cl
->pq_key
> q
->now
)
796 htb_safe_rb_erase(p
, q
->wait_pq
+ level
);
797 diff
= psched_tdiff_bounded(q
->now
, cl
->t_c
, cl
->mbuffer
);
798 htb_change_class_mode(q
, cl
, &diff
);
799 if (cl
->cmode
!= HTB_CAN_SEND
)
800 htb_add_to_wait_tree(q
, cl
, diff
);
803 printk(KERN_WARNING
"htb: too many events !\n");
804 return q
->now
+ PSCHED_TICKS_PER_SEC
/ 10;
807 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL
808 is no such one exists. */
809 static struct rb_node
*htb_id_find_next_upper(int prio
, struct rb_node
*n
,
812 struct rb_node
*r
= NULL
;
814 struct htb_class
*cl
=
815 rb_entry(n
, struct htb_class
, node
[prio
]);
816 if (id
== cl
->classid
)
819 if (id
> cl
->classid
) {
830 * htb_lookup_leaf - returns next leaf class in DRR order
832 * Find leaf where current feed pointers points to.
834 static struct htb_class
*htb_lookup_leaf(struct rb_root
*tree
, int prio
,
835 struct rb_node
**pptr
, u32
* pid
)
839 struct rb_node
*root
;
840 struct rb_node
**pptr
;
842 } stk
[TC_HTB_MAXDEPTH
], *sp
= stk
;
844 BUG_TRAP(tree
->rb_node
);
845 sp
->root
= tree
->rb_node
;
849 for (i
= 0; i
< 65535; i
++) {
850 if (!*sp
->pptr
&& *sp
->pid
) {
851 /* ptr was invalidated but id is valid - try to recover
852 the original or next ptr */
854 htb_id_find_next_upper(prio
, sp
->root
, *sp
->pid
);
856 *sp
->pid
= 0; /* ptr is valid now so that remove this hint as it
857 can become out of date quickly */
858 if (!*sp
->pptr
) { /* we are at right end; rewind & go up */
859 *sp
->pptr
= sp
->root
;
860 while ((*sp
->pptr
)->rb_left
)
861 *sp
->pptr
= (*sp
->pptr
)->rb_left
;
867 htb_next_rb_node(sp
->pptr
);
870 struct htb_class
*cl
;
871 cl
= rb_entry(*sp
->pptr
, struct htb_class
, node
[prio
]);
874 (++sp
)->root
= cl
->un
.inner
.feed
[prio
].rb_node
;
875 sp
->pptr
= cl
->un
.inner
.ptr
+ prio
;
876 sp
->pid
= cl
->un
.inner
.last_ptr_id
+ prio
;
883 /* dequeues packet at given priority and level; call only if
884 you are sure that there is active class at prio/level */
885 static struct sk_buff
*htb_dequeue_tree(struct htb_sched
*q
, int prio
,
888 struct sk_buff
*skb
= NULL
;
889 struct htb_class
*cl
, *start
;
890 /* look initial class up in the row */
891 start
= cl
= htb_lookup_leaf(q
->row
[level
] + prio
, prio
,
892 q
->ptr
[level
] + prio
,
893 q
->last_ptr_id
[level
] + prio
);
901 /* class can be empty - it is unlikely but can be true if leaf
902 qdisc drops packets in enqueue routine or if someone used
903 graft operation on the leaf since last dequeue;
904 simply deactivate and skip such class */
905 if (unlikely(cl
->un
.leaf
.q
->q
.qlen
== 0)) {
906 struct htb_class
*next
;
907 htb_deactivate(q
, cl
);
909 /* row/level might become empty */
910 if ((q
->row_mask
[level
] & (1 << prio
)) == 0)
913 next
= htb_lookup_leaf(q
->row
[level
] + prio
,
914 prio
, q
->ptr
[level
] + prio
,
915 q
->last_ptr_id
[level
] + prio
);
917 if (cl
== start
) /* fix start if we just deleted it */
923 skb
= cl
->un
.leaf
.q
->dequeue(cl
->un
.leaf
.q
);
924 if (likely(skb
!= NULL
))
928 "htb: class %X isn't work conserving ?!\n",
933 htb_next_rb_node((level
? cl
->parent
->un
.inner
.ptr
: q
->
935 cl
= htb_lookup_leaf(q
->row
[level
] + prio
, prio
,
936 q
->ptr
[level
] + prio
,
937 q
->last_ptr_id
[level
] + prio
);
939 } while (cl
!= start
);
941 if (likely(skb
!= NULL
)) {
942 if ((cl
->un
.leaf
.deficit
[level
] -= skb
->len
) < 0) {
943 cl
->un
.leaf
.deficit
[level
] += cl
->un
.leaf
.quantum
;
944 htb_next_rb_node((level
? cl
->parent
->un
.inner
.ptr
: q
->
947 /* this used to be after charge_class but this constelation
948 gives us slightly better performance */
949 if (!cl
->un
.leaf
.q
->q
.qlen
)
950 htb_deactivate(q
, cl
);
951 htb_charge_class(q
, cl
, level
, skb
->len
);
956 static struct sk_buff
*htb_dequeue(struct Qdisc
*sch
)
958 struct sk_buff
*skb
= NULL
;
959 struct htb_sched
*q
= qdisc_priv(sch
);
961 psched_time_t next_event
;
963 /* try to dequeue direct packets as high prio (!) to minimize cpu work */
964 skb
= __skb_dequeue(&q
->direct_queue
);
966 sch
->flags
&= ~TCQ_F_THROTTLED
;
973 q
->now
= psched_get_time();
975 next_event
= q
->now
+ 5 * PSCHED_TICKS_PER_SEC
;
977 for (level
= 0; level
< TC_HTB_MAXDEPTH
; level
++) {
978 /* common case optimization - skip event handler quickly */
982 if (q
->now
>= q
->near_ev_cache
[level
]) {
983 event
= htb_do_events(q
, level
);
985 event
= q
->now
+ PSCHED_TICKS_PER_SEC
;
986 q
->near_ev_cache
[level
] = event
;
988 event
= q
->near_ev_cache
[level
];
990 if (event
&& next_event
> event
)
993 m
= ~q
->row_mask
[level
];
994 while (m
!= (int)(-1)) {
997 skb
= htb_dequeue_tree(q
, prio
, level
);
998 if (likely(skb
!= NULL
)) {
1000 sch
->flags
&= ~TCQ_F_THROTTLED
;
1005 sch
->qstats
.overlimits
++;
1006 qdisc_watchdog_schedule(&q
->watchdog
, next_event
);
1011 /* try to drop from each class (by prio) until one succeed */
1012 static unsigned int htb_drop(struct Qdisc
*sch
)
1014 struct htb_sched
*q
= qdisc_priv(sch
);
1017 for (prio
= TC_HTB_NUMPRIO
- 1; prio
>= 0; prio
--) {
1018 struct list_head
*p
;
1019 list_for_each(p
, q
->drops
+ prio
) {
1020 struct htb_class
*cl
= list_entry(p
, struct htb_class
,
1023 if (cl
->un
.leaf
.q
->ops
->drop
&&
1024 (len
= cl
->un
.leaf
.q
->ops
->drop(cl
->un
.leaf
.q
))) {
1026 if (!cl
->un
.leaf
.q
->q
.qlen
)
1027 htb_deactivate(q
, cl
);
1035 /* reset all classes */
1036 /* always caled under BH & queue lock */
1037 static void htb_reset(struct Qdisc
*sch
)
1039 struct htb_sched
*q
= qdisc_priv(sch
);
1042 for (i
= 0; i
< HTB_HSIZE
; i
++) {
1043 struct hlist_node
*p
;
1044 struct htb_class
*cl
;
1046 hlist_for_each_entry(cl
, p
, q
->hash
+ i
, hlist
) {
1048 memset(&cl
->un
.inner
, 0, sizeof(cl
->un
.inner
));
1051 qdisc_reset(cl
->un
.leaf
.q
);
1052 INIT_LIST_HEAD(&cl
->un
.leaf
.drop_list
);
1054 cl
->prio_activity
= 0;
1055 cl
->cmode
= HTB_CAN_SEND
;
1059 qdisc_watchdog_cancel(&q
->watchdog
);
1060 __skb_queue_purge(&q
->direct_queue
);
1062 memset(q
->row
, 0, sizeof(q
->row
));
1063 memset(q
->row_mask
, 0, sizeof(q
->row_mask
));
1064 memset(q
->wait_pq
, 0, sizeof(q
->wait_pq
));
1065 memset(q
->ptr
, 0, sizeof(q
->ptr
));
1066 for (i
= 0; i
< TC_HTB_NUMPRIO
; i
++)
1067 INIT_LIST_HEAD(q
->drops
+ i
);
1070 static int htb_init(struct Qdisc
*sch
, struct rtattr
*opt
)
1072 struct htb_sched
*q
= qdisc_priv(sch
);
1073 struct rtattr
*tb
[TCA_HTB_INIT
];
1074 struct tc_htb_glob
*gopt
;
1076 if (!opt
|| rtattr_parse_nested(tb
, TCA_HTB_INIT
, opt
) ||
1077 tb
[TCA_HTB_INIT
- 1] == NULL
||
1078 RTA_PAYLOAD(tb
[TCA_HTB_INIT
- 1]) < sizeof(*gopt
)) {
1079 printk(KERN_ERR
"HTB: hey probably you have bad tc tool ?\n");
1082 gopt
= RTA_DATA(tb
[TCA_HTB_INIT
- 1]);
1083 if (gopt
->version
!= HTB_VER
>> 16) {
1085 "HTB: need tc/htb version %d (minor is %d), you have %d\n",
1086 HTB_VER
>> 16, HTB_VER
& 0xffff, gopt
->version
);
1090 INIT_LIST_HEAD(&q
->root
);
1091 for (i
= 0; i
< HTB_HSIZE
; i
++)
1092 INIT_HLIST_HEAD(q
->hash
+ i
);
1093 for (i
= 0; i
< TC_HTB_NUMPRIO
; i
++)
1094 INIT_LIST_HEAD(q
->drops
+ i
);
1096 qdisc_watchdog_init(&q
->watchdog
, sch
);
1097 skb_queue_head_init(&q
->direct_queue
);
1099 q
->direct_qlen
= sch
->dev
->tx_queue_len
;
1100 if (q
->direct_qlen
< 2) /* some devices have zero tx_queue_len */
1104 init_timer(&q
->rttim
);
1105 q
->rttim
.function
= htb_rate_timer
;
1106 q
->rttim
.data
= (unsigned long)sch
;
1107 q
->rttim
.expires
= jiffies
+ HZ
;
1108 add_timer(&q
->rttim
);
1110 if ((q
->rate2quantum
= gopt
->rate2quantum
) < 1)
1111 q
->rate2quantum
= 1;
1112 q
->defcls
= gopt
->defcls
;
1117 static int htb_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
1119 struct htb_sched
*q
= qdisc_priv(sch
);
1120 unsigned char *b
= skb_tail_pointer(skb
);
1122 struct tc_htb_glob gopt
;
1123 spin_lock_bh(&sch
->dev
->queue_lock
);
1124 gopt
.direct_pkts
= q
->direct_pkts
;
1126 gopt
.version
= HTB_VER
;
1127 gopt
.rate2quantum
= q
->rate2quantum
;
1128 gopt
.defcls
= q
->defcls
;
1130 rta
= (struct rtattr
*)b
;
1131 RTA_PUT(skb
, TCA_OPTIONS
, 0, NULL
);
1132 RTA_PUT(skb
, TCA_HTB_INIT
, sizeof(gopt
), &gopt
);
1133 rta
->rta_len
= skb_tail_pointer(skb
) - b
;
1134 spin_unlock_bh(&sch
->dev
->queue_lock
);
1137 spin_unlock_bh(&sch
->dev
->queue_lock
);
1138 nlmsg_trim(skb
, skb_tail_pointer(skb
));
1142 static int htb_dump_class(struct Qdisc
*sch
, unsigned long arg
,
1143 struct sk_buff
*skb
, struct tcmsg
*tcm
)
1145 struct htb_class
*cl
= (struct htb_class
*)arg
;
1146 unsigned char *b
= skb_tail_pointer(skb
);
1148 struct tc_htb_opt opt
;
1150 spin_lock_bh(&sch
->dev
->queue_lock
);
1151 tcm
->tcm_parent
= cl
->parent
? cl
->parent
->classid
: TC_H_ROOT
;
1152 tcm
->tcm_handle
= cl
->classid
;
1153 if (!cl
->level
&& cl
->un
.leaf
.q
)
1154 tcm
->tcm_info
= cl
->un
.leaf
.q
->handle
;
1156 rta
= (struct rtattr
*)b
;
1157 RTA_PUT(skb
, TCA_OPTIONS
, 0, NULL
);
1159 memset(&opt
, 0, sizeof(opt
));
1161 opt
.rate
= cl
->rate
->rate
;
1162 opt
.buffer
= cl
->buffer
;
1163 opt
.ceil
= cl
->ceil
->rate
;
1164 opt
.cbuffer
= cl
->cbuffer
;
1165 opt
.quantum
= cl
->un
.leaf
.quantum
;
1166 opt
.prio
= cl
->un
.leaf
.prio
;
1167 opt
.level
= cl
->level
;
1168 RTA_PUT(skb
, TCA_HTB_PARMS
, sizeof(opt
), &opt
);
1169 rta
->rta_len
= skb_tail_pointer(skb
) - b
;
1170 spin_unlock_bh(&sch
->dev
->queue_lock
);
1173 spin_unlock_bh(&sch
->dev
->queue_lock
);
1179 htb_dump_class_stats(struct Qdisc
*sch
, unsigned long arg
, struct gnet_dump
*d
)
1181 struct htb_class
*cl
= (struct htb_class
*)arg
;
1184 cl
->rate_est
.bps
= cl
->rate_bytes
/ (HTB_EWMAC
* HTB_HSIZE
);
1185 cl
->rate_est
.pps
= cl
->rate_packets
/ (HTB_EWMAC
* HTB_HSIZE
);
1188 if (!cl
->level
&& cl
->un
.leaf
.q
)
1189 cl
->qstats
.qlen
= cl
->un
.leaf
.q
->q
.qlen
;
1190 cl
->xstats
.tokens
= cl
->tokens
;
1191 cl
->xstats
.ctokens
= cl
->ctokens
;
1193 if (gnet_stats_copy_basic(d
, &cl
->bstats
) < 0 ||
1194 gnet_stats_copy_rate_est(d
, &cl
->rate_est
) < 0 ||
1195 gnet_stats_copy_queue(d
, &cl
->qstats
) < 0)
1198 return gnet_stats_copy_app(d
, &cl
->xstats
, sizeof(cl
->xstats
));
1201 static int htb_graft(struct Qdisc
*sch
, unsigned long arg
, struct Qdisc
*new,
1204 struct htb_class
*cl
= (struct htb_class
*)arg
;
1206 if (cl
&& !cl
->level
) {
1208 (new = qdisc_create_dflt(sch
->dev
, &pfifo_qdisc_ops
,
1213 if ((*old
= xchg(&cl
->un
.leaf
.q
, new)) != NULL
) {
1214 qdisc_tree_decrease_qlen(*old
, (*old
)->q
.qlen
);
1217 sch_tree_unlock(sch
);
1223 static struct Qdisc
*htb_leaf(struct Qdisc
*sch
, unsigned long arg
)
1225 struct htb_class
*cl
= (struct htb_class
*)arg
;
1226 return (cl
&& !cl
->level
) ? cl
->un
.leaf
.q
: NULL
;
1229 static void htb_qlen_notify(struct Qdisc
*sch
, unsigned long arg
)
1231 struct htb_class
*cl
= (struct htb_class
*)arg
;
1233 if (cl
->un
.leaf
.q
->q
.qlen
== 0)
1234 htb_deactivate(qdisc_priv(sch
), cl
);
1237 static unsigned long htb_get(struct Qdisc
*sch
, u32 classid
)
1239 struct htb_class
*cl
= htb_find(classid
, sch
);
1242 return (unsigned long)cl
;
1245 static inline int htb_parent_last_child(struct htb_class
*cl
)
1248 /* the root class */
1251 if (!(cl
->parent
->children
.next
== &cl
->sibling
&&
1252 cl
->parent
->children
.prev
== &cl
->sibling
))
1253 /* not the last child */
1259 static void htb_parent_to_leaf(struct htb_sched
*q
, struct htb_class
*cl
,
1260 struct Qdisc
*new_q
)
1262 struct htb_class
*parent
= cl
->parent
;
1264 BUG_TRAP(!cl
->level
&& cl
->un
.leaf
.q
&& !cl
->prio_activity
);
1266 if (parent
->cmode
!= HTB_CAN_SEND
)
1267 htb_safe_rb_erase(&parent
->pq_node
, q
->wait_pq
+ parent
->level
);
1270 memset(&parent
->un
.inner
, 0, sizeof(parent
->un
.inner
));
1271 INIT_LIST_HEAD(&parent
->un
.leaf
.drop_list
);
1272 parent
->un
.leaf
.q
= new_q
? new_q
: &noop_qdisc
;
1273 parent
->un
.leaf
.quantum
= parent
->quantum
;
1274 parent
->un
.leaf
.prio
= parent
->prio
;
1275 parent
->tokens
= parent
->buffer
;
1276 parent
->ctokens
= parent
->cbuffer
;
1277 parent
->t_c
= psched_get_time();
1278 parent
->cmode
= HTB_CAN_SEND
;
1281 static void htb_destroy_class(struct Qdisc
*sch
, struct htb_class
*cl
)
1283 struct htb_sched
*q
= qdisc_priv(sch
);
1286 BUG_TRAP(cl
->un
.leaf
.q
);
1287 qdisc_destroy(cl
->un
.leaf
.q
);
1289 qdisc_put_rtab(cl
->rate
);
1290 qdisc_put_rtab(cl
->ceil
);
1292 tcf_destroy_chain(cl
->filter_list
);
1294 while (!list_empty(&cl
->children
))
1295 htb_destroy_class(sch
, list_entry(cl
->children
.next
,
1296 struct htb_class
, sibling
));
1298 /* note: this delete may happen twice (see htb_delete) */
1299 hlist_del_init(&cl
->hlist
);
1300 list_del(&cl
->sibling
);
1302 if (cl
->prio_activity
)
1303 htb_deactivate(q
, cl
);
1305 if (cl
->cmode
!= HTB_CAN_SEND
)
1306 htb_safe_rb_erase(&cl
->pq_node
, q
->wait_pq
+ cl
->level
);
1311 /* always caled under BH & queue lock */
1312 static void htb_destroy(struct Qdisc
*sch
)
1314 struct htb_sched
*q
= qdisc_priv(sch
);
1316 qdisc_watchdog_cancel(&q
->watchdog
);
1318 del_timer_sync(&q
->rttim
);
1320 /* This line used to be after htb_destroy_class call below
1321 and surprisingly it worked in 2.4. But it must precede it
1322 because filter need its target class alive to be able to call
1323 unbind_filter on it (without Oops). */
1324 tcf_destroy_chain(q
->filter_list
);
1326 while (!list_empty(&q
->root
))
1327 htb_destroy_class(sch
, list_entry(q
->root
.next
,
1328 struct htb_class
, sibling
));
1330 __skb_queue_purge(&q
->direct_queue
);
1333 static int htb_delete(struct Qdisc
*sch
, unsigned long arg
)
1335 struct htb_sched
*q
= qdisc_priv(sch
);
1336 struct htb_class
*cl
= (struct htb_class
*)arg
;
1338 struct Qdisc
*new_q
= NULL
;
1341 // TODO: why don't allow to delete subtree ? references ? does
1342 // tc subsys quarantee us that in htb_destroy it holds no class
1343 // refs so that we can remove children safely there ?
1344 if (!list_empty(&cl
->children
) || cl
->filter_cnt
)
1347 if (!cl
->level
&& htb_parent_last_child(cl
)) {
1348 new_q
= qdisc_create_dflt(sch
->dev
, &pfifo_qdisc_ops
,
1349 cl
->parent
->classid
);
1356 qlen
= cl
->un
.leaf
.q
->q
.qlen
;
1357 qdisc_reset(cl
->un
.leaf
.q
);
1358 qdisc_tree_decrease_qlen(cl
->un
.leaf
.q
, qlen
);
1361 /* delete from hash and active; remainder in destroy_class */
1362 hlist_del_init(&cl
->hlist
);
1364 if (cl
->prio_activity
)
1365 htb_deactivate(q
, cl
);
1368 htb_parent_to_leaf(q
, cl
, new_q
);
1370 if (--cl
->refcnt
== 0)
1371 htb_destroy_class(sch
, cl
);
1373 sch_tree_unlock(sch
);
1377 static void htb_put(struct Qdisc
*sch
, unsigned long arg
)
1379 struct htb_class
*cl
= (struct htb_class
*)arg
;
1381 if (--cl
->refcnt
== 0)
1382 htb_destroy_class(sch
, cl
);
1385 static int htb_change_class(struct Qdisc
*sch
, u32 classid
,
1386 u32 parentid
, struct rtattr
**tca
,
1390 struct htb_sched
*q
= qdisc_priv(sch
);
1391 struct htb_class
*cl
= (struct htb_class
*)*arg
, *parent
;
1392 struct rtattr
*opt
= tca
[TCA_OPTIONS
- 1];
1393 struct qdisc_rate_table
*rtab
= NULL
, *ctab
= NULL
;
1394 struct rtattr
*tb
[TCA_HTB_RTAB
];
1395 struct tc_htb_opt
*hopt
;
1397 /* extract all subattrs from opt attr */
1398 if (!opt
|| rtattr_parse_nested(tb
, TCA_HTB_RTAB
, opt
) ||
1399 tb
[TCA_HTB_PARMS
- 1] == NULL
||
1400 RTA_PAYLOAD(tb
[TCA_HTB_PARMS
- 1]) < sizeof(*hopt
))
1403 parent
= parentid
== TC_H_ROOT
? NULL
: htb_find(parentid
, sch
);
1405 hopt
= RTA_DATA(tb
[TCA_HTB_PARMS
- 1]);
1407 rtab
= qdisc_get_rtab(&hopt
->rate
, tb
[TCA_HTB_RTAB
- 1]);
1408 ctab
= qdisc_get_rtab(&hopt
->ceil
, tb
[TCA_HTB_CTAB
- 1]);
1412 if (!cl
) { /* new class */
1413 struct Qdisc
*new_q
;
1416 /* check for valid classid */
1417 if (!classid
|| TC_H_MAJ(classid
^ sch
->handle
)
1418 || htb_find(classid
, sch
))
1421 /* check maximal depth */
1422 if (parent
&& parent
->parent
&& parent
->parent
->level
< 2) {
1423 printk(KERN_ERR
"htb: tree is too deep\n");
1427 if ((cl
= kzalloc(sizeof(*cl
), GFP_KERNEL
)) == NULL
)
1431 INIT_LIST_HEAD(&cl
->sibling
);
1432 INIT_HLIST_NODE(&cl
->hlist
);
1433 INIT_LIST_HEAD(&cl
->children
);
1434 INIT_LIST_HEAD(&cl
->un
.leaf
.drop_list
);
1435 RB_CLEAR_NODE(&cl
->pq_node
);
1437 for (prio
= 0; prio
< TC_HTB_NUMPRIO
; prio
++)
1438 RB_CLEAR_NODE(&cl
->node
[prio
]);
1440 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL)
1441 so that can't be used inside of sch_tree_lock
1442 -- thanks to Karlis Peisenieks */
1443 new_q
= qdisc_create_dflt(sch
->dev
, &pfifo_qdisc_ops
, classid
);
1445 if (parent
&& !parent
->level
) {
1446 unsigned int qlen
= parent
->un
.leaf
.q
->q
.qlen
;
1448 /* turn parent into inner node */
1449 qdisc_reset(parent
->un
.leaf
.q
);
1450 qdisc_tree_decrease_qlen(parent
->un
.leaf
.q
, qlen
);
1451 qdisc_destroy(parent
->un
.leaf
.q
);
1452 if (parent
->prio_activity
)
1453 htb_deactivate(q
, parent
);
1455 /* remove from evt list because of level change */
1456 if (parent
->cmode
!= HTB_CAN_SEND
) {
1457 htb_safe_rb_erase(&parent
->pq_node
, q
->wait_pq
);
1458 parent
->cmode
= HTB_CAN_SEND
;
1460 parent
->level
= (parent
->parent
? parent
->parent
->level
1461 : TC_HTB_MAXDEPTH
) - 1;
1462 memset(&parent
->un
.inner
, 0, sizeof(parent
->un
.inner
));
1464 /* leaf (we) needs elementary qdisc */
1465 cl
->un
.leaf
.q
= new_q
? new_q
: &noop_qdisc
;
1467 cl
->classid
= classid
;
1468 cl
->parent
= parent
;
1470 /* set class to be in HTB_CAN_SEND state */
1471 cl
->tokens
= hopt
->buffer
;
1472 cl
->ctokens
= hopt
->cbuffer
;
1473 cl
->mbuffer
= 60 * PSCHED_TICKS_PER_SEC
; /* 1min */
1474 cl
->t_c
= psched_get_time();
1475 cl
->cmode
= HTB_CAN_SEND
;
1477 /* attach to the hash list and parent's family */
1478 hlist_add_head(&cl
->hlist
, q
->hash
+ htb_hash(classid
));
1479 list_add_tail(&cl
->sibling
,
1480 parent
? &parent
->children
: &q
->root
);
1484 /* it used to be a nasty bug here, we have to check that node
1485 is really leaf before changing cl->un.leaf ! */
1487 cl
->un
.leaf
.quantum
= rtab
->rate
.rate
/ q
->rate2quantum
;
1488 if (!hopt
->quantum
&& cl
->un
.leaf
.quantum
< 1000) {
1490 "HTB: quantum of class %X is small. Consider r2q change.\n",
1492 cl
->un
.leaf
.quantum
= 1000;
1494 if (!hopt
->quantum
&& cl
->un
.leaf
.quantum
> 200000) {
1496 "HTB: quantum of class %X is big. Consider r2q change.\n",
1498 cl
->un
.leaf
.quantum
= 200000;
1501 cl
->un
.leaf
.quantum
= hopt
->quantum
;
1502 if ((cl
->un
.leaf
.prio
= hopt
->prio
) >= TC_HTB_NUMPRIO
)
1503 cl
->un
.leaf
.prio
= TC_HTB_NUMPRIO
- 1;
1505 /* backup for htb_parent_to_leaf */
1506 cl
->quantum
= cl
->un
.leaf
.quantum
;
1507 cl
->prio
= cl
->un
.leaf
.prio
;
1510 cl
->buffer
= hopt
->buffer
;
1511 cl
->cbuffer
= hopt
->cbuffer
;
1513 qdisc_put_rtab(cl
->rate
);
1516 qdisc_put_rtab(cl
->ceil
);
1518 sch_tree_unlock(sch
);
1520 *arg
= (unsigned long)cl
;
1525 qdisc_put_rtab(rtab
);
1527 qdisc_put_rtab(ctab
);
1531 static struct tcf_proto
**htb_find_tcf(struct Qdisc
*sch
, unsigned long arg
)
1533 struct htb_sched
*q
= qdisc_priv(sch
);
1534 struct htb_class
*cl
= (struct htb_class
*)arg
;
1535 struct tcf_proto
**fl
= cl
? &cl
->filter_list
: &q
->filter_list
;
1540 static unsigned long htb_bind_filter(struct Qdisc
*sch
, unsigned long parent
,
1543 struct htb_sched
*q
= qdisc_priv(sch
);
1544 struct htb_class
*cl
= htb_find(classid
, sch
);
1546 /*if (cl && !cl->level) return 0;
1547 The line above used to be there to prevent attaching filters to
1548 leaves. But at least tc_index filter uses this just to get class
1549 for other reasons so that we have to allow for it.
1551 19.6.2002 As Werner explained it is ok - bind filter is just
1552 another way to "lock" the class - unlike "get" this lock can
1553 be broken by class during destroy IIUC.
1559 return (unsigned long)cl
;
1562 static void htb_unbind_filter(struct Qdisc
*sch
, unsigned long arg
)
1564 struct htb_sched
*q
= qdisc_priv(sch
);
1565 struct htb_class
*cl
= (struct htb_class
*)arg
;
1573 static void htb_walk(struct Qdisc
*sch
, struct qdisc_walker
*arg
)
1575 struct htb_sched
*q
= qdisc_priv(sch
);
1581 for (i
= 0; i
< HTB_HSIZE
; i
++) {
1582 struct hlist_node
*p
;
1583 struct htb_class
*cl
;
1585 hlist_for_each_entry(cl
, p
, q
->hash
+ i
, hlist
) {
1586 if (arg
->count
< arg
->skip
) {
1590 if (arg
->fn(sch
, (unsigned long)cl
, arg
) < 0) {
1599 static struct Qdisc_class_ops htb_class_ops
= {
1602 .qlen_notify
= htb_qlen_notify
,
1605 .change
= htb_change_class
,
1606 .delete = htb_delete
,
1608 .tcf_chain
= htb_find_tcf
,
1609 .bind_tcf
= htb_bind_filter
,
1610 .unbind_tcf
= htb_unbind_filter
,
1611 .dump
= htb_dump_class
,
1612 .dump_stats
= htb_dump_class_stats
,
1615 static struct Qdisc_ops htb_qdisc_ops
= {
1617 .cl_ops
= &htb_class_ops
,
1619 .priv_size
= sizeof(struct htb_sched
),
1620 .enqueue
= htb_enqueue
,
1621 .dequeue
= htb_dequeue
,
1622 .requeue
= htb_requeue
,
1626 .destroy
= htb_destroy
,
1627 .change
= NULL
/* htb_change */,
1629 .owner
= THIS_MODULE
,
1632 static int __init
htb_module_init(void)
1634 return register_qdisc(&htb_qdisc_ops
);
1636 static void __exit
htb_module_exit(void)
1638 unregister_qdisc(&htb_qdisc_ops
);
1641 module_init(htb_module_init
)
1642 module_exit(htb_module_exit
)
1643 MODULE_LICENSE("GPL");