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/moduleparam.h>
32 #include <linux/types.h>
33 #include <linux/kernel.h>
34 #include <linux/string.h>
35 #include <linux/errno.h>
36 #include <linux/skbuff.h>
37 #include <linux/list.h>
38 #include <linux/compiler.h>
39 #include <linux/rbtree.h>
40 #include <net/netlink.h>
41 #include <net/pkt_sched.h>
45 ========================================================================
46 HTB is like TBF with multiple classes. It is also similar to CBQ because
47 it allows to assign priority to each class in hierarchy.
48 In fact it is another implementation of Floyd's formal sharing.
51 Each class is assigned level. Leaf has ALWAYS level 0 and root
52 classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level
53 one less than their parent.
56 #define HTB_HSIZE 16 /* classid hash size */
57 static int htb_hysteresis __read_mostly
= 0; /* whether to use mode hysteresis for speedup */
58 #define HTB_VER 0x30011 /* major must be matched with number suplied by TC as version */
60 #if HTB_VER >> 16 != TC_HTB_PROTOVER
61 #error "Mismatched sch_htb.c and pkt_sch.h"
64 /* Module parameter and sysfs export */
65 module_param (htb_hysteresis
, int, 0640);
66 MODULE_PARM_DESC(htb_hysteresis
, "Hysteresis mode, less CPU load, less accurate");
68 /* used internaly to keep status of single class */
70 HTB_CANT_SEND
, /* class can't send and can't borrow */
71 HTB_MAY_BORROW
, /* class can't send but may borrow */
72 HTB_CAN_SEND
/* class can send */
75 /* interior & leaf nodes; props specific to leaves are marked L: */
77 /* general class parameters */
79 struct gnet_stats_basic bstats
;
80 struct gnet_stats_queue qstats
;
81 struct gnet_stats_rate_est rate_est
;
82 struct tc_htb_xstats xstats
; /* our special stats */
83 int refcnt
; /* usage count of this class */
86 int level
; /* our level (see above) */
87 struct htb_class
*parent
; /* parent class */
88 struct hlist_node hlist
; /* classid hash list item */
89 struct list_head sibling
; /* sibling list item */
90 struct list_head children
; /* children list */
93 struct htb_class_leaf
{
98 int deficit
[TC_HTB_MAXDEPTH
];
99 struct list_head drop_list
;
101 struct htb_class_inner
{
102 struct rb_root feed
[TC_HTB_NUMPRIO
]; /* feed trees */
103 struct rb_node
*ptr
[TC_HTB_NUMPRIO
]; /* current class ptr */
104 /* When class changes from state 1->2 and disconnects from
105 parent's feed then we lost ptr value and start from the
106 first child again. Here we store classid of the
107 last valid ptr (used when ptr is NULL). */
108 u32 last_ptr_id
[TC_HTB_NUMPRIO
];
111 struct rb_node node
[TC_HTB_NUMPRIO
]; /* node for self or feed tree */
112 struct rb_node pq_node
; /* node for event queue */
113 psched_time_t pq_key
;
115 int prio_activity
; /* for which prios are we active */
116 enum htb_cmode cmode
; /* current mode of the class */
118 /* class attached filters */
119 struct tcf_proto
*filter_list
;
122 int warned
; /* only one warning about non work conserving .. */
124 /* token bucket parameters */
125 struct qdisc_rate_table
*rate
; /* rate table of the class itself */
126 struct qdisc_rate_table
*ceil
; /* ceiling rate (limits borrows too) */
127 long buffer
, cbuffer
; /* token bucket depth/rate */
128 psched_tdiff_t mbuffer
; /* max wait time */
129 long tokens
, ctokens
; /* current number of tokens */
130 psched_time_t t_c
; /* checkpoint time */
132 int prio
; /* For parent to leaf return possible here */
133 int quantum
; /* we do backup. Finally full replacement */
134 /* of un.leaf originals should be done. */
137 static inline long L2T(struct htb_class
*cl
, struct qdisc_rate_table
*rate
,
140 long result
= qdisc_l2t(rate
, size
);
145 struct list_head root
; /* root classes list */
146 struct hlist_head hash
[HTB_HSIZE
]; /* hashed by classid */
147 struct list_head drops
[TC_HTB_NUMPRIO
];/* active leaves (for drops) */
149 /* self list - roots of self generating tree */
150 struct rb_root row
[TC_HTB_MAXDEPTH
][TC_HTB_NUMPRIO
];
151 int row_mask
[TC_HTB_MAXDEPTH
];
152 struct rb_node
*ptr
[TC_HTB_MAXDEPTH
][TC_HTB_NUMPRIO
];
153 u32 last_ptr_id
[TC_HTB_MAXDEPTH
][TC_HTB_NUMPRIO
];
155 /* self wait list - roots of wait PQs per row */
156 struct rb_root wait_pq
[TC_HTB_MAXDEPTH
];
158 /* time of nearest event per level (row) */
159 psched_time_t near_ev_cache
[TC_HTB_MAXDEPTH
];
161 /* whether we hit non-work conserving class during this dequeue; we use */
162 int nwc_hit
; /* this to disable mindelay complaint in dequeue */
164 int defcls
; /* class where unclassified flows go to */
166 /* filters for qdisc itself */
167 struct tcf_proto
*filter_list
;
170 int rate2quantum
; /* quant = rate / rate2quantum */
171 psched_time_t now
; /* cached dequeue time */
172 struct qdisc_watchdog watchdog
;
174 /* non shaped skbs; let them go directly thru */
175 struct sk_buff_head direct_queue
;
176 int direct_qlen
; /* max qlen of above */
181 /* compute hash of size HTB_HSIZE for given handle */
182 static inline int htb_hash(u32 h
)
185 #error "Declare new hash for your HTB_HSIZE"
187 h
^= h
>> 8; /* stolen from cbq_hash */
192 /* find class in global hash table using given handle */
193 static inline struct htb_class
*htb_find(u32 handle
, struct Qdisc
*sch
)
195 struct htb_sched
*q
= qdisc_priv(sch
);
196 struct hlist_node
*p
;
197 struct htb_class
*cl
;
199 if (TC_H_MAJ(handle
) != sch
->handle
)
202 hlist_for_each_entry(cl
, p
, q
->hash
+ htb_hash(handle
), hlist
) {
203 if (cl
->classid
== handle
)
210 * htb_classify - classify a packet into class
212 * It returns NULL if the packet should be dropped or -1 if the packet
213 * should be passed directly thru. In all other cases leaf class is returned.
214 * We allow direct class selection by classid in priority. The we examine
215 * filters in qdisc and in inner nodes (if higher filter points to the inner
216 * node). If we end up with classid MAJOR:0 we enqueue the skb into special
217 * internal fifo (direct). These packets then go directly thru. If we still
218 * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessfull
219 * then finish and return direct queue.
221 #define HTB_DIRECT (struct htb_class*)-1
223 static struct htb_class
*htb_classify(struct sk_buff
*skb
, struct Qdisc
*sch
,
226 struct htb_sched
*q
= qdisc_priv(sch
);
227 struct htb_class
*cl
;
228 struct tcf_result res
;
229 struct tcf_proto
*tcf
;
232 /* allow to select class by setting skb->priority to valid classid;
233 note that nfmark can be used too by attaching filter fw with no
235 if (skb
->priority
== sch
->handle
)
236 return HTB_DIRECT
; /* X:0 (direct flow) selected */
237 if ((cl
= htb_find(skb
->priority
, sch
)) != NULL
&& cl
->level
== 0)
240 *qerr
= NET_XMIT_BYPASS
;
241 tcf
= q
->filter_list
;
242 while (tcf
&& (result
= tc_classify(skb
, tcf
, &res
)) >= 0) {
243 #ifdef CONFIG_NET_CLS_ACT
247 *qerr
= NET_XMIT_SUCCESS
;
252 if ((cl
= (void *)res
.class) == NULL
) {
253 if (res
.classid
== sch
->handle
)
254 return HTB_DIRECT
; /* X:0 (direct flow) */
255 if ((cl
= htb_find(res
.classid
, sch
)) == NULL
)
256 break; /* filter selected invalid classid */
259 return cl
; /* we hit leaf; return it */
261 /* we have got inner class; apply inner filter chain */
262 tcf
= cl
->filter_list
;
264 /* classification failed; try to use default class */
265 cl
= htb_find(TC_H_MAKE(TC_H_MAJ(sch
->handle
), q
->defcls
), sch
);
266 if (!cl
|| cl
->level
)
267 return HTB_DIRECT
; /* bad default .. this is safe bet */
272 * htb_add_to_id_tree - adds class to the round robin list
274 * Routine adds class to the list (actually tree) sorted by classid.
275 * Make sure that class is not already on such list for given prio.
277 static void htb_add_to_id_tree(struct rb_root
*root
,
278 struct htb_class
*cl
, int prio
)
280 struct rb_node
**p
= &root
->rb_node
, *parent
= NULL
;
285 c
= rb_entry(parent
, struct htb_class
, node
[prio
]);
287 if (cl
->classid
> c
->classid
)
288 p
= &parent
->rb_right
;
290 p
= &parent
->rb_left
;
292 rb_link_node(&cl
->node
[prio
], parent
, p
);
293 rb_insert_color(&cl
->node
[prio
], root
);
297 * htb_add_to_wait_tree - adds class to the event queue with delay
299 * The class is added to priority event queue to indicate that class will
300 * change its mode in cl->pq_key microseconds. Make sure that class is not
301 * already in the queue.
303 static void htb_add_to_wait_tree(struct htb_sched
*q
,
304 struct htb_class
*cl
, long delay
)
306 struct rb_node
**p
= &q
->wait_pq
[cl
->level
].rb_node
, *parent
= NULL
;
308 cl
->pq_key
= q
->now
+ delay
;
309 if (cl
->pq_key
== q
->now
)
312 /* update the nearest event cache */
313 if (q
->near_ev_cache
[cl
->level
] > cl
->pq_key
)
314 q
->near_ev_cache
[cl
->level
] = cl
->pq_key
;
319 c
= rb_entry(parent
, struct htb_class
, pq_node
);
320 if (cl
->pq_key
>= c
->pq_key
)
321 p
= &parent
->rb_right
;
323 p
= &parent
->rb_left
;
325 rb_link_node(&cl
->pq_node
, parent
, p
);
326 rb_insert_color(&cl
->pq_node
, &q
->wait_pq
[cl
->level
]);
330 * htb_next_rb_node - finds next node in binary tree
332 * When we are past last key we return NULL.
333 * Average complexity is 2 steps per call.
335 static inline void htb_next_rb_node(struct rb_node
**n
)
341 * htb_add_class_to_row - add class to its row
343 * The class is added to row at priorities marked in mask.
344 * It does nothing if mask == 0.
346 static inline void htb_add_class_to_row(struct htb_sched
*q
,
347 struct htb_class
*cl
, int mask
)
349 q
->row_mask
[cl
->level
] |= mask
;
351 int prio
= ffz(~mask
);
352 mask
&= ~(1 << prio
);
353 htb_add_to_id_tree(q
->row
[cl
->level
] + prio
, cl
, prio
);
357 /* If this triggers, it is a bug in this code, but it need not be fatal */
358 static void htb_safe_rb_erase(struct rb_node
*rb
, struct rb_root
*root
)
360 if (RB_EMPTY_NODE(rb
)) {
370 * htb_remove_class_from_row - removes class from its row
372 * The class is removed from row at priorities marked in mask.
373 * It does nothing if mask == 0.
375 static inline void htb_remove_class_from_row(struct htb_sched
*q
,
376 struct htb_class
*cl
, int mask
)
381 int prio
= ffz(~mask
);
383 mask
&= ~(1 << prio
);
384 if (q
->ptr
[cl
->level
][prio
] == cl
->node
+ prio
)
385 htb_next_rb_node(q
->ptr
[cl
->level
] + prio
);
387 htb_safe_rb_erase(cl
->node
+ prio
, q
->row
[cl
->level
] + prio
);
388 if (!q
->row
[cl
->level
][prio
].rb_node
)
391 q
->row_mask
[cl
->level
] &= ~m
;
395 * htb_activate_prios - creates active classe's feed chain
397 * The class is connected to ancestors and/or appropriate rows
398 * for priorities it is participating on. cl->cmode must be new
399 * (activated) mode. It does nothing if cl->prio_activity == 0.
401 static void htb_activate_prios(struct htb_sched
*q
, struct htb_class
*cl
)
403 struct htb_class
*p
= cl
->parent
;
404 long m
, mask
= cl
->prio_activity
;
406 while (cl
->cmode
== HTB_MAY_BORROW
&& p
&& mask
) {
412 if (p
->un
.inner
.feed
[prio
].rb_node
)
413 /* parent already has its feed in use so that
414 reset bit in mask as parent is already ok */
415 mask
&= ~(1 << prio
);
417 htb_add_to_id_tree(p
->un
.inner
.feed
+ prio
, cl
, prio
);
419 p
->prio_activity
|= mask
;
424 if (cl
->cmode
== HTB_CAN_SEND
&& mask
)
425 htb_add_class_to_row(q
, cl
, mask
);
429 * htb_deactivate_prios - remove class from feed chain
431 * cl->cmode must represent old mode (before deactivation). It does
432 * nothing if cl->prio_activity == 0. Class is removed from all feed
435 static void htb_deactivate_prios(struct htb_sched
*q
, struct htb_class
*cl
)
437 struct htb_class
*p
= cl
->parent
;
438 long m
, mask
= cl
->prio_activity
;
440 while (cl
->cmode
== HTB_MAY_BORROW
&& p
&& mask
) {
447 if (p
->un
.inner
.ptr
[prio
] == cl
->node
+ prio
) {
448 /* we are removing child which is pointed to from
449 parent feed - forget the pointer but remember
451 p
->un
.inner
.last_ptr_id
[prio
] = cl
->classid
;
452 p
->un
.inner
.ptr
[prio
] = NULL
;
455 htb_safe_rb_erase(cl
->node
+ prio
, p
->un
.inner
.feed
+ prio
);
457 if (!p
->un
.inner
.feed
[prio
].rb_node
)
461 p
->prio_activity
&= ~mask
;
466 if (cl
->cmode
== HTB_CAN_SEND
&& mask
)
467 htb_remove_class_from_row(q
, cl
, mask
);
470 static inline long htb_lowater(const struct htb_class
*cl
)
473 return cl
->cmode
!= HTB_CANT_SEND
? -cl
->cbuffer
: 0;
477 static inline long htb_hiwater(const struct htb_class
*cl
)
480 return cl
->cmode
== HTB_CAN_SEND
? -cl
->buffer
: 0;
487 * htb_class_mode - computes and returns current class mode
489 * It computes cl's mode at time cl->t_c+diff and returns it. If mode
490 * is not HTB_CAN_SEND then cl->pq_key is updated to time difference
491 * from now to time when cl will change its state.
492 * Also it is worth to note that class mode doesn't change simply
493 * at cl->{c,}tokens == 0 but there can rather be hysteresis of
494 * 0 .. -cl->{c,}buffer range. It is meant to limit number of
495 * mode transitions per time unit. The speed gain is about 1/6.
497 static inline enum htb_cmode
498 htb_class_mode(struct htb_class
*cl
, long *diff
)
502 if ((toks
= (cl
->ctokens
+ *diff
)) < htb_lowater(cl
)) {
504 return HTB_CANT_SEND
;
507 if ((toks
= (cl
->tokens
+ *diff
)) >= htb_hiwater(cl
))
511 return HTB_MAY_BORROW
;
515 * htb_change_class_mode - changes classe's mode
517 * This should be the only way how to change classe's mode under normal
518 * cirsumstances. Routine will update feed lists linkage, change mode
519 * and add class to the wait event queue if appropriate. New mode should
520 * be different from old one and cl->pq_key has to be valid if changing
521 * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree).
524 htb_change_class_mode(struct htb_sched
*q
, struct htb_class
*cl
, long *diff
)
526 enum htb_cmode new_mode
= htb_class_mode(cl
, diff
);
528 if (new_mode
== cl
->cmode
)
531 if (cl
->prio_activity
) { /* not necessary: speed optimization */
532 if (cl
->cmode
!= HTB_CANT_SEND
)
533 htb_deactivate_prios(q
, cl
);
534 cl
->cmode
= new_mode
;
535 if (new_mode
!= HTB_CANT_SEND
)
536 htb_activate_prios(q
, cl
);
538 cl
->cmode
= new_mode
;
542 * htb_activate - inserts leaf cl into appropriate active feeds
544 * Routine learns (new) priority of leaf and activates feed chain
545 * for the prio. It can be called on already active leaf safely.
546 * It also adds leaf into droplist.
548 static inline void htb_activate(struct htb_sched
*q
, struct htb_class
*cl
)
550 BUG_TRAP(!cl
->level
&& cl
->un
.leaf
.q
&& cl
->un
.leaf
.q
->q
.qlen
);
552 if (!cl
->prio_activity
) {
553 cl
->prio_activity
= 1 << (cl
->un
.leaf
.aprio
= cl
->un
.leaf
.prio
);
554 htb_activate_prios(q
, cl
);
555 list_add_tail(&cl
->un
.leaf
.drop_list
,
556 q
->drops
+ cl
->un
.leaf
.aprio
);
561 * htb_deactivate - remove leaf cl from active feeds
563 * Make sure that leaf is active. In the other words it can't be called
564 * with non-active leaf. It also removes class from the drop list.
566 static inline void htb_deactivate(struct htb_sched
*q
, struct htb_class
*cl
)
568 BUG_TRAP(cl
->prio_activity
);
570 htb_deactivate_prios(q
, cl
);
571 cl
->prio_activity
= 0;
572 list_del_init(&cl
->un
.leaf
.drop_list
);
575 static int htb_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
)
578 struct htb_sched
*q
= qdisc_priv(sch
);
579 struct htb_class
*cl
= htb_classify(skb
, sch
, &ret
);
581 if (cl
== HTB_DIRECT
) {
582 /* enqueue to helper queue */
583 if (q
->direct_queue
.qlen
< q
->direct_qlen
) {
584 __skb_queue_tail(&q
->direct_queue
, skb
);
589 return NET_XMIT_DROP
;
591 #ifdef CONFIG_NET_CLS_ACT
593 if (ret
== NET_XMIT_BYPASS
)
598 } else if ((ret
= cl
->un
.leaf
.q
->enqueue(skb
, cl
->un
.leaf
.q
)) !=
600 if (ret
== NET_XMIT_DROP
) {
606 cl
->bstats
.packets
+=
607 skb_is_gso(skb
)?skb_shinfo(skb
)->gso_segs
:1;
608 cl
->bstats
.bytes
+= skb
->len
;
613 sch
->bstats
.packets
+= skb_is_gso(skb
)?skb_shinfo(skb
)->gso_segs
:1;
614 sch
->bstats
.bytes
+= skb
->len
;
615 return NET_XMIT_SUCCESS
;
618 /* TODO: requeuing packet charges it to policers again !! */
619 static int htb_requeue(struct sk_buff
*skb
, struct Qdisc
*sch
)
622 struct htb_sched
*q
= qdisc_priv(sch
);
623 struct htb_class
*cl
= htb_classify(skb
, sch
, &ret
);
624 struct sk_buff
*tskb
;
626 if (cl
== HTB_DIRECT
) {
627 /* enqueue to helper queue */
628 if (q
->direct_queue
.qlen
< q
->direct_qlen
) {
629 __skb_queue_head(&q
->direct_queue
, skb
);
631 __skb_queue_head(&q
->direct_queue
, skb
);
632 tskb
= __skb_dequeue_tail(&q
->direct_queue
);
637 #ifdef CONFIG_NET_CLS_ACT
639 if (ret
== NET_XMIT_BYPASS
)
644 } else if ((ret
= cl
->un
.leaf
.q
->ops
->requeue(skb
, cl
->un
.leaf
.q
)) !=
646 if (ret
== NET_XMIT_DROP
) {
655 sch
->qstats
.requeues
++;
656 return NET_XMIT_SUCCESS
;
660 * htb_charge_class - charges amount "bytes" to leaf and ancestors
662 * Routine assumes that packet "bytes" long was dequeued from leaf cl
663 * borrowing from "level". It accounts bytes to ceil leaky bucket for
664 * leaf and all ancestors and to rate bucket for ancestors at levels
665 * "level" and higher. It also handles possible change of mode resulting
666 * from the update. Note that mode can also increase here (MAY_BORROW to
667 * CAN_SEND) because we can use more precise clock that event queue here.
668 * In such case we remove class from event queue first.
670 static void htb_charge_class(struct htb_sched
*q
, struct htb_class
*cl
,
671 int level
, struct sk_buff
*skb
)
673 int bytes
= skb
->len
;
675 enum htb_cmode old_mode
;
677 #define HTB_ACCNT(T,B,R) toks = diff + cl->T; \
678 if (toks > cl->B) toks = cl->B; \
679 toks -= L2T(cl, cl->R, bytes); \
680 if (toks <= -cl->mbuffer) toks = 1-cl->mbuffer; \
684 diff
= psched_tdiff_bounded(q
->now
, cl
->t_c
, cl
->mbuffer
);
685 if (cl
->level
>= level
) {
686 if (cl
->level
== level
)
688 HTB_ACCNT(tokens
, buffer
, rate
);
690 cl
->xstats
.borrows
++;
691 cl
->tokens
+= diff
; /* we moved t_c; update tokens */
693 HTB_ACCNT(ctokens
, cbuffer
, ceil
);
696 old_mode
= cl
->cmode
;
698 htb_change_class_mode(q
, cl
, &diff
);
699 if (old_mode
!= cl
->cmode
) {
700 if (old_mode
!= HTB_CAN_SEND
)
701 htb_safe_rb_erase(&cl
->pq_node
, q
->wait_pq
+ cl
->level
);
702 if (cl
->cmode
!= HTB_CAN_SEND
)
703 htb_add_to_wait_tree(q
, cl
, diff
);
706 /* update byte stats except for leaves which are already updated */
708 cl
->bstats
.bytes
+= bytes
;
709 cl
->bstats
.packets
+= skb_is_gso(skb
)?
710 skb_shinfo(skb
)->gso_segs
:1;
717 * htb_do_events - make mode changes to classes at the level
719 * Scans event queue for pending events and applies them. Returns time of
720 * next pending event (0 for no event in pq).
721 * Note: Applied are events whose have cl->pq_key <= q->now.
723 static psched_time_t
htb_do_events(struct htb_sched
*q
, int level
)
725 /* don't run for longer than 2 jiffies; 2 is used instead of
726 1 to simplify things when jiffy is going to be incremented
728 unsigned long stop_at
= jiffies
+ 2;
729 while (time_before(jiffies
, stop_at
)) {
730 struct htb_class
*cl
;
732 struct rb_node
*p
= rb_first(&q
->wait_pq
[level
]);
737 cl
= rb_entry(p
, struct htb_class
, pq_node
);
738 if (cl
->pq_key
> q
->now
)
741 htb_safe_rb_erase(p
, q
->wait_pq
+ level
);
742 diff
= psched_tdiff_bounded(q
->now
, cl
->t_c
, cl
->mbuffer
);
743 htb_change_class_mode(q
, cl
, &diff
);
744 if (cl
->cmode
!= HTB_CAN_SEND
)
745 htb_add_to_wait_tree(q
, cl
, diff
);
747 /* too much load - let's continue on next jiffie */
748 return q
->now
+ PSCHED_TICKS_PER_SEC
/ HZ
;
751 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL
752 is no such one exists. */
753 static struct rb_node
*htb_id_find_next_upper(int prio
, struct rb_node
*n
,
756 struct rb_node
*r
= NULL
;
758 struct htb_class
*cl
=
759 rb_entry(n
, struct htb_class
, node
[prio
]);
760 if (id
== cl
->classid
)
763 if (id
> cl
->classid
) {
774 * htb_lookup_leaf - returns next leaf class in DRR order
776 * Find leaf where current feed pointers points to.
778 static struct htb_class
*htb_lookup_leaf(struct rb_root
*tree
, int prio
,
779 struct rb_node
**pptr
, u32
* pid
)
783 struct rb_node
*root
;
784 struct rb_node
**pptr
;
786 } stk
[TC_HTB_MAXDEPTH
], *sp
= stk
;
788 BUG_TRAP(tree
->rb_node
);
789 sp
->root
= tree
->rb_node
;
793 for (i
= 0; i
< 65535; i
++) {
794 if (!*sp
->pptr
&& *sp
->pid
) {
795 /* ptr was invalidated but id is valid - try to recover
796 the original or next ptr */
798 htb_id_find_next_upper(prio
, sp
->root
, *sp
->pid
);
800 *sp
->pid
= 0; /* ptr is valid now so that remove this hint as it
801 can become out of date quickly */
802 if (!*sp
->pptr
) { /* we are at right end; rewind & go up */
803 *sp
->pptr
= sp
->root
;
804 while ((*sp
->pptr
)->rb_left
)
805 *sp
->pptr
= (*sp
->pptr
)->rb_left
;
811 htb_next_rb_node(sp
->pptr
);
814 struct htb_class
*cl
;
815 cl
= rb_entry(*sp
->pptr
, struct htb_class
, node
[prio
]);
818 (++sp
)->root
= cl
->un
.inner
.feed
[prio
].rb_node
;
819 sp
->pptr
= cl
->un
.inner
.ptr
+ prio
;
820 sp
->pid
= cl
->un
.inner
.last_ptr_id
+ prio
;
827 /* dequeues packet at given priority and level; call only if
828 you are sure that there is active class at prio/level */
829 static struct sk_buff
*htb_dequeue_tree(struct htb_sched
*q
, int prio
,
832 struct sk_buff
*skb
= NULL
;
833 struct htb_class
*cl
, *start
;
834 /* look initial class up in the row */
835 start
= cl
= htb_lookup_leaf(q
->row
[level
] + prio
, prio
,
836 q
->ptr
[level
] + prio
,
837 q
->last_ptr_id
[level
] + prio
);
845 /* class can be empty - it is unlikely but can be true if leaf
846 qdisc drops packets in enqueue routine or if someone used
847 graft operation on the leaf since last dequeue;
848 simply deactivate and skip such class */
849 if (unlikely(cl
->un
.leaf
.q
->q
.qlen
== 0)) {
850 struct htb_class
*next
;
851 htb_deactivate(q
, cl
);
853 /* row/level might become empty */
854 if ((q
->row_mask
[level
] & (1 << prio
)) == 0)
857 next
= htb_lookup_leaf(q
->row
[level
] + prio
,
858 prio
, q
->ptr
[level
] + prio
,
859 q
->last_ptr_id
[level
] + prio
);
861 if (cl
== start
) /* fix start if we just deleted it */
867 skb
= cl
->un
.leaf
.q
->dequeue(cl
->un
.leaf
.q
);
868 if (likely(skb
!= NULL
))
872 "htb: class %X isn't work conserving ?!\n",
877 htb_next_rb_node((level
? cl
->parent
->un
.inner
.ptr
: q
->
879 cl
= htb_lookup_leaf(q
->row
[level
] + prio
, prio
,
880 q
->ptr
[level
] + prio
,
881 q
->last_ptr_id
[level
] + prio
);
883 } while (cl
!= start
);
885 if (likely(skb
!= NULL
)) {
886 if ((cl
->un
.leaf
.deficit
[level
] -= skb
->len
) < 0) {
887 cl
->un
.leaf
.deficit
[level
] += cl
->un
.leaf
.quantum
;
888 htb_next_rb_node((level
? cl
->parent
->un
.inner
.ptr
: q
->
891 /* this used to be after charge_class but this constelation
892 gives us slightly better performance */
893 if (!cl
->un
.leaf
.q
->q
.qlen
)
894 htb_deactivate(q
, cl
);
895 htb_charge_class(q
, cl
, level
, skb
);
900 static struct sk_buff
*htb_dequeue(struct Qdisc
*sch
)
902 struct sk_buff
*skb
= NULL
;
903 struct htb_sched
*q
= qdisc_priv(sch
);
905 psched_time_t next_event
;
907 /* try to dequeue direct packets as high prio (!) to minimize cpu work */
908 skb
= __skb_dequeue(&q
->direct_queue
);
910 sch
->flags
&= ~TCQ_F_THROTTLED
;
917 q
->now
= psched_get_time();
919 next_event
= q
->now
+ 5 * PSCHED_TICKS_PER_SEC
;
921 for (level
= 0; level
< TC_HTB_MAXDEPTH
; level
++) {
922 /* common case optimization - skip event handler quickly */
926 if (q
->now
>= q
->near_ev_cache
[level
]) {
927 event
= htb_do_events(q
, level
);
929 event
= q
->now
+ PSCHED_TICKS_PER_SEC
;
930 q
->near_ev_cache
[level
] = event
;
932 event
= q
->near_ev_cache
[level
];
934 if (event
&& next_event
> event
)
937 m
= ~q
->row_mask
[level
];
938 while (m
!= (int)(-1)) {
941 skb
= htb_dequeue_tree(q
, prio
, level
);
942 if (likely(skb
!= NULL
)) {
944 sch
->flags
&= ~TCQ_F_THROTTLED
;
949 sch
->qstats
.overlimits
++;
950 qdisc_watchdog_schedule(&q
->watchdog
, next_event
);
955 /* try to drop from each class (by prio) until one succeed */
956 static unsigned int htb_drop(struct Qdisc
*sch
)
958 struct htb_sched
*q
= qdisc_priv(sch
);
961 for (prio
= TC_HTB_NUMPRIO
- 1; prio
>= 0; prio
--) {
963 list_for_each(p
, q
->drops
+ prio
) {
964 struct htb_class
*cl
= list_entry(p
, struct htb_class
,
967 if (cl
->un
.leaf
.q
->ops
->drop
&&
968 (len
= cl
->un
.leaf
.q
->ops
->drop(cl
->un
.leaf
.q
))) {
970 if (!cl
->un
.leaf
.q
->q
.qlen
)
971 htb_deactivate(q
, cl
);
979 /* reset all classes */
980 /* always caled under BH & queue lock */
981 static void htb_reset(struct Qdisc
*sch
)
983 struct htb_sched
*q
= qdisc_priv(sch
);
986 for (i
= 0; i
< HTB_HSIZE
; i
++) {
987 struct hlist_node
*p
;
988 struct htb_class
*cl
;
990 hlist_for_each_entry(cl
, p
, q
->hash
+ i
, hlist
) {
992 memset(&cl
->un
.inner
, 0, sizeof(cl
->un
.inner
));
995 qdisc_reset(cl
->un
.leaf
.q
);
996 INIT_LIST_HEAD(&cl
->un
.leaf
.drop_list
);
998 cl
->prio_activity
= 0;
999 cl
->cmode
= HTB_CAN_SEND
;
1003 qdisc_watchdog_cancel(&q
->watchdog
);
1004 __skb_queue_purge(&q
->direct_queue
);
1006 memset(q
->row
, 0, sizeof(q
->row
));
1007 memset(q
->row_mask
, 0, sizeof(q
->row_mask
));
1008 memset(q
->wait_pq
, 0, sizeof(q
->wait_pq
));
1009 memset(q
->ptr
, 0, sizeof(q
->ptr
));
1010 for (i
= 0; i
< TC_HTB_NUMPRIO
; i
++)
1011 INIT_LIST_HEAD(q
->drops
+ i
);
1014 static const struct nla_policy htb_policy
[TCA_HTB_MAX
+ 1] = {
1015 [TCA_HTB_PARMS
] = { .len
= sizeof(struct tc_htb_opt
) },
1016 [TCA_HTB_INIT
] = { .len
= sizeof(struct tc_htb_glob
) },
1017 [TCA_HTB_CTAB
] = { .type
= NLA_BINARY
, .len
= TC_RTAB_SIZE
},
1018 [TCA_HTB_RTAB
] = { .type
= NLA_BINARY
, .len
= TC_RTAB_SIZE
},
1021 static int htb_init(struct Qdisc
*sch
, struct nlattr
*opt
)
1023 struct htb_sched
*q
= qdisc_priv(sch
);
1024 struct nlattr
*tb
[TCA_HTB_INIT
+ 1];
1025 struct tc_htb_glob
*gopt
;
1032 err
= nla_parse_nested(tb
, TCA_HTB_INIT
, opt
, htb_policy
);
1036 if (tb
[TCA_HTB_INIT
] == NULL
) {
1037 printk(KERN_ERR
"HTB: hey probably you have bad tc tool ?\n");
1040 gopt
= nla_data(tb
[TCA_HTB_INIT
]);
1041 if (gopt
->version
!= HTB_VER
>> 16) {
1043 "HTB: need tc/htb version %d (minor is %d), you have %d\n",
1044 HTB_VER
>> 16, HTB_VER
& 0xffff, gopt
->version
);
1048 INIT_LIST_HEAD(&q
->root
);
1049 for (i
= 0; i
< HTB_HSIZE
; i
++)
1050 INIT_HLIST_HEAD(q
->hash
+ i
);
1051 for (i
= 0; i
< TC_HTB_NUMPRIO
; i
++)
1052 INIT_LIST_HEAD(q
->drops
+ i
);
1054 qdisc_watchdog_init(&q
->watchdog
, sch
);
1055 skb_queue_head_init(&q
->direct_queue
);
1057 q
->direct_qlen
= sch
->dev
->tx_queue_len
;
1058 if (q
->direct_qlen
< 2) /* some devices have zero tx_queue_len */
1061 if ((q
->rate2quantum
= gopt
->rate2quantum
) < 1)
1062 q
->rate2quantum
= 1;
1063 q
->defcls
= gopt
->defcls
;
1068 static int htb_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
1070 struct htb_sched
*q
= qdisc_priv(sch
);
1071 struct nlattr
*nest
;
1072 struct tc_htb_glob gopt
;
1074 spin_lock_bh(&sch
->dev
->queue_lock
);
1076 gopt
.direct_pkts
= q
->direct_pkts
;
1077 gopt
.version
= HTB_VER
;
1078 gopt
.rate2quantum
= q
->rate2quantum
;
1079 gopt
.defcls
= q
->defcls
;
1082 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
1084 goto nla_put_failure
;
1085 NLA_PUT(skb
, TCA_HTB_INIT
, sizeof(gopt
), &gopt
);
1086 nla_nest_end(skb
, nest
);
1088 spin_unlock_bh(&sch
->dev
->queue_lock
);
1092 spin_unlock_bh(&sch
->dev
->queue_lock
);
1093 nla_nest_cancel(skb
, nest
);
1097 static int htb_dump_class(struct Qdisc
*sch
, unsigned long arg
,
1098 struct sk_buff
*skb
, struct tcmsg
*tcm
)
1100 struct htb_class
*cl
= (struct htb_class
*)arg
;
1101 struct nlattr
*nest
;
1102 struct tc_htb_opt opt
;
1104 spin_lock_bh(&sch
->dev
->queue_lock
);
1105 tcm
->tcm_parent
= cl
->parent
? cl
->parent
->classid
: TC_H_ROOT
;
1106 tcm
->tcm_handle
= cl
->classid
;
1107 if (!cl
->level
&& cl
->un
.leaf
.q
)
1108 tcm
->tcm_info
= cl
->un
.leaf
.q
->handle
;
1110 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
1112 goto nla_put_failure
;
1114 memset(&opt
, 0, sizeof(opt
));
1116 opt
.rate
= cl
->rate
->rate
;
1117 opt
.buffer
= cl
->buffer
;
1118 opt
.ceil
= cl
->ceil
->rate
;
1119 opt
.cbuffer
= cl
->cbuffer
;
1120 opt
.quantum
= cl
->un
.leaf
.quantum
;
1121 opt
.prio
= cl
->un
.leaf
.prio
;
1122 opt
.level
= cl
->level
;
1123 NLA_PUT(skb
, TCA_HTB_PARMS
, sizeof(opt
), &opt
);
1125 nla_nest_end(skb
, nest
);
1126 spin_unlock_bh(&sch
->dev
->queue_lock
);
1130 spin_unlock_bh(&sch
->dev
->queue_lock
);
1131 nla_nest_cancel(skb
, nest
);
1136 htb_dump_class_stats(struct Qdisc
*sch
, unsigned long arg
, struct gnet_dump
*d
)
1138 struct htb_class
*cl
= (struct htb_class
*)arg
;
1140 if (!cl
->level
&& cl
->un
.leaf
.q
)
1141 cl
->qstats
.qlen
= cl
->un
.leaf
.q
->q
.qlen
;
1142 cl
->xstats
.tokens
= cl
->tokens
;
1143 cl
->xstats
.ctokens
= cl
->ctokens
;
1145 if (gnet_stats_copy_basic(d
, &cl
->bstats
) < 0 ||
1146 gnet_stats_copy_rate_est(d
, &cl
->rate_est
) < 0 ||
1147 gnet_stats_copy_queue(d
, &cl
->qstats
) < 0)
1150 return gnet_stats_copy_app(d
, &cl
->xstats
, sizeof(cl
->xstats
));
1153 static int htb_graft(struct Qdisc
*sch
, unsigned long arg
, struct Qdisc
*new,
1156 struct htb_class
*cl
= (struct htb_class
*)arg
;
1158 if (cl
&& !cl
->level
) {
1160 (new = qdisc_create_dflt(sch
->dev
, &pfifo_qdisc_ops
,
1165 if ((*old
= xchg(&cl
->un
.leaf
.q
, new)) != NULL
) {
1166 qdisc_tree_decrease_qlen(*old
, (*old
)->q
.qlen
);
1169 sch_tree_unlock(sch
);
1175 static struct Qdisc
*htb_leaf(struct Qdisc
*sch
, unsigned long arg
)
1177 struct htb_class
*cl
= (struct htb_class
*)arg
;
1178 return (cl
&& !cl
->level
) ? cl
->un
.leaf
.q
: NULL
;
1181 static void htb_qlen_notify(struct Qdisc
*sch
, unsigned long arg
)
1183 struct htb_class
*cl
= (struct htb_class
*)arg
;
1185 if (cl
->un
.leaf
.q
->q
.qlen
== 0)
1186 htb_deactivate(qdisc_priv(sch
), cl
);
1189 static unsigned long htb_get(struct Qdisc
*sch
, u32 classid
)
1191 struct htb_class
*cl
= htb_find(classid
, sch
);
1194 return (unsigned long)cl
;
1197 static inline int htb_parent_last_child(struct htb_class
*cl
)
1200 /* the root class */
1203 if (!(cl
->parent
->children
.next
== &cl
->sibling
&&
1204 cl
->parent
->children
.prev
== &cl
->sibling
))
1205 /* not the last child */
1211 static void htb_parent_to_leaf(struct htb_sched
*q
, struct htb_class
*cl
,
1212 struct Qdisc
*new_q
)
1214 struct htb_class
*parent
= cl
->parent
;
1216 BUG_TRAP(!cl
->level
&& cl
->un
.leaf
.q
&& !cl
->prio_activity
);
1218 if (parent
->cmode
!= HTB_CAN_SEND
)
1219 htb_safe_rb_erase(&parent
->pq_node
, q
->wait_pq
+ parent
->level
);
1222 memset(&parent
->un
.inner
, 0, sizeof(parent
->un
.inner
));
1223 INIT_LIST_HEAD(&parent
->un
.leaf
.drop_list
);
1224 parent
->un
.leaf
.q
= new_q
? new_q
: &noop_qdisc
;
1225 parent
->un
.leaf
.quantum
= parent
->quantum
;
1226 parent
->un
.leaf
.prio
= parent
->prio
;
1227 parent
->tokens
= parent
->buffer
;
1228 parent
->ctokens
= parent
->cbuffer
;
1229 parent
->t_c
= psched_get_time();
1230 parent
->cmode
= HTB_CAN_SEND
;
1233 static void htb_destroy_class(struct Qdisc
*sch
, struct htb_class
*cl
)
1235 struct htb_sched
*q
= qdisc_priv(sch
);
1238 BUG_TRAP(cl
->un
.leaf
.q
);
1239 qdisc_destroy(cl
->un
.leaf
.q
);
1241 gen_kill_estimator(&cl
->bstats
, &cl
->rate_est
);
1242 qdisc_put_rtab(cl
->rate
);
1243 qdisc_put_rtab(cl
->ceil
);
1245 tcf_destroy_chain(&cl
->filter_list
);
1247 while (!list_empty(&cl
->children
))
1248 htb_destroy_class(sch
, list_entry(cl
->children
.next
,
1249 struct htb_class
, sibling
));
1251 /* note: this delete may happen twice (see htb_delete) */
1252 hlist_del_init(&cl
->hlist
);
1253 list_del(&cl
->sibling
);
1255 if (cl
->prio_activity
)
1256 htb_deactivate(q
, cl
);
1258 if (cl
->cmode
!= HTB_CAN_SEND
)
1259 htb_safe_rb_erase(&cl
->pq_node
, q
->wait_pq
+ cl
->level
);
1264 /* always caled under BH & queue lock */
1265 static void htb_destroy(struct Qdisc
*sch
)
1267 struct htb_sched
*q
= qdisc_priv(sch
);
1269 qdisc_watchdog_cancel(&q
->watchdog
);
1270 /* This line used to be after htb_destroy_class call below
1271 and surprisingly it worked in 2.4. But it must precede it
1272 because filter need its target class alive to be able to call
1273 unbind_filter on it (without Oops). */
1274 tcf_destroy_chain(&q
->filter_list
);
1276 while (!list_empty(&q
->root
))
1277 htb_destroy_class(sch
, list_entry(q
->root
.next
,
1278 struct htb_class
, sibling
));
1280 __skb_queue_purge(&q
->direct_queue
);
1283 static int htb_delete(struct Qdisc
*sch
, unsigned long arg
)
1285 struct htb_sched
*q
= qdisc_priv(sch
);
1286 struct htb_class
*cl
= (struct htb_class
*)arg
;
1288 struct Qdisc
*new_q
= NULL
;
1291 // TODO: why don't allow to delete subtree ? references ? does
1292 // tc subsys quarantee us that in htb_destroy it holds no class
1293 // refs so that we can remove children safely there ?
1294 if (!list_empty(&cl
->children
) || cl
->filter_cnt
)
1297 if (!cl
->level
&& htb_parent_last_child(cl
)) {
1298 new_q
= qdisc_create_dflt(sch
->dev
, &pfifo_qdisc_ops
,
1299 cl
->parent
->classid
);
1306 qlen
= cl
->un
.leaf
.q
->q
.qlen
;
1307 qdisc_reset(cl
->un
.leaf
.q
);
1308 qdisc_tree_decrease_qlen(cl
->un
.leaf
.q
, qlen
);
1311 /* delete from hash and active; remainder in destroy_class */
1312 hlist_del_init(&cl
->hlist
);
1314 if (cl
->prio_activity
)
1315 htb_deactivate(q
, cl
);
1318 htb_parent_to_leaf(q
, cl
, new_q
);
1320 if (--cl
->refcnt
== 0)
1321 htb_destroy_class(sch
, cl
);
1323 sch_tree_unlock(sch
);
1327 static void htb_put(struct Qdisc
*sch
, unsigned long arg
)
1329 struct htb_class
*cl
= (struct htb_class
*)arg
;
1331 if (--cl
->refcnt
== 0)
1332 htb_destroy_class(sch
, cl
);
1335 static int htb_change_class(struct Qdisc
*sch
, u32 classid
,
1336 u32 parentid
, struct nlattr
**tca
,
1340 struct htb_sched
*q
= qdisc_priv(sch
);
1341 struct htb_class
*cl
= (struct htb_class
*)*arg
, *parent
;
1342 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
1343 struct qdisc_rate_table
*rtab
= NULL
, *ctab
= NULL
;
1344 struct nlattr
*tb
[TCA_HTB_RTAB
+ 1];
1345 struct tc_htb_opt
*hopt
;
1347 /* extract all subattrs from opt attr */
1351 err
= nla_parse_nested(tb
, TCA_HTB_RTAB
, opt
, htb_policy
);
1356 if (tb
[TCA_HTB_PARMS
] == NULL
)
1359 parent
= parentid
== TC_H_ROOT
? NULL
: htb_find(parentid
, sch
);
1361 hopt
= nla_data(tb
[TCA_HTB_PARMS
]);
1363 rtab
= qdisc_get_rtab(&hopt
->rate
, tb
[TCA_HTB_RTAB
]);
1364 ctab
= qdisc_get_rtab(&hopt
->ceil
, tb
[TCA_HTB_CTAB
]);
1368 if (!cl
) { /* new class */
1369 struct Qdisc
*new_q
;
1373 struct gnet_estimator opt
;
1376 .nla_len
= nla_attr_size(sizeof(est
.opt
)),
1377 .nla_type
= TCA_RATE
,
1380 /* 4s interval, 16s averaging constant */
1386 /* check for valid classid */
1387 if (!classid
|| TC_H_MAJ(classid
^ sch
->handle
)
1388 || htb_find(classid
, sch
))
1391 /* check maximal depth */
1392 if (parent
&& parent
->parent
&& parent
->parent
->level
< 2) {
1393 printk(KERN_ERR
"htb: tree is too deep\n");
1397 if ((cl
= kzalloc(sizeof(*cl
), GFP_KERNEL
)) == NULL
)
1400 gen_new_estimator(&cl
->bstats
, &cl
->rate_est
,
1401 &sch
->dev
->queue_lock
,
1402 tca
[TCA_RATE
] ? : &est
.nla
);
1404 INIT_LIST_HEAD(&cl
->sibling
);
1405 INIT_HLIST_NODE(&cl
->hlist
);
1406 INIT_LIST_HEAD(&cl
->children
);
1407 INIT_LIST_HEAD(&cl
->un
.leaf
.drop_list
);
1408 RB_CLEAR_NODE(&cl
->pq_node
);
1410 for (prio
= 0; prio
< TC_HTB_NUMPRIO
; prio
++)
1411 RB_CLEAR_NODE(&cl
->node
[prio
]);
1413 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL)
1414 so that can't be used inside of sch_tree_lock
1415 -- thanks to Karlis Peisenieks */
1416 new_q
= qdisc_create_dflt(sch
->dev
, &pfifo_qdisc_ops
, classid
);
1418 if (parent
&& !parent
->level
) {
1419 unsigned int qlen
= parent
->un
.leaf
.q
->q
.qlen
;
1421 /* turn parent into inner node */
1422 qdisc_reset(parent
->un
.leaf
.q
);
1423 qdisc_tree_decrease_qlen(parent
->un
.leaf
.q
, qlen
);
1424 qdisc_destroy(parent
->un
.leaf
.q
);
1425 if (parent
->prio_activity
)
1426 htb_deactivate(q
, parent
);
1428 /* remove from evt list because of level change */
1429 if (parent
->cmode
!= HTB_CAN_SEND
) {
1430 htb_safe_rb_erase(&parent
->pq_node
, q
->wait_pq
);
1431 parent
->cmode
= HTB_CAN_SEND
;
1433 parent
->level
= (parent
->parent
? parent
->parent
->level
1434 : TC_HTB_MAXDEPTH
) - 1;
1435 memset(&parent
->un
.inner
, 0, sizeof(parent
->un
.inner
));
1437 /* leaf (we) needs elementary qdisc */
1438 cl
->un
.leaf
.q
= new_q
? new_q
: &noop_qdisc
;
1440 cl
->classid
= classid
;
1441 cl
->parent
= parent
;
1443 /* set class to be in HTB_CAN_SEND state */
1444 cl
->tokens
= hopt
->buffer
;
1445 cl
->ctokens
= hopt
->cbuffer
;
1446 cl
->mbuffer
= 60 * PSCHED_TICKS_PER_SEC
; /* 1min */
1447 cl
->t_c
= psched_get_time();
1448 cl
->cmode
= HTB_CAN_SEND
;
1450 /* attach to the hash list and parent's family */
1451 hlist_add_head(&cl
->hlist
, q
->hash
+ htb_hash(classid
));
1452 list_add_tail(&cl
->sibling
,
1453 parent
? &parent
->children
: &q
->root
);
1456 gen_replace_estimator(&cl
->bstats
, &cl
->rate_est
,
1457 &sch
->dev
->queue_lock
,
1462 /* it used to be a nasty bug here, we have to check that node
1463 is really leaf before changing cl->un.leaf ! */
1465 cl
->un
.leaf
.quantum
= rtab
->rate
.rate
/ q
->rate2quantum
;
1466 if (!hopt
->quantum
&& cl
->un
.leaf
.quantum
< 1000) {
1468 "HTB: quantum of class %X is small. Consider r2q change.\n",
1470 cl
->un
.leaf
.quantum
= 1000;
1472 if (!hopt
->quantum
&& cl
->un
.leaf
.quantum
> 200000) {
1474 "HTB: quantum of class %X is big. Consider r2q change.\n",
1476 cl
->un
.leaf
.quantum
= 200000;
1479 cl
->un
.leaf
.quantum
= hopt
->quantum
;
1480 if ((cl
->un
.leaf
.prio
= hopt
->prio
) >= TC_HTB_NUMPRIO
)
1481 cl
->un
.leaf
.prio
= TC_HTB_NUMPRIO
- 1;
1483 /* backup for htb_parent_to_leaf */
1484 cl
->quantum
= cl
->un
.leaf
.quantum
;
1485 cl
->prio
= cl
->un
.leaf
.prio
;
1488 cl
->buffer
= hopt
->buffer
;
1489 cl
->cbuffer
= hopt
->cbuffer
;
1491 qdisc_put_rtab(cl
->rate
);
1494 qdisc_put_rtab(cl
->ceil
);
1496 sch_tree_unlock(sch
);
1498 *arg
= (unsigned long)cl
;
1503 qdisc_put_rtab(rtab
);
1505 qdisc_put_rtab(ctab
);
1509 static struct tcf_proto
**htb_find_tcf(struct Qdisc
*sch
, unsigned long arg
)
1511 struct htb_sched
*q
= qdisc_priv(sch
);
1512 struct htb_class
*cl
= (struct htb_class
*)arg
;
1513 struct tcf_proto
**fl
= cl
? &cl
->filter_list
: &q
->filter_list
;
1518 static unsigned long htb_bind_filter(struct Qdisc
*sch
, unsigned long parent
,
1521 struct htb_sched
*q
= qdisc_priv(sch
);
1522 struct htb_class
*cl
= htb_find(classid
, sch
);
1524 /*if (cl && !cl->level) return 0;
1525 The line above used to be there to prevent attaching filters to
1526 leaves. But at least tc_index filter uses this just to get class
1527 for other reasons so that we have to allow for it.
1529 19.6.2002 As Werner explained it is ok - bind filter is just
1530 another way to "lock" the class - unlike "get" this lock can
1531 be broken by class during destroy IIUC.
1537 return (unsigned long)cl
;
1540 static void htb_unbind_filter(struct Qdisc
*sch
, unsigned long arg
)
1542 struct htb_sched
*q
= qdisc_priv(sch
);
1543 struct htb_class
*cl
= (struct htb_class
*)arg
;
1551 static void htb_walk(struct Qdisc
*sch
, struct qdisc_walker
*arg
)
1553 struct htb_sched
*q
= qdisc_priv(sch
);
1559 for (i
= 0; i
< HTB_HSIZE
; i
++) {
1560 struct hlist_node
*p
;
1561 struct htb_class
*cl
;
1563 hlist_for_each_entry(cl
, p
, q
->hash
+ i
, hlist
) {
1564 if (arg
->count
< arg
->skip
) {
1568 if (arg
->fn(sch
, (unsigned long)cl
, arg
) < 0) {
1577 static const struct Qdisc_class_ops htb_class_ops
= {
1580 .qlen_notify
= htb_qlen_notify
,
1583 .change
= htb_change_class
,
1584 .delete = htb_delete
,
1586 .tcf_chain
= htb_find_tcf
,
1587 .bind_tcf
= htb_bind_filter
,
1588 .unbind_tcf
= htb_unbind_filter
,
1589 .dump
= htb_dump_class
,
1590 .dump_stats
= htb_dump_class_stats
,
1593 static struct Qdisc_ops htb_qdisc_ops __read_mostly
= {
1595 .cl_ops
= &htb_class_ops
,
1597 .priv_size
= sizeof(struct htb_sched
),
1598 .enqueue
= htb_enqueue
,
1599 .dequeue
= htb_dequeue
,
1600 .requeue
= htb_requeue
,
1604 .destroy
= htb_destroy
,
1605 .change
= NULL
/* htb_change */,
1607 .owner
= THIS_MODULE
,
1610 static int __init
htb_module_init(void)
1612 return register_qdisc(&htb_qdisc_ops
);
1614 static void __exit
htb_module_exit(void)
1616 unregister_qdisc(&htb_qdisc_ops
);
1619 module_init(htb_module_init
)
1620 module_exit(htb_module_exit
)
1621 MODULE_LICENSE("GPL");