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 #include <linux/module.h>
29 #include <linux/moduleparam.h>
30 #include <linux/types.h>
31 #include <linux/kernel.h>
32 #include <linux/string.h>
33 #include <linux/errno.h>
34 #include <linux/skbuff.h>
35 #include <linux/list.h>
36 #include <linux/compiler.h>
37 #include <linux/rbtree.h>
38 #include <linux/workqueue.h>
39 #include <linux/slab.h>
40 #include <net/netlink.h>
41 #include <net/sch_generic.h>
42 #include <net/pkt_sched.h>
46 ========================================================================
47 HTB is like TBF with multiple classes. It is also similar to CBQ because
48 it allows to assign priority to each class in hierarchy.
49 In fact it is another implementation of Floyd's formal sharing.
52 Each class is assigned level. Leaf has ALWAYS level 0 and root
53 classes have level TC_HTB_MAXDEPTH-1. Interior nodes has level
54 one less than their parent.
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 static int htb_rate_est
= 0; /* htb classes have a default rate estimator */
69 module_param(htb_rate_est
, int, 0640);
70 MODULE_PARM_DESC(htb_rate_est
, "setup a default rate estimator (4sec 16sec) for htb classes");
72 /* used internaly to keep status of single class */
74 HTB_CANT_SEND
, /* class can't send and can't borrow */
75 HTB_MAY_BORROW
, /* class can't send but may borrow */
76 HTB_CAN_SEND
/* class can send */
85 /* When class changes from state 1->2 and disconnects from
86 * parent's feed then we lost ptr value and start from the
87 * first child again. Here we store classid of the
88 * last valid ptr (used when ptr is NULL).
93 /* interior & leaf nodes; props specific to leaves are marked L:
94 * To reduce false sharing, place mostly read fields at beginning,
95 * and mostly written ones at the end.
98 struct Qdisc_class_common common
;
99 struct psched_ratecfg rate
;
100 struct psched_ratecfg ceil
;
101 s64 buffer
, cbuffer
;/* token bucket depth/rate */
102 s64 mbuffer
; /* max wait time */
103 u32 prio
; /* these two are used only by leaves... */
104 int quantum
; /* but stored for parent-to-leaf return */
106 struct tcf_proto
*filter_list
; /* class attached filters */
108 int refcnt
; /* usage count of this class */
110 int level
; /* our level (see above) */
111 unsigned int children
;
112 struct htb_class
*parent
; /* parent class */
114 struct gnet_stats_rate_est64 rate_est
;
117 * Written often fields
119 struct gnet_stats_basic_packed bstats
;
120 struct gnet_stats_queue qstats
;
121 struct tc_htb_xstats xstats
; /* our special stats */
123 /* token bucket parameters */
124 s64 tokens
, ctokens
;/* current number of tokens */
125 s64 t_c
; /* checkpoint time */
128 struct htb_class_leaf
{
129 struct list_head drop_list
;
130 int deficit
[TC_HTB_MAXDEPTH
];
133 struct htb_class_inner
{
134 struct htb_prio clprio
[TC_HTB_NUMPRIO
];
139 int prio_activity
; /* for which prios are we active */
140 enum htb_cmode cmode
; /* current mode of the class */
141 struct rb_node pq_node
; /* node for event queue */
142 struct rb_node node
[TC_HTB_NUMPRIO
]; /* node for self or feed tree */
146 struct rb_root wait_pq
;
147 struct htb_prio hprio
[TC_HTB_NUMPRIO
];
151 struct Qdisc_class_hash clhash
;
152 int defcls
; /* class where unclassified flows go to */
153 int rate2quantum
; /* quant = rate / rate2quantum */
155 /* filters for qdisc itself */
156 struct tcf_proto
*filter_list
;
158 #define HTB_WARN_TOOMANYEVENTS 0x1
159 unsigned int warned
; /* only one warning */
161 struct work_struct work
;
163 /* non shaped skbs; let them go directly thru */
164 struct sk_buff_head direct_queue
;
167 struct qdisc_watchdog watchdog
;
169 s64 now
; /* cached dequeue time */
170 struct list_head drops
[TC_HTB_NUMPRIO
];/* active leaves (for drops) */
172 /* time of nearest event per level (row) */
173 s64 near_ev_cache
[TC_HTB_MAXDEPTH
];
175 int row_mask
[TC_HTB_MAXDEPTH
];
177 struct htb_level hlevel
[TC_HTB_MAXDEPTH
];
180 /* find class in global hash table using given handle */
181 static inline struct htb_class
*htb_find(u32 handle
, struct Qdisc
*sch
)
183 struct htb_sched
*q
= qdisc_priv(sch
);
184 struct Qdisc_class_common
*clc
;
186 clc
= qdisc_class_find(&q
->clhash
, handle
);
189 return container_of(clc
, struct htb_class
, common
);
193 * htb_classify - classify a packet into class
195 * It returns NULL if the packet should be dropped or -1 if the packet
196 * should be passed directly thru. In all other cases leaf class is returned.
197 * We allow direct class selection by classid in priority. The we examine
198 * filters in qdisc and in inner nodes (if higher filter points to the inner
199 * node). If we end up with classid MAJOR:0 we enqueue the skb into special
200 * internal fifo (direct). These packets then go directly thru. If we still
201 * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessful
202 * then finish and return direct queue.
204 #define HTB_DIRECT ((struct htb_class *)-1L)
206 static struct htb_class
*htb_classify(struct sk_buff
*skb
, struct Qdisc
*sch
,
209 struct htb_sched
*q
= qdisc_priv(sch
);
210 struct htb_class
*cl
;
211 struct tcf_result res
;
212 struct tcf_proto
*tcf
;
215 /* allow to select class by setting skb->priority to valid classid;
216 * note that nfmark can be used too by attaching filter fw with no
219 if (skb
->priority
== sch
->handle
)
220 return HTB_DIRECT
; /* X:0 (direct flow) selected */
221 cl
= htb_find(skb
->priority
, sch
);
222 if (cl
&& cl
->level
== 0)
225 *qerr
= NET_XMIT_SUCCESS
| __NET_XMIT_BYPASS
;
226 tcf
= q
->filter_list
;
227 while (tcf
&& (result
= tc_classify(skb
, tcf
, &res
)) >= 0) {
228 #ifdef CONFIG_NET_CLS_ACT
232 *qerr
= NET_XMIT_SUCCESS
| __NET_XMIT_STOLEN
;
237 cl
= (void *)res
.class;
239 if (res
.classid
== sch
->handle
)
240 return HTB_DIRECT
; /* X:0 (direct flow) */
241 cl
= htb_find(res
.classid
, sch
);
243 break; /* filter selected invalid classid */
246 return cl
; /* we hit leaf; return it */
248 /* we have got inner class; apply inner filter chain */
249 tcf
= cl
->filter_list
;
251 /* classification failed; try to use default class */
252 cl
= htb_find(TC_H_MAKE(TC_H_MAJ(sch
->handle
), q
->defcls
), sch
);
253 if (!cl
|| cl
->level
)
254 return HTB_DIRECT
; /* bad default .. this is safe bet */
259 * htb_add_to_id_tree - adds class to the round robin list
261 * Routine adds class to the list (actually tree) sorted by classid.
262 * Make sure that class is not already on such list for given prio.
264 static void htb_add_to_id_tree(struct rb_root
*root
,
265 struct htb_class
*cl
, int prio
)
267 struct rb_node
**p
= &root
->rb_node
, *parent
= NULL
;
272 c
= rb_entry(parent
, struct htb_class
, node
[prio
]);
274 if (cl
->common
.classid
> c
->common
.classid
)
275 p
= &parent
->rb_right
;
277 p
= &parent
->rb_left
;
279 rb_link_node(&cl
->node
[prio
], parent
, p
);
280 rb_insert_color(&cl
->node
[prio
], root
);
284 * htb_add_to_wait_tree - adds class to the event queue with delay
286 * The class is added to priority event queue to indicate that class will
287 * change its mode in cl->pq_key microseconds. Make sure that class is not
288 * already in the queue.
290 static void htb_add_to_wait_tree(struct htb_sched
*q
,
291 struct htb_class
*cl
, s64 delay
)
293 struct rb_node
**p
= &q
->hlevel
[cl
->level
].wait_pq
.rb_node
, *parent
= NULL
;
295 cl
->pq_key
= q
->now
+ delay
;
296 if (cl
->pq_key
== q
->now
)
299 /* update the nearest event cache */
300 if (q
->near_ev_cache
[cl
->level
] > cl
->pq_key
)
301 q
->near_ev_cache
[cl
->level
] = cl
->pq_key
;
306 c
= rb_entry(parent
, struct htb_class
, pq_node
);
307 if (cl
->pq_key
>= c
->pq_key
)
308 p
= &parent
->rb_right
;
310 p
= &parent
->rb_left
;
312 rb_link_node(&cl
->pq_node
, parent
, p
);
313 rb_insert_color(&cl
->pq_node
, &q
->hlevel
[cl
->level
].wait_pq
);
317 * htb_next_rb_node - finds next node in binary tree
319 * When we are past last key we return NULL.
320 * Average complexity is 2 steps per call.
322 static inline void htb_next_rb_node(struct rb_node
**n
)
328 * htb_add_class_to_row - add class to its row
330 * The class is added to row at priorities marked in mask.
331 * It does nothing if mask == 0.
333 static inline void htb_add_class_to_row(struct htb_sched
*q
,
334 struct htb_class
*cl
, int mask
)
336 q
->row_mask
[cl
->level
] |= mask
;
338 int prio
= ffz(~mask
);
339 mask
&= ~(1 << prio
);
340 htb_add_to_id_tree(&q
->hlevel
[cl
->level
].hprio
[prio
].row
, cl
, prio
);
344 /* If this triggers, it is a bug in this code, but it need not be fatal */
345 static void htb_safe_rb_erase(struct rb_node
*rb
, struct rb_root
*root
)
347 if (RB_EMPTY_NODE(rb
)) {
357 * htb_remove_class_from_row - removes class from its row
359 * The class is removed from row at priorities marked in mask.
360 * It does nothing if mask == 0.
362 static inline void htb_remove_class_from_row(struct htb_sched
*q
,
363 struct htb_class
*cl
, int mask
)
366 struct htb_level
*hlevel
= &q
->hlevel
[cl
->level
];
369 int prio
= ffz(~mask
);
370 struct htb_prio
*hprio
= &hlevel
->hprio
[prio
];
372 mask
&= ~(1 << prio
);
373 if (hprio
->ptr
== cl
->node
+ prio
)
374 htb_next_rb_node(&hprio
->ptr
);
376 htb_safe_rb_erase(cl
->node
+ prio
, &hprio
->row
);
377 if (!hprio
->row
.rb_node
)
380 q
->row_mask
[cl
->level
] &= ~m
;
384 * htb_activate_prios - creates active classe's feed chain
386 * The class is connected to ancestors and/or appropriate rows
387 * for priorities it is participating on. cl->cmode must be new
388 * (activated) mode. It does nothing if cl->prio_activity == 0.
390 static void htb_activate_prios(struct htb_sched
*q
, struct htb_class
*cl
)
392 struct htb_class
*p
= cl
->parent
;
393 long m
, mask
= cl
->prio_activity
;
395 while (cl
->cmode
== HTB_MAY_BORROW
&& p
&& mask
) {
401 if (p
->un
.inner
.clprio
[prio
].feed
.rb_node
)
402 /* parent already has its feed in use so that
403 * reset bit in mask as parent is already ok
405 mask
&= ~(1 << prio
);
407 htb_add_to_id_tree(&p
->un
.inner
.clprio
[prio
].feed
, cl
, prio
);
409 p
->prio_activity
|= mask
;
414 if (cl
->cmode
== HTB_CAN_SEND
&& mask
)
415 htb_add_class_to_row(q
, cl
, mask
);
419 * htb_deactivate_prios - remove class from feed chain
421 * cl->cmode must represent old mode (before deactivation). It does
422 * nothing if cl->prio_activity == 0. Class is removed from all feed
425 static void htb_deactivate_prios(struct htb_sched
*q
, struct htb_class
*cl
)
427 struct htb_class
*p
= cl
->parent
;
428 long m
, mask
= cl
->prio_activity
;
430 while (cl
->cmode
== HTB_MAY_BORROW
&& p
&& mask
) {
437 if (p
->un
.inner
.clprio
[prio
].ptr
== cl
->node
+ prio
) {
438 /* we are removing child which is pointed to from
439 * parent feed - forget the pointer but remember
442 p
->un
.inner
.clprio
[prio
].last_ptr_id
= cl
->common
.classid
;
443 p
->un
.inner
.clprio
[prio
].ptr
= NULL
;
446 htb_safe_rb_erase(cl
->node
+ prio
,
447 &p
->un
.inner
.clprio
[prio
].feed
);
449 if (!p
->un
.inner
.clprio
[prio
].feed
.rb_node
)
453 p
->prio_activity
&= ~mask
;
458 if (cl
->cmode
== HTB_CAN_SEND
&& mask
)
459 htb_remove_class_from_row(q
, cl
, mask
);
462 static inline s64
htb_lowater(const struct htb_class
*cl
)
465 return cl
->cmode
!= HTB_CANT_SEND
? -cl
->cbuffer
: 0;
469 static inline s64
htb_hiwater(const struct htb_class
*cl
)
472 return cl
->cmode
== HTB_CAN_SEND
? -cl
->buffer
: 0;
479 * htb_class_mode - computes and returns current class mode
481 * It computes cl's mode at time cl->t_c+diff and returns it. If mode
482 * is not HTB_CAN_SEND then cl->pq_key is updated to time difference
483 * from now to time when cl will change its state.
484 * Also it is worth to note that class mode doesn't change simply
485 * at cl->{c,}tokens == 0 but there can rather be hysteresis of
486 * 0 .. -cl->{c,}buffer range. It is meant to limit number of
487 * mode transitions per time unit. The speed gain is about 1/6.
489 static inline enum htb_cmode
490 htb_class_mode(struct htb_class
*cl
, s64
*diff
)
494 if ((toks
= (cl
->ctokens
+ *diff
)) < htb_lowater(cl
)) {
496 return HTB_CANT_SEND
;
499 if ((toks
= (cl
->tokens
+ *diff
)) >= htb_hiwater(cl
))
503 return HTB_MAY_BORROW
;
507 * htb_change_class_mode - changes classe's mode
509 * This should be the only way how to change classe's mode under normal
510 * cirsumstances. Routine will update feed lists linkage, change mode
511 * and add class to the wait event queue if appropriate. New mode should
512 * be different from old one and cl->pq_key has to be valid if changing
513 * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree).
516 htb_change_class_mode(struct htb_sched
*q
, struct htb_class
*cl
, s64
*diff
)
518 enum htb_cmode new_mode
= htb_class_mode(cl
, diff
);
520 if (new_mode
== cl
->cmode
)
523 if (cl
->prio_activity
) { /* not necessary: speed optimization */
524 if (cl
->cmode
!= HTB_CANT_SEND
)
525 htb_deactivate_prios(q
, cl
);
526 cl
->cmode
= new_mode
;
527 if (new_mode
!= HTB_CANT_SEND
)
528 htb_activate_prios(q
, cl
);
530 cl
->cmode
= new_mode
;
534 * htb_activate - inserts leaf cl into appropriate active feeds
536 * Routine learns (new) priority of leaf and activates feed chain
537 * for the prio. It can be called on already active leaf safely.
538 * It also adds leaf into droplist.
540 static inline void htb_activate(struct htb_sched
*q
, struct htb_class
*cl
)
542 WARN_ON(cl
->level
|| !cl
->un
.leaf
.q
|| !cl
->un
.leaf
.q
->q
.qlen
);
544 if (!cl
->prio_activity
) {
545 cl
->prio_activity
= 1 << cl
->prio
;
546 htb_activate_prios(q
, cl
);
547 list_add_tail(&cl
->un
.leaf
.drop_list
,
548 q
->drops
+ cl
->prio
);
553 * htb_deactivate - remove leaf cl from active feeds
555 * Make sure that leaf is active. In the other words it can't be called
556 * with non-active leaf. It also removes class from the drop list.
558 static inline void htb_deactivate(struct htb_sched
*q
, struct htb_class
*cl
)
560 WARN_ON(!cl
->prio_activity
);
562 htb_deactivate_prios(q
, cl
);
563 cl
->prio_activity
= 0;
564 list_del_init(&cl
->un
.leaf
.drop_list
);
567 static int htb_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
)
569 int uninitialized_var(ret
);
570 struct htb_sched
*q
= qdisc_priv(sch
);
571 struct htb_class
*cl
= htb_classify(skb
, sch
, &ret
);
573 if (cl
== HTB_DIRECT
) {
574 /* enqueue to helper queue */
575 if (q
->direct_queue
.qlen
< q
->direct_qlen
) {
576 __skb_queue_tail(&q
->direct_queue
, skb
);
579 return qdisc_drop(skb
, sch
);
581 #ifdef CONFIG_NET_CLS_ACT
583 if (ret
& __NET_XMIT_BYPASS
)
588 } else if ((ret
= qdisc_enqueue(skb
, cl
->un
.leaf
.q
)) != NET_XMIT_SUCCESS
) {
589 if (net_xmit_drop_count(ret
)) {
599 return NET_XMIT_SUCCESS
;
602 static inline void htb_accnt_tokens(struct htb_class
*cl
, int bytes
, s64 diff
)
604 s64 toks
= diff
+ cl
->tokens
;
606 if (toks
> cl
->buffer
)
608 toks
-= (s64
) psched_l2t_ns(&cl
->rate
, bytes
);
609 if (toks
<= -cl
->mbuffer
)
610 toks
= 1 - cl
->mbuffer
;
615 static inline void htb_accnt_ctokens(struct htb_class
*cl
, int bytes
, s64 diff
)
617 s64 toks
= diff
+ cl
->ctokens
;
619 if (toks
> cl
->cbuffer
)
621 toks
-= (s64
) psched_l2t_ns(&cl
->ceil
, bytes
);
622 if (toks
<= -cl
->mbuffer
)
623 toks
= 1 - cl
->mbuffer
;
629 * htb_charge_class - charges amount "bytes" to leaf and ancestors
631 * Routine assumes that packet "bytes" long was dequeued from leaf cl
632 * borrowing from "level". It accounts bytes to ceil leaky bucket for
633 * leaf and all ancestors and to rate bucket for ancestors at levels
634 * "level" and higher. It also handles possible change of mode resulting
635 * from the update. Note that mode can also increase here (MAY_BORROW to
636 * CAN_SEND) because we can use more precise clock that event queue here.
637 * In such case we remove class from event queue first.
639 static void htb_charge_class(struct htb_sched
*q
, struct htb_class
*cl
,
640 int level
, struct sk_buff
*skb
)
642 int bytes
= qdisc_pkt_len(skb
);
643 enum htb_cmode old_mode
;
647 diff
= min_t(s64
, q
->now
- cl
->t_c
, cl
->mbuffer
);
648 if (cl
->level
>= level
) {
649 if (cl
->level
== level
)
651 htb_accnt_tokens(cl
, bytes
, diff
);
653 cl
->xstats
.borrows
++;
654 cl
->tokens
+= diff
; /* we moved t_c; update tokens */
656 htb_accnt_ctokens(cl
, bytes
, diff
);
659 old_mode
= cl
->cmode
;
661 htb_change_class_mode(q
, cl
, &diff
);
662 if (old_mode
!= cl
->cmode
) {
663 if (old_mode
!= HTB_CAN_SEND
)
664 htb_safe_rb_erase(&cl
->pq_node
, &q
->hlevel
[cl
->level
].wait_pq
);
665 if (cl
->cmode
!= HTB_CAN_SEND
)
666 htb_add_to_wait_tree(q
, cl
, diff
);
669 /* update basic stats except for leaves which are already updated */
671 bstats_update(&cl
->bstats
, skb
);
678 * htb_do_events - make mode changes to classes at the level
680 * Scans event queue for pending events and applies them. Returns time of
681 * next pending event (0 for no event in pq, q->now for too many events).
682 * Note: Applied are events whose have cl->pq_key <= q->now.
684 static s64
htb_do_events(struct htb_sched
*q
, const int level
,
687 /* don't run for longer than 2 jiffies; 2 is used instead of
688 * 1 to simplify things when jiffy is going to be incremented
691 unsigned long stop_at
= start
+ 2;
692 struct rb_root
*wait_pq
= &q
->hlevel
[level
].wait_pq
;
694 while (time_before(jiffies
, stop_at
)) {
695 struct htb_class
*cl
;
697 struct rb_node
*p
= rb_first(wait_pq
);
702 cl
= rb_entry(p
, struct htb_class
, pq_node
);
703 if (cl
->pq_key
> q
->now
)
706 htb_safe_rb_erase(p
, wait_pq
);
707 diff
= min_t(s64
, q
->now
- cl
->t_c
, cl
->mbuffer
);
708 htb_change_class_mode(q
, cl
, &diff
);
709 if (cl
->cmode
!= HTB_CAN_SEND
)
710 htb_add_to_wait_tree(q
, cl
, diff
);
713 /* too much load - let's continue after a break for scheduling */
714 if (!(q
->warned
& HTB_WARN_TOOMANYEVENTS
)) {
715 pr_warning("htb: too many events!\n");
716 q
->warned
|= HTB_WARN_TOOMANYEVENTS
;
722 /* Returns class->node+prio from id-tree where classe's id is >= id. NULL
723 * is no such one exists.
725 static struct rb_node
*htb_id_find_next_upper(int prio
, struct rb_node
*n
,
728 struct rb_node
*r
= NULL
;
730 struct htb_class
*cl
=
731 rb_entry(n
, struct htb_class
, node
[prio
]);
733 if (id
> cl
->common
.classid
) {
735 } else if (id
< cl
->common
.classid
) {
746 * htb_lookup_leaf - returns next leaf class in DRR order
748 * Find leaf where current feed pointers points to.
750 static struct htb_class
*htb_lookup_leaf(struct htb_prio
*hprio
, const int prio
)
754 struct rb_node
*root
;
755 struct rb_node
**pptr
;
757 } stk
[TC_HTB_MAXDEPTH
], *sp
= stk
;
759 BUG_ON(!hprio
->row
.rb_node
);
760 sp
->root
= hprio
->row
.rb_node
;
761 sp
->pptr
= &hprio
->ptr
;
762 sp
->pid
= &hprio
->last_ptr_id
;
764 for (i
= 0; i
< 65535; i
++) {
765 if (!*sp
->pptr
&& *sp
->pid
) {
766 /* ptr was invalidated but id is valid - try to recover
767 * the original or next ptr
770 htb_id_find_next_upper(prio
, sp
->root
, *sp
->pid
);
772 *sp
->pid
= 0; /* ptr is valid now so that remove this hint as it
773 * can become out of date quickly
775 if (!*sp
->pptr
) { /* we are at right end; rewind & go up */
776 *sp
->pptr
= sp
->root
;
777 while ((*sp
->pptr
)->rb_left
)
778 *sp
->pptr
= (*sp
->pptr
)->rb_left
;
785 htb_next_rb_node(sp
->pptr
);
788 struct htb_class
*cl
;
789 struct htb_prio
*clp
;
791 cl
= rb_entry(*sp
->pptr
, struct htb_class
, node
[prio
]);
794 clp
= &cl
->un
.inner
.clprio
[prio
];
795 (++sp
)->root
= clp
->feed
.rb_node
;
796 sp
->pptr
= &clp
->ptr
;
797 sp
->pid
= &clp
->last_ptr_id
;
804 /* dequeues packet at given priority and level; call only if
805 * you are sure that there is active class at prio/level
807 static struct sk_buff
*htb_dequeue_tree(struct htb_sched
*q
, const int prio
,
810 struct sk_buff
*skb
= NULL
;
811 struct htb_class
*cl
, *start
;
812 struct htb_level
*hlevel
= &q
->hlevel
[level
];
813 struct htb_prio
*hprio
= &hlevel
->hprio
[prio
];
815 /* look initial class up in the row */
816 start
= cl
= htb_lookup_leaf(hprio
, prio
);
823 /* class can be empty - it is unlikely but can be true if leaf
824 * qdisc drops packets in enqueue routine or if someone used
825 * graft operation on the leaf since last dequeue;
826 * simply deactivate and skip such class
828 if (unlikely(cl
->un
.leaf
.q
->q
.qlen
== 0)) {
829 struct htb_class
*next
;
830 htb_deactivate(q
, cl
);
832 /* row/level might become empty */
833 if ((q
->row_mask
[level
] & (1 << prio
)) == 0)
836 next
= htb_lookup_leaf(hprio
, prio
);
838 if (cl
== start
) /* fix start if we just deleted it */
844 skb
= cl
->un
.leaf
.q
->dequeue(cl
->un
.leaf
.q
);
845 if (likely(skb
!= NULL
))
848 qdisc_warn_nonwc("htb", cl
->un
.leaf
.q
);
849 htb_next_rb_node(level
? &cl
->parent
->un
.inner
.clprio
[prio
].ptr
:
850 &q
->hlevel
[0].hprio
[prio
].ptr
);
851 cl
= htb_lookup_leaf(hprio
, prio
);
853 } while (cl
!= start
);
855 if (likely(skb
!= NULL
)) {
856 bstats_update(&cl
->bstats
, skb
);
857 cl
->un
.leaf
.deficit
[level
] -= qdisc_pkt_len(skb
);
858 if (cl
->un
.leaf
.deficit
[level
] < 0) {
859 cl
->un
.leaf
.deficit
[level
] += cl
->quantum
;
860 htb_next_rb_node(level
? &cl
->parent
->un
.inner
.clprio
[prio
].ptr
:
861 &q
->hlevel
[0].hprio
[prio
].ptr
);
863 /* this used to be after charge_class but this constelation
864 * gives us slightly better performance
866 if (!cl
->un
.leaf
.q
->q
.qlen
)
867 htb_deactivate(q
, cl
);
868 htb_charge_class(q
, cl
, level
, skb
);
873 static struct sk_buff
*htb_dequeue(struct Qdisc
*sch
)
876 struct htb_sched
*q
= qdisc_priv(sch
);
879 unsigned long start_at
;
881 /* try to dequeue direct packets as high prio (!) to minimize cpu work */
882 skb
= __skb_dequeue(&q
->direct_queue
);
885 qdisc_bstats_update(sch
, skb
);
886 qdisc_unthrottled(sch
);
893 q
->now
= ktime_to_ns(ktime_get());
896 next_event
= q
->now
+ 5LLU * NSEC_PER_SEC
;
898 for (level
= 0; level
< TC_HTB_MAXDEPTH
; level
++) {
899 /* common case optimization - skip event handler quickly */
901 s64 event
= q
->near_ev_cache
[level
];
903 if (q
->now
>= event
) {
904 event
= htb_do_events(q
, level
, start_at
);
906 event
= q
->now
+ NSEC_PER_SEC
;
907 q
->near_ev_cache
[level
] = event
;
910 if (next_event
> event
)
913 m
= ~q
->row_mask
[level
];
914 while (m
!= (int)(-1)) {
918 skb
= htb_dequeue_tree(q
, prio
, level
);
919 if (likely(skb
!= NULL
))
923 sch
->qstats
.overlimits
++;
924 if (likely(next_event
> q
->now
)) {
925 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
926 &qdisc_root_sleeping(q
->watchdog
.qdisc
)->state
)) {
927 ktime_t time
= ns_to_ktime(next_event
);
928 qdisc_throttled(q
->watchdog
.qdisc
);
929 hrtimer_start(&q
->watchdog
.timer
, time
,
933 schedule_work(&q
->work
);
939 /* try to drop from each class (by prio) until one succeed */
940 static unsigned int htb_drop(struct Qdisc
*sch
)
942 struct htb_sched
*q
= qdisc_priv(sch
);
945 for (prio
= TC_HTB_NUMPRIO
- 1; prio
>= 0; prio
--) {
947 list_for_each(p
, q
->drops
+ prio
) {
948 struct htb_class
*cl
= list_entry(p
, struct htb_class
,
951 if (cl
->un
.leaf
.q
->ops
->drop
&&
952 (len
= cl
->un
.leaf
.q
->ops
->drop(cl
->un
.leaf
.q
))) {
954 if (!cl
->un
.leaf
.q
->q
.qlen
)
955 htb_deactivate(q
, cl
);
963 /* reset all classes */
964 /* always caled under BH & queue lock */
965 static void htb_reset(struct Qdisc
*sch
)
967 struct htb_sched
*q
= qdisc_priv(sch
);
968 struct htb_class
*cl
;
971 for (i
= 0; i
< q
->clhash
.hashsize
; i
++) {
972 hlist_for_each_entry(cl
, &q
->clhash
.hash
[i
], common
.hnode
) {
974 memset(&cl
->un
.inner
, 0, sizeof(cl
->un
.inner
));
977 qdisc_reset(cl
->un
.leaf
.q
);
978 INIT_LIST_HEAD(&cl
->un
.leaf
.drop_list
);
980 cl
->prio_activity
= 0;
981 cl
->cmode
= HTB_CAN_SEND
;
985 qdisc_watchdog_cancel(&q
->watchdog
);
986 __skb_queue_purge(&q
->direct_queue
);
988 memset(q
->hlevel
, 0, sizeof(q
->hlevel
));
989 memset(q
->row_mask
, 0, sizeof(q
->row_mask
));
990 for (i
= 0; i
< TC_HTB_NUMPRIO
; i
++)
991 INIT_LIST_HEAD(q
->drops
+ i
);
994 static const struct nla_policy htb_policy
[TCA_HTB_MAX
+ 1] = {
995 [TCA_HTB_PARMS
] = { .len
= sizeof(struct tc_htb_opt
) },
996 [TCA_HTB_INIT
] = { .len
= sizeof(struct tc_htb_glob
) },
997 [TCA_HTB_CTAB
] = { .type
= NLA_BINARY
, .len
= TC_RTAB_SIZE
},
998 [TCA_HTB_RTAB
] = { .type
= NLA_BINARY
, .len
= TC_RTAB_SIZE
},
999 [TCA_HTB_DIRECT_QLEN
] = { .type
= NLA_U32
},
1000 [TCA_HTB_RATE64
] = { .type
= NLA_U64
},
1001 [TCA_HTB_CEIL64
] = { .type
= NLA_U64
},
1004 static void htb_work_func(struct work_struct
*work
)
1006 struct htb_sched
*q
= container_of(work
, struct htb_sched
, work
);
1007 struct Qdisc
*sch
= q
->watchdog
.qdisc
;
1009 __netif_schedule(qdisc_root(sch
));
1012 static int htb_init(struct Qdisc
*sch
, struct nlattr
*opt
)
1014 struct htb_sched
*q
= qdisc_priv(sch
);
1015 struct nlattr
*tb
[TCA_HTB_MAX
+ 1];
1016 struct tc_htb_glob
*gopt
;
1023 err
= nla_parse_nested(tb
, TCA_HTB_MAX
, opt
, htb_policy
);
1027 if (!tb
[TCA_HTB_INIT
])
1030 gopt
= nla_data(tb
[TCA_HTB_INIT
]);
1031 if (gopt
->version
!= HTB_VER
>> 16)
1034 err
= qdisc_class_hash_init(&q
->clhash
);
1037 for (i
= 0; i
< TC_HTB_NUMPRIO
; i
++)
1038 INIT_LIST_HEAD(q
->drops
+ i
);
1040 qdisc_watchdog_init(&q
->watchdog
, sch
);
1041 INIT_WORK(&q
->work
, htb_work_func
);
1042 skb_queue_head_init(&q
->direct_queue
);
1044 if (tb
[TCA_HTB_DIRECT_QLEN
])
1045 q
->direct_qlen
= nla_get_u32(tb
[TCA_HTB_DIRECT_QLEN
]);
1047 q
->direct_qlen
= qdisc_dev(sch
)->tx_queue_len
;
1048 if (q
->direct_qlen
< 2) /* some devices have zero tx_queue_len */
1051 if ((q
->rate2quantum
= gopt
->rate2quantum
) < 1)
1052 q
->rate2quantum
= 1;
1053 q
->defcls
= gopt
->defcls
;
1058 static int htb_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
1060 spinlock_t
*root_lock
= qdisc_root_sleeping_lock(sch
);
1061 struct htb_sched
*q
= qdisc_priv(sch
);
1062 struct nlattr
*nest
;
1063 struct tc_htb_glob gopt
;
1065 spin_lock_bh(root_lock
);
1067 gopt
.direct_pkts
= q
->direct_pkts
;
1068 gopt
.version
= HTB_VER
;
1069 gopt
.rate2quantum
= q
->rate2quantum
;
1070 gopt
.defcls
= q
->defcls
;
1073 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
1075 goto nla_put_failure
;
1076 if (nla_put(skb
, TCA_HTB_INIT
, sizeof(gopt
), &gopt
) ||
1077 nla_put_u32(skb
, TCA_HTB_DIRECT_QLEN
, q
->direct_qlen
))
1078 goto nla_put_failure
;
1079 nla_nest_end(skb
, nest
);
1081 spin_unlock_bh(root_lock
);
1085 spin_unlock_bh(root_lock
);
1086 nla_nest_cancel(skb
, nest
);
1090 static int htb_dump_class(struct Qdisc
*sch
, unsigned long arg
,
1091 struct sk_buff
*skb
, struct tcmsg
*tcm
)
1093 struct htb_class
*cl
= (struct htb_class
*)arg
;
1094 spinlock_t
*root_lock
= qdisc_root_sleeping_lock(sch
);
1095 struct nlattr
*nest
;
1096 struct tc_htb_opt opt
;
1098 spin_lock_bh(root_lock
);
1099 tcm
->tcm_parent
= cl
->parent
? cl
->parent
->common
.classid
: TC_H_ROOT
;
1100 tcm
->tcm_handle
= cl
->common
.classid
;
1101 if (!cl
->level
&& cl
->un
.leaf
.q
)
1102 tcm
->tcm_info
= cl
->un
.leaf
.q
->handle
;
1104 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
1106 goto nla_put_failure
;
1108 memset(&opt
, 0, sizeof(opt
));
1110 psched_ratecfg_getrate(&opt
.rate
, &cl
->rate
);
1111 opt
.buffer
= PSCHED_NS2TICKS(cl
->buffer
);
1112 psched_ratecfg_getrate(&opt
.ceil
, &cl
->ceil
);
1113 opt
.cbuffer
= PSCHED_NS2TICKS(cl
->cbuffer
);
1114 opt
.quantum
= cl
->quantum
;
1115 opt
.prio
= cl
->prio
;
1116 opt
.level
= cl
->level
;
1117 if (nla_put(skb
, TCA_HTB_PARMS
, sizeof(opt
), &opt
))
1118 goto nla_put_failure
;
1119 if ((cl
->rate
.rate_bytes_ps
>= (1ULL << 32)) &&
1120 nla_put_u64(skb
, TCA_HTB_RATE64
, cl
->rate
.rate_bytes_ps
))
1121 goto nla_put_failure
;
1122 if ((cl
->ceil
.rate_bytes_ps
>= (1ULL << 32)) &&
1123 nla_put_u64(skb
, TCA_HTB_CEIL64
, cl
->ceil
.rate_bytes_ps
))
1124 goto nla_put_failure
;
1126 nla_nest_end(skb
, nest
);
1127 spin_unlock_bh(root_lock
);
1131 spin_unlock_bh(root_lock
);
1132 nla_nest_cancel(skb
, nest
);
1137 htb_dump_class_stats(struct Qdisc
*sch
, unsigned long arg
, struct gnet_dump
*d
)
1139 struct htb_class
*cl
= (struct htb_class
*)arg
;
1141 if (!cl
->level
&& cl
->un
.leaf
.q
)
1142 cl
->qstats
.qlen
= cl
->un
.leaf
.q
->q
.qlen
;
1143 cl
->xstats
.tokens
= PSCHED_NS2TICKS(cl
->tokens
);
1144 cl
->xstats
.ctokens
= PSCHED_NS2TICKS(cl
->ctokens
);
1146 if (gnet_stats_copy_basic(d
, &cl
->bstats
) < 0 ||
1147 gnet_stats_copy_rate_est(d
, NULL
, &cl
->rate_est
) < 0 ||
1148 gnet_stats_copy_queue(d
, &cl
->qstats
) < 0)
1151 return gnet_stats_copy_app(d
, &cl
->xstats
, sizeof(cl
->xstats
));
1154 static int htb_graft(struct Qdisc
*sch
, unsigned long arg
, struct Qdisc
*new,
1157 struct htb_class
*cl
= (struct htb_class
*)arg
;
1162 (new = qdisc_create_dflt(sch
->dev_queue
, &pfifo_qdisc_ops
,
1163 cl
->common
.classid
)) == NULL
)
1167 *old
= cl
->un
.leaf
.q
;
1168 cl
->un
.leaf
.q
= new;
1170 qdisc_tree_decrease_qlen(*old
, (*old
)->q
.qlen
);
1173 sch_tree_unlock(sch
);
1177 static struct Qdisc
*htb_leaf(struct Qdisc
*sch
, unsigned long arg
)
1179 struct htb_class
*cl
= (struct htb_class
*)arg
;
1180 return !cl
->level
? cl
->un
.leaf
.q
: NULL
;
1183 static void htb_qlen_notify(struct Qdisc
*sch
, unsigned long arg
)
1185 struct htb_class
*cl
= (struct htb_class
*)arg
;
1187 if (cl
->un
.leaf
.q
->q
.qlen
== 0)
1188 htb_deactivate(qdisc_priv(sch
), cl
);
1191 static unsigned long htb_get(struct Qdisc
*sch
, u32 classid
)
1193 struct htb_class
*cl
= htb_find(classid
, sch
);
1196 return (unsigned long)cl
;
1199 static inline int htb_parent_last_child(struct htb_class
*cl
)
1202 /* the root class */
1204 if (cl
->parent
->children
> 1)
1205 /* not the last child */
1210 static void htb_parent_to_leaf(struct htb_sched
*q
, struct htb_class
*cl
,
1211 struct Qdisc
*new_q
)
1213 struct htb_class
*parent
= cl
->parent
;
1215 WARN_ON(cl
->level
|| !cl
->un
.leaf
.q
|| cl
->prio_activity
);
1217 if (parent
->cmode
!= HTB_CAN_SEND
)
1218 htb_safe_rb_erase(&parent
->pq_node
,
1219 &q
->hlevel
[parent
->level
].wait_pq
);
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
->tokens
= parent
->buffer
;
1226 parent
->ctokens
= parent
->cbuffer
;
1227 parent
->t_c
= ktime_to_ns(ktime_get());
1228 parent
->cmode
= HTB_CAN_SEND
;
1231 static void htb_destroy_class(struct Qdisc
*sch
, struct htb_class
*cl
)
1234 WARN_ON(!cl
->un
.leaf
.q
);
1235 qdisc_destroy(cl
->un
.leaf
.q
);
1237 gen_kill_estimator(&cl
->bstats
, &cl
->rate_est
);
1238 tcf_destroy_chain(&cl
->filter_list
);
1242 static void htb_destroy(struct Qdisc
*sch
)
1244 struct htb_sched
*q
= qdisc_priv(sch
);
1245 struct hlist_node
*next
;
1246 struct htb_class
*cl
;
1249 cancel_work_sync(&q
->work
);
1250 qdisc_watchdog_cancel(&q
->watchdog
);
1251 /* This line used to be after htb_destroy_class call below
1252 * and surprisingly it worked in 2.4. But it must precede it
1253 * because filter need its target class alive to be able to call
1254 * unbind_filter on it (without Oops).
1256 tcf_destroy_chain(&q
->filter_list
);
1258 for (i
= 0; i
< q
->clhash
.hashsize
; i
++) {
1259 hlist_for_each_entry(cl
, &q
->clhash
.hash
[i
], common
.hnode
)
1260 tcf_destroy_chain(&cl
->filter_list
);
1262 for (i
= 0; i
< q
->clhash
.hashsize
; i
++) {
1263 hlist_for_each_entry_safe(cl
, next
, &q
->clhash
.hash
[i
],
1265 htb_destroy_class(sch
, cl
);
1267 qdisc_class_hash_destroy(&q
->clhash
);
1268 __skb_queue_purge(&q
->direct_queue
);
1271 static int htb_delete(struct Qdisc
*sch
, unsigned long arg
)
1273 struct htb_sched
*q
= qdisc_priv(sch
);
1274 struct htb_class
*cl
= (struct htb_class
*)arg
;
1276 struct Qdisc
*new_q
= NULL
;
1279 // TODO: why don't allow to delete subtree ? references ? does
1280 // tc subsys quarantee us that in htb_destroy it holds no class
1281 // refs so that we can remove children safely there ?
1282 if (cl
->children
|| cl
->filter_cnt
)
1285 if (!cl
->level
&& htb_parent_last_child(cl
)) {
1286 new_q
= qdisc_create_dflt(sch
->dev_queue
, &pfifo_qdisc_ops
,
1287 cl
->parent
->common
.classid
);
1294 qlen
= cl
->un
.leaf
.q
->q
.qlen
;
1295 qdisc_reset(cl
->un
.leaf
.q
);
1296 qdisc_tree_decrease_qlen(cl
->un
.leaf
.q
, qlen
);
1299 /* delete from hash and active; remainder in destroy_class */
1300 qdisc_class_hash_remove(&q
->clhash
, &cl
->common
);
1302 cl
->parent
->children
--;
1304 if (cl
->prio_activity
)
1305 htb_deactivate(q
, cl
);
1307 if (cl
->cmode
!= HTB_CAN_SEND
)
1308 htb_safe_rb_erase(&cl
->pq_node
,
1309 &q
->hlevel
[cl
->level
].wait_pq
);
1312 htb_parent_to_leaf(q
, cl
, new_q
);
1314 BUG_ON(--cl
->refcnt
== 0);
1316 * This shouldn't happen: we "hold" one cops->get() when called
1317 * from tc_ctl_tclass; the destroy method is done from cops->put().
1320 sch_tree_unlock(sch
);
1324 static void htb_put(struct Qdisc
*sch
, unsigned long arg
)
1326 struct htb_class
*cl
= (struct htb_class
*)arg
;
1328 if (--cl
->refcnt
== 0)
1329 htb_destroy_class(sch
, cl
);
1332 static int htb_change_class(struct Qdisc
*sch
, u32 classid
,
1333 u32 parentid
, struct nlattr
**tca
,
1337 struct htb_sched
*q
= qdisc_priv(sch
);
1338 struct htb_class
*cl
= (struct htb_class
*)*arg
, *parent
;
1339 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
1340 struct qdisc_rate_table
*rtab
= NULL
, *ctab
= NULL
;
1341 struct nlattr
*tb
[TCA_HTB_MAX
+ 1];
1342 struct tc_htb_opt
*hopt
;
1345 /* extract all subattrs from opt attr */
1349 err
= nla_parse_nested(tb
, TCA_HTB_MAX
, opt
, htb_policy
);
1354 if (tb
[TCA_HTB_PARMS
] == NULL
)
1357 parent
= parentid
== TC_H_ROOT
? NULL
: htb_find(parentid
, sch
);
1359 hopt
= nla_data(tb
[TCA_HTB_PARMS
]);
1360 if (!hopt
->rate
.rate
|| !hopt
->ceil
.rate
)
1363 /* Keeping backward compatible with rate_table based iproute2 tc */
1364 if (hopt
->rate
.linklayer
== TC_LINKLAYER_UNAWARE
) {
1365 rtab
= qdisc_get_rtab(&hopt
->rate
, tb
[TCA_HTB_RTAB
]);
1367 qdisc_put_rtab(rtab
);
1369 if (hopt
->ceil
.linklayer
== TC_LINKLAYER_UNAWARE
) {
1370 ctab
= qdisc_get_rtab(&hopt
->ceil
, tb
[TCA_HTB_CTAB
]);
1372 qdisc_put_rtab(ctab
);
1375 if (!cl
) { /* new class */
1376 struct Qdisc
*new_q
;
1380 struct gnet_estimator opt
;
1383 .nla_len
= nla_attr_size(sizeof(est
.opt
)),
1384 .nla_type
= TCA_RATE
,
1387 /* 4s interval, 16s averaging constant */
1393 /* check for valid classid */
1394 if (!classid
|| TC_H_MAJ(classid
^ sch
->handle
) ||
1395 htb_find(classid
, sch
))
1398 /* check maximal depth */
1399 if (parent
&& parent
->parent
&& parent
->parent
->level
< 2) {
1400 pr_err("htb: tree is too deep\n");
1404 cl
= kzalloc(sizeof(*cl
), GFP_KERNEL
);
1408 if (htb_rate_est
|| tca
[TCA_RATE
]) {
1409 err
= gen_new_estimator(&cl
->bstats
, &cl
->rate_est
,
1410 qdisc_root_sleeping_lock(sch
),
1411 tca
[TCA_RATE
] ? : &est
.nla
);
1420 INIT_LIST_HEAD(&cl
->un
.leaf
.drop_list
);
1421 RB_CLEAR_NODE(&cl
->pq_node
);
1423 for (prio
= 0; prio
< TC_HTB_NUMPRIO
; prio
++)
1424 RB_CLEAR_NODE(&cl
->node
[prio
]);
1426 /* create leaf qdisc early because it uses kmalloc(GFP_KERNEL)
1427 * so that can't be used inside of sch_tree_lock
1428 * -- thanks to Karlis Peisenieks
1430 new_q
= qdisc_create_dflt(sch
->dev_queue
,
1431 &pfifo_qdisc_ops
, classid
);
1433 if (parent
&& !parent
->level
) {
1434 unsigned int qlen
= parent
->un
.leaf
.q
->q
.qlen
;
1436 /* turn parent into inner node */
1437 qdisc_reset(parent
->un
.leaf
.q
);
1438 qdisc_tree_decrease_qlen(parent
->un
.leaf
.q
, qlen
);
1439 qdisc_destroy(parent
->un
.leaf
.q
);
1440 if (parent
->prio_activity
)
1441 htb_deactivate(q
, parent
);
1443 /* remove from evt list because of level change */
1444 if (parent
->cmode
!= HTB_CAN_SEND
) {
1445 htb_safe_rb_erase(&parent
->pq_node
, &q
->hlevel
[0].wait_pq
);
1446 parent
->cmode
= HTB_CAN_SEND
;
1448 parent
->level
= (parent
->parent
? parent
->parent
->level
1449 : TC_HTB_MAXDEPTH
) - 1;
1450 memset(&parent
->un
.inner
, 0, sizeof(parent
->un
.inner
));
1452 /* leaf (we) needs elementary qdisc */
1453 cl
->un
.leaf
.q
= new_q
? new_q
: &noop_qdisc
;
1455 cl
->common
.classid
= classid
;
1456 cl
->parent
= parent
;
1458 /* set class to be in HTB_CAN_SEND state */
1459 cl
->tokens
= PSCHED_TICKS2NS(hopt
->buffer
);
1460 cl
->ctokens
= PSCHED_TICKS2NS(hopt
->cbuffer
);
1461 cl
->mbuffer
= 60ULL * NSEC_PER_SEC
; /* 1min */
1462 cl
->t_c
= ktime_to_ns(ktime_get());
1463 cl
->cmode
= HTB_CAN_SEND
;
1465 /* attach to the hash list and parent's family */
1466 qdisc_class_hash_insert(&q
->clhash
, &cl
->common
);
1470 if (tca
[TCA_RATE
]) {
1471 err
= gen_replace_estimator(&cl
->bstats
, &cl
->rate_est
,
1472 qdisc_root_sleeping_lock(sch
),
1480 rate64
= tb
[TCA_HTB_RATE64
] ? nla_get_u64(tb
[TCA_HTB_RATE64
]) : 0;
1482 ceil64
= tb
[TCA_HTB_CEIL64
] ? nla_get_u64(tb
[TCA_HTB_CEIL64
]) : 0;
1484 psched_ratecfg_precompute(&cl
->rate
, &hopt
->rate
, rate64
);
1485 psched_ratecfg_precompute(&cl
->ceil
, &hopt
->ceil
, ceil64
);
1487 /* it used to be a nasty bug here, we have to check that node
1488 * is really leaf before changing cl->un.leaf !
1491 u64 quantum
= cl
->rate
.rate_bytes_ps
;
1493 do_div(quantum
, q
->rate2quantum
);
1494 cl
->quantum
= min_t(u64
, quantum
, INT_MAX
);
1496 if (!hopt
->quantum
&& cl
->quantum
< 1000) {
1498 "HTB: quantum of class %X is small. Consider r2q change.\n",
1499 cl
->common
.classid
);
1502 if (!hopt
->quantum
&& cl
->quantum
> 200000) {
1504 "HTB: quantum of class %X is big. Consider r2q change.\n",
1505 cl
->common
.classid
);
1506 cl
->quantum
= 200000;
1509 cl
->quantum
= hopt
->quantum
;
1510 if ((cl
->prio
= hopt
->prio
) >= TC_HTB_NUMPRIO
)
1511 cl
->prio
= TC_HTB_NUMPRIO
- 1;
1514 cl
->buffer
= PSCHED_TICKS2NS(hopt
->buffer
);
1515 cl
->cbuffer
= PSCHED_TICKS2NS(hopt
->cbuffer
);
1517 sch_tree_unlock(sch
);
1519 qdisc_class_hash_grow(sch
, &q
->clhash
);
1521 *arg
= (unsigned long)cl
;
1528 static struct tcf_proto
**htb_find_tcf(struct Qdisc
*sch
, unsigned long arg
)
1530 struct htb_sched
*q
= qdisc_priv(sch
);
1531 struct htb_class
*cl
= (struct htb_class
*)arg
;
1532 struct tcf_proto
**fl
= cl
? &cl
->filter_list
: &q
->filter_list
;
1537 static unsigned long htb_bind_filter(struct Qdisc
*sch
, unsigned long parent
,
1540 struct htb_class
*cl
= htb_find(classid
, sch
);
1542 /*if (cl && !cl->level) return 0;
1543 * The line above used to be there to prevent attaching filters to
1544 * leaves. But at least tc_index filter uses this just to get class
1545 * for other reasons so that we have to allow for it.
1547 * 19.6.2002 As Werner explained it is ok - bind filter is just
1548 * another way to "lock" the class - unlike "get" this lock can
1549 * be broken by class during destroy IIUC.
1553 return (unsigned long)cl
;
1556 static void htb_unbind_filter(struct Qdisc
*sch
, unsigned long arg
)
1558 struct htb_class
*cl
= (struct htb_class
*)arg
;
1564 static void htb_walk(struct Qdisc
*sch
, struct qdisc_walker
*arg
)
1566 struct htb_sched
*q
= qdisc_priv(sch
);
1567 struct htb_class
*cl
;
1573 for (i
= 0; i
< q
->clhash
.hashsize
; i
++) {
1574 hlist_for_each_entry(cl
, &q
->clhash
.hash
[i
], common
.hnode
) {
1575 if (arg
->count
< arg
->skip
) {
1579 if (arg
->fn(sch
, (unsigned long)cl
, arg
) < 0) {
1588 static const struct Qdisc_class_ops htb_class_ops
= {
1591 .qlen_notify
= htb_qlen_notify
,
1594 .change
= htb_change_class
,
1595 .delete = htb_delete
,
1597 .tcf_chain
= htb_find_tcf
,
1598 .bind_tcf
= htb_bind_filter
,
1599 .unbind_tcf
= htb_unbind_filter
,
1600 .dump
= htb_dump_class
,
1601 .dump_stats
= htb_dump_class_stats
,
1604 static struct Qdisc_ops htb_qdisc_ops __read_mostly
= {
1605 .cl_ops
= &htb_class_ops
,
1607 .priv_size
= sizeof(struct htb_sched
),
1608 .enqueue
= htb_enqueue
,
1609 .dequeue
= htb_dequeue
,
1610 .peek
= qdisc_peek_dequeued
,
1614 .destroy
= htb_destroy
,
1616 .owner
= THIS_MODULE
,
1619 static int __init
htb_module_init(void)
1621 return register_qdisc(&htb_qdisc_ops
);
1623 static void __exit
htb_module_exit(void)
1625 unregister_qdisc(&htb_qdisc_ops
);
1628 module_init(htb_module_init
)
1629 module_exit(htb_module_exit
)
1630 MODULE_LICENSE("GPL");