2 * net/sched/sch_tbf.c Token Bucket Filter queue.
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: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 * Dmitry Torokhov <dtor@mail.ru> - allow attaching inner qdiscs -
11 * original idea by Martin Devera
15 #include <linux/module.h>
16 #include <asm/uaccess.h>
17 #include <asm/system.h>
18 #include <linux/bitops.h>
19 #include <linux/types.h>
20 #include <linux/kernel.h>
21 #include <linux/jiffies.h>
22 #include <linux/string.h>
24 #include <linux/socket.h>
25 #include <linux/sockios.h>
27 #include <linux/errno.h>
28 #include <linux/interrupt.h>
29 #include <linux/if_ether.h>
30 #include <linux/inet.h>
31 #include <linux/netdevice.h>
32 #include <linux/etherdevice.h>
33 #include <linux/notifier.h>
35 #include <net/netlink.h>
36 #include <net/route.h>
37 #include <linux/skbuff.h>
39 #include <net/pkt_sched.h>
42 /* Simple Token Bucket Filter.
43 =======================================
53 A data flow obeys TBF with rate R and depth B, if for any
54 time interval t_i...t_f the number of transmitted bits
55 does not exceed B + R*(t_f-t_i).
57 Packetized version of this definition:
58 The sequence of packets of sizes s_i served at moments t_i
59 obeys TBF, if for any i<=k:
61 s_i+....+s_k <= B + R*(t_k - t_i)
66 Let N(t_i) be B/R initially and N(t) grow continuously with time as:
68 N(t+delta) = min{B/R, N(t) + delta}
70 If the first packet in queue has length S, it may be
71 transmitted only at the time t_* when S/R <= N(t_*),
72 and in this case N(t) jumps:
74 N(t_* + 0) = N(t_* - 0) - S/R.
78 Actually, QoS requires two TBF to be applied to a data stream.
79 One of them controls steady state burst size, another
80 one with rate P (peak rate) and depth M (equal to link MTU)
81 limits bursts at a smaller time scale.
83 It is easy to see that P>R, and B>M. If P is infinity, this double
84 TBF is equivalent to a single one.
86 When TBF works in reshaping mode, latency is estimated as:
88 lat = max ((L-B)/R, (L-M)/P)
94 If TBF throttles, it starts a watchdog timer, which will wake it up
95 when it is ready to transmit.
96 Note that the minimal timer resolution is 1/HZ.
97 If no new packets arrive during this period,
98 or if the device is not awaken by EOI for some previous packet,
99 TBF can stop its activity for 1/HZ.
102 This means, that with depth B, the maximal rate is
106 F.e. for 10Mbit ethernet and HZ=100 the minimal allowed B is ~10Kbytes.
108 Note that the peak rate TBF is much more tough: with MTU 1500
109 P_crit = 150Kbytes/sec. So, if you need greater peak
110 rates, use alpha with HZ=1000 :-)
112 With classful TBF, limit is just kept for backwards compatibility.
113 It is passed to the default bfifo qdisc - if the inner qdisc is
114 changed the limit is not effective anymore.
117 struct tbf_sched_data
120 u32 limit
; /* Maximal length of backlog: bytes */
121 u32 buffer
; /* Token bucket depth/rate: MUST BE >= MTU/B */
124 struct qdisc_rate_table
*R_tab
;
125 struct qdisc_rate_table
*P_tab
;
128 long tokens
; /* Current number of B tokens */
129 long ptokens
; /* Current number of P tokens */
130 psched_time_t t_c
; /* Time check-point */
131 struct Qdisc
*qdisc
; /* Inner qdisc, default - bfifo queue */
132 struct qdisc_watchdog watchdog
; /* Watchdog timer */
135 #define L2T(q,L) qdisc_l2t((q)->R_tab,L)
136 #define L2T_P(q,L) qdisc_l2t((q)->P_tab,L)
138 static int tbf_enqueue(struct sk_buff
*skb
, struct Qdisc
* sch
)
140 struct tbf_sched_data
*q
= qdisc_priv(sch
);
143 if (skb
->len
> q
->max_size
)
144 return qdisc_reshape_fail(skb
, sch
);
146 if ((ret
= q
->qdisc
->enqueue(skb
, q
->qdisc
)) != 0) {
152 sch
->bstats
.bytes
+= skb
->len
;
153 sch
->bstats
.packets
++;
157 static int tbf_requeue(struct sk_buff
*skb
, struct Qdisc
* sch
)
159 struct tbf_sched_data
*q
= qdisc_priv(sch
);
162 if ((ret
= q
->qdisc
->ops
->requeue(skb
, q
->qdisc
)) == 0) {
164 sch
->qstats
.requeues
++;
170 static unsigned int tbf_drop(struct Qdisc
* sch
)
172 struct tbf_sched_data
*q
= qdisc_priv(sch
);
173 unsigned int len
= 0;
175 if (q
->qdisc
->ops
->drop
&& (len
= q
->qdisc
->ops
->drop(q
->qdisc
)) != 0) {
182 static struct sk_buff
*tbf_dequeue(struct Qdisc
* sch
)
184 struct tbf_sched_data
*q
= qdisc_priv(sch
);
187 skb
= q
->qdisc
->dequeue(q
->qdisc
);
193 unsigned int len
= skb
->len
;
195 now
= psched_get_time();
196 toks
= psched_tdiff_bounded(now
, q
->t_c
, q
->buffer
);
199 ptoks
= toks
+ q
->ptokens
;
200 if (ptoks
> (long)q
->mtu
)
202 ptoks
-= L2T_P(q
, len
);
205 if (toks
> (long)q
->buffer
)
209 if ((toks
|ptoks
) >= 0) {
214 sch
->flags
&= ~TCQ_F_THROTTLED
;
218 qdisc_watchdog_schedule(&q
->watchdog
,
219 now
+ max_t(long, -toks
, -ptoks
));
221 /* Maybe we have a shorter packet in the queue,
222 which can be sent now. It sounds cool,
223 but, however, this is wrong in principle.
224 We MUST NOT reorder packets under these circumstances.
226 Really, if we split the flow into independent
227 subflows, it would be a very good solution.
228 This is the main idea of all FQ algorithms
229 (cf. CSZ, HPFQ, HFSC)
232 if (q
->qdisc
->ops
->requeue(skb
, q
->qdisc
) != NET_XMIT_SUCCESS
) {
233 /* When requeue fails skb is dropped */
234 qdisc_tree_decrease_qlen(q
->qdisc
, 1);
238 sch
->qstats
.overlimits
++;
243 static void tbf_reset(struct Qdisc
* sch
)
245 struct tbf_sched_data
*q
= qdisc_priv(sch
);
247 qdisc_reset(q
->qdisc
);
249 q
->t_c
= psched_get_time();
250 q
->tokens
= q
->buffer
;
252 qdisc_watchdog_cancel(&q
->watchdog
);
255 static struct Qdisc
*tbf_create_dflt_qdisc(struct Qdisc
*sch
, u32 limit
)
261 q
= qdisc_create_dflt(sch
->dev
, &bfifo_qdisc_ops
,
262 TC_H_MAKE(sch
->handle
, 1));
264 rta
= kmalloc(RTA_LENGTH(sizeof(struct tc_fifo_qopt
)), GFP_KERNEL
);
266 rta
->rta_type
= RTM_NEWQDISC
;
267 rta
->rta_len
= RTA_LENGTH(sizeof(struct tc_fifo_qopt
));
268 ((struct tc_fifo_qopt
*)RTA_DATA(rta
))->limit
= limit
;
270 ret
= q
->ops
->change(q
, rta
);
282 static int tbf_change(struct Qdisc
* sch
, struct rtattr
*opt
)
285 struct tbf_sched_data
*q
= qdisc_priv(sch
);
286 struct rtattr
*tb
[TCA_TBF_PTAB
];
287 struct tc_tbf_qopt
*qopt
;
288 struct qdisc_rate_table
*rtab
= NULL
;
289 struct qdisc_rate_table
*ptab
= NULL
;
290 struct Qdisc
*child
= NULL
;
293 if (rtattr_parse_nested(tb
, TCA_TBF_PTAB
, opt
) ||
294 tb
[TCA_TBF_PARMS
-1] == NULL
||
295 RTA_PAYLOAD(tb
[TCA_TBF_PARMS
-1]) < sizeof(*qopt
))
298 qopt
= RTA_DATA(tb
[TCA_TBF_PARMS
-1]);
299 rtab
= qdisc_get_rtab(&qopt
->rate
, tb
[TCA_TBF_RTAB
-1]);
303 if (qopt
->peakrate
.rate
) {
304 if (qopt
->peakrate
.rate
> qopt
->rate
.rate
)
305 ptab
= qdisc_get_rtab(&qopt
->peakrate
, tb
[TCA_TBF_PTAB
-1]);
310 for (n
= 0; n
< 256; n
++)
311 if (rtab
->data
[n
] > qopt
->buffer
) break;
312 max_size
= (n
<< qopt
->rate
.cell_log
)-1;
316 for (n
= 0; n
< 256; n
++)
317 if (ptab
->data
[n
] > qopt
->mtu
) break;
318 size
= (n
<< qopt
->peakrate
.cell_log
)-1;
319 if (size
< max_size
) max_size
= size
;
324 if (qopt
->limit
> 0) {
325 if ((child
= tbf_create_dflt_qdisc(sch
, qopt
->limit
)) == NULL
)
331 qdisc_tree_decrease_qlen(q
->qdisc
, q
->qdisc
->q
.qlen
);
332 qdisc_destroy(xchg(&q
->qdisc
, child
));
334 q
->limit
= qopt
->limit
;
336 q
->max_size
= max_size
;
337 q
->buffer
= qopt
->buffer
;
338 q
->tokens
= q
->buffer
;
340 rtab
= xchg(&q
->R_tab
, rtab
);
341 ptab
= xchg(&q
->P_tab
, ptab
);
342 sch_tree_unlock(sch
);
346 qdisc_put_rtab(rtab
);
348 qdisc_put_rtab(ptab
);
352 static int tbf_init(struct Qdisc
* sch
, struct rtattr
*opt
)
354 struct tbf_sched_data
*q
= qdisc_priv(sch
);
359 q
->t_c
= psched_get_time();
360 qdisc_watchdog_init(&q
->watchdog
, sch
);
361 q
->qdisc
= &noop_qdisc
;
363 return tbf_change(sch
, opt
);
366 static void tbf_destroy(struct Qdisc
*sch
)
368 struct tbf_sched_data
*q
= qdisc_priv(sch
);
370 qdisc_watchdog_cancel(&q
->watchdog
);
373 qdisc_put_rtab(q
->P_tab
);
375 qdisc_put_rtab(q
->R_tab
);
377 qdisc_destroy(q
->qdisc
);
380 static int tbf_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
382 struct tbf_sched_data
*q
= qdisc_priv(sch
);
383 unsigned char *b
= skb_tail_pointer(skb
);
385 struct tc_tbf_qopt opt
;
387 rta
= (struct rtattr
*)b
;
388 RTA_PUT(skb
, TCA_OPTIONS
, 0, NULL
);
390 opt
.limit
= q
->limit
;
391 opt
.rate
= q
->R_tab
->rate
;
393 opt
.peakrate
= q
->P_tab
->rate
;
395 memset(&opt
.peakrate
, 0, sizeof(opt
.peakrate
));
397 opt
.buffer
= q
->buffer
;
398 RTA_PUT(skb
, TCA_TBF_PARMS
, sizeof(opt
), &opt
);
399 rta
->rta_len
= skb_tail_pointer(skb
) - b
;
408 static int tbf_dump_class(struct Qdisc
*sch
, unsigned long cl
,
409 struct sk_buff
*skb
, struct tcmsg
*tcm
)
411 struct tbf_sched_data
*q
= qdisc_priv(sch
);
413 if (cl
!= 1) /* only one class */
416 tcm
->tcm_handle
|= TC_H_MIN(1);
417 tcm
->tcm_info
= q
->qdisc
->handle
;
422 static int tbf_graft(struct Qdisc
*sch
, unsigned long arg
, struct Qdisc
*new,
425 struct tbf_sched_data
*q
= qdisc_priv(sch
);
431 *old
= xchg(&q
->qdisc
, new);
432 qdisc_tree_decrease_qlen(*old
, (*old
)->q
.qlen
);
434 sch_tree_unlock(sch
);
439 static struct Qdisc
*tbf_leaf(struct Qdisc
*sch
, unsigned long arg
)
441 struct tbf_sched_data
*q
= qdisc_priv(sch
);
445 static unsigned long tbf_get(struct Qdisc
*sch
, u32 classid
)
450 static void tbf_put(struct Qdisc
*sch
, unsigned long arg
)
454 static int tbf_change_class(struct Qdisc
*sch
, u32 classid
, u32 parentid
,
455 struct rtattr
**tca
, unsigned long *arg
)
460 static int tbf_delete(struct Qdisc
*sch
, unsigned long arg
)
465 static void tbf_walk(struct Qdisc
*sch
, struct qdisc_walker
*walker
)
468 if (walker
->count
>= walker
->skip
)
469 if (walker
->fn(sch
, 1, walker
) < 0) {
477 static struct tcf_proto
**tbf_find_tcf(struct Qdisc
*sch
, unsigned long cl
)
482 static struct Qdisc_class_ops tbf_class_ops
=
488 .change
= tbf_change_class
,
489 .delete = tbf_delete
,
491 .tcf_chain
= tbf_find_tcf
,
492 .dump
= tbf_dump_class
,
495 static struct Qdisc_ops tbf_qdisc_ops
= {
497 .cl_ops
= &tbf_class_ops
,
499 .priv_size
= sizeof(struct tbf_sched_data
),
500 .enqueue
= tbf_enqueue
,
501 .dequeue
= tbf_dequeue
,
502 .requeue
= tbf_requeue
,
506 .destroy
= tbf_destroy
,
507 .change
= tbf_change
,
509 .owner
= THIS_MODULE
,
512 static int __init
tbf_module_init(void)
514 return register_qdisc(&tbf_qdisc_ops
);
517 static void __exit
tbf_module_exit(void)
519 unregister_qdisc(&tbf_qdisc_ops
);
521 module_init(tbf_module_init
)
522 module_exit(tbf_module_exit
)
523 MODULE_LICENSE("GPL");