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 <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/string.h>
19 #include <linux/errno.h>
20 #include <linux/skbuff.h>
21 #include <net/netlink.h>
22 #include <net/pkt_sched.h>
25 /* Simple Token Bucket Filter.
26 =======================================
36 A data flow obeys TBF with rate R and depth B, if for any
37 time interval t_i...t_f the number of transmitted bits
38 does not exceed B + R*(t_f-t_i).
40 Packetized version of this definition:
41 The sequence of packets of sizes s_i served at moments t_i
42 obeys TBF, if for any i<=k:
44 s_i+....+s_k <= B + R*(t_k - t_i)
49 Let N(t_i) be B/R initially and N(t) grow continuously with time as:
51 N(t+delta) = min{B/R, N(t) + delta}
53 If the first packet in queue has length S, it may be
54 transmitted only at the time t_* when S/R <= N(t_*),
55 and in this case N(t) jumps:
57 N(t_* + 0) = N(t_* - 0) - S/R.
61 Actually, QoS requires two TBF to be applied to a data stream.
62 One of them controls steady state burst size, another
63 one with rate P (peak rate) and depth M (equal to link MTU)
64 limits bursts at a smaller time scale.
66 It is easy to see that P>R, and B>M. If P is infinity, this double
67 TBF is equivalent to a single one.
69 When TBF works in reshaping mode, latency is estimated as:
71 lat = max ((L-B)/R, (L-M)/P)
77 If TBF throttles, it starts a watchdog timer, which will wake it up
78 when it is ready to transmit.
79 Note that the minimal timer resolution is 1/HZ.
80 If no new packets arrive during this period,
81 or if the device is not awaken by EOI for some previous packet,
82 TBF can stop its activity for 1/HZ.
85 This means, that with depth B, the maximal rate is
89 F.e. for 10Mbit ethernet and HZ=100 the minimal allowed B is ~10Kbytes.
91 Note that the peak rate TBF is much more tough: with MTU 1500
92 P_crit = 150Kbytes/sec. So, if you need greater peak
93 rates, use alpha with HZ=1000 :-)
95 With classful TBF, limit is just kept for backwards compatibility.
96 It is passed to the default bfifo qdisc - if the inner qdisc is
97 changed the limit is not effective anymore.
100 struct tbf_sched_data
103 u32 limit
; /* Maximal length of backlog: bytes */
104 u32 buffer
; /* Token bucket depth/rate: MUST BE >= MTU/B */
107 struct qdisc_rate_table
*R_tab
;
108 struct qdisc_rate_table
*P_tab
;
111 long tokens
; /* Current number of B tokens */
112 long ptokens
; /* Current number of P tokens */
113 psched_time_t t_c
; /* Time check-point */
114 struct Qdisc
*qdisc
; /* Inner qdisc, default - bfifo queue */
115 struct qdisc_watchdog watchdog
; /* Watchdog timer */
118 #define L2T(q,L) qdisc_l2t((q)->R_tab,L)
119 #define L2T_P(q,L) qdisc_l2t((q)->P_tab,L)
121 static int tbf_enqueue(struct sk_buff
*skb
, struct Qdisc
* sch
)
123 struct tbf_sched_data
*q
= qdisc_priv(sch
);
126 if (skb
->len
> q
->max_size
) {
128 #ifdef CONFIG_NET_CLS_ACT
129 if (sch
->reshape_fail
== NULL
|| sch
->reshape_fail(skb
, sch
))
133 return NET_XMIT_DROP
;
136 if ((ret
= q
->qdisc
->enqueue(skb
, q
->qdisc
)) != 0) {
142 sch
->bstats
.bytes
+= skb
->len
;
143 sch
->bstats
.packets
++;
147 static int tbf_requeue(struct sk_buff
*skb
, struct Qdisc
* sch
)
149 struct tbf_sched_data
*q
= qdisc_priv(sch
);
152 if ((ret
= q
->qdisc
->ops
->requeue(skb
, q
->qdisc
)) == 0) {
154 sch
->qstats
.requeues
++;
160 static unsigned int tbf_drop(struct Qdisc
* sch
)
162 struct tbf_sched_data
*q
= qdisc_priv(sch
);
163 unsigned int len
= 0;
165 if (q
->qdisc
->ops
->drop
&& (len
= q
->qdisc
->ops
->drop(q
->qdisc
)) != 0) {
172 static struct sk_buff
*tbf_dequeue(struct Qdisc
* sch
)
174 struct tbf_sched_data
*q
= qdisc_priv(sch
);
177 skb
= q
->qdisc
->dequeue(q
->qdisc
);
183 unsigned int len
= skb
->len
;
185 now
= psched_get_time();
186 toks
= psched_tdiff_bounded(now
, q
->t_c
, q
->buffer
);
189 ptoks
= toks
+ q
->ptokens
;
190 if (ptoks
> (long)q
->mtu
)
192 ptoks
-= L2T_P(q
, len
);
195 if (toks
> (long)q
->buffer
)
199 if ((toks
|ptoks
) >= 0) {
204 sch
->flags
&= ~TCQ_F_THROTTLED
;
208 qdisc_watchdog_schedule(&q
->watchdog
,
209 now
+ max_t(long, -toks
, -ptoks
));
211 /* Maybe we have a shorter packet in the queue,
212 which can be sent now. It sounds cool,
213 but, however, this is wrong in principle.
214 We MUST NOT reorder packets under these circumstances.
216 Really, if we split the flow into independent
217 subflows, it would be a very good solution.
218 This is the main idea of all FQ algorithms
219 (cf. CSZ, HPFQ, HFSC)
222 if (q
->qdisc
->ops
->requeue(skb
, q
->qdisc
) != NET_XMIT_SUCCESS
) {
223 /* When requeue fails skb is dropped */
224 qdisc_tree_decrease_qlen(q
->qdisc
, 1);
228 sch
->qstats
.overlimits
++;
233 static void tbf_reset(struct Qdisc
* sch
)
235 struct tbf_sched_data
*q
= qdisc_priv(sch
);
237 qdisc_reset(q
->qdisc
);
239 q
->t_c
= psched_get_time();
240 q
->tokens
= q
->buffer
;
242 qdisc_watchdog_cancel(&q
->watchdog
);
245 static const struct nla_policy tbf_policy
[TCA_TBF_MAX
+ 1] = {
246 [TCA_TBF_PARMS
] = { .len
= sizeof(struct tc_tbf_qopt
) },
247 [TCA_TBF_RTAB
] = { .type
= NLA_BINARY
, .len
= TC_RTAB_SIZE
},
248 [TCA_TBF_PTAB
] = { .type
= NLA_BINARY
, .len
= TC_RTAB_SIZE
},
251 static int tbf_change(struct Qdisc
* sch
, struct nlattr
*opt
)
254 struct tbf_sched_data
*q
= qdisc_priv(sch
);
255 struct nlattr
*tb
[TCA_TBF_PTAB
+ 1];
256 struct tc_tbf_qopt
*qopt
;
257 struct qdisc_rate_table
*rtab
= NULL
;
258 struct qdisc_rate_table
*ptab
= NULL
;
259 struct Qdisc
*child
= NULL
;
262 err
= nla_parse_nested(tb
, TCA_TBF_PTAB
, opt
, tbf_policy
);
267 if (tb
[TCA_TBF_PARMS
] == NULL
)
270 qopt
= nla_data(tb
[TCA_TBF_PARMS
]);
271 rtab
= qdisc_get_rtab(&qopt
->rate
, tb
[TCA_TBF_RTAB
]);
275 if (qopt
->peakrate
.rate
) {
276 if (qopt
->peakrate
.rate
> qopt
->rate
.rate
)
277 ptab
= qdisc_get_rtab(&qopt
->peakrate
, tb
[TCA_TBF_PTAB
]);
282 for (n
= 0; n
< 256; n
++)
283 if (rtab
->data
[n
] > qopt
->buffer
) break;
284 max_size
= (n
<< qopt
->rate
.cell_log
)-1;
288 for (n
= 0; n
< 256; n
++)
289 if (ptab
->data
[n
] > qopt
->mtu
) break;
290 size
= (n
<< qopt
->peakrate
.cell_log
)-1;
291 if (size
< max_size
) max_size
= size
;
296 if (qopt
->limit
> 0) {
297 child
= fifo_create_dflt(sch
, &bfifo_qdisc_ops
, qopt
->limit
);
299 err
= PTR_ERR(child
);
306 qdisc_tree_decrease_qlen(q
->qdisc
, q
->qdisc
->q
.qlen
);
307 qdisc_destroy(xchg(&q
->qdisc
, child
));
309 q
->limit
= qopt
->limit
;
311 q
->max_size
= max_size
;
312 q
->buffer
= qopt
->buffer
;
313 q
->tokens
= q
->buffer
;
315 rtab
= xchg(&q
->R_tab
, rtab
);
316 ptab
= xchg(&q
->P_tab
, ptab
);
317 sch_tree_unlock(sch
);
321 qdisc_put_rtab(rtab
);
323 qdisc_put_rtab(ptab
);
327 static int tbf_init(struct Qdisc
* sch
, struct nlattr
*opt
)
329 struct tbf_sched_data
*q
= qdisc_priv(sch
);
334 q
->t_c
= psched_get_time();
335 qdisc_watchdog_init(&q
->watchdog
, sch
);
336 q
->qdisc
= &noop_qdisc
;
338 return tbf_change(sch
, opt
);
341 static void tbf_destroy(struct Qdisc
*sch
)
343 struct tbf_sched_data
*q
= qdisc_priv(sch
);
345 qdisc_watchdog_cancel(&q
->watchdog
);
348 qdisc_put_rtab(q
->P_tab
);
350 qdisc_put_rtab(q
->R_tab
);
352 qdisc_destroy(q
->qdisc
);
355 static int tbf_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
357 struct tbf_sched_data
*q
= qdisc_priv(sch
);
359 struct tc_tbf_qopt opt
;
361 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
363 goto nla_put_failure
;
365 opt
.limit
= q
->limit
;
366 opt
.rate
= q
->R_tab
->rate
;
368 opt
.peakrate
= q
->P_tab
->rate
;
370 memset(&opt
.peakrate
, 0, sizeof(opt
.peakrate
));
372 opt
.buffer
= q
->buffer
;
373 NLA_PUT(skb
, TCA_TBF_PARMS
, sizeof(opt
), &opt
);
375 nla_nest_end(skb
, nest
);
379 nla_nest_cancel(skb
, nest
);
383 static int tbf_dump_class(struct Qdisc
*sch
, unsigned long cl
,
384 struct sk_buff
*skb
, struct tcmsg
*tcm
)
386 struct tbf_sched_data
*q
= qdisc_priv(sch
);
388 if (cl
!= 1) /* only one class */
391 tcm
->tcm_handle
|= TC_H_MIN(1);
392 tcm
->tcm_info
= q
->qdisc
->handle
;
397 static int tbf_graft(struct Qdisc
*sch
, unsigned long arg
, struct Qdisc
*new,
400 struct tbf_sched_data
*q
= qdisc_priv(sch
);
406 *old
= xchg(&q
->qdisc
, new);
407 qdisc_tree_decrease_qlen(*old
, (*old
)->q
.qlen
);
409 sch_tree_unlock(sch
);
414 static struct Qdisc
*tbf_leaf(struct Qdisc
*sch
, unsigned long arg
)
416 struct tbf_sched_data
*q
= qdisc_priv(sch
);
420 static unsigned long tbf_get(struct Qdisc
*sch
, u32 classid
)
425 static void tbf_put(struct Qdisc
*sch
, unsigned long arg
)
429 static int tbf_change_class(struct Qdisc
*sch
, u32 classid
, u32 parentid
,
430 struct nlattr
**tca
, unsigned long *arg
)
435 static int tbf_delete(struct Qdisc
*sch
, unsigned long arg
)
440 static void tbf_walk(struct Qdisc
*sch
, struct qdisc_walker
*walker
)
443 if (walker
->count
>= walker
->skip
)
444 if (walker
->fn(sch
, 1, walker
) < 0) {
452 static struct tcf_proto
**tbf_find_tcf(struct Qdisc
*sch
, unsigned long cl
)
457 static const struct Qdisc_class_ops tbf_class_ops
=
463 .change
= tbf_change_class
,
464 .delete = tbf_delete
,
466 .tcf_chain
= tbf_find_tcf
,
467 .dump
= tbf_dump_class
,
470 static struct Qdisc_ops tbf_qdisc_ops __read_mostly
= {
472 .cl_ops
= &tbf_class_ops
,
474 .priv_size
= sizeof(struct tbf_sched_data
),
475 .enqueue
= tbf_enqueue
,
476 .dequeue
= tbf_dequeue
,
477 .requeue
= tbf_requeue
,
481 .destroy
= tbf_destroy
,
482 .change
= tbf_change
,
484 .owner
= THIS_MODULE
,
487 static int __init
tbf_module_init(void)
489 return register_qdisc(&tbf_qdisc_ops
);
492 static void __exit
tbf_module_exit(void)
494 unregister_qdisc(&tbf_qdisc_ops
);
496 module_init(tbf_module_init
)
497 module_exit(tbf_module_exit
)
498 MODULE_LICENSE("GPL");