forked from torvalds/linux
-
Notifications
You must be signed in to change notification settings - Fork 0
/
sch_cbq.c
2062 lines (1723 loc) · 47.2 KB
/
sch_cbq.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
* net/sched/sch_cbq.c Class-Based Queueing discipline.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*
* Authors: Alexey Kuznetsov, <[email protected]>
*
*/
#include <linux/module.h>
#include <linux/slab.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/string.h>
#include <linux/errno.h>
#include <linux/skbuff.h>
#include <net/netlink.h>
#include <net/pkt_sched.h>
/* Class-Based Queueing (CBQ) algorithm.
=======================================
Sources: [1] Sally Floyd and Van Jacobson, "Link-sharing and Resource
Management Models for Packet Networks",
IEEE/ACM Transactions on Networking, Vol.3, No.4, 1995
[2] Sally Floyd, "Notes on CBQ and Guaranteed Service", 1995
[3] Sally Floyd, "Notes on Class-Based Queueing: Setting
Parameters", 1996
[4] Sally Floyd and Michael Speer, "Experimental Results
for Class-Based Queueing", 1998, not published.
-----------------------------------------------------------------------
Algorithm skeleton was taken from NS simulator cbq.cc.
If someone wants to check this code against the LBL version,
he should take into account that ONLY the skeleton was borrowed,
the implementation is different. Particularly:
--- The WRR algorithm is different. Our version looks more
reasonable (I hope) and works when quanta are allowed to be
less than MTU, which is always the case when real time classes
have small rates. Note, that the statement of [3] is
incomplete, delay may actually be estimated even if class
per-round allotment is less than MTU. Namely, if per-round
allotment is W*r_i, and r_1+...+r_k = r < 1
delay_i <= ([MTU/(W*r_i)]*W*r + W*r + k*MTU)/B
In the worst case we have IntServ estimate with D = W*r+k*MTU
and C = MTU*r. The proof (if correct at all) is trivial.
--- It seems that cbq-2.0 is not very accurate. At least, I cannot
interpret some places, which look like wrong translations
from NS. Anyone is advised to find these differences
and explain to me, why I am wrong 8).
--- Linux has no EOI event, so that we cannot estimate true class
idle time. Workaround is to consider the next dequeue event
as sign that previous packet is finished. This is wrong because of
internal device queueing, but on a permanently loaded link it is true.
Moreover, combined with clock integrator, this scheme looks
very close to an ideal solution. */
struct cbq_sched_data;
struct cbq_class {
struct Qdisc_class_common common;
struct cbq_class *next_alive; /* next class with backlog in this priority band */
/* Parameters */
unsigned char priority; /* class priority */
unsigned char priority2; /* priority to be used after overlimit */
unsigned char ewma_log; /* time constant for idle time calculation */
unsigned char ovl_strategy;
#ifdef CONFIG_NET_CLS_ACT
unsigned char police;
#endif
u32 defmap;
/* Link-sharing scheduler parameters */
long maxidle; /* Class parameters: see below. */
long offtime;
long minidle;
u32 avpkt;
struct qdisc_rate_table *R_tab;
/* Overlimit strategy parameters */
void (*overlimit)(struct cbq_class *cl);
psched_tdiff_t penalty;
/* General scheduler (WRR) parameters */
long allot;
long quantum; /* Allotment per WRR round */
long weight; /* Relative allotment: see below */
struct Qdisc *qdisc; /* Ptr to CBQ discipline */
struct cbq_class *split; /* Ptr to split node */
struct cbq_class *share; /* Ptr to LS parent in the class tree */
struct cbq_class *tparent; /* Ptr to tree parent in the class tree */
struct cbq_class *borrow; /* NULL if class is bandwidth limited;
parent otherwise */
struct cbq_class *sibling; /* Sibling chain */
struct cbq_class *children; /* Pointer to children chain */
struct Qdisc *q; /* Elementary queueing discipline */
/* Variables */
unsigned char cpriority; /* Effective priority */
unsigned char delayed;
unsigned char level; /* level of the class in hierarchy:
0 for leaf classes, and maximal
level of children + 1 for nodes.
*/
psched_time_t last; /* Last end of service */
psched_time_t undertime;
long avgidle;
long deficit; /* Saved deficit for WRR */
psched_time_t penalized;
struct gnet_stats_basic_packed bstats;
struct gnet_stats_queue qstats;
struct gnet_stats_rate_est64 rate_est;
struct tc_cbq_xstats xstats;
struct tcf_proto __rcu *filter_list;
int refcnt;
int filters;
struct cbq_class *defaults[TC_PRIO_MAX + 1];
};
struct cbq_sched_data {
struct Qdisc_class_hash clhash; /* Hash table of all classes */
int nclasses[TC_CBQ_MAXPRIO + 1];
unsigned int quanta[TC_CBQ_MAXPRIO + 1];
struct cbq_class link;
unsigned int activemask;
struct cbq_class *active[TC_CBQ_MAXPRIO + 1]; /* List of all classes
with backlog */
#ifdef CONFIG_NET_CLS_ACT
struct cbq_class *rx_class;
#endif
struct cbq_class *tx_class;
struct cbq_class *tx_borrowed;
int tx_len;
psched_time_t now; /* Cached timestamp */
unsigned int pmask;
struct hrtimer delay_timer;
struct qdisc_watchdog watchdog; /* Watchdog timer,
started when CBQ has
backlog, but cannot
transmit just now */
psched_tdiff_t wd_expires;
int toplevel;
u32 hgenerator;
};
#define L2T(cl, len) qdisc_l2t((cl)->R_tab, len)
static inline struct cbq_class *
cbq_class_lookup(struct cbq_sched_data *q, u32 classid)
{
struct Qdisc_class_common *clc;
clc = qdisc_class_find(&q->clhash, classid);
if (clc == NULL)
return NULL;
return container_of(clc, struct cbq_class, common);
}
#ifdef CONFIG_NET_CLS_ACT
static struct cbq_class *
cbq_reclassify(struct sk_buff *skb, struct cbq_class *this)
{
struct cbq_class *cl;
for (cl = this->tparent; cl; cl = cl->tparent) {
struct cbq_class *new = cl->defaults[TC_PRIO_BESTEFFORT];
if (new != NULL && new != this)
return new;
}
return NULL;
}
#endif
/* Classify packet. The procedure is pretty complicated, but
* it allows us to combine link sharing and priority scheduling
* transparently.
*
* Namely, you can put link sharing rules (f.e. route based) at root of CBQ,
* so that it resolves to split nodes. Then packets are classified
* by logical priority, or a more specific classifier may be attached
* to the split node.
*/
static struct cbq_class *
cbq_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
{
struct cbq_sched_data *q = qdisc_priv(sch);
struct cbq_class *head = &q->link;
struct cbq_class **defmap;
struct cbq_class *cl = NULL;
u32 prio = skb->priority;
struct tcf_proto *fl;
struct tcf_result res;
/*
* Step 1. If skb->priority points to one of our classes, use it.
*/
if (TC_H_MAJ(prio ^ sch->handle) == 0 &&
(cl = cbq_class_lookup(q, prio)) != NULL)
return cl;
*qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
for (;;) {
int result = 0;
defmap = head->defaults;
fl = rcu_dereference_bh(head->filter_list);
/*
* Step 2+n. Apply classifier.
*/
result = tc_classify_compat(skb, fl, &res);
if (!fl || result < 0)
goto fallback;
cl = (void *)res.class;
if (!cl) {
if (TC_H_MAJ(res.classid))
cl = cbq_class_lookup(q, res.classid);
else if ((cl = defmap[res.classid & TC_PRIO_MAX]) == NULL)
cl = defmap[TC_PRIO_BESTEFFORT];
if (cl == NULL)
goto fallback;
}
if (cl->level >= head->level)
goto fallback;
#ifdef CONFIG_NET_CLS_ACT
switch (result) {
case TC_ACT_QUEUED:
case TC_ACT_STOLEN:
*qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
case TC_ACT_SHOT:
return NULL;
case TC_ACT_RECLASSIFY:
return cbq_reclassify(skb, cl);
}
#endif
if (cl->level == 0)
return cl;
/*
* Step 3+n. If classifier selected a link sharing class,
* apply agency specific classifier.
* Repeat this procdure until we hit a leaf node.
*/
head = cl;
}
fallback:
cl = head;
/*
* Step 4. No success...
*/
if (TC_H_MAJ(prio) == 0 &&
!(cl = head->defaults[prio & TC_PRIO_MAX]) &&
!(cl = head->defaults[TC_PRIO_BESTEFFORT]))
return head;
return cl;
}
/*
* A packet has just been enqueued on the empty class.
* cbq_activate_class adds it to the tail of active class list
* of its priority band.
*/
static inline void cbq_activate_class(struct cbq_class *cl)
{
struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
int prio = cl->cpriority;
struct cbq_class *cl_tail;
cl_tail = q->active[prio];
q->active[prio] = cl;
if (cl_tail != NULL) {
cl->next_alive = cl_tail->next_alive;
cl_tail->next_alive = cl;
} else {
cl->next_alive = cl;
q->activemask |= (1<<prio);
}
}
/*
* Unlink class from active chain.
* Note that this same procedure is done directly in cbq_dequeue*
* during round-robin procedure.
*/
static void cbq_deactivate_class(struct cbq_class *this)
{
struct cbq_sched_data *q = qdisc_priv(this->qdisc);
int prio = this->cpriority;
struct cbq_class *cl;
struct cbq_class *cl_prev = q->active[prio];
do {
cl = cl_prev->next_alive;
if (cl == this) {
cl_prev->next_alive = cl->next_alive;
cl->next_alive = NULL;
if (cl == q->active[prio]) {
q->active[prio] = cl_prev;
if (cl == q->active[prio]) {
q->active[prio] = NULL;
q->activemask &= ~(1<<prio);
return;
}
}
return;
}
} while ((cl_prev = cl) != q->active[prio]);
}
static void
cbq_mark_toplevel(struct cbq_sched_data *q, struct cbq_class *cl)
{
int toplevel = q->toplevel;
if (toplevel > cl->level && !(qdisc_is_throttled(cl->q))) {
psched_time_t now = psched_get_time();
do {
if (cl->undertime < now) {
q->toplevel = cl->level;
return;
}
} while ((cl = cl->borrow) != NULL && toplevel > cl->level);
}
}
static int
cbq_enqueue(struct sk_buff *skb, struct Qdisc *sch)
{
struct cbq_sched_data *q = qdisc_priv(sch);
int uninitialized_var(ret);
struct cbq_class *cl = cbq_classify(skb, sch, &ret);
#ifdef CONFIG_NET_CLS_ACT
q->rx_class = cl;
#endif
if (cl == NULL) {
if (ret & __NET_XMIT_BYPASS)
qdisc_qstats_drop(sch);
kfree_skb(skb);
return ret;
}
#ifdef CONFIG_NET_CLS_ACT
cl->q->__parent = sch;
#endif
ret = qdisc_enqueue(skb, cl->q);
if (ret == NET_XMIT_SUCCESS) {
sch->q.qlen++;
cbq_mark_toplevel(q, cl);
if (!cl->next_alive)
cbq_activate_class(cl);
return ret;
}
if (net_xmit_drop_count(ret)) {
qdisc_qstats_drop(sch);
cbq_mark_toplevel(q, cl);
cl->qstats.drops++;
}
return ret;
}
/* Overlimit actions */
/* TC_CBQ_OVL_CLASSIC: (default) penalize leaf class by adding offtime */
static void cbq_ovl_classic(struct cbq_class *cl)
{
struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
psched_tdiff_t delay = cl->undertime - q->now;
if (!cl->delayed) {
delay += cl->offtime;
/*
* Class goes to sleep, so that it will have no
* chance to work avgidle. Let's forgive it 8)
*
* BTW cbq-2.0 has a crap in this
* place, apparently they forgot to shift it by cl->ewma_log.
*/
if (cl->avgidle < 0)
delay -= (-cl->avgidle) - ((-cl->avgidle) >> cl->ewma_log);
if (cl->avgidle < cl->minidle)
cl->avgidle = cl->minidle;
if (delay <= 0)
delay = 1;
cl->undertime = q->now + delay;
cl->xstats.overactions++;
cl->delayed = 1;
}
if (q->wd_expires == 0 || q->wd_expires > delay)
q->wd_expires = delay;
/* Dirty work! We must schedule wakeups based on
* real available rate, rather than leaf rate,
* which may be tiny (even zero).
*/
if (q->toplevel == TC_CBQ_MAXLEVEL) {
struct cbq_class *b;
psched_tdiff_t base_delay = q->wd_expires;
for (b = cl->borrow; b; b = b->borrow) {
delay = b->undertime - q->now;
if (delay < base_delay) {
if (delay <= 0)
delay = 1;
base_delay = delay;
}
}
q->wd_expires = base_delay;
}
}
/* TC_CBQ_OVL_RCLASSIC: penalize by offtime classes in hierarchy, when
* they go overlimit
*/
static void cbq_ovl_rclassic(struct cbq_class *cl)
{
struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
struct cbq_class *this = cl;
do {
if (cl->level > q->toplevel) {
cl = NULL;
break;
}
} while ((cl = cl->borrow) != NULL);
if (cl == NULL)
cl = this;
cbq_ovl_classic(cl);
}
/* TC_CBQ_OVL_DELAY: delay until it will go to underlimit */
static void cbq_ovl_delay(struct cbq_class *cl)
{
struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
psched_tdiff_t delay = cl->undertime - q->now;
if (test_bit(__QDISC_STATE_DEACTIVATED,
&qdisc_root_sleeping(cl->qdisc)->state))
return;
if (!cl->delayed) {
psched_time_t sched = q->now;
ktime_t expires;
delay += cl->offtime;
if (cl->avgidle < 0)
delay -= (-cl->avgidle) - ((-cl->avgidle) >> cl->ewma_log);
if (cl->avgidle < cl->minidle)
cl->avgidle = cl->minidle;
cl->undertime = q->now + delay;
if (delay > 0) {
sched += delay + cl->penalty;
cl->penalized = sched;
cl->cpriority = TC_CBQ_MAXPRIO;
q->pmask |= (1<<TC_CBQ_MAXPRIO);
expires = ns_to_ktime(PSCHED_TICKS2NS(sched));
if (hrtimer_try_to_cancel(&q->delay_timer) &&
ktime_to_ns(ktime_sub(
hrtimer_get_expires(&q->delay_timer),
expires)) > 0)
hrtimer_set_expires(&q->delay_timer, expires);
hrtimer_restart(&q->delay_timer);
cl->delayed = 1;
cl->xstats.overactions++;
return;
}
delay = 1;
}
if (q->wd_expires == 0 || q->wd_expires > delay)
q->wd_expires = delay;
}
/* TC_CBQ_OVL_LOWPRIO: penalize class by lowering its priority band */
static void cbq_ovl_lowprio(struct cbq_class *cl)
{
struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
cl->penalized = q->now + cl->penalty;
if (cl->cpriority != cl->priority2) {
cl->cpriority = cl->priority2;
q->pmask |= (1<<cl->cpriority);
cl->xstats.overactions++;
}
cbq_ovl_classic(cl);
}
/* TC_CBQ_OVL_DROP: penalize class by dropping */
static void cbq_ovl_drop(struct cbq_class *cl)
{
if (cl->q->ops->drop)
if (cl->q->ops->drop(cl->q))
cl->qdisc->q.qlen--;
cl->xstats.overactions++;
cbq_ovl_classic(cl);
}
static psched_tdiff_t cbq_undelay_prio(struct cbq_sched_data *q, int prio,
psched_time_t now)
{
struct cbq_class *cl;
struct cbq_class *cl_prev = q->active[prio];
psched_time_t sched = now;
if (cl_prev == NULL)
return 0;
do {
cl = cl_prev->next_alive;
if (now - cl->penalized > 0) {
cl_prev->next_alive = cl->next_alive;
cl->next_alive = NULL;
cl->cpriority = cl->priority;
cl->delayed = 0;
cbq_activate_class(cl);
if (cl == q->active[prio]) {
q->active[prio] = cl_prev;
if (cl == q->active[prio]) {
q->active[prio] = NULL;
return 0;
}
}
cl = cl_prev->next_alive;
} else if (sched - cl->penalized > 0)
sched = cl->penalized;
} while ((cl_prev = cl) != q->active[prio]);
return sched - now;
}
static enum hrtimer_restart cbq_undelay(struct hrtimer *timer)
{
struct cbq_sched_data *q = container_of(timer, struct cbq_sched_data,
delay_timer);
struct Qdisc *sch = q->watchdog.qdisc;
psched_time_t now;
psched_tdiff_t delay = 0;
unsigned int pmask;
now = psched_get_time();
pmask = q->pmask;
q->pmask = 0;
while (pmask) {
int prio = ffz(~pmask);
psched_tdiff_t tmp;
pmask &= ~(1<<prio);
tmp = cbq_undelay_prio(q, prio, now);
if (tmp > 0) {
q->pmask |= 1<<prio;
if (tmp < delay || delay == 0)
delay = tmp;
}
}
if (delay) {
ktime_t time;
time = ktime_set(0, 0);
time = ktime_add_ns(time, PSCHED_TICKS2NS(now + delay));
hrtimer_start(&q->delay_timer, time, HRTIMER_MODE_ABS_PINNED);
}
qdisc_unthrottled(sch);
__netif_schedule(qdisc_root(sch));
return HRTIMER_NORESTART;
}
#ifdef CONFIG_NET_CLS_ACT
static int cbq_reshape_fail(struct sk_buff *skb, struct Qdisc *child)
{
struct Qdisc *sch = child->__parent;
struct cbq_sched_data *q = qdisc_priv(sch);
struct cbq_class *cl = q->rx_class;
q->rx_class = NULL;
if (cl && (cl = cbq_reclassify(skb, cl)) != NULL) {
int ret;
cbq_mark_toplevel(q, cl);
q->rx_class = cl;
cl->q->__parent = sch;
ret = qdisc_enqueue(skb, cl->q);
if (ret == NET_XMIT_SUCCESS) {
sch->q.qlen++;
if (!cl->next_alive)
cbq_activate_class(cl);
return 0;
}
if (net_xmit_drop_count(ret))
qdisc_qstats_drop(sch);
return 0;
}
qdisc_qstats_drop(sch);
return -1;
}
#endif
/*
* It is mission critical procedure.
*
* We "regenerate" toplevel cutoff, if transmitting class
* has backlog and it is not regulated. It is not part of
* original CBQ description, but looks more reasonable.
* Probably, it is wrong. This question needs further investigation.
*/
static inline void
cbq_update_toplevel(struct cbq_sched_data *q, struct cbq_class *cl,
struct cbq_class *borrowed)
{
if (cl && q->toplevel >= borrowed->level) {
if (cl->q->q.qlen > 1) {
do {
if (borrowed->undertime == PSCHED_PASTPERFECT) {
q->toplevel = borrowed->level;
return;
}
} while ((borrowed = borrowed->borrow) != NULL);
}
#if 0
/* It is not necessary now. Uncommenting it
will save CPU cycles, but decrease fairness.
*/
q->toplevel = TC_CBQ_MAXLEVEL;
#endif
}
}
static void
cbq_update(struct cbq_sched_data *q)
{
struct cbq_class *this = q->tx_class;
struct cbq_class *cl = this;
int len = q->tx_len;
psched_time_t now;
q->tx_class = NULL;
/* Time integrator. We calculate EOS time
* by adding expected packet transmission time.
*/
now = q->now + L2T(&q->link, len);
for ( ; cl; cl = cl->share) {
long avgidle = cl->avgidle;
long idle;
cl->bstats.packets++;
cl->bstats.bytes += len;
/*
* (now - last) is total time between packet right edges.
* (last_pktlen/rate) is "virtual" busy time, so that
*
* idle = (now - last) - last_pktlen/rate
*/
idle = now - cl->last;
if ((unsigned long)idle > 128*1024*1024) {
avgidle = cl->maxidle;
} else {
idle -= L2T(cl, len);
/* true_avgidle := (1-W)*true_avgidle + W*idle,
* where W=2^{-ewma_log}. But cl->avgidle is scaled:
* cl->avgidle == true_avgidle/W,
* hence:
*/
avgidle += idle - (avgidle>>cl->ewma_log);
}
if (avgidle <= 0) {
/* Overlimit or at-limit */
if (avgidle < cl->minidle)
avgidle = cl->minidle;
cl->avgidle = avgidle;
/* Calculate expected time, when this class
* will be allowed to send.
* It will occur, when:
* (1-W)*true_avgidle + W*delay = 0, i.e.
* idle = (1/W - 1)*(-true_avgidle)
* or
* idle = (1 - W)*(-cl->avgidle);
*/
idle = (-avgidle) - ((-avgidle) >> cl->ewma_log);
/*
* That is not all.
* To maintain the rate allocated to the class,
* we add to undertime virtual clock,
* necessary to complete transmitted packet.
* (len/phys_bandwidth has been already passed
* to the moment of cbq_update)
*/
idle -= L2T(&q->link, len);
idle += L2T(cl, len);
cl->undertime = now + idle;
} else {
/* Underlimit */
cl->undertime = PSCHED_PASTPERFECT;
if (avgidle > cl->maxidle)
cl->avgidle = cl->maxidle;
else
cl->avgidle = avgidle;
}
if ((s64)(now - cl->last) > 0)
cl->last = now;
}
cbq_update_toplevel(q, this, q->tx_borrowed);
}
static inline struct cbq_class *
cbq_under_limit(struct cbq_class *cl)
{
struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
struct cbq_class *this_cl = cl;
if (cl->tparent == NULL)
return cl;
if (cl->undertime == PSCHED_PASTPERFECT || q->now >= cl->undertime) {
cl->delayed = 0;
return cl;
}
do {
/* It is very suspicious place. Now overlimit
* action is generated for not bounded classes
* only if link is completely congested.
* Though it is in agree with ancestor-only paradigm,
* it looks very stupid. Particularly,
* it means that this chunk of code will either
* never be called or result in strong amplification
* of burstiness. Dangerous, silly, and, however,
* no another solution exists.
*/
cl = cl->borrow;
if (!cl) {
this_cl->qstats.overlimits++;
this_cl->overlimit(this_cl);
return NULL;
}
if (cl->level > q->toplevel)
return NULL;
} while (cl->undertime != PSCHED_PASTPERFECT && q->now < cl->undertime);
cl->delayed = 0;
return cl;
}
static inline struct sk_buff *
cbq_dequeue_prio(struct Qdisc *sch, int prio)
{
struct cbq_sched_data *q = qdisc_priv(sch);
struct cbq_class *cl_tail, *cl_prev, *cl;
struct sk_buff *skb;
int deficit;
cl_tail = cl_prev = q->active[prio];
cl = cl_prev->next_alive;
do {
deficit = 0;
/* Start round */
do {
struct cbq_class *borrow = cl;
if (cl->q->q.qlen &&
(borrow = cbq_under_limit(cl)) == NULL)
goto skip_class;
if (cl->deficit <= 0) {
/* Class exhausted its allotment per
* this round. Switch to the next one.
*/
deficit = 1;
cl->deficit += cl->quantum;
goto next_class;
}
skb = cl->q->dequeue(cl->q);
/* Class did not give us any skb :-(
* It could occur even if cl->q->q.qlen != 0
* f.e. if cl->q == "tbf"
*/
if (skb == NULL)
goto skip_class;
cl->deficit -= qdisc_pkt_len(skb);
q->tx_class = cl;
q->tx_borrowed = borrow;
if (borrow != cl) {
#ifndef CBQ_XSTATS_BORROWS_BYTES
borrow->xstats.borrows++;
cl->xstats.borrows++;
#else
borrow->xstats.borrows += qdisc_pkt_len(skb);
cl->xstats.borrows += qdisc_pkt_len(skb);
#endif
}
q->tx_len = qdisc_pkt_len(skb);
if (cl->deficit <= 0) {
q->active[prio] = cl;
cl = cl->next_alive;
cl->deficit += cl->quantum;
}
return skb;
skip_class:
if (cl->q->q.qlen == 0 || prio != cl->cpriority) {
/* Class is empty or penalized.
* Unlink it from active chain.
*/
cl_prev->next_alive = cl->next_alive;
cl->next_alive = NULL;
/* Did cl_tail point to it? */
if (cl == cl_tail) {
/* Repair it! */
cl_tail = cl_prev;
/* Was it the last class in this band? */
if (cl == cl_tail) {
/* Kill the band! */
q->active[prio] = NULL;
q->activemask &= ~(1<<prio);
if (cl->q->q.qlen)
cbq_activate_class(cl);
return NULL;
}
q->active[prio] = cl_tail;
}
if (cl->q->q.qlen)
cbq_activate_class(cl);
cl = cl_prev;
}
next_class:
cl_prev = cl;
cl = cl->next_alive;
} while (cl_prev != cl_tail);
} while (deficit);
q->active[prio] = cl_prev;
return NULL;
}
static inline struct sk_buff *
cbq_dequeue_1(struct Qdisc *sch)
{
struct cbq_sched_data *q = qdisc_priv(sch);
struct sk_buff *skb;
unsigned int activemask;
activemask = q->activemask & 0xFF;
while (activemask) {
int prio = ffz(~activemask);
activemask &= ~(1<<prio);
skb = cbq_dequeue_prio(sch, prio);
if (skb)
return skb;
}
return NULL;
}
static struct sk_buff *
cbq_dequeue(struct Qdisc *sch)
{
struct sk_buff *skb;
struct cbq_sched_data *q = qdisc_priv(sch);
psched_time_t now;
now = psched_get_time();
if (q->tx_class)
cbq_update(q);
q->now = now;
for (;;) {
q->wd_expires = 0;
skb = cbq_dequeue_1(sch);
if (skb) {
qdisc_bstats_update(sch, skb);
sch->q.qlen--;
qdisc_unthrottled(sch);
return skb;
}
/* All the classes are overlimit.
*
* It is possible, if:
*
* 1. Scheduler is empty.
* 2. Toplevel cutoff inhibited borrowing.
* 3. Root class is overlimit.
*
* Reset 2d and 3d conditions and retry.
*
* Note, that NS and cbq-2.0 are buggy, peeking
* an arbitrary class is appropriate for ancestor-only
* sharing, but not for toplevel algorithm.
*
* Our version is better, but slower, because it requires
* two passes, but it is unavoidable with top-level sharing.
*/
if (q->toplevel == TC_CBQ_MAXLEVEL &&
q->link.undertime == PSCHED_PASTPERFECT)
break;
q->toplevel = TC_CBQ_MAXLEVEL;
q->link.undertime = PSCHED_PASTPERFECT;
}
/* No packets in scheduler or nobody wants to give them to us :-(
* Sigh... start watchdog timer in the last case.
*/
if (sch->q.qlen) {
qdisc_qstats_overlimit(sch);
if (q->wd_expires)
qdisc_watchdog_schedule(&q->watchdog,