summaryrefslogtreecommitdiff
path: root/net/sched/sch_cbq.c
blob: 481036f6b54e4730ee27fae6236277c64d3eaa1a (plain)
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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
/*
 * 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, <kuznet@ms2.inr.ac.ru>
 *
 */

#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>
#include <net/pkt_cls.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 */

	u32			defmap;

	/* Link-sharing scheduler parameters */
	long			maxidle;	/* Class parameters: see below. */
	long			offtime;
	long			minidle;
	u32			avpkt;
	struct qdisc_rate_table	*R_tab;

	/* 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 net_rate_estimator __rcu *rate_est;
	struct tc_cbq_xstats	xstats;

	struct tcf_proto __rcu	*filter_list;
	struct tcf_block	*block;

	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 = tcf_classify(skb, fl, &res, true);
		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:
		case TC_ACT_TRAP:
			*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) {
		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 sk_buff **to_free)
{
	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);
		__qdisc_drop(skb, to_free);
		return ret;
	}

	ret = qdisc_enqueue(skb, cl->q, to_free);
	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 action: penalize leaf class by adding offtime */
static void cbq_overlimit(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;
	}
}

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 = 0;
		time = ktime_add_ns(time, PSCHED_TICKS2NS(now + delay));
		hrtimer_start(&q->delay_timer, time, HRTIMER_MODE_ABS_PINNED);
	}

	__netif_schedule(qdisc_root(sch));
	return HRTIMER_NORESTART;
}

/*
 * 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++;
			cbq_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--;
			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,
						now + q->wd_expires);
	}
	return NULL;
}

/* CBQ class maintanance routines */

static void cbq_adjust_levels(struct cbq_class *this)
{
	if (this == NULL)
		return;

	do {
		int level = 0;
		struct cbq_class *cl;

		cl = this->children;
		if (cl) {
			do {
				if (cl->level > level)
					level = cl->level;
			} while ((cl = cl->sibling) != this->children);
		}
		this->level = level + 1;
	} while ((this = this->tparent) != NULL);
}

static void cbq_normalize_quanta(struct cbq_sched_data *q, int prio)
{
	struct cbq_class *cl;
	unsigned int h;

	if (q->quanta[prio] == 0)
		return;

	for (h = 0; h < q->clhash.hashsize; h++) {
		hlist_for_each_entry(cl, &q->clhash.hash[h], common.hnode) {
			/* BUGGGG... Beware! This expression suffer of
			 * arithmetic overflows!
			 */
			if (cl->priority == prio) {
				cl->quantum = (cl->weight*cl->allot*q->nclasses[prio])/
					q->quanta[prio];
			}
			if (cl->quantum <= 0 ||
			    cl->quantum > 32*qdisc_dev(cl->qdisc)->mtu) {
				pr_warn("CBQ: class %08x has bad quantum==%ld, repaired.\n",
					cl->common.classid, cl->quantum);
				cl->quantum = qdisc_dev(cl->qdisc)->mtu/2 + 1;
			}
		}
	}
}

static void cbq_sync_defmap(struct cbq_class *cl)
{
	struct cbq_sched_data *q = qdisc_priv(cl->qdisc);
	struct cbq_class *split = cl->split;
	unsigned int h;
	int i;

	if (split == NULL)
		return;

	for (i = 0; i <= TC_PRIO_MAX; i++) {
		if (split->defaults[i] == cl && !(cl->defmap & (1<<i)))
			split->defaults[i] = NULL;
	}

	for (i = 0; i <= TC_PRIO_MAX; i++) {
		int level = split->level;

		if (split->defaults[i])
			continue;

		for (h = 0; h < q->clhash.hashsize; h++) {
			struct cbq_class *c;

			hlist_for_each_entry(c, &q->clhash.hash[h],
					     common.hnode) {
				if (c->split == split && c->level < level &&
				    c->defmap & (1<<i)) {
					split->defaults[i] = c;
					level = c->level;
				}
			}
		}
	}
}

static void cbq_change_defmap(struct cbq_class *cl, u32 splitid, u32 def, u32 mask)
{
	struct cbq_class *split = NULL;

	if (splitid == 0) {
		split = cl->split;
		if (!split)
			return;
		splitid = split->common.classid;
	}

	if (split == NULL || split->common.classid != splitid) {
		for (split = cl->tparent; split; split = split->tparent)
			if (split->common.classid == splitid)
				break;
	}

	if (split == NULL)
		return;

	if (cl->split != split) {
		cl->defmap = 0;
		cbq_sync_defmap(cl);
		cl->split = split;
		cl->defmap = def & mask;
	} else
		cl->defmap = (cl->defmap & ~mask) | (def & mask);

	cbq_sync_defmap(cl);
}

static void cbq_unlink_class(struct cbq_class *this)
{
	struct cbq_class *cl, **clp;
	struct cbq_sched_data *q = qdisc_priv(this->qdisc);

	qdisc_class_hash_remove(&q->clhash, &this->common);

	if (this->tparent) {
		clp = &this->sibling;
		cl = *clp;
		do {
			if (cl == this) {
				*clp = cl->sibling;
				break;
			}
			clp = &cl->sibling;
		} while ((cl = *clp) != this->sibling);

		if (this->tparent->children == this) {
			this->tparent->children = this->sibling;
			if (this->sibling == this)
				this->tparent->children = NULL;
		}
	} else {
		WARN_ON(this->sibling != this);
	}
}

static void cbq_link_class(struct cbq_class *this)
{
	struct cbq_sched_data *q = qdisc_priv(this->qdisc);
	struct cbq_class *parent = this->tparent;

	this->sibling = this;
	qdisc_class_hash_insert(&q->clhash, &this->common);

	if (parent == NULL)
		return;

	if (parent->children == NULL) {
		parent->children = this;
	} else {
		this->sibling = parent->children->sibling;
		parent->children->sibling = this;
	}
}

static void
cbq_reset(struct Qdisc *sch)
{
	struct cbq_sched_data *q = qdisc_priv(sch);
	struct cbq_class *cl;
	int prio;
	unsigned int h;

	q->activemask = 0;
	q->pmask = 0;
	q->tx_class = NULL;
	q->tx_borrowed = NULL;
	qdisc_watchdog_cancel(&q->watchdog);
	hrtimer_cancel(&q->delay_timer);
	q->toplevel = TC_CBQ_MAXLEVEL;
	q->now = psched_get_time();

	for (prio = 0; prio <= TC_CBQ_MAXPRIO; prio++)
		q->active[prio] = NULL;

	for (h = 0; h < q->clhash.hashsize; h++) {
		hlist_for_each_entry(cl, &q->clhash.hash[h], common.hnode) {
			qdisc_reset(cl->q);

			cl->next_alive = NULL;
			cl->undertime = PSCHED_PASTPERFECT;
			cl->avgidle = cl->maxidle;
			cl->deficit = cl->quantum;
			cl->cpriority = cl->priority;
		}
	}
	sch->q.qlen = 0;
}


static int cbq_set_lss(struct cbq_class *cl, struct tc_cbq_lssopt *lss)
{
	if (lss->change & TCF_CBQ_LSS_FLAGS) {
		cl->share = (lss->flags & TCF_CBQ_LSS_ISOLATED) ? NULL : cl->tparent;
		cl->borrow = (lss->flags & TCF_CBQ_LSS_BOUNDED) ? NULL : cl->tparent;
	}
	if (lss->change & TCF_CBQ_LSS_EWMA)
		cl->ewma_log = lss->ewma_log;
	if (lss->change & TCF_CBQ_LSS_AVPKT)
		cl->avpkt = lss->avpkt;
	if (lss->change & TCF_CBQ_LSS_MINIDLE)
		cl->minidle = -(long)lss->minidle;
	if (lss->change & TCF_CBQ_LSS_MAXIDLE) {
		cl->maxidle = lss->maxidle;
		cl->avgidle = lss->maxidle;
	}
	if (lss->change & TCF_CBQ_LSS_OFFTIME)
		cl->offtime = lss->offtime;
	return 0;
}

static void cbq_rmprio(struct cbq_sched_data *q, struct cbq_class *cl)
{
	q->nclasses[cl->priority]--;
	q->quanta[cl->priority] -= cl->weight;
	cbq_normalize_quanta(q, cl->priority);
}

static void cbq_addprio(struct cbq_sched_data *q, struct cbq_class *cl)
{
	q->nclasses[cl->priority]++;
	q->quanta[cl->priority] += cl->weight;
	cbq_normalize_quanta(q, cl->priority);
}

static int cbq_set_wrr(struct cbq_class *cl, struct tc_cbq_wrropt *wrr)
{
	struct cbq_sched_data *q = qdisc_priv(cl->qdisc);

	if (wrr->allot)
		cl->allot = wrr->allot;
	if (wrr->weight)
		cl->weight = wrr->weight;
	if (wrr->priority) {
		cl->priority = wrr->priority - 1;
		cl->cpriority = cl->priority;
		if (cl->priority >= cl->priority2)
			cl->priority2 = TC_CBQ_MAXPRIO - 1;
	}

	cbq_addprio(q, cl);
	return 0;
}

static int cbq_set_fopt(struct cbq_class *cl, struct tc_cbq_fopt *fopt)
{
	cbq_change_defmap(cl, fopt->split, fopt->defmap, fopt->defchange);
	return 0;
}

static const struct nla_policy cbq_policy[TCA_CBQ_MAX + 1] = {
	[TCA_CBQ_LSSOPT]	= { .len = sizeof(struct tc_cbq_lssopt) },
	[TCA_CBQ_WRROPT]	= { .len = sizeof(struct tc_cbq_wrropt) },
	[TCA_CBQ_FOPT]		= { .len = sizeof(struct tc_cbq_fopt) },
	[TCA_CBQ_OVL_STRATEGY]	= { .len = sizeof(struct tc_cbq_ovl) },
	[TCA_CBQ_RATE]		= { .len = sizeof(struct tc_ratespec) },
	[TCA_CBQ_RTAB]		= { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
	[TCA_CBQ_POLICE]	= { .len = sizeof(struct tc_cbq_police) },
};

static int cbq_init(struct Qdisc *sch, struct nlattr *opt)
{
	struct cbq_sched_data *q = qdisc_priv(sch);
	struct nlattr *tb[TCA_CBQ_MAX + 1];
	struct tc_ratespec *r;
	int err;

	err = nla_parse_nested(tb, TCA_CBQ_MAX, opt, cbq_policy, NULL);
	if (err < 0)
		return err;

	if (tb[TCA_CBQ_RTAB] == NULL || tb[TCA_CBQ_RATE] == NULL)
		return -EINVAL;

	r = nla_data(tb[TCA_CBQ_RATE]);

	if ((q->link.R_tab = qdisc_get_rtab(r, tb[TCA_CBQ_RTAB])) == NULL)
		return -EINVAL;

	err = qdisc_class_hash_init(&q->clhash);
	if (err < 0)
		goto put_rtab;

	q->link.refcnt = 1;
	q->link.sibling = &q->link;
	q->link.common.classid = sch->handle;
	q->link.qdisc = sch;
	q->link.q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
				      sch->handle);
	if (!q->link.q)
		q->link.q = &noop_qdisc;
	else
		qdisc_hash_add(q->link.q, true);

	q->link.priority = TC_CBQ_MAXPRIO - 1;
	q->link.priority2 = TC_CBQ_MAXPRIO - 1;
	q->link.cpriority = TC_CBQ_MAXPRIO - 1;
	q->link.allot = psched_mtu(qdisc_dev(sch));
	q->link.quantum = q->link.allot;
	q->link.weight = q->link.R_tab->rate.rate;

	q->link.ewma_log = TC_CBQ_DEF_EWMA;
	q->link.avpkt = q->link.allot/2;
	q->link.minidle = -0x7FFFFFFF;

	qdisc_watchdog_init(&q->watchdog, sch);
	hrtimer_init(&q->delay_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
	q->delay_timer.function = cbq_undelay;
	q->toplevel = TC_CBQ_MAXLEVEL;
	q->now = psched_get_time();

	cbq_link_class(&q->link);

	if (tb[TCA_CBQ_LSSOPT])
		cbq_set_lss(&q->link, nla_data(tb[TCA_CBQ_LSSOPT]));

	cbq_addprio(q, &q->link);
	return 0;

put_rtab:
	qdisc_put_rtab(q->link.R_tab);
	return err;
}

static int cbq_dump_rate(struct sk_buff *skb, struct cbq_class *cl)
{
	unsigned char *b = skb_tail_pointer(skb);

	if (nla_put(skb, TCA_CBQ_RATE, sizeof(cl->R_tab->rate), &cl->R_tab->rate))
		goto nla_put_failure;
	return skb->len;

nla_put_failure:
	nlmsg_trim(skb, b);
	return -1;
}

static int cbq_dump_lss(struct sk_buff *skb, struct cbq_class *cl)
{
	unsigned char *b = skb_tail_pointer(skb);
	struct tc_cbq_lssopt opt;

	opt.flags = 0;
	if (cl->borrow == NULL)
		opt.flags |= TCF_CBQ_LSS_BOUNDED;
	if (cl->share == NULL)
		opt.flags |= TCF_CBQ_LSS_ISOLATED;
	opt.ewma_log = cl->ewma_log;
	opt.level = cl->level;
	opt.avpkt = cl->avpkt;
	opt.maxidle = cl->maxidle;
	opt.minidle = (u32)(-cl->minidle);
	opt.offtime = cl->offtime;
	opt.change = ~0;
	if (nla_put(skb, TCA_CBQ_LSSOPT, sizeof(opt), &opt))
		goto nla_put_failure;
	return skb->len;

nla_put_failure:
	nlmsg_trim(skb, b);
	return -1;
}

static int cbq_dump_wrr(struct sk_buff *skb, struct cbq_class *cl)
{
	unsigned char *b = skb_tail_pointer(skb);
	struct tc_cbq_wrropt opt;

	memset(&opt, 0, sizeof(opt));
	opt.flags = 0;
	opt.allot = cl->allot;
	opt.priority = cl->priority + 1;
	opt.cpriority = cl->cpriority + 1;
	opt.weight = cl->weight;
	if (nla_put(skb, TCA_CBQ_WRROPT, sizeof(opt), &opt))
		goto nla_put_failure;
	return skb->len;

nla_put_failure:
	nlmsg_trim(skb, b);
	return -1;
}

static int cbq_dump_fopt(struct sk_buff *skb, struct cbq_class *cl)
{
	unsigned char *b = skb_tail_pointer(skb);
	struct tc_cbq_fopt opt;

	if (cl->split || cl->defmap) {
		opt.split = cl->split ? cl->split->common.classid : 0;
		opt.defmap = cl->defmap;
		opt.defchange = ~0;
		if (nla_put(skb, TCA_CBQ_FOPT, sizeof(opt), &opt))
			goto nla_put_failure;
	}
	return skb->len;

nla_put_failure:
	nlmsg_trim(skb, b);
	return -1;
}

static int cbq_dump_attr(struct sk_buff *skb, struct cbq_class *cl)
{
	if (cbq_dump_lss(skb, cl) < 0 ||
	    cbq_dump_rate(skb, cl) < 0 ||
	    cbq_dump_wrr(skb, cl) < 0 ||
	    cbq_dump_fopt(skb, cl) < 0)
		return -1;
	return 0;
}

static int cbq_dump(struct Qdisc *sch, struct sk_buff *skb)
{
	struct cbq_sched_data *q = qdisc_priv(sch);
	struct nlattr *nest;

	nest = nla_nest_start(skb, TCA_OPTIONS);
	if (nest == NULL)
		goto nla_put_failure;
	if (cbq_dump_attr(skb, &q->link) < 0)
		goto nla_put_failure;
	return nla_nest_end(skb, nest);

nla_put_failure:
	nla_nest_cancel(skb, nest);
	return -1;
}

static int
cbq_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
{
	struct cbq_sched_data *q = qdisc_priv(sch);

	q->link.xstats.avgidle = q->link.avgidle;
	return gnet_stats_copy_app(d, &q->link.xstats, sizeof(q->link.xstats));
}

static int
cbq_dump_class(struct Qdisc *sch, unsigned long arg,
	       struct sk_buff *skb, struct tcmsg *tcm)
{
	struct cbq_class *cl = (struct cbq_class *)arg;
	struct nlattr *nest;

	if (cl->tparent)
		tcm->tcm_parent = cl->tparent->common.classid;
	else
		tcm->tcm_parent = TC_H_ROOT;
	tcm->tcm_handle = cl->common.classid;
	tcm->tcm_info = cl->q->handle;

	nest = nla_nest_start(skb, TCA_OPTIONS);
	if (nest == NULL)
		goto nla_put_failure;
	if (cbq_dump_attr(skb, cl) < 0)
		goto nla_put_failure;
	return nla_nest_end(skb, nest);

nla_put_failure:
	nla_nest_cancel(skb, nest);
	return -1;
}

static int
cbq_dump_class_stats(struct Qdisc *sch, unsigned long arg,
	struct gnet_dump *d)
{
	struct cbq_sched_data *q = qdisc_priv(sch);
	struct cbq_class *cl = (struct cbq_class *)arg;

	cl->xstats.avgidle = cl->avgidle;
	cl->xstats.undertime = 0;

	if (cl->undertime != PSCHED_PASTPERFECT)
		cl->xstats.undertime = cl->undertime - q->now;

	if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
				  d, NULL, &cl->bstats) < 0 ||
	    gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
	    gnet_stats_copy_queue(d, NULL, &cl->qstats, cl->q->q.qlen) < 0)
		return -1;

	return gnet_stats_copy_app(d, &cl->xstats, sizeof(cl->xstats));
}

static int cbq_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
		     struct Qdisc **old)
{
	struct cbq_class *cl = (struct cbq_class *)arg;

	if (new == NULL) {
		new = qdisc_create_dflt(sch->dev_queue,
					&pfifo_qdisc_ops, cl->common.classid);
		if (new == NULL)
			return -ENOBUFS;
	}

	*old = qdisc_replace(sch, new, &cl->q);
	return 0;
}

static struct Qdisc *cbq_leaf(struct Qdisc *sch, unsigned long arg)
{
	struct cbq_class *cl = (struct cbq_class *)arg;

	return cl->q;
}

static void cbq_qlen_notify(struct Qdisc *sch, unsigned long arg)
{
	struct cbq_class *cl = (struct cbq_class *)arg;

	if (cl->q->q.qlen == 0)
		cbq_deactivate_class(cl);
}

static unsigned long cbq_get(struct Qdisc *sch, u32 classid)
{
	struct cbq_sched_data *q = qdisc_priv(sch);
	struct cbq_class *cl = cbq_class_lookup(q, classid);

	if (cl) {
		cl->refcnt++;
		return (unsigned long)cl;
	}
	return 0;
}

static void cbq_destroy_class(struct Qdisc *sch, struct cbq_class *cl)
{
	struct cbq_sched_data *q = qdisc_priv(sch);

	WARN_ON(cl->filters);

	tcf_block_put(cl->block);
	qdisc_destroy(cl->q);
	qdisc_put_rtab(cl->R_tab);
	gen_kill_estimator(&cl->rate_est);
	if (cl != &q->link)
		kfree(cl);
}

static void cbq_destroy(struct Qdisc *sch)
{
	struct cbq_sched_data *q = qdisc_priv(sch);
	struct hlist_node *next;
	struct cbq_class *cl;
	unsigned int h;

#ifdef CONFIG_NET_CLS_ACT
	q->rx_class = NULL;
#endif
	/*
	 * Filters must be destroyed first because we don't destroy the
	 * classes from root to leafs which means that filters can still
	 * be bound to classes which have been destroyed already. --TGR '04
	 */
	for (h = 0; h < q->clhash.hashsize; h++) {
		hlist_for_each_entry(cl, &q->clhash.hash[h], common.hnode)
			tcf_block_put(cl->block);
	}
	for (h = 0; h < q->clhash.hashsize; h++) {
		hlist_for_each_entry_safe(cl, next, &q->clhash.hash[h],
					  common.hnode)
			cbq_destroy_class(sch, cl);
	}
	qdisc_class_hash_destroy(&q->clhash);
}

static void cbq_put(struct Qdisc *sch, unsigned long arg)
{
	struct cbq_class *cl = (struct cbq_class *)arg;

	if (--cl->refcnt == 0) {
#ifdef CONFIG_NET_CLS_ACT
		spinlock_t *root_lock = qdisc_root_sleeping_lock(sch);
		struct cbq_sched_data *q = qdisc_priv(sch);

		spin_lock_bh(root_lock);
		if (q->rx_class == cl)
			q->rx_class = NULL;
		spin_unlock_bh(root_lock);
#endif

		cbq_destroy_class(sch, cl);
	}
}

static int
cbq_change_class(struct Qdisc *sch, u32 classid, u32 parentid, struct nlattr **tca,
		 unsigned long *arg)
{
	int err;
	struct cbq_sched_data *q = qdisc_priv(sch);
	struct cbq_class *cl = (struct cbq_class *)*arg;
	struct nlattr *opt = tca[TCA_OPTIONS];
	struct nlattr *tb[TCA_CBQ_MAX + 1];
	struct cbq_class *parent;
	struct qdisc_rate_table *rtab = NULL;

	if (opt == NULL)
		return -EINVAL;

	err = nla_parse_nested(tb, TCA_CBQ_MAX, opt, cbq_policy, NULL);
	if (err < 0)
		return err;

	if (tb[TCA_CBQ_OVL_STRATEGY] || tb[TCA_CBQ_POLICE])
		return -EOPNOTSUPP;

	if (cl) {
		/* Check parent */
		if (parentid) {
			if (cl->tparent &&
			    cl->tparent->common.classid != parentid)
				return -EINVAL;
			if (!cl->tparent && parentid != TC_H_ROOT)
				return -EINVAL;
		}

		if (tb[TCA_CBQ_RATE]) {
			rtab = qdisc_get_rtab(nla_data(tb[TCA_CBQ_RATE]),
					      tb[TCA_CBQ_RTAB]);
			if (rtab == NULL)
				return -EINVAL;
		}

		if (tca[TCA_RATE]) {
			err = gen_replace_estimator(&cl->bstats, NULL,
						    &cl->rate_est,
						    NULL,
						    qdisc_root_sleeping_running(sch),
						    tca[TCA_RATE]);
			if (err) {
				qdisc_put_rtab(rtab);
				return err;
			}
		}

		/* Change class parameters */
		sch_tree_lock(sch);

		if (cl->next_alive != NULL)
			cbq_deactivate_class(cl);

		if (rtab) {
			qdisc_put_rtab(cl->R_tab);
			cl->R_tab = rtab;
		}

		if (tb[TCA_CBQ_LSSOPT])
			cbq_set_lss(cl, nla_data(tb[TCA_CBQ_LSSOPT]));

		if (tb[TCA_CBQ_WRROPT]) {
			cbq_rmprio(q, cl);
			cbq_set_wrr(cl, nla_data(tb[TCA_CBQ_WRROPT]));
		}

		if (tb[TCA_CBQ_FOPT])
			cbq_set_fopt(cl, nla_data(tb[TCA_CBQ_FOPT]));

		if (cl->q->q.qlen)
			cbq_activate_class(cl);

		sch_tree_unlock(sch);

		return 0;
	}

	if (parentid == TC_H_ROOT)
		return -EINVAL;

	if (tb[TCA_CBQ_WRROPT] == NULL || tb[TCA_CBQ_RATE] == NULL ||
	    tb[TCA_CBQ_LSSOPT] == NULL)
		return -EINVAL;

	rtab = qdisc_get_rtab(nla_data(tb[TCA_CBQ_RATE]), tb[TCA_CBQ_RTAB]);
	if (rtab == NULL)
		return -EINVAL;

	if (classid) {
		err = -EINVAL;
		if (TC_H_MAJ(classid ^ sch->handle) ||
		    cbq_class_lookup(q, classid))
			goto failure;
	} else {
		int i;
		classid = TC_H_MAKE(sch->handle, 0x8000);

		for (i = 0; i < 0x8000; i++) {
			if (++q->hgenerator >= 0x8000)
				q->hgenerator = 1;
			if (cbq_class_lookup(q, classid|q->hgenerator) == NULL)
				break;
		}
		err = -ENOSR;
		if (i >= 0x8000)
			goto failure;
		classid = classid|q->hgenerator;
	}

	parent = &q->link;
	if (parentid) {
		parent = cbq_class_lookup(q, parentid);
		err = -EINVAL;
		if (parent == NULL)
			goto failure;
	}

	err = -ENOBUFS;
	cl = kzalloc(sizeof(*cl), GFP_KERNEL);
	if (cl == NULL)
		goto failure;

	err = tcf_block_get(&cl->block, &cl->filter_list);
	if (err) {
		kfree(cl);
		return err;
	}

	if (tca[TCA_RATE]) {
		err = gen_new_estimator(&cl->bstats, NULL, &cl->rate_est,
					NULL,
					qdisc_root_sleeping_running(sch),
					tca[TCA_RATE]);
		if (err) {
			tcf_block_put(cl->block);
			kfree(cl);
			goto failure;
		}
	}

	cl->R_tab = rtab;
	rtab = NULL;
	cl->refcnt = 1;
	cl->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, classid);
	if (!cl->q)
		cl->q = &noop_qdisc;
	else
		qdisc_hash_add(cl->q, true);

	cl->common.classid = classid;
	cl->tparent = parent;
	cl->qdisc = sch;
	cl->allot = parent->allot;
	cl->quantum = cl->allot;
	cl->weight = cl->R_tab->rate.rate;

	sch_tree_lock(sch);
	cbq_link_class(cl);
	cl->borrow = cl->tparent;
	if (cl->tparent != &q->link)
		cl->share = cl->tparent;
	cbq_adjust_levels(parent);
	cl->minidle = -0x7FFFFFFF;
	cbq_set_lss(cl, nla_data(tb[TCA_CBQ_LSSOPT]));
	cbq_set_wrr(cl, nla_data(tb[TCA_CBQ_WRROPT]));
	if (cl->ewma_log == 0)
		cl->ewma_log = q->link.ewma_log;
	if (cl->maxidle == 0)
		cl->maxidle = q->link.maxidle;
	if (cl->avpkt == 0)
		cl->avpkt = q->link.avpkt;
	if (tb[TCA_CBQ_FOPT])
		cbq_set_fopt(cl, nla_data(tb[TCA_CBQ_FOPT]));
	sch_tree_unlock(sch);

	qdisc_class_hash_grow(sch, &q->clhash);

	*arg = (unsigned long)cl;
	return 0;

failure:
	qdisc_put_rtab(rtab);
	return err;
}

static int cbq_delete(struct Qdisc *sch, unsigned long arg)
{
	struct cbq_sched_data *q = qdisc_priv(sch);
	struct cbq_class *cl = (struct cbq_class *)arg;
	unsigned int qlen, backlog;

	if (cl->filters || cl->children || cl == &q->link)
		return -EBUSY;

	sch_tree_lock(sch);

	qlen = cl->q->q.qlen;
	backlog = cl->q->qstats.backlog;
	qdisc_reset(cl->q);
	qdisc_tree_reduce_backlog(cl->q, qlen, backlog);

	if (cl->next_alive)
		cbq_deactivate_class(cl);

	if (q->tx_borrowed == cl)
		q->tx_borrowed = q->tx_class;
	if (q->tx_class == cl) {
		q->tx_class = NULL;
		q->tx_borrowed = NULL;
	}
#ifdef CONFIG_NET_CLS_ACT
	if (q->rx_class == cl)
		q->rx_class = NULL;
#endif

	cbq_unlink_class(cl);
	cbq_adjust_levels(cl->tparent);
	cl->defmap = 0;
	cbq_sync_defmap(cl);

	cbq_rmprio(q, cl);
	sch_tree_unlock(sch);

	BUG_ON(--cl->refcnt == 0);
	/*
	 * This shouldn't happen: we "hold" one cops->get() when called
	 * from tc_ctl_tclass; the destroy method is done from cops->put().
	 */

	return 0;
}

static struct tcf_block *cbq_tcf_block(struct Qdisc *sch, unsigned long arg)
{
	struct cbq_sched_data *q = qdisc_priv(sch);
	struct cbq_class *cl = (struct cbq_class *)arg;

	if (cl == NULL)
		cl = &q->link;

	return cl->block;
}

static unsigned long cbq_bind_filter(struct Qdisc *sch, unsigned long parent,
				     u32 classid)
{
	struct cbq_sched_data *q = qdisc_priv(sch);
	struct cbq_class *p = (struct cbq_class *)parent;
	struct cbq_class *cl = cbq_class_lookup(q, classid);

	if (cl) {
		if (p && p->level <= cl->level)
			return 0;
		cl->filters++;
		return (unsigned long)cl;
	}
	return 0;
}

static void cbq_unbind_filter(struct Qdisc *sch, unsigned long arg)
{
	struct cbq_class *cl = (struct cbq_class *)arg;

	cl->filters--;
}

static void cbq_walk(struct Qdisc *sch, struct qdisc_walker *arg)
{
	struct cbq_sched_data *q = qdisc_priv(sch);
	struct cbq_class *cl;
	unsigned int h;

	if (arg->stop)
		return;

	for (h = 0; h < q->clhash.hashsize; h++) {
		hlist_for_each_entry(cl, &q->clhash.hash[h], common.hnode) {
			if (arg->count < arg->skip) {
				arg->count++;
				continue;
			}
			if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
				arg->stop = 1;
				return;
			}
			arg->count++;
		}
	}
}

static const struct Qdisc_class_ops cbq_class_ops = {
	.graft		=	cbq_graft,
	.leaf		=	cbq_leaf,
	.qlen_notify	=	cbq_qlen_notify,
	.get		=	cbq_get,
	.put		=	cbq_put,
	.change		=	cbq_change_class,
	.delete		=	cbq_delete,
	.walk		=	cbq_walk,
	.tcf_block	=	cbq_tcf_block,
	.bind_tcf	=	cbq_bind_filter,
	.unbind_tcf	=	cbq_unbind_filter,
	.dump		=	cbq_dump_class,
	.dump_stats	=	cbq_dump_class_stats,
};

static struct Qdisc_ops cbq_qdisc_ops __read_mostly = {
	.next		=	NULL,
	.cl_ops		=	&cbq_class_ops,
	.id		=	"cbq",
	.priv_size	=	sizeof(struct cbq_sched_data),
	.enqueue	=	cbq_enqueue,
	.dequeue	=	cbq_dequeue,
	.peek		=	qdisc_peek_dequeued,
	.init		=	cbq_init,
	.reset		=	cbq_reset,
	.destroy	=	cbq_destroy,
	.change		=	NULL,
	.dump		=	cbq_dump,
	.dump_stats	=	cbq_dump_stats,
	.owner		=	THIS_MODULE,
};

static int __init cbq_module_init(void)
{
	return register_qdisc(&cbq_qdisc_ops);
}
static void __exit cbq_module_exit(void)
{
	unregister_qdisc(&cbq_qdisc_ops);
}
module_init(cbq_module_init)
module_exit(cbq_module_exit)
MODULE_LICENSE("GPL");