pkt_sched.h revision dcfb7a77f8709125e97c313cb8ab6ec4d87468f4 
1#ifndef __LINUX_PKT_SCHED_H
2#define __LINUX_PKT_SCHED_H
3
4#include <linux/types.h>
5
6/* Logical priority bands not depending on specific packet scheduler.
7   Every scheduler will map them to real traffic classes, if it has
8   no more precise mechanism to classify packets.
9
10   These numbers have no special meaning, though their coincidence
11   with obsolete IPv6 values is not occasional :-). New IPv6 drafts
12   preferred full anarchy inspired by diffserv group.
13
14   Note: TC_PRIO_BESTEFFORT does not mean that it is the most unhappy
15   class, actually, as rule it will be handled with more care than
16   filler or even bulk.
17 */
18
19#define TC_PRIO_BESTEFFORT		0
20#define TC_PRIO_FILLER			1
21#define TC_PRIO_BULK			2
22#define TC_PRIO_INTERACTIVE_BULK	4
23#define TC_PRIO_INTERACTIVE		6
24#define TC_PRIO_CONTROL			7
25
26#define TC_PRIO_MAX			15
27
28/* Generic queue statistics, available for all the elements.
29   Particular schedulers may have also their private records.
30 */
31
32struct tc_stats
33{
34	__u64	bytes;			/* NUmber of enqueues bytes */
35	__u32	packets;		/* Number of enqueued packets	*/
36	__u32	drops;			/* Packets dropped because of lack of resources */
37	__u32	overlimits;		/* Number of throttle events when this
38					 * flow goes out of allocated bandwidth */
39	__u32	bps;			/* Current flow byte rate */
40	__u32	pps;			/* Current flow packet rate */
41	__u32	qlen;
42	__u32	backlog;
43};
44
45struct tc_estimator
46{
47	signed char	interval;
48	unsigned char	ewma_log;
49};
50
51/* "Handles"
52   ---------
53
54    All the traffic control objects have 32bit identifiers, or "handles".
55
56    They can be considered as opaque numbers from user API viewpoint,
57    but actually they always consist of two fields: major and
58    minor numbers, which are interpreted by kernel specially,
59    that may be used by applications, though not recommended.
60
61    F.e. qdisc handles always have minor number equal to zero,
62    classes (or flows) have major equal to parent qdisc major, and
63    minor uniquely identifying class inside qdisc.
64
65    Macros to manipulate handles:
66 */
67
68#define TC_H_MAJ_MASK (0xFFFF0000U)
69#define TC_H_MIN_MASK (0x0000FFFFU)
70#define TC_H_MAJ(h) ((h)&TC_H_MAJ_MASK)
71#define TC_H_MIN(h) ((h)&TC_H_MIN_MASK)
72#define TC_H_MAKE(maj,min) (((maj)&TC_H_MAJ_MASK)|((min)&TC_H_MIN_MASK))
73
74#define TC_H_UNSPEC	(0U)
75#define TC_H_ROOT	(0xFFFFFFFFU)
76#define TC_H_INGRESS    (0xFFFFFFF1U)
77
78struct tc_ratespec
79{
80	unsigned char	cell_log;
81	unsigned char	__reserved;
82	unsigned short	overhead;
83	short		cell_align;
84	unsigned short	mpu;
85	__u32		rate;
86};
87
88#define TC_RTAB_SIZE	1024
89
90struct tc_sizespec {
91	unsigned char	cell_log;
92	unsigned char	size_log;
93	short		cell_align;
94	int		overhead;
95	unsigned int	linklayer;
96	unsigned int	mpu;
97	unsigned int	mtu;
98	unsigned int	tsize;
99};
100
101enum {
102	TCA_STAB_UNSPEC,
103	TCA_STAB_BASE,
104	TCA_STAB_DATA,
105	__TCA_STAB_MAX
106};
107
108#define TCA_STAB_MAX (__TCA_STAB_MAX - 1)
109
110/* FIFO section */
111
112struct tc_fifo_qopt
113{
114	__u32	limit;	/* Queue length: bytes for bfifo, packets for pfifo */
115};
116
117/* PRIO section */
118
119#define TCQ_PRIO_BANDS	16
120#define TCQ_MIN_PRIO_BANDS 2
121
122struct tc_prio_qopt
123{
124	int	bands;			/* Number of bands */
125	__u8	priomap[TC_PRIO_MAX+1];	/* Map: logical priority -> PRIO band */
126};
127
128/* MULTIQ section */
129
130struct tc_multiq_qopt {
131	__u16	bands;			/* Number of bands */
132	__u16	max_bands;		/* Maximum number of queues */
133};
134
135/* TBF section */
136
137struct tc_tbf_qopt
138{
139	struct tc_ratespec rate;
140	struct tc_ratespec peakrate;
141	__u32		limit;
142	__u32		buffer;
143	__u32		mtu;
144};
145
146enum
147{
148	TCA_TBF_UNSPEC,
149	TCA_TBF_PARMS,
150	TCA_TBF_RTAB,
151	TCA_TBF_PTAB,
152	__TCA_TBF_MAX,
153};
154
155#define TCA_TBF_MAX (__TCA_TBF_MAX - 1)
156
157
158/* TEQL section */
159
160/* TEQL does not require any parameters */
161
162/* SFQ section */
163
164struct tc_sfq_qopt
165{
166	unsigned	quantum;	/* Bytes per round allocated to flow */
167	int		perturb_period;	/* Period of hash perturbation */
168	__u32		limit;		/* Maximal packets in queue */
169	unsigned	divisor;	/* Hash divisor  */
170	unsigned	flows;		/* Maximal number of flows  */
171};
172
173struct tc_sfq_xstats
174{
175	__s32		allot;
176};
177
178/*
179 *  NOTE: limit, divisor and flows are hardwired to code at the moment.
180 *
181 *	limit=flows=128, divisor=1024;
182 *
183 *	The only reason for this is efficiency, it is possible
184 *	to change these parameters in compile time.
185 */
186
187/* RED section */
188
189enum
190{
191	TCA_RED_UNSPEC,
192	TCA_RED_PARMS,
193	TCA_RED_STAB,
194	__TCA_RED_MAX,
195};
196
197#define TCA_RED_MAX (__TCA_RED_MAX - 1)
198
199struct tc_red_qopt
200{
201	__u32		limit;		/* HARD maximal queue length (bytes)	*/
202	__u32		qth_min;	/* Min average length threshold (bytes) */
203	__u32		qth_max;	/* Max average length threshold (bytes) */
204	unsigned char   Wlog;		/* log(W)		*/
205	unsigned char   Plog;		/* log(P_max/(qth_max-qth_min))	*/
206	unsigned char   Scell_log;	/* cell size for idle damping */
207	unsigned char	flags;
208#define TC_RED_ECN	1
209#define TC_RED_HARDDROP	2
210};
211
212struct tc_red_xstats
213{
214	__u32           early;          /* Early drops */
215	__u32           pdrop;          /* Drops due to queue limits */
216	__u32           other;          /* Drops due to drop() calls */
217	__u32           marked;         /* Marked packets */
218};
219
220/* GRED section */
221
222#define MAX_DPs 16
223
224enum
225{
226       TCA_GRED_UNSPEC,
227       TCA_GRED_PARMS,
228       TCA_GRED_STAB,
229       TCA_GRED_DPS,
230	   __TCA_GRED_MAX,
231};
232
233#define TCA_GRED_MAX (__TCA_GRED_MAX - 1)
234
235struct tc_gred_qopt
236{
237	__u32		limit;        /* HARD maximal queue length (bytes)    */
238	__u32		qth_min;      /* Min average length threshold (bytes) */
239	__u32		qth_max;      /* Max average length threshold (bytes) */
240	__u32		DP;           /* upto 2^32 DPs */
241	__u32		backlog;
242	__u32		qave;
243	__u32		forced;
244	__u32		early;
245	__u32		other;
246	__u32		pdrop;
247	__u8		Wlog;         /* log(W)               */
248	__u8		Plog;         /* log(P_max/(qth_max-qth_min)) */
249	__u8		Scell_log;    /* cell size for idle damping */
250	__u8		prio;         /* prio of this VQ */
251	__u32		packets;
252	__u32		bytesin;
253};
254
255/* gred setup */
256struct tc_gred_sopt
257{
258	__u32		DPs;
259	__u32		def_DP;
260	__u8		grio;
261	__u8		flags;
262	__u16		pad1;
263};
264
265/* HTB section */
266#define TC_HTB_NUMPRIO		8
267#define TC_HTB_MAXDEPTH		8
268#define TC_HTB_PROTOVER		3 /* the same as HTB and TC's major */
269
270struct tc_htb_opt
271{
272	struct tc_ratespec 	rate;
273	struct tc_ratespec 	ceil;
274	__u32	buffer;
275	__u32	cbuffer;
276	__u32	quantum;
277	__u32	level;		/* out only */
278	__u32	prio;
279};
280struct tc_htb_glob
281{
282	__u32 version;		/* to match HTB/TC */
283    	__u32 rate2quantum;	/* bps->quantum divisor */
284    	__u32 defcls;		/* default class number */
285	__u32 debug;		/* debug flags */
286
287	/* stats */
288	__u32 direct_pkts; /* count of non shapped packets */
289};
290enum
291{
292	TCA_HTB_UNSPEC,
293	TCA_HTB_PARMS,
294	TCA_HTB_INIT,
295	TCA_HTB_CTAB,
296	TCA_HTB_RTAB,
297	__TCA_HTB_MAX,
298};
299
300#define TCA_HTB_MAX (__TCA_HTB_MAX - 1)
301
302struct tc_htb_xstats
303{
304	__u32 lends;
305	__u32 borrows;
306	__u32 giants;	/* too big packets (rate will not be accurate) */
307	__u32 tokens;
308	__u32 ctokens;
309};
310
311/* HFSC section */
312
313struct tc_hfsc_qopt
314{
315	__u16	defcls;		/* default class */
316};
317
318struct tc_service_curve
319{
320	__u32	m1;		/* slope of the first segment in bps */
321	__u32	d;		/* x-projection of the first segment in us */
322	__u32	m2;		/* slope of the second segment in bps */
323};
324
325struct tc_hfsc_stats
326{
327	__u64	work;		/* total work done */
328	__u64	rtwork;		/* work done by real-time criteria */
329	__u32	period;		/* current period */
330	__u32	level;		/* class level in hierarchy */
331};
332
333enum
334{
335	TCA_HFSC_UNSPEC,
336	TCA_HFSC_RSC,
337	TCA_HFSC_FSC,
338	TCA_HFSC_USC,
339	__TCA_HFSC_MAX,
340};
341
342#define TCA_HFSC_MAX (__TCA_HFSC_MAX - 1)
343
344
345/* CBQ section */
346
347#define TC_CBQ_MAXPRIO		8
348#define TC_CBQ_MAXLEVEL		8
349#define TC_CBQ_DEF_EWMA		5
350
351struct tc_cbq_lssopt
352{
353	unsigned char	change;
354	unsigned char	flags;
355#define TCF_CBQ_LSS_BOUNDED	1
356#define TCF_CBQ_LSS_ISOLATED	2
357	unsigned char  	ewma_log;
358	unsigned char  	level;
359#define TCF_CBQ_LSS_FLAGS	1
360#define TCF_CBQ_LSS_EWMA	2
361#define TCF_CBQ_LSS_MAXIDLE	4
362#define TCF_CBQ_LSS_MINIDLE	8
363#define TCF_CBQ_LSS_OFFTIME	0x10
364#define TCF_CBQ_LSS_AVPKT	0x20
365	__u32		maxidle;
366	__u32		minidle;
367	__u32		offtime;
368	__u32		avpkt;
369};
370
371struct tc_cbq_wrropt
372{
373	unsigned char	flags;
374	unsigned char	priority;
375	unsigned char	cpriority;
376	unsigned char	__reserved;
377	__u32		allot;
378	__u32		weight;
379};
380
381struct tc_cbq_ovl
382{
383	unsigned char	strategy;
384#define	TC_CBQ_OVL_CLASSIC	0
385#define	TC_CBQ_OVL_DELAY	1
386#define	TC_CBQ_OVL_LOWPRIO	2
387#define	TC_CBQ_OVL_DROP		3
388#define	TC_CBQ_OVL_RCLASSIC	4
389	unsigned char	priority2;
390	__u16		pad;
391	__u32		penalty;
392};
393
394struct tc_cbq_police
395{
396	unsigned char	police;
397	unsigned char	__res1;
398	unsigned short	__res2;
399};
400
401struct tc_cbq_fopt
402{
403	__u32		split;
404	__u32		defmap;
405	__u32		defchange;
406};
407
408struct tc_cbq_xstats
409{
410	__u32		borrows;
411	__u32		overactions;
412	__s32		avgidle;
413	__s32		undertime;
414};
415
416enum
417{
418	TCA_CBQ_UNSPEC,
419	TCA_CBQ_LSSOPT,
420	TCA_CBQ_WRROPT,
421	TCA_CBQ_FOPT,
422	TCA_CBQ_OVL_STRATEGY,
423	TCA_CBQ_RATE,
424	TCA_CBQ_RTAB,
425	TCA_CBQ_POLICE,
426	__TCA_CBQ_MAX,
427};
428
429#define TCA_CBQ_MAX	(__TCA_CBQ_MAX - 1)
430
431/* dsmark section */
432
433enum {
434	TCA_DSMARK_UNSPEC,
435	TCA_DSMARK_INDICES,
436	TCA_DSMARK_DEFAULT_INDEX,
437	TCA_DSMARK_SET_TC_INDEX,
438	TCA_DSMARK_MASK,
439	TCA_DSMARK_VALUE,
440	__TCA_DSMARK_MAX,
441};
442
443#define TCA_DSMARK_MAX (__TCA_DSMARK_MAX - 1)
444
445/* ATM  section */
446
447enum {
448	TCA_ATM_UNSPEC,
449	TCA_ATM_FD,		/* file/socket descriptor */
450	TCA_ATM_PTR,		/* pointer to descriptor - later */
451	TCA_ATM_HDR,		/* LL header */
452	TCA_ATM_EXCESS,		/* excess traffic class (0 for CLP)  */
453	TCA_ATM_ADDR,		/* PVC address (for output only) */
454	TCA_ATM_STATE,		/* VC state (ATM_VS_*; for output only) */
455	__TCA_ATM_MAX,
456};
457
458#define TCA_ATM_MAX	(__TCA_ATM_MAX - 1)
459
460/* Network emulator */
461
462enum
463{
464	TCA_NETEM_UNSPEC,
465	TCA_NETEM_CORR,
466	TCA_NETEM_DELAY_DIST,
467	TCA_NETEM_REORDER,
468	TCA_NETEM_CORRUPT,
469	__TCA_NETEM_MAX,
470};
471
472#define TCA_NETEM_MAX (__TCA_NETEM_MAX - 1)
473
474struct tc_netem_qopt
475{
476	__u32	latency;	/* added delay (us) */
477	__u32   limit;		/* fifo limit (packets) */
478	__u32	loss;		/* random packet loss (0=none ~0=100%) */
479	__u32	gap;		/* re-ordering gap (0 for none) */
480	__u32   duplicate;	/* random packet dup  (0=none ~0=100%) */
481	__u32	jitter;		/* random jitter in latency (us) */
482};
483
484struct tc_netem_corr
485{
486	__u32	delay_corr;	/* delay correlation */
487	__u32	loss_corr;	/* packet loss correlation */
488	__u32	dup_corr;	/* duplicate correlation  */
489};
490
491struct tc_netem_reorder
492{
493	__u32	probability;
494	__u32	correlation;
495};
496
497struct tc_netem_corrupt
498{
499	__u32	probability;
500	__u32	correlation;
501};
502
503#define NETEM_DIST_SCALE	8192
504
505/* DRR */
506
507enum
508{
509	TCA_DRR_UNSPEC,
510	TCA_DRR_QUANTUM,
511	__TCA_DRR_MAX
512};
513
514#define TCA_DRR_MAX	(__TCA_DRR_MAX - 1)
515
516struct tc_drr_stats
517{
518	__u32	deficit;
519};
520
521#endif
522