net_sched: Add accessor function for packet length for qdiscs
[linux-2.6-block.git] / include / linux / pkt_sched.h
CommitLineData
1da177e4
LT
1#ifndef __LINUX_PKT_SCHED_H
2#define __LINUX_PKT_SCHED_H
3
4/* Logical priority bands not depending on specific packet scheduler.
5 Every scheduler will map them to real traffic classes, if it has
6 no more precise mechanism to classify packets.
7
8 These numbers have no special meaning, though their coincidence
9 with obsolete IPv6 values is not occasional :-). New IPv6 drafts
10 preferred full anarchy inspired by diffserv group.
11
12 Note: TC_PRIO_BESTEFFORT does not mean that it is the most unhappy
13 class, actually, as rule it will be handled with more care than
14 filler or even bulk.
15 */
16
17#define TC_PRIO_BESTEFFORT 0
18#define TC_PRIO_FILLER 1
19#define TC_PRIO_BULK 2
20#define TC_PRIO_INTERACTIVE_BULK 4
21#define TC_PRIO_INTERACTIVE 6
22#define TC_PRIO_CONTROL 7
23
24#define TC_PRIO_MAX 15
25
26/* Generic queue statistics, available for all the elements.
27 Particular schedulers may have also their private records.
28 */
29
30struct tc_stats
31{
32 __u64 bytes; /* NUmber of enqueues bytes */
33 __u32 packets; /* Number of enqueued packets */
34 __u32 drops; /* Packets dropped because of lack of resources */
35 __u32 overlimits; /* Number of throttle events when this
36 * flow goes out of allocated bandwidth */
37 __u32 bps; /* Current flow byte rate */
38 __u32 pps; /* Current flow packet rate */
39 __u32 qlen;
40 __u32 backlog;
41};
42
43struct tc_estimator
44{
45 signed char interval;
46 unsigned char ewma_log;
47};
48
49/* "Handles"
50 ---------
51
52 All the traffic control objects have 32bit identifiers, or "handles".
53
54 They can be considered as opaque numbers from user API viewpoint,
55 but actually they always consist of two fields: major and
56 minor numbers, which are interpreted by kernel specially,
57 that may be used by applications, though not recommended.
58
59 F.e. qdisc handles always have minor number equal to zero,
60 classes (or flows) have major equal to parent qdisc major, and
61 minor uniquely identifying class inside qdisc.
62
63 Macros to manipulate handles:
64 */
65
66#define TC_H_MAJ_MASK (0xFFFF0000U)
67#define TC_H_MIN_MASK (0x0000FFFFU)
68#define TC_H_MAJ(h) ((h)&TC_H_MAJ_MASK)
69#define TC_H_MIN(h) ((h)&TC_H_MIN_MASK)
70#define TC_H_MAKE(maj,min) (((maj)&TC_H_MAJ_MASK)|((min)&TC_H_MIN_MASK))
71
72#define TC_H_UNSPEC (0U)
73#define TC_H_ROOT (0xFFFFFFFFU)
74#define TC_H_INGRESS (0xFFFFFFF1U)
75
76struct tc_ratespec
77{
78 unsigned char cell_log;
79 unsigned char __reserved;
e08b0998
JDB
80 unsigned short overhead;
81 short cell_align;
1da177e4
LT
82 unsigned short mpu;
83 __u32 rate;
84};
85
5feb5e1a
PM
86#define TC_RTAB_SIZE 1024
87
1da177e4
LT
88/* FIFO section */
89
90struct tc_fifo_qopt
91{
92 __u32 limit; /* Queue length: bytes for bfifo, packets for pfifo */
93};
94
95/* PRIO section */
96
97#define TCQ_PRIO_BANDS 16
bdc450a0 98#define TCQ_MIN_PRIO_BANDS 2
1da177e4
LT
99
100struct tc_prio_qopt
101{
102 int bands; /* Number of bands */
103 __u8 priomap[TC_PRIO_MAX+1]; /* Map: logical priority -> PRIO band */
104};
105
106/* TBF section */
107
108struct tc_tbf_qopt
109{
110 struct tc_ratespec rate;
111 struct tc_ratespec peakrate;
112 __u32 limit;
113 __u32 buffer;
114 __u32 mtu;
115};
116
117enum
118{
119 TCA_TBF_UNSPEC,
120 TCA_TBF_PARMS,
121 TCA_TBF_RTAB,
122 TCA_TBF_PTAB,
123 __TCA_TBF_MAX,
124};
125
126#define TCA_TBF_MAX (__TCA_TBF_MAX - 1)
127
128
129/* TEQL section */
130
131/* TEQL does not require any parameters */
132
133/* SFQ section */
134
135struct tc_sfq_qopt
136{
137 unsigned quantum; /* Bytes per round allocated to flow */
138 int perturb_period; /* Period of hash perturbation */
139 __u32 limit; /* Maximal packets in queue */
140 unsigned divisor; /* Hash divisor */
141 unsigned flows; /* Maximal number of flows */
142};
143
94de78d1
PM
144struct tc_sfq_xstats
145{
146 __s32 allot;
147};
148
1da177e4
LT
149/*
150 * NOTE: limit, divisor and flows are hardwired to code at the moment.
151 *
152 * limit=flows=128, divisor=1024;
153 *
154 * The only reason for this is efficiency, it is possible
155 * to change these parameters in compile time.
156 */
157
158/* RED section */
159
160enum
161{
162 TCA_RED_UNSPEC,
163 TCA_RED_PARMS,
164 TCA_RED_STAB,
165 __TCA_RED_MAX,
166};
167
168#define TCA_RED_MAX (__TCA_RED_MAX - 1)
169
170struct tc_red_qopt
171{
172 __u32 limit; /* HARD maximal queue length (bytes) */
173 __u32 qth_min; /* Min average length threshold (bytes) */
174 __u32 qth_max; /* Max average length threshold (bytes) */
175 unsigned char Wlog; /* log(W) */
176 unsigned char Plog; /* log(P_max/(qth_max-qth_min)) */
177 unsigned char Scell_log; /* cell size for idle damping */
178 unsigned char flags;
179#define TC_RED_ECN 1
bdc450a0 180#define TC_RED_HARDDROP 2
1da177e4
LT
181};
182
183struct tc_red_xstats
184{
185 __u32 early; /* Early drops */
186 __u32 pdrop; /* Drops due to queue limits */
187 __u32 other; /* Drops due to drop() calls */
188 __u32 marked; /* Marked packets */
189};
190
191/* GRED section */
192
193#define MAX_DPs 16
194
195enum
196{
197 TCA_GRED_UNSPEC,
198 TCA_GRED_PARMS,
199 TCA_GRED_STAB,
200 TCA_GRED_DPS,
201 __TCA_GRED_MAX,
202};
203
204#define TCA_GRED_MAX (__TCA_GRED_MAX - 1)
205
1da177e4
LT
206struct tc_gred_qopt
207{
1e4dfaf9
TG
208 __u32 limit; /* HARD maximal queue length (bytes) */
209 __u32 qth_min; /* Min average length threshold (bytes) */
210 __u32 qth_max; /* Max average length threshold (bytes) */
211 __u32 DP; /* upto 2^32 DPs */
212 __u32 backlog;
213 __u32 qave;
214 __u32 forced;
215 __u32 early;
216 __u32 other;
217 __u32 pdrop;
218 __u8 Wlog; /* log(W) */
219 __u8 Plog; /* log(P_max/(qth_max-qth_min)) */
220 __u8 Scell_log; /* cell size for idle damping */
221 __u8 prio; /* prio of this VQ */
222 __u32 packets;
223 __u32 bytesin;
1da177e4 224};
1e4dfaf9 225
1da177e4
LT
226/* gred setup */
227struct tc_gred_sopt
228{
1e4dfaf9
TG
229 __u32 DPs;
230 __u32 def_DP;
231 __u8 grio;
b38c7eef
TG
232 __u8 flags;
233 __u16 pad1;
1da177e4
LT
234};
235
236/* HTB section */
237#define TC_HTB_NUMPRIO 8
238#define TC_HTB_MAXDEPTH 8
239#define TC_HTB_PROTOVER 3 /* the same as HTB and TC's major */
240
241struct tc_htb_opt
242{
243 struct tc_ratespec rate;
244 struct tc_ratespec ceil;
245 __u32 buffer;
246 __u32 cbuffer;
247 __u32 quantum;
248 __u32 level; /* out only */
249 __u32 prio;
250};
251struct tc_htb_glob
252{
253 __u32 version; /* to match HTB/TC */
254 __u32 rate2quantum; /* bps->quantum divisor */
255 __u32 defcls; /* default class number */
256 __u32 debug; /* debug flags */
257
258 /* stats */
259 __u32 direct_pkts; /* count of non shapped packets */
260};
261enum
262{
263 TCA_HTB_UNSPEC,
264 TCA_HTB_PARMS,
265 TCA_HTB_INIT,
266 TCA_HTB_CTAB,
267 TCA_HTB_RTAB,
268 __TCA_HTB_MAX,
269};
270
271#define TCA_HTB_MAX (__TCA_HTB_MAX - 1)
272
273struct tc_htb_xstats
274{
275 __u32 lends;
276 __u32 borrows;
277 __u32 giants; /* too big packets (rate will not be accurate) */
278 __u32 tokens;
279 __u32 ctokens;
280};
281
282/* HFSC section */
283
284struct tc_hfsc_qopt
285{
286 __u16 defcls; /* default class */
287};
288
289struct tc_service_curve
290{
291 __u32 m1; /* slope of the first segment in bps */
292 __u32 d; /* x-projection of the first segment in us */
293 __u32 m2; /* slope of the second segment in bps */
294};
295
296struct tc_hfsc_stats
297{
298 __u64 work; /* total work done */
299 __u64 rtwork; /* work done by real-time criteria */
300 __u32 period; /* current period */
301 __u32 level; /* class level in hierarchy */
302};
303
304enum
305{
306 TCA_HFSC_UNSPEC,
307 TCA_HFSC_RSC,
308 TCA_HFSC_FSC,
309 TCA_HFSC_USC,
310 __TCA_HFSC_MAX,
311};
312
313#define TCA_HFSC_MAX (__TCA_HFSC_MAX - 1)
314
315
316/* CBQ section */
317
318#define TC_CBQ_MAXPRIO 8
319#define TC_CBQ_MAXLEVEL 8
320#define TC_CBQ_DEF_EWMA 5
321
322struct tc_cbq_lssopt
323{
324 unsigned char change;
325 unsigned char flags;
326#define TCF_CBQ_LSS_BOUNDED 1
327#define TCF_CBQ_LSS_ISOLATED 2
328 unsigned char ewma_log;
329 unsigned char level;
330#define TCF_CBQ_LSS_FLAGS 1
331#define TCF_CBQ_LSS_EWMA 2
332#define TCF_CBQ_LSS_MAXIDLE 4
333#define TCF_CBQ_LSS_MINIDLE 8
334#define TCF_CBQ_LSS_OFFTIME 0x10
335#define TCF_CBQ_LSS_AVPKT 0x20
336 __u32 maxidle;
337 __u32 minidle;
338 __u32 offtime;
339 __u32 avpkt;
340};
341
342struct tc_cbq_wrropt
343{
344 unsigned char flags;
345 unsigned char priority;
346 unsigned char cpriority;
347 unsigned char __reserved;
348 __u32 allot;
349 __u32 weight;
350};
351
352struct tc_cbq_ovl
353{
354 unsigned char strategy;
355#define TC_CBQ_OVL_CLASSIC 0
356#define TC_CBQ_OVL_DELAY 1
357#define TC_CBQ_OVL_LOWPRIO 2
358#define TC_CBQ_OVL_DROP 3
359#define TC_CBQ_OVL_RCLASSIC 4
360 unsigned char priority2;
8a47077a 361 __u16 pad;
1da177e4
LT
362 __u32 penalty;
363};
364
365struct tc_cbq_police
366{
367 unsigned char police;
368 unsigned char __res1;
369 unsigned short __res2;
370};
371
372struct tc_cbq_fopt
373{
374 __u32 split;
375 __u32 defmap;
376 __u32 defchange;
377};
378
379struct tc_cbq_xstats
380{
381 __u32 borrows;
382 __u32 overactions;
383 __s32 avgidle;
384 __s32 undertime;
385};
386
387enum
388{
389 TCA_CBQ_UNSPEC,
390 TCA_CBQ_LSSOPT,
391 TCA_CBQ_WRROPT,
392 TCA_CBQ_FOPT,
393 TCA_CBQ_OVL_STRATEGY,
394 TCA_CBQ_RATE,
395 TCA_CBQ_RTAB,
396 TCA_CBQ_POLICE,
397 __TCA_CBQ_MAX,
398};
399
400#define TCA_CBQ_MAX (__TCA_CBQ_MAX - 1)
401
402/* dsmark section */
403
404enum {
405 TCA_DSMARK_UNSPEC,
406 TCA_DSMARK_INDICES,
407 TCA_DSMARK_DEFAULT_INDEX,
408 TCA_DSMARK_SET_TC_INDEX,
409 TCA_DSMARK_MASK,
410 TCA_DSMARK_VALUE,
411 __TCA_DSMARK_MAX,
412};
413
414#define TCA_DSMARK_MAX (__TCA_DSMARK_MAX - 1)
415
416/* ATM section */
417
418enum {
419 TCA_ATM_UNSPEC,
420 TCA_ATM_FD, /* file/socket descriptor */
421 TCA_ATM_PTR, /* pointer to descriptor - later */
422 TCA_ATM_HDR, /* LL header */
423 TCA_ATM_EXCESS, /* excess traffic class (0 for CLP) */
424 TCA_ATM_ADDR, /* PVC address (for output only) */
425 TCA_ATM_STATE, /* VC state (ATM_VS_*; for output only) */
426 __TCA_ATM_MAX,
427};
428
429#define TCA_ATM_MAX (__TCA_ATM_MAX - 1)
430
431/* Network emulator */
432
433enum
434{
435 TCA_NETEM_UNSPEC,
436 TCA_NETEM_CORR,
437 TCA_NETEM_DELAY_DIST,
0dca51d3 438 TCA_NETEM_REORDER,
c865e5d9 439 TCA_NETEM_CORRUPT,
1da177e4
LT
440 __TCA_NETEM_MAX,
441};
442
443#define TCA_NETEM_MAX (__TCA_NETEM_MAX - 1)
444
445struct tc_netem_qopt
446{
447 __u32 latency; /* added delay (us) */
448 __u32 limit; /* fifo limit (packets) */
449 __u32 loss; /* random packet loss (0=none ~0=100%) */
0dca51d3 450 __u32 gap; /* re-ordering gap (0 for none) */
1da177e4
LT
451 __u32 duplicate; /* random packet dup (0=none ~0=100%) */
452 __u32 jitter; /* random jitter in latency (us) */
453};
454
455struct tc_netem_corr
456{
457 __u32 delay_corr; /* delay correlation */
458 __u32 loss_corr; /* packet loss correlation */
459 __u32 dup_corr; /* duplicate correlation */
460};
461
0dca51d3
SH
462struct tc_netem_reorder
463{
464 __u32 probability;
465 __u32 correlation;
466};
467
c865e5d9
SH
468struct tc_netem_corrupt
469{
470 __u32 probability;
471 __u32 correlation;
472};
473
1da177e4
LT
474#define NETEM_DIST_SCALE 8192
475
476#endif