pkt_sched.h revision ab322673298bd0b8927cdd9d11f3d36af5941b93
156c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks#ifndef __LINUX_PKT_SCHED_H 256c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks#define __LINUX_PKT_SCHED_H 356c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks 456c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks#include <linux/types.h> 556c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks 656c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks/* Logical priority bands not depending on specific packet scheduler. 756c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks Every scheduler will map them to real traffic classes, if it has 856c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks no more precise mechanism to classify packets. 956c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks 1056c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks These numbers have no special meaning, though their coincidence 1156c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks with obsolete IPv6 values is not occasional :-). New IPv6 drafts 127df30109963092559d3760c0661a020f9daf1030The Android Open Source Project preferred full anarchy inspired by diffserv group. 137df30109963092559d3760c0661a020f9daf1030The Android Open Source Project 147df30109963092559d3760c0661a020f9daf1030The Android Open Source Project Note: TC_PRIO_BESTEFFORT does not mean that it is the most unhappy 157df30109963092559d3760c0661a020f9daf1030The Android Open Source Project class, actually, as rule it will be handled with more care than 167df30109963092559d3760c0661a020f9daf1030The Android Open Source Project filler or even bulk. 177df30109963092559d3760c0661a020f9daf1030The Android Open Source Project */ 187df30109963092559d3760c0661a020f9daf1030The Android Open Source Project 197df30109963092559d3760c0661a020f9daf1030The Android Open Source Project#define TC_PRIO_BESTEFFORT 0 207df30109963092559d3760c0661a020f9daf1030The Android Open Source Project#define TC_PRIO_FILLER 1 217df30109963092559d3760c0661a020f9daf1030The Android Open Source Project#define TC_PRIO_BULK 2 227df30109963092559d3760c0661a020f9daf1030The Android Open Source Project#define TC_PRIO_INTERACTIVE_BULK 4 237df30109963092559d3760c0661a020f9daf1030The Android Open Source Project#define TC_PRIO_INTERACTIVE 6 2456c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks#define TC_PRIO_CONTROL 7 2556c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks 2656c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks#define TC_PRIO_MAX 15 2756c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks 2856c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks/* Generic queue statistics, available for all the elements. 2956c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks Particular schedulers may have also their private records. 3056c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks */ 3156c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks 3256c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparksstruct tc_stats { 3356c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks __u64 bytes; /* NUmber of enqueues bytes */ 3456c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks __u32 packets; /* Number of enqueued packets */ 3556c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks __u32 drops; /* Packets dropped because of lack of resources */ 3656c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks __u32 overlimits; /* Number of throttle events when this 3756c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks * flow goes out of allocated bandwidth */ 3856c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks __u32 bps; /* Current flow byte rate */ 3956c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks __u32 pps; /* Current flow packet rate */ 4056c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks __u32 qlen; 4156c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks __u32 backlog; 4256c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks}; 4356c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks 4456c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparksstruct tc_estimator { 4556c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks signed char interval; 4656c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks unsigned char ewma_log; 4756c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks}; 4856c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks 4956c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks/* "Handles" 5056c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks --------- 5156c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks 5256c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks All the traffic control objects have 32bit identifiers, or "handles". 5356c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks 5456c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks They can be considered as opaque numbers from user API viewpoint, 5556c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks but actually they always consist of two fields: major and 5656c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks minor numbers, which are interpreted by kernel specially, 5756c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks that may be used by applications, though not recommended. 5856c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks 5956c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks F.e. qdisc handles always have minor number equal to zero, 6056c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks classes (or flows) have major equal to parent qdisc major, and 6156c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks minor uniquely identifying class inside qdisc. 6256c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks 6356c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks Macros to manipulate handles: 6456c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks */ 6556c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks 6656c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks#define TC_H_MAJ_MASK (0xFFFF0000U) 6756c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks#define TC_H_MIN_MASK (0x0000FFFFU) 6856c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks#define TC_H_MAJ(h) ((h)&TC_H_MAJ_MASK) 6956c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks#define TC_H_MIN(h) ((h)&TC_H_MIN_MASK) 7056c99cd2c2c1e6ab038dac5fced5b92ccf11ff6cDave Sparks#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 unsigned char cell_log; 78 unsigned char __reserved; 79 unsigned short overhead; 80 short cell_align; 81 unsigned short mpu; 82 __u32 rate; 83}; 84 85#define TC_RTAB_SIZE 1024 86 87struct tc_sizespec { 88 unsigned char cell_log; 89 unsigned char size_log; 90 short cell_align; 91 int overhead; 92 unsigned int linklayer; 93 unsigned int mpu; 94 unsigned int mtu; 95 unsigned int tsize; 96}; 97 98enum { 99 TCA_STAB_UNSPEC, 100 TCA_STAB_BASE, 101 TCA_STAB_DATA, 102 __TCA_STAB_MAX 103}; 104 105#define TCA_STAB_MAX (__TCA_STAB_MAX - 1) 106 107/* FIFO section */ 108 109struct tc_fifo_qopt { 110 __u32 limit; /* Queue length: bytes for bfifo, packets for pfifo */ 111}; 112 113/* PRIO section */ 114 115#define TCQ_PRIO_BANDS 16 116#define TCQ_MIN_PRIO_BANDS 2 117 118struct tc_prio_qopt { 119 int bands; /* Number of bands */ 120 __u8 priomap[TC_PRIO_MAX+1]; /* Map: logical priority -> PRIO band */ 121}; 122 123/* MULTIQ section */ 124 125struct tc_multiq_qopt { 126 __u16 bands; /* Number of bands */ 127 __u16 max_bands; /* Maximum number of queues */ 128}; 129 130/* TBF section */ 131 132struct tc_tbf_qopt { 133 struct tc_ratespec rate; 134 struct tc_ratespec peakrate; 135 __u32 limit; 136 __u32 buffer; 137 __u32 mtu; 138}; 139 140enum { 141 TCA_TBF_UNSPEC, 142 TCA_TBF_PARMS, 143 TCA_TBF_RTAB, 144 TCA_TBF_PTAB, 145 __TCA_TBF_MAX, 146}; 147 148#define TCA_TBF_MAX (__TCA_TBF_MAX - 1) 149 150 151/* TEQL section */ 152 153/* TEQL does not require any parameters */ 154 155/* SFQ section */ 156 157struct tc_sfq_qopt { 158 unsigned quantum; /* Bytes per round allocated to flow */ 159 int perturb_period; /* Period of hash perturbation */ 160 __u32 limit; /* Maximal packets in queue */ 161 unsigned divisor; /* Hash divisor */ 162 unsigned flows; /* Maximal number of flows */ 163}; 164 165struct tc_sfq_xstats { 166 __s32 allot; 167}; 168 169/* 170 * NOTE: limit, divisor and flows are hardwired to code at the moment. 171 * 172 * limit=flows=128, divisor=1024; 173 * 174 * The only reason for this is efficiency, it is possible 175 * to change these parameters in compile time. 176 */ 177 178/* RED section */ 179 180enum { 181 TCA_RED_UNSPEC, 182 TCA_RED_PARMS, 183 TCA_RED_STAB, 184 __TCA_RED_MAX, 185}; 186 187#define TCA_RED_MAX (__TCA_RED_MAX - 1) 188 189struct tc_red_qopt { 190 __u32 limit; /* HARD maximal queue length (bytes) */ 191 __u32 qth_min; /* Min average length threshold (bytes) */ 192 __u32 qth_max; /* Max average length threshold (bytes) */ 193 unsigned char Wlog; /* log(W) */ 194 unsigned char Plog; /* log(P_max/(qth_max-qth_min)) */ 195 unsigned char Scell_log; /* cell size for idle damping */ 196 unsigned char flags; 197#define TC_RED_ECN 1 198#define TC_RED_HARDDROP 2 199}; 200 201struct tc_red_xstats { 202 __u32 early; /* Early drops */ 203 __u32 pdrop; /* Drops due to queue limits */ 204 __u32 other; /* Drops due to drop() calls */ 205 __u32 marked; /* Marked packets */ 206}; 207 208/* GRED section */ 209 210#define MAX_DPs 16 211 212enum { 213 TCA_GRED_UNSPEC, 214 TCA_GRED_PARMS, 215 TCA_GRED_STAB, 216 TCA_GRED_DPS, 217 __TCA_GRED_MAX, 218}; 219 220#define TCA_GRED_MAX (__TCA_GRED_MAX - 1) 221 222struct tc_gred_qopt { 223 __u32 limit; /* HARD maximal queue length (bytes) */ 224 __u32 qth_min; /* Min average length threshold (bytes) */ 225 __u32 qth_max; /* Max average length threshold (bytes) */ 226 __u32 DP; /* upto 2^32 DPs */ 227 __u32 backlog; 228 __u32 qave; 229 __u32 forced; 230 __u32 early; 231 __u32 other; 232 __u32 pdrop; 233 __u8 Wlog; /* log(W) */ 234 __u8 Plog; /* log(P_max/(qth_max-qth_min)) */ 235 __u8 Scell_log; /* cell size for idle damping */ 236 __u8 prio; /* prio of this VQ */ 237 __u32 packets; 238 __u32 bytesin; 239}; 240 241/* gred setup */ 242struct tc_gred_sopt { 243 __u32 DPs; 244 __u32 def_DP; 245 __u8 grio; 246 __u8 flags; 247 __u16 pad1; 248}; 249 250/* HTB section */ 251#define TC_HTB_NUMPRIO 8 252#define TC_HTB_MAXDEPTH 8 253#define TC_HTB_PROTOVER 3 /* the same as HTB and TC's major */ 254 255struct tc_htb_opt { 256 struct tc_ratespec rate; 257 struct tc_ratespec ceil; 258 __u32 buffer; 259 __u32 cbuffer; 260 __u32 quantum; 261 __u32 level; /* out only */ 262 __u32 prio; 263}; 264struct tc_htb_glob { 265 __u32 version; /* to match HTB/TC */ 266 __u32 rate2quantum; /* bps->quantum divisor */ 267 __u32 defcls; /* default class number */ 268 __u32 debug; /* debug flags */ 269 270 /* stats */ 271 __u32 direct_pkts; /* count of non shapped packets */ 272}; 273enum { 274 TCA_HTB_UNSPEC, 275 TCA_HTB_PARMS, 276 TCA_HTB_INIT, 277 TCA_HTB_CTAB, 278 TCA_HTB_RTAB, 279 __TCA_HTB_MAX, 280}; 281 282#define TCA_HTB_MAX (__TCA_HTB_MAX - 1) 283 284struct tc_htb_xstats { 285 __u32 lends; 286 __u32 borrows; 287 __u32 giants; /* too big packets (rate will not be accurate) */ 288 __u32 tokens; 289 __u32 ctokens; 290}; 291 292/* HFSC section */ 293 294struct tc_hfsc_qopt { 295 __u16 defcls; /* default class */ 296}; 297 298struct tc_service_curve { 299 __u32 m1; /* slope of the first segment in bps */ 300 __u32 d; /* x-projection of the first segment in us */ 301 __u32 m2; /* slope of the second segment in bps */ 302}; 303 304struct tc_hfsc_stats { 305 __u64 work; /* total work done */ 306 __u64 rtwork; /* work done by real-time criteria */ 307 __u32 period; /* current period */ 308 __u32 level; /* class level in hierarchy */ 309}; 310 311enum { 312 TCA_HFSC_UNSPEC, 313 TCA_HFSC_RSC, 314 TCA_HFSC_FSC, 315 TCA_HFSC_USC, 316 __TCA_HFSC_MAX, 317}; 318 319#define TCA_HFSC_MAX (__TCA_HFSC_MAX - 1) 320 321 322/* CBQ section */ 323 324#define TC_CBQ_MAXPRIO 8 325#define TC_CBQ_MAXLEVEL 8 326#define TC_CBQ_DEF_EWMA 5 327 328struct tc_cbq_lssopt { 329 unsigned char change; 330 unsigned char flags; 331#define TCF_CBQ_LSS_BOUNDED 1 332#define TCF_CBQ_LSS_ISOLATED 2 333 unsigned char ewma_log; 334 unsigned char level; 335#define TCF_CBQ_LSS_FLAGS 1 336#define TCF_CBQ_LSS_EWMA 2 337#define TCF_CBQ_LSS_MAXIDLE 4 338#define TCF_CBQ_LSS_MINIDLE 8 339#define TCF_CBQ_LSS_OFFTIME 0x10 340#define TCF_CBQ_LSS_AVPKT 0x20 341 __u32 maxidle; 342 __u32 minidle; 343 __u32 offtime; 344 __u32 avpkt; 345}; 346 347struct tc_cbq_wrropt { 348 unsigned char flags; 349 unsigned char priority; 350 unsigned char cpriority; 351 unsigned char __reserved; 352 __u32 allot; 353 __u32 weight; 354}; 355 356struct tc_cbq_ovl { 357 unsigned char strategy; 358#define TC_CBQ_OVL_CLASSIC 0 359#define TC_CBQ_OVL_DELAY 1 360#define TC_CBQ_OVL_LOWPRIO 2 361#define TC_CBQ_OVL_DROP 3 362#define TC_CBQ_OVL_RCLASSIC 4 363 unsigned char priority2; 364 __u16 pad; 365 __u32 penalty; 366}; 367 368struct tc_cbq_police { 369 unsigned char police; 370 unsigned char __res1; 371 unsigned short __res2; 372}; 373 374struct tc_cbq_fopt { 375 __u32 split; 376 __u32 defmap; 377 __u32 defchange; 378}; 379 380struct tc_cbq_xstats { 381 __u32 borrows; 382 __u32 overactions; 383 __s32 avgidle; 384 __s32 undertime; 385}; 386 387enum { 388 TCA_CBQ_UNSPEC, 389 TCA_CBQ_LSSOPT, 390 TCA_CBQ_WRROPT, 391 TCA_CBQ_FOPT, 392 TCA_CBQ_OVL_STRATEGY, 393 TCA_CBQ_RATE, 394 TCA_CBQ_RTAB, 395 TCA_CBQ_POLICE, 396 __TCA_CBQ_MAX, 397}; 398 399#define TCA_CBQ_MAX (__TCA_CBQ_MAX - 1) 400 401/* dsmark section */ 402 403enum { 404 TCA_DSMARK_UNSPEC, 405 TCA_DSMARK_INDICES, 406 TCA_DSMARK_DEFAULT_INDEX, 407 TCA_DSMARK_SET_TC_INDEX, 408 TCA_DSMARK_MASK, 409 TCA_DSMARK_VALUE, 410 __TCA_DSMARK_MAX, 411}; 412 413#define TCA_DSMARK_MAX (__TCA_DSMARK_MAX - 1) 414 415/* ATM section */ 416 417enum { 418 TCA_ATM_UNSPEC, 419 TCA_ATM_FD, /* file/socket descriptor */ 420 TCA_ATM_PTR, /* pointer to descriptor - later */ 421 TCA_ATM_HDR, /* LL header */ 422 TCA_ATM_EXCESS, /* excess traffic class (0 for CLP) */ 423 TCA_ATM_ADDR, /* PVC address (for output only) */ 424 TCA_ATM_STATE, /* VC state (ATM_VS_*; for output only) */ 425 __TCA_ATM_MAX, 426}; 427 428#define TCA_ATM_MAX (__TCA_ATM_MAX - 1) 429 430/* Network emulator */ 431 432enum { 433 TCA_NETEM_UNSPEC, 434 TCA_NETEM_CORR, 435 TCA_NETEM_DELAY_DIST, 436 TCA_NETEM_REORDER, 437 TCA_NETEM_CORRUPT, 438 __TCA_NETEM_MAX, 439}; 440 441#define TCA_NETEM_MAX (__TCA_NETEM_MAX - 1) 442 443struct tc_netem_qopt { 444 __u32 latency; /* added delay (us) */ 445 __u32 limit; /* fifo limit (packets) */ 446 __u32 loss; /* random packet loss (0=none ~0=100%) */ 447 __u32 gap; /* re-ordering gap (0 for none) */ 448 __u32 duplicate; /* random packet dup (0=none ~0=100%) */ 449 __u32 jitter; /* random jitter in latency (us) */ 450}; 451 452struct tc_netem_corr { 453 __u32 delay_corr; /* delay correlation */ 454 __u32 loss_corr; /* packet loss correlation */ 455 __u32 dup_corr; /* duplicate correlation */ 456}; 457 458struct tc_netem_reorder { 459 __u32 probability; 460 __u32 correlation; 461}; 462 463struct tc_netem_corrupt { 464 __u32 probability; 465 __u32 correlation; 466}; 467 468#define NETEM_DIST_SCALE 8192 469 470/* DRR */ 471 472enum { 473 TCA_DRR_UNSPEC, 474 TCA_DRR_QUANTUM, 475 __TCA_DRR_MAX 476}; 477 478#define TCA_DRR_MAX (__TCA_DRR_MAX - 1) 479 480struct tc_drr_stats { 481 __u32 deficit; 482}; 483 484#endif 485