pkt_sched.h revision 1b1177ed5f3a8e34b1c7c7c79922c31448f212c9
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 __u64 bytes; /* Number of enqueued bytes */ 34 __u32 packets; /* Number of enqueued packets */ 35 __u32 drops; /* Packets dropped because of lack of resources */ 36 __u32 overlimits; /* Number of throttle events when this 37 * flow goes out of allocated bandwidth */ 38 __u32 bps; /* Current flow byte rate */ 39 __u32 pps; /* Current flow packet rate */ 40 __u32 qlen; 41 __u32 backlog; 42}; 43 44struct tc_estimator { 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 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; /* up to 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/* CHOKe section */ 251 252enum { 253 TCA_CHOKE_UNSPEC, 254 TCA_CHOKE_PARMS, 255 TCA_CHOKE_STAB, 256 __TCA_CHOKE_MAX, 257}; 258 259#define TCA_CHOKE_MAX (__TCA_CHOKE_MAX - 1) 260 261struct tc_choke_qopt { 262 __u32 limit; /* Hard queue length (packets) */ 263 __u32 qth_min; /* Min average threshold (packets) */ 264 __u32 qth_max; /* Max average threshold (packets) */ 265 unsigned char Wlog; /* log(W) */ 266 unsigned char Plog; /* log(P_max/(qth_max-qth_min)) */ 267 unsigned char Scell_log; /* cell size for idle damping */ 268 unsigned char flags; /* see RED flags */ 269}; 270 271struct tc_choke_xstats { 272 __u32 early; /* Early drops */ 273 __u32 pdrop; /* Drops due to queue limits */ 274 __u32 other; /* Drops due to drop() calls */ 275 __u32 marked; /* Marked packets */ 276 __u32 matched; /* Drops due to flow match */ 277}; 278 279/* HTB section */ 280#define TC_HTB_NUMPRIO 8 281#define TC_HTB_MAXDEPTH 8 282#define TC_HTB_PROTOVER 3 /* the same as HTB and TC's major */ 283 284struct tc_htb_opt { 285 struct tc_ratespec rate; 286 struct tc_ratespec ceil; 287 __u32 buffer; 288 __u32 cbuffer; 289 __u32 quantum; 290 __u32 level; /* out only */ 291 __u32 prio; 292}; 293struct tc_htb_glob { 294 __u32 version; /* to match HTB/TC */ 295 __u32 rate2quantum; /* bps->quantum divisor */ 296 __u32 defcls; /* default class number */ 297 __u32 debug; /* debug flags */ 298 299 /* stats */ 300 __u32 direct_pkts; /* count of non shaped packets */ 301}; 302enum { 303 TCA_HTB_UNSPEC, 304 TCA_HTB_PARMS, 305 TCA_HTB_INIT, 306 TCA_HTB_CTAB, 307 TCA_HTB_RTAB, 308 __TCA_HTB_MAX, 309}; 310 311#define TCA_HTB_MAX (__TCA_HTB_MAX - 1) 312 313struct tc_htb_xstats { 314 __u32 lends; 315 __u32 borrows; 316 __u32 giants; /* too big packets (rate will not be accurate) */ 317 __u32 tokens; 318 __u32 ctokens; 319}; 320 321/* HFSC section */ 322 323struct tc_hfsc_qopt { 324 __u16 defcls; /* default class */ 325}; 326 327struct tc_service_curve { 328 __u32 m1; /* slope of the first segment in bps */ 329 __u32 d; /* x-projection of the first segment in us */ 330 __u32 m2; /* slope of the second segment in bps */ 331}; 332 333struct tc_hfsc_stats { 334 __u64 work; /* total work done */ 335 __u64 rtwork; /* work done by real-time criteria */ 336 __u32 period; /* current period */ 337 __u32 level; /* class level in hierarchy */ 338}; 339 340enum { 341 TCA_HFSC_UNSPEC, 342 TCA_HFSC_RSC, 343 TCA_HFSC_FSC, 344 TCA_HFSC_USC, 345 __TCA_HFSC_MAX, 346}; 347 348#define TCA_HFSC_MAX (__TCA_HFSC_MAX - 1) 349 350 351/* CBQ section */ 352 353#define TC_CBQ_MAXPRIO 8 354#define TC_CBQ_MAXLEVEL 8 355#define TC_CBQ_DEF_EWMA 5 356 357struct tc_cbq_lssopt { 358 unsigned char change; 359 unsigned char flags; 360#define TCF_CBQ_LSS_BOUNDED 1 361#define TCF_CBQ_LSS_ISOLATED 2 362 unsigned char ewma_log; 363 unsigned char level; 364#define TCF_CBQ_LSS_FLAGS 1 365#define TCF_CBQ_LSS_EWMA 2 366#define TCF_CBQ_LSS_MAXIDLE 4 367#define TCF_CBQ_LSS_MINIDLE 8 368#define TCF_CBQ_LSS_OFFTIME 0x10 369#define TCF_CBQ_LSS_AVPKT 0x20 370 __u32 maxidle; 371 __u32 minidle; 372 __u32 offtime; 373 __u32 avpkt; 374}; 375 376struct tc_cbq_wrropt { 377 unsigned char flags; 378 unsigned char priority; 379 unsigned char cpriority; 380 unsigned char __reserved; 381 __u32 allot; 382 __u32 weight; 383}; 384 385struct tc_cbq_ovl { 386 unsigned char strategy; 387#define TC_CBQ_OVL_CLASSIC 0 388#define TC_CBQ_OVL_DELAY 1 389#define TC_CBQ_OVL_LOWPRIO 2 390#define TC_CBQ_OVL_DROP 3 391#define TC_CBQ_OVL_RCLASSIC 4 392 unsigned char priority2; 393 __u16 pad; 394 __u32 penalty; 395}; 396 397struct tc_cbq_police { 398 unsigned char police; 399 unsigned char __res1; 400 unsigned short __res2; 401}; 402 403struct tc_cbq_fopt { 404 __u32 split; 405 __u32 defmap; 406 __u32 defchange; 407}; 408 409struct tc_cbq_xstats { 410 __u32 borrows; 411 __u32 overactions; 412 __s32 avgidle; 413 __s32 undertime; 414}; 415 416enum { 417 TCA_CBQ_UNSPEC, 418 TCA_CBQ_LSSOPT, 419 TCA_CBQ_WRROPT, 420 TCA_CBQ_FOPT, 421 TCA_CBQ_OVL_STRATEGY, 422 TCA_CBQ_RATE, 423 TCA_CBQ_RTAB, 424 TCA_CBQ_POLICE, 425 __TCA_CBQ_MAX, 426}; 427 428#define TCA_CBQ_MAX (__TCA_CBQ_MAX - 1) 429 430/* dsmark section */ 431 432enum { 433 TCA_DSMARK_UNSPEC, 434 TCA_DSMARK_INDICES, 435 TCA_DSMARK_DEFAULT_INDEX, 436 TCA_DSMARK_SET_TC_INDEX, 437 TCA_DSMARK_MASK, 438 TCA_DSMARK_VALUE, 439 __TCA_DSMARK_MAX, 440}; 441 442#define TCA_DSMARK_MAX (__TCA_DSMARK_MAX - 1) 443 444/* ATM section */ 445 446enum { 447 TCA_ATM_UNSPEC, 448 TCA_ATM_FD, /* file/socket descriptor */ 449 TCA_ATM_PTR, /* pointer to descriptor - later */ 450 TCA_ATM_HDR, /* LL header */ 451 TCA_ATM_EXCESS, /* excess traffic class (0 for CLP) */ 452 TCA_ATM_ADDR, /* PVC address (for output only) */ 453 TCA_ATM_STATE, /* VC state (ATM_VS_*; for output only) */ 454 __TCA_ATM_MAX, 455}; 456 457#define TCA_ATM_MAX (__TCA_ATM_MAX - 1) 458 459/* Network emulator */ 460 461enum { 462 TCA_NETEM_UNSPEC, 463 TCA_NETEM_CORR, 464 TCA_NETEM_DELAY_DIST, 465 TCA_NETEM_REORDER, 466 TCA_NETEM_CORRUPT, 467 TCA_NETEM_LOSS, 468 __TCA_NETEM_MAX, 469}; 470 471#define TCA_NETEM_MAX (__TCA_NETEM_MAX - 1) 472 473struct tc_netem_qopt { 474 __u32 latency; /* added delay (us) */ 475 __u32 limit; /* fifo limit (packets) */ 476 __u32 loss; /* random packet loss (0=none ~0=100%) */ 477 __u32 gap; /* re-ordering gap (0 for none) */ 478 __u32 duplicate; /* random packet dup (0=none ~0=100%) */ 479 __u32 jitter; /* random jitter in latency (us) */ 480}; 481 482struct tc_netem_corr { 483 __u32 delay_corr; /* delay correlation */ 484 __u32 loss_corr; /* packet loss correlation */ 485 __u32 dup_corr; /* duplicate correlation */ 486}; 487 488struct tc_netem_reorder { 489 __u32 probability; 490 __u32 correlation; 491}; 492 493struct tc_netem_corrupt { 494 __u32 probability; 495 __u32 correlation; 496}; 497 498enum { 499 NETEM_LOSS_UNSPEC, 500 NETEM_LOSS_GI, /* General Intuitive - 4 state model */ 501 NETEM_LOSS_GE, /* Gilbert Elliot models */ 502 __NETEM_LOSS_MAX 503}; 504#define NETEM_LOSS_MAX (__NETEM_LOSS_MAX - 1) 505 506/* State transition probabilities for 4 state model */ 507struct tc_netem_gimodel { 508 __u32 p13; 509 __u32 p31; 510 __u32 p32; 511 __u32 p14; 512 __u32 p23; 513}; 514 515/* Gilbert-Elliot models */ 516struct tc_netem_gemodel { 517 __u32 p; 518 __u32 r; 519 __u32 h; 520 __u32 k1; 521}; 522 523#define NETEM_DIST_SCALE 8192 524#define NETEM_DIST_MAX 16384 525 526/* DRR */ 527 528enum { 529 TCA_DRR_UNSPEC, 530 TCA_DRR_QUANTUM, 531 __TCA_DRR_MAX 532}; 533 534#define TCA_DRR_MAX (__TCA_DRR_MAX - 1) 535 536struct tc_drr_stats { 537 __u32 deficit; 538}; 539 540/* MQPRIO */ 541#define TC_QOPT_BITMASK 15 542#define TC_QOPT_MAX_QUEUE 16 543 544struct tc_mqprio_qopt { 545 __u8 num_tc; 546 __u8 prio_tc_map[TC_QOPT_BITMASK + 1]; 547 __u8 hw; 548 __u16 count[TC_QOPT_MAX_QUEUE]; 549 __u16 offset[TC_QOPT_MAX_QUEUE]; 550}; 551 552/* SFB */ 553 554enum { 555 TCA_SFB_UNSPEC, 556 TCA_SFB_PARMS, 557 __TCA_SFB_MAX, 558}; 559 560#define TCA_SFB_MAX (__TCA_SFB_MAX - 1) 561 562/* 563 * Note: increment, decrement are Q0.16 fixed-point values. 564 */ 565struct tc_sfb_qopt { 566 __u32 rehash_interval; /* delay between hash move, in ms */ 567 __u32 warmup_time; /* double buffering warmup time in ms (warmup_time < rehash_interval) */ 568 __u32 max; /* max len of qlen_min */ 569 __u32 bin_size; /* maximum queue length per bin */ 570 __u32 increment; /* probability increment, (d1 in Blue) */ 571 __u32 decrement; /* probability decrement, (d2 in Blue) */ 572 __u32 limit; /* max SFB queue length */ 573 __u32 penalty_rate; /* inelastic flows are rate limited to 'rate' pps */ 574 __u32 penalty_burst; 575}; 576 577struct tc_sfb_xstats { 578 __u32 earlydrop; 579 __u32 penaltydrop; 580 __u32 bucketdrop; 581 __u32 queuedrop; 582 __u32 childdrop; /* drops in child qdisc */ 583 __u32 marked; 584 __u32 maxqlen; 585 __u32 maxprob; 586 __u32 avgprob; 587}; 588 589#define SFB_MAX_PROB 0xFFFF 590 591/* QFQ */ 592enum { 593 TCA_QFQ_UNSPEC, 594 TCA_QFQ_WEIGHT, 595 TCA_QFQ_LMAX, 596 __TCA_QFQ_MAX 597}; 598 599#define TCA_QFQ_MAX (__TCA_QFQ_MAX - 1) 600 601struct tc_qfq_stats { 602 __u32 weight; 603 __u32 lmax; 604}; 605 606#endif 607