pkt_sched.h revision 7878c0ba404c3688442d2362203d79fa108450e0
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_sfqred_stats { 166 __u32 prob_drop; /* Early drops, below max threshold */ 167 __u32 forced_drop; /* Early drops, after max threshold */ 168 __u32 prob_mark; /* Marked packets, below max threshold */ 169 __u32 forced_mark; /* Marked packets, after max threshold */ 170 __u32 prob_mark_head; /* Marked packets, below max threshold */ 171 __u32 forced_mark_head;/* Marked packets, after max threshold */ 172}; 173 174struct tc_sfq_qopt_v1 { 175 struct tc_sfq_qopt v0; 176 unsigned int depth; /* max number of packets per flow */ 177 unsigned int headdrop; 178/* SFQRED parameters */ 179 __u32 limit; /* HARD maximal flow queue length (bytes) */ 180 __u32 qth_min; /* Min average length threshold (bytes) */ 181 __u32 qth_max; /* Max average length threshold (bytes) */ 182 unsigned char Wlog; /* log(W) */ 183 unsigned char Plog; /* log(P_max/(qth_max-qth_min)) */ 184 unsigned char Scell_log; /* cell size for idle damping */ 185 unsigned char flags; 186 __u32 max_P; /* probability, high resolution */ 187/* SFQRED stats */ 188 struct tc_sfqred_stats stats; 189}; 190 191 192struct tc_sfq_xstats { 193 __s32 allot; 194}; 195 196/* RED section */ 197 198enum { 199 TCA_RED_UNSPEC, 200 TCA_RED_PARMS, 201 TCA_RED_STAB, 202 TCA_RED_MAX_P, 203 __TCA_RED_MAX, 204}; 205 206#define TCA_RED_MAX (__TCA_RED_MAX - 1) 207 208struct tc_red_qopt { 209 __u32 limit; /* HARD maximal queue length (bytes) */ 210 __u32 qth_min; /* Min average length threshold (bytes) */ 211 __u32 qth_max; /* Max average length threshold (bytes) */ 212 unsigned char Wlog; /* log(W) */ 213 unsigned char Plog; /* log(P_max/(qth_max-qth_min)) */ 214 unsigned char Scell_log; /* cell size for idle damping */ 215 unsigned char flags; 216#define TC_RED_ECN 1 217#define TC_RED_HARDDROP 2 218#define TC_RED_ADAPTATIVE 4 219}; 220 221struct tc_red_xstats { 222 __u32 early; /* Early drops */ 223 __u32 pdrop; /* Drops due to queue limits */ 224 __u32 other; /* Drops due to drop() calls */ 225 __u32 marked; /* Marked packets */ 226}; 227 228/* GRED section */ 229 230#define MAX_DPs 16 231 232enum { 233 TCA_GRED_UNSPEC, 234 TCA_GRED_PARMS, 235 TCA_GRED_STAB, 236 TCA_GRED_DPS, 237 TCA_GRED_MAX_P, 238 __TCA_GRED_MAX, 239}; 240 241#define TCA_GRED_MAX (__TCA_GRED_MAX - 1) 242 243struct tc_gred_qopt { 244 __u32 limit; /* HARD maximal queue length (bytes) */ 245 __u32 qth_min; /* Min average length threshold (bytes) */ 246 __u32 qth_max; /* Max average length threshold (bytes) */ 247 __u32 DP; /* up to 2^32 DPs */ 248 __u32 backlog; 249 __u32 qave; 250 __u32 forced; 251 __u32 early; 252 __u32 other; 253 __u32 pdrop; 254 __u8 Wlog; /* log(W) */ 255 __u8 Plog; /* log(P_max/(qth_max-qth_min)) */ 256 __u8 Scell_log; /* cell size for idle damping */ 257 __u8 prio; /* prio of this VQ */ 258 __u32 packets; 259 __u32 bytesin; 260}; 261 262/* gred setup */ 263struct tc_gred_sopt { 264 __u32 DPs; 265 __u32 def_DP; 266 __u8 grio; 267 __u8 flags; 268 __u16 pad1; 269}; 270 271/* CHOKe section */ 272 273enum { 274 TCA_CHOKE_UNSPEC, 275 TCA_CHOKE_PARMS, 276 TCA_CHOKE_STAB, 277 TCA_CHOKE_MAX_P, 278 __TCA_CHOKE_MAX, 279}; 280 281#define TCA_CHOKE_MAX (__TCA_CHOKE_MAX - 1) 282 283struct tc_choke_qopt { 284 __u32 limit; /* Hard queue length (packets) */ 285 __u32 qth_min; /* Min average threshold (packets) */ 286 __u32 qth_max; /* Max average threshold (packets) */ 287 unsigned char Wlog; /* log(W) */ 288 unsigned char Plog; /* log(P_max/(qth_max-qth_min)) */ 289 unsigned char Scell_log; /* cell size for idle damping */ 290 unsigned char flags; /* see RED flags */ 291}; 292 293struct tc_choke_xstats { 294 __u32 early; /* Early drops */ 295 __u32 pdrop; /* Drops due to queue limits */ 296 __u32 other; /* Drops due to drop() calls */ 297 __u32 marked; /* Marked packets */ 298 __u32 matched; /* Drops due to flow match */ 299}; 300 301/* HTB section */ 302#define TC_HTB_NUMPRIO 8 303#define TC_HTB_MAXDEPTH 8 304#define TC_HTB_PROTOVER 3 /* the same as HTB and TC's major */ 305 306struct tc_htb_opt { 307 struct tc_ratespec rate; 308 struct tc_ratespec ceil; 309 __u32 buffer; 310 __u32 cbuffer; 311 __u32 quantum; 312 __u32 level; /* out only */ 313 __u32 prio; 314}; 315struct tc_htb_glob { 316 __u32 version; /* to match HTB/TC */ 317 __u32 rate2quantum; /* bps->quantum divisor */ 318 __u32 defcls; /* default class number */ 319 __u32 debug; /* debug flags */ 320 321 /* stats */ 322 __u32 direct_pkts; /* count of non shaped packets */ 323}; 324enum { 325 TCA_HTB_UNSPEC, 326 TCA_HTB_PARMS, 327 TCA_HTB_INIT, 328 TCA_HTB_CTAB, 329 TCA_HTB_RTAB, 330 __TCA_HTB_MAX, 331}; 332 333#define TCA_HTB_MAX (__TCA_HTB_MAX - 1) 334 335struct tc_htb_xstats { 336 __u32 lends; 337 __u32 borrows; 338 __u32 giants; /* too big packets (rate will not be accurate) */ 339 __u32 tokens; 340 __u32 ctokens; 341}; 342 343/* HFSC section */ 344 345struct tc_hfsc_qopt { 346 __u16 defcls; /* default class */ 347}; 348 349struct tc_service_curve { 350 __u32 m1; /* slope of the first segment in bps */ 351 __u32 d; /* x-projection of the first segment in us */ 352 __u32 m2; /* slope of the second segment in bps */ 353}; 354 355struct tc_hfsc_stats { 356 __u64 work; /* total work done */ 357 __u64 rtwork; /* work done by real-time criteria */ 358 __u32 period; /* current period */ 359 __u32 level; /* class level in hierarchy */ 360}; 361 362enum { 363 TCA_HFSC_UNSPEC, 364 TCA_HFSC_RSC, 365 TCA_HFSC_FSC, 366 TCA_HFSC_USC, 367 __TCA_HFSC_MAX, 368}; 369 370#define TCA_HFSC_MAX (__TCA_HFSC_MAX - 1) 371 372 373/* CBQ section */ 374 375#define TC_CBQ_MAXPRIO 8 376#define TC_CBQ_MAXLEVEL 8 377#define TC_CBQ_DEF_EWMA 5 378 379struct tc_cbq_lssopt { 380 unsigned char change; 381 unsigned char flags; 382#define TCF_CBQ_LSS_BOUNDED 1 383#define TCF_CBQ_LSS_ISOLATED 2 384 unsigned char ewma_log; 385 unsigned char level; 386#define TCF_CBQ_LSS_FLAGS 1 387#define TCF_CBQ_LSS_EWMA 2 388#define TCF_CBQ_LSS_MAXIDLE 4 389#define TCF_CBQ_LSS_MINIDLE 8 390#define TCF_CBQ_LSS_OFFTIME 0x10 391#define TCF_CBQ_LSS_AVPKT 0x20 392 __u32 maxidle; 393 __u32 minidle; 394 __u32 offtime; 395 __u32 avpkt; 396}; 397 398struct tc_cbq_wrropt { 399 unsigned char flags; 400 unsigned char priority; 401 unsigned char cpriority; 402 unsigned char __reserved; 403 __u32 allot; 404 __u32 weight; 405}; 406 407struct tc_cbq_ovl { 408 unsigned char strategy; 409#define TC_CBQ_OVL_CLASSIC 0 410#define TC_CBQ_OVL_DELAY 1 411#define TC_CBQ_OVL_LOWPRIO 2 412#define TC_CBQ_OVL_DROP 3 413#define TC_CBQ_OVL_RCLASSIC 4 414 unsigned char priority2; 415 __u16 pad; 416 __u32 penalty; 417}; 418 419struct tc_cbq_police { 420 unsigned char police; 421 unsigned char __res1; 422 unsigned short __res2; 423}; 424 425struct tc_cbq_fopt { 426 __u32 split; 427 __u32 defmap; 428 __u32 defchange; 429}; 430 431struct tc_cbq_xstats { 432 __u32 borrows; 433 __u32 overactions; 434 __s32 avgidle; 435 __s32 undertime; 436}; 437 438enum { 439 TCA_CBQ_UNSPEC, 440 TCA_CBQ_LSSOPT, 441 TCA_CBQ_WRROPT, 442 TCA_CBQ_FOPT, 443 TCA_CBQ_OVL_STRATEGY, 444 TCA_CBQ_RATE, 445 TCA_CBQ_RTAB, 446 TCA_CBQ_POLICE, 447 __TCA_CBQ_MAX, 448}; 449 450#define TCA_CBQ_MAX (__TCA_CBQ_MAX - 1) 451 452/* dsmark section */ 453 454enum { 455 TCA_DSMARK_UNSPEC, 456 TCA_DSMARK_INDICES, 457 TCA_DSMARK_DEFAULT_INDEX, 458 TCA_DSMARK_SET_TC_INDEX, 459 TCA_DSMARK_MASK, 460 TCA_DSMARK_VALUE, 461 __TCA_DSMARK_MAX, 462}; 463 464#define TCA_DSMARK_MAX (__TCA_DSMARK_MAX - 1) 465 466/* ATM section */ 467 468enum { 469 TCA_ATM_UNSPEC, 470 TCA_ATM_FD, /* file/socket descriptor */ 471 TCA_ATM_PTR, /* pointer to descriptor - later */ 472 TCA_ATM_HDR, /* LL header */ 473 TCA_ATM_EXCESS, /* excess traffic class (0 for CLP) */ 474 TCA_ATM_ADDR, /* PVC address (for output only) */ 475 TCA_ATM_STATE, /* VC state (ATM_VS_*; for output only) */ 476 __TCA_ATM_MAX, 477}; 478 479#define TCA_ATM_MAX (__TCA_ATM_MAX - 1) 480 481/* Network emulator */ 482 483enum { 484 TCA_NETEM_UNSPEC, 485 TCA_NETEM_CORR, 486 TCA_NETEM_DELAY_DIST, 487 TCA_NETEM_REORDER, 488 TCA_NETEM_CORRUPT, 489 TCA_NETEM_LOSS, 490 TCA_NETEM_RATE, 491 __TCA_NETEM_MAX, 492}; 493 494#define TCA_NETEM_MAX (__TCA_NETEM_MAX - 1) 495 496struct tc_netem_qopt { 497 __u32 latency; /* added delay (us) */ 498 __u32 limit; /* fifo limit (packets) */ 499 __u32 loss; /* random packet loss (0=none ~0=100%) */ 500 __u32 gap; /* re-ordering gap (0 for none) */ 501 __u32 duplicate; /* random packet dup (0=none ~0=100%) */ 502 __u32 jitter; /* random jitter in latency (us) */ 503}; 504 505struct tc_netem_corr { 506 __u32 delay_corr; /* delay correlation */ 507 __u32 loss_corr; /* packet loss correlation */ 508 __u32 dup_corr; /* duplicate correlation */ 509}; 510 511struct tc_netem_reorder { 512 __u32 probability; 513 __u32 correlation; 514}; 515 516struct tc_netem_corrupt { 517 __u32 probability; 518 __u32 correlation; 519}; 520 521struct tc_netem_rate { 522 __u32 rate; /* byte/s */ 523 __s32 packet_overhead; 524 __u32 cell_size; 525 __s32 cell_overhead; 526}; 527 528enum { 529 NETEM_LOSS_UNSPEC, 530 NETEM_LOSS_GI, /* General Intuitive - 4 state model */ 531 NETEM_LOSS_GE, /* Gilbert Elliot models */ 532 __NETEM_LOSS_MAX 533}; 534#define NETEM_LOSS_MAX (__NETEM_LOSS_MAX - 1) 535 536/* State transition probabilities for 4 state model */ 537struct tc_netem_gimodel { 538 __u32 p13; 539 __u32 p31; 540 __u32 p32; 541 __u32 p14; 542 __u32 p23; 543}; 544 545/* Gilbert-Elliot models */ 546struct tc_netem_gemodel { 547 __u32 p; 548 __u32 r; 549 __u32 h; 550 __u32 k1; 551}; 552 553#define NETEM_DIST_SCALE 8192 554#define NETEM_DIST_MAX 16384 555 556/* DRR */ 557 558enum { 559 TCA_DRR_UNSPEC, 560 TCA_DRR_QUANTUM, 561 __TCA_DRR_MAX 562}; 563 564#define TCA_DRR_MAX (__TCA_DRR_MAX - 1) 565 566struct tc_drr_stats { 567 __u32 deficit; 568}; 569 570/* MQPRIO */ 571#define TC_QOPT_BITMASK 15 572#define TC_QOPT_MAX_QUEUE 16 573 574struct tc_mqprio_qopt { 575 __u8 num_tc; 576 __u8 prio_tc_map[TC_QOPT_BITMASK + 1]; 577 __u8 hw; 578 __u16 count[TC_QOPT_MAX_QUEUE]; 579 __u16 offset[TC_QOPT_MAX_QUEUE]; 580}; 581 582/* SFB */ 583 584enum { 585 TCA_SFB_UNSPEC, 586 TCA_SFB_PARMS, 587 __TCA_SFB_MAX, 588}; 589 590#define TCA_SFB_MAX (__TCA_SFB_MAX - 1) 591 592/* 593 * Note: increment, decrement are Q0.16 fixed-point values. 594 */ 595struct tc_sfb_qopt { 596 __u32 rehash_interval; /* delay between hash move, in ms */ 597 __u32 warmup_time; /* double buffering warmup time in ms (warmup_time < rehash_interval) */ 598 __u32 max; /* max len of qlen_min */ 599 __u32 bin_size; /* maximum queue length per bin */ 600 __u32 increment; /* probability increment, (d1 in Blue) */ 601 __u32 decrement; /* probability decrement, (d2 in Blue) */ 602 __u32 limit; /* max SFB queue length */ 603 __u32 penalty_rate; /* inelastic flows are rate limited to 'rate' pps */ 604 __u32 penalty_burst; 605}; 606 607struct tc_sfb_xstats { 608 __u32 earlydrop; 609 __u32 penaltydrop; 610 __u32 bucketdrop; 611 __u32 queuedrop; 612 __u32 childdrop; /* drops in child qdisc */ 613 __u32 marked; 614 __u32 maxqlen; 615 __u32 maxprob; 616 __u32 avgprob; 617}; 618 619#define SFB_MAX_PROB 0xFFFF 620 621/* QFQ */ 622enum { 623 TCA_QFQ_UNSPEC, 624 TCA_QFQ_WEIGHT, 625 TCA_QFQ_LMAX, 626 __TCA_QFQ_MAX 627}; 628 629#define TCA_QFQ_MAX (__TCA_QFQ_MAX - 1) 630 631struct tc_qfq_stats { 632 __u32 weight; 633 __u32 lmax; 634}; 635 636#endif 637