pkt_sched.h revision 29cea29df0dfe45562fd42673e26608d25180eee
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/* PLUG section */ 131 132#define TCQ_PLUG_BUFFER 0 133#define TCQ_PLUG_RELEASE_ONE 1 134#define TCQ_PLUG_RELEASE_INDEFINITE 2 135#define TCQ_PLUG_LIMIT 3 136 137struct tc_plug_qopt { 138 /* TCQ_PLUG_BUFFER: Inset a plug into the queue and 139 * buffer any incoming packets 140 * TCQ_PLUG_RELEASE_ONE: Dequeue packets from queue head 141 * to beginning of the next plug. 142 * TCQ_PLUG_RELEASE_INDEFINITE: Dequeue all packets from queue. 143 * Stop buffering packets until the next TCQ_PLUG_BUFFER 144 * command is received (just act as a pass-thru queue). 145 * TCQ_PLUG_LIMIT: Increase/decrease queue size 146 */ 147 int action; 148 __u32 limit; 149}; 150 151/* TBF section */ 152 153struct tc_tbf_qopt { 154 struct tc_ratespec rate; 155 struct tc_ratespec peakrate; 156 __u32 limit; 157 __u32 buffer; 158 __u32 mtu; 159}; 160 161enum { 162 TCA_TBF_UNSPEC, 163 TCA_TBF_PARMS, 164 TCA_TBF_RTAB, 165 TCA_TBF_PTAB, 166 __TCA_TBF_MAX, 167}; 168 169#define TCA_TBF_MAX (__TCA_TBF_MAX - 1) 170 171 172/* TEQL section */ 173 174/* TEQL does not require any parameters */ 175 176/* SFQ section */ 177 178struct tc_sfq_qopt { 179 unsigned quantum; /* Bytes per round allocated to flow */ 180 int perturb_period; /* Period of hash perturbation */ 181 __u32 limit; /* Maximal packets in queue */ 182 unsigned divisor; /* Hash divisor */ 183 unsigned flows; /* Maximal number of flows */ 184}; 185 186struct tc_sfqred_stats { 187 __u32 prob_drop; /* Early drops, below max threshold */ 188 __u32 forced_drop; /* Early drops, after max threshold */ 189 __u32 prob_mark; /* Marked packets, below max threshold */ 190 __u32 forced_mark; /* Marked packets, after max threshold */ 191 __u32 prob_mark_head; /* Marked packets, below max threshold */ 192 __u32 forced_mark_head;/* Marked packets, after max threshold */ 193}; 194 195struct tc_sfq_qopt_v1 { 196 struct tc_sfq_qopt v0; 197 unsigned int depth; /* max number of packets per flow */ 198 unsigned int headdrop; 199/* SFQRED parameters */ 200 __u32 limit; /* HARD maximal flow queue length (bytes) */ 201 __u32 qth_min; /* Min average length threshold (bytes) */ 202 __u32 qth_max; /* Max average length threshold (bytes) */ 203 unsigned char Wlog; /* log(W) */ 204 unsigned char Plog; /* log(P_max/(qth_max-qth_min)) */ 205 unsigned char Scell_log; /* cell size for idle damping */ 206 unsigned char flags; 207 __u32 max_P; /* probability, high resolution */ 208/* SFQRED stats */ 209 struct tc_sfqred_stats stats; 210}; 211 212 213struct tc_sfq_xstats { 214 __s32 allot; 215}; 216 217/* RED section */ 218 219enum { 220 TCA_RED_UNSPEC, 221 TCA_RED_PARMS, 222 TCA_RED_STAB, 223 TCA_RED_MAX_P, 224 __TCA_RED_MAX, 225}; 226 227#define TCA_RED_MAX (__TCA_RED_MAX - 1) 228 229struct tc_red_qopt { 230 __u32 limit; /* HARD maximal queue length (bytes) */ 231 __u32 qth_min; /* Min average length threshold (bytes) */ 232 __u32 qth_max; /* Max average length threshold (bytes) */ 233 unsigned char Wlog; /* log(W) */ 234 unsigned char Plog; /* log(P_max/(qth_max-qth_min)) */ 235 unsigned char Scell_log; /* cell size for idle damping */ 236 unsigned char flags; 237#define TC_RED_ECN 1 238#define TC_RED_HARDDROP 2 239#define TC_RED_ADAPTATIVE 4 240}; 241 242struct tc_red_xstats { 243 __u32 early; /* Early drops */ 244 __u32 pdrop; /* Drops due to queue limits */ 245 __u32 other; /* Drops due to drop() calls */ 246 __u32 marked; /* Marked packets */ 247}; 248 249/* GRED section */ 250 251#define MAX_DPs 16 252 253enum { 254 TCA_GRED_UNSPEC, 255 TCA_GRED_PARMS, 256 TCA_GRED_STAB, 257 TCA_GRED_DPS, 258 TCA_GRED_MAX_P, 259 __TCA_GRED_MAX, 260}; 261 262#define TCA_GRED_MAX (__TCA_GRED_MAX - 1) 263 264struct tc_gred_qopt { 265 __u32 limit; /* HARD maximal queue length (bytes) */ 266 __u32 qth_min; /* Min average length threshold (bytes) */ 267 __u32 qth_max; /* Max average length threshold (bytes) */ 268 __u32 DP; /* up to 2^32 DPs */ 269 __u32 backlog; 270 __u32 qave; 271 __u32 forced; 272 __u32 early; 273 __u32 other; 274 __u32 pdrop; 275 __u8 Wlog; /* log(W) */ 276 __u8 Plog; /* log(P_max/(qth_max-qth_min)) */ 277 __u8 Scell_log; /* cell size for idle damping */ 278 __u8 prio; /* prio of this VQ */ 279 __u32 packets; 280 __u32 bytesin; 281}; 282 283/* gred setup */ 284struct tc_gred_sopt { 285 __u32 DPs; 286 __u32 def_DP; 287 __u8 grio; 288 __u8 flags; 289 __u16 pad1; 290}; 291 292/* CHOKe section */ 293 294enum { 295 TCA_CHOKE_UNSPEC, 296 TCA_CHOKE_PARMS, 297 TCA_CHOKE_STAB, 298 TCA_CHOKE_MAX_P, 299 __TCA_CHOKE_MAX, 300}; 301 302#define TCA_CHOKE_MAX (__TCA_CHOKE_MAX - 1) 303 304struct tc_choke_qopt { 305 __u32 limit; /* Hard queue length (packets) */ 306 __u32 qth_min; /* Min average threshold (packets) */ 307 __u32 qth_max; /* Max average threshold (packets) */ 308 unsigned char Wlog; /* log(W) */ 309 unsigned char Plog; /* log(P_max/(qth_max-qth_min)) */ 310 unsigned char Scell_log; /* cell size for idle damping */ 311 unsigned char flags; /* see RED flags */ 312}; 313 314struct tc_choke_xstats { 315 __u32 early; /* Early drops */ 316 __u32 pdrop; /* Drops due to queue limits */ 317 __u32 other; /* Drops due to drop() calls */ 318 __u32 marked; /* Marked packets */ 319 __u32 matched; /* Drops due to flow match */ 320}; 321 322/* HTB section */ 323#define TC_HTB_NUMPRIO 8 324#define TC_HTB_MAXDEPTH 8 325#define TC_HTB_PROTOVER 3 /* the same as HTB and TC's major */ 326 327struct tc_htb_opt { 328 struct tc_ratespec rate; 329 struct tc_ratespec ceil; 330 __u32 buffer; 331 __u32 cbuffer; 332 __u32 quantum; 333 __u32 level; /* out only */ 334 __u32 prio; 335}; 336struct tc_htb_glob { 337 __u32 version; /* to match HTB/TC */ 338 __u32 rate2quantum; /* bps->quantum divisor */ 339 __u32 defcls; /* default class number */ 340 __u32 debug; /* debug flags */ 341 342 /* stats */ 343 __u32 direct_pkts; /* count of non shaped packets */ 344}; 345enum { 346 TCA_HTB_UNSPEC, 347 TCA_HTB_PARMS, 348 TCA_HTB_INIT, 349 TCA_HTB_CTAB, 350 TCA_HTB_RTAB, 351 __TCA_HTB_MAX, 352}; 353 354#define TCA_HTB_MAX (__TCA_HTB_MAX - 1) 355 356struct tc_htb_xstats { 357 __u32 lends; 358 __u32 borrows; 359 __u32 giants; /* too big packets (rate will not be accurate) */ 360 __u32 tokens; 361 __u32 ctokens; 362}; 363 364/* HFSC section */ 365 366struct tc_hfsc_qopt { 367 __u16 defcls; /* default class */ 368}; 369 370struct tc_service_curve { 371 __u32 m1; /* slope of the first segment in bps */ 372 __u32 d; /* x-projection of the first segment in us */ 373 __u32 m2; /* slope of the second segment in bps */ 374}; 375 376struct tc_hfsc_stats { 377 __u64 work; /* total work done */ 378 __u64 rtwork; /* work done by real-time criteria */ 379 __u32 period; /* current period */ 380 __u32 level; /* class level in hierarchy */ 381}; 382 383enum { 384 TCA_HFSC_UNSPEC, 385 TCA_HFSC_RSC, 386 TCA_HFSC_FSC, 387 TCA_HFSC_USC, 388 __TCA_HFSC_MAX, 389}; 390 391#define TCA_HFSC_MAX (__TCA_HFSC_MAX - 1) 392 393 394/* CBQ section */ 395 396#define TC_CBQ_MAXPRIO 8 397#define TC_CBQ_MAXLEVEL 8 398#define TC_CBQ_DEF_EWMA 5 399 400struct tc_cbq_lssopt { 401 unsigned char change; 402 unsigned char flags; 403#define TCF_CBQ_LSS_BOUNDED 1 404#define TCF_CBQ_LSS_ISOLATED 2 405 unsigned char ewma_log; 406 unsigned char level; 407#define TCF_CBQ_LSS_FLAGS 1 408#define TCF_CBQ_LSS_EWMA 2 409#define TCF_CBQ_LSS_MAXIDLE 4 410#define TCF_CBQ_LSS_MINIDLE 8 411#define TCF_CBQ_LSS_OFFTIME 0x10 412#define TCF_CBQ_LSS_AVPKT 0x20 413 __u32 maxidle; 414 __u32 minidle; 415 __u32 offtime; 416 __u32 avpkt; 417}; 418 419struct tc_cbq_wrropt { 420 unsigned char flags; 421 unsigned char priority; 422 unsigned char cpriority; 423 unsigned char __reserved; 424 __u32 allot; 425 __u32 weight; 426}; 427 428struct tc_cbq_ovl { 429 unsigned char strategy; 430#define TC_CBQ_OVL_CLASSIC 0 431#define TC_CBQ_OVL_DELAY 1 432#define TC_CBQ_OVL_LOWPRIO 2 433#define TC_CBQ_OVL_DROP 3 434#define TC_CBQ_OVL_RCLASSIC 4 435 unsigned char priority2; 436 __u16 pad; 437 __u32 penalty; 438}; 439 440struct tc_cbq_police { 441 unsigned char police; 442 unsigned char __res1; 443 unsigned short __res2; 444}; 445 446struct tc_cbq_fopt { 447 __u32 split; 448 __u32 defmap; 449 __u32 defchange; 450}; 451 452struct tc_cbq_xstats { 453 __u32 borrows; 454 __u32 overactions; 455 __s32 avgidle; 456 __s32 undertime; 457}; 458 459enum { 460 TCA_CBQ_UNSPEC, 461 TCA_CBQ_LSSOPT, 462 TCA_CBQ_WRROPT, 463 TCA_CBQ_FOPT, 464 TCA_CBQ_OVL_STRATEGY, 465 TCA_CBQ_RATE, 466 TCA_CBQ_RTAB, 467 TCA_CBQ_POLICE, 468 __TCA_CBQ_MAX, 469}; 470 471#define TCA_CBQ_MAX (__TCA_CBQ_MAX - 1) 472 473/* dsmark section */ 474 475enum { 476 TCA_DSMARK_UNSPEC, 477 TCA_DSMARK_INDICES, 478 TCA_DSMARK_DEFAULT_INDEX, 479 TCA_DSMARK_SET_TC_INDEX, 480 TCA_DSMARK_MASK, 481 TCA_DSMARK_VALUE, 482 __TCA_DSMARK_MAX, 483}; 484 485#define TCA_DSMARK_MAX (__TCA_DSMARK_MAX - 1) 486 487/* ATM section */ 488 489enum { 490 TCA_ATM_UNSPEC, 491 TCA_ATM_FD, /* file/socket descriptor */ 492 TCA_ATM_PTR, /* pointer to descriptor - later */ 493 TCA_ATM_HDR, /* LL header */ 494 TCA_ATM_EXCESS, /* excess traffic class (0 for CLP) */ 495 TCA_ATM_ADDR, /* PVC address (for output only) */ 496 TCA_ATM_STATE, /* VC state (ATM_VS_*; for output only) */ 497 __TCA_ATM_MAX, 498}; 499 500#define TCA_ATM_MAX (__TCA_ATM_MAX - 1) 501 502/* Network emulator */ 503 504enum { 505 TCA_NETEM_UNSPEC, 506 TCA_NETEM_CORR, 507 TCA_NETEM_DELAY_DIST, 508 TCA_NETEM_REORDER, 509 TCA_NETEM_CORRUPT, 510 TCA_NETEM_LOSS, 511 TCA_NETEM_RATE, 512 __TCA_NETEM_MAX, 513}; 514 515#define TCA_NETEM_MAX (__TCA_NETEM_MAX - 1) 516 517struct tc_netem_qopt { 518 __u32 latency; /* added delay (us) */ 519 __u32 limit; /* fifo limit (packets) */ 520 __u32 loss; /* random packet loss (0=none ~0=100%) */ 521 __u32 gap; /* re-ordering gap (0 for none) */ 522 __u32 duplicate; /* random packet dup (0=none ~0=100%) */ 523 __u32 jitter; /* random jitter in latency (us) */ 524}; 525 526struct tc_netem_corr { 527 __u32 delay_corr; /* delay correlation */ 528 __u32 loss_corr; /* packet loss correlation */ 529 __u32 dup_corr; /* duplicate correlation */ 530}; 531 532struct tc_netem_reorder { 533 __u32 probability; 534 __u32 correlation; 535}; 536 537struct tc_netem_corrupt { 538 __u32 probability; 539 __u32 correlation; 540}; 541 542struct tc_netem_rate { 543 __u32 rate; /* byte/s */ 544 __s32 packet_overhead; 545 __u32 cell_size; 546 __s32 cell_overhead; 547}; 548 549enum { 550 NETEM_LOSS_UNSPEC, 551 NETEM_LOSS_GI, /* General Intuitive - 4 state model */ 552 NETEM_LOSS_GE, /* Gilbert Elliot models */ 553 __NETEM_LOSS_MAX 554}; 555#define NETEM_LOSS_MAX (__NETEM_LOSS_MAX - 1) 556 557/* State transition probabilities for 4 state model */ 558struct tc_netem_gimodel { 559 __u32 p13; 560 __u32 p31; 561 __u32 p32; 562 __u32 p14; 563 __u32 p23; 564}; 565 566/* Gilbert-Elliot models */ 567struct tc_netem_gemodel { 568 __u32 p; 569 __u32 r; 570 __u32 h; 571 __u32 k1; 572}; 573 574#define NETEM_DIST_SCALE 8192 575#define NETEM_DIST_MAX 16384 576 577/* DRR */ 578 579enum { 580 TCA_DRR_UNSPEC, 581 TCA_DRR_QUANTUM, 582 __TCA_DRR_MAX 583}; 584 585#define TCA_DRR_MAX (__TCA_DRR_MAX - 1) 586 587struct tc_drr_stats { 588 __u32 deficit; 589}; 590 591/* MQPRIO */ 592#define TC_QOPT_BITMASK 15 593#define TC_QOPT_MAX_QUEUE 16 594 595struct tc_mqprio_qopt { 596 __u8 num_tc; 597 __u8 prio_tc_map[TC_QOPT_BITMASK + 1]; 598 __u8 hw; 599 __u16 count[TC_QOPT_MAX_QUEUE]; 600 __u16 offset[TC_QOPT_MAX_QUEUE]; 601}; 602 603/* SFB */ 604 605enum { 606 TCA_SFB_UNSPEC, 607 TCA_SFB_PARMS, 608 __TCA_SFB_MAX, 609}; 610 611#define TCA_SFB_MAX (__TCA_SFB_MAX - 1) 612 613/* 614 * Note: increment, decrement are Q0.16 fixed-point values. 615 */ 616struct tc_sfb_qopt { 617 __u32 rehash_interval; /* delay between hash move, in ms */ 618 __u32 warmup_time; /* double buffering warmup time in ms (warmup_time < rehash_interval) */ 619 __u32 max; /* max len of qlen_min */ 620 __u32 bin_size; /* maximum queue length per bin */ 621 __u32 increment; /* probability increment, (d1 in Blue) */ 622 __u32 decrement; /* probability decrement, (d2 in Blue) */ 623 __u32 limit; /* max SFB queue length */ 624 __u32 penalty_rate; /* inelastic flows are rate limited to 'rate' pps */ 625 __u32 penalty_burst; 626}; 627 628struct tc_sfb_xstats { 629 __u32 earlydrop; 630 __u32 penaltydrop; 631 __u32 bucketdrop; 632 __u32 queuedrop; 633 __u32 childdrop; /* drops in child qdisc */ 634 __u32 marked; 635 __u32 maxqlen; 636 __u32 maxprob; 637 __u32 avgprob; 638}; 639 640#define SFB_MAX_PROB 0xFFFF 641 642/* QFQ */ 643enum { 644 TCA_QFQ_UNSPEC, 645 TCA_QFQ_WEIGHT, 646 TCA_QFQ_LMAX, 647 __TCA_QFQ_MAX 648}; 649 650#define TCA_QFQ_MAX (__TCA_QFQ_MAX - 1) 651 652struct tc_qfq_stats { 653 __u32 weight; 654 __u32 lmax; 655}; 656 657#endif 658