release/src/linux/linux/include/linux/pkt_sched.h

   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
  30 struct 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 #ifdef __KERNEL__
  42         spinlock_t *lock;
  43 #endif
  44 };
  45
  46 struct tc_estimator
  47 {
  48         signed char     interval;
  49         unsigned char   ewma_log;
  50 };
  51
  52 /* "Handles"
  53    ---------
  54
  55     All the traffic control objects have 32bit identifiers, or "handles".
  56
  57     They can be considered as opaque numbers from user API viewpoint,
  58     but actually they always consist of two fields: major and
  59     minor numbers, which are interpreted by kernel specially,
  60     that may be used by applications, though not recommended.
  61
  62     F.e. qdisc handles always have minor number equal to zero,
  63     classes (or flows) have major equal to parent qdisc major, and
  64     minor uniquely identifying class inside qdisc.
  65
  66     Macros to manipulate handles:
  67  */
  68
  69 #define TC_H_MAJ_MASK (0xFFFF0000U)
  70 #define TC_H_MIN_MASK (0x0000FFFFU)
  71 #define TC_H_MAJ(h) ((h)&TC_H_MAJ_MASK)
  72 #define TC_H_MIN(h) ((h)&TC_H_MIN_MASK)
  73 #define TC_H_MAKE(maj,min) (((maj)&TC_H_MAJ_MASK)|((min)&TC_H_MIN_MASK))
  74
  75 #define TC_H_UNSPEC     (0U)
  76 #define TC_H_ROOT       (0xFFFFFFFFU)
  77 #define TC_H_INGRESS    (0xFFFFFFF1U)
  78
  79 struct tc_ratespec
  80 {
  81         unsigned char   cell_log;
  82         unsigned char   __reserved;
  83         unsigned short  feature;
  84         short           addend;
  85         unsigned short  mpu;
  86         __u32           rate;
  87 };
  88
  89 /* FIFO section */
  90
  91 struct tc_fifo_qopt
  92 {
  93         __u32   limit;  /* Queue length: bytes for bfifo, packets for pfifo */
  94 };
  95
  96 /* PRIO section */
  97
  98 #define TCQ_PRIO_BANDS  16
  99
 100 struct 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 /* CSZ section */
 107
 108 struct tc_csz_qopt
 109 {
 110         int             flows;          /* Maximal number of guaranteed flows */
 111         unsigned char   R_log;          /* Fixed point position for round number */
 112         unsigned char   delta_log;      /* Log of maximal managed time interval */
 113         __u8            priomap[TC_PRIO_MAX+1]; /* Map: logical priority -> CSZ band */
 114 };
 115
 116 struct tc_csz_copt
 117 {
 118         struct tc_ratespec slice;
 119         struct tc_ratespec rate;
 120         struct tc_ratespec peakrate;
 121         __u32           limit;
 122         __u32           buffer;
 123         __u32           mtu;
 124 };
 125
 126 enum
 127 {
 128         TCA_CSZ_UNSPEC,
 129         TCA_CSZ_PARMS,
 130         TCA_CSZ_RTAB,
 131         TCA_CSZ_PTAB,
 132 };
 133
 134 /* TBF section */
 135
 136 struct tc_tbf_qopt
 137 {
 138         struct tc_ratespec rate;
 139         struct tc_ratespec peakrate;
 140         __u32           limit;
 141         __u32           buffer;
 142         __u32           mtu;
 143 };
 144
 145 enum
 146 {
 147         TCA_TBF_UNSPEC,
 148         TCA_TBF_PARMS,
 149         TCA_TBF_RTAB,
 150         TCA_TBF_PTAB,
 151 };
 152
 153
 154 /* TEQL section */
 155
 156 /* TEQL does not require any parameters */
 157
 158 /* SFQ section */
 159
 160 struct tc_sfq_qopt
 161 {
 162         unsigned        quantum;        /* Bytes per round allocated to flow */
 163         int             perturb_period; /* Period of hash perturbation */
 164         __u32           limit;          /* Maximal packets in queue */
 165         unsigned        divisor;        /* Hash divisor  */
 166         unsigned        flows;          /* Maximal number of flows  */
 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  *      If you need to play with these values use esfq instead.
 178  */
 179
 180 /* ESFQ section */
 181
 182 enum
 183 {
 184         /* traditional */
 185         TCA_SFQ_HASH_CLASSIC,
 186         TCA_SFQ_HASH_DST,
 187         TCA_SFQ_HASH_SRC,
 188         TCA_SFQ_HASH_FWMARK,
 189         /* conntrack */
 190         TCA_SFQ_HASH_CTORIGDST,
 191         TCA_SFQ_HASH_CTORIGSRC,
 192         TCA_SFQ_HASH_CTREPLDST,
 193         TCA_SFQ_HASH_CTREPLSRC,
 194         TCA_SFQ_HASH_CTNATCHG,
 195 };
 196
 197 struct tc_esfq_qopt
 198 {
 199         unsigned        quantum;        /* Bytes per round allocated to flow */
 200         int             perturb_period; /* Period of hash perturbation */
 201         __u32           limit;          /* Maximal packets in queue */
 202         unsigned        divisor;        /* Hash divisor  */
 203         unsigned        flows;          /* Maximal number of flows  */
 204         unsigned        hash_kind;      /* Hash function to use for flow identification */
 205 };
 206
 207 /* RED section */
 208
 209 enum
 210 {
 211         TCA_RED_UNSPEC,
 212         TCA_RED_PARMS,
 213         TCA_RED_STAB,
 214 };
 215
 216 struct tc_red_qopt
 217 {
 218         __u32           limit;          /* HARD maximal queue length (bytes)    */
 219         __u32           qth_min;        /* Min average length threshold (bytes) */
 220         __u32           qth_max;        /* Max average length threshold (bytes) */
 221         unsigned char   Wlog;           /* log(W)               */
 222         unsigned char   Plog;           /* log(P_max/(qth_max-qth_min)) */
 223         unsigned char   Scell_log;      /* cell size for idle damping */
 224         unsigned char   flags;
 225 #define TC_RED_ECN      1
 226 };
 227
 228 struct tc_red_xstats
 229 {
 230         __u32           early;          /* Early drops */
 231         __u32           pdrop;          /* Drops due to queue limits */
 232         __u32           other;          /* Drops due to drop() calls */
 233         __u32           marked;         /* Marked packets */
 234 };
 235
 236 /* GRED section */
 237
 238 #define MAX_DPs 16
 239
 240 enum
 241 {
 242        TCA_GRED_UNSPEC,
 243        TCA_GRED_PARMS,
 244        TCA_GRED_STAB,
 245        TCA_GRED_DPS,
 246 };
 247
 248 #define TCA_SET_OFF TCA_GRED_PARMS
 249 struct tc_gred_qopt
 250 {
 251        __u32           limit;          /* HARD maximal queue length (bytes)
 252 */
 253        __u32           qth_min;        /* Min average length threshold (bytes)
 254 */
 255        __u32           qth_max;        /* Max average length threshold (bytes)
 256 */
 257        __u32           DP;             /* upto 2^32 DPs */
 258        __u32           backlog;
 259        __u32           qave;
 260        __u32           forced;
 261        __u32           early;
 262        __u32           other;
 263        __u32           pdrop;
 264
 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        __u8            prio;            /* prio of this VQ */
 269        __u32    packets;
 270        __u32    bytesin;
 271 };
 272 /* gred setup */
 273 struct tc_gred_sopt
 274 {
 275        __u32            DPs;
 276        __u32            def_DP;
 277        __u8             grio;
 278        __u8             pad1;
 279        __u16            pad2;
 280 };
 281
 282 /* HTB section */
 283 #define TC_HTB_NUMPRIO          11      /* Toastman */
 284 #define TC_HTB_MAXDEPTH         8
 285 #define TC_HTB_PROTOVER         3       /* the same as HTB and TC's major */
 286
 287 struct tc_htb_opt
 288 {
 289         struct tc_ratespec      rate;
 290         struct tc_ratespec      ceil;
 291         __u32   buffer;
 292         __u32   cbuffer;
 293         __u32   quantum;
 294         __u32   level;          /* out only */
 295         __u32   prio;
 296 };
 297 struct tc_htb_glob
 298 {
 299         __u32 version;          /* to match HTB/TC */
 300         __u32 rate2quantum;     /* bps->quantum divisor */
 301         __u32 defcls;           /* default class number */
 302         __u32 debug;            /* debug flags */
 303
 304         /* stats */
 305         __u32 direct_pkts; /* count of non shapped packets */
 306 };
 307 enum
 308 {
 309         TCA_HTB_UNSPEC,
 310         TCA_HTB_PARMS,
 311         TCA_HTB_INIT,
 312         TCA_HTB_CTAB,
 313         TCA_HTB_RTAB,
 314 };
 315 struct tc_htb_xstats
 316 {
 317         __u32 lends;
 318         __u32 borrows;
 319         __u32 giants;   /* too big packets (rate will not be accurate) */
 320         __u32 tokens;
 321         __u32 ctokens;
 322 };
 323
 324 /* HFSC section */
 325
 326 struct tc_hfsc_qopt
 327 {
 328         __u16   defcls;         /* default class */
 329 };
 330
 331 struct tc_service_curve
 332 {
 333         __u32   m1;             /* slope of the first segment in bps */
 334         __u32   d;              /* x-projection of the first segment in us */
 335         __u32   m2;             /* slope of the second segment in bps */
 336 };
 337
 338 struct tc_hfsc_stats
 339 {
 340         __u64   work;           /* total work done */
 341         __u64   rtwork;         /* work done by real-time criteria */
 342         __u32   period;         /* current period */
 343         __u32   level;          /* class level in hierarchy */
 344 };
 345
 346 enum
 347 {
 348         TCA_HFSC_UNSPEC,
 349         TCA_HFSC_RSC,
 350         TCA_HFSC_FSC,
 351         TCA_HFSC_USC,
 352         TCA_HFSC_MAX = TCA_HFSC_USC
 353 };
 354
 355 /* CBQ section */
 356
 357 #define TC_CBQ_MAXPRIO          11      /* Toastman */
 358 #define TC_CBQ_MAXLEVEL         8
 359 #define TC_CBQ_DEF_EWMA         5
 360
 361 struct tc_cbq_lssopt
 362 {
 363         unsigned char   change;
 364         unsigned char   flags;
 365 #define TCF_CBQ_LSS_BOUNDED     1
 366 #define TCF_CBQ_LSS_ISOLATED    2
 367         unsigned char   ewma_log;
 368         unsigned char   level;
 369 #define TCF_CBQ_LSS_FLAGS       1
 370 #define TCF_CBQ_LSS_EWMA        2
 371 #define TCF_CBQ_LSS_MAXIDLE     4
 372 #define TCF_CBQ_LSS_MINIDLE     8
 373 #define TCF_CBQ_LSS_OFFTIME     0x10
 374 #define TCF_CBQ_LSS_AVPKT       0x20
 375         __u32           maxidle;
 376         __u32           minidle;
 377         __u32           offtime;
 378         __u32           avpkt;
 379 };
 380
 381 struct tc_cbq_wrropt
 382 {
 383         unsigned char   flags;
 384         unsigned char   priority;
 385         unsigned char   cpriority;
 386         unsigned char   __reserved;
 387         __u32           allot;
 388         __u32           weight;
 389 };
 390
 391 struct tc_cbq_ovl
 392 {
 393         unsigned char   strategy;
 394 #define TC_CBQ_OVL_CLASSIC      0
 395 #define TC_CBQ_OVL_DELAY        1
 396 #define TC_CBQ_OVL_LOWPRIO      2
 397 #define TC_CBQ_OVL_DROP         3
 398 #define TC_CBQ_OVL_RCLASSIC     4
 399         unsigned char   priority2;
 400         __u16           pad;
 401         __u32           penalty;
 402 };
 403
 404 struct tc_cbq_police
 405 {
 406         unsigned char   police;
 407         unsigned char   __res1;
 408         unsigned short  __res2;
 409 };
 410
 411 struct tc_cbq_fopt
 412 {
 413         __u32           split;
 414         __u32           defmap;
 415         __u32           defchange;
 416 };
 417
 418 struct tc_cbq_xstats
 419 {
 420         __u32           borrows;
 421         __u32           overactions;
 422         __s32           avgidle;
 423         __s32           undertime;
 424 };
 425
 426 enum
 427 {
 428         TCA_CBQ_UNSPEC,
 429         TCA_CBQ_LSSOPT,
 430         TCA_CBQ_WRROPT,
 431         TCA_CBQ_FOPT,
 432         TCA_CBQ_OVL_STRATEGY,
 433         TCA_CBQ_RATE,
 434         TCA_CBQ_RTAB,
 435         TCA_CBQ_POLICE,
 436 };
 437
 438 #define TCA_CBQ_MAX     TCA_CBQ_POLICE
 439
 440 /* dsmark section */
 441
 442 enum {
 443         TCA_DSMARK_UNSPEC,
 444         TCA_DSMARK_INDICES,
 445         TCA_DSMARK_DEFAULT_INDEX,
 446         TCA_DSMARK_SET_TC_INDEX,
 447         TCA_DSMARK_MASK,
 448         TCA_DSMARK_VALUE
 449 };
 450
 451 #define TCA_DSMARK_MAX TCA_DSMARK_VALUE
 452
 453 /* ATM  section */
 454
 455 enum {
 456         TCA_ATM_UNSPEC,
 457         TCA_ATM_FD,             /* file/socket descriptor */
 458         TCA_ATM_PTR,            /* pointer to descriptor - later */
 459         TCA_ATM_HDR,            /* LL header */
 460         TCA_ATM_EXCESS,         /* excess traffic class (0 for CLP)  */
 461         TCA_ATM_ADDR,           /* PVC address (for output only) */
 462         TCA_ATM_STATE           /* VC state (ATM_VS_*; for output only) */
 463 };
 464
 465 #define TCA_ATM_MAX     TCA_ATM_STATE
 466
 467 /* Network emulator */
 468
 469 enum
 470 {
 471         TCA_NETEM_UNSPEC,
 472         TCA_NETEM_CORR,
 473         TCA_NETEM_DELAY_DIST,
 474 };
 475
 476 #define TCA_NETEM_MAX   TCA_NETEM_DELAY_DIST
 477
 478 struct tc_netem_qopt
 479 {
 480         __u32   latency;        /* added delay (us) */
 481         __u32   limit;          /* fifo limit (packets) */
 482         __u32   loss;           /* random packet loss (0=none ~0=100%) */
 483         __u32   gap;            /* re-ordering gap (0 for delay all) */
 484         __u32   duplicate;      /* random packet dup  (0=none ~0=100%) */
 485         __u32   jitter;         /* random jitter in latency (us) */
 486 };
 487
 488 struct tc_netem_corr
 489 {
 490         __u32   delay_corr;     /* delay correlation */
 491         __u32   loss_corr;      /* packet loss correlation */
 492         __u32   dup_corr;       /* duplicate correlation  */
 493 };
 494
 495 #define NETEM_DIST_SCALE        8192
 496
 497 #endif