net/sched/sch_gred.c

   1 /*
   2  * net/sched/sch_gred.c Generic Random Early Detection queue.
   3  *
   4  *
   5  *              This program is free software; you can redistribute it and/or
   6  *              modify it under the terms of the GNU General Public License
   7  *              as published by the Free Software Foundation; either version
   8  *              2 of the License, or (at your option) any later version.
   9  *
  10  * Authors:    J Hadi Salim (hadi@cyberus.ca) 1998-2002
  11  *
  12  *             991129: -  Bug fix with grio mode
  13  *                     - a better sing. AvgQ mode with Grio(WRED)
  14  *                     - A finer grained VQ dequeue based on sugestion
  15  *                       from Ren Liu
  16  *                     - More error checks
  17  *
  18  *  For all the glorious comments look at include/net/red.h
  19  */
  20
  21 #include <linux/config.h>
  22 #include <linux/module.h>
  23 #include <linux/types.h>
  24 #include <linux/kernel.h>
  25 #include <linux/netdevice.h>
  26 #include <linux/skbuff.h>
  27 #include <net/pkt_sched.h>
  28 #include <net/red.h>
  29
  30 #define GRED_DEF_PRIO (MAX_DPs / 2)
  31 #define GRED_VQ_MASK (MAX_DPs - 1)
  32
  33 struct gred_sched_data;
  34 struct gred_sched;
  35
  36 struct gred_sched_data
  37 {
  38         u32             limit;          /* HARD maximal queue length    */
  39         u32             DP;             /* the drop pramaters */
  40         u32             bytesin;        /* bytes seen on virtualQ so far*/
  41         u32             packetsin;      /* packets seen on virtualQ so far*/
  42         u32             backlog;        /* bytes on the virtualQ */
  43         u8              prio;           /* the prio of this vq */
  44
  45         struct red_parms parms;
  46         struct red_stats stats;
  47 };
  48
  49 enum {
  50         GRED_WRED_MODE = 1,
  51         GRED_RIO_MODE,
  52 };
  53
  54 struct gred_sched
  55 {
  56         struct gred_sched_data *tab[MAX_DPs];
  57         unsigned long   flags;
  58         u32             red_flags;
  59         u32             DPs;
  60         u32             def;
  61         struct red_parms wred_set;
  62 };
  63
  64 static inline int gred_wred_mode(struct gred_sched *table)
  65 {
  66         return test_bit(GRED_WRED_MODE, &table->flags);
  67 }
  68
  69 static inline void gred_enable_wred_mode(struct gred_sched *table)
  70 {
  71         __set_bit(GRED_WRED_MODE, &table->flags);
  72 }
  73
  74 static inline void gred_disable_wred_mode(struct gred_sched *table)
  75 {
  76         __clear_bit(GRED_WRED_MODE, &table->flags);
  77 }
  78
  79 static inline int gred_rio_mode(struct gred_sched *table)
  80 {
  81         return test_bit(GRED_RIO_MODE, &table->flags);
  82 }
  83
  84 static inline void gred_enable_rio_mode(struct gred_sched *table)
  85 {
  86         __set_bit(GRED_RIO_MODE, &table->flags);
  87 }
  88
  89 static inline void gred_disable_rio_mode(struct gred_sched *table)
  90 {
  91         __clear_bit(GRED_RIO_MODE, &table->flags);
  92 }
  93
  94 static inline int gred_wred_mode_check(struct Qdisc *sch)
  95 {
  96         struct gred_sched *table = qdisc_priv(sch);
  97         int i;
  98
  99         /* Really ugly O(n^2) but shouldn't be necessary too frequent. */
 100         for (i = 0; i < table->DPs; i++) {
 101                 struct gred_sched_data *q = table->tab[i];
 102                 int n;
 103
 104                 if (q == NULL)
 105                         continue;
 106
 107                 for (n = 0; n < table->DPs; n++)
 108                         if (table->tab[n] && table->tab[n] != q &&
 109                             table->tab[n]->prio == q->prio)
 110                                 return 1;
 111         }
 112
 113         return 0;
 114 }
 115
 116 static inline unsigned int gred_backlog(struct gred_sched *table,
 117                                         struct gred_sched_data *q,
 118                                         struct Qdisc *sch)
 119 {
 120         if (gred_wred_mode(table))
 121                 return sch->qstats.backlog;
 122         else
 123                 return q->backlog;
 124 }
 125
 126 static inline u16 tc_index_to_dp(struct sk_buff *skb)
 127 {
 128         return skb->tc_index & GRED_VQ_MASK;
 129 }
 130
 131 static inline void gred_load_wred_set(struct gred_sched *table,
 132                                       struct gred_sched_data *q)
 133 {
 134         q->parms.qavg = table->wred_set.qavg;
 135         q->parms.qidlestart = table->wred_set.qidlestart;
 136 }
 137
 138 static inline void gred_store_wred_set(struct gred_sched *table,
 139                                        struct gred_sched_data *q)
 140 {
 141         table->wred_set.qavg = q->parms.qavg;
 142 }
 143
 144 static inline int gred_use_ecn(struct gred_sched *t)
 145 {
 146         return t->red_flags & TC_RED_ECN;
 147 }
 148
 149 static inline int gred_use_harddrop(struct gred_sched *t)
 150 {
 151         return t->red_flags & TC_RED_HARDDROP;
 152 }
 153
 154 static int gred_enqueue(struct sk_buff *skb, struct Qdisc* sch)
 155 {
 156         struct gred_sched_data *q=NULL;
 157         struct gred_sched *t= qdisc_priv(sch);
 158         unsigned long qavg = 0;
 159         u16 dp = tc_index_to_dp(skb);
 160
 161         if (dp >= t->DPs  || (q = t->tab[dp]) == NULL) {
 162                 dp = t->def;
 163
 164                 if ((q = t->tab[dp]) == NULL) {
 165                         /* Pass through packets not assigned to a DP
 166                          * if no default DP has been configured. This
 167                          * allows for DP flows to be left untouched.
 168                          */
 169                         if (skb_queue_len(&sch->q) < sch->dev->tx_queue_len)
 170                                 return qdisc_enqueue_tail(skb, sch);
 171                         else
 172                                 goto drop;
 173                 }
 174
 175                 /* fix tc_index? --could be controvesial but needed for
 176                    requeueing */
 177                 skb->tc_index = (skb->tc_index & ~GRED_VQ_MASK) | dp;
 178         }
 179
 180         /* sum up all the qaves of prios <= to ours to get the new qave */
 181         if (!gred_wred_mode(t) && gred_rio_mode(t)) {
 182                 int i;
 183
 184                 for (i = 0; i < t->DPs; i++) {
 185                         if (t->tab[i] && t->tab[i]->prio < q->prio &&
 186                             !red_is_idling(&t->tab[i]->parms))
 187                                 qavg +=t->tab[i]->parms.qavg;
 188                 }
 189
 190         }
 191
 192         q->packetsin++;
 193         q->bytesin += skb->len;
 194
 195         if (gred_wred_mode(t))
 196                 gred_load_wred_set(t, q);
 197
 198         q->parms.qavg = red_calc_qavg(&q->parms, gred_backlog(t, q, sch));
 199
 200         if (red_is_idling(&q->parms))
 201                 red_end_of_idle_period(&q->parms);
 202
 203         if (gred_wred_mode(t))
 204                 gred_store_wred_set(t, q);
 205
 206         switch (red_action(&q->parms, q->parms.qavg + qavg)) {
 207                 case RED_DONT_MARK:
 208                         break;
 209
 210                 case RED_PROB_MARK:
 211                         sch->qstats.overlimits++;
 212                         if (!gred_use_ecn(t) || !INET_ECN_set_ce(skb)) {
 213                                 q->stats.prob_drop++;
 214                                 goto congestion_drop;
 215                         }
 216
 217                         q->stats.prob_mark++;
 218                         break;
 219
 220                 case RED_HARD_MARK:
 221                         sch->qstats.overlimits++;
 222                         if (gred_use_harddrop(t) || !gred_use_ecn(t) ||
 223                             !INET_ECN_set_ce(skb)) {
 224                                 q->stats.forced_drop++;
 225                                 goto congestion_drop;
 226                         }
 227                         q->stats.forced_mark++;
 228                         break;
 229         }
 230
 231         if (q->backlog + skb->len <= q->limit) {
 232                 q->backlog += skb->len;
 233                 return qdisc_enqueue_tail(skb, sch);
 234         }
 235
 236         q->stats.pdrop++;
 237 drop:
 238         return qdisc_drop(skb, sch);
 239
 240 congestion_drop:
 241         qdisc_drop(skb, sch);
 242         return NET_XMIT_CN;
 243 }
 244
 245 static int gred_requeue(struct sk_buff *skb, struct Qdisc* sch)
 246 {
 247         struct gred_sched *t = qdisc_priv(sch);
 248         struct gred_sched_data *q;
 249         u16 dp = tc_index_to_dp(skb);
 250
 251         if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
 252                 if (net_ratelimit())
 253                         printk(KERN_WARNING "GRED: Unable to relocate VQ 0x%x "
 254                                "for requeue, screwing up backlog.\n",
 255                                tc_index_to_dp(skb));
 256         } else {
 257                 if (red_is_idling(&q->parms))
 258                         red_end_of_idle_period(&q->parms);
 259                 q->backlog += skb->len;
 260         }
 261
 262         return qdisc_requeue(skb, sch);
 263 }
 264
 265 static struct sk_buff *gred_dequeue(struct Qdisc* sch)
 266 {
 267         struct sk_buff *skb;
 268         struct gred_sched *t = qdisc_priv(sch);
 269
 270         skb = qdisc_dequeue_head(sch);
 271
 272         if (skb) {
 273                 struct gred_sched_data *q;
 274                 u16 dp = tc_index_to_dp(skb);
 275
 276                 if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
 277                         if (net_ratelimit())
 278                                 printk(KERN_WARNING "GRED: Unable to relocate "
 279                                        "VQ 0x%x after dequeue, screwing up "
 280                                        "backlog.\n", tc_index_to_dp(skb));
 281                 } else {
 282                         q->backlog -= skb->len;
 283
 284                         if (!q->backlog && !gred_wred_mode(t))
 285                                 red_start_of_idle_period(&q->parms);
 286                 }
 287
 288                 return skb;
 289         }
 290
 291         if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
 292                 red_start_of_idle_period(&t->wred_set);
 293
 294         return NULL;
 295 }
 296
 297 static unsigned int gred_drop(struct Qdisc* sch)
 298 {
 299         struct sk_buff *skb;
 300         struct gred_sched *t = qdisc_priv(sch);
 301
 302         skb = qdisc_dequeue_tail(sch);
 303         if (skb) {
 304                 unsigned int len = skb->len;
 305                 struct gred_sched_data *q;
 306                 u16 dp = tc_index_to_dp(skb);
 307
 308                 if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
 309                         if (net_ratelimit())
 310                                 printk(KERN_WARNING "GRED: Unable to relocate "
 311                                        "VQ 0x%x while dropping, screwing up "
 312                                        "backlog.\n", tc_index_to_dp(skb));
 313                 } else {
 314                         q->backlog -= len;
 315                         q->stats.other++;
 316
 317                         if (!q->backlog && !gred_wred_mode(t))
 318                                 red_start_of_idle_period(&q->parms);
 319                 }
 320
 321                 qdisc_drop(skb, sch);
 322                 return len;
 323         }
 324
 325         if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
 326                 red_start_of_idle_period(&t->wred_set);
 327
 328         return 0;
 329
 330 }
 331
 332 static void gred_reset(struct Qdisc* sch)
 333 {
 334         int i;
 335         struct gred_sched *t = qdisc_priv(sch);
 336
 337         qdisc_reset_queue(sch);
 338
 339         for (i = 0; i < t->DPs; i++) {
 340                 struct gred_sched_data *q = t->tab[i];
 341
 342                 if (!q)
 343                         continue;
 344
 345                 red_restart(&q->parms);
 346                 q->backlog = 0;
 347         }
 348 }
 349
 350 static inline void gred_destroy_vq(struct gred_sched_data *q)
 351 {
 352         kfree(q);
 353 }
 354
 355 static inline int gred_change_table_def(struct Qdisc *sch, struct rtattr *dps)
 356 {
 357         struct gred_sched *table = qdisc_priv(sch);
 358         struct tc_gred_sopt *sopt;
 359         int i;
 360
 361         if (dps == NULL || RTA_PAYLOAD(dps) < sizeof(*sopt))
 362                 return -EINVAL;
 363
 364         sopt = RTA_DATA(dps);
 365
 366         if (sopt->DPs > MAX_DPs || sopt->DPs == 0 || sopt->def_DP >= sopt->DPs)
 367                 return -EINVAL;
 368
 369         sch_tree_lock(sch);
 370         table->DPs = sopt->DPs;
 371         table->def = sopt->def_DP;
 372         table->red_flags = sopt->flags;
 373
 374         /*
 375          * Every entry point to GRED is synchronized with the above code
 376          * and the DP is checked against DPs, i.e. shadowed VQs can no
 377          * longer be found so we can unlock right here.
 378          */
 379         sch_tree_unlock(sch);
 380
 381         if (sopt->grio) {
 382                 gred_enable_rio_mode(table);
 383                 gred_disable_wred_mode(table);
 384                 if (gred_wred_mode_check(sch))
 385                         gred_enable_wred_mode(table);
 386         } else {
 387                 gred_disable_rio_mode(table);
 388                 gred_disable_wred_mode(table);
 389         }
 390
 391         for (i = table->DPs; i < MAX_DPs; i++) {
 392                 if (table->tab[i]) {
 393                         printk(KERN_WARNING "GRED: Warning: Destroying "
 394                                "shadowed VQ 0x%x\n", i);
 395                         gred_destroy_vq(table->tab[i]);
 396                         table->tab[i] = NULL;
 397                 }
 398         }
 399
 400         return 0;
 401 }
 402
 403 static inline int gred_change_vq(struct Qdisc *sch, int dp,
 404                                  struct tc_gred_qopt *ctl, int prio, u8 *stab)
 405 {
 406         struct gred_sched *table = qdisc_priv(sch);
 407         struct gred_sched_data *q;
 408
 409         if (table->tab[dp] == NULL) {
 410                 table->tab[dp] = kmalloc(sizeof(*q), GFP_KERNEL);
 411                 if (table->tab[dp] == NULL)
 412                         return -ENOMEM;
 413                 memset(table->tab[dp], 0, sizeof(*q));
 414         }
 415
 416         q = table->tab[dp];
 417         q->DP = dp;
 418         q->prio = prio;
 419         q->limit = ctl->limit;
 420
 421         if (q->backlog == 0)
 422                 red_end_of_idle_period(&q->parms);
 423
 424         red_set_parms(&q->parms,
 425                       ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog,
 426                       ctl->Scell_log, stab);
 427
 428         return 0;
 429 }
 430
 431 static int gred_change(struct Qdisc *sch, struct rtattr *opt)
 432 {
 433         struct gred_sched *table = qdisc_priv(sch);
 434         struct tc_gred_qopt *ctl;
 435         struct rtattr *tb[TCA_GRED_MAX];
 436         int err = -EINVAL, prio = GRED_DEF_PRIO;
 437         u8 *stab;
 438
 439         if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
 440                 return -EINVAL;
 441
 442         if (tb[TCA_GRED_PARMS-1] == NULL && tb[TCA_GRED_STAB-1] == NULL)
 443                 return gred_change_table_def(sch, opt);
 444
 445         if (tb[TCA_GRED_PARMS-1] == NULL ||
 446             RTA_PAYLOAD(tb[TCA_GRED_PARMS-1]) < sizeof(*ctl) ||
 447             tb[TCA_GRED_STAB-1] == NULL ||
 448             RTA_PAYLOAD(tb[TCA_GRED_STAB-1]) < 256)
 449                 return -EINVAL;
 450
 451         ctl = RTA_DATA(tb[TCA_GRED_PARMS-1]);
 452         stab = RTA_DATA(tb[TCA_GRED_STAB-1]);
 453
 454         if (ctl->DP >= table->DPs)
 455                 goto errout;
 456
 457         if (gred_rio_mode(table)) {
 458                 if (ctl->prio == 0) {
 459                         int def_prio = GRED_DEF_PRIO;
 460
 461                         if (table->tab[table->def])
 462                                 def_prio = table->tab[table->def]->prio;
 463
 464                         printk(KERN_DEBUG "GRED: DP %u does not have a prio "
 465                                "setting default to %d\n", ctl->DP, def_prio);
 466
 467                         prio = def_prio;
 468                 } else
 469                         prio = ctl->prio;
 470         }
 471
 472         sch_tree_lock(sch);
 473
 474         err = gred_change_vq(sch, ctl->DP, ctl, prio, stab);
 475         if (err < 0)
 476                 goto errout_locked;
 477
 478         if (gred_rio_mode(table)) {
 479                 gred_disable_wred_mode(table);
 480                 if (gred_wred_mode_check(sch))
 481                         gred_enable_wred_mode(table);
 482         }
 483
 484         err = 0;
 485
 486 errout_locked:
 487         sch_tree_unlock(sch);
 488 errout:
 489         return err;
 490 }
 491
 492 static int gred_init(struct Qdisc *sch, struct rtattr *opt)
 493 {
 494         struct rtattr *tb[TCA_GRED_MAX];
 495
 496         if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
 497                 return -EINVAL;
 498
 499         if (tb[TCA_GRED_PARMS-1] || tb[TCA_GRED_STAB-1])
 500                 return -EINVAL;
 501
 502         return gred_change_table_def(sch, tb[TCA_GRED_DPS-1]);
 503 }
 504
 505 static int gred_dump(struct Qdisc *sch, struct sk_buff *skb)
 506 {
 507         struct gred_sched *table = qdisc_priv(sch);
 508         struct rtattr *parms, *opts = NULL;
 509         int i;
 510         struct tc_gred_sopt sopt = {
 511                 .DPs    = table->DPs,
 512                 .def_DP = table->def,
 513                 .grio   = gred_rio_mode(table),
 514                 .flags  = table->red_flags,
 515         };
 516
 517         opts = RTA_NEST(skb, TCA_OPTIONS);
 518         RTA_PUT(skb, TCA_GRED_DPS, sizeof(sopt), &sopt);
 519         parms = RTA_NEST(skb, TCA_GRED_PARMS);
 520
 521         for (i = 0; i < MAX_DPs; i++) {
 522                 struct gred_sched_data *q = table->tab[i];
 523                 struct tc_gred_qopt opt;
 524
 525                 memset(&opt, 0, sizeof(opt));
 526
 527                 if (!q) {
 528                         /* hack -- fix at some point with proper message
 529                            This is how we indicate to tc that there is no VQ
 530                            at this DP */
 531
 532                         opt.DP = MAX_DPs + i;
 533                         goto append_opt;
 534                 }
 535
 536                 opt.limit       = q->limit;
 537                 opt.DP          = q->DP;
 538                 opt.backlog     = q->backlog;
 539                 opt.prio        = q->prio;
 540                 opt.qth_min     = q->parms.qth_min >> q->parms.Wlog;
 541                 opt.qth_max     = q->parms.qth_max >> q->parms.Wlog;
 542                 opt.Wlog        = q->parms.Wlog;
 543                 opt.Plog        = q->parms.Plog;
 544                 opt.Scell_log   = q->parms.Scell_log;
 545                 opt.other       = q->stats.other;
 546                 opt.early       = q->stats.prob_drop;
 547                 opt.forced      = q->stats.forced_drop;
 548                 opt.pdrop       = q->stats.pdrop;
 549                 opt.packets     = q->packetsin;
 550                 opt.bytesin     = q->bytesin;
 551
 552                 if (gred_wred_mode(table)) {
 553                         q->parms.qidlestart =
 554                                 table->tab[table->def]->parms.qidlestart;
 555                         q->parms.qavg = table->tab[table->def]->parms.qavg;
 556                 }
 557
 558                 opt.qave = red_calc_qavg(&q->parms, q->parms.qavg);
 559
 560 append_opt:
 561                 RTA_APPEND(skb, sizeof(opt), &opt);
 562         }
 563
 564         RTA_NEST_END(skb, parms);
 565
 566         return RTA_NEST_END(skb, opts);
 567
 568 rtattr_failure:
 569         return RTA_NEST_CANCEL(skb, opts);
 570 }
 571
 572 static void gred_destroy(struct Qdisc *sch)
 573 {
 574         struct gred_sched *table = qdisc_priv(sch);
 575         int i;
 576
 577         for (i = 0; i < table->DPs; i++) {
 578                 if (table->tab[i])
 579                         gred_destroy_vq(table->tab[i]);
 580         }
 581 }
 582
 583 static struct Qdisc_ops gred_qdisc_ops = {
 584         .id             =       "gred",
 585         .priv_size      =       sizeof(struct gred_sched),
 586         .enqueue        =       gred_enqueue,
 587         .dequeue        =       gred_dequeue,
 588         .requeue        =       gred_requeue,
 589         .drop           =       gred_drop,
 590         .init           =       gred_init,
 591         .reset          =       gred_reset,
 592         .destroy        =       gred_destroy,
 593         .change         =       gred_change,
 594         .dump           =       gred_dump,
 595         .owner          =       THIS_MODULE,
 596 };
 597
 598 static int __init gred_module_init(void)
 599 {
 600         return register_qdisc(&gred_qdisc_ops);
 601 }
 602
 603 static void __exit gred_module_exit(void)
 604 {
 605         unregister_qdisc(&gred_qdisc_ops);
 606 }
 607
 608 module_init(gred_module_init)
 609 module_exit(gred_module_exit)
 610
 611 MODULE_LICENSE("GPL");