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  *
  19  *
  20  *  For all the glorious comments look at Alexey's sch_red.c
  21  */
  22
  23 #include <linux/config.h>
  24 #include <linux/module.h>
  25 #include <asm/uaccess.h>
  26 #include <asm/system.h>
  27 #include <linux/bitops.h>
  28 #include <linux/types.h>
  29 #include <linux/kernel.h>
  30 #include <linux/sched.h>
  31 #include <linux/string.h>
  32 #include <linux/mm.h>
  33 #include <linux/socket.h>
  34 #include <linux/sockios.h>
  35 #include <linux/in.h>
  36 #include <linux/errno.h>
  37 #include <linux/interrupt.h>
  38 #include <linux/if_ether.h>
  39 #include <linux/inet.h>
  40 #include <linux/netdevice.h>
  41 #include <linux/etherdevice.h>
  42 #include <linux/notifier.h>
  43 #include <net/ip.h>
  44 #include <net/route.h>
  45 #include <linux/skbuff.h>
  46 #include <net/sock.h>
  47 #include <net/pkt_sched.h>
  48 #include <net/red.h>
  49
  50 #if 1 /* control */
  51 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
  52 #else
  53 #define DPRINTK(format,args...)
  54 #endif
  55
  56 #if 0 /* data */
  57 #define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
  58 #else
  59 #define D2PRINTK(format,args...)
  60 #endif
  61
  62 #define GRED_DEF_PRIO (MAX_DPs / 2)
  63 #define GRED_VQ_MASK (MAX_DPs - 1)
  64
  65 struct gred_sched_data;
  66 struct gred_sched;
  67
  68 struct gred_sched_data
  69 {
  70         u32             limit;          /* HARD maximal queue length    */
  71         u32             DP;             /* the drop pramaters */
  72         u32             bytesin;        /* bytes seen on virtualQ so far*/
  73         u32             packetsin;      /* packets seen on virtualQ so far*/
  74         u32             backlog;        /* bytes on the virtualQ */
  75         u8              prio;        /* the prio of this vq */
  76
  77         struct red_parms parms;
  78         struct red_stats stats;
  79 };
  80
  81 enum {
  82         GRED_WRED_MODE = 1,
  83         GRED_RIO_MODE,
  84 };
  85
  86 struct gred_sched
  87 {
  88         struct gred_sched_data *tab[MAX_DPs];
  89         unsigned long   flags;
  90         u32             DPs;
  91         u32             def;
  92         struct red_parms wred_set;
  93 };
  94
  95 static inline int gred_wred_mode(struct gred_sched *table)
  96 {
  97         return test_bit(GRED_WRED_MODE, &table->flags);
  98 }
  99
 100 static inline void gred_enable_wred_mode(struct gred_sched *table)
 101 {
 102         __set_bit(GRED_WRED_MODE, &table->flags);
 103 }
 104
 105 static inline void gred_disable_wred_mode(struct gred_sched *table)
 106 {
 107         __clear_bit(GRED_WRED_MODE, &table->flags);
 108 }
 109
 110 static inline int gred_rio_mode(struct gred_sched *table)
 111 {
 112         return test_bit(GRED_RIO_MODE, &table->flags);
 113 }
 114
 115 static inline void gred_enable_rio_mode(struct gred_sched *table)
 116 {
 117         __set_bit(GRED_RIO_MODE, &table->flags);
 118 }
 119
 120 static inline void gred_disable_rio_mode(struct gred_sched *table)
 121 {
 122         __clear_bit(GRED_RIO_MODE, &table->flags);
 123 }
 124
 125 static inline int gred_wred_mode_check(struct Qdisc *sch)
 126 {
 127         struct gred_sched *table = qdisc_priv(sch);
 128         int i;
 129
 130         /* Really ugly O(n^2) but shouldn't be necessary too frequent. */
 131         for (i = 0; i < table->DPs; i++) {
 132                 struct gred_sched_data *q = table->tab[i];
 133                 int n;
 134
 135                 if (q == NULL)
 136                         continue;
 137
 138                 for (n = 0; n < table->DPs; n++)
 139                         if (table->tab[n] && table->tab[n] != q &&
 140                             table->tab[n]->prio == q->prio)
 141                                 return 1;
 142         }
 143
 144         return 0;
 145 }
 146
 147 static inline unsigned int gred_backlog(struct gred_sched *table,
 148                                         struct gred_sched_data *q,
 149                                         struct Qdisc *sch)
 150 {
 151         if (gred_wred_mode(table))
 152                 return sch->qstats.backlog;
 153         else
 154                 return q->backlog;
 155 }
 156
 157 static inline u16 tc_index_to_dp(struct sk_buff *skb)
 158 {
 159         return skb->tc_index & GRED_VQ_MASK;
 160 }
 161
 162 static inline void gred_load_wred_set(struct gred_sched *table,
 163                                       struct gred_sched_data *q)
 164 {
 165         q->parms.qavg = table->wred_set.qavg;
 166         q->parms.qidlestart = table->wred_set.qidlestart;
 167 }
 168
 169 static inline void gred_store_wred_set(struct gred_sched *table,
 170                                        struct gred_sched_data *q)
 171 {
 172         table->wred_set.qavg = q->parms.qavg;
 173 }
 174
 175 static int
 176 gred_enqueue(struct sk_buff *skb, struct Qdisc* sch)
 177 {
 178         struct gred_sched_data *q=NULL;
 179         struct gred_sched *t= qdisc_priv(sch);
 180         unsigned long qavg = 0;
 181         int i=0;
 182         u16 dp = tc_index_to_dp(skb);
 183
 184         if (dp >= t->DPs  || (q = t->tab[dp]) == NULL) {
 185                 dp = t->def;
 186
 187                 if ((q = t->tab[dp]) == NULL) {
 188                         /* Pass through packets not assigned to a DP
 189                          * if no default DP has been configured. This
 190                          * allows for DP flows to be left untouched.
 191                          */
 192                         if (skb_queue_len(&sch->q) < sch->dev->tx_queue_len)
 193                                 return qdisc_enqueue_tail(skb, sch);
 194                         else
 195                                 goto drop;
 196                 }
 197
 198                 /* fix tc_index? --could be controvesial but needed for
 199                    requeueing */
 200                 skb->tc_index = (skb->tc_index & ~GRED_VQ_MASK) | dp;
 201         }
 202
 203         /* sum up all the qaves of prios <= to ours to get the new qave*/
 204         if (!gred_wred_mode(t) && gred_rio_mode(t)) {
 205                 for (i=0;i<t->DPs;i++) {
 206                         if ((!t->tab[i]) || (i==q->DP))
 207                                 continue;
 208
 209                         if (t->tab[i]->prio < q->prio &&
 210                             !red_is_idling(&t->tab[i]->parms))
 211                                 qavg +=t->tab[i]->parms.qavg;
 212                 }
 213
 214         }
 215
 216         q->packetsin++;
 217         q->bytesin+=skb->len;
 218
 219         if (gred_wred_mode(t)) {
 220                 qavg = 0;
 221                 gred_load_wred_set(t, q);
 222         }
 223
 224         q->parms.qavg = red_calc_qavg(&q->parms, gred_backlog(t, q, sch));
 225
 226         if (red_is_idling(&q->parms))
 227                 red_end_of_idle_period(&q->parms);
 228
 229         if (gred_wred_mode(t))
 230                 gred_store_wred_set(t, q);
 231
 232         switch (red_action(&q->parms, q->parms.qavg + qavg)) {
 233                 case RED_DONT_MARK:
 234                         break;
 235
 236                 case RED_PROB_MARK:
 237                         sch->qstats.overlimits++;
 238                         q->stats.prob_drop++;
 239                         goto congestion_drop;
 240
 241                 case RED_HARD_MARK:
 242                         sch->qstats.overlimits++;
 243                         q->stats.forced_drop++;
 244                         goto congestion_drop;
 245         }
 246
 247         if (q->backlog + skb->len <= q->limit) {
 248                 q->backlog += skb->len;
 249                 return qdisc_enqueue_tail(skb, sch);
 250         }
 251
 252         q->stats.pdrop++;
 253 drop:
 254         return qdisc_drop(skb, sch);
 255
 256 congestion_drop:
 257         qdisc_drop(skb, sch);
 258         return NET_XMIT_CN;
 259 }
 260
 261 static int
 262 gred_requeue(struct sk_buff *skb, struct Qdisc* sch)
 263 {
 264         struct gred_sched *t = qdisc_priv(sch);
 265         struct gred_sched_data *q;
 266         u16 dp = tc_index_to_dp(skb);
 267
 268         if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
 269                 if (net_ratelimit())
 270                         printk(KERN_WARNING "GRED: Unable to relocate VQ 0x%x "
 271                                "for requeue, screwing up backlog.\n",
 272                                tc_index_to_dp(skb));
 273         } else {
 274                 if (red_is_idling(&q->parms))
 275                         red_end_of_idle_period(&q->parms);
 276                 q->backlog += skb->len;
 277         }
 278
 279         return qdisc_requeue(skb, sch);
 280 }
 281
 282 static struct sk_buff *
 283 gred_dequeue(struct Qdisc* sch)
 284 {
 285         struct sk_buff *skb;
 286         struct gred_sched_data *q;
 287         struct gred_sched *t= qdisc_priv(sch);
 288
 289         skb = qdisc_dequeue_head(sch);
 290
 291         if (skb) {
 292                 u16 dp = tc_index_to_dp(skb);
 293
 294                 if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
 295                         if (net_ratelimit())
 296                                 printk(KERN_WARNING "GRED: Unable to relocate "
 297                                        "VQ 0x%x after dequeue, screwing up "
 298                                        "backlog.\n", tc_index_to_dp(skb));
 299                 } else {
 300                         q->backlog -= skb->len;
 301
 302                         if (!q->backlog && !gred_wred_mode(t))
 303                                 red_start_of_idle_period(&q->parms);
 304                 }
 305
 306                 return skb;
 307         }
 308
 309         if (gred_wred_mode(t))
 310                 red_start_of_idle_period(&t->wred_set);
 311
 312         return NULL;
 313 }
 314
 315 static unsigned int gred_drop(struct Qdisc* sch)
 316 {
 317         struct sk_buff *skb;
 318
 319         struct gred_sched_data *q;
 320         struct gred_sched *t= qdisc_priv(sch);
 321
 322         skb = qdisc_dequeue_tail(sch);
 323         if (skb) {
 324                 unsigned int len = skb->len;
 325                 u16 dp = tc_index_to_dp(skb);
 326
 327                 if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
 328                         if (net_ratelimit())
 329                                 printk(KERN_WARNING "GRED: Unable to relocate "
 330                                        "VQ 0x%x while dropping, screwing up "
 331                                        "backlog.\n", tc_index_to_dp(skb));
 332                 } else {
 333                         q->backlog -= len;
 334                         q->stats.other++;
 335
 336                         if (!q->backlog && !gred_wred_mode(t))
 337                                 red_start_of_idle_period(&q->parms);
 338                 }
 339
 340                 qdisc_drop(skb, sch);
 341                 return len;
 342         }
 343
 344         if (gred_wred_mode(t))
 345                 red_start_of_idle_period(&t->wred_set);
 346
 347         return 0;
 348
 349 }
 350
 351 static void gred_reset(struct Qdisc* sch)
 352 {
 353         int i;
 354         struct gred_sched_data *q;
 355         struct gred_sched *t= qdisc_priv(sch);
 356
 357         qdisc_reset_queue(sch);
 358
 359         for (i=0;i<t->DPs;i++) {
 360                 q= t->tab[i];
 361                 if (!q)
 362                         continue;
 363                 red_restart(&q->parms);
 364                 q->backlog = 0;
 365         }
 366 }
 367
 368 static inline void gred_destroy_vq(struct gred_sched_data *q)
 369 {
 370         kfree(q);
 371 }
 372
 373 static inline int gred_change_table_def(struct Qdisc *sch, struct rtattr *dps)
 374 {
 375         struct gred_sched *table = qdisc_priv(sch);
 376         struct tc_gred_sopt *sopt;
 377         int i;
 378
 379         if (dps == NULL || RTA_PAYLOAD(dps) < sizeof(*sopt))
 380                 return -EINVAL;
 381
 382         sopt = RTA_DATA(dps);
 383
 384         if (sopt->DPs > MAX_DPs || sopt->DPs == 0 || sopt->def_DP >= sopt->DPs)
 385                 return -EINVAL;
 386
 387         sch_tree_lock(sch);
 388         table->DPs = sopt->DPs;
 389         table->def = sopt->def_DP;
 390
 391         /*
 392          * Every entry point to GRED is synchronized with the above code
 393          * and the DP is checked against DPs, i.e. shadowed VQs can no
 394          * longer be found so we can unlock right here.
 395          */
 396         sch_tree_unlock(sch);
 397
 398         if (sopt->grio) {
 399                 gred_enable_rio_mode(table);
 400                 gred_disable_wred_mode(table);
 401                 if (gred_wred_mode_check(sch))
 402                         gred_enable_wred_mode(table);
 403         } else {
 404                 gred_disable_rio_mode(table);
 405                 gred_disable_wred_mode(table);
 406         }
 407
 408         for (i = table->DPs; i < MAX_DPs; i++) {
 409                 if (table->tab[i]) {
 410                         printk(KERN_WARNING "GRED: Warning: Destroying "
 411                                "shadowed VQ 0x%x\n", i);
 412                         gred_destroy_vq(table->tab[i]);
 413                         table->tab[i] = NULL;
 414                 }
 415         }
 416
 417         return 0;
 418 }
 419
 420 static inline int gred_change_vq(struct Qdisc *sch, int dp,
 421                                  struct tc_gred_qopt *ctl, int prio, u8 *stab)
 422 {
 423         struct gred_sched *table = qdisc_priv(sch);
 424         struct gred_sched_data *q;
 425
 426         if (table->tab[dp] == NULL) {
 427                 table->tab[dp] = kmalloc(sizeof(*q), GFP_KERNEL);
 428                 if (table->tab[dp] == NULL)
 429                         return -ENOMEM;
 430                 memset(table->tab[dp], 0, sizeof(*q));
 431         }
 432
 433         q = table->tab[dp];
 434         q->DP = dp;
 435         q->prio = prio;
 436         q->limit = ctl->limit;
 437
 438         if (q->backlog == 0)
 439                 red_end_of_idle_period(&q->parms);
 440
 441         red_set_parms(&q->parms,
 442                       ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog,
 443                       ctl->Scell_log, stab);
 444
 445         return 0;
 446 }
 447
 448 static int gred_change(struct Qdisc *sch, struct rtattr *opt)
 449 {
 450         struct gred_sched *table = qdisc_priv(sch);
 451         struct tc_gred_qopt *ctl;
 452         struct rtattr *tb[TCA_GRED_MAX];
 453         int err = -EINVAL, prio = GRED_DEF_PRIO;
 454         u8 *stab;
 455
 456         if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
 457                 return -EINVAL;
 458
 459         if (tb[TCA_GRED_PARMS-1] == NULL && tb[TCA_GRED_STAB-1] == NULL)
 460                 return gred_change_table_def(sch, opt);
 461
 462         if (tb[TCA_GRED_PARMS-1] == NULL ||
 463             RTA_PAYLOAD(tb[TCA_GRED_PARMS-1]) < sizeof(*ctl) ||
 464             tb[TCA_GRED_STAB-1] == NULL ||
 465             RTA_PAYLOAD(tb[TCA_GRED_STAB-1]) < 256)
 466                 return -EINVAL;
 467
 468         ctl = RTA_DATA(tb[TCA_GRED_PARMS-1]);
 469         stab = RTA_DATA(tb[TCA_GRED_STAB-1]);
 470
 471         if (ctl->DP >= table->DPs)
 472                 goto errout;
 473
 474         if (gred_rio_mode(table)) {
 475                 if (ctl->prio == 0) {
 476                         int def_prio = GRED_DEF_PRIO;
 477
 478                         if (table->tab[table->def])
 479                                 def_prio = table->tab[table->def]->prio;
 480
 481                         printk(KERN_DEBUG "GRED: DP %u does not have a prio "
 482                                "setting default to %d\n", ctl->DP, def_prio);
 483
 484                         prio = def_prio;
 485                 } else
 486                         prio = ctl->prio;
 487         }
 488
 489         sch_tree_lock(sch);
 490
 491         err = gred_change_vq(sch, ctl->DP, ctl, prio, stab);
 492         if (err < 0)
 493                 goto errout_locked;
 494
 495         if (table->tab[table->def] == NULL) {
 496                 if (gred_rio_mode(table))
 497                         prio = table->tab[ctl->DP]->prio;
 498
 499                 err = gred_change_vq(sch, table->def, ctl, prio, stab);
 500                 if (err < 0)
 501                         goto errout_locked;
 502         }
 503
 504         if (gred_rio_mode(table)) {
 505                 gred_disable_wred_mode(table);
 506                 if (gred_wred_mode_check(sch))
 507                         gred_enable_wred_mode(table);
 508         }
 509
 510         err = 0;
 511
 512 errout_locked:
 513         sch_tree_unlock(sch);
 514 errout:
 515         return err;
 516 }
 517
 518 static int gred_init(struct Qdisc *sch, struct rtattr *opt)
 519 {
 520         struct rtattr *tb[TCA_GRED_MAX];
 521
 522         if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
 523                 return -EINVAL;
 524
 525         if (tb[TCA_GRED_PARMS-1] || tb[TCA_GRED_STAB-1])
 526                 return -EINVAL;
 527
 528         return gred_change_table_def(sch, tb[TCA_GRED_DPS-1]);
 529 }
 530
 531 static int gred_dump(struct Qdisc *sch, struct sk_buff *skb)
 532 {
 533         struct gred_sched *table = qdisc_priv(sch);
 534         struct rtattr *parms, *opts = NULL;
 535         int i;
 536         struct tc_gred_sopt sopt = {
 537                 .DPs    = table->DPs,
 538                 .def_DP = table->def,
 539                 .grio   = gred_rio_mode(table),
 540         };
 541
 542         opts = RTA_NEST(skb, TCA_OPTIONS);
 543         RTA_PUT(skb, TCA_GRED_DPS, sizeof(sopt), &sopt);
 544         parms = RTA_NEST(skb, TCA_GRED_PARMS);
 545
 546         for (i = 0; i < MAX_DPs; i++) {
 547                 struct gred_sched_data *q = table->tab[i];
 548                 struct tc_gred_qopt opt;
 549
 550                 memset(&opt, 0, sizeof(opt));
 551
 552                 if (!q) {
 553                         /* hack -- fix at some point with proper message
 554                            This is how we indicate to tc that there is no VQ
 555                            at this DP */
 556
 557                         opt.DP = MAX_DPs + i;
 558                         goto append_opt;
 559                 }
 560
 561                 opt.limit       = q->limit;
 562                 opt.DP          = q->DP;
 563                 opt.backlog     = q->backlog;
 564                 opt.prio        = q->prio;
 565                 opt.qth_min     = q->parms.qth_min >> q->parms.Wlog;
 566                 opt.qth_max     = q->parms.qth_max >> q->parms.Wlog;
 567                 opt.Wlog        = q->parms.Wlog;
 568                 opt.Plog        = q->parms.Plog;
 569                 opt.Scell_log   = q->parms.Scell_log;
 570                 opt.other       = q->stats.other;
 571                 opt.early       = q->stats.prob_drop;
 572                 opt.forced      = q->stats.forced_drop;
 573                 opt.pdrop       = q->stats.pdrop;
 574                 opt.packets     = q->packetsin;
 575                 opt.bytesin     = q->bytesin;
 576
 577                 if (gred_wred_mode(table)) {
 578                         q->parms.qidlestart =
 579                                 table->tab[table->def]->parms.qidlestart;
 580                         q->parms.qavg = table->tab[table->def]->parms.qavg;
 581                 }
 582
 583                 opt.qave = red_calc_qavg(&q->parms, q->parms.qavg);
 584
 585 append_opt:
 586                 RTA_APPEND(skb, sizeof(opt), &opt);
 587         }
 588
 589         RTA_NEST_END(skb, parms);
 590
 591         return RTA_NEST_END(skb, opts);
 592
 593 rtattr_failure:
 594         return RTA_NEST_CANCEL(skb, opts);
 595 }
 596
 597 static void gred_destroy(struct Qdisc *sch)
 598 {
 599         struct gred_sched *table = qdisc_priv(sch);
 600         int i;
 601
 602         for (i = 0;i < table->DPs; i++) {
 603                 if (table->tab[i])
 604                         gred_destroy_vq(table->tab[i]);
 605         }
 606 }
 607
 608 static struct Qdisc_ops gred_qdisc_ops = {
 609         .next           =       NULL,
 610         .cl_ops         =       NULL,
 611         .id             =       "gred",
 612         .priv_size      =       sizeof(struct gred_sched),
 613         .enqueue        =       gred_enqueue,
 614         .dequeue        =       gred_dequeue,
 615         .requeue        =       gred_requeue,
 616         .drop           =       gred_drop,
 617         .init           =       gred_init,
 618         .reset          =       gred_reset,
 619         .destroy        =       gred_destroy,
 620         .change         =       gred_change,
 621         .dump           =       gred_dump,
 622         .owner          =       THIS_MODULE,
 623 };
 624
 625 static int __init gred_module_init(void)
 626 {
 627         return register_qdisc(&gred_qdisc_ops);
 628 }
 629 static void __exit gred_module_exit(void)
 630 {
 631         unregister_qdisc(&gred_qdisc_ops);
 632 }
 633 module_init(gred_module_init)
 634 module_exit(gred_module_exit)
 635 MODULE_LICENSE("GPL");