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