net/sched/sch_red.c

   1 /*
   2  * net/sched/sch_red.c  Random Early Detection queue.
   3  *
   4  *              This program is free software; you can redistribute it and/or
   5  *              modify it under the terms of the GNU General Public License
   6  *              as published by the Free Software Foundation; either version
   7  *              2 of the License, or (at your option) any later version.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  *
  11  * Changes:
  12  * J Hadi Salim 980914: computation fixes
  13  * Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly.
  14  * J Hadi Salim 980816:  ECN support
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/skbuff.h>
  21 #include <net/pkt_sched.h>
  22 #include <net/inet_ecn.h>
  23 #include <net/red.h>
  24
  25
  26 /*      Parameters, settable by user:
  27         -----------------------------
  28
  29         limit           - bytes (must be > qth_max + burst)
  30
  31         Hard limit on queue length, should be chosen >qth_max
  32         to allow packet bursts. This parameter does not
  33         affect the algorithms behaviour and can be chosen
  34         arbitrarily high (well, less than ram size)
  35         Really, this limit will never be reached
  36         if RED works correctly.
  37  */
  38
  39 struct red_sched_data
  40 {
  41         u32                     limit;          /* HARD maximal queue length */
  42         unsigned char           flags;
  43         struct red_parms        parms;
  44         struct red_stats        stats;
  45         struct Qdisc            *qdisc;
  46 };
  47
  48 static inline int red_use_ecn(struct red_sched_data *q)
  49 {
  50         return q->flags & TC_RED_ECN;
  51 }
  52
  53 static inline int red_use_harddrop(struct red_sched_data *q)
  54 {
  55         return q->flags & TC_RED_HARDDROP;
  56 }
  57
  58 static int red_enqueue(struct sk_buff *skb, struct Qdisc* sch)
  59 {
  60         struct red_sched_data *q = qdisc_priv(sch);
  61         struct Qdisc *child = q->qdisc;
  62         int ret;
  63
  64         q->parms.qavg = red_calc_qavg(&q->parms, child->qstats.backlog);
  65
  66         if (red_is_idling(&q->parms))
  67                 red_end_of_idle_period(&q->parms);
  68
  69         switch (red_action(&q->parms, q->parms.qavg)) {
  70                 case RED_DONT_MARK:
  71                         break;
  72
  73                 case RED_PROB_MARK:
  74                         sch->qstats.overlimits++;
  75                         if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) {
  76                                 q->stats.prob_drop++;
  77                                 goto congestion_drop;
  78                         }
  79
  80                         q->stats.prob_mark++;
  81                         break;
  82
  83                 case RED_HARD_MARK:
  84                         sch->qstats.overlimits++;
  85                         if (red_use_harddrop(q) || !red_use_ecn(q) ||
  86                             !INET_ECN_set_ce(skb)) {
  87                                 q->stats.forced_drop++;
  88                                 goto congestion_drop;
  89                         }
  90
  91                         q->stats.forced_mark++;
  92                         break;
  93         }
  94
  95         ret = child->enqueue(skb, child);
  96         if (likely(ret == NET_XMIT_SUCCESS)) {
  97                 sch->bstats.bytes += skb->len;
  98                 sch->bstats.packets++;
  99                 sch->q.qlen++;
 100         } else {
 101                 q->stats.pdrop++;
 102                 sch->qstats.drops++;
 103         }
 104         return ret;
 105
 106 congestion_drop:
 107         qdisc_drop(skb, sch);
 108         return NET_XMIT_CN;
 109 }
 110
 111 static int red_requeue(struct sk_buff *skb, struct Qdisc* sch)
 112 {
 113         struct red_sched_data *q = qdisc_priv(sch);
 114         struct Qdisc *child = q->qdisc;
 115         int ret;
 116
 117         if (red_is_idling(&q->parms))
 118                 red_end_of_idle_period(&q->parms);
 119
 120         ret = child->ops->requeue(skb, child);
 121         if (likely(ret == NET_XMIT_SUCCESS)) {
 122                 sch->qstats.requeues++;
 123                 sch->q.qlen++;
 124         }
 125         return ret;
 126 }
 127
 128 static struct sk_buff * red_dequeue(struct Qdisc* sch)
 129 {
 130         struct sk_buff *skb;
 131         struct red_sched_data *q = qdisc_priv(sch);
 132         struct Qdisc *child = q->qdisc;
 133
 134         skb = child->dequeue(child);
 135         if (skb)
 136                 sch->q.qlen--;
 137         else if (!red_is_idling(&q->parms))
 138                 red_start_of_idle_period(&q->parms);
 139
 140         return skb;
 141 }
 142
 143 static unsigned int red_drop(struct Qdisc* sch)
 144 {
 145         struct red_sched_data *q = qdisc_priv(sch);
 146         struct Qdisc *child = q->qdisc;
 147         unsigned int len;
 148
 149         if (child->ops->drop && (len = child->ops->drop(child)) > 0) {
 150                 q->stats.other++;
 151                 sch->qstats.drops++;
 152                 sch->q.qlen--;
 153                 return len;
 154         }
 155
 156         if (!red_is_idling(&q->parms))
 157                 red_start_of_idle_period(&q->parms);
 158
 159         return 0;
 160 }
 161
 162 static void red_reset(struct Qdisc* sch)
 163 {
 164         struct red_sched_data *q = qdisc_priv(sch);
 165
 166         qdisc_reset(q->qdisc);
 167         sch->q.qlen = 0;
 168         red_restart(&q->parms);
 169 }
 170
 171 static void red_destroy(struct Qdisc *sch)
 172 {
 173         struct red_sched_data *q = qdisc_priv(sch);
 174         qdisc_destroy(q->qdisc);
 175 }
 176
 177 static struct Qdisc *red_create_dflt(struct Qdisc *sch, u32 limit)
 178 {
 179         struct Qdisc *q;
 180         struct nlattr *nla;
 181         int ret;
 182
 183         q = qdisc_create_dflt(sch->dev, &bfifo_qdisc_ops,
 184                               TC_H_MAKE(sch->handle, 1));
 185         if (q) {
 186                 nla = kmalloc(nla_attr_size(sizeof(struct tc_fifo_qopt)),
 187                               GFP_KERNEL);
 188                 if (nla) {
 189                         nla->nla_type = RTM_NEWQDISC;
 190                         nla->nla_len = nla_attr_size(sizeof(struct tc_fifo_qopt));
 191                         ((struct tc_fifo_qopt *)nla_data(nla))->limit = limit;
 192
 193                         ret = q->ops->change(q, nla);
 194                         kfree(nla);
 195
 196                         if (ret == 0)
 197                                 return q;
 198                 }
 199                 qdisc_destroy(q);
 200         }
 201         return NULL;
 202 }
 203
 204 static int red_change(struct Qdisc *sch, struct nlattr *opt)
 205 {
 206         struct red_sched_data *q = qdisc_priv(sch);
 207         struct nlattr *tb[TCA_RED_MAX + 1];
 208         struct tc_red_qopt *ctl;
 209         struct Qdisc *child = NULL;
 210         int err;
 211
 212         if (opt == NULL)
 213                 return -EINVAL;
 214
 215         err = nla_parse_nested(tb, TCA_RED_MAX, opt, NULL);
 216         if (err < 0)
 217                 return err;
 218
 219         if (tb[TCA_RED_PARMS] == NULL ||
 220             nla_len(tb[TCA_RED_PARMS]) < sizeof(*ctl) ||
 221             tb[TCA_RED_STAB] == NULL ||
 222             nla_len(tb[TCA_RED_STAB]) < RED_STAB_SIZE)
 223                 return -EINVAL;
 224
 225         ctl = nla_data(tb[TCA_RED_PARMS]);
 226
 227         if (ctl->limit > 0) {
 228                 child = red_create_dflt(sch, ctl->limit);
 229                 if (child == NULL)
 230                         return -ENOMEM;
 231         }
 232
 233         sch_tree_lock(sch);
 234         q->flags = ctl->flags;
 235         q->limit = ctl->limit;
 236         if (child) {
 237                 qdisc_tree_decrease_qlen(q->qdisc, q->qdisc->q.qlen);
 238                 qdisc_destroy(xchg(&q->qdisc, child));
 239         }
 240
 241         red_set_parms(&q->parms, ctl->qth_min, ctl->qth_max, ctl->Wlog,
 242                                  ctl->Plog, ctl->Scell_log,
 243                                  nla_data(tb[TCA_RED_STAB]));
 244
 245         if (skb_queue_empty(&sch->q))
 246                 red_end_of_idle_period(&q->parms);
 247
 248         sch_tree_unlock(sch);
 249         return 0;
 250 }
 251
 252 static int red_init(struct Qdisc* sch, struct nlattr *opt)
 253 {
 254         struct red_sched_data *q = qdisc_priv(sch);
 255
 256         q->qdisc = &noop_qdisc;
 257         return red_change(sch, opt);
 258 }
 259
 260 static int red_dump(struct Qdisc *sch, struct sk_buff *skb)
 261 {
 262         struct red_sched_data *q = qdisc_priv(sch);
 263         struct nlattr *opts = NULL;
 264         struct tc_red_qopt opt = {
 265                 .limit          = q->limit,
 266                 .flags          = q->flags,
 267                 .qth_min        = q->parms.qth_min >> q->parms.Wlog,
 268                 .qth_max        = q->parms.qth_max >> q->parms.Wlog,
 269                 .Wlog           = q->parms.Wlog,
 270                 .Plog           = q->parms.Plog,
 271                 .Scell_log      = q->parms.Scell_log,
 272         };
 273
 274         opts = nla_nest_start(skb, TCA_OPTIONS);
 275         if (opts == NULL)
 276                 goto nla_put_failure;
 277         NLA_PUT(skb, TCA_RED_PARMS, sizeof(opt), &opt);
 278         return nla_nest_end(skb, opts);
 279
 280 nla_put_failure:
 281         return nla_nest_cancel(skb, opts);
 282 }
 283
 284 static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
 285 {
 286         struct red_sched_data *q = qdisc_priv(sch);
 287         struct tc_red_xstats st = {
 288                 .early  = q->stats.prob_drop + q->stats.forced_drop,
 289                 .pdrop  = q->stats.pdrop,
 290                 .other  = q->stats.other,
 291                 .marked = q->stats.prob_mark + q->stats.forced_mark,
 292         };
 293
 294         return gnet_stats_copy_app(d, &st, sizeof(st));
 295 }
 296
 297 static int red_dump_class(struct Qdisc *sch, unsigned long cl,
 298                           struct sk_buff *skb, struct tcmsg *tcm)
 299 {
 300         struct red_sched_data *q = qdisc_priv(sch);
 301
 302         if (cl != 1)
 303                 return -ENOENT;
 304         tcm->tcm_handle |= TC_H_MIN(1);
 305         tcm->tcm_info = q->qdisc->handle;
 306         return 0;
 307 }
 308
 309 static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
 310                      struct Qdisc **old)
 311 {
 312         struct red_sched_data *q = qdisc_priv(sch);
 313
 314         if (new == NULL)
 315                 new = &noop_qdisc;
 316
 317         sch_tree_lock(sch);
 318         *old = xchg(&q->qdisc, new);
 319         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
 320         qdisc_reset(*old);
 321         sch_tree_unlock(sch);
 322         return 0;
 323 }
 324
 325 static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg)
 326 {
 327         struct red_sched_data *q = qdisc_priv(sch);
 328         return q->qdisc;
 329 }
 330
 331 static unsigned long red_get(struct Qdisc *sch, u32 classid)
 332 {
 333         return 1;
 334 }
 335
 336 static void red_put(struct Qdisc *sch, unsigned long arg)
 337 {
 338         return;
 339 }
 340
 341 static int red_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
 342                             struct nlattr **tca, unsigned long *arg)
 343 {
 344         return -ENOSYS;
 345 }
 346
 347 static int red_delete(struct Qdisc *sch, unsigned long cl)
 348 {
 349         return -ENOSYS;
 350 }
 351
 352 static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker)
 353 {
 354         if (!walker->stop) {
 355                 if (walker->count >= walker->skip)
 356                         if (walker->fn(sch, 1, walker) < 0) {
 357                                 walker->stop = 1;
 358                                 return;
 359                         }
 360                 walker->count++;
 361         }
 362 }
 363
 364 static struct tcf_proto **red_find_tcf(struct Qdisc *sch, unsigned long cl)
 365 {
 366         return NULL;
 367 }
 368
 369 static const struct Qdisc_class_ops red_class_ops = {
 370         .graft          =       red_graft,
 371         .leaf           =       red_leaf,
 372         .get            =       red_get,
 373         .put            =       red_put,
 374         .change         =       red_change_class,
 375         .delete         =       red_delete,
 376         .walk           =       red_walk,
 377         .tcf_chain      =       red_find_tcf,
 378         .dump           =       red_dump_class,
 379 };
 380
 381 static struct Qdisc_ops red_qdisc_ops __read_mostly = {
 382         .id             =       "red",
 383         .priv_size      =       sizeof(struct red_sched_data),
 384         .cl_ops         =       &red_class_ops,
 385         .enqueue        =       red_enqueue,
 386         .dequeue        =       red_dequeue,
 387         .requeue        =       red_requeue,
 388         .drop           =       red_drop,
 389         .init           =       red_init,
 390         .reset          =       red_reset,
 391         .destroy        =       red_destroy,
 392         .change         =       red_change,
 393         .dump           =       red_dump,
 394         .dump_stats     =       red_dump_stats,
 395         .owner          =       THIS_MODULE,
 396 };
 397
 398 static int __init red_module_init(void)
 399 {
 400         return register_qdisc(&red_qdisc_ops);
 401 }
 402
 403 static void __exit red_module_exit(void)
 404 {
 405         unregister_qdisc(&red_qdisc_ops);
 406 }
 407
 408 module_init(red_module_init)
 409 module_exit(red_module_exit)
 410
 411 MODULE_LICENSE("GPL");