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