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 = qdisc_enqueue(skb, child);
  96         if (likely(ret == NET_XMIT_SUCCESS)) {
  97                 qdisc_bstats_update(sch, skb);
  98                 sch->q.qlen++;
  99         } else if (net_xmit_drop_count(ret)) {
 100                 q->stats.pdrop++;
 101                 sch->qstats.drops++;
 102         }
 103         return ret;
 104
 105 congestion_drop:
 106         qdisc_drop(skb, sch);
 107         return NET_XMIT_CN;
 108 }
 109
 110 static struct sk_buff * red_dequeue(struct Qdisc* sch)
 111 {
 112         struct sk_buff *skb;
 113         struct red_sched_data *q = qdisc_priv(sch);
 114         struct Qdisc *child = q->qdisc;
 115
 116         skb = child->dequeue(child);
 117         if (skb)
 118                 sch->q.qlen--;
 119         else if (!red_is_idling(&q->parms))
 120                 red_start_of_idle_period(&q->parms);
 121
 122         return skb;
 123 }
 124
 125 static struct sk_buff * red_peek(struct Qdisc* sch)
 126 {
 127         struct red_sched_data *q = qdisc_priv(sch);
 128         struct Qdisc *child = q->qdisc;
 129
 130         return child->ops->peek(child);
 131 }
 132
 133 static unsigned int red_drop(struct Qdisc* sch)
 134 {
 135         struct red_sched_data *q = qdisc_priv(sch);
 136         struct Qdisc *child = q->qdisc;
 137         unsigned int len;
 138
 139         if (child->ops->drop && (len = child->ops->drop(child)) > 0) {
 140                 q->stats.other++;
 141                 sch->qstats.drops++;
 142                 sch->q.qlen--;
 143                 return len;
 144         }
 145
 146         if (!red_is_idling(&q->parms))
 147                 red_start_of_idle_period(&q->parms);
 148
 149         return 0;
 150 }
 151
 152 static void red_reset(struct Qdisc* sch)
 153 {
 154         struct red_sched_data *q = qdisc_priv(sch);
 155
 156         qdisc_reset(q->qdisc);
 157         sch->q.qlen = 0;
 158         red_restart(&q->parms);
 159 }
 160
 161 static void red_destroy(struct Qdisc *sch)
 162 {
 163         struct red_sched_data *q = qdisc_priv(sch);
 164         qdisc_destroy(q->qdisc);
 165 }
 166
 167 static const struct nla_policy red_policy[TCA_RED_MAX + 1] = {
 168         [TCA_RED_PARMS] = { .len = sizeof(struct tc_red_qopt) },
 169         [TCA_RED_STAB]  = { .len = RED_STAB_SIZE },
 170 };
 171
 172 static int red_change(struct Qdisc *sch, struct nlattr *opt)
 173 {
 174         struct red_sched_data *q = qdisc_priv(sch);
 175         struct nlattr *tb[TCA_RED_MAX + 1];
 176         struct tc_red_qopt *ctl;
 177         struct Qdisc *child = NULL;
 178         int err;
 179
 180         if (opt == NULL)
 181                 return -EINVAL;
 182
 183         err = nla_parse_nested(tb, TCA_RED_MAX, opt, red_policy);
 184         if (err < 0)
 185                 return err;
 186
 187         if (tb[TCA_RED_PARMS] == NULL ||
 188             tb[TCA_RED_STAB] == NULL)
 189                 return -EINVAL;
 190
 191         ctl = nla_data(tb[TCA_RED_PARMS]);
 192
 193         if (ctl->limit > 0) {
 194                 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit);
 195                 if (IS_ERR(child))
 196                         return PTR_ERR(child);
 197         }
 198
 199         sch_tree_lock(sch);
 200         q->flags = ctl->flags;
 201         q->limit = ctl->limit;
 202         if (child) {
 203                 qdisc_tree_decrease_qlen(q->qdisc, q->qdisc->q.qlen);
 204                 qdisc_destroy(q->qdisc);
 205                 q->qdisc = child;
 206         }
 207
 208         red_set_parms(&q->parms, ctl->qth_min, ctl->qth_max, ctl->Wlog,
 209                                  ctl->Plog, ctl->Scell_log,
 210                                  nla_data(tb[TCA_RED_STAB]));
 211
 212         if (skb_queue_empty(&sch->q))
 213                 red_end_of_idle_period(&q->parms);
 214
 215         sch_tree_unlock(sch);
 216         return 0;
 217 }
 218
 219 static int red_init(struct Qdisc* sch, struct nlattr *opt)
 220 {
 221         struct red_sched_data *q = qdisc_priv(sch);
 222
 223         q->qdisc = &noop_qdisc;
 224         return red_change(sch, opt);
 225 }
 226
 227 static int red_dump(struct Qdisc *sch, struct sk_buff *skb)
 228 {
 229         struct red_sched_data *q = qdisc_priv(sch);
 230         struct nlattr *opts = NULL;
 231         struct tc_red_qopt opt = {
 232                 .limit          = q->limit,
 233                 .flags          = q->flags,
 234                 .qth_min        = q->parms.qth_min >> q->parms.Wlog,
 235                 .qth_max        = q->parms.qth_max >> q->parms.Wlog,
 236                 .Wlog           = q->parms.Wlog,
 237                 .Plog           = q->parms.Plog,
 238                 .Scell_log      = q->parms.Scell_log,
 239         };
 240
 241         sch->qstats.backlog = q->qdisc->qstats.backlog;
 242         opts = nla_nest_start(skb, TCA_OPTIONS);
 243         if (opts == NULL)
 244                 goto nla_put_failure;
 245         NLA_PUT(skb, TCA_RED_PARMS, sizeof(opt), &opt);
 246         return nla_nest_end(skb, opts);
 247
 248 nla_put_failure:
 249         nla_nest_cancel(skb, opts);
 250         return -EMSGSIZE;
 251 }
 252
 253 static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
 254 {
 255         struct red_sched_data *q = qdisc_priv(sch);
 256         struct tc_red_xstats st = {
 257                 .early  = q->stats.prob_drop + q->stats.forced_drop,
 258                 .pdrop  = q->stats.pdrop,
 259                 .other  = q->stats.other,
 260                 .marked = q->stats.prob_mark + q->stats.forced_mark,
 261         };
 262
 263         return gnet_stats_copy_app(d, &st, sizeof(st));
 264 }
 265
 266 static int red_dump_class(struct Qdisc *sch, unsigned long cl,
 267                           struct sk_buff *skb, struct tcmsg *tcm)
 268 {
 269         struct red_sched_data *q = qdisc_priv(sch);
 270
 271         tcm->tcm_handle |= TC_H_MIN(1);
 272         tcm->tcm_info = q->qdisc->handle;
 273         return 0;
 274 }
 275
 276 static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
 277                      struct Qdisc **old)
 278 {
 279         struct red_sched_data *q = qdisc_priv(sch);
 280
 281         if (new == NULL)
 282                 new = &noop_qdisc;
 283
 284         sch_tree_lock(sch);
 285         *old = q->qdisc;
 286         q->qdisc = new;
 287         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
 288         qdisc_reset(*old);
 289         sch_tree_unlock(sch);
 290         return 0;
 291 }
 292
 293 static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg)
 294 {
 295         struct red_sched_data *q = qdisc_priv(sch);
 296         return q->qdisc;
 297 }
 298
 299 static unsigned long red_get(struct Qdisc *sch, u32 classid)
 300 {
 301         return 1;
 302 }
 303
 304 static void red_put(struct Qdisc *sch, unsigned long arg)
 305 {
 306 }
 307
 308 static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker)
 309 {
 310         if (!walker->stop) {
 311                 if (walker->count >= walker->skip)
 312                         if (walker->fn(sch, 1, walker) < 0) {
 313                                 walker->stop = 1;
 314                                 return;
 315                         }
 316                 walker->count++;
 317         }
 318 }
 319
 320 static const struct Qdisc_class_ops red_class_ops = {
 321         .graft          =       red_graft,
 322         .leaf           =       red_leaf,
 323         .get            =       red_get,
 324         .put            =       red_put,
 325         .walk           =       red_walk,
 326         .dump           =       red_dump_class,
 327 };
 328
 329 static struct Qdisc_ops red_qdisc_ops __read_mostly = {
 330         .id             =       "red",
 331         .priv_size      =       sizeof(struct red_sched_data),
 332         .cl_ops         =       &red_class_ops,
 333         .enqueue        =       red_enqueue,
 334         .dequeue        =       red_dequeue,
 335         .peek           =       red_peek,
 336         .drop           =       red_drop,
 337         .init           =       red_init,
 338         .reset          =       red_reset,
 339         .destroy        =       red_destroy,
 340         .change         =       red_change,
 341         .dump           =       red_dump,
 342         .dump_stats     =       red_dump_stats,
 343         .owner          =       THIS_MODULE,
 344 };
 345
 346 static int __init red_module_init(void)
 347 {
 348         return register_qdisc(&red_qdisc_ops);
 349 }
 350
 351 static void __exit red_module_exit(void)
 352 {
 353         unregister_qdisc(&red_qdisc_ops);
 354 }
 355
 356 module_init(red_module_init)
 357 module_exit(red_module_exit)
 358
 359 MODULE_LICENSE("GPL");