net/sched/sch_codel.c

   1 /*
   2  * Codel - The Controlled-Delay Active Queue Management algorithm
   3  *
   4  *  Copyright (C) 2011-2012 Kathleen Nichols <nichols@pollere.com>
   5  *  Copyright (C) 2011-2012 Van Jacobson <van@pollere.net>
   6  *
   7  *  Implemented on linux by :
   8  *  Copyright (C) 2012 Michael D. Taht <dave.taht@bufferbloat.net>
   9  *  Copyright (C) 2012 Eric Dumazet <edumazet@google.com>
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions, and the following disclaimer,
  16  *    without modification.
  17  * 2. Redistributions in binary form must reproduce the above copyright
  18  *    notice, this list of conditions and the following disclaimer in the
  19  *    documentation and/or other materials provided with the distribution.
  20  * 3. The names of the authors may not be used to endorse or promote products
  21  *    derived from this software without specific prior written permission.
  22  *
  23  * Alternatively, provided that this notice is retained in full, this
  24  * software may be distributed under the terms of the GNU General
  25  * Public License ("GPL") version 2, in which case the provisions of the
  26  * GPL apply INSTEAD OF those given above.
  27  *
  28  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  29  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  30  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  31  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  32  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  33  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  34  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  35  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  36  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  37  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  38  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
  39  * DAMAGE.
  40  *
  41  */
  42
  43 #include <linux/module.h>
  44 #include <linux/slab.h>
  45 #include <linux/types.h>
  46 #include <linux/kernel.h>
  47 #include <linux/errno.h>
  48 #include <linux/skbuff.h>
  49 #include <linux/prefetch.h>
  50 #include <net/pkt_sched.h>
  51 #include <net/codel.h>
  52
  53
  54 #define DEFAULT_CODEL_LIMIT 1000
  55
  56 struct codel_sched_data {
  57         struct codel_params     params;
  58         struct codel_vars       vars;
  59         struct codel_stats      stats;
  60         u32                     drop_overlimit;
  61 };
  62
  63 /* This is the specific function called from codel_dequeue()
  64  * to dequeue a packet from queue. Note: backlog is handled in
  65  * codel, we dont need to reduce it here.
  66  */
  67 static struct sk_buff *dequeue(struct codel_vars *vars, struct Qdisc *sch)
  68 {
  69         struct sk_buff *skb = __skb_dequeue(&sch->q);
  70
  71         prefetch(&skb->end); /* we'll need skb_shinfo() */
  72         return skb;
  73 }
  74
  75 static struct sk_buff *codel_qdisc_dequeue(struct Qdisc *sch)
  76 {
  77         struct codel_sched_data *q = qdisc_priv(sch);
  78         struct sk_buff *skb;
  79
  80         skb = codel_dequeue(sch, &q->params, &q->vars, &q->stats, dequeue);
  81
  82         /* We cant call qdisc_tree_decrease_qlen() if our qlen is 0,
  83          * or HTB crashes. Defer it for next round.
  84          */
  85         if (q->stats.drop_count && sch->q.qlen) {
  86                 qdisc_tree_decrease_qlen(sch, q->stats.drop_count);
  87                 q->stats.drop_count = 0;
  88         }
  89         if (skb)
  90                 qdisc_bstats_update(sch, skb);
  91         return skb;
  92 }
  93
  94 static int codel_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
  95 {
  96         struct codel_sched_data *q;
  97
  98         if (likely(qdisc_qlen(sch) < sch->limit)) {
  99                 codel_set_enqueue_time(skb);
 100                 return qdisc_enqueue_tail(skb, sch);
 101         }
 102         q = qdisc_priv(sch);
 103         q->drop_overlimit++;
 104         return qdisc_drop(skb, sch);
 105 }
 106
 107 static const struct nla_policy codel_policy[TCA_CODEL_MAX + 1] = {
 108         [TCA_CODEL_TARGET]      = { .type = NLA_U32 },
 109         [TCA_CODEL_LIMIT]       = { .type = NLA_U32 },
 110         [TCA_CODEL_INTERVAL]    = { .type = NLA_U32 },
 111         [TCA_CODEL_ECN]         = { .type = NLA_U32 },
 112 };
 113
 114 static int codel_change(struct Qdisc *sch, struct nlattr *opt)
 115 {
 116         struct codel_sched_data *q = qdisc_priv(sch);
 117         struct nlattr *tb[TCA_CODEL_MAX + 1];
 118         unsigned int qlen;
 119         int err;
 120
 121         if (!opt)
 122                 return -EINVAL;
 123
 124         err = nla_parse_nested(tb, TCA_CODEL_MAX, opt, codel_policy);
 125         if (err < 0)
 126                 return err;
 127
 128         sch_tree_lock(sch);
 129
 130         if (tb[TCA_CODEL_TARGET]) {
 131                 u32 target = nla_get_u32(tb[TCA_CODEL_TARGET]);
 132
 133                 q->params.target = ((u64)target * NSEC_PER_USEC) >> CODEL_SHIFT;
 134         }
 135
 136         if (tb[TCA_CODEL_INTERVAL]) {
 137                 u32 interval = nla_get_u32(tb[TCA_CODEL_INTERVAL]);
 138
 139                 q->params.interval = ((u64)interval * NSEC_PER_USEC) >> CODEL_SHIFT;
 140         }
 141
 142         if (tb[TCA_CODEL_LIMIT])
 143                 sch->limit = nla_get_u32(tb[TCA_CODEL_LIMIT]);
 144
 145         if (tb[TCA_CODEL_ECN])
 146                 q->params.ecn = !!nla_get_u32(tb[TCA_CODEL_ECN]);
 147
 148         qlen = sch->q.qlen;
 149         while (sch->q.qlen > sch->limit) {
 150                 struct sk_buff *skb = __skb_dequeue(&sch->q);
 151
 152                 sch->qstats.backlog -= qdisc_pkt_len(skb);
 153                 qdisc_drop(skb, sch);
 154         }
 155         qdisc_tree_decrease_qlen(sch, qlen - sch->q.qlen);
 156
 157         sch_tree_unlock(sch);
 158         return 0;
 159 }
 160
 161 static int codel_init(struct Qdisc *sch, struct nlattr *opt)
 162 {
 163         struct codel_sched_data *q = qdisc_priv(sch);
 164
 165         sch->limit = DEFAULT_CODEL_LIMIT;
 166
 167         codel_params_init(&q->params);
 168         codel_vars_init(&q->vars);
 169         codel_stats_init(&q->stats);
 170
 171         if (opt) {
 172                 int err = codel_change(sch, opt);
 173
 174                 if (err)
 175                         return err;
 176         }
 177
 178         if (sch->limit >= 1)
 179                 sch->flags |= TCQ_F_CAN_BYPASS;
 180         else
 181                 sch->flags &= ~TCQ_F_CAN_BYPASS;
 182
 183         return 0;
 184 }
 185
 186 static int codel_dump(struct Qdisc *sch, struct sk_buff *skb)
 187 {
 188         struct codel_sched_data *q = qdisc_priv(sch);
 189         struct nlattr *opts;
 190
 191         opts = nla_nest_start(skb, TCA_OPTIONS);
 192         if (opts == NULL)
 193                 goto nla_put_failure;
 194
 195         if (nla_put_u32(skb, TCA_CODEL_TARGET,
 196                         codel_time_to_us(q->params.target)) ||
 197             nla_put_u32(skb, TCA_CODEL_LIMIT,
 198                         sch->limit) ||
 199             nla_put_u32(skb, TCA_CODEL_INTERVAL,
 200                         codel_time_to_us(q->params.interval)) ||
 201             nla_put_u32(skb, TCA_CODEL_ECN,
 202                         q->params.ecn))
 203                 goto nla_put_failure;
 204
 205         return nla_nest_end(skb, opts);
 206
 207 nla_put_failure:
 208         nla_nest_cancel(skb, opts);
 209         return -1;
 210 }
 211
 212 static int codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
 213 {
 214         const struct codel_sched_data *q = qdisc_priv(sch);
 215         struct tc_codel_xstats st = {
 216                 .maxpacket      = q->stats.maxpacket,
 217                 .count          = q->vars.count,
 218                 .lastcount      = q->vars.lastcount,
 219                 .drop_overlimit = q->drop_overlimit,
 220                 .ldelay         = codel_time_to_us(q->vars.ldelay),
 221                 .dropping       = q->vars.dropping,
 222                 .ecn_mark       = q->stats.ecn_mark,
 223         };
 224
 225         if (q->vars.dropping) {
 226                 codel_tdiff_t delta = q->vars.drop_next - codel_get_time();
 227
 228                 if (delta >= 0)
 229                         st.drop_next = codel_time_to_us(delta);
 230                 else
 231                         st.drop_next = -codel_time_to_us(-delta);
 232         }
 233
 234         return gnet_stats_copy_app(d, &st, sizeof(st));
 235 }
 236
 237 static void codel_reset(struct Qdisc *sch)
 238 {
 239         struct codel_sched_data *q = qdisc_priv(sch);
 240
 241         qdisc_reset_queue(sch);
 242         codel_vars_init(&q->vars);
 243 }
 244
 245 static struct Qdisc_ops codel_qdisc_ops __read_mostly = {
 246         .id             =       "codel",
 247         .priv_size      =       sizeof(struct codel_sched_data),
 248
 249         .enqueue        =       codel_qdisc_enqueue,
 250         .dequeue        =       codel_qdisc_dequeue,
 251         .peek           =       qdisc_peek_dequeued,
 252         .init           =       codel_init,
 253         .reset          =       codel_reset,
 254         .change         =       codel_change,
 255         .dump           =       codel_dump,
 256         .dump_stats     =       codel_dump_stats,
 257         .owner          =       THIS_MODULE,
 258 };
 259
 260 static int __init codel_module_init(void)
 261 {
 262         return register_qdisc(&codel_qdisc_ops);
 263 }
 264
 265 static void __exit codel_module_exit(void)
 266 {
 267         unregister_qdisc(&codel_qdisc_ops);
 268 }
 269
 270 module_init(codel_module_init)
 271 module_exit(codel_module_exit)
 272
 273 MODULE_DESCRIPTION("Controlled Delay queue discipline");
 274 MODULE_AUTHOR("Dave Taht");
 275 MODULE_AUTHOR("Eric Dumazet");
 276 MODULE_LICENSE("Dual BSD/GPL");