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) 2012,2015 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_reduce_backlog() 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_reduce_backlog(sch, q->stats.drop_count, q->stats.drop_len);
  87                 q->stats.drop_count = 0;
  88                 q->stats.drop_len = 0;
  89         }
  90         if (skb)
  91                 qdisc_bstats_update(sch, skb);
  92         return skb;
  93 }
  94
  95 static int codel_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
  96 {
  97         struct codel_sched_data *q;
  98
  99         if (likely(qdisc_qlen(sch) < sch->limit)) {
 100                 codel_set_enqueue_time(skb);
 101                 return qdisc_enqueue_tail(skb, sch);
 102         }
 103         q = qdisc_priv(sch);
 104         q->drop_overlimit++;
 105         return qdisc_drop(skb, sch);
 106 }
 107
 108 static const struct nla_policy codel_policy[TCA_CODEL_MAX + 1] = {
 109         [TCA_CODEL_TARGET]      = { .type = NLA_U32 },
 110         [TCA_CODEL_LIMIT]       = { .type = NLA_U32 },
 111         [TCA_CODEL_INTERVAL]    = { .type = NLA_U32 },
 112         [TCA_CODEL_ECN]         = { .type = NLA_U32 },
 113         [TCA_CODEL_CE_THRESHOLD]= { .type = NLA_U32 },
 114 };
 115
 116 static int codel_change(struct Qdisc *sch, struct nlattr *opt)
 117 {
 118         struct codel_sched_data *q = qdisc_priv(sch);
 119         struct nlattr *tb[TCA_CODEL_MAX + 1];
 120         unsigned int qlen, dropped = 0;
 121         int err;
 122
 123         if (!opt)
 124                 return -EINVAL;
 125
 126         err = nla_parse_nested(tb, TCA_CODEL_MAX, opt, codel_policy);
 127         if (err < 0)
 128                 return err;
 129
 130         sch_tree_lock(sch);
 131
 132         if (tb[TCA_CODEL_TARGET]) {
 133                 u32 target = nla_get_u32(tb[TCA_CODEL_TARGET]);
 134
 135                 q->params.target = ((u64)target * NSEC_PER_USEC) >> CODEL_SHIFT;
 136         }
 137
 138         if (tb[TCA_CODEL_CE_THRESHOLD]) {
 139                 u64 val = nla_get_u32(tb[TCA_CODEL_CE_THRESHOLD]);
 140
 141                 q->params.ce_threshold = (val * NSEC_PER_USEC) >> CODEL_SHIFT;
 142         }
 143
 144         if (tb[TCA_CODEL_INTERVAL]) {
 145                 u32 interval = nla_get_u32(tb[TCA_CODEL_INTERVAL]);
 146
 147                 q->params.interval = ((u64)interval * NSEC_PER_USEC) >> CODEL_SHIFT;
 148         }
 149
 150         if (tb[TCA_CODEL_LIMIT])
 151                 sch->limit = nla_get_u32(tb[TCA_CODEL_LIMIT]);
 152
 153         if (tb[TCA_CODEL_ECN])
 154                 q->params.ecn = !!nla_get_u32(tb[TCA_CODEL_ECN]);
 155
 156         qlen = sch->q.qlen;
 157         while (sch->q.qlen > sch->limit) {
 158                 struct sk_buff *skb = __skb_dequeue(&sch->q);
 159
 160                 dropped += qdisc_pkt_len(skb);
 161                 qdisc_qstats_backlog_dec(sch, skb);
 162                 qdisc_drop(skb, sch);
 163         }
 164         qdisc_tree_reduce_backlog(sch, qlen - sch->q.qlen, dropped);
 165
 166         sch_tree_unlock(sch);
 167         return 0;
 168 }
 169
 170 static int codel_init(struct Qdisc *sch, struct nlattr *opt)
 171 {
 172         struct codel_sched_data *q = qdisc_priv(sch);
 173
 174         sch->limit = DEFAULT_CODEL_LIMIT;
 175
 176         codel_params_init(&q->params, sch);
 177         codel_vars_init(&q->vars);
 178         codel_stats_init(&q->stats);
 179
 180         if (opt) {
 181                 int err = codel_change(sch, opt);
 182
 183                 if (err)
 184                         return err;
 185         }
 186
 187         if (sch->limit >= 1)
 188                 sch->flags |= TCQ_F_CAN_BYPASS;
 189         else
 190                 sch->flags &= ~TCQ_F_CAN_BYPASS;
 191
 192         return 0;
 193 }
 194
 195 static int codel_dump(struct Qdisc *sch, struct sk_buff *skb)
 196 {
 197         struct codel_sched_data *q = qdisc_priv(sch);
 198         struct nlattr *opts;
 199
 200         opts = nla_nest_start(skb, TCA_OPTIONS);
 201         if (opts == NULL)
 202                 goto nla_put_failure;
 203
 204         if (nla_put_u32(skb, TCA_CODEL_TARGET,
 205                         codel_time_to_us(q->params.target)) ||
 206             nla_put_u32(skb, TCA_CODEL_LIMIT,
 207                         sch->limit) ||
 208             nla_put_u32(skb, TCA_CODEL_INTERVAL,
 209                         codel_time_to_us(q->params.interval)) ||
 210             nla_put_u32(skb, TCA_CODEL_ECN,
 211                         q->params.ecn))
 212                 goto nla_put_failure;
 213         if (q->params.ce_threshold != CODEL_DISABLED_THRESHOLD &&
 214             nla_put_u32(skb, TCA_CODEL_CE_THRESHOLD,
 215                         codel_time_to_us(q->params.ce_threshold)))
 216                 goto nla_put_failure;
 217         return nla_nest_end(skb, opts);
 218
 219 nla_put_failure:
 220         nla_nest_cancel(skb, opts);
 221         return -1;
 222 }
 223
 224 static int codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
 225 {
 226         const struct codel_sched_data *q = qdisc_priv(sch);
 227         struct tc_codel_xstats st = {
 228                 .maxpacket      = q->stats.maxpacket,
 229                 .count          = q->vars.count,
 230                 .lastcount      = q->vars.lastcount,
 231                 .drop_overlimit = q->drop_overlimit,
 232                 .ldelay         = codel_time_to_us(q->vars.ldelay),
 233                 .dropping       = q->vars.dropping,
 234                 .ecn_mark       = q->stats.ecn_mark,
 235                 .ce_mark        = q->stats.ce_mark,
 236         };
 237
 238         if (q->vars.dropping) {
 239                 codel_tdiff_t delta = q->vars.drop_next - codel_get_time();
 240
 241                 if (delta >= 0)
 242                         st.drop_next = codel_time_to_us(delta);
 243                 else
 244                         st.drop_next = -codel_time_to_us(-delta);
 245         }
 246
 247         return gnet_stats_copy_app(d, &st, sizeof(st));
 248 }
 249
 250 static void codel_reset(struct Qdisc *sch)
 251 {
 252         struct codel_sched_data *q = qdisc_priv(sch);
 253
 254         qdisc_reset_queue(sch);
 255         codel_vars_init(&q->vars);
 256 }
 257
 258 static struct Qdisc_ops codel_qdisc_ops __read_mostly = {
 259         .id             =       "codel",
 260         .priv_size      =       sizeof(struct codel_sched_data),
 261
 262         .enqueue        =       codel_qdisc_enqueue,
 263         .dequeue        =       codel_qdisc_dequeue,
 264         .peek           =       qdisc_peek_dequeued,
 265         .init           =       codel_init,
 266         .reset          =       codel_reset,
 267         .change         =       codel_change,
 268         .dump           =       codel_dump,
 269         .dump_stats     =       codel_dump_stats,
 270         .owner          =       THIS_MODULE,
 271 };
 272
 273 static int __init codel_module_init(void)
 274 {
 275         return register_qdisc(&codel_qdisc_ops);
 276 }
 277
 278 static void __exit codel_module_exit(void)
 279 {
 280         unregister_qdisc(&codel_qdisc_ops);
 281 }
 282
 283 module_init(codel_module_init)
 284 module_exit(codel_module_exit)
 285
 286 MODULE_DESCRIPTION("Controlled Delay queue discipline");
 287 MODULE_AUTHOR("Dave Taht");
 288 MODULE_AUTHOR("Eric Dumazet");
 289 MODULE_LICENSE("Dual BSD/GPL");