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 #include <net/codel_impl.h>
  53 #include <net/codel_qdisc.h>
  54
  55
  56 #define DEFAULT_CODEL_LIMIT 1000
  57
  58 struct codel_sched_data {
  59         struct codel_params     params;
  60         struct codel_vars       vars;
  61         struct codel_stats      stats;
  62         u32                     drop_overlimit;
  63 };
  64
  65 /* This is the specific function called from codel_dequeue()
  66  * to dequeue a packet from queue. Note: backlog is handled in
  67  * codel, we dont need to reduce it here.
  68  */
  69 static struct sk_buff *dequeue_func(struct codel_vars *vars, void *ctx)
  70 {
  71         struct Qdisc *sch = ctx;
  72         struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
  73
  74         if (skb)
  75                 sch->qstats.backlog -= qdisc_pkt_len(skb);
  76
  77         prefetch(&skb->end); /* we'll need skb_shinfo() */
  78         return skb;
  79 }
  80
  81 static void drop_func(struct sk_buff *skb, void *ctx)
  82 {
  83         struct Qdisc *sch = ctx;
  84
  85         kfree_skb(skb);
  86         qdisc_qstats_drop(sch);
  87 }
  88
  89 static struct sk_buff *codel_qdisc_dequeue(struct Qdisc *sch)
  90 {
  91         struct codel_sched_data *q = qdisc_priv(sch);
  92         struct sk_buff *skb;
  93
  94         skb = codel_dequeue(sch, &sch->qstats.backlog, &q->params, &q->vars,
  95                             &q->stats, qdisc_pkt_len, codel_get_enqueue_time,
  96                             drop_func, dequeue_func);
  97
  98         /* We cant call qdisc_tree_reduce_backlog() if our qlen is 0,
  99          * or HTB crashes. Defer it for next round.
 100          */
 101         if (q->stats.drop_count && sch->q.qlen) {
 102                 qdisc_tree_reduce_backlog(sch, q->stats.drop_count, q->stats.drop_len);
 103                 q->stats.drop_count = 0;
 104                 q->stats.drop_len = 0;
 105         }
 106         if (skb)
 107                 qdisc_bstats_update(sch, skb);
 108         return skb;
 109 }
 110
 111 static int codel_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 112                                struct sk_buff **to_free)
 113 {
 114         struct codel_sched_data *q;
 115
 116         if (likely(qdisc_qlen(sch) < sch->limit)) {
 117                 codel_set_enqueue_time(skb);
 118                 return qdisc_enqueue_tail(skb, sch);
 119         }
 120         q = qdisc_priv(sch);
 121         q->drop_overlimit++;
 122         return qdisc_drop(skb, sch, to_free);
 123 }
 124
 125 static const struct nla_policy codel_policy[TCA_CODEL_MAX + 1] = {
 126         [TCA_CODEL_TARGET]      = { .type = NLA_U32 },
 127         [TCA_CODEL_LIMIT]       = { .type = NLA_U32 },
 128         [TCA_CODEL_INTERVAL]    = { .type = NLA_U32 },
 129         [TCA_CODEL_ECN]         = { .type = NLA_U32 },
 130         [TCA_CODEL_CE_THRESHOLD]= { .type = NLA_U32 },
 131 };
 132
 133 static int codel_change(struct Qdisc *sch, struct nlattr *opt,
 134                         struct netlink_ext_ack *extack)
 135 {
 136         struct codel_sched_data *q = qdisc_priv(sch);
 137         struct nlattr *tb[TCA_CODEL_MAX + 1];
 138         unsigned int qlen, dropped = 0;
 139         int err;
 140
 141         if (!opt)
 142                 return -EINVAL;
 143
 144         err = nla_parse_nested(tb, TCA_CODEL_MAX, opt, codel_policy, NULL);
 145         if (err < 0)
 146                 return err;
 147
 148         sch_tree_lock(sch);
 149
 150         if (tb[TCA_CODEL_TARGET]) {
 151                 u32 target = nla_get_u32(tb[TCA_CODEL_TARGET]);
 152
 153                 q->params.target = ((u64)target * NSEC_PER_USEC) >> CODEL_SHIFT;
 154         }
 155
 156         if (tb[TCA_CODEL_CE_THRESHOLD]) {
 157                 u64 val = nla_get_u32(tb[TCA_CODEL_CE_THRESHOLD]);
 158
 159                 q->params.ce_threshold = (val * NSEC_PER_USEC) >> CODEL_SHIFT;
 160         }
 161
 162         if (tb[TCA_CODEL_INTERVAL]) {
 163                 u32 interval = nla_get_u32(tb[TCA_CODEL_INTERVAL]);
 164
 165                 q->params.interval = ((u64)interval * NSEC_PER_USEC) >> CODEL_SHIFT;
 166         }
 167
 168         if (tb[TCA_CODEL_LIMIT])
 169                 sch->limit = nla_get_u32(tb[TCA_CODEL_LIMIT]);
 170
 171         if (tb[TCA_CODEL_ECN])
 172                 q->params.ecn = !!nla_get_u32(tb[TCA_CODEL_ECN]);
 173
 174         qlen = sch->q.qlen;
 175         while (sch->q.qlen > sch->limit) {
 176                 struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
 177
 178                 dropped += qdisc_pkt_len(skb);
 179                 qdisc_qstats_backlog_dec(sch, skb);
 180                 rtnl_qdisc_drop(skb, sch);
 181         }
 182         qdisc_tree_reduce_backlog(sch, qlen - sch->q.qlen, dropped);
 183
 184         sch_tree_unlock(sch);
 185         return 0;
 186 }
 187
 188 static int codel_init(struct Qdisc *sch, struct nlattr *opt,
 189                       struct netlink_ext_ack *extack)
 190 {
 191         struct codel_sched_data *q = qdisc_priv(sch);
 192
 193         sch->limit = DEFAULT_CODEL_LIMIT;
 194
 195         codel_params_init(&q->params);
 196         codel_vars_init(&q->vars);
 197         codel_stats_init(&q->stats);
 198         q->params.mtu = psched_mtu(qdisc_dev(sch));
 199
 200         if (opt) {
 201                 int err = codel_change(sch, opt, extack);
 202
 203                 if (err)
 204                         return err;
 205         }
 206
 207         if (sch->limit >= 1)
 208                 sch->flags |= TCQ_F_CAN_BYPASS;
 209         else
 210                 sch->flags &= ~TCQ_F_CAN_BYPASS;
 211
 212         return 0;
 213 }
 214
 215 static int codel_dump(struct Qdisc *sch, struct sk_buff *skb)
 216 {
 217         struct codel_sched_data *q = qdisc_priv(sch);
 218         struct nlattr *opts;
 219
 220         opts = nla_nest_start(skb, TCA_OPTIONS);
 221         if (opts == NULL)
 222                 goto nla_put_failure;
 223
 224         if (nla_put_u32(skb, TCA_CODEL_TARGET,
 225                         codel_time_to_us(q->params.target)) ||
 226             nla_put_u32(skb, TCA_CODEL_LIMIT,
 227                         sch->limit) ||
 228             nla_put_u32(skb, TCA_CODEL_INTERVAL,
 229                         codel_time_to_us(q->params.interval)) ||
 230             nla_put_u32(skb, TCA_CODEL_ECN,
 231                         q->params.ecn))
 232                 goto nla_put_failure;
 233         if (q->params.ce_threshold != CODEL_DISABLED_THRESHOLD &&
 234             nla_put_u32(skb, TCA_CODEL_CE_THRESHOLD,
 235                         codel_time_to_us(q->params.ce_threshold)))
 236                 goto nla_put_failure;
 237         return nla_nest_end(skb, opts);
 238
 239 nla_put_failure:
 240         nla_nest_cancel(skb, opts);
 241         return -1;
 242 }
 243
 244 static int codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
 245 {
 246         const struct codel_sched_data *q = qdisc_priv(sch);
 247         struct tc_codel_xstats st = {
 248                 .maxpacket      = q->stats.maxpacket,
 249                 .count          = q->vars.count,
 250                 .lastcount      = q->vars.lastcount,
 251                 .drop_overlimit = q->drop_overlimit,
 252                 .ldelay         = codel_time_to_us(q->vars.ldelay),
 253                 .dropping       = q->vars.dropping,
 254                 .ecn_mark       = q->stats.ecn_mark,
 255                 .ce_mark        = q->stats.ce_mark,
 256         };
 257
 258         if (q->vars.dropping) {
 259                 codel_tdiff_t delta = q->vars.drop_next - codel_get_time();
 260
 261                 if (delta >= 0)
 262                         st.drop_next = codel_time_to_us(delta);
 263                 else
 264                         st.drop_next = -codel_time_to_us(-delta);
 265         }
 266
 267         return gnet_stats_copy_app(d, &st, sizeof(st));
 268 }
 269
 270 static void codel_reset(struct Qdisc *sch)
 271 {
 272         struct codel_sched_data *q = qdisc_priv(sch);
 273
 274         qdisc_reset_queue(sch);
 275         codel_vars_init(&q->vars);
 276 }
 277
 278 static struct Qdisc_ops codel_qdisc_ops __read_mostly = {
 279         .id             =       "codel",
 280         .priv_size      =       sizeof(struct codel_sched_data),
 281
 282         .enqueue        =       codel_qdisc_enqueue,
 283         .dequeue        =       codel_qdisc_dequeue,
 284         .peek           =       qdisc_peek_dequeued,
 285         .init           =       codel_init,
 286         .reset          =       codel_reset,
 287         .change         =       codel_change,
 288         .dump           =       codel_dump,
 289         .dump_stats     =       codel_dump_stats,
 290         .owner          =       THIS_MODULE,
 291 };
 292
 293 static int __init codel_module_init(void)
 294 {
 295         return register_qdisc(&codel_qdisc_ops);
 296 }
 297
 298 static void __exit codel_module_exit(void)
 299 {
 300         unregister_qdisc(&codel_qdisc_ops);
 301 }
 302
 303 module_init(codel_module_init)
 304 module_exit(codel_module_exit)
 305
 306 MODULE_DESCRIPTION("Controlled Delay queue discipline");
 307 MODULE_AUTHOR("Dave Taht");
 308 MODULE_AUTHOR("Eric Dumazet");
 309 MODULE_LICENSE("Dual BSD/GPL");