include/net/codel.h

   1 #ifndef __NET_SCHED_CODEL_H
   2 #define __NET_SCHED_CODEL_H
   3
   4 /*
   5  * Codel - The Controlled-Delay Active Queue Management algorithm
   6  *
   7  *  Copyright (C) 2011-2012 Kathleen Nichols <nichols@pollere.com>
   8  *  Copyright (C) 2011-2012 Van Jacobson <van@pollere.net>
   9  *  Copyright (C) 2012 Michael D. Taht <dave.taht@bufferbloat.net>
  10  *  Copyright (C) 2012,2015 Eric Dumazet <edumazet@google.com>
  11  *
  12  * Redistribution and use in source and binary forms, with or without
  13  * modification, are permitted provided that the following conditions
  14  * are met:
  15  * 1. Redistributions of source code must retain the above copyright
  16  *    notice, this list of conditions, and the following disclaimer,
  17  *    without modification.
  18  * 2. Redistributions in binary form must reproduce the above copyright
  19  *    notice, this list of conditions and the following disclaimer in the
  20  *    documentation and/or other materials provided with the distribution.
  21  * 3. The names of the authors may not be used to endorse or promote products
  22  *    derived from this software without specific prior written permission.
  23  *
  24  * Alternatively, provided that this notice is retained in full, this
  25  * software may be distributed under the terms of the GNU General
  26  * Public License ("GPL") version 2, in which case the provisions of the
  27  * GPL apply INSTEAD OF those given above.
  28  *
  29  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  30  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  31  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  32  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  33  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  34  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  35  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  36  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  37  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  38  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  39  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
  40  * DAMAGE.
  41  *
  42  */
  43
  44 #include <linux/types.h>
  45 #include <linux/ktime.h>
  46 #include <linux/skbuff.h>
  47 #include <net/pkt_sched.h>
  48 #include <net/inet_ecn.h>
  49
  50 /* Controlling Queue Delay (CoDel) algorithm
  51  * =========================================
  52  * Source : Kathleen Nichols and Van Jacobson
  53  * http://queue.acm.org/detail.cfm?id=2209336
  54  *
  55  * Implemented on linux by Dave Taht and Eric Dumazet
  56  */
  57
  58
  59 /* CoDel uses a 1024 nsec clock, encoded in u32
  60  * This gives a range of 2199 seconds, because of signed compares
  61  */
  62 typedef u32 codel_time_t;
  63 typedef s32 codel_tdiff_t;
  64 #define CODEL_SHIFT 10
  65 #define MS2TIME(a) ((a * NSEC_PER_MSEC) >> CODEL_SHIFT)
  66
  67 static inline codel_time_t codel_get_time(void)
  68 {
  69         u64 ns = ktime_get_ns();
  70
  71         return ns >> CODEL_SHIFT;
  72 }
  73
  74 /* Dealing with timer wrapping, according to RFC 1982, as desc in wikipedia:
  75  *  https://en.wikipedia.org/wiki/Serial_number_arithmetic#General_Solution
  76  * codel_time_after(a,b) returns true if the time a is after time b.
  77  */
  78 #define codel_time_after(a, b)                                          \
  79         (typecheck(codel_time_t, a) &&                                  \
  80          typecheck(codel_time_t, b) &&                                  \
  81          ((s32)((a) - (b)) > 0))
  82 #define codel_time_before(a, b)         codel_time_after(b, a)
  83
  84 #define codel_time_after_eq(a, b)                                       \
  85         (typecheck(codel_time_t, a) &&                                  \
  86          typecheck(codel_time_t, b) &&                                  \
  87          ((s32)((a) - (b)) >= 0))
  88 #define codel_time_before_eq(a, b)      codel_time_after_eq(b, a)
  89
  90 /* Qdiscs using codel plugin must use codel_skb_cb in their own cb[] */
  91 struct codel_skb_cb {
  92         codel_time_t enqueue_time;
  93 };
  94
  95 static struct codel_skb_cb *get_codel_cb(const struct sk_buff *skb)
  96 {
  97         qdisc_cb_private_validate(skb, sizeof(struct codel_skb_cb));
  98         return (struct codel_skb_cb *)qdisc_skb_cb(skb)->data;
  99 }
 100
 101 static codel_time_t codel_get_enqueue_time(const struct sk_buff *skb)
 102 {
 103         return get_codel_cb(skb)->enqueue_time;
 104 }
 105
 106 static void codel_set_enqueue_time(struct sk_buff *skb)
 107 {
 108         get_codel_cb(skb)->enqueue_time = codel_get_time();
 109 }
 110
 111 static inline u32 codel_time_to_us(codel_time_t val)
 112 {
 113         u64 valns = ((u64)val << CODEL_SHIFT);
 114
 115         do_div(valns, NSEC_PER_USEC);
 116         return (u32)valns;
 117 }
 118
 119 /**
 120  * struct codel_params - contains codel parameters
 121  * @target:     target queue size (in time units)
 122  * @ce_threshold:  threshold for marking packets with ECN CE
 123  * @interval:   width of moving time window
 124  * @mtu:        device mtu, or minimal queue backlog in bytes.
 125  * @ecn:        is Explicit Congestion Notification enabled
 126  */
 127 struct codel_params {
 128         codel_time_t    target;
 129         codel_time_t    ce_threshold;
 130         codel_time_t    interval;
 131         u32             mtu;
 132         bool            ecn;
 133 };
 134
 135 /**
 136  * struct codel_vars - contains codel variables
 137  * @count:              how many drops we've done since the last time we
 138  *                      entered dropping state
 139  * @lastcount:          count at entry to dropping state
 140  * @dropping:           set to true if in dropping state
 141  * @rec_inv_sqrt:       reciprocal value of sqrt(count) >> 1
 142  * @first_above_time:   when we went (or will go) continuously above target
 143  *                      for interval
 144  * @drop_next:          time to drop next packet, or when we dropped last
 145  * @ldelay:             sojourn time of last dequeued packet
 146  */
 147 struct codel_vars {
 148         u32             count;
 149         u32             lastcount;
 150         bool            dropping;
 151         u16             rec_inv_sqrt;
 152         codel_time_t    first_above_time;
 153         codel_time_t    drop_next;
 154         codel_time_t    ldelay;
 155 };
 156
 157 #define REC_INV_SQRT_BITS (8 * sizeof(u16)) /* or sizeof_in_bits(rec_inv_sqrt) */
 158 /* needed shift to get a Q0.32 number from rec_inv_sqrt */
 159 #define REC_INV_SQRT_SHIFT (32 - REC_INV_SQRT_BITS)
 160
 161 /**
 162  * struct codel_stats - contains codel shared variables and stats
 163  * @maxpacket:  largest packet we've seen so far
 164  * @drop_count: temp count of dropped packets in dequeue()
 165  * ecn_mark:    number of packets we ECN marked instead of dropping
 166  * ce_mark:     number of packets CE marked because sojourn time was above ce_threshold
 167  */
 168 struct codel_stats {
 169         u32             maxpacket;
 170         u32             drop_count;
 171         u32             ecn_mark;
 172         u32             ce_mark;
 173 };
 174
 175 #define CODEL_DISABLED_THRESHOLD INT_MAX
 176
 177 static void codel_params_init(struct codel_params *params,
 178                               const struct Qdisc *sch)
 179 {
 180         params->interval = MS2TIME(100);
 181         params->target = MS2TIME(5);
 182         params->mtu = psched_mtu(qdisc_dev(sch));
 183         params->ce_threshold = CODEL_DISABLED_THRESHOLD;
 184         params->ecn = false;
 185 }
 186
 187 static void codel_vars_init(struct codel_vars *vars)
 188 {
 189         memset(vars, 0, sizeof(*vars));
 190 }
 191
 192 static void codel_stats_init(struct codel_stats *stats)
 193 {
 194         stats->maxpacket = 0;
 195 }
 196
 197 /*
 198  * http://en.wikipedia.org/wiki/Methods_of_computing_square_roots#Iterative_methods_for_reciprocal_square_roots
 199  * new_invsqrt = (invsqrt / 2) * (3 - count * invsqrt^2)
 200  *
 201  * Here, invsqrt is a fixed point number (< 1.0), 32bit mantissa, aka Q0.32
 202  */
 203 static void codel_Newton_step(struct codel_vars *vars)
 204 {
 205         u32 invsqrt = ((u32)vars->rec_inv_sqrt) << REC_INV_SQRT_SHIFT;
 206         u32 invsqrt2 = ((u64)invsqrt * invsqrt) >> 32;
 207         u64 val = (3LL << 32) - ((u64)vars->count * invsqrt2);
 208
 209         val >>= 2; /* avoid overflow in following multiply */
 210         val = (val * invsqrt) >> (32 - 2 + 1);
 211
 212         vars->rec_inv_sqrt = val >> REC_INV_SQRT_SHIFT;
 213 }
 214
 215 /*
 216  * CoDel control_law is t + interval/sqrt(count)
 217  * We maintain in rec_inv_sqrt the reciprocal value of sqrt(count) to avoid
 218  * both sqrt() and divide operation.
 219  */
 220 static codel_time_t codel_control_law(codel_time_t t,
 221                                       codel_time_t interval,
 222                                       u32 rec_inv_sqrt)
 223 {
 224         return t + reciprocal_scale(interval, rec_inv_sqrt << REC_INV_SQRT_SHIFT);
 225 }
 226
 227 static bool codel_should_drop(const struct sk_buff *skb,
 228                               struct Qdisc *sch,
 229                               struct codel_vars *vars,
 230                               struct codel_params *params,
 231                               struct codel_stats *stats,
 232                               codel_time_t now)
 233 {
 234         bool ok_to_drop;
 235
 236         if (!skb) {
 237                 vars->first_above_time = 0;
 238                 return false;
 239         }
 240
 241         vars->ldelay = now - codel_get_enqueue_time(skb);
 242         sch->qstats.backlog -= qdisc_pkt_len(skb);
 243
 244         if (unlikely(qdisc_pkt_len(skb) > stats->maxpacket))
 245                 stats->maxpacket = qdisc_pkt_len(skb);
 246
 247         if (codel_time_before(vars->ldelay, params->target) ||
 248             sch->qstats.backlog <= params->mtu) {
 249                 /* went below - stay below for at least interval */
 250                 vars->first_above_time = 0;
 251                 return false;
 252         }
 253         ok_to_drop = false;
 254         if (vars->first_above_time == 0) {
 255                 /* just went above from below. If we stay above
 256                  * for at least interval we'll say it's ok to drop
 257                  */
 258                 vars->first_above_time = now + params->interval;
 259         } else if (codel_time_after(now, vars->first_above_time)) {
 260                 ok_to_drop = true;
 261         }
 262         return ok_to_drop;
 263 }
 264
 265 typedef struct sk_buff * (*codel_skb_dequeue_t)(struct codel_vars *vars,
 266                                                 struct Qdisc *sch);
 267
 268 static struct sk_buff *codel_dequeue(struct Qdisc *sch,
 269                                      struct codel_params *params,
 270                                      struct codel_vars *vars,
 271                                      struct codel_stats *stats,
 272                                      codel_skb_dequeue_t dequeue_func)
 273 {
 274         struct sk_buff *skb = dequeue_func(vars, sch);
 275         codel_time_t now;
 276         bool drop;
 277
 278         if (!skb) {
 279                 vars->dropping = false;
 280                 return skb;
 281         }
 282         now = codel_get_time();
 283         drop = codel_should_drop(skb, sch, vars, params, stats, now);
 284         if (vars->dropping) {
 285                 if (!drop) {
 286                         /* sojourn time below target - leave dropping state */
 287                         vars->dropping = false;
 288                 } else if (codel_time_after_eq(now, vars->drop_next)) {
 289                         /* It's time for the next drop. Drop the current
 290                          * packet and dequeue the next. The dequeue might
 291                          * take us out of dropping state.
 292                          * If not, schedule the next drop.
 293                          * A large backlog might result in drop rates so high
 294                          * that the next drop should happen now,
 295                          * hence the while loop.
 296                          */
 297                         while (vars->dropping &&
 298                                codel_time_after_eq(now, vars->drop_next)) {
 299                                 vars->count++; /* dont care of possible wrap
 300                                                 * since there is no more divide
 301                                                 */
 302                                 codel_Newton_step(vars);
 303                                 if (params->ecn && INET_ECN_set_ce(skb)) {
 304                                         stats->ecn_mark++;
 305                                         vars->drop_next =
 306                                                 codel_control_law(vars->drop_next,
 307                                                                   params->interval,
 308                                                                   vars->rec_inv_sqrt);
 309                                         goto end;
 310                                 }
 311                                 qdisc_drop(skb, sch);
 312                                 stats->drop_count++;
 313                                 skb = dequeue_func(vars, sch);
 314                                 if (!codel_should_drop(skb, sch,
 315                                                        vars, params, stats, now)) {
 316                                         /* leave dropping state */
 317                                         vars->dropping = false;
 318                                 } else {
 319                                         /* and schedule the next drop */
 320                                         vars->drop_next =
 321                                                 codel_control_law(vars->drop_next,
 322                                                                   params->interval,
 323                                                                   vars->rec_inv_sqrt);
 324                                 }
 325                         }
 326                 }
 327         } else if (drop) {
 328                 u32 delta;
 329
 330                 if (params->ecn && INET_ECN_set_ce(skb)) {
 331                         stats->ecn_mark++;
 332                 } else {
 333                         qdisc_drop(skb, sch);
 334                         stats->drop_count++;
 335
 336                         skb = dequeue_func(vars, sch);
 337                         drop = codel_should_drop(skb, sch, vars, params,
 338                                                  stats, now);
 339                 }
 340                 vars->dropping = true;
 341                 /* if min went above target close to when we last went below it
 342                  * assume that the drop rate that controlled the queue on the
 343                  * last cycle is a good starting point to control it now.
 344                  */
 345                 delta = vars->count - vars->lastcount;
 346                 if (delta > 1 &&
 347                     codel_time_before(now - vars->drop_next,
 348                                       16 * params->interval)) {
 349                         vars->count = delta;
 350                         /* we dont care if rec_inv_sqrt approximation
 351                          * is not very precise :
 352                          * Next Newton steps will correct it quadratically.
 353                          */
 354                         codel_Newton_step(vars);
 355                 } else {
 356                         vars->count = 1;
 357                         vars->rec_inv_sqrt = ~0U >> REC_INV_SQRT_SHIFT;
 358                 }
 359                 vars->lastcount = vars->count;
 360                 vars->drop_next = codel_control_law(now, params->interval,
 361                                                     vars->rec_inv_sqrt);
 362         }
 363 end:
 364         if (skb && codel_time_after(vars->ldelay, params->ce_threshold) &&
 365             INET_ECN_set_ce(skb))
 366                 stats->ce_mark++;
 367         return skb;
 368 }
 369 #endif