blob: d168aca115ccf0ff4c4ca98107327c533314c618 [file] [log] [blame]
Eric Dumazet76e3cc12012-05-10 07:51:25 +00001#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>
Eric Dumazet80ba92f2015-05-08 15:05:12 -070010 * Copyright (C) 2012,2015 Eric Dumazet <edumazet@google.com>
Eric Dumazet76e3cc12012-05-10 07:51:25 +000011 *
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 */
62typedef u32 codel_time_t;
63typedef s32 codel_tdiff_t;
64#define CODEL_SHIFT 10
65#define MS2TIME(a) ((a * NSEC_PER_MSEC) >> CODEL_SHIFT)
66
67static inline codel_time_t codel_get_time(void)
68{
Eric Dumazetd2de8752014-08-22 18:32:09 -070069 u64 ns = ktime_get_ns();
Eric Dumazet76e3cc12012-05-10 07:51:25 +000070
71 return ns >> CODEL_SHIFT;
72}
73
Jesper Dangaard Brouer1ba3aab2013-10-31 22:10:55 +010074/* 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)
Eric Dumazet76e3cc12012-05-10 07:51:25 +000089
90/* Qdiscs using codel plugin must use codel_skb_cb in their own cb[] */
91struct codel_skb_cb {
92 codel_time_t enqueue_time;
93};
94
95static 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
101static codel_time_t codel_get_enqueue_time(const struct sk_buff *skb)
102{
103 return get_codel_cb(skb)->enqueue_time;
104}
105
106static void codel_set_enqueue_time(struct sk_buff *skb)
107{
108 get_codel_cb(skb)->enqueue_time = codel_get_time();
109}
110
111static 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)
Eric Dumazet80ba92f2015-05-08 15:05:12 -0700122 * @ce_threshold: threshold for marking packets with ECN CE
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000123 * @interval: width of moving time window
Eric Dumazeta5d28092015-04-30 09:40:40 -0700124 * @mtu: device mtu, or minimal queue backlog in bytes.
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000125 * @ecn: is Explicit Congestion Notification enabled
126 */
127struct codel_params {
128 codel_time_t target;
Eric Dumazet80ba92f2015-05-08 15:05:12 -0700129 codel_time_t ce_threshold;
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000130 codel_time_t interval;
Eric Dumazeta5d28092015-04-30 09:40:40 -0700131 u32 mtu;
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000132 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
Eric Dumazet536edd62012-05-12 03:32:13 +0000141 * @rec_inv_sqrt: reciprocal value of sqrt(count) >> 1
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000142 * @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 */
147struct codel_vars {
148 u32 count;
149 u32 lastcount;
Eric Dumazet6ff272c2012-05-12 21:23:23 +0000150 bool dropping;
151 u16 rec_inv_sqrt;
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000152 codel_time_t first_above_time;
153 codel_time_t drop_next;
154 codel_time_t ldelay;
155};
156
Eric Dumazet6ff272c2012-05-12 21:23:23 +0000157#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
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000161/**
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()
WANG Cong2ccccf52016-02-25 14:55:01 -0800165 * @drop_len: bytes of dropped packets in dequeue()
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000166 * ecn_mark: number of packets we ECN marked instead of dropping
Eric Dumazet80ba92f2015-05-08 15:05:12 -0700167 * ce_mark: number of packets CE marked because sojourn time was above ce_threshold
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000168 */
169struct codel_stats {
170 u32 maxpacket;
171 u32 drop_count;
WANG Cong2ccccf52016-02-25 14:55:01 -0800172 u32 drop_len;
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000173 u32 ecn_mark;
Eric Dumazet80ba92f2015-05-08 15:05:12 -0700174 u32 ce_mark;
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000175};
176
Eric Dumazet80ba92f2015-05-08 15:05:12 -0700177#define CODEL_DISABLED_THRESHOLD INT_MAX
178
Eric Dumazeta5d28092015-04-30 09:40:40 -0700179static void codel_params_init(struct codel_params *params,
180 const struct Qdisc *sch)
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000181{
182 params->interval = MS2TIME(100);
183 params->target = MS2TIME(5);
Eric Dumazeta5d28092015-04-30 09:40:40 -0700184 params->mtu = psched_mtu(qdisc_dev(sch));
Eric Dumazet80ba92f2015-05-08 15:05:12 -0700185 params->ce_threshold = CODEL_DISABLED_THRESHOLD;
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000186 params->ecn = false;
187}
188
189static void codel_vars_init(struct codel_vars *vars)
190{
Eric Dumazet536edd62012-05-12 03:32:13 +0000191 memset(vars, 0, sizeof(*vars));
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000192}
193
194static void codel_stats_init(struct codel_stats *stats)
195{
Eric Dumazeta5d28092015-04-30 09:40:40 -0700196 stats->maxpacket = 0;
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000197}
198
Eric Dumazet536edd62012-05-12 03:32:13 +0000199/*
200 * http://en.wikipedia.org/wiki/Methods_of_computing_square_roots#Iterative_methods_for_reciprocal_square_roots
201 * new_invsqrt = (invsqrt / 2) * (3 - count * invsqrt^2)
202 *
Eric Dumazet6ff272c2012-05-12 21:23:23 +0000203 * Here, invsqrt is a fixed point number (< 1.0), 32bit mantissa, aka Q0.32
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000204 */
Eric Dumazet536edd62012-05-12 03:32:13 +0000205static void codel_Newton_step(struct codel_vars *vars)
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000206{
Eric Dumazet6ff272c2012-05-12 21:23:23 +0000207 u32 invsqrt = ((u32)vars->rec_inv_sqrt) << REC_INV_SQRT_SHIFT;
208 u32 invsqrt2 = ((u64)invsqrt * invsqrt) >> 32;
209 u64 val = (3LL << 32) - ((u64)vars->count * invsqrt2);
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000210
Eric Dumazet6ff272c2012-05-12 21:23:23 +0000211 val >>= 2; /* avoid overflow in following multiply */
212 val = (val * invsqrt) >> (32 - 2 + 1);
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000213
Eric Dumazet6ff272c2012-05-12 21:23:23 +0000214 vars->rec_inv_sqrt = val >> REC_INV_SQRT_SHIFT;
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000215}
216
Eric Dumazet536edd62012-05-12 03:32:13 +0000217/*
218 * CoDel control_law is t + interval/sqrt(count)
219 * We maintain in rec_inv_sqrt the reciprocal value of sqrt(count) to avoid
220 * both sqrt() and divide operation.
221 */
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000222static codel_time_t codel_control_law(codel_time_t t,
223 codel_time_t interval,
Eric Dumazet536edd62012-05-12 03:32:13 +0000224 u32 rec_inv_sqrt)
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000225{
Daniel Borkmann89770b02014-01-22 02:29:40 +0100226 return t + reciprocal_scale(interval, rec_inv_sqrt << REC_INV_SQRT_SHIFT);
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000227}
228
Eric Dumazet536edd62012-05-12 03:32:13 +0000229static bool codel_should_drop(const struct sk_buff *skb,
Eric Dumazet865ec552012-05-16 04:39:09 +0000230 struct Qdisc *sch,
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000231 struct codel_vars *vars,
232 struct codel_params *params,
233 struct codel_stats *stats,
234 codel_time_t now)
235{
236 bool ok_to_drop;
237
238 if (!skb) {
239 vars->first_above_time = 0;
240 return false;
241 }
242
243 vars->ldelay = now - codel_get_enqueue_time(skb);
Eric Dumazet865ec552012-05-16 04:39:09 +0000244 sch->qstats.backlog -= qdisc_pkt_len(skb);
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000245
246 if (unlikely(qdisc_pkt_len(skb) > stats->maxpacket))
247 stats->maxpacket = qdisc_pkt_len(skb);
248
249 if (codel_time_before(vars->ldelay, params->target) ||
Eric Dumazeta5d28092015-04-30 09:40:40 -0700250 sch->qstats.backlog <= params->mtu) {
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000251 /* went below - stay below for at least interval */
252 vars->first_above_time = 0;
253 return false;
254 }
255 ok_to_drop = false;
256 if (vars->first_above_time == 0) {
257 /* just went above from below. If we stay above
258 * for at least interval we'll say it's ok to drop
259 */
260 vars->first_above_time = now + params->interval;
261 } else if (codel_time_after(now, vars->first_above_time)) {
262 ok_to_drop = true;
263 }
264 return ok_to_drop;
265}
266
267typedef struct sk_buff * (*codel_skb_dequeue_t)(struct codel_vars *vars,
268 struct Qdisc *sch);
269
270static struct sk_buff *codel_dequeue(struct Qdisc *sch,
271 struct codel_params *params,
272 struct codel_vars *vars,
273 struct codel_stats *stats,
Eric Dumazet865ec552012-05-16 04:39:09 +0000274 codel_skb_dequeue_t dequeue_func)
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000275{
276 struct sk_buff *skb = dequeue_func(vars, sch);
277 codel_time_t now;
278 bool drop;
279
280 if (!skb) {
281 vars->dropping = false;
282 return skb;
283 }
284 now = codel_get_time();
Eric Dumazet865ec552012-05-16 04:39:09 +0000285 drop = codel_should_drop(skb, sch, vars, params, stats, now);
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000286 if (vars->dropping) {
287 if (!drop) {
288 /* sojourn time below target - leave dropping state */
289 vars->dropping = false;
290 } else if (codel_time_after_eq(now, vars->drop_next)) {
291 /* It's time for the next drop. Drop the current
292 * packet and dequeue the next. The dequeue might
293 * take us out of dropping state.
294 * If not, schedule the next drop.
295 * A large backlog might result in drop rates so high
296 * that the next drop should happen now,
297 * hence the while loop.
298 */
299 while (vars->dropping &&
300 codel_time_after_eq(now, vars->drop_next)) {
Eric Dumazet536edd62012-05-12 03:32:13 +0000301 vars->count++; /* dont care of possible wrap
302 * since there is no more divide
303 */
304 codel_Newton_step(vars);
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000305 if (params->ecn && INET_ECN_set_ce(skb)) {
306 stats->ecn_mark++;
307 vars->drop_next =
308 codel_control_law(vars->drop_next,
309 params->interval,
Eric Dumazet536edd62012-05-12 03:32:13 +0000310 vars->rec_inv_sqrt);
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000311 goto end;
312 }
WANG Cong2ccccf52016-02-25 14:55:01 -0800313 stats->drop_len += qdisc_pkt_len(skb);
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000314 qdisc_drop(skb, sch);
315 stats->drop_count++;
316 skb = dequeue_func(vars, sch);
Eric Dumazet865ec552012-05-16 04:39:09 +0000317 if (!codel_should_drop(skb, sch,
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000318 vars, params, stats, now)) {
319 /* leave dropping state */
320 vars->dropping = false;
321 } else {
322 /* and schedule the next drop */
323 vars->drop_next =
324 codel_control_law(vars->drop_next,
325 params->interval,
Eric Dumazet536edd62012-05-12 03:32:13 +0000326 vars->rec_inv_sqrt);
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000327 }
328 }
329 }
330 } else if (drop) {
Eric Dumazet2359a472012-07-29 20:52:21 +0000331 u32 delta;
332
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000333 if (params->ecn && INET_ECN_set_ce(skb)) {
334 stats->ecn_mark++;
335 } else {
WANG Cong2ccccf52016-02-25 14:55:01 -0800336 stats->drop_len += qdisc_pkt_len(skb);
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000337 qdisc_drop(skb, sch);
338 stats->drop_count++;
339
340 skb = dequeue_func(vars, sch);
Eric Dumazet865ec552012-05-16 04:39:09 +0000341 drop = codel_should_drop(skb, sch, vars, params,
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000342 stats, now);
343 }
344 vars->dropping = true;
345 /* if min went above target close to when we last went below it
346 * assume that the drop rate that controlled the queue on the
347 * last cycle is a good starting point to control it now.
348 */
Eric Dumazet2359a472012-07-29 20:52:21 +0000349 delta = vars->count - vars->lastcount;
350 if (delta > 1 &&
351 codel_time_before(now - vars->drop_next,
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000352 16 * params->interval)) {
Eric Dumazet2359a472012-07-29 20:52:21 +0000353 vars->count = delta;
Eric Dumazet536edd62012-05-12 03:32:13 +0000354 /* we dont care if rec_inv_sqrt approximation
355 * is not very precise :
356 * Next Newton steps will correct it quadratically.
357 */
358 codel_Newton_step(vars);
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000359 } else {
360 vars->count = 1;
Eric Dumazet6ff272c2012-05-12 21:23:23 +0000361 vars->rec_inv_sqrt = ~0U >> REC_INV_SQRT_SHIFT;
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000362 }
363 vars->lastcount = vars->count;
364 vars->drop_next = codel_control_law(now, params->interval,
Eric Dumazet536edd62012-05-12 03:32:13 +0000365 vars->rec_inv_sqrt);
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000366 }
367end:
Eric Dumazet80ba92f2015-05-08 15:05:12 -0700368 if (skb && codel_time_after(vars->ldelay, params->ce_threshold) &&
369 INET_ECN_set_ce(skb))
370 stats->ce_mark++;
Eric Dumazet76e3cc12012-05-10 07:51:25 +0000371 return skb;
372}
373#endif