aboutsummaryrefslogtreecommitdiffstats
path: root/net/sched/sch_pie.c
blob: 7197bcaa14ba7062a248dfe96d933f3f90ca0133 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
// SPDX-License-Identifier: GPL-2.0-only
/* Copyright (C) 2013 Cisco Systems, Inc, 2013.
 *
 * Author: Vijay Subramanian <vijaynsu@cisco.com>
 * Author: Mythili Prabhu <mysuryan@cisco.com>
 *
 * ECN support is added by Naeem Khademi <naeemk@ifi.uio.no>
 * University of Oslo, Norway.
 *
 * References:
 * RFC 8033: https://tools.ietf.org/html/rfc8033
 */

#include <linux/module.h>
#include <linux/slab.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/errno.h>
#include <linux/skbuff.h>
#include <net/pkt_sched.h>
#include <net/inet_ecn.h>
#include <net/pie.h>

/* private data for the Qdisc */
struct pie_sched_data {
	struct pie_params params;
	struct pie_vars vars;
	struct pie_stats stats;
	struct timer_list adapt_timer;
	struct Qdisc *sch;
};

static bool drop_early(struct Qdisc *sch, u32 packet_size)
{
	struct pie_sched_data *q = qdisc_priv(sch);
	u64 rnd;
	u64 local_prob = q->vars.prob;
	u32 mtu = psched_mtu(qdisc_dev(sch));

	/* If there is still burst allowance left skip random early drop */
	if (q->vars.burst_time > 0)
		return false;

	/* If current delay is less than half of target, and
	 * if drop prob is low already, disable early_drop
	 */
	if ((q->vars.qdelay < q->params.target / 2) &&
	    (q->vars.prob < MAX_PROB / 5))
		return false;

	/* If we have fewer than 2 mtu-sized packets, disable drop_early,
	 * similar to min_th in RED
	 */
	if (sch->qstats.backlog < 2 * mtu)
		return false;

	/* If bytemode is turned on, use packet size to compute new
	 * probablity. Smaller packets will have lower drop prob in this case
	 */
	if (q->params.bytemode && packet_size <= mtu)
		local_prob = (u64)packet_size * div_u64(local_prob, mtu);
	else
		local_prob = q->vars.prob;

	if (local_prob == 0) {
		q->vars.accu_prob = 0;
		q->vars.accu_prob_overflows = 0;
	}

	if (local_prob > MAX_PROB - q->vars.accu_prob)
		q->vars.accu_prob_overflows++;

	q->vars.accu_prob += local_prob;

	if (q->vars.accu_prob_overflows == 0 &&
	    q->vars.accu_prob < (MAX_PROB / 100) * 85)
		return false;
	if (q->vars.accu_prob_overflows == 8 &&
	    q->vars.accu_prob >= MAX_PROB / 2)
		return true;

	prandom_bytes(&rnd, 8);
	if (rnd < local_prob) {
		q->vars.accu_prob = 0;
		q->vars.accu_prob_overflows = 0;
		return true;
	}

	return false;
}

static int pie_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
			     struct sk_buff **to_free)
{
	struct pie_sched_data *q = qdisc_priv(sch);
	bool enqueue = false;

	if (unlikely(qdisc_qlen(sch) >= sch->limit)) {
		q->stats.overlimit++;
		goto out;
	}

	if (!drop_early(sch, skb->len)) {
		enqueue = true;
	} else if (q->params.ecn && (q->vars.prob <= MAX_PROB / 10) &&
		   INET_ECN_set_ce(skb)) {
		/* If packet is ecn capable, mark it if drop probability
		 * is lower than 10%, else drop it.
		 */
		q->stats.ecn_mark++;
		enqueue = true;
	}

	/* we can enqueue the packet */
	if (enqueue) {
		/* Set enqueue time only when dq_rate_estimator is disabled. */
		if (!q->params.dq_rate_estimator)
			pie_set_enqueue_time(skb);

		q->stats.packets_in++;
		if (qdisc_qlen(sch) > q->stats.maxq)
			q->stats.maxq = qdisc_qlen(sch);

		return qdisc_enqueue_tail(skb, sch);
	}

out:
	q->stats.dropped++;
	q->vars.accu_prob = 0;
	q->vars.accu_prob_overflows = 0;
	return qdisc_drop(skb, sch, to_free);
}

static const struct nla_policy pie_policy[TCA_PIE_MAX + 1] = {
	[TCA_PIE_TARGET] = {.type = NLA_U32},
	[TCA_PIE_LIMIT] = {.type = NLA_U32},
	[TCA_PIE_TUPDATE] = {.type = NLA_U32},
	[TCA_PIE_ALPHA] = {.type = NLA_U32},
	[TCA_PIE_BETA] = {.type = NLA_U32},
	[TCA_PIE_ECN] = {.type = NLA_U32},
	[TCA_PIE_BYTEMODE] = {.type = NLA_U32},
	[TCA_PIE_DQ_RATE_ESTIMATOR] = {.type = NLA_U32},
};

static int pie_change(struct Qdisc *sch, struct nlattr *opt,
		      struct netlink_ext_ack *extack)
{
	struct pie_sched_data *q = qdisc_priv(sch);
	struct nlattr *tb[TCA_PIE_MAX + 1];
	unsigned int qlen, dropped = 0;
	int err;

	if (!opt)
		return -EINVAL;

	err = nla_parse_nested_deprecated(tb, TCA_PIE_MAX, opt, pie_policy,
					  NULL);
	if (err < 0)
		return err;

	sch_tree_lock(sch);

	/* convert from microseconds to pschedtime */
	if (tb[TCA_PIE_TARGET]) {
		/* target is in us */
		u32 target = nla_get_u32(tb[TCA_PIE_TARGET]);

		/* convert to pschedtime */
		q->params.target = PSCHED_NS2TICKS((u64)target * NSEC_PER_USEC);
	}

	/* tupdate is in jiffies */
	if (tb[TCA_PIE_TUPDATE])
		q->params.tupdate =
			usecs_to_jiffies(nla_get_u32(tb[TCA_PIE_TUPDATE]));

	if (tb[TCA_PIE_LIMIT]) {
		u32 limit = nla_get_u32(tb[TCA_PIE_LIMIT]);

		q->params.limit = limit;
		sch->limit = limit;
	}

	if (tb[TCA_PIE_ALPHA])
		q->params.alpha = nla_get_u32(tb[TCA_PIE_ALPHA]);

	if (tb[TCA_PIE_BETA])
		q->params.beta = nla_get_u32(tb[TCA_PIE_BETA]);

	if (tb[TCA_PIE_ECN])
		q->params.ecn = nla_get_u32(tb[TCA_PIE_ECN]);

	if (tb[TCA_PIE_BYTEMODE])
		q->params.bytemode = nla_get_u32(tb[TCA_PIE_BYTEMODE]);

	if (tb[TCA_PIE_DQ_RATE_ESTIMATOR])
		q->params.dq_rate_estimator =
				nla_get_u32(tb[TCA_PIE_DQ_RATE_ESTIMATOR]);

	/* Drop excess packets if new limit is lower */
	qlen = sch->q.qlen;
	while (sch->q.qlen > sch->limit) {
		struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);

		dropped += qdisc_pkt_len(skb);
		qdisc_qstats_backlog_dec(sch, skb);
		rtnl_qdisc_drop(skb, sch);
	}
	qdisc_tree_reduce_backlog(sch, qlen - sch->q.qlen, dropped);

	sch_tree_unlock(sch);
	return 0;
}

static void pie_process_dequeue(struct Qdisc *sch, struct sk_buff *skb)
{
	struct pie_sched_data *q = qdisc_priv(sch);
	int qlen = sch->qstats.backlog;	/* current queue size in bytes */
	psched_time_t now = psched_get_time();
	u32 dtime = 0;

	/* If dq_rate_estimator is disabled, calculate qdelay using the
	 * packet timestamp.
	 */
	if (!q->params.dq_rate_estimator) {
		q->vars.qdelay = now - pie_get_enqueue_time(skb);

		if (q->vars.dq_tstamp != DTIME_INVALID)
			dtime = now - q->vars.dq_tstamp;

		q->vars.dq_tstamp = now;

		if (qlen == 0)
			q->vars.qdelay = 0;

		if (dtime == 0)
			return;

		goto burst_allowance_reduction;
	}

	/* If current queue is about 10 packets or more and dq_count is unset
	 * we have enough packets to calculate the drain rate. Save
	 * current time as dq_tstamp and start measurement cycle.
	 */
	if (qlen >= QUEUE_THRESHOLD && q->vars.dq_count == DQCOUNT_INVALID) {
		q->vars.dq_tstamp = psched_get_time();
		q->vars.dq_count = 0;
	}

	/* Calculate the average drain rate from this value.  If queue length
	 * has receded to a small value viz., <= QUEUE_THRESHOLD bytes,reset
	 * the dq_count to -1 as we don't have enough packets to calculate the
	 * drain rate anymore The following if block is entered only when we
	 * have a substantial queue built up (QUEUE_THRESHOLD bytes or more)
	 * and we calculate the drain rate for the threshold here.  dq_count is
	 * in bytes, time difference in psched_time, hence rate is in
	 * bytes/psched_time.
	 */
	if (q->vars.dq_count != DQCOUNT_INVALID) {
		q->vars.dq_count += skb->len;

		if (q->vars.dq_count >= QUEUE_THRESHOLD) {
			u32 count = q->vars.dq_count << PIE_SCALE;

			dtime = now - q->vars.dq_tstamp;

			if (dtime == 0)
				return;

			count = count / dtime;

			if (q->vars.avg_dq_rate == 0)
				q->vars.avg_dq_rate = count;
			else
				q->vars.avg_dq_rate =
				    (q->vars.avg_dq_rate -
				     (q->vars.avg_dq_rate >> 3)) + (count >> 3);

			/* If the queue has receded below the threshold, we hold
			 * on to the last drain rate calculated, else we reset
			 * dq_count to 0 to re-enter the if block when the next
			 * packet is dequeued
			 */
			if (qlen < QUEUE_THRESHOLD) {
				q->vars.dq_count = DQCOUNT_INVALID;
			} else {
				q->vars.dq_count = 0;
				q->vars.dq_tstamp = psched_get_time();
			}

			goto burst_allowance_reduction;
		}
	}

	return;

burst_allowance_reduction:
	if (q->vars.burst_time > 0) {
		if (q->vars.burst_time > dtime)
			q->vars.burst_time -= dtime;
		else
			q->vars.burst_time = 0;
	}
}

static void calculate_probability(struct Qdisc *sch)
{
	struct pie_sched_data *q = qdisc_priv(sch);
	u32 qlen = sch->qstats.backlog;	/* queue size in bytes */
	psched_time_t qdelay = 0;	/* in pschedtime */
	psched_time_t qdelay_old = 0;	/* in pschedtime */
	s64 delta = 0;		/* determines the change in probability */
	u64 oldprob;
	u64 alpha, beta;
	u32 power;
	bool update_prob = true;

	if (q->params.dq_rate_estimator) {
		qdelay_old = q->vars.qdelay;
		q->vars.qdelay_old = q->vars.qdelay;

		if (q->vars.avg_dq_rate > 0)
			qdelay = (qlen << PIE_SCALE) / q->vars.avg_dq_rate;
		else
			qdelay = 0;
	} else {
		qdelay = q->vars.qdelay;
		qdelay_old = q->vars.qdelay_old;
	}

	/* If qdelay is zero and qlen is not, it means qlen is very small, less
	 * than dequeue_rate, so we do not update probabilty in this round
	 */
	if (qdelay == 0 && qlen != 0)
		update_prob = false;

	/* In the algorithm, alpha and beta are between 0 and 2 with typical
	 * value for alpha as 0.125. In this implementation, we use values 0-32
	 * passed from user space to represent this. Also, alpha and beta have
	 * unit of HZ and need to be scaled before they can used to update
	 * probability. alpha/beta are updated locally below by scaling down
	 * by 16 to come to 0-2 range.
	 */
	alpha = ((u64)q->params.alpha * (MAX_PROB / PSCHED_TICKS_PER_SEC)) >> 4;
	beta = ((u64)q->params.beta * (MAX_PROB / PSCHED_TICKS_PER_SEC)) >> 4;

	/* We scale alpha and beta differently depending on how heavy the
	 * congestion is. Please see RFC 8033 for details.
	 */
	if (q->vars.prob < MAX_PROB / 10) {
		alpha >>= 1;
		beta >>= 1;

		power = 100;
		while (q->vars.prob < div_u64(MAX_PROB, power) &&
		       power <= 1000000) {
			alpha >>= 2;
			beta >>= 2;
			power *= 10;
		}
	}

	/* alpha and beta should be between 0 and 32, in multiples of 1/16 */
	delta += alpha * (u64)(qdelay - q->params.target);
	delta += beta * (u64)(qdelay - qdelay_old);

	oldprob = q->vars.prob;

	/* to ensure we increase probability in steps of no more than 2% */
	if (delta > (s64)(MAX_PROB / (100 / 2)) &&
	    q->vars.prob >= MAX_PROB / 10)
		delta = (MAX_PROB / 100) * 2;

	/* Non-linear drop:
	 * Tune drop probability to increase quickly for high delays(>= 250ms)
	 * 250ms is derived through experiments and provides error protection
	 */

	if (qdelay > (PSCHED_NS2TICKS(250 * NSEC_PER_MSEC)))
		delta += MAX_PROB / (100 / 2);

	q->vars.prob += delta;

	if (delta > 0) {
		/* prevent overflow */
		if (q->vars.prob < oldprob) {
			q->vars.prob = MAX_PROB;
			/* Prevent normalization error. If probability is at
			 * maximum value already, we normalize it here, and
			 * skip the check to do a non-linear drop in the next
			 * section.
			 */
			update_prob = false;
		}
	} else {
		/* prevent underflow */
		if (q->vars.prob > oldprob)
			q->vars.prob = 0;
	}

	/* Non-linear drop in probability: Reduce drop probability quickly if
	 * delay is 0 for 2 consecutive Tupdate periods.
	 */

	if (qdelay == 0 && qdelay_old == 0 && update_prob)
		/* Reduce drop probability to 98.4% */
		q->vars.prob -= q->vars.prob / 64u;

	q->vars.qdelay = qdelay;
	q->vars.qlen_old = qlen;

	/* We restart the measurement cycle if the following conditions are met
	 * 1. If the delay has been low for 2 consecutive Tupdate periods
	 * 2. Calculated drop probability is zero
	 * 3. If average dq_rate_estimator is enabled, we have atleast one
	 *    estimate for the avg_dq_rate ie., is a non-zero value
	 */
	if ((q->vars.qdelay < q->params.target / 2) &&
	    (q->vars.qdelay_old < q->params.target / 2) &&
	    q->vars.prob == 0 &&
	    (!q->params.dq_rate_estimator || q->vars.avg_dq_rate > 0)) {
		pie_vars_init(&q->vars);
	}

	if (!q->params.dq_rate_estimator)
		q->vars.qdelay_old = qdelay;
}

static void pie_timer(struct timer_list *t)
{
	struct pie_sched_data *q = from_timer(q, t, adapt_timer);
	struct Qdisc *sch = q->sch;
	spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch));

	spin_lock(root_lock);
	calculate_probability(sch);

	/* reset the timer to fire after 'tupdate'. tupdate is in jiffies. */
	if (q->params.tupdate)
		mod_timer(&q->adapt_timer, jiffies + q->params.tupdate);
	spin_unlock(root_lock);
}

static int pie_init(struct Qdisc *sch, struct nlattr *opt,
		    struct netlink_ext_ack *extack)
{
	struct pie_sched_data *q = qdisc_priv(sch);

	pie_params_init(&q->params);
	pie_vars_init(&q->vars);
	sch->limit = q->params.limit;

	q->sch = sch;
	timer_setup(&q->adapt_timer, pie_timer, 0);

	if (opt) {
		int err = pie_change(sch, opt, extack);

		if (err)
			return err;
	}

	mod_timer(&q->adapt_timer, jiffies + HZ / 2);
	return 0;
}

static int pie_dump(struct Qdisc *sch, struct sk_buff *skb)
{
	struct pie_sched_data *q = qdisc_priv(sch);
	struct nlattr *opts;

	opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
	if (!opts)
		goto nla_put_failure;

	/* convert target from pschedtime to us */
	if (nla_put_u32(skb, TCA_PIE_TARGET,
			((u32)PSCHED_TICKS2NS(q->params.target)) /
			NSEC_PER_USEC) ||
	    nla_put_u32(skb, TCA_PIE_LIMIT, sch->limit) ||
	    nla_put_u32(skb, TCA_PIE_TUPDATE,
			jiffies_to_usecs(q->params.tupdate)) ||
	    nla_put_u32(skb, TCA_PIE_ALPHA, q->params.alpha) ||
	    nla_put_u32(skb, TCA_PIE_BETA, q->params.beta) ||
	    nla_put_u32(skb, TCA_PIE_ECN, q->params.ecn) ||
	    nla_put_u32(skb, TCA_PIE_BYTEMODE, q->params.bytemode) ||
	    nla_put_u32(skb, TCA_PIE_DQ_RATE_ESTIMATOR,
			q->params.dq_rate_estimator))
		goto nla_put_failure;

	return nla_nest_end(skb, opts);

nla_put_failure:
	nla_nest_cancel(skb, opts);
	return -1;
}

static int pie_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
{
	struct pie_sched_data *q = qdisc_priv(sch);
	struct tc_pie_xstats st = {
		.prob		= q->vars.prob,
		.delay		= ((u32)PSCHED_TICKS2NS(q->vars.qdelay)) /
				   NSEC_PER_USEC,
		.packets_in	= q->stats.packets_in,
		.overlimit	= q->stats.overlimit,
		.maxq		= q->stats.maxq,
		.dropped	= q->stats.dropped,
		.ecn_mark	= q->stats.ecn_mark,
	};

	/* avg_dq_rate is only valid if dq_rate_estimator is enabled */
	st.dq_rate_estimating = q->params.dq_rate_estimator;

	/* unscale and return dq_rate in bytes per sec */
	if (q->params.dq_rate_estimator)
		st.avg_dq_rate = q->vars.avg_dq_rate *
				 (PSCHED_TICKS_PER_SEC) >> PIE_SCALE;

	return gnet_stats_copy_app(d, &st, sizeof(st));
}

static struct sk_buff *pie_qdisc_dequeue(struct Qdisc *sch)
{
	struct sk_buff *skb = qdisc_dequeue_head(sch);

	if (!skb)
		return NULL;

	pie_process_dequeue(sch, skb);
	return skb;
}

static void pie_reset(struct Qdisc *sch)
{
	struct pie_sched_data *q = qdisc_priv(sch);

	qdisc_reset_queue(sch);
	pie_vars_init(&q->vars);
}

static void pie_destroy(struct Qdisc *sch)
{
	struct pie_sched_data *q = qdisc_priv(sch);

	q->params.tupdate = 0;
	del_timer_sync(&q->adapt_timer);
}

static struct Qdisc_ops pie_qdisc_ops __read_mostly = {
	.id = "pie",
	.priv_size	= sizeof(struct pie_sched_data),
	.enqueue	= pie_qdisc_enqueue,
	.dequeue	= pie_qdisc_dequeue,
	.peek		= qdisc_peek_dequeued,
	.init		= pie_init,
	.destroy	= pie_destroy,
	.reset		= pie_reset,
	.change		= pie_change,
	.dump		= pie_dump,
	.dump_stats	= pie_dump_stats,
	.owner		= THIS_MODULE,
};

static int __init pie_module_init(void)
{
	return register_qdisc(&pie_qdisc_ops);
}

static void __exit pie_module_exit(void)
{
	unregister_qdisc(&pie_qdisc_ops);
}

module_init(pie_module_init);
module_exit(pie_module_exit);

MODULE_DESCRIPTION("Proportional Integral controller Enhanced (PIE) scheduler");
MODULE_AUTHOR("Vijay Subramanian");
MODULE_AUTHOR("Mythili Prabhu");
MODULE_LICENSE("GPL");