xref: /openbmc/linux/net/sched/sch_netem.c (revision 965f22bc)
1 /*
2  * net/sched/sch_netem.c	Network emulator
3  *
4  * 		This program is free software; you can redistribute it and/or
5  * 		modify it under the terms of the GNU General Public License
6  * 		as published by the Free Software Foundation; either version
7  * 		2 of the License.
8  *
9  *  		Many of the algorithms and ideas for this came from
10  *		NIST Net which is not copyrighted.
11  *
12  * Authors:	Stephen Hemminger <shemminger@osdl.org>
13  *		Catalin(ux aka Dino) BOIE <catab at umbrella dot ro>
14  */
15 
16 #include <linux/mm.h>
17 #include <linux/module.h>
18 #include <linux/slab.h>
19 #include <linux/types.h>
20 #include <linux/kernel.h>
21 #include <linux/errno.h>
22 #include <linux/skbuff.h>
23 #include <linux/vmalloc.h>
24 #include <linux/rtnetlink.h>
25 #include <linux/reciprocal_div.h>
26 #include <linux/rbtree.h>
27 
28 #include <net/netlink.h>
29 #include <net/pkt_sched.h>
30 #include <net/inet_ecn.h>
31 
32 #define VERSION "1.3"
33 
34 /*	Network Emulation Queuing algorithm.
35 	====================================
36 
37 	Sources: [1] Mark Carson, Darrin Santay, "NIST Net - A Linux-based
38 		 Network Emulation Tool
39 		 [2] Luigi Rizzo, DummyNet for FreeBSD
40 
41 	 ----------------------------------------------------------------
42 
43 	 This started out as a simple way to delay outgoing packets to
44 	 test TCP but has grown to include most of the functionality
45 	 of a full blown network emulator like NISTnet. It can delay
46 	 packets and add random jitter (and correlation). The random
47 	 distribution can be loaded from a table as well to provide
48 	 normal, Pareto, or experimental curves. Packet loss,
49 	 duplication, and reordering can also be emulated.
50 
51 	 This qdisc does not do classification that can be handled in
52 	 layering other disciplines.  It does not need to do bandwidth
53 	 control either since that can be handled by using token
54 	 bucket or other rate control.
55 
56      Correlated Loss Generator models
57 
58 	Added generation of correlated loss according to the
59 	"Gilbert-Elliot" model, a 4-state markov model.
60 
61 	References:
62 	[1] NetemCLG Home http://netgroup.uniroma2.it/NetemCLG
63 	[2] S. Salsano, F. Ludovici, A. Ordine, "Definition of a general
64 	and intuitive loss model for packet networks and its implementation
65 	in the Netem module in the Linux kernel", available in [1]
66 
67 	Authors: Stefano Salsano <stefano.salsano at uniroma2.it
68 		 Fabio Ludovici <fabio.ludovici at yahoo.it>
69 */
70 
71 struct disttable {
72 	u32  size;
73 	s16 table[0];
74 };
75 
76 struct netem_sched_data {
77 	/* internal t(ime)fifo qdisc uses t_root and sch->limit */
78 	struct rb_root t_root;
79 
80 	/* optional qdisc for classful handling (NULL at netem init) */
81 	struct Qdisc	*qdisc;
82 
83 	struct qdisc_watchdog watchdog;
84 
85 	s64 latency;
86 	s64 jitter;
87 
88 	u32 loss;
89 	u32 ecn;
90 	u32 limit;
91 	u32 counter;
92 	u32 gap;
93 	u32 duplicate;
94 	u32 reorder;
95 	u32 corrupt;
96 	u64 rate;
97 	s32 packet_overhead;
98 	u32 cell_size;
99 	struct reciprocal_value cell_size_reciprocal;
100 	s32 cell_overhead;
101 
102 	struct crndstate {
103 		u32 last;
104 		u32 rho;
105 	} delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor;
106 
107 	struct disttable *delay_dist;
108 
109 	enum  {
110 		CLG_RANDOM,
111 		CLG_4_STATES,
112 		CLG_GILB_ELL,
113 	} loss_model;
114 
115 	enum {
116 		TX_IN_GAP_PERIOD = 1,
117 		TX_IN_BURST_PERIOD,
118 		LOST_IN_GAP_PERIOD,
119 		LOST_IN_BURST_PERIOD,
120 	} _4_state_model;
121 
122 	enum {
123 		GOOD_STATE = 1,
124 		BAD_STATE,
125 	} GE_state_model;
126 
127 	/* Correlated Loss Generation models */
128 	struct clgstate {
129 		/* state of the Markov chain */
130 		u8 state;
131 
132 		/* 4-states and Gilbert-Elliot models */
133 		u32 a1;	/* p13 for 4-states or p for GE */
134 		u32 a2;	/* p31 for 4-states or r for GE */
135 		u32 a3;	/* p32 for 4-states or h for GE */
136 		u32 a4;	/* p14 for 4-states or 1-k for GE */
137 		u32 a5; /* p23 used only in 4-states */
138 	} clg;
139 
140 	struct tc_netem_slot slot_config;
141 	struct slotstate {
142 		u64 slot_next;
143 		s32 packets_left;
144 		s32 bytes_left;
145 	} slot;
146 
147 	struct disttable *slot_dist;
148 };
149 
150 /* Time stamp put into socket buffer control block
151  * Only valid when skbs are in our internal t(ime)fifo queue.
152  *
153  * As skb->rbnode uses same storage than skb->next, skb->prev and skb->tstamp,
154  * and skb->next & skb->prev are scratch space for a qdisc,
155  * we save skb->tstamp value in skb->cb[] before destroying it.
156  */
157 struct netem_skb_cb {
158 	u64	        time_to_send;
159 };
160 
161 static inline struct netem_skb_cb *netem_skb_cb(struct sk_buff *skb)
162 {
163 	/* we assume we can use skb next/prev/tstamp as storage for rb_node */
164 	qdisc_cb_private_validate(skb, sizeof(struct netem_skb_cb));
165 	return (struct netem_skb_cb *)qdisc_skb_cb(skb)->data;
166 }
167 
168 /* init_crandom - initialize correlated random number generator
169  * Use entropy source for initial seed.
170  */
171 static void init_crandom(struct crndstate *state, unsigned long rho)
172 {
173 	state->rho = rho;
174 	state->last = prandom_u32();
175 }
176 
177 /* get_crandom - correlated random number generator
178  * Next number depends on last value.
179  * rho is scaled to avoid floating point.
180  */
181 static u32 get_crandom(struct crndstate *state)
182 {
183 	u64 value, rho;
184 	unsigned long answer;
185 
186 	if (!state || state->rho == 0)	/* no correlation */
187 		return prandom_u32();
188 
189 	value = prandom_u32();
190 	rho = (u64)state->rho + 1;
191 	answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32;
192 	state->last = answer;
193 	return answer;
194 }
195 
196 /* loss_4state - 4-state model loss generator
197  * Generates losses according to the 4-state Markov chain adopted in
198  * the GI (General and Intuitive) loss model.
199  */
200 static bool loss_4state(struct netem_sched_data *q)
201 {
202 	struct clgstate *clg = &q->clg;
203 	u32 rnd = prandom_u32();
204 
205 	/*
206 	 * Makes a comparison between rnd and the transition
207 	 * probabilities outgoing from the current state, then decides the
208 	 * next state and if the next packet has to be transmitted or lost.
209 	 * The four states correspond to:
210 	 *   TX_IN_GAP_PERIOD => successfully transmitted packets within a gap period
211 	 *   LOST_IN_BURST_PERIOD => isolated losses within a gap period
212 	 *   LOST_IN_GAP_PERIOD => lost packets within a burst period
213 	 *   TX_IN_GAP_PERIOD => successfully transmitted packets within a burst period
214 	 */
215 	switch (clg->state) {
216 	case TX_IN_GAP_PERIOD:
217 		if (rnd < clg->a4) {
218 			clg->state = LOST_IN_BURST_PERIOD;
219 			return true;
220 		} else if (clg->a4 < rnd && rnd < clg->a1 + clg->a4) {
221 			clg->state = LOST_IN_GAP_PERIOD;
222 			return true;
223 		} else if (clg->a1 + clg->a4 < rnd) {
224 			clg->state = TX_IN_GAP_PERIOD;
225 		}
226 
227 		break;
228 	case TX_IN_BURST_PERIOD:
229 		if (rnd < clg->a5) {
230 			clg->state = LOST_IN_GAP_PERIOD;
231 			return true;
232 		} else {
233 			clg->state = TX_IN_BURST_PERIOD;
234 		}
235 
236 		break;
237 	case LOST_IN_GAP_PERIOD:
238 		if (rnd < clg->a3)
239 			clg->state = TX_IN_BURST_PERIOD;
240 		else if (clg->a3 < rnd && rnd < clg->a2 + clg->a3) {
241 			clg->state = TX_IN_GAP_PERIOD;
242 		} else if (clg->a2 + clg->a3 < rnd) {
243 			clg->state = LOST_IN_GAP_PERIOD;
244 			return true;
245 		}
246 		break;
247 	case LOST_IN_BURST_PERIOD:
248 		clg->state = TX_IN_GAP_PERIOD;
249 		break;
250 	}
251 
252 	return false;
253 }
254 
255 /* loss_gilb_ell - Gilbert-Elliot model loss generator
256  * Generates losses according to the Gilbert-Elliot loss model or
257  * its special cases  (Gilbert or Simple Gilbert)
258  *
259  * Makes a comparison between random number and the transition
260  * probabilities outgoing from the current state, then decides the
261  * next state. A second random number is extracted and the comparison
262  * with the loss probability of the current state decides if the next
263  * packet will be transmitted or lost.
264  */
265 static bool loss_gilb_ell(struct netem_sched_data *q)
266 {
267 	struct clgstate *clg = &q->clg;
268 
269 	switch (clg->state) {
270 	case GOOD_STATE:
271 		if (prandom_u32() < clg->a1)
272 			clg->state = BAD_STATE;
273 		if (prandom_u32() < clg->a4)
274 			return true;
275 		break;
276 	case BAD_STATE:
277 		if (prandom_u32() < clg->a2)
278 			clg->state = GOOD_STATE;
279 		if (prandom_u32() > clg->a3)
280 			return true;
281 	}
282 
283 	return false;
284 }
285 
286 static bool loss_event(struct netem_sched_data *q)
287 {
288 	switch (q->loss_model) {
289 	case CLG_RANDOM:
290 		/* Random packet drop 0 => none, ~0 => all */
291 		return q->loss && q->loss >= get_crandom(&q->loss_cor);
292 
293 	case CLG_4_STATES:
294 		/* 4state loss model algorithm (used also for GI model)
295 		* Extracts a value from the markov 4 state loss generator,
296 		* if it is 1 drops a packet and if needed writes the event in
297 		* the kernel logs
298 		*/
299 		return loss_4state(q);
300 
301 	case CLG_GILB_ELL:
302 		/* Gilbert-Elliot loss model algorithm
303 		* Extracts a value from the Gilbert-Elliot loss generator,
304 		* if it is 1 drops a packet and if needed writes the event in
305 		* the kernel logs
306 		*/
307 		return loss_gilb_ell(q);
308 	}
309 
310 	return false;	/* not reached */
311 }
312 
313 
314 /* tabledist - return a pseudo-randomly distributed value with mean mu and
315  * std deviation sigma.  Uses table lookup to approximate the desired
316  * distribution, and a uniformly-distributed pseudo-random source.
317  */
318 static s64 tabledist(s64 mu, s32 sigma,
319 		     struct crndstate *state,
320 		     const struct disttable *dist)
321 {
322 	s64 x;
323 	long t;
324 	u32 rnd;
325 
326 	if (sigma == 0)
327 		return mu;
328 
329 	rnd = get_crandom(state);
330 
331 	/* default uniform distribution */
332 	if (dist == NULL)
333 		return ((rnd % (2 * sigma)) + mu) - sigma;
334 
335 	t = dist->table[rnd % dist->size];
336 	x = (sigma % NETEM_DIST_SCALE) * t;
337 	if (x >= 0)
338 		x += NETEM_DIST_SCALE/2;
339 	else
340 		x -= NETEM_DIST_SCALE/2;
341 
342 	return  x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu;
343 }
344 
345 static u64 packet_time_ns(u64 len, const struct netem_sched_data *q)
346 {
347 	len += q->packet_overhead;
348 
349 	if (q->cell_size) {
350 		u32 cells = reciprocal_divide(len, q->cell_size_reciprocal);
351 
352 		if (len > cells * q->cell_size)	/* extra cell needed for remainder */
353 			cells++;
354 		len = cells * (q->cell_size + q->cell_overhead);
355 	}
356 
357 	return div64_u64(len * NSEC_PER_SEC, q->rate);
358 }
359 
360 static void tfifo_reset(struct Qdisc *sch)
361 {
362 	struct netem_sched_data *q = qdisc_priv(sch);
363 	struct rb_node *p = rb_first(&q->t_root);
364 
365 	while (p) {
366 		struct sk_buff *skb = rb_to_skb(p);
367 
368 		p = rb_next(p);
369 		rb_erase(&skb->rbnode, &q->t_root);
370 		rtnl_kfree_skbs(skb, skb);
371 	}
372 }
373 
374 static void tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch)
375 {
376 	struct netem_sched_data *q = qdisc_priv(sch);
377 	u64 tnext = netem_skb_cb(nskb)->time_to_send;
378 	struct rb_node **p = &q->t_root.rb_node, *parent = NULL;
379 
380 	while (*p) {
381 		struct sk_buff *skb;
382 
383 		parent = *p;
384 		skb = rb_to_skb(parent);
385 		if (tnext >= netem_skb_cb(skb)->time_to_send)
386 			p = &parent->rb_right;
387 		else
388 			p = &parent->rb_left;
389 	}
390 	rb_link_node(&nskb->rbnode, parent, p);
391 	rb_insert_color(&nskb->rbnode, &q->t_root);
392 	sch->q.qlen++;
393 }
394 
395 /* netem can't properly corrupt a megapacket (like we get from GSO), so instead
396  * when we statistically choose to corrupt one, we instead segment it, returning
397  * the first packet to be corrupted, and re-enqueue the remaining frames
398  */
399 static struct sk_buff *netem_segment(struct sk_buff *skb, struct Qdisc *sch,
400 				     struct sk_buff **to_free)
401 {
402 	struct sk_buff *segs;
403 	netdev_features_t features = netif_skb_features(skb);
404 
405 	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
406 
407 	if (IS_ERR_OR_NULL(segs)) {
408 		qdisc_drop(skb, sch, to_free);
409 		return NULL;
410 	}
411 	consume_skb(skb);
412 	return segs;
413 }
414 
415 static void netem_enqueue_skb_head(struct qdisc_skb_head *qh, struct sk_buff *skb)
416 {
417 	skb->next = qh->head;
418 
419 	if (!qh->head)
420 		qh->tail = skb;
421 	qh->head = skb;
422 	qh->qlen++;
423 }
424 
425 /*
426  * Insert one skb into qdisc.
427  * Note: parent depends on return value to account for queue length.
428  * 	NET_XMIT_DROP: queue length didn't change.
429  *      NET_XMIT_SUCCESS: one skb was queued.
430  */
431 static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch,
432 			 struct sk_buff **to_free)
433 {
434 	struct netem_sched_data *q = qdisc_priv(sch);
435 	/* We don't fill cb now as skb_unshare() may invalidate it */
436 	struct netem_skb_cb *cb;
437 	struct sk_buff *skb2;
438 	struct sk_buff *segs = NULL;
439 	unsigned int len = 0, last_len, prev_len = qdisc_pkt_len(skb);
440 	int nb = 0;
441 	int count = 1;
442 	int rc = NET_XMIT_SUCCESS;
443 
444 	/* Random duplication */
445 	if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor))
446 		++count;
447 
448 	/* Drop packet? */
449 	if (loss_event(q)) {
450 		if (q->ecn && INET_ECN_set_ce(skb))
451 			qdisc_qstats_drop(sch); /* mark packet */
452 		else
453 			--count;
454 	}
455 	if (count == 0) {
456 		qdisc_qstats_drop(sch);
457 		__qdisc_drop(skb, to_free);
458 		return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
459 	}
460 
461 	/* If a delay is expected, orphan the skb. (orphaning usually takes
462 	 * place at TX completion time, so _before_ the link transit delay)
463 	 */
464 	if (q->latency || q->jitter || q->rate)
465 		skb_orphan_partial(skb);
466 
467 	/*
468 	 * If we need to duplicate packet, then re-insert at top of the
469 	 * qdisc tree, since parent queuer expects that only one
470 	 * skb will be queued.
471 	 */
472 	if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) {
473 		struct Qdisc *rootq = qdisc_root(sch);
474 		u32 dupsave = q->duplicate; /* prevent duplicating a dup... */
475 
476 		q->duplicate = 0;
477 		rootq->enqueue(skb2, rootq, to_free);
478 		q->duplicate = dupsave;
479 	}
480 
481 	/*
482 	 * Randomized packet corruption.
483 	 * Make copy if needed since we are modifying
484 	 * If packet is going to be hardware checksummed, then
485 	 * do it now in software before we mangle it.
486 	 */
487 	if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) {
488 		if (skb_is_gso(skb)) {
489 			segs = netem_segment(skb, sch, to_free);
490 			if (!segs)
491 				return NET_XMIT_DROP;
492 		} else {
493 			segs = skb;
494 		}
495 
496 		skb = segs;
497 		segs = segs->next;
498 
499 		skb = skb_unshare(skb, GFP_ATOMIC);
500 		if (unlikely(!skb)) {
501 			qdisc_qstats_drop(sch);
502 			goto finish_segs;
503 		}
504 		if (skb->ip_summed == CHECKSUM_PARTIAL &&
505 		    skb_checksum_help(skb)) {
506 			qdisc_drop(skb, sch, to_free);
507 			goto finish_segs;
508 		}
509 
510 		skb->data[prandom_u32() % skb_headlen(skb)] ^=
511 			1<<(prandom_u32() % 8);
512 	}
513 
514 	if (unlikely(sch->q.qlen >= sch->limit))
515 		return qdisc_drop_all(skb, sch, to_free);
516 
517 	qdisc_qstats_backlog_inc(sch, skb);
518 
519 	cb = netem_skb_cb(skb);
520 	if (q->gap == 0 ||		/* not doing reordering */
521 	    q->counter < q->gap - 1 ||	/* inside last reordering gap */
522 	    q->reorder < get_crandom(&q->reorder_cor)) {
523 		u64 now;
524 		s64 delay;
525 
526 		delay = tabledist(q->latency, q->jitter,
527 				  &q->delay_cor, q->delay_dist);
528 
529 		now = ktime_get_ns();
530 
531 		if (q->rate) {
532 			struct netem_skb_cb *last = NULL;
533 
534 			if (sch->q.tail)
535 				last = netem_skb_cb(sch->q.tail);
536 			if (q->t_root.rb_node) {
537 				struct sk_buff *t_skb;
538 				struct netem_skb_cb *t_last;
539 
540 				t_skb = skb_rb_last(&q->t_root);
541 				t_last = netem_skb_cb(t_skb);
542 				if (!last ||
543 				    t_last->time_to_send > last->time_to_send) {
544 					last = t_last;
545 				}
546 			}
547 
548 			if (last) {
549 				/*
550 				 * Last packet in queue is reference point (now),
551 				 * calculate this time bonus and subtract
552 				 * from delay.
553 				 */
554 				delay -= last->time_to_send - now;
555 				delay = max_t(s64, 0, delay);
556 				now = last->time_to_send;
557 			}
558 
559 			delay += packet_time_ns(qdisc_pkt_len(skb), q);
560 		}
561 
562 		cb->time_to_send = now + delay;
563 		++q->counter;
564 		tfifo_enqueue(skb, sch);
565 	} else {
566 		/*
567 		 * Do re-ordering by putting one out of N packets at the front
568 		 * of the queue.
569 		 */
570 		cb->time_to_send = ktime_get_ns();
571 		q->counter = 0;
572 
573 		netem_enqueue_skb_head(&sch->q, skb);
574 		sch->qstats.requeues++;
575 	}
576 
577 finish_segs:
578 	if (segs) {
579 		while (segs) {
580 			skb2 = segs->next;
581 			segs->next = NULL;
582 			qdisc_skb_cb(segs)->pkt_len = segs->len;
583 			last_len = segs->len;
584 			rc = qdisc_enqueue(segs, sch, to_free);
585 			if (rc != NET_XMIT_SUCCESS) {
586 				if (net_xmit_drop_count(rc))
587 					qdisc_qstats_drop(sch);
588 			} else {
589 				nb++;
590 				len += last_len;
591 			}
592 			segs = skb2;
593 		}
594 		sch->q.qlen += nb;
595 		if (nb > 1)
596 			qdisc_tree_reduce_backlog(sch, 1 - nb, prev_len - len);
597 	}
598 	return NET_XMIT_SUCCESS;
599 }
600 
601 /* Delay the next round with a new future slot with a
602  * correct number of bytes and packets.
603  */
604 
605 static void get_slot_next(struct netem_sched_data *q, u64 now)
606 {
607 	s64 next_delay;
608 
609 	if (!q->slot_dist)
610 		next_delay = q->slot_config.min_delay +
611 				(prandom_u32() *
612 				 (q->slot_config.max_delay -
613 				  q->slot_config.min_delay) >> 32);
614 	else
615 		next_delay = tabledist(q->slot_config.dist_delay,
616 				       (s32)(q->slot_config.dist_jitter),
617 				       NULL, q->slot_dist);
618 
619 	q->slot.slot_next = now + next_delay;
620 	q->slot.packets_left = q->slot_config.max_packets;
621 	q->slot.bytes_left = q->slot_config.max_bytes;
622 }
623 
624 static struct sk_buff *netem_dequeue(struct Qdisc *sch)
625 {
626 	struct netem_sched_data *q = qdisc_priv(sch);
627 	struct sk_buff *skb;
628 	struct rb_node *p;
629 
630 tfifo_dequeue:
631 	skb = __qdisc_dequeue_head(&sch->q);
632 	if (skb) {
633 		qdisc_qstats_backlog_dec(sch, skb);
634 deliver:
635 		qdisc_bstats_update(sch, skb);
636 		return skb;
637 	}
638 	p = rb_first(&q->t_root);
639 	if (p) {
640 		u64 time_to_send;
641 		u64 now = ktime_get_ns();
642 
643 		skb = rb_to_skb(p);
644 
645 		/* if more time remaining? */
646 		time_to_send = netem_skb_cb(skb)->time_to_send;
647 		if (q->slot.slot_next && q->slot.slot_next < time_to_send)
648 			get_slot_next(q, now);
649 
650 		if (time_to_send <= now &&  q->slot.slot_next <= now) {
651 			rb_erase(p, &q->t_root);
652 			sch->q.qlen--;
653 			qdisc_qstats_backlog_dec(sch, skb);
654 			skb->next = NULL;
655 			skb->prev = NULL;
656 			/* skb->dev shares skb->rbnode area,
657 			 * we need to restore its value.
658 			 */
659 			skb->dev = qdisc_dev(sch);
660 
661 #ifdef CONFIG_NET_CLS_ACT
662 			/*
663 			 * If it's at ingress let's pretend the delay is
664 			 * from the network (tstamp will be updated).
665 			 */
666 			if (skb->tc_redirected && skb->tc_from_ingress)
667 				skb->tstamp = 0;
668 #endif
669 
670 			if (q->slot.slot_next) {
671 				q->slot.packets_left--;
672 				q->slot.bytes_left -= qdisc_pkt_len(skb);
673 				if (q->slot.packets_left <= 0 ||
674 				    q->slot.bytes_left <= 0)
675 					get_slot_next(q, now);
676 			}
677 
678 			if (q->qdisc) {
679 				unsigned int pkt_len = qdisc_pkt_len(skb);
680 				struct sk_buff *to_free = NULL;
681 				int err;
682 
683 				err = qdisc_enqueue(skb, q->qdisc, &to_free);
684 				kfree_skb_list(to_free);
685 				if (err != NET_XMIT_SUCCESS &&
686 				    net_xmit_drop_count(err)) {
687 					qdisc_qstats_drop(sch);
688 					qdisc_tree_reduce_backlog(sch, 1,
689 								  pkt_len);
690 				}
691 				goto tfifo_dequeue;
692 			}
693 			goto deliver;
694 		}
695 
696 		if (q->qdisc) {
697 			skb = q->qdisc->ops->dequeue(q->qdisc);
698 			if (skb)
699 				goto deliver;
700 		}
701 
702 		qdisc_watchdog_schedule_ns(&q->watchdog,
703 					   max(time_to_send,
704 					       q->slot.slot_next));
705 	}
706 
707 	if (q->qdisc) {
708 		skb = q->qdisc->ops->dequeue(q->qdisc);
709 		if (skb)
710 			goto deliver;
711 	}
712 	return NULL;
713 }
714 
715 static void netem_reset(struct Qdisc *sch)
716 {
717 	struct netem_sched_data *q = qdisc_priv(sch);
718 
719 	qdisc_reset_queue(sch);
720 	tfifo_reset(sch);
721 	if (q->qdisc)
722 		qdisc_reset(q->qdisc);
723 	qdisc_watchdog_cancel(&q->watchdog);
724 }
725 
726 static void dist_free(struct disttable *d)
727 {
728 	kvfree(d);
729 }
730 
731 /*
732  * Distribution data is a variable size payload containing
733  * signed 16 bit values.
734  */
735 
736 static int get_dist_table(struct Qdisc *sch, struct disttable **tbl,
737 			  const struct nlattr *attr)
738 {
739 	size_t n = nla_len(attr)/sizeof(__s16);
740 	const __s16 *data = nla_data(attr);
741 	spinlock_t *root_lock;
742 	struct disttable *d;
743 	int i;
744 
745 	if (n > NETEM_DIST_MAX)
746 		return -EINVAL;
747 
748 	d = kvmalloc(sizeof(struct disttable) + n * sizeof(s16), GFP_KERNEL);
749 	if (!d)
750 		return -ENOMEM;
751 
752 	d->size = n;
753 	for (i = 0; i < n; i++)
754 		d->table[i] = data[i];
755 
756 	root_lock = qdisc_root_sleeping_lock(sch);
757 
758 	spin_lock_bh(root_lock);
759 	swap(*tbl, d);
760 	spin_unlock_bh(root_lock);
761 
762 	dist_free(d);
763 	return 0;
764 }
765 
766 static void get_slot(struct netem_sched_data *q, const struct nlattr *attr)
767 {
768 	const struct tc_netem_slot *c = nla_data(attr);
769 
770 	q->slot_config = *c;
771 	if (q->slot_config.max_packets == 0)
772 		q->slot_config.max_packets = INT_MAX;
773 	if (q->slot_config.max_bytes == 0)
774 		q->slot_config.max_bytes = INT_MAX;
775 	q->slot.packets_left = q->slot_config.max_packets;
776 	q->slot.bytes_left = q->slot_config.max_bytes;
777 	if (q->slot_config.min_delay | q->slot_config.max_delay |
778 	    q->slot_config.dist_jitter)
779 		q->slot.slot_next = ktime_get_ns();
780 	else
781 		q->slot.slot_next = 0;
782 }
783 
784 static void get_correlation(struct netem_sched_data *q, const struct nlattr *attr)
785 {
786 	const struct tc_netem_corr *c = nla_data(attr);
787 
788 	init_crandom(&q->delay_cor, c->delay_corr);
789 	init_crandom(&q->loss_cor, c->loss_corr);
790 	init_crandom(&q->dup_cor, c->dup_corr);
791 }
792 
793 static void get_reorder(struct netem_sched_data *q, const struct nlattr *attr)
794 {
795 	const struct tc_netem_reorder *r = nla_data(attr);
796 
797 	q->reorder = r->probability;
798 	init_crandom(&q->reorder_cor, r->correlation);
799 }
800 
801 static void get_corrupt(struct netem_sched_data *q, const struct nlattr *attr)
802 {
803 	const struct tc_netem_corrupt *r = nla_data(attr);
804 
805 	q->corrupt = r->probability;
806 	init_crandom(&q->corrupt_cor, r->correlation);
807 }
808 
809 static void get_rate(struct netem_sched_data *q, const struct nlattr *attr)
810 {
811 	const struct tc_netem_rate *r = nla_data(attr);
812 
813 	q->rate = r->rate;
814 	q->packet_overhead = r->packet_overhead;
815 	q->cell_size = r->cell_size;
816 	q->cell_overhead = r->cell_overhead;
817 	if (q->cell_size)
818 		q->cell_size_reciprocal = reciprocal_value(q->cell_size);
819 	else
820 		q->cell_size_reciprocal = (struct reciprocal_value) { 0 };
821 }
822 
823 static int get_loss_clg(struct netem_sched_data *q, const struct nlattr *attr)
824 {
825 	const struct nlattr *la;
826 	int rem;
827 
828 	nla_for_each_nested(la, attr, rem) {
829 		u16 type = nla_type(la);
830 
831 		switch (type) {
832 		case NETEM_LOSS_GI: {
833 			const struct tc_netem_gimodel *gi = nla_data(la);
834 
835 			if (nla_len(la) < sizeof(struct tc_netem_gimodel)) {
836 				pr_info("netem: incorrect gi model size\n");
837 				return -EINVAL;
838 			}
839 
840 			q->loss_model = CLG_4_STATES;
841 
842 			q->clg.state = TX_IN_GAP_PERIOD;
843 			q->clg.a1 = gi->p13;
844 			q->clg.a2 = gi->p31;
845 			q->clg.a3 = gi->p32;
846 			q->clg.a4 = gi->p14;
847 			q->clg.a5 = gi->p23;
848 			break;
849 		}
850 
851 		case NETEM_LOSS_GE: {
852 			const struct tc_netem_gemodel *ge = nla_data(la);
853 
854 			if (nla_len(la) < sizeof(struct tc_netem_gemodel)) {
855 				pr_info("netem: incorrect ge model size\n");
856 				return -EINVAL;
857 			}
858 
859 			q->loss_model = CLG_GILB_ELL;
860 			q->clg.state = GOOD_STATE;
861 			q->clg.a1 = ge->p;
862 			q->clg.a2 = ge->r;
863 			q->clg.a3 = ge->h;
864 			q->clg.a4 = ge->k1;
865 			break;
866 		}
867 
868 		default:
869 			pr_info("netem: unknown loss type %u\n", type);
870 			return -EINVAL;
871 		}
872 	}
873 
874 	return 0;
875 }
876 
877 static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = {
878 	[TCA_NETEM_CORR]	= { .len = sizeof(struct tc_netem_corr) },
879 	[TCA_NETEM_REORDER]	= { .len = sizeof(struct tc_netem_reorder) },
880 	[TCA_NETEM_CORRUPT]	= { .len = sizeof(struct tc_netem_corrupt) },
881 	[TCA_NETEM_RATE]	= { .len = sizeof(struct tc_netem_rate) },
882 	[TCA_NETEM_LOSS]	= { .type = NLA_NESTED },
883 	[TCA_NETEM_ECN]		= { .type = NLA_U32 },
884 	[TCA_NETEM_RATE64]	= { .type = NLA_U64 },
885 	[TCA_NETEM_LATENCY64]	= { .type = NLA_S64 },
886 	[TCA_NETEM_JITTER64]	= { .type = NLA_S64 },
887 	[TCA_NETEM_SLOT]	= { .len = sizeof(struct tc_netem_slot) },
888 };
889 
890 static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla,
891 		      const struct nla_policy *policy, int len)
892 {
893 	int nested_len = nla_len(nla) - NLA_ALIGN(len);
894 
895 	if (nested_len < 0) {
896 		pr_info("netem: invalid attributes len %d\n", nested_len);
897 		return -EINVAL;
898 	}
899 
900 	if (nested_len >= nla_attr_size(0))
901 		return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len),
902 				 nested_len, policy, NULL);
903 
904 	memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1));
905 	return 0;
906 }
907 
908 /* Parse netlink message to set options */
909 static int netem_change(struct Qdisc *sch, struct nlattr *opt,
910 			struct netlink_ext_ack *extack)
911 {
912 	struct netem_sched_data *q = qdisc_priv(sch);
913 	struct nlattr *tb[TCA_NETEM_MAX + 1];
914 	struct tc_netem_qopt *qopt;
915 	struct clgstate old_clg;
916 	int old_loss_model = CLG_RANDOM;
917 	int ret;
918 
919 	if (opt == NULL)
920 		return -EINVAL;
921 
922 	qopt = nla_data(opt);
923 	ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt));
924 	if (ret < 0)
925 		return ret;
926 
927 	/* backup q->clg and q->loss_model */
928 	old_clg = q->clg;
929 	old_loss_model = q->loss_model;
930 
931 	if (tb[TCA_NETEM_LOSS]) {
932 		ret = get_loss_clg(q, tb[TCA_NETEM_LOSS]);
933 		if (ret) {
934 			q->loss_model = old_loss_model;
935 			return ret;
936 		}
937 	} else {
938 		q->loss_model = CLG_RANDOM;
939 	}
940 
941 	if (tb[TCA_NETEM_DELAY_DIST]) {
942 		ret = get_dist_table(sch, &q->delay_dist,
943 				     tb[TCA_NETEM_DELAY_DIST]);
944 		if (ret)
945 			goto get_table_failure;
946 	}
947 
948 	if (tb[TCA_NETEM_SLOT_DIST]) {
949 		ret = get_dist_table(sch, &q->slot_dist,
950 				     tb[TCA_NETEM_SLOT_DIST]);
951 		if (ret)
952 			goto get_table_failure;
953 	}
954 
955 	sch->limit = qopt->limit;
956 
957 	q->latency = PSCHED_TICKS2NS(qopt->latency);
958 	q->jitter = PSCHED_TICKS2NS(qopt->jitter);
959 	q->limit = qopt->limit;
960 	q->gap = qopt->gap;
961 	q->counter = 0;
962 	q->loss = qopt->loss;
963 	q->duplicate = qopt->duplicate;
964 
965 	/* for compatibility with earlier versions.
966 	 * if gap is set, need to assume 100% probability
967 	 */
968 	if (q->gap)
969 		q->reorder = ~0;
970 
971 	if (tb[TCA_NETEM_CORR])
972 		get_correlation(q, tb[TCA_NETEM_CORR]);
973 
974 	if (tb[TCA_NETEM_REORDER])
975 		get_reorder(q, tb[TCA_NETEM_REORDER]);
976 
977 	if (tb[TCA_NETEM_CORRUPT])
978 		get_corrupt(q, tb[TCA_NETEM_CORRUPT]);
979 
980 	if (tb[TCA_NETEM_RATE])
981 		get_rate(q, tb[TCA_NETEM_RATE]);
982 
983 	if (tb[TCA_NETEM_RATE64])
984 		q->rate = max_t(u64, q->rate,
985 				nla_get_u64(tb[TCA_NETEM_RATE64]));
986 
987 	if (tb[TCA_NETEM_LATENCY64])
988 		q->latency = nla_get_s64(tb[TCA_NETEM_LATENCY64]);
989 
990 	if (tb[TCA_NETEM_JITTER64])
991 		q->jitter = nla_get_s64(tb[TCA_NETEM_JITTER64]);
992 
993 	if (tb[TCA_NETEM_ECN])
994 		q->ecn = nla_get_u32(tb[TCA_NETEM_ECN]);
995 
996 	if (tb[TCA_NETEM_SLOT])
997 		get_slot(q, tb[TCA_NETEM_SLOT]);
998 
999 	return ret;
1000 
1001 get_table_failure:
1002 	/* recover clg and loss_model, in case of
1003 	 * q->clg and q->loss_model were modified
1004 	 * in get_loss_clg()
1005 	 */
1006 	q->clg = old_clg;
1007 	q->loss_model = old_loss_model;
1008 	return ret;
1009 }
1010 
1011 static int netem_init(struct Qdisc *sch, struct nlattr *opt,
1012 		      struct netlink_ext_ack *extack)
1013 {
1014 	struct netem_sched_data *q = qdisc_priv(sch);
1015 	int ret;
1016 
1017 	qdisc_watchdog_init(&q->watchdog, sch);
1018 
1019 	if (!opt)
1020 		return -EINVAL;
1021 
1022 	q->loss_model = CLG_RANDOM;
1023 	ret = netem_change(sch, opt, extack);
1024 	if (ret)
1025 		pr_info("netem: change failed\n");
1026 	return ret;
1027 }
1028 
1029 static void netem_destroy(struct Qdisc *sch)
1030 {
1031 	struct netem_sched_data *q = qdisc_priv(sch);
1032 
1033 	qdisc_watchdog_cancel(&q->watchdog);
1034 	if (q->qdisc)
1035 		qdisc_destroy(q->qdisc);
1036 	dist_free(q->delay_dist);
1037 	dist_free(q->slot_dist);
1038 }
1039 
1040 static int dump_loss_model(const struct netem_sched_data *q,
1041 			   struct sk_buff *skb)
1042 {
1043 	struct nlattr *nest;
1044 
1045 	nest = nla_nest_start(skb, TCA_NETEM_LOSS);
1046 	if (nest == NULL)
1047 		goto nla_put_failure;
1048 
1049 	switch (q->loss_model) {
1050 	case CLG_RANDOM:
1051 		/* legacy loss model */
1052 		nla_nest_cancel(skb, nest);
1053 		return 0;	/* no data */
1054 
1055 	case CLG_4_STATES: {
1056 		struct tc_netem_gimodel gi = {
1057 			.p13 = q->clg.a1,
1058 			.p31 = q->clg.a2,
1059 			.p32 = q->clg.a3,
1060 			.p14 = q->clg.a4,
1061 			.p23 = q->clg.a5,
1062 		};
1063 
1064 		if (nla_put(skb, NETEM_LOSS_GI, sizeof(gi), &gi))
1065 			goto nla_put_failure;
1066 		break;
1067 	}
1068 	case CLG_GILB_ELL: {
1069 		struct tc_netem_gemodel ge = {
1070 			.p = q->clg.a1,
1071 			.r = q->clg.a2,
1072 			.h = q->clg.a3,
1073 			.k1 = q->clg.a4,
1074 		};
1075 
1076 		if (nla_put(skb, NETEM_LOSS_GE, sizeof(ge), &ge))
1077 			goto nla_put_failure;
1078 		break;
1079 	}
1080 	}
1081 
1082 	nla_nest_end(skb, nest);
1083 	return 0;
1084 
1085 nla_put_failure:
1086 	nla_nest_cancel(skb, nest);
1087 	return -1;
1088 }
1089 
1090 static int netem_dump(struct Qdisc *sch, struct sk_buff *skb)
1091 {
1092 	const struct netem_sched_data *q = qdisc_priv(sch);
1093 	struct nlattr *nla = (struct nlattr *) skb_tail_pointer(skb);
1094 	struct tc_netem_qopt qopt;
1095 	struct tc_netem_corr cor;
1096 	struct tc_netem_reorder reorder;
1097 	struct tc_netem_corrupt corrupt;
1098 	struct tc_netem_rate rate;
1099 	struct tc_netem_slot slot;
1100 
1101 	qopt.latency = min_t(psched_tdiff_t, PSCHED_NS2TICKS(q->latency),
1102 			     UINT_MAX);
1103 	qopt.jitter = min_t(psched_tdiff_t, PSCHED_NS2TICKS(q->jitter),
1104 			    UINT_MAX);
1105 	qopt.limit = q->limit;
1106 	qopt.loss = q->loss;
1107 	qopt.gap = q->gap;
1108 	qopt.duplicate = q->duplicate;
1109 	if (nla_put(skb, TCA_OPTIONS, sizeof(qopt), &qopt))
1110 		goto nla_put_failure;
1111 
1112 	if (nla_put(skb, TCA_NETEM_LATENCY64, sizeof(q->latency), &q->latency))
1113 		goto nla_put_failure;
1114 
1115 	if (nla_put(skb, TCA_NETEM_JITTER64, sizeof(q->jitter), &q->jitter))
1116 		goto nla_put_failure;
1117 
1118 	cor.delay_corr = q->delay_cor.rho;
1119 	cor.loss_corr = q->loss_cor.rho;
1120 	cor.dup_corr = q->dup_cor.rho;
1121 	if (nla_put(skb, TCA_NETEM_CORR, sizeof(cor), &cor))
1122 		goto nla_put_failure;
1123 
1124 	reorder.probability = q->reorder;
1125 	reorder.correlation = q->reorder_cor.rho;
1126 	if (nla_put(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder))
1127 		goto nla_put_failure;
1128 
1129 	corrupt.probability = q->corrupt;
1130 	corrupt.correlation = q->corrupt_cor.rho;
1131 	if (nla_put(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt))
1132 		goto nla_put_failure;
1133 
1134 	if (q->rate >= (1ULL << 32)) {
1135 		if (nla_put_u64_64bit(skb, TCA_NETEM_RATE64, q->rate,
1136 				      TCA_NETEM_PAD))
1137 			goto nla_put_failure;
1138 		rate.rate = ~0U;
1139 	} else {
1140 		rate.rate = q->rate;
1141 	}
1142 	rate.packet_overhead = q->packet_overhead;
1143 	rate.cell_size = q->cell_size;
1144 	rate.cell_overhead = q->cell_overhead;
1145 	if (nla_put(skb, TCA_NETEM_RATE, sizeof(rate), &rate))
1146 		goto nla_put_failure;
1147 
1148 	if (q->ecn && nla_put_u32(skb, TCA_NETEM_ECN, q->ecn))
1149 		goto nla_put_failure;
1150 
1151 	if (dump_loss_model(q, skb) != 0)
1152 		goto nla_put_failure;
1153 
1154 	if (q->slot_config.min_delay | q->slot_config.max_delay |
1155 	    q->slot_config.dist_jitter) {
1156 		slot = q->slot_config;
1157 		if (slot.max_packets == INT_MAX)
1158 			slot.max_packets = 0;
1159 		if (slot.max_bytes == INT_MAX)
1160 			slot.max_bytes = 0;
1161 		if (nla_put(skb, TCA_NETEM_SLOT, sizeof(slot), &slot))
1162 			goto nla_put_failure;
1163 	}
1164 
1165 	return nla_nest_end(skb, nla);
1166 
1167 nla_put_failure:
1168 	nlmsg_trim(skb, nla);
1169 	return -1;
1170 }
1171 
1172 static int netem_dump_class(struct Qdisc *sch, unsigned long cl,
1173 			  struct sk_buff *skb, struct tcmsg *tcm)
1174 {
1175 	struct netem_sched_data *q = qdisc_priv(sch);
1176 
1177 	if (cl != 1 || !q->qdisc) 	/* only one class */
1178 		return -ENOENT;
1179 
1180 	tcm->tcm_handle |= TC_H_MIN(1);
1181 	tcm->tcm_info = q->qdisc->handle;
1182 
1183 	return 0;
1184 }
1185 
1186 static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1187 		     struct Qdisc **old, struct netlink_ext_ack *extack)
1188 {
1189 	struct netem_sched_data *q = qdisc_priv(sch);
1190 
1191 	*old = qdisc_replace(sch, new, &q->qdisc);
1192 	return 0;
1193 }
1194 
1195 static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg)
1196 {
1197 	struct netem_sched_data *q = qdisc_priv(sch);
1198 	return q->qdisc;
1199 }
1200 
1201 static unsigned long netem_find(struct Qdisc *sch, u32 classid)
1202 {
1203 	return 1;
1204 }
1205 
1206 static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker)
1207 {
1208 	if (!walker->stop) {
1209 		if (walker->count >= walker->skip)
1210 			if (walker->fn(sch, 1, walker) < 0) {
1211 				walker->stop = 1;
1212 				return;
1213 			}
1214 		walker->count++;
1215 	}
1216 }
1217 
1218 static const struct Qdisc_class_ops netem_class_ops = {
1219 	.graft		=	netem_graft,
1220 	.leaf		=	netem_leaf,
1221 	.find		=	netem_find,
1222 	.walk		=	netem_walk,
1223 	.dump		=	netem_dump_class,
1224 };
1225 
1226 static struct Qdisc_ops netem_qdisc_ops __read_mostly = {
1227 	.id		=	"netem",
1228 	.cl_ops		=	&netem_class_ops,
1229 	.priv_size	=	sizeof(struct netem_sched_data),
1230 	.enqueue	=	netem_enqueue,
1231 	.dequeue	=	netem_dequeue,
1232 	.peek		=	qdisc_peek_dequeued,
1233 	.init		=	netem_init,
1234 	.reset		=	netem_reset,
1235 	.destroy	=	netem_destroy,
1236 	.change		=	netem_change,
1237 	.dump		=	netem_dump,
1238 	.owner		=	THIS_MODULE,
1239 };
1240 
1241 
1242 static int __init netem_module_init(void)
1243 {
1244 	pr_info("netem: version " VERSION "\n");
1245 	return register_qdisc(&netem_qdisc_ops);
1246 }
1247 static void __exit netem_module_exit(void)
1248 {
1249 	unregister_qdisc(&netem_qdisc_ops);
1250 }
1251 module_init(netem_module_init)
1252 module_exit(netem_module_exit)
1253 MODULE_LICENSE("GPL");
1254