2 * net/sched/sch_netem.c Network emulator
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
9 * Many of the algorithms and ideas for this came from
10 * NIST Net which is not copyrighted.
12 * Authors: Stephen Hemminger <shemminger@osdl.org>
13 * Catalin(ux aka Dino) BOIE <catab at umbrella dot ro>
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>
28 #include <net/netlink.h>
29 #include <net/pkt_sched.h>
30 #include <net/inet_ecn.h>
34 /* Network Emulation Queuing algorithm.
35 ====================================
37 Sources: [1] Mark Carson, Darrin Santay, "NIST Net - A Linux-based
38 Network Emulation Tool
39 [2] Luigi Rizzo, DummyNet for FreeBSD
41 ----------------------------------------------------------------
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.
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.
56 Correlated Loss Generator models
58 Added generation of correlated loss according to the
59 "Gilbert-Elliot" model, a 4-state markov model.
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]
67 Authors: Stefano Salsano <stefano.salsano at uniroma2.it
68 Fabio Ludovici <fabio.ludovici at yahoo.it>
71 struct netem_sched_data {
72 /* internal t(ime)fifo qdisc uses t_root and sch->limit */
73 struct rb_root t_root;
75 /* optional qdisc for classful handling (NULL at netem init) */
78 struct qdisc_watchdog watchdog;
80 psched_tdiff_t latency;
81 psched_tdiff_t jitter;
94 struct reciprocal_value cell_size_reciprocal;
100 } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor;
114 TX_IN_GAP_PERIOD = 1,
117 LOST_IN_BURST_PERIOD,
125 /* Correlated Loss Generation models */
127 /* state of the Markov chain */
130 /* 4-states and Gilbert-Elliot models */
131 u32 a1; /* p13 for 4-states or p for GE */
132 u32 a2; /* p31 for 4-states or r for GE */
133 u32 a3; /* p32 for 4-states or h for GE */
134 u32 a4; /* p14 for 4-states or 1-k for GE */
135 u32 a5; /* p23 used only in 4-states */
140 /* Time stamp put into socket buffer control block
141 * Only valid when skbs are in our internal t(ime)fifo queue.
143 * As skb->rbnode uses same storage than skb->next, skb->prev and skb->tstamp,
144 * and skb->next & skb->prev are scratch space for a qdisc,
145 * we save skb->tstamp value in skb->cb[] before destroying it.
147 struct netem_skb_cb {
148 psched_time_t time_to_send;
153 static struct sk_buff *netem_rb_to_skb(struct rb_node *rb)
155 return container_of(rb, struct sk_buff, rbnode);
158 static inline struct netem_skb_cb *netem_skb_cb(struct sk_buff *skb)
160 /* we assume we can use skb next/prev/tstamp as storage for rb_node */
161 qdisc_cb_private_validate(skb, sizeof(struct netem_skb_cb));
162 return (struct netem_skb_cb *)qdisc_skb_cb(skb)->data;
165 /* init_crandom - initialize correlated random number generator
166 * Use entropy source for initial seed.
168 static void init_crandom(struct crndstate *state, unsigned long rho)
171 state->last = prandom_u32();
174 /* get_crandom - correlated random number generator
175 * Next number depends on last value.
176 * rho is scaled to avoid floating point.
178 static u32 get_crandom(struct crndstate *state)
181 unsigned long answer;
183 if (state->rho == 0) /* no correlation */
184 return prandom_u32();
186 value = prandom_u32();
187 rho = (u64)state->rho + 1;
188 answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32;
189 state->last = answer;
193 /* loss_4state - 4-state model loss generator
194 * Generates losses according to the 4-state Markov chain adopted in
195 * the GI (General and Intuitive) loss model.
197 static bool loss_4state(struct netem_sched_data *q)
199 struct clgstate *clg = &q->clg;
200 u32 rnd = prandom_u32();
203 * Makes a comparison between rnd and the transition
204 * probabilities outgoing from the current state, then decides the
205 * next state and if the next packet has to be transmitted or lost.
206 * The four states correspond to:
207 * TX_IN_GAP_PERIOD => successfully transmitted packets within a gap period
208 * LOST_IN_BURST_PERIOD => isolated losses within a gap period
209 * LOST_IN_GAP_PERIOD => lost packets within a burst period
210 * TX_IN_GAP_PERIOD => successfully transmitted packets within a burst period
212 switch (clg->state) {
213 case TX_IN_GAP_PERIOD:
215 clg->state = LOST_IN_BURST_PERIOD;
217 } else if (clg->a4 < rnd && rnd < clg->a1 + clg->a4) {
218 clg->state = LOST_IN_GAP_PERIOD;
220 } else if (clg->a1 + clg->a4 < rnd) {
221 clg->state = TX_IN_GAP_PERIOD;
225 case TX_IN_BURST_PERIOD:
227 clg->state = LOST_IN_GAP_PERIOD;
230 clg->state = TX_IN_BURST_PERIOD;
234 case LOST_IN_GAP_PERIOD:
236 clg->state = TX_IN_BURST_PERIOD;
237 else if (clg->a3 < rnd && rnd < clg->a2 + clg->a3) {
238 clg->state = TX_IN_GAP_PERIOD;
239 } else if (clg->a2 + clg->a3 < rnd) {
240 clg->state = LOST_IN_GAP_PERIOD;
244 case LOST_IN_BURST_PERIOD:
245 clg->state = TX_IN_GAP_PERIOD;
252 /* loss_gilb_ell - Gilbert-Elliot model loss generator
253 * Generates losses according to the Gilbert-Elliot loss model or
254 * its special cases (Gilbert or Simple Gilbert)
256 * Makes a comparison between random number and the transition
257 * probabilities outgoing from the current state, then decides the
258 * next state. A second random number is extracted and the comparison
259 * with the loss probability of the current state decides if the next
260 * packet will be transmitted or lost.
262 static bool loss_gilb_ell(struct netem_sched_data *q)
264 struct clgstate *clg = &q->clg;
266 switch (clg->state) {
268 if (prandom_u32() < clg->a1)
269 clg->state = BAD_STATE;
270 if (prandom_u32() < clg->a4)
274 if (prandom_u32() < clg->a2)
275 clg->state = GOOD_STATE;
276 if (prandom_u32() > clg->a3)
283 static bool loss_event(struct netem_sched_data *q)
285 switch (q->loss_model) {
287 /* Random packet drop 0 => none, ~0 => all */
288 return q->loss && q->loss >= get_crandom(&q->loss_cor);
291 /* 4state loss model algorithm (used also for GI model)
292 * Extracts a value from the markov 4 state loss generator,
293 * if it is 1 drops a packet and if needed writes the event in
296 return loss_4state(q);
299 /* Gilbert-Elliot loss model algorithm
300 * Extracts a value from the Gilbert-Elliot loss generator,
301 * if it is 1 drops a packet and if needed writes the event in
304 return loss_gilb_ell(q);
307 return false; /* not reached */
311 /* tabledist - return a pseudo-randomly distributed value with mean mu and
312 * std deviation sigma. Uses table lookup to approximate the desired
313 * distribution, and a uniformly-distributed pseudo-random source.
315 static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma,
316 struct crndstate *state,
317 const struct disttable *dist)
326 rnd = get_crandom(state);
328 /* default uniform distribution */
330 return (rnd % (2*sigma)) - sigma + mu;
332 t = dist->table[rnd % dist->size];
333 x = (sigma % NETEM_DIST_SCALE) * t;
335 x += NETEM_DIST_SCALE/2;
337 x -= NETEM_DIST_SCALE/2;
339 return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu;
342 static psched_time_t packet_len_2_sched_time(unsigned int len, struct netem_sched_data *q)
346 len += q->packet_overhead;
349 u32 cells = reciprocal_divide(len, q->cell_size_reciprocal);
351 if (len > cells * q->cell_size) /* extra cell needed for remainder */
353 len = cells * (q->cell_size + q->cell_overhead);
356 ticks = (u64)len * NSEC_PER_SEC;
358 do_div(ticks, q->rate);
359 return PSCHED_NS2TICKS(ticks);
362 static void tfifo_reset(struct Qdisc *sch)
364 struct netem_sched_data *q = qdisc_priv(sch);
367 while ((p = rb_first(&q->t_root))) {
368 struct sk_buff *skb = netem_rb_to_skb(p);
370 rb_erase(p, &q->t_root);
377 static void tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch)
379 struct netem_sched_data *q = qdisc_priv(sch);
380 psched_time_t tnext = netem_skb_cb(nskb)->time_to_send;
381 struct rb_node **p = &q->t_root.rb_node, *parent = NULL;
387 skb = netem_rb_to_skb(parent);
388 if (tnext >= netem_skb_cb(skb)->time_to_send)
389 p = &parent->rb_right;
391 p = &parent->rb_left;
393 rb_link_node(&nskb->rbnode, parent, p);
394 rb_insert_color(&nskb->rbnode, &q->t_root);
399 * Insert one skb into qdisc.
400 * Note: parent depends on return value to account for queue length.
401 * NET_XMIT_DROP: queue length didn't change.
402 * NET_XMIT_SUCCESS: one skb was queued.
404 static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch)
406 struct netem_sched_data *q = qdisc_priv(sch);
407 /* We don't fill cb now as skb_unshare() may invalidate it */
408 struct netem_skb_cb *cb;
409 struct sk_buff *skb2;
412 /* Random duplication */
413 if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor))
418 if (q->ecn && INET_ECN_set_ce(skb))
419 qdisc_qstats_drop(sch); /* mark packet */
424 qdisc_qstats_drop(sch);
426 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
429 /* If a delay is expected, orphan the skb. (orphaning usually takes
430 * place at TX completion time, so _before_ the link transit delay)
432 if (q->latency || q->jitter)
433 skb_orphan_partial(skb);
436 * If we need to duplicate packet, then re-insert at top of the
437 * qdisc tree, since parent queuer expects that only one
438 * skb will be queued.
440 if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) {
441 struct Qdisc *rootq = qdisc_root(sch);
442 u32 dupsave = q->duplicate; /* prevent duplicating a dup... */
445 qdisc_enqueue_root(skb2, rootq);
446 q->duplicate = dupsave;
450 * Randomized packet corruption.
451 * Make copy if needed since we are modifying
452 * If packet is going to be hardware checksummed, then
453 * do it now in software before we mangle it.
455 if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) {
456 if (!(skb = skb_unshare(skb, GFP_ATOMIC)) ||
457 (skb->ip_summed == CHECKSUM_PARTIAL &&
458 skb_checksum_help(skb)))
459 return qdisc_drop(skb, sch);
461 skb->data[prandom_u32() % skb_headlen(skb)] ^=
462 1<<(prandom_u32() % 8);
465 if (unlikely(skb_queue_len(&sch->q) >= sch->limit))
466 return qdisc_reshape_fail(skb, sch);
468 qdisc_qstats_backlog_inc(sch, skb);
470 cb = netem_skb_cb(skb);
471 if (q->gap == 0 || /* not doing reordering */
472 q->counter < q->gap - 1 || /* inside last reordering gap */
473 q->reorder < get_crandom(&q->reorder_cor)) {
475 psched_tdiff_t delay;
477 delay = tabledist(q->latency, q->jitter,
478 &q->delay_cor, q->delay_dist);
480 now = psched_get_time();
483 struct sk_buff *last;
485 if (!skb_queue_empty(&sch->q))
486 last = skb_peek_tail(&sch->q);
488 last = netem_rb_to_skb(rb_last(&q->t_root));
491 * Last packet in queue is reference point (now),
492 * calculate this time bonus and subtract
495 delay -= netem_skb_cb(last)->time_to_send - now;
496 delay = max_t(psched_tdiff_t, 0, delay);
497 now = netem_skb_cb(last)->time_to_send;
500 delay += packet_len_2_sched_time(qdisc_pkt_len(skb), q);
503 cb->time_to_send = now + delay;
504 cb->tstamp_save = skb->tstamp;
506 tfifo_enqueue(skb, sch);
509 * Do re-ordering by putting one out of N packets at the front
512 cb->time_to_send = psched_get_time();
515 __skb_queue_head(&sch->q, skb);
516 sch->qstats.requeues++;
519 return NET_XMIT_SUCCESS;
522 static unsigned int netem_drop(struct Qdisc *sch)
524 struct netem_sched_data *q = qdisc_priv(sch);
527 len = qdisc_queue_drop(sch);
530 struct rb_node *p = rb_first(&q->t_root);
533 struct sk_buff *skb = netem_rb_to_skb(p);
535 rb_erase(p, &q->t_root);
539 qdisc_qstats_backlog_dec(sch, skb);
543 if (!len && q->qdisc && q->qdisc->ops->drop)
544 len = q->qdisc->ops->drop(q->qdisc);
546 qdisc_qstats_drop(sch);
551 static struct sk_buff *netem_dequeue(struct Qdisc *sch)
553 struct netem_sched_data *q = qdisc_priv(sch);
557 if (qdisc_is_throttled(sch))
561 skb = __skb_dequeue(&sch->q);
564 qdisc_qstats_backlog_dec(sch, skb);
565 qdisc_unthrottled(sch);
566 qdisc_bstats_update(sch, skb);
569 p = rb_first(&q->t_root);
571 psched_time_t time_to_send;
573 skb = netem_rb_to_skb(p);
575 /* if more time remaining? */
576 time_to_send = netem_skb_cb(skb)->time_to_send;
577 if (time_to_send <= psched_get_time()) {
578 rb_erase(p, &q->t_root);
583 skb->tstamp = netem_skb_cb(skb)->tstamp_save;
585 #ifdef CONFIG_NET_CLS_ACT
587 * If it's at ingress let's pretend the delay is
588 * from the network (tstamp will be updated).
590 if (G_TC_FROM(skb->tc_verd) & AT_INGRESS)
591 skb->tstamp.tv64 = 0;
595 int err = qdisc_enqueue(skb, q->qdisc);
597 if (unlikely(err != NET_XMIT_SUCCESS)) {
598 if (net_xmit_drop_count(err)) {
599 qdisc_qstats_drop(sch);
600 qdisc_tree_decrease_qlen(sch, 1);
609 skb = q->qdisc->ops->dequeue(q->qdisc);
613 qdisc_watchdog_schedule(&q->watchdog, time_to_send);
617 skb = q->qdisc->ops->dequeue(q->qdisc);
624 static void netem_reset(struct Qdisc *sch)
626 struct netem_sched_data *q = qdisc_priv(sch);
628 qdisc_reset_queue(sch);
631 qdisc_reset(q->qdisc);
632 qdisc_watchdog_cancel(&q->watchdog);
635 static void dist_free(struct disttable *d)
641 * Distribution data is a variable size payload containing
642 * signed 16 bit values.
644 static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr)
646 struct netem_sched_data *q = qdisc_priv(sch);
647 size_t n = nla_len(attr)/sizeof(__s16);
648 const __s16 *data = nla_data(attr);
649 spinlock_t *root_lock;
654 if (n > NETEM_DIST_MAX)
657 s = sizeof(struct disttable) + n * sizeof(s16);
658 d = kmalloc(s, GFP_KERNEL | __GFP_NOWARN);
665 for (i = 0; i < n; i++)
666 d->table[i] = data[i];
668 root_lock = qdisc_root_sleeping_lock(sch);
670 spin_lock_bh(root_lock);
671 swap(q->delay_dist, d);
672 spin_unlock_bh(root_lock);
678 static void get_correlation(struct netem_sched_data *q, const struct nlattr *attr)
680 const struct tc_netem_corr *c = nla_data(attr);
682 init_crandom(&q->delay_cor, c->delay_corr);
683 init_crandom(&q->loss_cor, c->loss_corr);
684 init_crandom(&q->dup_cor, c->dup_corr);
687 static void get_reorder(struct netem_sched_data *q, const struct nlattr *attr)
689 const struct tc_netem_reorder *r = nla_data(attr);
691 q->reorder = r->probability;
692 init_crandom(&q->reorder_cor, r->correlation);
695 static void get_corrupt(struct netem_sched_data *q, const struct nlattr *attr)
697 const struct tc_netem_corrupt *r = nla_data(attr);
699 q->corrupt = r->probability;
700 init_crandom(&q->corrupt_cor, r->correlation);
703 static void get_rate(struct netem_sched_data *q, const struct nlattr *attr)
705 const struct tc_netem_rate *r = nla_data(attr);
708 q->packet_overhead = r->packet_overhead;
709 q->cell_size = r->cell_size;
710 q->cell_overhead = r->cell_overhead;
712 q->cell_size_reciprocal = reciprocal_value(q->cell_size);
714 q->cell_size_reciprocal = (struct reciprocal_value) { 0 };
717 static int get_loss_clg(struct netem_sched_data *q, const struct nlattr *attr)
719 const struct nlattr *la;
722 nla_for_each_nested(la, attr, rem) {
723 u16 type = nla_type(la);
726 case NETEM_LOSS_GI: {
727 const struct tc_netem_gimodel *gi = nla_data(la);
729 if (nla_len(la) < sizeof(struct tc_netem_gimodel)) {
730 pr_info("netem: incorrect gi model size\n");
734 q->loss_model = CLG_4_STATES;
736 q->clg.state = TX_IN_GAP_PERIOD;
745 case NETEM_LOSS_GE: {
746 const struct tc_netem_gemodel *ge = nla_data(la);
748 if (nla_len(la) < sizeof(struct tc_netem_gemodel)) {
749 pr_info("netem: incorrect ge model size\n");
753 q->loss_model = CLG_GILB_ELL;
754 q->clg.state = GOOD_STATE;
763 pr_info("netem: unknown loss type %u\n", type);
771 static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = {
772 [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) },
773 [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) },
774 [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) },
775 [TCA_NETEM_RATE] = { .len = sizeof(struct tc_netem_rate) },
776 [TCA_NETEM_LOSS] = { .type = NLA_NESTED },
777 [TCA_NETEM_ECN] = { .type = NLA_U32 },
778 [TCA_NETEM_RATE64] = { .type = NLA_U64 },
781 static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla,
782 const struct nla_policy *policy, int len)
784 int nested_len = nla_len(nla) - NLA_ALIGN(len);
786 if (nested_len < 0) {
787 pr_info("netem: invalid attributes len %d\n", nested_len);
791 if (nested_len >= nla_attr_size(0))
792 return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len),
795 memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1));
799 /* Parse netlink message to set options */
800 static int netem_change(struct Qdisc *sch, struct nlattr *opt)
802 struct netem_sched_data *q = qdisc_priv(sch);
803 struct nlattr *tb[TCA_NETEM_MAX + 1];
804 struct tc_netem_qopt *qopt;
805 struct clgstate old_clg;
806 int old_loss_model = CLG_RANDOM;
812 qopt = nla_data(opt);
813 ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt));
817 /* backup q->clg and q->loss_model */
819 old_loss_model = q->loss_model;
821 if (tb[TCA_NETEM_LOSS]) {
822 ret = get_loss_clg(q, tb[TCA_NETEM_LOSS]);
824 q->loss_model = old_loss_model;
828 q->loss_model = CLG_RANDOM;
831 if (tb[TCA_NETEM_DELAY_DIST]) {
832 ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]);
834 /* recover clg and loss_model, in case of
835 * q->clg and q->loss_model were modified
839 q->loss_model = old_loss_model;
844 sch->limit = qopt->limit;
846 q->latency = qopt->latency;
847 q->jitter = qopt->jitter;
848 q->limit = qopt->limit;
851 q->loss = qopt->loss;
852 q->duplicate = qopt->duplicate;
854 /* for compatibility with earlier versions.
855 * if gap is set, need to assume 100% probability
860 if (tb[TCA_NETEM_CORR])
861 get_correlation(q, tb[TCA_NETEM_CORR]);
863 if (tb[TCA_NETEM_REORDER])
864 get_reorder(q, tb[TCA_NETEM_REORDER]);
866 if (tb[TCA_NETEM_CORRUPT])
867 get_corrupt(q, tb[TCA_NETEM_CORRUPT]);
869 if (tb[TCA_NETEM_RATE])
870 get_rate(q, tb[TCA_NETEM_RATE]);
872 if (tb[TCA_NETEM_RATE64])
873 q->rate = max_t(u64, q->rate,
874 nla_get_u64(tb[TCA_NETEM_RATE64]));
876 if (tb[TCA_NETEM_ECN])
877 q->ecn = nla_get_u32(tb[TCA_NETEM_ECN]);
882 static int netem_init(struct Qdisc *sch, struct nlattr *opt)
884 struct netem_sched_data *q = qdisc_priv(sch);
890 qdisc_watchdog_init(&q->watchdog, sch);
892 q->loss_model = CLG_RANDOM;
893 ret = netem_change(sch, opt);
895 pr_info("netem: change failed\n");
899 static void netem_destroy(struct Qdisc *sch)
901 struct netem_sched_data *q = qdisc_priv(sch);
903 qdisc_watchdog_cancel(&q->watchdog);
905 qdisc_destroy(q->qdisc);
906 dist_free(q->delay_dist);
909 static int dump_loss_model(const struct netem_sched_data *q,
914 nest = nla_nest_start(skb, TCA_NETEM_LOSS);
916 goto nla_put_failure;
918 switch (q->loss_model) {
920 /* legacy loss model */
921 nla_nest_cancel(skb, nest);
922 return 0; /* no data */
925 struct tc_netem_gimodel gi = {
933 if (nla_put(skb, NETEM_LOSS_GI, sizeof(gi), &gi))
934 goto nla_put_failure;
938 struct tc_netem_gemodel ge = {
945 if (nla_put(skb, NETEM_LOSS_GE, sizeof(ge), &ge))
946 goto nla_put_failure;
951 nla_nest_end(skb, nest);
955 nla_nest_cancel(skb, nest);
959 static int netem_dump(struct Qdisc *sch, struct sk_buff *skb)
961 const struct netem_sched_data *q = qdisc_priv(sch);
962 struct nlattr *nla = (struct nlattr *) skb_tail_pointer(skb);
963 struct tc_netem_qopt qopt;
964 struct tc_netem_corr cor;
965 struct tc_netem_reorder reorder;
966 struct tc_netem_corrupt corrupt;
967 struct tc_netem_rate rate;
969 qopt.latency = q->latency;
970 qopt.jitter = q->jitter;
971 qopt.limit = q->limit;
974 qopt.duplicate = q->duplicate;
975 if (nla_put(skb, TCA_OPTIONS, sizeof(qopt), &qopt))
976 goto nla_put_failure;
978 cor.delay_corr = q->delay_cor.rho;
979 cor.loss_corr = q->loss_cor.rho;
980 cor.dup_corr = q->dup_cor.rho;
981 if (nla_put(skb, TCA_NETEM_CORR, sizeof(cor), &cor))
982 goto nla_put_failure;
984 reorder.probability = q->reorder;
985 reorder.correlation = q->reorder_cor.rho;
986 if (nla_put(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder))
987 goto nla_put_failure;
989 corrupt.probability = q->corrupt;
990 corrupt.correlation = q->corrupt_cor.rho;
991 if (nla_put(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt))
992 goto nla_put_failure;
994 if (q->rate >= (1ULL << 32)) {
995 if (nla_put_u64(skb, TCA_NETEM_RATE64, q->rate))
996 goto nla_put_failure;
1001 rate.packet_overhead = q->packet_overhead;
1002 rate.cell_size = q->cell_size;
1003 rate.cell_overhead = q->cell_overhead;
1004 if (nla_put(skb, TCA_NETEM_RATE, sizeof(rate), &rate))
1005 goto nla_put_failure;
1007 if (q->ecn && nla_put_u32(skb, TCA_NETEM_ECN, q->ecn))
1008 goto nla_put_failure;
1010 if (dump_loss_model(q, skb) != 0)
1011 goto nla_put_failure;
1013 return nla_nest_end(skb, nla);
1016 nlmsg_trim(skb, nla);
1020 static int netem_dump_class(struct Qdisc *sch, unsigned long cl,
1021 struct sk_buff *skb, struct tcmsg *tcm)
1023 struct netem_sched_data *q = qdisc_priv(sch);
1025 if (cl != 1 || !q->qdisc) /* only one class */
1028 tcm->tcm_handle |= TC_H_MIN(1);
1029 tcm->tcm_info = q->qdisc->handle;
1034 static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1037 struct netem_sched_data *q = qdisc_priv(sch);
1043 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
1046 sch_tree_unlock(sch);
1051 static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg)
1053 struct netem_sched_data *q = qdisc_priv(sch);
1057 static unsigned long netem_get(struct Qdisc *sch, u32 classid)
1062 static void netem_put(struct Qdisc *sch, unsigned long arg)
1066 static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker)
1068 if (!walker->stop) {
1069 if (walker->count >= walker->skip)
1070 if (walker->fn(sch, 1, walker) < 0) {
1078 static const struct Qdisc_class_ops netem_class_ops = {
1079 .graft = netem_graft,
1084 .dump = netem_dump_class,
1087 static struct Qdisc_ops netem_qdisc_ops __read_mostly = {
1089 .cl_ops = &netem_class_ops,
1090 .priv_size = sizeof(struct netem_sched_data),
1091 .enqueue = netem_enqueue,
1092 .dequeue = netem_dequeue,
1093 .peek = qdisc_peek_dequeued,
1096 .reset = netem_reset,
1097 .destroy = netem_destroy,
1098 .change = netem_change,
1100 .owner = THIS_MODULE,
1104 static int __init netem_module_init(void)
1106 pr_info("netem: version " VERSION "\n");
1107 return register_qdisc(&netem_qdisc_ops);
1109 static void __exit netem_module_exit(void)
1111 unregister_qdisc(&netem_qdisc_ops);
1113 module_init(netem_module_init)
1114 module_exit(netem_module_exit)
1115 MODULE_LICENSE("GPL");