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 rootq->enqueue(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);
563 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);
581 qdisc_qstats_backlog_dec(sch, skb);
584 skb->tstamp = netem_skb_cb(skb)->tstamp_save;
586 #ifdef CONFIG_NET_CLS_ACT
588 * If it's at ingress let's pretend the delay is
589 * from the network (tstamp will be updated).
591 if (G_TC_FROM(skb->tc_verd) & AT_INGRESS)
592 skb->tstamp.tv64 = 0;
596 int err = qdisc_enqueue(skb, q->qdisc);
598 if (unlikely(err != NET_XMIT_SUCCESS)) {
599 if (net_xmit_drop_count(err)) {
600 qdisc_qstats_drop(sch);
601 qdisc_tree_reduce_backlog(sch, 1,
611 skb = q->qdisc->ops->dequeue(q->qdisc);
615 qdisc_watchdog_schedule(&q->watchdog, time_to_send);
619 skb = q->qdisc->ops->dequeue(q->qdisc);
626 static void netem_reset(struct Qdisc *sch)
628 struct netem_sched_data *q = qdisc_priv(sch);
630 qdisc_reset_queue(sch);
633 qdisc_reset(q->qdisc);
634 qdisc_watchdog_cancel(&q->watchdog);
637 static void dist_free(struct disttable *d)
643 * Distribution data is a variable size payload containing
644 * signed 16 bit values.
646 static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr)
648 struct netem_sched_data *q = qdisc_priv(sch);
649 size_t n = nla_len(attr)/sizeof(__s16);
650 const __s16 *data = nla_data(attr);
651 spinlock_t *root_lock;
656 if (n > NETEM_DIST_MAX)
659 s = sizeof(struct disttable) + n * sizeof(s16);
660 d = kmalloc(s, GFP_KERNEL | __GFP_NOWARN);
667 for (i = 0; i < n; i++)
668 d->table[i] = data[i];
670 root_lock = qdisc_root_sleeping_lock(sch);
672 spin_lock_bh(root_lock);
673 swap(q->delay_dist, d);
674 spin_unlock_bh(root_lock);
680 static void get_correlation(struct netem_sched_data *q, const struct nlattr *attr)
682 const struct tc_netem_corr *c = nla_data(attr);
684 init_crandom(&q->delay_cor, c->delay_corr);
685 init_crandom(&q->loss_cor, c->loss_corr);
686 init_crandom(&q->dup_cor, c->dup_corr);
689 static void get_reorder(struct netem_sched_data *q, const struct nlattr *attr)
691 const struct tc_netem_reorder *r = nla_data(attr);
693 q->reorder = r->probability;
694 init_crandom(&q->reorder_cor, r->correlation);
697 static void get_corrupt(struct netem_sched_data *q, const struct nlattr *attr)
699 const struct tc_netem_corrupt *r = nla_data(attr);
701 q->corrupt = r->probability;
702 init_crandom(&q->corrupt_cor, r->correlation);
705 static void get_rate(struct netem_sched_data *q, const struct nlattr *attr)
707 const struct tc_netem_rate *r = nla_data(attr);
710 q->packet_overhead = r->packet_overhead;
711 q->cell_size = r->cell_size;
712 q->cell_overhead = r->cell_overhead;
714 q->cell_size_reciprocal = reciprocal_value(q->cell_size);
716 q->cell_size_reciprocal = (struct reciprocal_value) { 0 };
719 static int get_loss_clg(struct netem_sched_data *q, const struct nlattr *attr)
721 const struct nlattr *la;
724 nla_for_each_nested(la, attr, rem) {
725 u16 type = nla_type(la);
728 case NETEM_LOSS_GI: {
729 const struct tc_netem_gimodel *gi = nla_data(la);
731 if (nla_len(la) < sizeof(struct tc_netem_gimodel)) {
732 pr_info("netem: incorrect gi model size\n");
736 q->loss_model = CLG_4_STATES;
738 q->clg.state = TX_IN_GAP_PERIOD;
747 case NETEM_LOSS_GE: {
748 const struct tc_netem_gemodel *ge = nla_data(la);
750 if (nla_len(la) < sizeof(struct tc_netem_gemodel)) {
751 pr_info("netem: incorrect ge model size\n");
755 q->loss_model = CLG_GILB_ELL;
756 q->clg.state = GOOD_STATE;
765 pr_info("netem: unknown loss type %u\n", type);
773 static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = {
774 [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) },
775 [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) },
776 [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) },
777 [TCA_NETEM_RATE] = { .len = sizeof(struct tc_netem_rate) },
778 [TCA_NETEM_LOSS] = { .type = NLA_NESTED },
779 [TCA_NETEM_ECN] = { .type = NLA_U32 },
780 [TCA_NETEM_RATE64] = { .type = NLA_U64 },
783 static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla,
784 const struct nla_policy *policy, int len)
786 int nested_len = nla_len(nla) - NLA_ALIGN(len);
788 if (nested_len < 0) {
789 pr_info("netem: invalid attributes len %d\n", nested_len);
793 if (nested_len >= nla_attr_size(0))
794 return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len),
797 memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1));
801 /* Parse netlink message to set options */
802 static int netem_change(struct Qdisc *sch, struct nlattr *opt)
804 struct netem_sched_data *q = qdisc_priv(sch);
805 struct nlattr *tb[TCA_NETEM_MAX + 1];
806 struct tc_netem_qopt *qopt;
807 struct clgstate old_clg;
808 int old_loss_model = CLG_RANDOM;
814 qopt = nla_data(opt);
815 ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt));
819 /* backup q->clg and q->loss_model */
821 old_loss_model = q->loss_model;
823 if (tb[TCA_NETEM_LOSS]) {
824 ret = get_loss_clg(q, tb[TCA_NETEM_LOSS]);
826 q->loss_model = old_loss_model;
830 q->loss_model = CLG_RANDOM;
833 if (tb[TCA_NETEM_DELAY_DIST]) {
834 ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]);
836 /* recover clg and loss_model, in case of
837 * q->clg and q->loss_model were modified
841 q->loss_model = old_loss_model;
846 sch->limit = qopt->limit;
848 q->latency = qopt->latency;
849 q->jitter = qopt->jitter;
850 q->limit = qopt->limit;
853 q->loss = qopt->loss;
854 q->duplicate = qopt->duplicate;
856 /* for compatibility with earlier versions.
857 * if gap is set, need to assume 100% probability
862 if (tb[TCA_NETEM_CORR])
863 get_correlation(q, tb[TCA_NETEM_CORR]);
865 if (tb[TCA_NETEM_REORDER])
866 get_reorder(q, tb[TCA_NETEM_REORDER]);
868 if (tb[TCA_NETEM_CORRUPT])
869 get_corrupt(q, tb[TCA_NETEM_CORRUPT]);
871 if (tb[TCA_NETEM_RATE])
872 get_rate(q, tb[TCA_NETEM_RATE]);
874 if (tb[TCA_NETEM_RATE64])
875 q->rate = max_t(u64, q->rate,
876 nla_get_u64(tb[TCA_NETEM_RATE64]));
878 if (tb[TCA_NETEM_ECN])
879 q->ecn = nla_get_u32(tb[TCA_NETEM_ECN]);
884 static int netem_init(struct Qdisc *sch, struct nlattr *opt)
886 struct netem_sched_data *q = qdisc_priv(sch);
892 qdisc_watchdog_init(&q->watchdog, sch);
894 q->loss_model = CLG_RANDOM;
895 ret = netem_change(sch, opt);
897 pr_info("netem: change failed\n");
901 static void netem_destroy(struct Qdisc *sch)
903 struct netem_sched_data *q = qdisc_priv(sch);
905 qdisc_watchdog_cancel(&q->watchdog);
907 qdisc_destroy(q->qdisc);
908 dist_free(q->delay_dist);
911 static int dump_loss_model(const struct netem_sched_data *q,
916 nest = nla_nest_start(skb, TCA_NETEM_LOSS);
918 goto nla_put_failure;
920 switch (q->loss_model) {
922 /* legacy loss model */
923 nla_nest_cancel(skb, nest);
924 return 0; /* no data */
927 struct tc_netem_gimodel gi = {
935 if (nla_put(skb, NETEM_LOSS_GI, sizeof(gi), &gi))
936 goto nla_put_failure;
940 struct tc_netem_gemodel ge = {
947 if (nla_put(skb, NETEM_LOSS_GE, sizeof(ge), &ge))
948 goto nla_put_failure;
953 nla_nest_end(skb, nest);
957 nla_nest_cancel(skb, nest);
961 static int netem_dump(struct Qdisc *sch, struct sk_buff *skb)
963 const struct netem_sched_data *q = qdisc_priv(sch);
964 struct nlattr *nla = (struct nlattr *) skb_tail_pointer(skb);
965 struct tc_netem_qopt qopt;
966 struct tc_netem_corr cor;
967 struct tc_netem_reorder reorder;
968 struct tc_netem_corrupt corrupt;
969 struct tc_netem_rate rate;
971 qopt.latency = q->latency;
972 qopt.jitter = q->jitter;
973 qopt.limit = q->limit;
976 qopt.duplicate = q->duplicate;
977 if (nla_put(skb, TCA_OPTIONS, sizeof(qopt), &qopt))
978 goto nla_put_failure;
980 cor.delay_corr = q->delay_cor.rho;
981 cor.loss_corr = q->loss_cor.rho;
982 cor.dup_corr = q->dup_cor.rho;
983 if (nla_put(skb, TCA_NETEM_CORR, sizeof(cor), &cor))
984 goto nla_put_failure;
986 reorder.probability = q->reorder;
987 reorder.correlation = q->reorder_cor.rho;
988 if (nla_put(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder))
989 goto nla_put_failure;
991 corrupt.probability = q->corrupt;
992 corrupt.correlation = q->corrupt_cor.rho;
993 if (nla_put(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt))
994 goto nla_put_failure;
996 if (q->rate >= (1ULL << 32)) {
997 if (nla_put_u64(skb, TCA_NETEM_RATE64, q->rate))
998 goto nla_put_failure;
1001 rate.rate = q->rate;
1003 rate.packet_overhead = q->packet_overhead;
1004 rate.cell_size = q->cell_size;
1005 rate.cell_overhead = q->cell_overhead;
1006 if (nla_put(skb, TCA_NETEM_RATE, sizeof(rate), &rate))
1007 goto nla_put_failure;
1009 if (q->ecn && nla_put_u32(skb, TCA_NETEM_ECN, q->ecn))
1010 goto nla_put_failure;
1012 if (dump_loss_model(q, skb) != 0)
1013 goto nla_put_failure;
1015 return nla_nest_end(skb, nla);
1018 nlmsg_trim(skb, nla);
1022 static int netem_dump_class(struct Qdisc *sch, unsigned long cl,
1023 struct sk_buff *skb, struct tcmsg *tcm)
1025 struct netem_sched_data *q = qdisc_priv(sch);
1027 if (cl != 1 || !q->qdisc) /* only one class */
1030 tcm->tcm_handle |= TC_H_MIN(1);
1031 tcm->tcm_info = q->qdisc->handle;
1036 static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1039 struct netem_sched_data *q = qdisc_priv(sch);
1041 *old = qdisc_replace(sch, new, &q->qdisc);
1045 static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg)
1047 struct netem_sched_data *q = qdisc_priv(sch);
1051 static unsigned long netem_get(struct Qdisc *sch, u32 classid)
1056 static void netem_put(struct Qdisc *sch, unsigned long arg)
1060 static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker)
1062 if (!walker->stop) {
1063 if (walker->count >= walker->skip)
1064 if (walker->fn(sch, 1, walker) < 0) {
1072 static const struct Qdisc_class_ops netem_class_ops = {
1073 .graft = netem_graft,
1078 .dump = netem_dump_class,
1081 static struct Qdisc_ops netem_qdisc_ops __read_mostly = {
1083 .cl_ops = &netem_class_ops,
1084 .priv_size = sizeof(struct netem_sched_data),
1085 .enqueue = netem_enqueue,
1086 .dequeue = netem_dequeue,
1087 .peek = qdisc_peek_dequeued,
1090 .reset = netem_reset,
1091 .destroy = netem_destroy,
1092 .change = netem_change,
1094 .owner = THIS_MODULE,
1098 static int __init netem_module_init(void)
1100 pr_info("netem: version " VERSION "\n");
1101 return register_qdisc(&netem_qdisc_ops);
1103 static void __exit netem_module_exit(void)
1105 unregister_qdisc(&netem_qdisc_ops);
1107 module_init(netem_module_init)
1108 module_exit(netem_module_exit)
1109 MODULE_LICENSE("GPL");