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 u32 cell_size_reciprocal;
100 } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor;
113 /* Correlated Loss Generation models */
115 /* state of the Markov chain */
118 /* 4-states and Gilbert-Elliot models */
119 u32 a1; /* p13 for 4-states or p for GE */
120 u32 a2; /* p31 for 4-states or r for GE */
121 u32 a3; /* p32 for 4-states or h for GE */
122 u32 a4; /* p14 for 4-states or 1-k for GE */
123 u32 a5; /* p23 used only in 4-states */
128 /* Time stamp put into socket buffer control block
129 * Only valid when skbs are in our internal t(ime)fifo queue.
131 struct netem_skb_cb {
132 psched_time_t time_to_send;
136 /* Because space in skb->cb[] is tight, netem overloads skb->next/prev/tstamp
137 * to hold a rb_node structure.
139 * If struct sk_buff layout is changed, the following checks will complain.
141 static struct rb_node *netem_rb_node(struct sk_buff *skb)
143 BUILD_BUG_ON(offsetof(struct sk_buff, next) != 0);
144 BUILD_BUG_ON(offsetof(struct sk_buff, prev) !=
145 offsetof(struct sk_buff, next) + sizeof(skb->next));
146 BUILD_BUG_ON(offsetof(struct sk_buff, tstamp) !=
147 offsetof(struct sk_buff, prev) + sizeof(skb->prev));
148 BUILD_BUG_ON(sizeof(struct rb_node) > sizeof(skb->next) +
150 sizeof(skb->tstamp));
151 return (struct rb_node *)&skb->next;
154 static struct sk_buff *netem_rb_to_skb(struct rb_node *rb)
156 return (struct sk_buff *)rb;
159 static inline struct netem_skb_cb *netem_skb_cb(struct sk_buff *skb)
161 /* we assume we can use skb next/prev/tstamp as storage for rb_node */
162 qdisc_cb_private_validate(skb, sizeof(struct netem_skb_cb));
163 return (struct netem_skb_cb *)qdisc_skb_cb(skb)->data;
166 /* init_crandom - initialize correlated random number generator
167 * Use entropy source for initial seed.
169 static void init_crandom(struct crndstate *state, unsigned long rho)
172 state->last = prandom_u32();
175 /* get_crandom - correlated random number generator
176 * Next number depends on last value.
177 * rho is scaled to avoid floating point.
179 static u32 get_crandom(struct crndstate *state)
182 unsigned long answer;
184 if (state->rho == 0) /* no correlation */
185 return prandom_u32();
187 value = prandom_u32();
188 rho = (u64)state->rho + 1;
189 answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32;
190 state->last = answer;
194 /* loss_4state - 4-state model loss generator
195 * Generates losses according to the 4-state Markov chain adopted in
196 * the GI (General and Intuitive) loss model.
198 static bool loss_4state(struct netem_sched_data *q)
200 struct clgstate *clg = &q->clg;
201 u32 rnd = prandom_u32();
204 * Makes a comparison between rnd and the transition
205 * probabilities outgoing from the current state, then decides the
206 * next state and if the next packet has to be transmitted or lost.
207 * The four states correspond to:
208 * 1 => successfully transmitted packets within a gap period
209 * 4 => isolated losses within a gap period
210 * 3 => lost packets within a burst period
211 * 2 => successfully transmitted packets within a burst period
213 switch (clg->state) {
218 } else if (clg->a4 < rnd && rnd < clg->a1 + clg->a4) {
221 } else if (clg->a1 + clg->a4 < rnd)
236 else if (clg->a3 < rnd && rnd < clg->a2 + clg->a3) {
238 } else if (clg->a2 + clg->a3 < rnd) {
251 /* loss_gilb_ell - Gilbert-Elliot model loss generator
252 * Generates losses according to the Gilbert-Elliot loss model or
253 * its special cases (Gilbert or Simple Gilbert)
255 * Makes a comparison between random number and the transition
256 * probabilities outgoing from the current state, then decides the
257 * next state. A second random number is extracted and the comparison
258 * with the loss probability of the current state decides if the next
259 * packet will be transmitted or lost.
261 static bool loss_gilb_ell(struct netem_sched_data *q)
263 struct clgstate *clg = &q->clg;
265 switch (clg->state) {
267 if (prandom_u32() < clg->a1)
269 if (prandom_u32() < clg->a4)
273 if (prandom_u32() < clg->a2)
275 if (prandom_u32() > clg->a3)
282 static bool loss_event(struct netem_sched_data *q)
284 switch (q->loss_model) {
286 /* Random packet drop 0 => none, ~0 => all */
287 return q->loss && q->loss >= get_crandom(&q->loss_cor);
290 /* 4state loss model algorithm (used also for GI model)
291 * Extracts a value from the markov 4 state loss generator,
292 * if it is 1 drops a packet and if needed writes the event in
295 return loss_4state(q);
298 /* Gilbert-Elliot loss model algorithm
299 * Extracts a value from the Gilbert-Elliot loss generator,
300 * if it is 1 drops a packet and if needed writes the event in
303 return loss_gilb_ell(q);
306 return false; /* not reached */
310 /* tabledist - return a pseudo-randomly distributed value with mean mu and
311 * std deviation sigma. Uses table lookup to approximate the desired
312 * distribution, and a uniformly-distributed pseudo-random source.
314 static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma,
315 struct crndstate *state,
316 const struct disttable *dist)
325 rnd = get_crandom(state);
327 /* default uniform distribution */
329 return (rnd % (2*sigma)) - sigma + mu;
331 t = dist->table[rnd % dist->size];
332 x = (sigma % NETEM_DIST_SCALE) * t;
334 x += NETEM_DIST_SCALE/2;
336 x -= NETEM_DIST_SCALE/2;
338 return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu;
341 static psched_time_t packet_len_2_sched_time(unsigned int len, struct netem_sched_data *q)
345 len += q->packet_overhead;
348 u32 cells = reciprocal_divide(len, q->cell_size_reciprocal);
350 if (len > cells * q->cell_size) /* extra cell needed for remainder */
352 len = cells * (q->cell_size + q->cell_overhead);
355 ticks = (u64)len * NSEC_PER_SEC;
357 do_div(ticks, q->rate);
358 return PSCHED_NS2TICKS(ticks);
361 static void tfifo_reset(struct Qdisc *sch)
363 struct netem_sched_data *q = qdisc_priv(sch);
366 while ((p = rb_first(&q->t_root))) {
367 struct sk_buff *skb = netem_rb_to_skb(p);
369 rb_erase(p, &q->t_root);
376 static void tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch)
378 struct netem_sched_data *q = qdisc_priv(sch);
379 psched_time_t tnext = netem_skb_cb(nskb)->time_to_send;
380 struct rb_node **p = &q->t_root.rb_node, *parent = NULL;
386 skb = netem_rb_to_skb(parent);
387 if (tnext >= netem_skb_cb(skb)->time_to_send)
388 p = &parent->rb_right;
390 p = &parent->rb_left;
392 rb_link_node(netem_rb_node(nskb), parent, p);
393 rb_insert_color(netem_rb_node(nskb), &q->t_root);
398 * Insert one skb into qdisc.
399 * Note: parent depends on return value to account for queue length.
400 * NET_XMIT_DROP: queue length didn't change.
401 * NET_XMIT_SUCCESS: one skb was queued.
403 static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch)
405 struct netem_sched_data *q = qdisc_priv(sch);
406 /* We don't fill cb now as skb_unshare() may invalidate it */
407 struct netem_skb_cb *cb;
408 struct sk_buff *skb2;
411 /* Random duplication */
412 if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor))
417 if (q->ecn && INET_ECN_set_ce(skb))
418 sch->qstats.drops++; /* mark packet */
425 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
428 /* If a delay is expected, orphan the skb. (orphaning usually takes
429 * place at TX completion time, so _before_ the link transit delay)
431 if (q->latency || q->jitter)
432 skb_orphan_partial(skb);
435 * If we need to duplicate packet, then re-insert at top of the
436 * qdisc tree, since parent queuer expects that only one
437 * skb will be queued.
439 if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) {
440 struct Qdisc *rootq = qdisc_root(sch);
441 u32 dupsave = q->duplicate; /* prevent duplicating a dup... */
444 qdisc_enqueue_root(skb2, rootq);
445 q->duplicate = dupsave;
449 * Randomized packet corruption.
450 * Make copy if needed since we are modifying
451 * If packet is going to be hardware checksummed, then
452 * do it now in software before we mangle it.
454 if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) {
455 if (!(skb = skb_unshare(skb, GFP_ATOMIC)) ||
456 (skb->ip_summed == CHECKSUM_PARTIAL &&
457 skb_checksum_help(skb)))
458 return qdisc_drop(skb, sch);
460 skb->data[prandom_u32() % skb_headlen(skb)] ^=
461 1<<(prandom_u32() % 8);
464 if (unlikely(skb_queue_len(&sch->q) >= sch->limit))
465 return qdisc_reshape_fail(skb, sch);
467 sch->qstats.backlog += qdisc_pkt_len(skb);
469 cb = netem_skb_cb(skb);
470 if (q->gap == 0 || /* not doing reordering */
471 q->counter < q->gap - 1 || /* inside last reordering gap */
472 q->reorder < get_crandom(&q->reorder_cor)) {
474 psched_tdiff_t delay;
476 delay = tabledist(q->latency, q->jitter,
477 &q->delay_cor, q->delay_dist);
479 now = psched_get_time();
482 struct sk_buff *last;
484 if (!skb_queue_empty(&sch->q))
485 last = skb_peek_tail(&sch->q);
487 last = netem_rb_to_skb(rb_last(&q->t_root));
490 * Last packet in queue is reference point (now),
491 * calculate this time bonus and subtract
494 delay -= netem_skb_cb(last)->time_to_send - now;
495 delay = max_t(psched_tdiff_t, 0, delay);
496 now = netem_skb_cb(last)->time_to_send;
499 delay += packet_len_2_sched_time(qdisc_pkt_len(skb), q);
502 cb->time_to_send = now + delay;
503 cb->tstamp_save = skb->tstamp;
505 tfifo_enqueue(skb, sch);
508 * Do re-ordering by putting one out of N packets at the front
511 cb->time_to_send = psched_get_time();
514 __skb_queue_head(&sch->q, skb);
515 sch->qstats.requeues++;
518 return NET_XMIT_SUCCESS;
521 static unsigned int netem_drop(struct Qdisc *sch)
523 struct netem_sched_data *q = qdisc_priv(sch);
526 len = qdisc_queue_drop(sch);
529 struct rb_node *p = rb_first(&q->t_root);
532 struct sk_buff *skb = netem_rb_to_skb(p);
534 rb_erase(p, &q->t_root);
538 len = qdisc_pkt_len(skb);
539 sch->qstats.backlog -= len;
543 if (!len && q->qdisc && q->qdisc->ops->drop)
544 len = q->qdisc->ops->drop(q->qdisc);
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 sch->qstats.backlog -= qdisc_pkt_len(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)) {
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)
638 if (is_vmalloc_addr(d))
646 * Distribution data is a variable size payload containing
647 * signed 16 bit values.
649 static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr)
651 struct netem_sched_data *q = qdisc_priv(sch);
652 size_t n = nla_len(attr)/sizeof(__s16);
653 const __s16 *data = nla_data(attr);
654 spinlock_t *root_lock;
659 if (n > NETEM_DIST_MAX)
662 s = sizeof(struct disttable) + n * sizeof(s16);
663 d = kmalloc(s, GFP_KERNEL | __GFP_NOWARN);
670 for (i = 0; i < n; i++)
671 d->table[i] = data[i];
673 root_lock = qdisc_root_sleeping_lock(sch);
675 spin_lock_bh(root_lock);
676 swap(q->delay_dist, d);
677 spin_unlock_bh(root_lock);
683 static void get_correlation(struct Qdisc *sch, const struct nlattr *attr)
685 struct netem_sched_data *q = qdisc_priv(sch);
686 const struct tc_netem_corr *c = nla_data(attr);
688 init_crandom(&q->delay_cor, c->delay_corr);
689 init_crandom(&q->loss_cor, c->loss_corr);
690 init_crandom(&q->dup_cor, c->dup_corr);
693 static void get_reorder(struct Qdisc *sch, const struct nlattr *attr)
695 struct netem_sched_data *q = qdisc_priv(sch);
696 const struct tc_netem_reorder *r = nla_data(attr);
698 q->reorder = r->probability;
699 init_crandom(&q->reorder_cor, r->correlation);
702 static void get_corrupt(struct Qdisc *sch, const struct nlattr *attr)
704 struct netem_sched_data *q = qdisc_priv(sch);
705 const struct tc_netem_corrupt *r = nla_data(attr);
707 q->corrupt = r->probability;
708 init_crandom(&q->corrupt_cor, r->correlation);
711 static void get_rate(struct Qdisc *sch, const struct nlattr *attr)
713 struct netem_sched_data *q = qdisc_priv(sch);
714 const struct tc_netem_rate *r = nla_data(attr);
717 q->packet_overhead = r->packet_overhead;
718 q->cell_size = r->cell_size;
720 q->cell_size_reciprocal = reciprocal_value(q->cell_size);
721 q->cell_overhead = r->cell_overhead;
724 static int get_loss_clg(struct Qdisc *sch, const struct nlattr *attr)
726 struct netem_sched_data *q = qdisc_priv(sch);
727 const struct nlattr *la;
730 nla_for_each_nested(la, attr, rem) {
731 u16 type = nla_type(la);
734 case NETEM_LOSS_GI: {
735 const struct tc_netem_gimodel *gi = nla_data(la);
737 if (nla_len(la) < sizeof(struct tc_netem_gimodel)) {
738 pr_info("netem: incorrect gi model size\n");
742 q->loss_model = CLG_4_STATES;
753 case NETEM_LOSS_GE: {
754 const struct tc_netem_gemodel *ge = nla_data(la);
756 if (nla_len(la) < sizeof(struct tc_netem_gemodel)) {
757 pr_info("netem: incorrect ge model size\n");
761 q->loss_model = CLG_GILB_ELL;
771 pr_info("netem: unknown loss type %u\n", type);
779 static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = {
780 [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) },
781 [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) },
782 [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) },
783 [TCA_NETEM_RATE] = { .len = sizeof(struct tc_netem_rate) },
784 [TCA_NETEM_LOSS] = { .type = NLA_NESTED },
785 [TCA_NETEM_ECN] = { .type = NLA_U32 },
786 [TCA_NETEM_RATE64] = { .type = NLA_U64 },
789 static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla,
790 const struct nla_policy *policy, int len)
792 int nested_len = nla_len(nla) - NLA_ALIGN(len);
794 if (nested_len < 0) {
795 pr_info("netem: invalid attributes len %d\n", nested_len);
799 if (nested_len >= nla_attr_size(0))
800 return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len),
803 memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1));
807 /* Parse netlink message to set options */
808 static int netem_change(struct Qdisc *sch, struct nlattr *opt)
810 struct netem_sched_data *q = qdisc_priv(sch);
811 struct nlattr *tb[TCA_NETEM_MAX + 1];
812 struct tc_netem_qopt *qopt;
818 qopt = nla_data(opt);
819 ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt));
823 sch->limit = qopt->limit;
825 q->latency = qopt->latency;
826 q->jitter = qopt->jitter;
827 q->limit = qopt->limit;
830 q->loss = qopt->loss;
831 q->duplicate = qopt->duplicate;
833 /* for compatibility with earlier versions.
834 * if gap is set, need to assume 100% probability
839 if (tb[TCA_NETEM_CORR])
840 get_correlation(sch, tb[TCA_NETEM_CORR]);
842 if (tb[TCA_NETEM_DELAY_DIST]) {
843 ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]);
848 if (tb[TCA_NETEM_REORDER])
849 get_reorder(sch, tb[TCA_NETEM_REORDER]);
851 if (tb[TCA_NETEM_CORRUPT])
852 get_corrupt(sch, tb[TCA_NETEM_CORRUPT]);
854 if (tb[TCA_NETEM_RATE])
855 get_rate(sch, tb[TCA_NETEM_RATE]);
857 if (tb[TCA_NETEM_RATE64])
858 q->rate = max_t(u64, q->rate,
859 nla_get_u64(tb[TCA_NETEM_RATE64]));
861 if (tb[TCA_NETEM_ECN])
862 q->ecn = nla_get_u32(tb[TCA_NETEM_ECN]);
864 q->loss_model = CLG_RANDOM;
865 if (tb[TCA_NETEM_LOSS])
866 ret = get_loss_clg(sch, tb[TCA_NETEM_LOSS]);
871 static int netem_init(struct Qdisc *sch, struct nlattr *opt)
873 struct netem_sched_data *q = qdisc_priv(sch);
879 qdisc_watchdog_init(&q->watchdog, sch);
881 q->loss_model = CLG_RANDOM;
882 ret = netem_change(sch, opt);
884 pr_info("netem: change failed\n");
888 static void netem_destroy(struct Qdisc *sch)
890 struct netem_sched_data *q = qdisc_priv(sch);
892 qdisc_watchdog_cancel(&q->watchdog);
894 qdisc_destroy(q->qdisc);
895 dist_free(q->delay_dist);
898 static int dump_loss_model(const struct netem_sched_data *q,
903 nest = nla_nest_start(skb, TCA_NETEM_LOSS);
905 goto nla_put_failure;
907 switch (q->loss_model) {
909 /* legacy loss model */
910 nla_nest_cancel(skb, nest);
911 return 0; /* no data */
914 struct tc_netem_gimodel gi = {
922 if (nla_put(skb, NETEM_LOSS_GI, sizeof(gi), &gi))
923 goto nla_put_failure;
927 struct tc_netem_gemodel ge = {
934 if (nla_put(skb, NETEM_LOSS_GE, sizeof(ge), &ge))
935 goto nla_put_failure;
940 nla_nest_end(skb, nest);
944 nla_nest_cancel(skb, nest);
948 static int netem_dump(struct Qdisc *sch, struct sk_buff *skb)
950 const struct netem_sched_data *q = qdisc_priv(sch);
951 struct nlattr *nla = (struct nlattr *) skb_tail_pointer(skb);
952 struct tc_netem_qopt qopt;
953 struct tc_netem_corr cor;
954 struct tc_netem_reorder reorder;
955 struct tc_netem_corrupt corrupt;
956 struct tc_netem_rate rate;
958 qopt.latency = q->latency;
959 qopt.jitter = q->jitter;
960 qopt.limit = q->limit;
963 qopt.duplicate = q->duplicate;
964 if (nla_put(skb, TCA_OPTIONS, sizeof(qopt), &qopt))
965 goto nla_put_failure;
967 cor.delay_corr = q->delay_cor.rho;
968 cor.loss_corr = q->loss_cor.rho;
969 cor.dup_corr = q->dup_cor.rho;
970 if (nla_put(skb, TCA_NETEM_CORR, sizeof(cor), &cor))
971 goto nla_put_failure;
973 reorder.probability = q->reorder;
974 reorder.correlation = q->reorder_cor.rho;
975 if (nla_put(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder))
976 goto nla_put_failure;
978 corrupt.probability = q->corrupt;
979 corrupt.correlation = q->corrupt_cor.rho;
980 if (nla_put(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt))
981 goto nla_put_failure;
983 if (q->rate >= (1ULL << 32)) {
984 if (nla_put_u64(skb, TCA_NETEM_RATE64, q->rate))
985 goto nla_put_failure;
990 rate.packet_overhead = q->packet_overhead;
991 rate.cell_size = q->cell_size;
992 rate.cell_overhead = q->cell_overhead;
993 if (nla_put(skb, TCA_NETEM_RATE, sizeof(rate), &rate))
994 goto nla_put_failure;
996 if (q->ecn && nla_put_u32(skb, TCA_NETEM_ECN, q->ecn))
997 goto nla_put_failure;
999 if (dump_loss_model(q, skb) != 0)
1000 goto nla_put_failure;
1002 return nla_nest_end(skb, nla);
1005 nlmsg_trim(skb, nla);
1009 static int netem_dump_class(struct Qdisc *sch, unsigned long cl,
1010 struct sk_buff *skb, struct tcmsg *tcm)
1012 struct netem_sched_data *q = qdisc_priv(sch);
1014 if (cl != 1 || !q->qdisc) /* only one class */
1017 tcm->tcm_handle |= TC_H_MIN(1);
1018 tcm->tcm_info = q->qdisc->handle;
1023 static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1026 struct netem_sched_data *q = qdisc_priv(sch);
1032 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
1035 sch_tree_unlock(sch);
1040 static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg)
1042 struct netem_sched_data *q = qdisc_priv(sch);
1046 static unsigned long netem_get(struct Qdisc *sch, u32 classid)
1051 static void netem_put(struct Qdisc *sch, unsigned long arg)
1055 static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker)
1057 if (!walker->stop) {
1058 if (walker->count >= walker->skip)
1059 if (walker->fn(sch, 1, walker) < 0) {
1067 static const struct Qdisc_class_ops netem_class_ops = {
1068 .graft = netem_graft,
1073 .dump = netem_dump_class,
1076 static struct Qdisc_ops netem_qdisc_ops __read_mostly = {
1078 .cl_ops = &netem_class_ops,
1079 .priv_size = sizeof(struct netem_sched_data),
1080 .enqueue = netem_enqueue,
1081 .dequeue = netem_dequeue,
1082 .peek = qdisc_peek_dequeued,
1085 .reset = netem_reset,
1086 .destroy = netem_destroy,
1087 .change = netem_change,
1089 .owner = THIS_MODULE,
1093 static int __init netem_module_init(void)
1095 pr_info("netem: version " VERSION "\n");
1096 return register_qdisc(&netem_qdisc_ops);
1098 static void __exit netem_module_exit(void)
1100 unregister_qdisc(&netem_qdisc_ops);
1102 module_init(netem_module_init)
1103 module_exit(netem_module_exit)
1104 MODULE_LICENSE("GPL");