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);
371 rtnl_kfree_skbs(skb, skb);
375 static void tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch)
377 struct netem_sched_data *q = qdisc_priv(sch);
378 psched_time_t tnext = netem_skb_cb(nskb)->time_to_send;
379 struct rb_node **p = &q->t_root.rb_node, *parent = NULL;
385 skb = netem_rb_to_skb(parent);
386 if (tnext >= netem_skb_cb(skb)->time_to_send)
387 p = &parent->rb_right;
389 p = &parent->rb_left;
391 rb_link_node(&nskb->rbnode, parent, p);
392 rb_insert_color(&nskb->rbnode, &q->t_root);
396 /* netem can't properly corrupt a megapacket (like we get from GSO), so instead
397 * when we statistically choose to corrupt one, we instead segment it, returning
398 * the first packet to be corrupted, and re-enqueue the remaining frames
400 static struct sk_buff *netem_segment(struct sk_buff *skb, struct Qdisc *sch,
401 struct sk_buff **to_free)
403 struct sk_buff *segs;
404 netdev_features_t features = netif_skb_features(skb);
406 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
408 if (IS_ERR_OR_NULL(segs)) {
409 qdisc_drop(skb, sch, to_free);
416 static void netem_enqueue_skb_head(struct qdisc_skb_head *qh, struct sk_buff *skb)
418 skb->next = qh->head;
427 * Insert one skb into qdisc.
428 * Note: parent depends on return value to account for queue length.
429 * NET_XMIT_DROP: queue length didn't change.
430 * NET_XMIT_SUCCESS: one skb was queued.
432 static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch,
433 struct sk_buff **to_free)
435 struct netem_sched_data *q = qdisc_priv(sch);
436 /* We don't fill cb now as skb_unshare() may invalidate it */
437 struct netem_skb_cb *cb;
438 struct sk_buff *skb2;
439 struct sk_buff *segs = NULL;
440 unsigned int prev_len = qdisc_pkt_len(skb);
442 int rc = NET_XMIT_SUCCESS;
443 int rc_drop = NET_XMIT_DROP;
445 /* Do not fool qdisc_drop_all() */
448 /* Random duplication */
449 if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor))
454 if (q->ecn && INET_ECN_set_ce(skb))
455 qdisc_qstats_drop(sch); /* mark packet */
460 qdisc_qstats_drop(sch);
461 __qdisc_drop(skb, to_free);
462 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
465 /* If a delay is expected, orphan the skb. (orphaning usually takes
466 * place at TX completion time, so _before_ the link transit delay)
468 if (q->latency || q->jitter || q->rate)
469 skb_orphan_partial(skb);
472 * If we need to duplicate packet, then re-insert at top of the
473 * qdisc tree, since parent queuer expects that only one
474 * skb will be queued.
476 if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) {
477 struct Qdisc *rootq = qdisc_root_bh(sch);
478 u32 dupsave = q->duplicate; /* prevent duplicating a dup... */
481 rootq->enqueue(skb2, rootq, to_free);
482 q->duplicate = dupsave;
483 rc_drop = NET_XMIT_SUCCESS;
487 * Randomized packet corruption.
488 * Make copy if needed since we are modifying
489 * If packet is going to be hardware checksummed, then
490 * do it now in software before we mangle it.
492 if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) {
493 if (skb_is_gso(skb)) {
494 segs = netem_segment(skb, sch, to_free);
497 qdisc_skb_cb(segs)->pkt_len = segs->len;
505 skb = skb_unshare(skb, GFP_ATOMIC);
506 if (unlikely(!skb)) {
507 qdisc_qstats_drop(sch);
510 if (skb->ip_summed == CHECKSUM_PARTIAL &&
511 skb_checksum_help(skb)) {
512 qdisc_drop(skb, sch, to_free);
517 skb->data[prandom_u32() % skb_headlen(skb)] ^=
518 1<<(prandom_u32() % 8);
521 if (unlikely(sch->q.qlen >= sch->limit)) {
522 qdisc_drop_all(skb, sch, to_free);
526 qdisc_qstats_backlog_inc(sch, skb);
528 cb = netem_skb_cb(skb);
529 if (q->gap == 0 || /* not doing reordering */
530 q->counter < q->gap - 1 || /* inside last reordering gap */
531 q->reorder < get_crandom(&q->reorder_cor)) {
533 psched_tdiff_t delay;
535 delay = tabledist(q->latency, q->jitter,
536 &q->delay_cor, q->delay_dist);
538 now = psched_get_time();
541 struct netem_skb_cb *last = NULL;
544 last = netem_skb_cb(sch->q.tail);
545 if (q->t_root.rb_node) {
546 struct sk_buff *t_skb;
547 struct netem_skb_cb *t_last;
549 t_skb = netem_rb_to_skb(rb_last(&q->t_root));
550 t_last = netem_skb_cb(t_skb);
552 t_last->time_to_send > last->time_to_send) {
559 * Last packet in queue is reference point (now),
560 * calculate this time bonus and subtract
563 delay -= last->time_to_send - now;
564 delay = max_t(psched_tdiff_t, 0, delay);
565 now = last->time_to_send;
568 delay += packet_len_2_sched_time(qdisc_pkt_len(skb), q);
571 cb->time_to_send = now + delay;
572 cb->tstamp_save = skb->tstamp;
574 tfifo_enqueue(skb, sch);
577 * Do re-ordering by putting one out of N packets at the front
580 cb->time_to_send = psched_get_time();
583 netem_enqueue_skb_head(&sch->q, skb);
584 sch->qstats.requeues++;
589 unsigned int len, last_len;
592 len = skb ? skb->len : 0;
598 qdisc_skb_cb(segs)->pkt_len = segs->len;
599 last_len = segs->len;
600 rc = qdisc_enqueue(segs, sch, to_free);
601 if (rc != NET_XMIT_SUCCESS) {
602 if (net_xmit_drop_count(rc))
603 qdisc_qstats_drop(sch);
610 /* Parent qdiscs accounted for 1 skb of size @prev_len */
611 qdisc_tree_reduce_backlog(sch, -(nb - 1), -(len - prev_len));
613 return NET_XMIT_DROP;
615 return NET_XMIT_SUCCESS;
618 static struct sk_buff *netem_dequeue(struct Qdisc *sch)
620 struct netem_sched_data *q = qdisc_priv(sch);
625 skb = __qdisc_dequeue_head(&sch->q);
627 qdisc_qstats_backlog_dec(sch, skb);
629 qdisc_bstats_update(sch, skb);
632 p = rb_first(&q->t_root);
634 psched_time_t time_to_send;
636 skb = netem_rb_to_skb(p);
638 /* if more time remaining? */
639 time_to_send = netem_skb_cb(skb)->time_to_send;
640 if (time_to_send <= psched_get_time()) {
641 rb_erase(p, &q->t_root);
644 qdisc_qstats_backlog_dec(sch, skb);
647 skb->tstamp = netem_skb_cb(skb)->tstamp_save;
649 #ifdef CONFIG_NET_CLS_ACT
651 * If it's at ingress let's pretend the delay is
652 * from the network (tstamp will be updated).
654 if (G_TC_FROM(skb->tc_verd) & AT_INGRESS)
655 skb->tstamp.tv64 = 0;
659 unsigned int pkt_len = qdisc_pkt_len(skb);
660 struct sk_buff *to_free = NULL;
663 err = qdisc_enqueue(skb, q->qdisc, &to_free);
664 kfree_skb_list(to_free);
665 if (err != NET_XMIT_SUCCESS &&
666 net_xmit_drop_count(err)) {
667 qdisc_qstats_drop(sch);
668 qdisc_tree_reduce_backlog(sch, 1,
677 skb = q->qdisc->ops->dequeue(q->qdisc);
681 qdisc_watchdog_schedule(&q->watchdog, time_to_send);
685 skb = q->qdisc->ops->dequeue(q->qdisc);
692 static void netem_reset(struct Qdisc *sch)
694 struct netem_sched_data *q = qdisc_priv(sch);
696 qdisc_reset_queue(sch);
699 qdisc_reset(q->qdisc);
700 qdisc_watchdog_cancel(&q->watchdog);
703 static void dist_free(struct disttable *d)
709 * Distribution data is a variable size payload containing
710 * signed 16 bit values.
712 static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr)
714 struct netem_sched_data *q = qdisc_priv(sch);
715 size_t n = nla_len(attr)/sizeof(__s16);
716 const __s16 *data = nla_data(attr);
717 spinlock_t *root_lock;
722 if (!n || n > NETEM_DIST_MAX)
725 s = sizeof(struct disttable) + n * sizeof(s16);
726 d = kmalloc(s, GFP_KERNEL | __GFP_NOWARN);
733 for (i = 0; i < n; i++)
734 d->table[i] = data[i];
736 root_lock = qdisc_root_sleeping_lock(sch);
738 spin_lock_bh(root_lock);
739 swap(q->delay_dist, d);
740 spin_unlock_bh(root_lock);
746 static void get_correlation(struct netem_sched_data *q, const struct nlattr *attr)
748 const struct tc_netem_corr *c = nla_data(attr);
750 init_crandom(&q->delay_cor, c->delay_corr);
751 init_crandom(&q->loss_cor, c->loss_corr);
752 init_crandom(&q->dup_cor, c->dup_corr);
755 static void get_reorder(struct netem_sched_data *q, const struct nlattr *attr)
757 const struct tc_netem_reorder *r = nla_data(attr);
759 q->reorder = r->probability;
760 init_crandom(&q->reorder_cor, r->correlation);
763 static void get_corrupt(struct netem_sched_data *q, const struct nlattr *attr)
765 const struct tc_netem_corrupt *r = nla_data(attr);
767 q->corrupt = r->probability;
768 init_crandom(&q->corrupt_cor, r->correlation);
771 static void get_rate(struct netem_sched_data *q, const struct nlattr *attr)
773 const struct tc_netem_rate *r = nla_data(attr);
776 q->packet_overhead = r->packet_overhead;
777 q->cell_size = r->cell_size;
778 q->cell_overhead = r->cell_overhead;
780 q->cell_size_reciprocal = reciprocal_value(q->cell_size);
782 q->cell_size_reciprocal = (struct reciprocal_value) { 0 };
785 static int get_loss_clg(struct netem_sched_data *q, const struct nlattr *attr)
787 const struct nlattr *la;
790 nla_for_each_nested(la, attr, rem) {
791 u16 type = nla_type(la);
794 case NETEM_LOSS_GI: {
795 const struct tc_netem_gimodel *gi = nla_data(la);
797 if (nla_len(la) < sizeof(struct tc_netem_gimodel)) {
798 pr_info("netem: incorrect gi model size\n");
802 q->loss_model = CLG_4_STATES;
804 q->clg.state = TX_IN_GAP_PERIOD;
813 case NETEM_LOSS_GE: {
814 const struct tc_netem_gemodel *ge = nla_data(la);
816 if (nla_len(la) < sizeof(struct tc_netem_gemodel)) {
817 pr_info("netem: incorrect ge model size\n");
821 q->loss_model = CLG_GILB_ELL;
822 q->clg.state = GOOD_STATE;
831 pr_info("netem: unknown loss type %u\n", type);
839 static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = {
840 [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) },
841 [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) },
842 [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) },
843 [TCA_NETEM_RATE] = { .len = sizeof(struct tc_netem_rate) },
844 [TCA_NETEM_LOSS] = { .type = NLA_NESTED },
845 [TCA_NETEM_ECN] = { .type = NLA_U32 },
846 [TCA_NETEM_RATE64] = { .type = NLA_U64 },
849 static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla,
850 const struct nla_policy *policy, int len)
852 int nested_len = nla_len(nla) - NLA_ALIGN(len);
854 if (nested_len < 0) {
855 pr_info("netem: invalid attributes len %d\n", nested_len);
859 if (nested_len >= nla_attr_size(0))
860 return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len),
863 memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1));
867 /* Parse netlink message to set options */
868 static int netem_change(struct Qdisc *sch, struct nlattr *opt)
870 struct netem_sched_data *q = qdisc_priv(sch);
871 struct nlattr *tb[TCA_NETEM_MAX + 1];
872 struct tc_netem_qopt *qopt;
873 struct clgstate old_clg;
874 int old_loss_model = CLG_RANDOM;
880 qopt = nla_data(opt);
881 ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt));
885 /* backup q->clg and q->loss_model */
887 old_loss_model = q->loss_model;
889 if (tb[TCA_NETEM_LOSS]) {
890 ret = get_loss_clg(q, tb[TCA_NETEM_LOSS]);
892 q->loss_model = old_loss_model;
896 q->loss_model = CLG_RANDOM;
899 if (tb[TCA_NETEM_DELAY_DIST]) {
900 ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]);
902 /* recover clg and loss_model, in case of
903 * q->clg and q->loss_model were modified
907 q->loss_model = old_loss_model;
912 sch->limit = qopt->limit;
914 q->latency = qopt->latency;
915 q->jitter = qopt->jitter;
916 q->limit = qopt->limit;
919 q->loss = qopt->loss;
920 q->duplicate = qopt->duplicate;
922 /* for compatibility with earlier versions.
923 * if gap is set, need to assume 100% probability
928 if (tb[TCA_NETEM_CORR])
929 get_correlation(q, tb[TCA_NETEM_CORR]);
931 if (tb[TCA_NETEM_REORDER])
932 get_reorder(q, tb[TCA_NETEM_REORDER]);
934 if (tb[TCA_NETEM_CORRUPT])
935 get_corrupt(q, tb[TCA_NETEM_CORRUPT]);
937 if (tb[TCA_NETEM_RATE])
938 get_rate(q, tb[TCA_NETEM_RATE]);
940 if (tb[TCA_NETEM_RATE64])
941 q->rate = max_t(u64, q->rate,
942 nla_get_u64(tb[TCA_NETEM_RATE64]));
944 if (tb[TCA_NETEM_ECN])
945 q->ecn = nla_get_u32(tb[TCA_NETEM_ECN]);
950 static int netem_init(struct Qdisc *sch, struct nlattr *opt)
952 struct netem_sched_data *q = qdisc_priv(sch);
955 qdisc_watchdog_init(&q->watchdog, sch);
960 q->loss_model = CLG_RANDOM;
961 ret = netem_change(sch, opt);
963 pr_info("netem: change failed\n");
967 static void netem_destroy(struct Qdisc *sch)
969 struct netem_sched_data *q = qdisc_priv(sch);
971 qdisc_watchdog_cancel(&q->watchdog);
973 qdisc_destroy(q->qdisc);
974 dist_free(q->delay_dist);
977 static int dump_loss_model(const struct netem_sched_data *q,
982 nest = nla_nest_start(skb, TCA_NETEM_LOSS);
984 goto nla_put_failure;
986 switch (q->loss_model) {
988 /* legacy loss model */
989 nla_nest_cancel(skb, nest);
990 return 0; /* no data */
993 struct tc_netem_gimodel gi = {
1001 if (nla_put(skb, NETEM_LOSS_GI, sizeof(gi), &gi))
1002 goto nla_put_failure;
1005 case CLG_GILB_ELL: {
1006 struct tc_netem_gemodel ge = {
1013 if (nla_put(skb, NETEM_LOSS_GE, sizeof(ge), &ge))
1014 goto nla_put_failure;
1019 nla_nest_end(skb, nest);
1023 nla_nest_cancel(skb, nest);
1027 static int netem_dump(struct Qdisc *sch, struct sk_buff *skb)
1029 const struct netem_sched_data *q = qdisc_priv(sch);
1030 struct nlattr *nla = (struct nlattr *) skb_tail_pointer(skb);
1031 struct tc_netem_qopt qopt;
1032 struct tc_netem_corr cor;
1033 struct tc_netem_reorder reorder;
1034 struct tc_netem_corrupt corrupt;
1035 struct tc_netem_rate rate;
1037 qopt.latency = q->latency;
1038 qopt.jitter = q->jitter;
1039 qopt.limit = q->limit;
1040 qopt.loss = q->loss;
1042 qopt.duplicate = q->duplicate;
1043 if (nla_put(skb, TCA_OPTIONS, sizeof(qopt), &qopt))
1044 goto nla_put_failure;
1046 cor.delay_corr = q->delay_cor.rho;
1047 cor.loss_corr = q->loss_cor.rho;
1048 cor.dup_corr = q->dup_cor.rho;
1049 if (nla_put(skb, TCA_NETEM_CORR, sizeof(cor), &cor))
1050 goto nla_put_failure;
1052 reorder.probability = q->reorder;
1053 reorder.correlation = q->reorder_cor.rho;
1054 if (nla_put(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder))
1055 goto nla_put_failure;
1057 corrupt.probability = q->corrupt;
1058 corrupt.correlation = q->corrupt_cor.rho;
1059 if (nla_put(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt))
1060 goto nla_put_failure;
1062 if (q->rate >= (1ULL << 32)) {
1063 if (nla_put_u64_64bit(skb, TCA_NETEM_RATE64, q->rate,
1065 goto nla_put_failure;
1068 rate.rate = q->rate;
1070 rate.packet_overhead = q->packet_overhead;
1071 rate.cell_size = q->cell_size;
1072 rate.cell_overhead = q->cell_overhead;
1073 if (nla_put(skb, TCA_NETEM_RATE, sizeof(rate), &rate))
1074 goto nla_put_failure;
1076 if (q->ecn && nla_put_u32(skb, TCA_NETEM_ECN, q->ecn))
1077 goto nla_put_failure;
1079 if (dump_loss_model(q, skb) != 0)
1080 goto nla_put_failure;
1082 return nla_nest_end(skb, nla);
1085 nlmsg_trim(skb, nla);
1089 static int netem_dump_class(struct Qdisc *sch, unsigned long cl,
1090 struct sk_buff *skb, struct tcmsg *tcm)
1092 struct netem_sched_data *q = qdisc_priv(sch);
1094 if (cl != 1 || !q->qdisc) /* only one class */
1097 tcm->tcm_handle |= TC_H_MIN(1);
1098 tcm->tcm_info = q->qdisc->handle;
1103 static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1106 struct netem_sched_data *q = qdisc_priv(sch);
1108 *old = qdisc_replace(sch, new, &q->qdisc);
1112 static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg)
1114 struct netem_sched_data *q = qdisc_priv(sch);
1118 static unsigned long netem_get(struct Qdisc *sch, u32 classid)
1123 static void netem_put(struct Qdisc *sch, unsigned long arg)
1127 static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker)
1129 if (!walker->stop) {
1130 if (walker->count >= walker->skip)
1131 if (walker->fn(sch, 1, walker) < 0) {
1139 static const struct Qdisc_class_ops netem_class_ops = {
1140 .graft = netem_graft,
1145 .dump = netem_dump_class,
1148 static struct Qdisc_ops netem_qdisc_ops __read_mostly = {
1150 .cl_ops = &netem_class_ops,
1151 .priv_size = sizeof(struct netem_sched_data),
1152 .enqueue = netem_enqueue,
1153 .dequeue = netem_dequeue,
1154 .peek = qdisc_peek_dequeued,
1156 .reset = netem_reset,
1157 .destroy = netem_destroy,
1158 .change = netem_change,
1160 .owner = THIS_MODULE,
1164 static int __init netem_module_init(void)
1166 pr_info("netem: version " VERSION "\n");
1167 return register_qdisc(&netem_qdisc_ops);
1169 static void __exit netem_module_exit(void)
1171 unregister_qdisc(&netem_qdisc_ops);
1173 module_init(netem_module_init)
1174 module_exit(netem_module_exit)
1175 MODULE_LICENSE("GPL");