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 = net_random();
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 */
187 value = net_random();
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 = net_random();
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) {
221 } else if (clg->a1 < 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 (net_random() < clg->a1)
269 if (net_random() < clg->a4)
272 if (net_random() < clg->a2)
274 if (clg->a3 > net_random())
281 static bool loss_event(struct netem_sched_data *q)
283 switch (q->loss_model) {
285 /* Random packet drop 0 => none, ~0 => all */
286 return q->loss && q->loss >= get_crandom(&q->loss_cor);
289 /* 4state loss model algorithm (used also for GI model)
290 * Extracts a value from the markov 4 state loss generator,
291 * if it is 1 drops a packet and if needed writes the event in
294 return loss_4state(q);
297 /* Gilbert-Elliot loss model algorithm
298 * Extracts a value from the Gilbert-Elliot loss generator,
299 * if it is 1 drops a packet and if needed writes the event in
302 return loss_gilb_ell(q);
305 return false; /* not reached */
309 /* tabledist - return a pseudo-randomly distributed value with mean mu and
310 * std deviation sigma. Uses table lookup to approximate the desired
311 * distribution, and a uniformly-distributed pseudo-random source.
313 static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma,
314 struct crndstate *state,
315 const struct disttable *dist)
324 rnd = get_crandom(state);
326 /* default uniform distribution */
328 return (rnd % (2*sigma)) - sigma + mu;
330 t = dist->table[rnd % dist->size];
331 x = (sigma % NETEM_DIST_SCALE) * t;
333 x += NETEM_DIST_SCALE/2;
335 x -= NETEM_DIST_SCALE/2;
337 return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu;
340 static psched_time_t packet_len_2_sched_time(unsigned int len, struct netem_sched_data *q)
344 len += q->packet_overhead;
347 u32 cells = reciprocal_divide(len, q->cell_size_reciprocal);
349 if (len > cells * q->cell_size) /* extra cell needed for remainder */
351 len = cells * (q->cell_size + q->cell_overhead);
354 ticks = (u64)len * NSEC_PER_SEC;
356 do_div(ticks, q->rate);
357 return PSCHED_NS2TICKS(ticks);
360 static void tfifo_reset(struct Qdisc *sch)
362 struct netem_sched_data *q = qdisc_priv(sch);
365 while ((p = rb_first(&q->t_root))) {
366 struct sk_buff *skb = netem_rb_to_skb(p);
368 rb_erase(p, &q->t_root);
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(netem_rb_node(nskb), parent, p);
392 rb_insert_color(netem_rb_node(nskb), &q->t_root);
397 * Insert one skb into qdisc.
398 * Note: parent depends on return value to account for queue length.
399 * NET_XMIT_DROP: queue length didn't change.
400 * NET_XMIT_SUCCESS: one skb was queued.
402 static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch)
404 struct netem_sched_data *q = qdisc_priv(sch);
405 /* We don't fill cb now as skb_unshare() may invalidate it */
406 struct netem_skb_cb *cb;
407 struct sk_buff *skb2;
410 /* Random duplication */
411 if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor))
416 if (q->ecn && INET_ECN_set_ce(skb))
417 sch->qstats.drops++; /* mark packet */
424 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
427 /* If a delay is expected, orphan the skb. (orphaning usually takes
428 * place at TX completion time, so _before_ the link transit delay)
430 if (q->latency || q->jitter)
431 skb_orphan_partial(skb);
434 * If we need to duplicate packet, then re-insert at top of the
435 * qdisc tree, since parent queuer expects that only one
436 * skb will be queued.
438 if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) {
439 struct Qdisc *rootq = qdisc_root(sch);
440 u32 dupsave = q->duplicate; /* prevent duplicating a dup... */
443 qdisc_enqueue_root(skb2, rootq);
444 q->duplicate = dupsave;
448 * Randomized packet corruption.
449 * Make copy if needed since we are modifying
450 * If packet is going to be hardware checksummed, then
451 * do it now in software before we mangle it.
453 if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) {
454 if (!(skb = skb_unshare(skb, GFP_ATOMIC)) ||
455 (skb->ip_summed == CHECKSUM_PARTIAL &&
456 skb_checksum_help(skb)))
457 return qdisc_drop(skb, sch);
459 skb->data[net_random() % skb_headlen(skb)] ^= 1<<(net_random() % 8);
462 if (unlikely(skb_queue_len(&sch->q) >= sch->limit))
463 return qdisc_reshape_fail(skb, sch);
465 sch->qstats.backlog += qdisc_pkt_len(skb);
467 cb = netem_skb_cb(skb);
468 if (q->gap == 0 || /* not doing reordering */
469 q->counter < q->gap - 1 || /* inside last reordering gap */
470 q->reorder < get_crandom(&q->reorder_cor)) {
472 psched_tdiff_t delay;
474 delay = tabledist(q->latency, q->jitter,
475 &q->delay_cor, q->delay_dist);
477 now = psched_get_time();
480 struct sk_buff *last;
482 if (!skb_queue_empty(&sch->q))
483 last = skb_peek_tail(&sch->q);
485 last = netem_rb_to_skb(rb_last(&q->t_root));
488 * Last packet in queue is reference point (now),
489 * calculate this time bonus and subtract
492 delay -= netem_skb_cb(last)->time_to_send - now;
493 delay = max_t(psched_tdiff_t, 0, delay);
494 now = netem_skb_cb(last)->time_to_send;
497 delay += packet_len_2_sched_time(skb->len, q);
500 cb->time_to_send = now + delay;
501 cb->tstamp_save = skb->tstamp;
503 tfifo_enqueue(skb, sch);
506 * Do re-ordering by putting one out of N packets at the front
509 cb->time_to_send = psched_get_time();
512 __skb_queue_head(&sch->q, skb);
513 sch->qstats.requeues++;
516 return NET_XMIT_SUCCESS;
519 static unsigned int netem_drop(struct Qdisc *sch)
521 struct netem_sched_data *q = qdisc_priv(sch);
524 len = qdisc_queue_drop(sch);
527 struct rb_node *p = rb_first(&q->t_root);
530 struct sk_buff *skb = netem_rb_to_skb(p);
532 rb_erase(p, &q->t_root);
536 len = qdisc_pkt_len(skb);
537 sch->qstats.backlog -= len;
541 if (!len && q->qdisc && q->qdisc->ops->drop)
542 len = q->qdisc->ops->drop(q->qdisc);
549 static struct sk_buff *netem_dequeue(struct Qdisc *sch)
551 struct netem_sched_data *q = qdisc_priv(sch);
555 if (qdisc_is_throttled(sch))
559 skb = __skb_dequeue(&sch->q);
562 sch->qstats.backlog -= qdisc_pkt_len(skb);
563 qdisc_unthrottled(sch);
564 qdisc_bstats_update(sch, skb);
567 p = rb_first(&q->t_root);
569 psched_time_t time_to_send;
571 skb = netem_rb_to_skb(p);
573 /* if more time remaining? */
574 time_to_send = netem_skb_cb(skb)->time_to_send;
575 if (time_to_send <= psched_get_time()) {
576 rb_erase(p, &q->t_root);
581 skb->tstamp = netem_skb_cb(skb)->tstamp_save;
583 #ifdef CONFIG_NET_CLS_ACT
585 * If it's at ingress let's pretend the delay is
586 * from the network (tstamp will be updated).
588 if (G_TC_FROM(skb->tc_verd) & AT_INGRESS)
589 skb->tstamp.tv64 = 0;
593 int err = qdisc_enqueue(skb, q->qdisc);
595 if (unlikely(err != NET_XMIT_SUCCESS)) {
596 if (net_xmit_drop_count(err)) {
598 qdisc_tree_decrease_qlen(sch, 1);
607 skb = q->qdisc->ops->dequeue(q->qdisc);
611 qdisc_watchdog_schedule(&q->watchdog, time_to_send);
615 skb = q->qdisc->ops->dequeue(q->qdisc);
622 static void netem_reset(struct Qdisc *sch)
624 struct netem_sched_data *q = qdisc_priv(sch);
626 qdisc_reset_queue(sch);
629 qdisc_reset(q->qdisc);
630 qdisc_watchdog_cancel(&q->watchdog);
633 static void dist_free(struct disttable *d)
636 if (is_vmalloc_addr(d))
644 * Distribution data is a variable size payload containing
645 * signed 16 bit values.
647 static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr)
649 struct netem_sched_data *q = qdisc_priv(sch);
650 size_t n = nla_len(attr)/sizeof(__s16);
651 const __s16 *data = nla_data(attr);
652 spinlock_t *root_lock;
657 if (n > NETEM_DIST_MAX)
660 s = sizeof(struct disttable) + n * sizeof(s16);
661 d = kmalloc(s, GFP_KERNEL | __GFP_NOWARN);
668 for (i = 0; i < n; i++)
669 d->table[i] = data[i];
671 root_lock = qdisc_root_sleeping_lock(sch);
673 spin_lock_bh(root_lock);
674 swap(q->delay_dist, d);
675 spin_unlock_bh(root_lock);
681 static void get_correlation(struct Qdisc *sch, const struct nlattr *attr)
683 struct netem_sched_data *q = qdisc_priv(sch);
684 const struct tc_netem_corr *c = nla_data(attr);
686 init_crandom(&q->delay_cor, c->delay_corr);
687 init_crandom(&q->loss_cor, c->loss_corr);
688 init_crandom(&q->dup_cor, c->dup_corr);
691 static void get_reorder(struct Qdisc *sch, const struct nlattr *attr)
693 struct netem_sched_data *q = qdisc_priv(sch);
694 const struct tc_netem_reorder *r = nla_data(attr);
696 q->reorder = r->probability;
697 init_crandom(&q->reorder_cor, r->correlation);
700 static void get_corrupt(struct Qdisc *sch, const struct nlattr *attr)
702 struct netem_sched_data *q = qdisc_priv(sch);
703 const struct tc_netem_corrupt *r = nla_data(attr);
705 q->corrupt = r->probability;
706 init_crandom(&q->corrupt_cor, r->correlation);
709 static void get_rate(struct Qdisc *sch, const struct nlattr *attr)
711 struct netem_sched_data *q = qdisc_priv(sch);
712 const struct tc_netem_rate *r = nla_data(attr);
715 q->packet_overhead = r->packet_overhead;
716 q->cell_size = r->cell_size;
718 q->cell_size_reciprocal = reciprocal_value(q->cell_size);
719 q->cell_overhead = r->cell_overhead;
722 static int get_loss_clg(struct Qdisc *sch, const struct nlattr *attr)
724 struct netem_sched_data *q = qdisc_priv(sch);
725 const struct nlattr *la;
728 nla_for_each_nested(la, attr, rem) {
729 u16 type = nla_type(la);
732 case NETEM_LOSS_GI: {
733 const struct tc_netem_gimodel *gi = nla_data(la);
735 if (nla_len(la) < sizeof(struct tc_netem_gimodel)) {
736 pr_info("netem: incorrect gi model size\n");
740 q->loss_model = CLG_4_STATES;
751 case NETEM_LOSS_GE: {
752 const struct tc_netem_gemodel *ge = nla_data(la);
754 if (nla_len(la) < sizeof(struct tc_netem_gemodel)) {
755 pr_info("netem: incorrect ge model size\n");
759 q->loss_model = CLG_GILB_ELL;
769 pr_info("netem: unknown loss type %u\n", type);
777 static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = {
778 [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) },
779 [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) },
780 [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) },
781 [TCA_NETEM_RATE] = { .len = sizeof(struct tc_netem_rate) },
782 [TCA_NETEM_LOSS] = { .type = NLA_NESTED },
783 [TCA_NETEM_ECN] = { .type = NLA_U32 },
786 static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla,
787 const struct nla_policy *policy, int len)
789 int nested_len = nla_len(nla) - NLA_ALIGN(len);
791 if (nested_len < 0) {
792 pr_info("netem: invalid attributes len %d\n", nested_len);
796 if (nested_len >= nla_attr_size(0))
797 return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len),
800 memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1));
804 /* Parse netlink message to set options */
805 static int netem_change(struct Qdisc *sch, struct nlattr *opt)
807 struct netem_sched_data *q = qdisc_priv(sch);
808 struct nlattr *tb[TCA_NETEM_MAX + 1];
809 struct tc_netem_qopt *qopt;
815 qopt = nla_data(opt);
816 ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt));
820 sch->limit = qopt->limit;
822 q->latency = qopt->latency;
823 q->jitter = qopt->jitter;
824 q->limit = qopt->limit;
827 q->loss = qopt->loss;
828 q->duplicate = qopt->duplicate;
830 /* for compatibility with earlier versions.
831 * if gap is set, need to assume 100% probability
836 if (tb[TCA_NETEM_CORR])
837 get_correlation(sch, tb[TCA_NETEM_CORR]);
839 if (tb[TCA_NETEM_DELAY_DIST]) {
840 ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]);
845 if (tb[TCA_NETEM_REORDER])
846 get_reorder(sch, tb[TCA_NETEM_REORDER]);
848 if (tb[TCA_NETEM_CORRUPT])
849 get_corrupt(sch, tb[TCA_NETEM_CORRUPT]);
851 if (tb[TCA_NETEM_RATE])
852 get_rate(sch, tb[TCA_NETEM_RATE]);
854 if (tb[TCA_NETEM_ECN])
855 q->ecn = nla_get_u32(tb[TCA_NETEM_ECN]);
857 q->loss_model = CLG_RANDOM;
858 if (tb[TCA_NETEM_LOSS])
859 ret = get_loss_clg(sch, tb[TCA_NETEM_LOSS]);
864 static int netem_init(struct Qdisc *sch, struct nlattr *opt)
866 struct netem_sched_data *q = qdisc_priv(sch);
872 qdisc_watchdog_init(&q->watchdog, sch);
874 q->loss_model = CLG_RANDOM;
875 ret = netem_change(sch, opt);
877 pr_info("netem: change failed\n");
881 static void netem_destroy(struct Qdisc *sch)
883 struct netem_sched_data *q = qdisc_priv(sch);
885 qdisc_watchdog_cancel(&q->watchdog);
887 qdisc_destroy(q->qdisc);
888 dist_free(q->delay_dist);
891 static int dump_loss_model(const struct netem_sched_data *q,
896 nest = nla_nest_start(skb, TCA_NETEM_LOSS);
898 goto nla_put_failure;
900 switch (q->loss_model) {
902 /* legacy loss model */
903 nla_nest_cancel(skb, nest);
904 return 0; /* no data */
907 struct tc_netem_gimodel gi = {
915 if (nla_put(skb, NETEM_LOSS_GI, sizeof(gi), &gi))
916 goto nla_put_failure;
920 struct tc_netem_gemodel ge = {
927 if (nla_put(skb, NETEM_LOSS_GE, sizeof(ge), &ge))
928 goto nla_put_failure;
933 nla_nest_end(skb, nest);
937 nla_nest_cancel(skb, nest);
941 static int netem_dump(struct Qdisc *sch, struct sk_buff *skb)
943 const struct netem_sched_data *q = qdisc_priv(sch);
944 struct nlattr *nla = (struct nlattr *) skb_tail_pointer(skb);
945 struct tc_netem_qopt qopt;
946 struct tc_netem_corr cor;
947 struct tc_netem_reorder reorder;
948 struct tc_netem_corrupt corrupt;
949 struct tc_netem_rate rate;
951 qopt.latency = q->latency;
952 qopt.jitter = q->jitter;
953 qopt.limit = q->limit;
956 qopt.duplicate = q->duplicate;
957 if (nla_put(skb, TCA_OPTIONS, sizeof(qopt), &qopt))
958 goto nla_put_failure;
960 cor.delay_corr = q->delay_cor.rho;
961 cor.loss_corr = q->loss_cor.rho;
962 cor.dup_corr = q->dup_cor.rho;
963 if (nla_put(skb, TCA_NETEM_CORR, sizeof(cor), &cor))
964 goto nla_put_failure;
966 reorder.probability = q->reorder;
967 reorder.correlation = q->reorder_cor.rho;
968 if (nla_put(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder))
969 goto nla_put_failure;
971 corrupt.probability = q->corrupt;
972 corrupt.correlation = q->corrupt_cor.rho;
973 if (nla_put(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt))
974 goto nla_put_failure;
977 rate.packet_overhead = q->packet_overhead;
978 rate.cell_size = q->cell_size;
979 rate.cell_overhead = q->cell_overhead;
980 if (nla_put(skb, TCA_NETEM_RATE, sizeof(rate), &rate))
981 goto nla_put_failure;
983 if (q->ecn && nla_put_u32(skb, TCA_NETEM_ECN, q->ecn))
984 goto nla_put_failure;
986 if (dump_loss_model(q, skb) != 0)
987 goto nla_put_failure;
989 return nla_nest_end(skb, nla);
992 nlmsg_trim(skb, nla);
996 static int netem_dump_class(struct Qdisc *sch, unsigned long cl,
997 struct sk_buff *skb, struct tcmsg *tcm)
999 struct netem_sched_data *q = qdisc_priv(sch);
1001 if (cl != 1 || !q->qdisc) /* only one class */
1004 tcm->tcm_handle |= TC_H_MIN(1);
1005 tcm->tcm_info = q->qdisc->handle;
1010 static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1013 struct netem_sched_data *q = qdisc_priv(sch);
1019 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
1022 sch_tree_unlock(sch);
1027 static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg)
1029 struct netem_sched_data *q = qdisc_priv(sch);
1033 static unsigned long netem_get(struct Qdisc *sch, u32 classid)
1038 static void netem_put(struct Qdisc *sch, unsigned long arg)
1042 static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker)
1044 if (!walker->stop) {
1045 if (walker->count >= walker->skip)
1046 if (walker->fn(sch, 1, walker) < 0) {
1054 static const struct Qdisc_class_ops netem_class_ops = {
1055 .graft = netem_graft,
1060 .dump = netem_dump_class,
1063 static struct Qdisc_ops netem_qdisc_ops __read_mostly = {
1065 .cl_ops = &netem_class_ops,
1066 .priv_size = sizeof(struct netem_sched_data),
1067 .enqueue = netem_enqueue,
1068 .dequeue = netem_dequeue,
1069 .peek = qdisc_peek_dequeued,
1072 .reset = netem_reset,
1073 .destroy = netem_destroy,
1074 .change = netem_change,
1076 .owner = THIS_MODULE,
1080 static int __init netem_module_init(void)
1082 pr_info("netem: version " VERSION "\n");
1083 return register_qdisc(&netem_qdisc_ops);
1085 static void __exit netem_module_exit(void)
1087 unregister_qdisc(&netem_qdisc_ops);
1089 module_init(netem_module_init)
1090 module_exit(netem_module_exit)
1091 MODULE_LICENSE("GPL");