projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[deliverable/linux.git] / net / sched / sch_netem.c
1 /*
2 * net/sched/sch_netem.c Network emulator
3 *
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
7 * 2 of the License.
8 *
9 * Many of the algorithms and ideas for this came from
10 * NIST Net which is not copyrighted.
11 *
12 * Authors: Stephen Hemminger <shemminger@osdl.org>
13 * Catalin(ux aka Dino) BOIE <catab at umbrella dot ro>
14 */
15
16 #include <linux/mm.h>
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
26 #include <net/netlink.h>
27 #include <net/pkt_sched.h>
28
29 #define VERSION "1.3"
30
31 /* Network Emulation Queuing algorithm.
32 ====================================
33
34 Sources: [1] Mark Carson, Darrin Santay, "NIST Net - A Linux-based
35 Network Emulation Tool
36 [2] Luigi Rizzo, DummyNet for FreeBSD
37
38 ----------------------------------------------------------------
39
40 This started out as a simple way to delay outgoing packets to
41 test TCP but has grown to include most of the functionality
42 of a full blown network emulator like NISTnet. It can delay
43 packets and add random jitter (and correlation). The random
44 distribution can be loaded from a table as well to provide
45 normal, Pareto, or experimental curves. Packet loss,
46 duplication, and reordering can also be emulated.
47
48 This qdisc does not do classification that can be handled in
49 layering other disciplines. It does not need to do bandwidth
50 control either since that can be handled by using token
51 bucket or other rate control.
52
53 Correlated Loss Generator models
54
55 Added generation of correlated loss according to the
56 "Gilbert-Elliot" model, a 4-state markov model.
57
58 References:
59 [1] NetemCLG Home http://netgroup.uniroma2.it/NetemCLG
60 [2] S. Salsano, F. Ludovici, A. Ordine, "Definition of a general
61 and intuitive loss model for packet networks and its implementation
62 in the Netem module in the Linux kernel", available in [1]
63
64 Authors: Stefano Salsano <stefano.salsano at uniroma2.it
65 Fabio Ludovici <fabio.ludovici at yahoo.it>
66 */
67
68 struct netem_sched_data {
69 struct Qdisc *qdisc;
70 struct qdisc_watchdog watchdog;
71
72 psched_tdiff_t latency;
73 psched_tdiff_t jitter;
74
75 u32 loss;
76 u32 limit;
77 u32 counter;
78 u32 gap;
79 u32 duplicate;
80 u32 reorder;
81 u32 corrupt;
82
83 struct crndstate {
84 u32 last;
85 u32 rho;
86 } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor;
87
88 struct disttable {
89 u32 size;
90 s16 table[0];
91 } *delay_dist;
92
93 enum {
94 CLG_RANDOM,
95 CLG_4_STATES,
96 CLG_GILB_ELL,
97 } loss_model;
98
99 /* Correlated Loss Generation models */
100 struct clgstate {
101 /* state of the Markov chain */
102 u8 state;
103
104 /* 4-states and Gilbert-Elliot models */
105 u32 a1; /* p13 for 4-states or p for GE */
106 u32 a2; /* p31 for 4-states or r for GE */
107 u32 a3; /* p32 for 4-states or h for GE */
108 u32 a4; /* p14 for 4-states or 1-k for GE */
109 u32 a5; /* p23 used only in 4-states */
110 } clg;
111
112 };
113
114 /* Time stamp put into socket buffer control block */
115 struct netem_skb_cb {
116 psched_time_t time_to_send;
117 };
118
119 static inline struct netem_skb_cb *netem_skb_cb(struct sk_buff *skb)
120 {
121 BUILD_BUG_ON(sizeof(skb->cb) <
122 sizeof(struct qdisc_skb_cb) + sizeof(struct netem_skb_cb));
123 return (struct netem_skb_cb *)qdisc_skb_cb(skb)->data;
124 }
125
126 /* init_crandom - initialize correlated random number generator
127 * Use entropy source for initial seed.
128 */
129 static void init_crandom(struct crndstate *state, unsigned long rho)
130 {
131 state->rho = rho;
132 state->last = net_random();
133 }
134
135 /* get_crandom - correlated random number generator
136 * Next number depends on last value.
137 * rho is scaled to avoid floating point.
138 */
139 static u32 get_crandom(struct crndstate *state)
140 {
141 u64 value, rho;
142 unsigned long answer;
143
144 if (state->rho == 0) /* no correlation */
145 return net_random();
146
147 value = net_random();
148 rho = (u64)state->rho + 1;
149 answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32;
150 state->last = answer;
151 return answer;
152 }
153
154 /* loss_4state - 4-state model loss generator
155 * Generates losses according to the 4-state Markov chain adopted in
156 * the GI (General and Intuitive) loss model.
157 */
158 static bool loss_4state(struct netem_sched_data *q)
159 {
160 struct clgstate *clg = &q->clg;
161 u32 rnd = net_random();
162
163 /*
164 * Makes a comparison between rnd and the transition
165 * probabilities outgoing from the current state, then decides the
166 * next state and if the next packet has to be transmitted or lost.
167 * The four states correspond to:
168 * 1 => successfully transmitted packets within a gap period
169 * 4 => isolated losses within a gap period
170 * 3 => lost packets within a burst period
171 * 2 => successfully transmitted packets within a burst period
172 */
173 switch (clg->state) {
174 case 1:
175 if (rnd < clg->a4) {
176 clg->state = 4;
177 return true;
178 } else if (clg->a4 < rnd && rnd < clg->a1) {
179 clg->state = 3;
180 return true;
181 } else if (clg->a1 < rnd)
182 clg->state = 1;
183
184 break;
185 case 2:
186 if (rnd < clg->a5) {
187 clg->state = 3;
188 return true;
189 } else
190 clg->state = 2;
191
192 break;
193 case 3:
194 if (rnd < clg->a3)
195 clg->state = 2;
196 else if (clg->a3 < rnd && rnd < clg->a2 + clg->a3) {
197 clg->state = 1;
198 return true;
199 } else if (clg->a2 + clg->a3 < rnd) {
200 clg->state = 3;
201 return true;
202 }
203 break;
204 case 4:
205 clg->state = 1;
206 break;
207 }
208
209 return false;
210 }
211
212 /* loss_gilb_ell - Gilbert-Elliot model loss generator
213 * Generates losses according to the Gilbert-Elliot loss model or
214 * its special cases (Gilbert or Simple Gilbert)
215 *
216 * Makes a comparison between random number and the transition
217 * probabilities outgoing from the current state, then decides the
218 * next state. A second random number is extracted and the comparison
219 * with the loss probability of the current state decides if the next
220 * packet will be transmitted or lost.
221 */
222 static bool loss_gilb_ell(struct netem_sched_data *q)
223 {
224 struct clgstate *clg = &q->clg;
225
226 switch (clg->state) {
227 case 1:
228 if (net_random() < clg->a1)
229 clg->state = 2;
230 if (net_random() < clg->a4)
231 return true;
232 case 2:
233 if (net_random() < clg->a2)
234 clg->state = 1;
235 if (clg->a3 > net_random())
236 return true;
237 }
238
239 return false;
240 }
241
242 static bool loss_event(struct netem_sched_data *q)
243 {
244 switch (q->loss_model) {
245 case CLG_RANDOM:
246 /* Random packet drop 0 => none, ~0 => all */
247 return q->loss && q->loss >= get_crandom(&q->loss_cor);
248
249 case CLG_4_STATES:
250 /* 4state loss model algorithm (used also for GI model)
251 * Extracts a value from the markov 4 state loss generator,
252 * if it is 1 drops a packet and if needed writes the event in
253 * the kernel logs
254 */
255 return loss_4state(q);
256
257 case CLG_GILB_ELL:
258 /* Gilbert-Elliot loss model algorithm
259 * Extracts a value from the Gilbert-Elliot loss generator,
260 * if it is 1 drops a packet and if needed writes the event in
261 * the kernel logs
262 */
263 return loss_gilb_ell(q);
264 }
265
266 return false; /* not reached */
267 }
268
269
270 /* tabledist - return a pseudo-randomly distributed value with mean mu and
271 * std deviation sigma. Uses table lookup to approximate the desired
272 * distribution, and a uniformly-distributed pseudo-random source.
273 */
274 static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma,
275 struct crndstate *state,
276 const struct disttable *dist)
277 {
278 psched_tdiff_t x;
279 long t;
280 u32 rnd;
281
282 if (sigma == 0)
283 return mu;
284
285 rnd = get_crandom(state);
286
287 /* default uniform distribution */
288 if (dist == NULL)
289 return (rnd % (2*sigma)) - sigma + mu;
290
291 t = dist->table[rnd % dist->size];
292 x = (sigma % NETEM_DIST_SCALE) * t;
293 if (x >= 0)
294 x += NETEM_DIST_SCALE/2;
295 else
296 x -= NETEM_DIST_SCALE/2;
297
298 return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu;
299 }
300
301 /*
302 * Insert one skb into qdisc.
303 * Note: parent depends on return value to account for queue length.
304 * NET_XMIT_DROP: queue length didn't change.
305 * NET_XMIT_SUCCESS: one skb was queued.
306 */
307 static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch)
308 {
309 struct netem_sched_data *q = qdisc_priv(sch);
310 /* We don't fill cb now as skb_unshare() may invalidate it */
311 struct netem_skb_cb *cb;
312 struct sk_buff *skb2;
313 int ret;
314 int count = 1;
315
316 /* Random duplication */
317 if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor))
318 ++count;
319
320 /* Drop packet? */
321 if (loss_event(q))
322 --count;
323
324 if (count == 0) {
325 sch->qstats.drops++;
326 kfree_skb(skb);
327 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
328 }
329
330 skb_orphan(skb);
331
332 /*
333 * If we need to duplicate packet, then re-insert at top of the
334 * qdisc tree, since parent queuer expects that only one
335 * skb will be queued.
336 */
337 if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) {
338 struct Qdisc *rootq = qdisc_root(sch);
339 u32 dupsave = q->duplicate; /* prevent duplicating a dup... */
340 q->duplicate = 0;
341
342 qdisc_enqueue_root(skb2, rootq);
343 q->duplicate = dupsave;
344 }
345
346 /*
347 * Randomized packet corruption.
348 * Make copy if needed since we are modifying
349 * If packet is going to be hardware checksummed, then
350 * do it now in software before we mangle it.
351 */
352 if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) {
353 if (!(skb = skb_unshare(skb, GFP_ATOMIC)) ||
354 (skb->ip_summed == CHECKSUM_PARTIAL &&
355 skb_checksum_help(skb))) {
356 sch->qstats.drops++;
357 return NET_XMIT_DROP;
358 }
359
360 skb->data[net_random() % skb_headlen(skb)] ^= 1<<(net_random() % 8);
361 }
362
363 cb = netem_skb_cb(skb);
364 if (q->gap == 0 || /* not doing reordering */
365 q->counter < q->gap || /* inside last reordering gap */
366 q->reorder < get_crandom(&q->reorder_cor)) {
367 psched_time_t now;
368 psched_tdiff_t delay;
369
370 delay = tabledist(q->latency, q->jitter,
371 &q->delay_cor, q->delay_dist);
372
373 now = psched_get_time();
374 cb->time_to_send = now + delay;
375 ++q->counter;
376 ret = qdisc_enqueue(skb, q->qdisc);
377 } else {
378 /*
379 * Do re-ordering by putting one out of N packets at the front
380 * of the queue.
381 */
382 cb->time_to_send = psched_get_time();
383 q->counter = 0;
384
385 __skb_queue_head(&q->qdisc->q, skb);
386 q->qdisc->qstats.backlog += qdisc_pkt_len(skb);
387 q->qdisc->qstats.requeues++;
388 ret = NET_XMIT_SUCCESS;
389 }
390
391 if (ret != NET_XMIT_SUCCESS) {
392 if (net_xmit_drop_count(ret)) {
393 sch->qstats.drops++;
394 return ret;
395 }
396 }
397
398 sch->q.qlen++;
399 return NET_XMIT_SUCCESS;
400 }
401
402 static unsigned int netem_drop(struct Qdisc *sch)
403 {
404 struct netem_sched_data *q = qdisc_priv(sch);
405 unsigned int len = 0;
406
407 if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) {
408 sch->q.qlen--;
409 sch->qstats.drops++;
410 }
411 return len;
412 }
413
414 static struct sk_buff *netem_dequeue(struct Qdisc *sch)
415 {
416 struct netem_sched_data *q = qdisc_priv(sch);
417 struct sk_buff *skb;
418
419 if (qdisc_is_throttled(sch))
420 return NULL;
421
422 skb = q->qdisc->ops->peek(q->qdisc);
423 if (skb) {
424 const struct netem_skb_cb *cb = netem_skb_cb(skb);
425 psched_time_t now = psched_get_time();
426
427 /* if more time remaining? */
428 if (cb->time_to_send <= now) {
429 skb = qdisc_dequeue_peeked(q->qdisc);
430 if (unlikely(!skb))
431 return NULL;
432
433 #ifdef CONFIG_NET_CLS_ACT
434 /*
435 * If it's at ingress let's pretend the delay is
436 * from the network (tstamp will be updated).
437 */
438 if (G_TC_FROM(skb->tc_verd) & AT_INGRESS)
439 skb->tstamp.tv64 = 0;
440 #endif
441
442 sch->q.qlen--;
443 qdisc_unthrottled(sch);
444 qdisc_bstats_update(sch, skb);
445 return skb;
446 }
447
448 qdisc_watchdog_schedule(&q->watchdog, cb->time_to_send);
449 }
450
451 return NULL;
452 }
453
454 static void netem_reset(struct Qdisc *sch)
455 {
456 struct netem_sched_data *q = qdisc_priv(sch);
457
458 qdisc_reset(q->qdisc);
459 sch->q.qlen = 0;
460 qdisc_watchdog_cancel(&q->watchdog);
461 }
462
463 static void dist_free(struct disttable *d)
464 {
465 if (d) {
466 if (is_vmalloc_addr(d))
467 vfree(d);
468 else
469 kfree(d);
470 }
471 }
472
473 /*
474 * Distribution data is a variable size payload containing
475 * signed 16 bit values.
476 */
477 static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr)
478 {
479 struct netem_sched_data *q = qdisc_priv(sch);
480 size_t n = nla_len(attr)/sizeof(__s16);
481 const __s16 *data = nla_data(attr);
482 spinlock_t *root_lock;
483 struct disttable *d;
484 int i;
485 size_t s;
486
487 if (n > NETEM_DIST_MAX)
488 return -EINVAL;
489
490 s = sizeof(struct disttable) + n * sizeof(s16);
491 d = kmalloc(s, GFP_KERNEL | __GFP_NOWARN);
492 if (!d)
493 d = vmalloc(s);
494 if (!d)
495 return -ENOMEM;
496
497 d->size = n;
498 for (i = 0; i < n; i++)
499 d->table[i] = data[i];
500
501 root_lock = qdisc_root_sleeping_lock(sch);
502
503 spin_lock_bh(root_lock);
504 swap(q->delay_dist, d);
505 spin_unlock_bh(root_lock);
506
507 dist_free(d);
508 return 0;
509 }
510
511 static void get_correlation(struct Qdisc *sch, const struct nlattr *attr)
512 {
513 struct netem_sched_data *q = qdisc_priv(sch);
514 const struct tc_netem_corr *c = nla_data(attr);
515
516 init_crandom(&q->delay_cor, c->delay_corr);
517 init_crandom(&q->loss_cor, c->loss_corr);
518 init_crandom(&q->dup_cor, c->dup_corr);
519 }
520
521 static void get_reorder(struct Qdisc *sch, const struct nlattr *attr)
522 {
523 struct netem_sched_data *q = qdisc_priv(sch);
524 const struct tc_netem_reorder *r = nla_data(attr);
525
526 q->reorder = r->probability;
527 init_crandom(&q->reorder_cor, r->correlation);
528 }
529
530 static void get_corrupt(struct Qdisc *sch, const struct nlattr *attr)
531 {
532 struct netem_sched_data *q = qdisc_priv(sch);
533 const struct tc_netem_corrupt *r = nla_data(attr);
534
535 q->corrupt = r->probability;
536 init_crandom(&q->corrupt_cor, r->correlation);
537 }
538
539 static int get_loss_clg(struct Qdisc *sch, const struct nlattr *attr)
540 {
541 struct netem_sched_data *q = qdisc_priv(sch);
542 const struct nlattr *la;
543 int rem;
544
545 nla_for_each_nested(la, attr, rem) {
546 u16 type = nla_type(la);
547
548 switch(type) {
549 case NETEM_LOSS_GI: {
550 const struct tc_netem_gimodel *gi = nla_data(la);
551
552 if (nla_len(la) != sizeof(struct tc_netem_gimodel)) {
553 pr_info("netem: incorrect gi model size\n");
554 return -EINVAL;
555 }
556
557 q->loss_model = CLG_4_STATES;
558
559 q->clg.state = 1;
560 q->clg.a1 = gi->p13;
561 q->clg.a2 = gi->p31;
562 q->clg.a3 = gi->p32;
563 q->clg.a4 = gi->p14;
564 q->clg.a5 = gi->p23;
565 break;
566 }
567
568 case NETEM_LOSS_GE: {
569 const struct tc_netem_gemodel *ge = nla_data(la);
570
571 if (nla_len(la) != sizeof(struct tc_netem_gemodel)) {
572 pr_info("netem: incorrect gi model size\n");
573 return -EINVAL;
574 }
575
576 q->loss_model = CLG_GILB_ELL;
577 q->clg.state = 1;
578 q->clg.a1 = ge->p;
579 q->clg.a2 = ge->r;
580 q->clg.a3 = ge->h;
581 q->clg.a4 = ge->k1;
582 break;
583 }
584
585 default:
586 pr_info("netem: unknown loss type %u\n", type);
587 return -EINVAL;
588 }
589 }
590
591 return 0;
592 }
593
594 static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = {
595 [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) },
596 [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) },
597 [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) },
598 [TCA_NETEM_LOSS] = { .type = NLA_NESTED },
599 };
600
601 static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla,
602 const struct nla_policy *policy, int len)
603 {
604 int nested_len = nla_len(nla) - NLA_ALIGN(len);
605
606 if (nested_len < 0) {
607 pr_info("netem: invalid attributes len %d\n", nested_len);
608 return -EINVAL;
609 }
610
611 if (nested_len >= nla_attr_size(0))
612 return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len),
613 nested_len, policy);
614
615 memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1));
616 return 0;
617 }
618
619 /* Parse netlink message to set options */
620 static int netem_change(struct Qdisc *sch, struct nlattr *opt)
621 {
622 struct netem_sched_data *q = qdisc_priv(sch);
623 struct nlattr *tb[TCA_NETEM_MAX + 1];
624 struct tc_netem_qopt *qopt;
625 int ret;
626
627 if (opt == NULL)
628 return -EINVAL;
629
630 qopt = nla_data(opt);
631 ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt));
632 if (ret < 0)
633 return ret;
634
635 ret = fifo_set_limit(q->qdisc, qopt->limit);
636 if (ret) {
637 pr_info("netem: can't set fifo limit\n");
638 return ret;
639 }
640
641 q->latency = qopt->latency;
642 q->jitter = qopt->jitter;
643 q->limit = qopt->limit;
644 q->gap = qopt->gap;
645 q->counter = 0;
646 q->loss = qopt->loss;
647 q->duplicate = qopt->duplicate;
648
649 /* for compatibility with earlier versions.
650 * if gap is set, need to assume 100% probability
651 */
652 if (q->gap)
653 q->reorder = ~0;
654
655 if (tb[TCA_NETEM_CORR])
656 get_correlation(sch, tb[TCA_NETEM_CORR]);
657
658 if (tb[TCA_NETEM_DELAY_DIST]) {
659 ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]);
660 if (ret)
661 return ret;
662 }
663
664 if (tb[TCA_NETEM_REORDER])
665 get_reorder(sch, tb[TCA_NETEM_REORDER]);
666
667 if (tb[TCA_NETEM_CORRUPT])
668 get_corrupt(sch, tb[TCA_NETEM_CORRUPT]);
669
670 q->loss_model = CLG_RANDOM;
671 if (tb[TCA_NETEM_LOSS])
672 ret = get_loss_clg(sch, tb[TCA_NETEM_LOSS]);
673
674 return ret;
675 }
676
677 /*
678 * Special case version of FIFO queue for use by netem.
679 * It queues in order based on timestamps in skb's
680 */
681 struct fifo_sched_data {
682 u32 limit;
683 psched_time_t oldest;
684 };
685
686 static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch)
687 {
688 struct fifo_sched_data *q = qdisc_priv(sch);
689 struct sk_buff_head *list = &sch->q;
690 psched_time_t tnext = netem_skb_cb(nskb)->time_to_send;
691 struct sk_buff *skb;
692
693 if (likely(skb_queue_len(list) < q->limit)) {
694 /* Optimize for add at tail */
695 if (likely(skb_queue_empty(list) || tnext >= q->oldest)) {
696 q->oldest = tnext;
697 return qdisc_enqueue_tail(nskb, sch);
698 }
699
700 skb_queue_reverse_walk(list, skb) {
701 const struct netem_skb_cb *cb = netem_skb_cb(skb);
702
703 if (tnext >= cb->time_to_send)
704 break;
705 }
706
707 __skb_queue_after(list, skb, nskb);
708
709 sch->qstats.backlog += qdisc_pkt_len(nskb);
710
711 return NET_XMIT_SUCCESS;
712 }
713
714 return qdisc_reshape_fail(nskb, sch);
715 }
716
717 static int tfifo_init(struct Qdisc *sch, struct nlattr *opt)
718 {
719 struct fifo_sched_data *q = qdisc_priv(sch);
720
721 if (opt) {
722 struct tc_fifo_qopt *ctl = nla_data(opt);
723 if (nla_len(opt) < sizeof(*ctl))
724 return -EINVAL;
725
726 q->limit = ctl->limit;
727 } else
728 q->limit = max_t(u32, qdisc_dev(sch)->tx_queue_len, 1);
729
730 q->oldest = PSCHED_PASTPERFECT;
731 return 0;
732 }
733
734 static int tfifo_dump(struct Qdisc *sch, struct sk_buff *skb)
735 {
736 struct fifo_sched_data *q = qdisc_priv(sch);
737 struct tc_fifo_qopt opt = { .limit = q->limit };
738
739 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
740 return skb->len;
741
742 nla_put_failure:
743 return -1;
744 }
745
746 static struct Qdisc_ops tfifo_qdisc_ops __read_mostly = {
747 .id = "tfifo",
748 .priv_size = sizeof(struct fifo_sched_data),
749 .enqueue = tfifo_enqueue,
750 .dequeue = qdisc_dequeue_head,
751 .peek = qdisc_peek_head,
752 .drop = qdisc_queue_drop,
753 .init = tfifo_init,
754 .reset = qdisc_reset_queue,
755 .change = tfifo_init,
756 .dump = tfifo_dump,
757 };
758
759 static int netem_init(struct Qdisc *sch, struct nlattr *opt)
760 {
761 struct netem_sched_data *q = qdisc_priv(sch);
762 int ret;
763
764 if (!opt)
765 return -EINVAL;
766
767 qdisc_watchdog_init(&q->watchdog, sch);
768
769 q->loss_model = CLG_RANDOM;
770 q->qdisc = qdisc_create_dflt(sch->dev_queue, &tfifo_qdisc_ops,
771 TC_H_MAKE(sch->handle, 1));
772 if (!q->qdisc) {
773 pr_notice("netem: qdisc create tfifo qdisc failed\n");
774 return -ENOMEM;
775 }
776
777 ret = netem_change(sch, opt);
778 if (ret) {
779 pr_info("netem: change failed\n");
780 qdisc_destroy(q->qdisc);
781 }
782 return ret;
783 }
784
785 static void netem_destroy(struct Qdisc *sch)
786 {
787 struct netem_sched_data *q = qdisc_priv(sch);
788
789 qdisc_watchdog_cancel(&q->watchdog);
790 qdisc_destroy(q->qdisc);
791 dist_free(q->delay_dist);
792 }
793
794 static int dump_loss_model(const struct netem_sched_data *q,
795 struct sk_buff *skb)
796 {
797 struct nlattr *nest;
798
799 nest = nla_nest_start(skb, TCA_NETEM_LOSS);
800 if (nest == NULL)
801 goto nla_put_failure;
802
803 switch (q->loss_model) {
804 case CLG_RANDOM:
805 /* legacy loss model */
806 nla_nest_cancel(skb, nest);
807 return 0; /* no data */
808
809 case CLG_4_STATES: {
810 struct tc_netem_gimodel gi = {
811 .p13 = q->clg.a1,
812 .p31 = q->clg.a2,
813 .p32 = q->clg.a3,
814 .p14 = q->clg.a4,
815 .p23 = q->clg.a5,
816 };
817
818 NLA_PUT(skb, NETEM_LOSS_GI, sizeof(gi), &gi);
819 break;
820 }
821 case CLG_GILB_ELL: {
822 struct tc_netem_gemodel ge = {
823 .p = q->clg.a1,
824 .r = q->clg.a2,
825 .h = q->clg.a3,
826 .k1 = q->clg.a4,
827 };
828
829 NLA_PUT(skb, NETEM_LOSS_GE, sizeof(ge), &ge);
830 break;
831 }
832 }
833
834 nla_nest_end(skb, nest);
835 return 0;
836
837 nla_put_failure:
838 nla_nest_cancel(skb, nest);
839 return -1;
840 }
841
842 static int netem_dump(struct Qdisc *sch, struct sk_buff *skb)
843 {
844 const struct netem_sched_data *q = qdisc_priv(sch);
845 struct nlattr *nla = (struct nlattr *) skb_tail_pointer(skb);
846 struct tc_netem_qopt qopt;
847 struct tc_netem_corr cor;
848 struct tc_netem_reorder reorder;
849 struct tc_netem_corrupt corrupt;
850
851 qopt.latency = q->latency;
852 qopt.jitter = q->jitter;
853 qopt.limit = q->limit;
854 qopt.loss = q->loss;
855 qopt.gap = q->gap;
856 qopt.duplicate = q->duplicate;
857 NLA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt);
858
859 cor.delay_corr = q->delay_cor.rho;
860 cor.loss_corr = q->loss_cor.rho;
861 cor.dup_corr = q->dup_cor.rho;
862 NLA_PUT(skb, TCA_NETEM_CORR, sizeof(cor), &cor);
863
864 reorder.probability = q->reorder;
865 reorder.correlation = q->reorder_cor.rho;
866 NLA_PUT(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder);
867
868 corrupt.probability = q->corrupt;
869 corrupt.correlation = q->corrupt_cor.rho;
870 NLA_PUT(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt);
871
872 if (dump_loss_model(q, skb) != 0)
873 goto nla_put_failure;
874
875 return nla_nest_end(skb, nla);
876
877 nla_put_failure:
878 nlmsg_trim(skb, nla);
879 return -1;
880 }
881
882 static int netem_dump_class(struct Qdisc *sch, unsigned long cl,
883 struct sk_buff *skb, struct tcmsg *tcm)
884 {
885 struct netem_sched_data *q = qdisc_priv(sch);
886
887 if (cl != 1) /* only one class */
888 return -ENOENT;
889
890 tcm->tcm_handle |= TC_H_MIN(1);
891 tcm->tcm_info = q->qdisc->handle;
892
893 return 0;
894 }
895
896 static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
897 struct Qdisc **old)
898 {
899 struct netem_sched_data *q = qdisc_priv(sch);
900
901 if (new == NULL)
902 new = &noop_qdisc;
903
904 sch_tree_lock(sch);
905 *old = q->qdisc;
906 q->qdisc = new;
907 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
908 qdisc_reset(*old);
909 sch_tree_unlock(sch);
910
911 return 0;
912 }
913
914 static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg)
915 {
916 struct netem_sched_data *q = qdisc_priv(sch);
917 return q->qdisc;
918 }
919
920 static unsigned long netem_get(struct Qdisc *sch, u32 classid)
921 {
922 return 1;
923 }
924
925 static void netem_put(struct Qdisc *sch, unsigned long arg)
926 {
927 }
928
929 static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker)
930 {
931 if (!walker->stop) {
932 if (walker->count >= walker->skip)
933 if (walker->fn(sch, 1, walker) < 0) {
934 walker->stop = 1;
935 return;
936 }
937 walker->count++;
938 }
939 }
940
941 static const struct Qdisc_class_ops netem_class_ops = {
942 .graft = netem_graft,
943 .leaf = netem_leaf,
944 .get = netem_get,
945 .put = netem_put,
946 .walk = netem_walk,
947 .dump = netem_dump_class,
948 };
949
950 static struct Qdisc_ops netem_qdisc_ops __read_mostly = {
951 .id = "netem",
952 .cl_ops = &netem_class_ops,
953 .priv_size = sizeof(struct netem_sched_data),
954 .enqueue = netem_enqueue,
955 .dequeue = netem_dequeue,
956 .peek = qdisc_peek_dequeued,
957 .drop = netem_drop,
958 .init = netem_init,
959 .reset = netem_reset,
960 .destroy = netem_destroy,
961 .change = netem_change,
962 .dump = netem_dump,
963 .owner = THIS_MODULE,
964 };
965
966
967 static int __init netem_module_init(void)
968 {
969 pr_info("netem: version " VERSION "\n");
970 return register_qdisc(&netem_qdisc_ops);
971 }
972 static void __exit netem_module_exit(void)
973 {
974 unregister_qdisc(&netem_qdisc_ops);
975 }
976 module_init(netem_module_init)
977 module_exit(netem_module_exit)
978 MODULE_LICENSE("GPL");
Linux kernel - Deliverables
This page took 0.051118 seconds and 6 git commands to generate.