blob: 7228d30512c79466675276933268b6dcfc31972e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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, or (at your option) any later version.
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/config.h>
17#include <linux/module.h>
18#include <linux/bitops.h>
19#include <linux/types.h>
20#include <linux/kernel.h>
21#include <linux/errno.h>
22#include <linux/netdevice.h>
23#include <linux/skbuff.h>
24#include <linux/rtnetlink.h>
25
26#include <net/pkt_sched.h>
27
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -080028#define VERSION "1.2"
Stephen Hemmingereb229c42005-11-03 13:49:01 -080029
Linus Torvalds1da177e2005-04-16 15:20:36 -070030/* Network Emulation Queuing algorithm.
31 ====================================
32
33 Sources: [1] Mark Carson, Darrin Santay, "NIST Net - A Linux-based
34 Network Emulation Tool
35 [2] Luigi Rizzo, DummyNet for FreeBSD
36
37 ----------------------------------------------------------------
38
39 This started out as a simple way to delay outgoing packets to
40 test TCP but has grown to include most of the functionality
41 of a full blown network emulator like NISTnet. It can delay
42 packets and add random jitter (and correlation). The random
43 distribution can be loaded from a table as well to provide
44 normal, Pareto, or experimental curves. Packet loss,
45 duplication, and reordering can also be emulated.
46
47 This qdisc does not do classification that can be handled in
48 layering other disciplines. It does not need to do bandwidth
49 control either since that can be handled by using token
50 bucket or other rate control.
51
52 The simulator is limited by the Linux timer resolution
53 and will create packet bursts on the HZ boundary (1ms).
54*/
55
56struct netem_sched_data {
57 struct Qdisc *qdisc;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 struct timer_list timer;
59
60 u32 latency;
61 u32 loss;
62 u32 limit;
63 u32 counter;
64 u32 gap;
65 u32 jitter;
66 u32 duplicate;
Stephen Hemminger0dca51d2005-05-26 12:55:48 -070067 u32 reorder;
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -080068 u32 corrupt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
70 struct crndstate {
71 unsigned long last;
72 unsigned long rho;
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -080073 } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
75 struct disttable {
76 u32 size;
77 s16 table[0];
78 } *delay_dist;
79};
80
81/* Time stamp put into socket buffer control block */
82struct netem_skb_cb {
83 psched_time_t time_to_send;
84};
85
86/* init_crandom - initialize correlated random number generator
87 * Use entropy source for initial seed.
88 */
89static void init_crandom(struct crndstate *state, unsigned long rho)
90{
91 state->rho = rho;
92 state->last = net_random();
93}
94
95/* get_crandom - correlated random number generator
96 * Next number depends on last value.
97 * rho is scaled to avoid floating point.
98 */
99static unsigned long get_crandom(struct crndstate *state)
100{
101 u64 value, rho;
102 unsigned long answer;
103
104 if (state->rho == 0) /* no correllation */
105 return net_random();
106
107 value = net_random();
108 rho = (u64)state->rho + 1;
109 answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32;
110 state->last = answer;
111 return answer;
112}
113
114/* tabledist - return a pseudo-randomly distributed value with mean mu and
115 * std deviation sigma. Uses table lookup to approximate the desired
116 * distribution, and a uniformly-distributed pseudo-random source.
117 */
118static long tabledist(unsigned long mu, long sigma,
119 struct crndstate *state, const struct disttable *dist)
120{
121 long t, x;
122 unsigned long rnd;
123
124 if (sigma == 0)
125 return mu;
126
127 rnd = get_crandom(state);
128
129 /* default uniform distribution */
130 if (dist == NULL)
131 return (rnd % (2*sigma)) - sigma + mu;
132
133 t = dist->table[rnd % dist->size];
134 x = (sigma % NETEM_DIST_SCALE) * t;
135 if (x >= 0)
136 x += NETEM_DIST_SCALE/2;
137 else
138 x -= NETEM_DIST_SCALE/2;
139
140 return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu;
141}
142
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700143/*
144 * Insert one skb into qdisc.
145 * Note: parent depends on return value to account for queue length.
146 * NET_XMIT_DROP: queue length didn't change.
147 * NET_XMIT_SUCCESS: one skb was queued.
148 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch)
150{
151 struct netem_sched_data *q = qdisc_priv(sch);
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700152 struct netem_skb_cb *cb = (struct netem_skb_cb *)skb->cb;
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700153 struct sk_buff *skb2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 int ret;
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700155 int count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
Stephen Hemminger771018e2005-05-03 16:24:32 -0700157 pr_debug("netem_enqueue skb=%p\n", skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700159 /* Random duplication */
160 if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor))
161 ++count;
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 /* Random packet drop 0 => none, ~0 => all */
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700164 if (q->loss && q->loss >= get_crandom(&q->loss_cor))
165 --count;
166
167 if (count == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 sch->qstats.drops++;
169 kfree_skb(skb);
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700170 return NET_XMIT_DROP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 }
172
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700173 /*
174 * If we need to duplicate packet, then re-insert at top of the
175 * qdisc tree, since parent queuer expects that only one
176 * skb will be queued.
177 */
178 if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) {
179 struct Qdisc *rootq = sch->dev->qdisc;
180 u32 dupsave = q->duplicate; /* prevent duplicating a dup... */
181 q->duplicate = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700183 rootq->enqueue(skb2, rootq);
184 q->duplicate = dupsave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 }
186
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800187 /*
188 * Randomized packet corruption.
189 * Make copy if needed since we are modifying
190 * If packet is going to be hardware checksummed, then
191 * do it now in software before we mangle it.
192 */
193 if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) {
194 if (!(skb = skb_unshare(skb, GFP_ATOMIC))
195 || (skb->ip_summed == CHECKSUM_HW
196 && skb_checksum_help(skb, 0))) {
197 sch->qstats.drops++;
198 return NET_XMIT_DROP;
199 }
200
201 skb->data[net_random() % skb_headlen(skb)] ^= 1<<(net_random() % 8);
202 }
203
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700204 if (q->gap == 0 /* not doing reordering */
205 || q->counter < q->gap /* inside last reordering gap */
206 || q->reorder < get_crandom(&q->reorder_cor)) {
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700207 psched_time_t now;
Stephen Hemminger07aaa112005-11-03 13:43:07 -0800208 psched_tdiff_t delay;
209
210 delay = tabledist(q->latency, q->jitter,
211 &q->delay_cor, q->delay_dist);
212
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700213 PSCHED_GET_TIME(now);
Stephen Hemminger07aaa112005-11-03 13:43:07 -0800214 PSCHED_TADD2(now, delay, cb->time_to_send);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 ++q->counter;
216 ret = q->qdisc->enqueue(skb, q->qdisc);
217 } else {
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700218 /*
219 * Do re-ordering by putting one out of N packets at the front
220 * of the queue.
221 */
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700222 PSCHED_GET_TIME(cb->time_to_send);
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700223 q->counter = 0;
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700224 ret = q->qdisc->ops->requeue(skb, q->qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 }
226
227 if (likely(ret == NET_XMIT_SUCCESS)) {
228 sch->q.qlen++;
229 sch->bstats.bytes += skb->len;
230 sch->bstats.packets++;
231 } else
232 sch->qstats.drops++;
233
Stephen Hemmingerd5d75cd2005-05-03 16:24:57 -0700234 pr_debug("netem: enqueue ret %d\n", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 return ret;
236}
237
238/* Requeue packets but don't change time stamp */
239static int netem_requeue(struct sk_buff *skb, struct Qdisc *sch)
240{
241 struct netem_sched_data *q = qdisc_priv(sch);
242 int ret;
243
244 if ((ret = q->qdisc->ops->requeue(skb, q->qdisc)) == 0) {
245 sch->q.qlen++;
246 sch->qstats.requeues++;
247 }
248
249 return ret;
250}
251
252static unsigned int netem_drop(struct Qdisc* sch)
253{
254 struct netem_sched_data *q = qdisc_priv(sch);
Patrick McHardy6d037a22006-03-20 19:00:49 -0800255 unsigned int len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
Patrick McHardy6d037a22006-03-20 19:00:49 -0800257 if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 sch->q.qlen--;
259 sch->qstats.drops++;
260 }
261 return len;
262}
263
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264static struct sk_buff *netem_dequeue(struct Qdisc *sch)
265{
266 struct netem_sched_data *q = qdisc_priv(sch);
267 struct sk_buff *skb;
268
269 skb = q->qdisc->dequeue(q->qdisc);
Stephen Hemminger771018e2005-05-03 16:24:32 -0700270 if (skb) {
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700271 const struct netem_skb_cb *cb
272 = (const struct netem_skb_cb *)skb->cb;
273 psched_time_t now;
Stephen Hemminger771018e2005-05-03 16:24:32 -0700274
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700275 /* if more time remaining? */
276 PSCHED_GET_TIME(now);
Stephen Hemminger07aaa112005-11-03 13:43:07 -0800277
278 if (PSCHED_TLESS(cb->time_to_send, now)) {
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700279 pr_debug("netem_dequeue: return skb=%p\n", skb);
280 sch->q.qlen--;
281 sch->flags &= ~TCQ_F_THROTTLED;
282 return skb;
Stephen Hemminger07aaa112005-11-03 13:43:07 -0800283 } else {
284 psched_tdiff_t delay = PSCHED_TDIFF(cb->time_to_send, now);
285
286 if (q->qdisc->ops->requeue(skb, q->qdisc) != NET_XMIT_SUCCESS) {
287 sch->qstats.drops++;
288
289 /* After this qlen is confused */
290 printk(KERN_ERR "netem: queue discpline %s could not requeue\n",
291 q->qdisc->ops->id);
292
293 sch->q.qlen--;
294 }
295
296 mod_timer(&q->timer, jiffies + PSCHED_US2JIFFIE(delay));
297 sch->flags |= TCQ_F_THROTTLED;
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700298 }
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700299 }
300
301 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302}
303
304static void netem_watchdog(unsigned long arg)
305{
306 struct Qdisc *sch = (struct Qdisc *)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307
Stephen Hemminger771018e2005-05-03 16:24:32 -0700308 pr_debug("netem_watchdog qlen=%d\n", sch->q.qlen);
309 sch->flags &= ~TCQ_F_THROTTLED;
310 netif_schedule(sch->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311}
312
313static void netem_reset(struct Qdisc *sch)
314{
315 struct netem_sched_data *q = qdisc_priv(sch);
316
317 qdisc_reset(q->qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 sch->q.qlen = 0;
Stephen Hemminger771018e2005-05-03 16:24:32 -0700319 sch->flags &= ~TCQ_F_THROTTLED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 del_timer_sync(&q->timer);
321}
322
Stephen Hemminger300ce172005-10-30 13:47:34 -0800323/* Pass size change message down to embedded FIFO */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324static int set_fifo_limit(struct Qdisc *q, int limit)
325{
326 struct rtattr *rta;
327 int ret = -ENOMEM;
328
Stephen Hemminger300ce172005-10-30 13:47:34 -0800329 /* Hack to avoid sending change message to non-FIFO */
330 if (strncmp(q->ops->id + 1, "fifo", 4) != 0)
331 return 0;
332
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 rta = kmalloc(RTA_LENGTH(sizeof(struct tc_fifo_qopt)), GFP_KERNEL);
334 if (rta) {
335 rta->rta_type = RTM_NEWQDISC;
336 rta->rta_len = RTA_LENGTH(sizeof(struct tc_fifo_qopt));
337 ((struct tc_fifo_qopt *)RTA_DATA(rta))->limit = limit;
338
339 ret = q->ops->change(q, rta);
340 kfree(rta);
341 }
342 return ret;
343}
344
345/*
346 * Distribution data is a variable size payload containing
347 * signed 16 bit values.
348 */
349static int get_dist_table(struct Qdisc *sch, const struct rtattr *attr)
350{
351 struct netem_sched_data *q = qdisc_priv(sch);
352 unsigned long n = RTA_PAYLOAD(attr)/sizeof(__s16);
353 const __s16 *data = RTA_DATA(attr);
354 struct disttable *d;
355 int i;
356
357 if (n > 65536)
358 return -EINVAL;
359
360 d = kmalloc(sizeof(*d) + n*sizeof(d->table[0]), GFP_KERNEL);
361 if (!d)
362 return -ENOMEM;
363
364 d->size = n;
365 for (i = 0; i < n; i++)
366 d->table[i] = data[i];
367
368 spin_lock_bh(&sch->dev->queue_lock);
369 d = xchg(&q->delay_dist, d);
370 spin_unlock_bh(&sch->dev->queue_lock);
371
372 kfree(d);
373 return 0;
374}
375
376static int get_correlation(struct Qdisc *sch, const struct rtattr *attr)
377{
378 struct netem_sched_data *q = qdisc_priv(sch);
379 const struct tc_netem_corr *c = RTA_DATA(attr);
380
381 if (RTA_PAYLOAD(attr) != sizeof(*c))
382 return -EINVAL;
383
384 init_crandom(&q->delay_cor, c->delay_corr);
385 init_crandom(&q->loss_cor, c->loss_corr);
386 init_crandom(&q->dup_cor, c->dup_corr);
387 return 0;
388}
389
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700390static int get_reorder(struct Qdisc *sch, const struct rtattr *attr)
391{
392 struct netem_sched_data *q = qdisc_priv(sch);
393 const struct tc_netem_reorder *r = RTA_DATA(attr);
394
395 if (RTA_PAYLOAD(attr) != sizeof(*r))
396 return -EINVAL;
397
398 q->reorder = r->probability;
399 init_crandom(&q->reorder_cor, r->correlation);
400 return 0;
401}
402
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800403static int get_corrupt(struct Qdisc *sch, const struct rtattr *attr)
404{
405 struct netem_sched_data *q = qdisc_priv(sch);
406 const struct tc_netem_corrupt *r = RTA_DATA(attr);
407
408 if (RTA_PAYLOAD(attr) != sizeof(*r))
409 return -EINVAL;
410
411 q->corrupt = r->probability;
412 init_crandom(&q->corrupt_cor, r->correlation);
413 return 0;
414}
415
416/* Parse netlink message to set options */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417static int netem_change(struct Qdisc *sch, struct rtattr *opt)
418{
419 struct netem_sched_data *q = qdisc_priv(sch);
420 struct tc_netem_qopt *qopt;
421 int ret;
422
423 if (opt == NULL || RTA_PAYLOAD(opt) < sizeof(*qopt))
424 return -EINVAL;
425
426 qopt = RTA_DATA(opt);
427 ret = set_fifo_limit(q->qdisc, qopt->limit);
428 if (ret) {
429 pr_debug("netem: can't set fifo limit\n");
430 return ret;
431 }
432
433 q->latency = qopt->latency;
434 q->jitter = qopt->jitter;
435 q->limit = qopt->limit;
436 q->gap = qopt->gap;
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700437 q->counter = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 q->loss = qopt->loss;
439 q->duplicate = qopt->duplicate;
440
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700441 /* for compatiablity with earlier versions.
442 * if gap is set, need to assume 100% probablity
443 */
444 q->reorder = ~0;
445
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 /* Handle nested options after initial queue options.
447 * Should have put all options in nested format but too late now.
448 */
449 if (RTA_PAYLOAD(opt) > sizeof(*qopt)) {
450 struct rtattr *tb[TCA_NETEM_MAX];
451 if (rtattr_parse(tb, TCA_NETEM_MAX,
452 RTA_DATA(opt) + sizeof(*qopt),
453 RTA_PAYLOAD(opt) - sizeof(*qopt)))
454 return -EINVAL;
455
456 if (tb[TCA_NETEM_CORR-1]) {
457 ret = get_correlation(sch, tb[TCA_NETEM_CORR-1]);
458 if (ret)
459 return ret;
460 }
461
462 if (tb[TCA_NETEM_DELAY_DIST-1]) {
463 ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST-1]);
464 if (ret)
465 return ret;
466 }
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800467
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700468 if (tb[TCA_NETEM_REORDER-1]) {
469 ret = get_reorder(sch, tb[TCA_NETEM_REORDER-1]);
470 if (ret)
471 return ret;
472 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800474 if (tb[TCA_NETEM_CORRUPT-1]) {
475 ret = get_corrupt(sch, tb[TCA_NETEM_CORRUPT-1]);
476 if (ret)
477 return ret;
478 }
479 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
481 return 0;
482}
483
Stephen Hemminger300ce172005-10-30 13:47:34 -0800484/*
485 * Special case version of FIFO queue for use by netem.
486 * It queues in order based on timestamps in skb's
487 */
488struct fifo_sched_data {
489 u32 limit;
490};
491
492static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch)
493{
494 struct fifo_sched_data *q = qdisc_priv(sch);
495 struct sk_buff_head *list = &sch->q;
496 const struct netem_skb_cb *ncb
497 = (const struct netem_skb_cb *)nskb->cb;
498 struct sk_buff *skb;
499
500 if (likely(skb_queue_len(list) < q->limit)) {
501 skb_queue_reverse_walk(list, skb) {
502 const struct netem_skb_cb *cb
503 = (const struct netem_skb_cb *)skb->cb;
504
Andrea Bittauaa875162005-11-20 13:41:05 -0800505 if (!PSCHED_TLESS(ncb->time_to_send, cb->time_to_send))
Stephen Hemminger300ce172005-10-30 13:47:34 -0800506 break;
507 }
508
509 __skb_queue_after(list, skb, nskb);
510
511 sch->qstats.backlog += nskb->len;
512 sch->bstats.bytes += nskb->len;
513 sch->bstats.packets++;
514
515 return NET_XMIT_SUCCESS;
516 }
517
518 return qdisc_drop(nskb, sch);
519}
520
521static int tfifo_init(struct Qdisc *sch, struct rtattr *opt)
522{
523 struct fifo_sched_data *q = qdisc_priv(sch);
524
525 if (opt) {
526 struct tc_fifo_qopt *ctl = RTA_DATA(opt);
527 if (RTA_PAYLOAD(opt) < sizeof(*ctl))
528 return -EINVAL;
529
530 q->limit = ctl->limit;
531 } else
532 q->limit = max_t(u32, sch->dev->tx_queue_len, 1);
533
534 return 0;
535}
536
537static int tfifo_dump(struct Qdisc *sch, struct sk_buff *skb)
538{
539 struct fifo_sched_data *q = qdisc_priv(sch);
540 struct tc_fifo_qopt opt = { .limit = q->limit };
541
542 RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
543 return skb->len;
544
545rtattr_failure:
546 return -1;
547}
548
549static struct Qdisc_ops tfifo_qdisc_ops = {
550 .id = "tfifo",
551 .priv_size = sizeof(struct fifo_sched_data),
552 .enqueue = tfifo_enqueue,
553 .dequeue = qdisc_dequeue_head,
554 .requeue = qdisc_requeue,
555 .drop = qdisc_queue_drop,
556 .init = tfifo_init,
557 .reset = qdisc_reset_queue,
558 .change = tfifo_init,
559 .dump = tfifo_dump,
560};
561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562static int netem_init(struct Qdisc *sch, struct rtattr *opt)
563{
564 struct netem_sched_data *q = qdisc_priv(sch);
565 int ret;
566
567 if (!opt)
568 return -EINVAL;
569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 init_timer(&q->timer);
571 q->timer.function = netem_watchdog;
572 q->timer.data = (unsigned long) sch;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
Stephen Hemminger300ce172005-10-30 13:47:34 -0800574 q->qdisc = qdisc_create_dflt(sch->dev, &tfifo_qdisc_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 if (!q->qdisc) {
576 pr_debug("netem: qdisc create failed\n");
577 return -ENOMEM;
578 }
579
580 ret = netem_change(sch, opt);
581 if (ret) {
582 pr_debug("netem: change failed\n");
583 qdisc_destroy(q->qdisc);
584 }
585 return ret;
586}
587
588static void netem_destroy(struct Qdisc *sch)
589{
590 struct netem_sched_data *q = qdisc_priv(sch);
591
592 del_timer_sync(&q->timer);
593 qdisc_destroy(q->qdisc);
594 kfree(q->delay_dist);
595}
596
597static int netem_dump(struct Qdisc *sch, struct sk_buff *skb)
598{
599 const struct netem_sched_data *q = qdisc_priv(sch);
600 unsigned char *b = skb->tail;
601 struct rtattr *rta = (struct rtattr *) b;
602 struct tc_netem_qopt qopt;
603 struct tc_netem_corr cor;
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700604 struct tc_netem_reorder reorder;
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800605 struct tc_netem_corrupt corrupt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
607 qopt.latency = q->latency;
608 qopt.jitter = q->jitter;
609 qopt.limit = q->limit;
610 qopt.loss = q->loss;
611 qopt.gap = q->gap;
612 qopt.duplicate = q->duplicate;
613 RTA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt);
614
615 cor.delay_corr = q->delay_cor.rho;
616 cor.loss_corr = q->loss_cor.rho;
617 cor.dup_corr = q->dup_cor.rho;
618 RTA_PUT(skb, TCA_NETEM_CORR, sizeof(cor), &cor);
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700619
620 reorder.probability = q->reorder;
621 reorder.correlation = q->reorder_cor.rho;
622 RTA_PUT(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder);
623
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800624 corrupt.probability = q->corrupt;
625 corrupt.correlation = q->corrupt_cor.rho;
626 RTA_PUT(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt);
627
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 rta->rta_len = skb->tail - b;
629
630 return skb->len;
631
632rtattr_failure:
633 skb_trim(skb, b - skb->data);
634 return -1;
635}
636
637static int netem_dump_class(struct Qdisc *sch, unsigned long cl,
638 struct sk_buff *skb, struct tcmsg *tcm)
639{
640 struct netem_sched_data *q = qdisc_priv(sch);
641
642 if (cl != 1) /* only one class */
643 return -ENOENT;
644
645 tcm->tcm_handle |= TC_H_MIN(1);
646 tcm->tcm_info = q->qdisc->handle;
647
648 return 0;
649}
650
651static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
652 struct Qdisc **old)
653{
654 struct netem_sched_data *q = qdisc_priv(sch);
655
656 if (new == NULL)
657 new = &noop_qdisc;
658
659 sch_tree_lock(sch);
660 *old = xchg(&q->qdisc, new);
661 qdisc_reset(*old);
662 sch->q.qlen = 0;
663 sch_tree_unlock(sch);
664
665 return 0;
666}
667
668static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg)
669{
670 struct netem_sched_data *q = qdisc_priv(sch);
671 return q->qdisc;
672}
673
674static unsigned long netem_get(struct Qdisc *sch, u32 classid)
675{
676 return 1;
677}
678
679static void netem_put(struct Qdisc *sch, unsigned long arg)
680{
681}
682
683static int netem_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
684 struct rtattr **tca, unsigned long *arg)
685{
686 return -ENOSYS;
687}
688
689static int netem_delete(struct Qdisc *sch, unsigned long arg)
690{
691 return -ENOSYS;
692}
693
694static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker)
695{
696 if (!walker->stop) {
697 if (walker->count >= walker->skip)
698 if (walker->fn(sch, 1, walker) < 0) {
699 walker->stop = 1;
700 return;
701 }
702 walker->count++;
703 }
704}
705
706static struct tcf_proto **netem_find_tcf(struct Qdisc *sch, unsigned long cl)
707{
708 return NULL;
709}
710
711static struct Qdisc_class_ops netem_class_ops = {
712 .graft = netem_graft,
713 .leaf = netem_leaf,
714 .get = netem_get,
715 .put = netem_put,
716 .change = netem_change_class,
717 .delete = netem_delete,
718 .walk = netem_walk,
719 .tcf_chain = netem_find_tcf,
720 .dump = netem_dump_class,
721};
722
723static struct Qdisc_ops netem_qdisc_ops = {
724 .id = "netem",
725 .cl_ops = &netem_class_ops,
726 .priv_size = sizeof(struct netem_sched_data),
727 .enqueue = netem_enqueue,
728 .dequeue = netem_dequeue,
729 .requeue = netem_requeue,
730 .drop = netem_drop,
731 .init = netem_init,
732 .reset = netem_reset,
733 .destroy = netem_destroy,
734 .change = netem_change,
735 .dump = netem_dump,
736 .owner = THIS_MODULE,
737};
738
739
740static int __init netem_module_init(void)
741{
Stephen Hemmingereb229c42005-11-03 13:49:01 -0800742 pr_info("netem: version " VERSION "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 return register_qdisc(&netem_qdisc_ops);
744}
745static void __exit netem_module_exit(void)
746{
747 unregister_qdisc(&netem_qdisc_ops);
748}
749module_init(netem_module_init)
750module_exit(netem_module_exit)
751MODULE_LICENSE("GPL");