]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | |
798b6b19 | 7 | * 2 of the License. |
1da177e4 LT |
8 | * |
9 | * Many of the algorithms and ideas for this came from | |
10297b99 | 10 | * NIST Net which is not copyrighted. |
1da177e4 LT |
11 | * |
12 | * Authors: Stephen Hemminger <shemminger@osdl.org> | |
13 | * Catalin(ux aka Dino) BOIE <catab at umbrella dot ro> | |
14 | */ | |
15 | ||
1da177e4 LT |
16 | #include <linux/module.h> |
17 | #include <linux/bitops.h> | |
18 | #include <linux/types.h> | |
19 | #include <linux/kernel.h> | |
20 | #include <linux/errno.h> | |
21 | #include <linux/netdevice.h> | |
22 | #include <linux/skbuff.h> | |
23 | #include <linux/rtnetlink.h> | |
24 | ||
25 | #include <net/pkt_sched.h> | |
26 | ||
c865e5d9 | 27 | #define VERSION "1.2" |
eb229c4c | 28 | |
1da177e4 LT |
29 | /* Network Emulation Queuing algorithm. |
30 | ==================================== | |
31 | ||
32 | Sources: [1] Mark Carson, Darrin Santay, "NIST Net - A Linux-based | |
33 | Network Emulation Tool | |
34 | [2] Luigi Rizzo, DummyNet for FreeBSD | |
35 | ||
36 | ---------------------------------------------------------------- | |
37 | ||
38 | This started out as a simple way to delay outgoing packets to | |
39 | test TCP but has grown to include most of the functionality | |
40 | of a full blown network emulator like NISTnet. It can delay | |
41 | packets and add random jitter (and correlation). The random | |
42 | distribution can be loaded from a table as well to provide | |
43 | normal, Pareto, or experimental curves. Packet loss, | |
44 | duplication, and reordering can also be emulated. | |
45 | ||
46 | This qdisc does not do classification that can be handled in | |
47 | layering other disciplines. It does not need to do bandwidth | |
48 | control either since that can be handled by using token | |
49 | bucket or other rate control. | |
50 | ||
51 | The simulator is limited by the Linux timer resolution | |
52 | and will create packet bursts on the HZ boundary (1ms). | |
53 | */ | |
54 | ||
55 | struct netem_sched_data { | |
56 | struct Qdisc *qdisc; | |
59cb5c67 | 57 | struct qdisc_watchdog watchdog; |
1da177e4 LT |
58 | |
59 | u32 latency; | |
60 | u32 loss; | |
61 | u32 limit; | |
62 | u32 counter; | |
63 | u32 gap; | |
64 | u32 jitter; | |
65 | u32 duplicate; | |
0dca51d3 | 66 | u32 reorder; |
c865e5d9 | 67 | u32 corrupt; |
1da177e4 LT |
68 | |
69 | struct crndstate { | |
70 | unsigned long last; | |
71 | unsigned long rho; | |
c865e5d9 | 72 | } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor; |
1da177e4 LT |
73 | |
74 | struct disttable { | |
75 | u32 size; | |
76 | s16 table[0]; | |
77 | } *delay_dist; | |
78 | }; | |
79 | ||
80 | /* Time stamp put into socket buffer control block */ | |
81 | struct netem_skb_cb { | |
82 | psched_time_t time_to_send; | |
83 | }; | |
84 | ||
85 | /* init_crandom - initialize correlated random number generator | |
86 | * Use entropy source for initial seed. | |
87 | */ | |
88 | static void init_crandom(struct crndstate *state, unsigned long rho) | |
89 | { | |
90 | state->rho = rho; | |
91 | state->last = net_random(); | |
92 | } | |
93 | ||
94 | /* get_crandom - correlated random number generator | |
95 | * Next number depends on last value. | |
96 | * rho is scaled to avoid floating point. | |
97 | */ | |
98 | static unsigned long get_crandom(struct crndstate *state) | |
99 | { | |
100 | u64 value, rho; | |
101 | unsigned long answer; | |
102 | ||
103 | if (state->rho == 0) /* no correllation */ | |
104 | return net_random(); | |
105 | ||
106 | value = net_random(); | |
107 | rho = (u64)state->rho + 1; | |
108 | answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32; | |
109 | state->last = answer; | |
110 | return answer; | |
111 | } | |
112 | ||
113 | /* tabledist - return a pseudo-randomly distributed value with mean mu and | |
114 | * std deviation sigma. Uses table lookup to approximate the desired | |
115 | * distribution, and a uniformly-distributed pseudo-random source. | |
116 | */ | |
10297b99 | 117 | static long tabledist(unsigned long mu, long sigma, |
1da177e4 LT |
118 | struct crndstate *state, const struct disttable *dist) |
119 | { | |
120 | long t, x; | |
121 | unsigned long rnd; | |
122 | ||
123 | if (sigma == 0) | |
124 | return mu; | |
125 | ||
126 | rnd = get_crandom(state); | |
127 | ||
128 | /* default uniform distribution */ | |
10297b99 | 129 | if (dist == NULL) |
1da177e4 LT |
130 | return (rnd % (2*sigma)) - sigma + mu; |
131 | ||
132 | t = dist->table[rnd % dist->size]; | |
133 | x = (sigma % NETEM_DIST_SCALE) * t; | |
134 | if (x >= 0) | |
135 | x += NETEM_DIST_SCALE/2; | |
136 | else | |
137 | x -= NETEM_DIST_SCALE/2; | |
138 | ||
139 | return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu; | |
140 | } | |
141 | ||
0afb51e7 SH |
142 | /* |
143 | * Insert one skb into qdisc. | |
144 | * Note: parent depends on return value to account for queue length. | |
145 | * NET_XMIT_DROP: queue length didn't change. | |
146 | * NET_XMIT_SUCCESS: one skb was queued. | |
147 | */ | |
1da177e4 LT |
148 | static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
149 | { | |
150 | struct netem_sched_data *q = qdisc_priv(sch); | |
89e1df74 GC |
151 | /* We don't fill cb now as skb_unshare() may invalidate it */ |
152 | struct netem_skb_cb *cb; | |
0afb51e7 | 153 | struct sk_buff *skb2; |
1da177e4 | 154 | int ret; |
0afb51e7 | 155 | int count = 1; |
1da177e4 | 156 | |
771018e7 | 157 | pr_debug("netem_enqueue skb=%p\n", skb); |
1da177e4 | 158 | |
0afb51e7 SH |
159 | /* Random duplication */ |
160 | if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor)) | |
161 | ++count; | |
162 | ||
1da177e4 | 163 | /* Random packet drop 0 => none, ~0 => all */ |
0afb51e7 SH |
164 | if (q->loss && q->loss >= get_crandom(&q->loss_cor)) |
165 | --count; | |
166 | ||
167 | if (count == 0) { | |
1da177e4 LT |
168 | sch->qstats.drops++; |
169 | kfree_skb(skb); | |
89bbb0a3 | 170 | return NET_XMIT_BYPASS; |
1da177e4 LT |
171 | } |
172 | ||
4e8a5201 DM |
173 | skb_orphan(skb); |
174 | ||
0afb51e7 SH |
175 | /* |
176 | * If we need to duplicate packet, then re-insert at top of the | |
177 | * qdisc tree, since parent queuer expects that only one | |
178 | * skb will be queued. | |
179 | */ | |
180 | if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) { | |
181 | struct Qdisc *rootq = sch->dev->qdisc; | |
182 | u32 dupsave = q->duplicate; /* prevent duplicating a dup... */ | |
183 | q->duplicate = 0; | |
184 | ||
185 | rootq->enqueue(skb2, rootq); | |
186 | q->duplicate = dupsave; | |
1da177e4 LT |
187 | } |
188 | ||
c865e5d9 SH |
189 | /* |
190 | * Randomized packet corruption. | |
191 | * Make copy if needed since we are modifying | |
192 | * If packet is going to be hardware checksummed, then | |
193 | * do it now in software before we mangle it. | |
194 | */ | |
195 | if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) { | |
196 | if (!(skb = skb_unshare(skb, GFP_ATOMIC)) | |
84fa7933 PM |
197 | || (skb->ip_summed == CHECKSUM_PARTIAL |
198 | && skb_checksum_help(skb))) { | |
c865e5d9 SH |
199 | sch->qstats.drops++; |
200 | return NET_XMIT_DROP; | |
201 | } | |
202 | ||
203 | skb->data[net_random() % skb_headlen(skb)] ^= 1<<(net_random() % 8); | |
204 | } | |
205 | ||
89e1df74 | 206 | cb = (struct netem_skb_cb *)skb->cb; |
0dca51d3 SH |
207 | if (q->gap == 0 /* not doing reordering */ |
208 | || q->counter < q->gap /* inside last reordering gap */ | |
209 | || q->reorder < get_crandom(&q->reorder_cor)) { | |
0f9f32ac | 210 | psched_time_t now; |
07aaa115 SH |
211 | psched_tdiff_t delay; |
212 | ||
213 | delay = tabledist(q->latency, q->jitter, | |
214 | &q->delay_cor, q->delay_dist); | |
215 | ||
0f9f32ac | 216 | PSCHED_GET_TIME(now); |
07aaa115 | 217 | PSCHED_TADD2(now, delay, cb->time_to_send); |
1da177e4 LT |
218 | ++q->counter; |
219 | ret = q->qdisc->enqueue(skb, q->qdisc); | |
220 | } else { | |
10297b99 | 221 | /* |
0dca51d3 SH |
222 | * Do re-ordering by putting one out of N packets at the front |
223 | * of the queue. | |
224 | */ | |
0f9f32ac | 225 | PSCHED_GET_TIME(cb->time_to_send); |
0dca51d3 | 226 | q->counter = 0; |
0f9f32ac | 227 | ret = q->qdisc->ops->requeue(skb, q->qdisc); |
1da177e4 LT |
228 | } |
229 | ||
230 | if (likely(ret == NET_XMIT_SUCCESS)) { | |
231 | sch->q.qlen++; | |
232 | sch->bstats.bytes += skb->len; | |
233 | sch->bstats.packets++; | |
234 | } else | |
235 | sch->qstats.drops++; | |
236 | ||
d5d75cd6 | 237 | pr_debug("netem: enqueue ret %d\n", ret); |
1da177e4 LT |
238 | return ret; |
239 | } | |
240 | ||
241 | /* Requeue packets but don't change time stamp */ | |
242 | static int netem_requeue(struct sk_buff *skb, struct Qdisc *sch) | |
243 | { | |
244 | struct netem_sched_data *q = qdisc_priv(sch); | |
245 | int ret; | |
246 | ||
247 | if ((ret = q->qdisc->ops->requeue(skb, q->qdisc)) == 0) { | |
248 | sch->q.qlen++; | |
249 | sch->qstats.requeues++; | |
250 | } | |
251 | ||
252 | return ret; | |
253 | } | |
254 | ||
255 | static unsigned int netem_drop(struct Qdisc* sch) | |
256 | { | |
257 | struct netem_sched_data *q = qdisc_priv(sch); | |
6d037a26 | 258 | unsigned int len = 0; |
1da177e4 | 259 | |
6d037a26 | 260 | if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) { |
1da177e4 LT |
261 | sch->q.qlen--; |
262 | sch->qstats.drops++; | |
263 | } | |
264 | return len; | |
265 | } | |
266 | ||
1da177e4 LT |
267 | static struct sk_buff *netem_dequeue(struct Qdisc *sch) |
268 | { | |
269 | struct netem_sched_data *q = qdisc_priv(sch); | |
270 | struct sk_buff *skb; | |
271 | ||
272 | skb = q->qdisc->dequeue(q->qdisc); | |
771018e7 | 273 | if (skb) { |
0f9f32ac SH |
274 | const struct netem_skb_cb *cb |
275 | = (const struct netem_skb_cb *)skb->cb; | |
276 | psched_time_t now; | |
0f9f32ac SH |
277 | |
278 | /* if more time remaining? */ | |
279 | PSCHED_GET_TIME(now); | |
07aaa115 SH |
280 | |
281 | if (PSCHED_TLESS(cb->time_to_send, now)) { | |
0f9f32ac SH |
282 | pr_debug("netem_dequeue: return skb=%p\n", skb); |
283 | sch->q.qlen--; | |
284 | sch->flags &= ~TCQ_F_THROTTLED; | |
285 | return skb; | |
07aaa115 | 286 | } else { |
59cb5c67 | 287 | qdisc_watchdog_schedule(&q->watchdog, cb->time_to_send); |
0f9f32ac | 288 | |
07aaa115 | 289 | if (q->qdisc->ops->requeue(skb, q->qdisc) != NET_XMIT_SUCCESS) { |
e488eafc | 290 | qdisc_tree_decrease_qlen(q->qdisc, 1); |
07aaa115 | 291 | sch->qstats.drops++; |
07aaa115 SH |
292 | printk(KERN_ERR "netem: queue discpline %s could not requeue\n", |
293 | q->qdisc->ops->id); | |
07aaa115 | 294 | } |
07aaa115 | 295 | } |
0f9f32ac SH |
296 | } |
297 | ||
298 | return NULL; | |
1da177e4 LT |
299 | } |
300 | ||
1da177e4 LT |
301 | static void netem_reset(struct Qdisc *sch) |
302 | { | |
303 | struct netem_sched_data *q = qdisc_priv(sch); | |
304 | ||
305 | qdisc_reset(q->qdisc); | |
1da177e4 | 306 | sch->q.qlen = 0; |
59cb5c67 | 307 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
308 | } |
309 | ||
300ce174 | 310 | /* Pass size change message down to embedded FIFO */ |
1da177e4 LT |
311 | static int set_fifo_limit(struct Qdisc *q, int limit) |
312 | { | |
10297b99 | 313 | struct rtattr *rta; |
1da177e4 LT |
314 | int ret = -ENOMEM; |
315 | ||
300ce174 SH |
316 | /* Hack to avoid sending change message to non-FIFO */ |
317 | if (strncmp(q->ops->id + 1, "fifo", 4) != 0) | |
318 | return 0; | |
319 | ||
1da177e4 LT |
320 | rta = kmalloc(RTA_LENGTH(sizeof(struct tc_fifo_qopt)), GFP_KERNEL); |
321 | if (rta) { | |
322 | rta->rta_type = RTM_NEWQDISC; | |
10297b99 | 323 | rta->rta_len = RTA_LENGTH(sizeof(struct tc_fifo_qopt)); |
1da177e4 | 324 | ((struct tc_fifo_qopt *)RTA_DATA(rta))->limit = limit; |
10297b99 | 325 | |
1da177e4 LT |
326 | ret = q->ops->change(q, rta); |
327 | kfree(rta); | |
328 | } | |
329 | return ret; | |
330 | } | |
331 | ||
332 | /* | |
333 | * Distribution data is a variable size payload containing | |
334 | * signed 16 bit values. | |
335 | */ | |
336 | static int get_dist_table(struct Qdisc *sch, const struct rtattr *attr) | |
337 | { | |
338 | struct netem_sched_data *q = qdisc_priv(sch); | |
339 | unsigned long n = RTA_PAYLOAD(attr)/sizeof(__s16); | |
340 | const __s16 *data = RTA_DATA(attr); | |
341 | struct disttable *d; | |
342 | int i; | |
343 | ||
344 | if (n > 65536) | |
345 | return -EINVAL; | |
346 | ||
347 | d = kmalloc(sizeof(*d) + n*sizeof(d->table[0]), GFP_KERNEL); | |
348 | if (!d) | |
349 | return -ENOMEM; | |
350 | ||
351 | d->size = n; | |
352 | for (i = 0; i < n; i++) | |
353 | d->table[i] = data[i]; | |
10297b99 | 354 | |
1da177e4 LT |
355 | spin_lock_bh(&sch->dev->queue_lock); |
356 | d = xchg(&q->delay_dist, d); | |
357 | spin_unlock_bh(&sch->dev->queue_lock); | |
358 | ||
359 | kfree(d); | |
360 | return 0; | |
361 | } | |
362 | ||
363 | static int get_correlation(struct Qdisc *sch, const struct rtattr *attr) | |
364 | { | |
365 | struct netem_sched_data *q = qdisc_priv(sch); | |
366 | const struct tc_netem_corr *c = RTA_DATA(attr); | |
367 | ||
368 | if (RTA_PAYLOAD(attr) != sizeof(*c)) | |
369 | return -EINVAL; | |
370 | ||
371 | init_crandom(&q->delay_cor, c->delay_corr); | |
372 | init_crandom(&q->loss_cor, c->loss_corr); | |
373 | init_crandom(&q->dup_cor, c->dup_corr); | |
374 | return 0; | |
375 | } | |
376 | ||
0dca51d3 SH |
377 | static int get_reorder(struct Qdisc *sch, const struct rtattr *attr) |
378 | { | |
379 | struct netem_sched_data *q = qdisc_priv(sch); | |
380 | const struct tc_netem_reorder *r = RTA_DATA(attr); | |
381 | ||
382 | if (RTA_PAYLOAD(attr) != sizeof(*r)) | |
383 | return -EINVAL; | |
384 | ||
385 | q->reorder = r->probability; | |
386 | init_crandom(&q->reorder_cor, r->correlation); | |
387 | return 0; | |
388 | } | |
389 | ||
c865e5d9 SH |
390 | static int get_corrupt(struct Qdisc *sch, const struct rtattr *attr) |
391 | { | |
392 | struct netem_sched_data *q = qdisc_priv(sch); | |
393 | const struct tc_netem_corrupt *r = RTA_DATA(attr); | |
394 | ||
395 | if (RTA_PAYLOAD(attr) != sizeof(*r)) | |
396 | return -EINVAL; | |
397 | ||
398 | q->corrupt = r->probability; | |
399 | init_crandom(&q->corrupt_cor, r->correlation); | |
400 | return 0; | |
401 | } | |
402 | ||
403 | /* Parse netlink message to set options */ | |
1da177e4 LT |
404 | static int netem_change(struct Qdisc *sch, struct rtattr *opt) |
405 | { | |
406 | struct netem_sched_data *q = qdisc_priv(sch); | |
407 | struct tc_netem_qopt *qopt; | |
408 | int ret; | |
10297b99 | 409 | |
1da177e4 LT |
410 | if (opt == NULL || RTA_PAYLOAD(opt) < sizeof(*qopt)) |
411 | return -EINVAL; | |
412 | ||
413 | qopt = RTA_DATA(opt); | |
414 | ret = set_fifo_limit(q->qdisc, qopt->limit); | |
415 | if (ret) { | |
416 | pr_debug("netem: can't set fifo limit\n"); | |
417 | return ret; | |
418 | } | |
10297b99 | 419 | |
1da177e4 LT |
420 | q->latency = qopt->latency; |
421 | q->jitter = qopt->jitter; | |
422 | q->limit = qopt->limit; | |
423 | q->gap = qopt->gap; | |
0dca51d3 | 424 | q->counter = 0; |
1da177e4 LT |
425 | q->loss = qopt->loss; |
426 | q->duplicate = qopt->duplicate; | |
427 | ||
0dca51d3 SH |
428 | /* for compatiablity with earlier versions. |
429 | * if gap is set, need to assume 100% probablity | |
430 | */ | |
431 | q->reorder = ~0; | |
432 | ||
1da177e4 LT |
433 | /* Handle nested options after initial queue options. |
434 | * Should have put all options in nested format but too late now. | |
10297b99 | 435 | */ |
1da177e4 LT |
436 | if (RTA_PAYLOAD(opt) > sizeof(*qopt)) { |
437 | struct rtattr *tb[TCA_NETEM_MAX]; | |
10297b99 | 438 | if (rtattr_parse(tb, TCA_NETEM_MAX, |
1da177e4 LT |
439 | RTA_DATA(opt) + sizeof(*qopt), |
440 | RTA_PAYLOAD(opt) - sizeof(*qopt))) | |
441 | return -EINVAL; | |
442 | ||
443 | if (tb[TCA_NETEM_CORR-1]) { | |
444 | ret = get_correlation(sch, tb[TCA_NETEM_CORR-1]); | |
445 | if (ret) | |
446 | return ret; | |
447 | } | |
448 | ||
449 | if (tb[TCA_NETEM_DELAY_DIST-1]) { | |
450 | ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST-1]); | |
451 | if (ret) | |
452 | return ret; | |
453 | } | |
c865e5d9 | 454 | |
0dca51d3 SH |
455 | if (tb[TCA_NETEM_REORDER-1]) { |
456 | ret = get_reorder(sch, tb[TCA_NETEM_REORDER-1]); | |
457 | if (ret) | |
458 | return ret; | |
459 | } | |
1da177e4 | 460 | |
c865e5d9 SH |
461 | if (tb[TCA_NETEM_CORRUPT-1]) { |
462 | ret = get_corrupt(sch, tb[TCA_NETEM_CORRUPT-1]); | |
463 | if (ret) | |
464 | return ret; | |
465 | } | |
466 | } | |
1da177e4 LT |
467 | |
468 | return 0; | |
469 | } | |
470 | ||
300ce174 SH |
471 | /* |
472 | * Special case version of FIFO queue for use by netem. | |
473 | * It queues in order based on timestamps in skb's | |
474 | */ | |
475 | struct fifo_sched_data { | |
476 | u32 limit; | |
477 | }; | |
478 | ||
479 | static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch) | |
480 | { | |
481 | struct fifo_sched_data *q = qdisc_priv(sch); | |
482 | struct sk_buff_head *list = &sch->q; | |
483 | const struct netem_skb_cb *ncb | |
484 | = (const struct netem_skb_cb *)nskb->cb; | |
485 | struct sk_buff *skb; | |
486 | ||
487 | if (likely(skb_queue_len(list) < q->limit)) { | |
488 | skb_queue_reverse_walk(list, skb) { | |
489 | const struct netem_skb_cb *cb | |
490 | = (const struct netem_skb_cb *)skb->cb; | |
491 | ||
aa875166 | 492 | if (!PSCHED_TLESS(ncb->time_to_send, cb->time_to_send)) |
300ce174 SH |
493 | break; |
494 | } | |
495 | ||
496 | __skb_queue_after(list, skb, nskb); | |
497 | ||
498 | sch->qstats.backlog += nskb->len; | |
499 | sch->bstats.bytes += nskb->len; | |
500 | sch->bstats.packets++; | |
501 | ||
502 | return NET_XMIT_SUCCESS; | |
503 | } | |
504 | ||
505 | return qdisc_drop(nskb, sch); | |
506 | } | |
507 | ||
508 | static int tfifo_init(struct Qdisc *sch, struct rtattr *opt) | |
509 | { | |
510 | struct fifo_sched_data *q = qdisc_priv(sch); | |
511 | ||
512 | if (opt) { | |
513 | struct tc_fifo_qopt *ctl = RTA_DATA(opt); | |
514 | if (RTA_PAYLOAD(opt) < sizeof(*ctl)) | |
515 | return -EINVAL; | |
516 | ||
517 | q->limit = ctl->limit; | |
518 | } else | |
519 | q->limit = max_t(u32, sch->dev->tx_queue_len, 1); | |
520 | ||
521 | return 0; | |
522 | } | |
523 | ||
524 | static int tfifo_dump(struct Qdisc *sch, struct sk_buff *skb) | |
525 | { | |
526 | struct fifo_sched_data *q = qdisc_priv(sch); | |
527 | struct tc_fifo_qopt opt = { .limit = q->limit }; | |
528 | ||
529 | RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); | |
530 | return skb->len; | |
531 | ||
532 | rtattr_failure: | |
533 | return -1; | |
534 | } | |
535 | ||
536 | static struct Qdisc_ops tfifo_qdisc_ops = { | |
537 | .id = "tfifo", | |
538 | .priv_size = sizeof(struct fifo_sched_data), | |
539 | .enqueue = tfifo_enqueue, | |
540 | .dequeue = qdisc_dequeue_head, | |
541 | .requeue = qdisc_requeue, | |
542 | .drop = qdisc_queue_drop, | |
543 | .init = tfifo_init, | |
544 | .reset = qdisc_reset_queue, | |
545 | .change = tfifo_init, | |
546 | .dump = tfifo_dump, | |
547 | }; | |
548 | ||
1da177e4 LT |
549 | static int netem_init(struct Qdisc *sch, struct rtattr *opt) |
550 | { | |
551 | struct netem_sched_data *q = qdisc_priv(sch); | |
552 | int ret; | |
553 | ||
554 | if (!opt) | |
555 | return -EINVAL; | |
556 | ||
59cb5c67 | 557 | qdisc_watchdog_init(&q->watchdog, sch); |
1da177e4 | 558 | |
9f9afec4 PM |
559 | q->qdisc = qdisc_create_dflt(sch->dev, &tfifo_qdisc_ops, |
560 | TC_H_MAKE(sch->handle, 1)); | |
1da177e4 LT |
561 | if (!q->qdisc) { |
562 | pr_debug("netem: qdisc create failed\n"); | |
563 | return -ENOMEM; | |
564 | } | |
565 | ||
566 | ret = netem_change(sch, opt); | |
567 | if (ret) { | |
568 | pr_debug("netem: change failed\n"); | |
569 | qdisc_destroy(q->qdisc); | |
570 | } | |
571 | return ret; | |
572 | } | |
573 | ||
574 | static void netem_destroy(struct Qdisc *sch) | |
575 | { | |
576 | struct netem_sched_data *q = qdisc_priv(sch); | |
577 | ||
59cb5c67 | 578 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
579 | qdisc_destroy(q->qdisc); |
580 | kfree(q->delay_dist); | |
581 | } | |
582 | ||
583 | static int netem_dump(struct Qdisc *sch, struct sk_buff *skb) | |
584 | { | |
585 | const struct netem_sched_data *q = qdisc_priv(sch); | |
27a884dc | 586 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
587 | struct rtattr *rta = (struct rtattr *) b; |
588 | struct tc_netem_qopt qopt; | |
589 | struct tc_netem_corr cor; | |
0dca51d3 | 590 | struct tc_netem_reorder reorder; |
c865e5d9 | 591 | struct tc_netem_corrupt corrupt; |
1da177e4 LT |
592 | |
593 | qopt.latency = q->latency; | |
594 | qopt.jitter = q->jitter; | |
595 | qopt.limit = q->limit; | |
596 | qopt.loss = q->loss; | |
597 | qopt.gap = q->gap; | |
598 | qopt.duplicate = q->duplicate; | |
599 | RTA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt); | |
600 | ||
601 | cor.delay_corr = q->delay_cor.rho; | |
602 | cor.loss_corr = q->loss_cor.rho; | |
603 | cor.dup_corr = q->dup_cor.rho; | |
604 | RTA_PUT(skb, TCA_NETEM_CORR, sizeof(cor), &cor); | |
0dca51d3 SH |
605 | |
606 | reorder.probability = q->reorder; | |
607 | reorder.correlation = q->reorder_cor.rho; | |
608 | RTA_PUT(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder); | |
609 | ||
c865e5d9 SH |
610 | corrupt.probability = q->corrupt; |
611 | corrupt.correlation = q->corrupt_cor.rho; | |
612 | RTA_PUT(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt); | |
613 | ||
27a884dc | 614 | rta->rta_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
615 | |
616 | return skb->len; | |
617 | ||
618 | rtattr_failure: | |
619 | skb_trim(skb, b - skb->data); | |
620 | return -1; | |
621 | } | |
622 | ||
623 | static int netem_dump_class(struct Qdisc *sch, unsigned long cl, | |
624 | struct sk_buff *skb, struct tcmsg *tcm) | |
625 | { | |
626 | struct netem_sched_data *q = qdisc_priv(sch); | |
627 | ||
628 | if (cl != 1) /* only one class */ | |
629 | return -ENOENT; | |
630 | ||
631 | tcm->tcm_handle |= TC_H_MIN(1); | |
632 | tcm->tcm_info = q->qdisc->handle; | |
633 | ||
634 | return 0; | |
635 | } | |
636 | ||
637 | static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
638 | struct Qdisc **old) | |
639 | { | |
640 | struct netem_sched_data *q = qdisc_priv(sch); | |
641 | ||
642 | if (new == NULL) | |
643 | new = &noop_qdisc; | |
644 | ||
645 | sch_tree_lock(sch); | |
646 | *old = xchg(&q->qdisc, new); | |
5e50da01 | 647 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); |
1da177e4 | 648 | qdisc_reset(*old); |
1da177e4 LT |
649 | sch_tree_unlock(sch); |
650 | ||
651 | return 0; | |
652 | } | |
653 | ||
654 | static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg) | |
655 | { | |
656 | struct netem_sched_data *q = qdisc_priv(sch); | |
657 | return q->qdisc; | |
658 | } | |
659 | ||
660 | static unsigned long netem_get(struct Qdisc *sch, u32 classid) | |
661 | { | |
662 | return 1; | |
663 | } | |
664 | ||
665 | static void netem_put(struct Qdisc *sch, unsigned long arg) | |
666 | { | |
667 | } | |
668 | ||
10297b99 | 669 | static int netem_change_class(struct Qdisc *sch, u32 classid, u32 parentid, |
1da177e4 LT |
670 | struct rtattr **tca, unsigned long *arg) |
671 | { | |
672 | return -ENOSYS; | |
673 | } | |
674 | ||
675 | static int netem_delete(struct Qdisc *sch, unsigned long arg) | |
676 | { | |
677 | return -ENOSYS; | |
678 | } | |
679 | ||
680 | static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker) | |
681 | { | |
682 | if (!walker->stop) { | |
683 | if (walker->count >= walker->skip) | |
684 | if (walker->fn(sch, 1, walker) < 0) { | |
685 | walker->stop = 1; | |
686 | return; | |
687 | } | |
688 | walker->count++; | |
689 | } | |
690 | } | |
691 | ||
692 | static struct tcf_proto **netem_find_tcf(struct Qdisc *sch, unsigned long cl) | |
693 | { | |
694 | return NULL; | |
695 | } | |
696 | ||
697 | static struct Qdisc_class_ops netem_class_ops = { | |
698 | .graft = netem_graft, | |
699 | .leaf = netem_leaf, | |
700 | .get = netem_get, | |
701 | .put = netem_put, | |
702 | .change = netem_change_class, | |
703 | .delete = netem_delete, | |
704 | .walk = netem_walk, | |
705 | .tcf_chain = netem_find_tcf, | |
706 | .dump = netem_dump_class, | |
707 | }; | |
708 | ||
709 | static struct Qdisc_ops netem_qdisc_ops = { | |
710 | .id = "netem", | |
711 | .cl_ops = &netem_class_ops, | |
712 | .priv_size = sizeof(struct netem_sched_data), | |
713 | .enqueue = netem_enqueue, | |
714 | .dequeue = netem_dequeue, | |
715 | .requeue = netem_requeue, | |
716 | .drop = netem_drop, | |
717 | .init = netem_init, | |
718 | .reset = netem_reset, | |
719 | .destroy = netem_destroy, | |
720 | .change = netem_change, | |
721 | .dump = netem_dump, | |
722 | .owner = THIS_MODULE, | |
723 | }; | |
724 | ||
725 | ||
726 | static int __init netem_module_init(void) | |
727 | { | |
eb229c4c | 728 | pr_info("netem: version " VERSION "\n"); |
1da177e4 LT |
729 | return register_qdisc(&netem_qdisc_ops); |
730 | } | |
731 | static void __exit netem_module_exit(void) | |
732 | { | |
733 | unregister_qdisc(&netem_qdisc_ops); | |
734 | } | |
735 | module_init(netem_module_init) | |
736 | module_exit(netem_module_exit) | |
737 | MODULE_LICENSE("GPL"); |