]>
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 | ||
dc5fc579 | 25 | #include <net/netlink.h> |
1da177e4 LT |
26 | #include <net/pkt_sched.h> |
27 | ||
c865e5d9 | 28 | #define VERSION "1.2" |
eb229c4c | 29 | |
1da177e4 LT |
30 | /* 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 | ||
56 | struct netem_sched_data { | |
57 | struct Qdisc *qdisc; | |
59cb5c67 | 58 | struct qdisc_watchdog watchdog; |
1da177e4 LT |
59 | |
60 | u32 latency; | |
61 | u32 loss; | |
62 | u32 limit; | |
63 | u32 counter; | |
64 | u32 gap; | |
65 | u32 jitter; | |
66 | u32 duplicate; | |
0dca51d3 | 67 | u32 reorder; |
c865e5d9 | 68 | u32 corrupt; |
1da177e4 LT |
69 | |
70 | struct crndstate { | |
71 | unsigned long last; | |
72 | unsigned long rho; | |
c865e5d9 | 73 | } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor; |
1da177e4 LT |
74 | |
75 | struct disttable { | |
76 | u32 size; | |
77 | s16 table[0]; | |
78 | } *delay_dist; | |
79 | }; | |
80 | ||
81 | /* Time stamp put into socket buffer control block */ | |
82 | struct 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 | */ | |
89 | static 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 | */ | |
99 | static 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 | */ | |
10297b99 | 118 | static long tabledist(unsigned long mu, long sigma, |
1da177e4 LT |
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 */ | |
10297b99 | 130 | if (dist == NULL) |
1da177e4 LT |
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 | ||
0afb51e7 SH |
143 | /* |
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 | */ | |
1da177e4 LT |
149 | static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
150 | { | |
151 | struct netem_sched_data *q = qdisc_priv(sch); | |
89e1df74 GC |
152 | /* We don't fill cb now as skb_unshare() may invalidate it */ |
153 | struct netem_skb_cb *cb; | |
0afb51e7 | 154 | struct sk_buff *skb2; |
1da177e4 | 155 | int ret; |
0afb51e7 | 156 | int count = 1; |
1da177e4 | 157 | |
771018e7 | 158 | pr_debug("netem_enqueue skb=%p\n", skb); |
1da177e4 | 159 | |
0afb51e7 SH |
160 | /* Random duplication */ |
161 | if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor)) | |
162 | ++count; | |
163 | ||
1da177e4 | 164 | /* Random packet drop 0 => none, ~0 => all */ |
0afb51e7 SH |
165 | if (q->loss && q->loss >= get_crandom(&q->loss_cor)) |
166 | --count; | |
167 | ||
168 | if (count == 0) { | |
1da177e4 LT |
169 | sch->qstats.drops++; |
170 | kfree_skb(skb); | |
89bbb0a3 | 171 | return NET_XMIT_BYPASS; |
1da177e4 LT |
172 | } |
173 | ||
4e8a5201 DM |
174 | skb_orphan(skb); |
175 | ||
0afb51e7 SH |
176 | /* |
177 | * If we need to duplicate packet, then re-insert at top of the | |
178 | * qdisc tree, since parent queuer expects that only one | |
179 | * skb will be queued. | |
180 | */ | |
181 | if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) { | |
182 | struct Qdisc *rootq = sch->dev->qdisc; | |
183 | u32 dupsave = q->duplicate; /* prevent duplicating a dup... */ | |
184 | q->duplicate = 0; | |
185 | ||
186 | rootq->enqueue(skb2, rootq); | |
187 | q->duplicate = dupsave; | |
1da177e4 LT |
188 | } |
189 | ||
c865e5d9 SH |
190 | /* |
191 | * Randomized packet corruption. | |
192 | * Make copy if needed since we are modifying | |
193 | * If packet is going to be hardware checksummed, then | |
194 | * do it now in software before we mangle it. | |
195 | */ | |
196 | if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) { | |
197 | if (!(skb = skb_unshare(skb, GFP_ATOMIC)) | |
84fa7933 PM |
198 | || (skb->ip_summed == CHECKSUM_PARTIAL |
199 | && skb_checksum_help(skb))) { | |
c865e5d9 SH |
200 | sch->qstats.drops++; |
201 | return NET_XMIT_DROP; | |
202 | } | |
203 | ||
204 | skb->data[net_random() % skb_headlen(skb)] ^= 1<<(net_random() % 8); | |
205 | } | |
206 | ||
89e1df74 | 207 | cb = (struct netem_skb_cb *)skb->cb; |
0dca51d3 SH |
208 | if (q->gap == 0 /* not doing reordering */ |
209 | || q->counter < q->gap /* inside last reordering gap */ | |
210 | || q->reorder < get_crandom(&q->reorder_cor)) { | |
0f9f32ac | 211 | psched_time_t now; |
07aaa115 SH |
212 | psched_tdiff_t delay; |
213 | ||
214 | delay = tabledist(q->latency, q->jitter, | |
215 | &q->delay_cor, q->delay_dist); | |
216 | ||
0f9f32ac | 217 | PSCHED_GET_TIME(now); |
07aaa115 | 218 | PSCHED_TADD2(now, delay, cb->time_to_send); |
1da177e4 LT |
219 | ++q->counter; |
220 | ret = q->qdisc->enqueue(skb, q->qdisc); | |
221 | } else { | |
10297b99 | 222 | /* |
0dca51d3 SH |
223 | * Do re-ordering by putting one out of N packets at the front |
224 | * of the queue. | |
225 | */ | |
0f9f32ac | 226 | PSCHED_GET_TIME(cb->time_to_send); |
0dca51d3 | 227 | q->counter = 0; |
0f9f32ac | 228 | ret = q->qdisc->ops->requeue(skb, q->qdisc); |
1da177e4 LT |
229 | } |
230 | ||
231 | if (likely(ret == NET_XMIT_SUCCESS)) { | |
232 | sch->q.qlen++; | |
233 | sch->bstats.bytes += skb->len; | |
234 | sch->bstats.packets++; | |
235 | } else | |
236 | sch->qstats.drops++; | |
237 | ||
d5d75cd6 | 238 | pr_debug("netem: enqueue ret %d\n", ret); |
1da177e4 LT |
239 | return ret; |
240 | } | |
241 | ||
242 | /* Requeue packets but don't change time stamp */ | |
243 | static int netem_requeue(struct sk_buff *skb, struct Qdisc *sch) | |
244 | { | |
245 | struct netem_sched_data *q = qdisc_priv(sch); | |
246 | int ret; | |
247 | ||
248 | if ((ret = q->qdisc->ops->requeue(skb, q->qdisc)) == 0) { | |
249 | sch->q.qlen++; | |
250 | sch->qstats.requeues++; | |
251 | } | |
252 | ||
253 | return ret; | |
254 | } | |
255 | ||
256 | static unsigned int netem_drop(struct Qdisc* sch) | |
257 | { | |
258 | struct netem_sched_data *q = qdisc_priv(sch); | |
6d037a26 | 259 | unsigned int len = 0; |
1da177e4 | 260 | |
6d037a26 | 261 | if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) { |
1da177e4 LT |
262 | sch->q.qlen--; |
263 | sch->qstats.drops++; | |
264 | } | |
265 | return len; | |
266 | } | |
267 | ||
1da177e4 LT |
268 | static struct sk_buff *netem_dequeue(struct Qdisc *sch) |
269 | { | |
270 | struct netem_sched_data *q = qdisc_priv(sch); | |
271 | struct sk_buff *skb; | |
272 | ||
273 | skb = q->qdisc->dequeue(q->qdisc); | |
771018e7 | 274 | if (skb) { |
0f9f32ac SH |
275 | const struct netem_skb_cb *cb |
276 | = (const struct netem_skb_cb *)skb->cb; | |
277 | psched_time_t now; | |
0f9f32ac SH |
278 | |
279 | /* if more time remaining? */ | |
280 | PSCHED_GET_TIME(now); | |
07aaa115 SH |
281 | |
282 | if (PSCHED_TLESS(cb->time_to_send, now)) { | |
0f9f32ac SH |
283 | pr_debug("netem_dequeue: return skb=%p\n", skb); |
284 | sch->q.qlen--; | |
285 | sch->flags &= ~TCQ_F_THROTTLED; | |
286 | return skb; | |
07aaa115 | 287 | } else { |
59cb5c67 | 288 | qdisc_watchdog_schedule(&q->watchdog, cb->time_to_send); |
0f9f32ac | 289 | |
07aaa115 | 290 | if (q->qdisc->ops->requeue(skb, q->qdisc) != NET_XMIT_SUCCESS) { |
e488eafc | 291 | qdisc_tree_decrease_qlen(q->qdisc, 1); |
07aaa115 | 292 | sch->qstats.drops++; |
07aaa115 SH |
293 | printk(KERN_ERR "netem: queue discpline %s could not requeue\n", |
294 | q->qdisc->ops->id); | |
07aaa115 | 295 | } |
07aaa115 | 296 | } |
0f9f32ac SH |
297 | } |
298 | ||
299 | return NULL; | |
1da177e4 LT |
300 | } |
301 | ||
1da177e4 LT |
302 | static void netem_reset(struct Qdisc *sch) |
303 | { | |
304 | struct netem_sched_data *q = qdisc_priv(sch); | |
305 | ||
306 | qdisc_reset(q->qdisc); | |
1da177e4 | 307 | sch->q.qlen = 0; |
59cb5c67 | 308 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
309 | } |
310 | ||
300ce174 | 311 | /* Pass size change message down to embedded FIFO */ |
1da177e4 LT |
312 | static int set_fifo_limit(struct Qdisc *q, int limit) |
313 | { | |
10297b99 | 314 | struct rtattr *rta; |
1da177e4 LT |
315 | int ret = -ENOMEM; |
316 | ||
300ce174 SH |
317 | /* Hack to avoid sending change message to non-FIFO */ |
318 | if (strncmp(q->ops->id + 1, "fifo", 4) != 0) | |
319 | return 0; | |
320 | ||
1da177e4 LT |
321 | rta = kmalloc(RTA_LENGTH(sizeof(struct tc_fifo_qopt)), GFP_KERNEL); |
322 | if (rta) { | |
323 | rta->rta_type = RTM_NEWQDISC; | |
10297b99 | 324 | rta->rta_len = RTA_LENGTH(sizeof(struct tc_fifo_qopt)); |
1da177e4 | 325 | ((struct tc_fifo_qopt *)RTA_DATA(rta))->limit = limit; |
10297b99 | 326 | |
1da177e4 LT |
327 | ret = q->ops->change(q, rta); |
328 | kfree(rta); | |
329 | } | |
330 | return ret; | |
331 | } | |
332 | ||
333 | /* | |
334 | * Distribution data is a variable size payload containing | |
335 | * signed 16 bit values. | |
336 | */ | |
337 | static int get_dist_table(struct Qdisc *sch, const struct rtattr *attr) | |
338 | { | |
339 | struct netem_sched_data *q = qdisc_priv(sch); | |
340 | unsigned long n = RTA_PAYLOAD(attr)/sizeof(__s16); | |
341 | const __s16 *data = RTA_DATA(attr); | |
342 | struct disttable *d; | |
343 | int i; | |
344 | ||
345 | if (n > 65536) | |
346 | return -EINVAL; | |
347 | ||
348 | d = kmalloc(sizeof(*d) + n*sizeof(d->table[0]), GFP_KERNEL); | |
349 | if (!d) | |
350 | return -ENOMEM; | |
351 | ||
352 | d->size = n; | |
353 | for (i = 0; i < n; i++) | |
354 | d->table[i] = data[i]; | |
10297b99 | 355 | |
1da177e4 LT |
356 | spin_lock_bh(&sch->dev->queue_lock); |
357 | d = xchg(&q->delay_dist, d); | |
358 | spin_unlock_bh(&sch->dev->queue_lock); | |
359 | ||
360 | kfree(d); | |
361 | return 0; | |
362 | } | |
363 | ||
364 | static int get_correlation(struct Qdisc *sch, const struct rtattr *attr) | |
365 | { | |
366 | struct netem_sched_data *q = qdisc_priv(sch); | |
367 | const struct tc_netem_corr *c = RTA_DATA(attr); | |
368 | ||
369 | if (RTA_PAYLOAD(attr) != sizeof(*c)) | |
370 | return -EINVAL; | |
371 | ||
372 | init_crandom(&q->delay_cor, c->delay_corr); | |
373 | init_crandom(&q->loss_cor, c->loss_corr); | |
374 | init_crandom(&q->dup_cor, c->dup_corr); | |
375 | return 0; | |
376 | } | |
377 | ||
0dca51d3 SH |
378 | static int get_reorder(struct Qdisc *sch, const struct rtattr *attr) |
379 | { | |
380 | struct netem_sched_data *q = qdisc_priv(sch); | |
381 | const struct tc_netem_reorder *r = RTA_DATA(attr); | |
382 | ||
383 | if (RTA_PAYLOAD(attr) != sizeof(*r)) | |
384 | return -EINVAL; | |
385 | ||
386 | q->reorder = r->probability; | |
387 | init_crandom(&q->reorder_cor, r->correlation); | |
388 | return 0; | |
389 | } | |
390 | ||
c865e5d9 SH |
391 | static int get_corrupt(struct Qdisc *sch, const struct rtattr *attr) |
392 | { | |
393 | struct netem_sched_data *q = qdisc_priv(sch); | |
394 | const struct tc_netem_corrupt *r = RTA_DATA(attr); | |
395 | ||
396 | if (RTA_PAYLOAD(attr) != sizeof(*r)) | |
397 | return -EINVAL; | |
398 | ||
399 | q->corrupt = r->probability; | |
400 | init_crandom(&q->corrupt_cor, r->correlation); | |
401 | return 0; | |
402 | } | |
403 | ||
404 | /* Parse netlink message to set options */ | |
1da177e4 LT |
405 | static int netem_change(struct Qdisc *sch, struct rtattr *opt) |
406 | { | |
407 | struct netem_sched_data *q = qdisc_priv(sch); | |
408 | struct tc_netem_qopt *qopt; | |
409 | int ret; | |
10297b99 | 410 | |
1da177e4 LT |
411 | if (opt == NULL || RTA_PAYLOAD(opt) < sizeof(*qopt)) |
412 | return -EINVAL; | |
413 | ||
414 | qopt = RTA_DATA(opt); | |
415 | ret = set_fifo_limit(q->qdisc, qopt->limit); | |
416 | if (ret) { | |
417 | pr_debug("netem: can't set fifo limit\n"); | |
418 | return ret; | |
419 | } | |
10297b99 | 420 | |
1da177e4 LT |
421 | q->latency = qopt->latency; |
422 | q->jitter = qopt->jitter; | |
423 | q->limit = qopt->limit; | |
424 | q->gap = qopt->gap; | |
0dca51d3 | 425 | q->counter = 0; |
1da177e4 LT |
426 | q->loss = qopt->loss; |
427 | q->duplicate = qopt->duplicate; | |
428 | ||
0dca51d3 SH |
429 | /* for compatiablity with earlier versions. |
430 | * if gap is set, need to assume 100% probablity | |
431 | */ | |
a362e0a7 SH |
432 | if (q->gap) |
433 | q->reorder = ~0; | |
0dca51d3 | 434 | |
1da177e4 LT |
435 | /* Handle nested options after initial queue options. |
436 | * Should have put all options in nested format but too late now. | |
10297b99 | 437 | */ |
1da177e4 LT |
438 | if (RTA_PAYLOAD(opt) > sizeof(*qopt)) { |
439 | struct rtattr *tb[TCA_NETEM_MAX]; | |
10297b99 | 440 | if (rtattr_parse(tb, TCA_NETEM_MAX, |
1da177e4 LT |
441 | RTA_DATA(opt) + sizeof(*qopt), |
442 | RTA_PAYLOAD(opt) - sizeof(*qopt))) | |
443 | return -EINVAL; | |
444 | ||
445 | if (tb[TCA_NETEM_CORR-1]) { | |
446 | ret = get_correlation(sch, tb[TCA_NETEM_CORR-1]); | |
447 | if (ret) | |
448 | return ret; | |
449 | } | |
450 | ||
451 | if (tb[TCA_NETEM_DELAY_DIST-1]) { | |
452 | ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST-1]); | |
453 | if (ret) | |
454 | return ret; | |
455 | } | |
c865e5d9 | 456 | |
0dca51d3 SH |
457 | if (tb[TCA_NETEM_REORDER-1]) { |
458 | ret = get_reorder(sch, tb[TCA_NETEM_REORDER-1]); | |
459 | if (ret) | |
460 | return ret; | |
461 | } | |
1da177e4 | 462 | |
c865e5d9 SH |
463 | if (tb[TCA_NETEM_CORRUPT-1]) { |
464 | ret = get_corrupt(sch, tb[TCA_NETEM_CORRUPT-1]); | |
465 | if (ret) | |
466 | return ret; | |
467 | } | |
468 | } | |
1da177e4 LT |
469 | |
470 | return 0; | |
471 | } | |
472 | ||
300ce174 SH |
473 | /* |
474 | * Special case version of FIFO queue for use by netem. | |
475 | * It queues in order based on timestamps in skb's | |
476 | */ | |
477 | struct fifo_sched_data { | |
478 | u32 limit; | |
479 | }; | |
480 | ||
481 | static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch) | |
482 | { | |
483 | struct fifo_sched_data *q = qdisc_priv(sch); | |
484 | struct sk_buff_head *list = &sch->q; | |
485 | const struct netem_skb_cb *ncb | |
486 | = (const struct netem_skb_cb *)nskb->cb; | |
487 | struct sk_buff *skb; | |
488 | ||
489 | if (likely(skb_queue_len(list) < q->limit)) { | |
490 | skb_queue_reverse_walk(list, skb) { | |
491 | const struct netem_skb_cb *cb | |
492 | = (const struct netem_skb_cb *)skb->cb; | |
493 | ||
aa875166 | 494 | if (!PSCHED_TLESS(ncb->time_to_send, cb->time_to_send)) |
300ce174 SH |
495 | break; |
496 | } | |
497 | ||
498 | __skb_queue_after(list, skb, nskb); | |
499 | ||
500 | sch->qstats.backlog += nskb->len; | |
501 | sch->bstats.bytes += nskb->len; | |
502 | sch->bstats.packets++; | |
503 | ||
504 | return NET_XMIT_SUCCESS; | |
505 | } | |
506 | ||
507 | return qdisc_drop(nskb, sch); | |
508 | } | |
509 | ||
510 | static int tfifo_init(struct Qdisc *sch, struct rtattr *opt) | |
511 | { | |
512 | struct fifo_sched_data *q = qdisc_priv(sch); | |
513 | ||
514 | if (opt) { | |
515 | struct tc_fifo_qopt *ctl = RTA_DATA(opt); | |
516 | if (RTA_PAYLOAD(opt) < sizeof(*ctl)) | |
517 | return -EINVAL; | |
518 | ||
519 | q->limit = ctl->limit; | |
520 | } else | |
521 | q->limit = max_t(u32, sch->dev->tx_queue_len, 1); | |
522 | ||
523 | return 0; | |
524 | } | |
525 | ||
526 | static int tfifo_dump(struct Qdisc *sch, struct sk_buff *skb) | |
527 | { | |
528 | struct fifo_sched_data *q = qdisc_priv(sch); | |
529 | struct tc_fifo_qopt opt = { .limit = q->limit }; | |
530 | ||
531 | RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); | |
532 | return skb->len; | |
533 | ||
534 | rtattr_failure: | |
535 | return -1; | |
536 | } | |
537 | ||
538 | static struct Qdisc_ops tfifo_qdisc_ops = { | |
539 | .id = "tfifo", | |
540 | .priv_size = sizeof(struct fifo_sched_data), | |
541 | .enqueue = tfifo_enqueue, | |
542 | .dequeue = qdisc_dequeue_head, | |
543 | .requeue = qdisc_requeue, | |
544 | .drop = qdisc_queue_drop, | |
545 | .init = tfifo_init, | |
546 | .reset = qdisc_reset_queue, | |
547 | .change = tfifo_init, | |
548 | .dump = tfifo_dump, | |
549 | }; | |
550 | ||
1da177e4 LT |
551 | static int netem_init(struct Qdisc *sch, struct rtattr *opt) |
552 | { | |
553 | struct netem_sched_data *q = qdisc_priv(sch); | |
554 | int ret; | |
555 | ||
556 | if (!opt) | |
557 | return -EINVAL; | |
558 | ||
59cb5c67 | 559 | qdisc_watchdog_init(&q->watchdog, sch); |
1da177e4 | 560 | |
9f9afec4 PM |
561 | q->qdisc = qdisc_create_dflt(sch->dev, &tfifo_qdisc_ops, |
562 | TC_H_MAKE(sch->handle, 1)); | |
1da177e4 LT |
563 | if (!q->qdisc) { |
564 | pr_debug("netem: qdisc create failed\n"); | |
565 | return -ENOMEM; | |
566 | } | |
567 | ||
568 | ret = netem_change(sch, opt); | |
569 | if (ret) { | |
570 | pr_debug("netem: change failed\n"); | |
571 | qdisc_destroy(q->qdisc); | |
572 | } | |
573 | return ret; | |
574 | } | |
575 | ||
576 | static void netem_destroy(struct Qdisc *sch) | |
577 | { | |
578 | struct netem_sched_data *q = qdisc_priv(sch); | |
579 | ||
59cb5c67 | 580 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
581 | qdisc_destroy(q->qdisc); |
582 | kfree(q->delay_dist); | |
583 | } | |
584 | ||
585 | static int netem_dump(struct Qdisc *sch, struct sk_buff *skb) | |
586 | { | |
587 | const struct netem_sched_data *q = qdisc_priv(sch); | |
27a884dc | 588 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
589 | struct rtattr *rta = (struct rtattr *) b; |
590 | struct tc_netem_qopt qopt; | |
591 | struct tc_netem_corr cor; | |
0dca51d3 | 592 | struct tc_netem_reorder reorder; |
c865e5d9 | 593 | struct tc_netem_corrupt corrupt; |
1da177e4 LT |
594 | |
595 | qopt.latency = q->latency; | |
596 | qopt.jitter = q->jitter; | |
597 | qopt.limit = q->limit; | |
598 | qopt.loss = q->loss; | |
599 | qopt.gap = q->gap; | |
600 | qopt.duplicate = q->duplicate; | |
601 | RTA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt); | |
602 | ||
603 | cor.delay_corr = q->delay_cor.rho; | |
604 | cor.loss_corr = q->loss_cor.rho; | |
605 | cor.dup_corr = q->dup_cor.rho; | |
606 | RTA_PUT(skb, TCA_NETEM_CORR, sizeof(cor), &cor); | |
0dca51d3 SH |
607 | |
608 | reorder.probability = q->reorder; | |
609 | reorder.correlation = q->reorder_cor.rho; | |
610 | RTA_PUT(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder); | |
611 | ||
c865e5d9 SH |
612 | corrupt.probability = q->corrupt; |
613 | corrupt.correlation = q->corrupt_cor.rho; | |
614 | RTA_PUT(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt); | |
615 | ||
27a884dc | 616 | rta->rta_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
617 | |
618 | return skb->len; | |
619 | ||
620 | rtattr_failure: | |
dc5fc579 | 621 | nlmsg_trim(skb, b); |
1da177e4 LT |
622 | return -1; |
623 | } | |
624 | ||
625 | static int netem_dump_class(struct Qdisc *sch, unsigned long cl, | |
626 | struct sk_buff *skb, struct tcmsg *tcm) | |
627 | { | |
628 | struct netem_sched_data *q = qdisc_priv(sch); | |
629 | ||
630 | if (cl != 1) /* only one class */ | |
631 | return -ENOENT; | |
632 | ||
633 | tcm->tcm_handle |= TC_H_MIN(1); | |
634 | tcm->tcm_info = q->qdisc->handle; | |
635 | ||
636 | return 0; | |
637 | } | |
638 | ||
639 | static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
640 | struct Qdisc **old) | |
641 | { | |
642 | struct netem_sched_data *q = qdisc_priv(sch); | |
643 | ||
644 | if (new == NULL) | |
645 | new = &noop_qdisc; | |
646 | ||
647 | sch_tree_lock(sch); | |
648 | *old = xchg(&q->qdisc, new); | |
5e50da01 | 649 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); |
1da177e4 | 650 | qdisc_reset(*old); |
1da177e4 LT |
651 | sch_tree_unlock(sch); |
652 | ||
653 | return 0; | |
654 | } | |
655 | ||
656 | static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg) | |
657 | { | |
658 | struct netem_sched_data *q = qdisc_priv(sch); | |
659 | return q->qdisc; | |
660 | } | |
661 | ||
662 | static unsigned long netem_get(struct Qdisc *sch, u32 classid) | |
663 | { | |
664 | return 1; | |
665 | } | |
666 | ||
667 | static void netem_put(struct Qdisc *sch, unsigned long arg) | |
668 | { | |
669 | } | |
670 | ||
10297b99 | 671 | static int netem_change_class(struct Qdisc *sch, u32 classid, u32 parentid, |
1da177e4 LT |
672 | struct rtattr **tca, unsigned long *arg) |
673 | { | |
674 | return -ENOSYS; | |
675 | } | |
676 | ||
677 | static int netem_delete(struct Qdisc *sch, unsigned long arg) | |
678 | { | |
679 | return -ENOSYS; | |
680 | } | |
681 | ||
682 | static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker) | |
683 | { | |
684 | if (!walker->stop) { | |
685 | if (walker->count >= walker->skip) | |
686 | if (walker->fn(sch, 1, walker) < 0) { | |
687 | walker->stop = 1; | |
688 | return; | |
689 | } | |
690 | walker->count++; | |
691 | } | |
692 | } | |
693 | ||
694 | static struct tcf_proto **netem_find_tcf(struct Qdisc *sch, unsigned long cl) | |
695 | { | |
696 | return NULL; | |
697 | } | |
698 | ||
699 | static struct Qdisc_class_ops netem_class_ops = { | |
700 | .graft = netem_graft, | |
701 | .leaf = netem_leaf, | |
702 | .get = netem_get, | |
703 | .put = netem_put, | |
704 | .change = netem_change_class, | |
705 | .delete = netem_delete, | |
706 | .walk = netem_walk, | |
707 | .tcf_chain = netem_find_tcf, | |
708 | .dump = netem_dump_class, | |
709 | }; | |
710 | ||
711 | static struct Qdisc_ops netem_qdisc_ops = { | |
712 | .id = "netem", | |
713 | .cl_ops = &netem_class_ops, | |
714 | .priv_size = sizeof(struct netem_sched_data), | |
715 | .enqueue = netem_enqueue, | |
716 | .dequeue = netem_dequeue, | |
717 | .requeue = netem_requeue, | |
718 | .drop = netem_drop, | |
719 | .init = netem_init, | |
720 | .reset = netem_reset, | |
721 | .destroy = netem_destroy, | |
722 | .change = netem_change, | |
723 | .dump = netem_dump, | |
724 | .owner = THIS_MODULE, | |
725 | }; | |
726 | ||
727 | ||
728 | static int __init netem_module_init(void) | |
729 | { | |
eb229c4c | 730 | pr_info("netem: version " VERSION "\n"); |
1da177e4 LT |
731 | return register_qdisc(&netem_qdisc_ops); |
732 | } | |
733 | static void __exit netem_module_exit(void) | |
734 | { | |
735 | unregister_qdisc(&netem_qdisc_ops); | |
736 | } | |
737 | module_init(netem_module_init) | |
738 | module_exit(netem_module_exit) | |
739 | MODULE_LICENSE("GPL"); |