]>
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 | 59 | |
b407621c SH |
60 | psched_tdiff_t latency; |
61 | psched_tdiff_t jitter; | |
62 | ||
1da177e4 LT |
63 | u32 loss; |
64 | u32 limit; | |
65 | u32 counter; | |
66 | u32 gap; | |
1da177e4 | 67 | u32 duplicate; |
0dca51d3 | 68 | u32 reorder; |
c865e5d9 | 69 | u32 corrupt; |
1da177e4 LT |
70 | |
71 | struct crndstate { | |
b407621c SH |
72 | u32 last; |
73 | u32 rho; | |
c865e5d9 | 74 | } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor; |
1da177e4 LT |
75 | |
76 | struct disttable { | |
77 | u32 size; | |
78 | s16 table[0]; | |
79 | } *delay_dist; | |
80 | }; | |
81 | ||
82 | /* Time stamp put into socket buffer control block */ | |
83 | struct netem_skb_cb { | |
84 | psched_time_t time_to_send; | |
85 | }; | |
86 | ||
87 | /* init_crandom - initialize correlated random number generator | |
88 | * Use entropy source for initial seed. | |
89 | */ | |
90 | static void init_crandom(struct crndstate *state, unsigned long rho) | |
91 | { | |
92 | state->rho = rho; | |
93 | state->last = net_random(); | |
94 | } | |
95 | ||
96 | /* get_crandom - correlated random number generator | |
97 | * Next number depends on last value. | |
98 | * rho is scaled to avoid floating point. | |
99 | */ | |
b407621c | 100 | static u32 get_crandom(struct crndstate *state) |
1da177e4 LT |
101 | { |
102 | u64 value, rho; | |
103 | unsigned long answer; | |
104 | ||
bb2f8cc0 | 105 | if (state->rho == 0) /* no correlation */ |
1da177e4 LT |
106 | return net_random(); |
107 | ||
108 | value = net_random(); | |
109 | rho = (u64)state->rho + 1; | |
110 | answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32; | |
111 | state->last = answer; | |
112 | return answer; | |
113 | } | |
114 | ||
115 | /* tabledist - return a pseudo-randomly distributed value with mean mu and | |
116 | * std deviation sigma. Uses table lookup to approximate the desired | |
117 | * distribution, and a uniformly-distributed pseudo-random source. | |
118 | */ | |
b407621c SH |
119 | static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma, |
120 | struct crndstate *state, | |
121 | const struct disttable *dist) | |
1da177e4 | 122 | { |
b407621c SH |
123 | psched_tdiff_t x; |
124 | long t; | |
125 | u32 rnd; | |
1da177e4 LT |
126 | |
127 | if (sigma == 0) | |
128 | return mu; | |
129 | ||
130 | rnd = get_crandom(state); | |
131 | ||
132 | /* default uniform distribution */ | |
10297b99 | 133 | if (dist == NULL) |
1da177e4 LT |
134 | return (rnd % (2*sigma)) - sigma + mu; |
135 | ||
136 | t = dist->table[rnd % dist->size]; | |
137 | x = (sigma % NETEM_DIST_SCALE) * t; | |
138 | if (x >= 0) | |
139 | x += NETEM_DIST_SCALE/2; | |
140 | else | |
141 | x -= NETEM_DIST_SCALE/2; | |
142 | ||
143 | return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu; | |
144 | } | |
145 | ||
0afb51e7 SH |
146 | /* |
147 | * Insert one skb into qdisc. | |
148 | * Note: parent depends on return value to account for queue length. | |
149 | * NET_XMIT_DROP: queue length didn't change. | |
150 | * NET_XMIT_SUCCESS: one skb was queued. | |
151 | */ | |
1da177e4 LT |
152 | static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
153 | { | |
154 | struct netem_sched_data *q = qdisc_priv(sch); | |
89e1df74 GC |
155 | /* We don't fill cb now as skb_unshare() may invalidate it */ |
156 | struct netem_skb_cb *cb; | |
0afb51e7 | 157 | struct sk_buff *skb2; |
1da177e4 | 158 | int ret; |
0afb51e7 | 159 | int count = 1; |
1da177e4 | 160 | |
771018e7 | 161 | pr_debug("netem_enqueue skb=%p\n", skb); |
1da177e4 | 162 | |
0afb51e7 SH |
163 | /* Random duplication */ |
164 | if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor)) | |
165 | ++count; | |
166 | ||
1da177e4 | 167 | /* Random packet drop 0 => none, ~0 => all */ |
0afb51e7 SH |
168 | if (q->loss && q->loss >= get_crandom(&q->loss_cor)) |
169 | --count; | |
170 | ||
171 | if (count == 0) { | |
1da177e4 LT |
172 | sch->qstats.drops++; |
173 | kfree_skb(skb); | |
89bbb0a3 | 174 | return NET_XMIT_BYPASS; |
1da177e4 LT |
175 | } |
176 | ||
4e8a5201 DM |
177 | skb_orphan(skb); |
178 | ||
0afb51e7 SH |
179 | /* |
180 | * If we need to duplicate packet, then re-insert at top of the | |
181 | * qdisc tree, since parent queuer expects that only one | |
182 | * skb will be queued. | |
183 | */ | |
184 | if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) { | |
185 | struct Qdisc *rootq = sch->dev->qdisc; | |
186 | u32 dupsave = q->duplicate; /* prevent duplicating a dup... */ | |
187 | q->duplicate = 0; | |
188 | ||
189 | rootq->enqueue(skb2, rootq); | |
190 | q->duplicate = dupsave; | |
1da177e4 LT |
191 | } |
192 | ||
c865e5d9 SH |
193 | /* |
194 | * Randomized packet corruption. | |
195 | * Make copy if needed since we are modifying | |
196 | * If packet is going to be hardware checksummed, then | |
197 | * do it now in software before we mangle it. | |
198 | */ | |
199 | if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) { | |
200 | if (!(skb = skb_unshare(skb, GFP_ATOMIC)) | |
84fa7933 PM |
201 | || (skb->ip_summed == CHECKSUM_PARTIAL |
202 | && skb_checksum_help(skb))) { | |
c865e5d9 SH |
203 | sch->qstats.drops++; |
204 | return NET_XMIT_DROP; | |
205 | } | |
206 | ||
207 | skb->data[net_random() % skb_headlen(skb)] ^= 1<<(net_random() % 8); | |
208 | } | |
209 | ||
89e1df74 | 210 | cb = (struct netem_skb_cb *)skb->cb; |
0dca51d3 SH |
211 | if (q->gap == 0 /* not doing reordering */ |
212 | || q->counter < q->gap /* inside last reordering gap */ | |
213 | || q->reorder < get_crandom(&q->reorder_cor)) { | |
0f9f32ac | 214 | psched_time_t now; |
07aaa115 SH |
215 | psched_tdiff_t delay; |
216 | ||
217 | delay = tabledist(q->latency, q->jitter, | |
218 | &q->delay_cor, q->delay_dist); | |
219 | ||
0f9f32ac | 220 | PSCHED_GET_TIME(now); |
7c59e25f | 221 | cb->time_to_send = now + delay; |
1da177e4 LT |
222 | ++q->counter; |
223 | ret = q->qdisc->enqueue(skb, q->qdisc); | |
224 | } else { | |
10297b99 | 225 | /* |
0dca51d3 SH |
226 | * Do re-ordering by putting one out of N packets at the front |
227 | * of the queue. | |
228 | */ | |
0f9f32ac | 229 | PSCHED_GET_TIME(cb->time_to_send); |
0dca51d3 | 230 | q->counter = 0; |
0f9f32ac | 231 | ret = q->qdisc->ops->requeue(skb, q->qdisc); |
1da177e4 LT |
232 | } |
233 | ||
234 | if (likely(ret == NET_XMIT_SUCCESS)) { | |
235 | sch->q.qlen++; | |
236 | sch->bstats.bytes += skb->len; | |
237 | sch->bstats.packets++; | |
238 | } else | |
239 | sch->qstats.drops++; | |
240 | ||
d5d75cd6 | 241 | pr_debug("netem: enqueue ret %d\n", ret); |
1da177e4 LT |
242 | return ret; |
243 | } | |
244 | ||
245 | /* Requeue packets but don't change time stamp */ | |
246 | static int netem_requeue(struct sk_buff *skb, struct Qdisc *sch) | |
247 | { | |
248 | struct netem_sched_data *q = qdisc_priv(sch); | |
249 | int ret; | |
250 | ||
251 | if ((ret = q->qdisc->ops->requeue(skb, q->qdisc)) == 0) { | |
252 | sch->q.qlen++; | |
253 | sch->qstats.requeues++; | |
254 | } | |
255 | ||
256 | return ret; | |
257 | } | |
258 | ||
259 | static unsigned int netem_drop(struct Qdisc* sch) | |
260 | { | |
261 | struct netem_sched_data *q = qdisc_priv(sch); | |
6d037a26 | 262 | unsigned int len = 0; |
1da177e4 | 263 | |
6d037a26 | 264 | if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) { |
1da177e4 LT |
265 | sch->q.qlen--; |
266 | sch->qstats.drops++; | |
267 | } | |
268 | return len; | |
269 | } | |
270 | ||
1da177e4 LT |
271 | static struct sk_buff *netem_dequeue(struct Qdisc *sch) |
272 | { | |
273 | struct netem_sched_data *q = qdisc_priv(sch); | |
274 | struct sk_buff *skb; | |
275 | ||
11274e5a SH |
276 | smp_mb(); |
277 | if (sch->flags & TCQ_F_THROTTLED) | |
278 | return NULL; | |
279 | ||
1da177e4 | 280 | skb = q->qdisc->dequeue(q->qdisc); |
771018e7 | 281 | if (skb) { |
0f9f32ac SH |
282 | const struct netem_skb_cb *cb |
283 | = (const struct netem_skb_cb *)skb->cb; | |
284 | psched_time_t now; | |
0f9f32ac SH |
285 | |
286 | /* if more time remaining? */ | |
287 | PSCHED_GET_TIME(now); | |
07aaa115 | 288 | |
76d643cd | 289 | if (!PSCHED_TLESS(now, cb->time_to_send)) { |
0f9f32ac SH |
290 | pr_debug("netem_dequeue: return skb=%p\n", skb); |
291 | sch->q.qlen--; | |
0f9f32ac | 292 | return skb; |
07aaa115 | 293 | } |
11274e5a SH |
294 | |
295 | if (unlikely(q->qdisc->ops->requeue(skb, q->qdisc) != NET_XMIT_SUCCESS)) { | |
296 | qdisc_tree_decrease_qlen(q->qdisc, 1); | |
297 | sch->qstats.drops++; | |
298 | printk(KERN_ERR "netem: %s could not requeue\n", | |
299 | q->qdisc->ops->id); | |
300 | } | |
301 | ||
302 | qdisc_watchdog_schedule(&q->watchdog, cb->time_to_send); | |
0f9f32ac SH |
303 | } |
304 | ||
305 | return NULL; | |
1da177e4 LT |
306 | } |
307 | ||
1da177e4 LT |
308 | static void netem_reset(struct Qdisc *sch) |
309 | { | |
310 | struct netem_sched_data *q = qdisc_priv(sch); | |
311 | ||
312 | qdisc_reset(q->qdisc); | |
1da177e4 | 313 | sch->q.qlen = 0; |
59cb5c67 | 314 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
315 | } |
316 | ||
300ce174 | 317 | /* Pass size change message down to embedded FIFO */ |
1da177e4 LT |
318 | static int set_fifo_limit(struct Qdisc *q, int limit) |
319 | { | |
10297b99 | 320 | struct rtattr *rta; |
1da177e4 LT |
321 | int ret = -ENOMEM; |
322 | ||
300ce174 SH |
323 | /* Hack to avoid sending change message to non-FIFO */ |
324 | if (strncmp(q->ops->id + 1, "fifo", 4) != 0) | |
325 | return 0; | |
326 | ||
1da177e4 LT |
327 | rta = kmalloc(RTA_LENGTH(sizeof(struct tc_fifo_qopt)), GFP_KERNEL); |
328 | if (rta) { | |
329 | rta->rta_type = RTM_NEWQDISC; | |
10297b99 | 330 | rta->rta_len = RTA_LENGTH(sizeof(struct tc_fifo_qopt)); |
1da177e4 | 331 | ((struct tc_fifo_qopt *)RTA_DATA(rta))->limit = limit; |
10297b99 | 332 | |
1da177e4 LT |
333 | ret = q->ops->change(q, rta); |
334 | kfree(rta); | |
335 | } | |
336 | return ret; | |
337 | } | |
338 | ||
339 | /* | |
340 | * Distribution data is a variable size payload containing | |
341 | * signed 16 bit values. | |
342 | */ | |
343 | static int get_dist_table(struct Qdisc *sch, const struct rtattr *attr) | |
344 | { | |
345 | struct netem_sched_data *q = qdisc_priv(sch); | |
346 | unsigned long n = RTA_PAYLOAD(attr)/sizeof(__s16); | |
347 | const __s16 *data = RTA_DATA(attr); | |
348 | struct disttable *d; | |
349 | int i; | |
350 | ||
351 | if (n > 65536) | |
352 | return -EINVAL; | |
353 | ||
354 | d = kmalloc(sizeof(*d) + n*sizeof(d->table[0]), GFP_KERNEL); | |
355 | if (!d) | |
356 | return -ENOMEM; | |
357 | ||
358 | d->size = n; | |
359 | for (i = 0; i < n; i++) | |
360 | d->table[i] = data[i]; | |
10297b99 | 361 | |
1da177e4 LT |
362 | spin_lock_bh(&sch->dev->queue_lock); |
363 | d = xchg(&q->delay_dist, d); | |
364 | spin_unlock_bh(&sch->dev->queue_lock); | |
365 | ||
366 | kfree(d); | |
367 | return 0; | |
368 | } | |
369 | ||
370 | static int get_correlation(struct Qdisc *sch, const struct rtattr *attr) | |
371 | { | |
372 | struct netem_sched_data *q = qdisc_priv(sch); | |
373 | const struct tc_netem_corr *c = RTA_DATA(attr); | |
374 | ||
375 | if (RTA_PAYLOAD(attr) != sizeof(*c)) | |
376 | return -EINVAL; | |
377 | ||
378 | init_crandom(&q->delay_cor, c->delay_corr); | |
379 | init_crandom(&q->loss_cor, c->loss_corr); | |
380 | init_crandom(&q->dup_cor, c->dup_corr); | |
381 | return 0; | |
382 | } | |
383 | ||
0dca51d3 SH |
384 | static int get_reorder(struct Qdisc *sch, const struct rtattr *attr) |
385 | { | |
386 | struct netem_sched_data *q = qdisc_priv(sch); | |
387 | const struct tc_netem_reorder *r = RTA_DATA(attr); | |
388 | ||
389 | if (RTA_PAYLOAD(attr) != sizeof(*r)) | |
390 | return -EINVAL; | |
391 | ||
392 | q->reorder = r->probability; | |
393 | init_crandom(&q->reorder_cor, r->correlation); | |
394 | return 0; | |
395 | } | |
396 | ||
c865e5d9 SH |
397 | static int get_corrupt(struct Qdisc *sch, const struct rtattr *attr) |
398 | { | |
399 | struct netem_sched_data *q = qdisc_priv(sch); | |
400 | const struct tc_netem_corrupt *r = RTA_DATA(attr); | |
401 | ||
402 | if (RTA_PAYLOAD(attr) != sizeof(*r)) | |
403 | return -EINVAL; | |
404 | ||
405 | q->corrupt = r->probability; | |
406 | init_crandom(&q->corrupt_cor, r->correlation); | |
407 | return 0; | |
408 | } | |
409 | ||
410 | /* Parse netlink message to set options */ | |
1da177e4 LT |
411 | static int netem_change(struct Qdisc *sch, struct rtattr *opt) |
412 | { | |
413 | struct netem_sched_data *q = qdisc_priv(sch); | |
414 | struct tc_netem_qopt *qopt; | |
415 | int ret; | |
10297b99 | 416 | |
1da177e4 LT |
417 | if (opt == NULL || RTA_PAYLOAD(opt) < sizeof(*qopt)) |
418 | return -EINVAL; | |
419 | ||
420 | qopt = RTA_DATA(opt); | |
421 | ret = set_fifo_limit(q->qdisc, qopt->limit); | |
422 | if (ret) { | |
423 | pr_debug("netem: can't set fifo limit\n"); | |
424 | return ret; | |
425 | } | |
10297b99 | 426 | |
1da177e4 LT |
427 | q->latency = qopt->latency; |
428 | q->jitter = qopt->jitter; | |
429 | q->limit = qopt->limit; | |
430 | q->gap = qopt->gap; | |
0dca51d3 | 431 | q->counter = 0; |
1da177e4 LT |
432 | q->loss = qopt->loss; |
433 | q->duplicate = qopt->duplicate; | |
434 | ||
bb2f8cc0 SH |
435 | /* for compatibility with earlier versions. |
436 | * if gap is set, need to assume 100% probability | |
0dca51d3 | 437 | */ |
a362e0a7 SH |
438 | if (q->gap) |
439 | q->reorder = ~0; | |
0dca51d3 | 440 | |
1da177e4 LT |
441 | /* Handle nested options after initial queue options. |
442 | * Should have put all options in nested format but too late now. | |
10297b99 | 443 | */ |
1da177e4 LT |
444 | if (RTA_PAYLOAD(opt) > sizeof(*qopt)) { |
445 | struct rtattr *tb[TCA_NETEM_MAX]; | |
10297b99 | 446 | if (rtattr_parse(tb, TCA_NETEM_MAX, |
1da177e4 LT |
447 | RTA_DATA(opt) + sizeof(*qopt), |
448 | RTA_PAYLOAD(opt) - sizeof(*qopt))) | |
449 | return -EINVAL; | |
450 | ||
451 | if (tb[TCA_NETEM_CORR-1]) { | |
452 | ret = get_correlation(sch, tb[TCA_NETEM_CORR-1]); | |
453 | if (ret) | |
454 | return ret; | |
455 | } | |
456 | ||
457 | if (tb[TCA_NETEM_DELAY_DIST-1]) { | |
458 | ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST-1]); | |
459 | if (ret) | |
460 | return ret; | |
461 | } | |
c865e5d9 | 462 | |
0dca51d3 SH |
463 | if (tb[TCA_NETEM_REORDER-1]) { |
464 | ret = get_reorder(sch, tb[TCA_NETEM_REORDER-1]); | |
465 | if (ret) | |
466 | return ret; | |
467 | } | |
1da177e4 | 468 | |
c865e5d9 SH |
469 | if (tb[TCA_NETEM_CORRUPT-1]) { |
470 | ret = get_corrupt(sch, tb[TCA_NETEM_CORRUPT-1]); | |
471 | if (ret) | |
472 | return ret; | |
473 | } | |
474 | } | |
1da177e4 LT |
475 | |
476 | return 0; | |
477 | } | |
478 | ||
300ce174 SH |
479 | /* |
480 | * Special case version of FIFO queue for use by netem. | |
481 | * It queues in order based on timestamps in skb's | |
482 | */ | |
483 | struct fifo_sched_data { | |
484 | u32 limit; | |
075aa573 | 485 | psched_time_t oldest; |
300ce174 SH |
486 | }; |
487 | ||
488 | static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch) | |
489 | { | |
490 | struct fifo_sched_data *q = qdisc_priv(sch); | |
491 | struct sk_buff_head *list = &sch->q; | |
075aa573 | 492 | psched_time_t tnext = ((struct netem_skb_cb *)nskb->cb)->time_to_send; |
300ce174 SH |
493 | struct sk_buff *skb; |
494 | ||
495 | if (likely(skb_queue_len(list) < q->limit)) { | |
075aa573 SH |
496 | /* Optimize for add at tail */ |
497 | if (likely(skb_queue_empty(list) || !PSCHED_TLESS(tnext, q->oldest))) { | |
498 | q->oldest = tnext; | |
499 | return qdisc_enqueue_tail(nskb, sch); | |
500 | } | |
501 | ||
300ce174 SH |
502 | skb_queue_reverse_walk(list, skb) { |
503 | const struct netem_skb_cb *cb | |
504 | = (const struct netem_skb_cb *)skb->cb; | |
505 | ||
075aa573 | 506 | if (!PSCHED_TLESS(tnext, cb->time_to_send)) |
300ce174 SH |
507 | break; |
508 | } | |
509 | ||
510 | __skb_queue_after(list, skb, nskb); | |
511 | ||
512 | sch->qstats.backlog += nskb->len; | |
513 | sch->bstats.bytes += nskb->len; | |
514 | sch->bstats.packets++; | |
515 | ||
516 | return NET_XMIT_SUCCESS; | |
517 | } | |
518 | ||
075aa573 | 519 | return qdisc_reshape_fail(nskb, sch); |
300ce174 SH |
520 | } |
521 | ||
522 | static int tfifo_init(struct Qdisc *sch, struct rtattr *opt) | |
523 | { | |
524 | struct fifo_sched_data *q = qdisc_priv(sch); | |
525 | ||
526 | if (opt) { | |
527 | struct tc_fifo_qopt *ctl = RTA_DATA(opt); | |
528 | if (RTA_PAYLOAD(opt) < sizeof(*ctl)) | |
529 | return -EINVAL; | |
530 | ||
531 | q->limit = ctl->limit; | |
532 | } else | |
533 | q->limit = max_t(u32, sch->dev->tx_queue_len, 1); | |
534 | ||
075aa573 | 535 | PSCHED_SET_PASTPERFECT(q->oldest); |
300ce174 SH |
536 | return 0; |
537 | } | |
538 | ||
539 | static int tfifo_dump(struct Qdisc *sch, struct sk_buff *skb) | |
540 | { | |
541 | struct fifo_sched_data *q = qdisc_priv(sch); | |
542 | struct tc_fifo_qopt opt = { .limit = q->limit }; | |
543 | ||
544 | RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); | |
545 | return skb->len; | |
546 | ||
547 | rtattr_failure: | |
548 | return -1; | |
549 | } | |
550 | ||
551 | static struct Qdisc_ops tfifo_qdisc_ops = { | |
552 | .id = "tfifo", | |
553 | .priv_size = sizeof(struct fifo_sched_data), | |
554 | .enqueue = tfifo_enqueue, | |
555 | .dequeue = qdisc_dequeue_head, | |
556 | .requeue = qdisc_requeue, | |
557 | .drop = qdisc_queue_drop, | |
558 | .init = tfifo_init, | |
559 | .reset = qdisc_reset_queue, | |
560 | .change = tfifo_init, | |
561 | .dump = tfifo_dump, | |
562 | }; | |
563 | ||
1da177e4 LT |
564 | static int netem_init(struct Qdisc *sch, struct rtattr *opt) |
565 | { | |
566 | struct netem_sched_data *q = qdisc_priv(sch); | |
567 | int ret; | |
568 | ||
569 | if (!opt) | |
570 | return -EINVAL; | |
571 | ||
59cb5c67 | 572 | qdisc_watchdog_init(&q->watchdog, sch); |
1da177e4 | 573 | |
9f9afec4 PM |
574 | q->qdisc = qdisc_create_dflt(sch->dev, &tfifo_qdisc_ops, |
575 | TC_H_MAKE(sch->handle, 1)); | |
1da177e4 LT |
576 | if (!q->qdisc) { |
577 | pr_debug("netem: qdisc create failed\n"); | |
578 | return -ENOMEM; | |
579 | } | |
580 | ||
581 | ret = netem_change(sch, opt); | |
582 | if (ret) { | |
583 | pr_debug("netem: change failed\n"); | |
584 | qdisc_destroy(q->qdisc); | |
585 | } | |
586 | return ret; | |
587 | } | |
588 | ||
589 | static void netem_destroy(struct Qdisc *sch) | |
590 | { | |
591 | struct netem_sched_data *q = qdisc_priv(sch); | |
592 | ||
59cb5c67 | 593 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
594 | qdisc_destroy(q->qdisc); |
595 | kfree(q->delay_dist); | |
596 | } | |
597 | ||
598 | static int netem_dump(struct Qdisc *sch, struct sk_buff *skb) | |
599 | { | |
600 | const struct netem_sched_data *q = qdisc_priv(sch); | |
27a884dc | 601 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
602 | struct rtattr *rta = (struct rtattr *) b; |
603 | struct tc_netem_qopt qopt; | |
604 | struct tc_netem_corr cor; | |
0dca51d3 | 605 | struct tc_netem_reorder reorder; |
c865e5d9 | 606 | struct tc_netem_corrupt corrupt; |
1da177e4 LT |
607 | |
608 | qopt.latency = q->latency; | |
609 | qopt.jitter = q->jitter; | |
610 | qopt.limit = q->limit; | |
611 | qopt.loss = q->loss; | |
612 | qopt.gap = q->gap; | |
613 | qopt.duplicate = q->duplicate; | |
614 | RTA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt); | |
615 | ||
616 | cor.delay_corr = q->delay_cor.rho; | |
617 | cor.loss_corr = q->loss_cor.rho; | |
618 | cor.dup_corr = q->dup_cor.rho; | |
619 | RTA_PUT(skb, TCA_NETEM_CORR, sizeof(cor), &cor); | |
0dca51d3 SH |
620 | |
621 | reorder.probability = q->reorder; | |
622 | reorder.correlation = q->reorder_cor.rho; | |
623 | RTA_PUT(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder); | |
624 | ||
c865e5d9 SH |
625 | corrupt.probability = q->corrupt; |
626 | corrupt.correlation = q->corrupt_cor.rho; | |
627 | RTA_PUT(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt); | |
628 | ||
27a884dc | 629 | rta->rta_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
630 | |
631 | return skb->len; | |
632 | ||
633 | rtattr_failure: | |
dc5fc579 | 634 | nlmsg_trim(skb, b); |
1da177e4 LT |
635 | return -1; |
636 | } | |
637 | ||
638 | static int netem_dump_class(struct Qdisc *sch, unsigned long cl, | |
639 | struct sk_buff *skb, struct tcmsg *tcm) | |
640 | { | |
641 | struct netem_sched_data *q = qdisc_priv(sch); | |
642 | ||
643 | if (cl != 1) /* only one class */ | |
644 | return -ENOENT; | |
645 | ||
646 | tcm->tcm_handle |= TC_H_MIN(1); | |
647 | tcm->tcm_info = q->qdisc->handle; | |
648 | ||
649 | return 0; | |
650 | } | |
651 | ||
652 | static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
653 | struct Qdisc **old) | |
654 | { | |
655 | struct netem_sched_data *q = qdisc_priv(sch); | |
656 | ||
657 | if (new == NULL) | |
658 | new = &noop_qdisc; | |
659 | ||
660 | sch_tree_lock(sch); | |
661 | *old = xchg(&q->qdisc, new); | |
5e50da01 | 662 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); |
1da177e4 | 663 | qdisc_reset(*old); |
1da177e4 LT |
664 | sch_tree_unlock(sch); |
665 | ||
666 | return 0; | |
667 | } | |
668 | ||
669 | static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg) | |
670 | { | |
671 | struct netem_sched_data *q = qdisc_priv(sch); | |
672 | return q->qdisc; | |
673 | } | |
674 | ||
675 | static unsigned long netem_get(struct Qdisc *sch, u32 classid) | |
676 | { | |
677 | return 1; | |
678 | } | |
679 | ||
680 | static void netem_put(struct Qdisc *sch, unsigned long arg) | |
681 | { | |
682 | } | |
683 | ||
10297b99 | 684 | static int netem_change_class(struct Qdisc *sch, u32 classid, u32 parentid, |
1da177e4 LT |
685 | struct rtattr **tca, unsigned long *arg) |
686 | { | |
687 | return -ENOSYS; | |
688 | } | |
689 | ||
690 | static int netem_delete(struct Qdisc *sch, unsigned long arg) | |
691 | { | |
692 | return -ENOSYS; | |
693 | } | |
694 | ||
695 | static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker) | |
696 | { | |
697 | if (!walker->stop) { | |
698 | if (walker->count >= walker->skip) | |
699 | if (walker->fn(sch, 1, walker) < 0) { | |
700 | walker->stop = 1; | |
701 | return; | |
702 | } | |
703 | walker->count++; | |
704 | } | |
705 | } | |
706 | ||
707 | static struct tcf_proto **netem_find_tcf(struct Qdisc *sch, unsigned long cl) | |
708 | { | |
709 | return NULL; | |
710 | } | |
711 | ||
712 | static struct Qdisc_class_ops netem_class_ops = { | |
713 | .graft = netem_graft, | |
714 | .leaf = netem_leaf, | |
715 | .get = netem_get, | |
716 | .put = netem_put, | |
717 | .change = netem_change_class, | |
718 | .delete = netem_delete, | |
719 | .walk = netem_walk, | |
720 | .tcf_chain = netem_find_tcf, | |
721 | .dump = netem_dump_class, | |
722 | }; | |
723 | ||
724 | static struct Qdisc_ops netem_qdisc_ops = { | |
725 | .id = "netem", | |
726 | .cl_ops = &netem_class_ops, | |
727 | .priv_size = sizeof(struct netem_sched_data), | |
728 | .enqueue = netem_enqueue, | |
729 | .dequeue = netem_dequeue, | |
730 | .requeue = netem_requeue, | |
731 | .drop = netem_drop, | |
732 | .init = netem_init, | |
733 | .reset = netem_reset, | |
734 | .destroy = netem_destroy, | |
735 | .change = netem_change, | |
736 | .dump = netem_dump, | |
737 | .owner = THIS_MODULE, | |
738 | }; | |
739 | ||
740 | ||
741 | static int __init netem_module_init(void) | |
742 | { | |
eb229c4c | 743 | pr_info("netem: version " VERSION "\n"); |
1da177e4 LT |
744 | return register_qdisc(&netem_qdisc_ops); |
745 | } | |
746 | static void __exit netem_module_exit(void) | |
747 | { | |
748 | unregister_qdisc(&netem_qdisc_ops); | |
749 | } | |
750 | module_init(netem_module_init) | |
751 | module_exit(netem_module_exit) | |
752 | MODULE_LICENSE("GPL"); |