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 | 16 | #include <linux/module.h> |
1da177e4 LT |
17 | #include <linux/types.h> |
18 | #include <linux/kernel.h> | |
19 | #include <linux/errno.h> | |
1da177e4 LT |
20 | #include <linux/skbuff.h> |
21 | #include <linux/rtnetlink.h> | |
22 | ||
dc5fc579 | 23 | #include <net/netlink.h> |
1da177e4 LT |
24 | #include <net/pkt_sched.h> |
25 | ||
c865e5d9 | 26 | #define VERSION "1.2" |
eb229c4c | 27 | |
1da177e4 LT |
28 | /* Network Emulation Queuing algorithm. |
29 | ==================================== | |
30 | ||
31 | Sources: [1] Mark Carson, Darrin Santay, "NIST Net - A Linux-based | |
32 | Network Emulation Tool | |
33 | [2] Luigi Rizzo, DummyNet for FreeBSD | |
34 | ||
35 | ---------------------------------------------------------------- | |
36 | ||
37 | This started out as a simple way to delay outgoing packets to | |
38 | test TCP but has grown to include most of the functionality | |
39 | of a full blown network emulator like NISTnet. It can delay | |
40 | packets and add random jitter (and correlation). The random | |
41 | distribution can be loaded from a table as well to provide | |
42 | normal, Pareto, or experimental curves. Packet loss, | |
43 | duplication, and reordering can also be emulated. | |
44 | ||
45 | This qdisc does not do classification that can be handled in | |
46 | layering other disciplines. It does not need to do bandwidth | |
47 | control either since that can be handled by using token | |
48 | bucket or other rate control. | |
49 | ||
50 | The simulator is limited by the Linux timer resolution | |
51 | and will create packet bursts on the HZ boundary (1ms). | |
52 | */ | |
53 | ||
54 | struct netem_sched_data { | |
55 | struct Qdisc *qdisc; | |
59cb5c67 | 56 | struct qdisc_watchdog watchdog; |
1da177e4 | 57 | |
b407621c SH |
58 | psched_tdiff_t latency; |
59 | psched_tdiff_t jitter; | |
60 | ||
1da177e4 LT |
61 | u32 loss; |
62 | u32 limit; | |
63 | u32 counter; | |
64 | u32 gap; | |
1da177e4 | 65 | u32 duplicate; |
0dca51d3 | 66 | u32 reorder; |
c865e5d9 | 67 | u32 corrupt; |
1da177e4 LT |
68 | |
69 | struct crndstate { | |
b407621c SH |
70 | u32 last; |
71 | u32 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 | */ | |
b407621c | 98 | static u32 get_crandom(struct crndstate *state) |
1da177e4 LT |
99 | { |
100 | u64 value, rho; | |
101 | unsigned long answer; | |
102 | ||
bb2f8cc0 | 103 | if (state->rho == 0) /* no correlation */ |
1da177e4 LT |
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 | */ | |
b407621c SH |
117 | static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma, |
118 | struct crndstate *state, | |
119 | const struct disttable *dist) | |
1da177e4 | 120 | { |
b407621c SH |
121 | psched_tdiff_t x; |
122 | long t; | |
123 | u32 rnd; | |
1da177e4 LT |
124 | |
125 | if (sigma == 0) | |
126 | return mu; | |
127 | ||
128 | rnd = get_crandom(state); | |
129 | ||
130 | /* default uniform distribution */ | |
10297b99 | 131 | if (dist == NULL) |
1da177e4 LT |
132 | return (rnd % (2*sigma)) - sigma + mu; |
133 | ||
134 | t = dist->table[rnd % dist->size]; | |
135 | x = (sigma % NETEM_DIST_SCALE) * t; | |
136 | if (x >= 0) | |
137 | x += NETEM_DIST_SCALE/2; | |
138 | else | |
139 | x -= NETEM_DIST_SCALE/2; | |
140 | ||
141 | return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu; | |
142 | } | |
143 | ||
0afb51e7 SH |
144 | /* |
145 | * Insert one skb into qdisc. | |
146 | * Note: parent depends on return value to account for queue length. | |
147 | * NET_XMIT_DROP: queue length didn't change. | |
148 | * NET_XMIT_SUCCESS: one skb was queued. | |
149 | */ | |
1da177e4 LT |
150 | static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
151 | { | |
152 | struct netem_sched_data *q = qdisc_priv(sch); | |
89e1df74 GC |
153 | /* We don't fill cb now as skb_unshare() may invalidate it */ |
154 | struct netem_skb_cb *cb; | |
0afb51e7 | 155 | struct sk_buff *skb2; |
1da177e4 | 156 | int ret; |
0afb51e7 | 157 | int count = 1; |
1da177e4 | 158 | |
771018e7 | 159 | pr_debug("netem_enqueue skb=%p\n", skb); |
1da177e4 | 160 | |
0afb51e7 SH |
161 | /* Random duplication */ |
162 | if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor)) | |
163 | ++count; | |
164 | ||
1da177e4 | 165 | /* Random packet drop 0 => none, ~0 => all */ |
0afb51e7 SH |
166 | if (q->loss && q->loss >= get_crandom(&q->loss_cor)) |
167 | --count; | |
168 | ||
169 | if (count == 0) { | |
1da177e4 LT |
170 | sch->qstats.drops++; |
171 | kfree_skb(skb); | |
89bbb0a3 | 172 | return NET_XMIT_BYPASS; |
1da177e4 LT |
173 | } |
174 | ||
4e8a5201 DM |
175 | skb_orphan(skb); |
176 | ||
0afb51e7 SH |
177 | /* |
178 | * If we need to duplicate packet, then re-insert at top of the | |
179 | * qdisc tree, since parent queuer expects that only one | |
180 | * skb will be queued. | |
181 | */ | |
182 | if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) { | |
183 | struct Qdisc *rootq = sch->dev->qdisc; | |
184 | u32 dupsave = q->duplicate; /* prevent duplicating a dup... */ | |
185 | q->duplicate = 0; | |
186 | ||
187 | rootq->enqueue(skb2, rootq); | |
188 | q->duplicate = dupsave; | |
1da177e4 LT |
189 | } |
190 | ||
c865e5d9 SH |
191 | /* |
192 | * Randomized packet corruption. | |
193 | * Make copy if needed since we are modifying | |
194 | * If packet is going to be hardware checksummed, then | |
195 | * do it now in software before we mangle it. | |
196 | */ | |
197 | if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) { | |
198 | if (!(skb = skb_unshare(skb, GFP_ATOMIC)) | |
84fa7933 PM |
199 | || (skb->ip_summed == CHECKSUM_PARTIAL |
200 | && skb_checksum_help(skb))) { | |
c865e5d9 SH |
201 | sch->qstats.drops++; |
202 | return NET_XMIT_DROP; | |
203 | } | |
204 | ||
205 | skb->data[net_random() % skb_headlen(skb)] ^= 1<<(net_random() % 8); | |
206 | } | |
207 | ||
89e1df74 | 208 | cb = (struct netem_skb_cb *)skb->cb; |
0dca51d3 SH |
209 | if (q->gap == 0 /* not doing reordering */ |
210 | || q->counter < q->gap /* inside last reordering gap */ | |
211 | || q->reorder < get_crandom(&q->reorder_cor)) { | |
0f9f32ac | 212 | psched_time_t now; |
07aaa115 SH |
213 | psched_tdiff_t delay; |
214 | ||
215 | delay = tabledist(q->latency, q->jitter, | |
216 | &q->delay_cor, q->delay_dist); | |
217 | ||
3bebcda2 | 218 | now = psched_get_time(); |
7c59e25f | 219 | cb->time_to_send = now + delay; |
1da177e4 LT |
220 | ++q->counter; |
221 | ret = q->qdisc->enqueue(skb, q->qdisc); | |
222 | } else { | |
10297b99 | 223 | /* |
0dca51d3 SH |
224 | * Do re-ordering by putting one out of N packets at the front |
225 | * of the queue. | |
226 | */ | |
3bebcda2 | 227 | cb->time_to_send = psched_get_time(); |
0dca51d3 | 228 | q->counter = 0; |
0f9f32ac | 229 | ret = q->qdisc->ops->requeue(skb, q->qdisc); |
1da177e4 LT |
230 | } |
231 | ||
232 | if (likely(ret == NET_XMIT_SUCCESS)) { | |
233 | sch->q.qlen++; | |
234 | sch->bstats.bytes += skb->len; | |
235 | sch->bstats.packets++; | |
236 | } else | |
237 | sch->qstats.drops++; | |
238 | ||
d5d75cd6 | 239 | pr_debug("netem: enqueue ret %d\n", ret); |
1da177e4 LT |
240 | return ret; |
241 | } | |
242 | ||
243 | /* Requeue packets but don't change time stamp */ | |
244 | static int netem_requeue(struct sk_buff *skb, struct Qdisc *sch) | |
245 | { | |
246 | struct netem_sched_data *q = qdisc_priv(sch); | |
247 | int ret; | |
248 | ||
249 | if ((ret = q->qdisc->ops->requeue(skb, q->qdisc)) == 0) { | |
250 | sch->q.qlen++; | |
251 | sch->qstats.requeues++; | |
252 | } | |
253 | ||
254 | return ret; | |
255 | } | |
256 | ||
257 | static unsigned int netem_drop(struct Qdisc* sch) | |
258 | { | |
259 | struct netem_sched_data *q = qdisc_priv(sch); | |
6d037a26 | 260 | unsigned int len = 0; |
1da177e4 | 261 | |
6d037a26 | 262 | if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) { |
1da177e4 LT |
263 | sch->q.qlen--; |
264 | sch->qstats.drops++; | |
265 | } | |
266 | return len; | |
267 | } | |
268 | ||
1da177e4 LT |
269 | static struct sk_buff *netem_dequeue(struct Qdisc *sch) |
270 | { | |
271 | struct netem_sched_data *q = qdisc_priv(sch); | |
272 | struct sk_buff *skb; | |
273 | ||
11274e5a SH |
274 | smp_mb(); |
275 | if (sch->flags & TCQ_F_THROTTLED) | |
276 | return NULL; | |
277 | ||
1da177e4 | 278 | skb = q->qdisc->dequeue(q->qdisc); |
771018e7 | 279 | if (skb) { |
0f9f32ac SH |
280 | const struct netem_skb_cb *cb |
281 | = (const struct netem_skb_cb *)skb->cb; | |
3bebcda2 | 282 | psched_time_t now = psched_get_time(); |
0f9f32ac SH |
283 | |
284 | /* if more time remaining? */ | |
104e0878 | 285 | if (cb->time_to_send <= now) { |
0f9f32ac SH |
286 | pr_debug("netem_dequeue: return skb=%p\n", skb); |
287 | sch->q.qlen--; | |
0f9f32ac | 288 | return skb; |
07aaa115 | 289 | } |
11274e5a SH |
290 | |
291 | if (unlikely(q->qdisc->ops->requeue(skb, q->qdisc) != NET_XMIT_SUCCESS)) { | |
292 | qdisc_tree_decrease_qlen(q->qdisc, 1); | |
293 | sch->qstats.drops++; | |
294 | printk(KERN_ERR "netem: %s could not requeue\n", | |
295 | q->qdisc->ops->id); | |
296 | } | |
297 | ||
298 | qdisc_watchdog_schedule(&q->watchdog, cb->time_to_send); | |
0f9f32ac SH |
299 | } |
300 | ||
301 | return NULL; | |
1da177e4 LT |
302 | } |
303 | ||
1da177e4 LT |
304 | static void netem_reset(struct Qdisc *sch) |
305 | { | |
306 | struct netem_sched_data *q = qdisc_priv(sch); | |
307 | ||
308 | qdisc_reset(q->qdisc); | |
1da177e4 | 309 | sch->q.qlen = 0; |
59cb5c67 | 310 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
311 | } |
312 | ||
300ce174 | 313 | /* Pass size change message down to embedded FIFO */ |
1da177e4 LT |
314 | static int set_fifo_limit(struct Qdisc *q, int limit) |
315 | { | |
1e90474c | 316 | struct nlattr *nla; |
1da177e4 LT |
317 | int ret = -ENOMEM; |
318 | ||
300ce174 SH |
319 | /* Hack to avoid sending change message to non-FIFO */ |
320 | if (strncmp(q->ops->id + 1, "fifo", 4) != 0) | |
321 | return 0; | |
322 | ||
1e90474c PM |
323 | nla = kmalloc(nla_attr_size(sizeof(struct tc_fifo_qopt)), GFP_KERNEL); |
324 | if (nla) { | |
325 | nla->nla_type = RTM_NEWQDISC; | |
326 | nla->nla_len = nla_attr_size(sizeof(struct tc_fifo_qopt)); | |
327 | ((struct tc_fifo_qopt *)nla_data(nla))->limit = limit; | |
10297b99 | 328 | |
1e90474c PM |
329 | ret = q->ops->change(q, nla); |
330 | kfree(nla); | |
1da177e4 LT |
331 | } |
332 | return ret; | |
333 | } | |
334 | ||
335 | /* | |
336 | * Distribution data is a variable size payload containing | |
337 | * signed 16 bit values. | |
338 | */ | |
1e90474c | 339 | static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr) |
1da177e4 LT |
340 | { |
341 | struct netem_sched_data *q = qdisc_priv(sch); | |
1e90474c PM |
342 | unsigned long n = nla_len(attr)/sizeof(__s16); |
343 | const __s16 *data = nla_data(attr); | |
1da177e4 LT |
344 | struct disttable *d; |
345 | int i; | |
346 | ||
347 | if (n > 65536) | |
348 | return -EINVAL; | |
349 | ||
350 | d = kmalloc(sizeof(*d) + n*sizeof(d->table[0]), GFP_KERNEL); | |
351 | if (!d) | |
352 | return -ENOMEM; | |
353 | ||
354 | d->size = n; | |
355 | for (i = 0; i < n; i++) | |
356 | d->table[i] = data[i]; | |
10297b99 | 357 | |
1da177e4 LT |
358 | spin_lock_bh(&sch->dev->queue_lock); |
359 | d = xchg(&q->delay_dist, d); | |
360 | spin_unlock_bh(&sch->dev->queue_lock); | |
361 | ||
362 | kfree(d); | |
363 | return 0; | |
364 | } | |
365 | ||
1e90474c | 366 | static int get_correlation(struct Qdisc *sch, const struct nlattr *attr) |
1da177e4 LT |
367 | { |
368 | struct netem_sched_data *q = qdisc_priv(sch); | |
1e90474c | 369 | const struct tc_netem_corr *c = nla_data(attr); |
1da177e4 | 370 | |
1da177e4 LT |
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 | ||
1e90474c | 377 | static int get_reorder(struct Qdisc *sch, const struct nlattr *attr) |
0dca51d3 SH |
378 | { |
379 | struct netem_sched_data *q = qdisc_priv(sch); | |
1e90474c | 380 | const struct tc_netem_reorder *r = nla_data(attr); |
0dca51d3 | 381 | |
0dca51d3 SH |
382 | q->reorder = r->probability; |
383 | init_crandom(&q->reorder_cor, r->correlation); | |
384 | return 0; | |
385 | } | |
386 | ||
1e90474c | 387 | static int get_corrupt(struct Qdisc *sch, const struct nlattr *attr) |
c865e5d9 SH |
388 | { |
389 | struct netem_sched_data *q = qdisc_priv(sch); | |
1e90474c | 390 | const struct tc_netem_corrupt *r = nla_data(attr); |
c865e5d9 | 391 | |
c865e5d9 SH |
392 | q->corrupt = r->probability; |
393 | init_crandom(&q->corrupt_cor, r->correlation); | |
394 | return 0; | |
395 | } | |
396 | ||
27a3421e PM |
397 | static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = { |
398 | [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) }, | |
399 | [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) }, | |
400 | [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) }, | |
401 | }; | |
402 | ||
c865e5d9 | 403 | /* Parse netlink message to set options */ |
1e90474c | 404 | static int netem_change(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
405 | { |
406 | struct netem_sched_data *q = qdisc_priv(sch); | |
b03f4672 | 407 | struct nlattr *tb[TCA_NETEM_MAX + 1]; |
1da177e4 LT |
408 | struct tc_netem_qopt *qopt; |
409 | int ret; | |
10297b99 | 410 | |
b03f4672 | 411 | if (opt == NULL) |
1da177e4 LT |
412 | return -EINVAL; |
413 | ||
27a3421e PM |
414 | ret = nla_parse_nested_compat(tb, TCA_NETEM_MAX, opt, netem_policy, |
415 | qopt, sizeof(*qopt)); | |
b03f4672 PM |
416 | if (ret < 0) |
417 | return ret; | |
418 | ||
1da177e4 LT |
419 | ret = set_fifo_limit(q->qdisc, qopt->limit); |
420 | if (ret) { | |
421 | pr_debug("netem: can't set fifo limit\n"); | |
422 | return ret; | |
423 | } | |
10297b99 | 424 | |
1da177e4 LT |
425 | q->latency = qopt->latency; |
426 | q->jitter = qopt->jitter; | |
427 | q->limit = qopt->limit; | |
428 | q->gap = qopt->gap; | |
0dca51d3 | 429 | q->counter = 0; |
1da177e4 LT |
430 | q->loss = qopt->loss; |
431 | q->duplicate = qopt->duplicate; | |
432 | ||
bb2f8cc0 SH |
433 | /* for compatibility with earlier versions. |
434 | * if gap is set, need to assume 100% probability | |
0dca51d3 | 435 | */ |
a362e0a7 SH |
436 | if (q->gap) |
437 | q->reorder = ~0; | |
0dca51d3 | 438 | |
b03f4672 PM |
439 | if (tb[TCA_NETEM_CORR]) { |
440 | ret = get_correlation(sch, tb[TCA_NETEM_CORR]); | |
441 | if (ret) | |
442 | return ret; | |
443 | } | |
1da177e4 | 444 | |
b03f4672 PM |
445 | if (tb[TCA_NETEM_DELAY_DIST]) { |
446 | ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]); | |
447 | if (ret) | |
448 | return ret; | |
449 | } | |
c865e5d9 | 450 | |
b03f4672 PM |
451 | if (tb[TCA_NETEM_REORDER]) { |
452 | ret = get_reorder(sch, tb[TCA_NETEM_REORDER]); | |
453 | if (ret) | |
454 | return ret; | |
455 | } | |
1da177e4 | 456 | |
b03f4672 PM |
457 | if (tb[TCA_NETEM_CORRUPT]) { |
458 | ret = get_corrupt(sch, tb[TCA_NETEM_CORRUPT]); | |
459 | if (ret) | |
460 | return ret; | |
c865e5d9 | 461 | } |
1da177e4 LT |
462 | |
463 | return 0; | |
464 | } | |
465 | ||
300ce174 SH |
466 | /* |
467 | * Special case version of FIFO queue for use by netem. | |
468 | * It queues in order based on timestamps in skb's | |
469 | */ | |
470 | struct fifo_sched_data { | |
471 | u32 limit; | |
075aa573 | 472 | psched_time_t oldest; |
300ce174 SH |
473 | }; |
474 | ||
475 | static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch) | |
476 | { | |
477 | struct fifo_sched_data *q = qdisc_priv(sch); | |
478 | struct sk_buff_head *list = &sch->q; | |
075aa573 | 479 | psched_time_t tnext = ((struct netem_skb_cb *)nskb->cb)->time_to_send; |
300ce174 SH |
480 | struct sk_buff *skb; |
481 | ||
482 | if (likely(skb_queue_len(list) < q->limit)) { | |
075aa573 | 483 | /* Optimize for add at tail */ |
104e0878 | 484 | if (likely(skb_queue_empty(list) || tnext >= q->oldest)) { |
075aa573 SH |
485 | q->oldest = tnext; |
486 | return qdisc_enqueue_tail(nskb, sch); | |
487 | } | |
488 | ||
300ce174 SH |
489 | skb_queue_reverse_walk(list, skb) { |
490 | const struct netem_skb_cb *cb | |
491 | = (const struct netem_skb_cb *)skb->cb; | |
492 | ||
104e0878 | 493 | if (tnext >= cb->time_to_send) |
300ce174 SH |
494 | break; |
495 | } | |
496 | ||
497 | __skb_queue_after(list, skb, nskb); | |
498 | ||
499 | sch->qstats.backlog += nskb->len; | |
500 | sch->bstats.bytes += nskb->len; | |
501 | sch->bstats.packets++; | |
502 | ||
503 | return NET_XMIT_SUCCESS; | |
504 | } | |
505 | ||
075aa573 | 506 | return qdisc_reshape_fail(nskb, sch); |
300ce174 SH |
507 | } |
508 | ||
1e90474c | 509 | static int tfifo_init(struct Qdisc *sch, struct nlattr *opt) |
300ce174 SH |
510 | { |
511 | struct fifo_sched_data *q = qdisc_priv(sch); | |
512 | ||
513 | if (opt) { | |
1e90474c PM |
514 | struct tc_fifo_qopt *ctl = nla_data(opt); |
515 | if (nla_len(opt) < sizeof(*ctl)) | |
300ce174 SH |
516 | return -EINVAL; |
517 | ||
518 | q->limit = ctl->limit; | |
519 | } else | |
520 | q->limit = max_t(u32, sch->dev->tx_queue_len, 1); | |
521 | ||
a084980d | 522 | q->oldest = PSCHED_PASTPERFECT; |
300ce174 SH |
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 | ||
1e90474c | 531 | NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); |
300ce174 SH |
532 | return skb->len; |
533 | ||
1e90474c | 534 | nla_put_failure: |
300ce174 SH |
535 | return -1; |
536 | } | |
537 | ||
20fea08b | 538 | static struct Qdisc_ops tfifo_qdisc_ops __read_mostly = { |
300ce174 SH |
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 | ||
1e90474c | 551 | static int netem_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
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); |
1e90474c | 589 | struct nlattr *nla = (struct nlattr *) b; |
1da177e4 LT |
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; | |
1e90474c | 601 | NLA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt); |
1da177e4 LT |
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; | |
1e90474c | 606 | NLA_PUT(skb, TCA_NETEM_CORR, sizeof(cor), &cor); |
0dca51d3 SH |
607 | |
608 | reorder.probability = q->reorder; | |
609 | reorder.correlation = q->reorder_cor.rho; | |
1e90474c | 610 | NLA_PUT(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder); |
0dca51d3 | 611 | |
c865e5d9 SH |
612 | corrupt.probability = q->corrupt; |
613 | corrupt.correlation = q->corrupt_cor.rho; | |
1e90474c | 614 | NLA_PUT(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt); |
c865e5d9 | 615 | |
1e90474c | 616 | nla->nla_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
617 | |
618 | return skb->len; | |
619 | ||
1e90474c | 620 | nla_put_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, |
1e90474c | 672 | struct nlattr **tca, unsigned long *arg) |
1da177e4 LT |
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 | ||
20fea08b | 699 | static const struct Qdisc_class_ops netem_class_ops = { |
1da177e4 LT |
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 | ||
20fea08b | 711 | static struct Qdisc_ops netem_qdisc_ops __read_mostly = { |
1da177e4 LT |
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"); |