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 | ||
b7f080cf | 16 | #include <linux/mm.h> |
1da177e4 | 17 | #include <linux/module.h> |
5a0e3ad6 | 18 | #include <linux/slab.h> |
1da177e4 LT |
19 | #include <linux/types.h> |
20 | #include <linux/kernel.h> | |
21 | #include <linux/errno.h> | |
1da177e4 | 22 | #include <linux/skbuff.h> |
78776d3f | 23 | #include <linux/vmalloc.h> |
1da177e4 LT |
24 | #include <linux/rtnetlink.h> |
25 | ||
dc5fc579 | 26 | #include <net/netlink.h> |
1da177e4 LT |
27 | #include <net/pkt_sched.h> |
28 | ||
250a65f7 | 29 | #define VERSION "1.3" |
eb229c4c | 30 | |
1da177e4 LT |
31 | /* Network Emulation Queuing algorithm. |
32 | ==================================== | |
33 | ||
34 | Sources: [1] Mark Carson, Darrin Santay, "NIST Net - A Linux-based | |
35 | Network Emulation Tool | |
36 | [2] Luigi Rizzo, DummyNet for FreeBSD | |
37 | ||
38 | ---------------------------------------------------------------- | |
39 | ||
40 | This started out as a simple way to delay outgoing packets to | |
41 | test TCP but has grown to include most of the functionality | |
42 | of a full blown network emulator like NISTnet. It can delay | |
43 | packets and add random jitter (and correlation). The random | |
44 | distribution can be loaded from a table as well to provide | |
45 | normal, Pareto, or experimental curves. Packet loss, | |
46 | duplication, and reordering can also be emulated. | |
47 | ||
48 | This qdisc does not do classification that can be handled in | |
49 | layering other disciplines. It does not need to do bandwidth | |
50 | control either since that can be handled by using token | |
51 | bucket or other rate control. | |
661b7972 | 52 | |
53 | Correlated Loss Generator models | |
54 | ||
55 | Added generation of correlated loss according to the | |
56 | "Gilbert-Elliot" model, a 4-state markov model. | |
57 | ||
58 | References: | |
59 | [1] NetemCLG Home http://netgroup.uniroma2.it/NetemCLG | |
60 | [2] S. Salsano, F. Ludovici, A. Ordine, "Definition of a general | |
61 | and intuitive loss model for packet networks and its implementation | |
62 | in the Netem module in the Linux kernel", available in [1] | |
63 | ||
64 | Authors: Stefano Salsano <stefano.salsano at uniroma2.it | |
65 | Fabio Ludovici <fabio.ludovici at yahoo.it> | |
1da177e4 LT |
66 | */ |
67 | ||
68 | struct netem_sched_data { | |
69 | struct Qdisc *qdisc; | |
59cb5c67 | 70 | struct qdisc_watchdog watchdog; |
1da177e4 | 71 | |
b407621c SH |
72 | psched_tdiff_t latency; |
73 | psched_tdiff_t jitter; | |
74 | ||
1da177e4 LT |
75 | u32 loss; |
76 | u32 limit; | |
77 | u32 counter; | |
78 | u32 gap; | |
1da177e4 | 79 | u32 duplicate; |
0dca51d3 | 80 | u32 reorder; |
c865e5d9 | 81 | u32 corrupt; |
7bc0f28c | 82 | u32 rate; |
1da177e4 LT |
83 | |
84 | struct crndstate { | |
b407621c SH |
85 | u32 last; |
86 | u32 rho; | |
c865e5d9 | 87 | } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor; |
1da177e4 LT |
88 | |
89 | struct disttable { | |
90 | u32 size; | |
91 | s16 table[0]; | |
92 | } *delay_dist; | |
661b7972 | 93 | |
94 | enum { | |
95 | CLG_RANDOM, | |
96 | CLG_4_STATES, | |
97 | CLG_GILB_ELL, | |
98 | } loss_model; | |
99 | ||
100 | /* Correlated Loss Generation models */ | |
101 | struct clgstate { | |
102 | /* state of the Markov chain */ | |
103 | u8 state; | |
104 | ||
105 | /* 4-states and Gilbert-Elliot models */ | |
106 | u32 a1; /* p13 for 4-states or p for GE */ | |
107 | u32 a2; /* p31 for 4-states or r for GE */ | |
108 | u32 a3; /* p32 for 4-states or h for GE */ | |
109 | u32 a4; /* p14 for 4-states or 1-k for GE */ | |
110 | u32 a5; /* p23 used only in 4-states */ | |
111 | } clg; | |
112 | ||
1da177e4 LT |
113 | }; |
114 | ||
115 | /* Time stamp put into socket buffer control block */ | |
116 | struct netem_skb_cb { | |
117 | psched_time_t time_to_send; | |
118 | }; | |
119 | ||
5f86173b JK |
120 | static inline struct netem_skb_cb *netem_skb_cb(struct sk_buff *skb) |
121 | { | |
175f9c1b JK |
122 | BUILD_BUG_ON(sizeof(skb->cb) < |
123 | sizeof(struct qdisc_skb_cb) + sizeof(struct netem_skb_cb)); | |
124 | return (struct netem_skb_cb *)qdisc_skb_cb(skb)->data; | |
5f86173b JK |
125 | } |
126 | ||
1da177e4 LT |
127 | /* init_crandom - initialize correlated random number generator |
128 | * Use entropy source for initial seed. | |
129 | */ | |
130 | static void init_crandom(struct crndstate *state, unsigned long rho) | |
131 | { | |
132 | state->rho = rho; | |
133 | state->last = net_random(); | |
134 | } | |
135 | ||
136 | /* get_crandom - correlated random number generator | |
137 | * Next number depends on last value. | |
138 | * rho is scaled to avoid floating point. | |
139 | */ | |
b407621c | 140 | static u32 get_crandom(struct crndstate *state) |
1da177e4 LT |
141 | { |
142 | u64 value, rho; | |
143 | unsigned long answer; | |
144 | ||
bb2f8cc0 | 145 | if (state->rho == 0) /* no correlation */ |
1da177e4 LT |
146 | return net_random(); |
147 | ||
148 | value = net_random(); | |
149 | rho = (u64)state->rho + 1; | |
150 | answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32; | |
151 | state->last = answer; | |
152 | return answer; | |
153 | } | |
154 | ||
661b7972 | 155 | /* loss_4state - 4-state model loss generator |
156 | * Generates losses according to the 4-state Markov chain adopted in | |
157 | * the GI (General and Intuitive) loss model. | |
158 | */ | |
159 | static bool loss_4state(struct netem_sched_data *q) | |
160 | { | |
161 | struct clgstate *clg = &q->clg; | |
162 | u32 rnd = net_random(); | |
163 | ||
164 | /* | |
25985edc | 165 | * Makes a comparison between rnd and the transition |
661b7972 | 166 | * probabilities outgoing from the current state, then decides the |
167 | * next state and if the next packet has to be transmitted or lost. | |
168 | * The four states correspond to: | |
169 | * 1 => successfully transmitted packets within a gap period | |
170 | * 4 => isolated losses within a gap period | |
171 | * 3 => lost packets within a burst period | |
172 | * 2 => successfully transmitted packets within a burst period | |
173 | */ | |
174 | switch (clg->state) { | |
175 | case 1: | |
176 | if (rnd < clg->a4) { | |
177 | clg->state = 4; | |
178 | return true; | |
179 | } else if (clg->a4 < rnd && rnd < clg->a1) { | |
180 | clg->state = 3; | |
181 | return true; | |
182 | } else if (clg->a1 < rnd) | |
183 | clg->state = 1; | |
184 | ||
185 | break; | |
186 | case 2: | |
187 | if (rnd < clg->a5) { | |
188 | clg->state = 3; | |
189 | return true; | |
190 | } else | |
191 | clg->state = 2; | |
192 | ||
193 | break; | |
194 | case 3: | |
195 | if (rnd < clg->a3) | |
196 | clg->state = 2; | |
197 | else if (clg->a3 < rnd && rnd < clg->a2 + clg->a3) { | |
198 | clg->state = 1; | |
199 | return true; | |
200 | } else if (clg->a2 + clg->a3 < rnd) { | |
201 | clg->state = 3; | |
202 | return true; | |
203 | } | |
204 | break; | |
205 | case 4: | |
206 | clg->state = 1; | |
207 | break; | |
208 | } | |
209 | ||
210 | return false; | |
211 | } | |
212 | ||
213 | /* loss_gilb_ell - Gilbert-Elliot model loss generator | |
214 | * Generates losses according to the Gilbert-Elliot loss model or | |
215 | * its special cases (Gilbert or Simple Gilbert) | |
216 | * | |
25985edc | 217 | * Makes a comparison between random number and the transition |
661b7972 | 218 | * probabilities outgoing from the current state, then decides the |
25985edc | 219 | * next state. A second random number is extracted and the comparison |
661b7972 | 220 | * with the loss probability of the current state decides if the next |
221 | * packet will be transmitted or lost. | |
222 | */ | |
223 | static bool loss_gilb_ell(struct netem_sched_data *q) | |
224 | { | |
225 | struct clgstate *clg = &q->clg; | |
226 | ||
227 | switch (clg->state) { | |
228 | case 1: | |
229 | if (net_random() < clg->a1) | |
230 | clg->state = 2; | |
231 | if (net_random() < clg->a4) | |
232 | return true; | |
233 | case 2: | |
234 | if (net_random() < clg->a2) | |
235 | clg->state = 1; | |
236 | if (clg->a3 > net_random()) | |
237 | return true; | |
238 | } | |
239 | ||
240 | return false; | |
241 | } | |
242 | ||
243 | static bool loss_event(struct netem_sched_data *q) | |
244 | { | |
245 | switch (q->loss_model) { | |
246 | case CLG_RANDOM: | |
247 | /* Random packet drop 0 => none, ~0 => all */ | |
248 | return q->loss && q->loss >= get_crandom(&q->loss_cor); | |
249 | ||
250 | case CLG_4_STATES: | |
251 | /* 4state loss model algorithm (used also for GI model) | |
252 | * Extracts a value from the markov 4 state loss generator, | |
253 | * if it is 1 drops a packet and if needed writes the event in | |
254 | * the kernel logs | |
255 | */ | |
256 | return loss_4state(q); | |
257 | ||
258 | case CLG_GILB_ELL: | |
259 | /* Gilbert-Elliot loss model algorithm | |
260 | * Extracts a value from the Gilbert-Elliot loss generator, | |
261 | * if it is 1 drops a packet and if needed writes the event in | |
262 | * the kernel logs | |
263 | */ | |
264 | return loss_gilb_ell(q); | |
265 | } | |
266 | ||
267 | return false; /* not reached */ | |
268 | } | |
269 | ||
270 | ||
1da177e4 LT |
271 | /* tabledist - return a pseudo-randomly distributed value with mean mu and |
272 | * std deviation sigma. Uses table lookup to approximate the desired | |
273 | * distribution, and a uniformly-distributed pseudo-random source. | |
274 | */ | |
b407621c SH |
275 | static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma, |
276 | struct crndstate *state, | |
277 | const struct disttable *dist) | |
1da177e4 | 278 | { |
b407621c SH |
279 | psched_tdiff_t x; |
280 | long t; | |
281 | u32 rnd; | |
1da177e4 LT |
282 | |
283 | if (sigma == 0) | |
284 | return mu; | |
285 | ||
286 | rnd = get_crandom(state); | |
287 | ||
288 | /* default uniform distribution */ | |
10297b99 | 289 | if (dist == NULL) |
1da177e4 LT |
290 | return (rnd % (2*sigma)) - sigma + mu; |
291 | ||
292 | t = dist->table[rnd % dist->size]; | |
293 | x = (sigma % NETEM_DIST_SCALE) * t; | |
294 | if (x >= 0) | |
295 | x += NETEM_DIST_SCALE/2; | |
296 | else | |
297 | x -= NETEM_DIST_SCALE/2; | |
298 | ||
299 | return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu; | |
300 | } | |
301 | ||
7bc0f28c HPP |
302 | static psched_time_t packet_len_2_sched_time(unsigned int len, u32 rate) |
303 | { | |
fc33cc72 ED |
304 | u64 ticks = (u64)len * NSEC_PER_SEC; |
305 | ||
306 | do_div(ticks, rate); | |
307 | return PSCHED_NS2TICKS(ticks); | |
7bc0f28c HPP |
308 | } |
309 | ||
0afb51e7 SH |
310 | /* |
311 | * Insert one skb into qdisc. | |
312 | * Note: parent depends on return value to account for queue length. | |
313 | * NET_XMIT_DROP: queue length didn't change. | |
314 | * NET_XMIT_SUCCESS: one skb was queued. | |
315 | */ | |
1da177e4 LT |
316 | static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
317 | { | |
318 | struct netem_sched_data *q = qdisc_priv(sch); | |
89e1df74 GC |
319 | /* We don't fill cb now as skb_unshare() may invalidate it */ |
320 | struct netem_skb_cb *cb; | |
0afb51e7 | 321 | struct sk_buff *skb2; |
1da177e4 | 322 | int ret; |
0afb51e7 | 323 | int count = 1; |
1da177e4 | 324 | |
0afb51e7 SH |
325 | /* Random duplication */ |
326 | if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor)) | |
327 | ++count; | |
328 | ||
661b7972 | 329 | /* Drop packet? */ |
330 | if (loss_event(q)) | |
0afb51e7 SH |
331 | --count; |
332 | ||
333 | if (count == 0) { | |
1da177e4 LT |
334 | sch->qstats.drops++; |
335 | kfree_skb(skb); | |
c27f339a | 336 | return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
1da177e4 LT |
337 | } |
338 | ||
4e8a5201 DM |
339 | skb_orphan(skb); |
340 | ||
0afb51e7 SH |
341 | /* |
342 | * If we need to duplicate packet, then re-insert at top of the | |
343 | * qdisc tree, since parent queuer expects that only one | |
344 | * skb will be queued. | |
345 | */ | |
346 | if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) { | |
7698b4fc | 347 | struct Qdisc *rootq = qdisc_root(sch); |
0afb51e7 SH |
348 | u32 dupsave = q->duplicate; /* prevent duplicating a dup... */ |
349 | q->duplicate = 0; | |
350 | ||
5f86173b | 351 | qdisc_enqueue_root(skb2, rootq); |
0afb51e7 | 352 | q->duplicate = dupsave; |
1da177e4 LT |
353 | } |
354 | ||
c865e5d9 SH |
355 | /* |
356 | * Randomized packet corruption. | |
357 | * Make copy if needed since we are modifying | |
358 | * If packet is going to be hardware checksummed, then | |
359 | * do it now in software before we mangle it. | |
360 | */ | |
361 | if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) { | |
f64f9e71 JP |
362 | if (!(skb = skb_unshare(skb, GFP_ATOMIC)) || |
363 | (skb->ip_summed == CHECKSUM_PARTIAL && | |
364 | skb_checksum_help(skb))) { | |
c865e5d9 SH |
365 | sch->qstats.drops++; |
366 | return NET_XMIT_DROP; | |
367 | } | |
368 | ||
369 | skb->data[net_random() % skb_headlen(skb)] ^= 1<<(net_random() % 8); | |
370 | } | |
371 | ||
5f86173b | 372 | cb = netem_skb_cb(skb); |
cc7ec456 ED |
373 | if (q->gap == 0 || /* not doing reordering */ |
374 | q->counter < q->gap || /* inside last reordering gap */ | |
f64f9e71 | 375 | q->reorder < get_crandom(&q->reorder_cor)) { |
0f9f32ac | 376 | psched_time_t now; |
07aaa115 SH |
377 | psched_tdiff_t delay; |
378 | ||
379 | delay = tabledist(q->latency, q->jitter, | |
380 | &q->delay_cor, q->delay_dist); | |
381 | ||
3bebcda2 | 382 | now = psched_get_time(); |
7bc0f28c HPP |
383 | |
384 | if (q->rate) { | |
385 | struct sk_buff_head *list = &q->qdisc->q; | |
386 | ||
387 | delay += packet_len_2_sched_time(skb->len, q->rate); | |
388 | ||
389 | if (!skb_queue_empty(list)) { | |
390 | /* | |
391 | * Last packet in queue is reference point (now). | |
392 | * First packet in queue is already in flight, | |
393 | * calculate this time bonus and substract | |
394 | * from delay. | |
395 | */ | |
396 | delay -= now - netem_skb_cb(skb_peek(list))->time_to_send; | |
397 | now = netem_skb_cb(skb_peek_tail(list))->time_to_send; | |
398 | } | |
399 | } | |
400 | ||
7c59e25f | 401 | cb->time_to_send = now + delay; |
1da177e4 | 402 | ++q->counter; |
5f86173b | 403 | ret = qdisc_enqueue(skb, q->qdisc); |
1da177e4 | 404 | } else { |
10297b99 | 405 | /* |
0dca51d3 SH |
406 | * Do re-ordering by putting one out of N packets at the front |
407 | * of the queue. | |
408 | */ | |
3bebcda2 | 409 | cb->time_to_send = psched_get_time(); |
0dca51d3 | 410 | q->counter = 0; |
8ba25dad JP |
411 | |
412 | __skb_queue_head(&q->qdisc->q, skb); | |
413 | q->qdisc->qstats.backlog += qdisc_pkt_len(skb); | |
414 | q->qdisc->qstats.requeues++; | |
415 | ret = NET_XMIT_SUCCESS; | |
1da177e4 LT |
416 | } |
417 | ||
10f6dfcf | 418 | if (ret != NET_XMIT_SUCCESS) { |
419 | if (net_xmit_drop_count(ret)) { | |
420 | sch->qstats.drops++; | |
421 | return ret; | |
422 | } | |
378a2f09 | 423 | } |
1da177e4 | 424 | |
10f6dfcf | 425 | sch->q.qlen++; |
426 | return NET_XMIT_SUCCESS; | |
1da177e4 LT |
427 | } |
428 | ||
cc7ec456 | 429 | static unsigned int netem_drop(struct Qdisc *sch) |
1da177e4 LT |
430 | { |
431 | struct netem_sched_data *q = qdisc_priv(sch); | |
6d037a26 | 432 | unsigned int len = 0; |
1da177e4 | 433 | |
6d037a26 | 434 | if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) { |
1da177e4 LT |
435 | sch->q.qlen--; |
436 | sch->qstats.drops++; | |
437 | } | |
438 | return len; | |
439 | } | |
440 | ||
1da177e4 LT |
441 | static struct sk_buff *netem_dequeue(struct Qdisc *sch) |
442 | { | |
443 | struct netem_sched_data *q = qdisc_priv(sch); | |
444 | struct sk_buff *skb; | |
445 | ||
fd245a4a | 446 | if (qdisc_is_throttled(sch)) |
11274e5a SH |
447 | return NULL; |
448 | ||
03c05f0d | 449 | skb = q->qdisc->ops->peek(q->qdisc); |
771018e7 | 450 | if (skb) { |
5f86173b | 451 | const struct netem_skb_cb *cb = netem_skb_cb(skb); |
3bebcda2 | 452 | psched_time_t now = psched_get_time(); |
0f9f32ac SH |
453 | |
454 | /* if more time remaining? */ | |
104e0878 | 455 | if (cb->time_to_send <= now) { |
77be155c JP |
456 | skb = qdisc_dequeue_peeked(q->qdisc); |
457 | if (unlikely(!skb)) | |
03c05f0d JP |
458 | return NULL; |
459 | ||
8caf1539 JP |
460 | #ifdef CONFIG_NET_CLS_ACT |
461 | /* | |
462 | * If it's at ingress let's pretend the delay is | |
463 | * from the network (tstamp will be updated). | |
464 | */ | |
465 | if (G_TC_FROM(skb->tc_verd) & AT_INGRESS) | |
466 | skb->tstamp.tv64 = 0; | |
467 | #endif | |
10f6dfcf | 468 | |
0f9f32ac | 469 | sch->q.qlen--; |
10f6dfcf | 470 | qdisc_unthrottled(sch); |
471 | qdisc_bstats_update(sch, skb); | |
0f9f32ac | 472 | return skb; |
07aaa115 | 473 | } |
11274e5a | 474 | |
11274e5a | 475 | qdisc_watchdog_schedule(&q->watchdog, cb->time_to_send); |
0f9f32ac SH |
476 | } |
477 | ||
478 | return NULL; | |
1da177e4 LT |
479 | } |
480 | ||
1da177e4 LT |
481 | static void netem_reset(struct Qdisc *sch) |
482 | { | |
483 | struct netem_sched_data *q = qdisc_priv(sch); | |
484 | ||
485 | qdisc_reset(q->qdisc); | |
1da177e4 | 486 | sch->q.qlen = 0; |
59cb5c67 | 487 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
488 | } |
489 | ||
6373a9a2 | 490 | static void dist_free(struct disttable *d) |
491 | { | |
492 | if (d) { | |
493 | if (is_vmalloc_addr(d)) | |
494 | vfree(d); | |
495 | else | |
496 | kfree(d); | |
497 | } | |
498 | } | |
499 | ||
1da177e4 LT |
500 | /* |
501 | * Distribution data is a variable size payload containing | |
502 | * signed 16 bit values. | |
503 | */ | |
1e90474c | 504 | static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr) |
1da177e4 LT |
505 | { |
506 | struct netem_sched_data *q = qdisc_priv(sch); | |
6373a9a2 | 507 | size_t n = nla_len(attr)/sizeof(__s16); |
1e90474c | 508 | const __s16 *data = nla_data(attr); |
7698b4fc | 509 | spinlock_t *root_lock; |
1da177e4 LT |
510 | struct disttable *d; |
511 | int i; | |
6373a9a2 | 512 | size_t s; |
1da177e4 | 513 | |
df173bda | 514 | if (n > NETEM_DIST_MAX) |
1da177e4 LT |
515 | return -EINVAL; |
516 | ||
6373a9a2 | 517 | s = sizeof(struct disttable) + n * sizeof(s16); |
518 | d = kmalloc(s, GFP_KERNEL); | |
519 | if (!d) | |
520 | d = vmalloc(s); | |
1da177e4 LT |
521 | if (!d) |
522 | return -ENOMEM; | |
523 | ||
524 | d->size = n; | |
525 | for (i = 0; i < n; i++) | |
526 | d->table[i] = data[i]; | |
10297b99 | 527 | |
102396ae | 528 | root_lock = qdisc_root_sleeping_lock(sch); |
7698b4fc DM |
529 | |
530 | spin_lock_bh(root_lock); | |
6373a9a2 | 531 | dist_free(q->delay_dist); |
b94c8afc | 532 | q->delay_dist = d; |
7698b4fc | 533 | spin_unlock_bh(root_lock); |
1da177e4 LT |
534 | return 0; |
535 | } | |
536 | ||
265eb67f | 537 | static void get_correlation(struct Qdisc *sch, const struct nlattr *attr) |
1da177e4 LT |
538 | { |
539 | struct netem_sched_data *q = qdisc_priv(sch); | |
1e90474c | 540 | const struct tc_netem_corr *c = nla_data(attr); |
1da177e4 | 541 | |
1da177e4 LT |
542 | init_crandom(&q->delay_cor, c->delay_corr); |
543 | init_crandom(&q->loss_cor, c->loss_corr); | |
544 | init_crandom(&q->dup_cor, c->dup_corr); | |
1da177e4 LT |
545 | } |
546 | ||
265eb67f | 547 | static void get_reorder(struct Qdisc *sch, const struct nlattr *attr) |
0dca51d3 SH |
548 | { |
549 | struct netem_sched_data *q = qdisc_priv(sch); | |
1e90474c | 550 | const struct tc_netem_reorder *r = nla_data(attr); |
0dca51d3 | 551 | |
0dca51d3 SH |
552 | q->reorder = r->probability; |
553 | init_crandom(&q->reorder_cor, r->correlation); | |
0dca51d3 SH |
554 | } |
555 | ||
265eb67f | 556 | static void get_corrupt(struct Qdisc *sch, const struct nlattr *attr) |
c865e5d9 SH |
557 | { |
558 | struct netem_sched_data *q = qdisc_priv(sch); | |
1e90474c | 559 | const struct tc_netem_corrupt *r = nla_data(attr); |
c865e5d9 | 560 | |
c865e5d9 SH |
561 | q->corrupt = r->probability; |
562 | init_crandom(&q->corrupt_cor, r->correlation); | |
c865e5d9 SH |
563 | } |
564 | ||
7bc0f28c HPP |
565 | static void get_rate(struct Qdisc *sch, const struct nlattr *attr) |
566 | { | |
567 | struct netem_sched_data *q = qdisc_priv(sch); | |
568 | const struct tc_netem_rate *r = nla_data(attr); | |
569 | ||
570 | q->rate = r->rate; | |
571 | } | |
572 | ||
661b7972 | 573 | static int get_loss_clg(struct Qdisc *sch, const struct nlattr *attr) |
574 | { | |
575 | struct netem_sched_data *q = qdisc_priv(sch); | |
576 | const struct nlattr *la; | |
577 | int rem; | |
578 | ||
579 | nla_for_each_nested(la, attr, rem) { | |
580 | u16 type = nla_type(la); | |
581 | ||
582 | switch(type) { | |
583 | case NETEM_LOSS_GI: { | |
584 | const struct tc_netem_gimodel *gi = nla_data(la); | |
585 | ||
586 | if (nla_len(la) != sizeof(struct tc_netem_gimodel)) { | |
587 | pr_info("netem: incorrect gi model size\n"); | |
588 | return -EINVAL; | |
589 | } | |
590 | ||
591 | q->loss_model = CLG_4_STATES; | |
592 | ||
593 | q->clg.state = 1; | |
594 | q->clg.a1 = gi->p13; | |
595 | q->clg.a2 = gi->p31; | |
596 | q->clg.a3 = gi->p32; | |
597 | q->clg.a4 = gi->p14; | |
598 | q->clg.a5 = gi->p23; | |
599 | break; | |
600 | } | |
601 | ||
602 | case NETEM_LOSS_GE: { | |
603 | const struct tc_netem_gemodel *ge = nla_data(la); | |
604 | ||
605 | if (nla_len(la) != sizeof(struct tc_netem_gemodel)) { | |
606 | pr_info("netem: incorrect gi model size\n"); | |
607 | return -EINVAL; | |
608 | } | |
609 | ||
610 | q->loss_model = CLG_GILB_ELL; | |
611 | q->clg.state = 1; | |
612 | q->clg.a1 = ge->p; | |
613 | q->clg.a2 = ge->r; | |
614 | q->clg.a3 = ge->h; | |
615 | q->clg.a4 = ge->k1; | |
616 | break; | |
617 | } | |
618 | ||
619 | default: | |
620 | pr_info("netem: unknown loss type %u\n", type); | |
621 | return -EINVAL; | |
622 | } | |
623 | } | |
624 | ||
625 | return 0; | |
626 | } | |
627 | ||
27a3421e PM |
628 | static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = { |
629 | [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) }, | |
630 | [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) }, | |
631 | [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) }, | |
7bc0f28c | 632 | [TCA_NETEM_RATE] = { .len = sizeof(struct tc_netem_rate) }, |
661b7972 | 633 | [TCA_NETEM_LOSS] = { .type = NLA_NESTED }, |
27a3421e PM |
634 | }; |
635 | ||
2c10b32b TG |
636 | static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla, |
637 | const struct nla_policy *policy, int len) | |
638 | { | |
639 | int nested_len = nla_len(nla) - NLA_ALIGN(len); | |
640 | ||
661b7972 | 641 | if (nested_len < 0) { |
642 | pr_info("netem: invalid attributes len %d\n", nested_len); | |
2c10b32b | 643 | return -EINVAL; |
661b7972 | 644 | } |
645 | ||
2c10b32b TG |
646 | if (nested_len >= nla_attr_size(0)) |
647 | return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len), | |
648 | nested_len, policy); | |
661b7972 | 649 | |
2c10b32b TG |
650 | memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1)); |
651 | return 0; | |
652 | } | |
653 | ||
c865e5d9 | 654 | /* Parse netlink message to set options */ |
1e90474c | 655 | static int netem_change(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
656 | { |
657 | struct netem_sched_data *q = qdisc_priv(sch); | |
b03f4672 | 658 | struct nlattr *tb[TCA_NETEM_MAX + 1]; |
1da177e4 LT |
659 | struct tc_netem_qopt *qopt; |
660 | int ret; | |
10297b99 | 661 | |
b03f4672 | 662 | if (opt == NULL) |
1da177e4 LT |
663 | return -EINVAL; |
664 | ||
2c10b32b TG |
665 | qopt = nla_data(opt); |
666 | ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt)); | |
b03f4672 PM |
667 | if (ret < 0) |
668 | return ret; | |
669 | ||
fb0305ce | 670 | ret = fifo_set_limit(q->qdisc, qopt->limit); |
1da177e4 | 671 | if (ret) { |
250a65f7 | 672 | pr_info("netem: can't set fifo limit\n"); |
1da177e4 LT |
673 | return ret; |
674 | } | |
10297b99 | 675 | |
1da177e4 LT |
676 | q->latency = qopt->latency; |
677 | q->jitter = qopt->jitter; | |
678 | q->limit = qopt->limit; | |
679 | q->gap = qopt->gap; | |
0dca51d3 | 680 | q->counter = 0; |
1da177e4 LT |
681 | q->loss = qopt->loss; |
682 | q->duplicate = qopt->duplicate; | |
683 | ||
bb2f8cc0 SH |
684 | /* for compatibility with earlier versions. |
685 | * if gap is set, need to assume 100% probability | |
0dca51d3 | 686 | */ |
a362e0a7 SH |
687 | if (q->gap) |
688 | q->reorder = ~0; | |
0dca51d3 | 689 | |
265eb67f SH |
690 | if (tb[TCA_NETEM_CORR]) |
691 | get_correlation(sch, tb[TCA_NETEM_CORR]); | |
1da177e4 | 692 | |
b03f4672 PM |
693 | if (tb[TCA_NETEM_DELAY_DIST]) { |
694 | ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]); | |
695 | if (ret) | |
696 | return ret; | |
697 | } | |
c865e5d9 | 698 | |
265eb67f SH |
699 | if (tb[TCA_NETEM_REORDER]) |
700 | get_reorder(sch, tb[TCA_NETEM_REORDER]); | |
1da177e4 | 701 | |
265eb67f SH |
702 | if (tb[TCA_NETEM_CORRUPT]) |
703 | get_corrupt(sch, tb[TCA_NETEM_CORRUPT]); | |
1da177e4 | 704 | |
7bc0f28c HPP |
705 | if (tb[TCA_NETEM_RATE]) |
706 | get_rate(sch, tb[TCA_NETEM_RATE]); | |
707 | ||
661b7972 | 708 | q->loss_model = CLG_RANDOM; |
709 | if (tb[TCA_NETEM_LOSS]) | |
710 | ret = get_loss_clg(sch, tb[TCA_NETEM_LOSS]); | |
711 | ||
712 | return ret; | |
1da177e4 LT |
713 | } |
714 | ||
300ce174 SH |
715 | /* |
716 | * Special case version of FIFO queue for use by netem. | |
717 | * It queues in order based on timestamps in skb's | |
718 | */ | |
719 | struct fifo_sched_data { | |
720 | u32 limit; | |
075aa573 | 721 | psched_time_t oldest; |
300ce174 SH |
722 | }; |
723 | ||
724 | static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch) | |
725 | { | |
726 | struct fifo_sched_data *q = qdisc_priv(sch); | |
727 | struct sk_buff_head *list = &sch->q; | |
5f86173b | 728 | psched_time_t tnext = netem_skb_cb(nskb)->time_to_send; |
300ce174 SH |
729 | struct sk_buff *skb; |
730 | ||
731 | if (likely(skb_queue_len(list) < q->limit)) { | |
075aa573 | 732 | /* Optimize for add at tail */ |
104e0878 | 733 | if (likely(skb_queue_empty(list) || tnext >= q->oldest)) { |
075aa573 SH |
734 | q->oldest = tnext; |
735 | return qdisc_enqueue_tail(nskb, sch); | |
736 | } | |
737 | ||
300ce174 | 738 | skb_queue_reverse_walk(list, skb) { |
5f86173b | 739 | const struct netem_skb_cb *cb = netem_skb_cb(skb); |
300ce174 | 740 | |
104e0878 | 741 | if (tnext >= cb->time_to_send) |
300ce174 SH |
742 | break; |
743 | } | |
744 | ||
745 | __skb_queue_after(list, skb, nskb); | |
746 | ||
0abf77e5 | 747 | sch->qstats.backlog += qdisc_pkt_len(nskb); |
300ce174 SH |
748 | |
749 | return NET_XMIT_SUCCESS; | |
750 | } | |
751 | ||
075aa573 | 752 | return qdisc_reshape_fail(nskb, sch); |
300ce174 SH |
753 | } |
754 | ||
1e90474c | 755 | static int tfifo_init(struct Qdisc *sch, struct nlattr *opt) |
300ce174 SH |
756 | { |
757 | struct fifo_sched_data *q = qdisc_priv(sch); | |
758 | ||
759 | if (opt) { | |
1e90474c PM |
760 | struct tc_fifo_qopt *ctl = nla_data(opt); |
761 | if (nla_len(opt) < sizeof(*ctl)) | |
300ce174 SH |
762 | return -EINVAL; |
763 | ||
764 | q->limit = ctl->limit; | |
765 | } else | |
5ce2d488 | 766 | q->limit = max_t(u32, qdisc_dev(sch)->tx_queue_len, 1); |
300ce174 | 767 | |
a084980d | 768 | q->oldest = PSCHED_PASTPERFECT; |
300ce174 SH |
769 | return 0; |
770 | } | |
771 | ||
772 | static int tfifo_dump(struct Qdisc *sch, struct sk_buff *skb) | |
773 | { | |
774 | struct fifo_sched_data *q = qdisc_priv(sch); | |
775 | struct tc_fifo_qopt opt = { .limit = q->limit }; | |
776 | ||
1e90474c | 777 | NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); |
300ce174 SH |
778 | return skb->len; |
779 | ||
1e90474c | 780 | nla_put_failure: |
300ce174 SH |
781 | return -1; |
782 | } | |
783 | ||
20fea08b | 784 | static struct Qdisc_ops tfifo_qdisc_ops __read_mostly = { |
300ce174 SH |
785 | .id = "tfifo", |
786 | .priv_size = sizeof(struct fifo_sched_data), | |
787 | .enqueue = tfifo_enqueue, | |
788 | .dequeue = qdisc_dequeue_head, | |
8e3af978 | 789 | .peek = qdisc_peek_head, |
300ce174 SH |
790 | .drop = qdisc_queue_drop, |
791 | .init = tfifo_init, | |
792 | .reset = qdisc_reset_queue, | |
793 | .change = tfifo_init, | |
794 | .dump = tfifo_dump, | |
795 | }; | |
796 | ||
1e90474c | 797 | static int netem_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
798 | { |
799 | struct netem_sched_data *q = qdisc_priv(sch); | |
800 | int ret; | |
801 | ||
802 | if (!opt) | |
803 | return -EINVAL; | |
804 | ||
59cb5c67 | 805 | qdisc_watchdog_init(&q->watchdog, sch); |
1da177e4 | 806 | |
661b7972 | 807 | q->loss_model = CLG_RANDOM; |
3511c913 | 808 | q->qdisc = qdisc_create_dflt(sch->dev_queue, &tfifo_qdisc_ops, |
9f9afec4 | 809 | TC_H_MAKE(sch->handle, 1)); |
1da177e4 | 810 | if (!q->qdisc) { |
250a65f7 | 811 | pr_notice("netem: qdisc create tfifo qdisc failed\n"); |
1da177e4 LT |
812 | return -ENOMEM; |
813 | } | |
814 | ||
815 | ret = netem_change(sch, opt); | |
816 | if (ret) { | |
250a65f7 | 817 | pr_info("netem: change failed\n"); |
1da177e4 LT |
818 | qdisc_destroy(q->qdisc); |
819 | } | |
820 | return ret; | |
821 | } | |
822 | ||
823 | static void netem_destroy(struct Qdisc *sch) | |
824 | { | |
825 | struct netem_sched_data *q = qdisc_priv(sch); | |
826 | ||
59cb5c67 | 827 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 | 828 | qdisc_destroy(q->qdisc); |
6373a9a2 | 829 | dist_free(q->delay_dist); |
1da177e4 LT |
830 | } |
831 | ||
661b7972 | 832 | static int dump_loss_model(const struct netem_sched_data *q, |
833 | struct sk_buff *skb) | |
834 | { | |
835 | struct nlattr *nest; | |
836 | ||
837 | nest = nla_nest_start(skb, TCA_NETEM_LOSS); | |
838 | if (nest == NULL) | |
839 | goto nla_put_failure; | |
840 | ||
841 | switch (q->loss_model) { | |
842 | case CLG_RANDOM: | |
843 | /* legacy loss model */ | |
844 | nla_nest_cancel(skb, nest); | |
845 | return 0; /* no data */ | |
846 | ||
847 | case CLG_4_STATES: { | |
848 | struct tc_netem_gimodel gi = { | |
849 | .p13 = q->clg.a1, | |
850 | .p31 = q->clg.a2, | |
851 | .p32 = q->clg.a3, | |
852 | .p14 = q->clg.a4, | |
853 | .p23 = q->clg.a5, | |
854 | }; | |
855 | ||
856 | NLA_PUT(skb, NETEM_LOSS_GI, sizeof(gi), &gi); | |
857 | break; | |
858 | } | |
859 | case CLG_GILB_ELL: { | |
860 | struct tc_netem_gemodel ge = { | |
861 | .p = q->clg.a1, | |
862 | .r = q->clg.a2, | |
863 | .h = q->clg.a3, | |
864 | .k1 = q->clg.a4, | |
865 | }; | |
866 | ||
867 | NLA_PUT(skb, NETEM_LOSS_GE, sizeof(ge), &ge); | |
868 | break; | |
869 | } | |
870 | } | |
871 | ||
872 | nla_nest_end(skb, nest); | |
873 | return 0; | |
874 | ||
875 | nla_put_failure: | |
876 | nla_nest_cancel(skb, nest); | |
877 | return -1; | |
878 | } | |
879 | ||
1da177e4 LT |
880 | static int netem_dump(struct Qdisc *sch, struct sk_buff *skb) |
881 | { | |
882 | const struct netem_sched_data *q = qdisc_priv(sch); | |
861d7f74 | 883 | struct nlattr *nla = (struct nlattr *) skb_tail_pointer(skb); |
1da177e4 LT |
884 | struct tc_netem_qopt qopt; |
885 | struct tc_netem_corr cor; | |
0dca51d3 | 886 | struct tc_netem_reorder reorder; |
c865e5d9 | 887 | struct tc_netem_corrupt corrupt; |
7bc0f28c | 888 | struct tc_netem_rate rate; |
1da177e4 LT |
889 | |
890 | qopt.latency = q->latency; | |
891 | qopt.jitter = q->jitter; | |
892 | qopt.limit = q->limit; | |
893 | qopt.loss = q->loss; | |
894 | qopt.gap = q->gap; | |
895 | qopt.duplicate = q->duplicate; | |
1e90474c | 896 | NLA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt); |
1da177e4 LT |
897 | |
898 | cor.delay_corr = q->delay_cor.rho; | |
899 | cor.loss_corr = q->loss_cor.rho; | |
900 | cor.dup_corr = q->dup_cor.rho; | |
1e90474c | 901 | NLA_PUT(skb, TCA_NETEM_CORR, sizeof(cor), &cor); |
0dca51d3 SH |
902 | |
903 | reorder.probability = q->reorder; | |
904 | reorder.correlation = q->reorder_cor.rho; | |
1e90474c | 905 | NLA_PUT(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder); |
0dca51d3 | 906 | |
c865e5d9 SH |
907 | corrupt.probability = q->corrupt; |
908 | corrupt.correlation = q->corrupt_cor.rho; | |
1e90474c | 909 | NLA_PUT(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt); |
c865e5d9 | 910 | |
7bc0f28c HPP |
911 | rate.rate = q->rate; |
912 | NLA_PUT(skb, TCA_NETEM_RATE, sizeof(rate), &rate); | |
913 | ||
661b7972 | 914 | if (dump_loss_model(q, skb) != 0) |
915 | goto nla_put_failure; | |
916 | ||
861d7f74 | 917 | return nla_nest_end(skb, nla); |
1da177e4 | 918 | |
1e90474c | 919 | nla_put_failure: |
861d7f74 | 920 | nlmsg_trim(skb, nla); |
1da177e4 LT |
921 | return -1; |
922 | } | |
923 | ||
10f6dfcf | 924 | static int netem_dump_class(struct Qdisc *sch, unsigned long cl, |
925 | struct sk_buff *skb, struct tcmsg *tcm) | |
926 | { | |
927 | struct netem_sched_data *q = qdisc_priv(sch); | |
928 | ||
929 | if (cl != 1) /* only one class */ | |
930 | return -ENOENT; | |
931 | ||
932 | tcm->tcm_handle |= TC_H_MIN(1); | |
933 | tcm->tcm_info = q->qdisc->handle; | |
934 | ||
935 | return 0; | |
936 | } | |
937 | ||
938 | static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
939 | struct Qdisc **old) | |
940 | { | |
941 | struct netem_sched_data *q = qdisc_priv(sch); | |
942 | ||
943 | if (new == NULL) | |
944 | new = &noop_qdisc; | |
945 | ||
946 | sch_tree_lock(sch); | |
947 | *old = q->qdisc; | |
948 | q->qdisc = new; | |
949 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); | |
950 | qdisc_reset(*old); | |
951 | sch_tree_unlock(sch); | |
952 | ||
953 | return 0; | |
954 | } | |
955 | ||
956 | static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg) | |
957 | { | |
958 | struct netem_sched_data *q = qdisc_priv(sch); | |
959 | return q->qdisc; | |
960 | } | |
961 | ||
962 | static unsigned long netem_get(struct Qdisc *sch, u32 classid) | |
963 | { | |
964 | return 1; | |
965 | } | |
966 | ||
967 | static void netem_put(struct Qdisc *sch, unsigned long arg) | |
968 | { | |
969 | } | |
970 | ||
971 | static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker) | |
972 | { | |
973 | if (!walker->stop) { | |
974 | if (walker->count >= walker->skip) | |
975 | if (walker->fn(sch, 1, walker) < 0) { | |
976 | walker->stop = 1; | |
977 | return; | |
978 | } | |
979 | walker->count++; | |
980 | } | |
981 | } | |
982 | ||
983 | static const struct Qdisc_class_ops netem_class_ops = { | |
984 | .graft = netem_graft, | |
985 | .leaf = netem_leaf, | |
986 | .get = netem_get, | |
987 | .put = netem_put, | |
988 | .walk = netem_walk, | |
989 | .dump = netem_dump_class, | |
990 | }; | |
991 | ||
20fea08b | 992 | static struct Qdisc_ops netem_qdisc_ops __read_mostly = { |
1da177e4 | 993 | .id = "netem", |
10f6dfcf | 994 | .cl_ops = &netem_class_ops, |
1da177e4 LT |
995 | .priv_size = sizeof(struct netem_sched_data), |
996 | .enqueue = netem_enqueue, | |
997 | .dequeue = netem_dequeue, | |
77be155c | 998 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
999 | .drop = netem_drop, |
1000 | .init = netem_init, | |
1001 | .reset = netem_reset, | |
1002 | .destroy = netem_destroy, | |
1003 | .change = netem_change, | |
1004 | .dump = netem_dump, | |
1005 | .owner = THIS_MODULE, | |
1006 | }; | |
1007 | ||
1008 | ||
1009 | static int __init netem_module_init(void) | |
1010 | { | |
eb229c4c | 1011 | pr_info("netem: version " VERSION "\n"); |
1da177e4 LT |
1012 | return register_qdisc(&netem_qdisc_ops); |
1013 | } | |
1014 | static void __exit netem_module_exit(void) | |
1015 | { | |
1016 | unregister_qdisc(&netem_qdisc_ops); | |
1017 | } | |
1018 | module_init(netem_module_init) | |
1019 | module_exit(netem_module_exit) | |
1020 | MODULE_LICENSE("GPL"); |