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