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