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 | 24 | #include <linux/rtnetlink.h> |
90b41a1c | 25 | #include <linux/reciprocal_div.h> |
1da177e4 | 26 | |
dc5fc579 | 27 | #include <net/netlink.h> |
1da177e4 LT |
28 | #include <net/pkt_sched.h> |
29 | ||
250a65f7 | 30 | #define VERSION "1.3" |
eb229c4c | 31 | |
1da177e4 LT |
32 | /* Network Emulation Queuing algorithm. |
33 | ==================================== | |
34 | ||
35 | Sources: [1] Mark Carson, Darrin Santay, "NIST Net - A Linux-based | |
36 | Network Emulation Tool | |
37 | [2] Luigi Rizzo, DummyNet for FreeBSD | |
38 | ||
39 | ---------------------------------------------------------------- | |
40 | ||
41 | This started out as a simple way to delay outgoing packets to | |
42 | test TCP but has grown to include most of the functionality | |
43 | of a full blown network emulator like NISTnet. It can delay | |
44 | packets and add random jitter (and correlation). The random | |
45 | distribution can be loaded from a table as well to provide | |
46 | normal, Pareto, or experimental curves. Packet loss, | |
47 | duplication, and reordering can also be emulated. | |
48 | ||
49 | This qdisc does not do classification that can be handled in | |
50 | layering other disciplines. It does not need to do bandwidth | |
51 | control either since that can be handled by using token | |
52 | bucket or other rate control. | |
661b7972 | 53 | |
54 | Correlated Loss Generator models | |
55 | ||
56 | Added generation of correlated loss according to the | |
57 | "Gilbert-Elliot" model, a 4-state markov model. | |
58 | ||
59 | References: | |
60 | [1] NetemCLG Home http://netgroup.uniroma2.it/NetemCLG | |
61 | [2] S. Salsano, F. Ludovici, A. Ordine, "Definition of a general | |
62 | and intuitive loss model for packet networks and its implementation | |
63 | in the Netem module in the Linux kernel", available in [1] | |
64 | ||
65 | Authors: Stefano Salsano <stefano.salsano at uniroma2.it | |
66 | Fabio Ludovici <fabio.ludovici at yahoo.it> | |
1da177e4 LT |
67 | */ |
68 | ||
69 | struct netem_sched_data { | |
50612537 ED |
70 | /* internal t(ime)fifo qdisc uses sch->q and sch->limit */ |
71 | ||
72 | /* optional qdisc for classful handling (NULL at netem init) */ | |
1da177e4 | 73 | struct Qdisc *qdisc; |
50612537 | 74 | |
59cb5c67 | 75 | struct qdisc_watchdog watchdog; |
1da177e4 | 76 | |
b407621c SH |
77 | psched_tdiff_t latency; |
78 | psched_tdiff_t jitter; | |
79 | ||
1da177e4 LT |
80 | u32 loss; |
81 | u32 limit; | |
82 | u32 counter; | |
83 | u32 gap; | |
1da177e4 | 84 | u32 duplicate; |
0dca51d3 | 85 | u32 reorder; |
c865e5d9 | 86 | u32 corrupt; |
7bc0f28c | 87 | u32 rate; |
90b41a1c HPP |
88 | s32 packet_overhead; |
89 | u32 cell_size; | |
90 | u32 cell_size_reciprocal; | |
91 | s32 cell_overhead; | |
1da177e4 LT |
92 | |
93 | struct crndstate { | |
b407621c SH |
94 | u32 last; |
95 | u32 rho; | |
c865e5d9 | 96 | } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor; |
1da177e4 LT |
97 | |
98 | struct disttable { | |
99 | u32 size; | |
100 | s16 table[0]; | |
101 | } *delay_dist; | |
661b7972 | 102 | |
103 | enum { | |
104 | CLG_RANDOM, | |
105 | CLG_4_STATES, | |
106 | CLG_GILB_ELL, | |
107 | } loss_model; | |
108 | ||
109 | /* Correlated Loss Generation models */ | |
110 | struct clgstate { | |
111 | /* state of the Markov chain */ | |
112 | u8 state; | |
113 | ||
114 | /* 4-states and Gilbert-Elliot models */ | |
115 | u32 a1; /* p13 for 4-states or p for GE */ | |
116 | u32 a2; /* p31 for 4-states or r for GE */ | |
117 | u32 a3; /* p32 for 4-states or h for GE */ | |
118 | u32 a4; /* p14 for 4-states or 1-k for GE */ | |
119 | u32 a5; /* p23 used only in 4-states */ | |
120 | } clg; | |
121 | ||
1da177e4 LT |
122 | }; |
123 | ||
50612537 ED |
124 | /* Time stamp put into socket buffer control block |
125 | * Only valid when skbs are in our internal t(ime)fifo queue. | |
126 | */ | |
1da177e4 LT |
127 | struct netem_skb_cb { |
128 | psched_time_t time_to_send; | |
129 | }; | |
130 | ||
5f86173b JK |
131 | static inline struct netem_skb_cb *netem_skb_cb(struct sk_buff *skb) |
132 | { | |
16bda13d | 133 | qdisc_cb_private_validate(skb, sizeof(struct netem_skb_cb)); |
175f9c1b | 134 | return (struct netem_skb_cb *)qdisc_skb_cb(skb)->data; |
5f86173b JK |
135 | } |
136 | ||
1da177e4 LT |
137 | /* init_crandom - initialize correlated random number generator |
138 | * Use entropy source for initial seed. | |
139 | */ | |
140 | static void init_crandom(struct crndstate *state, unsigned long rho) | |
141 | { | |
142 | state->rho = rho; | |
143 | state->last = net_random(); | |
144 | } | |
145 | ||
146 | /* get_crandom - correlated random number generator | |
147 | * Next number depends on last value. | |
148 | * rho is scaled to avoid floating point. | |
149 | */ | |
b407621c | 150 | static u32 get_crandom(struct crndstate *state) |
1da177e4 LT |
151 | { |
152 | u64 value, rho; | |
153 | unsigned long answer; | |
154 | ||
bb2f8cc0 | 155 | if (state->rho == 0) /* no correlation */ |
1da177e4 LT |
156 | return net_random(); |
157 | ||
158 | value = net_random(); | |
159 | rho = (u64)state->rho + 1; | |
160 | answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32; | |
161 | state->last = answer; | |
162 | return answer; | |
163 | } | |
164 | ||
661b7972 | 165 | /* loss_4state - 4-state model loss generator |
166 | * Generates losses according to the 4-state Markov chain adopted in | |
167 | * the GI (General and Intuitive) loss model. | |
168 | */ | |
169 | static bool loss_4state(struct netem_sched_data *q) | |
170 | { | |
171 | struct clgstate *clg = &q->clg; | |
172 | u32 rnd = net_random(); | |
173 | ||
174 | /* | |
25985edc | 175 | * Makes a comparison between rnd and the transition |
661b7972 | 176 | * probabilities outgoing from the current state, then decides the |
177 | * next state and if the next packet has to be transmitted or lost. | |
178 | * The four states correspond to: | |
179 | * 1 => successfully transmitted packets within a gap period | |
180 | * 4 => isolated losses within a gap period | |
181 | * 3 => lost packets within a burst period | |
182 | * 2 => successfully transmitted packets within a burst period | |
183 | */ | |
184 | switch (clg->state) { | |
185 | case 1: | |
186 | if (rnd < clg->a4) { | |
187 | clg->state = 4; | |
188 | return true; | |
189 | } else if (clg->a4 < rnd && rnd < clg->a1) { | |
190 | clg->state = 3; | |
191 | return true; | |
192 | } else if (clg->a1 < rnd) | |
193 | clg->state = 1; | |
194 | ||
195 | break; | |
196 | case 2: | |
197 | if (rnd < clg->a5) { | |
198 | clg->state = 3; | |
199 | return true; | |
200 | } else | |
201 | clg->state = 2; | |
202 | ||
203 | break; | |
204 | case 3: | |
205 | if (rnd < clg->a3) | |
206 | clg->state = 2; | |
207 | else if (clg->a3 < rnd && rnd < clg->a2 + clg->a3) { | |
208 | clg->state = 1; | |
209 | return true; | |
210 | } else if (clg->a2 + clg->a3 < rnd) { | |
211 | clg->state = 3; | |
212 | return true; | |
213 | } | |
214 | break; | |
215 | case 4: | |
216 | clg->state = 1; | |
217 | break; | |
218 | } | |
219 | ||
220 | return false; | |
221 | } | |
222 | ||
223 | /* loss_gilb_ell - Gilbert-Elliot model loss generator | |
224 | * Generates losses according to the Gilbert-Elliot loss model or | |
225 | * its special cases (Gilbert or Simple Gilbert) | |
226 | * | |
25985edc | 227 | * Makes a comparison between random number and the transition |
661b7972 | 228 | * probabilities outgoing from the current state, then decides the |
25985edc | 229 | * next state. A second random number is extracted and the comparison |
661b7972 | 230 | * with the loss probability of the current state decides if the next |
231 | * packet will be transmitted or lost. | |
232 | */ | |
233 | static bool loss_gilb_ell(struct netem_sched_data *q) | |
234 | { | |
235 | struct clgstate *clg = &q->clg; | |
236 | ||
237 | switch (clg->state) { | |
238 | case 1: | |
239 | if (net_random() < clg->a1) | |
240 | clg->state = 2; | |
241 | if (net_random() < clg->a4) | |
242 | return true; | |
243 | case 2: | |
244 | if (net_random() < clg->a2) | |
245 | clg->state = 1; | |
246 | if (clg->a3 > net_random()) | |
247 | return true; | |
248 | } | |
249 | ||
250 | return false; | |
251 | } | |
252 | ||
253 | static bool loss_event(struct netem_sched_data *q) | |
254 | { | |
255 | switch (q->loss_model) { | |
256 | case CLG_RANDOM: | |
257 | /* Random packet drop 0 => none, ~0 => all */ | |
258 | return q->loss && q->loss >= get_crandom(&q->loss_cor); | |
259 | ||
260 | case CLG_4_STATES: | |
261 | /* 4state loss model algorithm (used also for GI model) | |
262 | * Extracts a value from the markov 4 state loss generator, | |
263 | * if it is 1 drops a packet and if needed writes the event in | |
264 | * the kernel logs | |
265 | */ | |
266 | return loss_4state(q); | |
267 | ||
268 | case CLG_GILB_ELL: | |
269 | /* Gilbert-Elliot loss model algorithm | |
270 | * Extracts a value from the Gilbert-Elliot loss generator, | |
271 | * if it is 1 drops a packet and if needed writes the event in | |
272 | * the kernel logs | |
273 | */ | |
274 | return loss_gilb_ell(q); | |
275 | } | |
276 | ||
277 | return false; /* not reached */ | |
278 | } | |
279 | ||
280 | ||
1da177e4 LT |
281 | /* tabledist - return a pseudo-randomly distributed value with mean mu and |
282 | * std deviation sigma. Uses table lookup to approximate the desired | |
283 | * distribution, and a uniformly-distributed pseudo-random source. | |
284 | */ | |
b407621c SH |
285 | static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma, |
286 | struct crndstate *state, | |
287 | const struct disttable *dist) | |
1da177e4 | 288 | { |
b407621c SH |
289 | psched_tdiff_t x; |
290 | long t; | |
291 | u32 rnd; | |
1da177e4 LT |
292 | |
293 | if (sigma == 0) | |
294 | return mu; | |
295 | ||
296 | rnd = get_crandom(state); | |
297 | ||
298 | /* default uniform distribution */ | |
10297b99 | 299 | if (dist == NULL) |
1da177e4 LT |
300 | return (rnd % (2*sigma)) - sigma + mu; |
301 | ||
302 | t = dist->table[rnd % dist->size]; | |
303 | x = (sigma % NETEM_DIST_SCALE) * t; | |
304 | if (x >= 0) | |
305 | x += NETEM_DIST_SCALE/2; | |
306 | else | |
307 | x -= NETEM_DIST_SCALE/2; | |
308 | ||
309 | return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu; | |
310 | } | |
311 | ||
90b41a1c | 312 | static psched_time_t packet_len_2_sched_time(unsigned int len, struct netem_sched_data *q) |
7bc0f28c | 313 | { |
90b41a1c | 314 | u64 ticks; |
fc33cc72 | 315 | |
90b41a1c HPP |
316 | len += q->packet_overhead; |
317 | ||
318 | if (q->cell_size) { | |
319 | u32 cells = reciprocal_divide(len, q->cell_size_reciprocal); | |
320 | ||
321 | if (len > cells * q->cell_size) /* extra cell needed for remainder */ | |
322 | cells++; | |
323 | len = cells * (q->cell_size + q->cell_overhead); | |
324 | } | |
325 | ||
326 | ticks = (u64)len * NSEC_PER_SEC; | |
327 | ||
328 | do_div(ticks, q->rate); | |
fc33cc72 | 329 | return PSCHED_NS2TICKS(ticks); |
7bc0f28c HPP |
330 | } |
331 | ||
50612537 ED |
332 | static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch) |
333 | { | |
334 | struct sk_buff_head *list = &sch->q; | |
335 | psched_time_t tnext = netem_skb_cb(nskb)->time_to_send; | |
336 | struct sk_buff *skb; | |
337 | ||
338 | if (likely(skb_queue_len(list) < sch->limit)) { | |
339 | skb = skb_peek_tail(list); | |
340 | /* Optimize for add at tail */ | |
341 | if (likely(!skb || tnext >= netem_skb_cb(skb)->time_to_send)) | |
342 | return qdisc_enqueue_tail(nskb, sch); | |
343 | ||
344 | skb_queue_reverse_walk(list, skb) { | |
345 | if (tnext >= netem_skb_cb(skb)->time_to_send) | |
346 | break; | |
347 | } | |
348 | ||
349 | __skb_queue_after(list, skb, nskb); | |
350 | sch->qstats.backlog += qdisc_pkt_len(nskb); | |
351 | return NET_XMIT_SUCCESS; | |
352 | } | |
353 | ||
354 | return qdisc_reshape_fail(nskb, sch); | |
355 | } | |
356 | ||
0afb51e7 SH |
357 | /* |
358 | * Insert one skb into qdisc. | |
359 | * Note: parent depends on return value to account for queue length. | |
360 | * NET_XMIT_DROP: queue length didn't change. | |
361 | * NET_XMIT_SUCCESS: one skb was queued. | |
362 | */ | |
1da177e4 LT |
363 | static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
364 | { | |
365 | struct netem_sched_data *q = qdisc_priv(sch); | |
89e1df74 GC |
366 | /* We don't fill cb now as skb_unshare() may invalidate it */ |
367 | struct netem_skb_cb *cb; | |
0afb51e7 | 368 | struct sk_buff *skb2; |
1da177e4 | 369 | int ret; |
0afb51e7 | 370 | int count = 1; |
1da177e4 | 371 | |
0afb51e7 SH |
372 | /* Random duplication */ |
373 | if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor)) | |
374 | ++count; | |
375 | ||
661b7972 | 376 | /* Drop packet? */ |
377 | if (loss_event(q)) | |
0afb51e7 SH |
378 | --count; |
379 | ||
380 | if (count == 0) { | |
1da177e4 LT |
381 | sch->qstats.drops++; |
382 | kfree_skb(skb); | |
c27f339a | 383 | return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
1da177e4 LT |
384 | } |
385 | ||
4e8a5201 DM |
386 | skb_orphan(skb); |
387 | ||
0afb51e7 SH |
388 | /* |
389 | * If we need to duplicate packet, then re-insert at top of the | |
390 | * qdisc tree, since parent queuer expects that only one | |
391 | * skb will be queued. | |
392 | */ | |
393 | if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) { | |
7698b4fc | 394 | struct Qdisc *rootq = qdisc_root(sch); |
0afb51e7 SH |
395 | u32 dupsave = q->duplicate; /* prevent duplicating a dup... */ |
396 | q->duplicate = 0; | |
397 | ||
5f86173b | 398 | qdisc_enqueue_root(skb2, rootq); |
0afb51e7 | 399 | q->duplicate = dupsave; |
1da177e4 LT |
400 | } |
401 | ||
c865e5d9 SH |
402 | /* |
403 | * Randomized packet corruption. | |
404 | * Make copy if needed since we are modifying | |
405 | * If packet is going to be hardware checksummed, then | |
406 | * do it now in software before we mangle it. | |
407 | */ | |
408 | if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) { | |
f64f9e71 JP |
409 | if (!(skb = skb_unshare(skb, GFP_ATOMIC)) || |
410 | (skb->ip_summed == CHECKSUM_PARTIAL && | |
411 | skb_checksum_help(skb))) { | |
c865e5d9 SH |
412 | sch->qstats.drops++; |
413 | return NET_XMIT_DROP; | |
414 | } | |
415 | ||
416 | skb->data[net_random() % skb_headlen(skb)] ^= 1<<(net_random() % 8); | |
417 | } | |
418 | ||
5f86173b | 419 | cb = netem_skb_cb(skb); |
cc7ec456 | 420 | if (q->gap == 0 || /* not doing reordering */ |
a42b4799 | 421 | q->counter < q->gap - 1 || /* inside last reordering gap */ |
f64f9e71 | 422 | q->reorder < get_crandom(&q->reorder_cor)) { |
0f9f32ac | 423 | psched_time_t now; |
07aaa115 SH |
424 | psched_tdiff_t delay; |
425 | ||
426 | delay = tabledist(q->latency, q->jitter, | |
427 | &q->delay_cor, q->delay_dist); | |
428 | ||
3bebcda2 | 429 | now = psched_get_time(); |
7bc0f28c HPP |
430 | |
431 | if (q->rate) { | |
50612537 | 432 | struct sk_buff_head *list = &sch->q; |
7bc0f28c | 433 | |
90b41a1c | 434 | delay += packet_len_2_sched_time(skb->len, q); |
7bc0f28c HPP |
435 | |
436 | if (!skb_queue_empty(list)) { | |
437 | /* | |
438 | * Last packet in queue is reference point (now). | |
439 | * First packet in queue is already in flight, | |
440 | * calculate this time bonus and substract | |
441 | * from delay. | |
442 | */ | |
443 | delay -= now - netem_skb_cb(skb_peek(list))->time_to_send; | |
444 | now = netem_skb_cb(skb_peek_tail(list))->time_to_send; | |
445 | } | |
446 | } | |
447 | ||
7c59e25f | 448 | cb->time_to_send = now + delay; |
1da177e4 | 449 | ++q->counter; |
50612537 | 450 | ret = tfifo_enqueue(skb, sch); |
1da177e4 | 451 | } else { |
10297b99 | 452 | /* |
0dca51d3 SH |
453 | * Do re-ordering by putting one out of N packets at the front |
454 | * of the queue. | |
455 | */ | |
3bebcda2 | 456 | cb->time_to_send = psched_get_time(); |
0dca51d3 | 457 | q->counter = 0; |
8ba25dad | 458 | |
50612537 | 459 | __skb_queue_head(&sch->q, skb); |
eb101924 HPP |
460 | sch->qstats.backlog += qdisc_pkt_len(skb); |
461 | sch->qstats.requeues++; | |
8ba25dad | 462 | ret = NET_XMIT_SUCCESS; |
1da177e4 LT |
463 | } |
464 | ||
10f6dfcf | 465 | if (ret != NET_XMIT_SUCCESS) { |
466 | if (net_xmit_drop_count(ret)) { | |
467 | sch->qstats.drops++; | |
468 | return ret; | |
469 | } | |
378a2f09 | 470 | } |
1da177e4 | 471 | |
10f6dfcf | 472 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
473 | } |
474 | ||
cc7ec456 | 475 | static unsigned int netem_drop(struct Qdisc *sch) |
1da177e4 LT |
476 | { |
477 | struct netem_sched_data *q = qdisc_priv(sch); | |
50612537 | 478 | unsigned int len; |
1da177e4 | 479 | |
50612537 ED |
480 | len = qdisc_queue_drop(sch); |
481 | if (!len && q->qdisc && q->qdisc->ops->drop) | |
482 | len = q->qdisc->ops->drop(q->qdisc); | |
483 | if (len) | |
1da177e4 | 484 | sch->qstats.drops++; |
50612537 | 485 | |
1da177e4 LT |
486 | return len; |
487 | } | |
488 | ||
1da177e4 LT |
489 | static struct sk_buff *netem_dequeue(struct Qdisc *sch) |
490 | { | |
491 | struct netem_sched_data *q = qdisc_priv(sch); | |
492 | struct sk_buff *skb; | |
493 | ||
fd245a4a | 494 | if (qdisc_is_throttled(sch)) |
11274e5a SH |
495 | return NULL; |
496 | ||
50612537 ED |
497 | tfifo_dequeue: |
498 | skb = qdisc_peek_head(sch); | |
771018e7 | 499 | if (skb) { |
5f86173b | 500 | const struct netem_skb_cb *cb = netem_skb_cb(skb); |
0f9f32ac SH |
501 | |
502 | /* if more time remaining? */ | |
50612537 | 503 | if (cb->time_to_send <= psched_get_time()) { |
cd961c2c ED |
504 | __skb_unlink(skb, &sch->q); |
505 | sch->qstats.backlog -= qdisc_pkt_len(skb); | |
03c05f0d | 506 | |
8caf1539 JP |
507 | #ifdef CONFIG_NET_CLS_ACT |
508 | /* | |
509 | * If it's at ingress let's pretend the delay is | |
510 | * from the network (tstamp will be updated). | |
511 | */ | |
512 | if (G_TC_FROM(skb->tc_verd) & AT_INGRESS) | |
513 | skb->tstamp.tv64 = 0; | |
514 | #endif | |
10f6dfcf | 515 | |
50612537 ED |
516 | if (q->qdisc) { |
517 | int err = qdisc_enqueue(skb, q->qdisc); | |
518 | ||
519 | if (unlikely(err != NET_XMIT_SUCCESS)) { | |
520 | if (net_xmit_drop_count(err)) { | |
521 | sch->qstats.drops++; | |
522 | qdisc_tree_decrease_qlen(sch, 1); | |
523 | } | |
524 | } | |
525 | goto tfifo_dequeue; | |
526 | } | |
527 | deliver: | |
10f6dfcf | 528 | qdisc_unthrottled(sch); |
529 | qdisc_bstats_update(sch, skb); | |
0f9f32ac | 530 | return skb; |
07aaa115 | 531 | } |
11274e5a | 532 | |
50612537 ED |
533 | if (q->qdisc) { |
534 | skb = q->qdisc->ops->dequeue(q->qdisc); | |
535 | if (skb) | |
536 | goto deliver; | |
537 | } | |
11274e5a | 538 | qdisc_watchdog_schedule(&q->watchdog, cb->time_to_send); |
0f9f32ac SH |
539 | } |
540 | ||
50612537 ED |
541 | if (q->qdisc) { |
542 | skb = q->qdisc->ops->dequeue(q->qdisc); | |
543 | if (skb) | |
544 | goto deliver; | |
545 | } | |
0f9f32ac | 546 | return NULL; |
1da177e4 LT |
547 | } |
548 | ||
1da177e4 LT |
549 | static void netem_reset(struct Qdisc *sch) |
550 | { | |
551 | struct netem_sched_data *q = qdisc_priv(sch); | |
552 | ||
50612537 ED |
553 | qdisc_reset_queue(sch); |
554 | if (q->qdisc) | |
555 | qdisc_reset(q->qdisc); | |
59cb5c67 | 556 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
557 | } |
558 | ||
6373a9a2 | 559 | static void dist_free(struct disttable *d) |
560 | { | |
561 | if (d) { | |
562 | if (is_vmalloc_addr(d)) | |
563 | vfree(d); | |
564 | else | |
565 | kfree(d); | |
566 | } | |
567 | } | |
568 | ||
1da177e4 LT |
569 | /* |
570 | * Distribution data is a variable size payload containing | |
571 | * signed 16 bit values. | |
572 | */ | |
1e90474c | 573 | static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr) |
1da177e4 LT |
574 | { |
575 | struct netem_sched_data *q = qdisc_priv(sch); | |
6373a9a2 | 576 | size_t n = nla_len(attr)/sizeof(__s16); |
1e90474c | 577 | const __s16 *data = nla_data(attr); |
7698b4fc | 578 | spinlock_t *root_lock; |
1da177e4 LT |
579 | struct disttable *d; |
580 | int i; | |
6373a9a2 | 581 | size_t s; |
1da177e4 | 582 | |
df173bda | 583 | if (n > NETEM_DIST_MAX) |
1da177e4 LT |
584 | return -EINVAL; |
585 | ||
6373a9a2 | 586 | s = sizeof(struct disttable) + n * sizeof(s16); |
bb52c7ac | 587 | d = kmalloc(s, GFP_KERNEL | __GFP_NOWARN); |
6373a9a2 | 588 | if (!d) |
589 | d = vmalloc(s); | |
1da177e4 LT |
590 | if (!d) |
591 | return -ENOMEM; | |
592 | ||
593 | d->size = n; | |
594 | for (i = 0; i < n; i++) | |
595 | d->table[i] = data[i]; | |
10297b99 | 596 | |
102396ae | 597 | root_lock = qdisc_root_sleeping_lock(sch); |
7698b4fc DM |
598 | |
599 | spin_lock_bh(root_lock); | |
bb52c7ac | 600 | swap(q->delay_dist, d); |
7698b4fc | 601 | spin_unlock_bh(root_lock); |
bb52c7ac ED |
602 | |
603 | dist_free(d); | |
1da177e4 LT |
604 | return 0; |
605 | } | |
606 | ||
265eb67f | 607 | static void get_correlation(struct Qdisc *sch, const struct nlattr *attr) |
1da177e4 LT |
608 | { |
609 | struct netem_sched_data *q = qdisc_priv(sch); | |
1e90474c | 610 | const struct tc_netem_corr *c = nla_data(attr); |
1da177e4 | 611 | |
1da177e4 LT |
612 | init_crandom(&q->delay_cor, c->delay_corr); |
613 | init_crandom(&q->loss_cor, c->loss_corr); | |
614 | init_crandom(&q->dup_cor, c->dup_corr); | |
1da177e4 LT |
615 | } |
616 | ||
265eb67f | 617 | static void get_reorder(struct Qdisc *sch, const struct nlattr *attr) |
0dca51d3 SH |
618 | { |
619 | struct netem_sched_data *q = qdisc_priv(sch); | |
1e90474c | 620 | const struct tc_netem_reorder *r = nla_data(attr); |
0dca51d3 | 621 | |
0dca51d3 SH |
622 | q->reorder = r->probability; |
623 | init_crandom(&q->reorder_cor, r->correlation); | |
0dca51d3 SH |
624 | } |
625 | ||
265eb67f | 626 | static void get_corrupt(struct Qdisc *sch, const struct nlattr *attr) |
c865e5d9 SH |
627 | { |
628 | struct netem_sched_data *q = qdisc_priv(sch); | |
1e90474c | 629 | const struct tc_netem_corrupt *r = nla_data(attr); |
c865e5d9 | 630 | |
c865e5d9 SH |
631 | q->corrupt = r->probability; |
632 | init_crandom(&q->corrupt_cor, r->correlation); | |
c865e5d9 SH |
633 | } |
634 | ||
7bc0f28c HPP |
635 | static void get_rate(struct Qdisc *sch, const struct nlattr *attr) |
636 | { | |
637 | struct netem_sched_data *q = qdisc_priv(sch); | |
638 | const struct tc_netem_rate *r = nla_data(attr); | |
639 | ||
640 | q->rate = r->rate; | |
90b41a1c HPP |
641 | q->packet_overhead = r->packet_overhead; |
642 | q->cell_size = r->cell_size; | |
643 | if (q->cell_size) | |
644 | q->cell_size_reciprocal = reciprocal_value(q->cell_size); | |
645 | q->cell_overhead = r->cell_overhead; | |
7bc0f28c HPP |
646 | } |
647 | ||
661b7972 | 648 | static int get_loss_clg(struct Qdisc *sch, const struct nlattr *attr) |
649 | { | |
650 | struct netem_sched_data *q = qdisc_priv(sch); | |
651 | const struct nlattr *la; | |
652 | int rem; | |
653 | ||
654 | nla_for_each_nested(la, attr, rem) { | |
655 | u16 type = nla_type(la); | |
656 | ||
657 | switch(type) { | |
658 | case NETEM_LOSS_GI: { | |
659 | const struct tc_netem_gimodel *gi = nla_data(la); | |
660 | ||
2494654d | 661 | if (nla_len(la) < sizeof(struct tc_netem_gimodel)) { |
661b7972 | 662 | pr_info("netem: incorrect gi model size\n"); |
663 | return -EINVAL; | |
664 | } | |
665 | ||
666 | q->loss_model = CLG_4_STATES; | |
667 | ||
668 | q->clg.state = 1; | |
669 | q->clg.a1 = gi->p13; | |
670 | q->clg.a2 = gi->p31; | |
671 | q->clg.a3 = gi->p32; | |
672 | q->clg.a4 = gi->p14; | |
673 | q->clg.a5 = gi->p23; | |
674 | break; | |
675 | } | |
676 | ||
677 | case NETEM_LOSS_GE: { | |
678 | const struct tc_netem_gemodel *ge = nla_data(la); | |
679 | ||
2494654d | 680 | if (nla_len(la) < sizeof(struct tc_netem_gemodel)) { |
681 | pr_info("netem: incorrect ge model size\n"); | |
661b7972 | 682 | return -EINVAL; |
683 | } | |
684 | ||
685 | q->loss_model = CLG_GILB_ELL; | |
686 | q->clg.state = 1; | |
687 | q->clg.a1 = ge->p; | |
688 | q->clg.a2 = ge->r; | |
689 | q->clg.a3 = ge->h; | |
690 | q->clg.a4 = ge->k1; | |
691 | break; | |
692 | } | |
693 | ||
694 | default: | |
695 | pr_info("netem: unknown loss type %u\n", type); | |
696 | return -EINVAL; | |
697 | } | |
698 | } | |
699 | ||
700 | return 0; | |
701 | } | |
702 | ||
27a3421e PM |
703 | static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = { |
704 | [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) }, | |
705 | [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) }, | |
706 | [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) }, | |
7bc0f28c | 707 | [TCA_NETEM_RATE] = { .len = sizeof(struct tc_netem_rate) }, |
661b7972 | 708 | [TCA_NETEM_LOSS] = { .type = NLA_NESTED }, |
27a3421e PM |
709 | }; |
710 | ||
2c10b32b TG |
711 | static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla, |
712 | const struct nla_policy *policy, int len) | |
713 | { | |
714 | int nested_len = nla_len(nla) - NLA_ALIGN(len); | |
715 | ||
661b7972 | 716 | if (nested_len < 0) { |
717 | pr_info("netem: invalid attributes len %d\n", nested_len); | |
2c10b32b | 718 | return -EINVAL; |
661b7972 | 719 | } |
720 | ||
2c10b32b TG |
721 | if (nested_len >= nla_attr_size(0)) |
722 | return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len), | |
723 | nested_len, policy); | |
661b7972 | 724 | |
2c10b32b TG |
725 | memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1)); |
726 | return 0; | |
727 | } | |
728 | ||
c865e5d9 | 729 | /* Parse netlink message to set options */ |
1e90474c | 730 | static int netem_change(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
731 | { |
732 | struct netem_sched_data *q = qdisc_priv(sch); | |
b03f4672 | 733 | struct nlattr *tb[TCA_NETEM_MAX + 1]; |
1da177e4 LT |
734 | struct tc_netem_qopt *qopt; |
735 | int ret; | |
10297b99 | 736 | |
b03f4672 | 737 | if (opt == NULL) |
1da177e4 LT |
738 | return -EINVAL; |
739 | ||
2c10b32b TG |
740 | qopt = nla_data(opt); |
741 | ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt)); | |
b03f4672 PM |
742 | if (ret < 0) |
743 | return ret; | |
744 | ||
50612537 | 745 | sch->limit = qopt->limit; |
10297b99 | 746 | |
1da177e4 LT |
747 | q->latency = qopt->latency; |
748 | q->jitter = qopt->jitter; | |
749 | q->limit = qopt->limit; | |
750 | q->gap = qopt->gap; | |
0dca51d3 | 751 | q->counter = 0; |
1da177e4 LT |
752 | q->loss = qopt->loss; |
753 | q->duplicate = qopt->duplicate; | |
754 | ||
bb2f8cc0 SH |
755 | /* for compatibility with earlier versions. |
756 | * if gap is set, need to assume 100% probability | |
0dca51d3 | 757 | */ |
a362e0a7 SH |
758 | if (q->gap) |
759 | q->reorder = ~0; | |
0dca51d3 | 760 | |
265eb67f SH |
761 | if (tb[TCA_NETEM_CORR]) |
762 | get_correlation(sch, tb[TCA_NETEM_CORR]); | |
1da177e4 | 763 | |
b03f4672 PM |
764 | if (tb[TCA_NETEM_DELAY_DIST]) { |
765 | ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]); | |
766 | if (ret) | |
767 | return ret; | |
768 | } | |
c865e5d9 | 769 | |
265eb67f SH |
770 | if (tb[TCA_NETEM_REORDER]) |
771 | get_reorder(sch, tb[TCA_NETEM_REORDER]); | |
1da177e4 | 772 | |
265eb67f SH |
773 | if (tb[TCA_NETEM_CORRUPT]) |
774 | get_corrupt(sch, tb[TCA_NETEM_CORRUPT]); | |
1da177e4 | 775 | |
7bc0f28c HPP |
776 | if (tb[TCA_NETEM_RATE]) |
777 | get_rate(sch, tb[TCA_NETEM_RATE]); | |
778 | ||
661b7972 | 779 | q->loss_model = CLG_RANDOM; |
780 | if (tb[TCA_NETEM_LOSS]) | |
781 | ret = get_loss_clg(sch, tb[TCA_NETEM_LOSS]); | |
782 | ||
783 | return ret; | |
1da177e4 LT |
784 | } |
785 | ||
1e90474c | 786 | static int netem_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
787 | { |
788 | struct netem_sched_data *q = qdisc_priv(sch); | |
789 | int ret; | |
790 | ||
791 | if (!opt) | |
792 | return -EINVAL; | |
793 | ||
59cb5c67 | 794 | qdisc_watchdog_init(&q->watchdog, sch); |
1da177e4 | 795 | |
661b7972 | 796 | q->loss_model = CLG_RANDOM; |
1da177e4 | 797 | ret = netem_change(sch, opt); |
50612537 | 798 | if (ret) |
250a65f7 | 799 | pr_info("netem: change failed\n"); |
1da177e4 LT |
800 | return ret; |
801 | } | |
802 | ||
803 | static void netem_destroy(struct Qdisc *sch) | |
804 | { | |
805 | struct netem_sched_data *q = qdisc_priv(sch); | |
806 | ||
59cb5c67 | 807 | qdisc_watchdog_cancel(&q->watchdog); |
50612537 ED |
808 | if (q->qdisc) |
809 | qdisc_destroy(q->qdisc); | |
6373a9a2 | 810 | dist_free(q->delay_dist); |
1da177e4 LT |
811 | } |
812 | ||
661b7972 | 813 | static int dump_loss_model(const struct netem_sched_data *q, |
814 | struct sk_buff *skb) | |
815 | { | |
816 | struct nlattr *nest; | |
817 | ||
818 | nest = nla_nest_start(skb, TCA_NETEM_LOSS); | |
819 | if (nest == NULL) | |
820 | goto nla_put_failure; | |
821 | ||
822 | switch (q->loss_model) { | |
823 | case CLG_RANDOM: | |
824 | /* legacy loss model */ | |
825 | nla_nest_cancel(skb, nest); | |
826 | return 0; /* no data */ | |
827 | ||
828 | case CLG_4_STATES: { | |
829 | struct tc_netem_gimodel gi = { | |
830 | .p13 = q->clg.a1, | |
831 | .p31 = q->clg.a2, | |
832 | .p32 = q->clg.a3, | |
833 | .p14 = q->clg.a4, | |
834 | .p23 = q->clg.a5, | |
835 | }; | |
836 | ||
837 | NLA_PUT(skb, NETEM_LOSS_GI, sizeof(gi), &gi); | |
838 | break; | |
839 | } | |
840 | case CLG_GILB_ELL: { | |
841 | struct tc_netem_gemodel ge = { | |
842 | .p = q->clg.a1, | |
843 | .r = q->clg.a2, | |
844 | .h = q->clg.a3, | |
845 | .k1 = q->clg.a4, | |
846 | }; | |
847 | ||
848 | NLA_PUT(skb, NETEM_LOSS_GE, sizeof(ge), &ge); | |
849 | break; | |
850 | } | |
851 | } | |
852 | ||
853 | nla_nest_end(skb, nest); | |
854 | return 0; | |
855 | ||
856 | nla_put_failure: | |
857 | nla_nest_cancel(skb, nest); | |
858 | return -1; | |
859 | } | |
860 | ||
1da177e4 LT |
861 | static int netem_dump(struct Qdisc *sch, struct sk_buff *skb) |
862 | { | |
863 | const struct netem_sched_data *q = qdisc_priv(sch); | |
861d7f74 | 864 | struct nlattr *nla = (struct nlattr *) skb_tail_pointer(skb); |
1da177e4 LT |
865 | struct tc_netem_qopt qopt; |
866 | struct tc_netem_corr cor; | |
0dca51d3 | 867 | struct tc_netem_reorder reorder; |
c865e5d9 | 868 | struct tc_netem_corrupt corrupt; |
7bc0f28c | 869 | struct tc_netem_rate rate; |
1da177e4 LT |
870 | |
871 | qopt.latency = q->latency; | |
872 | qopt.jitter = q->jitter; | |
873 | qopt.limit = q->limit; | |
874 | qopt.loss = q->loss; | |
875 | qopt.gap = q->gap; | |
876 | qopt.duplicate = q->duplicate; | |
1e90474c | 877 | NLA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt); |
1da177e4 LT |
878 | |
879 | cor.delay_corr = q->delay_cor.rho; | |
880 | cor.loss_corr = q->loss_cor.rho; | |
881 | cor.dup_corr = q->dup_cor.rho; | |
1e90474c | 882 | NLA_PUT(skb, TCA_NETEM_CORR, sizeof(cor), &cor); |
0dca51d3 SH |
883 | |
884 | reorder.probability = q->reorder; | |
885 | reorder.correlation = q->reorder_cor.rho; | |
1e90474c | 886 | NLA_PUT(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder); |
0dca51d3 | 887 | |
c865e5d9 SH |
888 | corrupt.probability = q->corrupt; |
889 | corrupt.correlation = q->corrupt_cor.rho; | |
1e90474c | 890 | NLA_PUT(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt); |
c865e5d9 | 891 | |
7bc0f28c | 892 | rate.rate = q->rate; |
90b41a1c HPP |
893 | rate.packet_overhead = q->packet_overhead; |
894 | rate.cell_size = q->cell_size; | |
895 | rate.cell_overhead = q->cell_overhead; | |
7bc0f28c HPP |
896 | NLA_PUT(skb, TCA_NETEM_RATE, sizeof(rate), &rate); |
897 | ||
661b7972 | 898 | if (dump_loss_model(q, skb) != 0) |
899 | goto nla_put_failure; | |
900 | ||
861d7f74 | 901 | return nla_nest_end(skb, nla); |
1da177e4 | 902 | |
1e90474c | 903 | nla_put_failure: |
861d7f74 | 904 | nlmsg_trim(skb, nla); |
1da177e4 LT |
905 | return -1; |
906 | } | |
907 | ||
10f6dfcf | 908 | static int netem_dump_class(struct Qdisc *sch, unsigned long cl, |
909 | struct sk_buff *skb, struct tcmsg *tcm) | |
910 | { | |
911 | struct netem_sched_data *q = qdisc_priv(sch); | |
912 | ||
50612537 | 913 | if (cl != 1 || !q->qdisc) /* only one class */ |
10f6dfcf | 914 | return -ENOENT; |
915 | ||
916 | tcm->tcm_handle |= TC_H_MIN(1); | |
917 | tcm->tcm_info = q->qdisc->handle; | |
918 | ||
919 | return 0; | |
920 | } | |
921 | ||
922 | static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
923 | struct Qdisc **old) | |
924 | { | |
925 | struct netem_sched_data *q = qdisc_priv(sch); | |
926 | ||
10f6dfcf | 927 | sch_tree_lock(sch); |
928 | *old = q->qdisc; | |
929 | q->qdisc = new; | |
50612537 ED |
930 | if (*old) { |
931 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); | |
932 | qdisc_reset(*old); | |
933 | } | |
10f6dfcf | 934 | sch_tree_unlock(sch); |
935 | ||
936 | return 0; | |
937 | } | |
938 | ||
939 | static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg) | |
940 | { | |
941 | struct netem_sched_data *q = qdisc_priv(sch); | |
942 | return q->qdisc; | |
943 | } | |
944 | ||
945 | static unsigned long netem_get(struct Qdisc *sch, u32 classid) | |
946 | { | |
947 | return 1; | |
948 | } | |
949 | ||
950 | static void netem_put(struct Qdisc *sch, unsigned long arg) | |
951 | { | |
952 | } | |
953 | ||
954 | static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker) | |
955 | { | |
956 | if (!walker->stop) { | |
957 | if (walker->count >= walker->skip) | |
958 | if (walker->fn(sch, 1, walker) < 0) { | |
959 | walker->stop = 1; | |
960 | return; | |
961 | } | |
962 | walker->count++; | |
963 | } | |
964 | } | |
965 | ||
966 | static const struct Qdisc_class_ops netem_class_ops = { | |
967 | .graft = netem_graft, | |
968 | .leaf = netem_leaf, | |
969 | .get = netem_get, | |
970 | .put = netem_put, | |
971 | .walk = netem_walk, | |
972 | .dump = netem_dump_class, | |
973 | }; | |
974 | ||
20fea08b | 975 | static struct Qdisc_ops netem_qdisc_ops __read_mostly = { |
1da177e4 | 976 | .id = "netem", |
10f6dfcf | 977 | .cl_ops = &netem_class_ops, |
1da177e4 LT |
978 | .priv_size = sizeof(struct netem_sched_data), |
979 | .enqueue = netem_enqueue, | |
980 | .dequeue = netem_dequeue, | |
77be155c | 981 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
982 | .drop = netem_drop, |
983 | .init = netem_init, | |
984 | .reset = netem_reset, | |
985 | .destroy = netem_destroy, | |
986 | .change = netem_change, | |
987 | .dump = netem_dump, | |
988 | .owner = THIS_MODULE, | |
989 | }; | |
990 | ||
991 | ||
992 | static int __init netem_module_init(void) | |
993 | { | |
eb229c4c | 994 | pr_info("netem: version " VERSION "\n"); |
1da177e4 LT |
995 | return register_qdisc(&netem_qdisc_ops); |
996 | } | |
997 | static void __exit netem_module_exit(void) | |
998 | { | |
999 | unregister_qdisc(&netem_qdisc_ops); | |
1000 | } | |
1001 | module_init(netem_module_init) | |
1002 | module_exit(netem_module_exit) | |
1003 | MODULE_LICENSE("GPL"); |