Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * net/sched/sch_netem.c Network emulator | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
798b6b19 | 7 | * 2 of the License. |
1da177e4 LT |
8 | * |
9 | * Many of the algorithms and ideas for this came from | |
10297b99 | 10 | * NIST Net which is not copyrighted. |
1da177e4 LT |
11 | * |
12 | * Authors: Stephen Hemminger <shemminger@osdl.org> | |
13 | * Catalin(ux aka Dino) BOIE <catab at umbrella dot ro> | |
14 | */ | |
15 | ||
1da177e4 | 16 | #include <linux/module.h> |
1da177e4 LT |
17 | #include <linux/types.h> |
18 | #include <linux/kernel.h> | |
19 | #include <linux/errno.h> | |
1da177e4 LT |
20 | #include <linux/skbuff.h> |
21 | #include <linux/rtnetlink.h> | |
22 | ||
dc5fc579 | 23 | #include <net/netlink.h> |
1da177e4 LT |
24 | #include <net/pkt_sched.h> |
25 | ||
c865e5d9 | 26 | #define VERSION "1.2" |
eb229c4c | 27 | |
1da177e4 LT |
28 | /* Network Emulation Queuing algorithm. |
29 | ==================================== | |
30 | ||
31 | Sources: [1] Mark Carson, Darrin Santay, "NIST Net - A Linux-based | |
32 | Network Emulation Tool | |
33 | [2] Luigi Rizzo, DummyNet for FreeBSD | |
34 | ||
35 | ---------------------------------------------------------------- | |
36 | ||
37 | This started out as a simple way to delay outgoing packets to | |
38 | test TCP but has grown to include most of the functionality | |
39 | of a full blown network emulator like NISTnet. It can delay | |
40 | packets and add random jitter (and correlation). The random | |
41 | distribution can be loaded from a table as well to provide | |
42 | normal, Pareto, or experimental curves. Packet loss, | |
43 | duplication, and reordering can also be emulated. | |
44 | ||
45 | This qdisc does not do classification that can be handled in | |
46 | layering other disciplines. It does not need to do bandwidth | |
47 | control either since that can be handled by using token | |
48 | bucket or other rate control. | |
49 | ||
50 | The simulator is limited by the Linux timer resolution | |
51 | and will create packet bursts on the HZ boundary (1ms). | |
52 | */ | |
53 | ||
54 | struct netem_sched_data { | |
55 | struct Qdisc *qdisc; | |
59cb5c67 | 56 | struct qdisc_watchdog watchdog; |
1da177e4 | 57 | |
b407621c SH |
58 | psched_tdiff_t latency; |
59 | psched_tdiff_t jitter; | |
60 | ||
1da177e4 LT |
61 | u32 loss; |
62 | u32 limit; | |
63 | u32 counter; | |
64 | u32 gap; | |
1da177e4 | 65 | u32 duplicate; |
0dca51d3 | 66 | u32 reorder; |
c865e5d9 | 67 | u32 corrupt; |
1da177e4 LT |
68 | |
69 | struct crndstate { | |
b407621c SH |
70 | u32 last; |
71 | u32 rho; | |
c865e5d9 | 72 | } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor; |
1da177e4 LT |
73 | |
74 | struct disttable { | |
75 | u32 size; | |
76 | s16 table[0]; | |
77 | } *delay_dist; | |
78 | }; | |
79 | ||
80 | /* Time stamp put into socket buffer control block */ | |
81 | struct netem_skb_cb { | |
82 | psched_time_t time_to_send; | |
83 | }; | |
84 | ||
5f86173b JK |
85 | static inline struct netem_skb_cb *netem_skb_cb(struct sk_buff *skb) |
86 | { | |
175f9c1b JK |
87 | BUILD_BUG_ON(sizeof(skb->cb) < |
88 | sizeof(struct qdisc_skb_cb) + sizeof(struct netem_skb_cb)); | |
89 | return (struct netem_skb_cb *)qdisc_skb_cb(skb)->data; | |
5f86173b JK |
90 | } |
91 | ||
1da177e4 LT |
92 | /* init_crandom - initialize correlated random number generator |
93 | * Use entropy source for initial seed. | |
94 | */ | |
95 | static void init_crandom(struct crndstate *state, unsigned long rho) | |
96 | { | |
97 | state->rho = rho; | |
98 | state->last = net_random(); | |
99 | } | |
100 | ||
101 | /* get_crandom - correlated random number generator | |
102 | * Next number depends on last value. | |
103 | * rho is scaled to avoid floating point. | |
104 | */ | |
b407621c | 105 | static u32 get_crandom(struct crndstate *state) |
1da177e4 LT |
106 | { |
107 | u64 value, rho; | |
108 | unsigned long answer; | |
109 | ||
bb2f8cc0 | 110 | if (state->rho == 0) /* no correlation */ |
1da177e4 LT |
111 | return net_random(); |
112 | ||
113 | value = net_random(); | |
114 | rho = (u64)state->rho + 1; | |
115 | answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32; | |
116 | state->last = answer; | |
117 | return answer; | |
118 | } | |
119 | ||
120 | /* tabledist - return a pseudo-randomly distributed value with mean mu and | |
121 | * std deviation sigma. Uses table lookup to approximate the desired | |
122 | * distribution, and a uniformly-distributed pseudo-random source. | |
123 | */ | |
b407621c SH |
124 | static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma, |
125 | struct crndstate *state, | |
126 | const struct disttable *dist) | |
1da177e4 | 127 | { |
b407621c SH |
128 | psched_tdiff_t x; |
129 | long t; | |
130 | u32 rnd; | |
1da177e4 LT |
131 | |
132 | if (sigma == 0) | |
133 | return mu; | |
134 | ||
135 | rnd = get_crandom(state); | |
136 | ||
137 | /* default uniform distribution */ | |
10297b99 | 138 | if (dist == NULL) |
1da177e4 LT |
139 | return (rnd % (2*sigma)) - sigma + mu; |
140 | ||
141 | t = dist->table[rnd % dist->size]; | |
142 | x = (sigma % NETEM_DIST_SCALE) * t; | |
143 | if (x >= 0) | |
144 | x += NETEM_DIST_SCALE/2; | |
145 | else | |
146 | x -= NETEM_DIST_SCALE/2; | |
147 | ||
148 | return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu; | |
149 | } | |
150 | ||
0afb51e7 SH |
151 | /* |
152 | * Insert one skb into qdisc. | |
153 | * Note: parent depends on return value to account for queue length. | |
154 | * NET_XMIT_DROP: queue length didn't change. | |
155 | * NET_XMIT_SUCCESS: one skb was queued. | |
156 | */ | |
1da177e4 LT |
157 | static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
158 | { | |
159 | struct netem_sched_data *q = qdisc_priv(sch); | |
89e1df74 GC |
160 | /* We don't fill cb now as skb_unshare() may invalidate it */ |
161 | struct netem_skb_cb *cb; | |
0afb51e7 | 162 | struct sk_buff *skb2; |
1da177e4 | 163 | int ret; |
0afb51e7 | 164 | int count = 1; |
1da177e4 | 165 | |
771018e7 | 166 | pr_debug("netem_enqueue skb=%p\n", skb); |
1da177e4 | 167 | |
0afb51e7 SH |
168 | /* Random duplication */ |
169 | if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor)) | |
170 | ++count; | |
171 | ||
1da177e4 | 172 | /* Random packet drop 0 => none, ~0 => all */ |
0afb51e7 SH |
173 | if (q->loss && q->loss >= get_crandom(&q->loss_cor)) |
174 | --count; | |
175 | ||
176 | if (count == 0) { | |
1da177e4 LT |
177 | sch->qstats.drops++; |
178 | kfree_skb(skb); | |
89bbb0a3 | 179 | return NET_XMIT_BYPASS; |
1da177e4 LT |
180 | } |
181 | ||
4e8a5201 DM |
182 | skb_orphan(skb); |
183 | ||
0afb51e7 SH |
184 | /* |
185 | * If we need to duplicate packet, then re-insert at top of the | |
186 | * qdisc tree, since parent queuer expects that only one | |
187 | * skb will be queued. | |
188 | */ | |
189 | if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) { | |
7698b4fc | 190 | struct Qdisc *rootq = qdisc_root(sch); |
0afb51e7 SH |
191 | u32 dupsave = q->duplicate; /* prevent duplicating a dup... */ |
192 | q->duplicate = 0; | |
193 | ||
5f86173b | 194 | qdisc_enqueue_root(skb2, rootq); |
0afb51e7 | 195 | q->duplicate = dupsave; |
1da177e4 LT |
196 | } |
197 | ||
c865e5d9 SH |
198 | /* |
199 | * Randomized packet corruption. | |
200 | * Make copy if needed since we are modifying | |
201 | * If packet is going to be hardware checksummed, then | |
202 | * do it now in software before we mangle it. | |
203 | */ | |
204 | if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) { | |
205 | if (!(skb = skb_unshare(skb, GFP_ATOMIC)) | |
84fa7933 PM |
206 | || (skb->ip_summed == CHECKSUM_PARTIAL |
207 | && skb_checksum_help(skb))) { | |
c865e5d9 SH |
208 | sch->qstats.drops++; |
209 | return NET_XMIT_DROP; | |
210 | } | |
211 | ||
212 | skb->data[net_random() % skb_headlen(skb)] ^= 1<<(net_random() % 8); | |
213 | } | |
214 | ||
5f86173b | 215 | cb = netem_skb_cb(skb); |
0dca51d3 SH |
216 | if (q->gap == 0 /* not doing reordering */ |
217 | || q->counter < q->gap /* inside last reordering gap */ | |
218 | || q->reorder < get_crandom(&q->reorder_cor)) { | |
0f9f32ac | 219 | psched_time_t now; |
07aaa115 SH |
220 | psched_tdiff_t delay; |
221 | ||
222 | delay = tabledist(q->latency, q->jitter, | |
223 | &q->delay_cor, q->delay_dist); | |
224 | ||
3bebcda2 | 225 | now = psched_get_time(); |
7c59e25f | 226 | cb->time_to_send = now + delay; |
1da177e4 | 227 | ++q->counter; |
5f86173b | 228 | ret = qdisc_enqueue(skb, q->qdisc); |
1da177e4 | 229 | } else { |
10297b99 | 230 | /* |
0dca51d3 SH |
231 | * Do re-ordering by putting one out of N packets at the front |
232 | * of the queue. | |
233 | */ | |
3bebcda2 | 234 | cb->time_to_send = psched_get_time(); |
0dca51d3 | 235 | q->counter = 0; |
0f9f32ac | 236 | ret = q->qdisc->ops->requeue(skb, q->qdisc); |
1da177e4 LT |
237 | } |
238 | ||
239 | if (likely(ret == NET_XMIT_SUCCESS)) { | |
240 | sch->q.qlen++; | |
0abf77e5 | 241 | sch->bstats.bytes += qdisc_pkt_len(skb); |
1da177e4 | 242 | sch->bstats.packets++; |
378a2f09 | 243 | } else if (net_xmit_drop_count(ret)) { |
1da177e4 | 244 | sch->qstats.drops++; |
378a2f09 | 245 | } |
1da177e4 | 246 | |
d5d75cd6 | 247 | pr_debug("netem: enqueue ret %d\n", ret); |
1da177e4 LT |
248 | return ret; |
249 | } | |
250 | ||
251 | /* Requeue packets but don't change time stamp */ | |
252 | static int netem_requeue(struct sk_buff *skb, struct Qdisc *sch) | |
253 | { | |
254 | struct netem_sched_data *q = qdisc_priv(sch); | |
255 | int ret; | |
256 | ||
257 | if ((ret = q->qdisc->ops->requeue(skb, q->qdisc)) == 0) { | |
258 | sch->q.qlen++; | |
259 | sch->qstats.requeues++; | |
260 | } | |
261 | ||
262 | return ret; | |
263 | } | |
264 | ||
265 | static unsigned int netem_drop(struct Qdisc* sch) | |
266 | { | |
267 | struct netem_sched_data *q = qdisc_priv(sch); | |
6d037a26 | 268 | unsigned int len = 0; |
1da177e4 | 269 | |
6d037a26 | 270 | if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) { |
1da177e4 LT |
271 | sch->q.qlen--; |
272 | sch->qstats.drops++; | |
273 | } | |
274 | return len; | |
275 | } | |
276 | ||
1da177e4 LT |
277 | static struct sk_buff *netem_dequeue(struct Qdisc *sch) |
278 | { | |
279 | struct netem_sched_data *q = qdisc_priv(sch); | |
280 | struct sk_buff *skb; | |
281 | ||
11274e5a SH |
282 | smp_mb(); |
283 | if (sch->flags & TCQ_F_THROTTLED) | |
284 | return NULL; | |
285 | ||
1da177e4 | 286 | skb = q->qdisc->dequeue(q->qdisc); |
771018e7 | 287 | if (skb) { |
5f86173b | 288 | const struct netem_skb_cb *cb = netem_skb_cb(skb); |
3bebcda2 | 289 | psched_time_t now = psched_get_time(); |
0f9f32ac SH |
290 | |
291 | /* if more time remaining? */ | |
104e0878 | 292 | if (cb->time_to_send <= now) { |
0f9f32ac SH |
293 | pr_debug("netem_dequeue: return skb=%p\n", skb); |
294 | sch->q.qlen--; | |
0f9f32ac | 295 | return skb; |
07aaa115 | 296 | } |
11274e5a SH |
297 | |
298 | if (unlikely(q->qdisc->ops->requeue(skb, q->qdisc) != NET_XMIT_SUCCESS)) { | |
299 | qdisc_tree_decrease_qlen(q->qdisc, 1); | |
300 | sch->qstats.drops++; | |
301 | printk(KERN_ERR "netem: %s could not requeue\n", | |
302 | q->qdisc->ops->id); | |
303 | } | |
304 | ||
305 | qdisc_watchdog_schedule(&q->watchdog, cb->time_to_send); | |
0f9f32ac SH |
306 | } |
307 | ||
308 | return NULL; | |
1da177e4 LT |
309 | } |
310 | ||
1da177e4 LT |
311 | static void netem_reset(struct Qdisc *sch) |
312 | { | |
313 | struct netem_sched_data *q = qdisc_priv(sch); | |
314 | ||
315 | qdisc_reset(q->qdisc); | |
1da177e4 | 316 | sch->q.qlen = 0; |
59cb5c67 | 317 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
318 | } |
319 | ||
1da177e4 LT |
320 | /* |
321 | * Distribution data is a variable size payload containing | |
322 | * signed 16 bit values. | |
323 | */ | |
1e90474c | 324 | static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr) |
1da177e4 LT |
325 | { |
326 | struct netem_sched_data *q = qdisc_priv(sch); | |
1e90474c PM |
327 | unsigned long n = nla_len(attr)/sizeof(__s16); |
328 | const __s16 *data = nla_data(attr); | |
7698b4fc | 329 | spinlock_t *root_lock; |
1da177e4 LT |
330 | struct disttable *d; |
331 | int i; | |
332 | ||
333 | if (n > 65536) | |
334 | return -EINVAL; | |
335 | ||
336 | d = kmalloc(sizeof(*d) + n*sizeof(d->table[0]), GFP_KERNEL); | |
337 | if (!d) | |
338 | return -ENOMEM; | |
339 | ||
340 | d->size = n; | |
341 | for (i = 0; i < n; i++) | |
342 | d->table[i] = data[i]; | |
10297b99 | 343 | |
7698b4fc DM |
344 | root_lock = qdisc_root_lock(sch); |
345 | ||
346 | spin_lock_bh(root_lock); | |
1da177e4 | 347 | d = xchg(&q->delay_dist, d); |
7698b4fc | 348 | spin_unlock_bh(root_lock); |
1da177e4 LT |
349 | |
350 | kfree(d); | |
351 | return 0; | |
352 | } | |
353 | ||
1e90474c | 354 | static int get_correlation(struct Qdisc *sch, const struct nlattr *attr) |
1da177e4 LT |
355 | { |
356 | struct netem_sched_data *q = qdisc_priv(sch); | |
1e90474c | 357 | const struct tc_netem_corr *c = nla_data(attr); |
1da177e4 | 358 | |
1da177e4 LT |
359 | init_crandom(&q->delay_cor, c->delay_corr); |
360 | init_crandom(&q->loss_cor, c->loss_corr); | |
361 | init_crandom(&q->dup_cor, c->dup_corr); | |
362 | return 0; | |
363 | } | |
364 | ||
1e90474c | 365 | static int get_reorder(struct Qdisc *sch, const struct nlattr *attr) |
0dca51d3 SH |
366 | { |
367 | struct netem_sched_data *q = qdisc_priv(sch); | |
1e90474c | 368 | const struct tc_netem_reorder *r = nla_data(attr); |
0dca51d3 | 369 | |
0dca51d3 SH |
370 | q->reorder = r->probability; |
371 | init_crandom(&q->reorder_cor, r->correlation); | |
372 | return 0; | |
373 | } | |
374 | ||
1e90474c | 375 | static int get_corrupt(struct Qdisc *sch, const struct nlattr *attr) |
c865e5d9 SH |
376 | { |
377 | struct netem_sched_data *q = qdisc_priv(sch); | |
1e90474c | 378 | const struct tc_netem_corrupt *r = nla_data(attr); |
c865e5d9 | 379 | |
c865e5d9 SH |
380 | q->corrupt = r->probability; |
381 | init_crandom(&q->corrupt_cor, r->correlation); | |
382 | return 0; | |
383 | } | |
384 | ||
27a3421e PM |
385 | static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = { |
386 | [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) }, | |
387 | [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) }, | |
388 | [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) }, | |
389 | }; | |
390 | ||
c865e5d9 | 391 | /* Parse netlink message to set options */ |
1e90474c | 392 | static int netem_change(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
393 | { |
394 | struct netem_sched_data *q = qdisc_priv(sch); | |
b03f4672 | 395 | struct nlattr *tb[TCA_NETEM_MAX + 1]; |
1da177e4 LT |
396 | struct tc_netem_qopt *qopt; |
397 | int ret; | |
10297b99 | 398 | |
b03f4672 | 399 | if (opt == NULL) |
1da177e4 LT |
400 | return -EINVAL; |
401 | ||
27a3421e PM |
402 | ret = nla_parse_nested_compat(tb, TCA_NETEM_MAX, opt, netem_policy, |
403 | qopt, sizeof(*qopt)); | |
b03f4672 PM |
404 | if (ret < 0) |
405 | return ret; | |
406 | ||
fb0305ce | 407 | ret = fifo_set_limit(q->qdisc, qopt->limit); |
1da177e4 LT |
408 | if (ret) { |
409 | pr_debug("netem: can't set fifo limit\n"); | |
410 | return ret; | |
411 | } | |
10297b99 | 412 | |
1da177e4 LT |
413 | q->latency = qopt->latency; |
414 | q->jitter = qopt->jitter; | |
415 | q->limit = qopt->limit; | |
416 | q->gap = qopt->gap; | |
0dca51d3 | 417 | q->counter = 0; |
1da177e4 LT |
418 | q->loss = qopt->loss; |
419 | q->duplicate = qopt->duplicate; | |
420 | ||
bb2f8cc0 SH |
421 | /* for compatibility with earlier versions. |
422 | * if gap is set, need to assume 100% probability | |
0dca51d3 | 423 | */ |
a362e0a7 SH |
424 | if (q->gap) |
425 | q->reorder = ~0; | |
0dca51d3 | 426 | |
b03f4672 PM |
427 | if (tb[TCA_NETEM_CORR]) { |
428 | ret = get_correlation(sch, tb[TCA_NETEM_CORR]); | |
429 | if (ret) | |
430 | return ret; | |
431 | } | |
1da177e4 | 432 | |
b03f4672 PM |
433 | if (tb[TCA_NETEM_DELAY_DIST]) { |
434 | ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]); | |
435 | if (ret) | |
436 | return ret; | |
437 | } | |
c865e5d9 | 438 | |
b03f4672 PM |
439 | if (tb[TCA_NETEM_REORDER]) { |
440 | ret = get_reorder(sch, tb[TCA_NETEM_REORDER]); | |
441 | if (ret) | |
442 | return ret; | |
443 | } | |
1da177e4 | 444 | |
b03f4672 PM |
445 | if (tb[TCA_NETEM_CORRUPT]) { |
446 | ret = get_corrupt(sch, tb[TCA_NETEM_CORRUPT]); | |
447 | if (ret) | |
448 | return ret; | |
c865e5d9 | 449 | } |
1da177e4 LT |
450 | |
451 | return 0; | |
452 | } | |
453 | ||
300ce174 SH |
454 | /* |
455 | * Special case version of FIFO queue for use by netem. | |
456 | * It queues in order based on timestamps in skb's | |
457 | */ | |
458 | struct fifo_sched_data { | |
459 | u32 limit; | |
075aa573 | 460 | psched_time_t oldest; |
300ce174 SH |
461 | }; |
462 | ||
463 | static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch) | |
464 | { | |
465 | struct fifo_sched_data *q = qdisc_priv(sch); | |
466 | struct sk_buff_head *list = &sch->q; | |
5f86173b | 467 | psched_time_t tnext = netem_skb_cb(nskb)->time_to_send; |
300ce174 SH |
468 | struct sk_buff *skb; |
469 | ||
470 | if (likely(skb_queue_len(list) < q->limit)) { | |
075aa573 | 471 | /* Optimize for add at tail */ |
104e0878 | 472 | if (likely(skb_queue_empty(list) || tnext >= q->oldest)) { |
075aa573 SH |
473 | q->oldest = tnext; |
474 | return qdisc_enqueue_tail(nskb, sch); | |
475 | } | |
476 | ||
300ce174 | 477 | skb_queue_reverse_walk(list, skb) { |
5f86173b | 478 | const struct netem_skb_cb *cb = netem_skb_cb(skb); |
300ce174 | 479 | |
104e0878 | 480 | if (tnext >= cb->time_to_send) |
300ce174 SH |
481 | break; |
482 | } | |
483 | ||
484 | __skb_queue_after(list, skb, nskb); | |
485 | ||
0abf77e5 JK |
486 | sch->qstats.backlog += qdisc_pkt_len(nskb); |
487 | sch->bstats.bytes += qdisc_pkt_len(nskb); | |
300ce174 SH |
488 | sch->bstats.packets++; |
489 | ||
490 | return NET_XMIT_SUCCESS; | |
491 | } | |
492 | ||
075aa573 | 493 | return qdisc_reshape_fail(nskb, sch); |
300ce174 SH |
494 | } |
495 | ||
1e90474c | 496 | static int tfifo_init(struct Qdisc *sch, struct nlattr *opt) |
300ce174 SH |
497 | { |
498 | struct fifo_sched_data *q = qdisc_priv(sch); | |
499 | ||
500 | if (opt) { | |
1e90474c PM |
501 | struct tc_fifo_qopt *ctl = nla_data(opt); |
502 | if (nla_len(opt) < sizeof(*ctl)) | |
300ce174 SH |
503 | return -EINVAL; |
504 | ||
505 | q->limit = ctl->limit; | |
506 | } else | |
5ce2d488 | 507 | q->limit = max_t(u32, qdisc_dev(sch)->tx_queue_len, 1); |
300ce174 | 508 | |
a084980d | 509 | q->oldest = PSCHED_PASTPERFECT; |
300ce174 SH |
510 | return 0; |
511 | } | |
512 | ||
513 | static int tfifo_dump(struct Qdisc *sch, struct sk_buff *skb) | |
514 | { | |
515 | struct fifo_sched_data *q = qdisc_priv(sch); | |
516 | struct tc_fifo_qopt opt = { .limit = q->limit }; | |
517 | ||
1e90474c | 518 | NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); |
300ce174 SH |
519 | return skb->len; |
520 | ||
1e90474c | 521 | nla_put_failure: |
300ce174 SH |
522 | return -1; |
523 | } | |
524 | ||
20fea08b | 525 | static struct Qdisc_ops tfifo_qdisc_ops __read_mostly = { |
300ce174 SH |
526 | .id = "tfifo", |
527 | .priv_size = sizeof(struct fifo_sched_data), | |
528 | .enqueue = tfifo_enqueue, | |
529 | .dequeue = qdisc_dequeue_head, | |
530 | .requeue = qdisc_requeue, | |
531 | .drop = qdisc_queue_drop, | |
532 | .init = tfifo_init, | |
533 | .reset = qdisc_reset_queue, | |
534 | .change = tfifo_init, | |
535 | .dump = tfifo_dump, | |
536 | }; | |
537 | ||
1e90474c | 538 | static int netem_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
539 | { |
540 | struct netem_sched_data *q = qdisc_priv(sch); | |
541 | int ret; | |
542 | ||
543 | if (!opt) | |
544 | return -EINVAL; | |
545 | ||
59cb5c67 | 546 | qdisc_watchdog_init(&q->watchdog, sch); |
1da177e4 | 547 | |
5ce2d488 | 548 | q->qdisc = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue, |
bb949fbd | 549 | &tfifo_qdisc_ops, |
9f9afec4 | 550 | TC_H_MAKE(sch->handle, 1)); |
1da177e4 LT |
551 | if (!q->qdisc) { |
552 | pr_debug("netem: qdisc create failed\n"); | |
553 | return -ENOMEM; | |
554 | } | |
555 | ||
556 | ret = netem_change(sch, opt); | |
557 | if (ret) { | |
558 | pr_debug("netem: change failed\n"); | |
559 | qdisc_destroy(q->qdisc); | |
560 | } | |
561 | return ret; | |
562 | } | |
563 | ||
564 | static void netem_destroy(struct Qdisc *sch) | |
565 | { | |
566 | struct netem_sched_data *q = qdisc_priv(sch); | |
567 | ||
59cb5c67 | 568 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
569 | qdisc_destroy(q->qdisc); |
570 | kfree(q->delay_dist); | |
571 | } | |
572 | ||
573 | static int netem_dump(struct Qdisc *sch, struct sk_buff *skb) | |
574 | { | |
575 | const struct netem_sched_data *q = qdisc_priv(sch); | |
27a884dc | 576 | unsigned char *b = skb_tail_pointer(skb); |
1e90474c | 577 | struct nlattr *nla = (struct nlattr *) b; |
1da177e4 LT |
578 | struct tc_netem_qopt qopt; |
579 | struct tc_netem_corr cor; | |
0dca51d3 | 580 | struct tc_netem_reorder reorder; |
c865e5d9 | 581 | struct tc_netem_corrupt corrupt; |
1da177e4 LT |
582 | |
583 | qopt.latency = q->latency; | |
584 | qopt.jitter = q->jitter; | |
585 | qopt.limit = q->limit; | |
586 | qopt.loss = q->loss; | |
587 | qopt.gap = q->gap; | |
588 | qopt.duplicate = q->duplicate; | |
1e90474c | 589 | NLA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt); |
1da177e4 LT |
590 | |
591 | cor.delay_corr = q->delay_cor.rho; | |
592 | cor.loss_corr = q->loss_cor.rho; | |
593 | cor.dup_corr = q->dup_cor.rho; | |
1e90474c | 594 | NLA_PUT(skb, TCA_NETEM_CORR, sizeof(cor), &cor); |
0dca51d3 SH |
595 | |
596 | reorder.probability = q->reorder; | |
597 | reorder.correlation = q->reorder_cor.rho; | |
1e90474c | 598 | NLA_PUT(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder); |
0dca51d3 | 599 | |
c865e5d9 SH |
600 | corrupt.probability = q->corrupt; |
601 | corrupt.correlation = q->corrupt_cor.rho; | |
1e90474c | 602 | NLA_PUT(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt); |
c865e5d9 | 603 | |
1e90474c | 604 | nla->nla_len = skb_tail_pointer(skb) - b; |
1da177e4 LT |
605 | |
606 | return skb->len; | |
607 | ||
1e90474c | 608 | nla_put_failure: |
dc5fc579 | 609 | nlmsg_trim(skb, b); |
1da177e4 LT |
610 | return -1; |
611 | } | |
612 | ||
613 | static int netem_dump_class(struct Qdisc *sch, unsigned long cl, | |
614 | struct sk_buff *skb, struct tcmsg *tcm) | |
615 | { | |
616 | struct netem_sched_data *q = qdisc_priv(sch); | |
617 | ||
618 | if (cl != 1) /* only one class */ | |
619 | return -ENOENT; | |
620 | ||
621 | tcm->tcm_handle |= TC_H_MIN(1); | |
622 | tcm->tcm_info = q->qdisc->handle; | |
623 | ||
624 | return 0; | |
625 | } | |
626 | ||
627 | static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
628 | struct Qdisc **old) | |
629 | { | |
630 | struct netem_sched_data *q = qdisc_priv(sch); | |
631 | ||
632 | if (new == NULL) | |
633 | new = &noop_qdisc; | |
634 | ||
635 | sch_tree_lock(sch); | |
636 | *old = xchg(&q->qdisc, new); | |
5e50da01 | 637 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); |
1da177e4 | 638 | qdisc_reset(*old); |
1da177e4 LT |
639 | sch_tree_unlock(sch); |
640 | ||
641 | return 0; | |
642 | } | |
643 | ||
644 | static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg) | |
645 | { | |
646 | struct netem_sched_data *q = qdisc_priv(sch); | |
647 | return q->qdisc; | |
648 | } | |
649 | ||
650 | static unsigned long netem_get(struct Qdisc *sch, u32 classid) | |
651 | { | |
652 | return 1; | |
653 | } | |
654 | ||
655 | static void netem_put(struct Qdisc *sch, unsigned long arg) | |
656 | { | |
657 | } | |
658 | ||
10297b99 | 659 | static int netem_change_class(struct Qdisc *sch, u32 classid, u32 parentid, |
1e90474c | 660 | struct nlattr **tca, unsigned long *arg) |
1da177e4 LT |
661 | { |
662 | return -ENOSYS; | |
663 | } | |
664 | ||
665 | static int netem_delete(struct Qdisc *sch, unsigned long arg) | |
666 | { | |
667 | return -ENOSYS; | |
668 | } | |
669 | ||
670 | static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker) | |
671 | { | |
672 | if (!walker->stop) { | |
673 | if (walker->count >= walker->skip) | |
674 | if (walker->fn(sch, 1, walker) < 0) { | |
675 | walker->stop = 1; | |
676 | return; | |
677 | } | |
678 | walker->count++; | |
679 | } | |
680 | } | |
681 | ||
682 | static struct tcf_proto **netem_find_tcf(struct Qdisc *sch, unsigned long cl) | |
683 | { | |
684 | return NULL; | |
685 | } | |
686 | ||
20fea08b | 687 | static const struct Qdisc_class_ops netem_class_ops = { |
1da177e4 LT |
688 | .graft = netem_graft, |
689 | .leaf = netem_leaf, | |
690 | .get = netem_get, | |
691 | .put = netem_put, | |
692 | .change = netem_change_class, | |
693 | .delete = netem_delete, | |
694 | .walk = netem_walk, | |
695 | .tcf_chain = netem_find_tcf, | |
696 | .dump = netem_dump_class, | |
697 | }; | |
698 | ||
20fea08b | 699 | static struct Qdisc_ops netem_qdisc_ops __read_mostly = { |
1da177e4 LT |
700 | .id = "netem", |
701 | .cl_ops = &netem_class_ops, | |
702 | .priv_size = sizeof(struct netem_sched_data), | |
703 | .enqueue = netem_enqueue, | |
704 | .dequeue = netem_dequeue, | |
705 | .requeue = netem_requeue, | |
706 | .drop = netem_drop, | |
707 | .init = netem_init, | |
708 | .reset = netem_reset, | |
709 | .destroy = netem_destroy, | |
710 | .change = netem_change, | |
711 | .dump = netem_dump, | |
712 | .owner = THIS_MODULE, | |
713 | }; | |
714 | ||
715 | ||
716 | static int __init netem_module_init(void) | |
717 | { | |
eb229c4c | 718 | pr_info("netem: version " VERSION "\n"); |
1da177e4 LT |
719 | return register_qdisc(&netem_qdisc_ops); |
720 | } | |
721 | static void __exit netem_module_exit(void) | |
722 | { | |
723 | unregister_qdisc(&netem_qdisc_ops); | |
724 | } | |
725 | module_init(netem_module_init) | |
726 | module_exit(netem_module_exit) | |
727 | MODULE_LICENSE("GPL"); |