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> |
aec0a40a | 26 | #include <linux/rbtree.h> |
1da177e4 | 27 | |
dc5fc579 | 28 | #include <net/netlink.h> |
1da177e4 | 29 | #include <net/pkt_sched.h> |
e4ae004b | 30 | #include <net/inet_ecn.h> |
1da177e4 | 31 | |
250a65f7 | 32 | #define VERSION "1.3" |
eb229c4c | 33 | |
1da177e4 LT |
34 | /* Network Emulation Queuing algorithm. |
35 | ==================================== | |
36 | ||
37 | Sources: [1] Mark Carson, Darrin Santay, "NIST Net - A Linux-based | |
38 | Network Emulation Tool | |
39 | [2] Luigi Rizzo, DummyNet for FreeBSD | |
40 | ||
41 | ---------------------------------------------------------------- | |
42 | ||
43 | This started out as a simple way to delay outgoing packets to | |
44 | test TCP but has grown to include most of the functionality | |
45 | of a full blown network emulator like NISTnet. It can delay | |
46 | packets and add random jitter (and correlation). The random | |
47 | distribution can be loaded from a table as well to provide | |
48 | normal, Pareto, or experimental curves. Packet loss, | |
49 | duplication, and reordering can also be emulated. | |
50 | ||
51 | This qdisc does not do classification that can be handled in | |
52 | layering other disciplines. It does not need to do bandwidth | |
53 | control either since that can be handled by using token | |
54 | bucket or other rate control. | |
661b7972 | 55 | |
56 | Correlated Loss Generator models | |
57 | ||
58 | Added generation of correlated loss according to the | |
59 | "Gilbert-Elliot" model, a 4-state markov model. | |
60 | ||
61 | References: | |
62 | [1] NetemCLG Home http://netgroup.uniroma2.it/NetemCLG | |
63 | [2] S. Salsano, F. Ludovici, A. Ordine, "Definition of a general | |
64 | and intuitive loss model for packet networks and its implementation | |
65 | in the Netem module in the Linux kernel", available in [1] | |
66 | ||
67 | Authors: Stefano Salsano <stefano.salsano at uniroma2.it | |
68 | Fabio Ludovici <fabio.ludovici at yahoo.it> | |
1da177e4 LT |
69 | */ |
70 | ||
71 | struct netem_sched_data { | |
aec0a40a ED |
72 | /* internal t(ime)fifo qdisc uses t_root and sch->limit */ |
73 | struct rb_root t_root; | |
50612537 ED |
74 | |
75 | /* optional qdisc for classful handling (NULL at netem init) */ | |
1da177e4 | 76 | struct Qdisc *qdisc; |
50612537 | 77 | |
59cb5c67 | 78 | struct qdisc_watchdog watchdog; |
1da177e4 | 79 | |
b407621c SH |
80 | psched_tdiff_t latency; |
81 | psched_tdiff_t jitter; | |
82 | ||
1da177e4 | 83 | u32 loss; |
e4ae004b | 84 | u32 ecn; |
1da177e4 LT |
85 | u32 limit; |
86 | u32 counter; | |
87 | u32 gap; | |
1da177e4 | 88 | u32 duplicate; |
0dca51d3 | 89 | u32 reorder; |
c865e5d9 | 90 | u32 corrupt; |
6a031f67 | 91 | u64 rate; |
90b41a1c HPP |
92 | s32 packet_overhead; |
93 | u32 cell_size; | |
809fa972 | 94 | struct reciprocal_value cell_size_reciprocal; |
90b41a1c | 95 | s32 cell_overhead; |
1da177e4 LT |
96 | |
97 | struct crndstate { | |
b407621c SH |
98 | u32 last; |
99 | u32 rho; | |
c865e5d9 | 100 | } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor; |
1da177e4 LT |
101 | |
102 | struct disttable { | |
103 | u32 size; | |
104 | s16 table[0]; | |
105 | } *delay_dist; | |
661b7972 | 106 | |
107 | enum { | |
108 | CLG_RANDOM, | |
109 | CLG_4_STATES, | |
110 | CLG_GILB_ELL, | |
111 | } loss_model; | |
112 | ||
a6e2fe17 YY |
113 | enum { |
114 | TX_IN_GAP_PERIOD = 1, | |
115 | TX_IN_BURST_PERIOD, | |
116 | LOST_IN_GAP_PERIOD, | |
117 | LOST_IN_BURST_PERIOD, | |
118 | } _4_state_model; | |
119 | ||
c045a734 YY |
120 | enum { |
121 | GOOD_STATE = 1, | |
122 | BAD_STATE, | |
123 | } GE_state_model; | |
124 | ||
661b7972 | 125 | /* Correlated Loss Generation models */ |
126 | struct clgstate { | |
127 | /* state of the Markov chain */ | |
128 | u8 state; | |
129 | ||
130 | /* 4-states and Gilbert-Elliot models */ | |
131 | u32 a1; /* p13 for 4-states or p for GE */ | |
132 | u32 a2; /* p31 for 4-states or r for GE */ | |
133 | u32 a3; /* p32 for 4-states or h for GE */ | |
134 | u32 a4; /* p14 for 4-states or 1-k for GE */ | |
135 | u32 a5; /* p23 used only in 4-states */ | |
136 | } clg; | |
137 | ||
1da177e4 LT |
138 | }; |
139 | ||
50612537 ED |
140 | /* Time stamp put into socket buffer control block |
141 | * Only valid when skbs are in our internal t(ime)fifo queue. | |
56b17425 ED |
142 | * |
143 | * As skb->rbnode uses same storage than skb->next, skb->prev and skb->tstamp, | |
144 | * and skb->next & skb->prev are scratch space for a qdisc, | |
145 | * we save skb->tstamp value in skb->cb[] before destroying it. | |
50612537 | 146 | */ |
1da177e4 LT |
147 | struct netem_skb_cb { |
148 | psched_time_t time_to_send; | |
aec0a40a | 149 | ktime_t tstamp_save; |
1da177e4 LT |
150 | }; |
151 | ||
aec0a40a ED |
152 | |
153 | static struct sk_buff *netem_rb_to_skb(struct rb_node *rb) | |
154 | { | |
56b17425 | 155 | return container_of(rb, struct sk_buff, rbnode); |
aec0a40a ED |
156 | } |
157 | ||
5f86173b JK |
158 | static inline struct netem_skb_cb *netem_skb_cb(struct sk_buff *skb) |
159 | { | |
aec0a40a | 160 | /* we assume we can use skb next/prev/tstamp as storage for rb_node */ |
16bda13d | 161 | qdisc_cb_private_validate(skb, sizeof(struct netem_skb_cb)); |
175f9c1b | 162 | return (struct netem_skb_cb *)qdisc_skb_cb(skb)->data; |
5f86173b JK |
163 | } |
164 | ||
1da177e4 LT |
165 | /* init_crandom - initialize correlated random number generator |
166 | * Use entropy source for initial seed. | |
167 | */ | |
168 | static void init_crandom(struct crndstate *state, unsigned long rho) | |
169 | { | |
170 | state->rho = rho; | |
63862b5b | 171 | state->last = prandom_u32(); |
1da177e4 LT |
172 | } |
173 | ||
174 | /* get_crandom - correlated random number generator | |
175 | * Next number depends on last value. | |
176 | * rho is scaled to avoid floating point. | |
177 | */ | |
b407621c | 178 | static u32 get_crandom(struct crndstate *state) |
1da177e4 LT |
179 | { |
180 | u64 value, rho; | |
181 | unsigned long answer; | |
182 | ||
bb2f8cc0 | 183 | if (state->rho == 0) /* no correlation */ |
63862b5b | 184 | return prandom_u32(); |
1da177e4 | 185 | |
63862b5b | 186 | value = prandom_u32(); |
1da177e4 LT |
187 | rho = (u64)state->rho + 1; |
188 | answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32; | |
189 | state->last = answer; | |
190 | return answer; | |
191 | } | |
192 | ||
661b7972 | 193 | /* loss_4state - 4-state model loss generator |
194 | * Generates losses according to the 4-state Markov chain adopted in | |
195 | * the GI (General and Intuitive) loss model. | |
196 | */ | |
197 | static bool loss_4state(struct netem_sched_data *q) | |
198 | { | |
199 | struct clgstate *clg = &q->clg; | |
63862b5b | 200 | u32 rnd = prandom_u32(); |
661b7972 | 201 | |
202 | /* | |
25985edc | 203 | * Makes a comparison between rnd and the transition |
661b7972 | 204 | * probabilities outgoing from the current state, then decides the |
205 | * next state and if the next packet has to be transmitted or lost. | |
206 | * The four states correspond to: | |
a6e2fe17 YY |
207 | * TX_IN_GAP_PERIOD => successfully transmitted packets within a gap period |
208 | * LOST_IN_BURST_PERIOD => isolated losses within a gap period | |
209 | * LOST_IN_GAP_PERIOD => lost packets within a burst period | |
210 | * TX_IN_GAP_PERIOD => successfully transmitted packets within a burst period | |
661b7972 | 211 | */ |
212 | switch (clg->state) { | |
a6e2fe17 | 213 | case TX_IN_GAP_PERIOD: |
661b7972 | 214 | if (rnd < clg->a4) { |
a6e2fe17 | 215 | clg->state = LOST_IN_BURST_PERIOD; |
661b7972 | 216 | return true; |
ab6c27be | 217 | } else if (clg->a4 < rnd && rnd < clg->a1 + clg->a4) { |
a6e2fe17 | 218 | clg->state = LOST_IN_GAP_PERIOD; |
661b7972 | 219 | return true; |
a6e2fe17 YY |
220 | } else if (clg->a1 + clg->a4 < rnd) { |
221 | clg->state = TX_IN_GAP_PERIOD; | |
222 | } | |
661b7972 | 223 | |
224 | break; | |
a6e2fe17 | 225 | case TX_IN_BURST_PERIOD: |
661b7972 | 226 | if (rnd < clg->a5) { |
a6e2fe17 | 227 | clg->state = LOST_IN_GAP_PERIOD; |
661b7972 | 228 | return true; |
a6e2fe17 YY |
229 | } else { |
230 | clg->state = TX_IN_BURST_PERIOD; | |
231 | } | |
661b7972 | 232 | |
233 | break; | |
a6e2fe17 | 234 | case LOST_IN_GAP_PERIOD: |
661b7972 | 235 | if (rnd < clg->a3) |
a6e2fe17 | 236 | clg->state = TX_IN_BURST_PERIOD; |
661b7972 | 237 | else if (clg->a3 < rnd && rnd < clg->a2 + clg->a3) { |
a6e2fe17 | 238 | clg->state = TX_IN_GAP_PERIOD; |
661b7972 | 239 | } else if (clg->a2 + clg->a3 < rnd) { |
a6e2fe17 | 240 | clg->state = LOST_IN_GAP_PERIOD; |
661b7972 | 241 | return true; |
242 | } | |
243 | break; | |
a6e2fe17 YY |
244 | case LOST_IN_BURST_PERIOD: |
245 | clg->state = TX_IN_GAP_PERIOD; | |
661b7972 | 246 | break; |
247 | } | |
248 | ||
249 | return false; | |
250 | } | |
251 | ||
252 | /* loss_gilb_ell - Gilbert-Elliot model loss generator | |
253 | * Generates losses according to the Gilbert-Elliot loss model or | |
254 | * its special cases (Gilbert or Simple Gilbert) | |
255 | * | |
25985edc | 256 | * Makes a comparison between random number and the transition |
661b7972 | 257 | * probabilities outgoing from the current state, then decides the |
25985edc | 258 | * next state. A second random number is extracted and the comparison |
661b7972 | 259 | * with the loss probability of the current state decides if the next |
260 | * packet will be transmitted or lost. | |
261 | */ | |
262 | static bool loss_gilb_ell(struct netem_sched_data *q) | |
263 | { | |
264 | struct clgstate *clg = &q->clg; | |
265 | ||
266 | switch (clg->state) { | |
c045a734 | 267 | case GOOD_STATE: |
63862b5b | 268 | if (prandom_u32() < clg->a1) |
c045a734 | 269 | clg->state = BAD_STATE; |
63862b5b | 270 | if (prandom_u32() < clg->a4) |
661b7972 | 271 | return true; |
7c2781fa | 272 | break; |
c045a734 | 273 | case BAD_STATE: |
63862b5b | 274 | if (prandom_u32() < clg->a2) |
c045a734 | 275 | clg->state = GOOD_STATE; |
63862b5b | 276 | if (prandom_u32() > clg->a3) |
661b7972 | 277 | return true; |
278 | } | |
279 | ||
280 | return false; | |
281 | } | |
282 | ||
283 | static bool loss_event(struct netem_sched_data *q) | |
284 | { | |
285 | switch (q->loss_model) { | |
286 | case CLG_RANDOM: | |
287 | /* Random packet drop 0 => none, ~0 => all */ | |
288 | return q->loss && q->loss >= get_crandom(&q->loss_cor); | |
289 | ||
290 | case CLG_4_STATES: | |
291 | /* 4state loss model algorithm (used also for GI model) | |
292 | * Extracts a value from the markov 4 state loss generator, | |
293 | * if it is 1 drops a packet and if needed writes the event in | |
294 | * the kernel logs | |
295 | */ | |
296 | return loss_4state(q); | |
297 | ||
298 | case CLG_GILB_ELL: | |
299 | /* Gilbert-Elliot loss model algorithm | |
300 | * Extracts a value from the Gilbert-Elliot loss generator, | |
301 | * if it is 1 drops a packet and if needed writes the event in | |
302 | * the kernel logs | |
303 | */ | |
304 | return loss_gilb_ell(q); | |
305 | } | |
306 | ||
307 | return false; /* not reached */ | |
308 | } | |
309 | ||
310 | ||
1da177e4 LT |
311 | /* tabledist - return a pseudo-randomly distributed value with mean mu and |
312 | * std deviation sigma. Uses table lookup to approximate the desired | |
313 | * distribution, and a uniformly-distributed pseudo-random source. | |
314 | */ | |
b407621c SH |
315 | static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma, |
316 | struct crndstate *state, | |
317 | const struct disttable *dist) | |
1da177e4 | 318 | { |
b407621c SH |
319 | psched_tdiff_t x; |
320 | long t; | |
321 | u32 rnd; | |
1da177e4 LT |
322 | |
323 | if (sigma == 0) | |
324 | return mu; | |
325 | ||
326 | rnd = get_crandom(state); | |
327 | ||
328 | /* default uniform distribution */ | |
10297b99 | 329 | if (dist == NULL) |
1da177e4 LT |
330 | return (rnd % (2*sigma)) - sigma + mu; |
331 | ||
332 | t = dist->table[rnd % dist->size]; | |
333 | x = (sigma % NETEM_DIST_SCALE) * t; | |
334 | if (x >= 0) | |
335 | x += NETEM_DIST_SCALE/2; | |
336 | else | |
337 | x -= NETEM_DIST_SCALE/2; | |
338 | ||
339 | return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu; | |
340 | } | |
341 | ||
90b41a1c | 342 | static psched_time_t packet_len_2_sched_time(unsigned int len, struct netem_sched_data *q) |
7bc0f28c | 343 | { |
90b41a1c | 344 | u64 ticks; |
fc33cc72 | 345 | |
90b41a1c HPP |
346 | len += q->packet_overhead; |
347 | ||
348 | if (q->cell_size) { | |
349 | u32 cells = reciprocal_divide(len, q->cell_size_reciprocal); | |
350 | ||
351 | if (len > cells * q->cell_size) /* extra cell needed for remainder */ | |
352 | cells++; | |
353 | len = cells * (q->cell_size + q->cell_overhead); | |
354 | } | |
355 | ||
356 | ticks = (u64)len * NSEC_PER_SEC; | |
357 | ||
358 | do_div(ticks, q->rate); | |
fc33cc72 | 359 | return PSCHED_NS2TICKS(ticks); |
7bc0f28c HPP |
360 | } |
361 | ||
ff704050 | 362 | static void tfifo_reset(struct Qdisc *sch) |
363 | { | |
364 | struct netem_sched_data *q = qdisc_priv(sch); | |
365 | struct rb_node *p; | |
366 | ||
367 | while ((p = rb_first(&q->t_root))) { | |
368 | struct sk_buff *skb = netem_rb_to_skb(p); | |
369 | ||
370 | rb_erase(p, &q->t_root); | |
371 | skb->next = NULL; | |
372 | skb->prev = NULL; | |
373 | kfree_skb(skb); | |
374 | } | |
375 | } | |
376 | ||
960fb66e | 377 | static void tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch) |
50612537 | 378 | { |
aec0a40a | 379 | struct netem_sched_data *q = qdisc_priv(sch); |
50612537 | 380 | psched_time_t tnext = netem_skb_cb(nskb)->time_to_send; |
aec0a40a | 381 | struct rb_node **p = &q->t_root.rb_node, *parent = NULL; |
50612537 | 382 | |
aec0a40a ED |
383 | while (*p) { |
384 | struct sk_buff *skb; | |
50612537 | 385 | |
aec0a40a ED |
386 | parent = *p; |
387 | skb = netem_rb_to_skb(parent); | |
960fb66e | 388 | if (tnext >= netem_skb_cb(skb)->time_to_send) |
aec0a40a ED |
389 | p = &parent->rb_right; |
390 | else | |
391 | p = &parent->rb_left; | |
50612537 | 392 | } |
56b17425 ED |
393 | rb_link_node(&nskb->rbnode, parent, p); |
394 | rb_insert_color(&nskb->rbnode, &q->t_root); | |
aec0a40a | 395 | sch->q.qlen++; |
50612537 ED |
396 | } |
397 | ||
0afb51e7 SH |
398 | /* |
399 | * Insert one skb into qdisc. | |
400 | * Note: parent depends on return value to account for queue length. | |
401 | * NET_XMIT_DROP: queue length didn't change. | |
402 | * NET_XMIT_SUCCESS: one skb was queued. | |
403 | */ | |
1da177e4 LT |
404 | static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
405 | { | |
406 | struct netem_sched_data *q = qdisc_priv(sch); | |
89e1df74 GC |
407 | /* We don't fill cb now as skb_unshare() may invalidate it */ |
408 | struct netem_skb_cb *cb; | |
0afb51e7 | 409 | struct sk_buff *skb2; |
0afb51e7 | 410 | int count = 1; |
1da177e4 | 411 | |
0afb51e7 SH |
412 | /* Random duplication */ |
413 | if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor)) | |
414 | ++count; | |
415 | ||
661b7972 | 416 | /* Drop packet? */ |
e4ae004b ED |
417 | if (loss_event(q)) { |
418 | if (q->ecn && INET_ECN_set_ce(skb)) | |
25331d6c | 419 | qdisc_qstats_drop(sch); /* mark packet */ |
e4ae004b ED |
420 | else |
421 | --count; | |
422 | } | |
0afb51e7 | 423 | if (count == 0) { |
25331d6c | 424 | qdisc_qstats_drop(sch); |
1da177e4 | 425 | kfree_skb(skb); |
c27f339a | 426 | return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
1da177e4 LT |
427 | } |
428 | ||
5a308f40 ED |
429 | /* If a delay is expected, orphan the skb. (orphaning usually takes |
430 | * place at TX completion time, so _before_ the link transit delay) | |
5a308f40 ED |
431 | */ |
432 | if (q->latency || q->jitter) | |
f2f872f9 | 433 | skb_orphan_partial(skb); |
4e8a5201 | 434 | |
0afb51e7 SH |
435 | /* |
436 | * If we need to duplicate packet, then re-insert at top of the | |
437 | * qdisc tree, since parent queuer expects that only one | |
438 | * skb will be queued. | |
439 | */ | |
440 | if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) { | |
7698b4fc | 441 | struct Qdisc *rootq = qdisc_root(sch); |
0afb51e7 | 442 | u32 dupsave = q->duplicate; /* prevent duplicating a dup... */ |
0afb51e7 | 443 | |
b396cca6 ED |
444 | q->duplicate = 0; |
445 | rootq->enqueue(skb2, rootq); | |
0afb51e7 | 446 | q->duplicate = dupsave; |
1da177e4 LT |
447 | } |
448 | ||
c865e5d9 SH |
449 | /* |
450 | * Randomized packet corruption. | |
451 | * Make copy if needed since we are modifying | |
452 | * If packet is going to be hardware checksummed, then | |
453 | * do it now in software before we mangle it. | |
454 | */ | |
455 | if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) { | |
f64f9e71 JP |
456 | if (!(skb = skb_unshare(skb, GFP_ATOMIC)) || |
457 | (skb->ip_summed == CHECKSUM_PARTIAL && | |
116a0fc3 ED |
458 | skb_checksum_help(skb))) |
459 | return qdisc_drop(skb, sch); | |
c865e5d9 | 460 | |
63862b5b AH |
461 | skb->data[prandom_u32() % skb_headlen(skb)] ^= |
462 | 1<<(prandom_u32() % 8); | |
c865e5d9 SH |
463 | } |
464 | ||
960fb66e ED |
465 | if (unlikely(skb_queue_len(&sch->q) >= sch->limit)) |
466 | return qdisc_reshape_fail(skb, sch); | |
467 | ||
25331d6c | 468 | qdisc_qstats_backlog_inc(sch, skb); |
960fb66e | 469 | |
5f86173b | 470 | cb = netem_skb_cb(skb); |
cc7ec456 | 471 | if (q->gap == 0 || /* not doing reordering */ |
a42b4799 | 472 | q->counter < q->gap - 1 || /* inside last reordering gap */ |
f64f9e71 | 473 | q->reorder < get_crandom(&q->reorder_cor)) { |
0f9f32ac | 474 | psched_time_t now; |
07aaa115 SH |
475 | psched_tdiff_t delay; |
476 | ||
477 | delay = tabledist(q->latency, q->jitter, | |
478 | &q->delay_cor, q->delay_dist); | |
479 | ||
3bebcda2 | 480 | now = psched_get_time(); |
7bc0f28c HPP |
481 | |
482 | if (q->rate) { | |
aec0a40a | 483 | struct sk_buff *last; |
7bc0f28c | 484 | |
aec0a40a ED |
485 | if (!skb_queue_empty(&sch->q)) |
486 | last = skb_peek_tail(&sch->q); | |
487 | else | |
488 | last = netem_rb_to_skb(rb_last(&q->t_root)); | |
489 | if (last) { | |
7bc0f28c | 490 | /* |
a13d3104 JN |
491 | * Last packet in queue is reference point (now), |
492 | * calculate this time bonus and subtract | |
7bc0f28c HPP |
493 | * from delay. |
494 | */ | |
aec0a40a | 495 | delay -= netem_skb_cb(last)->time_to_send - now; |
a13d3104 | 496 | delay = max_t(psched_tdiff_t, 0, delay); |
aec0a40a | 497 | now = netem_skb_cb(last)->time_to_send; |
7bc0f28c | 498 | } |
a13d3104 | 499 | |
8cfd88d6 | 500 | delay += packet_len_2_sched_time(qdisc_pkt_len(skb), q); |
7bc0f28c HPP |
501 | } |
502 | ||
7c59e25f | 503 | cb->time_to_send = now + delay; |
aec0a40a | 504 | cb->tstamp_save = skb->tstamp; |
1da177e4 | 505 | ++q->counter; |
960fb66e | 506 | tfifo_enqueue(skb, sch); |
1da177e4 | 507 | } else { |
10297b99 | 508 | /* |
0dca51d3 SH |
509 | * Do re-ordering by putting one out of N packets at the front |
510 | * of the queue. | |
511 | */ | |
3bebcda2 | 512 | cb->time_to_send = psched_get_time(); |
0dca51d3 | 513 | q->counter = 0; |
8ba25dad | 514 | |
50612537 | 515 | __skb_queue_head(&sch->q, skb); |
eb101924 | 516 | sch->qstats.requeues++; |
378a2f09 | 517 | } |
1da177e4 | 518 | |
10f6dfcf | 519 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
520 | } |
521 | ||
cc7ec456 | 522 | static unsigned int netem_drop(struct Qdisc *sch) |
1da177e4 LT |
523 | { |
524 | struct netem_sched_data *q = qdisc_priv(sch); | |
50612537 | 525 | unsigned int len; |
1da177e4 | 526 | |
50612537 | 527 | len = qdisc_queue_drop(sch); |
aec0a40a ED |
528 | |
529 | if (!len) { | |
530 | struct rb_node *p = rb_first(&q->t_root); | |
531 | ||
532 | if (p) { | |
533 | struct sk_buff *skb = netem_rb_to_skb(p); | |
534 | ||
535 | rb_erase(p, &q->t_root); | |
536 | sch->q.qlen--; | |
537 | skb->next = NULL; | |
538 | skb->prev = NULL; | |
25331d6c | 539 | qdisc_qstats_backlog_dec(sch, skb); |
aec0a40a ED |
540 | kfree_skb(skb); |
541 | } | |
542 | } | |
50612537 ED |
543 | if (!len && q->qdisc && q->qdisc->ops->drop) |
544 | len = q->qdisc->ops->drop(q->qdisc); | |
545 | if (len) | |
25331d6c | 546 | qdisc_qstats_drop(sch); |
50612537 | 547 | |
1da177e4 LT |
548 | return len; |
549 | } | |
550 | ||
1da177e4 LT |
551 | static struct sk_buff *netem_dequeue(struct Qdisc *sch) |
552 | { | |
553 | struct netem_sched_data *q = qdisc_priv(sch); | |
554 | struct sk_buff *skb; | |
aec0a40a | 555 | struct rb_node *p; |
1da177e4 | 556 | |
fd245a4a | 557 | if (qdisc_is_throttled(sch)) |
11274e5a SH |
558 | return NULL; |
559 | ||
50612537 | 560 | tfifo_dequeue: |
aec0a40a | 561 | skb = __skb_dequeue(&sch->q); |
771018e7 | 562 | if (skb) { |
25331d6c | 563 | qdisc_qstats_backlog_dec(sch, skb); |
0ad2a836 | 564 | deliver: |
aec0a40a ED |
565 | qdisc_unthrottled(sch); |
566 | qdisc_bstats_update(sch, skb); | |
567 | return skb; | |
568 | } | |
569 | p = rb_first(&q->t_root); | |
570 | if (p) { | |
36b7bfe0 ED |
571 | psched_time_t time_to_send; |
572 | ||
aec0a40a | 573 | skb = netem_rb_to_skb(p); |
0f9f32ac SH |
574 | |
575 | /* if more time remaining? */ | |
36b7bfe0 ED |
576 | time_to_send = netem_skb_cb(skb)->time_to_send; |
577 | if (time_to_send <= psched_get_time()) { | |
aec0a40a ED |
578 | rb_erase(p, &q->t_root); |
579 | ||
580 | sch->q.qlen--; | |
0ad2a836 | 581 | qdisc_qstats_backlog_dec(sch, skb); |
aec0a40a ED |
582 | skb->next = NULL; |
583 | skb->prev = NULL; | |
584 | skb->tstamp = netem_skb_cb(skb)->tstamp_save; | |
03c05f0d | 585 | |
8caf1539 JP |
586 | #ifdef CONFIG_NET_CLS_ACT |
587 | /* | |
588 | * If it's at ingress let's pretend the delay is | |
589 | * from the network (tstamp will be updated). | |
590 | */ | |
591 | if (G_TC_FROM(skb->tc_verd) & AT_INGRESS) | |
592 | skb->tstamp.tv64 = 0; | |
593 | #endif | |
10f6dfcf | 594 | |
50612537 ED |
595 | if (q->qdisc) { |
596 | int err = qdisc_enqueue(skb, q->qdisc); | |
597 | ||
598 | if (unlikely(err != NET_XMIT_SUCCESS)) { | |
599 | if (net_xmit_drop_count(err)) { | |
25331d6c | 600 | qdisc_qstats_drop(sch); |
50612537 ED |
601 | qdisc_tree_decrease_qlen(sch, 1); |
602 | } | |
603 | } | |
604 | goto tfifo_dequeue; | |
605 | } | |
aec0a40a | 606 | goto deliver; |
07aaa115 | 607 | } |
11274e5a | 608 | |
50612537 ED |
609 | if (q->qdisc) { |
610 | skb = q->qdisc->ops->dequeue(q->qdisc); | |
611 | if (skb) | |
612 | goto deliver; | |
613 | } | |
36b7bfe0 | 614 | qdisc_watchdog_schedule(&q->watchdog, time_to_send); |
0f9f32ac SH |
615 | } |
616 | ||
50612537 ED |
617 | if (q->qdisc) { |
618 | skb = q->qdisc->ops->dequeue(q->qdisc); | |
619 | if (skb) | |
620 | goto deliver; | |
621 | } | |
0f9f32ac | 622 | return NULL; |
1da177e4 LT |
623 | } |
624 | ||
1da177e4 LT |
625 | static void netem_reset(struct Qdisc *sch) |
626 | { | |
627 | struct netem_sched_data *q = qdisc_priv(sch); | |
628 | ||
50612537 | 629 | qdisc_reset_queue(sch); |
ff704050 | 630 | tfifo_reset(sch); |
50612537 ED |
631 | if (q->qdisc) |
632 | qdisc_reset(q->qdisc); | |
59cb5c67 | 633 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
634 | } |
635 | ||
6373a9a2 | 636 | static void dist_free(struct disttable *d) |
637 | { | |
4cb28970 | 638 | kvfree(d); |
6373a9a2 | 639 | } |
640 | ||
1da177e4 LT |
641 | /* |
642 | * Distribution data is a variable size payload containing | |
643 | * signed 16 bit values. | |
644 | */ | |
1e90474c | 645 | static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr) |
1da177e4 LT |
646 | { |
647 | struct netem_sched_data *q = qdisc_priv(sch); | |
6373a9a2 | 648 | size_t n = nla_len(attr)/sizeof(__s16); |
1e90474c | 649 | const __s16 *data = nla_data(attr); |
7698b4fc | 650 | spinlock_t *root_lock; |
1da177e4 LT |
651 | struct disttable *d; |
652 | int i; | |
6373a9a2 | 653 | size_t s; |
1da177e4 | 654 | |
df173bda | 655 | if (n > NETEM_DIST_MAX) |
1da177e4 LT |
656 | return -EINVAL; |
657 | ||
6373a9a2 | 658 | s = sizeof(struct disttable) + n * sizeof(s16); |
bb52c7ac | 659 | d = kmalloc(s, GFP_KERNEL | __GFP_NOWARN); |
6373a9a2 | 660 | if (!d) |
661 | d = vmalloc(s); | |
1da177e4 LT |
662 | if (!d) |
663 | return -ENOMEM; | |
664 | ||
665 | d->size = n; | |
666 | for (i = 0; i < n; i++) | |
667 | d->table[i] = data[i]; | |
10297b99 | 668 | |
102396ae | 669 | root_lock = qdisc_root_sleeping_lock(sch); |
7698b4fc DM |
670 | |
671 | spin_lock_bh(root_lock); | |
bb52c7ac | 672 | swap(q->delay_dist, d); |
7698b4fc | 673 | spin_unlock_bh(root_lock); |
bb52c7ac ED |
674 | |
675 | dist_free(d); | |
1da177e4 LT |
676 | return 0; |
677 | } | |
678 | ||
49545a77 | 679 | static void get_correlation(struct netem_sched_data *q, const struct nlattr *attr) |
1da177e4 | 680 | { |
1e90474c | 681 | const struct tc_netem_corr *c = nla_data(attr); |
1da177e4 | 682 | |
1da177e4 LT |
683 | init_crandom(&q->delay_cor, c->delay_corr); |
684 | init_crandom(&q->loss_cor, c->loss_corr); | |
685 | init_crandom(&q->dup_cor, c->dup_corr); | |
1da177e4 LT |
686 | } |
687 | ||
49545a77 | 688 | static void get_reorder(struct netem_sched_data *q, const struct nlattr *attr) |
0dca51d3 | 689 | { |
1e90474c | 690 | const struct tc_netem_reorder *r = nla_data(attr); |
0dca51d3 | 691 | |
0dca51d3 SH |
692 | q->reorder = r->probability; |
693 | init_crandom(&q->reorder_cor, r->correlation); | |
0dca51d3 SH |
694 | } |
695 | ||
49545a77 | 696 | static void get_corrupt(struct netem_sched_data *q, const struct nlattr *attr) |
c865e5d9 | 697 | { |
1e90474c | 698 | const struct tc_netem_corrupt *r = nla_data(attr); |
c865e5d9 | 699 | |
c865e5d9 SH |
700 | q->corrupt = r->probability; |
701 | init_crandom(&q->corrupt_cor, r->correlation); | |
c865e5d9 SH |
702 | } |
703 | ||
49545a77 | 704 | static void get_rate(struct netem_sched_data *q, const struct nlattr *attr) |
7bc0f28c | 705 | { |
7bc0f28c HPP |
706 | const struct tc_netem_rate *r = nla_data(attr); |
707 | ||
708 | q->rate = r->rate; | |
90b41a1c HPP |
709 | q->packet_overhead = r->packet_overhead; |
710 | q->cell_size = r->cell_size; | |
809fa972 | 711 | q->cell_overhead = r->cell_overhead; |
90b41a1c HPP |
712 | if (q->cell_size) |
713 | q->cell_size_reciprocal = reciprocal_value(q->cell_size); | |
809fa972 HFS |
714 | else |
715 | q->cell_size_reciprocal = (struct reciprocal_value) { 0 }; | |
7bc0f28c HPP |
716 | } |
717 | ||
49545a77 | 718 | static int get_loss_clg(struct netem_sched_data *q, const struct nlattr *attr) |
661b7972 | 719 | { |
661b7972 | 720 | const struct nlattr *la; |
721 | int rem; | |
722 | ||
723 | nla_for_each_nested(la, attr, rem) { | |
724 | u16 type = nla_type(la); | |
725 | ||
833fa743 | 726 | switch (type) { |
661b7972 | 727 | case NETEM_LOSS_GI: { |
728 | const struct tc_netem_gimodel *gi = nla_data(la); | |
729 | ||
2494654d | 730 | if (nla_len(la) < sizeof(struct tc_netem_gimodel)) { |
661b7972 | 731 | pr_info("netem: incorrect gi model size\n"); |
732 | return -EINVAL; | |
733 | } | |
734 | ||
735 | q->loss_model = CLG_4_STATES; | |
736 | ||
3fbac2a8 | 737 | q->clg.state = TX_IN_GAP_PERIOD; |
661b7972 | 738 | q->clg.a1 = gi->p13; |
739 | q->clg.a2 = gi->p31; | |
740 | q->clg.a3 = gi->p32; | |
741 | q->clg.a4 = gi->p14; | |
742 | q->clg.a5 = gi->p23; | |
743 | break; | |
744 | } | |
745 | ||
746 | case NETEM_LOSS_GE: { | |
747 | const struct tc_netem_gemodel *ge = nla_data(la); | |
748 | ||
2494654d | 749 | if (nla_len(la) < sizeof(struct tc_netem_gemodel)) { |
750 | pr_info("netem: incorrect ge model size\n"); | |
661b7972 | 751 | return -EINVAL; |
752 | } | |
753 | ||
754 | q->loss_model = CLG_GILB_ELL; | |
3fbac2a8 | 755 | q->clg.state = GOOD_STATE; |
661b7972 | 756 | q->clg.a1 = ge->p; |
757 | q->clg.a2 = ge->r; | |
758 | q->clg.a3 = ge->h; | |
759 | q->clg.a4 = ge->k1; | |
760 | break; | |
761 | } | |
762 | ||
763 | default: | |
764 | pr_info("netem: unknown loss type %u\n", type); | |
765 | return -EINVAL; | |
766 | } | |
767 | } | |
768 | ||
769 | return 0; | |
770 | } | |
771 | ||
27a3421e PM |
772 | static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = { |
773 | [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) }, | |
774 | [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) }, | |
775 | [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) }, | |
7bc0f28c | 776 | [TCA_NETEM_RATE] = { .len = sizeof(struct tc_netem_rate) }, |
661b7972 | 777 | [TCA_NETEM_LOSS] = { .type = NLA_NESTED }, |
e4ae004b | 778 | [TCA_NETEM_ECN] = { .type = NLA_U32 }, |
6a031f67 | 779 | [TCA_NETEM_RATE64] = { .type = NLA_U64 }, |
27a3421e PM |
780 | }; |
781 | ||
2c10b32b TG |
782 | static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla, |
783 | const struct nla_policy *policy, int len) | |
784 | { | |
785 | int nested_len = nla_len(nla) - NLA_ALIGN(len); | |
786 | ||
661b7972 | 787 | if (nested_len < 0) { |
788 | pr_info("netem: invalid attributes len %d\n", nested_len); | |
2c10b32b | 789 | return -EINVAL; |
661b7972 | 790 | } |
791 | ||
2c10b32b TG |
792 | if (nested_len >= nla_attr_size(0)) |
793 | return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len), | |
794 | nested_len, policy); | |
661b7972 | 795 | |
2c10b32b TG |
796 | memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1)); |
797 | return 0; | |
798 | } | |
799 | ||
c865e5d9 | 800 | /* Parse netlink message to set options */ |
1e90474c | 801 | static int netem_change(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
802 | { |
803 | struct netem_sched_data *q = qdisc_priv(sch); | |
b03f4672 | 804 | struct nlattr *tb[TCA_NETEM_MAX + 1]; |
1da177e4 | 805 | struct tc_netem_qopt *qopt; |
54a4b05c YY |
806 | struct clgstate old_clg; |
807 | int old_loss_model = CLG_RANDOM; | |
1da177e4 | 808 | int ret; |
10297b99 | 809 | |
b03f4672 | 810 | if (opt == NULL) |
1da177e4 LT |
811 | return -EINVAL; |
812 | ||
2c10b32b TG |
813 | qopt = nla_data(opt); |
814 | ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt)); | |
b03f4672 PM |
815 | if (ret < 0) |
816 | return ret; | |
817 | ||
54a4b05c YY |
818 | /* backup q->clg and q->loss_model */ |
819 | old_clg = q->clg; | |
820 | old_loss_model = q->loss_model; | |
821 | ||
822 | if (tb[TCA_NETEM_LOSS]) { | |
49545a77 | 823 | ret = get_loss_clg(q, tb[TCA_NETEM_LOSS]); |
54a4b05c YY |
824 | if (ret) { |
825 | q->loss_model = old_loss_model; | |
826 | return ret; | |
827 | } | |
828 | } else { | |
829 | q->loss_model = CLG_RANDOM; | |
830 | } | |
831 | ||
832 | if (tb[TCA_NETEM_DELAY_DIST]) { | |
833 | ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]); | |
834 | if (ret) { | |
835 | /* recover clg and loss_model, in case of | |
836 | * q->clg and q->loss_model were modified | |
837 | * in get_loss_clg() | |
838 | */ | |
839 | q->clg = old_clg; | |
840 | q->loss_model = old_loss_model; | |
841 | return ret; | |
842 | } | |
843 | } | |
844 | ||
50612537 | 845 | sch->limit = qopt->limit; |
10297b99 | 846 | |
1da177e4 LT |
847 | q->latency = qopt->latency; |
848 | q->jitter = qopt->jitter; | |
849 | q->limit = qopt->limit; | |
850 | q->gap = qopt->gap; | |
0dca51d3 | 851 | q->counter = 0; |
1da177e4 LT |
852 | q->loss = qopt->loss; |
853 | q->duplicate = qopt->duplicate; | |
854 | ||
bb2f8cc0 SH |
855 | /* for compatibility with earlier versions. |
856 | * if gap is set, need to assume 100% probability | |
0dca51d3 | 857 | */ |
a362e0a7 SH |
858 | if (q->gap) |
859 | q->reorder = ~0; | |
0dca51d3 | 860 | |
265eb67f | 861 | if (tb[TCA_NETEM_CORR]) |
49545a77 | 862 | get_correlation(q, tb[TCA_NETEM_CORR]); |
1da177e4 | 863 | |
265eb67f | 864 | if (tb[TCA_NETEM_REORDER]) |
49545a77 | 865 | get_reorder(q, tb[TCA_NETEM_REORDER]); |
1da177e4 | 866 | |
265eb67f | 867 | if (tb[TCA_NETEM_CORRUPT]) |
49545a77 | 868 | get_corrupt(q, tb[TCA_NETEM_CORRUPT]); |
1da177e4 | 869 | |
7bc0f28c | 870 | if (tb[TCA_NETEM_RATE]) |
49545a77 | 871 | get_rate(q, tb[TCA_NETEM_RATE]); |
7bc0f28c | 872 | |
6a031f67 YY |
873 | if (tb[TCA_NETEM_RATE64]) |
874 | q->rate = max_t(u64, q->rate, | |
875 | nla_get_u64(tb[TCA_NETEM_RATE64])); | |
876 | ||
e4ae004b ED |
877 | if (tb[TCA_NETEM_ECN]) |
878 | q->ecn = nla_get_u32(tb[TCA_NETEM_ECN]); | |
879 | ||
661b7972 | 880 | return ret; |
1da177e4 LT |
881 | } |
882 | ||
1e90474c | 883 | static int netem_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
884 | { |
885 | struct netem_sched_data *q = qdisc_priv(sch); | |
886 | int ret; | |
887 | ||
888 | if (!opt) | |
889 | return -EINVAL; | |
890 | ||
59cb5c67 | 891 | qdisc_watchdog_init(&q->watchdog, sch); |
1da177e4 | 892 | |
661b7972 | 893 | q->loss_model = CLG_RANDOM; |
1da177e4 | 894 | ret = netem_change(sch, opt); |
50612537 | 895 | if (ret) |
250a65f7 | 896 | pr_info("netem: change failed\n"); |
1da177e4 LT |
897 | return ret; |
898 | } | |
899 | ||
900 | static void netem_destroy(struct Qdisc *sch) | |
901 | { | |
902 | struct netem_sched_data *q = qdisc_priv(sch); | |
903 | ||
59cb5c67 | 904 | qdisc_watchdog_cancel(&q->watchdog); |
50612537 ED |
905 | if (q->qdisc) |
906 | qdisc_destroy(q->qdisc); | |
6373a9a2 | 907 | dist_free(q->delay_dist); |
1da177e4 LT |
908 | } |
909 | ||
661b7972 | 910 | static int dump_loss_model(const struct netem_sched_data *q, |
911 | struct sk_buff *skb) | |
912 | { | |
913 | struct nlattr *nest; | |
914 | ||
915 | nest = nla_nest_start(skb, TCA_NETEM_LOSS); | |
916 | if (nest == NULL) | |
917 | goto nla_put_failure; | |
918 | ||
919 | switch (q->loss_model) { | |
920 | case CLG_RANDOM: | |
921 | /* legacy loss model */ | |
922 | nla_nest_cancel(skb, nest); | |
923 | return 0; /* no data */ | |
924 | ||
925 | case CLG_4_STATES: { | |
926 | struct tc_netem_gimodel gi = { | |
927 | .p13 = q->clg.a1, | |
928 | .p31 = q->clg.a2, | |
929 | .p32 = q->clg.a3, | |
930 | .p14 = q->clg.a4, | |
931 | .p23 = q->clg.a5, | |
932 | }; | |
933 | ||
1b34ec43 DM |
934 | if (nla_put(skb, NETEM_LOSS_GI, sizeof(gi), &gi)) |
935 | goto nla_put_failure; | |
661b7972 | 936 | break; |
937 | } | |
938 | case CLG_GILB_ELL: { | |
939 | struct tc_netem_gemodel ge = { | |
940 | .p = q->clg.a1, | |
941 | .r = q->clg.a2, | |
942 | .h = q->clg.a3, | |
943 | .k1 = q->clg.a4, | |
944 | }; | |
945 | ||
1b34ec43 DM |
946 | if (nla_put(skb, NETEM_LOSS_GE, sizeof(ge), &ge)) |
947 | goto nla_put_failure; | |
661b7972 | 948 | break; |
949 | } | |
950 | } | |
951 | ||
952 | nla_nest_end(skb, nest); | |
953 | return 0; | |
954 | ||
955 | nla_put_failure: | |
956 | nla_nest_cancel(skb, nest); | |
957 | return -1; | |
958 | } | |
959 | ||
1da177e4 LT |
960 | static int netem_dump(struct Qdisc *sch, struct sk_buff *skb) |
961 | { | |
962 | const struct netem_sched_data *q = qdisc_priv(sch); | |
861d7f74 | 963 | struct nlattr *nla = (struct nlattr *) skb_tail_pointer(skb); |
1da177e4 LT |
964 | struct tc_netem_qopt qopt; |
965 | struct tc_netem_corr cor; | |
0dca51d3 | 966 | struct tc_netem_reorder reorder; |
c865e5d9 | 967 | struct tc_netem_corrupt corrupt; |
7bc0f28c | 968 | struct tc_netem_rate rate; |
1da177e4 LT |
969 | |
970 | qopt.latency = q->latency; | |
971 | qopt.jitter = q->jitter; | |
972 | qopt.limit = q->limit; | |
973 | qopt.loss = q->loss; | |
974 | qopt.gap = q->gap; | |
975 | qopt.duplicate = q->duplicate; | |
1b34ec43 DM |
976 | if (nla_put(skb, TCA_OPTIONS, sizeof(qopt), &qopt)) |
977 | goto nla_put_failure; | |
1da177e4 LT |
978 | |
979 | cor.delay_corr = q->delay_cor.rho; | |
980 | cor.loss_corr = q->loss_cor.rho; | |
981 | cor.dup_corr = q->dup_cor.rho; | |
1b34ec43 DM |
982 | if (nla_put(skb, TCA_NETEM_CORR, sizeof(cor), &cor)) |
983 | goto nla_put_failure; | |
0dca51d3 SH |
984 | |
985 | reorder.probability = q->reorder; | |
986 | reorder.correlation = q->reorder_cor.rho; | |
1b34ec43 DM |
987 | if (nla_put(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder)) |
988 | goto nla_put_failure; | |
0dca51d3 | 989 | |
c865e5d9 SH |
990 | corrupt.probability = q->corrupt; |
991 | corrupt.correlation = q->corrupt_cor.rho; | |
1b34ec43 DM |
992 | if (nla_put(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt)) |
993 | goto nla_put_failure; | |
c865e5d9 | 994 | |
6a031f67 YY |
995 | if (q->rate >= (1ULL << 32)) { |
996 | if (nla_put_u64(skb, TCA_NETEM_RATE64, q->rate)) | |
997 | goto nla_put_failure; | |
998 | rate.rate = ~0U; | |
999 | } else { | |
1000 | rate.rate = q->rate; | |
1001 | } | |
90b41a1c HPP |
1002 | rate.packet_overhead = q->packet_overhead; |
1003 | rate.cell_size = q->cell_size; | |
1004 | rate.cell_overhead = q->cell_overhead; | |
1b34ec43 DM |
1005 | if (nla_put(skb, TCA_NETEM_RATE, sizeof(rate), &rate)) |
1006 | goto nla_put_failure; | |
7bc0f28c | 1007 | |
e4ae004b ED |
1008 | if (q->ecn && nla_put_u32(skb, TCA_NETEM_ECN, q->ecn)) |
1009 | goto nla_put_failure; | |
1010 | ||
661b7972 | 1011 | if (dump_loss_model(q, skb) != 0) |
1012 | goto nla_put_failure; | |
1013 | ||
861d7f74 | 1014 | return nla_nest_end(skb, nla); |
1da177e4 | 1015 | |
1e90474c | 1016 | nla_put_failure: |
861d7f74 | 1017 | nlmsg_trim(skb, nla); |
1da177e4 LT |
1018 | return -1; |
1019 | } | |
1020 | ||
10f6dfcf | 1021 | static int netem_dump_class(struct Qdisc *sch, unsigned long cl, |
1022 | struct sk_buff *skb, struct tcmsg *tcm) | |
1023 | { | |
1024 | struct netem_sched_data *q = qdisc_priv(sch); | |
1025 | ||
50612537 | 1026 | if (cl != 1 || !q->qdisc) /* only one class */ |
10f6dfcf | 1027 | return -ENOENT; |
1028 | ||
1029 | tcm->tcm_handle |= TC_H_MIN(1); | |
1030 | tcm->tcm_info = q->qdisc->handle; | |
1031 | ||
1032 | return 0; | |
1033 | } | |
1034 | ||
1035 | static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
1036 | struct Qdisc **old) | |
1037 | { | |
1038 | struct netem_sched_data *q = qdisc_priv(sch); | |
1039 | ||
10f6dfcf | 1040 | sch_tree_lock(sch); |
1041 | *old = q->qdisc; | |
1042 | q->qdisc = new; | |
50612537 ED |
1043 | if (*old) { |
1044 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); | |
1045 | qdisc_reset(*old); | |
1046 | } | |
10f6dfcf | 1047 | sch_tree_unlock(sch); |
1048 | ||
1049 | return 0; | |
1050 | } | |
1051 | ||
1052 | static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg) | |
1053 | { | |
1054 | struct netem_sched_data *q = qdisc_priv(sch); | |
1055 | return q->qdisc; | |
1056 | } | |
1057 | ||
1058 | static unsigned long netem_get(struct Qdisc *sch, u32 classid) | |
1059 | { | |
1060 | return 1; | |
1061 | } | |
1062 | ||
1063 | static void netem_put(struct Qdisc *sch, unsigned long arg) | |
1064 | { | |
1065 | } | |
1066 | ||
1067 | static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker) | |
1068 | { | |
1069 | if (!walker->stop) { | |
1070 | if (walker->count >= walker->skip) | |
1071 | if (walker->fn(sch, 1, walker) < 0) { | |
1072 | walker->stop = 1; | |
1073 | return; | |
1074 | } | |
1075 | walker->count++; | |
1076 | } | |
1077 | } | |
1078 | ||
1079 | static const struct Qdisc_class_ops netem_class_ops = { | |
1080 | .graft = netem_graft, | |
1081 | .leaf = netem_leaf, | |
1082 | .get = netem_get, | |
1083 | .put = netem_put, | |
1084 | .walk = netem_walk, | |
1085 | .dump = netem_dump_class, | |
1086 | }; | |
1087 | ||
20fea08b | 1088 | static struct Qdisc_ops netem_qdisc_ops __read_mostly = { |
1da177e4 | 1089 | .id = "netem", |
10f6dfcf | 1090 | .cl_ops = &netem_class_ops, |
1da177e4 LT |
1091 | .priv_size = sizeof(struct netem_sched_data), |
1092 | .enqueue = netem_enqueue, | |
1093 | .dequeue = netem_dequeue, | |
77be155c | 1094 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
1095 | .drop = netem_drop, |
1096 | .init = netem_init, | |
1097 | .reset = netem_reset, | |
1098 | .destroy = netem_destroy, | |
1099 | .change = netem_change, | |
1100 | .dump = netem_dump, | |
1101 | .owner = THIS_MODULE, | |
1102 | }; | |
1103 | ||
1104 | ||
1105 | static int __init netem_module_init(void) | |
1106 | { | |
eb229c4c | 1107 | pr_info("netem: version " VERSION "\n"); |
1da177e4 LT |
1108 | return register_qdisc(&netem_qdisc_ops); |
1109 | } | |
1110 | static void __exit netem_module_exit(void) | |
1111 | { | |
1112 | unregister_qdisc(&netem_qdisc_ops); | |
1113 | } | |
1114 | module_init(netem_module_init) | |
1115 | module_exit(netem_module_exit) | |
1116 | MODULE_LICENSE("GPL"); |