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 | ||
661b7972 | 120 | /* Correlated Loss Generation models */ |
121 | struct clgstate { | |
122 | /* state of the Markov chain */ | |
123 | u8 state; | |
124 | ||
125 | /* 4-states and Gilbert-Elliot models */ | |
126 | u32 a1; /* p13 for 4-states or p for GE */ | |
127 | u32 a2; /* p31 for 4-states or r for GE */ | |
128 | u32 a3; /* p32 for 4-states or h for GE */ | |
129 | u32 a4; /* p14 for 4-states or 1-k for GE */ | |
130 | u32 a5; /* p23 used only in 4-states */ | |
131 | } clg; | |
132 | ||
1da177e4 LT |
133 | }; |
134 | ||
50612537 ED |
135 | /* Time stamp put into socket buffer control block |
136 | * Only valid when skbs are in our internal t(ime)fifo queue. | |
137 | */ | |
1da177e4 LT |
138 | struct netem_skb_cb { |
139 | psched_time_t time_to_send; | |
aec0a40a | 140 | ktime_t tstamp_save; |
1da177e4 LT |
141 | }; |
142 | ||
aec0a40a ED |
143 | /* Because space in skb->cb[] is tight, netem overloads skb->next/prev/tstamp |
144 | * to hold a rb_node structure. | |
145 | * | |
146 | * If struct sk_buff layout is changed, the following checks will complain. | |
147 | */ | |
148 | static struct rb_node *netem_rb_node(struct sk_buff *skb) | |
149 | { | |
150 | BUILD_BUG_ON(offsetof(struct sk_buff, next) != 0); | |
151 | BUILD_BUG_ON(offsetof(struct sk_buff, prev) != | |
152 | offsetof(struct sk_buff, next) + sizeof(skb->next)); | |
153 | BUILD_BUG_ON(offsetof(struct sk_buff, tstamp) != | |
154 | offsetof(struct sk_buff, prev) + sizeof(skb->prev)); | |
155 | BUILD_BUG_ON(sizeof(struct rb_node) > sizeof(skb->next) + | |
156 | sizeof(skb->prev) + | |
157 | sizeof(skb->tstamp)); | |
158 | return (struct rb_node *)&skb->next; | |
159 | } | |
160 | ||
161 | static struct sk_buff *netem_rb_to_skb(struct rb_node *rb) | |
162 | { | |
163 | return (struct sk_buff *)rb; | |
164 | } | |
165 | ||
5f86173b JK |
166 | static inline struct netem_skb_cb *netem_skb_cb(struct sk_buff *skb) |
167 | { | |
aec0a40a | 168 | /* we assume we can use skb next/prev/tstamp as storage for rb_node */ |
16bda13d | 169 | qdisc_cb_private_validate(skb, sizeof(struct netem_skb_cb)); |
175f9c1b | 170 | return (struct netem_skb_cb *)qdisc_skb_cb(skb)->data; |
5f86173b JK |
171 | } |
172 | ||
1da177e4 LT |
173 | /* init_crandom - initialize correlated random number generator |
174 | * Use entropy source for initial seed. | |
175 | */ | |
176 | static void init_crandom(struct crndstate *state, unsigned long rho) | |
177 | { | |
178 | state->rho = rho; | |
63862b5b | 179 | state->last = prandom_u32(); |
1da177e4 LT |
180 | } |
181 | ||
182 | /* get_crandom - correlated random number generator | |
183 | * Next number depends on last value. | |
184 | * rho is scaled to avoid floating point. | |
185 | */ | |
b407621c | 186 | static u32 get_crandom(struct crndstate *state) |
1da177e4 LT |
187 | { |
188 | u64 value, rho; | |
189 | unsigned long answer; | |
190 | ||
bb2f8cc0 | 191 | if (state->rho == 0) /* no correlation */ |
63862b5b | 192 | return prandom_u32(); |
1da177e4 | 193 | |
63862b5b | 194 | value = prandom_u32(); |
1da177e4 LT |
195 | rho = (u64)state->rho + 1; |
196 | answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32; | |
197 | state->last = answer; | |
198 | return answer; | |
199 | } | |
200 | ||
661b7972 | 201 | /* loss_4state - 4-state model loss generator |
202 | * Generates losses according to the 4-state Markov chain adopted in | |
203 | * the GI (General and Intuitive) loss model. | |
204 | */ | |
205 | static bool loss_4state(struct netem_sched_data *q) | |
206 | { | |
207 | struct clgstate *clg = &q->clg; | |
63862b5b | 208 | u32 rnd = prandom_u32(); |
661b7972 | 209 | |
210 | /* | |
25985edc | 211 | * Makes a comparison between rnd and the transition |
661b7972 | 212 | * probabilities outgoing from the current state, then decides the |
213 | * next state and if the next packet has to be transmitted or lost. | |
214 | * The four states correspond to: | |
a6e2fe17 YY |
215 | * TX_IN_GAP_PERIOD => successfully transmitted packets within a gap period |
216 | * LOST_IN_BURST_PERIOD => isolated losses within a gap period | |
217 | * LOST_IN_GAP_PERIOD => lost packets within a burst period | |
218 | * TX_IN_GAP_PERIOD => successfully transmitted packets within a burst period | |
661b7972 | 219 | */ |
220 | switch (clg->state) { | |
a6e2fe17 | 221 | case TX_IN_GAP_PERIOD: |
661b7972 | 222 | if (rnd < clg->a4) { |
a6e2fe17 | 223 | clg->state = LOST_IN_BURST_PERIOD; |
661b7972 | 224 | return true; |
ab6c27be | 225 | } else if (clg->a4 < rnd && rnd < clg->a1 + clg->a4) { |
a6e2fe17 | 226 | clg->state = LOST_IN_GAP_PERIOD; |
661b7972 | 227 | return true; |
a6e2fe17 YY |
228 | } else if (clg->a1 + clg->a4 < rnd) { |
229 | clg->state = TX_IN_GAP_PERIOD; | |
230 | } | |
661b7972 | 231 | |
232 | break; | |
a6e2fe17 | 233 | case TX_IN_BURST_PERIOD: |
661b7972 | 234 | if (rnd < clg->a5) { |
a6e2fe17 | 235 | clg->state = LOST_IN_GAP_PERIOD; |
661b7972 | 236 | return true; |
a6e2fe17 YY |
237 | } else { |
238 | clg->state = TX_IN_BURST_PERIOD; | |
239 | } | |
661b7972 | 240 | |
241 | break; | |
a6e2fe17 | 242 | case LOST_IN_GAP_PERIOD: |
661b7972 | 243 | if (rnd < clg->a3) |
a6e2fe17 | 244 | clg->state = TX_IN_BURST_PERIOD; |
661b7972 | 245 | else if (clg->a3 < rnd && rnd < clg->a2 + clg->a3) { |
a6e2fe17 | 246 | clg->state = TX_IN_GAP_PERIOD; |
661b7972 | 247 | } else if (clg->a2 + clg->a3 < rnd) { |
a6e2fe17 | 248 | clg->state = LOST_IN_GAP_PERIOD; |
661b7972 | 249 | return true; |
250 | } | |
251 | break; | |
a6e2fe17 YY |
252 | case LOST_IN_BURST_PERIOD: |
253 | clg->state = TX_IN_GAP_PERIOD; | |
661b7972 | 254 | break; |
255 | } | |
256 | ||
257 | return false; | |
258 | } | |
259 | ||
260 | /* loss_gilb_ell - Gilbert-Elliot model loss generator | |
261 | * Generates losses according to the Gilbert-Elliot loss model or | |
262 | * its special cases (Gilbert or Simple Gilbert) | |
263 | * | |
25985edc | 264 | * Makes a comparison between random number and the transition |
661b7972 | 265 | * probabilities outgoing from the current state, then decides the |
25985edc | 266 | * next state. A second random number is extracted and the comparison |
661b7972 | 267 | * with the loss probability of the current state decides if the next |
268 | * packet will be transmitted or lost. | |
269 | */ | |
270 | static bool loss_gilb_ell(struct netem_sched_data *q) | |
271 | { | |
272 | struct clgstate *clg = &q->clg; | |
273 | ||
274 | switch (clg->state) { | |
275 | case 1: | |
63862b5b | 276 | if (prandom_u32() < clg->a1) |
661b7972 | 277 | clg->state = 2; |
63862b5b | 278 | if (prandom_u32() < clg->a4) |
661b7972 | 279 | return true; |
7c2781fa | 280 | break; |
661b7972 | 281 | case 2: |
63862b5b | 282 | if (prandom_u32() < clg->a2) |
661b7972 | 283 | clg->state = 1; |
63862b5b | 284 | if (prandom_u32() > clg->a3) |
661b7972 | 285 | return true; |
286 | } | |
287 | ||
288 | return false; | |
289 | } | |
290 | ||
291 | static bool loss_event(struct netem_sched_data *q) | |
292 | { | |
293 | switch (q->loss_model) { | |
294 | case CLG_RANDOM: | |
295 | /* Random packet drop 0 => none, ~0 => all */ | |
296 | return q->loss && q->loss >= get_crandom(&q->loss_cor); | |
297 | ||
298 | case CLG_4_STATES: | |
299 | /* 4state loss model algorithm (used also for GI model) | |
300 | * Extracts a value from the markov 4 state 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_4state(q); | |
305 | ||
306 | case CLG_GILB_ELL: | |
307 | /* Gilbert-Elliot loss model algorithm | |
308 | * Extracts a value from the Gilbert-Elliot loss generator, | |
309 | * if it is 1 drops a packet and if needed writes the event in | |
310 | * the kernel logs | |
311 | */ | |
312 | return loss_gilb_ell(q); | |
313 | } | |
314 | ||
315 | return false; /* not reached */ | |
316 | } | |
317 | ||
318 | ||
1da177e4 LT |
319 | /* tabledist - return a pseudo-randomly distributed value with mean mu and |
320 | * std deviation sigma. Uses table lookup to approximate the desired | |
321 | * distribution, and a uniformly-distributed pseudo-random source. | |
322 | */ | |
b407621c SH |
323 | static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma, |
324 | struct crndstate *state, | |
325 | const struct disttable *dist) | |
1da177e4 | 326 | { |
b407621c SH |
327 | psched_tdiff_t x; |
328 | long t; | |
329 | u32 rnd; | |
1da177e4 LT |
330 | |
331 | if (sigma == 0) | |
332 | return mu; | |
333 | ||
334 | rnd = get_crandom(state); | |
335 | ||
336 | /* default uniform distribution */ | |
10297b99 | 337 | if (dist == NULL) |
1da177e4 LT |
338 | return (rnd % (2*sigma)) - sigma + mu; |
339 | ||
340 | t = dist->table[rnd % dist->size]; | |
341 | x = (sigma % NETEM_DIST_SCALE) * t; | |
342 | if (x >= 0) | |
343 | x += NETEM_DIST_SCALE/2; | |
344 | else | |
345 | x -= NETEM_DIST_SCALE/2; | |
346 | ||
347 | return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu; | |
348 | } | |
349 | ||
90b41a1c | 350 | static psched_time_t packet_len_2_sched_time(unsigned int len, struct netem_sched_data *q) |
7bc0f28c | 351 | { |
90b41a1c | 352 | u64 ticks; |
fc33cc72 | 353 | |
90b41a1c HPP |
354 | len += q->packet_overhead; |
355 | ||
356 | if (q->cell_size) { | |
357 | u32 cells = reciprocal_divide(len, q->cell_size_reciprocal); | |
358 | ||
359 | if (len > cells * q->cell_size) /* extra cell needed for remainder */ | |
360 | cells++; | |
361 | len = cells * (q->cell_size + q->cell_overhead); | |
362 | } | |
363 | ||
364 | ticks = (u64)len * NSEC_PER_SEC; | |
365 | ||
366 | do_div(ticks, q->rate); | |
fc33cc72 | 367 | return PSCHED_NS2TICKS(ticks); |
7bc0f28c HPP |
368 | } |
369 | ||
ff704050 | 370 | static void tfifo_reset(struct Qdisc *sch) |
371 | { | |
372 | struct netem_sched_data *q = qdisc_priv(sch); | |
373 | struct rb_node *p; | |
374 | ||
375 | while ((p = rb_first(&q->t_root))) { | |
376 | struct sk_buff *skb = netem_rb_to_skb(p); | |
377 | ||
378 | rb_erase(p, &q->t_root); | |
379 | skb->next = NULL; | |
380 | skb->prev = NULL; | |
381 | kfree_skb(skb); | |
382 | } | |
383 | } | |
384 | ||
960fb66e | 385 | static void tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch) |
50612537 | 386 | { |
aec0a40a | 387 | struct netem_sched_data *q = qdisc_priv(sch); |
50612537 | 388 | psched_time_t tnext = netem_skb_cb(nskb)->time_to_send; |
aec0a40a | 389 | struct rb_node **p = &q->t_root.rb_node, *parent = NULL; |
50612537 | 390 | |
aec0a40a ED |
391 | while (*p) { |
392 | struct sk_buff *skb; | |
50612537 | 393 | |
aec0a40a ED |
394 | parent = *p; |
395 | skb = netem_rb_to_skb(parent); | |
960fb66e | 396 | if (tnext >= netem_skb_cb(skb)->time_to_send) |
aec0a40a ED |
397 | p = &parent->rb_right; |
398 | else | |
399 | p = &parent->rb_left; | |
50612537 | 400 | } |
aec0a40a ED |
401 | rb_link_node(netem_rb_node(nskb), parent, p); |
402 | rb_insert_color(netem_rb_node(nskb), &q->t_root); | |
403 | sch->q.qlen++; | |
50612537 ED |
404 | } |
405 | ||
0afb51e7 SH |
406 | /* |
407 | * Insert one skb into qdisc. | |
408 | * Note: parent depends on return value to account for queue length. | |
409 | * NET_XMIT_DROP: queue length didn't change. | |
410 | * NET_XMIT_SUCCESS: one skb was queued. | |
411 | */ | |
1da177e4 LT |
412 | static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
413 | { | |
414 | struct netem_sched_data *q = qdisc_priv(sch); | |
89e1df74 GC |
415 | /* We don't fill cb now as skb_unshare() may invalidate it */ |
416 | struct netem_skb_cb *cb; | |
0afb51e7 | 417 | struct sk_buff *skb2; |
0afb51e7 | 418 | int count = 1; |
1da177e4 | 419 | |
0afb51e7 SH |
420 | /* Random duplication */ |
421 | if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor)) | |
422 | ++count; | |
423 | ||
661b7972 | 424 | /* Drop packet? */ |
e4ae004b ED |
425 | if (loss_event(q)) { |
426 | if (q->ecn && INET_ECN_set_ce(skb)) | |
427 | sch->qstats.drops++; /* mark packet */ | |
428 | else | |
429 | --count; | |
430 | } | |
0afb51e7 | 431 | if (count == 0) { |
1da177e4 LT |
432 | sch->qstats.drops++; |
433 | kfree_skb(skb); | |
c27f339a | 434 | return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
1da177e4 LT |
435 | } |
436 | ||
5a308f40 ED |
437 | /* If a delay is expected, orphan the skb. (orphaning usually takes |
438 | * place at TX completion time, so _before_ the link transit delay) | |
5a308f40 ED |
439 | */ |
440 | if (q->latency || q->jitter) | |
f2f872f9 | 441 | skb_orphan_partial(skb); |
4e8a5201 | 442 | |
0afb51e7 SH |
443 | /* |
444 | * If we need to duplicate packet, then re-insert at top of the | |
445 | * qdisc tree, since parent queuer expects that only one | |
446 | * skb will be queued. | |
447 | */ | |
448 | if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) { | |
7698b4fc | 449 | struct Qdisc *rootq = qdisc_root(sch); |
0afb51e7 SH |
450 | u32 dupsave = q->duplicate; /* prevent duplicating a dup... */ |
451 | q->duplicate = 0; | |
452 | ||
5f86173b | 453 | qdisc_enqueue_root(skb2, rootq); |
0afb51e7 | 454 | q->duplicate = dupsave; |
1da177e4 LT |
455 | } |
456 | ||
c865e5d9 SH |
457 | /* |
458 | * Randomized packet corruption. | |
459 | * Make copy if needed since we are modifying | |
460 | * If packet is going to be hardware checksummed, then | |
461 | * do it now in software before we mangle it. | |
462 | */ | |
463 | if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) { | |
f64f9e71 JP |
464 | if (!(skb = skb_unshare(skb, GFP_ATOMIC)) || |
465 | (skb->ip_summed == CHECKSUM_PARTIAL && | |
116a0fc3 ED |
466 | skb_checksum_help(skb))) |
467 | return qdisc_drop(skb, sch); | |
c865e5d9 | 468 | |
63862b5b AH |
469 | skb->data[prandom_u32() % skb_headlen(skb)] ^= |
470 | 1<<(prandom_u32() % 8); | |
c865e5d9 SH |
471 | } |
472 | ||
960fb66e ED |
473 | if (unlikely(skb_queue_len(&sch->q) >= sch->limit)) |
474 | return qdisc_reshape_fail(skb, sch); | |
475 | ||
476 | sch->qstats.backlog += qdisc_pkt_len(skb); | |
477 | ||
5f86173b | 478 | cb = netem_skb_cb(skb); |
cc7ec456 | 479 | if (q->gap == 0 || /* not doing reordering */ |
a42b4799 | 480 | q->counter < q->gap - 1 || /* inside last reordering gap */ |
f64f9e71 | 481 | q->reorder < get_crandom(&q->reorder_cor)) { |
0f9f32ac | 482 | psched_time_t now; |
07aaa115 SH |
483 | psched_tdiff_t delay; |
484 | ||
485 | delay = tabledist(q->latency, q->jitter, | |
486 | &q->delay_cor, q->delay_dist); | |
487 | ||
3bebcda2 | 488 | now = psched_get_time(); |
7bc0f28c HPP |
489 | |
490 | if (q->rate) { | |
aec0a40a | 491 | struct sk_buff *last; |
7bc0f28c | 492 | |
aec0a40a ED |
493 | if (!skb_queue_empty(&sch->q)) |
494 | last = skb_peek_tail(&sch->q); | |
495 | else | |
496 | last = netem_rb_to_skb(rb_last(&q->t_root)); | |
497 | if (last) { | |
7bc0f28c | 498 | /* |
a13d3104 JN |
499 | * Last packet in queue is reference point (now), |
500 | * calculate this time bonus and subtract | |
7bc0f28c HPP |
501 | * from delay. |
502 | */ | |
aec0a40a | 503 | delay -= netem_skb_cb(last)->time_to_send - now; |
a13d3104 | 504 | delay = max_t(psched_tdiff_t, 0, delay); |
aec0a40a | 505 | now = netem_skb_cb(last)->time_to_send; |
7bc0f28c | 506 | } |
a13d3104 | 507 | |
8cfd88d6 | 508 | delay += packet_len_2_sched_time(qdisc_pkt_len(skb), q); |
7bc0f28c HPP |
509 | } |
510 | ||
7c59e25f | 511 | cb->time_to_send = now + delay; |
aec0a40a | 512 | cb->tstamp_save = skb->tstamp; |
1da177e4 | 513 | ++q->counter; |
960fb66e | 514 | tfifo_enqueue(skb, sch); |
1da177e4 | 515 | } else { |
10297b99 | 516 | /* |
0dca51d3 SH |
517 | * Do re-ordering by putting one out of N packets at the front |
518 | * of the queue. | |
519 | */ | |
3bebcda2 | 520 | cb->time_to_send = psched_get_time(); |
0dca51d3 | 521 | q->counter = 0; |
8ba25dad | 522 | |
50612537 | 523 | __skb_queue_head(&sch->q, skb); |
eb101924 | 524 | sch->qstats.requeues++; |
378a2f09 | 525 | } |
1da177e4 | 526 | |
10f6dfcf | 527 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
528 | } |
529 | ||
cc7ec456 | 530 | static unsigned int netem_drop(struct Qdisc *sch) |
1da177e4 LT |
531 | { |
532 | struct netem_sched_data *q = qdisc_priv(sch); | |
50612537 | 533 | unsigned int len; |
1da177e4 | 534 | |
50612537 | 535 | len = qdisc_queue_drop(sch); |
aec0a40a ED |
536 | |
537 | if (!len) { | |
538 | struct rb_node *p = rb_first(&q->t_root); | |
539 | ||
540 | if (p) { | |
541 | struct sk_buff *skb = netem_rb_to_skb(p); | |
542 | ||
543 | rb_erase(p, &q->t_root); | |
544 | sch->q.qlen--; | |
545 | skb->next = NULL; | |
546 | skb->prev = NULL; | |
547 | len = qdisc_pkt_len(skb); | |
638a52b8 | 548 | sch->qstats.backlog -= len; |
aec0a40a ED |
549 | kfree_skb(skb); |
550 | } | |
551 | } | |
50612537 ED |
552 | if (!len && q->qdisc && q->qdisc->ops->drop) |
553 | len = q->qdisc->ops->drop(q->qdisc); | |
554 | if (len) | |
1da177e4 | 555 | sch->qstats.drops++; |
50612537 | 556 | |
1da177e4 LT |
557 | return len; |
558 | } | |
559 | ||
1da177e4 LT |
560 | static struct sk_buff *netem_dequeue(struct Qdisc *sch) |
561 | { | |
562 | struct netem_sched_data *q = qdisc_priv(sch); | |
563 | struct sk_buff *skb; | |
aec0a40a | 564 | struct rb_node *p; |
1da177e4 | 565 | |
fd245a4a | 566 | if (qdisc_is_throttled(sch)) |
11274e5a SH |
567 | return NULL; |
568 | ||
50612537 | 569 | tfifo_dequeue: |
aec0a40a | 570 | skb = __skb_dequeue(&sch->q); |
771018e7 | 571 | if (skb) { |
aec0a40a ED |
572 | deliver: |
573 | sch->qstats.backlog -= qdisc_pkt_len(skb); | |
574 | qdisc_unthrottled(sch); | |
575 | qdisc_bstats_update(sch, skb); | |
576 | return skb; | |
577 | } | |
578 | p = rb_first(&q->t_root); | |
579 | if (p) { | |
36b7bfe0 ED |
580 | psched_time_t time_to_send; |
581 | ||
aec0a40a | 582 | skb = netem_rb_to_skb(p); |
0f9f32ac SH |
583 | |
584 | /* if more time remaining? */ | |
36b7bfe0 ED |
585 | time_to_send = netem_skb_cb(skb)->time_to_send; |
586 | if (time_to_send <= psched_get_time()) { | |
aec0a40a ED |
587 | rb_erase(p, &q->t_root); |
588 | ||
589 | sch->q.qlen--; | |
590 | skb->next = NULL; | |
591 | skb->prev = NULL; | |
592 | skb->tstamp = netem_skb_cb(skb)->tstamp_save; | |
03c05f0d | 593 | |
8caf1539 JP |
594 | #ifdef CONFIG_NET_CLS_ACT |
595 | /* | |
596 | * If it's at ingress let's pretend the delay is | |
597 | * from the network (tstamp will be updated). | |
598 | */ | |
599 | if (G_TC_FROM(skb->tc_verd) & AT_INGRESS) | |
600 | skb->tstamp.tv64 = 0; | |
601 | #endif | |
10f6dfcf | 602 | |
50612537 ED |
603 | if (q->qdisc) { |
604 | int err = qdisc_enqueue(skb, q->qdisc); | |
605 | ||
606 | if (unlikely(err != NET_XMIT_SUCCESS)) { | |
607 | if (net_xmit_drop_count(err)) { | |
608 | sch->qstats.drops++; | |
609 | qdisc_tree_decrease_qlen(sch, 1); | |
610 | } | |
611 | } | |
612 | goto tfifo_dequeue; | |
613 | } | |
aec0a40a | 614 | goto deliver; |
07aaa115 | 615 | } |
11274e5a | 616 | |
50612537 ED |
617 | if (q->qdisc) { |
618 | skb = q->qdisc->ops->dequeue(q->qdisc); | |
619 | if (skb) | |
620 | goto deliver; | |
621 | } | |
36b7bfe0 | 622 | qdisc_watchdog_schedule(&q->watchdog, time_to_send); |
0f9f32ac SH |
623 | } |
624 | ||
50612537 ED |
625 | if (q->qdisc) { |
626 | skb = q->qdisc->ops->dequeue(q->qdisc); | |
627 | if (skb) | |
628 | goto deliver; | |
629 | } | |
0f9f32ac | 630 | return NULL; |
1da177e4 LT |
631 | } |
632 | ||
1da177e4 LT |
633 | static void netem_reset(struct Qdisc *sch) |
634 | { | |
635 | struct netem_sched_data *q = qdisc_priv(sch); | |
636 | ||
50612537 | 637 | qdisc_reset_queue(sch); |
ff704050 | 638 | tfifo_reset(sch); |
50612537 ED |
639 | if (q->qdisc) |
640 | qdisc_reset(q->qdisc); | |
59cb5c67 | 641 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
642 | } |
643 | ||
6373a9a2 | 644 | static void dist_free(struct disttable *d) |
645 | { | |
646 | if (d) { | |
647 | if (is_vmalloc_addr(d)) | |
648 | vfree(d); | |
649 | else | |
650 | kfree(d); | |
651 | } | |
652 | } | |
653 | ||
1da177e4 LT |
654 | /* |
655 | * Distribution data is a variable size payload containing | |
656 | * signed 16 bit values. | |
657 | */ | |
1e90474c | 658 | static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr) |
1da177e4 LT |
659 | { |
660 | struct netem_sched_data *q = qdisc_priv(sch); | |
6373a9a2 | 661 | size_t n = nla_len(attr)/sizeof(__s16); |
1e90474c | 662 | const __s16 *data = nla_data(attr); |
7698b4fc | 663 | spinlock_t *root_lock; |
1da177e4 LT |
664 | struct disttable *d; |
665 | int i; | |
6373a9a2 | 666 | size_t s; |
1da177e4 | 667 | |
df173bda | 668 | if (n > NETEM_DIST_MAX) |
1da177e4 LT |
669 | return -EINVAL; |
670 | ||
6373a9a2 | 671 | s = sizeof(struct disttable) + n * sizeof(s16); |
bb52c7ac | 672 | d = kmalloc(s, GFP_KERNEL | __GFP_NOWARN); |
6373a9a2 | 673 | if (!d) |
674 | d = vmalloc(s); | |
1da177e4 LT |
675 | if (!d) |
676 | return -ENOMEM; | |
677 | ||
678 | d->size = n; | |
679 | for (i = 0; i < n; i++) | |
680 | d->table[i] = data[i]; | |
10297b99 | 681 | |
102396ae | 682 | root_lock = qdisc_root_sleeping_lock(sch); |
7698b4fc DM |
683 | |
684 | spin_lock_bh(root_lock); | |
bb52c7ac | 685 | swap(q->delay_dist, d); |
7698b4fc | 686 | spin_unlock_bh(root_lock); |
bb52c7ac ED |
687 | |
688 | dist_free(d); | |
1da177e4 LT |
689 | return 0; |
690 | } | |
691 | ||
49545a77 | 692 | static void get_correlation(struct netem_sched_data *q, const struct nlattr *attr) |
1da177e4 | 693 | { |
1e90474c | 694 | const struct tc_netem_corr *c = nla_data(attr); |
1da177e4 | 695 | |
1da177e4 LT |
696 | init_crandom(&q->delay_cor, c->delay_corr); |
697 | init_crandom(&q->loss_cor, c->loss_corr); | |
698 | init_crandom(&q->dup_cor, c->dup_corr); | |
1da177e4 LT |
699 | } |
700 | ||
49545a77 | 701 | static void get_reorder(struct netem_sched_data *q, const struct nlattr *attr) |
0dca51d3 | 702 | { |
1e90474c | 703 | const struct tc_netem_reorder *r = nla_data(attr); |
0dca51d3 | 704 | |
0dca51d3 SH |
705 | q->reorder = r->probability; |
706 | init_crandom(&q->reorder_cor, r->correlation); | |
0dca51d3 SH |
707 | } |
708 | ||
49545a77 | 709 | static void get_corrupt(struct netem_sched_data *q, const struct nlattr *attr) |
c865e5d9 | 710 | { |
1e90474c | 711 | const struct tc_netem_corrupt *r = nla_data(attr); |
c865e5d9 | 712 | |
c865e5d9 SH |
713 | q->corrupt = r->probability; |
714 | init_crandom(&q->corrupt_cor, r->correlation); | |
c865e5d9 SH |
715 | } |
716 | ||
49545a77 | 717 | static void get_rate(struct netem_sched_data *q, const struct nlattr *attr) |
7bc0f28c | 718 | { |
7bc0f28c HPP |
719 | const struct tc_netem_rate *r = nla_data(attr); |
720 | ||
721 | q->rate = r->rate; | |
90b41a1c HPP |
722 | q->packet_overhead = r->packet_overhead; |
723 | q->cell_size = r->cell_size; | |
809fa972 | 724 | q->cell_overhead = r->cell_overhead; |
90b41a1c HPP |
725 | if (q->cell_size) |
726 | q->cell_size_reciprocal = reciprocal_value(q->cell_size); | |
809fa972 HFS |
727 | else |
728 | q->cell_size_reciprocal = (struct reciprocal_value) { 0 }; | |
7bc0f28c HPP |
729 | } |
730 | ||
49545a77 | 731 | static int get_loss_clg(struct netem_sched_data *q, const struct nlattr *attr) |
661b7972 | 732 | { |
661b7972 | 733 | const struct nlattr *la; |
734 | int rem; | |
735 | ||
736 | nla_for_each_nested(la, attr, rem) { | |
737 | u16 type = nla_type(la); | |
738 | ||
833fa743 | 739 | switch (type) { |
661b7972 | 740 | case NETEM_LOSS_GI: { |
741 | const struct tc_netem_gimodel *gi = nla_data(la); | |
742 | ||
2494654d | 743 | if (nla_len(la) < sizeof(struct tc_netem_gimodel)) { |
661b7972 | 744 | pr_info("netem: incorrect gi model size\n"); |
745 | return -EINVAL; | |
746 | } | |
747 | ||
748 | q->loss_model = CLG_4_STATES; | |
749 | ||
750 | q->clg.state = 1; | |
751 | q->clg.a1 = gi->p13; | |
752 | q->clg.a2 = gi->p31; | |
753 | q->clg.a3 = gi->p32; | |
754 | q->clg.a4 = gi->p14; | |
755 | q->clg.a5 = gi->p23; | |
756 | break; | |
757 | } | |
758 | ||
759 | case NETEM_LOSS_GE: { | |
760 | const struct tc_netem_gemodel *ge = nla_data(la); | |
761 | ||
2494654d | 762 | if (nla_len(la) < sizeof(struct tc_netem_gemodel)) { |
763 | pr_info("netem: incorrect ge model size\n"); | |
661b7972 | 764 | return -EINVAL; |
765 | } | |
766 | ||
767 | q->loss_model = CLG_GILB_ELL; | |
768 | q->clg.state = 1; | |
769 | q->clg.a1 = ge->p; | |
770 | q->clg.a2 = ge->r; | |
771 | q->clg.a3 = ge->h; | |
772 | q->clg.a4 = ge->k1; | |
773 | break; | |
774 | } | |
775 | ||
776 | default: | |
777 | pr_info("netem: unknown loss type %u\n", type); | |
778 | return -EINVAL; | |
779 | } | |
780 | } | |
781 | ||
782 | return 0; | |
783 | } | |
784 | ||
27a3421e PM |
785 | static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = { |
786 | [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) }, | |
787 | [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) }, | |
788 | [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) }, | |
7bc0f28c | 789 | [TCA_NETEM_RATE] = { .len = sizeof(struct tc_netem_rate) }, |
661b7972 | 790 | [TCA_NETEM_LOSS] = { .type = NLA_NESTED }, |
e4ae004b | 791 | [TCA_NETEM_ECN] = { .type = NLA_U32 }, |
6a031f67 | 792 | [TCA_NETEM_RATE64] = { .type = NLA_U64 }, |
27a3421e PM |
793 | }; |
794 | ||
2c10b32b TG |
795 | static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla, |
796 | const struct nla_policy *policy, int len) | |
797 | { | |
798 | int nested_len = nla_len(nla) - NLA_ALIGN(len); | |
799 | ||
661b7972 | 800 | if (nested_len < 0) { |
801 | pr_info("netem: invalid attributes len %d\n", nested_len); | |
2c10b32b | 802 | return -EINVAL; |
661b7972 | 803 | } |
804 | ||
2c10b32b TG |
805 | if (nested_len >= nla_attr_size(0)) |
806 | return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len), | |
807 | nested_len, policy); | |
661b7972 | 808 | |
2c10b32b TG |
809 | memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1)); |
810 | return 0; | |
811 | } | |
812 | ||
c865e5d9 | 813 | /* Parse netlink message to set options */ |
1e90474c | 814 | static int netem_change(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
815 | { |
816 | struct netem_sched_data *q = qdisc_priv(sch); | |
b03f4672 | 817 | struct nlattr *tb[TCA_NETEM_MAX + 1]; |
1da177e4 | 818 | struct tc_netem_qopt *qopt; |
54a4b05c YY |
819 | struct clgstate old_clg; |
820 | int old_loss_model = CLG_RANDOM; | |
1da177e4 | 821 | int ret; |
10297b99 | 822 | |
b03f4672 | 823 | if (opt == NULL) |
1da177e4 LT |
824 | return -EINVAL; |
825 | ||
2c10b32b TG |
826 | qopt = nla_data(opt); |
827 | ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt)); | |
b03f4672 PM |
828 | if (ret < 0) |
829 | return ret; | |
830 | ||
54a4b05c YY |
831 | /* backup q->clg and q->loss_model */ |
832 | old_clg = q->clg; | |
833 | old_loss_model = q->loss_model; | |
834 | ||
835 | if (tb[TCA_NETEM_LOSS]) { | |
49545a77 | 836 | ret = get_loss_clg(q, tb[TCA_NETEM_LOSS]); |
54a4b05c YY |
837 | if (ret) { |
838 | q->loss_model = old_loss_model; | |
839 | return ret; | |
840 | } | |
841 | } else { | |
842 | q->loss_model = CLG_RANDOM; | |
843 | } | |
844 | ||
845 | if (tb[TCA_NETEM_DELAY_DIST]) { | |
846 | ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]); | |
847 | if (ret) { | |
848 | /* recover clg and loss_model, in case of | |
849 | * q->clg and q->loss_model were modified | |
850 | * in get_loss_clg() | |
851 | */ | |
852 | q->clg = old_clg; | |
853 | q->loss_model = old_loss_model; | |
854 | return ret; | |
855 | } | |
856 | } | |
857 | ||
50612537 | 858 | sch->limit = qopt->limit; |
10297b99 | 859 | |
1da177e4 LT |
860 | q->latency = qopt->latency; |
861 | q->jitter = qopt->jitter; | |
862 | q->limit = qopt->limit; | |
863 | q->gap = qopt->gap; | |
0dca51d3 | 864 | q->counter = 0; |
1da177e4 LT |
865 | q->loss = qopt->loss; |
866 | q->duplicate = qopt->duplicate; | |
867 | ||
bb2f8cc0 SH |
868 | /* for compatibility with earlier versions. |
869 | * if gap is set, need to assume 100% probability | |
0dca51d3 | 870 | */ |
a362e0a7 SH |
871 | if (q->gap) |
872 | q->reorder = ~0; | |
0dca51d3 | 873 | |
265eb67f | 874 | if (tb[TCA_NETEM_CORR]) |
49545a77 | 875 | get_correlation(q, tb[TCA_NETEM_CORR]); |
1da177e4 | 876 | |
265eb67f | 877 | if (tb[TCA_NETEM_REORDER]) |
49545a77 | 878 | get_reorder(q, tb[TCA_NETEM_REORDER]); |
1da177e4 | 879 | |
265eb67f | 880 | if (tb[TCA_NETEM_CORRUPT]) |
49545a77 | 881 | get_corrupt(q, tb[TCA_NETEM_CORRUPT]); |
1da177e4 | 882 | |
7bc0f28c | 883 | if (tb[TCA_NETEM_RATE]) |
49545a77 | 884 | get_rate(q, tb[TCA_NETEM_RATE]); |
7bc0f28c | 885 | |
6a031f67 YY |
886 | if (tb[TCA_NETEM_RATE64]) |
887 | q->rate = max_t(u64, q->rate, | |
888 | nla_get_u64(tb[TCA_NETEM_RATE64])); | |
889 | ||
e4ae004b ED |
890 | if (tb[TCA_NETEM_ECN]) |
891 | q->ecn = nla_get_u32(tb[TCA_NETEM_ECN]); | |
892 | ||
661b7972 | 893 | return ret; |
1da177e4 LT |
894 | } |
895 | ||
1e90474c | 896 | static int netem_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
897 | { |
898 | struct netem_sched_data *q = qdisc_priv(sch); | |
899 | int ret; | |
900 | ||
901 | if (!opt) | |
902 | return -EINVAL; | |
903 | ||
59cb5c67 | 904 | qdisc_watchdog_init(&q->watchdog, sch); |
1da177e4 | 905 | |
661b7972 | 906 | q->loss_model = CLG_RANDOM; |
1da177e4 | 907 | ret = netem_change(sch, opt); |
50612537 | 908 | if (ret) |
250a65f7 | 909 | pr_info("netem: change failed\n"); |
1da177e4 LT |
910 | return ret; |
911 | } | |
912 | ||
913 | static void netem_destroy(struct Qdisc *sch) | |
914 | { | |
915 | struct netem_sched_data *q = qdisc_priv(sch); | |
916 | ||
59cb5c67 | 917 | qdisc_watchdog_cancel(&q->watchdog); |
50612537 ED |
918 | if (q->qdisc) |
919 | qdisc_destroy(q->qdisc); | |
6373a9a2 | 920 | dist_free(q->delay_dist); |
1da177e4 LT |
921 | } |
922 | ||
661b7972 | 923 | static int dump_loss_model(const struct netem_sched_data *q, |
924 | struct sk_buff *skb) | |
925 | { | |
926 | struct nlattr *nest; | |
927 | ||
928 | nest = nla_nest_start(skb, TCA_NETEM_LOSS); | |
929 | if (nest == NULL) | |
930 | goto nla_put_failure; | |
931 | ||
932 | switch (q->loss_model) { | |
933 | case CLG_RANDOM: | |
934 | /* legacy loss model */ | |
935 | nla_nest_cancel(skb, nest); | |
936 | return 0; /* no data */ | |
937 | ||
938 | case CLG_4_STATES: { | |
939 | struct tc_netem_gimodel gi = { | |
940 | .p13 = q->clg.a1, | |
941 | .p31 = q->clg.a2, | |
942 | .p32 = q->clg.a3, | |
943 | .p14 = q->clg.a4, | |
944 | .p23 = q->clg.a5, | |
945 | }; | |
946 | ||
1b34ec43 DM |
947 | if (nla_put(skb, NETEM_LOSS_GI, sizeof(gi), &gi)) |
948 | goto nla_put_failure; | |
661b7972 | 949 | break; |
950 | } | |
951 | case CLG_GILB_ELL: { | |
952 | struct tc_netem_gemodel ge = { | |
953 | .p = q->clg.a1, | |
954 | .r = q->clg.a2, | |
955 | .h = q->clg.a3, | |
956 | .k1 = q->clg.a4, | |
957 | }; | |
958 | ||
1b34ec43 DM |
959 | if (nla_put(skb, NETEM_LOSS_GE, sizeof(ge), &ge)) |
960 | goto nla_put_failure; | |
661b7972 | 961 | break; |
962 | } | |
963 | } | |
964 | ||
965 | nla_nest_end(skb, nest); | |
966 | return 0; | |
967 | ||
968 | nla_put_failure: | |
969 | nla_nest_cancel(skb, nest); | |
970 | return -1; | |
971 | } | |
972 | ||
1da177e4 LT |
973 | static int netem_dump(struct Qdisc *sch, struct sk_buff *skb) |
974 | { | |
975 | const struct netem_sched_data *q = qdisc_priv(sch); | |
861d7f74 | 976 | struct nlattr *nla = (struct nlattr *) skb_tail_pointer(skb); |
1da177e4 LT |
977 | struct tc_netem_qopt qopt; |
978 | struct tc_netem_corr cor; | |
0dca51d3 | 979 | struct tc_netem_reorder reorder; |
c865e5d9 | 980 | struct tc_netem_corrupt corrupt; |
7bc0f28c | 981 | struct tc_netem_rate rate; |
1da177e4 LT |
982 | |
983 | qopt.latency = q->latency; | |
984 | qopt.jitter = q->jitter; | |
985 | qopt.limit = q->limit; | |
986 | qopt.loss = q->loss; | |
987 | qopt.gap = q->gap; | |
988 | qopt.duplicate = q->duplicate; | |
1b34ec43 DM |
989 | if (nla_put(skb, TCA_OPTIONS, sizeof(qopt), &qopt)) |
990 | goto nla_put_failure; | |
1da177e4 LT |
991 | |
992 | cor.delay_corr = q->delay_cor.rho; | |
993 | cor.loss_corr = q->loss_cor.rho; | |
994 | cor.dup_corr = q->dup_cor.rho; | |
1b34ec43 DM |
995 | if (nla_put(skb, TCA_NETEM_CORR, sizeof(cor), &cor)) |
996 | goto nla_put_failure; | |
0dca51d3 SH |
997 | |
998 | reorder.probability = q->reorder; | |
999 | reorder.correlation = q->reorder_cor.rho; | |
1b34ec43 DM |
1000 | if (nla_put(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder)) |
1001 | goto nla_put_failure; | |
0dca51d3 | 1002 | |
c865e5d9 SH |
1003 | corrupt.probability = q->corrupt; |
1004 | corrupt.correlation = q->corrupt_cor.rho; | |
1b34ec43 DM |
1005 | if (nla_put(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt)) |
1006 | goto nla_put_failure; | |
c865e5d9 | 1007 | |
6a031f67 YY |
1008 | if (q->rate >= (1ULL << 32)) { |
1009 | if (nla_put_u64(skb, TCA_NETEM_RATE64, q->rate)) | |
1010 | goto nla_put_failure; | |
1011 | rate.rate = ~0U; | |
1012 | } else { | |
1013 | rate.rate = q->rate; | |
1014 | } | |
90b41a1c HPP |
1015 | rate.packet_overhead = q->packet_overhead; |
1016 | rate.cell_size = q->cell_size; | |
1017 | rate.cell_overhead = q->cell_overhead; | |
1b34ec43 DM |
1018 | if (nla_put(skb, TCA_NETEM_RATE, sizeof(rate), &rate)) |
1019 | goto nla_put_failure; | |
7bc0f28c | 1020 | |
e4ae004b ED |
1021 | if (q->ecn && nla_put_u32(skb, TCA_NETEM_ECN, q->ecn)) |
1022 | goto nla_put_failure; | |
1023 | ||
661b7972 | 1024 | if (dump_loss_model(q, skb) != 0) |
1025 | goto nla_put_failure; | |
1026 | ||
861d7f74 | 1027 | return nla_nest_end(skb, nla); |
1da177e4 | 1028 | |
1e90474c | 1029 | nla_put_failure: |
861d7f74 | 1030 | nlmsg_trim(skb, nla); |
1da177e4 LT |
1031 | return -1; |
1032 | } | |
1033 | ||
10f6dfcf | 1034 | static int netem_dump_class(struct Qdisc *sch, unsigned long cl, |
1035 | struct sk_buff *skb, struct tcmsg *tcm) | |
1036 | { | |
1037 | struct netem_sched_data *q = qdisc_priv(sch); | |
1038 | ||
50612537 | 1039 | if (cl != 1 || !q->qdisc) /* only one class */ |
10f6dfcf | 1040 | return -ENOENT; |
1041 | ||
1042 | tcm->tcm_handle |= TC_H_MIN(1); | |
1043 | tcm->tcm_info = q->qdisc->handle; | |
1044 | ||
1045 | return 0; | |
1046 | } | |
1047 | ||
1048 | static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
1049 | struct Qdisc **old) | |
1050 | { | |
1051 | struct netem_sched_data *q = qdisc_priv(sch); | |
1052 | ||
10f6dfcf | 1053 | sch_tree_lock(sch); |
1054 | *old = q->qdisc; | |
1055 | q->qdisc = new; | |
50612537 ED |
1056 | if (*old) { |
1057 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); | |
1058 | qdisc_reset(*old); | |
1059 | } | |
10f6dfcf | 1060 | sch_tree_unlock(sch); |
1061 | ||
1062 | return 0; | |
1063 | } | |
1064 | ||
1065 | static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg) | |
1066 | { | |
1067 | struct netem_sched_data *q = qdisc_priv(sch); | |
1068 | return q->qdisc; | |
1069 | } | |
1070 | ||
1071 | static unsigned long netem_get(struct Qdisc *sch, u32 classid) | |
1072 | { | |
1073 | return 1; | |
1074 | } | |
1075 | ||
1076 | static void netem_put(struct Qdisc *sch, unsigned long arg) | |
1077 | { | |
1078 | } | |
1079 | ||
1080 | static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker) | |
1081 | { | |
1082 | if (!walker->stop) { | |
1083 | if (walker->count >= walker->skip) | |
1084 | if (walker->fn(sch, 1, walker) < 0) { | |
1085 | walker->stop = 1; | |
1086 | return; | |
1087 | } | |
1088 | walker->count++; | |
1089 | } | |
1090 | } | |
1091 | ||
1092 | static const struct Qdisc_class_ops netem_class_ops = { | |
1093 | .graft = netem_graft, | |
1094 | .leaf = netem_leaf, | |
1095 | .get = netem_get, | |
1096 | .put = netem_put, | |
1097 | .walk = netem_walk, | |
1098 | .dump = netem_dump_class, | |
1099 | }; | |
1100 | ||
20fea08b | 1101 | static struct Qdisc_ops netem_qdisc_ops __read_mostly = { |
1da177e4 | 1102 | .id = "netem", |
10f6dfcf | 1103 | .cl_ops = &netem_class_ops, |
1da177e4 LT |
1104 | .priv_size = sizeof(struct netem_sched_data), |
1105 | .enqueue = netem_enqueue, | |
1106 | .dequeue = netem_dequeue, | |
77be155c | 1107 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
1108 | .drop = netem_drop, |
1109 | .init = netem_init, | |
1110 | .reset = netem_reset, | |
1111 | .destroy = netem_destroy, | |
1112 | .change = netem_change, | |
1113 | .dump = netem_dump, | |
1114 | .owner = THIS_MODULE, | |
1115 | }; | |
1116 | ||
1117 | ||
1118 | static int __init netem_module_init(void) | |
1119 | { | |
eb229c4c | 1120 | pr_info("netem: version " VERSION "\n"); |
1da177e4 LT |
1121 | return register_qdisc(&netem_qdisc_ops); |
1122 | } | |
1123 | static void __exit netem_module_exit(void) | |
1124 | { | |
1125 | unregister_qdisc(&netem_qdisc_ops); | |
1126 | } | |
1127 | module_init(netem_module_init) | |
1128 | module_exit(netem_module_exit) | |
1129 | MODULE_LICENSE("GPL"); |