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 | ||
0a9fe5c3 YS |
71 | struct disttable { |
72 | u32 size; | |
73 | s16 table[0]; | |
74 | }; | |
75 | ||
1da177e4 | 76 | struct netem_sched_data { |
aec0a40a ED |
77 | /* internal t(ime)fifo qdisc uses t_root and sch->limit */ |
78 | struct rb_root t_root; | |
50612537 ED |
79 | |
80 | /* optional qdisc for classful handling (NULL at netem init) */ | |
1da177e4 | 81 | struct Qdisc *qdisc; |
50612537 | 82 | |
59cb5c67 | 83 | struct qdisc_watchdog watchdog; |
1da177e4 | 84 | |
112f9cb6 DT |
85 | s64 latency; |
86 | s64 jitter; | |
b407621c | 87 | |
1da177e4 | 88 | u32 loss; |
e4ae004b | 89 | u32 ecn; |
1da177e4 LT |
90 | u32 limit; |
91 | u32 counter; | |
92 | u32 gap; | |
1da177e4 | 93 | u32 duplicate; |
0dca51d3 | 94 | u32 reorder; |
c865e5d9 | 95 | u32 corrupt; |
6a031f67 | 96 | u64 rate; |
90b41a1c HPP |
97 | s32 packet_overhead; |
98 | u32 cell_size; | |
809fa972 | 99 | struct reciprocal_value cell_size_reciprocal; |
90b41a1c | 100 | s32 cell_overhead; |
1da177e4 LT |
101 | |
102 | struct crndstate { | |
b407621c SH |
103 | u32 last; |
104 | u32 rho; | |
c865e5d9 | 105 | } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor; |
1da177e4 | 106 | |
0a9fe5c3 | 107 | struct disttable *delay_dist; |
661b7972 | 108 | |
109 | enum { | |
110 | CLG_RANDOM, | |
111 | CLG_4_STATES, | |
112 | CLG_GILB_ELL, | |
113 | } loss_model; | |
114 | ||
a6e2fe17 YY |
115 | enum { |
116 | TX_IN_GAP_PERIOD = 1, | |
117 | TX_IN_BURST_PERIOD, | |
118 | LOST_IN_GAP_PERIOD, | |
119 | LOST_IN_BURST_PERIOD, | |
120 | } _4_state_model; | |
121 | ||
c045a734 YY |
122 | enum { |
123 | GOOD_STATE = 1, | |
124 | BAD_STATE, | |
125 | } GE_state_model; | |
126 | ||
661b7972 | 127 | /* Correlated Loss Generation models */ |
128 | struct clgstate { | |
129 | /* state of the Markov chain */ | |
130 | u8 state; | |
131 | ||
132 | /* 4-states and Gilbert-Elliot models */ | |
133 | u32 a1; /* p13 for 4-states or p for GE */ | |
134 | u32 a2; /* p31 for 4-states or r for GE */ | |
135 | u32 a3; /* p32 for 4-states or h for GE */ | |
136 | u32 a4; /* p14 for 4-states or 1-k for GE */ | |
137 | u32 a5; /* p23 used only in 4-states */ | |
138 | } clg; | |
139 | ||
836af83b DT |
140 | struct tc_netem_slot slot_config; |
141 | struct slotstate { | |
142 | u64 slot_next; | |
143 | s32 packets_left; | |
144 | s32 bytes_left; | |
145 | } slot; | |
146 | ||
0a9fe5c3 | 147 | struct disttable *slot_dist; |
1da177e4 LT |
148 | }; |
149 | ||
50612537 ED |
150 | /* Time stamp put into socket buffer control block |
151 | * Only valid when skbs are in our internal t(ime)fifo queue. | |
56b17425 ED |
152 | * |
153 | * As skb->rbnode uses same storage than skb->next, skb->prev and skb->tstamp, | |
154 | * and skb->next & skb->prev are scratch space for a qdisc, | |
155 | * we save skb->tstamp value in skb->cb[] before destroying it. | |
50612537 | 156 | */ |
1da177e4 | 157 | struct netem_skb_cb { |
112f9cb6 | 158 | u64 time_to_send; |
1da177e4 LT |
159 | }; |
160 | ||
5f86173b JK |
161 | static inline struct netem_skb_cb *netem_skb_cb(struct sk_buff *skb) |
162 | { | |
aec0a40a | 163 | /* we assume we can use skb next/prev/tstamp as storage for rb_node */ |
16bda13d | 164 | qdisc_cb_private_validate(skb, sizeof(struct netem_skb_cb)); |
175f9c1b | 165 | return (struct netem_skb_cb *)qdisc_skb_cb(skb)->data; |
5f86173b JK |
166 | } |
167 | ||
1da177e4 LT |
168 | /* init_crandom - initialize correlated random number generator |
169 | * Use entropy source for initial seed. | |
170 | */ | |
171 | static void init_crandom(struct crndstate *state, unsigned long rho) | |
172 | { | |
173 | state->rho = rho; | |
63862b5b | 174 | state->last = prandom_u32(); |
1da177e4 LT |
175 | } |
176 | ||
177 | /* get_crandom - correlated random number generator | |
178 | * Next number depends on last value. | |
179 | * rho is scaled to avoid floating point. | |
180 | */ | |
b407621c | 181 | static u32 get_crandom(struct crndstate *state) |
1da177e4 LT |
182 | { |
183 | u64 value, rho; | |
184 | unsigned long answer; | |
185 | ||
0a9fe5c3 | 186 | if (!state || state->rho == 0) /* no correlation */ |
63862b5b | 187 | return prandom_u32(); |
1da177e4 | 188 | |
63862b5b | 189 | value = prandom_u32(); |
1da177e4 LT |
190 | rho = (u64)state->rho + 1; |
191 | answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32; | |
192 | state->last = answer; | |
193 | return answer; | |
194 | } | |
195 | ||
661b7972 | 196 | /* loss_4state - 4-state model loss generator |
197 | * Generates losses according to the 4-state Markov chain adopted in | |
198 | * the GI (General and Intuitive) loss model. | |
199 | */ | |
200 | static bool loss_4state(struct netem_sched_data *q) | |
201 | { | |
202 | struct clgstate *clg = &q->clg; | |
63862b5b | 203 | u32 rnd = prandom_u32(); |
661b7972 | 204 | |
205 | /* | |
25985edc | 206 | * Makes a comparison between rnd and the transition |
661b7972 | 207 | * probabilities outgoing from the current state, then decides the |
208 | * next state and if the next packet has to be transmitted or lost. | |
209 | * The four states correspond to: | |
a6e2fe17 YY |
210 | * TX_IN_GAP_PERIOD => successfully transmitted packets within a gap period |
211 | * LOST_IN_BURST_PERIOD => isolated losses within a gap period | |
212 | * LOST_IN_GAP_PERIOD => lost packets within a burst period | |
213 | * TX_IN_GAP_PERIOD => successfully transmitted packets within a burst period | |
661b7972 | 214 | */ |
215 | switch (clg->state) { | |
a6e2fe17 | 216 | case TX_IN_GAP_PERIOD: |
661b7972 | 217 | if (rnd < clg->a4) { |
a6e2fe17 | 218 | clg->state = LOST_IN_BURST_PERIOD; |
661b7972 | 219 | return true; |
ab6c27be | 220 | } else if (clg->a4 < rnd && rnd < clg->a1 + clg->a4) { |
a6e2fe17 | 221 | clg->state = LOST_IN_GAP_PERIOD; |
661b7972 | 222 | return true; |
a6e2fe17 YY |
223 | } else if (clg->a1 + clg->a4 < rnd) { |
224 | clg->state = TX_IN_GAP_PERIOD; | |
225 | } | |
661b7972 | 226 | |
227 | break; | |
a6e2fe17 | 228 | case TX_IN_BURST_PERIOD: |
661b7972 | 229 | if (rnd < clg->a5) { |
a6e2fe17 | 230 | clg->state = LOST_IN_GAP_PERIOD; |
661b7972 | 231 | return true; |
a6e2fe17 YY |
232 | } else { |
233 | clg->state = TX_IN_BURST_PERIOD; | |
234 | } | |
661b7972 | 235 | |
236 | break; | |
a6e2fe17 | 237 | case LOST_IN_GAP_PERIOD: |
661b7972 | 238 | if (rnd < clg->a3) |
a6e2fe17 | 239 | clg->state = TX_IN_BURST_PERIOD; |
661b7972 | 240 | else if (clg->a3 < rnd && rnd < clg->a2 + clg->a3) { |
a6e2fe17 | 241 | clg->state = TX_IN_GAP_PERIOD; |
661b7972 | 242 | } else if (clg->a2 + clg->a3 < rnd) { |
a6e2fe17 | 243 | clg->state = LOST_IN_GAP_PERIOD; |
661b7972 | 244 | return true; |
245 | } | |
246 | break; | |
a6e2fe17 YY |
247 | case LOST_IN_BURST_PERIOD: |
248 | clg->state = TX_IN_GAP_PERIOD; | |
661b7972 | 249 | break; |
250 | } | |
251 | ||
252 | return false; | |
253 | } | |
254 | ||
255 | /* loss_gilb_ell - Gilbert-Elliot model loss generator | |
256 | * Generates losses according to the Gilbert-Elliot loss model or | |
257 | * its special cases (Gilbert or Simple Gilbert) | |
258 | * | |
25985edc | 259 | * Makes a comparison between random number and the transition |
661b7972 | 260 | * probabilities outgoing from the current state, then decides the |
25985edc | 261 | * next state. A second random number is extracted and the comparison |
661b7972 | 262 | * with the loss probability of the current state decides if the next |
263 | * packet will be transmitted or lost. | |
264 | */ | |
265 | static bool loss_gilb_ell(struct netem_sched_data *q) | |
266 | { | |
267 | struct clgstate *clg = &q->clg; | |
268 | ||
269 | switch (clg->state) { | |
c045a734 | 270 | case GOOD_STATE: |
63862b5b | 271 | if (prandom_u32() < clg->a1) |
c045a734 | 272 | clg->state = BAD_STATE; |
63862b5b | 273 | if (prandom_u32() < clg->a4) |
661b7972 | 274 | return true; |
7c2781fa | 275 | break; |
c045a734 | 276 | case BAD_STATE: |
63862b5b | 277 | if (prandom_u32() < clg->a2) |
c045a734 | 278 | clg->state = GOOD_STATE; |
63862b5b | 279 | if (prandom_u32() > clg->a3) |
661b7972 | 280 | return true; |
281 | } | |
282 | ||
283 | return false; | |
284 | } | |
285 | ||
286 | static bool loss_event(struct netem_sched_data *q) | |
287 | { | |
288 | switch (q->loss_model) { | |
289 | case CLG_RANDOM: | |
290 | /* Random packet drop 0 => none, ~0 => all */ | |
291 | return q->loss && q->loss >= get_crandom(&q->loss_cor); | |
292 | ||
293 | case CLG_4_STATES: | |
294 | /* 4state loss model algorithm (used also for GI model) | |
295 | * Extracts a value from the markov 4 state loss generator, | |
296 | * if it is 1 drops a packet and if needed writes the event in | |
297 | * the kernel logs | |
298 | */ | |
299 | return loss_4state(q); | |
300 | ||
301 | case CLG_GILB_ELL: | |
302 | /* Gilbert-Elliot loss model algorithm | |
303 | * Extracts a value from the Gilbert-Elliot loss generator, | |
304 | * if it is 1 drops a packet and if needed writes the event in | |
305 | * the kernel logs | |
306 | */ | |
307 | return loss_gilb_ell(q); | |
308 | } | |
309 | ||
310 | return false; /* not reached */ | |
311 | } | |
312 | ||
313 | ||
1da177e4 LT |
314 | /* tabledist - return a pseudo-randomly distributed value with mean mu and |
315 | * std deviation sigma. Uses table lookup to approximate the desired | |
316 | * distribution, and a uniformly-distributed pseudo-random source. | |
317 | */ | |
9b0ed891 | 318 | static s64 tabledist(s64 mu, s32 sigma, |
112f9cb6 | 319 | struct crndstate *state, |
9b0ed891 | 320 | const struct disttable *dist) |
1da177e4 | 321 | { |
112f9cb6 | 322 | s64 x; |
b407621c SH |
323 | long t; |
324 | u32 rnd; | |
1da177e4 LT |
325 | |
326 | if (sigma == 0) | |
327 | return mu; | |
328 | ||
329 | rnd = get_crandom(state); | |
330 | ||
331 | /* default uniform distribution */ | |
10297b99 | 332 | if (dist == NULL) |
043e337f | 333 | return ((rnd % (2 * sigma)) + mu) - sigma; |
1da177e4 LT |
334 | |
335 | t = dist->table[rnd % dist->size]; | |
336 | x = (sigma % NETEM_DIST_SCALE) * t; | |
337 | if (x >= 0) | |
338 | x += NETEM_DIST_SCALE/2; | |
339 | else | |
340 | x -= NETEM_DIST_SCALE/2; | |
341 | ||
342 | return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu; | |
343 | } | |
344 | ||
bce552fd | 345 | static u64 packet_time_ns(u64 len, const struct netem_sched_data *q) |
7bc0f28c | 346 | { |
90b41a1c HPP |
347 | len += q->packet_overhead; |
348 | ||
349 | if (q->cell_size) { | |
350 | u32 cells = reciprocal_divide(len, q->cell_size_reciprocal); | |
351 | ||
352 | if (len > cells * q->cell_size) /* extra cell needed for remainder */ | |
353 | cells++; | |
354 | len = cells * (q->cell_size + q->cell_overhead); | |
355 | } | |
bce552fd SH |
356 | |
357 | return div64_u64(len * NSEC_PER_SEC, q->rate); | |
7bc0f28c HPP |
358 | } |
359 | ||
ff704050 | 360 | static void tfifo_reset(struct Qdisc *sch) |
361 | { | |
362 | struct netem_sched_data *q = qdisc_priv(sch); | |
3aa605f2 | 363 | struct rb_node *p = rb_first(&q->t_root); |
ff704050 | 364 | |
3aa605f2 | 365 | while (p) { |
18a4c0ea | 366 | struct sk_buff *skb = rb_to_skb(p); |
ff704050 | 367 | |
3aa605f2 ED |
368 | p = rb_next(p); |
369 | rb_erase(&skb->rbnode, &q->t_root); | |
2f08a9a1 | 370 | rtnl_kfree_skbs(skb, skb); |
ff704050 | 371 | } |
372 | } | |
373 | ||
960fb66e | 374 | static void tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch) |
50612537 | 375 | { |
aec0a40a | 376 | struct netem_sched_data *q = qdisc_priv(sch); |
112f9cb6 | 377 | u64 tnext = netem_skb_cb(nskb)->time_to_send; |
aec0a40a | 378 | struct rb_node **p = &q->t_root.rb_node, *parent = NULL; |
50612537 | 379 | |
aec0a40a ED |
380 | while (*p) { |
381 | struct sk_buff *skb; | |
50612537 | 382 | |
aec0a40a | 383 | parent = *p; |
18a4c0ea | 384 | skb = rb_to_skb(parent); |
960fb66e | 385 | if (tnext >= netem_skb_cb(skb)->time_to_send) |
aec0a40a ED |
386 | p = &parent->rb_right; |
387 | else | |
388 | p = &parent->rb_left; | |
50612537 | 389 | } |
56b17425 ED |
390 | rb_link_node(&nskb->rbnode, parent, p); |
391 | rb_insert_color(&nskb->rbnode, &q->t_root); | |
aec0a40a | 392 | sch->q.qlen++; |
50612537 ED |
393 | } |
394 | ||
6071bd1a NH |
395 | /* netem can't properly corrupt a megapacket (like we get from GSO), so instead |
396 | * when we statistically choose to corrupt one, we instead segment it, returning | |
397 | * the first packet to be corrupted, and re-enqueue the remaining frames | |
398 | */ | |
520ac30f ED |
399 | static struct sk_buff *netem_segment(struct sk_buff *skb, struct Qdisc *sch, |
400 | struct sk_buff **to_free) | |
6071bd1a NH |
401 | { |
402 | struct sk_buff *segs; | |
403 | netdev_features_t features = netif_skb_features(skb); | |
404 | ||
405 | segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK); | |
406 | ||
407 | if (IS_ERR_OR_NULL(segs)) { | |
520ac30f | 408 | qdisc_drop(skb, sch, to_free); |
6071bd1a NH |
409 | return NULL; |
410 | } | |
411 | consume_skb(skb); | |
412 | return segs; | |
413 | } | |
414 | ||
0afb51e7 SH |
415 | /* |
416 | * Insert one skb into qdisc. | |
417 | * Note: parent depends on return value to account for queue length. | |
418 | * NET_XMIT_DROP: queue length didn't change. | |
419 | * NET_XMIT_SUCCESS: one skb was queued. | |
420 | */ | |
520ac30f ED |
421 | static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch, |
422 | struct sk_buff **to_free) | |
1da177e4 LT |
423 | { |
424 | struct netem_sched_data *q = qdisc_priv(sch); | |
89e1df74 GC |
425 | /* We don't fill cb now as skb_unshare() may invalidate it */ |
426 | struct netem_skb_cb *cb; | |
0afb51e7 | 427 | struct sk_buff *skb2; |
6071bd1a NH |
428 | struct sk_buff *segs = NULL; |
429 | unsigned int len = 0, last_len, prev_len = qdisc_pkt_len(skb); | |
430 | int nb = 0; | |
0afb51e7 | 431 | int count = 1; |
6071bd1a | 432 | int rc = NET_XMIT_SUCCESS; |
1da177e4 | 433 | |
0afb51e7 SH |
434 | /* Random duplication */ |
435 | if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor)) | |
436 | ++count; | |
437 | ||
661b7972 | 438 | /* Drop packet? */ |
e4ae004b ED |
439 | if (loss_event(q)) { |
440 | if (q->ecn && INET_ECN_set_ce(skb)) | |
25331d6c | 441 | qdisc_qstats_drop(sch); /* mark packet */ |
e4ae004b ED |
442 | else |
443 | --count; | |
444 | } | |
0afb51e7 | 445 | if (count == 0) { |
25331d6c | 446 | qdisc_qstats_drop(sch); |
520ac30f | 447 | __qdisc_drop(skb, to_free); |
c27f339a | 448 | return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
1da177e4 LT |
449 | } |
450 | ||
5a308f40 ED |
451 | /* If a delay is expected, orphan the skb. (orphaning usually takes |
452 | * place at TX completion time, so _before_ the link transit delay) | |
5a308f40 | 453 | */ |
5080f39e | 454 | if (q->latency || q->jitter || q->rate) |
f2f872f9 | 455 | skb_orphan_partial(skb); |
4e8a5201 | 456 | |
0afb51e7 SH |
457 | /* |
458 | * If we need to duplicate packet, then re-insert at top of the | |
459 | * qdisc tree, since parent queuer expects that only one | |
460 | * skb will be queued. | |
461 | */ | |
462 | if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) { | |
7698b4fc | 463 | struct Qdisc *rootq = qdisc_root(sch); |
0afb51e7 | 464 | u32 dupsave = q->duplicate; /* prevent duplicating a dup... */ |
0afb51e7 | 465 | |
b396cca6 | 466 | q->duplicate = 0; |
520ac30f | 467 | rootq->enqueue(skb2, rootq, to_free); |
0afb51e7 | 468 | q->duplicate = dupsave; |
1da177e4 LT |
469 | } |
470 | ||
c865e5d9 SH |
471 | /* |
472 | * Randomized packet corruption. | |
473 | * Make copy if needed since we are modifying | |
474 | * If packet is going to be hardware checksummed, then | |
475 | * do it now in software before we mangle it. | |
476 | */ | |
477 | if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) { | |
6071bd1a | 478 | if (skb_is_gso(skb)) { |
520ac30f | 479 | segs = netem_segment(skb, sch, to_free); |
6071bd1a NH |
480 | if (!segs) |
481 | return NET_XMIT_DROP; | |
482 | } else { | |
483 | segs = skb; | |
484 | } | |
485 | ||
486 | skb = segs; | |
487 | segs = segs->next; | |
488 | ||
8a6e9c67 ED |
489 | skb = skb_unshare(skb, GFP_ATOMIC); |
490 | if (unlikely(!skb)) { | |
491 | qdisc_qstats_drop(sch); | |
492 | goto finish_segs; | |
493 | } | |
494 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
495 | skb_checksum_help(skb)) { | |
496 | qdisc_drop(skb, sch, to_free); | |
6071bd1a NH |
497 | goto finish_segs; |
498 | } | |
c865e5d9 | 499 | |
63862b5b AH |
500 | skb->data[prandom_u32() % skb_headlen(skb)] ^= |
501 | 1<<(prandom_u32() % 8); | |
c865e5d9 SH |
502 | } |
503 | ||
97d0678f | 504 | if (unlikely(sch->q.qlen >= sch->limit)) |
35d889d1 | 505 | return qdisc_drop_all(skb, sch, to_free); |
960fb66e | 506 | |
25331d6c | 507 | qdisc_qstats_backlog_inc(sch, skb); |
960fb66e | 508 | |
5f86173b | 509 | cb = netem_skb_cb(skb); |
cc7ec456 | 510 | if (q->gap == 0 || /* not doing reordering */ |
a42b4799 | 511 | q->counter < q->gap - 1 || /* inside last reordering gap */ |
f64f9e71 | 512 | q->reorder < get_crandom(&q->reorder_cor)) { |
112f9cb6 DT |
513 | u64 now; |
514 | s64 delay; | |
07aaa115 SH |
515 | |
516 | delay = tabledist(q->latency, q->jitter, | |
517 | &q->delay_cor, q->delay_dist); | |
518 | ||
112f9cb6 | 519 | now = ktime_get_ns(); |
7bc0f28c HPP |
520 | |
521 | if (q->rate) { | |
5080f39e NU |
522 | struct netem_skb_cb *last = NULL; |
523 | ||
524 | if (sch->q.tail) | |
525 | last = netem_skb_cb(sch->q.tail); | |
526 | if (q->t_root.rb_node) { | |
527 | struct sk_buff *t_skb; | |
528 | struct netem_skb_cb *t_last; | |
529 | ||
18a4c0ea | 530 | t_skb = skb_rb_last(&q->t_root); |
5080f39e NU |
531 | t_last = netem_skb_cb(t_skb); |
532 | if (!last || | |
533 | t_last->time_to_send > last->time_to_send) { | |
534 | last = t_last; | |
535 | } | |
536 | } | |
7bc0f28c | 537 | |
aec0a40a | 538 | if (last) { |
7bc0f28c | 539 | /* |
a13d3104 JN |
540 | * Last packet in queue is reference point (now), |
541 | * calculate this time bonus and subtract | |
7bc0f28c HPP |
542 | * from delay. |
543 | */ | |
5080f39e | 544 | delay -= last->time_to_send - now; |
112f9cb6 | 545 | delay = max_t(s64, 0, delay); |
5080f39e | 546 | now = last->time_to_send; |
7bc0f28c | 547 | } |
a13d3104 | 548 | |
bce552fd | 549 | delay += packet_time_ns(qdisc_pkt_len(skb), q); |
7bc0f28c HPP |
550 | } |
551 | ||
7c59e25f | 552 | cb->time_to_send = now + delay; |
1da177e4 | 553 | ++q->counter; |
960fb66e | 554 | tfifo_enqueue(skb, sch); |
1da177e4 | 555 | } else { |
10297b99 | 556 | /* |
0dca51d3 SH |
557 | * Do re-ordering by putting one out of N packets at the front |
558 | * of the queue. | |
559 | */ | |
112f9cb6 | 560 | cb->time_to_send = ktime_get_ns(); |
0dca51d3 | 561 | q->counter = 0; |
8ba25dad | 562 | |
59697730 | 563 | __qdisc_enqueue_head(skb, &sch->q); |
eb101924 | 564 | sch->qstats.requeues++; |
378a2f09 | 565 | } |
1da177e4 | 566 | |
6071bd1a NH |
567 | finish_segs: |
568 | if (segs) { | |
569 | while (segs) { | |
570 | skb2 = segs->next; | |
a8305bff | 571 | skb_mark_not_on_list(segs); |
6071bd1a NH |
572 | qdisc_skb_cb(segs)->pkt_len = segs->len; |
573 | last_len = segs->len; | |
520ac30f | 574 | rc = qdisc_enqueue(segs, sch, to_free); |
6071bd1a NH |
575 | if (rc != NET_XMIT_SUCCESS) { |
576 | if (net_xmit_drop_count(rc)) | |
577 | qdisc_qstats_drop(sch); | |
578 | } else { | |
579 | nb++; | |
580 | len += last_len; | |
581 | } | |
582 | segs = skb2; | |
583 | } | |
584 | sch->q.qlen += nb; | |
585 | if (nb > 1) | |
586 | qdisc_tree_reduce_backlog(sch, 1 - nb, prev_len - len); | |
587 | } | |
10f6dfcf | 588 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
589 | } |
590 | ||
836af83b DT |
591 | /* Delay the next round with a new future slot with a |
592 | * correct number of bytes and packets. | |
593 | */ | |
594 | ||
595 | static void get_slot_next(struct netem_sched_data *q, u64 now) | |
596 | { | |
0a9fe5c3 YS |
597 | s64 next_delay; |
598 | ||
599 | if (!q->slot_dist) | |
600 | next_delay = q->slot_config.min_delay + | |
601 | (prandom_u32() * | |
602 | (q->slot_config.max_delay - | |
603 | q->slot_config.min_delay) >> 32); | |
604 | else | |
605 | next_delay = tabledist(q->slot_config.dist_delay, | |
606 | (s32)(q->slot_config.dist_jitter), | |
607 | NULL, q->slot_dist); | |
608 | ||
609 | q->slot.slot_next = now + next_delay; | |
836af83b DT |
610 | q->slot.packets_left = q->slot_config.max_packets; |
611 | q->slot.bytes_left = q->slot_config.max_bytes; | |
612 | } | |
613 | ||
1da177e4 LT |
614 | static struct sk_buff *netem_dequeue(struct Qdisc *sch) |
615 | { | |
616 | struct netem_sched_data *q = qdisc_priv(sch); | |
617 | struct sk_buff *skb; | |
aec0a40a | 618 | struct rb_node *p; |
1da177e4 | 619 | |
50612537 | 620 | tfifo_dequeue: |
ed760cb8 | 621 | skb = __qdisc_dequeue_head(&sch->q); |
771018e7 | 622 | if (skb) { |
25331d6c | 623 | qdisc_qstats_backlog_dec(sch, skb); |
0ad2a836 | 624 | deliver: |
aec0a40a ED |
625 | qdisc_bstats_update(sch, skb); |
626 | return skb; | |
627 | } | |
628 | p = rb_first(&q->t_root); | |
629 | if (p) { | |
112f9cb6 | 630 | u64 time_to_send; |
836af83b | 631 | u64 now = ktime_get_ns(); |
36b7bfe0 | 632 | |
18a4c0ea | 633 | skb = rb_to_skb(p); |
0f9f32ac SH |
634 | |
635 | /* if more time remaining? */ | |
36b7bfe0 | 636 | time_to_send = netem_skb_cb(skb)->time_to_send; |
836af83b DT |
637 | if (q->slot.slot_next && q->slot.slot_next < time_to_send) |
638 | get_slot_next(q, now); | |
aec0a40a | 639 | |
836af83b DT |
640 | if (time_to_send <= now && q->slot.slot_next <= now) { |
641 | rb_erase(p, &q->t_root); | |
aec0a40a | 642 | sch->q.qlen--; |
0ad2a836 | 643 | qdisc_qstats_backlog_dec(sch, skb); |
aec0a40a ED |
644 | skb->next = NULL; |
645 | skb->prev = NULL; | |
bffa72cf ED |
646 | /* skb->dev shares skb->rbnode area, |
647 | * we need to restore its value. | |
648 | */ | |
649 | skb->dev = qdisc_dev(sch); | |
03c05f0d | 650 | |
836af83b DT |
651 | if (q->slot.slot_next) { |
652 | q->slot.packets_left--; | |
653 | q->slot.bytes_left -= qdisc_pkt_len(skb); | |
654 | if (q->slot.packets_left <= 0 || | |
655 | q->slot.bytes_left <= 0) | |
656 | get_slot_next(q, now); | |
657 | } | |
658 | ||
50612537 | 659 | if (q->qdisc) { |
21de12ee | 660 | unsigned int pkt_len = qdisc_pkt_len(skb); |
520ac30f ED |
661 | struct sk_buff *to_free = NULL; |
662 | int err; | |
50612537 | 663 | |
520ac30f ED |
664 | err = qdisc_enqueue(skb, q->qdisc, &to_free); |
665 | kfree_skb_list(to_free); | |
21de12ee ED |
666 | if (err != NET_XMIT_SUCCESS && |
667 | net_xmit_drop_count(err)) { | |
668 | qdisc_qstats_drop(sch); | |
669 | qdisc_tree_reduce_backlog(sch, 1, | |
670 | pkt_len); | |
50612537 ED |
671 | } |
672 | goto tfifo_dequeue; | |
673 | } | |
aec0a40a | 674 | goto deliver; |
07aaa115 | 675 | } |
11274e5a | 676 | |
50612537 ED |
677 | if (q->qdisc) { |
678 | skb = q->qdisc->ops->dequeue(q->qdisc); | |
679 | if (skb) | |
680 | goto deliver; | |
681 | } | |
836af83b DT |
682 | |
683 | qdisc_watchdog_schedule_ns(&q->watchdog, | |
684 | max(time_to_send, | |
685 | q->slot.slot_next)); | |
0f9f32ac SH |
686 | } |
687 | ||
50612537 ED |
688 | if (q->qdisc) { |
689 | skb = q->qdisc->ops->dequeue(q->qdisc); | |
690 | if (skb) | |
691 | goto deliver; | |
692 | } | |
0f9f32ac | 693 | return NULL; |
1da177e4 LT |
694 | } |
695 | ||
1da177e4 LT |
696 | static void netem_reset(struct Qdisc *sch) |
697 | { | |
698 | struct netem_sched_data *q = qdisc_priv(sch); | |
699 | ||
50612537 | 700 | qdisc_reset_queue(sch); |
ff704050 | 701 | tfifo_reset(sch); |
50612537 ED |
702 | if (q->qdisc) |
703 | qdisc_reset(q->qdisc); | |
59cb5c67 | 704 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
705 | } |
706 | ||
6373a9a2 | 707 | static void dist_free(struct disttable *d) |
708 | { | |
4cb28970 | 709 | kvfree(d); |
6373a9a2 | 710 | } |
711 | ||
1da177e4 LT |
712 | /* |
713 | * Distribution data is a variable size payload containing | |
714 | * signed 16 bit values. | |
715 | */ | |
836af83b | 716 | |
0a9fe5c3 YS |
717 | static int get_dist_table(struct Qdisc *sch, struct disttable **tbl, |
718 | const struct nlattr *attr) | |
1da177e4 | 719 | { |
6373a9a2 | 720 | size_t n = nla_len(attr)/sizeof(__s16); |
1e90474c | 721 | const __s16 *data = nla_data(attr); |
7698b4fc | 722 | spinlock_t *root_lock; |
1da177e4 LT |
723 | struct disttable *d; |
724 | int i; | |
725 | ||
df173bda | 726 | if (n > NETEM_DIST_MAX) |
1da177e4 LT |
727 | return -EINVAL; |
728 | ||
752ade68 | 729 | d = kvmalloc(sizeof(struct disttable) + n * sizeof(s16), GFP_KERNEL); |
1da177e4 LT |
730 | if (!d) |
731 | return -ENOMEM; | |
732 | ||
733 | d->size = n; | |
734 | for (i = 0; i < n; i++) | |
735 | d->table[i] = data[i]; | |
10297b99 | 736 | |
102396ae | 737 | root_lock = qdisc_root_sleeping_lock(sch); |
7698b4fc DM |
738 | |
739 | spin_lock_bh(root_lock); | |
0a9fe5c3 | 740 | swap(*tbl, d); |
7698b4fc | 741 | spin_unlock_bh(root_lock); |
bb52c7ac ED |
742 | |
743 | dist_free(d); | |
1da177e4 LT |
744 | return 0; |
745 | } | |
746 | ||
836af83b DT |
747 | static void get_slot(struct netem_sched_data *q, const struct nlattr *attr) |
748 | { | |
749 | const struct tc_netem_slot *c = nla_data(attr); | |
750 | ||
751 | q->slot_config = *c; | |
752 | if (q->slot_config.max_packets == 0) | |
753 | q->slot_config.max_packets = INT_MAX; | |
754 | if (q->slot_config.max_bytes == 0) | |
755 | q->slot_config.max_bytes = INT_MAX; | |
756 | q->slot.packets_left = q->slot_config.max_packets; | |
757 | q->slot.bytes_left = q->slot_config.max_bytes; | |
0a9fe5c3 YS |
758 | if (q->slot_config.min_delay | q->slot_config.max_delay | |
759 | q->slot_config.dist_jitter) | |
836af83b DT |
760 | q->slot.slot_next = ktime_get_ns(); |
761 | else | |
762 | q->slot.slot_next = 0; | |
763 | } | |
764 | ||
49545a77 | 765 | static void get_correlation(struct netem_sched_data *q, const struct nlattr *attr) |
1da177e4 | 766 | { |
1e90474c | 767 | const struct tc_netem_corr *c = nla_data(attr); |
1da177e4 | 768 | |
1da177e4 LT |
769 | init_crandom(&q->delay_cor, c->delay_corr); |
770 | init_crandom(&q->loss_cor, c->loss_corr); | |
771 | init_crandom(&q->dup_cor, c->dup_corr); | |
1da177e4 LT |
772 | } |
773 | ||
49545a77 | 774 | static void get_reorder(struct netem_sched_data *q, const struct nlattr *attr) |
0dca51d3 | 775 | { |
1e90474c | 776 | const struct tc_netem_reorder *r = nla_data(attr); |
0dca51d3 | 777 | |
0dca51d3 SH |
778 | q->reorder = r->probability; |
779 | init_crandom(&q->reorder_cor, r->correlation); | |
0dca51d3 SH |
780 | } |
781 | ||
49545a77 | 782 | static void get_corrupt(struct netem_sched_data *q, const struct nlattr *attr) |
c865e5d9 | 783 | { |
1e90474c | 784 | const struct tc_netem_corrupt *r = nla_data(attr); |
c865e5d9 | 785 | |
c865e5d9 SH |
786 | q->corrupt = r->probability; |
787 | init_crandom(&q->corrupt_cor, r->correlation); | |
c865e5d9 SH |
788 | } |
789 | ||
49545a77 | 790 | static void get_rate(struct netem_sched_data *q, const struct nlattr *attr) |
7bc0f28c | 791 | { |
7bc0f28c HPP |
792 | const struct tc_netem_rate *r = nla_data(attr); |
793 | ||
794 | q->rate = r->rate; | |
90b41a1c HPP |
795 | q->packet_overhead = r->packet_overhead; |
796 | q->cell_size = r->cell_size; | |
809fa972 | 797 | q->cell_overhead = r->cell_overhead; |
90b41a1c HPP |
798 | if (q->cell_size) |
799 | q->cell_size_reciprocal = reciprocal_value(q->cell_size); | |
809fa972 HFS |
800 | else |
801 | q->cell_size_reciprocal = (struct reciprocal_value) { 0 }; | |
7bc0f28c HPP |
802 | } |
803 | ||
49545a77 | 804 | static int get_loss_clg(struct netem_sched_data *q, const struct nlattr *attr) |
661b7972 | 805 | { |
661b7972 | 806 | const struct nlattr *la; |
807 | int rem; | |
808 | ||
809 | nla_for_each_nested(la, attr, rem) { | |
810 | u16 type = nla_type(la); | |
811 | ||
833fa743 | 812 | switch (type) { |
661b7972 | 813 | case NETEM_LOSS_GI: { |
814 | const struct tc_netem_gimodel *gi = nla_data(la); | |
815 | ||
2494654d | 816 | if (nla_len(la) < sizeof(struct tc_netem_gimodel)) { |
661b7972 | 817 | pr_info("netem: incorrect gi model size\n"); |
818 | return -EINVAL; | |
819 | } | |
820 | ||
821 | q->loss_model = CLG_4_STATES; | |
822 | ||
3fbac2a8 | 823 | q->clg.state = TX_IN_GAP_PERIOD; |
661b7972 | 824 | q->clg.a1 = gi->p13; |
825 | q->clg.a2 = gi->p31; | |
826 | q->clg.a3 = gi->p32; | |
827 | q->clg.a4 = gi->p14; | |
828 | q->clg.a5 = gi->p23; | |
829 | break; | |
830 | } | |
831 | ||
832 | case NETEM_LOSS_GE: { | |
833 | const struct tc_netem_gemodel *ge = nla_data(la); | |
834 | ||
2494654d | 835 | if (nla_len(la) < sizeof(struct tc_netem_gemodel)) { |
836 | pr_info("netem: incorrect ge model size\n"); | |
661b7972 | 837 | return -EINVAL; |
838 | } | |
839 | ||
840 | q->loss_model = CLG_GILB_ELL; | |
3fbac2a8 | 841 | q->clg.state = GOOD_STATE; |
661b7972 | 842 | q->clg.a1 = ge->p; |
843 | q->clg.a2 = ge->r; | |
844 | q->clg.a3 = ge->h; | |
845 | q->clg.a4 = ge->k1; | |
846 | break; | |
847 | } | |
848 | ||
849 | default: | |
850 | pr_info("netem: unknown loss type %u\n", type); | |
851 | return -EINVAL; | |
852 | } | |
853 | } | |
854 | ||
855 | return 0; | |
856 | } | |
857 | ||
27a3421e PM |
858 | static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = { |
859 | [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) }, | |
860 | [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) }, | |
861 | [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) }, | |
7bc0f28c | 862 | [TCA_NETEM_RATE] = { .len = sizeof(struct tc_netem_rate) }, |
661b7972 | 863 | [TCA_NETEM_LOSS] = { .type = NLA_NESTED }, |
e4ae004b | 864 | [TCA_NETEM_ECN] = { .type = NLA_U32 }, |
6a031f67 | 865 | [TCA_NETEM_RATE64] = { .type = NLA_U64 }, |
99803171 DT |
866 | [TCA_NETEM_LATENCY64] = { .type = NLA_S64 }, |
867 | [TCA_NETEM_JITTER64] = { .type = NLA_S64 }, | |
836af83b | 868 | [TCA_NETEM_SLOT] = { .len = sizeof(struct tc_netem_slot) }, |
27a3421e PM |
869 | }; |
870 | ||
2c10b32b TG |
871 | static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla, |
872 | const struct nla_policy *policy, int len) | |
873 | { | |
874 | int nested_len = nla_len(nla) - NLA_ALIGN(len); | |
875 | ||
661b7972 | 876 | if (nested_len < 0) { |
877 | pr_info("netem: invalid attributes len %d\n", nested_len); | |
2c10b32b | 878 | return -EINVAL; |
661b7972 | 879 | } |
880 | ||
2c10b32b TG |
881 | if (nested_len >= nla_attr_size(0)) |
882 | return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len), | |
fceb6435 | 883 | nested_len, policy, NULL); |
661b7972 | 884 | |
2c10b32b TG |
885 | memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1)); |
886 | return 0; | |
887 | } | |
888 | ||
c865e5d9 | 889 | /* Parse netlink message to set options */ |
2030721c AA |
890 | static int netem_change(struct Qdisc *sch, struct nlattr *opt, |
891 | struct netlink_ext_ack *extack) | |
1da177e4 LT |
892 | { |
893 | struct netem_sched_data *q = qdisc_priv(sch); | |
b03f4672 | 894 | struct nlattr *tb[TCA_NETEM_MAX + 1]; |
1da177e4 | 895 | struct tc_netem_qopt *qopt; |
54a4b05c YY |
896 | struct clgstate old_clg; |
897 | int old_loss_model = CLG_RANDOM; | |
1da177e4 | 898 | int ret; |
10297b99 | 899 | |
b03f4672 | 900 | if (opt == NULL) |
1da177e4 LT |
901 | return -EINVAL; |
902 | ||
2c10b32b TG |
903 | qopt = nla_data(opt); |
904 | ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt)); | |
b03f4672 PM |
905 | if (ret < 0) |
906 | return ret; | |
907 | ||
54a4b05c YY |
908 | /* backup q->clg and q->loss_model */ |
909 | old_clg = q->clg; | |
910 | old_loss_model = q->loss_model; | |
911 | ||
912 | if (tb[TCA_NETEM_LOSS]) { | |
49545a77 | 913 | ret = get_loss_clg(q, tb[TCA_NETEM_LOSS]); |
54a4b05c YY |
914 | if (ret) { |
915 | q->loss_model = old_loss_model; | |
916 | return ret; | |
917 | } | |
918 | } else { | |
919 | q->loss_model = CLG_RANDOM; | |
920 | } | |
921 | ||
922 | if (tb[TCA_NETEM_DELAY_DIST]) { | |
0a9fe5c3 YS |
923 | ret = get_dist_table(sch, &q->delay_dist, |
924 | tb[TCA_NETEM_DELAY_DIST]); | |
925 | if (ret) | |
926 | goto get_table_failure; | |
927 | } | |
928 | ||
929 | if (tb[TCA_NETEM_SLOT_DIST]) { | |
930 | ret = get_dist_table(sch, &q->slot_dist, | |
931 | tb[TCA_NETEM_SLOT_DIST]); | |
932 | if (ret) | |
933 | goto get_table_failure; | |
54a4b05c YY |
934 | } |
935 | ||
50612537 | 936 | sch->limit = qopt->limit; |
10297b99 | 937 | |
112f9cb6 DT |
938 | q->latency = PSCHED_TICKS2NS(qopt->latency); |
939 | q->jitter = PSCHED_TICKS2NS(qopt->jitter); | |
1da177e4 LT |
940 | q->limit = qopt->limit; |
941 | q->gap = qopt->gap; | |
0dca51d3 | 942 | q->counter = 0; |
1da177e4 LT |
943 | q->loss = qopt->loss; |
944 | q->duplicate = qopt->duplicate; | |
945 | ||
bb2f8cc0 SH |
946 | /* for compatibility with earlier versions. |
947 | * if gap is set, need to assume 100% probability | |
0dca51d3 | 948 | */ |
a362e0a7 SH |
949 | if (q->gap) |
950 | q->reorder = ~0; | |
0dca51d3 | 951 | |
265eb67f | 952 | if (tb[TCA_NETEM_CORR]) |
49545a77 | 953 | get_correlation(q, tb[TCA_NETEM_CORR]); |
1da177e4 | 954 | |
265eb67f | 955 | if (tb[TCA_NETEM_REORDER]) |
49545a77 | 956 | get_reorder(q, tb[TCA_NETEM_REORDER]); |
1da177e4 | 957 | |
265eb67f | 958 | if (tb[TCA_NETEM_CORRUPT]) |
49545a77 | 959 | get_corrupt(q, tb[TCA_NETEM_CORRUPT]); |
1da177e4 | 960 | |
7bc0f28c | 961 | if (tb[TCA_NETEM_RATE]) |
49545a77 | 962 | get_rate(q, tb[TCA_NETEM_RATE]); |
7bc0f28c | 963 | |
6a031f67 YY |
964 | if (tb[TCA_NETEM_RATE64]) |
965 | q->rate = max_t(u64, q->rate, | |
966 | nla_get_u64(tb[TCA_NETEM_RATE64])); | |
967 | ||
99803171 DT |
968 | if (tb[TCA_NETEM_LATENCY64]) |
969 | q->latency = nla_get_s64(tb[TCA_NETEM_LATENCY64]); | |
970 | ||
971 | if (tb[TCA_NETEM_JITTER64]) | |
972 | q->jitter = nla_get_s64(tb[TCA_NETEM_JITTER64]); | |
973 | ||
e4ae004b ED |
974 | if (tb[TCA_NETEM_ECN]) |
975 | q->ecn = nla_get_u32(tb[TCA_NETEM_ECN]); | |
976 | ||
836af83b DT |
977 | if (tb[TCA_NETEM_SLOT]) |
978 | get_slot(q, tb[TCA_NETEM_SLOT]); | |
979 | ||
661b7972 | 980 | return ret; |
0a9fe5c3 YS |
981 | |
982 | get_table_failure: | |
983 | /* recover clg and loss_model, in case of | |
984 | * q->clg and q->loss_model were modified | |
985 | * in get_loss_clg() | |
986 | */ | |
987 | q->clg = old_clg; | |
988 | q->loss_model = old_loss_model; | |
989 | return ret; | |
1da177e4 LT |
990 | } |
991 | ||
e63d7dfd AA |
992 | static int netem_init(struct Qdisc *sch, struct nlattr *opt, |
993 | struct netlink_ext_ack *extack) | |
1da177e4 LT |
994 | { |
995 | struct netem_sched_data *q = qdisc_priv(sch); | |
996 | int ret; | |
997 | ||
634576a1 NA |
998 | qdisc_watchdog_init(&q->watchdog, sch); |
999 | ||
1da177e4 LT |
1000 | if (!opt) |
1001 | return -EINVAL; | |
1002 | ||
661b7972 | 1003 | q->loss_model = CLG_RANDOM; |
2030721c | 1004 | ret = netem_change(sch, opt, extack); |
50612537 | 1005 | if (ret) |
250a65f7 | 1006 | pr_info("netem: change failed\n"); |
1da177e4 LT |
1007 | return ret; |
1008 | } | |
1009 | ||
1010 | static void netem_destroy(struct Qdisc *sch) | |
1011 | { | |
1012 | struct netem_sched_data *q = qdisc_priv(sch); | |
1013 | ||
59cb5c67 | 1014 | qdisc_watchdog_cancel(&q->watchdog); |
50612537 | 1015 | if (q->qdisc) |
86bd446b | 1016 | qdisc_put(q->qdisc); |
6373a9a2 | 1017 | dist_free(q->delay_dist); |
0a9fe5c3 | 1018 | dist_free(q->slot_dist); |
1da177e4 LT |
1019 | } |
1020 | ||
661b7972 | 1021 | static int dump_loss_model(const struct netem_sched_data *q, |
1022 | struct sk_buff *skb) | |
1023 | { | |
1024 | struct nlattr *nest; | |
1025 | ||
1026 | nest = nla_nest_start(skb, TCA_NETEM_LOSS); | |
1027 | if (nest == NULL) | |
1028 | goto nla_put_failure; | |
1029 | ||
1030 | switch (q->loss_model) { | |
1031 | case CLG_RANDOM: | |
1032 | /* legacy loss model */ | |
1033 | nla_nest_cancel(skb, nest); | |
1034 | return 0; /* no data */ | |
1035 | ||
1036 | case CLG_4_STATES: { | |
1037 | struct tc_netem_gimodel gi = { | |
1038 | .p13 = q->clg.a1, | |
1039 | .p31 = q->clg.a2, | |
1040 | .p32 = q->clg.a3, | |
1041 | .p14 = q->clg.a4, | |
1042 | .p23 = q->clg.a5, | |
1043 | }; | |
1044 | ||
1b34ec43 DM |
1045 | if (nla_put(skb, NETEM_LOSS_GI, sizeof(gi), &gi)) |
1046 | goto nla_put_failure; | |
661b7972 | 1047 | break; |
1048 | } | |
1049 | case CLG_GILB_ELL: { | |
1050 | struct tc_netem_gemodel ge = { | |
1051 | .p = q->clg.a1, | |
1052 | .r = q->clg.a2, | |
1053 | .h = q->clg.a3, | |
1054 | .k1 = q->clg.a4, | |
1055 | }; | |
1056 | ||
1b34ec43 DM |
1057 | if (nla_put(skb, NETEM_LOSS_GE, sizeof(ge), &ge)) |
1058 | goto nla_put_failure; | |
661b7972 | 1059 | break; |
1060 | } | |
1061 | } | |
1062 | ||
1063 | nla_nest_end(skb, nest); | |
1064 | return 0; | |
1065 | ||
1066 | nla_put_failure: | |
1067 | nla_nest_cancel(skb, nest); | |
1068 | return -1; | |
1069 | } | |
1070 | ||
1da177e4 LT |
1071 | static int netem_dump(struct Qdisc *sch, struct sk_buff *skb) |
1072 | { | |
1073 | const struct netem_sched_data *q = qdisc_priv(sch); | |
861d7f74 | 1074 | struct nlattr *nla = (struct nlattr *) skb_tail_pointer(skb); |
1da177e4 LT |
1075 | struct tc_netem_qopt qopt; |
1076 | struct tc_netem_corr cor; | |
0dca51d3 | 1077 | struct tc_netem_reorder reorder; |
c865e5d9 | 1078 | struct tc_netem_corrupt corrupt; |
7bc0f28c | 1079 | struct tc_netem_rate rate; |
836af83b | 1080 | struct tc_netem_slot slot; |
1da177e4 | 1081 | |
112f9cb6 DT |
1082 | qopt.latency = min_t(psched_tdiff_t, PSCHED_NS2TICKS(q->latency), |
1083 | UINT_MAX); | |
1084 | qopt.jitter = min_t(psched_tdiff_t, PSCHED_NS2TICKS(q->jitter), | |
1085 | UINT_MAX); | |
1da177e4 LT |
1086 | qopt.limit = q->limit; |
1087 | qopt.loss = q->loss; | |
1088 | qopt.gap = q->gap; | |
1089 | qopt.duplicate = q->duplicate; | |
1b34ec43 DM |
1090 | if (nla_put(skb, TCA_OPTIONS, sizeof(qopt), &qopt)) |
1091 | goto nla_put_failure; | |
1da177e4 | 1092 | |
99803171 DT |
1093 | if (nla_put(skb, TCA_NETEM_LATENCY64, sizeof(q->latency), &q->latency)) |
1094 | goto nla_put_failure; | |
1095 | ||
1096 | if (nla_put(skb, TCA_NETEM_JITTER64, sizeof(q->jitter), &q->jitter)) | |
1097 | goto nla_put_failure; | |
1098 | ||
1da177e4 LT |
1099 | cor.delay_corr = q->delay_cor.rho; |
1100 | cor.loss_corr = q->loss_cor.rho; | |
1101 | cor.dup_corr = q->dup_cor.rho; | |
1b34ec43 DM |
1102 | if (nla_put(skb, TCA_NETEM_CORR, sizeof(cor), &cor)) |
1103 | goto nla_put_failure; | |
0dca51d3 SH |
1104 | |
1105 | reorder.probability = q->reorder; | |
1106 | reorder.correlation = q->reorder_cor.rho; | |
1b34ec43 DM |
1107 | if (nla_put(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder)) |
1108 | goto nla_put_failure; | |
0dca51d3 | 1109 | |
c865e5d9 SH |
1110 | corrupt.probability = q->corrupt; |
1111 | corrupt.correlation = q->corrupt_cor.rho; | |
1b34ec43 DM |
1112 | if (nla_put(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt)) |
1113 | goto nla_put_failure; | |
c865e5d9 | 1114 | |
6a031f67 | 1115 | if (q->rate >= (1ULL << 32)) { |
2a51c1e8 ND |
1116 | if (nla_put_u64_64bit(skb, TCA_NETEM_RATE64, q->rate, |
1117 | TCA_NETEM_PAD)) | |
6a031f67 YY |
1118 | goto nla_put_failure; |
1119 | rate.rate = ~0U; | |
1120 | } else { | |
1121 | rate.rate = q->rate; | |
1122 | } | |
90b41a1c HPP |
1123 | rate.packet_overhead = q->packet_overhead; |
1124 | rate.cell_size = q->cell_size; | |
1125 | rate.cell_overhead = q->cell_overhead; | |
1b34ec43 DM |
1126 | if (nla_put(skb, TCA_NETEM_RATE, sizeof(rate), &rate)) |
1127 | goto nla_put_failure; | |
7bc0f28c | 1128 | |
e4ae004b ED |
1129 | if (q->ecn && nla_put_u32(skb, TCA_NETEM_ECN, q->ecn)) |
1130 | goto nla_put_failure; | |
1131 | ||
661b7972 | 1132 | if (dump_loss_model(q, skb) != 0) |
1133 | goto nla_put_failure; | |
1134 | ||
0a9fe5c3 YS |
1135 | if (q->slot_config.min_delay | q->slot_config.max_delay | |
1136 | q->slot_config.dist_jitter) { | |
836af83b DT |
1137 | slot = q->slot_config; |
1138 | if (slot.max_packets == INT_MAX) | |
1139 | slot.max_packets = 0; | |
1140 | if (slot.max_bytes == INT_MAX) | |
1141 | slot.max_bytes = 0; | |
1142 | if (nla_put(skb, TCA_NETEM_SLOT, sizeof(slot), &slot)) | |
1143 | goto nla_put_failure; | |
1144 | } | |
1145 | ||
861d7f74 | 1146 | return nla_nest_end(skb, nla); |
1da177e4 | 1147 | |
1e90474c | 1148 | nla_put_failure: |
861d7f74 | 1149 | nlmsg_trim(skb, nla); |
1da177e4 LT |
1150 | return -1; |
1151 | } | |
1152 | ||
10f6dfcf | 1153 | static int netem_dump_class(struct Qdisc *sch, unsigned long cl, |
1154 | struct sk_buff *skb, struct tcmsg *tcm) | |
1155 | { | |
1156 | struct netem_sched_data *q = qdisc_priv(sch); | |
1157 | ||
50612537 | 1158 | if (cl != 1 || !q->qdisc) /* only one class */ |
10f6dfcf | 1159 | return -ENOENT; |
1160 | ||
1161 | tcm->tcm_handle |= TC_H_MIN(1); | |
1162 | tcm->tcm_info = q->qdisc->handle; | |
1163 | ||
1164 | return 0; | |
1165 | } | |
1166 | ||
1167 | static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
653d6fd6 | 1168 | struct Qdisc **old, struct netlink_ext_ack *extack) |
10f6dfcf | 1169 | { |
1170 | struct netem_sched_data *q = qdisc_priv(sch); | |
1171 | ||
86a7996c | 1172 | *old = qdisc_replace(sch, new, &q->qdisc); |
10f6dfcf | 1173 | return 0; |
1174 | } | |
1175 | ||
1176 | static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg) | |
1177 | { | |
1178 | struct netem_sched_data *q = qdisc_priv(sch); | |
1179 | return q->qdisc; | |
1180 | } | |
1181 | ||
143976ce | 1182 | static unsigned long netem_find(struct Qdisc *sch, u32 classid) |
10f6dfcf | 1183 | { |
1184 | return 1; | |
1185 | } | |
1186 | ||
10f6dfcf | 1187 | static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker) |
1188 | { | |
1189 | if (!walker->stop) { | |
1190 | if (walker->count >= walker->skip) | |
1191 | if (walker->fn(sch, 1, walker) < 0) { | |
1192 | walker->stop = 1; | |
1193 | return; | |
1194 | } | |
1195 | walker->count++; | |
1196 | } | |
1197 | } | |
1198 | ||
1199 | static const struct Qdisc_class_ops netem_class_ops = { | |
1200 | .graft = netem_graft, | |
1201 | .leaf = netem_leaf, | |
143976ce | 1202 | .find = netem_find, |
10f6dfcf | 1203 | .walk = netem_walk, |
1204 | .dump = netem_dump_class, | |
1205 | }; | |
1206 | ||
20fea08b | 1207 | static struct Qdisc_ops netem_qdisc_ops __read_mostly = { |
1da177e4 | 1208 | .id = "netem", |
10f6dfcf | 1209 | .cl_ops = &netem_class_ops, |
1da177e4 LT |
1210 | .priv_size = sizeof(struct netem_sched_data), |
1211 | .enqueue = netem_enqueue, | |
1212 | .dequeue = netem_dequeue, | |
77be155c | 1213 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
1214 | .init = netem_init, |
1215 | .reset = netem_reset, | |
1216 | .destroy = netem_destroy, | |
1217 | .change = netem_change, | |
1218 | .dump = netem_dump, | |
1219 | .owner = THIS_MODULE, | |
1220 | }; | |
1221 | ||
1222 | ||
1223 | static int __init netem_module_init(void) | |
1224 | { | |
eb229c4c | 1225 | pr_info("netem: version " VERSION "\n"); |
1da177e4 LT |
1226 | return register_qdisc(&netem_qdisc_ops); |
1227 | } | |
1228 | static void __exit netem_module_exit(void) | |
1229 | { | |
1230 | unregister_qdisc(&netem_qdisc_ops); | |
1231 | } | |
1232 | module_init(netem_module_init) | |
1233 | module_exit(netem_module_exit) | |
1234 | MODULE_LICENSE("GPL"); |