]>
Commit | Line | Data |
---|---|---|
84a14ae8 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * net/sched/sch_netem.c Network emulator | |
4 | * | |
1da177e4 | 5 | * Many of the algorithms and ideas for this came from |
10297b99 | 6 | * NIST Net which is not copyrighted. |
1da177e4 LT |
7 | * |
8 | * Authors: Stephen Hemminger <shemminger@osdl.org> | |
9 | * Catalin(ux aka Dino) BOIE <catab at umbrella dot ro> | |
10 | */ | |
11 | ||
b7f080cf | 12 | #include <linux/mm.h> |
1da177e4 | 13 | #include <linux/module.h> |
5a0e3ad6 | 14 | #include <linux/slab.h> |
1da177e4 LT |
15 | #include <linux/types.h> |
16 | #include <linux/kernel.h> | |
17 | #include <linux/errno.h> | |
1da177e4 | 18 | #include <linux/skbuff.h> |
78776d3f | 19 | #include <linux/vmalloc.h> |
1da177e4 | 20 | #include <linux/rtnetlink.h> |
90b41a1c | 21 | #include <linux/reciprocal_div.h> |
aec0a40a | 22 | #include <linux/rbtree.h> |
1da177e4 | 23 | |
dc5fc579 | 24 | #include <net/netlink.h> |
1da177e4 | 25 | #include <net/pkt_sched.h> |
e4ae004b | 26 | #include <net/inet_ecn.h> |
1da177e4 | 27 | |
250a65f7 | 28 | #define VERSION "1.3" |
eb229c4c | 29 | |
1da177e4 LT |
30 | /* Network Emulation Queuing algorithm. |
31 | ==================================== | |
32 | ||
33 | Sources: [1] Mark Carson, Darrin Santay, "NIST Net - A Linux-based | |
34 | Network Emulation Tool | |
35 | [2] Luigi Rizzo, DummyNet for FreeBSD | |
36 | ||
37 | ---------------------------------------------------------------- | |
38 | ||
39 | This started out as a simple way to delay outgoing packets to | |
40 | test TCP but has grown to include most of the functionality | |
41 | of a full blown network emulator like NISTnet. It can delay | |
42 | packets and add random jitter (and correlation). The random | |
43 | distribution can be loaded from a table as well to provide | |
44 | normal, Pareto, or experimental curves. Packet loss, | |
45 | duplication, and reordering can also be emulated. | |
46 | ||
47 | This qdisc does not do classification that can be handled in | |
48 | layering other disciplines. It does not need to do bandwidth | |
49 | control either since that can be handled by using token | |
50 | bucket or other rate control. | |
661b7972 | 51 | |
52 | Correlated Loss Generator models | |
53 | ||
54 | Added generation of correlated loss according to the | |
55 | "Gilbert-Elliot" model, a 4-state markov model. | |
56 | ||
57 | References: | |
58 | [1] NetemCLG Home http://netgroup.uniroma2.it/NetemCLG | |
59 | [2] S. Salsano, F. Ludovici, A. Ordine, "Definition of a general | |
60 | and intuitive loss model for packet networks and its implementation | |
61 | in the Netem module in the Linux kernel", available in [1] | |
62 | ||
63 | Authors: Stefano Salsano <stefano.salsano at uniroma2.it | |
64 | Fabio Ludovici <fabio.ludovici at yahoo.it> | |
1da177e4 LT |
65 | */ |
66 | ||
0a9fe5c3 YS |
67 | struct disttable { |
68 | u32 size; | |
69 | s16 table[0]; | |
70 | }; | |
71 | ||
1da177e4 | 72 | struct netem_sched_data { |
aec0a40a ED |
73 | /* internal t(ime)fifo qdisc uses t_root and sch->limit */ |
74 | struct rb_root t_root; | |
50612537 | 75 | |
d66280b1 PO |
76 | /* a linear queue; reduces rbtree rebalancing when jitter is low */ |
77 | struct sk_buff *t_head; | |
78 | struct sk_buff *t_tail; | |
79 | ||
50612537 | 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 | } |
d66280b1 PO |
372 | |
373 | rtnl_kfree_skbs(q->t_head, q->t_tail); | |
374 | q->t_head = NULL; | |
375 | q->t_tail = NULL; | |
ff704050 | 376 | } |
377 | ||
960fb66e | 378 | static void tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch) |
50612537 | 379 | { |
aec0a40a | 380 | struct netem_sched_data *q = qdisc_priv(sch); |
112f9cb6 | 381 | u64 tnext = netem_skb_cb(nskb)->time_to_send; |
50612537 | 382 | |
d66280b1 PO |
383 | if (!q->t_tail || tnext >= netem_skb_cb(q->t_tail)->time_to_send) { |
384 | if (q->t_tail) | |
385 | q->t_tail->next = nskb; | |
aec0a40a | 386 | else |
d66280b1 PO |
387 | q->t_head = nskb; |
388 | q->t_tail = nskb; | |
389 | } else { | |
390 | struct rb_node **p = &q->t_root.rb_node, *parent = NULL; | |
391 | ||
392 | while (*p) { | |
393 | struct sk_buff *skb; | |
394 | ||
395 | parent = *p; | |
396 | skb = rb_to_skb(parent); | |
397 | if (tnext >= netem_skb_cb(skb)->time_to_send) | |
398 | p = &parent->rb_right; | |
399 | else | |
400 | p = &parent->rb_left; | |
401 | } | |
402 | rb_link_node(&nskb->rbnode, parent, p); | |
403 | rb_insert_color(&nskb->rbnode, &q->t_root); | |
50612537 | 404 | } |
aec0a40a | 405 | sch->q.qlen++; |
50612537 ED |
406 | } |
407 | ||
6071bd1a NH |
408 | /* netem can't properly corrupt a megapacket (like we get from GSO), so instead |
409 | * when we statistically choose to corrupt one, we instead segment it, returning | |
410 | * the first packet to be corrupted, and re-enqueue the remaining frames | |
411 | */ | |
520ac30f ED |
412 | static struct sk_buff *netem_segment(struct sk_buff *skb, struct Qdisc *sch, |
413 | struct sk_buff **to_free) | |
6071bd1a NH |
414 | { |
415 | struct sk_buff *segs; | |
416 | netdev_features_t features = netif_skb_features(skb); | |
417 | ||
418 | segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK); | |
419 | ||
420 | if (IS_ERR_OR_NULL(segs)) { | |
520ac30f | 421 | qdisc_drop(skb, sch, to_free); |
6071bd1a NH |
422 | return NULL; |
423 | } | |
424 | consume_skb(skb); | |
425 | return segs; | |
426 | } | |
427 | ||
0afb51e7 SH |
428 | /* |
429 | * Insert one skb into qdisc. | |
430 | * Note: parent depends on return value to account for queue length. | |
431 | * NET_XMIT_DROP: queue length didn't change. | |
432 | * NET_XMIT_SUCCESS: one skb was queued. | |
433 | */ | |
520ac30f ED |
434 | static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch, |
435 | struct sk_buff **to_free) | |
1da177e4 LT |
436 | { |
437 | struct netem_sched_data *q = qdisc_priv(sch); | |
89e1df74 GC |
438 | /* We don't fill cb now as skb_unshare() may invalidate it */ |
439 | struct netem_skb_cb *cb; | |
0afb51e7 | 440 | struct sk_buff *skb2; |
6071bd1a | 441 | struct sk_buff *segs = NULL; |
177b8007 | 442 | unsigned int prev_len = qdisc_pkt_len(skb); |
0afb51e7 | 443 | int count = 1; |
6071bd1a | 444 | int rc = NET_XMIT_SUCCESS; |
5845f706 | 445 | int rc_drop = NET_XMIT_DROP; |
1da177e4 | 446 | |
9410d386 CP |
447 | /* Do not fool qdisc_drop_all() */ |
448 | skb->prev = NULL; | |
449 | ||
0afb51e7 SH |
450 | /* Random duplication */ |
451 | if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor)) | |
452 | ++count; | |
453 | ||
661b7972 | 454 | /* Drop packet? */ |
e4ae004b ED |
455 | if (loss_event(q)) { |
456 | if (q->ecn && INET_ECN_set_ce(skb)) | |
25331d6c | 457 | qdisc_qstats_drop(sch); /* mark packet */ |
e4ae004b ED |
458 | else |
459 | --count; | |
460 | } | |
0afb51e7 | 461 | if (count == 0) { |
25331d6c | 462 | qdisc_qstats_drop(sch); |
520ac30f | 463 | __qdisc_drop(skb, to_free); |
c27f339a | 464 | return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
1da177e4 LT |
465 | } |
466 | ||
5a308f40 ED |
467 | /* If a delay is expected, orphan the skb. (orphaning usually takes |
468 | * place at TX completion time, so _before_ the link transit delay) | |
5a308f40 | 469 | */ |
5080f39e | 470 | if (q->latency || q->jitter || q->rate) |
f2f872f9 | 471 | skb_orphan_partial(skb); |
4e8a5201 | 472 | |
0afb51e7 SH |
473 | /* |
474 | * If we need to duplicate packet, then re-insert at top of the | |
475 | * qdisc tree, since parent queuer expects that only one | |
476 | * skb will be queued. | |
477 | */ | |
478 | if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) { | |
159d2c7d | 479 | struct Qdisc *rootq = qdisc_root_bh(sch); |
0afb51e7 | 480 | u32 dupsave = q->duplicate; /* prevent duplicating a dup... */ |
0afb51e7 | 481 | |
b396cca6 | 482 | q->duplicate = 0; |
520ac30f | 483 | rootq->enqueue(skb2, rootq, to_free); |
0afb51e7 | 484 | q->duplicate = dupsave; |
5845f706 | 485 | rc_drop = NET_XMIT_SUCCESS; |
1da177e4 LT |
486 | } |
487 | ||
c865e5d9 SH |
488 | /* |
489 | * Randomized packet corruption. | |
490 | * Make copy if needed since we are modifying | |
491 | * If packet is going to be hardware checksummed, then | |
492 | * do it now in software before we mangle it. | |
493 | */ | |
494 | if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) { | |
6071bd1a | 495 | if (skb_is_gso(skb)) { |
3e14c383 JK |
496 | skb = netem_segment(skb, sch, to_free); |
497 | if (!skb) | |
5845f706 | 498 | return rc_drop; |
3e14c383 JK |
499 | segs = skb->next; |
500 | skb_mark_not_on_list(skb); | |
501 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
6071bd1a NH |
502 | } |
503 | ||
8a6e9c67 ED |
504 | skb = skb_unshare(skb, GFP_ATOMIC); |
505 | if (unlikely(!skb)) { | |
506 | qdisc_qstats_drop(sch); | |
507 | goto finish_segs; | |
508 | } | |
509 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
510 | skb_checksum_help(skb)) { | |
511 | qdisc_drop(skb, sch, to_free); | |
a7fa12d1 | 512 | skb = NULL; |
6071bd1a NH |
513 | goto finish_segs; |
514 | } | |
c865e5d9 | 515 | |
63862b5b AH |
516 | skb->data[prandom_u32() % skb_headlen(skb)] ^= |
517 | 1<<(prandom_u32() % 8); | |
c865e5d9 SH |
518 | } |
519 | ||
5845f706 | 520 | if (unlikely(sch->q.qlen >= sch->limit)) { |
3e14c383 JK |
521 | /* re-link segs, so that qdisc_drop_all() frees them all */ |
522 | skb->next = segs; | |
5845f706 SL |
523 | qdisc_drop_all(skb, sch, to_free); |
524 | return rc_drop; | |
525 | } | |
960fb66e | 526 | |
25331d6c | 527 | qdisc_qstats_backlog_inc(sch, skb); |
960fb66e | 528 | |
5f86173b | 529 | cb = netem_skb_cb(skb); |
cc7ec456 | 530 | if (q->gap == 0 || /* not doing reordering */ |
a42b4799 | 531 | q->counter < q->gap - 1 || /* inside last reordering gap */ |
f64f9e71 | 532 | q->reorder < get_crandom(&q->reorder_cor)) { |
112f9cb6 DT |
533 | u64 now; |
534 | s64 delay; | |
07aaa115 SH |
535 | |
536 | delay = tabledist(q->latency, q->jitter, | |
537 | &q->delay_cor, q->delay_dist); | |
538 | ||
112f9cb6 | 539 | now = ktime_get_ns(); |
7bc0f28c HPP |
540 | |
541 | if (q->rate) { | |
5080f39e NU |
542 | struct netem_skb_cb *last = NULL; |
543 | ||
544 | if (sch->q.tail) | |
545 | last = netem_skb_cb(sch->q.tail); | |
546 | if (q->t_root.rb_node) { | |
547 | struct sk_buff *t_skb; | |
548 | struct netem_skb_cb *t_last; | |
549 | ||
18a4c0ea | 550 | t_skb = skb_rb_last(&q->t_root); |
5080f39e NU |
551 | t_last = netem_skb_cb(t_skb); |
552 | if (!last || | |
d66280b1 PO |
553 | t_last->time_to_send > last->time_to_send) |
554 | last = t_last; | |
555 | } | |
556 | if (q->t_tail) { | |
557 | struct netem_skb_cb *t_last = | |
558 | netem_skb_cb(q->t_tail); | |
559 | ||
560 | if (!last || | |
561 | t_last->time_to_send > last->time_to_send) | |
5080f39e | 562 | last = t_last; |
5080f39e | 563 | } |
7bc0f28c | 564 | |
aec0a40a | 565 | if (last) { |
7bc0f28c | 566 | /* |
a13d3104 JN |
567 | * Last packet in queue is reference point (now), |
568 | * calculate this time bonus and subtract | |
7bc0f28c HPP |
569 | * from delay. |
570 | */ | |
5080f39e | 571 | delay -= last->time_to_send - now; |
112f9cb6 | 572 | delay = max_t(s64, 0, delay); |
5080f39e | 573 | now = last->time_to_send; |
7bc0f28c | 574 | } |
a13d3104 | 575 | |
bce552fd | 576 | delay += packet_time_ns(qdisc_pkt_len(skb), q); |
7bc0f28c HPP |
577 | } |
578 | ||
7c59e25f | 579 | cb->time_to_send = now + delay; |
1da177e4 | 580 | ++q->counter; |
960fb66e | 581 | tfifo_enqueue(skb, sch); |
1da177e4 | 582 | } else { |
10297b99 | 583 | /* |
0dca51d3 SH |
584 | * Do re-ordering by putting one out of N packets at the front |
585 | * of the queue. | |
586 | */ | |
112f9cb6 | 587 | cb->time_to_send = ktime_get_ns(); |
0dca51d3 | 588 | q->counter = 0; |
8ba25dad | 589 | |
59697730 | 590 | __qdisc_enqueue_head(skb, &sch->q); |
eb101924 | 591 | sch->qstats.requeues++; |
378a2f09 | 592 | } |
1da177e4 | 593 | |
6071bd1a NH |
594 | finish_segs: |
595 | if (segs) { | |
177b8007 | 596 | unsigned int len, last_len; |
a7fa12d1 | 597 | int nb; |
177b8007 | 598 | |
a7fa12d1 JK |
599 | len = skb ? skb->len : 0; |
600 | nb = skb ? 1 : 0; | |
177b8007 | 601 | |
6071bd1a NH |
602 | while (segs) { |
603 | skb2 = segs->next; | |
a8305bff | 604 | skb_mark_not_on_list(segs); |
6071bd1a NH |
605 | qdisc_skb_cb(segs)->pkt_len = segs->len; |
606 | last_len = segs->len; | |
520ac30f | 607 | rc = qdisc_enqueue(segs, sch, to_free); |
6071bd1a NH |
608 | if (rc != NET_XMIT_SUCCESS) { |
609 | if (net_xmit_drop_count(rc)) | |
610 | qdisc_qstats_drop(sch); | |
611 | } else { | |
612 | nb++; | |
613 | len += last_len; | |
614 | } | |
615 | segs = skb2; | |
616 | } | |
a7fa12d1 JK |
617 | /* Parent qdiscs accounted for 1 skb of size @prev_len */ |
618 | qdisc_tree_reduce_backlog(sch, -(nb - 1), -(len - prev_len)); | |
e0ad032e JK |
619 | } else if (!skb) { |
620 | return NET_XMIT_DROP; | |
6071bd1a | 621 | } |
10f6dfcf | 622 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
623 | } |
624 | ||
836af83b DT |
625 | /* Delay the next round with a new future slot with a |
626 | * correct number of bytes and packets. | |
627 | */ | |
628 | ||
629 | static void get_slot_next(struct netem_sched_data *q, u64 now) | |
630 | { | |
0a9fe5c3 YS |
631 | s64 next_delay; |
632 | ||
633 | if (!q->slot_dist) | |
634 | next_delay = q->slot_config.min_delay + | |
635 | (prandom_u32() * | |
636 | (q->slot_config.max_delay - | |
637 | q->slot_config.min_delay) >> 32); | |
638 | else | |
639 | next_delay = tabledist(q->slot_config.dist_delay, | |
640 | (s32)(q->slot_config.dist_jitter), | |
641 | NULL, q->slot_dist); | |
642 | ||
643 | q->slot.slot_next = now + next_delay; | |
836af83b DT |
644 | q->slot.packets_left = q->slot_config.max_packets; |
645 | q->slot.bytes_left = q->slot_config.max_bytes; | |
646 | } | |
647 | ||
d66280b1 PO |
648 | static struct sk_buff *netem_peek(struct netem_sched_data *q) |
649 | { | |
650 | struct sk_buff *skb = skb_rb_first(&q->t_root); | |
651 | u64 t1, t2; | |
652 | ||
653 | if (!skb) | |
654 | return q->t_head; | |
655 | if (!q->t_head) | |
656 | return skb; | |
657 | ||
658 | t1 = netem_skb_cb(skb)->time_to_send; | |
659 | t2 = netem_skb_cb(q->t_head)->time_to_send; | |
660 | if (t1 < t2) | |
661 | return skb; | |
662 | return q->t_head; | |
663 | } | |
664 | ||
665 | static void netem_erase_head(struct netem_sched_data *q, struct sk_buff *skb) | |
666 | { | |
667 | if (skb == q->t_head) { | |
668 | q->t_head = skb->next; | |
669 | if (!q->t_head) | |
670 | q->t_tail = NULL; | |
671 | } else { | |
672 | rb_erase(&skb->rbnode, &q->t_root); | |
673 | } | |
674 | } | |
675 | ||
1da177e4 LT |
676 | static struct sk_buff *netem_dequeue(struct Qdisc *sch) |
677 | { | |
678 | struct netem_sched_data *q = qdisc_priv(sch); | |
679 | struct sk_buff *skb; | |
680 | ||
50612537 | 681 | tfifo_dequeue: |
ed760cb8 | 682 | skb = __qdisc_dequeue_head(&sch->q); |
771018e7 | 683 | if (skb) { |
25331d6c | 684 | qdisc_qstats_backlog_dec(sch, skb); |
0ad2a836 | 685 | deliver: |
aec0a40a ED |
686 | qdisc_bstats_update(sch, skb); |
687 | return skb; | |
688 | } | |
d66280b1 PO |
689 | skb = netem_peek(q); |
690 | if (skb) { | |
112f9cb6 | 691 | u64 time_to_send; |
836af83b | 692 | u64 now = ktime_get_ns(); |
36b7bfe0 | 693 | |
0f9f32ac | 694 | /* if more time remaining? */ |
36b7bfe0 | 695 | time_to_send = netem_skb_cb(skb)->time_to_send; |
836af83b DT |
696 | if (q->slot.slot_next && q->slot.slot_next < time_to_send) |
697 | get_slot_next(q, now); | |
aec0a40a | 698 | |
d66280b1 PO |
699 | if (time_to_send <= now && q->slot.slot_next <= now) { |
700 | netem_erase_head(q, skb); | |
aec0a40a | 701 | sch->q.qlen--; |
0ad2a836 | 702 | qdisc_qstats_backlog_dec(sch, skb); |
aec0a40a ED |
703 | skb->next = NULL; |
704 | skb->prev = NULL; | |
bffa72cf ED |
705 | /* skb->dev shares skb->rbnode area, |
706 | * we need to restore its value. | |
707 | */ | |
708 | skb->dev = qdisc_dev(sch); | |
03c05f0d | 709 | |
836af83b DT |
710 | if (q->slot.slot_next) { |
711 | q->slot.packets_left--; | |
712 | q->slot.bytes_left -= qdisc_pkt_len(skb); | |
713 | if (q->slot.packets_left <= 0 || | |
714 | q->slot.bytes_left <= 0) | |
715 | get_slot_next(q, now); | |
716 | } | |
717 | ||
50612537 | 718 | if (q->qdisc) { |
21de12ee | 719 | unsigned int pkt_len = qdisc_pkt_len(skb); |
520ac30f ED |
720 | struct sk_buff *to_free = NULL; |
721 | int err; | |
50612537 | 722 | |
520ac30f ED |
723 | err = qdisc_enqueue(skb, q->qdisc, &to_free); |
724 | kfree_skb_list(to_free); | |
21de12ee ED |
725 | if (err != NET_XMIT_SUCCESS && |
726 | net_xmit_drop_count(err)) { | |
727 | qdisc_qstats_drop(sch); | |
728 | qdisc_tree_reduce_backlog(sch, 1, | |
729 | pkt_len); | |
50612537 ED |
730 | } |
731 | goto tfifo_dequeue; | |
732 | } | |
aec0a40a | 733 | goto deliver; |
07aaa115 | 734 | } |
11274e5a | 735 | |
50612537 ED |
736 | if (q->qdisc) { |
737 | skb = q->qdisc->ops->dequeue(q->qdisc); | |
738 | if (skb) | |
739 | goto deliver; | |
740 | } | |
836af83b DT |
741 | |
742 | qdisc_watchdog_schedule_ns(&q->watchdog, | |
743 | max(time_to_send, | |
744 | q->slot.slot_next)); | |
0f9f32ac SH |
745 | } |
746 | ||
50612537 ED |
747 | if (q->qdisc) { |
748 | skb = q->qdisc->ops->dequeue(q->qdisc); | |
749 | if (skb) | |
750 | goto deliver; | |
751 | } | |
0f9f32ac | 752 | return NULL; |
1da177e4 LT |
753 | } |
754 | ||
1da177e4 LT |
755 | static void netem_reset(struct Qdisc *sch) |
756 | { | |
757 | struct netem_sched_data *q = qdisc_priv(sch); | |
758 | ||
50612537 | 759 | qdisc_reset_queue(sch); |
ff704050 | 760 | tfifo_reset(sch); |
50612537 ED |
761 | if (q->qdisc) |
762 | qdisc_reset(q->qdisc); | |
59cb5c67 | 763 | qdisc_watchdog_cancel(&q->watchdog); |
1da177e4 LT |
764 | } |
765 | ||
6373a9a2 | 766 | static void dist_free(struct disttable *d) |
767 | { | |
4cb28970 | 768 | kvfree(d); |
6373a9a2 | 769 | } |
770 | ||
1da177e4 LT |
771 | /* |
772 | * Distribution data is a variable size payload containing | |
773 | * signed 16 bit values. | |
774 | */ | |
836af83b | 775 | |
0a9fe5c3 YS |
776 | static int get_dist_table(struct Qdisc *sch, struct disttable **tbl, |
777 | const struct nlattr *attr) | |
1da177e4 | 778 | { |
6373a9a2 | 779 | size_t n = nla_len(attr)/sizeof(__s16); |
1e90474c | 780 | const __s16 *data = nla_data(attr); |
7698b4fc | 781 | spinlock_t *root_lock; |
1da177e4 LT |
782 | struct disttable *d; |
783 | int i; | |
784 | ||
b41d936b | 785 | if (!n || n > NETEM_DIST_MAX) |
1da177e4 LT |
786 | return -EINVAL; |
787 | ||
752ade68 | 788 | d = kvmalloc(sizeof(struct disttable) + n * sizeof(s16), GFP_KERNEL); |
1da177e4 LT |
789 | if (!d) |
790 | return -ENOMEM; | |
791 | ||
792 | d->size = n; | |
793 | for (i = 0; i < n; i++) | |
794 | d->table[i] = data[i]; | |
10297b99 | 795 | |
102396ae | 796 | root_lock = qdisc_root_sleeping_lock(sch); |
7698b4fc DM |
797 | |
798 | spin_lock_bh(root_lock); | |
0a9fe5c3 | 799 | swap(*tbl, d); |
7698b4fc | 800 | spin_unlock_bh(root_lock); |
bb52c7ac ED |
801 | |
802 | dist_free(d); | |
1da177e4 LT |
803 | return 0; |
804 | } | |
805 | ||
836af83b DT |
806 | static void get_slot(struct netem_sched_data *q, const struct nlattr *attr) |
807 | { | |
808 | const struct tc_netem_slot *c = nla_data(attr); | |
809 | ||
810 | q->slot_config = *c; | |
811 | if (q->slot_config.max_packets == 0) | |
812 | q->slot_config.max_packets = INT_MAX; | |
813 | if (q->slot_config.max_bytes == 0) | |
814 | q->slot_config.max_bytes = INT_MAX; | |
815 | q->slot.packets_left = q->slot_config.max_packets; | |
816 | q->slot.bytes_left = q->slot_config.max_bytes; | |
0a9fe5c3 YS |
817 | if (q->slot_config.min_delay | q->slot_config.max_delay | |
818 | q->slot_config.dist_jitter) | |
836af83b DT |
819 | q->slot.slot_next = ktime_get_ns(); |
820 | else | |
821 | q->slot.slot_next = 0; | |
822 | } | |
823 | ||
49545a77 | 824 | static void get_correlation(struct netem_sched_data *q, const struct nlattr *attr) |
1da177e4 | 825 | { |
1e90474c | 826 | const struct tc_netem_corr *c = nla_data(attr); |
1da177e4 | 827 | |
1da177e4 LT |
828 | init_crandom(&q->delay_cor, c->delay_corr); |
829 | init_crandom(&q->loss_cor, c->loss_corr); | |
830 | init_crandom(&q->dup_cor, c->dup_corr); | |
1da177e4 LT |
831 | } |
832 | ||
49545a77 | 833 | static void get_reorder(struct netem_sched_data *q, const struct nlattr *attr) |
0dca51d3 | 834 | { |
1e90474c | 835 | const struct tc_netem_reorder *r = nla_data(attr); |
0dca51d3 | 836 | |
0dca51d3 SH |
837 | q->reorder = r->probability; |
838 | init_crandom(&q->reorder_cor, r->correlation); | |
0dca51d3 SH |
839 | } |
840 | ||
49545a77 | 841 | static void get_corrupt(struct netem_sched_data *q, const struct nlattr *attr) |
c865e5d9 | 842 | { |
1e90474c | 843 | const struct tc_netem_corrupt *r = nla_data(attr); |
c865e5d9 | 844 | |
c865e5d9 SH |
845 | q->corrupt = r->probability; |
846 | init_crandom(&q->corrupt_cor, r->correlation); | |
c865e5d9 SH |
847 | } |
848 | ||
49545a77 | 849 | static void get_rate(struct netem_sched_data *q, const struct nlattr *attr) |
7bc0f28c | 850 | { |
7bc0f28c HPP |
851 | const struct tc_netem_rate *r = nla_data(attr); |
852 | ||
853 | q->rate = r->rate; | |
90b41a1c HPP |
854 | q->packet_overhead = r->packet_overhead; |
855 | q->cell_size = r->cell_size; | |
809fa972 | 856 | q->cell_overhead = r->cell_overhead; |
90b41a1c HPP |
857 | if (q->cell_size) |
858 | q->cell_size_reciprocal = reciprocal_value(q->cell_size); | |
809fa972 HFS |
859 | else |
860 | q->cell_size_reciprocal = (struct reciprocal_value) { 0 }; | |
7bc0f28c HPP |
861 | } |
862 | ||
49545a77 | 863 | static int get_loss_clg(struct netem_sched_data *q, const struct nlattr *attr) |
661b7972 | 864 | { |
661b7972 | 865 | const struct nlattr *la; |
866 | int rem; | |
867 | ||
868 | nla_for_each_nested(la, attr, rem) { | |
869 | u16 type = nla_type(la); | |
870 | ||
833fa743 | 871 | switch (type) { |
661b7972 | 872 | case NETEM_LOSS_GI: { |
873 | const struct tc_netem_gimodel *gi = nla_data(la); | |
874 | ||
2494654d | 875 | if (nla_len(la) < sizeof(struct tc_netem_gimodel)) { |
661b7972 | 876 | pr_info("netem: incorrect gi model size\n"); |
877 | return -EINVAL; | |
878 | } | |
879 | ||
880 | q->loss_model = CLG_4_STATES; | |
881 | ||
3fbac2a8 | 882 | q->clg.state = TX_IN_GAP_PERIOD; |
661b7972 | 883 | q->clg.a1 = gi->p13; |
884 | q->clg.a2 = gi->p31; | |
885 | q->clg.a3 = gi->p32; | |
886 | q->clg.a4 = gi->p14; | |
887 | q->clg.a5 = gi->p23; | |
888 | break; | |
889 | } | |
890 | ||
891 | case NETEM_LOSS_GE: { | |
892 | const struct tc_netem_gemodel *ge = nla_data(la); | |
893 | ||
2494654d | 894 | if (nla_len(la) < sizeof(struct tc_netem_gemodel)) { |
895 | pr_info("netem: incorrect ge model size\n"); | |
661b7972 | 896 | return -EINVAL; |
897 | } | |
898 | ||
899 | q->loss_model = CLG_GILB_ELL; | |
3fbac2a8 | 900 | q->clg.state = GOOD_STATE; |
661b7972 | 901 | q->clg.a1 = ge->p; |
902 | q->clg.a2 = ge->r; | |
903 | q->clg.a3 = ge->h; | |
904 | q->clg.a4 = ge->k1; | |
905 | break; | |
906 | } | |
907 | ||
908 | default: | |
909 | pr_info("netem: unknown loss type %u\n", type); | |
910 | return -EINVAL; | |
911 | } | |
912 | } | |
913 | ||
914 | return 0; | |
915 | } | |
916 | ||
27a3421e PM |
917 | static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = { |
918 | [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) }, | |
919 | [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) }, | |
920 | [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) }, | |
7bc0f28c | 921 | [TCA_NETEM_RATE] = { .len = sizeof(struct tc_netem_rate) }, |
661b7972 | 922 | [TCA_NETEM_LOSS] = { .type = NLA_NESTED }, |
e4ae004b | 923 | [TCA_NETEM_ECN] = { .type = NLA_U32 }, |
6a031f67 | 924 | [TCA_NETEM_RATE64] = { .type = NLA_U64 }, |
99803171 DT |
925 | [TCA_NETEM_LATENCY64] = { .type = NLA_S64 }, |
926 | [TCA_NETEM_JITTER64] = { .type = NLA_S64 }, | |
836af83b | 927 | [TCA_NETEM_SLOT] = { .len = sizeof(struct tc_netem_slot) }, |
27a3421e PM |
928 | }; |
929 | ||
2c10b32b TG |
930 | static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla, |
931 | const struct nla_policy *policy, int len) | |
932 | { | |
933 | int nested_len = nla_len(nla) - NLA_ALIGN(len); | |
934 | ||
661b7972 | 935 | if (nested_len < 0) { |
936 | pr_info("netem: invalid attributes len %d\n", nested_len); | |
2c10b32b | 937 | return -EINVAL; |
661b7972 | 938 | } |
939 | ||
2c10b32b | 940 | if (nested_len >= nla_attr_size(0)) |
8cb08174 JB |
941 | return nla_parse_deprecated(tb, maxtype, |
942 | nla_data(nla) + NLA_ALIGN(len), | |
943 | nested_len, policy, NULL); | |
661b7972 | 944 | |
2c10b32b TG |
945 | memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1)); |
946 | return 0; | |
947 | } | |
948 | ||
c865e5d9 | 949 | /* Parse netlink message to set options */ |
2030721c AA |
950 | static int netem_change(struct Qdisc *sch, struct nlattr *opt, |
951 | struct netlink_ext_ack *extack) | |
1da177e4 LT |
952 | { |
953 | struct netem_sched_data *q = qdisc_priv(sch); | |
b03f4672 | 954 | struct nlattr *tb[TCA_NETEM_MAX + 1]; |
1da177e4 | 955 | struct tc_netem_qopt *qopt; |
54a4b05c YY |
956 | struct clgstate old_clg; |
957 | int old_loss_model = CLG_RANDOM; | |
1da177e4 | 958 | int ret; |
10297b99 | 959 | |
b03f4672 | 960 | if (opt == NULL) |
1da177e4 LT |
961 | return -EINVAL; |
962 | ||
2c10b32b TG |
963 | qopt = nla_data(opt); |
964 | ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt)); | |
b03f4672 PM |
965 | if (ret < 0) |
966 | return ret; | |
967 | ||
54a4b05c YY |
968 | /* backup q->clg and q->loss_model */ |
969 | old_clg = q->clg; | |
970 | old_loss_model = q->loss_model; | |
971 | ||
972 | if (tb[TCA_NETEM_LOSS]) { | |
49545a77 | 973 | ret = get_loss_clg(q, tb[TCA_NETEM_LOSS]); |
54a4b05c YY |
974 | if (ret) { |
975 | q->loss_model = old_loss_model; | |
976 | return ret; | |
977 | } | |
978 | } else { | |
979 | q->loss_model = CLG_RANDOM; | |
980 | } | |
981 | ||
982 | if (tb[TCA_NETEM_DELAY_DIST]) { | |
0a9fe5c3 YS |
983 | ret = get_dist_table(sch, &q->delay_dist, |
984 | tb[TCA_NETEM_DELAY_DIST]); | |
985 | if (ret) | |
986 | goto get_table_failure; | |
987 | } | |
988 | ||
989 | if (tb[TCA_NETEM_SLOT_DIST]) { | |
990 | ret = get_dist_table(sch, &q->slot_dist, | |
991 | tb[TCA_NETEM_SLOT_DIST]); | |
992 | if (ret) | |
993 | goto get_table_failure; | |
54a4b05c YY |
994 | } |
995 | ||
50612537 | 996 | sch->limit = qopt->limit; |
10297b99 | 997 | |
112f9cb6 DT |
998 | q->latency = PSCHED_TICKS2NS(qopt->latency); |
999 | q->jitter = PSCHED_TICKS2NS(qopt->jitter); | |
1da177e4 LT |
1000 | q->limit = qopt->limit; |
1001 | q->gap = qopt->gap; | |
0dca51d3 | 1002 | q->counter = 0; |
1da177e4 LT |
1003 | q->loss = qopt->loss; |
1004 | q->duplicate = qopt->duplicate; | |
1005 | ||
bb2f8cc0 SH |
1006 | /* for compatibility with earlier versions. |
1007 | * if gap is set, need to assume 100% probability | |
0dca51d3 | 1008 | */ |
a362e0a7 SH |
1009 | if (q->gap) |
1010 | q->reorder = ~0; | |
0dca51d3 | 1011 | |
265eb67f | 1012 | if (tb[TCA_NETEM_CORR]) |
49545a77 | 1013 | get_correlation(q, tb[TCA_NETEM_CORR]); |
1da177e4 | 1014 | |
265eb67f | 1015 | if (tb[TCA_NETEM_REORDER]) |
49545a77 | 1016 | get_reorder(q, tb[TCA_NETEM_REORDER]); |
1da177e4 | 1017 | |
265eb67f | 1018 | if (tb[TCA_NETEM_CORRUPT]) |
49545a77 | 1019 | get_corrupt(q, tb[TCA_NETEM_CORRUPT]); |
1da177e4 | 1020 | |
7bc0f28c | 1021 | if (tb[TCA_NETEM_RATE]) |
49545a77 | 1022 | get_rate(q, tb[TCA_NETEM_RATE]); |
7bc0f28c | 1023 | |
6a031f67 YY |
1024 | if (tb[TCA_NETEM_RATE64]) |
1025 | q->rate = max_t(u64, q->rate, | |
1026 | nla_get_u64(tb[TCA_NETEM_RATE64])); | |
1027 | ||
99803171 DT |
1028 | if (tb[TCA_NETEM_LATENCY64]) |
1029 | q->latency = nla_get_s64(tb[TCA_NETEM_LATENCY64]); | |
1030 | ||
1031 | if (tb[TCA_NETEM_JITTER64]) | |
1032 | q->jitter = nla_get_s64(tb[TCA_NETEM_JITTER64]); | |
1033 | ||
e4ae004b ED |
1034 | if (tb[TCA_NETEM_ECN]) |
1035 | q->ecn = nla_get_u32(tb[TCA_NETEM_ECN]); | |
1036 | ||
836af83b DT |
1037 | if (tb[TCA_NETEM_SLOT]) |
1038 | get_slot(q, tb[TCA_NETEM_SLOT]); | |
1039 | ||
661b7972 | 1040 | return ret; |
0a9fe5c3 YS |
1041 | |
1042 | get_table_failure: | |
1043 | /* recover clg and loss_model, in case of | |
1044 | * q->clg and q->loss_model were modified | |
1045 | * in get_loss_clg() | |
1046 | */ | |
1047 | q->clg = old_clg; | |
1048 | q->loss_model = old_loss_model; | |
1049 | return ret; | |
1da177e4 LT |
1050 | } |
1051 | ||
e63d7dfd AA |
1052 | static int netem_init(struct Qdisc *sch, struct nlattr *opt, |
1053 | struct netlink_ext_ack *extack) | |
1da177e4 LT |
1054 | { |
1055 | struct netem_sched_data *q = qdisc_priv(sch); | |
1056 | int ret; | |
1057 | ||
634576a1 NA |
1058 | qdisc_watchdog_init(&q->watchdog, sch); |
1059 | ||
1da177e4 LT |
1060 | if (!opt) |
1061 | return -EINVAL; | |
1062 | ||
661b7972 | 1063 | q->loss_model = CLG_RANDOM; |
2030721c | 1064 | ret = netem_change(sch, opt, extack); |
50612537 | 1065 | if (ret) |
250a65f7 | 1066 | pr_info("netem: change failed\n"); |
1da177e4 LT |
1067 | return ret; |
1068 | } | |
1069 | ||
1070 | static void netem_destroy(struct Qdisc *sch) | |
1071 | { | |
1072 | struct netem_sched_data *q = qdisc_priv(sch); | |
1073 | ||
59cb5c67 | 1074 | qdisc_watchdog_cancel(&q->watchdog); |
50612537 | 1075 | if (q->qdisc) |
86bd446b | 1076 | qdisc_put(q->qdisc); |
6373a9a2 | 1077 | dist_free(q->delay_dist); |
0a9fe5c3 | 1078 | dist_free(q->slot_dist); |
1da177e4 LT |
1079 | } |
1080 | ||
661b7972 | 1081 | static int dump_loss_model(const struct netem_sched_data *q, |
1082 | struct sk_buff *skb) | |
1083 | { | |
1084 | struct nlattr *nest; | |
1085 | ||
ae0be8de | 1086 | nest = nla_nest_start_noflag(skb, TCA_NETEM_LOSS); |
661b7972 | 1087 | if (nest == NULL) |
1088 | goto nla_put_failure; | |
1089 | ||
1090 | switch (q->loss_model) { | |
1091 | case CLG_RANDOM: | |
1092 | /* legacy loss model */ | |
1093 | nla_nest_cancel(skb, nest); | |
1094 | return 0; /* no data */ | |
1095 | ||
1096 | case CLG_4_STATES: { | |
1097 | struct tc_netem_gimodel gi = { | |
1098 | .p13 = q->clg.a1, | |
1099 | .p31 = q->clg.a2, | |
1100 | .p32 = q->clg.a3, | |
1101 | .p14 = q->clg.a4, | |
1102 | .p23 = q->clg.a5, | |
1103 | }; | |
1104 | ||
1b34ec43 DM |
1105 | if (nla_put(skb, NETEM_LOSS_GI, sizeof(gi), &gi)) |
1106 | goto nla_put_failure; | |
661b7972 | 1107 | break; |
1108 | } | |
1109 | case CLG_GILB_ELL: { | |
1110 | struct tc_netem_gemodel ge = { | |
1111 | .p = q->clg.a1, | |
1112 | .r = q->clg.a2, | |
1113 | .h = q->clg.a3, | |
1114 | .k1 = q->clg.a4, | |
1115 | }; | |
1116 | ||
1b34ec43 DM |
1117 | if (nla_put(skb, NETEM_LOSS_GE, sizeof(ge), &ge)) |
1118 | goto nla_put_failure; | |
661b7972 | 1119 | break; |
1120 | } | |
1121 | } | |
1122 | ||
1123 | nla_nest_end(skb, nest); | |
1124 | return 0; | |
1125 | ||
1126 | nla_put_failure: | |
1127 | nla_nest_cancel(skb, nest); | |
1128 | return -1; | |
1129 | } | |
1130 | ||
1da177e4 LT |
1131 | static int netem_dump(struct Qdisc *sch, struct sk_buff *skb) |
1132 | { | |
1133 | const struct netem_sched_data *q = qdisc_priv(sch); | |
861d7f74 | 1134 | struct nlattr *nla = (struct nlattr *) skb_tail_pointer(skb); |
1da177e4 LT |
1135 | struct tc_netem_qopt qopt; |
1136 | struct tc_netem_corr cor; | |
0dca51d3 | 1137 | struct tc_netem_reorder reorder; |
c865e5d9 | 1138 | struct tc_netem_corrupt corrupt; |
7bc0f28c | 1139 | struct tc_netem_rate rate; |
836af83b | 1140 | struct tc_netem_slot slot; |
1da177e4 | 1141 | |
112f9cb6 DT |
1142 | qopt.latency = min_t(psched_tdiff_t, PSCHED_NS2TICKS(q->latency), |
1143 | UINT_MAX); | |
1144 | qopt.jitter = min_t(psched_tdiff_t, PSCHED_NS2TICKS(q->jitter), | |
1145 | UINT_MAX); | |
1da177e4 LT |
1146 | qopt.limit = q->limit; |
1147 | qopt.loss = q->loss; | |
1148 | qopt.gap = q->gap; | |
1149 | qopt.duplicate = q->duplicate; | |
1b34ec43 DM |
1150 | if (nla_put(skb, TCA_OPTIONS, sizeof(qopt), &qopt)) |
1151 | goto nla_put_failure; | |
1da177e4 | 1152 | |
99803171 DT |
1153 | if (nla_put(skb, TCA_NETEM_LATENCY64, sizeof(q->latency), &q->latency)) |
1154 | goto nla_put_failure; | |
1155 | ||
1156 | if (nla_put(skb, TCA_NETEM_JITTER64, sizeof(q->jitter), &q->jitter)) | |
1157 | goto nla_put_failure; | |
1158 | ||
1da177e4 LT |
1159 | cor.delay_corr = q->delay_cor.rho; |
1160 | cor.loss_corr = q->loss_cor.rho; | |
1161 | cor.dup_corr = q->dup_cor.rho; | |
1b34ec43 DM |
1162 | if (nla_put(skb, TCA_NETEM_CORR, sizeof(cor), &cor)) |
1163 | goto nla_put_failure; | |
0dca51d3 SH |
1164 | |
1165 | reorder.probability = q->reorder; | |
1166 | reorder.correlation = q->reorder_cor.rho; | |
1b34ec43 DM |
1167 | if (nla_put(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder)) |
1168 | goto nla_put_failure; | |
0dca51d3 | 1169 | |
c865e5d9 SH |
1170 | corrupt.probability = q->corrupt; |
1171 | corrupt.correlation = q->corrupt_cor.rho; | |
1b34ec43 DM |
1172 | if (nla_put(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt)) |
1173 | goto nla_put_failure; | |
c865e5d9 | 1174 | |
6a031f67 | 1175 | if (q->rate >= (1ULL << 32)) { |
2a51c1e8 ND |
1176 | if (nla_put_u64_64bit(skb, TCA_NETEM_RATE64, q->rate, |
1177 | TCA_NETEM_PAD)) | |
6a031f67 YY |
1178 | goto nla_put_failure; |
1179 | rate.rate = ~0U; | |
1180 | } else { | |
1181 | rate.rate = q->rate; | |
1182 | } | |
90b41a1c HPP |
1183 | rate.packet_overhead = q->packet_overhead; |
1184 | rate.cell_size = q->cell_size; | |
1185 | rate.cell_overhead = q->cell_overhead; | |
1b34ec43 DM |
1186 | if (nla_put(skb, TCA_NETEM_RATE, sizeof(rate), &rate)) |
1187 | goto nla_put_failure; | |
7bc0f28c | 1188 | |
e4ae004b ED |
1189 | if (q->ecn && nla_put_u32(skb, TCA_NETEM_ECN, q->ecn)) |
1190 | goto nla_put_failure; | |
1191 | ||
661b7972 | 1192 | if (dump_loss_model(q, skb) != 0) |
1193 | goto nla_put_failure; | |
1194 | ||
0a9fe5c3 YS |
1195 | if (q->slot_config.min_delay | q->slot_config.max_delay | |
1196 | q->slot_config.dist_jitter) { | |
836af83b DT |
1197 | slot = q->slot_config; |
1198 | if (slot.max_packets == INT_MAX) | |
1199 | slot.max_packets = 0; | |
1200 | if (slot.max_bytes == INT_MAX) | |
1201 | slot.max_bytes = 0; | |
1202 | if (nla_put(skb, TCA_NETEM_SLOT, sizeof(slot), &slot)) | |
1203 | goto nla_put_failure; | |
1204 | } | |
1205 | ||
861d7f74 | 1206 | return nla_nest_end(skb, nla); |
1da177e4 | 1207 | |
1e90474c | 1208 | nla_put_failure: |
861d7f74 | 1209 | nlmsg_trim(skb, nla); |
1da177e4 LT |
1210 | return -1; |
1211 | } | |
1212 | ||
10f6dfcf | 1213 | static int netem_dump_class(struct Qdisc *sch, unsigned long cl, |
1214 | struct sk_buff *skb, struct tcmsg *tcm) | |
1215 | { | |
1216 | struct netem_sched_data *q = qdisc_priv(sch); | |
1217 | ||
50612537 | 1218 | if (cl != 1 || !q->qdisc) /* only one class */ |
10f6dfcf | 1219 | return -ENOENT; |
1220 | ||
1221 | tcm->tcm_handle |= TC_H_MIN(1); | |
1222 | tcm->tcm_info = q->qdisc->handle; | |
1223 | ||
1224 | return 0; | |
1225 | } | |
1226 | ||
1227 | static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
653d6fd6 | 1228 | struct Qdisc **old, struct netlink_ext_ack *extack) |
10f6dfcf | 1229 | { |
1230 | struct netem_sched_data *q = qdisc_priv(sch); | |
1231 | ||
86a7996c | 1232 | *old = qdisc_replace(sch, new, &q->qdisc); |
10f6dfcf | 1233 | return 0; |
1234 | } | |
1235 | ||
1236 | static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg) | |
1237 | { | |
1238 | struct netem_sched_data *q = qdisc_priv(sch); | |
1239 | return q->qdisc; | |
1240 | } | |
1241 | ||
143976ce | 1242 | static unsigned long netem_find(struct Qdisc *sch, u32 classid) |
10f6dfcf | 1243 | { |
1244 | return 1; | |
1245 | } | |
1246 | ||
10f6dfcf | 1247 | static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker) |
1248 | { | |
1249 | if (!walker->stop) { | |
1250 | if (walker->count >= walker->skip) | |
1251 | if (walker->fn(sch, 1, walker) < 0) { | |
1252 | walker->stop = 1; | |
1253 | return; | |
1254 | } | |
1255 | walker->count++; | |
1256 | } | |
1257 | } | |
1258 | ||
1259 | static const struct Qdisc_class_ops netem_class_ops = { | |
1260 | .graft = netem_graft, | |
1261 | .leaf = netem_leaf, | |
143976ce | 1262 | .find = netem_find, |
10f6dfcf | 1263 | .walk = netem_walk, |
1264 | .dump = netem_dump_class, | |
1265 | }; | |
1266 | ||
20fea08b | 1267 | static struct Qdisc_ops netem_qdisc_ops __read_mostly = { |
1da177e4 | 1268 | .id = "netem", |
10f6dfcf | 1269 | .cl_ops = &netem_class_ops, |
1da177e4 LT |
1270 | .priv_size = sizeof(struct netem_sched_data), |
1271 | .enqueue = netem_enqueue, | |
1272 | .dequeue = netem_dequeue, | |
77be155c | 1273 | .peek = qdisc_peek_dequeued, |
1da177e4 LT |
1274 | .init = netem_init, |
1275 | .reset = netem_reset, | |
1276 | .destroy = netem_destroy, | |
1277 | .change = netem_change, | |
1278 | .dump = netem_dump, | |
1279 | .owner = THIS_MODULE, | |
1280 | }; | |
1281 | ||
1282 | ||
1283 | static int __init netem_module_init(void) | |
1284 | { | |
eb229c4c | 1285 | pr_info("netem: version " VERSION "\n"); |
1da177e4 LT |
1286 | return register_qdisc(&netem_qdisc_ops); |
1287 | } | |
1288 | static void __exit netem_module_exit(void) | |
1289 | { | |
1290 | unregister_qdisc(&netem_qdisc_ops); | |
1291 | } | |
1292 | module_init(netem_module_init) | |
1293 | module_exit(netem_module_exit) | |
1294 | MODULE_LICENSE("GPL"); |