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 7 * 2 of the License. 8 * 9 * Many of the algorithms and ideas for this came from 10 * NIST Net which is not copyrighted. 11 * 12 * Authors: Stephen Hemminger <shemminger@osdl.org> 13 * Catalin(ux aka Dino) BOIE <catab at umbrella dot ro> 14 */ 15 16 #include <linux/mm.h> 17 #include <linux/module.h> 18 #include <linux/slab.h> 19 #include <linux/types.h> 20 #include <linux/kernel.h> 21 #include <linux/errno.h> 22 #include <linux/skbuff.h> 23 #include <linux/vmalloc.h> 24 #include <linux/rtnetlink.h> 25 #include <linux/reciprocal_div.h> 26 27 #include <net/netlink.h> 28 #include <net/pkt_sched.h> 29 30 #define VERSION "1.3" 31 32 /* Network Emulation Queuing algorithm. 33 ==================================== 34 35 Sources: [1] Mark Carson, Darrin Santay, "NIST Net - A Linux-based 36 Network Emulation Tool 37 [2] Luigi Rizzo, DummyNet for FreeBSD 38 39 ---------------------------------------------------------------- 40 41 This started out as a simple way to delay outgoing packets to 42 test TCP but has grown to include most of the functionality 43 of a full blown network emulator like NISTnet. It can delay 44 packets and add random jitter (and correlation). The random 45 distribution can be loaded from a table as well to provide 46 normal, Pareto, or experimental curves. Packet loss, 47 duplication, and reordering can also be emulated. 48 49 This qdisc does not do classification that can be handled in 50 layering other disciplines. It does not need to do bandwidth 51 control either since that can be handled by using token 52 bucket or other rate control. 53 54 Correlated Loss Generator models 55 56 Added generation of correlated loss according to the 57 "Gilbert-Elliot" model, a 4-state markov model. 58 59 References: 60 [1] NetemCLG Home http://netgroup.uniroma2.it/NetemCLG 61 [2] S. Salsano, F. Ludovici, A. Ordine, "Definition of a general 62 and intuitive loss model for packet networks and its implementation 63 in the Netem module in the Linux kernel", available in [1] 64 65 Authors: Stefano Salsano <stefano.salsano at uniroma2.it 66 Fabio Ludovici <fabio.ludovici at yahoo.it> 67 */ 68 69 struct netem_sched_data { 70 /* internal t(ime)fifo qdisc uses sch->q and sch->limit */ 71 72 /* optional qdisc for classful handling (NULL at netem init) */ 73 struct Qdisc *qdisc; 74 75 struct qdisc_watchdog watchdog; 76 77 psched_tdiff_t latency; 78 psched_tdiff_t jitter; 79 80 u32 loss; 81 u32 limit; 82 u32 counter; 83 u32 gap; 84 u32 duplicate; 85 u32 reorder; 86 u32 corrupt; 87 u32 rate; 88 s32 packet_overhead; 89 u32 cell_size; 90 u32 cell_size_reciprocal; 91 s32 cell_overhead; 92 93 struct crndstate { 94 u32 last; 95 u32 rho; 96 } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor; 97 98 struct disttable { 99 u32 size; 100 s16 table[0]; 101 } *delay_dist; 102 103 enum { 104 CLG_RANDOM, 105 CLG_4_STATES, 106 CLG_GILB_ELL, 107 } loss_model; 108 109 /* Correlated Loss Generation models */ 110 struct clgstate { 111 /* state of the Markov chain */ 112 u8 state; 113 114 /* 4-states and Gilbert-Elliot models */ 115 u32 a1; /* p13 for 4-states or p for GE */ 116 u32 a2; /* p31 for 4-states or r for GE */ 117 u32 a3; /* p32 for 4-states or h for GE */ 118 u32 a4; /* p14 for 4-states or 1-k for GE */ 119 u32 a5; /* p23 used only in 4-states */ 120 } clg; 121 122 }; 123 124 /* Time stamp put into socket buffer control block 125 * Only valid when skbs are in our internal t(ime)fifo queue. 126 */ 127 struct netem_skb_cb { 128 psched_time_t time_to_send; 129 }; 130 131 static inline struct netem_skb_cb *netem_skb_cb(struct sk_buff *skb) 132 { 133 qdisc_cb_private_validate(skb, sizeof(struct netem_skb_cb)); 134 return (struct netem_skb_cb *)qdisc_skb_cb(skb)->data; 135 } 136 137 /* init_crandom - initialize correlated random number generator 138 * Use entropy source for initial seed. 139 */ 140 static void init_crandom(struct crndstate *state, unsigned long rho) 141 { 142 state->rho = rho; 143 state->last = net_random(); 144 } 145 146 /* get_crandom - correlated random number generator 147 * Next number depends on last value. 148 * rho is scaled to avoid floating point. 149 */ 150 static u32 get_crandom(struct crndstate *state) 151 { 152 u64 value, rho; 153 unsigned long answer; 154 155 if (state->rho == 0) /* no correlation */ 156 return net_random(); 157 158 value = net_random(); 159 rho = (u64)state->rho + 1; 160 answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32; 161 state->last = answer; 162 return answer; 163 } 164 165 /* loss_4state - 4-state model loss generator 166 * Generates losses according to the 4-state Markov chain adopted in 167 * the GI (General and Intuitive) loss model. 168 */ 169 static bool loss_4state(struct netem_sched_data *q) 170 { 171 struct clgstate *clg = &q->clg; 172 u32 rnd = net_random(); 173 174 /* 175 * Makes a comparison between rnd and the transition 176 * probabilities outgoing from the current state, then decides the 177 * next state and if the next packet has to be transmitted or lost. 178 * The four states correspond to: 179 * 1 => successfully transmitted packets within a gap period 180 * 4 => isolated losses within a gap period 181 * 3 => lost packets within a burst period 182 * 2 => successfully transmitted packets within a burst period 183 */ 184 switch (clg->state) { 185 case 1: 186 if (rnd < clg->a4) { 187 clg->state = 4; 188 return true; 189 } else if (clg->a4 < rnd && rnd < clg->a1) { 190 clg->state = 3; 191 return true; 192 } else if (clg->a1 < rnd) 193 clg->state = 1; 194 195 break; 196 case 2: 197 if (rnd < clg->a5) { 198 clg->state = 3; 199 return true; 200 } else 201 clg->state = 2; 202 203 break; 204 case 3: 205 if (rnd < clg->a3) 206 clg->state = 2; 207 else if (clg->a3 < rnd && rnd < clg->a2 + clg->a3) { 208 clg->state = 1; 209 return true; 210 } else if (clg->a2 + clg->a3 < rnd) { 211 clg->state = 3; 212 return true; 213 } 214 break; 215 case 4: 216 clg->state = 1; 217 break; 218 } 219 220 return false; 221 } 222 223 /* loss_gilb_ell - Gilbert-Elliot model loss generator 224 * Generates losses according to the Gilbert-Elliot loss model or 225 * its special cases (Gilbert or Simple Gilbert) 226 * 227 * Makes a comparison between random number and the transition 228 * probabilities outgoing from the current state, then decides the 229 * next state. A second random number is extracted and the comparison 230 * with the loss probability of the current state decides if the next 231 * packet will be transmitted or lost. 232 */ 233 static bool loss_gilb_ell(struct netem_sched_data *q) 234 { 235 struct clgstate *clg = &q->clg; 236 237 switch (clg->state) { 238 case 1: 239 if (net_random() < clg->a1) 240 clg->state = 2; 241 if (net_random() < clg->a4) 242 return true; 243 case 2: 244 if (net_random() < clg->a2) 245 clg->state = 1; 246 if (clg->a3 > net_random()) 247 return true; 248 } 249 250 return false; 251 } 252 253 static bool loss_event(struct netem_sched_data *q) 254 { 255 switch (q->loss_model) { 256 case CLG_RANDOM: 257 /* Random packet drop 0 => none, ~0 => all */ 258 return q->loss && q->loss >= get_crandom(&q->loss_cor); 259 260 case CLG_4_STATES: 261 /* 4state loss model algorithm (used also for GI model) 262 * Extracts a value from the markov 4 state loss generator, 263 * if it is 1 drops a packet and if needed writes the event in 264 * the kernel logs 265 */ 266 return loss_4state(q); 267 268 case CLG_GILB_ELL: 269 /* Gilbert-Elliot loss model algorithm 270 * Extracts a value from the Gilbert-Elliot loss generator, 271 * if it is 1 drops a packet and if needed writes the event in 272 * the kernel logs 273 */ 274 return loss_gilb_ell(q); 275 } 276 277 return false; /* not reached */ 278 } 279 280 281 /* tabledist - return a pseudo-randomly distributed value with mean mu and 282 * std deviation sigma. Uses table lookup to approximate the desired 283 * distribution, and a uniformly-distributed pseudo-random source. 284 */ 285 static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma, 286 struct crndstate *state, 287 const struct disttable *dist) 288 { 289 psched_tdiff_t x; 290 long t; 291 u32 rnd; 292 293 if (sigma == 0) 294 return mu; 295 296 rnd = get_crandom(state); 297 298 /* default uniform distribution */ 299 if (dist == NULL) 300 return (rnd % (2*sigma)) - sigma + mu; 301 302 t = dist->table[rnd % dist->size]; 303 x = (sigma % NETEM_DIST_SCALE) * t; 304 if (x >= 0) 305 x += NETEM_DIST_SCALE/2; 306 else 307 x -= NETEM_DIST_SCALE/2; 308 309 return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu; 310 } 311 312 static psched_time_t packet_len_2_sched_time(unsigned int len, struct netem_sched_data *q) 313 { 314 u64 ticks; 315 316 len += q->packet_overhead; 317 318 if (q->cell_size) { 319 u32 cells = reciprocal_divide(len, q->cell_size_reciprocal); 320 321 if (len > cells * q->cell_size) /* extra cell needed for remainder */ 322 cells++; 323 len = cells * (q->cell_size + q->cell_overhead); 324 } 325 326 ticks = (u64)len * NSEC_PER_SEC; 327 328 do_div(ticks, q->rate); 329 return PSCHED_NS2TICKS(ticks); 330 } 331 332 static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch) 333 { 334 struct sk_buff_head *list = &sch->q; 335 psched_time_t tnext = netem_skb_cb(nskb)->time_to_send; 336 struct sk_buff *skb; 337 338 if (likely(skb_queue_len(list) < sch->limit)) { 339 skb = skb_peek_tail(list); 340 /* Optimize for add at tail */ 341 if (likely(!skb || tnext >= netem_skb_cb(skb)->time_to_send)) 342 return qdisc_enqueue_tail(nskb, sch); 343 344 skb_queue_reverse_walk(list, skb) { 345 if (tnext >= netem_skb_cb(skb)->time_to_send) 346 break; 347 } 348 349 __skb_queue_after(list, skb, nskb); 350 sch->qstats.backlog += qdisc_pkt_len(nskb); 351 return NET_XMIT_SUCCESS; 352 } 353 354 return qdisc_reshape_fail(nskb, sch); 355 } 356 357 /* 358 * Insert one skb into qdisc. 359 * Note: parent depends on return value to account for queue length. 360 * NET_XMIT_DROP: queue length didn't change. 361 * NET_XMIT_SUCCESS: one skb was queued. 362 */ 363 static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch) 364 { 365 struct netem_sched_data *q = qdisc_priv(sch); 366 /* We don't fill cb now as skb_unshare() may invalidate it */ 367 struct netem_skb_cb *cb; 368 struct sk_buff *skb2; 369 int ret; 370 int count = 1; 371 372 /* Random duplication */ 373 if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor)) 374 ++count; 375 376 /* Drop packet? */ 377 if (loss_event(q)) 378 --count; 379 380 if (count == 0) { 381 sch->qstats.drops++; 382 kfree_skb(skb); 383 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 384 } 385 386 skb_orphan(skb); 387 388 /* 389 * If we need to duplicate packet, then re-insert at top of the 390 * qdisc tree, since parent queuer expects that only one 391 * skb will be queued. 392 */ 393 if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) { 394 struct Qdisc *rootq = qdisc_root(sch); 395 u32 dupsave = q->duplicate; /* prevent duplicating a dup... */ 396 q->duplicate = 0; 397 398 qdisc_enqueue_root(skb2, rootq); 399 q->duplicate = dupsave; 400 } 401 402 /* 403 * Randomized packet corruption. 404 * Make copy if needed since we are modifying 405 * If packet is going to be hardware checksummed, then 406 * do it now in software before we mangle it. 407 */ 408 if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) { 409 if (!(skb = skb_unshare(skb, GFP_ATOMIC)) || 410 (skb->ip_summed == CHECKSUM_PARTIAL && 411 skb_checksum_help(skb))) 412 return qdisc_drop(skb, sch); 413 414 skb->data[net_random() % skb_headlen(skb)] ^= 1<<(net_random() % 8); 415 } 416 417 cb = netem_skb_cb(skb); 418 if (q->gap == 0 || /* not doing reordering */ 419 q->counter < q->gap - 1 || /* inside last reordering gap */ 420 q->reorder < get_crandom(&q->reorder_cor)) { 421 psched_time_t now; 422 psched_tdiff_t delay; 423 424 delay = tabledist(q->latency, q->jitter, 425 &q->delay_cor, q->delay_dist); 426 427 now = psched_get_time(); 428 429 if (q->rate) { 430 struct sk_buff_head *list = &sch->q; 431 432 delay += packet_len_2_sched_time(skb->len, q); 433 434 if (!skb_queue_empty(list)) { 435 /* 436 * Last packet in queue is reference point (now). 437 * First packet in queue is already in flight, 438 * calculate this time bonus and substract 439 * from delay. 440 */ 441 delay -= now - netem_skb_cb(skb_peek(list))->time_to_send; 442 now = netem_skb_cb(skb_peek_tail(list))->time_to_send; 443 } 444 } 445 446 cb->time_to_send = now + delay; 447 ++q->counter; 448 ret = tfifo_enqueue(skb, sch); 449 } else { 450 /* 451 * Do re-ordering by putting one out of N packets at the front 452 * of the queue. 453 */ 454 cb->time_to_send = psched_get_time(); 455 q->counter = 0; 456 457 __skb_queue_head(&sch->q, skb); 458 sch->qstats.backlog += qdisc_pkt_len(skb); 459 sch->qstats.requeues++; 460 ret = NET_XMIT_SUCCESS; 461 } 462 463 if (ret != NET_XMIT_SUCCESS) { 464 if (net_xmit_drop_count(ret)) { 465 sch->qstats.drops++; 466 return ret; 467 } 468 } 469 470 return NET_XMIT_SUCCESS; 471 } 472 473 static unsigned int netem_drop(struct Qdisc *sch) 474 { 475 struct netem_sched_data *q = qdisc_priv(sch); 476 unsigned int len; 477 478 len = qdisc_queue_drop(sch); 479 if (!len && q->qdisc && q->qdisc->ops->drop) 480 len = q->qdisc->ops->drop(q->qdisc); 481 if (len) 482 sch->qstats.drops++; 483 484 return len; 485 } 486 487 static struct sk_buff *netem_dequeue(struct Qdisc *sch) 488 { 489 struct netem_sched_data *q = qdisc_priv(sch); 490 struct sk_buff *skb; 491 492 if (qdisc_is_throttled(sch)) 493 return NULL; 494 495 tfifo_dequeue: 496 skb = qdisc_peek_head(sch); 497 if (skb) { 498 const struct netem_skb_cb *cb = netem_skb_cb(skb); 499 500 /* if more time remaining? */ 501 if (cb->time_to_send <= psched_get_time()) { 502 __skb_unlink(skb, &sch->q); 503 sch->qstats.backlog -= qdisc_pkt_len(skb); 504 505 #ifdef CONFIG_NET_CLS_ACT 506 /* 507 * If it's at ingress let's pretend the delay is 508 * from the network (tstamp will be updated). 509 */ 510 if (G_TC_FROM(skb->tc_verd) & AT_INGRESS) 511 skb->tstamp.tv64 = 0; 512 #endif 513 514 if (q->qdisc) { 515 int err = qdisc_enqueue(skb, q->qdisc); 516 517 if (unlikely(err != NET_XMIT_SUCCESS)) { 518 if (net_xmit_drop_count(err)) { 519 sch->qstats.drops++; 520 qdisc_tree_decrease_qlen(sch, 1); 521 } 522 } 523 goto tfifo_dequeue; 524 } 525 deliver: 526 qdisc_unthrottled(sch); 527 qdisc_bstats_update(sch, skb); 528 return skb; 529 } 530 531 if (q->qdisc) { 532 skb = q->qdisc->ops->dequeue(q->qdisc); 533 if (skb) 534 goto deliver; 535 } 536 qdisc_watchdog_schedule(&q->watchdog, cb->time_to_send); 537 } 538 539 if (q->qdisc) { 540 skb = q->qdisc->ops->dequeue(q->qdisc); 541 if (skb) 542 goto deliver; 543 } 544 return NULL; 545 } 546 547 static void netem_reset(struct Qdisc *sch) 548 { 549 struct netem_sched_data *q = qdisc_priv(sch); 550 551 qdisc_reset_queue(sch); 552 if (q->qdisc) 553 qdisc_reset(q->qdisc); 554 qdisc_watchdog_cancel(&q->watchdog); 555 } 556 557 static void dist_free(struct disttable *d) 558 { 559 if (d) { 560 if (is_vmalloc_addr(d)) 561 vfree(d); 562 else 563 kfree(d); 564 } 565 } 566 567 /* 568 * Distribution data is a variable size payload containing 569 * signed 16 bit values. 570 */ 571 static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr) 572 { 573 struct netem_sched_data *q = qdisc_priv(sch); 574 size_t n = nla_len(attr)/sizeof(__s16); 575 const __s16 *data = nla_data(attr); 576 spinlock_t *root_lock; 577 struct disttable *d; 578 int i; 579 size_t s; 580 581 if (n > NETEM_DIST_MAX) 582 return -EINVAL; 583 584 s = sizeof(struct disttable) + n * sizeof(s16); 585 d = kmalloc(s, GFP_KERNEL | __GFP_NOWARN); 586 if (!d) 587 d = vmalloc(s); 588 if (!d) 589 return -ENOMEM; 590 591 d->size = n; 592 for (i = 0; i < n; i++) 593 d->table[i] = data[i]; 594 595 root_lock = qdisc_root_sleeping_lock(sch); 596 597 spin_lock_bh(root_lock); 598 swap(q->delay_dist, d); 599 spin_unlock_bh(root_lock); 600 601 dist_free(d); 602 return 0; 603 } 604 605 static void get_correlation(struct Qdisc *sch, const struct nlattr *attr) 606 { 607 struct netem_sched_data *q = qdisc_priv(sch); 608 const struct tc_netem_corr *c = nla_data(attr); 609 610 init_crandom(&q->delay_cor, c->delay_corr); 611 init_crandom(&q->loss_cor, c->loss_corr); 612 init_crandom(&q->dup_cor, c->dup_corr); 613 } 614 615 static void get_reorder(struct Qdisc *sch, const struct nlattr *attr) 616 { 617 struct netem_sched_data *q = qdisc_priv(sch); 618 const struct tc_netem_reorder *r = nla_data(attr); 619 620 q->reorder = r->probability; 621 init_crandom(&q->reorder_cor, r->correlation); 622 } 623 624 static void get_corrupt(struct Qdisc *sch, const struct nlattr *attr) 625 { 626 struct netem_sched_data *q = qdisc_priv(sch); 627 const struct tc_netem_corrupt *r = nla_data(attr); 628 629 q->corrupt = r->probability; 630 init_crandom(&q->corrupt_cor, r->correlation); 631 } 632 633 static void get_rate(struct Qdisc *sch, const struct nlattr *attr) 634 { 635 struct netem_sched_data *q = qdisc_priv(sch); 636 const struct tc_netem_rate *r = nla_data(attr); 637 638 q->rate = r->rate; 639 q->packet_overhead = r->packet_overhead; 640 q->cell_size = r->cell_size; 641 if (q->cell_size) 642 q->cell_size_reciprocal = reciprocal_value(q->cell_size); 643 q->cell_overhead = r->cell_overhead; 644 } 645 646 static int get_loss_clg(struct Qdisc *sch, const struct nlattr *attr) 647 { 648 struct netem_sched_data *q = qdisc_priv(sch); 649 const struct nlattr *la; 650 int rem; 651 652 nla_for_each_nested(la, attr, rem) { 653 u16 type = nla_type(la); 654 655 switch(type) { 656 case NETEM_LOSS_GI: { 657 const struct tc_netem_gimodel *gi = nla_data(la); 658 659 if (nla_len(la) < sizeof(struct tc_netem_gimodel)) { 660 pr_info("netem: incorrect gi model size\n"); 661 return -EINVAL; 662 } 663 664 q->loss_model = CLG_4_STATES; 665 666 q->clg.state = 1; 667 q->clg.a1 = gi->p13; 668 q->clg.a2 = gi->p31; 669 q->clg.a3 = gi->p32; 670 q->clg.a4 = gi->p14; 671 q->clg.a5 = gi->p23; 672 break; 673 } 674 675 case NETEM_LOSS_GE: { 676 const struct tc_netem_gemodel *ge = nla_data(la); 677 678 if (nla_len(la) < sizeof(struct tc_netem_gemodel)) { 679 pr_info("netem: incorrect ge model size\n"); 680 return -EINVAL; 681 } 682 683 q->loss_model = CLG_GILB_ELL; 684 q->clg.state = 1; 685 q->clg.a1 = ge->p; 686 q->clg.a2 = ge->r; 687 q->clg.a3 = ge->h; 688 q->clg.a4 = ge->k1; 689 break; 690 } 691 692 default: 693 pr_info("netem: unknown loss type %u\n", type); 694 return -EINVAL; 695 } 696 } 697 698 return 0; 699 } 700 701 static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = { 702 [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) }, 703 [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) }, 704 [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) }, 705 [TCA_NETEM_RATE] = { .len = sizeof(struct tc_netem_rate) }, 706 [TCA_NETEM_LOSS] = { .type = NLA_NESTED }, 707 }; 708 709 static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla, 710 const struct nla_policy *policy, int len) 711 { 712 int nested_len = nla_len(nla) - NLA_ALIGN(len); 713 714 if (nested_len < 0) { 715 pr_info("netem: invalid attributes len %d\n", nested_len); 716 return -EINVAL; 717 } 718 719 if (nested_len >= nla_attr_size(0)) 720 return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len), 721 nested_len, policy); 722 723 memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1)); 724 return 0; 725 } 726 727 /* Parse netlink message to set options */ 728 static int netem_change(struct Qdisc *sch, struct nlattr *opt) 729 { 730 struct netem_sched_data *q = qdisc_priv(sch); 731 struct nlattr *tb[TCA_NETEM_MAX + 1]; 732 struct tc_netem_qopt *qopt; 733 int ret; 734 735 if (opt == NULL) 736 return -EINVAL; 737 738 qopt = nla_data(opt); 739 ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt)); 740 if (ret < 0) 741 return ret; 742 743 sch->limit = qopt->limit; 744 745 q->latency = qopt->latency; 746 q->jitter = qopt->jitter; 747 q->limit = qopt->limit; 748 q->gap = qopt->gap; 749 q->counter = 0; 750 q->loss = qopt->loss; 751 q->duplicate = qopt->duplicate; 752 753 /* for compatibility with earlier versions. 754 * if gap is set, need to assume 100% probability 755 */ 756 if (q->gap) 757 q->reorder = ~0; 758 759 if (tb[TCA_NETEM_CORR]) 760 get_correlation(sch, tb[TCA_NETEM_CORR]); 761 762 if (tb[TCA_NETEM_DELAY_DIST]) { 763 ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]); 764 if (ret) 765 return ret; 766 } 767 768 if (tb[TCA_NETEM_REORDER]) 769 get_reorder(sch, tb[TCA_NETEM_REORDER]); 770 771 if (tb[TCA_NETEM_CORRUPT]) 772 get_corrupt(sch, tb[TCA_NETEM_CORRUPT]); 773 774 if (tb[TCA_NETEM_RATE]) 775 get_rate(sch, tb[TCA_NETEM_RATE]); 776 777 q->loss_model = CLG_RANDOM; 778 if (tb[TCA_NETEM_LOSS]) 779 ret = get_loss_clg(sch, tb[TCA_NETEM_LOSS]); 780 781 return ret; 782 } 783 784 static int netem_init(struct Qdisc *sch, struct nlattr *opt) 785 { 786 struct netem_sched_data *q = qdisc_priv(sch); 787 int ret; 788 789 if (!opt) 790 return -EINVAL; 791 792 qdisc_watchdog_init(&q->watchdog, sch); 793 794 q->loss_model = CLG_RANDOM; 795 ret = netem_change(sch, opt); 796 if (ret) 797 pr_info("netem: change failed\n"); 798 return ret; 799 } 800 801 static void netem_destroy(struct Qdisc *sch) 802 { 803 struct netem_sched_data *q = qdisc_priv(sch); 804 805 qdisc_watchdog_cancel(&q->watchdog); 806 if (q->qdisc) 807 qdisc_destroy(q->qdisc); 808 dist_free(q->delay_dist); 809 } 810 811 static int dump_loss_model(const struct netem_sched_data *q, 812 struct sk_buff *skb) 813 { 814 struct nlattr *nest; 815 816 nest = nla_nest_start(skb, TCA_NETEM_LOSS); 817 if (nest == NULL) 818 goto nla_put_failure; 819 820 switch (q->loss_model) { 821 case CLG_RANDOM: 822 /* legacy loss model */ 823 nla_nest_cancel(skb, nest); 824 return 0; /* no data */ 825 826 case CLG_4_STATES: { 827 struct tc_netem_gimodel gi = { 828 .p13 = q->clg.a1, 829 .p31 = q->clg.a2, 830 .p32 = q->clg.a3, 831 .p14 = q->clg.a4, 832 .p23 = q->clg.a5, 833 }; 834 835 NLA_PUT(skb, NETEM_LOSS_GI, sizeof(gi), &gi); 836 break; 837 } 838 case CLG_GILB_ELL: { 839 struct tc_netem_gemodel ge = { 840 .p = q->clg.a1, 841 .r = q->clg.a2, 842 .h = q->clg.a3, 843 .k1 = q->clg.a4, 844 }; 845 846 NLA_PUT(skb, NETEM_LOSS_GE, sizeof(ge), &ge); 847 break; 848 } 849 } 850 851 nla_nest_end(skb, nest); 852 return 0; 853 854 nla_put_failure: 855 nla_nest_cancel(skb, nest); 856 return -1; 857 } 858 859 static int netem_dump(struct Qdisc *sch, struct sk_buff *skb) 860 { 861 const struct netem_sched_data *q = qdisc_priv(sch); 862 struct nlattr *nla = (struct nlattr *) skb_tail_pointer(skb); 863 struct tc_netem_qopt qopt; 864 struct tc_netem_corr cor; 865 struct tc_netem_reorder reorder; 866 struct tc_netem_corrupt corrupt; 867 struct tc_netem_rate rate; 868 869 qopt.latency = q->latency; 870 qopt.jitter = q->jitter; 871 qopt.limit = q->limit; 872 qopt.loss = q->loss; 873 qopt.gap = q->gap; 874 qopt.duplicate = q->duplicate; 875 NLA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt); 876 877 cor.delay_corr = q->delay_cor.rho; 878 cor.loss_corr = q->loss_cor.rho; 879 cor.dup_corr = q->dup_cor.rho; 880 NLA_PUT(skb, TCA_NETEM_CORR, sizeof(cor), &cor); 881 882 reorder.probability = q->reorder; 883 reorder.correlation = q->reorder_cor.rho; 884 NLA_PUT(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder); 885 886 corrupt.probability = q->corrupt; 887 corrupt.correlation = q->corrupt_cor.rho; 888 NLA_PUT(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt); 889 890 rate.rate = q->rate; 891 rate.packet_overhead = q->packet_overhead; 892 rate.cell_size = q->cell_size; 893 rate.cell_overhead = q->cell_overhead; 894 NLA_PUT(skb, TCA_NETEM_RATE, sizeof(rate), &rate); 895 896 if (dump_loss_model(q, skb) != 0) 897 goto nla_put_failure; 898 899 return nla_nest_end(skb, nla); 900 901 nla_put_failure: 902 nlmsg_trim(skb, nla); 903 return -1; 904 } 905 906 static int netem_dump_class(struct Qdisc *sch, unsigned long cl, 907 struct sk_buff *skb, struct tcmsg *tcm) 908 { 909 struct netem_sched_data *q = qdisc_priv(sch); 910 911 if (cl != 1 || !q->qdisc) /* only one class */ 912 return -ENOENT; 913 914 tcm->tcm_handle |= TC_H_MIN(1); 915 tcm->tcm_info = q->qdisc->handle; 916 917 return 0; 918 } 919 920 static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 921 struct Qdisc **old) 922 { 923 struct netem_sched_data *q = qdisc_priv(sch); 924 925 sch_tree_lock(sch); 926 *old = q->qdisc; 927 q->qdisc = new; 928 if (*old) { 929 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); 930 qdisc_reset(*old); 931 } 932 sch_tree_unlock(sch); 933 934 return 0; 935 } 936 937 static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg) 938 { 939 struct netem_sched_data *q = qdisc_priv(sch); 940 return q->qdisc; 941 } 942 943 static unsigned long netem_get(struct Qdisc *sch, u32 classid) 944 { 945 return 1; 946 } 947 948 static void netem_put(struct Qdisc *sch, unsigned long arg) 949 { 950 } 951 952 static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker) 953 { 954 if (!walker->stop) { 955 if (walker->count >= walker->skip) 956 if (walker->fn(sch, 1, walker) < 0) { 957 walker->stop = 1; 958 return; 959 } 960 walker->count++; 961 } 962 } 963 964 static const struct Qdisc_class_ops netem_class_ops = { 965 .graft = netem_graft, 966 .leaf = netem_leaf, 967 .get = netem_get, 968 .put = netem_put, 969 .walk = netem_walk, 970 .dump = netem_dump_class, 971 }; 972 973 static struct Qdisc_ops netem_qdisc_ops __read_mostly = { 974 .id = "netem", 975 .cl_ops = &netem_class_ops, 976 .priv_size = sizeof(struct netem_sched_data), 977 .enqueue = netem_enqueue, 978 .dequeue = netem_dequeue, 979 .peek = qdisc_peek_dequeued, 980 .drop = netem_drop, 981 .init = netem_init, 982 .reset = netem_reset, 983 .destroy = netem_destroy, 984 .change = netem_change, 985 .dump = netem_dump, 986 .owner = THIS_MODULE, 987 }; 988 989 990 static int __init netem_module_init(void) 991 { 992 pr_info("netem: version " VERSION "\n"); 993 return register_qdisc(&netem_qdisc_ops); 994 } 995 static void __exit netem_module_exit(void) 996 { 997 unregister_qdisc(&netem_qdisc_ops); 998 } 999 module_init(netem_module_init) 1000 module_exit(netem_module_exit) 1001 MODULE_LICENSE("GPL"); 1002