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/module.h> 17#include <linux/slab.h> 18#include <linux/types.h> 19#include <linux/kernel.h> 20#include <linux/errno.h> 21#include <linux/skbuff.h> 22#include <linux/rtnetlink.h> 23 24#include <net/netlink.h> 25#include <net/pkt_sched.h> 26 27#define VERSION "1.2" 28 29/* Network Emulation Queuing algorithm. 30 ==================================== 31 32 Sources: [1] Mark Carson, Darrin Santay, "NIST Net - A Linux-based 33 Network Emulation Tool 34 [2] Luigi Rizzo, DummyNet for FreeBSD 35 36 ---------------------------------------------------------------- 37 38 This started out as a simple way to delay outgoing packets to 39 test TCP but has grown to include most of the functionality 40 of a full blown network emulator like NISTnet. It can delay 41 packets and add random jitter (and correlation). The random 42 distribution can be loaded from a table as well to provide 43 normal, Pareto, or experimental curves. Packet loss, 44 duplication, and reordering can also be emulated. 45 46 This qdisc does not do classification that can be handled in 47 layering other disciplines. It does not need to do bandwidth 48 control either since that can be handled by using token 49 bucket or other rate control. 50*/ 51 52struct netem_sched_data { 53 struct Qdisc *qdisc; 54 struct qdisc_watchdog watchdog; 55 56 psched_tdiff_t latency; 57 psched_tdiff_t jitter; 58 59 u32 loss; 60 u32 limit; 61 u32 counter; 62 u32 gap; 63 u32 duplicate; 64 u32 reorder; 65 u32 corrupt; 66 67 struct crndstate { 68 u32 last; 69 u32 rho; 70 } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor; 71 72 struct disttable { 73 u32 size; 74 s16 table[0]; 75 } *delay_dist; 76}; 77 78/* Time stamp put into socket buffer control block */ 79struct netem_skb_cb { 80 psched_time_t time_to_send; 81}; 82 83static inline struct netem_skb_cb *netem_skb_cb(struct sk_buff *skb) 84{ 85 BUILD_BUG_ON(sizeof(skb->cb) < 86 sizeof(struct qdisc_skb_cb) + sizeof(struct netem_skb_cb)); 87 return (struct netem_skb_cb *)qdisc_skb_cb(skb)->data; 88} 89 90/* init_crandom - initialize correlated random number generator 91 * Use entropy source for initial seed. 92 */ 93static void init_crandom(struct crndstate *state, unsigned long rho) 94{ 95 state->rho = rho; 96 state->last = net_random(); 97} 98 99/* get_crandom - correlated random number generator 100 * Next number depends on last value. 101 * rho is scaled to avoid floating point. 102 */ 103static u32 get_crandom(struct crndstate *state) 104{ 105 u64 value, rho; 106 unsigned long answer; 107 108 if (state->rho == 0) /* no correlation */ 109 return net_random(); 110 111 value = net_random(); 112 rho = (u64)state->rho + 1; 113 answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32; 114 state->last = answer; 115 return answer; 116} 117 118/* tabledist - return a pseudo-randomly distributed value with mean mu and 119 * std deviation sigma. Uses table lookup to approximate the desired 120 * distribution, and a uniformly-distributed pseudo-random source. 121 */ 122static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma, 123 struct crndstate *state, 124 const struct disttable *dist) 125{ 126 psched_tdiff_t x; 127 long t; 128 u32 rnd; 129 130 if (sigma == 0) 131 return mu; 132 133 rnd = get_crandom(state); 134 135 /* default uniform distribution */ 136 if (dist == NULL) 137 return (rnd % (2*sigma)) - sigma + mu; 138 139 t = dist->table[rnd % dist->size]; 140 x = (sigma % NETEM_DIST_SCALE) * t; 141 if (x >= 0) 142 x += NETEM_DIST_SCALE/2; 143 else 144 x -= NETEM_DIST_SCALE/2; 145 146 return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu; 147} 148 149/* 150 * Insert one skb into qdisc. 151 * Note: parent depends on return value to account for queue length. 152 * NET_XMIT_DROP: queue length didn't change. 153 * NET_XMIT_SUCCESS: one skb was queued. 154 */ 155static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch) 156{ 157 struct netem_sched_data *q = qdisc_priv(sch); 158 /* We don't fill cb now as skb_unshare() may invalidate it */ 159 struct netem_skb_cb *cb; 160 struct sk_buff *skb2; 161 int ret; 162 int count = 1; 163 164 pr_debug("netem_enqueue skb=%p\n", skb); 165 166 /* Random duplication */ 167 if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor)) 168 ++count; 169 170 /* Random packet drop 0 => none, ~0 => all */ 171 if (q->loss && q->loss >= get_crandom(&q->loss_cor)) 172 --count; 173 174 if (count == 0) { 175 sch->qstats.drops++; 176 kfree_skb(skb); 177 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 178 } 179 180 skb_orphan(skb); 181 182 /* 183 * If we need to duplicate packet, then re-insert at top of the 184 * qdisc tree, since parent queuer expects that only one 185 * skb will be queued. 186 */ 187 if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) { 188 struct Qdisc *rootq = qdisc_root(sch); 189 u32 dupsave = q->duplicate; /* prevent duplicating a dup... */ 190 q->duplicate = 0; 191 192 qdisc_enqueue_root(skb2, rootq); 193 q->duplicate = dupsave; 194 } 195 196 /* 197 * Randomized packet corruption. 198 * Make copy if needed since we are modifying 199 * If packet is going to be hardware checksummed, then 200 * do it now in software before we mangle it. 201 */ 202 if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) { 203 if (!(skb = skb_unshare(skb, GFP_ATOMIC)) || 204 (skb->ip_summed == CHECKSUM_PARTIAL && 205 skb_checksum_help(skb))) { 206 sch->qstats.drops++; 207 return NET_XMIT_DROP; 208 } 209 210 skb->data[net_random() % skb_headlen(skb)] ^= 1<<(net_random() % 8); 211 } 212 213 cb = netem_skb_cb(skb); 214 if (q->gap == 0 || /* not doing reordering */ 215 q->counter < q->gap || /* inside last reordering gap */ 216 q->reorder < get_crandom(&q->reorder_cor)) { 217 psched_time_t now; 218 psched_tdiff_t delay; 219 220 delay = tabledist(q->latency, q->jitter, 221 &q->delay_cor, q->delay_dist); 222 223 now = psched_get_time(); 224 cb->time_to_send = now + delay; 225 ++q->counter; 226 ret = qdisc_enqueue(skb, q->qdisc); 227 } else { 228 /* 229 * Do re-ordering by putting one out of N packets at the front 230 * of the queue. 231 */ 232 cb->time_to_send = psched_get_time(); 233 q->counter = 0; 234 235 __skb_queue_head(&q->qdisc->q, skb); 236 q->qdisc->qstats.backlog += qdisc_pkt_len(skb); 237 q->qdisc->qstats.requeues++; 238 ret = NET_XMIT_SUCCESS; 239 } 240 241 if (likely(ret == NET_XMIT_SUCCESS)) { 242 sch->q.qlen++; 243 sch->bstats.bytes += qdisc_pkt_len(skb); 244 sch->bstats.packets++; 245 } else if (net_xmit_drop_count(ret)) { 246 sch->qstats.drops++; 247 } 248 249 pr_debug("netem: enqueue ret %d\n", ret); 250 return ret; 251} 252 253static unsigned int netem_drop(struct Qdisc* sch) 254{ 255 struct netem_sched_data *q = qdisc_priv(sch); 256 unsigned int len = 0; 257 258 if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) { 259 sch->q.qlen--; 260 sch->qstats.drops++; 261 } 262 return len; 263} 264 265static struct sk_buff *netem_dequeue(struct Qdisc *sch) 266{ 267 struct netem_sched_data *q = qdisc_priv(sch); 268 struct sk_buff *skb; 269 270 if (sch->flags & TCQ_F_THROTTLED) 271 return NULL; 272 273 skb = q->qdisc->ops->peek(q->qdisc); 274 if (skb) { 275 const struct netem_skb_cb *cb = netem_skb_cb(skb); 276 psched_time_t now = psched_get_time(); 277 278 /* if more time remaining? */ 279 if (cb->time_to_send <= now) { 280 skb = qdisc_dequeue_peeked(q->qdisc); 281 if (unlikely(!skb)) 282 return NULL; 283 284#ifdef CONFIG_NET_CLS_ACT 285 /* 286 * If it's at ingress let's pretend the delay is 287 * from the network (tstamp will be updated). 288 */ 289 if (G_TC_FROM(skb->tc_verd) & AT_INGRESS) 290 skb->tstamp.tv64 = 0; 291#endif 292 pr_debug("netem_dequeue: return skb=%p\n", skb); 293 sch->q.qlen--; 294 return skb; 295 } 296 297 qdisc_watchdog_schedule(&q->watchdog, cb->time_to_send); 298 } 299 300 return NULL; 301} 302 303static void netem_reset(struct Qdisc *sch) 304{ 305 struct netem_sched_data *q = qdisc_priv(sch); 306 307 qdisc_reset(q->qdisc); 308 sch->q.qlen = 0; 309 qdisc_watchdog_cancel(&q->watchdog); 310} 311 312/* 313 * Distribution data is a variable size payload containing 314 * signed 16 bit values. 315 */ 316static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr) 317{ 318 struct netem_sched_data *q = qdisc_priv(sch); 319 unsigned long n = nla_len(attr)/sizeof(__s16); 320 const __s16 *data = nla_data(attr); 321 spinlock_t *root_lock; 322 struct disttable *d; 323 int i; 324 325 if (n > 65536) 326 return -EINVAL; 327 328 d = kmalloc(sizeof(*d) + n*sizeof(d->table[0]), GFP_KERNEL); 329 if (!d) 330 return -ENOMEM; 331 332 d->size = n; 333 for (i = 0; i < n; i++) 334 d->table[i] = data[i]; 335 336 root_lock = qdisc_root_sleeping_lock(sch); 337 338 spin_lock_bh(root_lock); 339 kfree(q->delay_dist); 340 q->delay_dist = d; 341 spin_unlock_bh(root_lock); 342 return 0; 343} 344 345static void get_correlation(struct Qdisc *sch, const struct nlattr *attr) 346{ 347 struct netem_sched_data *q = qdisc_priv(sch); 348 const struct tc_netem_corr *c = nla_data(attr); 349 350 init_crandom(&q->delay_cor, c->delay_corr); 351 init_crandom(&q->loss_cor, c->loss_corr); 352 init_crandom(&q->dup_cor, c->dup_corr); 353} 354 355static void get_reorder(struct Qdisc *sch, const struct nlattr *attr) 356{ 357 struct netem_sched_data *q = qdisc_priv(sch); 358 const struct tc_netem_reorder *r = nla_data(attr); 359 360 q->reorder = r->probability; 361 init_crandom(&q->reorder_cor, r->correlation); 362} 363 364static void get_corrupt(struct Qdisc *sch, const struct nlattr *attr) 365{ 366 struct netem_sched_data *q = qdisc_priv(sch); 367 const struct tc_netem_corrupt *r = nla_data(attr); 368 369 q->corrupt = r->probability; 370 init_crandom(&q->corrupt_cor, r->correlation); 371} 372 373static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = { 374 [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) }, 375 [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) }, 376 [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) }, 377}; 378 379static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla, 380 const struct nla_policy *policy, int len) 381{ 382 int nested_len = nla_len(nla) - NLA_ALIGN(len); 383 384 if (nested_len < 0) 385 return -EINVAL; 386 if (nested_len >= nla_attr_size(0)) 387 return nla_parse(tb, maxtype, nla_data(nla) + NLA_ALIGN(len), 388 nested_len, policy); 389 memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1)); 390 return 0; 391} 392 393/* Parse netlink message to set options */ 394static int netem_change(struct Qdisc *sch, struct nlattr *opt) 395{ 396 struct netem_sched_data *q = qdisc_priv(sch); 397 struct nlattr *tb[TCA_NETEM_MAX + 1]; 398 struct tc_netem_qopt *qopt; 399 int ret; 400 401 if (opt == NULL) 402 return -EINVAL; 403 404 qopt = nla_data(opt); 405 ret = parse_attr(tb, TCA_NETEM_MAX, opt, netem_policy, sizeof(*qopt)); 406 if (ret < 0) 407 return ret; 408 409 ret = fifo_set_limit(q->qdisc, qopt->limit); 410 if (ret) { 411 pr_debug("netem: can't set fifo limit\n"); 412 return ret; 413 } 414 415 q->latency = qopt->latency; 416 q->jitter = qopt->jitter; 417 q->limit = qopt->limit; 418 q->gap = qopt->gap; 419 q->counter = 0; 420 q->loss = qopt->loss; 421 q->duplicate = qopt->duplicate; 422 423 /* for compatibility with earlier versions. 424 * if gap is set, need to assume 100% probability 425 */ 426 if (q->gap) 427 q->reorder = ~0; 428 429 if (tb[TCA_NETEM_CORR]) 430 get_correlation(sch, tb[TCA_NETEM_CORR]); 431 432 if (tb[TCA_NETEM_DELAY_DIST]) { 433 ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]); 434 if (ret) 435 return ret; 436 } 437 438 if (tb[TCA_NETEM_REORDER]) 439 get_reorder(sch, tb[TCA_NETEM_REORDER]); 440 441 if (tb[TCA_NETEM_CORRUPT]) 442 get_corrupt(sch, tb[TCA_NETEM_CORRUPT]); 443 444 return 0; 445} 446 447/* 448 * Special case version of FIFO queue for use by netem. 449 * It queues in order based on timestamps in skb's 450 */ 451struct fifo_sched_data { 452 u32 limit; 453 psched_time_t oldest; 454}; 455 456static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch) 457{ 458 struct fifo_sched_data *q = qdisc_priv(sch); 459 struct sk_buff_head *list = &sch->q; 460 psched_time_t tnext = netem_skb_cb(nskb)->time_to_send; 461 struct sk_buff *skb; 462 463 if (likely(skb_queue_len(list) < q->limit)) { 464 /* Optimize for add at tail */ 465 if (likely(skb_queue_empty(list) || tnext >= q->oldest)) { 466 q->oldest = tnext; 467 return qdisc_enqueue_tail(nskb, sch); 468 } 469 470 skb_queue_reverse_walk(list, skb) { 471 const struct netem_skb_cb *cb = netem_skb_cb(skb); 472 473 if (tnext >= cb->time_to_send) 474 break; 475 } 476 477 __skb_queue_after(list, skb, nskb); 478 479 sch->qstats.backlog += qdisc_pkt_len(nskb); 480 sch->bstats.bytes += qdisc_pkt_len(nskb); 481 sch->bstats.packets++; 482 483 return NET_XMIT_SUCCESS; 484 } 485 486 return qdisc_reshape_fail(nskb, sch); 487} 488 489static int tfifo_init(struct Qdisc *sch, struct nlattr *opt) 490{ 491 struct fifo_sched_data *q = qdisc_priv(sch); 492 493 if (opt) { 494 struct tc_fifo_qopt *ctl = nla_data(opt); 495 if (nla_len(opt) < sizeof(*ctl)) 496 return -EINVAL; 497 498 q->limit = ctl->limit; 499 } else 500 q->limit = max_t(u32, qdisc_dev(sch)->tx_queue_len, 1); 501 502 q->oldest = PSCHED_PASTPERFECT; 503 return 0; 504} 505 506static int tfifo_dump(struct Qdisc *sch, struct sk_buff *skb) 507{ 508 struct fifo_sched_data *q = qdisc_priv(sch); 509 struct tc_fifo_qopt opt = { .limit = q->limit }; 510 511 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); 512 return skb->len; 513 514nla_put_failure: 515 return -1; 516} 517 518static struct Qdisc_ops tfifo_qdisc_ops __read_mostly = { 519 .id = "tfifo", 520 .priv_size = sizeof(struct fifo_sched_data), 521 .enqueue = tfifo_enqueue, 522 .dequeue = qdisc_dequeue_head, 523 .peek = qdisc_peek_head, 524 .drop = qdisc_queue_drop, 525 .init = tfifo_init, 526 .reset = qdisc_reset_queue, 527 .change = tfifo_init, 528 .dump = tfifo_dump, 529}; 530 531static int netem_init(struct Qdisc *sch, struct nlattr *opt) 532{ 533 struct netem_sched_data *q = qdisc_priv(sch); 534 int ret; 535 536 if (!opt) 537 return -EINVAL; 538 539 qdisc_watchdog_init(&q->watchdog, sch); 540 541 q->qdisc = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue, 542 &tfifo_qdisc_ops, 543 TC_H_MAKE(sch->handle, 1)); 544 if (!q->qdisc) { 545 pr_debug("netem: qdisc create failed\n"); 546 return -ENOMEM; 547 } 548 549 ret = netem_change(sch, opt); 550 if (ret) { 551 pr_debug("netem: change failed\n"); 552 qdisc_destroy(q->qdisc); 553 } 554 return ret; 555} 556 557static void netem_destroy(struct Qdisc *sch) 558{ 559 struct netem_sched_data *q = qdisc_priv(sch); 560 561 qdisc_watchdog_cancel(&q->watchdog); 562 qdisc_destroy(q->qdisc); 563 kfree(q->delay_dist); 564} 565 566static int netem_dump(struct Qdisc *sch, struct sk_buff *skb) 567{ 568 const struct netem_sched_data *q = qdisc_priv(sch); 569 unsigned char *b = skb_tail_pointer(skb); 570 struct nlattr *nla = (struct nlattr *) b; 571 struct tc_netem_qopt qopt; 572 struct tc_netem_corr cor; 573 struct tc_netem_reorder reorder; 574 struct tc_netem_corrupt corrupt; 575 576 qopt.latency = q->latency; 577 qopt.jitter = q->jitter; 578 qopt.limit = q->limit; 579 qopt.loss = q->loss; 580 qopt.gap = q->gap; 581 qopt.duplicate = q->duplicate; 582 NLA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt); 583 584 cor.delay_corr = q->delay_cor.rho; 585 cor.loss_corr = q->loss_cor.rho; 586 cor.dup_corr = q->dup_cor.rho; 587 NLA_PUT(skb, TCA_NETEM_CORR, sizeof(cor), &cor); 588 589 reorder.probability = q->reorder; 590 reorder.correlation = q->reorder_cor.rho; 591 NLA_PUT(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder); 592 593 corrupt.probability = q->corrupt; 594 corrupt.correlation = q->corrupt_cor.rho; 595 NLA_PUT(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt); 596 597 nla->nla_len = skb_tail_pointer(skb) - b; 598 599 return skb->len; 600 601nla_put_failure: 602 nlmsg_trim(skb, b); 603 return -1; 604} 605 606static struct Qdisc_ops netem_qdisc_ops __read_mostly = { 607 .id = "netem", 608 .priv_size = sizeof(struct netem_sched_data), 609 .enqueue = netem_enqueue, 610 .dequeue = netem_dequeue, 611 .peek = qdisc_peek_dequeued, 612 .drop = netem_drop, 613 .init = netem_init, 614 .reset = netem_reset, 615 .destroy = netem_destroy, 616 .change = netem_change, 617 .dump = netem_dump, 618 .owner = THIS_MODULE, 619}; 620 621 622static int __init netem_module_init(void) 623{ 624 pr_info("netem: version " VERSION "\n"); 625 return register_qdisc(&netem_qdisc_ops); 626} 627static void __exit netem_module_exit(void) 628{ 629 unregister_qdisc(&netem_qdisc_ops); 630} 631module_init(netem_module_init) 632module_exit(netem_module_exit) 633MODULE_LICENSE("GPL"); 634