1 /*- 2 * Copyright (c) 2010 Luigi Rizzo, Riccardo Panicucci, Universita` di Pisa 3 * All rights reserved 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 /* 28 * Dummynet portions related to packet handling. 29 */ 30 #include <sys/cdefs.h> 31 __FBSDID("$FreeBSD$"); 32 33 #include "opt_inet6.h" 34 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/malloc.h> 38 #include <sys/mbuf.h> 39 #include <sys/kernel.h> 40 #include <sys/lock.h> 41 #include <sys/module.h> 42 #include <sys/mutex.h> 43 #include <sys/priv.h> 44 #include <sys/proc.h> 45 #include <sys/rwlock.h> 46 #include <sys/socket.h> 47 #include <sys/time.h> 48 #include <sys/sysctl.h> 49 50 #include <net/if.h> /* IFNAMSIZ, struct ifaddr, ifq head, lock.h mutex.h */ 51 #include <net/netisr.h> 52 #include <net/vnet.h> 53 54 #include <netinet/in.h> 55 #include <netinet/ip.h> /* ip_len, ip_off */ 56 #include <netinet/ip_var.h> /* ip_output(), IP_FORWARDING */ 57 #include <netinet/ip_fw.h> 58 #include <netinet/ip_dummynet.h> 59 #include <netinet/if_ether.h> /* various ether_* routines */ 60 #include <netinet/ip6.h> /* for ip6_input, ip6_output prototypes */ 61 #include <netinet6/ip6_var.h> 62 63 #include <netpfil/ipfw/ip_fw_private.h> 64 #include <netpfil/ipfw/dn_heap.h> 65 #include <netpfil/ipfw/ip_dn_private.h> 66 #include <netpfil/ipfw/dn_sched.h> 67 68 /* 69 * We keep a private variable for the simulation time, but we could 70 * probably use an existing one ("softticks" in sys/kern/kern_timeout.c) 71 * instead of dn_cfg.curr_time 72 */ 73 74 struct dn_parms dn_cfg; 75 //VNET_DEFINE(struct dn_parms, _base_dn_cfg); 76 77 static long tick_last; /* Last tick duration (usec). */ 78 static long tick_delta; /* Last vs standard tick diff (usec). */ 79 static long tick_delta_sum; /* Accumulated tick difference (usec).*/ 80 static long tick_adjustment; /* Tick adjustments done. */ 81 static long tick_lost; /* Lost(coalesced) ticks number. */ 82 /* Adjusted vs non-adjusted curr_time difference (ticks). */ 83 static long tick_diff; 84 85 static unsigned long io_pkt; 86 static unsigned long io_pkt_fast; 87 static unsigned long io_pkt_drop; 88 89 /* 90 * We use a heap to store entities for which we have pending timer events. 91 * The heap is checked at every tick and all entities with expired events 92 * are extracted. 93 */ 94 95 MALLOC_DEFINE(M_DUMMYNET, "dummynet", "dummynet heap"); 96 97 extern void (*bridge_dn_p)(struct mbuf *, struct ifnet *); 98 99 #ifdef SYSCTL_NODE 100 101 /* 102 * Because of the way the SYSBEGIN/SYSEND macros work on other 103 * platforms, there should not be functions between them. 104 * So keep the handlers outside the block. 105 */ 106 static int 107 sysctl_hash_size(SYSCTL_HANDLER_ARGS) 108 { 109 int error, value; 110 111 value = dn_cfg.hash_size; 112 error = sysctl_handle_int(oidp, &value, 0, req); 113 if (error != 0 || req->newptr == NULL) 114 return (error); 115 if (value < 16 || value > 65536) 116 return (EINVAL); 117 dn_cfg.hash_size = value; 118 return (0); 119 } 120 121 static int 122 sysctl_limits(SYSCTL_HANDLER_ARGS) 123 { 124 int error; 125 long value; 126 127 if (arg2 != 0) 128 value = dn_cfg.slot_limit; 129 else 130 value = dn_cfg.byte_limit; 131 error = sysctl_handle_long(oidp, &value, 0, req); 132 133 if (error != 0 || req->newptr == NULL) 134 return (error); 135 if (arg2 != 0) { 136 if (value < 1) 137 return (EINVAL); 138 dn_cfg.slot_limit = value; 139 } else { 140 if (value < 1500) 141 return (EINVAL); 142 dn_cfg.byte_limit = value; 143 } 144 return (0); 145 } 146 147 SYSBEGIN(f4) 148 149 SYSCTL_DECL(_net_inet); 150 SYSCTL_DECL(_net_inet_ip); 151 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, dummynet, CTLFLAG_RW, 0, "Dummynet"); 152 153 /* wrapper to pass dn_cfg fields to SYSCTL_* */ 154 //#define DC(x) (&(VNET_NAME(_base_dn_cfg).x)) 155 #define DC(x) (&(dn_cfg.x)) 156 /* parameters */ 157 158 159 SYSCTL_PROC(_net_inet_ip_dummynet, OID_AUTO, hash_size, 160 CTLTYPE_INT | CTLFLAG_RW, 0, 0, sysctl_hash_size, 161 "I", "Default hash table size"); 162 163 164 SYSCTL_PROC(_net_inet_ip_dummynet, OID_AUTO, pipe_slot_limit, 165 CTLTYPE_LONG | CTLFLAG_RW, 0, 1, sysctl_limits, 166 "L", "Upper limit in slots for pipe queue."); 167 SYSCTL_PROC(_net_inet_ip_dummynet, OID_AUTO, pipe_byte_limit, 168 CTLTYPE_LONG | CTLFLAG_RW, 0, 0, sysctl_limits, 169 "L", "Upper limit in bytes for pipe queue."); 170 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, io_fast, 171 CTLFLAG_RW, DC(io_fast), 0, "Enable fast dummynet io."); 172 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, debug, 173 CTLFLAG_RW, DC(debug), 0, "Dummynet debug level"); 174 175 /* RED parameters */ 176 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_lookup_depth, 177 CTLFLAG_RD, DC(red_lookup_depth), 0, "Depth of RED lookup table"); 178 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_avg_pkt_size, 179 CTLFLAG_RD, DC(red_avg_pkt_size), 0, "RED Medium packet size"); 180 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_max_pkt_size, 181 CTLFLAG_RD, DC(red_max_pkt_size), 0, "RED Max packet size"); 182 183 /* time adjustment */ 184 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_delta, 185 CTLFLAG_RD, &tick_delta, 0, "Last vs standard tick difference (usec)."); 186 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_delta_sum, 187 CTLFLAG_RD, &tick_delta_sum, 0, "Accumulated tick difference (usec)."); 188 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_adjustment, 189 CTLFLAG_RD, &tick_adjustment, 0, "Tick adjustments done."); 190 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_diff, 191 CTLFLAG_RD, &tick_diff, 0, 192 "Adjusted vs non-adjusted curr_time difference (ticks)."); 193 SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_lost, 194 CTLFLAG_RD, &tick_lost, 0, 195 "Number of ticks coalesced by dummynet taskqueue."); 196 197 /* Drain parameters */ 198 SYSCTL_UINT(_net_inet_ip_dummynet, OID_AUTO, expire, 199 CTLFLAG_RW, DC(expire), 0, "Expire empty queues/pipes"); 200 SYSCTL_UINT(_net_inet_ip_dummynet, OID_AUTO, expire_cycle, 201 CTLFLAG_RD, DC(expire_cycle), 0, "Expire cycle for queues/pipes"); 202 203 /* statistics */ 204 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, schk_count, 205 CTLFLAG_RD, DC(schk_count), 0, "Number of schedulers"); 206 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, si_count, 207 CTLFLAG_RD, DC(si_count), 0, "Number of scheduler instances"); 208 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, fsk_count, 209 CTLFLAG_RD, DC(fsk_count), 0, "Number of flowsets"); 210 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, queue_count, 211 CTLFLAG_RD, DC(queue_count), 0, "Number of queues"); 212 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt, 213 CTLFLAG_RD, &io_pkt, 0, 214 "Number of packets passed to dummynet."); 215 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt_fast, 216 CTLFLAG_RD, &io_pkt_fast, 0, 217 "Number of packets bypassed dummynet scheduler."); 218 SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt_drop, 219 CTLFLAG_RD, &io_pkt_drop, 0, 220 "Number of packets dropped by dummynet."); 221 #undef DC 222 SYSEND 223 224 #endif 225 226 static void dummynet_send(struct mbuf *); 227 228 /* 229 * Packets processed by dummynet have an mbuf tag associated with 230 * them that carries their dummynet state. 231 * Outside dummynet, only the 'rule' field is relevant, and it must 232 * be at the beginning of the structure. 233 */ 234 struct dn_pkt_tag { 235 struct ipfw_rule_ref rule; /* matching rule */ 236 237 /* second part, dummynet specific */ 238 int dn_dir; /* action when packet comes out.*/ 239 /* see ip_fw_private.h */ 240 uint64_t output_time; /* when the pkt is due for delivery*/ 241 struct ifnet *ifp; /* interface, for ip_output */ 242 struct _ip6dn_args ip6opt; /* XXX ipv6 options */ 243 }; 244 245 /* 246 * Return the mbuf tag holding the dummynet state (it should 247 * be the first one on the list). 248 */ 249 static struct dn_pkt_tag * 250 dn_tag_get(struct mbuf *m) 251 { 252 struct m_tag *mtag = m_tag_first(m); 253 KASSERT(mtag != NULL && 254 mtag->m_tag_cookie == MTAG_ABI_COMPAT && 255 mtag->m_tag_id == PACKET_TAG_DUMMYNET, 256 ("packet on dummynet queue w/o dummynet tag!")); 257 return (struct dn_pkt_tag *)(mtag+1); 258 } 259 260 static inline void 261 mq_append(struct mq *q, struct mbuf *m) 262 { 263 if (q->head == NULL) 264 q->head = m; 265 else 266 q->tail->m_nextpkt = m; 267 q->tail = m; 268 m->m_nextpkt = NULL; 269 } 270 271 /* 272 * Dispose a list of packet. Use a functions so if we need to do 273 * more work, this is a central point to do it. 274 */ 275 void dn_free_pkts(struct mbuf *mnext) 276 { 277 struct mbuf *m; 278 279 while ((m = mnext) != NULL) { 280 mnext = m->m_nextpkt; 281 FREE_PKT(m); 282 } 283 } 284 285 static int 286 red_drops (struct dn_queue *q, int len) 287 { 288 /* 289 * RED algorithm 290 * 291 * RED calculates the average queue size (avg) using a low-pass filter 292 * with an exponential weighted (w_q) moving average: 293 * avg <- (1-w_q) * avg + w_q * q_size 294 * where q_size is the queue length (measured in bytes or * packets). 295 * 296 * If q_size == 0, we compute the idle time for the link, and set 297 * avg = (1 - w_q)^(idle/s) 298 * where s is the time needed for transmitting a medium-sized packet. 299 * 300 * Now, if avg < min_th the packet is enqueued. 301 * If avg > max_th the packet is dropped. Otherwise, the packet is 302 * dropped with probability P function of avg. 303 */ 304 305 struct dn_fsk *fs = q->fs; 306 int64_t p_b = 0; 307 308 /* Queue in bytes or packets? */ 309 uint32_t q_size = (fs->fs.flags & DN_QSIZE_BYTES) ? 310 q->ni.len_bytes : q->ni.length; 311 312 /* Average queue size estimation. */ 313 if (q_size != 0) { 314 /* Queue is not empty, avg <- avg + (q_size - avg) * w_q */ 315 int diff = SCALE(q_size) - q->avg; 316 int64_t v = SCALE_MUL((int64_t)diff, (int64_t)fs->w_q); 317 318 q->avg += (int)v; 319 } else { 320 /* 321 * Queue is empty, find for how long the queue has been 322 * empty and use a lookup table for computing 323 * (1 - * w_q)^(idle_time/s) where s is the time to send a 324 * (small) packet. 325 * XXX check wraps... 326 */ 327 if (q->avg) { 328 u_int t = div64((dn_cfg.curr_time - q->q_time), fs->lookup_step); 329 330 q->avg = (t < fs->lookup_depth) ? 331 SCALE_MUL(q->avg, fs->w_q_lookup[t]) : 0; 332 } 333 } 334 335 /* Should i drop? */ 336 if (q->avg < fs->min_th) { 337 q->count = -1; 338 return (0); /* accept packet */ 339 } 340 if (q->avg >= fs->max_th) { /* average queue >= max threshold */ 341 if (fs->fs.flags & DN_IS_GENTLE_RED) { 342 /* 343 * According to Gentle-RED, if avg is greater than 344 * max_th the packet is dropped with a probability 345 * p_b = c_3 * avg - c_4 346 * where c_3 = (1 - max_p) / max_th 347 * c_4 = 1 - 2 * max_p 348 */ 349 p_b = SCALE_MUL((int64_t)fs->c_3, (int64_t)q->avg) - 350 fs->c_4; 351 } else { 352 q->count = -1; 353 return (1); 354 } 355 } else if (q->avg > fs->min_th) { 356 /* 357 * We compute p_b using the linear dropping function 358 * p_b = c_1 * avg - c_2 359 * where c_1 = max_p / (max_th - min_th) 360 * c_2 = max_p * min_th / (max_th - min_th) 361 */ 362 p_b = SCALE_MUL((int64_t)fs->c_1, (int64_t)q->avg) - fs->c_2; 363 } 364 365 if (fs->fs.flags & DN_QSIZE_BYTES) 366 p_b = div64((p_b * len) , fs->max_pkt_size); 367 if (++q->count == 0) 368 q->random = random() & 0xffff; 369 else { 370 /* 371 * q->count counts packets arrived since last drop, so a greater 372 * value of q->count means a greater packet drop probability. 373 */ 374 if (SCALE_MUL(p_b, SCALE((int64_t)q->count)) > q->random) { 375 q->count = 0; 376 /* After a drop we calculate a new random value. */ 377 q->random = random() & 0xffff; 378 return (1); /* drop */ 379 } 380 } 381 /* End of RED algorithm. */ 382 383 return (0); /* accept */ 384 385 } 386 387 /* 388 * Enqueue a packet in q, subject to space and queue management policy 389 * (whose parameters are in q->fs). 390 * Update stats for the queue and the scheduler. 391 * Return 0 on success, 1 on drop. The packet is consumed anyways. 392 */ 393 int 394 dn_enqueue(struct dn_queue *q, struct mbuf* m, int drop) 395 { 396 struct dn_fs *f; 397 struct dn_flow *ni; /* stats for scheduler instance */ 398 uint64_t len; 399 400 if (q->fs == NULL || q->_si == NULL) { 401 printf("%s fs %p si %p, dropping\n", 402 __FUNCTION__, q->fs, q->_si); 403 FREE_PKT(m); 404 return 1; 405 } 406 f = &(q->fs->fs); 407 ni = &q->_si->ni; 408 len = m->m_pkthdr.len; 409 /* Update statistics, then check reasons to drop pkt. */ 410 q->ni.tot_bytes += len; 411 q->ni.tot_pkts++; 412 ni->tot_bytes += len; 413 ni->tot_pkts++; 414 if (drop) 415 goto drop; 416 if (f->plr && random() < f->plr) 417 goto drop; 418 if (f->flags & DN_IS_RED && red_drops(q, m->m_pkthdr.len)) 419 goto drop; 420 if (f->flags & DN_QSIZE_BYTES) { 421 if (q->ni.len_bytes > f->qsize) 422 goto drop; 423 } else if (q->ni.length >= f->qsize) { 424 goto drop; 425 } 426 mq_append(&q->mq, m); 427 q->ni.length++; 428 q->ni.len_bytes += len; 429 ni->length++; 430 ni->len_bytes += len; 431 return 0; 432 433 drop: 434 io_pkt_drop++; 435 q->ni.drops++; 436 ni->drops++; 437 FREE_PKT(m); 438 return 1; 439 } 440 441 /* 442 * Fetch packets from the delay line which are due now. If there are 443 * leftover packets, reinsert the delay line in the heap. 444 * Runs under scheduler lock. 445 */ 446 static void 447 transmit_event(struct mq *q, struct delay_line *dline, uint64_t now) 448 { 449 struct mbuf *m; 450 struct dn_pkt_tag *pkt = NULL; 451 452 dline->oid.subtype = 0; /* not in heap */ 453 while ((m = dline->mq.head) != NULL) { 454 pkt = dn_tag_get(m); 455 if (!DN_KEY_LEQ(pkt->output_time, now)) 456 break; 457 dline->mq.head = m->m_nextpkt; 458 mq_append(q, m); 459 } 460 if (m != NULL) { 461 dline->oid.subtype = 1; /* in heap */ 462 heap_insert(&dn_cfg.evheap, pkt->output_time, dline); 463 } 464 } 465 466 /* 467 * Convert the additional MAC overheads/delays into an equivalent 468 * number of bits for the given data rate. The samples are 469 * in milliseconds so we need to divide by 1000. 470 */ 471 static uint64_t 472 extra_bits(struct mbuf *m, struct dn_schk *s) 473 { 474 int index; 475 uint64_t bits; 476 struct dn_profile *pf = s->profile; 477 478 if (!pf || pf->samples_no == 0) 479 return 0; 480 index = random() % pf->samples_no; 481 bits = div64((uint64_t)pf->samples[index] * s->link.bandwidth, 1000); 482 if (index >= pf->loss_level) { 483 struct dn_pkt_tag *dt = dn_tag_get(m); 484 if (dt) 485 dt->dn_dir = DIR_DROP; 486 } 487 return bits; 488 } 489 490 /* 491 * Send traffic from a scheduler instance due by 'now'. 492 * Return a pointer to the head of the queue. 493 */ 494 static struct mbuf * 495 serve_sched(struct mq *q, struct dn_sch_inst *si, uint64_t now) 496 { 497 struct mq def_q; 498 struct dn_schk *s = si->sched; 499 struct mbuf *m = NULL; 500 int delay_line_idle = (si->dline.mq.head == NULL); 501 int done, bw; 502 503 if (q == NULL) { 504 q = &def_q; 505 q->head = NULL; 506 } 507 508 bw = s->link.bandwidth; 509 si->kflags &= ~DN_ACTIVE; 510 511 if (bw > 0) 512 si->credit += (now - si->sched_time) * bw; 513 else 514 si->credit = 0; 515 si->sched_time = now; 516 done = 0; 517 while (si->credit >= 0 && (m = s->fp->dequeue(si)) != NULL) { 518 uint64_t len_scaled; 519 520 done++; 521 len_scaled = (bw == 0) ? 0 : hz * 522 (m->m_pkthdr.len * 8 + extra_bits(m, s)); 523 si->credit -= len_scaled; 524 /* Move packet in the delay line */ 525 dn_tag_get(m)->output_time = dn_cfg.curr_time + s->link.delay ; 526 mq_append(&si->dline.mq, m); 527 } 528 529 /* 530 * If credit >= 0 the instance is idle, mark time. 531 * Otherwise put back in the heap, and adjust the output 532 * time of the last inserted packet, m, which was too early. 533 */ 534 if (si->credit >= 0) { 535 si->idle_time = now; 536 } else { 537 uint64_t t; 538 KASSERT (bw > 0, ("bw=0 and credit<0 ?")); 539 t = div64(bw - 1 - si->credit, bw); 540 if (m) 541 dn_tag_get(m)->output_time += t; 542 si->kflags |= DN_ACTIVE; 543 heap_insert(&dn_cfg.evheap, now + t, si); 544 } 545 if (delay_line_idle && done) 546 transmit_event(q, &si->dline, now); 547 return q->head; 548 } 549 550 /* 551 * The timer handler for dummynet. Time is computed in ticks, but 552 * but the code is tolerant to the actual rate at which this is called. 553 * Once complete, the function reschedules itself for the next tick. 554 */ 555 void 556 dummynet_task(void *context, int pending) 557 { 558 struct timeval t; 559 struct mq q = { NULL, NULL }; /* queue to accumulate results */ 560 561 CURVNET_SET((struct vnet *)context); 562 563 DN_BH_WLOCK(); 564 565 /* Update number of lost(coalesced) ticks. */ 566 tick_lost += pending - 1; 567 568 getmicrouptime(&t); 569 /* Last tick duration (usec). */ 570 tick_last = (t.tv_sec - dn_cfg.prev_t.tv_sec) * 1000000 + 571 (t.tv_usec - dn_cfg.prev_t.tv_usec); 572 /* Last tick vs standard tick difference (usec). */ 573 tick_delta = (tick_last * hz - 1000000) / hz; 574 /* Accumulated tick difference (usec). */ 575 tick_delta_sum += tick_delta; 576 577 dn_cfg.prev_t = t; 578 579 /* 580 * Adjust curr_time if the accumulated tick difference is 581 * greater than the 'standard' tick. Since curr_time should 582 * be monotonically increasing, we do positive adjustments 583 * as required, and throttle curr_time in case of negative 584 * adjustment. 585 */ 586 dn_cfg.curr_time++; 587 if (tick_delta_sum - tick >= 0) { 588 int diff = tick_delta_sum / tick; 589 590 dn_cfg.curr_time += diff; 591 tick_diff += diff; 592 tick_delta_sum %= tick; 593 tick_adjustment++; 594 } else if (tick_delta_sum + tick <= 0) { 595 dn_cfg.curr_time--; 596 tick_diff--; 597 tick_delta_sum += tick; 598 tick_adjustment++; 599 } 600 601 /* serve pending events, accumulate in q */ 602 for (;;) { 603 struct dn_id *p; /* generic parameter to handler */ 604 605 if (dn_cfg.evheap.elements == 0 || 606 DN_KEY_LT(dn_cfg.curr_time, HEAP_TOP(&dn_cfg.evheap)->key)) 607 break; 608 p = HEAP_TOP(&dn_cfg.evheap)->object; 609 heap_extract(&dn_cfg.evheap, NULL); 610 611 if (p->type == DN_SCH_I) { 612 serve_sched(&q, (struct dn_sch_inst *)p, dn_cfg.curr_time); 613 } else { /* extracted a delay line */ 614 transmit_event(&q, (struct delay_line *)p, dn_cfg.curr_time); 615 } 616 } 617 if (dn_cfg.expire && ++dn_cfg.expire_cycle >= dn_cfg.expire) { 618 dn_cfg.expire_cycle = 0; 619 dn_drain_scheduler(); 620 dn_drain_queue(); 621 } 622 623 DN_BH_WUNLOCK(); 624 dn_reschedule(); 625 if (q.head != NULL) 626 dummynet_send(q.head); 627 CURVNET_RESTORE(); 628 } 629 630 /* 631 * forward a chain of packets to the proper destination. 632 * This runs outside the dummynet lock. 633 */ 634 static void 635 dummynet_send(struct mbuf *m) 636 { 637 struct mbuf *n; 638 639 for (; m != NULL; m = n) { 640 struct ifnet *ifp = NULL; /* gcc 3.4.6 complains */ 641 struct m_tag *tag; 642 int dst; 643 644 n = m->m_nextpkt; 645 m->m_nextpkt = NULL; 646 tag = m_tag_first(m); 647 if (tag == NULL) { /* should not happen */ 648 dst = DIR_DROP; 649 } else { 650 struct dn_pkt_tag *pkt = dn_tag_get(m); 651 /* extract the dummynet info, rename the tag 652 * to carry reinject info. 653 */ 654 dst = pkt->dn_dir; 655 ifp = pkt->ifp; 656 tag->m_tag_cookie = MTAG_IPFW_RULE; 657 tag->m_tag_id = 0; 658 } 659 660 switch (dst) { 661 case DIR_OUT: 662 ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL); 663 break ; 664 665 case DIR_IN : 666 netisr_dispatch(NETISR_IP, m); 667 break; 668 669 #ifdef INET6 670 case DIR_IN | PROTO_IPV6: 671 netisr_dispatch(NETISR_IPV6, m); 672 break; 673 674 case DIR_OUT | PROTO_IPV6: 675 ip6_output(m, NULL, NULL, IPV6_FORWARDING, NULL, NULL, NULL); 676 break; 677 #endif 678 679 case DIR_FWD | PROTO_IFB: /* DN_TO_IFB_FWD: */ 680 if (bridge_dn_p != NULL) 681 ((*bridge_dn_p)(m, ifp)); 682 else 683 printf("dummynet: if_bridge not loaded\n"); 684 685 break; 686 687 case DIR_IN | PROTO_LAYER2: /* DN_TO_ETH_DEMUX: */ 688 /* 689 * The Ethernet code assumes the Ethernet header is 690 * contiguous in the first mbuf header. 691 * Insure this is true. 692 */ 693 if (m->m_len < ETHER_HDR_LEN && 694 (m = m_pullup(m, ETHER_HDR_LEN)) == NULL) { 695 printf("dummynet/ether: pullup failed, " 696 "dropping packet\n"); 697 break; 698 } 699 ether_demux(m->m_pkthdr.rcvif, m); 700 break; 701 702 case DIR_OUT | PROTO_LAYER2: /* N_TO_ETH_OUT: */ 703 ether_output_frame(ifp, m); 704 break; 705 706 case DIR_DROP: 707 /* drop the packet after some time */ 708 FREE_PKT(m); 709 break; 710 711 default: 712 printf("dummynet: bad switch %d!\n", dst); 713 FREE_PKT(m); 714 break; 715 } 716 } 717 } 718 719 static inline int 720 tag_mbuf(struct mbuf *m, int dir, struct ip_fw_args *fwa) 721 { 722 struct dn_pkt_tag *dt; 723 struct m_tag *mtag; 724 725 mtag = m_tag_get(PACKET_TAG_DUMMYNET, 726 sizeof(*dt), M_NOWAIT | M_ZERO); 727 if (mtag == NULL) 728 return 1; /* Cannot allocate packet header. */ 729 m_tag_prepend(m, mtag); /* Attach to mbuf chain. */ 730 dt = (struct dn_pkt_tag *)(mtag + 1); 731 dt->rule = fwa->rule; 732 dt->rule.info &= IPFW_ONEPASS; /* only keep this info */ 733 dt->dn_dir = dir; 734 dt->ifp = fwa->oif; 735 /* dt->output tame is updated as we move through */ 736 dt->output_time = dn_cfg.curr_time; 737 return 0; 738 } 739 740 741 /* 742 * dummynet hook for packets. 743 * We use the argument to locate the flowset fs and the sched_set sch 744 * associated to it. The we apply flow_mask and sched_mask to 745 * determine the queue and scheduler instances. 746 * 747 * dir where shall we send the packet after dummynet. 748 * *m0 the mbuf with the packet 749 * ifp the 'ifp' parameter from the caller. 750 * NULL in ip_input, destination interface in ip_output, 751 */ 752 int 753 dummynet_io(struct mbuf **m0, int dir, struct ip_fw_args *fwa) 754 { 755 struct mbuf *m = *m0; 756 struct dn_fsk *fs = NULL; 757 struct dn_sch_inst *si; 758 struct dn_queue *q = NULL; /* default */ 759 760 int fs_id = (fwa->rule.info & IPFW_INFO_MASK) + 761 ((fwa->rule.info & IPFW_IS_PIPE) ? 2*DN_MAX_ID : 0); 762 DN_BH_WLOCK(); 763 io_pkt++; 764 /* we could actually tag outside the lock, but who cares... */ 765 if (tag_mbuf(m, dir, fwa)) 766 goto dropit; 767 if (dn_cfg.busy) { 768 /* if the upper half is busy doing something expensive, 769 * lets queue the packet and move forward 770 */ 771 mq_append(&dn_cfg.pending, m); 772 m = *m0 = NULL; /* consumed */ 773 goto done; /* already active, nothing to do */ 774 } 775 /* XXX locate_flowset could be optimised with a direct ref. */ 776 fs = dn_ht_find(dn_cfg.fshash, fs_id, 0, NULL); 777 if (fs == NULL) 778 goto dropit; /* This queue/pipe does not exist! */ 779 if (fs->sched == NULL) /* should not happen */ 780 goto dropit; 781 /* find scheduler instance, possibly applying sched_mask */ 782 si = ipdn_si_find(fs->sched, &(fwa->f_id)); 783 if (si == NULL) 784 goto dropit; 785 /* 786 * If the scheduler supports multiple queues, find the right one 787 * (otherwise it will be ignored by enqueue). 788 */ 789 if (fs->sched->fp->flags & DN_MULTIQUEUE) { 790 q = ipdn_q_find(fs, si, &(fwa->f_id)); 791 if (q == NULL) 792 goto dropit; 793 } 794 if (fs->sched->fp->enqueue(si, q, m)) { 795 /* packet was dropped by enqueue() */ 796 m = *m0 = NULL; 797 goto dropit; 798 } 799 800 if (si->kflags & DN_ACTIVE) { 801 m = *m0 = NULL; /* consumed */ 802 goto done; /* already active, nothing to do */ 803 } 804 805 /* compute the initial allowance */ 806 if (si->idle_time < dn_cfg.curr_time) { 807 /* Do this only on the first packet on an idle pipe */ 808 struct dn_link *p = &fs->sched->link; 809 810 si->sched_time = dn_cfg.curr_time; 811 si->credit = dn_cfg.io_fast ? p->bandwidth : 0; 812 if (p->burst) { 813 uint64_t burst = (dn_cfg.curr_time - si->idle_time) * p->bandwidth; 814 if (burst > p->burst) 815 burst = p->burst; 816 si->credit += burst; 817 } 818 } 819 /* pass through scheduler and delay line */ 820 m = serve_sched(NULL, si, dn_cfg.curr_time); 821 822 /* optimization -- pass it back to ipfw for immediate send */ 823 /* XXX Don't call dummynet_send() if scheduler return the packet 824 * just enqueued. This avoid a lock order reversal. 825 * 826 */ 827 if (/*dn_cfg.io_fast &&*/ m == *m0 && (dir & PROTO_LAYER2) == 0 ) { 828 /* fast io, rename the tag * to carry reinject info. */ 829 struct m_tag *tag = m_tag_first(m); 830 831 tag->m_tag_cookie = MTAG_IPFW_RULE; 832 tag->m_tag_id = 0; 833 io_pkt_fast++; 834 if (m->m_nextpkt != NULL) { 835 printf("dummynet: fast io: pkt chain detected!\n"); 836 m->m_nextpkt = NULL; 837 } 838 m = NULL; 839 } else { 840 *m0 = NULL; 841 } 842 done: 843 DN_BH_WUNLOCK(); 844 if (m) 845 dummynet_send(m); 846 return 0; 847 848 dropit: 849 io_pkt_drop++; 850 DN_BH_WUNLOCK(); 851 if (m) 852 FREE_PKT(m); 853 *m0 = NULL; 854 return (fs && (fs->fs.flags & DN_NOERROR)) ? 0 : ENOBUFS; 855 } 856