1 /* 2 * Copyright (c) 1998-2002 Luigi Rizzo, Universita` di Pisa 3 * Portions Copyright (c) 2000 Akamba Corp. 4 * All rights reserved 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 * 27 * $FreeBSD: src/sys/netinet/ip_dummynet.c,v 1.24.2.22 2003/05/13 09:31:06 maxim Exp $ 28 * $DragonFly: src/sys/net/dummynet/ip_dummynet.c,v 1.13 2005/06/03 18:20:36 swildner Exp $ 29 */ 30 31 #if !defined(KLD_MODULE) 32 #include "opt_ipfw.h" /* for IPFW2 definition */ 33 #endif 34 35 #define DEB(x) 36 #define DDB(x) x 37 38 /* 39 * This module implements IP dummynet, a bandwidth limiter/delay emulator 40 * used in conjunction with the ipfw package. 41 * Description of the data structures used is in ip_dummynet.h 42 * Here you mainly find the following blocks of code: 43 * + variable declarations; 44 * + heap management functions; 45 * + scheduler and dummynet functions; 46 * + configuration and initialization. 47 * 48 * NOTA BENE: critical sections are protected by splimp()/splx() 49 * pairs. One would think that splnet() is enough as for most of 50 * the netinet code, but it is not so because when used with 51 * bridging, dummynet is invoked at splimp(). 52 * 53 * Most important Changes: 54 * 55 * 011004: KLDable 56 * 010124: Fixed WF2Q behaviour 57 * 010122: Fixed spl protection. 58 * 000601: WF2Q support 59 * 000106: large rewrite, use heaps to handle very many pipes. 60 * 980513: initial release 61 * 62 * include files marked with XXX are probably not needed 63 */ 64 65 #include <sys/param.h> 66 #include <sys/systm.h> 67 #include <sys/malloc.h> 68 #include <sys/mbuf.h> 69 #include <sys/kernel.h> 70 #include <sys/module.h> 71 #include <sys/proc.h> 72 #include <sys/socket.h> 73 #include <sys/socketvar.h> 74 #include <sys/time.h> 75 #include <sys/sysctl.h> 76 #include <sys/thread2.h> 77 #include <net/if.h> 78 #include <net/route.h> 79 #include <netinet/in.h> 80 #include <netinet/in_systm.h> 81 #include <netinet/in_var.h> 82 #include <netinet/ip.h> 83 #include <net/ipfw/ip_fw.h> 84 #include "ip_dummynet.h" 85 #include <netinet/ip_var.h> 86 87 #include <netinet/if_ether.h> /* for struct arpcom */ 88 #include <net/bridge/bridge.h> 89 90 /* 91 * We keep a private variable for the simulation time, but we could 92 * probably use an existing one ("softticks" in sys/kern/kern_timer.c) 93 */ 94 static dn_key curr_time = 0 ; /* current simulation time */ 95 96 static int dn_hash_size = 64 ; /* default hash size */ 97 98 /* statistics on number of queue searches and search steps */ 99 static int searches, search_steps ; 100 static int pipe_expire = 1 ; /* expire queue if empty */ 101 static int dn_max_ratio = 16 ; /* max queues/buckets ratio */ 102 103 static int red_lookup_depth = 256; /* RED - default lookup table depth */ 104 static int red_avg_pkt_size = 512; /* RED - default medium packet size */ 105 static int red_max_pkt_size = 1500; /* RED - default max packet size */ 106 107 /* 108 * Three heaps contain queues and pipes that the scheduler handles: 109 * 110 * ready_heap contains all dn_flow_queue related to fixed-rate pipes. 111 * 112 * wfq_ready_heap contains the pipes associated with WF2Q flows 113 * 114 * extract_heap contains pipes associated with delay lines. 115 * 116 */ 117 118 MALLOC_DEFINE(M_DUMMYNET, "dummynet", "dummynet heap"); 119 120 static struct dn_heap ready_heap, extract_heap, wfq_ready_heap ; 121 122 static int heap_init(struct dn_heap *h, int size) ; 123 static int heap_insert (struct dn_heap *h, dn_key key1, void *p); 124 static void heap_extract(struct dn_heap *h, void *obj); 125 126 static void transmit_event(struct dn_pipe *pipe); 127 static void ready_event(struct dn_flow_queue *q); 128 129 static struct dn_pipe *all_pipes = NULL ; /* list of all pipes */ 130 static struct dn_flow_set *all_flow_sets = NULL ;/* list of all flow_sets */ 131 132 static struct callout dn_timeout; 133 134 #ifdef SYSCTL_NODE 135 SYSCTL_NODE(_net_inet_ip, OID_AUTO, dummynet, 136 CTLFLAG_RW, 0, "Dummynet"); 137 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, hash_size, 138 CTLFLAG_RW, &dn_hash_size, 0, "Default hash table size"); 139 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, curr_time, 140 CTLFLAG_RD, &curr_time, 0, "Current tick"); 141 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, ready_heap, 142 CTLFLAG_RD, &ready_heap.size, 0, "Size of ready heap"); 143 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, extract_heap, 144 CTLFLAG_RD, &extract_heap.size, 0, "Size of extract heap"); 145 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, searches, 146 CTLFLAG_RD, &searches, 0, "Number of queue searches"); 147 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, search_steps, 148 CTLFLAG_RD, &search_steps, 0, "Number of queue search steps"); 149 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, expire, 150 CTLFLAG_RW, &pipe_expire, 0, "Expire queue if empty"); 151 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, max_chain_len, 152 CTLFLAG_RW, &dn_max_ratio, 0, 153 "Max ratio between dynamic queues and buckets"); 154 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_lookup_depth, 155 CTLFLAG_RD, &red_lookup_depth, 0, "Depth of RED lookup table"); 156 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_avg_pkt_size, 157 CTLFLAG_RD, &red_avg_pkt_size, 0, "RED Medium packet size"); 158 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_max_pkt_size, 159 CTLFLAG_RD, &red_max_pkt_size, 0, "RED Max packet size"); 160 #endif 161 162 static int config_pipe(struct dn_pipe *p); 163 static int ip_dn_ctl(struct sockopt *sopt); 164 165 static void rt_unref(struct rtentry *); 166 static void dummynet(void *); 167 static void dummynet_flush(void); 168 void dummynet_drain(void); 169 static ip_dn_io_t dummynet_io; 170 static void dn_rule_delete(void *); 171 172 int if_tx_rdy(struct ifnet *ifp); 173 174 static void 175 rt_unref(struct rtentry *rt) 176 { 177 if (rt == NULL) 178 return ; 179 if (rt->rt_refcnt <= 0) 180 printf("-- warning, refcnt now %ld, decreasing\n", rt->rt_refcnt); 181 RTFREE(rt); 182 } 183 184 /* 185 * Heap management functions. 186 * 187 * In the heap, first node is element 0. Children of i are 2i+1 and 2i+2. 188 * Some macros help finding parent/children so we can optimize them. 189 * 190 * heap_init() is called to expand the heap when needed. 191 * Increment size in blocks of 16 entries. 192 * XXX failure to allocate a new element is a pretty bad failure 193 * as we basically stall a whole queue forever!! 194 * Returns 1 on error, 0 on success 195 */ 196 #define HEAP_FATHER(x) ( ( (x) - 1 ) / 2 ) 197 #define HEAP_LEFT(x) ( 2*(x) + 1 ) 198 #define HEAP_IS_LEFT(x) ( (x) & 1 ) 199 #define HEAP_RIGHT(x) ( 2*(x) + 2 ) 200 #define HEAP_SWAP(a, b, buffer) { buffer = a ; a = b ; b = buffer ; } 201 #define HEAP_INCREMENT 15 202 203 static int 204 heap_init(struct dn_heap *h, int new_size) 205 { 206 struct dn_heap_entry *p; 207 208 if (h->size >= new_size ) { 209 printf("heap_init, Bogus call, have %d want %d\n", 210 h->size, new_size); 211 return 0 ; 212 } 213 new_size = (new_size + HEAP_INCREMENT ) & ~HEAP_INCREMENT ; 214 p = malloc(new_size * sizeof(*p), M_DUMMYNET, M_WAITOK | M_ZERO); 215 if (h->size > 0) { 216 bcopy(h->p, p, h->size * sizeof(*p) ); 217 free(h->p, M_DUMMYNET); 218 } 219 h->p = p ; 220 h->size = new_size ; 221 return 0 ; 222 } 223 224 /* 225 * Insert element in heap. Normally, p != NULL, we insert p in 226 * a new position and bubble up. If p == NULL, then the element is 227 * already in place, and key is the position where to start the 228 * bubble-up. 229 * Returns 1 on failure (cannot allocate new heap entry) 230 * 231 * If offset > 0 the position (index, int) of the element in the heap is 232 * also stored in the element itself at the given offset in bytes. 233 */ 234 #define SET_OFFSET(heap, node) \ 235 if (heap->offset > 0) \ 236 *((int *)((char *)(heap->p[node].object) + heap->offset)) = node ; 237 /* 238 * RESET_OFFSET is used for sanity checks. It sets offset to an invalid value. 239 */ 240 #define RESET_OFFSET(heap, node) \ 241 if (heap->offset > 0) \ 242 *((int *)((char *)(heap->p[node].object) + heap->offset)) = -1 ; 243 static int 244 heap_insert(struct dn_heap *h, dn_key key1, void *p) 245 { 246 int son = h->elements ; 247 248 if (p == NULL) /* data already there, set starting point */ 249 son = key1 ; 250 else { /* insert new element at the end, possibly resize */ 251 son = h->elements ; 252 if (son == h->size) /* need resize... */ 253 if (heap_init(h, h->elements+1) ) 254 return 1 ; /* failure... */ 255 h->p[son].object = p ; 256 h->p[son].key = key1 ; 257 h->elements++ ; 258 } 259 while (son > 0) { /* bubble up */ 260 int father = HEAP_FATHER(son) ; 261 struct dn_heap_entry tmp ; 262 263 if (DN_KEY_LT( h->p[father].key, h->p[son].key ) ) 264 break ; /* found right position */ 265 /* son smaller than father, swap and repeat */ 266 HEAP_SWAP(h->p[son], h->p[father], tmp) ; 267 SET_OFFSET(h, son); 268 son = father ; 269 } 270 SET_OFFSET(h, son); 271 return 0 ; 272 } 273 274 /* 275 * remove top element from heap, or obj if obj != NULL 276 */ 277 static void 278 heap_extract(struct dn_heap *h, void *obj) 279 { 280 int child, father, max = h->elements - 1 ; 281 282 if (max < 0) { 283 printf("warning, extract from empty heap 0x%p\n", h); 284 return ; 285 } 286 father = 0 ; /* default: move up smallest child */ 287 if (obj != NULL) { /* extract specific element, index is at offset */ 288 if (h->offset <= 0) 289 panic("*** heap_extract from middle not supported on this heap!!!\n"); 290 father = *((int *)((char *)obj + h->offset)) ; 291 if (father < 0 || father >= h->elements) { 292 printf("dummynet: heap_extract, father %d out of bound 0..%d\n", 293 father, h->elements); 294 panic("heap_extract"); 295 } 296 } 297 RESET_OFFSET(h, father); 298 child = HEAP_LEFT(father) ; /* left child */ 299 while (child <= max) { /* valid entry */ 300 if (child != max && DN_KEY_LT(h->p[child+1].key, h->p[child].key) ) 301 child = child+1 ; /* take right child, otherwise left */ 302 h->p[father] = h->p[child] ; 303 SET_OFFSET(h, father); 304 father = child ; 305 child = HEAP_LEFT(child) ; /* left child for next loop */ 306 } 307 h->elements-- ; 308 if (father != max) { 309 /* 310 * Fill hole with last entry and bubble up, reusing the insert code 311 */ 312 h->p[father] = h->p[max] ; 313 heap_insert(h, father, NULL); /* this one cannot fail */ 314 } 315 } 316 317 #if 0 318 /* 319 * change object position and update references 320 * XXX this one is never used! 321 */ 322 static void 323 heap_move(struct dn_heap *h, dn_key new_key, void *object) 324 { 325 int temp; 326 int i ; 327 int max = h->elements-1 ; 328 struct dn_heap_entry buf ; 329 330 if (h->offset <= 0) 331 panic("cannot move items on this heap"); 332 333 i = *((int *)((char *)object + h->offset)); 334 if (DN_KEY_LT(new_key, h->p[i].key) ) { /* must move up */ 335 h->p[i].key = new_key ; 336 for (; i>0 && DN_KEY_LT(new_key, h->p[(temp = HEAP_FATHER(i))].key) ; 337 i = temp ) { /* bubble up */ 338 HEAP_SWAP(h->p[i], h->p[temp], buf) ; 339 SET_OFFSET(h, i); 340 } 341 } else { /* must move down */ 342 h->p[i].key = new_key ; 343 while ( (temp = HEAP_LEFT(i)) <= max ) { /* found left child */ 344 if ((temp != max) && DN_KEY_GT(h->p[temp].key, h->p[temp+1].key)) 345 temp++ ; /* select child with min key */ 346 if (DN_KEY_GT(new_key, h->p[temp].key)) { /* go down */ 347 HEAP_SWAP(h->p[i], h->p[temp], buf) ; 348 SET_OFFSET(h, i); 349 } else 350 break ; 351 i = temp ; 352 } 353 } 354 SET_OFFSET(h, i); 355 } 356 #endif /* heap_move, unused */ 357 358 /* 359 * heapify() will reorganize data inside an array to maintain the 360 * heap property. It is needed when we delete a bunch of entries. 361 */ 362 static void 363 heapify(struct dn_heap *h) 364 { 365 int i ; 366 367 for (i = 0 ; i < h->elements ; i++ ) 368 heap_insert(h, i , NULL) ; 369 } 370 371 /* 372 * cleanup the heap and free data structure 373 */ 374 static void 375 heap_free(struct dn_heap *h) 376 { 377 if (h->size >0 ) 378 free(h->p, M_DUMMYNET); 379 bzero(h, sizeof(*h) ); 380 } 381 382 /* 383 * --- end of heap management functions --- 384 */ 385 386 /* 387 * Scheduler functions: 388 * 389 * transmit_event() is called when the delay-line needs to enter 390 * the scheduler, either because of existing pkts getting ready, 391 * or new packets entering the queue. The event handled is the delivery 392 * time of the packet. 393 * 394 * ready_event() does something similar with fixed-rate queues, and the 395 * event handled is the finish time of the head pkt. 396 * 397 * wfq_ready_event() does something similar with WF2Q queues, and the 398 * event handled is the start time of the head pkt. 399 * 400 * In all cases, we make sure that the data structures are consistent 401 * before passing pkts out, because this might trigger recursive 402 * invocations of the procedures. 403 */ 404 static void 405 transmit_event(struct dn_pipe *pipe) 406 { 407 struct dn_pkt *pkt ; 408 409 while ( (pkt = pipe->head) && DN_KEY_LEQ(pkt->output_time, curr_time) ) { 410 /* 411 * first unlink, then call procedures, since ip_input() can invoke 412 * ip_output() and viceversa, thus causing nested calls 413 */ 414 pipe->head = DN_NEXT(pkt) ; 415 416 /* 417 * The actual mbuf is preceded by a struct dn_pkt, resembling an mbuf 418 * (NOT A REAL one, just a small block of malloc'ed memory) with 419 * m_type = MT_TAG, m_flags = PACKET_TAG_DUMMYNET 420 * dn_m (m_next) = actual mbuf to be processed by ip_input/output 421 * and some other fields. 422 * The block IS FREED HERE because it contains parameters passed 423 * to the called routine. 424 */ 425 switch (pkt->dn_dir) { 426 case DN_TO_IP_OUT: 427 (void)ip_output((struct mbuf *)pkt, NULL, NULL, 0, NULL, NULL); 428 rt_unref (pkt->ro.ro_rt) ; 429 break ; 430 431 case DN_TO_IP_IN : 432 ip_input((struct mbuf *)pkt) ; 433 break ; 434 435 case DN_TO_BDG_FWD : 436 if (!BDG_LOADED) { 437 /* somebody unloaded the bridge module. Drop pkt */ 438 printf("-- dropping bridged packet trapped in pipe--\n"); 439 m_freem(pkt->dn_m); 440 break; 441 } /* fallthrough */ 442 case DN_TO_ETH_DEMUX: 443 { 444 struct mbuf *m = (struct mbuf *)pkt ; 445 struct ether_header *eh; 446 447 if (pkt->dn_m->m_len < ETHER_HDR_LEN && 448 (pkt->dn_m = m_pullup(pkt->dn_m, ETHER_HDR_LEN)) == NULL) { 449 printf("dummynet/bridge: pullup fail, dropping pkt\n"); 450 break; 451 } 452 /* 453 * same as ether_input, make eh be a pointer into the mbuf 454 */ 455 eh = mtod(pkt->dn_m, struct ether_header *); 456 m_adj(pkt->dn_m, ETHER_HDR_LEN); 457 /* 458 * bdg_forward() wants a pointer to the pseudo-mbuf-header, but 459 * on return it will supply the pointer to the actual packet 460 * (originally pkt->dn_m, but could be something else now) if 461 * it has not consumed it. 462 */ 463 if (pkt->dn_dir == DN_TO_BDG_FWD) { 464 m = bdg_forward_ptr(m, eh, pkt->ifp); 465 if (m) 466 m_freem(m); 467 } else 468 ether_demux(NULL, eh, m); /* which consumes the mbuf */ 469 } 470 break ; 471 case DN_TO_ETH_OUT: 472 ether_output_frame(pkt->ifp, (struct mbuf *)pkt); 473 break; 474 475 default: 476 printf("dummynet: bad switch %d!\n", pkt->dn_dir); 477 m_freem(pkt->dn_m); 478 break ; 479 } 480 free(pkt, M_DUMMYNET); 481 } 482 /* if there are leftover packets, put into the heap for next event */ 483 if ( (pkt = pipe->head) ) 484 heap_insert(&extract_heap, pkt->output_time, pipe ) ; 485 /* XXX should check errors on heap_insert, by draining the 486 * whole pipe p and hoping in the future we are more successful 487 */ 488 } 489 490 /* 491 * the following macro computes how many ticks we have to wait 492 * before being able to transmit a packet. The credit is taken from 493 * either a pipe (WF2Q) or a flow_queue (per-flow queueing) 494 */ 495 #define SET_TICKS(pkt, q, p) \ 496 (pkt->dn_m->m_pkthdr.len*8*hz - (q)->numbytes + p->bandwidth - 1 ) / \ 497 p->bandwidth ; 498 499 /* 500 * extract pkt from queue, compute output time (could be now) 501 * and put into delay line (p_queue) 502 */ 503 static void 504 move_pkt(struct dn_pkt *pkt, struct dn_flow_queue *q, 505 struct dn_pipe *p, int len) 506 { 507 q->head = DN_NEXT(pkt) ; 508 q->len-- ; 509 q->len_bytes -= len ; 510 511 pkt->output_time = curr_time + p->delay ; 512 513 if (p->head == NULL) 514 p->head = pkt; 515 else 516 DN_NEXT_NC(p->tail) = (struct mbuf *)pkt; 517 p->tail = pkt; 518 DN_NEXT_NC(p->tail) = NULL; 519 } 520 521 /* 522 * ready_event() is invoked every time the queue must enter the 523 * scheduler, either because the first packet arrives, or because 524 * a previously scheduled event fired. 525 * On invokation, drain as many pkts as possible (could be 0) and then 526 * if there are leftover packets reinsert the pkt in the scheduler. 527 */ 528 static void 529 ready_event(struct dn_flow_queue *q) 530 { 531 struct dn_pkt *pkt; 532 struct dn_pipe *p = q->fs->pipe ; 533 int p_was_empty ; 534 535 if (p == NULL) { 536 printf("ready_event- pipe is gone\n"); 537 return ; 538 } 539 p_was_empty = (p->head == NULL) ; 540 541 /* 542 * schedule fixed-rate queues linked to this pipe: 543 * Account for the bw accumulated since last scheduling, then 544 * drain as many pkts as allowed by q->numbytes and move to 545 * the delay line (in p) computing output time. 546 * bandwidth==0 (no limit) means we can drain the whole queue, 547 * setting len_scaled = 0 does the job. 548 */ 549 q->numbytes += ( curr_time - q->sched_time ) * p->bandwidth; 550 while ( (pkt = q->head) != NULL ) { 551 int len = pkt->dn_m->m_pkthdr.len; 552 int len_scaled = p->bandwidth ? len*8*hz : 0 ; 553 if (len_scaled > q->numbytes ) 554 break ; 555 q->numbytes -= len_scaled ; 556 move_pkt(pkt, q, p, len); 557 } 558 /* 559 * If we have more packets queued, schedule next ready event 560 * (can only occur when bandwidth != 0, otherwise we would have 561 * flushed the whole queue in the previous loop). 562 * To this purpose we record the current time and compute how many 563 * ticks to go for the finish time of the packet. 564 */ 565 if ( (pkt = q->head) != NULL ) { /* this implies bandwidth != 0 */ 566 dn_key t = SET_TICKS(pkt, q, p); /* ticks i have to wait */ 567 q->sched_time = curr_time ; 568 heap_insert(&ready_heap, curr_time + t, (void *)q ); 569 /* XXX should check errors on heap_insert, and drain the whole 570 * queue on error hoping next time we are luckier. 571 */ 572 } else { /* RED needs to know when the queue becomes empty */ 573 q->q_time = curr_time; 574 q->numbytes = 0; 575 } 576 /* 577 * If the delay line was empty call transmit_event(p) now. 578 * Otherwise, the scheduler will take care of it. 579 */ 580 if (p_was_empty) 581 transmit_event(p); 582 } 583 584 /* 585 * Called when we can transmit packets on WF2Q queues. Take pkts out of 586 * the queues at their start time, and enqueue into the delay line. 587 * Packets are drained until p->numbytes < 0. As long as 588 * len_scaled >= p->numbytes, the packet goes into the delay line 589 * with a deadline p->delay. For the last packet, if p->numbytes<0, 590 * there is an additional delay. 591 */ 592 static void 593 ready_event_wfq(struct dn_pipe *p) 594 { 595 int p_was_empty = (p->head == NULL) ; 596 struct dn_heap *sch = &(p->scheduler_heap); 597 struct dn_heap *neh = &(p->not_eligible_heap) ; 598 599 if (p->if_name[0] == 0) /* tx clock is simulated */ 600 p->numbytes += ( curr_time - p->sched_time ) * p->bandwidth; 601 else { /* tx clock is for real, the ifq must be empty or this is a NOP */ 602 if (p->ifp && p->ifp->if_snd.ifq_head != NULL) 603 return ; 604 else { 605 DEB(printf("pipe %d ready from %s --\n", 606 p->pipe_nr, p->if_name);) 607 } 608 } 609 610 /* 611 * While we have backlogged traffic AND credit, we need to do 612 * something on the queue. 613 */ 614 while ( p->numbytes >=0 && (sch->elements>0 || neh->elements >0) ) { 615 if (sch->elements > 0) { /* have some eligible pkts to send out */ 616 struct dn_flow_queue *q = sch->p[0].object ; 617 struct dn_pkt *pkt = q->head; 618 struct dn_flow_set *fs = q->fs; 619 u_int64_t len = pkt->dn_m->m_pkthdr.len; 620 int len_scaled = p->bandwidth ? len*8*hz : 0 ; 621 622 heap_extract(sch, NULL); /* remove queue from heap */ 623 p->numbytes -= len_scaled ; 624 move_pkt(pkt, q, p, len); 625 626 p->V += (len<<MY_M) / p->sum ; /* update V */ 627 q->S = q->F ; /* update start time */ 628 if (q->len == 0) { /* Flow not backlogged any more */ 629 fs->backlogged-- ; 630 heap_insert(&(p->idle_heap), q->F, q); 631 } else { /* still backlogged */ 632 /* 633 * update F and position in backlogged queue, then 634 * put flow in not_eligible_heap (we will fix this later). 635 */ 636 len = (q->head)->dn_m->m_pkthdr.len; 637 q->F += (len<<MY_M)/(u_int64_t) fs->weight ; 638 if (DN_KEY_LEQ(q->S, p->V)) 639 heap_insert(neh, q->S, q); 640 else 641 heap_insert(sch, q->F, q); 642 } 643 } 644 /* 645 * now compute V = max(V, min(S_i)). Remember that all elements in sch 646 * have by definition S_i <= V so if sch is not empty, V is surely 647 * the max and we must not update it. Conversely, if sch is empty 648 * we only need to look at neh. 649 */ 650 if (sch->elements == 0 && neh->elements > 0) 651 p->V = MAX64 ( p->V, neh->p[0].key ); 652 /* move from neh to sch any packets that have become eligible */ 653 while (neh->elements > 0 && DN_KEY_LEQ(neh->p[0].key, p->V) ) { 654 struct dn_flow_queue *q = neh->p[0].object ; 655 heap_extract(neh, NULL); 656 heap_insert(sch, q->F, q); 657 } 658 659 if (p->if_name[0] != '\0') {/* tx clock is from a real thing */ 660 p->numbytes = -1 ; /* mark not ready for I/O */ 661 break ; 662 } 663 } 664 if (sch->elements == 0 && neh->elements == 0 && p->numbytes >= 0 665 && p->idle_heap.elements > 0) { 666 /* 667 * no traffic and no events scheduled. We can get rid of idle-heap. 668 */ 669 int i ; 670 671 for (i = 0 ; i < p->idle_heap.elements ; i++) { 672 struct dn_flow_queue *q = p->idle_heap.p[i].object ; 673 674 q->F = 0 ; 675 q->S = q->F + 1 ; 676 } 677 p->sum = 0 ; 678 p->V = 0 ; 679 p->idle_heap.elements = 0 ; 680 } 681 /* 682 * If we are getting clocks from dummynet (not a real interface) and 683 * If we are under credit, schedule the next ready event. 684 * Also fix the delivery time of the last packet. 685 */ 686 if (p->if_name[0]==0 && p->numbytes < 0) { /* this implies bandwidth >0 */ 687 dn_key t=0 ; /* number of ticks i have to wait */ 688 689 if (p->bandwidth > 0) 690 t = ( p->bandwidth -1 - p->numbytes) / p->bandwidth ; 691 p->tail->output_time += t ; 692 p->sched_time = curr_time ; 693 heap_insert(&wfq_ready_heap, curr_time + t, (void *)p); 694 /* XXX should check errors on heap_insert, and drain the whole 695 * queue on error hoping next time we are luckier. 696 */ 697 } 698 /* 699 * If the delay line was empty call transmit_event(p) now. 700 * Otherwise, the scheduler will take care of it. 701 */ 702 if (p_was_empty) 703 transmit_event(p); 704 } 705 706 /* 707 * This is called once per tick, or HZ times per second. It is used to 708 * increment the current tick counter and schedule expired events. 709 */ 710 static void 711 dummynet(void * __unused unused) 712 { 713 void *p ; /* generic parameter to handler */ 714 struct dn_heap *h ; 715 struct dn_heap *heaps[3]; 716 int i; 717 struct dn_pipe *pe ; 718 719 heaps[0] = &ready_heap ; /* fixed-rate queues */ 720 heaps[1] = &wfq_ready_heap ; /* wfq queues */ 721 heaps[2] = &extract_heap ; /* delay line */ 722 crit_enter(); /* see note on top, splnet() is not enough */ 723 curr_time++ ; 724 for (i=0; i < 3 ; i++) { 725 h = heaps[i]; 726 while (h->elements > 0 && DN_KEY_LEQ(h->p[0].key, curr_time) ) { 727 DDB(if (h->p[0].key > curr_time) 728 printf("-- dummynet: warning, heap %d is %d ticks late\n", 729 i, (int)(curr_time - h->p[0].key));) 730 p = h->p[0].object ; /* store a copy before heap_extract */ 731 heap_extract(h, NULL); /* need to extract before processing */ 732 if (i == 0) 733 ready_event(p) ; 734 else if (i == 1) { 735 struct dn_pipe *pipe = p; 736 if (pipe->if_name[0] != '\0') 737 printf("*** bad ready_event_wfq for pipe %s\n", 738 pipe->if_name); 739 else 740 ready_event_wfq(p) ; 741 } else 742 transmit_event(p); 743 } 744 } 745 /* sweep pipes trying to expire idle flow_queues */ 746 for (pe = all_pipes; pe ; pe = pe->next ) 747 if (pe->idle_heap.elements > 0 && 748 DN_KEY_LT(pe->idle_heap.p[0].key, pe->V) ) { 749 struct dn_flow_queue *q = pe->idle_heap.p[0].object ; 750 751 heap_extract(&(pe->idle_heap), NULL); 752 q->S = q->F + 1 ; /* mark timestamp as invalid */ 753 pe->sum -= q->fs->weight ; 754 } 755 crit_exit(); 756 callout_reset(&dn_timeout, 1, dummynet, NULL); 757 } 758 759 /* 760 * called by an interface when tx_rdy occurs. 761 */ 762 int 763 if_tx_rdy(struct ifnet *ifp) 764 { 765 struct dn_pipe *p; 766 767 for (p = all_pipes; p ; p = p->next ) 768 if (p->ifp == ifp) 769 break ; 770 if (p == NULL) { 771 for (p = all_pipes; p ; p = p->next ) 772 if (!strcmp(p->if_name, ifp->if_xname) ) { 773 p->ifp = ifp ; 774 DEB(printf("++ tx rdy from %s (now found)\n", ifp->if_xname);) 775 break ; 776 } 777 } 778 if (p != NULL) { 779 DEB(printf("++ tx rdy from %s - qlen %d\n", ifp->if_xname, 780 ifp->if_snd.ifq_len);) 781 p->numbytes = 0 ; /* mark ready for I/O */ 782 ready_event_wfq(p); 783 } 784 return 0; 785 } 786 787 /* 788 * Unconditionally expire empty queues in case of shortage. 789 * Returns the number of queues freed. 790 */ 791 static int 792 expire_queues(struct dn_flow_set *fs) 793 { 794 struct dn_flow_queue *q, *prev ; 795 int i, initial_elements = fs->rq_elements ; 796 797 if (fs->last_expired == time_second) 798 return 0 ; 799 fs->last_expired = time_second ; 800 for (i = 0 ; i <= fs->rq_size ; i++) /* last one is overflow */ 801 for (prev=NULL, q = fs->rq[i] ; q != NULL ; ) 802 if (q->head != NULL || q->S != q->F+1) { 803 prev = q ; 804 q = q->next ; 805 } else { /* entry is idle, expire it */ 806 struct dn_flow_queue *old_q = q ; 807 808 if (prev != NULL) 809 prev->next = q = q->next ; 810 else 811 fs->rq[i] = q = q->next ; 812 fs->rq_elements-- ; 813 free(old_q, M_DUMMYNET); 814 } 815 return initial_elements - fs->rq_elements ; 816 } 817 818 /* 819 * If room, create a new queue and put at head of slot i; 820 * otherwise, create or use the default queue. 821 */ 822 static struct dn_flow_queue * 823 create_queue(struct dn_flow_set *fs, int i) 824 { 825 struct dn_flow_queue *q ; 826 827 if (fs->rq_elements > fs->rq_size * dn_max_ratio && 828 expire_queues(fs) == 0) { 829 /* 830 * No way to get room, use or create overflow queue. 831 */ 832 i = fs->rq_size ; 833 if ( fs->rq[i] != NULL ) 834 return fs->rq[i] ; 835 } 836 q = malloc(sizeof(*q), M_DUMMYNET, M_WAITOK | M_ZERO); 837 q->fs = fs ; 838 q->hash_slot = i ; 839 q->next = fs->rq[i] ; 840 q->S = q->F + 1; /* hack - mark timestamp as invalid */ 841 fs->rq[i] = q ; 842 fs->rq_elements++ ; 843 return q ; 844 } 845 846 /* 847 * Given a flow_set and a pkt in last_pkt, find a matching queue 848 * after appropriate masking. The queue is moved to front 849 * so that further searches take less time. 850 */ 851 static struct dn_flow_queue * 852 find_queue(struct dn_flow_set *fs, struct ipfw_flow_id *id) 853 { 854 int i = 0 ; /* we need i and q for new allocations */ 855 struct dn_flow_queue *q, *prev; 856 857 if ( !(fs->flags_fs & DN_HAVE_FLOW_MASK) ) 858 q = fs->rq[0] ; 859 else { 860 /* first, do the masking */ 861 id->dst_ip &= fs->flow_mask.dst_ip ; 862 id->src_ip &= fs->flow_mask.src_ip ; 863 id->dst_port &= fs->flow_mask.dst_port ; 864 id->src_port &= fs->flow_mask.src_port ; 865 id->proto &= fs->flow_mask.proto ; 866 id->flags = 0 ; /* we don't care about this one */ 867 /* then, hash function */ 868 i = ( (id->dst_ip) & 0xffff ) ^ 869 ( (id->dst_ip >> 15) & 0xffff ) ^ 870 ( (id->src_ip << 1) & 0xffff ) ^ 871 ( (id->src_ip >> 16 ) & 0xffff ) ^ 872 (id->dst_port << 1) ^ (id->src_port) ^ 873 (id->proto ); 874 i = i % fs->rq_size ; 875 /* finally, scan the current list for a match */ 876 searches++ ; 877 for (prev=NULL, q = fs->rq[i] ; q ; ) { 878 search_steps++; 879 if (id->dst_ip == q->id.dst_ip && 880 id->src_ip == q->id.src_ip && 881 id->dst_port == q->id.dst_port && 882 id->src_port == q->id.src_port && 883 id->proto == q->id.proto && 884 id->flags == q->id.flags) 885 break ; /* found */ 886 else if (pipe_expire && q->head == NULL && q->S == q->F+1 ) { 887 /* entry is idle and not in any heap, expire it */ 888 struct dn_flow_queue *old_q = q ; 889 890 if (prev != NULL) 891 prev->next = q = q->next ; 892 else 893 fs->rq[i] = q = q->next ; 894 fs->rq_elements-- ; 895 free(old_q, M_DUMMYNET); 896 continue ; 897 } 898 prev = q ; 899 q = q->next ; 900 } 901 if (q && prev != NULL) { /* found and not in front */ 902 prev->next = q->next ; 903 q->next = fs->rq[i] ; 904 fs->rq[i] = q ; 905 } 906 } 907 if (q == NULL) { /* no match, need to allocate a new entry */ 908 q = create_queue(fs, i); 909 if (q != NULL) 910 q->id = *id ; 911 } 912 return q ; 913 } 914 915 static int 916 red_drops(struct dn_flow_set *fs, struct dn_flow_queue *q, int len) 917 { 918 /* 919 * RED algorithm 920 * 921 * RED calculates the average queue size (avg) using a low-pass filter 922 * with an exponential weighted (w_q) moving average: 923 * avg <- (1-w_q) * avg + w_q * q_size 924 * where q_size is the queue length (measured in bytes or * packets). 925 * 926 * If q_size == 0, we compute the idle time for the link, and set 927 * avg = (1 - w_q)^(idle/s) 928 * where s is the time needed for transmitting a medium-sized packet. 929 * 930 * Now, if avg < min_th the packet is enqueued. 931 * If avg > max_th the packet is dropped. Otherwise, the packet is 932 * dropped with probability P function of avg. 933 * 934 */ 935 936 int64_t p_b = 0; 937 /* queue in bytes or packets ? */ 938 u_int q_size = (fs->flags_fs & DN_QSIZE_IS_BYTES) ? q->len_bytes : q->len; 939 940 DEB(printf("\n%d q: %2u ", (int) curr_time, q_size);) 941 942 /* average queue size estimation */ 943 if (q_size != 0) { 944 /* 945 * queue is not empty, avg <- avg + (q_size - avg) * w_q 946 */ 947 int diff = SCALE(q_size) - q->avg; 948 int64_t v = SCALE_MUL((int64_t) diff, (int64_t) fs->w_q); 949 950 q->avg += (int) v; 951 } else { 952 /* 953 * queue is empty, find for how long the queue has been 954 * empty and use a lookup table for computing 955 * (1 - * w_q)^(idle_time/s) where s is the time to send a 956 * (small) packet. 957 * XXX check wraps... 958 */ 959 if (q->avg) { 960 u_int t = (curr_time - q->q_time) / fs->lookup_step; 961 962 q->avg = (t < fs->lookup_depth) ? 963 SCALE_MUL(q->avg, fs->w_q_lookup[t]) : 0; 964 } 965 } 966 DEB(printf("avg: %u ", SCALE_VAL(q->avg));) 967 968 /* should i drop ? */ 969 970 if (q->avg < fs->min_th) { 971 q->count = -1; 972 return 0; /* accept packet ; */ 973 } 974 if (q->avg >= fs->max_th) { /* average queue >= max threshold */ 975 if (fs->flags_fs & DN_IS_GENTLE_RED) { 976 /* 977 * According to Gentle-RED, if avg is greater than max_th the 978 * packet is dropped with a probability 979 * p_b = c_3 * avg - c_4 980 * where c_3 = (1 - max_p) / max_th, and c_4 = 1 - 2 * max_p 981 */ 982 p_b = SCALE_MUL((int64_t) fs->c_3, (int64_t) q->avg) - fs->c_4; 983 } else { 984 q->count = -1; 985 printf("- drop"); 986 return 1 ; 987 } 988 } else if (q->avg > fs->min_th) { 989 /* 990 * we compute p_b using the linear dropping function p_b = c_1 * 991 * avg - c_2, where c_1 = max_p / (max_th - min_th), and c_2 = 992 * max_p * min_th / (max_th - min_th) 993 */ 994 p_b = SCALE_MUL((int64_t) fs->c_1, (int64_t) q->avg) - fs->c_2; 995 } 996 if (fs->flags_fs & DN_QSIZE_IS_BYTES) 997 p_b = (p_b * len) / fs->max_pkt_size; 998 if (++q->count == 0) 999 q->random = random() & 0xffff; 1000 else { 1001 /* 1002 * q->count counts packets arrived since last drop, so a greater 1003 * value of q->count means a greater packet drop probability. 1004 */ 1005 if (SCALE_MUL(p_b, SCALE((int64_t) q->count)) > q->random) { 1006 q->count = 0; 1007 DEB(printf("- red drop");) 1008 /* after a drop we calculate a new random value */ 1009 q->random = random() & 0xffff; 1010 return 1; /* drop */ 1011 } 1012 } 1013 /* end of RED algorithm */ 1014 return 0 ; /* accept */ 1015 } 1016 1017 static __inline 1018 struct dn_flow_set * 1019 locate_flowset(int pipe_nr, struct ip_fw *rule) 1020 { 1021 #if IPFW2 1022 struct dn_flow_set *fs; 1023 ipfw_insn *cmd = rule->cmd + rule->act_ofs; 1024 1025 if (cmd->opcode == O_LOG) 1026 cmd += F_LEN(cmd); 1027 fs = ((ipfw_insn_pipe *)cmd)->pipe_ptr; 1028 1029 if (fs != NULL) 1030 return fs; 1031 1032 if (cmd->opcode == O_QUEUE) 1033 #else /* !IPFW2 */ 1034 struct dn_flow_set *fs = NULL ; 1035 1036 if ( (rule->fw_flg & IP_FW_F_COMMAND) == IP_FW_F_QUEUE ) 1037 #endif /* !IPFW2 */ 1038 for (fs=all_flow_sets; fs && fs->fs_nr != pipe_nr; fs=fs->next) 1039 ; 1040 else { 1041 struct dn_pipe *p1; 1042 for (p1 = all_pipes; p1 && p1->pipe_nr != pipe_nr; p1 = p1->next) 1043 ; 1044 if (p1 != NULL) 1045 fs = &(p1->fs) ; 1046 } 1047 /* record for the future */ 1048 #if IPFW2 1049 ((ipfw_insn_pipe *)cmd)->pipe_ptr = fs; 1050 #else 1051 if (fs != NULL) 1052 rule->pipe_ptr = fs; 1053 #endif 1054 return fs ; 1055 } 1056 1057 /* 1058 * dummynet hook for packets. Below 'pipe' is a pipe or a queue 1059 * depending on whether WF2Q or fixed bw is used. 1060 * 1061 * pipe_nr pipe or queue the packet is destined for. 1062 * dir where shall we send the packet after dummynet. 1063 * m the mbuf with the packet 1064 * ifp the 'ifp' parameter from the caller. 1065 * NULL in ip_input, destination interface in ip_output, 1066 * real_dst in bdg_forward 1067 * ro route parameter (only used in ip_output, NULL otherwise) 1068 * dst destination address, only used by ip_output 1069 * rule matching rule, in case of multiple passes 1070 * flags flags from the caller, only used in ip_output 1071 * 1072 */ 1073 static int 1074 dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa) 1075 { 1076 struct dn_pkt *pkt; 1077 struct dn_flow_set *fs; 1078 struct dn_pipe *pipe ; 1079 u_int64_t len = m->m_pkthdr.len ; 1080 struct dn_flow_queue *q = NULL ; 1081 int is_pipe; 1082 1083 crit_enter(); 1084 #if IPFW2 1085 ipfw_insn *cmd = fwa->rule->cmd + fwa->rule->act_ofs; 1086 1087 if (cmd->opcode == O_LOG) 1088 cmd += F_LEN(cmd); 1089 is_pipe = (cmd->opcode == O_PIPE); 1090 #else 1091 is_pipe = (fwa->rule->fw_flg & IP_FW_F_COMMAND) == IP_FW_F_PIPE; 1092 #endif 1093 1094 pipe_nr &= 0xffff ; 1095 1096 /* 1097 * this is a dummynet rule, so we expect a O_PIPE or O_QUEUE rule 1098 */ 1099 fs = locate_flowset(pipe_nr, fwa->rule); 1100 if (fs == NULL) 1101 goto dropit ; /* this queue/pipe does not exist! */ 1102 pipe = fs->pipe ; 1103 if (pipe == NULL) { /* must be a queue, try find a matching pipe */ 1104 for (pipe = all_pipes; pipe && pipe->pipe_nr != fs->parent_nr; 1105 pipe = pipe->next) 1106 ; 1107 if (pipe != NULL) 1108 fs->pipe = pipe ; 1109 else { 1110 printf("No pipe %d for queue %d, drop pkt\n", 1111 fs->parent_nr, fs->fs_nr); 1112 goto dropit ; 1113 } 1114 } 1115 q = find_queue(fs, &(fwa->f_id)); 1116 if ( q == NULL ) 1117 goto dropit ; /* cannot allocate queue */ 1118 /* 1119 * update statistics, then check reasons to drop pkt 1120 */ 1121 q->tot_bytes += len ; 1122 q->tot_pkts++ ; 1123 if ( fs->plr && random() < fs->plr ) 1124 goto dropit ; /* random pkt drop */ 1125 if ( fs->flags_fs & DN_QSIZE_IS_BYTES) { 1126 if (q->len_bytes > fs->qsize) 1127 goto dropit ; /* queue size overflow */ 1128 } else { 1129 if (q->len >= fs->qsize) 1130 goto dropit ; /* queue count overflow */ 1131 } 1132 if ( fs->flags_fs & DN_IS_RED && red_drops(fs, q, len) ) 1133 goto dropit ; 1134 1135 /* XXX expensive to zero, see if we can remove it*/ 1136 pkt = malloc(sizeof (*pkt), M_DUMMYNET, M_INTWAIT | M_ZERO | M_NULLOK); 1137 if (pkt == NULL) 1138 goto dropit; /* cannot allocate packet header */ 1139 1140 /* ok, i can handle the pkt now... */ 1141 /* build and enqueue packet + parameters */ 1142 pkt->hdr.mh_type = MT_TAG; 1143 pkt->hdr.mh_flags = PACKET_TAG_DUMMYNET; 1144 pkt->rule = fwa->rule ; 1145 DN_NEXT_NC(pkt) = NULL; 1146 pkt->dn_m = m; 1147 pkt->dn_dir = dir ; 1148 1149 pkt->ifp = fwa->oif; 1150 if (dir == DN_TO_IP_OUT) { 1151 /* 1152 * We need to copy *ro because for ICMP pkts (and maybe others) 1153 * the caller passed a pointer into the stack; dst might also be 1154 * a pointer into *ro so it needs to be updated. 1155 */ 1156 pkt->ro = *(fwa->ro); 1157 if (fwa->ro->ro_rt) 1158 fwa->ro->ro_rt->rt_refcnt++ ; 1159 if (fwa->dst == (struct sockaddr_in *)&fwa->ro->ro_dst) /* dst points into ro */ 1160 fwa->dst = (struct sockaddr_in *)&(pkt->ro.ro_dst) ; 1161 1162 pkt->dn_dst = fwa->dst; 1163 pkt->flags = fwa->flags; 1164 } 1165 if (q->head == NULL) 1166 q->head = pkt; 1167 else 1168 DN_NEXT_NC(q->tail) = (struct mbuf *)pkt; 1169 q->tail = pkt; 1170 q->len++; 1171 q->len_bytes += len ; 1172 1173 if ( q->head != pkt ) /* flow was not idle, we are done */ 1174 goto done; 1175 /* 1176 * If we reach this point the flow was previously idle, so we need 1177 * to schedule it. This involves different actions for fixed-rate or 1178 * WF2Q queues. 1179 */ 1180 if (is_pipe) { 1181 /* 1182 * Fixed-rate queue: just insert into the ready_heap. 1183 */ 1184 dn_key t = 0 ; 1185 if (pipe->bandwidth) 1186 t = SET_TICKS(pkt, q, pipe); 1187 q->sched_time = curr_time ; 1188 if (t == 0) /* must process it now */ 1189 ready_event( q ); 1190 else 1191 heap_insert(&ready_heap, curr_time + t , q ); 1192 } else { 1193 /* 1194 * WF2Q. First, compute start time S: if the flow was idle (S=F+1) 1195 * set S to the virtual time V for the controlling pipe, and update 1196 * the sum of weights for the pipe; otherwise, remove flow from 1197 * idle_heap and set S to max(F,V). 1198 * Second, compute finish time F = S + len/weight. 1199 * Third, if pipe was idle, update V=max(S, V). 1200 * Fourth, count one more backlogged flow. 1201 */ 1202 if (DN_KEY_GT(q->S, q->F)) { /* means timestamps are invalid */ 1203 q->S = pipe->V ; 1204 pipe->sum += fs->weight ; /* add weight of new queue */ 1205 } else { 1206 heap_extract(&(pipe->idle_heap), q); 1207 q->S = MAX64(q->F, pipe->V ) ; 1208 } 1209 q->F = q->S + ( len<<MY_M )/(u_int64_t) fs->weight; 1210 1211 if (pipe->not_eligible_heap.elements == 0 && 1212 pipe->scheduler_heap.elements == 0) 1213 pipe->V = MAX64 ( q->S, pipe->V ); 1214 fs->backlogged++ ; 1215 /* 1216 * Look at eligibility. A flow is not eligibile if S>V (when 1217 * this happens, it means that there is some other flow already 1218 * scheduled for the same pipe, so the scheduler_heap cannot be 1219 * empty). If the flow is not eligible we just store it in the 1220 * not_eligible_heap. Otherwise, we store in the scheduler_heap 1221 * and possibly invoke ready_event_wfq() right now if there is 1222 * leftover credit. 1223 * Note that for all flows in scheduler_heap (SCH), S_i <= V, 1224 * and for all flows in not_eligible_heap (NEH), S_i > V . 1225 * So when we need to compute max( V, min(S_i) ) forall i in SCH+NEH, 1226 * we only need to look into NEH. 1227 */ 1228 if (DN_KEY_GT(q->S, pipe->V) ) { /* not eligible */ 1229 if (pipe->scheduler_heap.elements == 0) 1230 printf("++ ouch! not eligible but empty scheduler!\n"); 1231 heap_insert(&(pipe->not_eligible_heap), q->S, q); 1232 } else { 1233 heap_insert(&(pipe->scheduler_heap), q->F, q); 1234 if (pipe->numbytes >= 0) { /* pipe is idle */ 1235 if (pipe->scheduler_heap.elements != 1) 1236 printf("*** OUCH! pipe should have been idle!\n"); 1237 DEB(printf("Waking up pipe %d at %d\n", 1238 pipe->pipe_nr, (int)(q->F >> MY_M)); ) 1239 pipe->sched_time = curr_time ; 1240 ready_event_wfq(pipe); 1241 } 1242 } 1243 } 1244 done: 1245 crit_exit(); 1246 return 0; 1247 1248 dropit: 1249 crit_exit(); 1250 if (q) 1251 q->drops++ ; 1252 m_freem(m); 1253 return ( (fs && (fs->flags_fs & DN_NOERROR)) ? 0 : ENOBUFS); 1254 } 1255 1256 /* 1257 * Below, the rt_unref is only needed when (pkt->dn_dir == DN_TO_IP_OUT) 1258 * Doing this would probably save us the initial bzero of dn_pkt 1259 */ 1260 #define DN_FREE_PKT(pkt) { \ 1261 struct dn_pkt *n = pkt ; \ 1262 rt_unref ( n->ro.ro_rt ) ; \ 1263 m_freem(n->dn_m); \ 1264 pkt = DN_NEXT(n) ; \ 1265 free(n, M_DUMMYNET) ; } 1266 1267 /* 1268 * Dispose all packets and flow_queues on a flow_set. 1269 * If all=1, also remove red lookup table and other storage, 1270 * including the descriptor itself. 1271 * For the one in dn_pipe MUST also cleanup ready_heap... 1272 */ 1273 static void 1274 purge_flow_set(struct dn_flow_set *fs, int all) 1275 { 1276 struct dn_pkt *pkt ; 1277 struct dn_flow_queue *q, *qn ; 1278 int i ; 1279 1280 for (i = 0 ; i <= fs->rq_size ; i++ ) { 1281 for (q = fs->rq[i] ; q ; q = qn ) { 1282 for (pkt = q->head ; pkt ; ) 1283 DN_FREE_PKT(pkt) ; 1284 qn = q->next ; 1285 free(q, M_DUMMYNET); 1286 } 1287 fs->rq[i] = NULL ; 1288 } 1289 fs->rq_elements = 0 ; 1290 if (all) { 1291 /* RED - free lookup table */ 1292 if (fs->w_q_lookup) 1293 free(fs->w_q_lookup, M_DUMMYNET); 1294 if (fs->rq) 1295 free(fs->rq, M_DUMMYNET); 1296 /* if this fs is not part of a pipe, free it */ 1297 if (fs->pipe && fs != &(fs->pipe->fs) ) 1298 free(fs, M_DUMMYNET); 1299 } 1300 } 1301 1302 /* 1303 * Dispose all packets queued on a pipe (not a flow_set). 1304 * Also free all resources associated to a pipe, which is about 1305 * to be deleted. 1306 */ 1307 static void 1308 purge_pipe(struct dn_pipe *pipe) 1309 { 1310 struct dn_pkt *pkt ; 1311 1312 purge_flow_set( &(pipe->fs), 1 ); 1313 1314 for (pkt = pipe->head ; pkt ; ) 1315 DN_FREE_PKT(pkt) ; 1316 1317 heap_free( &(pipe->scheduler_heap) ); 1318 heap_free( &(pipe->not_eligible_heap) ); 1319 heap_free( &(pipe->idle_heap) ); 1320 } 1321 1322 /* 1323 * Delete all pipes and heaps returning memory. Must also 1324 * remove references from all ipfw rules to all pipes. 1325 */ 1326 static void 1327 dummynet_flush() 1328 { 1329 struct dn_pipe *curr_p, *p ; 1330 struct dn_flow_set *fs, *curr_fs; 1331 1332 crit_enter(); 1333 1334 /* remove all references to pipes ...*/ 1335 flush_pipe_ptrs(NULL); 1336 /* prevent future matches... */ 1337 p = all_pipes ; 1338 all_pipes = NULL ; 1339 fs = all_flow_sets ; 1340 all_flow_sets = NULL ; 1341 /* and free heaps so we don't have unwanted events */ 1342 heap_free(&ready_heap); 1343 heap_free(&wfq_ready_heap); 1344 heap_free(&extract_heap); 1345 crit_exit(); 1346 /* 1347 * Now purge all queued pkts and delete all pipes 1348 */ 1349 /* scan and purge all flow_sets. */ 1350 for ( ; fs ; ) { 1351 curr_fs = fs ; 1352 fs = fs->next ; 1353 purge_flow_set(curr_fs, 1); 1354 } 1355 for ( ; p ; ) { 1356 purge_pipe(p); 1357 curr_p = p ; 1358 p = p->next ; 1359 free(curr_p, M_DUMMYNET); 1360 } 1361 } 1362 1363 1364 extern struct ip_fw *ip_fw_default_rule ; 1365 static void 1366 dn_rule_delete_fs(struct dn_flow_set *fs, void *r) 1367 { 1368 int i ; 1369 struct dn_flow_queue *q ; 1370 struct dn_pkt *pkt ; 1371 1372 for (i = 0 ; i <= fs->rq_size ; i++) /* last one is ovflow */ 1373 for (q = fs->rq[i] ; q ; q = q->next ) 1374 for (pkt = q->head ; pkt ; pkt = DN_NEXT(pkt) ) 1375 if (pkt->rule == r) 1376 pkt->rule = ip_fw_default_rule ; 1377 } 1378 /* 1379 * when a firewall rule is deleted, scan all queues and remove the flow-id 1380 * from packets matching this rule. 1381 */ 1382 void 1383 dn_rule_delete(void *r) 1384 { 1385 struct dn_pipe *p ; 1386 struct dn_pkt *pkt ; 1387 struct dn_flow_set *fs ; 1388 1389 /* 1390 * If the rule references a queue (dn_flow_set), then scan 1391 * the flow set, otherwise scan pipes. Should do either, but doing 1392 * both does not harm. 1393 */ 1394 for ( fs = all_flow_sets ; fs ; fs = fs->next ) 1395 dn_rule_delete_fs(fs, r); 1396 for ( p = all_pipes ; p ; p = p->next ) { 1397 fs = &(p->fs) ; 1398 dn_rule_delete_fs(fs, r); 1399 for (pkt = p->head ; pkt ; pkt = DN_NEXT(pkt) ) 1400 if (pkt->rule == r) 1401 pkt->rule = ip_fw_default_rule ; 1402 } 1403 } 1404 1405 /* 1406 * setup RED parameters 1407 */ 1408 static int 1409 config_red(struct dn_flow_set *p, struct dn_flow_set * x) 1410 { 1411 int i; 1412 1413 x->w_q = p->w_q; 1414 x->min_th = SCALE(p->min_th); 1415 x->max_th = SCALE(p->max_th); 1416 x->max_p = p->max_p; 1417 1418 x->c_1 = p->max_p / (p->max_th - p->min_th); 1419 x->c_2 = SCALE_MUL(x->c_1, SCALE(p->min_th)); 1420 if (x->flags_fs & DN_IS_GENTLE_RED) { 1421 x->c_3 = (SCALE(1) - p->max_p) / p->max_th; 1422 x->c_4 = (SCALE(1) - 2 * p->max_p); 1423 } 1424 1425 /* if the lookup table already exist, free and create it again */ 1426 if (x->w_q_lookup) { 1427 free(x->w_q_lookup, M_DUMMYNET); 1428 x->w_q_lookup = NULL ; 1429 } 1430 if (red_lookup_depth == 0) { 1431 printf("\nnet.inet.ip.dummynet.red_lookup_depth must be > 0"); 1432 free(x, M_DUMMYNET); 1433 return EINVAL; 1434 } 1435 x->lookup_depth = red_lookup_depth; 1436 x->w_q_lookup = malloc(x->lookup_depth * sizeof(int), 1437 M_DUMMYNET, M_WAITOK); 1438 1439 /* fill the lookup table with (1 - w_q)^x */ 1440 x->lookup_step = p->lookup_step ; 1441 x->lookup_weight = p->lookup_weight ; 1442 x->w_q_lookup[0] = SCALE(1) - x->w_q; 1443 for (i = 1; i < x->lookup_depth; i++) 1444 x->w_q_lookup[i] = SCALE_MUL(x->w_q_lookup[i - 1], x->lookup_weight); 1445 if (red_avg_pkt_size < 1) 1446 red_avg_pkt_size = 512 ; 1447 x->avg_pkt_size = red_avg_pkt_size ; 1448 if (red_max_pkt_size < 1) 1449 red_max_pkt_size = 1500 ; 1450 x->max_pkt_size = red_max_pkt_size ; 1451 return 0 ; 1452 } 1453 1454 static int 1455 alloc_hash(struct dn_flow_set *x, struct dn_flow_set *pfs) 1456 { 1457 if (x->flags_fs & DN_HAVE_FLOW_MASK) { /* allocate some slots */ 1458 int l = pfs->rq_size; 1459 1460 if (l == 0) 1461 l = dn_hash_size; 1462 if (l < 4) 1463 l = 4; 1464 else if (l > DN_MAX_HASH_SIZE) 1465 l = DN_MAX_HASH_SIZE; 1466 x->rq_size = l; 1467 } else /* one is enough for null mask */ 1468 x->rq_size = 1; 1469 x->rq = malloc((1 + x->rq_size) * sizeof(struct dn_flow_queue *), 1470 M_DUMMYNET, M_WAITOK | M_ZERO); 1471 x->rq_elements = 0; 1472 return 0 ; 1473 } 1474 1475 static void 1476 set_fs_parms(struct dn_flow_set *x, struct dn_flow_set *src) 1477 { 1478 x->flags_fs = src->flags_fs; 1479 x->qsize = src->qsize; 1480 x->plr = src->plr; 1481 x->flow_mask = src->flow_mask; 1482 if (x->flags_fs & DN_QSIZE_IS_BYTES) { 1483 if (x->qsize > 1024*1024) 1484 x->qsize = 1024*1024 ; 1485 } else { 1486 if (x->qsize == 0) 1487 x->qsize = 50 ; 1488 if (x->qsize > 100) 1489 x->qsize = 50 ; 1490 } 1491 /* configuring RED */ 1492 if ( x->flags_fs & DN_IS_RED ) 1493 config_red(src, x) ; /* XXX should check errors */ 1494 } 1495 1496 /* 1497 * setup pipe or queue parameters. 1498 */ 1499 1500 static int 1501 config_pipe(struct dn_pipe *p) 1502 { 1503 int i, s; 1504 struct dn_flow_set *pfs = &(p->fs); 1505 struct dn_flow_queue *q; 1506 1507 /* 1508 * The config program passes parameters as follows: 1509 * bw = bits/second (0 means no limits), 1510 * delay = ms, must be translated into ticks. 1511 * qsize = slots/bytes 1512 */ 1513 p->delay = ( p->delay * hz ) / 1000 ; 1514 /* We need either a pipe number or a flow_set number */ 1515 if (p->pipe_nr == 0 && pfs->fs_nr == 0) 1516 return EINVAL ; 1517 if (p->pipe_nr != 0 && pfs->fs_nr != 0) 1518 return EINVAL ; 1519 if (p->pipe_nr != 0) { /* this is a pipe */ 1520 struct dn_pipe *x, *a, *b; 1521 /* locate pipe */ 1522 for (a = NULL , b = all_pipes ; b && b->pipe_nr < p->pipe_nr ; 1523 a = b , b = b->next) ; 1524 1525 if (b == NULL || b->pipe_nr != p->pipe_nr) { /* new pipe */ 1526 x = malloc(sizeof(struct dn_pipe), M_DUMMYNET, M_WAITOK | M_ZERO); 1527 x->pipe_nr = p->pipe_nr; 1528 x->fs.pipe = x ; 1529 /* idle_heap is the only one from which we extract from the middle. 1530 */ 1531 x->idle_heap.size = x->idle_heap.elements = 0 ; 1532 x->idle_heap.offset=OFFSET_OF(struct dn_flow_queue, heap_pos); 1533 } else { 1534 x = b; 1535 crit_enter(); 1536 /* Flush accumulated credit for all queues */ 1537 for (i = 0; i <= x->fs.rq_size; i++) 1538 for (q = x->fs.rq[i]; q; q = q->next) 1539 q->numbytes = 0; 1540 crit_exit(); 1541 } 1542 1543 crit_enter(); 1544 x->bandwidth = p->bandwidth ; 1545 x->numbytes = 0; /* just in case... */ 1546 bcopy(p->if_name, x->if_name, sizeof(p->if_name) ); 1547 x->ifp = NULL ; /* reset interface ptr */ 1548 x->delay = p->delay ; 1549 set_fs_parms(&(x->fs), pfs); 1550 1551 1552 if ( x->fs.rq == NULL ) { /* a new pipe */ 1553 s = alloc_hash(&(x->fs), pfs) ; 1554 if (s) { 1555 free(x, M_DUMMYNET); 1556 return s ; 1557 } 1558 x->next = b ; 1559 if (a == NULL) 1560 all_pipes = x ; 1561 else 1562 a->next = x ; 1563 } 1564 crit_exit(); 1565 } else { /* config queue */ 1566 struct dn_flow_set *x, *a, *b ; 1567 1568 /* locate flow_set */ 1569 for (a=NULL, b=all_flow_sets ; b && b->fs_nr < pfs->fs_nr ; 1570 a = b , b = b->next) ; 1571 1572 if (b == NULL || b->fs_nr != pfs->fs_nr) { /* new */ 1573 if (pfs->parent_nr == 0) /* need link to a pipe */ 1574 return EINVAL ; 1575 x = malloc(sizeof(struct dn_flow_set), M_DUMMYNET, M_WAITOK|M_ZERO); 1576 x->fs_nr = pfs->fs_nr; 1577 x->parent_nr = pfs->parent_nr; 1578 x->weight = pfs->weight ; 1579 if (x->weight == 0) 1580 x->weight = 1 ; 1581 else if (x->weight > 100) 1582 x->weight = 100 ; 1583 } else { 1584 /* Change parent pipe not allowed; must delete and recreate */ 1585 if (pfs->parent_nr != 0 && b->parent_nr != pfs->parent_nr) 1586 return EINVAL ; 1587 x = b; 1588 } 1589 crit_enter(); 1590 set_fs_parms(x, pfs); 1591 1592 if ( x->rq == NULL ) { /* a new flow_set */ 1593 s = alloc_hash(x, pfs) ; 1594 if (s) { 1595 free(x, M_DUMMYNET); 1596 return s ; 1597 } 1598 x->next = b; 1599 if (a == NULL) 1600 all_flow_sets = x; 1601 else 1602 a->next = x; 1603 } 1604 crit_exit(); 1605 } 1606 return 0 ; 1607 } 1608 1609 /* 1610 * Helper function to remove from a heap queues which are linked to 1611 * a flow_set about to be deleted. 1612 */ 1613 static void 1614 fs_remove_from_heap(struct dn_heap *h, struct dn_flow_set *fs) 1615 { 1616 int i = 0, found = 0 ; 1617 for (; i < h->elements ;) 1618 if ( ((struct dn_flow_queue *)h->p[i].object)->fs == fs) { 1619 h->elements-- ; 1620 h->p[i] = h->p[h->elements] ; 1621 found++ ; 1622 } else 1623 i++ ; 1624 if (found) 1625 heapify(h); 1626 } 1627 1628 /* 1629 * helper function to remove a pipe from a heap (can be there at most once) 1630 */ 1631 static void 1632 pipe_remove_from_heap(struct dn_heap *h, struct dn_pipe *p) 1633 { 1634 if (h->elements > 0) { 1635 int i = 0 ; 1636 for (i=0; i < h->elements ; i++ ) { 1637 if (h->p[i].object == p) { /* found it */ 1638 h->elements-- ; 1639 h->p[i] = h->p[h->elements] ; 1640 heapify(h); 1641 break ; 1642 } 1643 } 1644 } 1645 } 1646 1647 /* 1648 * drain all queues. Called in case of severe mbuf shortage. 1649 */ 1650 void 1651 dummynet_drain() 1652 { 1653 struct dn_flow_set *fs; 1654 struct dn_pipe *p; 1655 struct dn_pkt *pkt; 1656 1657 heap_free(&ready_heap); 1658 heap_free(&wfq_ready_heap); 1659 heap_free(&extract_heap); 1660 /* remove all references to this pipe from flow_sets */ 1661 for (fs = all_flow_sets; fs; fs= fs->next ) 1662 purge_flow_set(fs, 0); 1663 1664 for (p = all_pipes; p; p= p->next ) { 1665 purge_flow_set(&(p->fs), 0); 1666 for (pkt = p->head ; pkt ; ) 1667 DN_FREE_PKT(pkt) ; 1668 p->head = p->tail = NULL ; 1669 } 1670 } 1671 1672 /* 1673 * Fully delete a pipe or a queue, cleaning up associated info. 1674 */ 1675 static int 1676 delete_pipe(struct dn_pipe *p) 1677 { 1678 if (p->pipe_nr == 0 && p->fs.fs_nr == 0) 1679 return EINVAL ; 1680 if (p->pipe_nr != 0 && p->fs.fs_nr != 0) 1681 return EINVAL ; 1682 if (p->pipe_nr != 0) { /* this is an old-style pipe */ 1683 struct dn_pipe *a, *b; 1684 struct dn_flow_set *fs; 1685 1686 /* locate pipe */ 1687 for (a = NULL , b = all_pipes ; b && b->pipe_nr < p->pipe_nr ; 1688 a = b , b = b->next) ; 1689 if (b == NULL || (b->pipe_nr != p->pipe_nr) ) 1690 return EINVAL ; /* not found */ 1691 1692 crit_enter(); 1693 1694 /* unlink from list of pipes */ 1695 if (a == NULL) 1696 all_pipes = b->next ; 1697 else 1698 a->next = b->next ; 1699 /* remove references to this pipe from the ip_fw rules. */ 1700 flush_pipe_ptrs(&(b->fs)); 1701 1702 /* remove all references to this pipe from flow_sets */ 1703 for (fs = all_flow_sets; fs; fs= fs->next ) 1704 if (fs->pipe == b) { 1705 printf("++ ref to pipe %d from fs %d\n", 1706 p->pipe_nr, fs->fs_nr); 1707 fs->pipe = NULL ; 1708 purge_flow_set(fs, 0); 1709 } 1710 fs_remove_from_heap(&ready_heap, &(b->fs)); 1711 purge_pipe(b); /* remove all data associated to this pipe */ 1712 /* remove reference to here from extract_heap and wfq_ready_heap */ 1713 pipe_remove_from_heap(&extract_heap, b); 1714 pipe_remove_from_heap(&wfq_ready_heap, b); 1715 crit_exit(); 1716 free(b, M_DUMMYNET); 1717 } else { /* this is a WF2Q queue (dn_flow_set) */ 1718 struct dn_flow_set *a, *b; 1719 1720 /* locate set */ 1721 for (a = NULL, b = all_flow_sets ; b && b->fs_nr < p->fs.fs_nr ; 1722 a = b , b = b->next) ; 1723 if (b == NULL || (b->fs_nr != p->fs.fs_nr) ) 1724 return EINVAL ; /* not found */ 1725 1726 crit_enter(); 1727 if (a == NULL) 1728 all_flow_sets = b->next ; 1729 else 1730 a->next = b->next ; 1731 /* remove references to this flow_set from the ip_fw rules. */ 1732 flush_pipe_ptrs(b); 1733 1734 if (b->pipe != NULL) { 1735 /* Update total weight on parent pipe and cleanup parent heaps */ 1736 b->pipe->sum -= b->weight * b->backlogged ; 1737 fs_remove_from_heap(&(b->pipe->not_eligible_heap), b); 1738 fs_remove_from_heap(&(b->pipe->scheduler_heap), b); 1739 #if 1 /* XXX should i remove from idle_heap as well ? */ 1740 fs_remove_from_heap(&(b->pipe->idle_heap), b); 1741 #endif 1742 } 1743 purge_flow_set(b, 1); 1744 crit_exit(); 1745 } 1746 return 0 ; 1747 } 1748 1749 /* 1750 * helper function used to copy data from kernel in DUMMYNET_GET 1751 */ 1752 static char * 1753 dn_copy_set(struct dn_flow_set *set, char *bp) 1754 { 1755 int i, copied = 0 ; 1756 struct dn_flow_queue *q, *qp = (struct dn_flow_queue *)bp; 1757 1758 for (i = 0 ; i <= set->rq_size ; i++) 1759 for (q = set->rq[i] ; q ; q = q->next, qp++ ) { 1760 if (q->hash_slot != i) 1761 printf("++ at %d: wrong slot (have %d, " 1762 "should be %d)\n", copied, q->hash_slot, i); 1763 if (q->fs != set) 1764 printf("++ at %d: wrong fs ptr (have %p, should be %p)\n", 1765 i, q->fs, set); 1766 copied++ ; 1767 bcopy(q, qp, sizeof( *q ) ); 1768 /* cleanup pointers */ 1769 qp->next = NULL ; 1770 qp->head = qp->tail = NULL ; 1771 qp->fs = NULL ; 1772 } 1773 if (copied != set->rq_elements) 1774 printf("++ wrong count, have %d should be %d\n", 1775 copied, set->rq_elements); 1776 return (char *)qp ; 1777 } 1778 1779 static int 1780 dummynet_get(struct sockopt *sopt) 1781 { 1782 char *buf, *bp ; /* bp is the "copy-pointer" */ 1783 size_t size ; 1784 struct dn_flow_set *set ; 1785 struct dn_pipe *p ; 1786 int error=0 ; 1787 1788 crit_enter(); 1789 /* 1790 * compute size of data structures: list of pipes and flow_sets. 1791 */ 1792 for (p = all_pipes, size = 0 ; p ; p = p->next ) 1793 size += sizeof( *p ) + 1794 p->fs.rq_elements * sizeof(struct dn_flow_queue); 1795 for (set = all_flow_sets ; set ; set = set->next ) 1796 size += sizeof ( *set ) + 1797 set->rq_elements * sizeof(struct dn_flow_queue); 1798 buf = malloc(size, M_TEMP, M_WAITOK); 1799 for (p = all_pipes, bp = buf ; p ; p = p->next ) { 1800 struct dn_pipe *pipe_bp = (struct dn_pipe *)bp ; 1801 1802 /* 1803 * copy pipe descriptor into *bp, convert delay back to ms, 1804 * then copy the flow_set descriptor(s) one at a time. 1805 * After each flow_set, copy the queue descriptor it owns. 1806 */ 1807 bcopy(p, bp, sizeof( *p ) ); 1808 pipe_bp->delay = (pipe_bp->delay * 1000) / hz ; 1809 /* 1810 * XXX the following is a hack based on ->next being the 1811 * first field in dn_pipe and dn_flow_set. The correct 1812 * solution would be to move the dn_flow_set to the beginning 1813 * of struct dn_pipe. 1814 */ 1815 pipe_bp->next = (struct dn_pipe *)DN_IS_PIPE ; 1816 /* clean pointers */ 1817 pipe_bp->head = pipe_bp->tail = NULL ; 1818 pipe_bp->fs.next = NULL ; 1819 pipe_bp->fs.pipe = NULL ; 1820 pipe_bp->fs.rq = NULL ; 1821 1822 bp += sizeof( *p ) ; 1823 bp = dn_copy_set( &(p->fs), bp ); 1824 } 1825 for (set = all_flow_sets ; set ; set = set->next ) { 1826 struct dn_flow_set *fs_bp = (struct dn_flow_set *)bp ; 1827 bcopy(set, bp, sizeof( *set ) ); 1828 /* XXX same hack as above */ 1829 fs_bp->next = (struct dn_flow_set *)DN_IS_QUEUE ; 1830 fs_bp->pipe = NULL ; 1831 fs_bp->rq = NULL ; 1832 bp += sizeof( *set ) ; 1833 bp = dn_copy_set( set, bp ); 1834 } 1835 crit_exit(); 1836 error = sooptcopyout(sopt, buf, size); 1837 free(buf, M_TEMP); 1838 return error ; 1839 } 1840 1841 /* 1842 * Handler for the various dummynet socket options (get, flush, config, del) 1843 */ 1844 static int 1845 ip_dn_ctl(struct sockopt *sopt) 1846 { 1847 int error = 0 ; 1848 struct dn_pipe *p, tmp_pipe; 1849 1850 /* Disallow sets in really-really secure mode. */ 1851 if (sopt->sopt_dir == SOPT_SET) { 1852 #if defined(__FreeBSD__) && __FreeBSD_version >= 500034 1853 error = securelevel_ge(sopt->sopt_td->td_ucred, 3); 1854 if (error) 1855 return (error); 1856 #else 1857 if (securelevel >= 3) 1858 return (EPERM); 1859 #endif 1860 } 1861 1862 switch (sopt->sopt_name) { 1863 default : 1864 printf("ip_dn_ctl -- unknown option %d", sopt->sopt_name); 1865 return EINVAL ; 1866 1867 case IP_DUMMYNET_GET : 1868 error = dummynet_get(sopt); 1869 break ; 1870 1871 case IP_DUMMYNET_FLUSH : 1872 dummynet_flush() ; 1873 break ; 1874 1875 case IP_DUMMYNET_CONFIGURE : 1876 p = &tmp_pipe ; 1877 error = sooptcopyin(sopt, p, sizeof *p, sizeof *p); 1878 if (error) 1879 break ; 1880 error = config_pipe(p); 1881 break ; 1882 1883 case IP_DUMMYNET_DEL : /* remove a pipe or queue */ 1884 p = &tmp_pipe ; 1885 error = sooptcopyin(sopt, p, sizeof *p, sizeof *p); 1886 if (error) 1887 break ; 1888 1889 error = delete_pipe(p); 1890 break ; 1891 } 1892 return error ; 1893 } 1894 1895 static void 1896 ip_dn_init(void) 1897 { 1898 printf("DUMMYNET initialized (011031)\n"); 1899 all_pipes = NULL ; 1900 all_flow_sets = NULL ; 1901 ready_heap.size = ready_heap.elements = 0 ; 1902 ready_heap.offset = 0 ; 1903 1904 wfq_ready_heap.size = wfq_ready_heap.elements = 0 ; 1905 wfq_ready_heap.offset = 0 ; 1906 1907 extract_heap.size = extract_heap.elements = 0 ; 1908 extract_heap.offset = 0 ; 1909 ip_dn_ctl_ptr = ip_dn_ctl; 1910 ip_dn_io_ptr = dummynet_io; 1911 ip_dn_ruledel_ptr = dn_rule_delete; 1912 callout_init(&dn_timeout); 1913 callout_reset(&dn_timeout, 1, dummynet, NULL); 1914 } 1915 1916 static int 1917 dummynet_modevent(module_t mod, int type, void *data) 1918 { 1919 switch (type) { 1920 case MOD_LOAD: 1921 crit_enter(); 1922 if (DUMMYNET_LOADED) { 1923 crit_exit(); 1924 printf("DUMMYNET already loaded\n"); 1925 return EEXIST ; 1926 } 1927 ip_dn_init(); 1928 crit_exit(); 1929 break; 1930 1931 case MOD_UNLOAD: 1932 #if !defined(KLD_MODULE) 1933 printf("dummynet statically compiled, cannot unload\n"); 1934 return EINVAL ; 1935 #else 1936 crit_enter(); 1937 callout_stop(&dn_timeout); 1938 dummynet_flush(); 1939 ip_dn_ctl_ptr = NULL; 1940 ip_dn_io_ptr = NULL; 1941 ip_dn_ruledel_ptr = NULL; 1942 crit_exit(); 1943 #endif 1944 break ; 1945 default: 1946 break ; 1947 } 1948 return 0 ; 1949 } 1950 1951 static moduledata_t dummynet_mod = { 1952 "dummynet", 1953 dummynet_modevent, 1954 NULL 1955 }; 1956 DECLARE_MODULE(dummynet, dummynet_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 1957 MODULE_DEPEND(dummynet, ipfw, 1, 1, 1); 1958 MODULE_VERSION(dummynet, 1); 1959