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 1.8 2004/04/22 04:21:58 dillon 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 <net/if.h> 77 #include <net/route.h> 78 #include <netinet/in.h> 79 #include <netinet/in_systm.h> 80 #include <netinet/in_var.h> 81 #include <netinet/ip.h> 82 #include <net/ipfw/ip_fw.h> 83 #include "ip_dummynet.h" 84 #include <netinet/ip_var.h> 85 86 #include <netinet/if_ether.h> /* for struct arpcom */ 87 #include <net/bridge/bridge.h> 88 89 /* 90 * We keep a private variable for the simulation time, but we could 91 * probably use an existing one ("softticks" in sys/kern/kern_timer.c) 92 */ 93 static dn_key curr_time = 0 ; /* current simulation time */ 94 95 static int dn_hash_size = 64 ; /* default hash size */ 96 97 /* statistics on number of queue searches and search steps */ 98 static int searches, search_steps ; 99 static int pipe_expire = 1 ; /* expire queue if empty */ 100 static int dn_max_ratio = 16 ; /* max queues/buckets ratio */ 101 102 static int red_lookup_depth = 256; /* RED - default lookup table depth */ 103 static int red_avg_pkt_size = 512; /* RED - default medium packet size */ 104 static int red_max_pkt_size = 1500; /* RED - default max packet size */ 105 106 /* 107 * Three heaps contain queues and pipes that the scheduler handles: 108 * 109 * ready_heap contains all dn_flow_queue related to fixed-rate pipes. 110 * 111 * wfq_ready_heap contains the pipes associated with WF2Q flows 112 * 113 * extract_heap contains pipes associated with delay lines. 114 * 115 */ 116 117 MALLOC_DEFINE(M_DUMMYNET, "dummynet", "dummynet heap"); 118 119 static struct dn_heap ready_heap, extract_heap, wfq_ready_heap ; 120 121 static int heap_init(struct dn_heap *h, int size) ; 122 static int heap_insert (struct dn_heap *h, dn_key key1, void *p); 123 static void heap_extract(struct dn_heap *h, void *obj); 124 125 static void transmit_event(struct dn_pipe *pipe); 126 static void ready_event(struct dn_flow_queue *q); 127 128 static struct dn_pipe *all_pipes = NULL ; /* list of all pipes */ 129 static struct dn_flow_set *all_flow_sets = NULL ;/* list of all flow_sets */ 130 131 static struct callout_handle dn_timeout; 132 133 #ifdef SYSCTL_NODE 134 SYSCTL_NODE(_net_inet_ip, OID_AUTO, dummynet, 135 CTLFLAG_RW, 0, "Dummynet"); 136 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, hash_size, 137 CTLFLAG_RW, &dn_hash_size, 0, "Default hash table size"); 138 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, curr_time, 139 CTLFLAG_RD, &curr_time, 0, "Current tick"); 140 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, ready_heap, 141 CTLFLAG_RD, &ready_heap.size, 0, "Size of ready heap"); 142 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, extract_heap, 143 CTLFLAG_RD, &extract_heap.size, 0, "Size of extract heap"); 144 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, searches, 145 CTLFLAG_RD, &searches, 0, "Number of queue searches"); 146 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, search_steps, 147 CTLFLAG_RD, &search_steps, 0, "Number of queue search steps"); 148 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, expire, 149 CTLFLAG_RW, &pipe_expire, 0, "Expire queue if empty"); 150 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, max_chain_len, 151 CTLFLAG_RW, &dn_max_ratio, 0, 152 "Max ratio between dynamic queues and buckets"); 153 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_lookup_depth, 154 CTLFLAG_RD, &red_lookup_depth, 0, "Depth of RED lookup table"); 155 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_avg_pkt_size, 156 CTLFLAG_RD, &red_avg_pkt_size, 0, "RED Medium packet size"); 157 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_max_pkt_size, 158 CTLFLAG_RD, &red_max_pkt_size, 0, "RED Max packet size"); 159 #endif 160 161 static int config_pipe(struct dn_pipe *p); 162 static int ip_dn_ctl(struct sockopt *sopt); 163 164 static void rt_unref(struct rtentry *); 165 static void dummynet(void *); 166 static void dummynet_flush(void); 167 void dummynet_drain(void); 168 static ip_dn_io_t dummynet_io; 169 static void dn_rule_delete(void *); 170 171 int if_tx_rdy(struct ifnet *ifp); 172 173 static void 174 rt_unref(struct rtentry *rt) 175 { 176 if (rt == NULL) 177 return ; 178 if (rt->rt_refcnt <= 0) 179 printf("-- warning, refcnt now %ld, decreasing\n", rt->rt_refcnt); 180 RTFREE(rt); 181 } 182 183 /* 184 * Heap management functions. 185 * 186 * In the heap, first node is element 0. Children of i are 2i+1 and 2i+2. 187 * Some macros help finding parent/children so we can optimize them. 188 * 189 * heap_init() is called to expand the heap when needed. 190 * Increment size in blocks of 16 entries. 191 * XXX failure to allocate a new element is a pretty bad failure 192 * as we basically stall a whole queue forever!! 193 * Returns 1 on error, 0 on success 194 */ 195 #define HEAP_FATHER(x) ( ( (x) - 1 ) / 2 ) 196 #define HEAP_LEFT(x) ( 2*(x) + 1 ) 197 #define HEAP_IS_LEFT(x) ( (x) & 1 ) 198 #define HEAP_RIGHT(x) ( 2*(x) + 2 ) 199 #define HEAP_SWAP(a, b, buffer) { buffer = a ; a = b ; b = buffer ; } 200 #define HEAP_INCREMENT 15 201 202 static int 203 heap_init(struct dn_heap *h, int new_size) 204 { 205 struct dn_heap_entry *p; 206 207 if (h->size >= new_size ) { 208 printf("heap_init, Bogus call, have %d want %d\n", 209 h->size, new_size); 210 return 0 ; 211 } 212 new_size = (new_size + HEAP_INCREMENT ) & ~HEAP_INCREMENT ; 213 p = malloc(new_size * sizeof(*p), M_DUMMYNET, M_WAITOK | M_ZERO); 214 if (h->size > 0) { 215 bcopy(h->p, p, h->size * sizeof(*p) ); 216 free(h->p, M_DUMMYNET); 217 } 218 h->p = p ; 219 h->size = new_size ; 220 return 0 ; 221 } 222 223 /* 224 * Insert element in heap. Normally, p != NULL, we insert p in 225 * a new position and bubble up. If p == NULL, then the element is 226 * already in place, and key is the position where to start the 227 * bubble-up. 228 * Returns 1 on failure (cannot allocate new heap entry) 229 * 230 * If offset > 0 the position (index, int) of the element in the heap is 231 * also stored in the element itself at the given offset in bytes. 232 */ 233 #define SET_OFFSET(heap, node) \ 234 if (heap->offset > 0) \ 235 *((int *)((char *)(heap->p[node].object) + heap->offset)) = node ; 236 /* 237 * RESET_OFFSET is used for sanity checks. It sets offset to an invalid value. 238 */ 239 #define RESET_OFFSET(heap, node) \ 240 if (heap->offset > 0) \ 241 *((int *)((char *)(heap->p[node].object) + heap->offset)) = -1 ; 242 static int 243 heap_insert(struct dn_heap *h, dn_key key1, void *p) 244 { 245 int son = h->elements ; 246 247 if (p == NULL) /* data already there, set starting point */ 248 son = key1 ; 249 else { /* insert new element at the end, possibly resize */ 250 son = h->elements ; 251 if (son == h->size) /* need resize... */ 252 if (heap_init(h, h->elements+1) ) 253 return 1 ; /* failure... */ 254 h->p[son].object = p ; 255 h->p[son].key = key1 ; 256 h->elements++ ; 257 } 258 while (son > 0) { /* bubble up */ 259 int father = HEAP_FATHER(son) ; 260 struct dn_heap_entry tmp ; 261 262 if (DN_KEY_LT( h->p[father].key, h->p[son].key ) ) 263 break ; /* found right position */ 264 /* son smaller than father, swap and repeat */ 265 HEAP_SWAP(h->p[son], h->p[father], tmp) ; 266 SET_OFFSET(h, son); 267 son = father ; 268 } 269 SET_OFFSET(h, son); 270 return 0 ; 271 } 272 273 /* 274 * remove top element from heap, or obj if obj != NULL 275 */ 276 static void 277 heap_extract(struct dn_heap *h, void *obj) 278 { 279 int child, father, max = h->elements - 1 ; 280 281 if (max < 0) { 282 printf("warning, extract from empty heap 0x%p\n", h); 283 return ; 284 } 285 father = 0 ; /* default: move up smallest child */ 286 if (obj != NULL) { /* extract specific element, index is at offset */ 287 if (h->offset <= 0) 288 panic("*** heap_extract from middle not supported on this heap!!!\n"); 289 father = *((int *)((char *)obj + h->offset)) ; 290 if (father < 0 || father >= h->elements) { 291 printf("dummynet: heap_extract, father %d out of bound 0..%d\n", 292 father, h->elements); 293 panic("heap_extract"); 294 } 295 } 296 RESET_OFFSET(h, father); 297 child = HEAP_LEFT(father) ; /* left child */ 298 while (child <= max) { /* valid entry */ 299 if (child != max && DN_KEY_LT(h->p[child+1].key, h->p[child].key) ) 300 child = child+1 ; /* take right child, otherwise left */ 301 h->p[father] = h->p[child] ; 302 SET_OFFSET(h, father); 303 father = child ; 304 child = HEAP_LEFT(child) ; /* left child for next loop */ 305 } 306 h->elements-- ; 307 if (father != max) { 308 /* 309 * Fill hole with last entry and bubble up, reusing the insert code 310 */ 311 h->p[father] = h->p[max] ; 312 heap_insert(h, father, NULL); /* this one cannot fail */ 313 } 314 } 315 316 #if 0 317 /* 318 * change object position and update references 319 * XXX this one is never used! 320 */ 321 static void 322 heap_move(struct dn_heap *h, dn_key new_key, void *object) 323 { 324 int temp; 325 int i ; 326 int max = h->elements-1 ; 327 struct dn_heap_entry buf ; 328 329 if (h->offset <= 0) 330 panic("cannot move items on this heap"); 331 332 i = *((int *)((char *)object + h->offset)); 333 if (DN_KEY_LT(new_key, h->p[i].key) ) { /* must move up */ 334 h->p[i].key = new_key ; 335 for (; i>0 && DN_KEY_LT(new_key, h->p[(temp = HEAP_FATHER(i))].key) ; 336 i = temp ) { /* bubble up */ 337 HEAP_SWAP(h->p[i], h->p[temp], buf) ; 338 SET_OFFSET(h, i); 339 } 340 } else { /* must move down */ 341 h->p[i].key = new_key ; 342 while ( (temp = HEAP_LEFT(i)) <= max ) { /* found left child */ 343 if ((temp != max) && DN_KEY_GT(h->p[temp].key, h->p[temp+1].key)) 344 temp++ ; /* select child with min key */ 345 if (DN_KEY_GT(new_key, h->p[temp].key)) { /* go down */ 346 HEAP_SWAP(h->p[i], h->p[temp], buf) ; 347 SET_OFFSET(h, i); 348 } else 349 break ; 350 i = temp ; 351 } 352 } 353 SET_OFFSET(h, i); 354 } 355 #endif /* heap_move, unused */ 356 357 /* 358 * heapify() will reorganize data inside an array to maintain the 359 * heap property. It is needed when we delete a bunch of entries. 360 */ 361 static void 362 heapify(struct dn_heap *h) 363 { 364 int i ; 365 366 for (i = 0 ; i < h->elements ; i++ ) 367 heap_insert(h, i , NULL) ; 368 } 369 370 /* 371 * cleanup the heap and free data structure 372 */ 373 static void 374 heap_free(struct dn_heap *h) 375 { 376 if (h->size >0 ) 377 free(h->p, M_DUMMYNET); 378 bzero(h, sizeof(*h) ); 379 } 380 381 /* 382 * --- end of heap management functions --- 383 */ 384 385 /* 386 * Scheduler functions: 387 * 388 * transmit_event() is called when the delay-line needs to enter 389 * the scheduler, either because of existing pkts getting ready, 390 * or new packets entering the queue. The event handled is the delivery 391 * time of the packet. 392 * 393 * ready_event() does something similar with fixed-rate queues, and the 394 * event handled is the finish time of the head pkt. 395 * 396 * wfq_ready_event() does something similar with WF2Q queues, and the 397 * event handled is the start time of the head pkt. 398 * 399 * In all cases, we make sure that the data structures are consistent 400 * before passing pkts out, because this might trigger recursive 401 * invocations of the procedures. 402 */ 403 static void 404 transmit_event(struct dn_pipe *pipe) 405 { 406 struct dn_pkt *pkt ; 407 408 while ( (pkt = pipe->head) && DN_KEY_LEQ(pkt->output_time, curr_time) ) { 409 /* 410 * first unlink, then call procedures, since ip_input() can invoke 411 * ip_output() and viceversa, thus causing nested calls 412 */ 413 pipe->head = DN_NEXT(pkt) ; 414 415 /* 416 * The actual mbuf is preceded by a struct dn_pkt, resembling an mbuf 417 * (NOT A REAL one, just a small block of malloc'ed memory) with 418 * m_type = MT_TAG, m_flags = PACKET_TAG_DUMMYNET 419 * dn_m (m_next) = actual mbuf to be processed by ip_input/output 420 * and some other fields. 421 * The block IS FREED HERE because it contains parameters passed 422 * to the called routine. 423 */ 424 switch (pkt->dn_dir) { 425 case DN_TO_IP_OUT: 426 (void)ip_output((struct mbuf *)pkt, NULL, NULL, 0, NULL, NULL); 427 rt_unref (pkt->ro.ro_rt) ; 428 break ; 429 430 case DN_TO_IP_IN : 431 ip_input((struct mbuf *)pkt) ; 432 break ; 433 434 case DN_TO_BDG_FWD : 435 if (!BDG_LOADED) { 436 /* somebody unloaded the bridge module. Drop pkt */ 437 printf("-- dropping bridged packet trapped in pipe--\n"); 438 m_freem(pkt->dn_m); 439 break; 440 } /* fallthrough */ 441 case DN_TO_ETH_DEMUX: 442 { 443 struct mbuf *m = (struct mbuf *)pkt ; 444 struct ether_header *eh; 445 446 if (pkt->dn_m->m_len < ETHER_HDR_LEN && 447 (pkt->dn_m = m_pullup(pkt->dn_m, ETHER_HDR_LEN)) == NULL) { 448 printf("dummynet/bridge: pullup fail, dropping pkt\n"); 449 break; 450 } 451 /* 452 * same as ether_input, make eh be a pointer into the mbuf 453 */ 454 eh = mtod(pkt->dn_m, struct ether_header *); 455 m_adj(pkt->dn_m, ETHER_HDR_LEN); 456 /* 457 * bdg_forward() wants a pointer to the pseudo-mbuf-header, but 458 * on return it will supply the pointer to the actual packet 459 * (originally pkt->dn_m, but could be something else now) if 460 * it has not consumed it. 461 */ 462 if (pkt->dn_dir == DN_TO_BDG_FWD) { 463 m = bdg_forward_ptr(m, eh, pkt->ifp); 464 if (m) 465 m_freem(m); 466 } else 467 ether_demux(NULL, eh, m); /* which consumes the mbuf */ 468 } 469 break ; 470 case DN_TO_ETH_OUT: 471 ether_output_frame(pkt->ifp, (struct mbuf *)pkt); 472 break; 473 474 default: 475 printf("dummynet: bad switch %d!\n", pkt->dn_dir); 476 m_freem(pkt->dn_m); 477 break ; 478 } 479 free(pkt, M_DUMMYNET); 480 } 481 /* if there are leftover packets, put into the heap for next event */ 482 if ( (pkt = pipe->head) ) 483 heap_insert(&extract_heap, pkt->output_time, pipe ) ; 484 /* XXX should check errors on heap_insert, by draining the 485 * whole pipe p and hoping in the future we are more successful 486 */ 487 } 488 489 /* 490 * the following macro computes how many ticks we have to wait 491 * before being able to transmit a packet. The credit is taken from 492 * either a pipe (WF2Q) or a flow_queue (per-flow queueing) 493 */ 494 #define SET_TICKS(pkt, q, p) \ 495 (pkt->dn_m->m_pkthdr.len*8*hz - (q)->numbytes + p->bandwidth - 1 ) / \ 496 p->bandwidth ; 497 498 /* 499 * extract pkt from queue, compute output time (could be now) 500 * and put into delay line (p_queue) 501 */ 502 static void 503 move_pkt(struct dn_pkt *pkt, struct dn_flow_queue *q, 504 struct dn_pipe *p, int len) 505 { 506 q->head = DN_NEXT(pkt) ; 507 q->len-- ; 508 q->len_bytes -= len ; 509 510 pkt->output_time = curr_time + p->delay ; 511 512 if (p->head == NULL) 513 p->head = pkt; 514 else 515 DN_NEXT(p->tail) = pkt; 516 p->tail = pkt; 517 DN_NEXT(p->tail) = NULL; 518 } 519 520 /* 521 * ready_event() is invoked every time the queue must enter the 522 * scheduler, either because the first packet arrives, or because 523 * a previously scheduled event fired. 524 * On invokation, drain as many pkts as possible (could be 0) and then 525 * if there are leftover packets reinsert the pkt in the scheduler. 526 */ 527 static void 528 ready_event(struct dn_flow_queue *q) 529 { 530 struct dn_pkt *pkt; 531 struct dn_pipe *p = q->fs->pipe ; 532 int p_was_empty ; 533 534 if (p == NULL) { 535 printf("ready_event- pipe is gone\n"); 536 return ; 537 } 538 p_was_empty = (p->head == NULL) ; 539 540 /* 541 * schedule fixed-rate queues linked to this pipe: 542 * Account for the bw accumulated since last scheduling, then 543 * drain as many pkts as allowed by q->numbytes and move to 544 * the delay line (in p) computing output time. 545 * bandwidth==0 (no limit) means we can drain the whole queue, 546 * setting len_scaled = 0 does the job. 547 */ 548 q->numbytes += ( curr_time - q->sched_time ) * p->bandwidth; 549 while ( (pkt = q->head) != NULL ) { 550 int len = pkt->dn_m->m_pkthdr.len; 551 int len_scaled = p->bandwidth ? len*8*hz : 0 ; 552 if (len_scaled > q->numbytes ) 553 break ; 554 q->numbytes -= len_scaled ; 555 move_pkt(pkt, q, p, len); 556 } 557 /* 558 * If we have more packets queued, schedule next ready event 559 * (can only occur when bandwidth != 0, otherwise we would have 560 * flushed the whole queue in the previous loop). 561 * To this purpose we record the current time and compute how many 562 * ticks to go for the finish time of the packet. 563 */ 564 if ( (pkt = q->head) != NULL ) { /* this implies bandwidth != 0 */ 565 dn_key t = SET_TICKS(pkt, q, p); /* ticks i have to wait */ 566 q->sched_time = curr_time ; 567 heap_insert(&ready_heap, curr_time + t, (void *)q ); 568 /* XXX should check errors on heap_insert, and drain the whole 569 * queue on error hoping next time we are luckier. 570 */ 571 } else { /* RED needs to know when the queue becomes empty */ 572 q->q_time = curr_time; 573 q->numbytes = 0; 574 } 575 /* 576 * If the delay line was empty call transmit_event(p) now. 577 * Otherwise, the scheduler will take care of it. 578 */ 579 if (p_was_empty) 580 transmit_event(p); 581 } 582 583 /* 584 * Called when we can transmit packets on WF2Q queues. Take pkts out of 585 * the queues at their start time, and enqueue into the delay line. 586 * Packets are drained until p->numbytes < 0. As long as 587 * len_scaled >= p->numbytes, the packet goes into the delay line 588 * with a deadline p->delay. For the last packet, if p->numbytes<0, 589 * there is an additional delay. 590 */ 591 static void 592 ready_event_wfq(struct dn_pipe *p) 593 { 594 int p_was_empty = (p->head == NULL) ; 595 struct dn_heap *sch = &(p->scheduler_heap); 596 struct dn_heap *neh = &(p->not_eligible_heap) ; 597 598 if (p->if_name[0] == 0) /* tx clock is simulated */ 599 p->numbytes += ( curr_time - p->sched_time ) * p->bandwidth; 600 else { /* tx clock is for real, the ifq must be empty or this is a NOP */ 601 if (p->ifp && p->ifp->if_snd.ifq_head != NULL) 602 return ; 603 else { 604 DEB(printf("pipe %d ready from %s --\n", 605 p->pipe_nr, p->if_name);) 606 } 607 } 608 609 /* 610 * While we have backlogged traffic AND credit, we need to do 611 * something on the queue. 612 */ 613 while ( p->numbytes >=0 && (sch->elements>0 || neh->elements >0) ) { 614 if (sch->elements > 0) { /* have some eligible pkts to send out */ 615 struct dn_flow_queue *q = sch->p[0].object ; 616 struct dn_pkt *pkt = q->head; 617 struct dn_flow_set *fs = q->fs; 618 u_int64_t len = pkt->dn_m->m_pkthdr.len; 619 int len_scaled = p->bandwidth ? len*8*hz : 0 ; 620 621 heap_extract(sch, NULL); /* remove queue from heap */ 622 p->numbytes -= len_scaled ; 623 move_pkt(pkt, q, p, len); 624 625 p->V += (len<<MY_M) / p->sum ; /* update V */ 626 q->S = q->F ; /* update start time */ 627 if (q->len == 0) { /* Flow not backlogged any more */ 628 fs->backlogged-- ; 629 heap_insert(&(p->idle_heap), q->F, q); 630 } else { /* still backlogged */ 631 /* 632 * update F and position in backlogged queue, then 633 * put flow in not_eligible_heap (we will fix this later). 634 */ 635 len = (q->head)->dn_m->m_pkthdr.len; 636 q->F += (len<<MY_M)/(u_int64_t) fs->weight ; 637 if (DN_KEY_LEQ(q->S, p->V)) 638 heap_insert(neh, q->S, q); 639 else 640 heap_insert(sch, q->F, q); 641 } 642 } 643 /* 644 * now compute V = max(V, min(S_i)). Remember that all elements in sch 645 * have by definition S_i <= V so if sch is not empty, V is surely 646 * the max and we must not update it. Conversely, if sch is empty 647 * we only need to look at neh. 648 */ 649 if (sch->elements == 0 && neh->elements > 0) 650 p->V = MAX64 ( p->V, neh->p[0].key ); 651 /* move from neh to sch any packets that have become eligible */ 652 while (neh->elements > 0 && DN_KEY_LEQ(neh->p[0].key, p->V) ) { 653 struct dn_flow_queue *q = neh->p[0].object ; 654 heap_extract(neh, NULL); 655 heap_insert(sch, q->F, q); 656 } 657 658 if (p->if_name[0] != '\0') {/* tx clock is from a real thing */ 659 p->numbytes = -1 ; /* mark not ready for I/O */ 660 break ; 661 } 662 } 663 if (sch->elements == 0 && neh->elements == 0 && p->numbytes >= 0 664 && p->idle_heap.elements > 0) { 665 /* 666 * no traffic and no events scheduled. We can get rid of idle-heap. 667 */ 668 int i ; 669 670 for (i = 0 ; i < p->idle_heap.elements ; i++) { 671 struct dn_flow_queue *q = p->idle_heap.p[i].object ; 672 673 q->F = 0 ; 674 q->S = q->F + 1 ; 675 } 676 p->sum = 0 ; 677 p->V = 0 ; 678 p->idle_heap.elements = 0 ; 679 } 680 /* 681 * If we are getting clocks from dummynet (not a real interface) and 682 * If we are under credit, schedule the next ready event. 683 * Also fix the delivery time of the last packet. 684 */ 685 if (p->if_name[0]==0 && p->numbytes < 0) { /* this implies bandwidth >0 */ 686 dn_key t=0 ; /* number of ticks i have to wait */ 687 688 if (p->bandwidth > 0) 689 t = ( p->bandwidth -1 - p->numbytes) / p->bandwidth ; 690 p->tail->output_time += t ; 691 p->sched_time = curr_time ; 692 heap_insert(&wfq_ready_heap, curr_time + t, (void *)p); 693 /* XXX should check errors on heap_insert, and drain the whole 694 * queue on error hoping next time we are luckier. 695 */ 696 } 697 /* 698 * If the delay line was empty call transmit_event(p) now. 699 * Otherwise, the scheduler will take care of it. 700 */ 701 if (p_was_empty) 702 transmit_event(p); 703 } 704 705 /* 706 * This is called once per tick, or HZ times per second. It is used to 707 * increment the current tick counter and schedule expired events. 708 */ 709 static void 710 dummynet(void * __unused unused) 711 { 712 void *p ; /* generic parameter to handler */ 713 struct dn_heap *h ; 714 int s ; 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 s = splimp(); /* 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 splx(s); 756 dn_timeout = timeout(dummynet, NULL, 1); 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 s = splimp(); 1082 int is_pipe; 1083 #if IPFW2 1084 ipfw_insn *cmd = fwa->rule->cmd + fwa->rule->act_ofs; 1085 1086 if (cmd->opcode == O_LOG) 1087 cmd += F_LEN(cmd); 1088 is_pipe = (cmd->opcode == O_PIPE); 1089 #else 1090 is_pipe = (fwa->rule->fw_flg & IP_FW_F_COMMAND) == IP_FW_F_PIPE; 1091 #endif 1092 1093 pipe_nr &= 0xffff ; 1094 1095 /* 1096 * this is a dummynet rule, so we expect a O_PIPE or O_QUEUE rule 1097 */ 1098 fs = locate_flowset(pipe_nr, fwa->rule); 1099 if (fs == NULL) 1100 goto dropit ; /* this queue/pipe does not exist! */ 1101 pipe = fs->pipe ; 1102 if (pipe == NULL) { /* must be a queue, try find a matching pipe */ 1103 for (pipe = all_pipes; pipe && pipe->pipe_nr != fs->parent_nr; 1104 pipe = pipe->next) 1105 ; 1106 if (pipe != NULL) 1107 fs->pipe = pipe ; 1108 else { 1109 printf("No pipe %d for queue %d, drop pkt\n", 1110 fs->parent_nr, fs->fs_nr); 1111 goto dropit ; 1112 } 1113 } 1114 q = find_queue(fs, &(fwa->f_id)); 1115 if ( q == NULL ) 1116 goto dropit ; /* cannot allocate queue */ 1117 /* 1118 * update statistics, then check reasons to drop pkt 1119 */ 1120 q->tot_bytes += len ; 1121 q->tot_pkts++ ; 1122 if ( fs->plr && random() < fs->plr ) 1123 goto dropit ; /* random pkt drop */ 1124 if ( fs->flags_fs & DN_QSIZE_IS_BYTES) { 1125 if (q->len_bytes > fs->qsize) 1126 goto dropit ; /* queue size overflow */ 1127 } else { 1128 if (q->len >= fs->qsize) 1129 goto dropit ; /* queue count overflow */ 1130 } 1131 if ( fs->flags_fs & DN_IS_RED && red_drops(fs, q, len) ) 1132 goto dropit ; 1133 1134 /* XXX expensive to zero, see if we can remove it*/ 1135 pkt = malloc(sizeof (*pkt), M_DUMMYNET, M_INTWAIT | M_ZERO | M_NULLOK); 1136 if (pkt == NULL) 1137 goto dropit; /* cannot allocate packet header */ 1138 1139 /* ok, i can handle the pkt now... */ 1140 /* build and enqueue packet + parameters */ 1141 pkt->hdr.mh_type = MT_TAG; 1142 pkt->hdr.mh_flags = PACKET_TAG_DUMMYNET; 1143 pkt->rule = fwa->rule ; 1144 DN_NEXT(pkt) = NULL; 1145 pkt->dn_m = m; 1146 pkt->dn_dir = dir ; 1147 1148 pkt->ifp = fwa->oif; 1149 if (dir == DN_TO_IP_OUT) { 1150 /* 1151 * We need to copy *ro because for ICMP pkts (and maybe others) 1152 * the caller passed a pointer into the stack; dst might also be 1153 * a pointer into *ro so it needs to be updated. 1154 */ 1155 pkt->ro = *(fwa->ro); 1156 if (fwa->ro->ro_rt) 1157 fwa->ro->ro_rt->rt_refcnt++ ; 1158 if (fwa->dst == (struct sockaddr_in *)&fwa->ro->ro_dst) /* dst points into ro */ 1159 fwa->dst = (struct sockaddr_in *)&(pkt->ro.ro_dst) ; 1160 1161 pkt->dn_dst = fwa->dst; 1162 pkt->flags = fwa->flags; 1163 } 1164 if (q->head == NULL) 1165 q->head = pkt; 1166 else 1167 DN_NEXT(q->tail) = pkt; 1168 q->tail = pkt; 1169 q->len++; 1170 q->len_bytes += len ; 1171 1172 if ( q->head != pkt ) /* flow was not idle, we are done */ 1173 goto done; 1174 /* 1175 * If we reach this point the flow was previously idle, so we need 1176 * to schedule it. This involves different actions for fixed-rate or 1177 * WF2Q queues. 1178 */ 1179 if (is_pipe) { 1180 /* 1181 * Fixed-rate queue: just insert into the ready_heap. 1182 */ 1183 dn_key t = 0 ; 1184 if (pipe->bandwidth) 1185 t = SET_TICKS(pkt, q, pipe); 1186 q->sched_time = curr_time ; 1187 if (t == 0) /* must process it now */ 1188 ready_event( q ); 1189 else 1190 heap_insert(&ready_heap, curr_time + t , q ); 1191 } else { 1192 /* 1193 * WF2Q. First, compute start time S: if the flow was idle (S=F+1) 1194 * set S to the virtual time V for the controlling pipe, and update 1195 * the sum of weights for the pipe; otherwise, remove flow from 1196 * idle_heap and set S to max(F,V). 1197 * Second, compute finish time F = S + len/weight. 1198 * Third, if pipe was idle, update V=max(S, V). 1199 * Fourth, count one more backlogged flow. 1200 */ 1201 if (DN_KEY_GT(q->S, q->F)) { /* means timestamps are invalid */ 1202 q->S = pipe->V ; 1203 pipe->sum += fs->weight ; /* add weight of new queue */ 1204 } else { 1205 heap_extract(&(pipe->idle_heap), q); 1206 q->S = MAX64(q->F, pipe->V ) ; 1207 } 1208 q->F = q->S + ( len<<MY_M )/(u_int64_t) fs->weight; 1209 1210 if (pipe->not_eligible_heap.elements == 0 && 1211 pipe->scheduler_heap.elements == 0) 1212 pipe->V = MAX64 ( q->S, pipe->V ); 1213 fs->backlogged++ ; 1214 /* 1215 * Look at eligibility. A flow is not eligibile if S>V (when 1216 * this happens, it means that there is some other flow already 1217 * scheduled for the same pipe, so the scheduler_heap cannot be 1218 * empty). If the flow is not eligible we just store it in the 1219 * not_eligible_heap. Otherwise, we store in the scheduler_heap 1220 * and possibly invoke ready_event_wfq() right now if there is 1221 * leftover credit. 1222 * Note that for all flows in scheduler_heap (SCH), S_i <= V, 1223 * and for all flows in not_eligible_heap (NEH), S_i > V . 1224 * So when we need to compute max( V, min(S_i) ) forall i in SCH+NEH, 1225 * we only need to look into NEH. 1226 */ 1227 if (DN_KEY_GT(q->S, pipe->V) ) { /* not eligible */ 1228 if (pipe->scheduler_heap.elements == 0) 1229 printf("++ ouch! not eligible but empty scheduler!\n"); 1230 heap_insert(&(pipe->not_eligible_heap), q->S, q); 1231 } else { 1232 heap_insert(&(pipe->scheduler_heap), q->F, q); 1233 if (pipe->numbytes >= 0) { /* pipe is idle */ 1234 if (pipe->scheduler_heap.elements != 1) 1235 printf("*** OUCH! pipe should have been idle!\n"); 1236 DEB(printf("Waking up pipe %d at %d\n", 1237 pipe->pipe_nr, (int)(q->F >> MY_M)); ) 1238 pipe->sched_time = curr_time ; 1239 ready_event_wfq(pipe); 1240 } 1241 } 1242 } 1243 done: 1244 splx(s); 1245 return 0; 1246 1247 dropit: 1248 splx(s); 1249 if (q) 1250 q->drops++ ; 1251 m_freem(m); 1252 return ( (fs && (fs->flags_fs & DN_NOERROR)) ? 0 : ENOBUFS); 1253 } 1254 1255 /* 1256 * Below, the rt_unref is only needed when (pkt->dn_dir == DN_TO_IP_OUT) 1257 * Doing this would probably save us the initial bzero of dn_pkt 1258 */ 1259 #define DN_FREE_PKT(pkt) { \ 1260 struct dn_pkt *n = pkt ; \ 1261 rt_unref ( n->ro.ro_rt ) ; \ 1262 m_freem(n->dn_m); \ 1263 pkt = DN_NEXT(n) ; \ 1264 free(n, M_DUMMYNET) ; } 1265 1266 /* 1267 * Dispose all packets and flow_queues on a flow_set. 1268 * If all=1, also remove red lookup table and other storage, 1269 * including the descriptor itself. 1270 * For the one in dn_pipe MUST also cleanup ready_heap... 1271 */ 1272 static void 1273 purge_flow_set(struct dn_flow_set *fs, int all) 1274 { 1275 struct dn_pkt *pkt ; 1276 struct dn_flow_queue *q, *qn ; 1277 int i ; 1278 1279 for (i = 0 ; i <= fs->rq_size ; i++ ) { 1280 for (q = fs->rq[i] ; q ; q = qn ) { 1281 for (pkt = q->head ; pkt ; ) 1282 DN_FREE_PKT(pkt) ; 1283 qn = q->next ; 1284 free(q, M_DUMMYNET); 1285 } 1286 fs->rq[i] = NULL ; 1287 } 1288 fs->rq_elements = 0 ; 1289 if (all) { 1290 /* RED - free lookup table */ 1291 if (fs->w_q_lookup) 1292 free(fs->w_q_lookup, M_DUMMYNET); 1293 if (fs->rq) 1294 free(fs->rq, M_DUMMYNET); 1295 /* if this fs is not part of a pipe, free it */ 1296 if (fs->pipe && fs != &(fs->pipe->fs) ) 1297 free(fs, M_DUMMYNET); 1298 } 1299 } 1300 1301 /* 1302 * Dispose all packets queued on a pipe (not a flow_set). 1303 * Also free all resources associated to a pipe, which is about 1304 * to be deleted. 1305 */ 1306 static void 1307 purge_pipe(struct dn_pipe *pipe) 1308 { 1309 struct dn_pkt *pkt ; 1310 1311 purge_flow_set( &(pipe->fs), 1 ); 1312 1313 for (pkt = pipe->head ; pkt ; ) 1314 DN_FREE_PKT(pkt) ; 1315 1316 heap_free( &(pipe->scheduler_heap) ); 1317 heap_free( &(pipe->not_eligible_heap) ); 1318 heap_free( &(pipe->idle_heap) ); 1319 } 1320 1321 /* 1322 * Delete all pipes and heaps returning memory. Must also 1323 * remove references from all ipfw rules to all pipes. 1324 */ 1325 static void 1326 dummynet_flush() 1327 { 1328 struct dn_pipe *curr_p, *p ; 1329 struct dn_flow_set *fs, *curr_fs; 1330 int s ; 1331 1332 s = splimp() ; 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 splx(s) ; 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 s = splimp(); 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 splx(s); 1541 } 1542 1543 s = splimp(); 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 splx(s); 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 s = splimp(); 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 splx(s); 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 int s ; 1679 1680 if (p->pipe_nr == 0 && p->fs.fs_nr == 0) 1681 return EINVAL ; 1682 if (p->pipe_nr != 0 && p->fs.fs_nr != 0) 1683 return EINVAL ; 1684 if (p->pipe_nr != 0) { /* this is an old-style pipe */ 1685 struct dn_pipe *a, *b; 1686 struct dn_flow_set *fs; 1687 1688 /* locate pipe */ 1689 for (a = NULL , b = all_pipes ; b && b->pipe_nr < p->pipe_nr ; 1690 a = b , b = b->next) ; 1691 if (b == NULL || (b->pipe_nr != p->pipe_nr) ) 1692 return EINVAL ; /* not found */ 1693 1694 s = splimp() ; 1695 1696 /* unlink from list of pipes */ 1697 if (a == NULL) 1698 all_pipes = b->next ; 1699 else 1700 a->next = b->next ; 1701 /* remove references to this pipe from the ip_fw rules. */ 1702 flush_pipe_ptrs(&(b->fs)); 1703 1704 /* remove all references to this pipe from flow_sets */ 1705 for (fs = all_flow_sets; fs; fs= fs->next ) 1706 if (fs->pipe == b) { 1707 printf("++ ref to pipe %d from fs %d\n", 1708 p->pipe_nr, fs->fs_nr); 1709 fs->pipe = NULL ; 1710 purge_flow_set(fs, 0); 1711 } 1712 fs_remove_from_heap(&ready_heap, &(b->fs)); 1713 purge_pipe(b); /* remove all data associated to this pipe */ 1714 /* remove reference to here from extract_heap and wfq_ready_heap */ 1715 pipe_remove_from_heap(&extract_heap, b); 1716 pipe_remove_from_heap(&wfq_ready_heap, b); 1717 splx(s); 1718 free(b, M_DUMMYNET); 1719 } else { /* this is a WF2Q queue (dn_flow_set) */ 1720 struct dn_flow_set *a, *b; 1721 1722 /* locate set */ 1723 for (a = NULL, b = all_flow_sets ; b && b->fs_nr < p->fs.fs_nr ; 1724 a = b , b = b->next) ; 1725 if (b == NULL || (b->fs_nr != p->fs.fs_nr) ) 1726 return EINVAL ; /* not found */ 1727 1728 s = splimp() ; 1729 if (a == NULL) 1730 all_flow_sets = b->next ; 1731 else 1732 a->next = b->next ; 1733 /* remove references to this flow_set from the ip_fw rules. */ 1734 flush_pipe_ptrs(b); 1735 1736 if (b->pipe != NULL) { 1737 /* Update total weight on parent pipe and cleanup parent heaps */ 1738 b->pipe->sum -= b->weight * b->backlogged ; 1739 fs_remove_from_heap(&(b->pipe->not_eligible_heap), b); 1740 fs_remove_from_heap(&(b->pipe->scheduler_heap), b); 1741 #if 1 /* XXX should i remove from idle_heap as well ? */ 1742 fs_remove_from_heap(&(b->pipe->idle_heap), b); 1743 #endif 1744 } 1745 purge_flow_set(b, 1); 1746 splx(s); 1747 } 1748 return 0 ; 1749 } 1750 1751 /* 1752 * helper function used to copy data from kernel in DUMMYNET_GET 1753 */ 1754 static char * 1755 dn_copy_set(struct dn_flow_set *set, char *bp) 1756 { 1757 int i, copied = 0 ; 1758 struct dn_flow_queue *q, *qp = (struct dn_flow_queue *)bp; 1759 1760 for (i = 0 ; i <= set->rq_size ; i++) 1761 for (q = set->rq[i] ; q ; q = q->next, qp++ ) { 1762 if (q->hash_slot != i) 1763 printf("++ at %d: wrong slot (have %d, " 1764 "should be %d)\n", copied, q->hash_slot, i); 1765 if (q->fs != set) 1766 printf("++ at %d: wrong fs ptr (have %p, should be %p)\n", 1767 i, q->fs, set); 1768 copied++ ; 1769 bcopy(q, qp, sizeof( *q ) ); 1770 /* cleanup pointers */ 1771 qp->next = NULL ; 1772 qp->head = qp->tail = NULL ; 1773 qp->fs = NULL ; 1774 } 1775 if (copied != set->rq_elements) 1776 printf("++ wrong count, have %d should be %d\n", 1777 copied, set->rq_elements); 1778 return (char *)qp ; 1779 } 1780 1781 static int 1782 dummynet_get(struct sockopt *sopt) 1783 { 1784 char *buf, *bp ; /* bp is the "copy-pointer" */ 1785 size_t size ; 1786 struct dn_flow_set *set ; 1787 struct dn_pipe *p ; 1788 int s, error=0 ; 1789 1790 s = splimp(); 1791 /* 1792 * compute size of data structures: list of pipes and flow_sets. 1793 */ 1794 for (p = all_pipes, size = 0 ; p ; p = p->next ) 1795 size += sizeof( *p ) + 1796 p->fs.rq_elements * sizeof(struct dn_flow_queue); 1797 for (set = all_flow_sets ; set ; set = set->next ) 1798 size += sizeof ( *set ) + 1799 set->rq_elements * sizeof(struct dn_flow_queue); 1800 buf = malloc(size, M_TEMP, M_WAITOK); 1801 for (p = all_pipes, bp = buf ; p ; p = p->next ) { 1802 struct dn_pipe *pipe_bp = (struct dn_pipe *)bp ; 1803 1804 /* 1805 * copy pipe descriptor into *bp, convert delay back to ms, 1806 * then copy the flow_set descriptor(s) one at a time. 1807 * After each flow_set, copy the queue descriptor it owns. 1808 */ 1809 bcopy(p, bp, sizeof( *p ) ); 1810 pipe_bp->delay = (pipe_bp->delay * 1000) / hz ; 1811 /* 1812 * XXX the following is a hack based on ->next being the 1813 * first field in dn_pipe and dn_flow_set. The correct 1814 * solution would be to move the dn_flow_set to the beginning 1815 * of struct dn_pipe. 1816 */ 1817 pipe_bp->next = (struct dn_pipe *)DN_IS_PIPE ; 1818 /* clean pointers */ 1819 pipe_bp->head = pipe_bp->tail = NULL ; 1820 pipe_bp->fs.next = NULL ; 1821 pipe_bp->fs.pipe = NULL ; 1822 pipe_bp->fs.rq = NULL ; 1823 1824 bp += sizeof( *p ) ; 1825 bp = dn_copy_set( &(p->fs), bp ); 1826 } 1827 for (set = all_flow_sets ; set ; set = set->next ) { 1828 struct dn_flow_set *fs_bp = (struct dn_flow_set *)bp ; 1829 bcopy(set, bp, sizeof( *set ) ); 1830 /* XXX same hack as above */ 1831 fs_bp->next = (struct dn_flow_set *)DN_IS_QUEUE ; 1832 fs_bp->pipe = NULL ; 1833 fs_bp->rq = NULL ; 1834 bp += sizeof( *set ) ; 1835 bp = dn_copy_set( set, bp ); 1836 } 1837 splx(s); 1838 error = sooptcopyout(sopt, buf, size); 1839 free(buf, M_TEMP); 1840 return error ; 1841 } 1842 1843 /* 1844 * Handler for the various dummynet socket options (get, flush, config, del) 1845 */ 1846 static int 1847 ip_dn_ctl(struct sockopt *sopt) 1848 { 1849 int error = 0 ; 1850 struct dn_pipe *p, tmp_pipe; 1851 1852 /* Disallow sets in really-really secure mode. */ 1853 if (sopt->sopt_dir == SOPT_SET) { 1854 #if defined(__FreeBSD__) && __FreeBSD_version >= 500034 1855 error = securelevel_ge(sopt->sopt_td->td_ucred, 3); 1856 if (error) 1857 return (error); 1858 #else 1859 if (securelevel >= 3) 1860 return (EPERM); 1861 #endif 1862 } 1863 1864 switch (sopt->sopt_name) { 1865 default : 1866 printf("ip_dn_ctl -- unknown option %d", sopt->sopt_name); 1867 return EINVAL ; 1868 1869 case IP_DUMMYNET_GET : 1870 error = dummynet_get(sopt); 1871 break ; 1872 1873 case IP_DUMMYNET_FLUSH : 1874 dummynet_flush() ; 1875 break ; 1876 1877 case IP_DUMMYNET_CONFIGURE : 1878 p = &tmp_pipe ; 1879 error = sooptcopyin(sopt, p, sizeof *p, sizeof *p); 1880 if (error) 1881 break ; 1882 error = config_pipe(p); 1883 break ; 1884 1885 case IP_DUMMYNET_DEL : /* remove a pipe or queue */ 1886 p = &tmp_pipe ; 1887 error = sooptcopyin(sopt, p, sizeof *p, sizeof *p); 1888 if (error) 1889 break ; 1890 1891 error = delete_pipe(p); 1892 break ; 1893 } 1894 return error ; 1895 } 1896 1897 static void 1898 ip_dn_init(void) 1899 { 1900 printf("DUMMYNET initialized (011031)\n"); 1901 all_pipes = NULL ; 1902 all_flow_sets = NULL ; 1903 ready_heap.size = ready_heap.elements = 0 ; 1904 ready_heap.offset = 0 ; 1905 1906 wfq_ready_heap.size = wfq_ready_heap.elements = 0 ; 1907 wfq_ready_heap.offset = 0 ; 1908 1909 extract_heap.size = extract_heap.elements = 0 ; 1910 extract_heap.offset = 0 ; 1911 ip_dn_ctl_ptr = ip_dn_ctl; 1912 ip_dn_io_ptr = dummynet_io; 1913 ip_dn_ruledel_ptr = dn_rule_delete; 1914 bzero(&dn_timeout, sizeof(struct callout_handle)); 1915 dn_timeout = timeout(dummynet, NULL, 1); 1916 } 1917 1918 static int 1919 dummynet_modevent(module_t mod, int type, void *data) 1920 { 1921 int s; 1922 switch (type) { 1923 case MOD_LOAD: 1924 s = splimp(); 1925 if (DUMMYNET_LOADED) { 1926 splx(s); 1927 printf("DUMMYNET already loaded\n"); 1928 return EEXIST ; 1929 } 1930 ip_dn_init(); 1931 splx(s); 1932 break; 1933 1934 case MOD_UNLOAD: 1935 #if !defined(KLD_MODULE) 1936 printf("dummynet statically compiled, cannot unload\n"); 1937 return EINVAL ; 1938 #else 1939 s = splimp(); 1940 untimeout(dummynet, NULL, dn_timeout); 1941 dummynet_flush(); 1942 ip_dn_ctl_ptr = NULL; 1943 ip_dn_io_ptr = NULL; 1944 ip_dn_ruledel_ptr = NULL; 1945 splx(s); 1946 #endif 1947 break ; 1948 default: 1949 break ; 1950 } 1951 return 0 ; 1952 } 1953 1954 static moduledata_t dummynet_mod = { 1955 "dummynet", 1956 dummynet_modevent, 1957 NULL 1958 }; 1959 DECLARE_MODULE(dummynet, dummynet_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 1960 MODULE_DEPEND(dummynet, ipfw, 1, 1, 1); 1961 MODULE_VERSION(dummynet, 1); 1962