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 1.17 2006/01/14 11:05:17 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/oldbridge/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 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 /* which consumes the mbuf */ 469 ether_demux(NULL, eh, m); 470 } 471 } 472 break ; 473 case DN_TO_ETH_OUT: 474 ether_output_frame(pkt->ifp, (struct mbuf *)pkt); 475 break; 476 477 default: 478 printf("dummynet: bad switch %d!\n", pkt->dn_dir); 479 m_freem(pkt->dn_m); 480 break ; 481 } 482 free(pkt, M_DUMMYNET); 483 } 484 /* if there are leftover packets, put into the heap for next event */ 485 if ( (pkt = pipe->head) ) 486 heap_insert(&extract_heap, pkt->output_time, pipe ) ; 487 /* XXX should check errors on heap_insert, by draining the 488 * whole pipe p and hoping in the future we are more successful 489 */ 490 } 491 492 /* 493 * the following macro computes how many ticks we have to wait 494 * before being able to transmit a packet. The credit is taken from 495 * either a pipe (WF2Q) or a flow_queue (per-flow queueing) 496 */ 497 #define SET_TICKS(pkt, q, p) \ 498 (pkt->dn_m->m_pkthdr.len*8*hz - (q)->numbytes + p->bandwidth - 1 ) / \ 499 p->bandwidth ; 500 501 /* 502 * extract pkt from queue, compute output time (could be now) 503 * and put into delay line (p_queue) 504 */ 505 static void 506 move_pkt(struct dn_pkt *pkt, struct dn_flow_queue *q, 507 struct dn_pipe *p, int len) 508 { 509 q->head = DN_NEXT(pkt) ; 510 q->len-- ; 511 q->len_bytes -= len ; 512 513 pkt->output_time = curr_time + p->delay ; 514 515 if (p->head == NULL) 516 p->head = pkt; 517 else 518 DN_NEXT_NC(p->tail) = (struct mbuf *)pkt; 519 p->tail = pkt; 520 DN_NEXT_NC(p->tail) = NULL; 521 } 522 523 /* 524 * ready_event() is invoked every time the queue must enter the 525 * scheduler, either because the first packet arrives, or because 526 * a previously scheduled event fired. 527 * On invokation, drain as many pkts as possible (could be 0) and then 528 * if there are leftover packets reinsert the pkt in the scheduler. 529 */ 530 static void 531 ready_event(struct dn_flow_queue *q) 532 { 533 struct dn_pkt *pkt; 534 struct dn_pipe *p = q->fs->pipe ; 535 int p_was_empty ; 536 537 if (p == NULL) { 538 printf("ready_event- pipe is gone\n"); 539 return ; 540 } 541 p_was_empty = (p->head == NULL) ; 542 543 /* 544 * schedule fixed-rate queues linked to this pipe: 545 * Account for the bw accumulated since last scheduling, then 546 * drain as many pkts as allowed by q->numbytes and move to 547 * the delay line (in p) computing output time. 548 * bandwidth==0 (no limit) means we can drain the whole queue, 549 * setting len_scaled = 0 does the job. 550 */ 551 q->numbytes += ( curr_time - q->sched_time ) * p->bandwidth; 552 while ( (pkt = q->head) != NULL ) { 553 int len = pkt->dn_m->m_pkthdr.len; 554 int len_scaled = p->bandwidth ? len*8*hz : 0 ; 555 if (len_scaled > q->numbytes ) 556 break ; 557 q->numbytes -= len_scaled ; 558 move_pkt(pkt, q, p, len); 559 } 560 /* 561 * If we have more packets queued, schedule next ready event 562 * (can only occur when bandwidth != 0, otherwise we would have 563 * flushed the whole queue in the previous loop). 564 * To this purpose we record the current time and compute how many 565 * ticks to go for the finish time of the packet. 566 */ 567 if ( (pkt = q->head) != NULL ) { /* this implies bandwidth != 0 */ 568 dn_key t = SET_TICKS(pkt, q, p); /* ticks i have to wait */ 569 q->sched_time = curr_time ; 570 heap_insert(&ready_heap, curr_time + t, (void *)q ); 571 /* XXX should check errors on heap_insert, and drain the whole 572 * queue on error hoping next time we are luckier. 573 */ 574 } else { /* RED needs to know when the queue becomes empty */ 575 q->q_time = curr_time; 576 q->numbytes = 0; 577 } 578 /* 579 * If the delay line was empty call transmit_event(p) now. 580 * Otherwise, the scheduler will take care of it. 581 */ 582 if (p_was_empty) 583 transmit_event(p); 584 } 585 586 /* 587 * Called when we can transmit packets on WF2Q queues. Take pkts out of 588 * the queues at their start time, and enqueue into the delay line. 589 * Packets are drained until p->numbytes < 0. As long as 590 * len_scaled >= p->numbytes, the packet goes into the delay line 591 * with a deadline p->delay. For the last packet, if p->numbytes<0, 592 * there is an additional delay. 593 */ 594 static void 595 ready_event_wfq(struct dn_pipe *p) 596 { 597 int p_was_empty = (p->head == NULL) ; 598 struct dn_heap *sch = &(p->scheduler_heap); 599 struct dn_heap *neh = &(p->not_eligible_heap) ; 600 601 if (p->if_name[0] == 0) /* tx clock is simulated */ 602 p->numbytes += ( curr_time - p->sched_time ) * p->bandwidth; 603 else { /* tx clock is for real, the ifq must be empty or this is a NOP */ 604 if (p->ifp && p->ifp->if_snd.ifq_head != NULL) 605 return ; 606 else { 607 DEB(printf("pipe %d ready from %s --\n", 608 p->pipe_nr, p->if_name);) 609 } 610 } 611 612 /* 613 * While we have backlogged traffic AND credit, we need to do 614 * something on the queue. 615 */ 616 while ( p->numbytes >=0 && (sch->elements>0 || neh->elements >0) ) { 617 if (sch->elements > 0) { /* have some eligible pkts to send out */ 618 struct dn_flow_queue *q = sch->p[0].object ; 619 struct dn_pkt *pkt = q->head; 620 struct dn_flow_set *fs = q->fs; 621 u_int64_t len = pkt->dn_m->m_pkthdr.len; 622 int len_scaled = p->bandwidth ? len*8*hz : 0 ; 623 624 heap_extract(sch, NULL); /* remove queue from heap */ 625 p->numbytes -= len_scaled ; 626 move_pkt(pkt, q, p, len); 627 628 p->V += (len<<MY_M) / p->sum ; /* update V */ 629 q->S = q->F ; /* update start time */ 630 if (q->len == 0) { /* Flow not backlogged any more */ 631 fs->backlogged-- ; 632 heap_insert(&(p->idle_heap), q->F, q); 633 } else { /* still backlogged */ 634 /* 635 * update F and position in backlogged queue, then 636 * put flow in not_eligible_heap (we will fix this later). 637 */ 638 len = (q->head)->dn_m->m_pkthdr.len; 639 q->F += (len<<MY_M)/(u_int64_t) fs->weight ; 640 if (DN_KEY_LEQ(q->S, p->V)) 641 heap_insert(neh, q->S, q); 642 else 643 heap_insert(sch, q->F, q); 644 } 645 } 646 /* 647 * now compute V = max(V, min(S_i)). Remember that all elements in sch 648 * have by definition S_i <= V so if sch is not empty, V is surely 649 * the max and we must not update it. Conversely, if sch is empty 650 * we only need to look at neh. 651 */ 652 if (sch->elements == 0 && neh->elements > 0) 653 p->V = MAX64 ( p->V, neh->p[0].key ); 654 /* move from neh to sch any packets that have become eligible */ 655 while (neh->elements > 0 && DN_KEY_LEQ(neh->p[0].key, p->V) ) { 656 struct dn_flow_queue *q = neh->p[0].object ; 657 heap_extract(neh, NULL); 658 heap_insert(sch, q->F, q); 659 } 660 661 if (p->if_name[0] != '\0') {/* tx clock is from a real thing */ 662 p->numbytes = -1 ; /* mark not ready for I/O */ 663 break ; 664 } 665 } 666 if (sch->elements == 0 && neh->elements == 0 && p->numbytes >= 0 667 && p->idle_heap.elements > 0) { 668 /* 669 * no traffic and no events scheduled. We can get rid of idle-heap. 670 */ 671 int i ; 672 673 for (i = 0 ; i < p->idle_heap.elements ; i++) { 674 struct dn_flow_queue *q = p->idle_heap.p[i].object ; 675 676 q->F = 0 ; 677 q->S = q->F + 1 ; 678 } 679 p->sum = 0 ; 680 p->V = 0 ; 681 p->idle_heap.elements = 0 ; 682 } 683 /* 684 * If we are getting clocks from dummynet (not a real interface) and 685 * If we are under credit, schedule the next ready event. 686 * Also fix the delivery time of the last packet. 687 */ 688 if (p->if_name[0]==0 && p->numbytes < 0) { /* this implies bandwidth >0 */ 689 dn_key t=0 ; /* number of ticks i have to wait */ 690 691 if (p->bandwidth > 0) 692 t = ( p->bandwidth -1 - p->numbytes) / p->bandwidth ; 693 p->tail->output_time += t ; 694 p->sched_time = curr_time ; 695 heap_insert(&wfq_ready_heap, curr_time + t, (void *)p); 696 /* XXX should check errors on heap_insert, and drain the whole 697 * queue on error hoping next time we are luckier. 698 */ 699 } 700 /* 701 * If the delay line was empty call transmit_event(p) now. 702 * Otherwise, the scheduler will take care of it. 703 */ 704 if (p_was_empty) 705 transmit_event(p); 706 } 707 708 /* 709 * This is called once per tick, or HZ times per second. It is used to 710 * increment the current tick counter and schedule expired events. 711 */ 712 static void 713 dummynet(void * __unused unused) 714 { 715 void *p ; /* generic parameter to handler */ 716 struct dn_heap *h ; 717 struct dn_heap *heaps[3]; 718 int i; 719 struct dn_pipe *pe ; 720 721 heaps[0] = &ready_heap ; /* fixed-rate queues */ 722 heaps[1] = &wfq_ready_heap ; /* wfq queues */ 723 heaps[2] = &extract_heap ; /* delay line */ 724 crit_enter(); /* see note on top, splnet() is not enough */ 725 curr_time++ ; 726 for (i=0; i < 3 ; i++) { 727 h = heaps[i]; 728 while (h->elements > 0 && DN_KEY_LEQ(h->p[0].key, curr_time) ) { 729 DDB(if (h->p[0].key > curr_time) 730 printf("-- dummynet: warning, heap %d is %d ticks late\n", 731 i, (int)(curr_time - h->p[0].key));) 732 p = h->p[0].object ; /* store a copy before heap_extract */ 733 heap_extract(h, NULL); /* need to extract before processing */ 734 if (i == 0) 735 ready_event(p) ; 736 else if (i == 1) { 737 struct dn_pipe *pipe = p; 738 if (pipe->if_name[0] != '\0') 739 printf("*** bad ready_event_wfq for pipe %s\n", 740 pipe->if_name); 741 else 742 ready_event_wfq(p) ; 743 } else 744 transmit_event(p); 745 } 746 } 747 /* sweep pipes trying to expire idle flow_queues */ 748 for (pe = all_pipes; pe ; pe = pe->next ) 749 if (pe->idle_heap.elements > 0 && 750 DN_KEY_LT(pe->idle_heap.p[0].key, pe->V) ) { 751 struct dn_flow_queue *q = pe->idle_heap.p[0].object ; 752 753 heap_extract(&(pe->idle_heap), NULL); 754 q->S = q->F + 1 ; /* mark timestamp as invalid */ 755 pe->sum -= q->fs->weight ; 756 } 757 crit_exit(); 758 callout_reset(&dn_timeout, 1, dummynet, NULL); 759 } 760 761 /* 762 * called by an interface when tx_rdy occurs. 763 */ 764 int 765 if_tx_rdy(struct ifnet *ifp) 766 { 767 struct dn_pipe *p; 768 769 for (p = all_pipes; p ; p = p->next ) 770 if (p->ifp == ifp) 771 break ; 772 if (p == NULL) { 773 for (p = all_pipes; p ; p = p->next ) 774 if (!strcmp(p->if_name, ifp->if_xname) ) { 775 p->ifp = ifp ; 776 DEB(printf("++ tx rdy from %s (now found)\n", ifp->if_xname);) 777 break ; 778 } 779 } 780 if (p != NULL) { 781 DEB(printf("++ tx rdy from %s - qlen %d\n", ifp->if_xname, 782 ifp->if_snd.ifq_len);) 783 p->numbytes = 0 ; /* mark ready for I/O */ 784 ready_event_wfq(p); 785 } 786 return 0; 787 } 788 789 /* 790 * Unconditionally expire empty queues in case of shortage. 791 * Returns the number of queues freed. 792 */ 793 static int 794 expire_queues(struct dn_flow_set *fs) 795 { 796 struct dn_flow_queue *q, *prev ; 797 int i, initial_elements = fs->rq_elements ; 798 799 if (fs->last_expired == time_second) 800 return 0 ; 801 fs->last_expired = time_second ; 802 for (i = 0 ; i <= fs->rq_size ; i++) /* last one is overflow */ 803 for (prev=NULL, q = fs->rq[i] ; q != NULL ; ) 804 if (q->head != NULL || q->S != q->F+1) { 805 prev = q ; 806 q = q->next ; 807 } else { /* entry is idle, expire it */ 808 struct dn_flow_queue *old_q = q ; 809 810 if (prev != NULL) 811 prev->next = q = q->next ; 812 else 813 fs->rq[i] = q = q->next ; 814 fs->rq_elements-- ; 815 free(old_q, M_DUMMYNET); 816 } 817 return initial_elements - fs->rq_elements ; 818 } 819 820 /* 821 * If room, create a new queue and put at head of slot i; 822 * otherwise, create or use the default queue. 823 */ 824 static struct dn_flow_queue * 825 create_queue(struct dn_flow_set *fs, int i) 826 { 827 struct dn_flow_queue *q ; 828 829 if (fs->rq_elements > fs->rq_size * dn_max_ratio && 830 expire_queues(fs) == 0) { 831 /* 832 * No way to get room, use or create overflow queue. 833 */ 834 i = fs->rq_size ; 835 if ( fs->rq[i] != NULL ) 836 return fs->rq[i] ; 837 } 838 q = malloc(sizeof(*q), M_DUMMYNET, M_WAITOK | M_ZERO); 839 q->fs = fs ; 840 q->hash_slot = i ; 841 q->next = fs->rq[i] ; 842 q->S = q->F + 1; /* hack - mark timestamp as invalid */ 843 fs->rq[i] = q ; 844 fs->rq_elements++ ; 845 return q ; 846 } 847 848 /* 849 * Given a flow_set and a pkt in last_pkt, find a matching queue 850 * after appropriate masking. The queue is moved to front 851 * so that further searches take less time. 852 */ 853 static struct dn_flow_queue * 854 find_queue(struct dn_flow_set *fs, struct ipfw_flow_id *id) 855 { 856 int i = 0 ; /* we need i and q for new allocations */ 857 struct dn_flow_queue *q, *prev; 858 859 if ( !(fs->flags_fs & DN_HAVE_FLOW_MASK) ) 860 q = fs->rq[0] ; 861 else { 862 /* first, do the masking */ 863 id->dst_ip &= fs->flow_mask.dst_ip ; 864 id->src_ip &= fs->flow_mask.src_ip ; 865 id->dst_port &= fs->flow_mask.dst_port ; 866 id->src_port &= fs->flow_mask.src_port ; 867 id->proto &= fs->flow_mask.proto ; 868 id->flags = 0 ; /* we don't care about this one */ 869 /* then, hash function */ 870 i = ( (id->dst_ip) & 0xffff ) ^ 871 ( (id->dst_ip >> 15) & 0xffff ) ^ 872 ( (id->src_ip << 1) & 0xffff ) ^ 873 ( (id->src_ip >> 16 ) & 0xffff ) ^ 874 (id->dst_port << 1) ^ (id->src_port) ^ 875 (id->proto ); 876 i = i % fs->rq_size ; 877 /* finally, scan the current list for a match */ 878 searches++ ; 879 for (prev=NULL, q = fs->rq[i] ; q ; ) { 880 search_steps++; 881 if (id->dst_ip == q->id.dst_ip && 882 id->src_ip == q->id.src_ip && 883 id->dst_port == q->id.dst_port && 884 id->src_port == q->id.src_port && 885 id->proto == q->id.proto && 886 id->flags == q->id.flags) 887 break ; /* found */ 888 else if (pipe_expire && q->head == NULL && q->S == q->F+1 ) { 889 /* entry is idle and not in any heap, expire it */ 890 struct dn_flow_queue *old_q = q ; 891 892 if (prev != NULL) 893 prev->next = q = q->next ; 894 else 895 fs->rq[i] = q = q->next ; 896 fs->rq_elements-- ; 897 free(old_q, M_DUMMYNET); 898 continue ; 899 } 900 prev = q ; 901 q = q->next ; 902 } 903 if (q && prev != NULL) { /* found and not in front */ 904 prev->next = q->next ; 905 q->next = fs->rq[i] ; 906 fs->rq[i] = q ; 907 } 908 } 909 if (q == NULL) { /* no match, need to allocate a new entry */ 910 q = create_queue(fs, i); 911 if (q != NULL) 912 q->id = *id ; 913 } 914 return q ; 915 } 916 917 static int 918 red_drops(struct dn_flow_set *fs, struct dn_flow_queue *q, int len) 919 { 920 /* 921 * RED algorithm 922 * 923 * RED calculates the average queue size (avg) using a low-pass filter 924 * with an exponential weighted (w_q) moving average: 925 * avg <- (1-w_q) * avg + w_q * q_size 926 * where q_size is the queue length (measured in bytes or * packets). 927 * 928 * If q_size == 0, we compute the idle time for the link, and set 929 * avg = (1 - w_q)^(idle/s) 930 * where s is the time needed for transmitting a medium-sized packet. 931 * 932 * Now, if avg < min_th the packet is enqueued. 933 * If avg > max_th the packet is dropped. Otherwise, the packet is 934 * dropped with probability P function of avg. 935 * 936 */ 937 938 int64_t p_b = 0; 939 /* queue in bytes or packets ? */ 940 u_int q_size = (fs->flags_fs & DN_QSIZE_IS_BYTES) ? q->len_bytes : q->len; 941 942 DEB(printf("\n%d q: %2u ", (int) curr_time, q_size);) 943 944 /* average queue size estimation */ 945 if (q_size != 0) { 946 /* 947 * queue is not empty, avg <- avg + (q_size - avg) * w_q 948 */ 949 int diff = SCALE(q_size) - q->avg; 950 int64_t v = SCALE_MUL((int64_t) diff, (int64_t) fs->w_q); 951 952 q->avg += (int) v; 953 } else { 954 /* 955 * queue is empty, find for how long the queue has been 956 * empty and use a lookup table for computing 957 * (1 - * w_q)^(idle_time/s) where s is the time to send a 958 * (small) packet. 959 * XXX check wraps... 960 */ 961 if (q->avg) { 962 u_int t = (curr_time - q->q_time) / fs->lookup_step; 963 964 q->avg = (t < fs->lookup_depth) ? 965 SCALE_MUL(q->avg, fs->w_q_lookup[t]) : 0; 966 } 967 } 968 DEB(printf("avg: %u ", SCALE_VAL(q->avg));) 969 970 /* should i drop ? */ 971 972 if (q->avg < fs->min_th) { 973 q->count = -1; 974 return 0; /* accept packet ; */ 975 } 976 if (q->avg >= fs->max_th) { /* average queue >= max threshold */ 977 if (fs->flags_fs & DN_IS_GENTLE_RED) { 978 /* 979 * According to Gentle-RED, if avg is greater than max_th the 980 * packet is dropped with a probability 981 * p_b = c_3 * avg - c_4 982 * where c_3 = (1 - max_p) / max_th, and c_4 = 1 - 2 * max_p 983 */ 984 p_b = SCALE_MUL((int64_t) fs->c_3, (int64_t) q->avg) - fs->c_4; 985 } else { 986 q->count = -1; 987 printf("- drop"); 988 return 1 ; 989 } 990 } else if (q->avg > fs->min_th) { 991 /* 992 * we compute p_b using the linear dropping function p_b = c_1 * 993 * avg - c_2, where c_1 = max_p / (max_th - min_th), and c_2 = 994 * max_p * min_th / (max_th - min_th) 995 */ 996 p_b = SCALE_MUL((int64_t) fs->c_1, (int64_t) q->avg) - fs->c_2; 997 } 998 if (fs->flags_fs & DN_QSIZE_IS_BYTES) 999 p_b = (p_b * len) / fs->max_pkt_size; 1000 if (++q->count == 0) 1001 q->random = random() & 0xffff; 1002 else { 1003 /* 1004 * q->count counts packets arrived since last drop, so a greater 1005 * value of q->count means a greater packet drop probability. 1006 */ 1007 if (SCALE_MUL(p_b, SCALE((int64_t) q->count)) > q->random) { 1008 q->count = 0; 1009 DEB(printf("- red drop");) 1010 /* after a drop we calculate a new random value */ 1011 q->random = random() & 0xffff; 1012 return 1; /* drop */ 1013 } 1014 } 1015 /* end of RED algorithm */ 1016 return 0 ; /* accept */ 1017 } 1018 1019 static __inline 1020 struct dn_flow_set * 1021 locate_flowset(int pipe_nr, struct ip_fw *rule) 1022 { 1023 #if IPFW2 1024 struct dn_flow_set *fs; 1025 ipfw_insn *cmd = rule->cmd + rule->act_ofs; 1026 1027 if (cmd->opcode == O_LOG) 1028 cmd += F_LEN(cmd); 1029 fs = ((ipfw_insn_pipe *)cmd)->pipe_ptr; 1030 1031 if (fs != NULL) 1032 return fs; 1033 1034 if (cmd->opcode == O_QUEUE) 1035 #else /* !IPFW2 */ 1036 struct dn_flow_set *fs = NULL ; 1037 1038 if ( (rule->fw_flg & IP_FW_F_COMMAND) == IP_FW_F_QUEUE ) 1039 #endif /* !IPFW2 */ 1040 for (fs=all_flow_sets; fs && fs->fs_nr != pipe_nr; fs=fs->next) 1041 ; 1042 else { 1043 struct dn_pipe *p1; 1044 for (p1 = all_pipes; p1 && p1->pipe_nr != pipe_nr; p1 = p1->next) 1045 ; 1046 if (p1 != NULL) 1047 fs = &(p1->fs) ; 1048 } 1049 /* record for the future */ 1050 #if IPFW2 1051 ((ipfw_insn_pipe *)cmd)->pipe_ptr = fs; 1052 #else 1053 if (fs != NULL) 1054 rule->pipe_ptr = fs; 1055 #endif 1056 return fs ; 1057 } 1058 1059 /* 1060 * dummynet hook for packets. Below 'pipe' is a pipe or a queue 1061 * depending on whether WF2Q or fixed bw is used. 1062 * 1063 * pipe_nr pipe or queue the packet is destined for. 1064 * dir where shall we send the packet after dummynet. 1065 * m the mbuf with the packet 1066 * ifp the 'ifp' parameter from the caller. 1067 * NULL in ip_input, destination interface in ip_output, 1068 * real_dst in bdg_forward 1069 * ro route parameter (only used in ip_output, NULL otherwise) 1070 * dst destination address, only used by ip_output 1071 * rule matching rule, in case of multiple passes 1072 * flags flags from the caller, only used in ip_output 1073 * 1074 */ 1075 static int 1076 dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa) 1077 { 1078 struct dn_pkt *pkt; 1079 struct dn_flow_set *fs; 1080 struct dn_pipe *pipe ; 1081 u_int64_t len = m->m_pkthdr.len ; 1082 struct dn_flow_queue *q = NULL ; 1083 int is_pipe; 1084 1085 crit_enter(); 1086 #if IPFW2 1087 ipfw_insn *cmd = fwa->rule->cmd + fwa->rule->act_ofs; 1088 1089 if (cmd->opcode == O_LOG) 1090 cmd += F_LEN(cmd); 1091 is_pipe = (cmd->opcode == O_PIPE); 1092 #else 1093 is_pipe = (fwa->rule->fw_flg & IP_FW_F_COMMAND) == IP_FW_F_PIPE; 1094 #endif 1095 1096 pipe_nr &= 0xffff ; 1097 1098 /* 1099 * this is a dummynet rule, so we expect a O_PIPE or O_QUEUE rule 1100 */ 1101 fs = locate_flowset(pipe_nr, fwa->rule); 1102 if (fs == NULL) 1103 goto dropit ; /* this queue/pipe does not exist! */ 1104 pipe = fs->pipe ; 1105 if (pipe == NULL) { /* must be a queue, try find a matching pipe */ 1106 for (pipe = all_pipes; pipe && pipe->pipe_nr != fs->parent_nr; 1107 pipe = pipe->next) 1108 ; 1109 if (pipe != NULL) 1110 fs->pipe = pipe ; 1111 else { 1112 printf("No pipe %d for queue %d, drop pkt\n", 1113 fs->parent_nr, fs->fs_nr); 1114 goto dropit ; 1115 } 1116 } 1117 q = find_queue(fs, &(fwa->f_id)); 1118 if ( q == NULL ) 1119 goto dropit ; /* cannot allocate queue */ 1120 /* 1121 * update statistics, then check reasons to drop pkt 1122 */ 1123 q->tot_bytes += len ; 1124 q->tot_pkts++ ; 1125 if ( fs->plr && random() < fs->plr ) 1126 goto dropit ; /* random pkt drop */ 1127 if ( fs->flags_fs & DN_QSIZE_IS_BYTES) { 1128 if (q->len_bytes > fs->qsize) 1129 goto dropit ; /* queue size overflow */ 1130 } else { 1131 if (q->len >= fs->qsize) 1132 goto dropit ; /* queue count overflow */ 1133 } 1134 if ( fs->flags_fs & DN_IS_RED && red_drops(fs, q, len) ) 1135 goto dropit ; 1136 1137 /* XXX expensive to zero, see if we can remove it*/ 1138 pkt = malloc(sizeof (*pkt), M_DUMMYNET, M_INTWAIT | M_ZERO | M_NULLOK); 1139 if (pkt == NULL) 1140 goto dropit; /* cannot allocate packet header */ 1141 1142 /* ok, i can handle the pkt now... */ 1143 /* build and enqueue packet + parameters */ 1144 pkt->hdr.mh_type = MT_TAG; 1145 pkt->hdr.mh_flags = PACKET_TAG_DUMMYNET; 1146 pkt->rule = fwa->rule ; 1147 DN_NEXT_NC(pkt) = NULL; 1148 pkt->dn_m = m; 1149 pkt->dn_dir = dir ; 1150 1151 pkt->ifp = fwa->oif; 1152 if (dir == DN_TO_IP_OUT) { 1153 /* 1154 * We need to copy *ro because for ICMP pkts (and maybe others) 1155 * the caller passed a pointer into the stack; dst might also be 1156 * a pointer into *ro so it needs to be updated. 1157 */ 1158 pkt->ro = *(fwa->ro); 1159 if (fwa->ro->ro_rt) 1160 fwa->ro->ro_rt->rt_refcnt++ ; 1161 if (fwa->dst == (struct sockaddr_in *)&fwa->ro->ro_dst) /* dst points into ro */ 1162 fwa->dst = (struct sockaddr_in *)&(pkt->ro.ro_dst) ; 1163 1164 pkt->dn_dst = fwa->dst; 1165 pkt->flags = fwa->flags; 1166 } 1167 if (q->head == NULL) 1168 q->head = pkt; 1169 else 1170 DN_NEXT_NC(q->tail) = (struct mbuf *)pkt; 1171 q->tail = pkt; 1172 q->len++; 1173 q->len_bytes += len ; 1174 1175 if ( q->head != pkt ) /* flow was not idle, we are done */ 1176 goto done; 1177 /* 1178 * If we reach this point the flow was previously idle, so we need 1179 * to schedule it. This involves different actions for fixed-rate or 1180 * WF2Q queues. 1181 */ 1182 if (is_pipe) { 1183 /* 1184 * Fixed-rate queue: just insert into the ready_heap. 1185 */ 1186 dn_key t = 0 ; 1187 if (pipe->bandwidth) 1188 t = SET_TICKS(pkt, q, pipe); 1189 q->sched_time = curr_time ; 1190 if (t == 0) /* must process it now */ 1191 ready_event( q ); 1192 else 1193 heap_insert(&ready_heap, curr_time + t , q ); 1194 } else { 1195 /* 1196 * WF2Q. First, compute start time S: if the flow was idle (S=F+1) 1197 * set S to the virtual time V for the controlling pipe, and update 1198 * the sum of weights for the pipe; otherwise, remove flow from 1199 * idle_heap and set S to max(F,V). 1200 * Second, compute finish time F = S + len/weight. 1201 * Third, if pipe was idle, update V=max(S, V). 1202 * Fourth, count one more backlogged flow. 1203 */ 1204 if (DN_KEY_GT(q->S, q->F)) { /* means timestamps are invalid */ 1205 q->S = pipe->V ; 1206 pipe->sum += fs->weight ; /* add weight of new queue */ 1207 } else { 1208 heap_extract(&(pipe->idle_heap), q); 1209 q->S = MAX64(q->F, pipe->V ) ; 1210 } 1211 q->F = q->S + ( len<<MY_M )/(u_int64_t) fs->weight; 1212 1213 if (pipe->not_eligible_heap.elements == 0 && 1214 pipe->scheduler_heap.elements == 0) 1215 pipe->V = MAX64 ( q->S, pipe->V ); 1216 fs->backlogged++ ; 1217 /* 1218 * Look at eligibility. A flow is not eligibile if S>V (when 1219 * this happens, it means that there is some other flow already 1220 * scheduled for the same pipe, so the scheduler_heap cannot be 1221 * empty). If the flow is not eligible we just store it in the 1222 * not_eligible_heap. Otherwise, we store in the scheduler_heap 1223 * and possibly invoke ready_event_wfq() right now if there is 1224 * leftover credit. 1225 * Note that for all flows in scheduler_heap (SCH), S_i <= V, 1226 * and for all flows in not_eligible_heap (NEH), S_i > V . 1227 * So when we need to compute max( V, min(S_i) ) forall i in SCH+NEH, 1228 * we only need to look into NEH. 1229 */ 1230 if (DN_KEY_GT(q->S, pipe->V) ) { /* not eligible */ 1231 if (pipe->scheduler_heap.elements == 0) 1232 printf("++ ouch! not eligible but empty scheduler!\n"); 1233 heap_insert(&(pipe->not_eligible_heap), q->S, q); 1234 } else { 1235 heap_insert(&(pipe->scheduler_heap), q->F, q); 1236 if (pipe->numbytes >= 0) { /* pipe is idle */ 1237 if (pipe->scheduler_heap.elements != 1) 1238 printf("*** OUCH! pipe should have been idle!\n"); 1239 DEB(printf("Waking up pipe %d at %d\n", 1240 pipe->pipe_nr, (int)(q->F >> MY_M)); ) 1241 pipe->sched_time = curr_time ; 1242 ready_event_wfq(pipe); 1243 } 1244 } 1245 } 1246 done: 1247 crit_exit(); 1248 return 0; 1249 1250 dropit: 1251 crit_exit(); 1252 if (q) 1253 q->drops++ ; 1254 m_freem(m); 1255 return ( (fs && (fs->flags_fs & DN_NOERROR)) ? 0 : ENOBUFS); 1256 } 1257 1258 /* 1259 * Below, the rt_unref is only needed when (pkt->dn_dir == DN_TO_IP_OUT) 1260 * Doing this would probably save us the initial bzero of dn_pkt 1261 */ 1262 #define DN_FREE_PKT(pkt) { \ 1263 struct dn_pkt *n = pkt ; \ 1264 rt_unref ( n->ro.ro_rt ) ; \ 1265 m_freem(n->dn_m); \ 1266 pkt = DN_NEXT(n) ; \ 1267 free(n, M_DUMMYNET) ; } 1268 1269 /* 1270 * Dispose all packets and flow_queues on a flow_set. 1271 * If all=1, also remove red lookup table and other storage, 1272 * including the descriptor itself. 1273 * For the one in dn_pipe MUST also cleanup ready_heap... 1274 */ 1275 static void 1276 purge_flow_set(struct dn_flow_set *fs, int all) 1277 { 1278 struct dn_pkt *pkt ; 1279 struct dn_flow_queue *q, *qn ; 1280 int i ; 1281 1282 for (i = 0 ; i <= fs->rq_size ; i++ ) { 1283 for (q = fs->rq[i] ; q ; q = qn ) { 1284 for (pkt = q->head ; pkt ; ) 1285 DN_FREE_PKT(pkt) ; 1286 qn = q->next ; 1287 free(q, M_DUMMYNET); 1288 } 1289 fs->rq[i] = NULL ; 1290 } 1291 fs->rq_elements = 0 ; 1292 if (all) { 1293 /* RED - free lookup table */ 1294 if (fs->w_q_lookup) 1295 free(fs->w_q_lookup, M_DUMMYNET); 1296 if (fs->rq) 1297 free(fs->rq, M_DUMMYNET); 1298 /* if this fs is not part of a pipe, free it */ 1299 if (fs->pipe && fs != &(fs->pipe->fs) ) 1300 free(fs, M_DUMMYNET); 1301 } 1302 } 1303 1304 /* 1305 * Dispose all packets queued on a pipe (not a flow_set). 1306 * Also free all resources associated to a pipe, which is about 1307 * to be deleted. 1308 */ 1309 static void 1310 purge_pipe(struct dn_pipe *pipe) 1311 { 1312 struct dn_pkt *pkt ; 1313 1314 purge_flow_set( &(pipe->fs), 1 ); 1315 1316 for (pkt = pipe->head ; pkt ; ) 1317 DN_FREE_PKT(pkt) ; 1318 1319 heap_free( &(pipe->scheduler_heap) ); 1320 heap_free( &(pipe->not_eligible_heap) ); 1321 heap_free( &(pipe->idle_heap) ); 1322 } 1323 1324 /* 1325 * Delete all pipes and heaps returning memory. Must also 1326 * remove references from all ipfw rules to all pipes. 1327 */ 1328 static void 1329 dummynet_flush(void) 1330 { 1331 struct dn_pipe *curr_p, *p ; 1332 struct dn_flow_set *fs, *curr_fs; 1333 1334 crit_enter(); 1335 1336 /* remove all references to pipes ...*/ 1337 flush_pipe_ptrs(NULL); 1338 /* prevent future matches... */ 1339 p = all_pipes ; 1340 all_pipes = NULL ; 1341 fs = all_flow_sets ; 1342 all_flow_sets = NULL ; 1343 /* and free heaps so we don't have unwanted events */ 1344 heap_free(&ready_heap); 1345 heap_free(&wfq_ready_heap); 1346 heap_free(&extract_heap); 1347 crit_exit(); 1348 /* 1349 * Now purge all queued pkts and delete all pipes 1350 */ 1351 /* scan and purge all flow_sets. */ 1352 for ( ; fs ; ) { 1353 curr_fs = fs ; 1354 fs = fs->next ; 1355 purge_flow_set(curr_fs, 1); 1356 } 1357 for ( ; p ; ) { 1358 purge_pipe(p); 1359 curr_p = p ; 1360 p = p->next ; 1361 free(curr_p, M_DUMMYNET); 1362 } 1363 } 1364 1365 1366 extern struct ip_fw *ip_fw_default_rule ; 1367 static void 1368 dn_rule_delete_fs(struct dn_flow_set *fs, void *r) 1369 { 1370 int i ; 1371 struct dn_flow_queue *q ; 1372 struct dn_pkt *pkt ; 1373 1374 for (i = 0 ; i <= fs->rq_size ; i++) /* last one is ovflow */ 1375 for (q = fs->rq[i] ; q ; q = q->next ) 1376 for (pkt = q->head ; pkt ; pkt = DN_NEXT(pkt) ) 1377 if (pkt->rule == r) 1378 pkt->rule = ip_fw_default_rule ; 1379 } 1380 /* 1381 * when a firewall rule is deleted, scan all queues and remove the flow-id 1382 * from packets matching this rule. 1383 */ 1384 void 1385 dn_rule_delete(void *r) 1386 { 1387 struct dn_pipe *p ; 1388 struct dn_pkt *pkt ; 1389 struct dn_flow_set *fs ; 1390 1391 /* 1392 * If the rule references a queue (dn_flow_set), then scan 1393 * the flow set, otherwise scan pipes. Should do either, but doing 1394 * both does not harm. 1395 */ 1396 for ( fs = all_flow_sets ; fs ; fs = fs->next ) 1397 dn_rule_delete_fs(fs, r); 1398 for ( p = all_pipes ; p ; p = p->next ) { 1399 fs = &(p->fs) ; 1400 dn_rule_delete_fs(fs, r); 1401 for (pkt = p->head ; pkt ; pkt = DN_NEXT(pkt) ) 1402 if (pkt->rule == r) 1403 pkt->rule = ip_fw_default_rule ; 1404 } 1405 } 1406 1407 /* 1408 * setup RED parameters 1409 */ 1410 static int 1411 config_red(struct dn_flow_set *p, struct dn_flow_set * x) 1412 { 1413 int i; 1414 1415 x->w_q = p->w_q; 1416 x->min_th = SCALE(p->min_th); 1417 x->max_th = SCALE(p->max_th); 1418 x->max_p = p->max_p; 1419 1420 x->c_1 = p->max_p / (p->max_th - p->min_th); 1421 x->c_2 = SCALE_MUL(x->c_1, SCALE(p->min_th)); 1422 if (x->flags_fs & DN_IS_GENTLE_RED) { 1423 x->c_3 = (SCALE(1) - p->max_p) / p->max_th; 1424 x->c_4 = (SCALE(1) - 2 * p->max_p); 1425 } 1426 1427 /* if the lookup table already exist, free and create it again */ 1428 if (x->w_q_lookup) { 1429 free(x->w_q_lookup, M_DUMMYNET); 1430 x->w_q_lookup = NULL ; 1431 } 1432 if (red_lookup_depth == 0) { 1433 printf("\nnet.inet.ip.dummynet.red_lookup_depth must be > 0"); 1434 free(x, M_DUMMYNET); 1435 return EINVAL; 1436 } 1437 x->lookup_depth = red_lookup_depth; 1438 x->w_q_lookup = malloc(x->lookup_depth * sizeof(int), 1439 M_DUMMYNET, M_WAITOK); 1440 1441 /* fill the lookup table with (1 - w_q)^x */ 1442 x->lookup_step = p->lookup_step ; 1443 x->lookup_weight = p->lookup_weight ; 1444 x->w_q_lookup[0] = SCALE(1) - x->w_q; 1445 for (i = 1; i < x->lookup_depth; i++) 1446 x->w_q_lookup[i] = SCALE_MUL(x->w_q_lookup[i - 1], x->lookup_weight); 1447 if (red_avg_pkt_size < 1) 1448 red_avg_pkt_size = 512 ; 1449 x->avg_pkt_size = red_avg_pkt_size ; 1450 if (red_max_pkt_size < 1) 1451 red_max_pkt_size = 1500 ; 1452 x->max_pkt_size = red_max_pkt_size ; 1453 return 0 ; 1454 } 1455 1456 static int 1457 alloc_hash(struct dn_flow_set *x, struct dn_flow_set *pfs) 1458 { 1459 if (x->flags_fs & DN_HAVE_FLOW_MASK) { /* allocate some slots */ 1460 int l = pfs->rq_size; 1461 1462 if (l == 0) 1463 l = dn_hash_size; 1464 if (l < 4) 1465 l = 4; 1466 else if (l > DN_MAX_HASH_SIZE) 1467 l = DN_MAX_HASH_SIZE; 1468 x->rq_size = l; 1469 } else /* one is enough for null mask */ 1470 x->rq_size = 1; 1471 x->rq = malloc((1 + x->rq_size) * sizeof(struct dn_flow_queue *), 1472 M_DUMMYNET, M_WAITOK | M_ZERO); 1473 x->rq_elements = 0; 1474 return 0 ; 1475 } 1476 1477 static void 1478 set_fs_parms(struct dn_flow_set *x, struct dn_flow_set *src) 1479 { 1480 x->flags_fs = src->flags_fs; 1481 x->qsize = src->qsize; 1482 x->plr = src->plr; 1483 x->flow_mask = src->flow_mask; 1484 if (x->flags_fs & DN_QSIZE_IS_BYTES) { 1485 if (x->qsize > 1024*1024) 1486 x->qsize = 1024*1024 ; 1487 } else { 1488 if (x->qsize == 0) 1489 x->qsize = 50 ; 1490 if (x->qsize > 100) 1491 x->qsize = 50 ; 1492 } 1493 /* configuring RED */ 1494 if ( x->flags_fs & DN_IS_RED ) 1495 config_red(src, x) ; /* XXX should check errors */ 1496 } 1497 1498 /* 1499 * setup pipe or queue parameters. 1500 */ 1501 1502 static int 1503 config_pipe(struct dn_pipe *p) 1504 { 1505 int i, s; 1506 struct dn_flow_set *pfs = &(p->fs); 1507 struct dn_flow_queue *q; 1508 1509 /* 1510 * The config program passes parameters as follows: 1511 * bw = bits/second (0 means no limits), 1512 * delay = ms, must be translated into ticks. 1513 * qsize = slots/bytes 1514 */ 1515 p->delay = ( p->delay * hz ) / 1000 ; 1516 /* We need either a pipe number or a flow_set number */ 1517 if (p->pipe_nr == 0 && pfs->fs_nr == 0) 1518 return EINVAL ; 1519 if (p->pipe_nr != 0 && pfs->fs_nr != 0) 1520 return EINVAL ; 1521 if (p->pipe_nr != 0) { /* this is a pipe */ 1522 struct dn_pipe *x, *a, *b; 1523 /* locate pipe */ 1524 for (a = NULL , b = all_pipes ; b && b->pipe_nr < p->pipe_nr ; 1525 a = b , b = b->next) ; 1526 1527 if (b == NULL || b->pipe_nr != p->pipe_nr) { /* new pipe */ 1528 x = malloc(sizeof(struct dn_pipe), M_DUMMYNET, M_WAITOK | M_ZERO); 1529 x->pipe_nr = p->pipe_nr; 1530 x->fs.pipe = x ; 1531 /* idle_heap is the only one from which we extract from the middle. 1532 */ 1533 x->idle_heap.size = x->idle_heap.elements = 0 ; 1534 x->idle_heap.offset=OFFSET_OF(struct dn_flow_queue, heap_pos); 1535 } else { 1536 x = b; 1537 crit_enter(); 1538 /* Flush accumulated credit for all queues */ 1539 for (i = 0; i <= x->fs.rq_size; i++) 1540 for (q = x->fs.rq[i]; q; q = q->next) 1541 q->numbytes = 0; 1542 crit_exit(); 1543 } 1544 1545 crit_enter(); 1546 x->bandwidth = p->bandwidth ; 1547 x->numbytes = 0; /* just in case... */ 1548 bcopy(p->if_name, x->if_name, sizeof(p->if_name) ); 1549 x->ifp = NULL ; /* reset interface ptr */ 1550 x->delay = p->delay ; 1551 set_fs_parms(&(x->fs), pfs); 1552 1553 1554 if ( x->fs.rq == NULL ) { /* a new pipe */ 1555 s = alloc_hash(&(x->fs), pfs) ; 1556 if (s) { 1557 free(x, M_DUMMYNET); 1558 return s ; 1559 } 1560 x->next = b ; 1561 if (a == NULL) 1562 all_pipes = x ; 1563 else 1564 a->next = x ; 1565 } 1566 crit_exit(); 1567 } else { /* config queue */ 1568 struct dn_flow_set *x, *a, *b ; 1569 1570 /* locate flow_set */ 1571 for (a=NULL, b=all_flow_sets ; b && b->fs_nr < pfs->fs_nr ; 1572 a = b , b = b->next) ; 1573 1574 if (b == NULL || b->fs_nr != pfs->fs_nr) { /* new */ 1575 if (pfs->parent_nr == 0) /* need link to a pipe */ 1576 return EINVAL ; 1577 x = malloc(sizeof(struct dn_flow_set), M_DUMMYNET, M_WAITOK|M_ZERO); 1578 x->fs_nr = pfs->fs_nr; 1579 x->parent_nr = pfs->parent_nr; 1580 x->weight = pfs->weight ; 1581 if (x->weight == 0) 1582 x->weight = 1 ; 1583 else if (x->weight > 100) 1584 x->weight = 100 ; 1585 } else { 1586 /* Change parent pipe not allowed; must delete and recreate */ 1587 if (pfs->parent_nr != 0 && b->parent_nr != pfs->parent_nr) 1588 return EINVAL ; 1589 x = b; 1590 } 1591 crit_enter(); 1592 set_fs_parms(x, pfs); 1593 1594 if ( x->rq == NULL ) { /* a new flow_set */ 1595 s = alloc_hash(x, pfs) ; 1596 if (s) { 1597 free(x, M_DUMMYNET); 1598 return s ; 1599 } 1600 x->next = b; 1601 if (a == NULL) 1602 all_flow_sets = x; 1603 else 1604 a->next = x; 1605 } 1606 crit_exit(); 1607 } 1608 return 0 ; 1609 } 1610 1611 /* 1612 * Helper function to remove from a heap queues which are linked to 1613 * a flow_set about to be deleted. 1614 */ 1615 static void 1616 fs_remove_from_heap(struct dn_heap *h, struct dn_flow_set *fs) 1617 { 1618 int i = 0, found = 0 ; 1619 for (; i < h->elements ;) 1620 if ( ((struct dn_flow_queue *)h->p[i].object)->fs == fs) { 1621 h->elements-- ; 1622 h->p[i] = h->p[h->elements] ; 1623 found++ ; 1624 } else 1625 i++ ; 1626 if (found) 1627 heapify(h); 1628 } 1629 1630 /* 1631 * helper function to remove a pipe from a heap (can be there at most once) 1632 */ 1633 static void 1634 pipe_remove_from_heap(struct dn_heap *h, struct dn_pipe *p) 1635 { 1636 if (h->elements > 0) { 1637 int i = 0 ; 1638 for (i=0; i < h->elements ; i++ ) { 1639 if (h->p[i].object == p) { /* found it */ 1640 h->elements-- ; 1641 h->p[i] = h->p[h->elements] ; 1642 heapify(h); 1643 break ; 1644 } 1645 } 1646 } 1647 } 1648 1649 /* 1650 * drain all queues. Called in case of severe mbuf shortage. 1651 */ 1652 void 1653 dummynet_drain(void) 1654 { 1655 struct dn_flow_set *fs; 1656 struct dn_pipe *p; 1657 struct dn_pkt *pkt; 1658 1659 heap_free(&ready_heap); 1660 heap_free(&wfq_ready_heap); 1661 heap_free(&extract_heap); 1662 /* remove all references to this pipe from flow_sets */ 1663 for (fs = all_flow_sets; fs; fs= fs->next ) 1664 purge_flow_set(fs, 0); 1665 1666 for (p = all_pipes; p; p= p->next ) { 1667 purge_flow_set(&(p->fs), 0); 1668 for (pkt = p->head ; pkt ; ) 1669 DN_FREE_PKT(pkt) ; 1670 p->head = p->tail = NULL ; 1671 } 1672 } 1673 1674 /* 1675 * Fully delete a pipe or a queue, cleaning up associated info. 1676 */ 1677 static int 1678 delete_pipe(struct dn_pipe *p) 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 crit_enter(); 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 crit_exit(); 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 crit_enter(); 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 crit_exit(); 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 error=0 ; 1789 1790 crit_enter(); 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 crit_exit(); 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 callout_init(&dn_timeout); 1915 callout_reset(&dn_timeout, 1, dummynet, NULL); 1916 } 1917 1918 static int 1919 dummynet_modevent(module_t mod, int type, void *data) 1920 { 1921 switch (type) { 1922 case MOD_LOAD: 1923 crit_enter(); 1924 if (DUMMYNET_LOADED) { 1925 crit_exit(); 1926 printf("DUMMYNET already loaded\n"); 1927 return EEXIST ; 1928 } 1929 ip_dn_init(); 1930 crit_exit(); 1931 break; 1932 1933 case MOD_UNLOAD: 1934 #if !defined(KLD_MODULE) 1935 printf("dummynet statically compiled, cannot unload\n"); 1936 return EINVAL ; 1937 #else 1938 crit_enter(); 1939 callout_stop(&dn_timeout); 1940 dummynet_flush(); 1941 ip_dn_ctl_ptr = NULL; 1942 ip_dn_io_ptr = NULL; 1943 ip_dn_ruledel_ptr = NULL; 1944 crit_exit(); 1945 #endif 1946 break ; 1947 default: 1948 break ; 1949 } 1950 return 0 ; 1951 } 1952 1953 static moduledata_t dummynet_mod = { 1954 "dummynet", 1955 dummynet_modevent, 1956 NULL 1957 }; 1958 DECLARE_MODULE(dummynet, dummynet_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 1959 MODULE_DEPEND(dummynet, ipfw, 1, 1, 1); 1960 MODULE_VERSION(dummynet, 1); 1961