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 1.26 2007/10/20 09:08:28 sephe 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/systimer.h> 77 #include <sys/thread2.h> 78 #include <net/if.h> 79 #include <net/route.h> 80 #include <netinet/in.h> 81 #include <netinet/in_systm.h> 82 #include <netinet/in_var.h> 83 #include <netinet/ip.h> 84 #include <net/ipfw/ip_fw.h> 85 #include <net/dummynet/ip_dummynet.h> 86 #include <netinet/ip_var.h> 87 #include <net/netmsg2.h> 88 89 #include <netinet/if_ether.h> /* for struct arpcom */ 90 91 #ifndef DUMMYNET_CALLOUT_FREQ_MAX 92 #define DUMMYNET_CALLOUT_FREQ_MAX 30000 93 #endif 94 95 /* 96 * We keep a private variable for the simulation time, but we could 97 * probably use an existing one ("softticks" in sys/kern/kern_timer.c) 98 */ 99 static dn_key curr_time = 0 ; /* current simulation time */ 100 101 static int dn_hash_size = 64 ; /* default hash size */ 102 103 /* statistics on number of queue searches and search steps */ 104 static int searches, search_steps ; 105 static int pipe_expire = 1 ; /* expire queue if empty */ 106 static int dn_max_ratio = 16 ; /* max queues/buckets ratio */ 107 108 static int red_lookup_depth = 256; /* RED - default lookup table depth */ 109 static int red_avg_pkt_size = 512; /* RED - default medium packet size */ 110 static int red_max_pkt_size = 1500; /* RED - default max packet size */ 111 112 /* 113 * Three heaps contain queues and pipes that the scheduler handles: 114 * 115 * ready_heap contains all dn_flow_queue related to fixed-rate pipes. 116 * 117 * wfq_ready_heap contains the pipes associated with WF2Q flows 118 * 119 * extract_heap contains pipes associated with delay lines. 120 * 121 */ 122 123 MALLOC_DEFINE(M_DUMMYNET, "dummynet", "dummynet heap"); 124 125 static struct dn_heap ready_heap, extract_heap, wfq_ready_heap ; 126 127 static int heap_init(struct dn_heap *h, int size) ; 128 static int heap_insert (struct dn_heap *h, dn_key key1, void *p); 129 static void heap_extract(struct dn_heap *h, void *obj); 130 131 static void transmit_event(struct dn_pipe *pipe); 132 static void ready_event(struct dn_flow_queue *q); 133 134 static int sysctl_dn_hz(SYSCTL_HANDLER_ARGS); 135 136 static struct dn_pipe *all_pipes = NULL ; /* list of all pipes */ 137 static struct dn_flow_set *all_flow_sets = NULL ;/* list of all flow_sets */ 138 139 static struct netmsg dn_netmsg; 140 static struct systimer dn_clock; 141 static int dn_hz = 1000; 142 static int dn_cpu = 0; /* TODO tunable */ 143 144 SYSCTL_NODE(_net_inet_ip, OID_AUTO, dummynet, 145 CTLFLAG_RW, 0, "Dummynet"); 146 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, hash_size, 147 CTLFLAG_RW, &dn_hash_size, 0, "Default hash table size"); 148 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, curr_time, 149 CTLFLAG_RD, &curr_time, 0, "Current tick"); 150 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, ready_heap, 151 CTLFLAG_RD, &ready_heap.size, 0, "Size of ready heap"); 152 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, extract_heap, 153 CTLFLAG_RD, &extract_heap.size, 0, "Size of extract heap"); 154 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, searches, 155 CTLFLAG_RD, &searches, 0, "Number of queue searches"); 156 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, search_steps, 157 CTLFLAG_RD, &search_steps, 0, "Number of queue search steps"); 158 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, expire, 159 CTLFLAG_RW, &pipe_expire, 0, "Expire queue if empty"); 160 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, max_chain_len, 161 CTLFLAG_RW, &dn_max_ratio, 0, 162 "Max ratio between dynamic queues and buckets"); 163 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_lookup_depth, 164 CTLFLAG_RD, &red_lookup_depth, 0, "Depth of RED lookup table"); 165 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_avg_pkt_size, 166 CTLFLAG_RD, &red_avg_pkt_size, 0, "RED Medium packet size"); 167 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_max_pkt_size, 168 CTLFLAG_RD, &red_max_pkt_size, 0, "RED Max packet size"); 169 SYSCTL_PROC(_net_inet_ip_dummynet, OID_AUTO, hz, CTLTYPE_INT | CTLFLAG_RW, 170 0, 0, sysctl_dn_hz, "I", "Dummynet callout frequency"); 171 172 static int config_pipe(struct dn_pipe *p); 173 static int ip_dn_ctl(struct sockopt *sopt); 174 175 static void rt_unref(struct rtentry *); 176 static void dummynet_clock(systimer_t, struct intrframe *); 177 static void dummynet(struct netmsg *); 178 static void dummynet_flush(void); 179 void dummynet_drain(void); 180 static ip_dn_io_t dummynet_io; 181 static void dn_rule_delete(void *); 182 183 int if_tx_rdy(struct ifnet *ifp); 184 185 static void 186 rt_unref(struct rtentry *rt) 187 { 188 if (rt == NULL) 189 return ; 190 if (rt->rt_refcnt <= 0) 191 kprintf("-- warning, refcnt now %ld, decreasing\n", rt->rt_refcnt); 192 RTFREE(rt); 193 } 194 195 /* 196 * Heap management functions. 197 * 198 * In the heap, first node is element 0. Children of i are 2i+1 and 2i+2. 199 * Some macros help finding parent/children so we can optimize them. 200 * 201 * heap_init() is called to expand the heap when needed. 202 * Increment size in blocks of 16 entries. 203 * XXX failure to allocate a new element is a pretty bad failure 204 * as we basically stall a whole queue forever!! 205 * Returns 1 on error, 0 on success 206 */ 207 #define HEAP_FATHER(x) ( ( (x) - 1 ) / 2 ) 208 #define HEAP_LEFT(x) ( 2*(x) + 1 ) 209 #define HEAP_IS_LEFT(x) ( (x) & 1 ) 210 #define HEAP_RIGHT(x) ( 2*(x) + 2 ) 211 #define HEAP_SWAP(a, b, buffer) { buffer = a ; a = b ; b = buffer ; } 212 #define HEAP_INCREMENT 15 213 214 static int 215 heap_init(struct dn_heap *h, int new_size) 216 { 217 struct dn_heap_entry *p; 218 219 if (h->size >= new_size ) { 220 kprintf("heap_init, Bogus call, have %d want %d\n", 221 h->size, new_size); 222 return 0 ; 223 } 224 new_size = (new_size + HEAP_INCREMENT ) & ~HEAP_INCREMENT ; 225 p = kmalloc(new_size * sizeof(*p), M_DUMMYNET, M_WAITOK | M_ZERO); 226 if (h->size > 0) { 227 bcopy(h->p, p, h->size * sizeof(*p) ); 228 kfree(h->p, M_DUMMYNET); 229 } 230 h->p = p ; 231 h->size = new_size ; 232 return 0 ; 233 } 234 235 /* 236 * Insert element in heap. Normally, p != NULL, we insert p in 237 * a new position and bubble up. If p == NULL, then the element is 238 * already in place, and key is the position where to start the 239 * bubble-up. 240 * Returns 1 on failure (cannot allocate new heap entry) 241 * 242 * If offset > 0 the position (index, int) of the element in the heap is 243 * also stored in the element itself at the given offset in bytes. 244 */ 245 #define SET_OFFSET(heap, node) \ 246 if (heap->offset > 0) \ 247 *((int *)((char *)(heap->p[node].object) + heap->offset)) = node ; 248 /* 249 * RESET_OFFSET is used for sanity checks. It sets offset to an invalid value. 250 */ 251 #define RESET_OFFSET(heap, node) \ 252 if (heap->offset > 0) \ 253 *((int *)((char *)(heap->p[node].object) + heap->offset)) = -1 ; 254 static int 255 heap_insert(struct dn_heap *h, dn_key key1, void *p) 256 { 257 int son = h->elements ; 258 259 if (p == NULL) /* data already there, set starting point */ 260 son = key1 ; 261 else { /* insert new element at the end, possibly resize */ 262 son = h->elements ; 263 if (son == h->size) /* need resize... */ 264 if (heap_init(h, h->elements+1) ) 265 return 1 ; /* failure... */ 266 h->p[son].object = p ; 267 h->p[son].key = key1 ; 268 h->elements++ ; 269 } 270 while (son > 0) { /* bubble up */ 271 int father = HEAP_FATHER(son) ; 272 struct dn_heap_entry tmp ; 273 274 if (DN_KEY_LT( h->p[father].key, h->p[son].key ) ) 275 break ; /* found right position */ 276 /* son smaller than father, swap and repeat */ 277 HEAP_SWAP(h->p[son], h->p[father], tmp) ; 278 SET_OFFSET(h, son); 279 son = father ; 280 } 281 SET_OFFSET(h, son); 282 return 0 ; 283 } 284 285 /* 286 * remove top element from heap, or obj if obj != NULL 287 */ 288 static void 289 heap_extract(struct dn_heap *h, void *obj) 290 { 291 int child, father, max = h->elements - 1 ; 292 293 if (max < 0) { 294 kprintf("warning, extract from empty heap 0x%p\n", h); 295 return ; 296 } 297 father = 0 ; /* default: move up smallest child */ 298 if (obj != NULL) { /* extract specific element, index is at offset */ 299 if (h->offset <= 0) 300 panic("*** heap_extract from middle not supported on this heap!!!\n"); 301 father = *((int *)((char *)obj + h->offset)) ; 302 if (father < 0 || father >= h->elements) { 303 kprintf("dummynet: heap_extract, father %d out of bound 0..%d\n", 304 father, h->elements); 305 panic("heap_extract"); 306 } 307 } 308 RESET_OFFSET(h, father); 309 child = HEAP_LEFT(father) ; /* left child */ 310 while (child <= max) { /* valid entry */ 311 if (child != max && DN_KEY_LT(h->p[child+1].key, h->p[child].key) ) 312 child = child+1 ; /* take right child, otherwise left */ 313 h->p[father] = h->p[child] ; 314 SET_OFFSET(h, father); 315 father = child ; 316 child = HEAP_LEFT(child) ; /* left child for next loop */ 317 } 318 h->elements-- ; 319 if (father != max) { 320 /* 321 * Fill hole with last entry and bubble up, reusing the insert code 322 */ 323 h->p[father] = h->p[max] ; 324 heap_insert(h, father, NULL); /* this one cannot fail */ 325 } 326 } 327 328 #if 0 329 /* 330 * change object position and update references 331 * XXX this one is never used! 332 */ 333 static void 334 heap_move(struct dn_heap *h, dn_key new_key, void *object) 335 { 336 int temp; 337 int i ; 338 int max = h->elements-1 ; 339 struct dn_heap_entry buf ; 340 341 if (h->offset <= 0) 342 panic("cannot move items on this heap"); 343 344 i = *((int *)((char *)object + h->offset)); 345 if (DN_KEY_LT(new_key, h->p[i].key) ) { /* must move up */ 346 h->p[i].key = new_key ; 347 for (; i>0 && DN_KEY_LT(new_key, h->p[(temp = HEAP_FATHER(i))].key) ; 348 i = temp ) { /* bubble up */ 349 HEAP_SWAP(h->p[i], h->p[temp], buf) ; 350 SET_OFFSET(h, i); 351 } 352 } else { /* must move down */ 353 h->p[i].key = new_key ; 354 while ( (temp = HEAP_LEFT(i)) <= max ) { /* found left child */ 355 if ((temp != max) && DN_KEY_GT(h->p[temp].key, h->p[temp+1].key)) 356 temp++ ; /* select child with min key */ 357 if (DN_KEY_GT(new_key, h->p[temp].key)) { /* go down */ 358 HEAP_SWAP(h->p[i], h->p[temp], buf) ; 359 SET_OFFSET(h, i); 360 } else 361 break ; 362 i = temp ; 363 } 364 } 365 SET_OFFSET(h, i); 366 } 367 #endif /* heap_move, unused */ 368 369 /* 370 * heapify() will reorganize data inside an array to maintain the 371 * heap property. It is needed when we delete a bunch of entries. 372 */ 373 static void 374 heapify(struct dn_heap *h) 375 { 376 int i ; 377 378 for (i = 0 ; i < h->elements ; i++ ) 379 heap_insert(h, i , NULL) ; 380 } 381 382 /* 383 * cleanup the heap and free data structure 384 */ 385 static void 386 heap_free(struct dn_heap *h) 387 { 388 if (h->size >0 ) 389 kfree(h->p, M_DUMMYNET); 390 bzero(h, sizeof(*h) ); 391 } 392 393 /* 394 * --- end of heap management functions --- 395 */ 396 397 /* 398 * Scheduler functions: 399 * 400 * transmit_event() is called when the delay-line needs to enter 401 * the scheduler, either because of existing pkts getting ready, 402 * or new packets entering the queue. The event handled is the delivery 403 * time of the packet. 404 * 405 * ready_event() does something similar with fixed-rate queues, and the 406 * event handled is the finish time of the head pkt. 407 * 408 * wfq_ready_event() does something similar with WF2Q queues, and the 409 * event handled is the start time of the head pkt. 410 * 411 * In all cases, we make sure that the data structures are consistent 412 * before passing pkts out, because this might trigger recursive 413 * invocations of the procedures. 414 */ 415 static void 416 transmit_event(struct dn_pipe *pipe) 417 { 418 struct dn_pkt *pkt ; 419 420 while ( (pkt = pipe->head) && DN_KEY_LEQ(pkt->output_time, curr_time) ) { 421 /* 422 * first unlink, then call procedures, since ip_input() can invoke 423 * ip_output() and viceversa, thus causing nested calls 424 */ 425 pipe->head = pkt->dn_next; 426 427 /* 428 * The actual mbuf is preceded by a struct dn_pkt, resembling an mbuf 429 * (NOT A REAL one, just a small block of malloc'ed memory) with 430 * m_type = MT_TAG, m_flags = PACKET_TAG_DUMMYNET 431 * dn_m (m_next) = actual mbuf to be processed by ip_input/output 432 * and some other fields. 433 * The block IS FREED HERE because it contains parameters passed 434 * to the called routine. 435 */ 436 switch (pkt->dn_dir) { 437 case DN_TO_IP_OUT: 438 ip_output((struct mbuf *)pkt, NULL, NULL, 0, NULL, NULL); 439 rt_unref (pkt->ro.ro_rt) ; 440 break ; 441 442 case DN_TO_IP_IN : 443 ip_input((struct mbuf *)pkt) ; 444 break ; 445 446 case DN_TO_ETH_DEMUX: 447 { 448 struct mbuf *m = (struct mbuf *)pkt ; 449 struct ether_header *eh; 450 451 if (pkt->dn_m->m_len < ETHER_HDR_LEN && 452 (pkt->dn_m = m_pullup(pkt->dn_m, ETHER_HDR_LEN)) == NULL) { 453 kprintf("dummynet: pullup fail, dropping pkt\n"); 454 break; 455 } 456 /* 457 * same as ether_input, make eh be a pointer into the mbuf 458 */ 459 eh = mtod(pkt->dn_m, struct ether_header *); 460 m_adj(pkt->dn_m, ETHER_HDR_LEN); 461 /* which consumes the mbuf */ 462 ether_demux(NULL, eh, m); 463 } 464 break ; 465 case DN_TO_ETH_OUT: 466 ether_output_frame(pkt->ifp, (struct mbuf *)pkt); 467 break; 468 469 default: 470 kprintf("dummynet: bad switch %d!\n", pkt->dn_dir); 471 m_freem(pkt->dn_m); 472 break ; 473 } 474 kfree(pkt, M_DUMMYNET); 475 } 476 /* if there are leftover packets, put into the heap for next event */ 477 if ( (pkt = pipe->head) ) 478 heap_insert(&extract_heap, pkt->output_time, pipe ) ; 479 /* XXX should check errors on heap_insert, by draining the 480 * whole pipe p and hoping in the future we are more successful 481 */ 482 } 483 484 /* 485 * the following macro computes how many ticks we have to wait 486 * before being able to transmit a packet. The credit is taken from 487 * either a pipe (WF2Q) or a flow_queue (per-flow queueing) 488 */ 489 #define SET_TICKS(pkt, q, p) \ 490 (pkt->dn_m->m_pkthdr.len*8*dn_hz - (q)->numbytes + p->bandwidth - 1 ) / \ 491 p->bandwidth ; 492 493 /* 494 * extract pkt from queue, compute output time (could be now) 495 * and put into delay line (p_queue) 496 */ 497 static void 498 move_pkt(struct dn_pkt *pkt, struct dn_flow_queue *q, 499 struct dn_pipe *p, int len) 500 { 501 q->head = pkt->dn_next; 502 q->len-- ; 503 q->len_bytes -= len ; 504 505 pkt->output_time = curr_time + p->delay ; 506 507 if (p->head == NULL) 508 p->head = pkt; 509 else 510 p->tail->dn_next = pkt; 511 p->tail = pkt; 512 p->tail->dn_next = NULL; 513 } 514 515 /* 516 * ready_event() is invoked every time the queue must enter the 517 * scheduler, either because the first packet arrives, or because 518 * a previously scheduled event fired. 519 * On invokation, drain as many pkts as possible (could be 0) and then 520 * if there are leftover packets reinsert the pkt in the scheduler. 521 */ 522 static void 523 ready_event(struct dn_flow_queue *q) 524 { 525 struct dn_pkt *pkt; 526 struct dn_pipe *p = q->fs->pipe ; 527 int p_was_empty ; 528 529 if (p == NULL) { 530 kprintf("ready_event- pipe is gone\n"); 531 return ; 532 } 533 p_was_empty = (p->head == NULL) ; 534 535 /* 536 * schedule fixed-rate queues linked to this pipe: 537 * Account for the bw accumulated since last scheduling, then 538 * drain as many pkts as allowed by q->numbytes and move to 539 * the delay line (in p) computing output time. 540 * bandwidth==0 (no limit) means we can drain the whole queue, 541 * setting len_scaled = 0 does the job. 542 */ 543 q->numbytes += ( curr_time - q->sched_time ) * p->bandwidth; 544 while ( (pkt = q->head) != NULL ) { 545 int len = pkt->dn_m->m_pkthdr.len; 546 int len_scaled = p->bandwidth ? len*8*dn_hz : 0 ; 547 if (len_scaled > q->numbytes ) 548 break ; 549 q->numbytes -= len_scaled ; 550 move_pkt(pkt, q, p, len); 551 } 552 /* 553 * If we have more packets queued, schedule next ready event 554 * (can only occur when bandwidth != 0, otherwise we would have 555 * flushed the whole queue in the previous loop). 556 * To this purpose we record the current time and compute how many 557 * ticks to go for the finish time of the packet. 558 */ 559 if ( (pkt = q->head) != NULL ) { /* this implies bandwidth != 0 */ 560 dn_key t = SET_TICKS(pkt, q, p); /* ticks i have to wait */ 561 q->sched_time = curr_time ; 562 heap_insert(&ready_heap, curr_time + t, (void *)q ); 563 /* XXX should check errors on heap_insert, and drain the whole 564 * queue on error hoping next time we are luckier. 565 */ 566 } else { /* RED needs to know when the queue becomes empty */ 567 q->q_time = curr_time; 568 q->numbytes = 0; 569 } 570 /* 571 * If the delay line was empty call transmit_event(p) now. 572 * Otherwise, the scheduler will take care of it. 573 */ 574 if (p_was_empty) 575 transmit_event(p); 576 } 577 578 /* 579 * Called when we can transmit packets on WF2Q queues. Take pkts out of 580 * the queues at their start time, and enqueue into the delay line. 581 * Packets are drained until p->numbytes < 0. As long as 582 * len_scaled >= p->numbytes, the packet goes into the delay line 583 * with a deadline p->delay. For the last packet, if p->numbytes<0, 584 * there is an additional delay. 585 */ 586 static void 587 ready_event_wfq(struct dn_pipe *p) 588 { 589 int p_was_empty = (p->head == NULL) ; 590 struct dn_heap *sch = &(p->scheduler_heap); 591 struct dn_heap *neh = &(p->not_eligible_heap) ; 592 593 if (p->if_name[0] == 0) /* tx clock is simulated */ 594 p->numbytes += ( curr_time - p->sched_time ) * p->bandwidth; 595 else { /* tx clock is for real, the ifq must be empty or this is a NOP */ 596 if (p->ifp && p->ifp->if_snd.ifq_head != NULL) 597 return ; 598 else { 599 DEB(kprintf("pipe %d ready from %s --\n", 600 p->pipe_nr, p->if_name);) 601 } 602 } 603 604 /* 605 * While we have backlogged traffic AND credit, we need to do 606 * something on the queue. 607 */ 608 while ( p->numbytes >=0 && (sch->elements>0 || neh->elements >0) ) { 609 if (sch->elements > 0) { /* have some eligible pkts to send out */ 610 struct dn_flow_queue *q = sch->p[0].object ; 611 struct dn_pkt *pkt = q->head; 612 struct dn_flow_set *fs = q->fs; 613 u_int64_t len = pkt->dn_m->m_pkthdr.len; 614 int len_scaled = p->bandwidth ? len*8*dn_hz : 0 ; 615 616 heap_extract(sch, NULL); /* remove queue from heap */ 617 p->numbytes -= len_scaled ; 618 move_pkt(pkt, q, p, len); 619 620 p->V += (len<<MY_M) / p->sum ; /* update V */ 621 q->S = q->F ; /* update start time */ 622 if (q->len == 0) { /* Flow not backlogged any more */ 623 fs->backlogged-- ; 624 heap_insert(&(p->idle_heap), q->F, q); 625 } else { /* still backlogged */ 626 /* 627 * update F and position in backlogged queue, then 628 * put flow in not_eligible_heap (we will fix this later). 629 */ 630 len = (q->head)->dn_m->m_pkthdr.len; 631 q->F += (len<<MY_M)/(u_int64_t) fs->weight ; 632 if (DN_KEY_LEQ(q->S, p->V)) 633 heap_insert(neh, q->S, q); 634 else 635 heap_insert(sch, q->F, q); 636 } 637 } 638 /* 639 * now compute V = max(V, min(S_i)). Remember that all elements in sch 640 * have by definition S_i <= V so if sch is not empty, V is surely 641 * the max and we must not update it. Conversely, if sch is empty 642 * we only need to look at neh. 643 */ 644 if (sch->elements == 0 && neh->elements > 0) 645 p->V = MAX64 ( p->V, neh->p[0].key ); 646 /* move from neh to sch any packets that have become eligible */ 647 while (neh->elements > 0 && DN_KEY_LEQ(neh->p[0].key, p->V) ) { 648 struct dn_flow_queue *q = neh->p[0].object ; 649 heap_extract(neh, NULL); 650 heap_insert(sch, q->F, q); 651 } 652 653 if (p->if_name[0] != '\0') {/* tx clock is from a real thing */ 654 p->numbytes = -1 ; /* mark not ready for I/O */ 655 break ; 656 } 657 } 658 if (sch->elements == 0 && neh->elements == 0 && p->numbytes >= 0 659 && p->idle_heap.elements > 0) { 660 /* 661 * no traffic and no events scheduled. We can get rid of idle-heap. 662 */ 663 int i ; 664 665 for (i = 0 ; i < p->idle_heap.elements ; i++) { 666 struct dn_flow_queue *q = p->idle_heap.p[i].object ; 667 668 q->F = 0 ; 669 q->S = q->F + 1 ; 670 } 671 p->sum = 0 ; 672 p->V = 0 ; 673 p->idle_heap.elements = 0 ; 674 } 675 /* 676 * If we are getting clocks from dummynet (not a real interface) and 677 * If we are under credit, schedule the next ready event. 678 * Also fix the delivery time of the last packet. 679 */ 680 if (p->if_name[0]==0 && p->numbytes < 0) { /* this implies bandwidth >0 */ 681 dn_key t=0 ; /* number of ticks i have to wait */ 682 683 if (p->bandwidth > 0) 684 t = ( p->bandwidth -1 - p->numbytes) / p->bandwidth ; 685 p->tail->output_time += t ; 686 p->sched_time = curr_time ; 687 heap_insert(&wfq_ready_heap, curr_time + t, (void *)p); 688 /* XXX should check errors on heap_insert, and drain the whole 689 * queue on error hoping next time we are luckier. 690 */ 691 } 692 /* 693 * If the delay line was empty call transmit_event(p) now. 694 * Otherwise, the scheduler will take care of it. 695 */ 696 if (p_was_empty) 697 transmit_event(p); 698 } 699 700 /* 701 * This is called once per tick, or HZ times per second. It is used to 702 * increment the current tick counter and schedule expired events. 703 */ 704 static void 705 dummynet(struct netmsg *msg) 706 { 707 void *p ; /* generic parameter to handler */ 708 struct dn_heap *h ; 709 struct dn_heap *heaps[3]; 710 int i; 711 struct dn_pipe *pe ; 712 713 heaps[0] = &ready_heap ; /* fixed-rate queues */ 714 heaps[1] = &wfq_ready_heap ; /* wfq queues */ 715 heaps[2] = &extract_heap ; /* delay line */ 716 crit_enter(); /* see note on top, splnet() is not enough */ 717 718 /* Reply ASAP */ 719 lwkt_replymsg(&msg->nm_lmsg, 0); 720 721 curr_time++ ; 722 for (i=0; i < 3 ; i++) { 723 h = heaps[i]; 724 while (h->elements > 0 && DN_KEY_LEQ(h->p[0].key, curr_time) ) { 725 DDB(if (h->p[0].key > curr_time) 726 kprintf("-- dummynet: warning, heap %d is %d ticks late\n", 727 i, (int)(curr_time - h->p[0].key));) 728 p = h->p[0].object ; /* store a copy before heap_extract */ 729 heap_extract(h, NULL); /* need to extract before processing */ 730 if (i == 0) 731 ready_event(p) ; 732 else if (i == 1) { 733 struct dn_pipe *pipe = p; 734 if (pipe->if_name[0] != '\0') 735 kprintf("*** bad ready_event_wfq for pipe %s\n", 736 pipe->if_name); 737 else 738 ready_event_wfq(p) ; 739 } else 740 transmit_event(p); 741 } 742 } 743 /* sweep pipes trying to expire idle flow_queues */ 744 for (pe = all_pipes; pe ; pe = pe->next ) 745 if (pe->idle_heap.elements > 0 && 746 DN_KEY_LT(pe->idle_heap.p[0].key, pe->V) ) { 747 struct dn_flow_queue *q = pe->idle_heap.p[0].object ; 748 749 heap_extract(&(pe->idle_heap), NULL); 750 q->S = q->F + 1 ; /* mark timestamp as invalid */ 751 pe->sum -= q->fs->weight ; 752 } 753 crit_exit(); 754 } 755 756 /* 757 * called by an interface when tx_rdy occurs. 758 */ 759 int 760 if_tx_rdy(struct ifnet *ifp) 761 { 762 struct dn_pipe *p; 763 764 for (p = all_pipes; p ; p = p->next ) 765 if (p->ifp == ifp) 766 break ; 767 if (p == NULL) { 768 for (p = all_pipes; p ; p = p->next ) 769 if (!strcmp(p->if_name, ifp->if_xname) ) { 770 p->ifp = ifp ; 771 DEB(kprintf("++ tx rdy from %s (now found)\n", ifp->if_xname);) 772 break ; 773 } 774 } 775 if (p != NULL) { 776 DEB(kprintf("++ tx rdy from %s - qlen %d\n", ifp->if_xname, 777 ifp->if_snd.ifq_len);) 778 p->numbytes = 0 ; /* mark ready for I/O */ 779 ready_event_wfq(p); 780 } 781 return 0; 782 } 783 784 /* 785 * Unconditionally expire empty queues in case of shortage. 786 * Returns the number of queues freed. 787 */ 788 static int 789 expire_queues(struct dn_flow_set *fs) 790 { 791 struct dn_flow_queue *q, *prev ; 792 int i, initial_elements = fs->rq_elements ; 793 794 if (fs->last_expired == time_second) 795 return 0 ; 796 fs->last_expired = time_second ; 797 for (i = 0 ; i <= fs->rq_size ; i++) /* last one is overflow */ 798 for (prev=NULL, q = fs->rq[i] ; q != NULL ; ) 799 if (q->head != NULL || q->S != q->F+1) { 800 prev = q ; 801 q = q->next ; 802 } else { /* entry is idle, expire it */ 803 struct dn_flow_queue *old_q = q ; 804 805 if (prev != NULL) 806 prev->next = q = q->next ; 807 else 808 fs->rq[i] = q = q->next ; 809 fs->rq_elements-- ; 810 kfree(old_q, M_DUMMYNET); 811 } 812 return initial_elements - fs->rq_elements ; 813 } 814 815 /* 816 * If room, create a new queue and put at head of slot i; 817 * otherwise, create or use the default queue. 818 */ 819 static struct dn_flow_queue * 820 create_queue(struct dn_flow_set *fs, int i) 821 { 822 struct dn_flow_queue *q ; 823 824 if (fs->rq_elements > fs->rq_size * dn_max_ratio && 825 expire_queues(fs) == 0) { 826 /* 827 * No way to get room, use or create overflow queue. 828 */ 829 i = fs->rq_size ; 830 if ( fs->rq[i] != NULL ) 831 return fs->rq[i] ; 832 } 833 q = kmalloc(sizeof(*q), M_DUMMYNET, M_WAITOK | M_ZERO); 834 q->fs = fs ; 835 q->hash_slot = i ; 836 q->next = fs->rq[i] ; 837 q->S = q->F + 1; /* hack - mark timestamp as invalid */ 838 fs->rq[i] = q ; 839 fs->rq_elements++ ; 840 return q ; 841 } 842 843 /* 844 * Given a flow_set and a pkt in last_pkt, find a matching queue 845 * after appropriate masking. The queue is moved to front 846 * so that further searches take less time. 847 */ 848 static struct dn_flow_queue * 849 find_queue(struct dn_flow_set *fs, struct ipfw_flow_id *id) 850 { 851 int i = 0 ; /* we need i and q for new allocations */ 852 struct dn_flow_queue *q, *prev; 853 854 if ( !(fs->flags_fs & DN_HAVE_FLOW_MASK) ) 855 q = fs->rq[0] ; 856 else { 857 /* first, do the masking */ 858 id->dst_ip &= fs->flow_mask.dst_ip ; 859 id->src_ip &= fs->flow_mask.src_ip ; 860 id->dst_port &= fs->flow_mask.dst_port ; 861 id->src_port &= fs->flow_mask.src_port ; 862 id->proto &= fs->flow_mask.proto ; 863 id->flags = 0 ; /* we don't care about this one */ 864 /* then, hash function */ 865 i = ( (id->dst_ip) & 0xffff ) ^ 866 ( (id->dst_ip >> 15) & 0xffff ) ^ 867 ( (id->src_ip << 1) & 0xffff ) ^ 868 ( (id->src_ip >> 16 ) & 0xffff ) ^ 869 (id->dst_port << 1) ^ (id->src_port) ^ 870 (id->proto ); 871 i = i % fs->rq_size ; 872 /* finally, scan the current list for a match */ 873 searches++ ; 874 for (prev=NULL, q = fs->rq[i] ; q ; ) { 875 search_steps++; 876 if (id->dst_ip == q->id.dst_ip && 877 id->src_ip == q->id.src_ip && 878 id->dst_port == q->id.dst_port && 879 id->src_port == q->id.src_port && 880 id->proto == q->id.proto && 881 id->flags == q->id.flags) 882 break ; /* found */ 883 else if (pipe_expire && q->head == NULL && q->S == q->F+1 ) { 884 /* entry is idle and not in any heap, expire it */ 885 struct dn_flow_queue *old_q = q ; 886 887 if (prev != NULL) 888 prev->next = q = q->next ; 889 else 890 fs->rq[i] = q = q->next ; 891 fs->rq_elements-- ; 892 kfree(old_q, M_DUMMYNET); 893 continue ; 894 } 895 prev = q ; 896 q = q->next ; 897 } 898 if (q && prev != NULL) { /* found and not in front */ 899 prev->next = q->next ; 900 q->next = fs->rq[i] ; 901 fs->rq[i] = q ; 902 } 903 } 904 if (q == NULL) { /* no match, need to allocate a new entry */ 905 q = create_queue(fs, i); 906 if (q != NULL) 907 q->id = *id ; 908 } 909 return q ; 910 } 911 912 static int 913 red_drops(struct dn_flow_set *fs, struct dn_flow_queue *q, int len) 914 { 915 /* 916 * RED algorithm 917 * 918 * RED calculates the average queue size (avg) using a low-pass filter 919 * with an exponential weighted (w_q) moving average: 920 * avg <- (1-w_q) * avg + w_q * q_size 921 * where q_size is the queue length (measured in bytes or * packets). 922 * 923 * If q_size == 0, we compute the idle time for the link, and set 924 * avg = (1 - w_q)^(idle/s) 925 * where s is the time needed for transmitting a medium-sized packet. 926 * 927 * Now, if avg < min_th the packet is enqueued. 928 * If avg > max_th the packet is dropped. Otherwise, the packet is 929 * dropped with probability P function of avg. 930 * 931 */ 932 933 int64_t p_b = 0; 934 /* queue in bytes or packets ? */ 935 u_int q_size = (fs->flags_fs & DN_QSIZE_IS_BYTES) ? q->len_bytes : q->len; 936 937 DEB(kprintf("\n%d q: %2u ", (int) curr_time, q_size);) 938 939 /* average queue size estimation */ 940 if (q_size != 0) { 941 /* 942 * queue is not empty, avg <- avg + (q_size - avg) * w_q 943 */ 944 int diff = SCALE(q_size) - q->avg; 945 int64_t v = SCALE_MUL((int64_t) diff, (int64_t) fs->w_q); 946 947 q->avg += (int) v; 948 } else { 949 /* 950 * queue is empty, find for how long the queue has been 951 * empty and use a lookup table for computing 952 * (1 - * w_q)^(idle_time/s) where s is the time to send a 953 * (small) packet. 954 * XXX check wraps... 955 */ 956 if (q->avg) { 957 u_int t = (curr_time - q->q_time) / fs->lookup_step; 958 959 q->avg = (t < fs->lookup_depth) ? 960 SCALE_MUL(q->avg, fs->w_q_lookup[t]) : 0; 961 } 962 } 963 DEB(kprintf("avg: %u ", SCALE_VAL(q->avg));) 964 965 /* should i drop ? */ 966 967 if (q->avg < fs->min_th) { 968 q->count = -1; 969 return 0; /* accept packet ; */ 970 } 971 if (q->avg >= fs->max_th) { /* average queue >= max threshold */ 972 if (fs->flags_fs & DN_IS_GENTLE_RED) { 973 /* 974 * According to Gentle-RED, if avg is greater than max_th the 975 * packet is dropped with a probability 976 * p_b = c_3 * avg - c_4 977 * where c_3 = (1 - max_p) / max_th, and c_4 = 1 - 2 * max_p 978 */ 979 p_b = SCALE_MUL((int64_t) fs->c_3, (int64_t) q->avg) - fs->c_4; 980 } else { 981 q->count = -1; 982 kprintf("- drop"); 983 return 1 ; 984 } 985 } else if (q->avg > fs->min_th) { 986 /* 987 * we compute p_b using the linear dropping function p_b = c_1 * 988 * avg - c_2, where c_1 = max_p / (max_th - min_th), and c_2 = 989 * max_p * min_th / (max_th - min_th) 990 */ 991 p_b = SCALE_MUL((int64_t) fs->c_1, (int64_t) q->avg) - fs->c_2; 992 } 993 if (fs->flags_fs & DN_QSIZE_IS_BYTES) 994 p_b = (p_b * len) / fs->max_pkt_size; 995 if (++q->count == 0) 996 q->random = krandom() & 0xffff; 997 else { 998 /* 999 * q->count counts packets arrived since last drop, so a greater 1000 * value of q->count means a greater packet drop probability. 1001 */ 1002 if (SCALE_MUL(p_b, SCALE((int64_t) q->count)) > q->random) { 1003 q->count = 0; 1004 DEB(kprintf("- red drop");) 1005 /* after a drop we calculate a new random value */ 1006 q->random = krandom() & 0xffff; 1007 return 1; /* drop */ 1008 } 1009 } 1010 /* end of RED algorithm */ 1011 return 0 ; /* accept */ 1012 } 1013 1014 static __inline 1015 struct dn_flow_set * 1016 locate_flowset(int pipe_nr, struct ip_fw *rule) 1017 { 1018 struct dn_flow_set *fs; 1019 ipfw_insn *cmd = rule->cmd + rule->act_ofs; 1020 1021 if (cmd->opcode == O_LOG) 1022 cmd += F_LEN(cmd); 1023 fs = ((ipfw_insn_pipe *)cmd)->pipe_ptr; 1024 1025 if (fs != NULL) 1026 return fs; 1027 1028 if (cmd->opcode == O_QUEUE) 1029 for (fs=all_flow_sets; fs && fs->fs_nr != pipe_nr; fs=fs->next) 1030 ; 1031 else { 1032 struct dn_pipe *p1; 1033 for (p1 = all_pipes; p1 && p1->pipe_nr != pipe_nr; p1 = p1->next) 1034 ; 1035 if (p1 != NULL) 1036 fs = &(p1->fs) ; 1037 } 1038 /* record for the future */ 1039 ((ipfw_insn_pipe *)cmd)->pipe_ptr = fs; 1040 return fs ; 1041 } 1042 1043 /* 1044 * dummynet hook for packets. Below 'pipe' is a pipe or a queue 1045 * depending on whether WF2Q or fixed bw is used. 1046 * 1047 * pipe_nr pipe or queue the packet is destined for. 1048 * dir where shall we send the packet after dummynet. 1049 * m the mbuf with the packet 1050 * ifp the 'ifp' parameter from the caller. 1051 * NULL in ip_input, destination interface in ip_output 1052 * ro route parameter (only used in ip_output, NULL otherwise) 1053 * dst destination address, only used by ip_output 1054 * rule matching rule, in case of multiple passes 1055 * flags flags from the caller, only used in ip_output 1056 * 1057 */ 1058 static int 1059 dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa) 1060 { 1061 struct dn_pkt *pkt; 1062 struct dn_flow_set *fs; 1063 struct dn_pipe *pipe ; 1064 u_int64_t len = m->m_pkthdr.len ; 1065 struct dn_flow_queue *q = NULL ; 1066 int is_pipe; 1067 1068 crit_enter(); 1069 ipfw_insn *cmd = fwa->rule->cmd + fwa->rule->act_ofs; 1070 1071 if (cmd->opcode == O_LOG) 1072 cmd += F_LEN(cmd); 1073 is_pipe = (cmd->opcode == O_PIPE); 1074 1075 pipe_nr &= 0xffff ; 1076 1077 /* 1078 * this is a dummynet rule, so we expect a O_PIPE or O_QUEUE rule 1079 */ 1080 fs = locate_flowset(pipe_nr, fwa->rule); 1081 if (fs == NULL) 1082 goto dropit ; /* this queue/pipe does not exist! */ 1083 pipe = fs->pipe ; 1084 if (pipe == NULL) { /* must be a queue, try find a matching pipe */ 1085 for (pipe = all_pipes; pipe && pipe->pipe_nr != fs->parent_nr; 1086 pipe = pipe->next) 1087 ; 1088 if (pipe != NULL) 1089 fs->pipe = pipe ; 1090 else { 1091 kprintf("No pipe %d for queue %d, drop pkt\n", 1092 fs->parent_nr, fs->fs_nr); 1093 goto dropit ; 1094 } 1095 } 1096 q = find_queue(fs, &(fwa->f_id)); 1097 if ( q == NULL ) 1098 goto dropit ; /* cannot allocate queue */ 1099 /* 1100 * update statistics, then check reasons to drop pkt 1101 */ 1102 q->tot_bytes += len ; 1103 q->tot_pkts++ ; 1104 if ( fs->plr && krandom() < fs->plr ) 1105 goto dropit ; /* random pkt drop */ 1106 if ( fs->flags_fs & DN_QSIZE_IS_BYTES) { 1107 if (q->len_bytes > fs->qsize) 1108 goto dropit ; /* queue size overflow */ 1109 } else { 1110 if (q->len >= fs->qsize) 1111 goto dropit ; /* queue count overflow */ 1112 } 1113 if ( fs->flags_fs & DN_IS_RED && red_drops(fs, q, len) ) 1114 goto dropit ; 1115 1116 /* XXX expensive to zero, see if we can remove it*/ 1117 pkt = kmalloc(sizeof (*pkt), M_DUMMYNET, M_INTWAIT | M_ZERO | M_NULLOK); 1118 if (pkt == NULL) 1119 goto dropit; /* cannot allocate packet header */ 1120 1121 /* ok, i can handle the pkt now... */ 1122 /* build and enqueue packet + parameters */ 1123 pkt->hdr.mh_type = MT_TAG; 1124 pkt->hdr.mh_flags = PACKET_TAG_DUMMYNET; 1125 pkt->rule = fwa->rule ; 1126 pkt->dn_next = NULL; 1127 pkt->dn_m = m; 1128 pkt->dn_dir = dir ; 1129 1130 pkt->ifp = fwa->oif; 1131 if (dir == DN_TO_IP_OUT) { 1132 /* 1133 * We need to copy *ro because for ICMP pkts (and maybe others) 1134 * the caller passed a pointer into the stack; dst might also be 1135 * a pointer into *ro so it needs to be updated. 1136 */ 1137 pkt->ro = *(fwa->ro); 1138 if (fwa->ro->ro_rt) 1139 fwa->ro->ro_rt->rt_refcnt++ ; 1140 if (fwa->dst == (struct sockaddr_in *)&fwa->ro->ro_dst) /* dst points into ro */ 1141 fwa->dst = (struct sockaddr_in *)&(pkt->ro.ro_dst) ; 1142 1143 pkt->dn_dst = fwa->dst; 1144 pkt->flags = fwa->flags; 1145 } 1146 if (q->head == NULL) 1147 q->head = pkt; 1148 else 1149 q->tail->dn_next = pkt; 1150 q->tail = pkt; 1151 q->len++; 1152 q->len_bytes += len ; 1153 1154 if ( q->head != pkt ) /* flow was not idle, we are done */ 1155 goto done; 1156 /* 1157 * If we reach this point the flow was previously idle, so we need 1158 * to schedule it. This involves different actions for fixed-rate or 1159 * WF2Q queues. 1160 */ 1161 if (is_pipe) { 1162 /* 1163 * Fixed-rate queue: just insert into the ready_heap. 1164 */ 1165 dn_key t = 0 ; 1166 if (pipe->bandwidth) 1167 t = SET_TICKS(pkt, q, pipe); 1168 q->sched_time = curr_time ; 1169 if (t == 0) /* must process it now */ 1170 ready_event( q ); 1171 else 1172 heap_insert(&ready_heap, curr_time + t , q ); 1173 } else { 1174 /* 1175 * WF2Q. First, compute start time S: if the flow was idle (S=F+1) 1176 * set S to the virtual time V for the controlling pipe, and update 1177 * the sum of weights for the pipe; otherwise, remove flow from 1178 * idle_heap and set S to max(F,V). 1179 * Second, compute finish time F = S + len/weight. 1180 * Third, if pipe was idle, update V=max(S, V). 1181 * Fourth, count one more backlogged flow. 1182 */ 1183 if (DN_KEY_GT(q->S, q->F)) { /* means timestamps are invalid */ 1184 q->S = pipe->V ; 1185 pipe->sum += fs->weight ; /* add weight of new queue */ 1186 } else { 1187 heap_extract(&(pipe->idle_heap), q); 1188 q->S = MAX64(q->F, pipe->V ) ; 1189 } 1190 q->F = q->S + ( len<<MY_M )/(u_int64_t) fs->weight; 1191 1192 if (pipe->not_eligible_heap.elements == 0 && 1193 pipe->scheduler_heap.elements == 0) 1194 pipe->V = MAX64 ( q->S, pipe->V ); 1195 fs->backlogged++ ; 1196 /* 1197 * Look at eligibility. A flow is not eligibile if S>V (when 1198 * this happens, it means that there is some other flow already 1199 * scheduled for the same pipe, so the scheduler_heap cannot be 1200 * empty). If the flow is not eligible we just store it in the 1201 * not_eligible_heap. Otherwise, we store in the scheduler_heap 1202 * and possibly invoke ready_event_wfq() right now if there is 1203 * leftover credit. 1204 * Note that for all flows in scheduler_heap (SCH), S_i <= V, 1205 * and for all flows in not_eligible_heap (NEH), S_i > V . 1206 * So when we need to compute max( V, min(S_i) ) forall i in SCH+NEH, 1207 * we only need to look into NEH. 1208 */ 1209 if (DN_KEY_GT(q->S, pipe->V) ) { /* not eligible */ 1210 if (pipe->scheduler_heap.elements == 0) 1211 kprintf("++ ouch! not eligible but empty scheduler!\n"); 1212 heap_insert(&(pipe->not_eligible_heap), q->S, q); 1213 } else { 1214 heap_insert(&(pipe->scheduler_heap), q->F, q); 1215 if (pipe->numbytes >= 0) { /* pipe is idle */ 1216 if (pipe->scheduler_heap.elements != 1) 1217 kprintf("*** OUCH! pipe should have been idle!\n"); 1218 DEB(kprintf("Waking up pipe %d at %d\n", 1219 pipe->pipe_nr, (int)(q->F >> MY_M)); ) 1220 pipe->sched_time = curr_time ; 1221 ready_event_wfq(pipe); 1222 } 1223 } 1224 } 1225 done: 1226 crit_exit(); 1227 return 0; 1228 1229 dropit: 1230 crit_exit(); 1231 if (q) 1232 q->drops++ ; 1233 m_freem(m); 1234 return ( (fs && (fs->flags_fs & DN_NOERROR)) ? 0 : ENOBUFS); 1235 } 1236 1237 /* 1238 * Below, the rt_unref is only needed when (pkt->dn_dir == DN_TO_IP_OUT) 1239 * Doing this would probably save us the initial bzero of dn_pkt 1240 */ 1241 #define DN_FREE_PKT(pkt) { \ 1242 struct dn_pkt *n = pkt ; \ 1243 rt_unref ( n->ro.ro_rt ) ; \ 1244 m_freem(n->dn_m); \ 1245 pkt = n->dn_next; \ 1246 kfree(n, M_DUMMYNET) ; } 1247 1248 /* 1249 * Dispose all packets and flow_queues on a flow_set. 1250 * If all=1, also remove red lookup table and other storage, 1251 * including the descriptor itself. 1252 * For the one in dn_pipe MUST also cleanup ready_heap... 1253 */ 1254 static void 1255 purge_flow_set(struct dn_flow_set *fs, int all) 1256 { 1257 struct dn_pkt *pkt ; 1258 struct dn_flow_queue *q, *qn ; 1259 int i ; 1260 1261 for (i = 0 ; i <= fs->rq_size ; i++ ) { 1262 for (q = fs->rq[i] ; q ; q = qn ) { 1263 for (pkt = q->head ; pkt ; ) 1264 DN_FREE_PKT(pkt) ; 1265 qn = q->next ; 1266 kfree(q, M_DUMMYNET); 1267 } 1268 fs->rq[i] = NULL ; 1269 } 1270 fs->rq_elements = 0 ; 1271 if (all) { 1272 /* RED - free lookup table */ 1273 if (fs->w_q_lookup) 1274 kfree(fs->w_q_lookup, M_DUMMYNET); 1275 if (fs->rq) 1276 kfree(fs->rq, M_DUMMYNET); 1277 /* if this fs is not part of a pipe, free it */ 1278 if (fs->pipe && fs != &(fs->pipe->fs) ) 1279 kfree(fs, M_DUMMYNET); 1280 } 1281 } 1282 1283 /* 1284 * Dispose all packets queued on a pipe (not a flow_set). 1285 * Also free all resources associated to a pipe, which is about 1286 * to be deleted. 1287 */ 1288 static void 1289 purge_pipe(struct dn_pipe *pipe) 1290 { 1291 struct dn_pkt *pkt ; 1292 1293 purge_flow_set( &(pipe->fs), 1 ); 1294 1295 for (pkt = pipe->head ; pkt ; ) 1296 DN_FREE_PKT(pkt) ; 1297 1298 heap_free( &(pipe->scheduler_heap) ); 1299 heap_free( &(pipe->not_eligible_heap) ); 1300 heap_free( &(pipe->idle_heap) ); 1301 } 1302 1303 /* 1304 * Delete all pipes and heaps returning memory. Must also 1305 * remove references from all ipfw rules to all pipes. 1306 */ 1307 static void 1308 dummynet_flush(void) 1309 { 1310 struct dn_pipe *curr_p, *p ; 1311 struct dn_flow_set *fs, *curr_fs; 1312 1313 crit_enter(); 1314 1315 /* remove all references to pipes ...*/ 1316 flush_pipe_ptrs(NULL); 1317 /* prevent future matches... */ 1318 p = all_pipes ; 1319 all_pipes = NULL ; 1320 fs = all_flow_sets ; 1321 all_flow_sets = NULL ; 1322 /* and free heaps so we don't have unwanted events */ 1323 heap_free(&ready_heap); 1324 heap_free(&wfq_ready_heap); 1325 heap_free(&extract_heap); 1326 crit_exit(); 1327 /* 1328 * Now purge all queued pkts and delete all pipes 1329 */ 1330 /* scan and purge all flow_sets. */ 1331 for ( ; fs ; ) { 1332 curr_fs = fs ; 1333 fs = fs->next ; 1334 purge_flow_set(curr_fs, 1); 1335 } 1336 for ( ; p ; ) { 1337 purge_pipe(p); 1338 curr_p = p ; 1339 p = p->next ; 1340 kfree(curr_p, M_DUMMYNET); 1341 } 1342 } 1343 1344 1345 extern struct ip_fw *ip_fw_default_rule ; 1346 static void 1347 dn_rule_delete_fs(struct dn_flow_set *fs, void *r) 1348 { 1349 int i ; 1350 struct dn_flow_queue *q ; 1351 struct dn_pkt *pkt ; 1352 1353 for (i = 0 ; i <= fs->rq_size ; i++) /* last one is ovflow */ 1354 for (q = fs->rq[i] ; q ; q = q->next ) 1355 for (pkt = q->head; pkt; pkt = pkt->dn_next) 1356 if (pkt->rule == r) 1357 pkt->rule = ip_fw_default_rule ; 1358 } 1359 /* 1360 * when a firewall rule is deleted, scan all queues and remove the flow-id 1361 * from packets matching this rule. 1362 */ 1363 void 1364 dn_rule_delete(void *r) 1365 { 1366 struct dn_pipe *p ; 1367 struct dn_pkt *pkt ; 1368 struct dn_flow_set *fs ; 1369 1370 /* 1371 * If the rule references a queue (dn_flow_set), then scan 1372 * the flow set, otherwise scan pipes. Should do either, but doing 1373 * both does not harm. 1374 */ 1375 for ( fs = all_flow_sets ; fs ; fs = fs->next ) 1376 dn_rule_delete_fs(fs, r); 1377 for ( p = all_pipes ; p ; p = p->next ) { 1378 fs = &(p->fs) ; 1379 dn_rule_delete_fs(fs, r); 1380 for (pkt = p->head; pkt; pkt = pkt->dn_next) 1381 if (pkt->rule == r) 1382 pkt->rule = ip_fw_default_rule ; 1383 } 1384 } 1385 1386 /* 1387 * setup RED parameters 1388 */ 1389 static int 1390 config_red(struct dn_flow_set *p, struct dn_flow_set * x) 1391 { 1392 int i; 1393 1394 x->w_q = p->w_q; 1395 x->min_th = SCALE(p->min_th); 1396 x->max_th = SCALE(p->max_th); 1397 x->max_p = p->max_p; 1398 1399 x->c_1 = p->max_p / (p->max_th - p->min_th); 1400 x->c_2 = SCALE_MUL(x->c_1, SCALE(p->min_th)); 1401 if (x->flags_fs & DN_IS_GENTLE_RED) { 1402 x->c_3 = (SCALE(1) - p->max_p) / p->max_th; 1403 x->c_4 = (SCALE(1) - 2 * p->max_p); 1404 } 1405 1406 /* if the lookup table already exist, free and create it again */ 1407 if (x->w_q_lookup) { 1408 kfree(x->w_q_lookup, M_DUMMYNET); 1409 x->w_q_lookup = NULL ; 1410 } 1411 if (red_lookup_depth == 0) { 1412 kprintf("\nnet.inet.ip.dummynet.red_lookup_depth must be > 0"); 1413 kfree(x, M_DUMMYNET); 1414 return EINVAL; 1415 } 1416 x->lookup_depth = red_lookup_depth; 1417 x->w_q_lookup = kmalloc(x->lookup_depth * sizeof(int), 1418 M_DUMMYNET, M_WAITOK); 1419 1420 /* fill the lookup table with (1 - w_q)^x */ 1421 x->lookup_step = p->lookup_step ; 1422 x->lookup_weight = p->lookup_weight ; 1423 x->w_q_lookup[0] = SCALE(1) - x->w_q; 1424 for (i = 1; i < x->lookup_depth; i++) 1425 x->w_q_lookup[i] = SCALE_MUL(x->w_q_lookup[i - 1], x->lookup_weight); 1426 if (red_avg_pkt_size < 1) 1427 red_avg_pkt_size = 512 ; 1428 x->avg_pkt_size = red_avg_pkt_size ; 1429 if (red_max_pkt_size < 1) 1430 red_max_pkt_size = 1500 ; 1431 x->max_pkt_size = red_max_pkt_size ; 1432 return 0 ; 1433 } 1434 1435 static int 1436 alloc_hash(struct dn_flow_set *x, struct dn_flow_set *pfs) 1437 { 1438 if (x->flags_fs & DN_HAVE_FLOW_MASK) { /* allocate some slots */ 1439 int l = pfs->rq_size; 1440 1441 if (l == 0) 1442 l = dn_hash_size; 1443 if (l < 4) 1444 l = 4; 1445 else if (l > DN_MAX_HASH_SIZE) 1446 l = DN_MAX_HASH_SIZE; 1447 x->rq_size = l; 1448 } else /* one is enough for null mask */ 1449 x->rq_size = 1; 1450 x->rq = kmalloc((1 + x->rq_size) * sizeof(struct dn_flow_queue *), 1451 M_DUMMYNET, M_WAITOK | M_ZERO); 1452 x->rq_elements = 0; 1453 return 0 ; 1454 } 1455 1456 static void 1457 set_fs_parms(struct dn_flow_set *x, struct dn_flow_set *src) 1458 { 1459 x->flags_fs = src->flags_fs; 1460 x->qsize = src->qsize; 1461 x->plr = src->plr; 1462 x->flow_mask = src->flow_mask; 1463 if (x->flags_fs & DN_QSIZE_IS_BYTES) { 1464 if (x->qsize > 1024*1024) 1465 x->qsize = 1024*1024 ; 1466 } else { 1467 if (x->qsize == 0) 1468 x->qsize = 50 ; 1469 if (x->qsize > 100) 1470 x->qsize = 50 ; 1471 } 1472 /* configuring RED */ 1473 if ( x->flags_fs & DN_IS_RED ) 1474 config_red(src, x) ; /* XXX should check errors */ 1475 } 1476 1477 /* 1478 * setup pipe or queue parameters. 1479 */ 1480 1481 static int 1482 config_pipe(struct dn_pipe *p) 1483 { 1484 int i, s; 1485 struct dn_flow_set *pfs = &(p->fs); 1486 struct dn_flow_queue *q; 1487 1488 /* 1489 * The config program passes parameters as follows: 1490 * bw = bits/second (0 means no limits), 1491 * delay = ms, must be translated into ticks. 1492 * qsize = slots/bytes 1493 */ 1494 p->delay = ( p->delay * dn_hz ) / 1000 ; 1495 /* We need either a pipe number or a flow_set number */ 1496 if (p->pipe_nr == 0 && pfs->fs_nr == 0) 1497 return EINVAL ; 1498 if (p->pipe_nr != 0 && pfs->fs_nr != 0) 1499 return EINVAL ; 1500 if (p->pipe_nr != 0) { /* this is a pipe */ 1501 struct dn_pipe *x, *a, *b; 1502 /* locate pipe */ 1503 for (a = NULL , b = all_pipes ; b && b->pipe_nr < p->pipe_nr ; 1504 a = b , b = b->next) ; 1505 1506 if (b == NULL || b->pipe_nr != p->pipe_nr) { /* new pipe */ 1507 x = kmalloc(sizeof(struct dn_pipe), M_DUMMYNET, M_WAITOK | M_ZERO); 1508 x->pipe_nr = p->pipe_nr; 1509 x->fs.pipe = x ; 1510 /* idle_heap is the only one from which we extract from the middle. 1511 */ 1512 x->idle_heap.size = x->idle_heap.elements = 0 ; 1513 x->idle_heap.offset = __offsetof(struct dn_flow_queue, heap_pos); 1514 } else { 1515 x = b; 1516 crit_enter(); 1517 /* Flush accumulated credit for all queues */ 1518 for (i = 0; i <= x->fs.rq_size; i++) 1519 for (q = x->fs.rq[i]; q; q = q->next) 1520 q->numbytes = 0; 1521 crit_exit(); 1522 } 1523 1524 crit_enter(); 1525 x->bandwidth = p->bandwidth ; 1526 x->numbytes = 0; /* just in case... */ 1527 bcopy(p->if_name, x->if_name, sizeof(p->if_name) ); 1528 x->ifp = NULL ; /* reset interface ptr */ 1529 x->delay = p->delay ; 1530 set_fs_parms(&(x->fs), pfs); 1531 1532 1533 if ( x->fs.rq == NULL ) { /* a new pipe */ 1534 s = alloc_hash(&(x->fs), pfs) ; 1535 if (s) { 1536 kfree(x, M_DUMMYNET); 1537 return s ; 1538 } 1539 x->next = b ; 1540 if (a == NULL) 1541 all_pipes = x ; 1542 else 1543 a->next = x ; 1544 } 1545 crit_exit(); 1546 } else { /* config queue */ 1547 struct dn_flow_set *x, *a, *b ; 1548 1549 /* locate flow_set */ 1550 for (a=NULL, b=all_flow_sets ; b && b->fs_nr < pfs->fs_nr ; 1551 a = b , b = b->next) ; 1552 1553 if (b == NULL || b->fs_nr != pfs->fs_nr) { /* new */ 1554 if (pfs->parent_nr == 0) /* need link to a pipe */ 1555 return EINVAL ; 1556 x = kmalloc(sizeof(struct dn_flow_set), M_DUMMYNET, M_WAITOK|M_ZERO); 1557 x->fs_nr = pfs->fs_nr; 1558 x->parent_nr = pfs->parent_nr; 1559 x->weight = pfs->weight ; 1560 if (x->weight == 0) 1561 x->weight = 1 ; 1562 else if (x->weight > 100) 1563 x->weight = 100 ; 1564 } else { 1565 /* Change parent pipe not allowed; must delete and recreate */ 1566 if (pfs->parent_nr != 0 && b->parent_nr != pfs->parent_nr) 1567 return EINVAL ; 1568 x = b; 1569 } 1570 crit_enter(); 1571 set_fs_parms(x, pfs); 1572 1573 if ( x->rq == NULL ) { /* a new flow_set */ 1574 s = alloc_hash(x, pfs) ; 1575 if (s) { 1576 kfree(x, M_DUMMYNET); 1577 return s ; 1578 } 1579 x->next = b; 1580 if (a == NULL) 1581 all_flow_sets = x; 1582 else 1583 a->next = x; 1584 } 1585 crit_exit(); 1586 } 1587 return 0 ; 1588 } 1589 1590 /* 1591 * Helper function to remove from a heap queues which are linked to 1592 * a flow_set about to be deleted. 1593 */ 1594 static void 1595 fs_remove_from_heap(struct dn_heap *h, struct dn_flow_set *fs) 1596 { 1597 int i = 0, found = 0 ; 1598 for (; i < h->elements ;) 1599 if ( ((struct dn_flow_queue *)h->p[i].object)->fs == fs) { 1600 h->elements-- ; 1601 h->p[i] = h->p[h->elements] ; 1602 found++ ; 1603 } else 1604 i++ ; 1605 if (found) 1606 heapify(h); 1607 } 1608 1609 /* 1610 * helper function to remove a pipe from a heap (can be there at most once) 1611 */ 1612 static void 1613 pipe_remove_from_heap(struct dn_heap *h, struct dn_pipe *p) 1614 { 1615 if (h->elements > 0) { 1616 int i = 0 ; 1617 for (i=0; i < h->elements ; i++ ) { 1618 if (h->p[i].object == p) { /* found it */ 1619 h->elements-- ; 1620 h->p[i] = h->p[h->elements] ; 1621 heapify(h); 1622 break ; 1623 } 1624 } 1625 } 1626 } 1627 1628 /* 1629 * drain all queues. Called in case of severe mbuf shortage. 1630 */ 1631 void 1632 dummynet_drain(void) 1633 { 1634 struct dn_flow_set *fs; 1635 struct dn_pipe *p; 1636 struct dn_pkt *pkt; 1637 1638 heap_free(&ready_heap); 1639 heap_free(&wfq_ready_heap); 1640 heap_free(&extract_heap); 1641 /* remove all references to this pipe from flow_sets */ 1642 for (fs = all_flow_sets; fs; fs= fs->next ) 1643 purge_flow_set(fs, 0); 1644 1645 for (p = all_pipes; p; p= p->next ) { 1646 purge_flow_set(&(p->fs), 0); 1647 for (pkt = p->head ; pkt ; ) 1648 DN_FREE_PKT(pkt) ; 1649 p->head = p->tail = NULL ; 1650 } 1651 } 1652 1653 /* 1654 * Fully delete a pipe or a queue, cleaning up associated info. 1655 */ 1656 static int 1657 delete_pipe(struct dn_pipe *p) 1658 { 1659 if (p->pipe_nr == 0 && p->fs.fs_nr == 0) 1660 return EINVAL ; 1661 if (p->pipe_nr != 0 && p->fs.fs_nr != 0) 1662 return EINVAL ; 1663 if (p->pipe_nr != 0) { /* this is an old-style pipe */ 1664 struct dn_pipe *a, *b; 1665 struct dn_flow_set *fs; 1666 1667 /* locate pipe */ 1668 for (a = NULL , b = all_pipes ; b && b->pipe_nr < p->pipe_nr ; 1669 a = b , b = b->next) ; 1670 if (b == NULL || (b->pipe_nr != p->pipe_nr) ) 1671 return EINVAL ; /* not found */ 1672 1673 crit_enter(); 1674 1675 /* unlink from list of pipes */ 1676 if (a == NULL) 1677 all_pipes = b->next ; 1678 else 1679 a->next = b->next ; 1680 /* remove references to this pipe from the ip_fw rules. */ 1681 flush_pipe_ptrs(&(b->fs)); 1682 1683 /* remove all references to this pipe from flow_sets */ 1684 for (fs = all_flow_sets; fs; fs= fs->next ) 1685 if (fs->pipe == b) { 1686 kprintf("++ ref to pipe %d from fs %d\n", 1687 p->pipe_nr, fs->fs_nr); 1688 fs->pipe = NULL ; 1689 purge_flow_set(fs, 0); 1690 } 1691 fs_remove_from_heap(&ready_heap, &(b->fs)); 1692 purge_pipe(b); /* remove all data associated to this pipe */ 1693 /* remove reference to here from extract_heap and wfq_ready_heap */ 1694 pipe_remove_from_heap(&extract_heap, b); 1695 pipe_remove_from_heap(&wfq_ready_heap, b); 1696 crit_exit(); 1697 kfree(b, M_DUMMYNET); 1698 } else { /* this is a WF2Q queue (dn_flow_set) */ 1699 struct dn_flow_set *a, *b; 1700 1701 /* locate set */ 1702 for (a = NULL, b = all_flow_sets ; b && b->fs_nr < p->fs.fs_nr ; 1703 a = b , b = b->next) ; 1704 if (b == NULL || (b->fs_nr != p->fs.fs_nr) ) 1705 return EINVAL ; /* not found */ 1706 1707 crit_enter(); 1708 if (a == NULL) 1709 all_flow_sets = b->next ; 1710 else 1711 a->next = b->next ; 1712 /* remove references to this flow_set from the ip_fw rules. */ 1713 flush_pipe_ptrs(b); 1714 1715 if (b->pipe != NULL) { 1716 /* Update total weight on parent pipe and cleanup parent heaps */ 1717 b->pipe->sum -= b->weight * b->backlogged ; 1718 fs_remove_from_heap(&(b->pipe->not_eligible_heap), b); 1719 fs_remove_from_heap(&(b->pipe->scheduler_heap), b); 1720 #if 1 /* XXX should i remove from idle_heap as well ? */ 1721 fs_remove_from_heap(&(b->pipe->idle_heap), b); 1722 #endif 1723 } 1724 purge_flow_set(b, 1); 1725 crit_exit(); 1726 } 1727 return 0 ; 1728 } 1729 1730 /* 1731 * helper function used to copy data from kernel in DUMMYNET_GET 1732 */ 1733 static char * 1734 dn_copy_set(struct dn_flow_set *set, char *bp) 1735 { 1736 int i, copied = 0 ; 1737 struct dn_flow_queue *q, *qp = (struct dn_flow_queue *)bp; 1738 1739 for (i = 0 ; i <= set->rq_size ; i++) 1740 for (q = set->rq[i] ; q ; q = q->next, qp++ ) { 1741 if (q->hash_slot != i) 1742 kprintf("++ at %d: wrong slot (have %d, " 1743 "should be %d)\n", copied, q->hash_slot, i); 1744 if (q->fs != set) 1745 kprintf("++ at %d: wrong fs ptr (have %p, should be %p)\n", 1746 i, q->fs, set); 1747 copied++ ; 1748 bcopy(q, qp, sizeof( *q ) ); 1749 /* cleanup pointers */ 1750 qp->next = NULL ; 1751 qp->head = qp->tail = NULL ; 1752 qp->fs = NULL ; 1753 } 1754 if (copied != set->rq_elements) 1755 kprintf("++ wrong count, have %d should be %d\n", 1756 copied, set->rq_elements); 1757 return (char *)qp ; 1758 } 1759 1760 static int 1761 dummynet_get(struct sockopt *sopt) 1762 { 1763 char *buf, *bp ; /* bp is the "copy-pointer" */ 1764 size_t size ; 1765 struct dn_flow_set *set ; 1766 struct dn_pipe *p ; 1767 int error=0 ; 1768 1769 crit_enter(); 1770 /* 1771 * compute size of data structures: list of pipes and flow_sets. 1772 */ 1773 for (p = all_pipes, size = 0 ; p ; p = p->next ) 1774 size += sizeof( *p ) + 1775 p->fs.rq_elements * sizeof(struct dn_flow_queue); 1776 for (set = all_flow_sets ; set ; set = set->next ) 1777 size += sizeof ( *set ) + 1778 set->rq_elements * sizeof(struct dn_flow_queue); 1779 buf = kmalloc(size, M_TEMP, M_WAITOK); 1780 for (p = all_pipes, bp = buf ; p ; p = p->next ) { 1781 struct dn_pipe *pipe_bp = (struct dn_pipe *)bp ; 1782 1783 /* 1784 * copy pipe descriptor into *bp, convert delay back to ms, 1785 * then copy the flow_set descriptor(s) one at a time. 1786 * After each flow_set, copy the queue descriptor it owns. 1787 */ 1788 bcopy(p, bp, sizeof( *p ) ); 1789 pipe_bp->delay = (pipe_bp->delay * 1000) / dn_hz ; 1790 /* 1791 * XXX the following is a hack based on ->next being the 1792 * first field in dn_pipe and dn_flow_set. The correct 1793 * solution would be to move the dn_flow_set to the beginning 1794 * of struct dn_pipe. 1795 */ 1796 pipe_bp->next = (struct dn_pipe *)DN_IS_PIPE ; 1797 /* clean pointers */ 1798 pipe_bp->head = pipe_bp->tail = NULL ; 1799 pipe_bp->fs.next = NULL ; 1800 pipe_bp->fs.pipe = NULL ; 1801 pipe_bp->fs.rq = NULL ; 1802 1803 bp += sizeof( *p ) ; 1804 bp = dn_copy_set( &(p->fs), bp ); 1805 } 1806 for (set = all_flow_sets ; set ; set = set->next ) { 1807 struct dn_flow_set *fs_bp = (struct dn_flow_set *)bp ; 1808 bcopy(set, bp, sizeof( *set ) ); 1809 /* XXX same hack as above */ 1810 fs_bp->next = (struct dn_flow_set *)DN_IS_QUEUE ; 1811 fs_bp->pipe = NULL ; 1812 fs_bp->rq = NULL ; 1813 bp += sizeof( *set ) ; 1814 bp = dn_copy_set( set, bp ); 1815 } 1816 crit_exit(); 1817 error = sooptcopyout(sopt, buf, size); 1818 kfree(buf, M_TEMP); 1819 return error ; 1820 } 1821 1822 /* 1823 * Handler for the various dummynet socket options (get, flush, config, del) 1824 */ 1825 static int 1826 ip_dn_ctl(struct sockopt *sopt) 1827 { 1828 int error = 0 ; 1829 struct dn_pipe *p, tmp_pipe; 1830 1831 /* Disallow sets in really-really secure mode. */ 1832 if (sopt->sopt_dir == SOPT_SET) { 1833 #if defined(__FreeBSD__) && __FreeBSD_version >= 500034 1834 error = securelevel_ge(sopt->sopt_td->td_ucred, 3); 1835 if (error) 1836 return (error); 1837 #else 1838 if (securelevel >= 3) 1839 return (EPERM); 1840 #endif 1841 } 1842 1843 switch (sopt->sopt_name) { 1844 default : 1845 kprintf("ip_dn_ctl -- unknown option %d", sopt->sopt_name); 1846 return EINVAL ; 1847 1848 case IP_DUMMYNET_GET : 1849 error = dummynet_get(sopt); 1850 break ; 1851 1852 case IP_DUMMYNET_FLUSH : 1853 dummynet_flush() ; 1854 break ; 1855 1856 case IP_DUMMYNET_CONFIGURE : 1857 p = &tmp_pipe ; 1858 error = sooptcopyin(sopt, p, sizeof *p, sizeof *p); 1859 if (error) 1860 break ; 1861 error = config_pipe(p); 1862 break ; 1863 1864 case IP_DUMMYNET_DEL : /* remove a pipe or queue */ 1865 p = &tmp_pipe ; 1866 error = sooptcopyin(sopt, p, sizeof *p, sizeof *p); 1867 if (error) 1868 break ; 1869 1870 error = delete_pipe(p); 1871 break ; 1872 } 1873 return error ; 1874 } 1875 1876 static void 1877 dummynet_clock(systimer_t info __unused, struct intrframe *frame __unused) 1878 { 1879 KASSERT(mycpu->gd_cpuid == dn_cpu, 1880 ("systimer comes on a different cpu!\n")); 1881 1882 crit_enter(); 1883 if (dn_netmsg.nm_lmsg.ms_flags & MSGF_DONE) 1884 lwkt_sendmsg(cpu_portfn(mycpu->gd_cpuid), &dn_netmsg.nm_lmsg); 1885 crit_exit(); 1886 } 1887 1888 static int 1889 sysctl_dn_hz(SYSCTL_HANDLER_ARGS) 1890 { 1891 int error, val; 1892 1893 val = dn_hz; 1894 error = sysctl_handle_int(oidp, &val, 0, req); 1895 if (error || req->newptr == NULL) 1896 return error; 1897 if (val <= 0) 1898 return EINVAL; 1899 else if (val > DUMMYNET_CALLOUT_FREQ_MAX) 1900 val = DUMMYNET_CALLOUT_FREQ_MAX; 1901 1902 crit_enter(); 1903 dn_hz = val; 1904 systimer_adjust_periodic(&dn_clock, val); 1905 crit_exit(); 1906 1907 return 0; 1908 } 1909 1910 static void 1911 ip_dn_register_systimer(struct netmsg *msg) 1912 { 1913 systimer_init_periodic_nq(&dn_clock, dummynet_clock, NULL, dn_hz); 1914 lwkt_replymsg(&msg->nm_lmsg, 0); 1915 } 1916 1917 static void 1918 ip_dn_deregister_systimer(struct netmsg *msg) 1919 { 1920 systimer_del(&dn_clock); 1921 lwkt_replymsg(&msg->nm_lmsg, 0); 1922 } 1923 1924 static void 1925 ip_dn_init(void) 1926 { 1927 struct netmsg smsg; 1928 lwkt_port_t port; 1929 1930 kprintf("DUMMYNET initialized (011031)\n"); 1931 all_pipes = NULL ; 1932 all_flow_sets = NULL ; 1933 ready_heap.size = ready_heap.elements = 0 ; 1934 ready_heap.offset = 0 ; 1935 1936 wfq_ready_heap.size = wfq_ready_heap.elements = 0 ; 1937 wfq_ready_heap.offset = 0 ; 1938 1939 extract_heap.size = extract_heap.elements = 0 ; 1940 extract_heap.offset = 0 ; 1941 ip_dn_ctl_ptr = ip_dn_ctl; 1942 ip_dn_io_ptr = dummynet_io; 1943 ip_dn_ruledel_ptr = dn_rule_delete; 1944 1945 netmsg_init(&dn_netmsg, &netisr_adone_rport, 0, dummynet); 1946 1947 netmsg_init(&smsg, &curthread->td_msgport, 0, ip_dn_register_systimer); 1948 port = cpu_portfn(dn_cpu); 1949 lwkt_domsg(port, &smsg.nm_lmsg, 0); 1950 } 1951 1952 static void 1953 ip_dn_stop(void) 1954 { 1955 struct netmsg smsg; 1956 lwkt_port_t port; 1957 1958 netmsg_init(&smsg, &curthread->td_msgport, 0, ip_dn_deregister_systimer); 1959 port = cpu_portfn(dn_cpu); 1960 lwkt_domsg(port, &smsg.nm_lmsg, 0); 1961 1962 dummynet_flush(); 1963 ip_dn_ctl_ptr = NULL; 1964 ip_dn_io_ptr = NULL; 1965 ip_dn_ruledel_ptr = NULL; 1966 1967 netmsg_service_sync(); 1968 } 1969 1970 static int 1971 dummynet_modevent(module_t mod, int type, void *data) 1972 { 1973 switch (type) { 1974 case MOD_LOAD: 1975 crit_enter(); 1976 if (DUMMYNET_LOADED) { 1977 crit_exit(); 1978 kprintf("DUMMYNET already loaded\n"); 1979 return EEXIST ; 1980 } 1981 ip_dn_init(); 1982 crit_exit(); 1983 break; 1984 1985 case MOD_UNLOAD: 1986 #if !defined(KLD_MODULE) 1987 kprintf("dummynet statically compiled, cannot unload\n"); 1988 return EINVAL ; 1989 #else 1990 crit_enter(); 1991 ip_dn_stop(); 1992 crit_exit(); 1993 #endif 1994 break ; 1995 default: 1996 break ; 1997 } 1998 return 0 ; 1999 } 2000 2001 static moduledata_t dummynet_mod = { 2002 "dummynet", 2003 dummynet_modevent, 2004 NULL 2005 }; 2006 DECLARE_MODULE(dummynet, dummynet_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 2007 MODULE_DEPEND(dummynet, ipfw, 1, 1, 1); 2008 MODULE_VERSION(dummynet, 1); 2009