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 */ 29 30 #include "opt_ipdn.h" 31 32 /* 33 * This module implements IP dummynet, a bandwidth limiter/delay emulator. 34 * Description of the data structures used is in ip_dummynet.h 35 * Here you mainly find the following blocks of code: 36 * + variable declarations; 37 * + heap management functions; 38 * + scheduler and dummynet functions; 39 * + configuration and initialization. 40 * 41 * Most important Changes: 42 * 43 * 011004: KLDable 44 * 010124: Fixed WF2Q behaviour 45 * 010122: Fixed spl protection. 46 * 000601: WF2Q support 47 * 000106: Large rewrite, use heaps to handle very many pipes. 48 * 980513: Initial release 49 */ 50 51 #include <sys/param.h> 52 #include <sys/kernel.h> 53 #include <sys/malloc.h> 54 #include <sys/mbuf.h> 55 #include <sys/socketvar.h> 56 #include <sys/sysctl.h> 57 #include <sys/systimer.h> 58 #include <sys/thread2.h> 59 60 #include <net/ethernet.h> 61 #include <net/netmsg2.h> 62 #include <net/netisr2.h> 63 #include <net/route.h> 64 65 #include <net/if.h> 66 #include <netinet/in_var.h> 67 #include <netinet/ip_var.h> 68 69 #include <net/dummynet/ip_dummynet.h> 70 71 #ifdef DUMMYNET_DEBUG 72 #define DPRINTF(fmt, ...) kprintf(fmt, __VA_ARGS__) 73 #else 74 #define DPRINTF(fmt, ...) ((void)0) 75 #endif 76 77 #ifndef DN_CALLOUT_FREQ_MAX 78 #define DN_CALLOUT_FREQ_MAX 10000 79 #endif 80 81 /* 82 * The maximum/minimum hash table size for queues. 83 * These values must be a power of 2. 84 */ 85 #define DN_MIN_HASH_SIZE 4 86 #define DN_MAX_HASH_SIZE 65536 87 88 /* 89 * Some macros are used to compare key values and handle wraparounds. 90 * MAX64 returns the largest of two key values. 91 */ 92 #define DN_KEY_LT(a, b) ((int64_t)((a) - (b)) < 0) 93 #define DN_KEY_LEQ(a, b) ((int64_t)((a) - (b)) <= 0) 94 #define DN_KEY_GT(a, b) ((int64_t)((a) - (b)) > 0) 95 #define DN_KEY_GEQ(a, b) ((int64_t)((a) - (b)) >= 0) 96 #define MAX64(x, y) ((((int64_t)((y) - (x))) > 0) ? (y) : (x)) 97 98 #define DN_NR_HASH_MAX 16 99 #define DN_NR_HASH_MASK (DN_NR_HASH_MAX - 1) 100 #define DN_NR_HASH(nr) \ 101 ((((nr) >> 12) ^ ((nr) >> 8) ^ ((nr) >> 4) ^ (nr)) & DN_NR_HASH_MASK) 102 103 MALLOC_DEFINE(M_DUMMYNET, "dummynet", "dummynet heap"); 104 105 extern int ip_dn_cpu; 106 107 static dn_key curr_time = 0; /* current simulation time */ 108 static int dn_hash_size = 64; /* default hash size */ 109 static int pipe_expire = 1; /* expire queue if empty */ 110 static int dn_max_ratio = 16; /* max queues/buckets ratio */ 111 112 /* 113 * Statistics on number of queue searches and search steps 114 */ 115 static int searches; 116 static int search_steps; 117 118 /* 119 * RED parameters 120 */ 121 static int red_lookup_depth = 256; /* default lookup table depth */ 122 static int red_avg_pkt_size = 512; /* default medium packet size */ 123 static int red_max_pkt_size = 1500;/* default max packet size */ 124 125 /* 126 * Three heaps contain queues and pipes that the scheduler handles: 127 * 128 * + ready_heap contains all dn_flow_queue related to fixed-rate pipes. 129 * + wfq_ready_heap contains the pipes associated with WF2Q flows 130 * + extract_heap contains pipes associated with delay lines. 131 */ 132 static struct dn_heap ready_heap; 133 static struct dn_heap extract_heap; 134 static struct dn_heap wfq_ready_heap; 135 136 static struct dn_pipe_head pipe_table[DN_NR_HASH_MAX]; 137 static struct dn_flowset_head flowset_table[DN_NR_HASH_MAX]; 138 139 /* 140 * Variables for dummynet systimer 141 */ 142 static struct netmsg_base dn_netmsg; 143 static struct systimer dn_clock; 144 static int dn_hz = 1000; 145 146 static int sysctl_dn_hz(SYSCTL_HANDLER_ARGS); 147 148 SYSCTL_DECL(_net_inet_ip_dummynet); 149 150 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, hash_size, CTLFLAG_RW, 151 &dn_hash_size, 0, "Default hash table size"); 152 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, curr_time, CTLFLAG_RD, 153 &curr_time, 0, "Current tick"); 154 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, expire, CTLFLAG_RW, 155 &pipe_expire, 0, "Expire queue if empty"); 156 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, max_chain_len, CTLFLAG_RW, 157 &dn_max_ratio, 0, "Max ratio between dynamic queues and buckets"); 158 159 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, ready_heap, CTLFLAG_RD, 160 &ready_heap.size, 0, "Size of ready heap"); 161 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, extract_heap, CTLFLAG_RD, 162 &extract_heap.size, 0, "Size of extract heap"); 163 164 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, searches, CTLFLAG_RD, 165 &searches, 0, "Number of queue searches"); 166 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, search_steps, CTLFLAG_RD, 167 &search_steps, 0, "Number of queue search steps"); 168 169 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_lookup_depth, CTLFLAG_RD, 170 &red_lookup_depth, 0, "Depth of RED lookup table"); 171 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_avg_pkt_size, CTLFLAG_RD, 172 &red_avg_pkt_size, 0, "RED Medium packet size"); 173 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_max_pkt_size, CTLFLAG_RD, 174 &red_max_pkt_size, 0, "RED Max packet size"); 175 176 SYSCTL_PROC(_net_inet_ip_dummynet, OID_AUTO, hz, CTLTYPE_INT | CTLFLAG_RW, 177 0, 0, sysctl_dn_hz, "I", "Dummynet callout frequency"); 178 179 static int heap_init(struct dn_heap *, int); 180 static int heap_insert(struct dn_heap *, dn_key, void *); 181 static void heap_extract(struct dn_heap *, void *); 182 183 static void transmit_event(struct dn_pipe *); 184 static void ready_event(struct dn_flow_queue *); 185 static void ready_event_wfq(struct dn_pipe *); 186 187 static int config_pipe(struct dn_ioc_pipe *); 188 static void dummynet_flush(void); 189 190 static void dummynet_clock(systimer_t, int, struct intrframe *); 191 static void dummynet(netmsg_t); 192 193 static struct dn_pipe *dn_find_pipe(int); 194 static struct dn_flow_set *dn_locate_flowset(int, int); 195 196 typedef void (*dn_pipe_iter_t)(struct dn_pipe *, void *); 197 static void dn_iterate_pipe(dn_pipe_iter_t, void *); 198 199 typedef void (*dn_flowset_iter_t)(struct dn_flow_set *, void *); 200 static void dn_iterate_flowset(dn_flowset_iter_t, void *); 201 202 static ip_dn_io_t dummynet_io; 203 static ip_dn_ctl_t dummynet_ctl; 204 205 /* 206 * Heap management functions. 207 * 208 * In the heap, first node is element 0. Children of i are 2i+1 and 2i+2. 209 * Some macros help finding parent/children so we can optimize them. 210 * 211 * heap_init() is called to expand the heap when needed. 212 * Increment size in blocks of 16 entries. 213 * XXX failure to allocate a new element is a pretty bad failure 214 * as we basically stall a whole queue forever!! 215 * Returns 1 on error, 0 on success 216 */ 217 #define HEAP_FATHER(x) (((x) - 1) / 2) 218 #define HEAP_LEFT(x) (2*(x) + 1) 219 #define HEAP_IS_LEFT(x) ((x) & 1) 220 #define HEAP_RIGHT(x) (2*(x) + 2) 221 #define HEAP_SWAP(a, b, buffer) { buffer = a; a = b; b = buffer; } 222 #define HEAP_INCREMENT 15 223 224 static int 225 heap_init(struct dn_heap *h, int new_size) 226 { 227 struct dn_heap_entry *p; 228 229 if (h->size >= new_size) { 230 kprintf("%s, Bogus call, have %d want %d\n", __func__, 231 h->size, new_size); 232 return 0; 233 } 234 235 new_size = (new_size + HEAP_INCREMENT) & ~HEAP_INCREMENT; 236 p = kmalloc(new_size * sizeof(*p), M_DUMMYNET, M_WAITOK | M_ZERO); 237 if (h->size > 0) { 238 bcopy(h->p, p, h->size * sizeof(*p)); 239 kfree(h->p, M_DUMMYNET); 240 } 241 h->p = p; 242 h->size = new_size; 243 return 0; 244 } 245 246 /* 247 * Insert element in heap. Normally, p != NULL, we insert p in 248 * a new position and bubble up. If p == NULL, then the element is 249 * already in place, and key is the position where to start the 250 * bubble-up. 251 * Returns 1 on failure (cannot allocate new heap entry) 252 * 253 * If offset > 0 the position (index, int) of the element in the heap is 254 * also stored in the element itself at the given offset in bytes. 255 */ 256 #define SET_OFFSET(heap, node) \ 257 if (heap->offset > 0) \ 258 *((int *)((char *)(heap->p[node].object) + heap->offset)) = node; 259 260 /* 261 * RESET_OFFSET is used for sanity checks. It sets offset to an invalid value. 262 */ 263 #define RESET_OFFSET(heap, node) \ 264 if (heap->offset > 0) \ 265 *((int *)((char *)(heap->p[node].object) + heap->offset)) = -1; 266 267 static int 268 heap_insert(struct dn_heap *h, dn_key key1, void *p) 269 { 270 int son; 271 272 if (p == NULL) { /* Data already there, set starting point */ 273 son = key1; 274 } else { /* Insert new element at the end, possibly resize */ 275 son = h->elements; 276 if (son == h->size) { /* Need resize... */ 277 if (heap_init(h, h->elements + 1)) 278 return 1; /* Failure... */ 279 } 280 h->p[son].object = p; 281 h->p[son].key = key1; 282 h->elements++; 283 } 284 285 while (son > 0) { /* Bubble up */ 286 int father = HEAP_FATHER(son); 287 struct dn_heap_entry tmp; 288 289 if (DN_KEY_LT(h->p[father].key, h->p[son].key)) 290 break; /* Found right position */ 291 292 /* 'son' smaller than 'father', swap and repeat */ 293 HEAP_SWAP(h->p[son], h->p[father], tmp); 294 SET_OFFSET(h, son); 295 son = father; 296 } 297 SET_OFFSET(h, son); 298 return 0; 299 } 300 301 /* 302 * Remove top element from heap, or obj if obj != NULL 303 */ 304 static void 305 heap_extract(struct dn_heap *h, void *obj) 306 { 307 int child, father, max = h->elements - 1; 308 309 if (max < 0) { 310 kprintf("warning, extract from empty heap 0x%p\n", h); 311 return; 312 } 313 314 father = 0; /* Default: move up smallest child */ 315 if (obj != NULL) { /* Extract specific element, index is at offset */ 316 if (h->offset <= 0) 317 panic("%s from middle not supported on this heap!!!", __func__); 318 319 father = *((int *)((char *)obj + h->offset)); 320 if (father < 0 || father >= h->elements) { 321 panic("%s father %d out of bound 0..%d", __func__, 322 father, h->elements); 323 } 324 } 325 RESET_OFFSET(h, father); 326 327 child = HEAP_LEFT(father); /* Left child */ 328 while (child <= max) { /* Valid entry */ 329 if (child != max && DN_KEY_LT(h->p[child + 1].key, h->p[child].key)) 330 child = child + 1; /* Take right child, otherwise left */ 331 h->p[father] = h->p[child]; 332 SET_OFFSET(h, father); 333 father = child; 334 child = HEAP_LEFT(child); /* Left child for next loop */ 335 } 336 h->elements--; 337 if (father != max) { 338 /* 339 * Fill hole with last entry and bubble up, reusing the insert code 340 */ 341 h->p[father] = h->p[max]; 342 heap_insert(h, father, NULL); /* This one cannot fail */ 343 } 344 } 345 346 /* 347 * heapify() will reorganize data inside an array to maintain the 348 * heap property. It is needed when we delete a bunch of entries. 349 */ 350 static void 351 heapify(struct dn_heap *h) 352 { 353 int i; 354 355 for (i = 0; i < h->elements; i++) 356 heap_insert(h, i , NULL); 357 } 358 359 /* 360 * Cleanup the heap and free data structure 361 */ 362 static void 363 heap_free(struct dn_heap *h) 364 { 365 if (h->size > 0) 366 kfree(h->p, M_DUMMYNET); 367 bzero(h, sizeof(*h)); 368 } 369 370 /* 371 * --- End of heap management functions --- 372 */ 373 374 /* 375 * Scheduler functions: 376 * 377 * transmit_event() is called when the delay-line needs to enter 378 * the scheduler, either because of existing pkts getting ready, 379 * or new packets entering the queue. The event handled is the delivery 380 * time of the packet. 381 * 382 * ready_event() does something similar with fixed-rate queues, and the 383 * event handled is the finish time of the head pkt. 384 * 385 * ready_event_wfq() does something similar with WF2Q queues, and the 386 * event handled is the start time of the head pkt. 387 * 388 * In all cases, we make sure that the data structures are consistent 389 * before passing pkts out, because this might trigger recursive 390 * invocations of the procedures. 391 */ 392 static void 393 transmit_event(struct dn_pipe *pipe) 394 { 395 struct dn_pkt *pkt; 396 397 while ((pkt = TAILQ_FIRST(&pipe->p_queue)) && 398 DN_KEY_LEQ(pkt->output_time, curr_time)) { 399 TAILQ_REMOVE(&pipe->p_queue, pkt, dn_next); 400 ip_dn_packet_redispatch(pkt); 401 } 402 403 /* 404 * If there are leftover packets, put into the heap for next event 405 */ 406 if ((pkt = TAILQ_FIRST(&pipe->p_queue)) != NULL) { 407 /* 408 * XXX should check errors on heap_insert, by draining the 409 * whole pipe and hoping in the future we are more successful 410 */ 411 heap_insert(&extract_heap, pkt->output_time, pipe); 412 } 413 } 414 415 /* 416 * The following macro computes how many ticks we have to wait 417 * before being able to transmit a packet. The credit is taken from 418 * either a pipe (WF2Q) or a flow_queue (per-flow queueing) 419 */ 420 #define SET_TICKS(pkt, q, p) \ 421 (pkt->dn_m->m_pkthdr.len*8*dn_hz - (q)->numbytes + p->bandwidth - 1 ) / \ 422 p->bandwidth; 423 424 /* 425 * Extract pkt from queue, compute output time (could be now) 426 * and put into delay line (p_queue) 427 */ 428 static void 429 move_pkt(struct dn_pkt *pkt, struct dn_flow_queue *q, 430 struct dn_pipe *p, int len) 431 { 432 TAILQ_REMOVE(&q->queue, pkt, dn_next); 433 q->len--; 434 q->len_bytes -= len; 435 436 pkt->output_time = curr_time + p->delay; 437 438 TAILQ_INSERT_TAIL(&p->p_queue, pkt, dn_next); 439 } 440 441 /* 442 * ready_event() is invoked every time the queue must enter the 443 * scheduler, either because the first packet arrives, or because 444 * a previously scheduled event fired. 445 * On invokation, drain as many pkts as possible (could be 0) and then 446 * if there are leftover packets reinsert the pkt in the scheduler. 447 */ 448 static void 449 ready_event(struct dn_flow_queue *q) 450 { 451 struct dn_pkt *pkt; 452 struct dn_pipe *p = q->fs->pipe; 453 int p_was_empty; 454 455 if (p == NULL) { 456 kprintf("ready_event- pipe is gone\n"); 457 return; 458 } 459 p_was_empty = TAILQ_EMPTY(&p->p_queue); 460 461 /* 462 * Schedule fixed-rate queues linked to this pipe: 463 * Account for the bw accumulated since last scheduling, then 464 * drain as many pkts as allowed by q->numbytes and move to 465 * the delay line (in p) computing output time. 466 * bandwidth==0 (no limit) means we can drain the whole queue, 467 * setting len_scaled = 0 does the job. 468 */ 469 q->numbytes += (curr_time - q->sched_time) * p->bandwidth; 470 while ((pkt = TAILQ_FIRST(&q->queue)) != NULL) { 471 int len = pkt->dn_m->m_pkthdr.len; 472 int len_scaled = p->bandwidth ? len*8*dn_hz : 0; 473 474 if (len_scaled > q->numbytes) 475 break; 476 q->numbytes -= len_scaled; 477 move_pkt(pkt, q, p, len); 478 } 479 480 /* 481 * If we have more packets queued, schedule next ready event 482 * (can only occur when bandwidth != 0, otherwise we would have 483 * flushed the whole queue in the previous loop). 484 * To this purpose we record the current time and compute how many 485 * ticks to go for the finish time of the packet. 486 */ 487 if ((pkt = TAILQ_FIRST(&q->queue)) != NULL) { 488 /* This implies bandwidth != 0 */ 489 dn_key t = SET_TICKS(pkt, q, p); /* ticks i have to wait */ 490 491 q->sched_time = curr_time; 492 493 /* 494 * XXX should check errors on heap_insert, and drain the whole 495 * queue on error hoping next time we are luckier. 496 */ 497 heap_insert(&ready_heap, curr_time + t, q); 498 } else { /* RED needs to know when the queue becomes empty */ 499 q->q_time = curr_time; 500 q->numbytes = 0; 501 } 502 503 /* 504 * If the delay line was empty call transmit_event(p) now. 505 * Otherwise, the scheduler will take care of it. 506 */ 507 if (p_was_empty) 508 transmit_event(p); 509 } 510 511 /* 512 * Called when we can transmit packets on WF2Q queues. Take pkts out of 513 * the queues at their start time, and enqueue into the delay line. 514 * Packets are drained until p->numbytes < 0. As long as 515 * len_scaled >= p->numbytes, the packet goes into the delay line 516 * with a deadline p->delay. For the last packet, if p->numbytes < 0, 517 * there is an additional delay. 518 */ 519 static void 520 ready_event_wfq(struct dn_pipe *p) 521 { 522 int p_was_empty = TAILQ_EMPTY(&p->p_queue); 523 struct dn_heap *sch = &p->scheduler_heap; 524 struct dn_heap *neh = &p->not_eligible_heap; 525 526 p->numbytes += (curr_time - p->sched_time) * p->bandwidth; 527 528 /* 529 * While we have backlogged traffic AND credit, we need to do 530 * something on the queue. 531 */ 532 while (p->numbytes >= 0 && (sch->elements > 0 || neh->elements > 0)) { 533 if (sch->elements > 0) { /* Have some eligible pkts to send out */ 534 struct dn_flow_queue *q = sch->p[0].object; 535 struct dn_pkt *pkt = TAILQ_FIRST(&q->queue); 536 struct dn_flow_set *fs = q->fs; 537 uint64_t len = pkt->dn_m->m_pkthdr.len; 538 int len_scaled = p->bandwidth ? len*8*dn_hz : 0; 539 540 heap_extract(sch, NULL); /* Remove queue from heap */ 541 p->numbytes -= len_scaled; 542 move_pkt(pkt, q, p, len); 543 544 p->V += (len << MY_M) / p->sum; /* Update V */ 545 q->S = q->F; /* Update start time */ 546 547 if (q->len == 0) { /* Flow not backlogged any more */ 548 fs->backlogged--; 549 heap_insert(&p->idle_heap, q->F, q); 550 } else { /* Still backlogged */ 551 /* 552 * Update F and position in backlogged queue, then 553 * put flow in not_eligible_heap (we will fix this later). 554 */ 555 len = TAILQ_FIRST(&q->queue)->dn_m->m_pkthdr.len; 556 q->F += (len << MY_M) / (uint64_t)fs->weight; 557 if (DN_KEY_LEQ(q->S, p->V)) 558 heap_insert(neh, q->S, q); 559 else 560 heap_insert(sch, q->F, q); 561 } 562 } 563 564 /* 565 * Now compute V = max(V, min(S_i)). Remember that all elements in 566 * sch have by definition S_i <= V so if sch is not empty, V is surely 567 * the max and we must not update it. Conversely, if sch is empty 568 * we only need to look at neh. 569 */ 570 if (sch->elements == 0 && neh->elements > 0) 571 p->V = MAX64(p->V, neh->p[0].key); 572 573 /* 574 * Move from neh to sch any packets that have become eligible 575 */ 576 while (neh->elements > 0 && DN_KEY_LEQ(neh->p[0].key, p->V)) { 577 struct dn_flow_queue *q = neh->p[0].object; 578 579 heap_extract(neh, NULL); 580 heap_insert(sch, q->F, q); 581 } 582 } 583 584 if (sch->elements == 0 && neh->elements == 0 && p->numbytes >= 0 && 585 p->idle_heap.elements > 0) { 586 /* 587 * No traffic and no events scheduled. We can get rid of idle-heap. 588 */ 589 int i; 590 591 for (i = 0; i < p->idle_heap.elements; i++) { 592 struct dn_flow_queue *q = p->idle_heap.p[i].object; 593 594 q->F = 0; 595 q->S = q->F + 1; 596 } 597 p->sum = 0; 598 p->V = 0; 599 p->idle_heap.elements = 0; 600 } 601 602 /* 603 * If we are getting clocks from dummynet and if we are under credit, 604 * schedule the next ready event. 605 * Also fix the delivery time of the last packet. 606 */ 607 if (p->numbytes < 0) { /* This implies bandwidth>0 */ 608 dn_key t = 0; /* Number of ticks i have to wait */ 609 610 if (p->bandwidth > 0) 611 t = (p->bandwidth - 1 - p->numbytes) / p->bandwidth; 612 TAILQ_LAST(&p->p_queue, dn_pkt_queue)->output_time += t; 613 p->sched_time = curr_time; 614 615 /* 616 * XXX should check errors on heap_insert, and drain the whole 617 * queue on error hoping next time we are luckier. 618 */ 619 heap_insert(&wfq_ready_heap, curr_time + t, p); 620 } 621 622 /* 623 * If the delay line was empty call transmit_event(p) now. 624 * Otherwise, the scheduler will take care of it. 625 */ 626 if (p_was_empty) 627 transmit_event(p); 628 } 629 630 static void 631 dn_expire_pipe_cb(struct dn_pipe *pipe, void *dummy __unused) 632 { 633 if (pipe->idle_heap.elements > 0 && 634 DN_KEY_LT(pipe->idle_heap.p[0].key, pipe->V)) { 635 struct dn_flow_queue *q = pipe->idle_heap.p[0].object; 636 637 heap_extract(&pipe->idle_heap, NULL); 638 q->S = q->F + 1; /* Mark timestamp as invalid */ 639 pipe->sum -= q->fs->weight; 640 } 641 } 642 643 /* 644 * This is called once per tick, or dn_hz times per second. It is used to 645 * increment the current tick counter and schedule expired events. 646 */ 647 static void 648 dummynet(netmsg_t msg) 649 { 650 void *p; 651 struct dn_heap *h; 652 struct dn_heap *heaps[3]; 653 int i; 654 655 heaps[0] = &ready_heap; /* Fixed-rate queues */ 656 heaps[1] = &wfq_ready_heap; /* WF2Q queues */ 657 heaps[2] = &extract_heap; /* Delay line */ 658 659 /* Reply ASAP */ 660 crit_enter(); 661 lwkt_replymsg(&msg->lmsg, 0); 662 crit_exit(); 663 664 curr_time++; 665 for (i = 0; i < 3; i++) { 666 h = heaps[i]; 667 while (h->elements > 0 && DN_KEY_LEQ(h->p[0].key, curr_time)) { 668 if (h->p[0].key > curr_time) { 669 kprintf("-- dummynet: warning, heap %d is %d ticks late\n", 670 i, (int)(curr_time - h->p[0].key)); 671 } 672 673 p = h->p[0].object; /* Store a copy before heap_extract */ 674 heap_extract(h, NULL); /* Need to extract before processing */ 675 676 if (i == 0) 677 ready_event(p); 678 else if (i == 1) 679 ready_event_wfq(p); 680 else 681 transmit_event(p); 682 } 683 } 684 685 /* Sweep pipes trying to expire idle flow_queues */ 686 dn_iterate_pipe(dn_expire_pipe_cb, NULL); 687 } 688 689 /* 690 * Unconditionally expire empty queues in case of shortage. 691 * Returns the number of queues freed. 692 */ 693 static int 694 expire_queues(struct dn_flow_set *fs) 695 { 696 int i, initial_elements = fs->rq_elements; 697 698 if (fs->last_expired == time_uptime) 699 return 0; 700 701 fs->last_expired = time_uptime; 702 703 for (i = 0; i <= fs->rq_size; i++) { /* Last one is overflow */ 704 struct dn_flow_queue *q, *qn; 705 706 LIST_FOREACH_MUTABLE(q, &fs->rq[i], q_link, qn) { 707 if (!TAILQ_EMPTY(&q->queue) || q->S != q->F + 1) 708 continue; 709 710 /* 711 * Entry is idle, expire it 712 */ 713 LIST_REMOVE(q, q_link); 714 kfree(q, M_DUMMYNET); 715 716 KASSERT(fs->rq_elements > 0, 717 ("invalid rq_elements %d", fs->rq_elements)); 718 fs->rq_elements--; 719 } 720 } 721 return initial_elements - fs->rq_elements; 722 } 723 724 /* 725 * If room, create a new queue and put at head of slot i; 726 * otherwise, create or use the default queue. 727 */ 728 static struct dn_flow_queue * 729 create_queue(struct dn_flow_set *fs, int i) 730 { 731 struct dn_flow_queue *q; 732 733 if (fs->rq_elements > fs->rq_size * dn_max_ratio && 734 expire_queues(fs) == 0) { 735 /* 736 * No way to get room, use or create overflow queue. 737 */ 738 i = fs->rq_size; 739 if (!LIST_EMPTY(&fs->rq[i])) 740 return LIST_FIRST(&fs->rq[i]); 741 } 742 743 q = kmalloc(sizeof(*q), M_DUMMYNET, M_INTWAIT | M_NULLOK | M_ZERO); 744 if (q == NULL) 745 return NULL; 746 747 q->fs = fs; 748 q->hash_slot = i; 749 q->S = q->F + 1; /* hack - mark timestamp as invalid */ 750 TAILQ_INIT(&q->queue); 751 752 LIST_INSERT_HEAD(&fs->rq[i], q, q_link); 753 fs->rq_elements++; 754 755 return q; 756 } 757 758 /* 759 * Given a flow_set and a pkt in last_pkt, find a matching queue 760 * after appropriate masking. The queue is moved to front 761 * so that further searches take less time. 762 */ 763 static struct dn_flow_queue * 764 find_queue(struct dn_flow_set *fs, struct dn_flow_id *id) 765 { 766 struct dn_flow_queue *q; 767 int i = 0; 768 769 if (!(fs->flags_fs & DN_HAVE_FLOW_MASK)) { 770 q = LIST_FIRST(&fs->rq[0]); 771 } else { 772 struct dn_flow_queue *qn; 773 774 /* First, do the masking */ 775 id->fid_dst_ip &= fs->flow_mask.fid_dst_ip; 776 id->fid_src_ip &= fs->flow_mask.fid_src_ip; 777 id->fid_dst_port &= fs->flow_mask.fid_dst_port; 778 id->fid_src_port &= fs->flow_mask.fid_src_port; 779 id->fid_proto &= fs->flow_mask.fid_proto; 780 id->fid_flags = 0; /* we don't care about this one */ 781 782 /* Then, hash function */ 783 i = ((id->fid_dst_ip) & 0xffff) ^ 784 ((id->fid_dst_ip >> 15) & 0xffff) ^ 785 ((id->fid_src_ip << 1) & 0xffff) ^ 786 ((id->fid_src_ip >> 16 ) & 0xffff) ^ 787 (id->fid_dst_port << 1) ^ (id->fid_src_port) ^ 788 (id->fid_proto); 789 i = i % fs->rq_size; 790 791 /* 792 * Finally, scan the current list for a match and 793 * expire idle flow queues 794 */ 795 searches++; 796 LIST_FOREACH_MUTABLE(q, &fs->rq[i], q_link, qn) { 797 search_steps++; 798 if (id->fid_dst_ip == q->id.fid_dst_ip && 799 id->fid_src_ip == q->id.fid_src_ip && 800 id->fid_dst_port == q->id.fid_dst_port && 801 id->fid_src_port == q->id.fid_src_port && 802 id->fid_proto == q->id.fid_proto && 803 id->fid_flags == q->id.fid_flags) { 804 break; /* Found */ 805 } else if (pipe_expire && TAILQ_EMPTY(&q->queue) && 806 q->S == q->F + 1) { 807 /* 808 * Entry is idle and not in any heap, expire it 809 */ 810 LIST_REMOVE(q, q_link); 811 kfree(q, M_DUMMYNET); 812 813 KASSERT(fs->rq_elements > 0, 814 ("invalid rq_elements %d", fs->rq_elements)); 815 fs->rq_elements--; 816 } 817 } 818 if (q && LIST_FIRST(&fs->rq[i]) != q) { /* Found and not in front */ 819 LIST_REMOVE(q, q_link); 820 LIST_INSERT_HEAD(&fs->rq[i], q, q_link); 821 } 822 } 823 if (q == NULL) { /* No match, need to allocate a new entry */ 824 q = create_queue(fs, i); 825 if (q != NULL) 826 q->id = *id; 827 } 828 return q; 829 } 830 831 static int 832 red_drops(struct dn_flow_set *fs, struct dn_flow_queue *q, int len) 833 { 834 /* 835 * RED algorithm 836 * 837 * RED calculates the average queue size (avg) using a low-pass filter 838 * with an exponential weighted (w_q) moving average: 839 * avg <- (1-w_q) * avg + w_q * q_size 840 * where q_size is the queue length (measured in bytes or * packets). 841 * 842 * If q_size == 0, we compute the idle time for the link, and set 843 * avg = (1 - w_q)^(idle/s) 844 * where s is the time needed for transmitting a medium-sized packet. 845 * 846 * Now, if avg < min_th the packet is enqueued. 847 * If avg > max_th the packet is dropped. Otherwise, the packet is 848 * dropped with probability P function of avg. 849 */ 850 851 int64_t p_b = 0; 852 u_int q_size = (fs->flags_fs & DN_QSIZE_IS_BYTES) ? q->len_bytes : q->len; 853 854 DPRINTF("\n%d q: %2u ", (int)curr_time, q_size); 855 856 /* Average queue size estimation */ 857 if (q_size != 0) { 858 /* 859 * Queue is not empty, avg <- avg + (q_size - avg) * w_q 860 */ 861 int diff = SCALE(q_size) - q->avg; 862 int64_t v = SCALE_MUL((int64_t)diff, (int64_t)fs->w_q); 863 864 q->avg += (int)v; 865 } else { 866 /* 867 * Queue is empty, find for how long the queue has been 868 * empty and use a lookup table for computing 869 * (1 - * w_q)^(idle_time/s) where s is the time to send a 870 * (small) packet. 871 * XXX check wraps... 872 */ 873 if (q->avg) { 874 u_int t = (curr_time - q->q_time) / fs->lookup_step; 875 876 q->avg = (t < fs->lookup_depth) ? 877 SCALE_MUL(q->avg, fs->w_q_lookup[t]) : 0; 878 } 879 } 880 DPRINTF("avg: %u ", SCALE_VAL(q->avg)); 881 882 /* Should i drop? */ 883 884 if (q->avg < fs->min_th) { 885 /* Accept packet */ 886 q->count = -1; 887 return 0; 888 } 889 890 if (q->avg >= fs->max_th) { /* Average queue >= Max threshold */ 891 if (fs->flags_fs & DN_IS_GENTLE_RED) { 892 /* 893 * According to Gentle-RED, if avg is greater than max_th the 894 * packet is dropped with a probability 895 * p_b = c_3 * avg - c_4 896 * where c_3 = (1 - max_p) / max_th, and c_4 = 1 - 2 * max_p 897 */ 898 p_b = SCALE_MUL((int64_t)fs->c_3, (int64_t)q->avg) - fs->c_4; 899 } else { 900 q->count = -1; 901 kprintf("- drop\n"); 902 return 1; 903 } 904 } else if (q->avg > fs->min_th) { 905 /* 906 * We compute p_b using the linear dropping function p_b = c_1 * 907 * avg - c_2, where c_1 = max_p / (max_th - min_th), and c_2 = 908 * max_p * min_th / (max_th - min_th) 909 */ 910 p_b = SCALE_MUL((int64_t)fs->c_1, (int64_t)q->avg) - fs->c_2; 911 } 912 if (fs->flags_fs & DN_QSIZE_IS_BYTES) 913 p_b = (p_b * len) / fs->max_pkt_size; 914 915 if (++q->count == 0) { 916 q->random = krandom() & 0xffff; 917 } else { 918 /* 919 * q->count counts packets arrived since last drop, so a greater 920 * value of q->count means a greater packet drop probability. 921 */ 922 if (SCALE_MUL(p_b, SCALE((int64_t)q->count)) > q->random) { 923 q->count = 0; 924 DPRINTF("%s", "- red drop"); 925 /* After a drop we calculate a new random value */ 926 q->random = krandom() & 0xffff; 927 return 1; /* Drop */ 928 } 929 } 930 /* End of RED algorithm */ 931 return 0; /* Accept */ 932 } 933 934 static void 935 dn_iterate_pipe(dn_pipe_iter_t func, void *arg) 936 { 937 int i; 938 939 for (i = 0; i < DN_NR_HASH_MAX; ++i) { 940 struct dn_pipe_head *pipe_hdr = &pipe_table[i]; 941 struct dn_pipe *pipe, *pipe_next; 942 943 LIST_FOREACH_MUTABLE(pipe, pipe_hdr, p_link, pipe_next) 944 func(pipe, arg); 945 } 946 } 947 948 static void 949 dn_iterate_flowset(dn_flowset_iter_t func, void *arg) 950 { 951 int i; 952 953 for (i = 0; i < DN_NR_HASH_MAX; ++i) { 954 struct dn_flowset_head *fs_hdr = &flowset_table[i]; 955 struct dn_flow_set *fs, *fs_next; 956 957 LIST_FOREACH_MUTABLE(fs, fs_hdr, fs_link, fs_next) 958 func(fs, arg); 959 } 960 } 961 962 static struct dn_pipe * 963 dn_find_pipe(int pipe_nr) 964 { 965 struct dn_pipe_head *pipe_hdr; 966 struct dn_pipe *p; 967 968 pipe_hdr = &pipe_table[DN_NR_HASH(pipe_nr)]; 969 LIST_FOREACH(p, pipe_hdr, p_link) { 970 if (p->pipe_nr == pipe_nr) 971 break; 972 } 973 return p; 974 } 975 976 static struct dn_flow_set * 977 dn_find_flowset(int fs_nr) 978 { 979 struct dn_flowset_head *fs_hdr; 980 struct dn_flow_set *fs; 981 982 fs_hdr = &flowset_table[DN_NR_HASH(fs_nr)]; 983 LIST_FOREACH(fs, fs_hdr, fs_link) { 984 if (fs->fs_nr == fs_nr) 985 break; 986 } 987 return fs; 988 } 989 990 static struct dn_flow_set * 991 dn_locate_flowset(int pipe_nr, int is_pipe) 992 { 993 struct dn_flow_set *fs = NULL; 994 995 if (!is_pipe) { 996 fs = dn_find_flowset(pipe_nr); 997 } else { 998 struct dn_pipe *p; 999 1000 p = dn_find_pipe(pipe_nr); 1001 if (p != NULL) 1002 fs = &p->fs; 1003 } 1004 return fs; 1005 } 1006 1007 /* 1008 * Dummynet hook for packets. Below 'pipe' is a pipe or a queue 1009 * depending on whether WF2Q or fixed bw is used. 1010 * 1011 * pipe_nr pipe or queue the packet is destined for. 1012 * dir where shall we send the packet after dummynet. 1013 * m the mbuf with the packet 1014 * fwa->oif the 'ifp' parameter from the caller. 1015 * NULL in ip_input, destination interface in ip_output 1016 * fwa->ro route parameter (only used in ip_output, NULL otherwise) 1017 * fwa->dst destination address, only used by ip_output 1018 * fwa->rule matching rule, in case of multiple passes 1019 * fwa->flags flags from the caller, only used in ip_output 1020 */ 1021 static int 1022 dummynet_io(struct mbuf *m) 1023 { 1024 struct dn_pkt *pkt; 1025 struct m_tag *tag; 1026 struct dn_flow_set *fs; 1027 struct dn_pipe *pipe; 1028 uint64_t len = m->m_pkthdr.len; 1029 struct dn_flow_queue *q = NULL; 1030 int is_pipe, pipe_nr; 1031 1032 tag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); 1033 pkt = m_tag_data(tag); 1034 1035 is_pipe = pkt->dn_flags & DN_FLAGS_IS_PIPE; 1036 pipe_nr = pkt->pipe_nr; 1037 1038 /* 1039 * This is a dummynet rule, so we expect a O_PIPE or O_QUEUE rule 1040 */ 1041 fs = dn_locate_flowset(pipe_nr, is_pipe); 1042 if (fs == NULL) 1043 goto dropit; /* This queue/pipe does not exist! */ 1044 1045 pipe = fs->pipe; 1046 if (pipe == NULL) { /* Must be a queue, try find a matching pipe */ 1047 pipe = dn_find_pipe(fs->parent_nr); 1048 if (pipe != NULL) { 1049 fs->pipe = pipe; 1050 } else { 1051 kprintf("No pipe %d for queue %d, drop pkt\n", 1052 fs->parent_nr, fs->fs_nr); 1053 goto dropit; 1054 } 1055 } 1056 1057 q = find_queue(fs, &pkt->id); 1058 if (q == NULL) 1059 goto dropit; /* Cannot allocate queue */ 1060 1061 /* 1062 * Update statistics, then check reasons to drop pkt 1063 */ 1064 q->tot_bytes += len; 1065 q->tot_pkts++; 1066 1067 if (fs->plr && krandom() < fs->plr) 1068 goto dropit; /* Random pkt drop */ 1069 1070 if (fs->flags_fs & DN_QSIZE_IS_BYTES) { 1071 if (q->len_bytes > fs->qsize) 1072 goto dropit; /* Queue size overflow */ 1073 } else { 1074 if (q->len >= fs->qsize) 1075 goto dropit; /* Queue count overflow */ 1076 } 1077 1078 if ((fs->flags_fs & DN_IS_RED) && red_drops(fs, q, len)) 1079 goto dropit; 1080 1081 TAILQ_INSERT_TAIL(&q->queue, pkt, dn_next); 1082 q->len++; 1083 q->len_bytes += len; 1084 1085 if (TAILQ_FIRST(&q->queue) != pkt) /* Flow was not idle, we are done */ 1086 goto done; 1087 1088 /* 1089 * If we reach this point the flow was previously idle, so we need 1090 * to schedule it. This involves different actions for fixed-rate 1091 * or WF2Q queues. 1092 */ 1093 if (is_pipe) { 1094 /* 1095 * Fixed-rate queue: just insert into the ready_heap. 1096 */ 1097 dn_key t = 0; 1098 1099 if (pipe->bandwidth) 1100 t = SET_TICKS(pkt, q, pipe); 1101 1102 q->sched_time = curr_time; 1103 if (t == 0) /* Must process it now */ 1104 ready_event(q); 1105 else 1106 heap_insert(&ready_heap, curr_time + t, q); 1107 } else { 1108 /* 1109 * WF2Q: 1110 * First, compute start time S: if the flow was idle (S=F+1) 1111 * set S to the virtual time V for the controlling pipe, and update 1112 * the sum of weights for the pipe; otherwise, remove flow from 1113 * idle_heap and set S to max(F, V). 1114 * Second, compute finish time F = S + len/weight. 1115 * Third, if pipe was idle, update V = max(S, V). 1116 * Fourth, count one more backlogged flow. 1117 */ 1118 if (DN_KEY_GT(q->S, q->F)) { /* Means timestamps are invalid */ 1119 q->S = pipe->V; 1120 pipe->sum += fs->weight; /* Add weight of new queue */ 1121 } else { 1122 heap_extract(&pipe->idle_heap, q); 1123 q->S = MAX64(q->F, pipe->V); 1124 } 1125 q->F = q->S + (len << MY_M) / (uint64_t)fs->weight; 1126 1127 if (pipe->not_eligible_heap.elements == 0 && 1128 pipe->scheduler_heap.elements == 0) 1129 pipe->V = MAX64(q->S, pipe->V); 1130 1131 fs->backlogged++; 1132 1133 /* 1134 * Look at eligibility. A flow is not eligibile if S>V (when 1135 * this happens, it means that there is some other flow already 1136 * scheduled for the same pipe, so the scheduler_heap cannot be 1137 * empty). If the flow is not eligible we just store it in the 1138 * not_eligible_heap. Otherwise, we store in the scheduler_heap 1139 * and possibly invoke ready_event_wfq() right now if there is 1140 * leftover credit. 1141 * Note that for all flows in scheduler_heap (SCH), S_i <= V, 1142 * and for all flows in not_eligible_heap (NEH), S_i > V. 1143 * So when we need to compute max(V, min(S_i)) forall i in SCH+NEH, 1144 * we only need to look into NEH. 1145 */ 1146 if (DN_KEY_GT(q->S, pipe->V)) { /* Not eligible */ 1147 if (pipe->scheduler_heap.elements == 0) 1148 kprintf("++ ouch! not eligible but empty scheduler!\n"); 1149 heap_insert(&pipe->not_eligible_heap, q->S, q); 1150 } else { 1151 heap_insert(&pipe->scheduler_heap, q->F, q); 1152 if (pipe->numbytes >= 0) { /* Pipe is idle */ 1153 if (pipe->scheduler_heap.elements != 1) 1154 kprintf("*** OUCH! pipe should have been idle!\n"); 1155 DPRINTF("Waking up pipe %d at %d\n", 1156 pipe->pipe_nr, (int)(q->F >> MY_M)); 1157 pipe->sched_time = curr_time; 1158 ready_event_wfq(pipe); 1159 } 1160 } 1161 } 1162 done: 1163 return 0; 1164 1165 dropit: 1166 if (q) 1167 q->drops++; 1168 return ENOBUFS; 1169 } 1170 1171 /* 1172 * Dispose all packets and flow_queues on a flow_set. 1173 * If all=1, also remove red lookup table and other storage, 1174 * including the descriptor itself. 1175 * For the one in dn_pipe MUST also cleanup ready_heap... 1176 */ 1177 static void 1178 purge_flow_set(struct dn_flow_set *fs, int all) 1179 { 1180 int i; 1181 #ifdef INVARIANTS 1182 int rq_elements = 0; 1183 #endif 1184 1185 for (i = 0; i <= fs->rq_size; i++) { 1186 struct dn_flow_queue *q; 1187 1188 while ((q = LIST_FIRST(&fs->rq[i])) != NULL) { 1189 struct dn_pkt *pkt; 1190 1191 while ((pkt = TAILQ_FIRST(&q->queue)) != NULL) { 1192 TAILQ_REMOVE(&q->queue, pkt, dn_next); 1193 ip_dn_packet_free(pkt); 1194 } 1195 1196 LIST_REMOVE(q, q_link); 1197 kfree(q, M_DUMMYNET); 1198 1199 #ifdef INVARIANTS 1200 rq_elements++; 1201 #endif 1202 } 1203 } 1204 KASSERT(rq_elements == fs->rq_elements, 1205 ("# rq elements mismatch, freed %d, total %d", 1206 rq_elements, fs->rq_elements)); 1207 fs->rq_elements = 0; 1208 1209 if (all) { 1210 /* RED - free lookup table */ 1211 if (fs->w_q_lookup) 1212 kfree(fs->w_q_lookup, M_DUMMYNET); 1213 1214 if (fs->rq) 1215 kfree(fs->rq, M_DUMMYNET); 1216 1217 /* 1218 * If this fs is not part of a pipe, free it 1219 * 1220 * fs->pipe == NULL could happen, if 'fs' is a WF2Q and 1221 * - No packet belongs to that flow set is delivered by 1222 * dummynet_io(), i.e. parent pipe is not installed yet. 1223 * - Parent pipe is deleted. 1224 */ 1225 if (fs->pipe == NULL || (fs->pipe && fs != &fs->pipe->fs)) 1226 kfree(fs, M_DUMMYNET); 1227 } 1228 } 1229 1230 /* 1231 * Dispose all packets queued on a pipe (not a flow_set). 1232 * Also free all resources associated to a pipe, which is about 1233 * to be deleted. 1234 */ 1235 static void 1236 purge_pipe(struct dn_pipe *pipe) 1237 { 1238 struct dn_pkt *pkt; 1239 1240 purge_flow_set(&pipe->fs, 1); 1241 1242 while ((pkt = TAILQ_FIRST(&pipe->p_queue)) != NULL) { 1243 TAILQ_REMOVE(&pipe->p_queue, pkt, dn_next); 1244 ip_dn_packet_free(pkt); 1245 } 1246 1247 heap_free(&pipe->scheduler_heap); 1248 heap_free(&pipe->not_eligible_heap); 1249 heap_free(&pipe->idle_heap); 1250 } 1251 1252 /* 1253 * Delete all pipes and heaps returning memory. 1254 */ 1255 static void 1256 dummynet_flush(void) 1257 { 1258 struct dn_pipe_head pipe_list; 1259 struct dn_flowset_head fs_list; 1260 struct dn_pipe *p; 1261 struct dn_flow_set *fs; 1262 int i; 1263 1264 /* 1265 * Prevent future matches... 1266 */ 1267 LIST_INIT(&pipe_list); 1268 for (i = 0; i < DN_NR_HASH_MAX; ++i) { 1269 struct dn_pipe_head *pipe_hdr = &pipe_table[i]; 1270 1271 while ((p = LIST_FIRST(pipe_hdr)) != NULL) { 1272 LIST_REMOVE(p, p_link); 1273 LIST_INSERT_HEAD(&pipe_list, p, p_link); 1274 } 1275 } 1276 1277 LIST_INIT(&fs_list); 1278 for (i = 0; i < DN_NR_HASH_MAX; ++i) { 1279 struct dn_flowset_head *fs_hdr = &flowset_table[i]; 1280 1281 while ((fs = LIST_FIRST(fs_hdr)) != NULL) { 1282 LIST_REMOVE(fs, fs_link); 1283 LIST_INSERT_HEAD(&fs_list, fs, fs_link); 1284 } 1285 } 1286 1287 /* Free heaps so we don't have unwanted events */ 1288 heap_free(&ready_heap); 1289 heap_free(&wfq_ready_heap); 1290 heap_free(&extract_heap); 1291 1292 /* 1293 * Now purge all queued pkts and delete all pipes 1294 */ 1295 /* Scan and purge all flow_sets. */ 1296 while ((fs = LIST_FIRST(&fs_list)) != NULL) { 1297 LIST_REMOVE(fs, fs_link); 1298 purge_flow_set(fs, 1); 1299 } 1300 1301 while ((p = LIST_FIRST(&pipe_list)) != NULL) { 1302 LIST_REMOVE(p, p_link); 1303 purge_pipe(p); 1304 kfree(p, M_DUMMYNET); 1305 } 1306 } 1307 1308 /* 1309 * setup RED parameters 1310 */ 1311 static int 1312 config_red(const struct dn_ioc_flowset *ioc_fs, struct dn_flow_set *x) 1313 { 1314 int i; 1315 1316 x->w_q = ioc_fs->w_q; 1317 x->min_th = SCALE(ioc_fs->min_th); 1318 x->max_th = SCALE(ioc_fs->max_th); 1319 x->max_p = ioc_fs->max_p; 1320 1321 x->c_1 = ioc_fs->max_p / (ioc_fs->max_th - ioc_fs->min_th); 1322 x->c_2 = SCALE_MUL(x->c_1, SCALE(ioc_fs->min_th)); 1323 if (x->flags_fs & DN_IS_GENTLE_RED) { 1324 x->c_3 = (SCALE(1) - ioc_fs->max_p) / ioc_fs->max_th; 1325 x->c_4 = (SCALE(1) - 2 * ioc_fs->max_p); 1326 } 1327 1328 /* If the lookup table already exist, free and create it again */ 1329 if (x->w_q_lookup) { 1330 kfree(x->w_q_lookup, M_DUMMYNET); 1331 x->w_q_lookup = NULL ; 1332 } 1333 1334 if (red_lookup_depth == 0) { 1335 kprintf("net.inet.ip.dummynet.red_lookup_depth must be > 0\n"); 1336 kfree(x, M_DUMMYNET); 1337 return EINVAL; 1338 } 1339 x->lookup_depth = red_lookup_depth; 1340 x->w_q_lookup = kmalloc(x->lookup_depth * sizeof(int), 1341 M_DUMMYNET, M_WAITOK); 1342 1343 /* Fill the lookup table with (1 - w_q)^x */ 1344 x->lookup_step = ioc_fs->lookup_step; 1345 x->lookup_weight = ioc_fs->lookup_weight; 1346 1347 x->w_q_lookup[0] = SCALE(1) - x->w_q; 1348 for (i = 1; i < x->lookup_depth; i++) 1349 x->w_q_lookup[i] = SCALE_MUL(x->w_q_lookup[i - 1], x->lookup_weight); 1350 1351 if (red_avg_pkt_size < 1) 1352 red_avg_pkt_size = 512; 1353 x->avg_pkt_size = red_avg_pkt_size; 1354 1355 if (red_max_pkt_size < 1) 1356 red_max_pkt_size = 1500; 1357 x->max_pkt_size = red_max_pkt_size; 1358 1359 return 0; 1360 } 1361 1362 static void 1363 alloc_hash(struct dn_flow_set *x, const struct dn_ioc_flowset *ioc_fs) 1364 { 1365 int i, alloc_size; 1366 1367 if (x->flags_fs & DN_HAVE_FLOW_MASK) { 1368 int l = ioc_fs->rq_size; 1369 1370 /* Allocate some slots */ 1371 if (l == 0) 1372 l = dn_hash_size; 1373 1374 if (l < DN_MIN_HASH_SIZE) 1375 l = DN_MIN_HASH_SIZE; 1376 else if (l > DN_MAX_HASH_SIZE) 1377 l = DN_MAX_HASH_SIZE; 1378 1379 x->rq_size = l; 1380 } else { 1381 /* One is enough for null mask */ 1382 x->rq_size = 1; 1383 } 1384 alloc_size = x->rq_size + 1; 1385 1386 x->rq = kmalloc(alloc_size * sizeof(struct dn_flowqueue_head), 1387 M_DUMMYNET, M_WAITOK | M_ZERO); 1388 x->rq_elements = 0; 1389 1390 for (i = 0; i < alloc_size; ++i) 1391 LIST_INIT(&x->rq[i]); 1392 } 1393 1394 static void 1395 set_flowid_parms(struct dn_flow_id *id, const struct dn_ioc_flowid *ioc_id) 1396 { 1397 id->fid_dst_ip = ioc_id->u.ip.dst_ip; 1398 id->fid_src_ip = ioc_id->u.ip.src_ip; 1399 id->fid_dst_port = ioc_id->u.ip.dst_port; 1400 id->fid_src_port = ioc_id->u.ip.src_port; 1401 id->fid_proto = ioc_id->u.ip.proto; 1402 id->fid_flags = ioc_id->u.ip.flags; 1403 } 1404 1405 static void 1406 set_fs_parms(struct dn_flow_set *x, const struct dn_ioc_flowset *ioc_fs) 1407 { 1408 x->flags_fs = ioc_fs->flags_fs; 1409 x->qsize = ioc_fs->qsize; 1410 x->plr = ioc_fs->plr; 1411 set_flowid_parms(&x->flow_mask, &ioc_fs->flow_mask); 1412 if (x->flags_fs & DN_QSIZE_IS_BYTES) { 1413 if (x->qsize > 1024 * 1024) 1414 x->qsize = 1024 * 1024; 1415 } else { 1416 if (x->qsize == 0 || x->qsize > 100) 1417 x->qsize = 50; 1418 } 1419 1420 /* Configuring RED */ 1421 if (x->flags_fs & DN_IS_RED) 1422 config_red(ioc_fs, x); /* XXX should check errors */ 1423 } 1424 1425 /* 1426 * setup pipe or queue parameters. 1427 */ 1428 1429 static int 1430 config_pipe(struct dn_ioc_pipe *ioc_pipe) 1431 { 1432 struct dn_ioc_flowset *ioc_fs = &ioc_pipe->fs; 1433 int error; 1434 1435 /* 1436 * The config program passes parameters as follows: 1437 * bw bits/second (0 means no limits) 1438 * delay ms (must be translated into ticks) 1439 * qsize slots or bytes 1440 */ 1441 ioc_pipe->delay = (ioc_pipe->delay * dn_hz) / 1000; 1442 1443 /* 1444 * We need either a pipe number or a flow_set number 1445 */ 1446 if (ioc_pipe->pipe_nr == 0 && ioc_fs->fs_nr == 0) 1447 return EINVAL; 1448 if (ioc_pipe->pipe_nr != 0 && ioc_fs->fs_nr != 0) 1449 return EINVAL; 1450 1451 /* 1452 * Validate pipe number 1453 */ 1454 if (ioc_pipe->pipe_nr > DN_PIPE_NR_MAX || ioc_pipe->pipe_nr < 0) 1455 return EINVAL; 1456 1457 error = EINVAL; 1458 if (ioc_pipe->pipe_nr != 0) { /* This is a pipe */ 1459 struct dn_pipe *x, *p; 1460 1461 /* Locate pipe */ 1462 p = dn_find_pipe(ioc_pipe->pipe_nr); 1463 1464 if (p == NULL) { /* New pipe */ 1465 x = kmalloc(sizeof(struct dn_pipe), M_DUMMYNET, M_WAITOK | M_ZERO); 1466 x->pipe_nr = ioc_pipe->pipe_nr; 1467 x->fs.pipe = x; 1468 TAILQ_INIT(&x->p_queue); 1469 1470 /* 1471 * idle_heap is the only one from which we extract from the middle. 1472 */ 1473 x->idle_heap.size = x->idle_heap.elements = 0; 1474 x->idle_heap.offset = __offsetof(struct dn_flow_queue, heap_pos); 1475 } else { 1476 int i; 1477 1478 x = p; 1479 1480 /* Flush accumulated credit for all queues */ 1481 for (i = 0; i <= x->fs.rq_size; i++) { 1482 struct dn_flow_queue *q; 1483 1484 LIST_FOREACH(q, &x->fs.rq[i], q_link) 1485 q->numbytes = 0; 1486 } 1487 } 1488 1489 x->bandwidth = ioc_pipe->bandwidth; 1490 x->numbytes = 0; /* Just in case... */ 1491 x->delay = ioc_pipe->delay; 1492 1493 set_fs_parms(&x->fs, ioc_fs); 1494 1495 if (x->fs.rq == NULL) { /* A new pipe */ 1496 struct dn_pipe_head *pipe_hdr; 1497 1498 alloc_hash(&x->fs, ioc_fs); 1499 1500 pipe_hdr = &pipe_table[DN_NR_HASH(x->pipe_nr)]; 1501 LIST_INSERT_HEAD(pipe_hdr, x, p_link); 1502 } 1503 } else { /* Config flow_set */ 1504 struct dn_flow_set *x, *fs; 1505 1506 /* Locate flow_set */ 1507 fs = dn_find_flowset(ioc_fs->fs_nr); 1508 1509 if (fs == NULL) { /* New flow_set */ 1510 if (ioc_fs->parent_nr == 0) /* Need link to a pipe */ 1511 goto back; 1512 1513 x = kmalloc(sizeof(struct dn_flow_set), M_DUMMYNET, 1514 M_WAITOK | M_ZERO); 1515 x->fs_nr = ioc_fs->fs_nr; 1516 x->parent_nr = ioc_fs->parent_nr; 1517 x->weight = ioc_fs->weight; 1518 if (x->weight == 0) 1519 x->weight = 1; 1520 else if (x->weight > 100) 1521 x->weight = 100; 1522 } else { 1523 /* Change parent pipe not allowed; must delete and recreate */ 1524 if (ioc_fs->parent_nr != 0 && fs->parent_nr != ioc_fs->parent_nr) 1525 goto back; 1526 x = fs; 1527 } 1528 1529 set_fs_parms(x, ioc_fs); 1530 1531 if (x->rq == NULL) { /* A new flow_set */ 1532 struct dn_flowset_head *fs_hdr; 1533 1534 alloc_hash(x, ioc_fs); 1535 1536 fs_hdr = &flowset_table[DN_NR_HASH(x->fs_nr)]; 1537 LIST_INSERT_HEAD(fs_hdr, x, fs_link); 1538 } 1539 } 1540 error = 0; 1541 1542 back: 1543 return error; 1544 } 1545 1546 /* 1547 * Helper function to remove from a heap queues which are linked to 1548 * a flow_set about to be deleted. 1549 */ 1550 static void 1551 fs_remove_from_heap(struct dn_heap *h, struct dn_flow_set *fs) 1552 { 1553 int i = 0, found = 0; 1554 1555 while (i < h->elements) { 1556 if (((struct dn_flow_queue *)h->p[i].object)->fs == fs) { 1557 h->elements--; 1558 h->p[i] = h->p[h->elements]; 1559 found++; 1560 } else { 1561 i++; 1562 } 1563 } 1564 if (found) 1565 heapify(h); 1566 } 1567 1568 /* 1569 * helper function to remove a pipe from a heap (can be there at most once) 1570 */ 1571 static void 1572 pipe_remove_from_heap(struct dn_heap *h, struct dn_pipe *p) 1573 { 1574 if (h->elements > 0) { 1575 int i; 1576 1577 for (i = 0; i < h->elements; i++) { 1578 if (h->p[i].object == p) { /* found it */ 1579 h->elements--; 1580 h->p[i] = h->p[h->elements]; 1581 heapify(h); 1582 break; 1583 } 1584 } 1585 } 1586 } 1587 1588 static void 1589 dn_unref_pipe_cb(struct dn_flow_set *fs, void *pipe0) 1590 { 1591 struct dn_pipe *pipe = pipe0; 1592 1593 if (fs->pipe == pipe) { 1594 kprintf("++ ref to pipe %d from fs %d\n", 1595 pipe->pipe_nr, fs->fs_nr); 1596 fs->pipe = NULL; 1597 purge_flow_set(fs, 0); 1598 } 1599 } 1600 1601 /* 1602 * Fully delete a pipe or a queue, cleaning up associated info. 1603 */ 1604 static int 1605 delete_pipe(const struct dn_ioc_pipe *ioc_pipe) 1606 { 1607 struct dn_pipe *p; 1608 int error; 1609 1610 if (ioc_pipe->pipe_nr == 0 && ioc_pipe->fs.fs_nr == 0) 1611 return EINVAL; 1612 if (ioc_pipe->pipe_nr != 0 && ioc_pipe->fs.fs_nr != 0) 1613 return EINVAL; 1614 1615 if (ioc_pipe->pipe_nr > DN_NR_HASH_MAX || ioc_pipe->pipe_nr < 0) 1616 return EINVAL; 1617 1618 error = EINVAL; 1619 if (ioc_pipe->pipe_nr != 0) { /* This is an old-style pipe */ 1620 /* Locate pipe */ 1621 p = dn_find_pipe(ioc_pipe->pipe_nr); 1622 if (p == NULL) 1623 goto back; /* Not found */ 1624 1625 /* Unlink from pipe hash table */ 1626 LIST_REMOVE(p, p_link); 1627 1628 /* Remove all references to this pipe from flow_sets */ 1629 dn_iterate_flowset(dn_unref_pipe_cb, p); 1630 1631 fs_remove_from_heap(&ready_heap, &p->fs); 1632 purge_pipe(p); /* Remove all data associated to this pipe */ 1633 1634 /* Remove reference to here from extract_heap and wfq_ready_heap */ 1635 pipe_remove_from_heap(&extract_heap, p); 1636 pipe_remove_from_heap(&wfq_ready_heap, p); 1637 1638 kfree(p, M_DUMMYNET); 1639 } else { /* This is a WF2Q queue (dn_flow_set) */ 1640 struct dn_flow_set *fs; 1641 1642 /* Locate flow_set */ 1643 fs = dn_find_flowset(ioc_pipe->fs.fs_nr); 1644 if (fs == NULL) 1645 goto back; /* Not found */ 1646 1647 LIST_REMOVE(fs, fs_link); 1648 1649 if ((p = fs->pipe) != NULL) { 1650 /* Update total weight on parent pipe and cleanup parent heaps */ 1651 p->sum -= fs->weight * fs->backlogged; 1652 fs_remove_from_heap(&p->not_eligible_heap, fs); 1653 fs_remove_from_heap(&p->scheduler_heap, fs); 1654 #if 1 /* XXX should i remove from idle_heap as well ? */ 1655 fs_remove_from_heap(&p->idle_heap, fs); 1656 #endif 1657 } 1658 purge_flow_set(fs, 1); 1659 } 1660 error = 0; 1661 1662 back: 1663 return error; 1664 } 1665 1666 /* 1667 * helper function used to copy data from kernel in DUMMYNET_GET 1668 */ 1669 static void 1670 dn_copy_flowid(const struct dn_flow_id *id, struct dn_ioc_flowid *ioc_id) 1671 { 1672 ioc_id->type = ETHERTYPE_IP; 1673 ioc_id->u.ip.dst_ip = id->fid_dst_ip; 1674 ioc_id->u.ip.src_ip = id->fid_src_ip; 1675 ioc_id->u.ip.dst_port = id->fid_dst_port; 1676 ioc_id->u.ip.src_port = id->fid_src_port; 1677 ioc_id->u.ip.proto = id->fid_proto; 1678 ioc_id->u.ip.flags = id->fid_flags; 1679 } 1680 1681 static void * 1682 dn_copy_flowqueues(const struct dn_flow_set *fs, void *bp) 1683 { 1684 struct dn_ioc_flowqueue *ioc_fq = bp; 1685 int i, copied = 0; 1686 1687 for (i = 0; i <= fs->rq_size; i++) { 1688 const struct dn_flow_queue *q; 1689 1690 LIST_FOREACH(q, &fs->rq[i], q_link) { 1691 if (q->hash_slot != i) { /* XXX ASSERT */ 1692 kprintf("++ at %d: wrong slot (have %d, " 1693 "should be %d)\n", copied, q->hash_slot, i); 1694 } 1695 if (q->fs != fs) { /* XXX ASSERT */ 1696 kprintf("++ at %d: wrong fs ptr (have %p, should be %p)\n", 1697 i, q->fs, fs); 1698 } 1699 1700 copied++; 1701 1702 ioc_fq->len = q->len; 1703 ioc_fq->len_bytes = q->len_bytes; 1704 ioc_fq->tot_pkts = q->tot_pkts; 1705 ioc_fq->tot_bytes = q->tot_bytes; 1706 ioc_fq->drops = q->drops; 1707 ioc_fq->hash_slot = q->hash_slot; 1708 ioc_fq->S = q->S; 1709 ioc_fq->F = q->F; 1710 dn_copy_flowid(&q->id, &ioc_fq->id); 1711 1712 ioc_fq++; 1713 } 1714 } 1715 1716 if (copied != fs->rq_elements) { /* XXX ASSERT */ 1717 kprintf("++ wrong count, have %d should be %d\n", 1718 copied, fs->rq_elements); 1719 } 1720 return ioc_fq; 1721 } 1722 1723 static void 1724 dn_copy_flowset(const struct dn_flow_set *fs, struct dn_ioc_flowset *ioc_fs, 1725 u_short fs_type) 1726 { 1727 ioc_fs->fs_type = fs_type; 1728 1729 ioc_fs->fs_nr = fs->fs_nr; 1730 ioc_fs->flags_fs = fs->flags_fs; 1731 ioc_fs->parent_nr = fs->parent_nr; 1732 1733 ioc_fs->weight = fs->weight; 1734 ioc_fs->qsize = fs->qsize; 1735 ioc_fs->plr = fs->plr; 1736 1737 ioc_fs->rq_size = fs->rq_size; 1738 ioc_fs->rq_elements = fs->rq_elements; 1739 1740 ioc_fs->w_q = fs->w_q; 1741 ioc_fs->max_th = fs->max_th; 1742 ioc_fs->min_th = fs->min_th; 1743 ioc_fs->max_p = fs->max_p; 1744 1745 dn_copy_flowid(&fs->flow_mask, &ioc_fs->flow_mask); 1746 } 1747 1748 static void 1749 dn_calc_pipe_size_cb(struct dn_pipe *pipe, void *sz) 1750 { 1751 size_t *size = sz; 1752 1753 *size += sizeof(struct dn_ioc_pipe) + 1754 pipe->fs.rq_elements * sizeof(struct dn_ioc_flowqueue); 1755 } 1756 1757 static void 1758 dn_calc_fs_size_cb(struct dn_flow_set *fs, void *sz) 1759 { 1760 size_t *size = sz; 1761 1762 *size += sizeof(struct dn_ioc_flowset) + 1763 fs->rq_elements * sizeof(struct dn_ioc_flowqueue); 1764 } 1765 1766 static void 1767 dn_copyout_pipe_cb(struct dn_pipe *pipe, void *bp0) 1768 { 1769 char **bp = bp0; 1770 struct dn_ioc_pipe *ioc_pipe = (struct dn_ioc_pipe *)(*bp); 1771 1772 /* 1773 * Copy flow set descriptor associated with this pipe 1774 */ 1775 dn_copy_flowset(&pipe->fs, &ioc_pipe->fs, DN_IS_PIPE); 1776 1777 /* 1778 * Copy pipe descriptor 1779 */ 1780 ioc_pipe->bandwidth = pipe->bandwidth; 1781 ioc_pipe->pipe_nr = pipe->pipe_nr; 1782 ioc_pipe->V = pipe->V; 1783 /* Convert delay to milliseconds */ 1784 ioc_pipe->delay = (pipe->delay * 1000) / dn_hz; 1785 1786 /* 1787 * Copy flow queue descriptors 1788 */ 1789 *bp += sizeof(*ioc_pipe); 1790 *bp = dn_copy_flowqueues(&pipe->fs, *bp); 1791 } 1792 1793 static void 1794 dn_copyout_fs_cb(struct dn_flow_set *fs, void *bp0) 1795 { 1796 char **bp = bp0; 1797 struct dn_ioc_flowset *ioc_fs = (struct dn_ioc_flowset *)(*bp); 1798 1799 /* 1800 * Copy flow set descriptor 1801 */ 1802 dn_copy_flowset(fs, ioc_fs, DN_IS_QUEUE); 1803 1804 /* 1805 * Copy flow queue descriptors 1806 */ 1807 *bp += sizeof(*ioc_fs); 1808 *bp = dn_copy_flowqueues(fs, *bp); 1809 } 1810 1811 static int 1812 dummynet_get(struct dn_sopt *dn_sopt) 1813 { 1814 char *buf, *bp; 1815 size_t size = 0; 1816 1817 /* 1818 * Compute size of data structures: list of pipes and flow_sets. 1819 */ 1820 dn_iterate_pipe(dn_calc_pipe_size_cb, &size); 1821 dn_iterate_flowset(dn_calc_fs_size_cb, &size); 1822 1823 /* 1824 * Copyout pipe/flow_set/flow_queue 1825 */ 1826 bp = buf = kmalloc(size, M_TEMP, M_WAITOK | M_ZERO); 1827 dn_iterate_pipe(dn_copyout_pipe_cb, &bp); 1828 dn_iterate_flowset(dn_copyout_fs_cb, &bp); 1829 1830 /* Temp memory will be freed by caller */ 1831 dn_sopt->dn_sopt_arg = buf; 1832 dn_sopt->dn_sopt_arglen = size; 1833 return 0; 1834 } 1835 1836 /* 1837 * Handler for the various dummynet socket options (get, flush, config, del) 1838 */ 1839 static int 1840 dummynet_ctl(struct dn_sopt *dn_sopt) 1841 { 1842 int error = 0; 1843 1844 switch (dn_sopt->dn_sopt_name) { 1845 case IP_DUMMYNET_GET: 1846 error = dummynet_get(dn_sopt); 1847 break; 1848 1849 case IP_DUMMYNET_FLUSH: 1850 dummynet_flush(); 1851 break; 1852 1853 case IP_DUMMYNET_CONFIGURE: 1854 KKASSERT(dn_sopt->dn_sopt_arglen == sizeof(struct dn_ioc_pipe)); 1855 error = config_pipe(dn_sopt->dn_sopt_arg); 1856 break; 1857 1858 case IP_DUMMYNET_DEL: /* Remove a pipe or flow_set */ 1859 KKASSERT(dn_sopt->dn_sopt_arglen == sizeof(struct dn_ioc_pipe)); 1860 error = delete_pipe(dn_sopt->dn_sopt_arg); 1861 break; 1862 1863 default: 1864 kprintf("%s -- unknown option %d\n", __func__, dn_sopt->dn_sopt_name); 1865 error = EINVAL; 1866 break; 1867 } 1868 return error; 1869 } 1870 1871 static void 1872 dummynet_clock(systimer_t info __unused, int in_ipi __unused, 1873 struct intrframe *frame __unused) 1874 { 1875 KASSERT(mycpuid == ip_dn_cpu, 1876 ("dummynet systimer comes on cpu%d, should be %d!", 1877 mycpuid, ip_dn_cpu)); 1878 1879 crit_enter(); 1880 if (DUMMYNET_LOADED && (dn_netmsg.lmsg.ms_flags & MSGF_DONE)) 1881 lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), &dn_netmsg.lmsg); 1882 crit_exit(); 1883 } 1884 1885 static int 1886 sysctl_dn_hz(SYSCTL_HANDLER_ARGS) 1887 { 1888 int error, val, origcpu; 1889 1890 val = dn_hz; 1891 error = sysctl_handle_int(oidp, &val, 0, req); 1892 if (error || req->newptr == NULL) 1893 return error; 1894 if (val <= 0) 1895 return EINVAL; 1896 else if (val > DN_CALLOUT_FREQ_MAX) 1897 val = DN_CALLOUT_FREQ_MAX; 1898 1899 origcpu = mycpuid; 1900 lwkt_migratecpu(ip_dn_cpu); 1901 1902 crit_enter(); 1903 dn_hz = val; 1904 systimer_adjust_periodic(&dn_clock, val); 1905 crit_exit(); 1906 1907 lwkt_migratecpu(origcpu); 1908 1909 return 0; 1910 } 1911 1912 static void 1913 ip_dn_init_dispatch(netmsg_t msg) 1914 { 1915 int i, error = 0; 1916 1917 KASSERT(mycpuid == ip_dn_cpu, 1918 ("%s runs on cpu%d, instead of cpu%d", __func__, 1919 mycpuid, ip_dn_cpu)); 1920 1921 crit_enter(); 1922 1923 if (DUMMYNET_LOADED) { 1924 kprintf("DUMMYNET already loaded\n"); 1925 error = EEXIST; 1926 goto back; 1927 } 1928 1929 kprintf("DUMMYNET initialized (011031)\n"); 1930 1931 for (i = 0; i < DN_NR_HASH_MAX; ++i) 1932 LIST_INIT(&pipe_table[i]); 1933 1934 for (i = 0; i < DN_NR_HASH_MAX; ++i) 1935 LIST_INIT(&flowset_table[i]); 1936 1937 ready_heap.size = ready_heap.elements = 0; 1938 ready_heap.offset = 0; 1939 1940 wfq_ready_heap.size = wfq_ready_heap.elements = 0; 1941 wfq_ready_heap.offset = 0; 1942 1943 extract_heap.size = extract_heap.elements = 0; 1944 extract_heap.offset = 0; 1945 1946 ip_dn_ctl_ptr = dummynet_ctl; 1947 ip_dn_io_ptr = dummynet_io; 1948 1949 netmsg_init(&dn_netmsg, NULL, &netisr_adone_rport, 1950 0, dummynet); 1951 systimer_init_periodic_nq(&dn_clock, dummynet_clock, NULL, dn_hz); 1952 1953 back: 1954 crit_exit(); 1955 lwkt_replymsg(&msg->lmsg, error); 1956 } 1957 1958 static int 1959 ip_dn_init(void) 1960 { 1961 struct netmsg_base smsg; 1962 1963 if (ip_dn_cpu >= ncpus) { 1964 kprintf("%s: CPU%d does not exist, switch to CPU0\n", 1965 __func__, ip_dn_cpu); 1966 ip_dn_cpu = 0; 1967 } 1968 1969 netmsg_init(&smsg, NULL, &curthread->td_msgport, 1970 0, ip_dn_init_dispatch); 1971 lwkt_domsg(netisr_cpuport(ip_dn_cpu), &smsg.lmsg, 0); 1972 return smsg.lmsg.ms_error; 1973 } 1974 1975 #ifdef KLD_MODULE 1976 1977 static void 1978 ip_dn_stop_dispatch(netmsg_t msg) 1979 { 1980 crit_enter(); 1981 1982 dummynet_flush(); 1983 1984 ip_dn_ctl_ptr = NULL; 1985 ip_dn_io_ptr = NULL; 1986 1987 systimer_del(&dn_clock); 1988 1989 crit_exit(); 1990 lwkt_replymsg(&msg->lmsg, 0); 1991 } 1992 1993 1994 static void 1995 ip_dn_stop(void) 1996 { 1997 struct netmsg_base smsg; 1998 1999 netmsg_init(&smsg, NULL, &curthread->td_msgport, 2000 0, ip_dn_stop_dispatch); 2001 lwkt_domsg(netisr_cpuport(ip_dn_cpu), &smsg.lmsg, 0); 2002 2003 netmsg_service_sync(); 2004 } 2005 2006 #endif /* KLD_MODULE */ 2007 2008 static int 2009 dummynet_modevent(module_t mod, int type, void *data) 2010 { 2011 switch (type) { 2012 case MOD_LOAD: 2013 return ip_dn_init(); 2014 2015 case MOD_UNLOAD: 2016 #ifndef KLD_MODULE 2017 kprintf("dummynet statically compiled, cannot unload\n"); 2018 return EINVAL; 2019 #else 2020 ip_dn_stop(); 2021 #endif 2022 break; 2023 2024 default: 2025 break; 2026 } 2027 return 0; 2028 } 2029 2030 static moduledata_t dummynet_mod = { 2031 "dummynet", 2032 dummynet_modevent, 2033 NULL 2034 }; 2035 DECLARE_MODULE(dummynet, dummynet_mod, SI_SUB_PROTO_END, SI_ORDER_ANY); 2036 MODULE_VERSION(dummynet, 1); 2037