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 #ifdef _KERNEL_VIRTUAL 145 static int dn_hz = 100; 146 #else 147 static int dn_hz = 1000; 148 #endif 149 static int dn_count; 150 static int dn_running; 151 static struct lock dn_lock = LOCK_INITIALIZER("dnlk", 0, 0); 152 153 static int sysctl_dn_hz(SYSCTL_HANDLER_ARGS); 154 155 SYSCTL_DECL(_net_inet_ip_dummynet); 156 157 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, hash_size, CTLFLAG_RW, 158 &dn_hash_size, 0, "Default hash table size"); 159 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, curr_time, CTLFLAG_RD, 160 &curr_time, 0, "Current tick"); 161 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, expire, CTLFLAG_RW, 162 &pipe_expire, 0, "Expire queue if empty"); 163 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, max_chain_len, CTLFLAG_RW, 164 &dn_max_ratio, 0, "Max ratio between dynamic queues and buckets"); 165 166 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, ready_heap, CTLFLAG_RD, 167 &ready_heap.size, 0, "Size of ready heap"); 168 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, extract_heap, CTLFLAG_RD, 169 &extract_heap.size, 0, "Size of extract heap"); 170 171 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, searches, CTLFLAG_RD, 172 &searches, 0, "Number of queue searches"); 173 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, search_steps, CTLFLAG_RD, 174 &search_steps, 0, "Number of queue search steps"); 175 176 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_lookup_depth, CTLFLAG_RD, 177 &red_lookup_depth, 0, "Depth of RED lookup table"); 178 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_avg_pkt_size, CTLFLAG_RD, 179 &red_avg_pkt_size, 0, "RED Medium packet size"); 180 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_max_pkt_size, CTLFLAG_RD, 181 &red_max_pkt_size, 0, "RED Max packet size"); 182 183 SYSCTL_PROC(_net_inet_ip_dummynet, OID_AUTO, hz, CTLTYPE_INT | CTLFLAG_RW, 184 0, 0, sysctl_dn_hz, "I", "Dummynet callout frequency"); 185 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, running, CTLFLAG_RD, 186 &dn_running, 0, "Dummynet Active"); 187 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, count, CTLFLAG_RD, 188 &dn_count, 0, "Dummynet pipe+flow count"); 189 190 static int heap_init(struct dn_heap *, int); 191 static int heap_insert(struct dn_heap *, dn_key, void *); 192 static void heap_extract(struct dn_heap *, void *); 193 194 static void transmit_event(struct dn_pipe *); 195 static void ready_event(struct dn_flow_queue *); 196 static void ready_event_wfq(struct dn_pipe *); 197 198 static int config_pipe(struct dn_ioc_pipe *); 199 static void dummynet_flush(void); 200 201 static void dummynet_clock(systimer_t, int, struct intrframe *); 202 static void dummynet(netmsg_t); 203 204 static struct dn_pipe *dn_find_pipe(int); 205 static struct dn_flow_set *dn_locate_flowset(int, int); 206 207 typedef void (*dn_pipe_iter_t)(struct dn_pipe *, void *); 208 static void dn_iterate_pipe(dn_pipe_iter_t, void *); 209 210 typedef void (*dn_flowset_iter_t)(struct dn_flow_set *, void *); 211 static void dn_iterate_flowset(dn_flowset_iter_t, void *); 212 213 static ip_dn_io_t dummynet_io; 214 static ip_dn_ctl_t dummynet_ctl; 215 216 /* 217 * Heap management functions. 218 * 219 * In the heap, first node is element 0. Children of i are 2i+1 and 2i+2. 220 * Some macros help finding parent/children so we can optimize them. 221 * 222 * heap_init() is called to expand the heap when needed. 223 * Increment size in blocks of 16 entries. 224 * XXX failure to allocate a new element is a pretty bad failure 225 * as we basically stall a whole queue forever!! 226 * Returns 1 on error, 0 on success 227 */ 228 #define HEAP_FATHER(x) (((x) - 1) / 2) 229 #define HEAP_LEFT(x) (2*(x) + 1) 230 #define HEAP_IS_LEFT(x) ((x) & 1) 231 #define HEAP_RIGHT(x) (2*(x) + 2) 232 #define HEAP_SWAP(a, b, buffer) { buffer = a; a = b; b = buffer; } 233 #define HEAP_INCREMENT 15 234 235 static int 236 heap_init(struct dn_heap *h, int new_size) 237 { 238 struct dn_heap_entry *p; 239 240 if (h->size >= new_size) { 241 kprintf("%s, Bogus call, have %d want %d\n", __func__, 242 h->size, new_size); 243 return 0; 244 } 245 246 new_size = (new_size + HEAP_INCREMENT) & ~HEAP_INCREMENT; 247 p = kmalloc(new_size * sizeof(*p), M_DUMMYNET, M_WAITOK | M_ZERO); 248 if (h->size > 0) { 249 bcopy(h->p, p, h->size * sizeof(*p)); 250 kfree(h->p, M_DUMMYNET); 251 } 252 h->p = p; 253 h->size = new_size; 254 return 0; 255 } 256 257 /* 258 * Insert element in heap. Normally, p != NULL, we insert p in 259 * a new position and bubble up. If p == NULL, then the element is 260 * already in place, and key is the position where to start the 261 * bubble-up. 262 * Returns 1 on failure (cannot allocate new heap entry) 263 * 264 * If offset > 0 the position (index, int) of the element in the heap is 265 * also stored in the element itself at the given offset in bytes. 266 */ 267 #define SET_OFFSET(heap, node) \ 268 if (heap->offset > 0) \ 269 *((int *)((char *)(heap->p[node].object) + heap->offset)) = node; 270 271 /* 272 * RESET_OFFSET is used for sanity checks. It sets offset to an invalid value. 273 */ 274 #define RESET_OFFSET(heap, node) \ 275 if (heap->offset > 0) \ 276 *((int *)((char *)(heap->p[node].object) + heap->offset)) = -1; 277 278 static int 279 heap_insert(struct dn_heap *h, dn_key key1, void *p) 280 { 281 int son; 282 283 if (p == NULL) { /* Data already there, set starting point */ 284 son = key1; 285 } else { /* Insert new element at the end, possibly resize */ 286 son = h->elements; 287 if (son == h->size) { /* Need resize... */ 288 if (heap_init(h, h->elements + 1)) 289 return 1; /* Failure... */ 290 } 291 h->p[son].object = p; 292 h->p[son].key = key1; 293 h->elements++; 294 } 295 296 while (son > 0) { /* Bubble up */ 297 int father = HEAP_FATHER(son); 298 struct dn_heap_entry tmp; 299 300 if (DN_KEY_LT(h->p[father].key, h->p[son].key)) 301 break; /* Found right position */ 302 303 /* 'son' smaller than 'father', swap and repeat */ 304 HEAP_SWAP(h->p[son], h->p[father], tmp); 305 SET_OFFSET(h, son); 306 son = father; 307 } 308 SET_OFFSET(h, son); 309 return 0; 310 } 311 312 /* 313 * Remove top element from heap, or obj if obj != NULL 314 */ 315 static void 316 heap_extract(struct dn_heap *h, void *obj) 317 { 318 int child, father, max = h->elements - 1; 319 320 if (max < 0) { 321 kprintf("warning, extract from empty heap 0x%p\n", h); 322 return; 323 } 324 325 father = 0; /* Default: move up smallest child */ 326 if (obj != NULL) { /* Extract specific element, index is at offset */ 327 if (h->offset <= 0) 328 panic("%s from middle not supported on this heap!!!", __func__); 329 330 father = *((int *)((char *)obj + h->offset)); 331 if (father < 0 || father >= h->elements) { 332 panic("%s father %d out of bound 0..%d", __func__, 333 father, h->elements); 334 } 335 } 336 RESET_OFFSET(h, father); 337 338 child = HEAP_LEFT(father); /* Left child */ 339 while (child <= max) { /* Valid entry */ 340 if (child != max && DN_KEY_LT(h->p[child + 1].key, h->p[child].key)) 341 child = child + 1; /* Take right child, otherwise left */ 342 h->p[father] = h->p[child]; 343 SET_OFFSET(h, father); 344 father = child; 345 child = HEAP_LEFT(child); /* Left child for next loop */ 346 } 347 h->elements--; 348 if (father != max) { 349 /* 350 * Fill hole with last entry and bubble up, reusing the insert code 351 */ 352 h->p[father] = h->p[max]; 353 heap_insert(h, father, NULL); /* This one cannot fail */ 354 } 355 } 356 357 /* 358 * heapify() will reorganize data inside an array to maintain the 359 * heap property. It is needed when we delete a bunch of entries. 360 */ 361 static void 362 heapify(struct dn_heap *h) 363 { 364 int i; 365 366 for (i = 0; i < h->elements; i++) 367 heap_insert(h, i , NULL); 368 } 369 370 /* 371 * Cleanup the heap and free data structure 372 */ 373 static void 374 heap_free(struct dn_heap *h) 375 { 376 if (h->size > 0) 377 kfree(h->p, M_DUMMYNET); 378 bzero(h, sizeof(*h)); 379 } 380 381 /* 382 * --- End of heap management functions --- 383 */ 384 385 /* 386 * Scheduler functions: 387 * 388 * transmit_event() is called when the delay-line needs to enter 389 * the scheduler, either because of existing pkts getting ready, 390 * or new packets entering the queue. The event handled is the delivery 391 * time of the packet. 392 * 393 * ready_event() does something similar with fixed-rate queues, and the 394 * event handled is the finish time of the head pkt. 395 * 396 * ready_event_wfq() does something similar with WF2Q queues, and the 397 * event handled is the start time of the head pkt. 398 * 399 * In all cases, we make sure that the data structures are consistent 400 * before passing pkts out, because this might trigger recursive 401 * invocations of the procedures. 402 */ 403 static void 404 transmit_event(struct dn_pipe *pipe) 405 { 406 struct dn_pkt *pkt; 407 408 while ((pkt = TAILQ_FIRST(&pipe->p_queue)) && 409 DN_KEY_LEQ(pkt->output_time, curr_time)) { 410 TAILQ_REMOVE(&pipe->p_queue, pkt, dn_next); 411 ip_dn_packet_redispatch(pkt); 412 } 413 414 /* 415 * If there are leftover packets, put into the heap for next event 416 */ 417 if ((pkt = TAILQ_FIRST(&pipe->p_queue)) != NULL) { 418 /* 419 * XXX should check errors on heap_insert, by draining the 420 * whole pipe and hoping in the future we are more successful 421 */ 422 heap_insert(&extract_heap, pkt->output_time, pipe); 423 } 424 } 425 426 /* 427 * The following macro computes how many ticks we have to wait 428 * before being able to transmit a packet. The credit is taken from 429 * either a pipe (WF2Q) or a flow_queue (per-flow queueing) 430 */ 431 #define SET_TICKS(pkt, q, p) \ 432 (pkt->dn_m->m_pkthdr.len*8*dn_hz - (q)->numbytes + p->bandwidth - 1 ) / \ 433 p->bandwidth; 434 435 /* 436 * Extract pkt from queue, compute output time (could be now) 437 * and put into delay line (p_queue) 438 */ 439 static void 440 move_pkt(struct dn_pkt *pkt, struct dn_flow_queue *q, 441 struct dn_pipe *p, int len) 442 { 443 TAILQ_REMOVE(&q->queue, pkt, dn_next); 444 q->len--; 445 q->len_bytes -= len; 446 447 pkt->output_time = curr_time + p->delay; 448 449 TAILQ_INSERT_TAIL(&p->p_queue, pkt, dn_next); 450 } 451 452 /* 453 * ready_event() is invoked every time the queue must enter the 454 * scheduler, either because the first packet arrives, or because 455 * a previously scheduled event fired. 456 * On invokation, drain as many pkts as possible (could be 0) and then 457 * if there are leftover packets reinsert the pkt in the scheduler. 458 */ 459 static void 460 ready_event(struct dn_flow_queue *q) 461 { 462 struct dn_pkt *pkt; 463 struct dn_pipe *p = q->fs->pipe; 464 int p_was_empty; 465 466 if (p == NULL) { 467 kprintf("ready_event- pipe is gone\n"); 468 return; 469 } 470 p_was_empty = TAILQ_EMPTY(&p->p_queue); 471 472 /* 473 * Schedule fixed-rate queues linked to this pipe: 474 * Account for the bw accumulated since last scheduling, then 475 * drain as many pkts as allowed by q->numbytes and move to 476 * the delay line (in p) computing output time. 477 * bandwidth==0 (no limit) means we can drain the whole queue, 478 * setting len_scaled = 0 does the job. 479 */ 480 q->numbytes += (curr_time - q->sched_time) * p->bandwidth; 481 while ((pkt = TAILQ_FIRST(&q->queue)) != NULL) { 482 int len = pkt->dn_m->m_pkthdr.len; 483 int len_scaled = p->bandwidth ? len*8*dn_hz : 0; 484 485 if (len_scaled > q->numbytes) 486 break; 487 q->numbytes -= len_scaled; 488 move_pkt(pkt, q, p, len); 489 } 490 491 /* 492 * If we have more packets queued, schedule next ready event 493 * (can only occur when bandwidth != 0, otherwise we would have 494 * flushed the whole queue in the previous loop). 495 * To this purpose we record the current time and compute how many 496 * ticks to go for the finish time of the packet. 497 */ 498 if ((pkt = TAILQ_FIRST(&q->queue)) != NULL) { 499 /* This implies bandwidth != 0 */ 500 dn_key t = SET_TICKS(pkt, q, p); /* ticks i have to wait */ 501 502 q->sched_time = curr_time; 503 504 /* 505 * XXX should check errors on heap_insert, and drain the whole 506 * queue on error hoping next time we are luckier. 507 */ 508 heap_insert(&ready_heap, curr_time + t, q); 509 } else { /* RED needs to know when the queue becomes empty */ 510 q->q_time = curr_time; 511 q->numbytes = 0; 512 } 513 514 /* 515 * If the delay line was empty call transmit_event(p) now. 516 * Otherwise, the scheduler will take care of it. 517 */ 518 if (p_was_empty) 519 transmit_event(p); 520 } 521 522 /* 523 * Called when we can transmit packets on WF2Q queues. Take pkts out of 524 * the queues at their start time, and enqueue into the delay line. 525 * Packets are drained until p->numbytes < 0. As long as 526 * len_scaled >= p->numbytes, the packet goes into the delay line 527 * with a deadline p->delay. For the last packet, if p->numbytes < 0, 528 * there is an additional delay. 529 */ 530 static void 531 ready_event_wfq(struct dn_pipe *p) 532 { 533 int p_was_empty = TAILQ_EMPTY(&p->p_queue); 534 struct dn_heap *sch = &p->scheduler_heap; 535 struct dn_heap *neh = &p->not_eligible_heap; 536 537 p->numbytes += (curr_time - p->sched_time) * p->bandwidth; 538 539 /* 540 * While we have backlogged traffic AND credit, we need to do 541 * something on the queue. 542 */ 543 while (p->numbytes >= 0 && (sch->elements > 0 || neh->elements > 0)) { 544 if (sch->elements > 0) { /* Have some eligible pkts to send out */ 545 struct dn_flow_queue *q = sch->p[0].object; 546 struct dn_pkt *pkt = TAILQ_FIRST(&q->queue); 547 struct dn_flow_set *fs = q->fs; 548 uint64_t len = pkt->dn_m->m_pkthdr.len; 549 int len_scaled = p->bandwidth ? len*8*dn_hz : 0; 550 551 heap_extract(sch, NULL); /* Remove queue from heap */ 552 p->numbytes -= len_scaled; 553 move_pkt(pkt, q, p, len); 554 555 p->V += (len << MY_M) / p->sum; /* Update V */ 556 q->S = q->F; /* Update start time */ 557 558 if (q->len == 0) { /* Flow not backlogged any more */ 559 fs->backlogged--; 560 heap_insert(&p->idle_heap, q->F, q); 561 } else { /* Still backlogged */ 562 /* 563 * Update F and position in backlogged queue, then 564 * put flow in not_eligible_heap (we will fix this later). 565 */ 566 len = TAILQ_FIRST(&q->queue)->dn_m->m_pkthdr.len; 567 q->F += (len << MY_M) / (uint64_t)fs->weight; 568 if (DN_KEY_LEQ(q->S, p->V)) 569 heap_insert(neh, q->S, q); 570 else 571 heap_insert(sch, q->F, q); 572 } 573 } 574 575 /* 576 * Now compute V = max(V, min(S_i)). Remember that all elements in 577 * sch have by definition S_i <= V so if sch is not empty, V is surely 578 * the max and we must not update it. Conversely, if sch is empty 579 * we only need to look at neh. 580 */ 581 if (sch->elements == 0 && neh->elements > 0) 582 p->V = MAX64(p->V, neh->p[0].key); 583 584 /* 585 * Move from neh to sch any packets that have become eligible 586 */ 587 while (neh->elements > 0 && DN_KEY_LEQ(neh->p[0].key, p->V)) { 588 struct dn_flow_queue *q = neh->p[0].object; 589 590 heap_extract(neh, NULL); 591 heap_insert(sch, q->F, q); 592 } 593 } 594 595 if (sch->elements == 0 && neh->elements == 0 && p->numbytes >= 0 && 596 p->idle_heap.elements > 0) { 597 /* 598 * No traffic and no events scheduled. We can get rid of idle-heap. 599 */ 600 int i; 601 602 for (i = 0; i < p->idle_heap.elements; i++) { 603 struct dn_flow_queue *q = p->idle_heap.p[i].object; 604 605 q->F = 0; 606 q->S = q->F + 1; 607 } 608 p->sum = 0; 609 p->V = 0; 610 p->idle_heap.elements = 0; 611 } 612 613 /* 614 * If we are getting clocks from dummynet and if we are under credit, 615 * schedule the next ready event. 616 * Also fix the delivery time of the last packet. 617 */ 618 if (p->numbytes < 0) { /* This implies bandwidth>0 */ 619 dn_key t = 0; /* Number of ticks i have to wait */ 620 621 if (p->bandwidth > 0) 622 t = (p->bandwidth - 1 - p->numbytes) / p->bandwidth; 623 TAILQ_LAST(&p->p_queue, dn_pkt_queue)->output_time += t; 624 p->sched_time = curr_time; 625 626 /* 627 * XXX should check errors on heap_insert, and drain the whole 628 * queue on error hoping next time we are luckier. 629 */ 630 heap_insert(&wfq_ready_heap, curr_time + t, p); 631 } 632 633 /* 634 * If the delay line was empty call transmit_event(p) now. 635 * Otherwise, the scheduler will take care of it. 636 */ 637 if (p_was_empty) 638 transmit_event(p); 639 } 640 641 static void 642 dn_expire_pipe_cb(struct dn_pipe *pipe, void *dummy __unused) 643 { 644 if (pipe->idle_heap.elements > 0 && 645 DN_KEY_LT(pipe->idle_heap.p[0].key, pipe->V)) { 646 struct dn_flow_queue *q = pipe->idle_heap.p[0].object; 647 648 heap_extract(&pipe->idle_heap, NULL); 649 q->S = q->F + 1; /* Mark timestamp as invalid */ 650 pipe->sum -= q->fs->weight; 651 } 652 } 653 654 /* 655 * This is called once per tick, or dn_hz times per second. It is used to 656 * increment the current tick counter and schedule expired events. 657 */ 658 static void 659 dummynet(netmsg_t msg) 660 { 661 void *p; 662 struct dn_heap *h; 663 struct dn_heap *heaps[3]; 664 int i; 665 666 heaps[0] = &ready_heap; /* Fixed-rate queues */ 667 heaps[1] = &wfq_ready_heap; /* WF2Q queues */ 668 heaps[2] = &extract_heap; /* Delay line */ 669 670 /* Reply ASAP */ 671 crit_enter(); 672 lwkt_replymsg(&msg->lmsg, 0); 673 crit_exit(); 674 675 curr_time++; 676 for (i = 0; i < 3; i++) { 677 h = heaps[i]; 678 while (h->elements > 0 && DN_KEY_LEQ(h->p[0].key, curr_time)) { 679 if (h->p[0].key > curr_time) { 680 kprintf("-- dummynet: warning, heap %d is %d ticks late\n", 681 i, (int)(curr_time - h->p[0].key)); 682 } 683 684 p = h->p[0].object; /* Store a copy before heap_extract */ 685 heap_extract(h, NULL); /* Need to extract before processing */ 686 687 if (i == 0) 688 ready_event(p); 689 else if (i == 1) 690 ready_event_wfq(p); 691 else 692 transmit_event(p); 693 } 694 } 695 696 /* Sweep pipes trying to expire idle flow_queues */ 697 dn_iterate_pipe(dn_expire_pipe_cb, NULL); 698 } 699 700 /* 701 * Unconditionally expire empty queues in case of shortage. 702 * Returns the number of queues freed. 703 */ 704 static int 705 expire_queues(struct dn_flow_set *fs) 706 { 707 int i, initial_elements = fs->rq_elements; 708 709 if (fs->last_expired == time_uptime) 710 return 0; 711 712 fs->last_expired = time_uptime; 713 714 for (i = 0; i <= fs->rq_size; i++) { /* Last one is overflow */ 715 struct dn_flow_queue *q, *qn; 716 717 LIST_FOREACH_MUTABLE(q, &fs->rq[i], q_link, qn) { 718 if (!TAILQ_EMPTY(&q->queue) || q->S != q->F + 1) 719 continue; 720 721 /* 722 * Entry is idle, expire it 723 */ 724 LIST_REMOVE(q, q_link); 725 kfree(q, M_DUMMYNET); 726 727 KASSERT(fs->rq_elements > 0, 728 ("invalid rq_elements %d", fs->rq_elements)); 729 fs->rq_elements--; 730 } 731 } 732 return initial_elements - fs->rq_elements; 733 } 734 735 /* 736 * If room, create a new queue and put at head of slot i; 737 * otherwise, create or use the default queue. 738 */ 739 static struct dn_flow_queue * 740 create_queue(struct dn_flow_set *fs, int i) 741 { 742 struct dn_flow_queue *q; 743 744 if (fs->rq_elements > fs->rq_size * dn_max_ratio && 745 expire_queues(fs) == 0) { 746 /* 747 * No way to get room, use or create overflow queue. 748 */ 749 i = fs->rq_size; 750 if (!LIST_EMPTY(&fs->rq[i])) 751 return LIST_FIRST(&fs->rq[i]); 752 } 753 754 q = kmalloc(sizeof(*q), M_DUMMYNET, M_INTWAIT | M_NULLOK | M_ZERO); 755 if (q == NULL) 756 return NULL; 757 758 q->fs = fs; 759 q->hash_slot = i; 760 q->S = q->F + 1; /* hack - mark timestamp as invalid */ 761 TAILQ_INIT(&q->queue); 762 763 LIST_INSERT_HEAD(&fs->rq[i], q, q_link); 764 fs->rq_elements++; 765 766 return q; 767 } 768 769 /* 770 * Given a flow_set and a pkt in last_pkt, find a matching queue 771 * after appropriate masking. The queue is moved to front 772 * so that further searches take less time. 773 */ 774 static struct dn_flow_queue * 775 find_queue(struct dn_flow_set *fs, struct dn_flow_id *id) 776 { 777 struct dn_flow_queue *q; 778 int i = 0; 779 780 if (!(fs->flags_fs & DN_HAVE_FLOW_MASK)) { 781 q = LIST_FIRST(&fs->rq[0]); 782 } else { 783 struct dn_flow_queue *qn; 784 785 /* First, do the masking */ 786 id->fid_dst_ip &= fs->flow_mask.fid_dst_ip; 787 id->fid_src_ip &= fs->flow_mask.fid_src_ip; 788 id->fid_dst_port &= fs->flow_mask.fid_dst_port; 789 id->fid_src_port &= fs->flow_mask.fid_src_port; 790 id->fid_proto &= fs->flow_mask.fid_proto; 791 id->fid_flags = 0; /* we don't care about this one */ 792 793 /* Then, hash function */ 794 i = ((id->fid_dst_ip) & 0xffff) ^ 795 ((id->fid_dst_ip >> 15) & 0xffff) ^ 796 ((id->fid_src_ip << 1) & 0xffff) ^ 797 ((id->fid_src_ip >> 16 ) & 0xffff) ^ 798 (id->fid_dst_port << 1) ^ (id->fid_src_port) ^ 799 (id->fid_proto); 800 i = i % fs->rq_size; 801 802 /* 803 * Finally, scan the current list for a match and 804 * expire idle flow queues 805 */ 806 searches++; 807 LIST_FOREACH_MUTABLE(q, &fs->rq[i], q_link, qn) { 808 search_steps++; 809 if (id->fid_dst_ip == q->id.fid_dst_ip && 810 id->fid_src_ip == q->id.fid_src_ip && 811 id->fid_dst_port == q->id.fid_dst_port && 812 id->fid_src_port == q->id.fid_src_port && 813 id->fid_proto == q->id.fid_proto && 814 id->fid_flags == q->id.fid_flags) { 815 break; /* Found */ 816 } else if (pipe_expire && TAILQ_EMPTY(&q->queue) && 817 q->S == q->F + 1) { 818 /* 819 * Entry is idle and not in any heap, expire it 820 */ 821 LIST_REMOVE(q, q_link); 822 kfree(q, M_DUMMYNET); 823 824 KASSERT(fs->rq_elements > 0, 825 ("invalid rq_elements %d", fs->rq_elements)); 826 fs->rq_elements--; 827 } 828 } 829 if (q && LIST_FIRST(&fs->rq[i]) != q) { /* Found and not in front */ 830 LIST_REMOVE(q, q_link); 831 LIST_INSERT_HEAD(&fs->rq[i], q, q_link); 832 } 833 } 834 if (q == NULL) { /* No match, need to allocate a new entry */ 835 q = create_queue(fs, i); 836 if (q != NULL) 837 q->id = *id; 838 } 839 return q; 840 } 841 842 static int 843 red_drops(struct dn_flow_set *fs, struct dn_flow_queue *q, int len) 844 { 845 /* 846 * RED algorithm 847 * 848 * RED calculates the average queue size (avg) using a low-pass filter 849 * with an exponential weighted (w_q) moving average: 850 * avg <- (1-w_q) * avg + w_q * q_size 851 * where q_size is the queue length (measured in bytes or * packets). 852 * 853 * If q_size == 0, we compute the idle time for the link, and set 854 * avg = (1 - w_q)^(idle/s) 855 * where s is the time needed for transmitting a medium-sized packet. 856 * 857 * Now, if avg < min_th the packet is enqueued. 858 * If avg > max_th the packet is dropped. Otherwise, the packet is 859 * dropped with probability P function of avg. 860 */ 861 862 int64_t p_b = 0; 863 u_int q_size = (fs->flags_fs & DN_QSIZE_IS_BYTES) ? q->len_bytes : q->len; 864 865 DPRINTF("\n%d q: %2u ", (int)curr_time, q_size); 866 867 /* Average queue size estimation */ 868 if (q_size != 0) { 869 /* 870 * Queue is not empty, avg <- avg + (q_size - avg) * w_q 871 */ 872 int diff = SCALE(q_size) - q->avg; 873 int64_t v = SCALE_MUL((int64_t)diff, (int64_t)fs->w_q); 874 875 q->avg += (int)v; 876 } else { 877 /* 878 * Queue is empty, find for how long the queue has been 879 * empty and use a lookup table for computing 880 * (1 - * w_q)^(idle_time/s) where s is the time to send a 881 * (small) packet. 882 * XXX check wraps... 883 */ 884 if (q->avg) { 885 u_int t = (curr_time - q->q_time) / fs->lookup_step; 886 887 q->avg = (t < fs->lookup_depth) ? 888 SCALE_MUL(q->avg, fs->w_q_lookup[t]) : 0; 889 } 890 } 891 DPRINTF("avg: %u ", SCALE_VAL(q->avg)); 892 893 /* Should i drop? */ 894 895 if (q->avg < fs->min_th) { 896 /* Accept packet */ 897 q->count = -1; 898 return 0; 899 } 900 901 if (q->avg >= fs->max_th) { /* Average queue >= Max threshold */ 902 if (fs->flags_fs & DN_IS_GENTLE_RED) { 903 /* 904 * According to Gentle-RED, if avg is greater than max_th the 905 * packet is dropped with a probability 906 * p_b = c_3 * avg - c_4 907 * where c_3 = (1 - max_p) / max_th, and c_4 = 1 - 2 * max_p 908 */ 909 p_b = SCALE_MUL((int64_t)fs->c_3, (int64_t)q->avg) - fs->c_4; 910 } else { 911 q->count = -1; 912 kprintf("- drop\n"); 913 return 1; 914 } 915 } else if (q->avg > fs->min_th) { 916 /* 917 * We compute p_b using the linear dropping function p_b = c_1 * 918 * avg - c_2, where c_1 = max_p / (max_th - min_th), and c_2 = 919 * max_p * min_th / (max_th - min_th) 920 */ 921 p_b = SCALE_MUL((int64_t)fs->c_1, (int64_t)q->avg) - fs->c_2; 922 } 923 if (fs->flags_fs & DN_QSIZE_IS_BYTES) 924 p_b = (p_b * len) / fs->max_pkt_size; 925 926 if (++q->count == 0) { 927 q->random = krandom() & 0xffff; 928 } else { 929 /* 930 * q->count counts packets arrived since last drop, so a greater 931 * value of q->count means a greater packet drop probability. 932 */ 933 if (SCALE_MUL(p_b, SCALE((int64_t)q->count)) > q->random) { 934 q->count = 0; 935 DPRINTF("%s", "- red drop"); 936 /* After a drop we calculate a new random value */ 937 q->random = krandom() & 0xffff; 938 return 1; /* Drop */ 939 } 940 } 941 /* End of RED algorithm */ 942 return 0; /* Accept */ 943 } 944 945 static void 946 dn_iterate_pipe(dn_pipe_iter_t func, void *arg) 947 { 948 int i; 949 950 for (i = 0; i < DN_NR_HASH_MAX; ++i) { 951 struct dn_pipe_head *pipe_hdr = &pipe_table[i]; 952 struct dn_pipe *pipe, *pipe_next; 953 954 LIST_FOREACH_MUTABLE(pipe, pipe_hdr, p_link, pipe_next) 955 func(pipe, arg); 956 } 957 } 958 959 static void 960 dn_iterate_flowset(dn_flowset_iter_t func, void *arg) 961 { 962 int i; 963 964 for (i = 0; i < DN_NR_HASH_MAX; ++i) { 965 struct dn_flowset_head *fs_hdr = &flowset_table[i]; 966 struct dn_flow_set *fs, *fs_next; 967 968 LIST_FOREACH_MUTABLE(fs, fs_hdr, fs_link, fs_next) 969 func(fs, arg); 970 } 971 } 972 973 static struct dn_pipe * 974 dn_find_pipe(int pipe_nr) 975 { 976 struct dn_pipe_head *pipe_hdr; 977 struct dn_pipe *p; 978 979 pipe_hdr = &pipe_table[DN_NR_HASH(pipe_nr)]; 980 LIST_FOREACH(p, pipe_hdr, p_link) { 981 if (p->pipe_nr == pipe_nr) 982 break; 983 } 984 return p; 985 } 986 987 static struct dn_flow_set * 988 dn_find_flowset(int fs_nr) 989 { 990 struct dn_flowset_head *fs_hdr; 991 struct dn_flow_set *fs; 992 993 fs_hdr = &flowset_table[DN_NR_HASH(fs_nr)]; 994 LIST_FOREACH(fs, fs_hdr, fs_link) { 995 if (fs->fs_nr == fs_nr) 996 break; 997 } 998 return fs; 999 } 1000 1001 static struct dn_flow_set * 1002 dn_locate_flowset(int pipe_nr, int is_pipe) 1003 { 1004 struct dn_flow_set *fs = NULL; 1005 1006 if (!is_pipe) { 1007 fs = dn_find_flowset(pipe_nr); 1008 } else { 1009 struct dn_pipe *p; 1010 1011 p = dn_find_pipe(pipe_nr); 1012 if (p != NULL) 1013 fs = &p->fs; 1014 } 1015 return fs; 1016 } 1017 1018 /* 1019 * Dummynet hook for packets. Below 'pipe' is a pipe or a queue 1020 * depending on whether WF2Q or fixed bw is used. 1021 * 1022 * pipe_nr pipe or queue the packet is destined for. 1023 * dir where shall we send the packet after dummynet. 1024 * m the mbuf with the packet 1025 * fwa->oif the 'ifp' parameter from the caller. 1026 * NULL in ip_input, destination interface in ip_output 1027 * fwa->ro route parameter (only used in ip_output, NULL otherwise) 1028 * fwa->dst destination address, only used by ip_output 1029 * fwa->rule matching rule, in case of multiple passes 1030 * fwa->flags flags from the caller, only used in ip_output 1031 */ 1032 static int 1033 dummynet_io(struct mbuf *m) 1034 { 1035 struct dn_pkt *pkt; 1036 struct m_tag *tag; 1037 struct dn_flow_set *fs; 1038 struct dn_pipe *pipe; 1039 uint64_t len = m->m_pkthdr.len; 1040 struct dn_flow_queue *q = NULL; 1041 int is_pipe, pipe_nr; 1042 1043 tag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); 1044 pkt = m_tag_data(tag); 1045 1046 is_pipe = pkt->dn_flags & DN_FLAGS_IS_PIPE; 1047 pipe_nr = pkt->pipe_nr; 1048 1049 /* 1050 * This is a dummynet rule, so we expect a O_PIPE or O_QUEUE rule 1051 */ 1052 fs = dn_locate_flowset(pipe_nr, is_pipe); 1053 if (fs == NULL) 1054 goto dropit; /* This queue/pipe does not exist! */ 1055 1056 pipe = fs->pipe; 1057 if (pipe == NULL) { /* Must be a queue, try find a matching pipe */ 1058 pipe = dn_find_pipe(fs->parent_nr); 1059 if (pipe != NULL) { 1060 fs->pipe = pipe; 1061 } else { 1062 kprintf("No pipe %d for queue %d, drop pkt\n", 1063 fs->parent_nr, fs->fs_nr); 1064 goto dropit; 1065 } 1066 } 1067 1068 q = find_queue(fs, &pkt->id); 1069 if (q == NULL) 1070 goto dropit; /* Cannot allocate queue */ 1071 1072 /* 1073 * Update statistics, then check reasons to drop pkt 1074 */ 1075 q->tot_bytes += len; 1076 q->tot_pkts++; 1077 1078 if (fs->plr && krandom() < fs->plr) 1079 goto dropit; /* Random pkt drop */ 1080 1081 if (fs->flags_fs & DN_QSIZE_IS_BYTES) { 1082 if (q->len_bytes > fs->qsize) 1083 goto dropit; /* Queue size overflow */ 1084 } else { 1085 if (q->len >= fs->qsize) 1086 goto dropit; /* Queue count overflow */ 1087 } 1088 1089 if ((fs->flags_fs & DN_IS_RED) && red_drops(fs, q, len)) 1090 goto dropit; 1091 1092 TAILQ_INSERT_TAIL(&q->queue, pkt, dn_next); 1093 q->len++; 1094 q->len_bytes += len; 1095 1096 if (TAILQ_FIRST(&q->queue) != pkt) /* Flow was not idle, we are done */ 1097 goto done; 1098 1099 /* 1100 * If we reach this point the flow was previously idle, so we need 1101 * to schedule it. This involves different actions for fixed-rate 1102 * or WF2Q queues. 1103 */ 1104 if (is_pipe) { 1105 /* 1106 * Fixed-rate queue: just insert into the ready_heap. 1107 */ 1108 dn_key t = 0; 1109 1110 if (pipe->bandwidth) 1111 t = SET_TICKS(pkt, q, pipe); 1112 1113 q->sched_time = curr_time; 1114 if (t == 0) /* Must process it now */ 1115 ready_event(q); 1116 else 1117 heap_insert(&ready_heap, curr_time + t, q); 1118 } else { 1119 /* 1120 * WF2Q: 1121 * First, compute start time S: if the flow was idle (S=F+1) 1122 * set S to the virtual time V for the controlling pipe, and update 1123 * the sum of weights for the pipe; otherwise, remove flow from 1124 * idle_heap and set S to max(F, V). 1125 * Second, compute finish time F = S + len/weight. 1126 * Third, if pipe was idle, update V = max(S, V). 1127 * Fourth, count one more backlogged flow. 1128 */ 1129 if (DN_KEY_GT(q->S, q->F)) { /* Means timestamps are invalid */ 1130 q->S = pipe->V; 1131 pipe->sum += fs->weight; /* Add weight of new queue */ 1132 } else { 1133 heap_extract(&pipe->idle_heap, q); 1134 q->S = MAX64(q->F, pipe->V); 1135 } 1136 q->F = q->S + (len << MY_M) / (uint64_t)fs->weight; 1137 1138 if (pipe->not_eligible_heap.elements == 0 && 1139 pipe->scheduler_heap.elements == 0) 1140 pipe->V = MAX64(q->S, pipe->V); 1141 1142 fs->backlogged++; 1143 1144 /* 1145 * Look at eligibility. A flow is not eligibile if S>V (when 1146 * this happens, it means that there is some other flow already 1147 * scheduled for the same pipe, so the scheduler_heap cannot be 1148 * empty). If the flow is not eligible we just store it in the 1149 * not_eligible_heap. Otherwise, we store in the scheduler_heap 1150 * and possibly invoke ready_event_wfq() right now if there is 1151 * leftover credit. 1152 * Note that for all flows in scheduler_heap (SCH), S_i <= V, 1153 * and for all flows in not_eligible_heap (NEH), S_i > V. 1154 * So when we need to compute max(V, min(S_i)) forall i in SCH+NEH, 1155 * we only need to look into NEH. 1156 */ 1157 if (DN_KEY_GT(q->S, pipe->V)) { /* Not eligible */ 1158 if (pipe->scheduler_heap.elements == 0) 1159 kprintf("++ ouch! not eligible but empty scheduler!\n"); 1160 heap_insert(&pipe->not_eligible_heap, q->S, q); 1161 } else { 1162 heap_insert(&pipe->scheduler_heap, q->F, q); 1163 if (pipe->numbytes >= 0) { /* Pipe is idle */ 1164 if (pipe->scheduler_heap.elements != 1) 1165 kprintf("*** OUCH! pipe should have been idle!\n"); 1166 DPRINTF("Waking up pipe %d at %d\n", 1167 pipe->pipe_nr, (int)(q->F >> MY_M)); 1168 pipe->sched_time = curr_time; 1169 ready_event_wfq(pipe); 1170 } 1171 } 1172 } 1173 done: 1174 return 0; 1175 1176 dropit: 1177 if (q) 1178 q->drops++; 1179 return ENOBUFS; 1180 } 1181 1182 /* 1183 * Dispose all packets and flow_queues on a flow_set. 1184 * If all=1, also remove red lookup table and other storage, 1185 * including the descriptor itself. 1186 * For the one in dn_pipe MUST also cleanup ready_heap... 1187 */ 1188 static void 1189 purge_flow_set(struct dn_flow_set *fs, int all) 1190 { 1191 int i; 1192 #ifdef INVARIANTS 1193 int rq_elements = 0; 1194 #endif 1195 1196 for (i = 0; i <= fs->rq_size; i++) { 1197 struct dn_flow_queue *q; 1198 1199 while ((q = LIST_FIRST(&fs->rq[i])) != NULL) { 1200 struct dn_pkt *pkt; 1201 1202 while ((pkt = TAILQ_FIRST(&q->queue)) != NULL) { 1203 TAILQ_REMOVE(&q->queue, pkt, dn_next); 1204 ip_dn_packet_free(pkt); 1205 } 1206 1207 LIST_REMOVE(q, q_link); 1208 kfree(q, M_DUMMYNET); 1209 1210 #ifdef INVARIANTS 1211 rq_elements++; 1212 #endif 1213 } 1214 } 1215 KASSERT(rq_elements == fs->rq_elements, 1216 ("# rq elements mismatch, freed %d, total %d", 1217 rq_elements, fs->rq_elements)); 1218 fs->rq_elements = 0; 1219 1220 if (all) { 1221 /* RED - free lookup table */ 1222 if (fs->w_q_lookup) 1223 kfree(fs->w_q_lookup, M_DUMMYNET); 1224 1225 if (fs->rq) 1226 kfree(fs->rq, M_DUMMYNET); 1227 1228 /* 1229 * If this fs is not part of a pipe, free it 1230 * 1231 * fs->pipe == NULL could happen, if 'fs' is a WF2Q and 1232 * - No packet belongs to that flow set is delivered by 1233 * dummynet_io(), i.e. parent pipe is not installed yet. 1234 * - Parent pipe is deleted. 1235 */ 1236 if (fs->pipe == NULL || (fs->pipe && fs != &fs->pipe->fs)) 1237 kfree(fs, M_DUMMYNET); 1238 } 1239 } 1240 1241 /* 1242 * Dispose all packets queued on a pipe (not a flow_set). 1243 * Also free all resources associated to a pipe, which is about 1244 * to be deleted. 1245 */ 1246 static void 1247 purge_pipe(struct dn_pipe *pipe) 1248 { 1249 struct dn_pkt *pkt; 1250 1251 purge_flow_set(&pipe->fs, 1); 1252 1253 while ((pkt = TAILQ_FIRST(&pipe->p_queue)) != NULL) { 1254 TAILQ_REMOVE(&pipe->p_queue, pkt, dn_next); 1255 ip_dn_packet_free(pkt); 1256 } 1257 1258 heap_free(&pipe->scheduler_heap); 1259 heap_free(&pipe->not_eligible_heap); 1260 heap_free(&pipe->idle_heap); 1261 } 1262 1263 /* 1264 * Delete all pipes and heaps returning memory. 1265 */ 1266 static void 1267 dummynet_flush(void) 1268 { 1269 struct dn_pipe_head pipe_list; 1270 struct dn_flowset_head fs_list; 1271 struct dn_pipe *p; 1272 struct dn_flow_set *fs; 1273 int i; 1274 1275 lockmgr(&dn_lock, LK_EXCLUSIVE); 1276 1277 /* 1278 * Prevent future matches... 1279 */ 1280 LIST_INIT(&pipe_list); 1281 for (i = 0; i < DN_NR_HASH_MAX; ++i) { 1282 struct dn_pipe_head *pipe_hdr = &pipe_table[i]; 1283 1284 while ((p = LIST_FIRST(pipe_hdr)) != NULL) { 1285 LIST_REMOVE(p, p_link); 1286 LIST_INSERT_HEAD(&pipe_list, p, p_link); 1287 --dn_count; 1288 } 1289 } 1290 1291 LIST_INIT(&fs_list); 1292 for (i = 0; i < DN_NR_HASH_MAX; ++i) { 1293 struct dn_flowset_head *fs_hdr = &flowset_table[i]; 1294 1295 while ((fs = LIST_FIRST(fs_hdr)) != NULL) { 1296 LIST_REMOVE(fs, fs_link); 1297 LIST_INSERT_HEAD(&fs_list, fs, fs_link); 1298 --dn_count; 1299 } 1300 } 1301 1302 /* Free heaps so we don't have unwanted events */ 1303 heap_free(&ready_heap); 1304 heap_free(&wfq_ready_heap); 1305 heap_free(&extract_heap); 1306 1307 /* 1308 * Now purge all queued pkts and delete all pipes 1309 */ 1310 /* Scan and purge all flow_sets. */ 1311 while ((fs = LIST_FIRST(&fs_list)) != NULL) { 1312 LIST_REMOVE(fs, fs_link); 1313 purge_flow_set(fs, 1); 1314 } 1315 1316 while ((p = LIST_FIRST(&pipe_list)) != NULL) { 1317 LIST_REMOVE(p, p_link); 1318 purge_pipe(p); 1319 kfree(p, M_DUMMYNET); 1320 } 1321 1322 /* 1323 * Everything has been cleaned out, clear the run state. 1324 */ 1325 KKASSERT(dn_count == 0); 1326 if (dn_running) { 1327 systimer_del(&dn_clock); 1328 dn_running = 0; 1329 } 1330 lockmgr(&dn_lock, LK_RELEASE); 1331 } 1332 1333 /* 1334 * setup RED parameters 1335 */ 1336 static int 1337 config_red(const struct dn_ioc_flowset *ioc_fs, struct dn_flow_set *x) 1338 { 1339 int i; 1340 1341 x->w_q = ioc_fs->w_q; 1342 x->min_th = SCALE(ioc_fs->min_th); 1343 x->max_th = SCALE(ioc_fs->max_th); 1344 x->max_p = ioc_fs->max_p; 1345 1346 x->c_1 = ioc_fs->max_p / (ioc_fs->max_th - ioc_fs->min_th); 1347 x->c_2 = SCALE_MUL(x->c_1, SCALE(ioc_fs->min_th)); 1348 if (x->flags_fs & DN_IS_GENTLE_RED) { 1349 x->c_3 = (SCALE(1) - ioc_fs->max_p) / ioc_fs->max_th; 1350 x->c_4 = (SCALE(1) - 2 * ioc_fs->max_p); 1351 } 1352 1353 /* If the lookup table already exist, free and create it again */ 1354 if (x->w_q_lookup) { 1355 kfree(x->w_q_lookup, M_DUMMYNET); 1356 x->w_q_lookup = NULL ; 1357 } 1358 1359 if (red_lookup_depth == 0) { 1360 kprintf("net.inet.ip.dummynet.red_lookup_depth must be > 0\n"); 1361 kfree(x, M_DUMMYNET); 1362 return EINVAL; 1363 } 1364 x->lookup_depth = red_lookup_depth; 1365 x->w_q_lookup = kmalloc(x->lookup_depth * sizeof(int), 1366 M_DUMMYNET, M_WAITOK); 1367 1368 /* Fill the lookup table with (1 - w_q)^x */ 1369 x->lookup_step = ioc_fs->lookup_step; 1370 x->lookup_weight = ioc_fs->lookup_weight; 1371 1372 x->w_q_lookup[0] = SCALE(1) - x->w_q; 1373 for (i = 1; i < x->lookup_depth; i++) 1374 x->w_q_lookup[i] = SCALE_MUL(x->w_q_lookup[i - 1], x->lookup_weight); 1375 1376 if (red_avg_pkt_size < 1) 1377 red_avg_pkt_size = 512; 1378 x->avg_pkt_size = red_avg_pkt_size; 1379 1380 if (red_max_pkt_size < 1) 1381 red_max_pkt_size = 1500; 1382 x->max_pkt_size = red_max_pkt_size; 1383 1384 return 0; 1385 } 1386 1387 static void 1388 alloc_hash(struct dn_flow_set *x, const struct dn_ioc_flowset *ioc_fs) 1389 { 1390 int i, alloc_size; 1391 1392 if (x->flags_fs & DN_HAVE_FLOW_MASK) { 1393 int l = ioc_fs->rq_size; 1394 1395 /* Allocate some slots */ 1396 if (l == 0) 1397 l = dn_hash_size; 1398 1399 if (l < DN_MIN_HASH_SIZE) 1400 l = DN_MIN_HASH_SIZE; 1401 else if (l > DN_MAX_HASH_SIZE) 1402 l = DN_MAX_HASH_SIZE; 1403 1404 x->rq_size = l; 1405 } else { 1406 /* One is enough for null mask */ 1407 x->rq_size = 1; 1408 } 1409 alloc_size = x->rq_size + 1; 1410 1411 x->rq = kmalloc(alloc_size * sizeof(struct dn_flowqueue_head), 1412 M_DUMMYNET, M_WAITOK | M_ZERO); 1413 x->rq_elements = 0; 1414 1415 for (i = 0; i < alloc_size; ++i) 1416 LIST_INIT(&x->rq[i]); 1417 } 1418 1419 static void 1420 set_flowid_parms(struct dn_flow_id *id, const struct dn_ioc_flowid *ioc_id) 1421 { 1422 id->fid_dst_ip = ioc_id->u.ip.dst_ip; 1423 id->fid_src_ip = ioc_id->u.ip.src_ip; 1424 id->fid_dst_port = ioc_id->u.ip.dst_port; 1425 id->fid_src_port = ioc_id->u.ip.src_port; 1426 id->fid_proto = ioc_id->u.ip.proto; 1427 id->fid_flags = ioc_id->u.ip.flags; 1428 } 1429 1430 static void 1431 set_fs_parms(struct dn_flow_set *x, const struct dn_ioc_flowset *ioc_fs) 1432 { 1433 x->flags_fs = ioc_fs->flags_fs; 1434 x->qsize = ioc_fs->qsize; 1435 x->plr = ioc_fs->plr; 1436 set_flowid_parms(&x->flow_mask, &ioc_fs->flow_mask); 1437 if (x->flags_fs & DN_QSIZE_IS_BYTES) { 1438 if (x->qsize > 1024 * 1024) 1439 x->qsize = 1024 * 1024; 1440 } else { 1441 if (x->qsize == 0 || x->qsize > 100) 1442 x->qsize = 50; 1443 } 1444 1445 /* Configuring RED */ 1446 if (x->flags_fs & DN_IS_RED) 1447 config_red(ioc_fs, x); /* XXX should check errors */ 1448 } 1449 1450 /* 1451 * setup pipe or queue parameters. 1452 */ 1453 1454 static int 1455 config_pipe(struct dn_ioc_pipe *ioc_pipe) 1456 { 1457 struct dn_ioc_flowset *ioc_fs = &ioc_pipe->fs; 1458 int error; 1459 1460 /* 1461 * The config program passes parameters as follows: 1462 * bw bits/second (0 means no limits) 1463 * delay ms (must be translated into ticks) 1464 * qsize slots or bytes 1465 */ 1466 ioc_pipe->delay = (ioc_pipe->delay * dn_hz) / 1000; 1467 1468 /* 1469 * We need either a pipe number or a flow_set number 1470 */ 1471 if (ioc_pipe->pipe_nr == 0 && ioc_fs->fs_nr == 0) 1472 return EINVAL; 1473 if (ioc_pipe->pipe_nr != 0 && ioc_fs->fs_nr != 0) 1474 return EINVAL; 1475 1476 /* 1477 * Validate pipe number 1478 */ 1479 if (ioc_pipe->pipe_nr > DN_PIPE_NR_MAX || ioc_pipe->pipe_nr < 0) 1480 return EINVAL; 1481 1482 lockmgr(&dn_lock, LK_EXCLUSIVE); 1483 error = EINVAL; 1484 1485 if (ioc_pipe->pipe_nr != 0) { /* This is a pipe */ 1486 struct dn_pipe *x, *p; 1487 1488 /* Locate pipe */ 1489 p = dn_find_pipe(ioc_pipe->pipe_nr); 1490 1491 if (p == NULL) { /* New pipe */ 1492 x = kmalloc(sizeof(struct dn_pipe), M_DUMMYNET, M_WAITOK | M_ZERO); 1493 x->pipe_nr = ioc_pipe->pipe_nr; 1494 x->fs.pipe = x; 1495 TAILQ_INIT(&x->p_queue); 1496 1497 /* 1498 * idle_heap is the only one from which we extract from the middle. 1499 */ 1500 x->idle_heap.size = x->idle_heap.elements = 0; 1501 x->idle_heap.offset = __offsetof(struct dn_flow_queue, heap_pos); 1502 } else { 1503 int i; 1504 1505 x = p; 1506 1507 /* Flush accumulated credit for all queues */ 1508 for (i = 0; i <= x->fs.rq_size; i++) { 1509 struct dn_flow_queue *q; 1510 1511 LIST_FOREACH(q, &x->fs.rq[i], q_link) 1512 q->numbytes = 0; 1513 } 1514 } 1515 1516 x->bandwidth = ioc_pipe->bandwidth; 1517 x->numbytes = 0; /* Just in case... */ 1518 x->delay = ioc_pipe->delay; 1519 1520 set_fs_parms(&x->fs, ioc_fs); 1521 1522 if (x->fs.rq == NULL) { /* A new pipe */ 1523 struct dn_pipe_head *pipe_hdr; 1524 1525 alloc_hash(&x->fs, ioc_fs); 1526 1527 pipe_hdr = &pipe_table[DN_NR_HASH(x->pipe_nr)]; 1528 LIST_INSERT_HEAD(pipe_hdr, x, p_link); 1529 ++dn_count; 1530 } 1531 } else { /* Config flow_set */ 1532 struct dn_flow_set *x, *fs; 1533 1534 /* Locate flow_set */ 1535 fs = dn_find_flowset(ioc_fs->fs_nr); 1536 1537 if (fs == NULL) { /* New flow_set */ 1538 if (ioc_fs->parent_nr == 0) /* Need link to a pipe */ 1539 goto back; 1540 1541 x = kmalloc(sizeof(struct dn_flow_set), M_DUMMYNET, 1542 M_WAITOK | M_ZERO); 1543 x->fs_nr = ioc_fs->fs_nr; 1544 x->parent_nr = ioc_fs->parent_nr; 1545 x->weight = ioc_fs->weight; 1546 if (x->weight == 0) 1547 x->weight = 1; 1548 else if (x->weight > 100) 1549 x->weight = 100; 1550 } else { 1551 /* Change parent pipe not allowed; must delete and recreate */ 1552 if (ioc_fs->parent_nr != 0 && fs->parent_nr != ioc_fs->parent_nr) 1553 goto back; 1554 x = fs; 1555 } 1556 1557 set_fs_parms(x, ioc_fs); 1558 1559 if (x->rq == NULL) { /* A new flow_set */ 1560 struct dn_flowset_head *fs_hdr; 1561 1562 alloc_hash(x, ioc_fs); 1563 1564 fs_hdr = &flowset_table[DN_NR_HASH(x->fs_nr)]; 1565 LIST_INSERT_HEAD(fs_hdr, x, fs_link); 1566 ++dn_count; 1567 } 1568 } 1569 error = 0; 1570 1571 /* 1572 * We have at least one entry, set run state and start the systimer 1573 * poll if necessary. 1574 */ 1575 if (dn_running == 0) { 1576 dn_running = 1; 1577 systimer_init_periodic_nq(&dn_clock, dummynet_clock, NULL, dn_hz); 1578 } 1579 1580 back: 1581 lockmgr(&dn_lock, LK_RELEASE); 1582 return error; 1583 } 1584 1585 /* 1586 * Helper function to remove from a heap queues which are linked to 1587 * a flow_set about to be deleted. 1588 */ 1589 static void 1590 fs_remove_from_heap(struct dn_heap *h, struct dn_flow_set *fs) 1591 { 1592 int i = 0, found = 0; 1593 1594 while (i < h->elements) { 1595 if (((struct dn_flow_queue *)h->p[i].object)->fs == fs) { 1596 h->elements--; 1597 h->p[i] = h->p[h->elements]; 1598 found++; 1599 } else { 1600 i++; 1601 } 1602 } 1603 if (found) 1604 heapify(h); 1605 } 1606 1607 /* 1608 * helper function to remove a pipe from a heap (can be there at most once) 1609 */ 1610 static void 1611 pipe_remove_from_heap(struct dn_heap *h, struct dn_pipe *p) 1612 { 1613 if (h->elements > 0) { 1614 int i; 1615 1616 for (i = 0; i < h->elements; i++) { 1617 if (h->p[i].object == p) { /* found it */ 1618 h->elements--; 1619 h->p[i] = h->p[h->elements]; 1620 heapify(h); 1621 break; 1622 } 1623 } 1624 } 1625 } 1626 1627 static void 1628 dn_unref_pipe_cb(struct dn_flow_set *fs, void *pipe0) 1629 { 1630 struct dn_pipe *pipe = pipe0; 1631 1632 if (fs->pipe == pipe) { 1633 kprintf("++ ref to pipe %d from fs %d\n", 1634 pipe->pipe_nr, fs->fs_nr); 1635 fs->pipe = NULL; 1636 purge_flow_set(fs, 0); 1637 } 1638 } 1639 1640 /* 1641 * Fully delete a pipe or a queue, cleaning up associated info. 1642 */ 1643 static int 1644 delete_pipe(const struct dn_ioc_pipe *ioc_pipe) 1645 { 1646 struct dn_pipe *p; 1647 int error; 1648 1649 if (ioc_pipe->pipe_nr == 0 && ioc_pipe->fs.fs_nr == 0) 1650 return EINVAL; 1651 if (ioc_pipe->pipe_nr != 0 && ioc_pipe->fs.fs_nr != 0) 1652 return EINVAL; 1653 1654 if (ioc_pipe->pipe_nr > DN_NR_HASH_MAX || ioc_pipe->pipe_nr < 0) 1655 return EINVAL; 1656 1657 lockmgr(&dn_lock, LK_EXCLUSIVE); 1658 1659 error = EINVAL; 1660 if (ioc_pipe->pipe_nr != 0) { /* This is an old-style pipe */ 1661 /* Locate pipe */ 1662 p = dn_find_pipe(ioc_pipe->pipe_nr); 1663 if (p == NULL) 1664 goto back; /* Not found */ 1665 1666 /* Unlink from pipe hash table */ 1667 LIST_REMOVE(p, p_link); 1668 --dn_count; 1669 1670 /* Remove all references to this pipe from flow_sets */ 1671 dn_iterate_flowset(dn_unref_pipe_cb, p); 1672 1673 fs_remove_from_heap(&ready_heap, &p->fs); 1674 purge_pipe(p); /* Remove all data associated to this pipe */ 1675 1676 /* Remove reference to here from extract_heap and wfq_ready_heap */ 1677 pipe_remove_from_heap(&extract_heap, p); 1678 pipe_remove_from_heap(&wfq_ready_heap, p); 1679 1680 kfree(p, M_DUMMYNET); 1681 } else { /* This is a WF2Q queue (dn_flow_set) */ 1682 struct dn_flow_set *fs; 1683 1684 /* Locate flow_set */ 1685 fs = dn_find_flowset(ioc_pipe->fs.fs_nr); 1686 if (fs == NULL) 1687 goto back; /* Not found */ 1688 1689 LIST_REMOVE(fs, fs_link); 1690 --dn_count; 1691 1692 if ((p = fs->pipe) != NULL) { 1693 /* Update total weight on parent pipe and cleanup parent heaps */ 1694 p->sum -= fs->weight * fs->backlogged; 1695 fs_remove_from_heap(&p->not_eligible_heap, fs); 1696 fs_remove_from_heap(&p->scheduler_heap, fs); 1697 #if 1 /* XXX should i remove from idle_heap as well ? */ 1698 fs_remove_from_heap(&p->idle_heap, fs); 1699 #endif 1700 } 1701 purge_flow_set(fs, 1); 1702 } 1703 error = 0; 1704 1705 /* 1706 * If there are no more pipes or flow-sets, clear the run state. 1707 */ 1708 if (dn_count == 0 && dn_running) { 1709 systimer_del(&dn_clock); 1710 dn_running = 0; 1711 } 1712 back: 1713 lockmgr(&dn_lock, LK_RELEASE); 1714 1715 return error; 1716 } 1717 1718 /* 1719 * helper function used to copy data from kernel in DUMMYNET_GET 1720 */ 1721 static void 1722 dn_copy_flowid(const struct dn_flow_id *id, struct dn_ioc_flowid *ioc_id) 1723 { 1724 ioc_id->type = ETHERTYPE_IP; 1725 ioc_id->u.ip.dst_ip = id->fid_dst_ip; 1726 ioc_id->u.ip.src_ip = id->fid_src_ip; 1727 ioc_id->u.ip.dst_port = id->fid_dst_port; 1728 ioc_id->u.ip.src_port = id->fid_src_port; 1729 ioc_id->u.ip.proto = id->fid_proto; 1730 ioc_id->u.ip.flags = id->fid_flags; 1731 } 1732 1733 static void * 1734 dn_copy_flowqueues(const struct dn_flow_set *fs, void *bp) 1735 { 1736 struct dn_ioc_flowqueue *ioc_fq = bp; 1737 int i, copied = 0; 1738 1739 for (i = 0; i <= fs->rq_size; i++) { 1740 const struct dn_flow_queue *q; 1741 1742 LIST_FOREACH(q, &fs->rq[i], q_link) { 1743 if (q->hash_slot != i) { /* XXX ASSERT */ 1744 kprintf("++ at %d: wrong slot (have %d, " 1745 "should be %d)\n", copied, q->hash_slot, i); 1746 } 1747 if (q->fs != fs) { /* XXX ASSERT */ 1748 kprintf("++ at %d: wrong fs ptr (have %p, should be %p)\n", 1749 i, q->fs, fs); 1750 } 1751 1752 copied++; 1753 1754 ioc_fq->len = q->len; 1755 ioc_fq->len_bytes = q->len_bytes; 1756 ioc_fq->tot_pkts = q->tot_pkts; 1757 ioc_fq->tot_bytes = q->tot_bytes; 1758 ioc_fq->drops = q->drops; 1759 ioc_fq->hash_slot = q->hash_slot; 1760 ioc_fq->S = q->S; 1761 ioc_fq->F = q->F; 1762 dn_copy_flowid(&q->id, &ioc_fq->id); 1763 1764 ioc_fq++; 1765 } 1766 } 1767 1768 if (copied != fs->rq_elements) { /* XXX ASSERT */ 1769 kprintf("++ wrong count, have %d should be %d\n", 1770 copied, fs->rq_elements); 1771 } 1772 return ioc_fq; 1773 } 1774 1775 static void 1776 dn_copy_flowset(const struct dn_flow_set *fs, struct dn_ioc_flowset *ioc_fs, 1777 u_short fs_type) 1778 { 1779 ioc_fs->fs_type = fs_type; 1780 1781 ioc_fs->fs_nr = fs->fs_nr; 1782 ioc_fs->flags_fs = fs->flags_fs; 1783 ioc_fs->parent_nr = fs->parent_nr; 1784 1785 ioc_fs->weight = fs->weight; 1786 ioc_fs->qsize = fs->qsize; 1787 ioc_fs->plr = fs->plr; 1788 1789 ioc_fs->rq_size = fs->rq_size; 1790 ioc_fs->rq_elements = fs->rq_elements; 1791 1792 ioc_fs->w_q = fs->w_q; 1793 ioc_fs->max_th = fs->max_th; 1794 ioc_fs->min_th = fs->min_th; 1795 ioc_fs->max_p = fs->max_p; 1796 1797 dn_copy_flowid(&fs->flow_mask, &ioc_fs->flow_mask); 1798 } 1799 1800 static void 1801 dn_calc_pipe_size_cb(struct dn_pipe *pipe, void *sz) 1802 { 1803 size_t *size = sz; 1804 1805 *size += sizeof(struct dn_ioc_pipe) + 1806 pipe->fs.rq_elements * sizeof(struct dn_ioc_flowqueue); 1807 } 1808 1809 static void 1810 dn_calc_fs_size_cb(struct dn_flow_set *fs, void *sz) 1811 { 1812 size_t *size = sz; 1813 1814 *size += sizeof(struct dn_ioc_flowset) + 1815 fs->rq_elements * sizeof(struct dn_ioc_flowqueue); 1816 } 1817 1818 static void 1819 dn_copyout_pipe_cb(struct dn_pipe *pipe, void *bp0) 1820 { 1821 char **bp = bp0; 1822 struct dn_ioc_pipe *ioc_pipe = (struct dn_ioc_pipe *)(*bp); 1823 1824 /* 1825 * Copy flow set descriptor associated with this pipe 1826 */ 1827 dn_copy_flowset(&pipe->fs, &ioc_pipe->fs, DN_IS_PIPE); 1828 1829 /* 1830 * Copy pipe descriptor 1831 */ 1832 ioc_pipe->bandwidth = pipe->bandwidth; 1833 ioc_pipe->pipe_nr = pipe->pipe_nr; 1834 ioc_pipe->V = pipe->V; 1835 /* Convert delay to milliseconds */ 1836 ioc_pipe->delay = (pipe->delay * 1000) / dn_hz; 1837 1838 /* 1839 * Copy flow queue descriptors 1840 */ 1841 *bp += sizeof(*ioc_pipe); 1842 *bp = dn_copy_flowqueues(&pipe->fs, *bp); 1843 } 1844 1845 static void 1846 dn_copyout_fs_cb(struct dn_flow_set *fs, void *bp0) 1847 { 1848 char **bp = bp0; 1849 struct dn_ioc_flowset *ioc_fs = (struct dn_ioc_flowset *)(*bp); 1850 1851 /* 1852 * Copy flow set descriptor 1853 */ 1854 dn_copy_flowset(fs, ioc_fs, DN_IS_QUEUE); 1855 1856 /* 1857 * Copy flow queue descriptors 1858 */ 1859 *bp += sizeof(*ioc_fs); 1860 *bp = dn_copy_flowqueues(fs, *bp); 1861 } 1862 1863 static int 1864 dummynet_get(struct dn_sopt *dn_sopt) 1865 { 1866 char *buf, *bp; 1867 size_t size = 0; 1868 1869 /* 1870 * Compute size of data structures: list of pipes and flow_sets. 1871 */ 1872 dn_iterate_pipe(dn_calc_pipe_size_cb, &size); 1873 dn_iterate_flowset(dn_calc_fs_size_cb, &size); 1874 1875 /* 1876 * Copyout pipe/flow_set/flow_queue 1877 */ 1878 bp = buf = kmalloc(size, M_TEMP, M_WAITOK | M_ZERO); 1879 dn_iterate_pipe(dn_copyout_pipe_cb, &bp); 1880 dn_iterate_flowset(dn_copyout_fs_cb, &bp); 1881 1882 /* Temp memory will be freed by caller */ 1883 dn_sopt->dn_sopt_arg = buf; 1884 dn_sopt->dn_sopt_arglen = size; 1885 return 0; 1886 } 1887 1888 /* 1889 * Handler for the various dummynet socket options (get, flush, config, del) 1890 */ 1891 static int 1892 dummynet_ctl(struct dn_sopt *dn_sopt) 1893 { 1894 int error = 0; 1895 1896 switch (dn_sopt->dn_sopt_name) { 1897 case IP_DUMMYNET_GET: 1898 error = dummynet_get(dn_sopt); 1899 break; 1900 1901 case IP_DUMMYNET_FLUSH: 1902 dummynet_flush(); 1903 break; 1904 1905 case IP_DUMMYNET_CONFIGURE: 1906 KKASSERT(dn_sopt->dn_sopt_arglen == sizeof(struct dn_ioc_pipe)); 1907 error = config_pipe(dn_sopt->dn_sopt_arg); 1908 break; 1909 1910 case IP_DUMMYNET_DEL: /* Remove a pipe or flow_set */ 1911 KKASSERT(dn_sopt->dn_sopt_arglen == sizeof(struct dn_ioc_pipe)); 1912 error = delete_pipe(dn_sopt->dn_sopt_arg); 1913 break; 1914 1915 default: 1916 kprintf("%s -- unknown option %d\n", __func__, dn_sopt->dn_sopt_name); 1917 error = EINVAL; 1918 break; 1919 } 1920 return error; 1921 } 1922 1923 static void 1924 dummynet_clock(systimer_t info __unused, int in_ipi __unused, 1925 struct intrframe *frame __unused) 1926 { 1927 KASSERT(mycpuid == ip_dn_cpu, 1928 ("dummynet systimer comes on cpu%d, should be %d!", 1929 mycpuid, ip_dn_cpu)); 1930 1931 crit_enter(); 1932 if (DUMMYNET_LOADED && (dn_netmsg.lmsg.ms_flags & MSGF_DONE)) 1933 lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), &dn_netmsg.lmsg); 1934 crit_exit(); 1935 } 1936 1937 static int 1938 sysctl_dn_hz(SYSCTL_HANDLER_ARGS) 1939 { 1940 int error, val, origcpu; 1941 1942 val = dn_hz; 1943 error = sysctl_handle_int(oidp, &val, 0, req); 1944 if (error || req->newptr == NULL) 1945 return error; 1946 if (val <= 0) 1947 return EINVAL; 1948 else if (val > DN_CALLOUT_FREQ_MAX) 1949 val = DN_CALLOUT_FREQ_MAX; 1950 1951 origcpu = mycpuid; 1952 lwkt_migratecpu(ip_dn_cpu); 1953 1954 lockmgr(&dn_lock, LK_EXCLUSIVE); 1955 crit_enter(); 1956 dn_hz = val; 1957 if (dn_running) 1958 systimer_adjust_periodic(&dn_clock, val); 1959 crit_exit(); 1960 lockmgr(&dn_lock, LK_RELEASE); 1961 1962 lwkt_migratecpu(origcpu); 1963 1964 return 0; 1965 } 1966 1967 static void 1968 ip_dn_init_dispatch(netmsg_t msg) 1969 { 1970 int i, error = 0; 1971 1972 KASSERT(mycpuid == ip_dn_cpu, 1973 ("%s runs on cpu%d, instead of cpu%d", __func__, 1974 mycpuid, ip_dn_cpu)); 1975 1976 crit_enter(); 1977 1978 if (DUMMYNET_LOADED) { 1979 kprintf("DUMMYNET already loaded\n"); 1980 error = EEXIST; 1981 goto back; 1982 } 1983 1984 kprintf("DUMMYNET initialized (011031)\n"); 1985 1986 for (i = 0; i < DN_NR_HASH_MAX; ++i) 1987 LIST_INIT(&pipe_table[i]); 1988 1989 for (i = 0; i < DN_NR_HASH_MAX; ++i) 1990 LIST_INIT(&flowset_table[i]); 1991 1992 ready_heap.size = ready_heap.elements = 0; 1993 ready_heap.offset = 0; 1994 1995 wfq_ready_heap.size = wfq_ready_heap.elements = 0; 1996 wfq_ready_heap.offset = 0; 1997 1998 extract_heap.size = extract_heap.elements = 0; 1999 extract_heap.offset = 0; 2000 2001 ip_dn_ctl_ptr = dummynet_ctl; 2002 ip_dn_io_ptr = dummynet_io; 2003 2004 netmsg_init(&dn_netmsg, NULL, &netisr_adone_rport, 2005 0, dummynet); 2006 2007 KKASSERT(dn_running == 0); 2008 #if 0 2009 /* REMOVED, initialized on first insertion */ 2010 systimer_init_periodic_nq(&dn_clock, dummynet_clock, NULL, dn_hz); 2011 #endif 2012 2013 back: 2014 crit_exit(); 2015 lwkt_replymsg(&msg->lmsg, error); 2016 } 2017 2018 static int 2019 ip_dn_init(void) 2020 { 2021 struct netmsg_base smsg; 2022 2023 if (ip_dn_cpu >= ncpus) { 2024 kprintf("%s: CPU%d does not exist, switch to CPU0\n", 2025 __func__, ip_dn_cpu); 2026 ip_dn_cpu = 0; 2027 } 2028 2029 netmsg_init(&smsg, NULL, &curthread->td_msgport, 2030 0, ip_dn_init_dispatch); 2031 lwkt_domsg(netisr_cpuport(ip_dn_cpu), &smsg.lmsg, 0); 2032 return smsg.lmsg.ms_error; 2033 } 2034 2035 #ifdef KLD_MODULE 2036 2037 static void 2038 ip_dn_stop_dispatch(netmsg_t msg) 2039 { 2040 crit_enter(); 2041 2042 dummynet_flush(); 2043 2044 ip_dn_ctl_ptr = NULL; 2045 ip_dn_io_ptr = NULL; 2046 KKASSERT(dn_running == 0); 2047 2048 crit_exit(); 2049 lwkt_replymsg(&msg->lmsg, 0); 2050 } 2051 2052 2053 static void 2054 ip_dn_stop(void) 2055 { 2056 struct netmsg_base smsg; 2057 2058 netmsg_init(&smsg, NULL, &curthread->td_msgport, 2059 0, ip_dn_stop_dispatch); 2060 lwkt_domsg(netisr_cpuport(ip_dn_cpu), &smsg.lmsg, 0); 2061 2062 netmsg_service_sync(); 2063 } 2064 2065 #endif /* KLD_MODULE */ 2066 2067 static int 2068 dummynet_modevent(module_t mod, int type, void *data) 2069 { 2070 switch (type) { 2071 case MOD_LOAD: 2072 return ip_dn_init(); 2073 2074 case MOD_UNLOAD: 2075 #ifndef KLD_MODULE 2076 kprintf("dummynet statically compiled, cannot unload\n"); 2077 return EINVAL; 2078 #else 2079 ip_dn_stop(); 2080 #endif 2081 break; 2082 2083 default: 2084 break; 2085 } 2086 return 0; 2087 } 2088 2089 static moduledata_t dummynet_mod = { 2090 "dummynet", 2091 dummynet_modevent, 2092 NULL 2093 }; 2094 DECLARE_MODULE(dummynet, dummynet_mod, SI_SUB_PROTO_END, SI_ORDER_ANY); 2095 MODULE_VERSION(dummynet, 1); 2096