dummynet.c revision 193500
1/* 2 * Copyright (c) 2002-2003 Luigi Rizzo 3 * Copyright (c) 1996 Alex Nash, Paul Traina, Poul-Henning Kamp 4 * Copyright (c) 1994 Ugen J.S.Antsilevich 5 * 6 * Idea and grammar partially left from: 7 * Copyright (c) 1993 Daniel Boulet 8 * 9 * Redistribution and use in source forms, with and without modification, 10 * are permitted provided that this entire comment appears intact. 11 * 12 * Redistribution in binary form may occur without any restrictions. 13 * Obviously, it would be nice if you gave credit where credit is due 14 * but requiring it would be too onerous. 15 * 16 * This software is provided ``AS IS'' without any warranties of any kind. 17 * 18 * NEW command line interface for IP firewall facility 19 * 20 * $FreeBSD: head/sbin/ipfw/dummynet.c 193500 2009-06-05 13:11:34Z luigi $ 21 * 22 * dummynet support 23 */ 24 25#include <sys/types.h> 26#include <sys/socket.h> 27#include <sys/queue.h> 28/* XXX there are several sysctl leftover here */ 29#include <sys/sysctl.h> 30 31#include "ipfw2.h" 32 33#include <ctype.h> 34#include <err.h> 35#include <netdb.h> 36#include <stdio.h> 37#include <stdlib.h> 38#include <string.h> 39#include <sysexits.h> 40 41#include <net/if.h> 42#include <netinet/in.h> 43#include <netinet/ip_fw.h> 44#include <netinet/ip_dummynet.h> 45#include <arpa/inet.h> /* inet_ntoa */ 46 47static struct _s_x dummynet_params[] = { 48 { "plr", TOK_PLR }, 49 { "noerror", TOK_NOERROR }, 50 { "buckets", TOK_BUCKETS }, 51 { "dst-ip", TOK_DSTIP }, 52 { "src-ip", TOK_SRCIP }, 53 { "dst-port", TOK_DSTPORT }, 54 { "src-port", TOK_SRCPORT }, 55 { "proto", TOK_PROTO }, 56 { "weight", TOK_WEIGHT }, 57 { "all", TOK_ALL }, 58 { "mask", TOK_MASK }, 59 { "droptail", TOK_DROPTAIL }, 60 { "red", TOK_RED }, 61 { "gred", TOK_GRED }, 62 { "bw", TOK_BW }, 63 { "bandwidth", TOK_BW }, 64 { "delay", TOK_DELAY }, 65 { "pipe", TOK_PIPE }, 66 { "queue", TOK_QUEUE }, 67 { "flow-id", TOK_FLOWID}, 68 { "dst-ipv6", TOK_DSTIP6}, 69 { "dst-ip6", TOK_DSTIP6}, 70 { "src-ipv6", TOK_SRCIP6}, 71 { "src-ip6", TOK_SRCIP6}, 72 { "profile", TOK_PIPE_PROFILE}, 73 { "dummynet-params", TOK_NULL }, 74 { NULL, 0 } /* terminator */ 75}; 76 77static int 78sort_q(const void *pa, const void *pb) 79{ 80 int rev = (co.do_sort < 0); 81 int field = rev ? -co.do_sort : co.do_sort; 82 long long res = 0; 83 const struct dn_flow_queue *a = pa; 84 const struct dn_flow_queue *b = pb; 85 86 switch (field) { 87 case 1: /* pkts */ 88 res = a->len - b->len; 89 break; 90 case 2: /* bytes */ 91 res = a->len_bytes - b->len_bytes; 92 break; 93 94 case 3: /* tot pkts */ 95 res = a->tot_pkts - b->tot_pkts; 96 break; 97 98 case 4: /* tot bytes */ 99 res = a->tot_bytes - b->tot_bytes; 100 break; 101 } 102 if (res < 0) 103 res = -1; 104 if (res > 0) 105 res = 1; 106 return (int)(rev ? res : -res); 107} 108 109static void 110list_queues(struct dn_flow_set *fs, struct dn_flow_queue *q) 111{ 112 int l; 113 int index_printed, indexes = 0; 114 char buff[255]; 115 struct protoent *pe; 116 117 if (fs->rq_elements == 0) 118 return; 119 120 if (co.do_sort != 0) 121 heapsort(q, fs->rq_elements, sizeof *q, sort_q); 122 123 /* Print IPv4 flows */ 124 index_printed = 0; 125 for (l = 0; l < fs->rq_elements; l++) { 126 struct in_addr ina; 127 128 /* XXX: Should check for IPv4 flows */ 129 if (IS_IP6_FLOW_ID(&(q[l].id))) 130 continue; 131 132 if (!index_printed) { 133 index_printed = 1; 134 if (indexes > 0) /* currently a no-op */ 135 printf("\n"); 136 indexes++; 137 printf(" " 138 "mask: 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n", 139 fs->flow_mask.proto, 140 fs->flow_mask.src_ip, fs->flow_mask.src_port, 141 fs->flow_mask.dst_ip, fs->flow_mask.dst_port); 142 143 printf("BKT Prot ___Source IP/port____ " 144 "____Dest. IP/port____ " 145 "Tot_pkt/bytes Pkt/Byte Drp\n"); 146 } 147 148 printf("%3d ", q[l].hash_slot); 149 pe = getprotobynumber(q[l].id.proto); 150 if (pe) 151 printf("%-4s ", pe->p_name); 152 else 153 printf("%4u ", q[l].id.proto); 154 ina.s_addr = htonl(q[l].id.src_ip); 155 printf("%15s/%-5d ", 156 inet_ntoa(ina), q[l].id.src_port); 157 ina.s_addr = htonl(q[l].id.dst_ip); 158 printf("%15s/%-5d ", 159 inet_ntoa(ina), q[l].id.dst_port); 160 printf("%4llu %8llu %2u %4u %3u\n", 161 align_uint64(&q[l].tot_pkts), 162 align_uint64(&q[l].tot_bytes), 163 q[l].len, q[l].len_bytes, q[l].drops); 164 if (co.verbose) 165 printf(" S %20llu F %20llu\n", 166 align_uint64(&q[l].S), align_uint64(&q[l].F)); 167 } 168 169 /* Print IPv6 flows */ 170 index_printed = 0; 171 for (l = 0; l < fs->rq_elements; l++) { 172 if (!IS_IP6_FLOW_ID(&(q[l].id))) 173 continue; 174 175 if (!index_printed) { 176 index_printed = 1; 177 if (indexes > 0) 178 printf("\n"); 179 indexes++; 180 printf("\n mask: proto: 0x%02x, flow_id: 0x%08x, ", 181 fs->flow_mask.proto, fs->flow_mask.flow_id6); 182 inet_ntop(AF_INET6, &(fs->flow_mask.src_ip6), 183 buff, sizeof(buff)); 184 printf("%s/0x%04x -> ", buff, fs->flow_mask.src_port); 185 inet_ntop( AF_INET6, &(fs->flow_mask.dst_ip6), 186 buff, sizeof(buff) ); 187 printf("%s/0x%04x\n", buff, fs->flow_mask.dst_port); 188 189 printf("BKT ___Prot___ _flow-id_ " 190 "______________Source IPv6/port_______________ " 191 "_______________Dest. IPv6/port_______________ " 192 "Tot_pkt/bytes Pkt/Byte Drp\n"); 193 } 194 printf("%3d ", q[l].hash_slot); 195 pe = getprotobynumber(q[l].id.proto); 196 if (pe != NULL) 197 printf("%9s ", pe->p_name); 198 else 199 printf("%9u ", q[l].id.proto); 200 printf("%7d %39s/%-5d ", q[l].id.flow_id6, 201 inet_ntop(AF_INET6, &(q[l].id.src_ip6), buff, sizeof(buff)), 202 q[l].id.src_port); 203 printf(" %39s/%-5d ", 204 inet_ntop(AF_INET6, &(q[l].id.dst_ip6), buff, sizeof(buff)), 205 q[l].id.dst_port); 206 printf(" %4llu %8llu %2u %4u %3u\n", 207 align_uint64(&q[l].tot_pkts), 208 align_uint64(&q[l].tot_bytes), 209 q[l].len, q[l].len_bytes, q[l].drops); 210 if (co.verbose) 211 printf(" S %20llu F %20llu\n", 212 align_uint64(&q[l].S), 213 align_uint64(&q[l].F)); 214 } 215} 216 217static void 218print_flowset_parms(struct dn_flow_set *fs, char *prefix) 219{ 220 int l; 221 char qs[30]; 222 char plr[30]; 223 char red[90]; /* Display RED parameters */ 224 225 l = fs->qsize; 226 if (fs->flags_fs & DN_QSIZE_IS_BYTES) { 227 if (l >= 8192) 228 sprintf(qs, "%d KB", l / 1024); 229 else 230 sprintf(qs, "%d B", l); 231 } else 232 sprintf(qs, "%3d sl.", l); 233 if (fs->plr) 234 sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff)); 235 else 236 plr[0] = '\0'; 237 if (fs->flags_fs & DN_IS_RED) /* RED parameters */ 238 sprintf(red, 239 "\n\t %cRED w_q %f min_th %d max_th %d max_p %f", 240 (fs->flags_fs & DN_IS_GENTLE_RED) ? 'G' : ' ', 241 1.0 * fs->w_q / (double)(1 << SCALE_RED), 242 SCALE_VAL(fs->min_th), 243 SCALE_VAL(fs->max_th), 244 1.0 * fs->max_p / (double)(1 << SCALE_RED)); 245 else 246 sprintf(red, "droptail"); 247 248 printf("%s %s%s %d queues (%d buckets) %s\n", 249 prefix, qs, plr, fs->rq_elements, fs->rq_size, red); 250} 251 252static void 253print_extra_delay_parms(struct dn_pipe *p, char *prefix) 254{ 255 double loss; 256 if (p->samples_no <= 0) 257 return; 258 259 loss = p->loss_level; 260 loss /= p->samples_no; 261 printf("%s profile: name \"%s\" loss %f samples %d\n", 262 prefix, p->name, loss, p->samples_no); 263} 264 265void 266ipfw_list_pipes(void *data, uint nbytes, int ac, char *av[]) 267{ 268 int rulenum; 269 void *next = data; 270 struct dn_pipe *p = (struct dn_pipe *) data; 271 struct dn_flow_set *fs; 272 struct dn_flow_queue *q; 273 int l; 274 275 if (ac > 0) 276 rulenum = strtoul(*av++, NULL, 10); 277 else 278 rulenum = 0; 279 for (; nbytes >= sizeof *p; p = (struct dn_pipe *)next) { 280 double b = p->bandwidth; 281 char buf[30]; 282 char prefix[80]; 283 284 if (SLIST_NEXT(p, next) != (struct dn_pipe *)DN_IS_PIPE) 285 break; /* done with pipes, now queues */ 286 287 /* 288 * compute length, as pipe have variable size 289 */ 290 l = sizeof(*p) + p->fs.rq_elements * sizeof(*q); 291 next = (char *)p + l; 292 nbytes -= l; 293 294 if ((rulenum != 0 && rulenum != p->pipe_nr) || co.do_pipe == 2) 295 continue; 296 297 /* 298 * Print rate (or clocking interface) 299 */ 300 if (p->if_name[0] != '\0') 301 sprintf(buf, "%s", p->if_name); 302 else if (b == 0) 303 sprintf(buf, "unlimited"); 304 else if (b >= 1000000) 305 sprintf(buf, "%7.3f Mbit/s", b/1000000); 306 else if (b >= 1000) 307 sprintf(buf, "%7.3f Kbit/s", b/1000); 308 else 309 sprintf(buf, "%7.3f bit/s ", b); 310 311 sprintf(prefix, "%05d: %s %4d ms ", 312 p->pipe_nr, buf, p->delay); 313 314 print_extra_delay_parms(p, prefix); 315 316 print_flowset_parms(&(p->fs), prefix); 317 318 q = (struct dn_flow_queue *)(p+1); 319 list_queues(&(p->fs), q); 320 } 321 for (fs = next; nbytes >= sizeof *fs; fs = next) { 322 char prefix[80]; 323 324 if (SLIST_NEXT(fs, next) != (struct dn_flow_set *)DN_IS_QUEUE) 325 break; 326 l = sizeof(*fs) + fs->rq_elements * sizeof(*q); 327 next = (char *)fs + l; 328 nbytes -= l; 329 330 if (rulenum != 0 && ((rulenum != fs->fs_nr && co.do_pipe == 2) || 331 (rulenum != fs->parent_nr && co.do_pipe == 1))) { 332 continue; 333 } 334 335 q = (struct dn_flow_queue *)(fs+1); 336 sprintf(prefix, "q%05d: weight %d pipe %d ", 337 fs->fs_nr, fs->weight, fs->parent_nr); 338 print_flowset_parms(fs, prefix); 339 list_queues(fs, q); 340 } 341} 342 343/* 344 * Delete pipe or queue i 345 */ 346int 347ipfw_delete_pipe(int pipe_or_queue, int i) 348{ 349 struct dn_pipe p; 350 351 memset(&p, 0, sizeof p); 352 if (pipe_or_queue == 1) 353 p.pipe_nr = i; /* pipe */ 354 else 355 p.fs.fs_nr = i; /* queue */ 356 i = do_cmd(IP_DUMMYNET_DEL, &p, sizeof p); 357 if (i) { 358 i = 1; 359 warn("rule %u: setsockopt(IP_DUMMYNET_DEL)", i); 360 } 361 return i; 362} 363 364/* 365 * Code to parse delay profiles. 366 * 367 * Some link types introduce extra delays in the transmission 368 * of a packet, e.g. because of MAC level framing, contention on 369 * the use of the channel, MAC level retransmissions and so on. 370 * From our point of view, the channel is effectively unavailable 371 * for this extra time, which is constant or variable depending 372 * on the link type. Additionally, packets may be dropped after this 373 * time (e.g. on a wireless link after too many retransmissions). 374 * We can model the additional delay with an empirical curve 375 * that represents its distribution. 376 * 377 * cumulative probability 378 * 1.0 ^ 379 * | 380 * L +-- loss-level x 381 * | ****** 382 * | * 383 * | ***** 384 * | * 385 * | ** 386 * | * 387 * +-------*-------------------> 388 * delay 389 * 390 * The empirical curve may have both vertical and horizontal lines. 391 * Vertical lines represent constant delay for a range of 392 * probabilities; horizontal lines correspond to a discontinuty 393 * in the delay distribution: the pipe will use the largest delay 394 * for a given probability. 395 * 396 * To pass the curve to dummynet, we must store the parameters 397 * in a file as described below, and issue the command 398 * 399 * ipfw pipe <n> config ... bw XXX profile <filename> ... 400 * 401 * The file format is the following, with whitespace acting as 402 * a separator and '#' indicating the beginning a comment: 403 * 404 * samples N 405 * the number of samples used in the internal 406 * representation (2..1024; default 100); 407 * 408 * loss-level L 409 * The probability above which packets are lost. 410 * (0.0 <= L <= 1.0, default 1.0 i.e. no loss); 411 * 412 * name identifier 413 * Optional a name (listed by "ipfw pipe show") 414 * to identify the distribution; 415 * 416 * "delay prob" | "prob delay" 417 * One of these two lines is mandatory and defines 418 * the format of the following lines with data points. 419 * 420 * XXX YYY 421 * 2 or more lines representing points in the curve, 422 * with either delay or probability first, according 423 * to the chosen format. 424 * The unit for delay is milliseconds. 425 * 426 * Data points does not need to be ordered or equal to the number 427 * specified in the "samples" line. ipfw will sort and interpolate 428 * the curve as needed. 429 * 430 * Example of a profile file: 431 432 name bla_bla_bla 433 samples 100 434 loss-level 0.86 435 prob delay 436 0 200 # minimum overhead is 200ms 437 0.5 200 438 0.5 300 439 0.8 1000 440 0.9 1300 441 1 1300 442 443 * Internally, we will convert the curve to a fixed number of 444 * samples, and when it is time to transmit a packet we will 445 * model the extra delay as extra bits in the packet. 446 * 447 */ 448 449#define ED_MAX_LINE_LEN 256+ED_MAX_NAME_LEN 450#define ED_TOK_SAMPLES "samples" 451#define ED_TOK_LOSS "loss-level" 452#define ED_TOK_NAME "name" 453#define ED_TOK_DELAY "delay" 454#define ED_TOK_PROB "prob" 455#define ED_SEPARATORS " \t\n" 456#define ED_MIN_SAMPLES_NO 2 457 458/* 459 * returns 1 if s is a non-negative number, with at least one '.' 460 */ 461static int 462is_valid_number(const char *s) 463{ 464 int i, dots_found = 0; 465 int len = strlen(s); 466 467 for (i = 0; i<len; ++i) 468 if (!isdigit(s[i]) && (s[i] !='.' || ++dots_found > 1)) 469 return 0; 470 return 1; 471} 472 473struct point { 474 double prob; 475 double delay; 476}; 477 478int 479compare_points(const void *vp1, const void *vp2) 480{ 481 const struct point *p1 = vp1; 482 const struct point *p2 = vp2; 483 double res = 0; 484 485 res = p1->prob - p2->prob; 486 if (res == 0) 487 res = p1->delay - p2->delay; 488 if (res < 0) 489 return -1; 490 else if (res > 0) 491 return 1; 492 else 493 return 0; 494} 495 496#define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno 497 498static void 499load_extra_delays(const char *filename, struct dn_pipe *p) 500{ 501 char line[ED_MAX_LINE_LEN]; 502 FILE *f; 503 int lineno = 0; 504 int i; 505 506 int samples = -1; 507 double loss = -1.0; 508 char profile_name[ED_MAX_NAME_LEN]; 509 int delay_first = -1; 510 int do_points = 0; 511 struct point points[ED_MAX_SAMPLES_NO]; 512 int points_no = 0; 513 514 profile_name[0] = '\0'; 515 f = fopen(filename, "r"); 516 if (f == NULL) 517 err(EX_UNAVAILABLE, "fopen: %s", filename); 518 519 while (fgets(line, ED_MAX_LINE_LEN, f)) { /* read commands */ 520 char *s, *cur = line, *name = NULL, *arg = NULL; 521 522 ++lineno; 523 524 /* parse the line */ 525 while (cur) { 526 s = strsep(&cur, ED_SEPARATORS); 527 if (s == NULL || *s == '#') 528 break; 529 if (*s == '\0') 530 continue; 531 if (arg) 532 errx(ED_EFMT("too many arguments")); 533 if (name == NULL) 534 name = s; 535 else 536 arg = s; 537 } 538 if (name == NULL) /* empty line */ 539 continue; 540 if (arg == NULL) 541 errx(ED_EFMT("missing arg for %s"), name); 542 543 if (!strcasecmp(name, ED_TOK_SAMPLES)) { 544 if (samples > 0) 545 errx(ED_EFMT("duplicate ``samples'' line")); 546 if (atoi(arg) <=0) 547 errx(ED_EFMT("invalid number of samples")); 548 samples = atoi(arg); 549 if (samples>ED_MAX_SAMPLES_NO) 550 errx(ED_EFMT("too many samples, maximum is %d"), 551 ED_MAX_SAMPLES_NO); 552 do_points = 0; 553 } else if (!strcasecmp(name, ED_TOK_LOSS)) { 554 if (loss != -1.0) 555 errx(ED_EFMT("duplicated token: %s"), name); 556 if (!is_valid_number(arg)) 557 errx(ED_EFMT("invalid %s"), arg); 558 loss = atof(arg); 559 if (loss > 1) 560 errx(ED_EFMT("%s greater than 1.0"), name); 561 do_points = 0; 562 } else if (!strcasecmp(name, ED_TOK_NAME)) { 563 if (profile_name[0] != '\0') 564 errx(ED_EFMT("duplicated token: %s"), name); 565 strncpy(profile_name, arg, sizeof(profile_name) - 1); 566 profile_name[sizeof(profile_name)-1] = '\0'; 567 do_points = 0; 568 } else if (!strcasecmp(name, ED_TOK_DELAY)) { 569 if (do_points) 570 errx(ED_EFMT("duplicated token: %s"), name); 571 delay_first = 1; 572 do_points = 1; 573 } else if (!strcasecmp(name, ED_TOK_PROB)) { 574 if (do_points) 575 errx(ED_EFMT("duplicated token: %s"), name); 576 delay_first = 0; 577 do_points = 1; 578 } else if (do_points) { 579 if (!is_valid_number(name) || !is_valid_number(arg)) 580 errx(ED_EFMT("invalid point found")); 581 if (delay_first) { 582 points[points_no].delay = atof(name); 583 points[points_no].prob = atof(arg); 584 } else { 585 points[points_no].delay = atof(arg); 586 points[points_no].prob = atof(name); 587 } 588 if (points[points_no].prob > 1.0) 589 errx(ED_EFMT("probability greater than 1.0")); 590 ++points_no; 591 } else { 592 errx(ED_EFMT("unrecognised command '%s'"), name); 593 } 594 } 595 596 if (samples == -1) { 597 warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES); 598 samples = 100; 599 } 600 601 if (loss == -1.0) { 602 warnx("'%s' not found, assuming no loss", ED_TOK_LOSS); 603 loss = 1; 604 } 605 606 /* make sure that there are enough points. */ 607 if (points_no < ED_MIN_SAMPLES_NO) 608 errx(ED_EFMT("too few samples, need at least %d"), 609 ED_MIN_SAMPLES_NO); 610 611 qsort(points, points_no, sizeof(struct point), compare_points); 612 613 /* interpolation */ 614 for (i = 0; i<points_no-1; ++i) { 615 double y1 = points[i].prob * samples; 616 double x1 = points[i].delay; 617 double y2 = points[i+1].prob * samples; 618 double x2 = points[i+1].delay; 619 620 int index = y1; 621 int stop = y2; 622 623 if (x1 == x2) { 624 for (; index<stop; ++index) 625 p->samples[index] = x1; 626 } else { 627 double m = (y2-y1)/(x2-x1); 628 double c = y1 - m*x1; 629 for (; index<stop ; ++index) 630 p->samples[index] = (index - c)/m; 631 } 632 } 633 p->samples_no = samples; 634 p->loss_level = loss * samples; 635 strncpy(p->name, profile_name, sizeof(p->name)); 636} 637 638void 639ipfw_config_pipe(int ac, char **av) 640{ 641 int samples[ED_MAX_SAMPLES_NO]; 642 struct dn_pipe p; 643 int i; 644 char *end; 645 void *par = NULL; 646 647 memset(&p, 0, sizeof p); 648 649 av++; ac--; 650 /* Pipe number */ 651 if (ac && isdigit(**av)) { 652 i = atoi(*av); av++; ac--; 653 if (co.do_pipe == 1) 654 p.pipe_nr = i; 655 else 656 p.fs.fs_nr = i; 657 } 658 while (ac > 0) { 659 double d; 660 int tok = match_token(dummynet_params, *av); 661 ac--; av++; 662 663 switch(tok) { 664 case TOK_NOERROR: 665 p.fs.flags_fs |= DN_NOERROR; 666 break; 667 668 case TOK_PLR: 669 NEED1("plr needs argument 0..1\n"); 670 d = strtod(av[0], NULL); 671 if (d > 1) 672 d = 1; 673 else if (d < 0) 674 d = 0; 675 p.fs.plr = (int)(d*0x7fffffff); 676 ac--; av++; 677 break; 678 679 case TOK_QUEUE: 680 NEED1("queue needs queue size\n"); 681 end = NULL; 682 p.fs.qsize = strtoul(av[0], &end, 0); 683 if (*end == 'K' || *end == 'k') { 684 p.fs.flags_fs |= DN_QSIZE_IS_BYTES; 685 p.fs.qsize *= 1024; 686 } else if (*end == 'B' || 687 _substrcmp2(end, "by", "bytes") == 0) { 688 p.fs.flags_fs |= DN_QSIZE_IS_BYTES; 689 } 690 ac--; av++; 691 break; 692 693 case TOK_BUCKETS: 694 NEED1("buckets needs argument\n"); 695 p.fs.rq_size = strtoul(av[0], NULL, 0); 696 ac--; av++; 697 break; 698 699 case TOK_MASK: 700 NEED1("mask needs mask specifier\n"); 701 /* 702 * per-flow queue, mask is dst_ip, dst_port, 703 * src_ip, src_port, proto measured in bits 704 */ 705 par = NULL; 706 707 bzero(&p.fs.flow_mask, sizeof(p.fs.flow_mask)); 708 end = NULL; 709 710 while (ac >= 1) { 711 uint32_t *p32 = NULL; 712 uint16_t *p16 = NULL; 713 uint32_t *p20 = NULL; 714 struct in6_addr *pa6 = NULL; 715 uint32_t a; 716 717 tok = match_token(dummynet_params, *av); 718 ac--; av++; 719 switch(tok) { 720 case TOK_ALL: 721 /* 722 * special case, all bits significant 723 */ 724 p.fs.flow_mask.dst_ip = ~0; 725 p.fs.flow_mask.src_ip = ~0; 726 p.fs.flow_mask.dst_port = ~0; 727 p.fs.flow_mask.src_port = ~0; 728 p.fs.flow_mask.proto = ~0; 729 n2mask(&(p.fs.flow_mask.dst_ip6), 128); 730 n2mask(&(p.fs.flow_mask.src_ip6), 128); 731 p.fs.flow_mask.flow_id6 = ~0; 732 p.fs.flags_fs |= DN_HAVE_FLOW_MASK; 733 goto end_mask; 734 735 case TOK_DSTIP: 736 p32 = &p.fs.flow_mask.dst_ip; 737 break; 738 739 case TOK_SRCIP: 740 p32 = &p.fs.flow_mask.src_ip; 741 break; 742 743 case TOK_DSTIP6: 744 pa6 = &(p.fs.flow_mask.dst_ip6); 745 break; 746 747 case TOK_SRCIP6: 748 pa6 = &(p.fs.flow_mask.src_ip6); 749 break; 750 751 case TOK_FLOWID: 752 p20 = &p.fs.flow_mask.flow_id6; 753 break; 754 755 case TOK_DSTPORT: 756 p16 = &p.fs.flow_mask.dst_port; 757 break; 758 759 case TOK_SRCPORT: 760 p16 = &p.fs.flow_mask.src_port; 761 break; 762 763 case TOK_PROTO: 764 break; 765 766 default: 767 ac++; av--; /* backtrack */ 768 goto end_mask; 769 } 770 if (ac < 1) 771 errx(EX_USAGE, "mask: value missing"); 772 if (*av[0] == '/') { 773 a = strtoul(av[0]+1, &end, 0); 774 if (pa6 == NULL) 775 a = (a == 32) ? ~0 : (1 << a) - 1; 776 } else 777 a = strtoul(av[0], &end, 0); 778 if (p32 != NULL) 779 *p32 = a; 780 else if (p16 != NULL) { 781 if (a > 0xFFFF) 782 errx(EX_DATAERR, 783 "port mask must be 16 bit"); 784 *p16 = (uint16_t)a; 785 } else if (p20 != NULL) { 786 if (a > 0xfffff) 787 errx(EX_DATAERR, 788 "flow_id mask must be 20 bit"); 789 *p20 = (uint32_t)a; 790 } else if (pa6 != NULL) { 791 if (a > 128) 792 errx(EX_DATAERR, 793 "in6addr invalid mask len"); 794 else 795 n2mask(pa6, a); 796 } else { 797 if (a > 0xFF) 798 errx(EX_DATAERR, 799 "proto mask must be 8 bit"); 800 p.fs.flow_mask.proto = (uint8_t)a; 801 } 802 if (a != 0) 803 p.fs.flags_fs |= DN_HAVE_FLOW_MASK; 804 ac--; av++; 805 } /* end while, config masks */ 806end_mask: 807 break; 808 809 case TOK_RED: 810 case TOK_GRED: 811 NEED1("red/gred needs w_q/min_th/max_th/max_p\n"); 812 p.fs.flags_fs |= DN_IS_RED; 813 if (tok == TOK_GRED) 814 p.fs.flags_fs |= DN_IS_GENTLE_RED; 815 /* 816 * the format for parameters is w_q/min_th/max_th/max_p 817 */ 818 if ((end = strsep(&av[0], "/"))) { 819 double w_q = strtod(end, NULL); 820 if (w_q > 1 || w_q <= 0) 821 errx(EX_DATAERR, "0 < w_q <= 1"); 822 p.fs.w_q = (int) (w_q * (1 << SCALE_RED)); 823 } 824 if ((end = strsep(&av[0], "/"))) { 825 p.fs.min_th = strtoul(end, &end, 0); 826 if (*end == 'K' || *end == 'k') 827 p.fs.min_th *= 1024; 828 } 829 if ((end = strsep(&av[0], "/"))) { 830 p.fs.max_th = strtoul(end, &end, 0); 831 if (*end == 'K' || *end == 'k') 832 p.fs.max_th *= 1024; 833 } 834 if ((end = strsep(&av[0], "/"))) { 835 double max_p = strtod(end, NULL); 836 if (max_p > 1 || max_p <= 0) 837 errx(EX_DATAERR, "0 < max_p <= 1"); 838 p.fs.max_p = (int)(max_p * (1 << SCALE_RED)); 839 } 840 ac--; av++; 841 break; 842 843 case TOK_DROPTAIL: 844 p.fs.flags_fs &= ~(DN_IS_RED|DN_IS_GENTLE_RED); 845 break; 846 847 case TOK_BW: 848 NEED1("bw needs bandwidth or interface\n"); 849 if (co.do_pipe != 1) 850 errx(EX_DATAERR, "bandwidth only valid for pipes"); 851 /* 852 * set clocking interface or bandwidth value 853 */ 854 if (av[0][0] >= 'a' && av[0][0] <= 'z') { 855 int l = sizeof(p.if_name)-1; 856 /* interface name */ 857 strncpy(p.if_name, av[0], l); 858 p.if_name[l] = '\0'; 859 p.bandwidth = 0; 860 } else { 861 p.if_name[0] = '\0'; 862 p.bandwidth = strtoul(av[0], &end, 0); 863 if (*end == 'K' || *end == 'k') { 864 end++; 865 p.bandwidth *= 1000; 866 } else if (*end == 'M') { 867 end++; 868 p.bandwidth *= 1000000; 869 } 870 if ((*end == 'B' && 871 _substrcmp2(end, "Bi", "Bit/s") != 0) || 872 _substrcmp2(end, "by", "bytes") == 0) 873 p.bandwidth *= 8; 874 if (p.bandwidth < 0) 875 errx(EX_DATAERR, "bandwidth too large"); 876 } 877 ac--; av++; 878 break; 879 880 case TOK_DELAY: 881 if (co.do_pipe != 1) 882 errx(EX_DATAERR, "delay only valid for pipes"); 883 NEED1("delay needs argument 0..10000ms\n"); 884 p.delay = strtoul(av[0], NULL, 0); 885 ac--; av++; 886 break; 887 888 case TOK_WEIGHT: 889 if (co.do_pipe == 1) 890 errx(EX_DATAERR,"weight only valid for queues"); 891 NEED1("weight needs argument 0..100\n"); 892 p.fs.weight = strtoul(av[0], &end, 0); 893 ac--; av++; 894 break; 895 896 case TOK_PIPE: 897 if (co.do_pipe == 1) 898 errx(EX_DATAERR,"pipe only valid for queues"); 899 NEED1("pipe needs pipe_number\n"); 900 p.fs.parent_nr = strtoul(av[0], &end, 0); 901 ac--; av++; 902 break; 903 904 case TOK_PIPE_PROFILE: 905 if (co.do_pipe != 1) 906 errx(EX_DATAERR, "extra delay only valid for pipes"); 907 NEED1("extra delay needs the file name\n"); 908 p.samples = &samples[0]; 909 load_extra_delays(av[0], &p); 910 --ac; ++av; 911 break; 912 913 default: 914 errx(EX_DATAERR, "unrecognised option ``%s''", av[-1]); 915 } 916 } 917 if (co.do_pipe == 1) { 918 if (p.pipe_nr == 0) 919 errx(EX_DATAERR, "pipe_nr must be > 0"); 920 if (p.delay > 10000) 921 errx(EX_DATAERR, "delay must be < 10000"); 922 if (p.samples_no > 0 && p.bandwidth == 0) 923 errx(EX_DATAERR, 924 "profile requires a bandwidth limit"); 925 } else { /* co.do_pipe == 2, queue */ 926 if (p.fs.parent_nr == 0) 927 errx(EX_DATAERR, "pipe must be > 0"); 928 if (p.fs.weight >100) 929 errx(EX_DATAERR, "weight must be <= 100"); 930 } 931 if (p.fs.flags_fs & DN_QSIZE_IS_BYTES) { 932 size_t len; 933 long limit; 934 935 len = sizeof(limit); 936 if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit", 937 &limit, &len, NULL, 0) == -1) 938 limit = 1024*1024; 939 if (p.fs.qsize > limit) 940 errx(EX_DATAERR, "queue size must be < %ldB", limit); 941 } else { 942 size_t len; 943 long limit; 944 945 len = sizeof(limit); 946 if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit", 947 &limit, &len, NULL, 0) == -1) 948 limit = 100; 949 if (p.fs.qsize > limit) 950 errx(EX_DATAERR, "2 <= queue size <= %ld", limit); 951 } 952 if (p.fs.flags_fs & DN_IS_RED) { 953 size_t len; 954 int lookup_depth, avg_pkt_size; 955 double s, idle, weight, w_q; 956 struct clockinfo ck; 957 int t; 958 959 if (p.fs.min_th >= p.fs.max_th) 960 errx(EX_DATAERR, "min_th %d must be < than max_th %d", 961 p.fs.min_th, p.fs.max_th); 962 if (p.fs.max_th == 0) 963 errx(EX_DATAERR, "max_th must be > 0"); 964 965 len = sizeof(int); 966 if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth", 967 &lookup_depth, &len, NULL, 0) == -1) 968 errx(1, "sysctlbyname(\"%s\")", 969 "net.inet.ip.dummynet.red_lookup_depth"); 970 if (lookup_depth == 0) 971 errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth" 972 " must be greater than zero"); 973 974 len = sizeof(int); 975 if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size", 976 &avg_pkt_size, &len, NULL, 0) == -1) 977 978 errx(1, "sysctlbyname(\"%s\")", 979 "net.inet.ip.dummynet.red_avg_pkt_size"); 980 if (avg_pkt_size == 0) 981 errx(EX_DATAERR, 982 "net.inet.ip.dummynet.red_avg_pkt_size must" 983 " be greater than zero"); 984 985 len = sizeof(struct clockinfo); 986 if (sysctlbyname("kern.clockrate", &ck, &len, NULL, 0) == -1) 987 errx(1, "sysctlbyname(\"%s\")", "kern.clockrate"); 988 989 /* 990 * Ticks needed for sending a medium-sized packet. 991 * Unfortunately, when we are configuring a WF2Q+ queue, we 992 * do not have bandwidth information, because that is stored 993 * in the parent pipe, and also we have multiple queues 994 * competing for it. So we set s=0, which is not very 995 * correct. But on the other hand, why do we want RED with 996 * WF2Q+ ? 997 */ 998 if (p.bandwidth==0) /* this is a WF2Q+ queue */ 999 s = 0; 1000 else 1001 s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth; 1002 1003 /* 1004 * max idle time (in ticks) before avg queue size becomes 0. 1005 * NOTA: (3/w_q) is approx the value x so that 1006 * (1-w_q)^x < 10^-3. 1007 */ 1008 w_q = ((double)p.fs.w_q) / (1 << SCALE_RED); 1009 idle = s * 3. / w_q; 1010 p.fs.lookup_step = (int)idle / lookup_depth; 1011 if (!p.fs.lookup_step) 1012 p.fs.lookup_step = 1; 1013 weight = 1 - w_q; 1014 for (t = p.fs.lookup_step; t > 1; --t) 1015 weight *= 1 - w_q; 1016 p.fs.lookup_weight = (int)(weight * (1 << SCALE_RED)); 1017 } 1018 if (p.samples_no <= 0) { 1019 i = do_cmd(IP_DUMMYNET_CONFIGURE, &p, sizeof p); 1020 } else { 1021 struct dn_pipe_max pm; 1022 int len = sizeof(pm); 1023 1024 memcpy(&pm.pipe, &p, sizeof(pm.pipe)); 1025 memcpy(&pm.samples, samples, sizeof(pm.samples)); 1026 1027 i = do_cmd(IP_DUMMYNET_CONFIGURE, &pm, len); 1028 } 1029 1030 if (i) 1031 err(1, "setsockopt(%s)", "IP_DUMMYNET_CONFIGURE"); 1032} 1033