1 /* 2 * Copyright (c) 2002-2003,2010 Luigi Rizzo 3 * 4 * Redistribution and use in source forms, with and without modification, 5 * are permitted provided that this entire comment appears intact. 6 * 7 * Redistribution in binary form may occur without any restrictions. 8 * Obviously, it would be nice if you gave credit where credit is due 9 * but requiring it would be too onerous. 10 * 11 * This software is provided ``AS IS'' without any warranties of any kind. 12 * 13 * $FreeBSD$ 14 * 15 * dummynet support 16 */ 17 18 #include <sys/types.h> 19 #include <sys/socket.h> 20 /* XXX there are several sysctl leftover here */ 21 #include <sys/sysctl.h> 22 23 #include "ipfw2.h" 24 25 #include <ctype.h> 26 #include <err.h> 27 #include <errno.h> 28 #include <libutil.h> 29 #include <netdb.h> 30 #include <stdio.h> 31 #include <stdlib.h> 32 #include <string.h> 33 #include <sysexits.h> 34 35 #include <net/if.h> 36 #include <netinet/in.h> 37 #include <netinet/ip_fw.h> 38 #include <netinet/ip_dummynet.h> 39 #include <arpa/inet.h> /* inet_ntoa */ 40 41 42 static struct _s_x dummynet_params[] = { 43 { "plr", TOK_PLR }, 44 { "noerror", TOK_NOERROR }, 45 { "buckets", TOK_BUCKETS }, 46 { "dst-ip", TOK_DSTIP }, 47 { "src-ip", TOK_SRCIP }, 48 { "dst-port", TOK_DSTPORT }, 49 { "src-port", TOK_SRCPORT }, 50 { "proto", TOK_PROTO }, 51 { "weight", TOK_WEIGHT }, 52 { "lmax", TOK_LMAX }, 53 { "maxlen", TOK_LMAX }, 54 { "all", TOK_ALL }, 55 { "mask", TOK_MASK }, /* alias for both */ 56 { "sched_mask", TOK_SCHED_MASK }, 57 { "flow_mask", TOK_FLOW_MASK }, 58 { "droptail", TOK_DROPTAIL }, 59 { "red", TOK_RED }, 60 { "gred", TOK_GRED }, 61 { "bw", TOK_BW }, 62 { "bandwidth", TOK_BW }, 63 { "delay", TOK_DELAY }, 64 { "link", TOK_LINK }, 65 { "pipe", TOK_PIPE }, 66 { "queue", TOK_QUEUE }, 67 { "flowset", TOK_FLOWSET }, 68 { "sched", TOK_SCHED }, 69 { "pri", TOK_PRI }, 70 { "priority", TOK_PRI }, 71 { "type", TOK_TYPE }, 72 { "flow-id", TOK_FLOWID}, 73 { "dst-ipv6", TOK_DSTIP6}, 74 { "dst-ip6", TOK_DSTIP6}, 75 { "src-ipv6", TOK_SRCIP6}, 76 { "src-ip6", TOK_SRCIP6}, 77 { "profile", TOK_PROFILE}, 78 { "burst", TOK_BURST}, 79 { "dummynet-params", TOK_NULL }, 80 { NULL, 0 } /* terminator */ 81 }; 82 83 #define O_NEXT(p, len) ((void *)((char *)p + len)) 84 85 static void 86 oid_fill(struct dn_id *oid, int len, int type, uintptr_t id) 87 { 88 oid->len = len; 89 oid->type = type; 90 oid->subtype = 0; 91 oid->id = id; 92 } 93 94 /* make room in the buffer and move the pointer forward */ 95 static void * 96 o_next(struct dn_id **o, int len, int type) 97 { 98 struct dn_id *ret = *o; 99 oid_fill(ret, len, type, 0); 100 *o = O_NEXT(*o, len); 101 return ret; 102 } 103 104 #if 0 105 static int 106 sort_q(void *arg, const void *pa, const void *pb) 107 { 108 int rev = (co.do_sort < 0); 109 int field = rev ? -co.do_sort : co.do_sort; 110 long long res = 0; 111 const struct dn_flow_queue *a = pa; 112 const struct dn_flow_queue *b = pb; 113 114 switch (field) { 115 case 1: /* pkts */ 116 res = a->len - b->len; 117 break; 118 case 2: /* bytes */ 119 res = a->len_bytes - b->len_bytes; 120 break; 121 122 case 3: /* tot pkts */ 123 res = a->tot_pkts - b->tot_pkts; 124 break; 125 126 case 4: /* tot bytes */ 127 res = a->tot_bytes - b->tot_bytes; 128 break; 129 } 130 if (res < 0) 131 res = -1; 132 if (res > 0) 133 res = 1; 134 return (int)(rev ? res : -res); 135 } 136 #endif 137 138 /* print a mask and header for the subsequent list of flows */ 139 static void 140 print_mask(struct ipfw_flow_id *id) 141 { 142 if (!IS_IP6_FLOW_ID(id)) { 143 printf(" " 144 "mask: %s 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n", 145 id->extra ? "queue," : "", 146 id->proto, 147 id->src_ip, id->src_port, 148 id->dst_ip, id->dst_port); 149 150 printf("BKT Prot ___Source IP/port____ " 151 "____Dest. IP/port____ " 152 "Tot_pkt/bytes Pkt/Byte Drp\n"); 153 } else { 154 char buf[255]; 155 printf("\n mask: %sproto: 0x%02x, flow_id: 0x%08x, ", 156 id->extra ? "queue," : "", 157 id->proto, id->flow_id6); 158 inet_ntop(AF_INET6, &(id->src_ip6), buf, sizeof(buf)); 159 printf("%s/0x%04x -> ", buf, id->src_port); 160 inet_ntop(AF_INET6, &(id->dst_ip6), buf, sizeof(buf)); 161 printf("%s/0x%04x\n", buf, id->dst_port); 162 163 printf("BKT ___Prot___ _flow-id_ " 164 "______________Source IPv6/port_______________ " 165 "_______________Dest. IPv6/port_______________ " 166 "Tot_pkt/bytes Pkt/Byte Drp\n"); 167 } 168 } 169 170 static void 171 list_flow(struct dn_flow *ni) 172 { 173 char buff[255]; 174 struct protoent *pe; 175 struct in_addr ina; 176 struct ipfw_flow_id *id = &ni->fid; 177 178 pe = getprotobynumber(id->proto); 179 /* XXX: Should check for IPv4 flows */ 180 printf("%3u%c", (ni->oid.id) & 0xff, 181 id->extra ? '*' : ' '); 182 if (!IS_IP6_FLOW_ID(id)) { 183 if (pe) 184 printf("%-4s ", pe->p_name); 185 else 186 printf("%4u ", id->proto); 187 ina.s_addr = htonl(id->src_ip); 188 printf("%15s/%-5d ", 189 inet_ntoa(ina), id->src_port); 190 ina.s_addr = htonl(id->dst_ip); 191 printf("%15s/%-5d ", 192 inet_ntoa(ina), id->dst_port); 193 } else { 194 /* Print IPv6 flows */ 195 if (pe != NULL) 196 printf("%9s ", pe->p_name); 197 else 198 printf("%9u ", id->proto); 199 printf("%7d %39s/%-5d ", id->flow_id6, 200 inet_ntop(AF_INET6, &(id->src_ip6), buff, sizeof(buff)), 201 id->src_port); 202 printf(" %39s/%-5d ", 203 inet_ntop(AF_INET6, &(id->dst_ip6), buff, sizeof(buff)), 204 id->dst_port); 205 } 206 pr_u64(&ni->tot_pkts, 4); 207 pr_u64(&ni->tot_bytes, 8); 208 printf("%2u %4u %3u\n", 209 ni->length, ni->len_bytes, ni->drops); 210 } 211 212 static void 213 print_flowset_parms(struct dn_fs *fs, char *prefix) 214 { 215 int l; 216 char qs[30]; 217 char plr[30]; 218 char red[90]; /* Display RED parameters */ 219 220 l = fs->qsize; 221 if (fs->flags & DN_QSIZE_BYTES) { 222 if (l >= 8192) 223 sprintf(qs, "%d KB", l / 1024); 224 else 225 sprintf(qs, "%d B", l); 226 } else 227 sprintf(qs, "%3d sl.", l); 228 if (fs->plr) 229 sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff)); 230 else 231 plr[0] = '\0'; 232 233 if (fs->flags & DN_IS_RED) /* RED parameters */ 234 sprintf(red, 235 "\n\t %cRED w_q %f min_th %d max_th %d max_p %f", 236 (fs->flags & DN_IS_GENTLE_RED) ? 'G' : ' ', 237 1.0 * fs->w_q / (double)(1 << SCALE_RED), 238 fs->min_th, 239 fs->max_th, 240 1.0 * fs->max_p / (double)(1 << SCALE_RED)); 241 else 242 sprintf(red, "droptail"); 243 244 if (prefix[0]) { 245 printf("%s %s%s %d queues (%d buckets) %s\n", 246 prefix, qs, plr, fs->oid.id, fs->buckets, red); 247 prefix[0] = '\0'; 248 } else { 249 printf("q%05d %s%s %d flows (%d buckets) sched %d " 250 "weight %d lmax %d pri %d %s\n", 251 fs->fs_nr, qs, plr, fs->oid.id, fs->buckets, 252 fs->sched_nr, fs->par[0], fs->par[1], fs->par[2], red); 253 if (fs->flags & DN_HAVE_MASK) 254 print_mask(&fs->flow_mask); 255 } 256 } 257 258 static void 259 print_extra_delay_parms(struct dn_profile *p) 260 { 261 double loss; 262 if (p->samples_no <= 0) 263 return; 264 265 loss = p->loss_level; 266 loss /= p->samples_no; 267 printf("\t profile: name \"%s\" loss %f samples %d\n", 268 p->name, loss, p->samples_no); 269 } 270 271 static void 272 flush_buf(char *buf) 273 { 274 if (buf[0]) 275 printf("%s\n", buf); 276 buf[0] = '\0'; 277 } 278 279 /* 280 * generic list routine. We expect objects in a specific order, i.e. 281 * PIPES AND SCHEDULERS: 282 * link; scheduler; internal flowset if any; instances 283 * we can tell a pipe from the number. 284 * 285 * FLOWSETS: 286 * flowset; queues; 287 * link i (int queue); scheduler i; si(i) { flowsets() : queues } 288 */ 289 static void 290 list_pipes(struct dn_id *oid, struct dn_id *end) 291 { 292 char buf[160]; /* pending buffer */ 293 294 buf[0] = '\0'; 295 for (; oid != end; oid = O_NEXT(oid, oid->len)) { 296 if (oid->len < sizeof(*oid)) 297 errx(1, "invalid oid len %d\n", oid->len); 298 299 switch (oid->type) { 300 default: 301 flush_buf(buf); 302 printf("unrecognized object %d size %d\n", oid->type, oid->len); 303 break; 304 case DN_TEXT: /* list of attached flowsets */ 305 { 306 int i, l; 307 struct { 308 struct dn_id id; 309 uint32_t p[0]; 310 } *d = (void *)oid; 311 l = (oid->len - sizeof(*oid))/sizeof(d->p[0]); 312 if (l == 0) 313 break; 314 printf(" Children flowsets: "); 315 for (i = 0; i < l; i++) 316 printf("%u ", d->p[i]); 317 printf("\n"); 318 break; 319 } 320 case DN_CMD_GET: 321 if (co.verbose) 322 printf("answer for cmd %d, len %d\n", oid->type, oid->id); 323 break; 324 case DN_SCH: { 325 struct dn_sch *s = (struct dn_sch *)oid; 326 flush_buf(buf); 327 printf(" sched %d type %s flags 0x%x %d buckets %d active\n", 328 s->sched_nr, 329 s->name, s->flags, s->buckets, s->oid.id); 330 if (s->flags & DN_HAVE_MASK) 331 print_mask(&s->sched_mask); 332 } 333 break; 334 335 case DN_FLOW: 336 list_flow((struct dn_flow *)oid); 337 break; 338 339 case DN_LINK: { 340 struct dn_link *p = (struct dn_link *)oid; 341 double b = p->bandwidth; 342 char bwbuf[30]; 343 char burst[5 + 7]; 344 345 /* This starts a new object so flush buffer */ 346 flush_buf(buf); 347 /* data rate */ 348 if (b == 0) 349 sprintf(bwbuf, "unlimited "); 350 else if (b >= 1000000) 351 sprintf(bwbuf, "%7.3f Mbit/s", b/1000000); 352 else if (b >= 1000) 353 sprintf(bwbuf, "%7.3f Kbit/s", b/1000); 354 else 355 sprintf(bwbuf, "%7.3f bit/s ", b); 356 357 if (humanize_number(burst, sizeof(burst), p->burst, 358 "", HN_AUTOSCALE, 0) < 0 || co.verbose) 359 sprintf(burst, "%d", (int)p->burst); 360 sprintf(buf, "%05d: %s %4d ms burst %s", 361 p->link_nr % DN_MAX_ID, bwbuf, p->delay, burst); 362 } 363 break; 364 365 case DN_FS: 366 print_flowset_parms((struct dn_fs *)oid, buf); 367 break; 368 case DN_PROFILE: 369 flush_buf(buf); 370 print_extra_delay_parms((struct dn_profile *)oid); 371 } 372 flush_buf(buf); // XXX does it really go here ? 373 } 374 } 375 376 /* 377 * Delete pipe, queue or scheduler i 378 */ 379 int 380 ipfw_delete_pipe(int do_pipe, int i) 381 { 382 struct { 383 struct dn_id oid; 384 uintptr_t a[1]; /* add more if we want a list */ 385 } cmd; 386 oid_fill((void *)&cmd, sizeof(cmd), DN_CMD_DELETE, DN_API_VERSION); 387 cmd.oid.subtype = (do_pipe == 1) ? DN_LINK : 388 ( (do_pipe == 2) ? DN_FS : DN_SCH); 389 cmd.a[0] = i; 390 i = do_cmd(IP_DUMMYNET3, &cmd, cmd.oid.len); 391 if (i) { 392 i = 1; 393 warn("rule %u: setsockopt(IP_DUMMYNET_DEL)", i); 394 } 395 return i; 396 } 397 398 /* 399 * Code to parse delay profiles. 400 * 401 * Some link types introduce extra delays in the transmission 402 * of a packet, e.g. because of MAC level framing, contention on 403 * the use of the channel, MAC level retransmissions and so on. 404 * From our point of view, the channel is effectively unavailable 405 * for this extra time, which is constant or variable depending 406 * on the link type. Additionally, packets may be dropped after this 407 * time (e.g. on a wireless link after too many retransmissions). 408 * We can model the additional delay with an empirical curve 409 * that represents its distribution. 410 * 411 * cumulative probability 412 * 1.0 ^ 413 * | 414 * L +-- loss-level x 415 * | ****** 416 * | * 417 * | ***** 418 * | * 419 * | ** 420 * | * 421 * +-------*-------------------> 422 * delay 423 * 424 * The empirical curve may have both vertical and horizontal lines. 425 * Vertical lines represent constant delay for a range of 426 * probabilities; horizontal lines correspond to a discontinuty 427 * in the delay distribution: the link will use the largest delay 428 * for a given probability. 429 * 430 * To pass the curve to dummynet, we must store the parameters 431 * in a file as described below, and issue the command 432 * 433 * ipfw pipe <n> config ... bw XXX profile <filename> ... 434 * 435 * The file format is the following, with whitespace acting as 436 * a separator and '#' indicating the beginning a comment: 437 * 438 * samples N 439 * the number of samples used in the internal 440 * representation (2..1024; default 100); 441 * 442 * loss-level L 443 * The probability above which packets are lost. 444 * (0.0 <= L <= 1.0, default 1.0 i.e. no loss); 445 * 446 * name identifier 447 * Optional a name (listed by "ipfw pipe show") 448 * to identify the distribution; 449 * 450 * "delay prob" | "prob delay" 451 * One of these two lines is mandatory and defines 452 * the format of the following lines with data points. 453 * 454 * XXX YYY 455 * 2 or more lines representing points in the curve, 456 * with either delay or probability first, according 457 * to the chosen format. 458 * The unit for delay is milliseconds. 459 * 460 * Data points does not need to be ordered or equal to the number 461 * specified in the "samples" line. ipfw will sort and interpolate 462 * the curve as needed. 463 * 464 * Example of a profile file: 465 466 name bla_bla_bla 467 samples 100 468 loss-level 0.86 469 prob delay 470 0 200 # minimum overhead is 200ms 471 0.5 200 472 0.5 300 473 0.8 1000 474 0.9 1300 475 1 1300 476 477 * Internally, we will convert the curve to a fixed number of 478 * samples, and when it is time to transmit a packet we will 479 * model the extra delay as extra bits in the packet. 480 * 481 */ 482 483 #define ED_MAX_LINE_LEN 256+ED_MAX_NAME_LEN 484 #define ED_TOK_SAMPLES "samples" 485 #define ED_TOK_LOSS "loss-level" 486 #define ED_TOK_NAME "name" 487 #define ED_TOK_DELAY "delay" 488 #define ED_TOK_PROB "prob" 489 #define ED_TOK_BW "bw" 490 #define ED_SEPARATORS " \t\n" 491 #define ED_MIN_SAMPLES_NO 2 492 493 /* 494 * returns 1 if s is a non-negative number, with at least one '.' 495 */ 496 static int 497 is_valid_number(const char *s) 498 { 499 int i, dots_found = 0; 500 int len = strlen(s); 501 502 for (i = 0; i<len; ++i) 503 if (!isdigit(s[i]) && (s[i] !='.' || ++dots_found > 1)) 504 return 0; 505 return 1; 506 } 507 508 /* 509 * Take as input a string describing a bandwidth value 510 * and return the numeric bandwidth value. 511 * set clocking interface or bandwidth value 512 */ 513 static void 514 read_bandwidth(char *arg, int *bandwidth, char *if_name, int namelen) 515 { 516 if (*bandwidth != -1) 517 warnx("duplicate token, override bandwidth value!"); 518 519 if (arg[0] >= 'a' && arg[0] <= 'z') { 520 if (!if_name) { 521 errx(1, "no if support"); 522 } 523 if (namelen >= IFNAMSIZ) 524 warn("interface name truncated"); 525 namelen--; 526 /* interface name */ 527 strncpy(if_name, arg, namelen); 528 if_name[namelen] = '\0'; 529 *bandwidth = 0; 530 } else { /* read bandwidth value */ 531 int bw; 532 char *end = NULL; 533 534 bw = strtoul(arg, &end, 0); 535 if (*end == 'K' || *end == 'k') { 536 end++; 537 bw *= 1000; 538 } else if (*end == 'M' || *end == 'm') { 539 end++; 540 bw *= 1000000; 541 } 542 if ((*end == 'B' && 543 _substrcmp2(end, "Bi", "Bit/s") != 0) || 544 _substrcmp2(end, "by", "bytes") == 0) 545 bw *= 8; 546 547 if (bw < 0) 548 errx(EX_DATAERR, "bandwidth too large"); 549 550 *bandwidth = bw; 551 if (if_name) 552 if_name[0] = '\0'; 553 } 554 } 555 556 struct point { 557 double prob; 558 double delay; 559 }; 560 561 static int 562 compare_points(const void *vp1, const void *vp2) 563 { 564 const struct point *p1 = vp1; 565 const struct point *p2 = vp2; 566 double res = 0; 567 568 res = p1->prob - p2->prob; 569 if (res == 0) 570 res = p1->delay - p2->delay; 571 if (res < 0) 572 return -1; 573 else if (res > 0) 574 return 1; 575 else 576 return 0; 577 } 578 579 #define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno 580 581 static void 582 load_extra_delays(const char *filename, struct dn_profile *p, 583 struct dn_link *link) 584 { 585 char line[ED_MAX_LINE_LEN]; 586 FILE *f; 587 int lineno = 0; 588 int i; 589 590 int samples = -1; 591 double loss = -1.0; 592 char profile_name[ED_MAX_NAME_LEN]; 593 int delay_first = -1; 594 int do_points = 0; 595 struct point points[ED_MAX_SAMPLES_NO]; 596 int points_no = 0; 597 598 /* XXX link never NULL? */ 599 p->link_nr = link->link_nr; 600 601 profile_name[0] = '\0'; 602 f = fopen(filename, "r"); 603 if (f == NULL) 604 err(EX_UNAVAILABLE, "fopen: %s", filename); 605 606 while (fgets(line, ED_MAX_LINE_LEN, f)) { /* read commands */ 607 char *s, *cur = line, *name = NULL, *arg = NULL; 608 609 ++lineno; 610 611 /* parse the line */ 612 while (cur) { 613 s = strsep(&cur, ED_SEPARATORS); 614 if (s == NULL || *s == '#') 615 break; 616 if (*s == '\0') 617 continue; 618 if (arg) 619 errx(ED_EFMT("too many arguments")); 620 if (name == NULL) 621 name = s; 622 else 623 arg = s; 624 } 625 if (name == NULL) /* empty line */ 626 continue; 627 if (arg == NULL) 628 errx(ED_EFMT("missing arg for %s"), name); 629 630 if (!strcasecmp(name, ED_TOK_SAMPLES)) { 631 if (samples > 0) 632 errx(ED_EFMT("duplicate ``samples'' line")); 633 if (atoi(arg) <=0) 634 errx(ED_EFMT("invalid number of samples")); 635 samples = atoi(arg); 636 if (samples>ED_MAX_SAMPLES_NO) 637 errx(ED_EFMT("too many samples, maximum is %d"), 638 ED_MAX_SAMPLES_NO); 639 do_points = 0; 640 } else if (!strcasecmp(name, ED_TOK_BW)) { 641 char buf[IFNAMSIZ]; 642 read_bandwidth(arg, &link->bandwidth, buf, sizeof(buf)); 643 } else if (!strcasecmp(name, ED_TOK_LOSS)) { 644 if (loss != -1.0) 645 errx(ED_EFMT("duplicated token: %s"), name); 646 if (!is_valid_number(arg)) 647 errx(ED_EFMT("invalid %s"), arg); 648 loss = atof(arg); 649 if (loss > 1) 650 errx(ED_EFMT("%s greater than 1.0"), name); 651 do_points = 0; 652 } else if (!strcasecmp(name, ED_TOK_NAME)) { 653 if (profile_name[0] != '\0') 654 errx(ED_EFMT("duplicated token: %s"), name); 655 strncpy(profile_name, arg, sizeof(profile_name) - 1); 656 profile_name[sizeof(profile_name)-1] = '\0'; 657 do_points = 0; 658 } else if (!strcasecmp(name, ED_TOK_DELAY)) { 659 if (do_points) 660 errx(ED_EFMT("duplicated token: %s"), name); 661 delay_first = 1; 662 do_points = 1; 663 } else if (!strcasecmp(name, ED_TOK_PROB)) { 664 if (do_points) 665 errx(ED_EFMT("duplicated token: %s"), name); 666 delay_first = 0; 667 do_points = 1; 668 } else if (do_points) { 669 if (!is_valid_number(name) || !is_valid_number(arg)) 670 errx(ED_EFMT("invalid point found")); 671 if (delay_first) { 672 points[points_no].delay = atof(name); 673 points[points_no].prob = atof(arg); 674 } else { 675 points[points_no].delay = atof(arg); 676 points[points_no].prob = atof(name); 677 } 678 if (points[points_no].prob > 1.0) 679 errx(ED_EFMT("probability greater than 1.0")); 680 ++points_no; 681 } else { 682 errx(ED_EFMT("unrecognised command '%s'"), name); 683 } 684 } 685 686 fclose (f); 687 688 if (samples == -1) { 689 warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES); 690 samples = 100; 691 } 692 693 if (loss == -1.0) { 694 warnx("'%s' not found, assuming no loss", ED_TOK_LOSS); 695 loss = 1; 696 } 697 698 /* make sure that there are enough points. */ 699 if (points_no < ED_MIN_SAMPLES_NO) 700 errx(ED_EFMT("too few samples, need at least %d"), 701 ED_MIN_SAMPLES_NO); 702 703 qsort(points, points_no, sizeof(struct point), compare_points); 704 705 /* interpolation */ 706 for (i = 0; i<points_no-1; ++i) { 707 double y1 = points[i].prob * samples; 708 double x1 = points[i].delay; 709 double y2 = points[i+1].prob * samples; 710 double x2 = points[i+1].delay; 711 712 int ix = y1; 713 int stop = y2; 714 715 if (x1 == x2) { 716 for (; ix<stop; ++ix) 717 p->samples[ix] = x1; 718 } else { 719 double m = (y2-y1)/(x2-x1); 720 double c = y1 - m*x1; 721 for (; ix<stop ; ++ix) 722 p->samples[ix] = (ix - c)/m; 723 } 724 } 725 p->samples_no = samples; 726 p->loss_level = loss * samples; 727 strncpy(p->name, profile_name, sizeof(p->name)); 728 } 729 730 /* 731 * configuration of pipes, schedulers, flowsets. 732 * When we configure a new scheduler, an empty pipe is created, so: 733 * 734 * do_pipe = 1 -> "pipe N config ..." only for backward compatibility 735 * sched N+Delta type fifo sched_mask ... 736 * pipe N+Delta <parameters> 737 * flowset N+Delta pipe N+Delta (no parameters) 738 * sched N type wf2q+ sched_mask ... 739 * pipe N <parameters> 740 * 741 * do_pipe = 2 -> flowset N config 742 * flowset N parameters 743 * 744 * do_pipe = 3 -> sched N config 745 * sched N parameters (default no pipe) 746 * optional Pipe N config ... 747 * pipe ==> 748 */ 749 void 750 ipfw_config_pipe(int ac, char **av) 751 { 752 int i, j; 753 char *end; 754 void *par = NULL; 755 struct dn_id *buf, *base; 756 struct dn_sch *sch = NULL; 757 struct dn_link *p = NULL; 758 struct dn_fs *fs = NULL; 759 struct dn_profile *pf = NULL; 760 struct ipfw_flow_id *mask = NULL; 761 int lmax; 762 uint32_t _foo = 0, *flags = &_foo , *buckets = &_foo; 763 764 /* 765 * allocate space for 1 header, 766 * 1 scheduler, 1 link, 1 flowset, 1 profile 767 */ 768 lmax = sizeof(struct dn_id); /* command header */ 769 lmax += sizeof(struct dn_sch) + sizeof(struct dn_link) + 770 sizeof(struct dn_fs) + sizeof(struct dn_profile); 771 772 av++; ac--; 773 /* Pipe number */ 774 if (ac && isdigit(**av)) { 775 i = atoi(*av); av++; ac--; 776 } else 777 i = -1; 778 if (i <= 0) 779 errx(EX_USAGE, "need a pipe/flowset/sched number"); 780 base = buf = safe_calloc(1, lmax); 781 /* all commands start with a 'CONFIGURE' and a version */ 782 o_next(&buf, sizeof(struct dn_id), DN_CMD_CONFIG); 783 base->id = DN_API_VERSION; 784 785 switch (co.do_pipe) { 786 case 1: /* "pipe N config ..." */ 787 /* Allocate space for the WF2Q+ scheduler, its link 788 * and the FIFO flowset. Set the number, but leave 789 * the scheduler subtype and other parameters to 0 790 * so the kernel will use appropriate defaults. 791 * XXX todo: add a flag to record if a parameter 792 * is actually configured. 793 * If we do a 'pipe config' mask -> sched_mask. 794 * The FIFO scheduler and link are derived from the 795 * WF2Q+ one in the kernel. 796 */ 797 sch = o_next(&buf, sizeof(*sch), DN_SCH); 798 p = o_next(&buf, sizeof(*p), DN_LINK); 799 fs = o_next(&buf, sizeof(*fs), DN_FS); 800 801 sch->sched_nr = i; 802 sch->oid.subtype = 0; /* defaults to WF2Q+ */ 803 mask = &sch->sched_mask; 804 flags = &sch->flags; 805 buckets = &sch->buckets; 806 *flags |= DN_PIPE_CMD; 807 808 p->link_nr = i; 809 810 /* This flowset is only for the FIFO scheduler */ 811 fs->fs_nr = i + 2*DN_MAX_ID; 812 fs->sched_nr = i + DN_MAX_ID; 813 break; 814 815 case 2: /* "queue N config ... " */ 816 fs = o_next(&buf, sizeof(*fs), DN_FS); 817 fs->fs_nr = i; 818 mask = &fs->flow_mask; 819 flags = &fs->flags; 820 buckets = &fs->buckets; 821 break; 822 823 case 3: /* "sched N config ..." */ 824 sch = o_next(&buf, sizeof(*sch), DN_SCH); 825 fs = o_next(&buf, sizeof(*fs), DN_FS); 826 sch->sched_nr = i; 827 mask = &sch->sched_mask; 828 flags = &sch->flags; 829 buckets = &sch->buckets; 830 /* fs is used only with !MULTIQUEUE schedulers */ 831 fs->fs_nr = i + DN_MAX_ID; 832 fs->sched_nr = i; 833 break; 834 } 835 /* set to -1 those fields for which we want to reuse existing 836 * values from the kernel. 837 * Also, *_nr and subtype = 0 mean reuse the value from the kernel. 838 * XXX todo: support reuse of the mask. 839 */ 840 if (p) 841 p->bandwidth = -1; 842 for (j = 0; j < sizeof(fs->par)/sizeof(fs->par[0]); j++) 843 fs->par[j] = -1; 844 while (ac > 0) { 845 double d; 846 int tok = match_token(dummynet_params, *av); 847 ac--; av++; 848 849 switch(tok) { 850 case TOK_NOERROR: 851 NEED(fs, "noerror is only for pipes"); 852 fs->flags |= DN_NOERROR; 853 break; 854 855 case TOK_PLR: 856 NEED(fs, "plr is only for pipes"); 857 NEED1("plr needs argument 0..1\n"); 858 d = strtod(av[0], NULL); 859 if (d > 1) 860 d = 1; 861 else if (d < 0) 862 d = 0; 863 fs->plr = (int)(d*0x7fffffff); 864 ac--; av++; 865 break; 866 867 case TOK_QUEUE: 868 NEED(fs, "queue is only for pipes or flowsets"); 869 NEED1("queue needs queue size\n"); 870 end = NULL; 871 fs->qsize = strtoul(av[0], &end, 0); 872 if (*end == 'K' || *end == 'k') { 873 fs->flags |= DN_QSIZE_BYTES; 874 fs->qsize *= 1024; 875 } else if (*end == 'B' || 876 _substrcmp2(end, "by", "bytes") == 0) { 877 fs->flags |= DN_QSIZE_BYTES; 878 } 879 ac--; av++; 880 break; 881 882 case TOK_BUCKETS: 883 NEED(fs, "buckets is only for pipes or flowsets"); 884 NEED1("buckets needs argument\n"); 885 *buckets = strtoul(av[0], NULL, 0); 886 ac--; av++; 887 break; 888 889 case TOK_FLOW_MASK: 890 case TOK_SCHED_MASK: 891 case TOK_MASK: 892 NEED(mask, "tok_mask"); 893 NEED1("mask needs mask specifier\n"); 894 /* 895 * per-flow queue, mask is dst_ip, dst_port, 896 * src_ip, src_port, proto measured in bits 897 */ 898 par = NULL; 899 900 bzero(mask, sizeof(*mask)); 901 end = NULL; 902 903 while (ac >= 1) { 904 uint32_t *p32 = NULL; 905 uint16_t *p16 = NULL; 906 uint32_t *p20 = NULL; 907 struct in6_addr *pa6 = NULL; 908 uint32_t a; 909 910 tok = match_token(dummynet_params, *av); 911 ac--; av++; 912 switch(tok) { 913 case TOK_ALL: 914 /* 915 * special case, all bits significant 916 * except 'extra' (the queue number) 917 */ 918 mask->dst_ip = ~0; 919 mask->src_ip = ~0; 920 mask->dst_port = ~0; 921 mask->src_port = ~0; 922 mask->proto = ~0; 923 n2mask(&mask->dst_ip6, 128); 924 n2mask(&mask->src_ip6, 128); 925 mask->flow_id6 = ~0; 926 *flags |= DN_HAVE_MASK; 927 goto end_mask; 928 929 case TOK_QUEUE: 930 mask->extra = ~0; 931 *flags |= DN_HAVE_MASK; 932 goto end_mask; 933 934 case TOK_DSTIP: 935 mask->addr_type = 4; 936 p32 = &mask->dst_ip; 937 break; 938 939 case TOK_SRCIP: 940 mask->addr_type = 4; 941 p32 = &mask->src_ip; 942 break; 943 944 case TOK_DSTIP6: 945 mask->addr_type = 6; 946 pa6 = &mask->dst_ip6; 947 break; 948 949 case TOK_SRCIP6: 950 mask->addr_type = 6; 951 pa6 = &mask->src_ip6; 952 break; 953 954 case TOK_FLOWID: 955 mask->addr_type = 6; 956 p20 = &mask->flow_id6; 957 break; 958 959 case TOK_DSTPORT: 960 p16 = &mask->dst_port; 961 break; 962 963 case TOK_SRCPORT: 964 p16 = &mask->src_port; 965 break; 966 967 case TOK_PROTO: 968 break; 969 970 default: 971 ac++; av--; /* backtrack */ 972 goto end_mask; 973 } 974 if (ac < 1) 975 errx(EX_USAGE, "mask: value missing"); 976 if (*av[0] == '/') { 977 a = strtoul(av[0]+1, &end, 0); 978 if (pa6 == NULL) 979 a = (a == 32) ? ~0 : (1 << a) - 1; 980 } else 981 a = strtoul(av[0], &end, 0); 982 if (p32 != NULL) 983 *p32 = a; 984 else if (p16 != NULL) { 985 if (a > 0xFFFF) 986 errx(EX_DATAERR, 987 "port mask must be 16 bit"); 988 *p16 = (uint16_t)a; 989 } else if (p20 != NULL) { 990 if (a > 0xfffff) 991 errx(EX_DATAERR, 992 "flow_id mask must be 20 bit"); 993 *p20 = (uint32_t)a; 994 } else if (pa6 != NULL) { 995 if (a > 128) 996 errx(EX_DATAERR, 997 "in6addr invalid mask len"); 998 else 999 n2mask(pa6, a); 1000 } else { 1001 if (a > 0xFF) 1002 errx(EX_DATAERR, 1003 "proto mask must be 8 bit"); 1004 mask->proto = (uint8_t)a; 1005 } 1006 if (a != 0) 1007 *flags |= DN_HAVE_MASK; 1008 ac--; av++; 1009 } /* end while, config masks */ 1010 end_mask: 1011 break; 1012 1013 case TOK_RED: 1014 case TOK_GRED: 1015 NEED1("red/gred needs w_q/min_th/max_th/max_p\n"); 1016 fs->flags |= DN_IS_RED; 1017 if (tok == TOK_GRED) 1018 fs->flags |= DN_IS_GENTLE_RED; 1019 /* 1020 * the format for parameters is w_q/min_th/max_th/max_p 1021 */ 1022 if ((end = strsep(&av[0], "/"))) { 1023 double w_q = strtod(end, NULL); 1024 if (w_q > 1 || w_q <= 0) 1025 errx(EX_DATAERR, "0 < w_q <= 1"); 1026 fs->w_q = (int) (w_q * (1 << SCALE_RED)); 1027 } 1028 if ((end = strsep(&av[0], "/"))) { 1029 fs->min_th = strtoul(end, &end, 0); 1030 if (*end == 'K' || *end == 'k') 1031 fs->min_th *= 1024; 1032 } 1033 if ((end = strsep(&av[0], "/"))) { 1034 fs->max_th = strtoul(end, &end, 0); 1035 if (*end == 'K' || *end == 'k') 1036 fs->max_th *= 1024; 1037 } 1038 if ((end = strsep(&av[0], "/"))) { 1039 double max_p = strtod(end, NULL); 1040 if (max_p > 1 || max_p <= 0) 1041 errx(EX_DATAERR, "0 < max_p <= 1"); 1042 fs->max_p = (int)(max_p * (1 << SCALE_RED)); 1043 } 1044 ac--; av++; 1045 break; 1046 1047 case TOK_DROPTAIL: 1048 NEED(fs, "droptail is only for flowsets"); 1049 fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED); 1050 break; 1051 1052 case TOK_BW: 1053 NEED(p, "bw is only for links"); 1054 NEED1("bw needs bandwidth or interface\n"); 1055 read_bandwidth(av[0], &p->bandwidth, NULL, 0); 1056 ac--; av++; 1057 break; 1058 1059 case TOK_DELAY: 1060 NEED(p, "delay is only for links"); 1061 NEED1("delay needs argument 0..10000ms\n"); 1062 p->delay = strtoul(av[0], NULL, 0); 1063 ac--; av++; 1064 break; 1065 1066 case TOK_TYPE: { 1067 int l; 1068 NEED(sch, "type is only for schedulers"); 1069 NEED1("type needs a string"); 1070 l = strlen(av[0]); 1071 if (l == 0 || l > 15) 1072 errx(1, "type %s too long\n", av[0]); 1073 strcpy(sch->name, av[0]); 1074 sch->oid.subtype = 0; /* use string */ 1075 ac--; av++; 1076 break; 1077 } 1078 1079 case TOK_WEIGHT: 1080 NEED(fs, "weight is only for flowsets"); 1081 NEED1("weight needs argument\n"); 1082 fs->par[0] = strtol(av[0], &end, 0); 1083 ac--; av++; 1084 break; 1085 1086 case TOK_LMAX: 1087 NEED(fs, "lmax is only for flowsets"); 1088 NEED1("lmax needs argument\n"); 1089 fs->par[1] = strtol(av[0], &end, 0); 1090 ac--; av++; 1091 break; 1092 1093 case TOK_PRI: 1094 NEED(fs, "priority is only for flowsets"); 1095 NEED1("priority needs argument\n"); 1096 fs->par[2] = strtol(av[0], &end, 0); 1097 ac--; av++; 1098 break; 1099 1100 case TOK_SCHED: 1101 case TOK_PIPE: 1102 NEED(fs, "pipe/sched"); 1103 NEED1("pipe/link/sched needs number\n"); 1104 fs->sched_nr = strtoul(av[0], &end, 0); 1105 ac--; av++; 1106 break; 1107 1108 case TOK_PROFILE: 1109 NEED((!pf), "profile already set"); 1110 NEED(p, "profile"); 1111 { 1112 NEED1("extra delay needs the file name\n"); 1113 pf = o_next(&buf, sizeof(*pf), DN_PROFILE); 1114 load_extra_delays(av[0], pf, p); //XXX can't fail? 1115 --ac; ++av; 1116 } 1117 break; 1118 1119 case TOK_BURST: 1120 NEED(p, "burst"); 1121 NEED1("burst needs argument\n"); 1122 errno = 0; 1123 if (expand_number(av[0], (int64_t *)&p->burst) < 0) 1124 if (errno != ERANGE) 1125 errx(EX_DATAERR, 1126 "burst: invalid argument"); 1127 if (errno || p->burst > (1ULL << 48) - 1) 1128 errx(EX_DATAERR, 1129 "burst: out of range (0..2^48-1)"); 1130 ac--; av++; 1131 break; 1132 1133 default: 1134 errx(EX_DATAERR, "unrecognised option ``%s''", av[-1]); 1135 } 1136 } 1137 1138 /* check validity of parameters */ 1139 if (p) { 1140 if (p->delay > 10000) 1141 errx(EX_DATAERR, "delay must be < 10000"); 1142 if (p->bandwidth == -1) 1143 p->bandwidth = 0; 1144 } 1145 if (fs) { 1146 /* XXX accept a 0 scheduler to keep the default */ 1147 if (fs->flags & DN_QSIZE_BYTES) { 1148 size_t len; 1149 long limit; 1150 1151 len = sizeof(limit); 1152 if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit", 1153 &limit, &len, NULL, 0) == -1) 1154 limit = 1024*1024; 1155 if (fs->qsize > limit) 1156 errx(EX_DATAERR, "queue size must be < %ldB", limit); 1157 } else { 1158 size_t len; 1159 long limit; 1160 1161 len = sizeof(limit); 1162 if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit", 1163 &limit, &len, NULL, 0) == -1) 1164 limit = 100; 1165 if (fs->qsize > limit) 1166 errx(EX_DATAERR, "2 <= queue size <= %ld", limit); 1167 } 1168 1169 if (fs->flags & DN_IS_RED) { 1170 size_t len; 1171 int lookup_depth, avg_pkt_size; 1172 double w_q; 1173 1174 if (fs->min_th >= fs->max_th) 1175 errx(EX_DATAERR, "min_th %d must be < than max_th %d", 1176 fs->min_th, fs->max_th); 1177 if (fs->max_th == 0) 1178 errx(EX_DATAERR, "max_th must be > 0"); 1179 1180 len = sizeof(int); 1181 if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth", 1182 &lookup_depth, &len, NULL, 0) == -1) 1183 lookup_depth = 256; 1184 if (lookup_depth == 0) 1185 errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth" 1186 " must be greater than zero"); 1187 1188 len = sizeof(int); 1189 if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size", 1190 &avg_pkt_size, &len, NULL, 0) == -1) 1191 avg_pkt_size = 512; 1192 1193 if (avg_pkt_size == 0) 1194 errx(EX_DATAERR, 1195 "net.inet.ip.dummynet.red_avg_pkt_size must" 1196 " be greater than zero"); 1197 1198 /* 1199 * Ticks needed for sending a medium-sized packet. 1200 * Unfortunately, when we are configuring a WF2Q+ queue, we 1201 * do not have bandwidth information, because that is stored 1202 * in the parent pipe, and also we have multiple queues 1203 * competing for it. So we set s=0, which is not very 1204 * correct. But on the other hand, why do we want RED with 1205 * WF2Q+ ? 1206 */ 1207 #if 0 1208 if (p.bandwidth==0) /* this is a WF2Q+ queue */ 1209 s = 0; 1210 else 1211 s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth; 1212 #endif 1213 /* 1214 * max idle time (in ticks) before avg queue size becomes 0. 1215 * NOTA: (3/w_q) is approx the value x so that 1216 * (1-w_q)^x < 10^-3. 1217 */ 1218 w_q = ((double)fs->w_q) / (1 << SCALE_RED); 1219 #if 0 // go in kernel 1220 idle = s * 3. / w_q; 1221 fs->lookup_step = (int)idle / lookup_depth; 1222 if (!fs->lookup_step) 1223 fs->lookup_step = 1; 1224 weight = 1 - w_q; 1225 for (t = fs->lookup_step; t > 1; --t) 1226 weight *= 1 - w_q; 1227 fs->lookup_weight = (int)(weight * (1 << SCALE_RED)); 1228 #endif 1229 } 1230 } 1231 1232 i = do_cmd(IP_DUMMYNET3, base, (char *)buf - (char *)base); 1233 1234 if (i) 1235 err(1, "setsockopt(%s)", "IP_DUMMYNET_CONFIGURE"); 1236 } 1237 1238 void 1239 dummynet_flush(void) 1240 { 1241 struct dn_id oid; 1242 oid_fill(&oid, sizeof(oid), DN_CMD_FLUSH, DN_API_VERSION); 1243 do_cmd(IP_DUMMYNET3, &oid, oid.len); 1244 } 1245 1246 /* Parse input for 'ipfw [pipe|sched|queue] show [range list]' 1247 * Returns the number of ranges, and possibly stores them 1248 * in the array v of size len. 1249 */ 1250 static int 1251 parse_range(int ac, char *av[], uint32_t *v, int len) 1252 { 1253 int n = 0; 1254 char *endptr, *s; 1255 uint32_t base[2]; 1256 1257 if (v == NULL || len < 2) { 1258 v = base; 1259 len = 2; 1260 } 1261 1262 for (s = *av; s != NULL; av++, ac--) { 1263 v[0] = strtoul(s, &endptr, 10); 1264 v[1] = (*endptr != '-') ? v[0] : 1265 strtoul(endptr+1, &endptr, 10); 1266 if (*endptr == '\0') { /* prepare for next round */ 1267 s = (ac > 0) ? *(av+1) : NULL; 1268 } else { 1269 if (*endptr != ',') { 1270 warn("invalid number: %s", s); 1271 s = ++endptr; 1272 continue; 1273 } 1274 /* continue processing from here */ 1275 s = ++endptr; 1276 ac++; 1277 av--; 1278 } 1279 if (v[1] < v[0] || 1280 v[1] < 0 || v[1] >= DN_MAX_ID-1 || 1281 v[0] < 0 || v[1] >= DN_MAX_ID-1) { 1282 continue; /* invalid entry */ 1283 } 1284 n++; 1285 /* translate if 'pipe list' */ 1286 if (co.do_pipe == 1) { 1287 v[0] += DN_MAX_ID; 1288 v[1] += DN_MAX_ID; 1289 } 1290 v = (n*2 < len) ? v + 2 : base; 1291 } 1292 return n; 1293 } 1294 1295 /* main entry point for dummynet list functions. co.do_pipe indicates 1296 * which function we want to support. 1297 * av may contain filtering arguments, either individual entries 1298 * or ranges, or lists (space or commas are valid separators). 1299 * Format for a range can be n1-n2 or n3 n4 n5 ... 1300 * In a range n1 must be <= n2, otherwise the range is ignored. 1301 * A number 'n4' is translate in a range 'n4-n4' 1302 * All number must be > 0 and < DN_MAX_ID-1 1303 */ 1304 void 1305 dummynet_list(int ac, char *av[], int show_counters) 1306 { 1307 struct dn_id *oid, *x = NULL; 1308 int ret, i, l; 1309 int n; /* # of ranges */ 1310 int buflen; 1311 int max_size; /* largest obj passed up */ 1312 1313 ac--; 1314 av++; /* skip 'list' | 'show' word */ 1315 1316 n = parse_range(ac, av, NULL, 0); /* Count # of ranges. */ 1317 1318 /* Allocate space to store ranges */ 1319 l = sizeof(*oid) + sizeof(uint32_t) * n * 2; 1320 oid = safe_calloc(1, l); 1321 oid_fill(oid, l, DN_CMD_GET, DN_API_VERSION); 1322 1323 if (n > 0) /* store ranges in idx */ 1324 parse_range(ac, av, (uint32_t *)(oid + 1), n*2); 1325 /* 1326 * Compute the size of the largest object returned. If the 1327 * response leaves at least this much spare space in the 1328 * buffer, then surely the response is complete; otherwise 1329 * there might be a risk of truncation and we will need to 1330 * retry with a larger buffer. 1331 * XXX don't bother with smaller structs. 1332 */ 1333 max_size = sizeof(struct dn_fs); 1334 if (max_size < sizeof(struct dn_sch)) 1335 max_size = sizeof(struct dn_sch); 1336 if (max_size < sizeof(struct dn_flow)) 1337 max_size = sizeof(struct dn_flow); 1338 1339 switch (co.do_pipe) { 1340 case 1: 1341 oid->subtype = DN_LINK; /* list pipe */ 1342 break; 1343 case 2: 1344 oid->subtype = DN_FS; /* list queue */ 1345 break; 1346 case 3: 1347 oid->subtype = DN_SCH; /* list sched */ 1348 break; 1349 } 1350 1351 /* 1352 * Ask the kernel an estimate of the required space (result 1353 * in oid.id), unless we are requesting a subset of objects, 1354 * in which case the kernel does not give an exact answer. 1355 * In any case, space might grow in the meantime due to the 1356 * creation of new queues, so we must be prepared to retry. 1357 */ 1358 if (n > 0) { 1359 buflen = 4*1024; 1360 } else { 1361 ret = do_cmd(-IP_DUMMYNET3, oid, (uintptr_t)&l); 1362 if (ret != 0 || oid->id <= sizeof(*oid)) 1363 goto done; 1364 buflen = oid->id + max_size; 1365 oid->len = sizeof(*oid); /* restore */ 1366 } 1367 /* Try a few times, until the buffer fits */ 1368 for (i = 0; i < 20; i++) { 1369 l = buflen; 1370 x = safe_realloc(x, l); 1371 bcopy(oid, x, oid->len); 1372 ret = do_cmd(-IP_DUMMYNET3, x, (uintptr_t)&l); 1373 if (ret != 0 || x->id <= sizeof(*oid)) 1374 goto done; /* no response */ 1375 if (l + max_size <= buflen) 1376 break; /* ok */ 1377 buflen *= 2; /* double for next attempt */ 1378 } 1379 list_pipes(x, O_NEXT(x, l)); 1380 done: 1381 if (x) 1382 free(x); 1383 free(oid); 1384 } 1385