ipfw/dummynet.c

   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 187769 2009-01-27 11:06:59Z 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
  47 static 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
  77 /*
  78  * XXX to be updated to the new version,
  79  * without the global struct command_opts variable
  80  */
  81 static int
  82 sort_q(void * to_be_done, const void *pa, const void *pb)
  83 {
  84         int rev = (co.do_sort < 0);
  85         int field = rev ? -co.do_sort : co.do_sort;
  86         long long res = 0;
  87         const struct dn_flow_queue *a = pa;
  88         const struct dn_flow_queue *b = pb;
  89
  90         switch (field) {
  91         case 1: /* pkts */
  92                 res = a->len - b->len;
  93                 break;
  94         case 2: /* bytes */
  95                 res = a->len_bytes - b->len_bytes;
  96                 break;
  97
  98         case 3: /* tot pkts */
  99                 res = a->tot_pkts - b->tot_pkts;
 100                 break;
 101
 102         case 4: /* tot bytes */
 103                 res = a->tot_bytes - b->tot_bytes;
 104                 break;
 105         }
 106         if (res < 0)
 107                 res = -1;
 108         if (res > 0)
 109                 res = 1;
 110         return (int)(rev ? res : -res);
 111 }
 112
 113 static void
 114 list_queues(struct dn_flow_set *fs, struct dn_flow_queue *q)
 115 {
 116         int l;
 117         int index_printed, indexes = 0;
 118         char buff[255];
 119         struct protoent *pe;
 120
 121         if (fs->rq_elements == 0)
 122                 return;
 123
 124         if (co.do_sort != 0)
 125                 qsort_r(q, fs->rq_elements, sizeof *q, NULL, sort_q);
 126
 127         /* Print IPv4 flows */
 128         index_printed = 0;
 129         for (l = 0; l < fs->rq_elements; l++) {
 130                 struct in_addr ina;
 131
 132                 /* XXX: Should check for IPv4 flows */
 133                 if (IS_IP6_FLOW_ID(&(q[l].id)))
 134                         continue;
 135
 136                 if (!index_printed) {
 137                         index_printed = 1;
 138                         if (indexes > 0)        /* currently a no-op */
 139                                 printf("\n");
 140                         indexes++;
 141                         printf("    "
 142                             "mask: 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n",
 143                             fs->flow_mask.proto,
 144                             fs->flow_mask.src_ip, fs->flow_mask.src_port,
 145                             fs->flow_mask.dst_ip, fs->flow_mask.dst_port);
 146
 147                         printf("BKT Prot ___Source IP/port____ "
 148                             "____Dest. IP/port____ "
 149                             "Tot_pkt/bytes Pkt/Byte Drp\n");
 150                 }
 151
 152                 printf("%3d ", q[l].hash_slot);
 153                 pe = getprotobynumber(q[l].id.proto);
 154                 if (pe)
 155                         printf("%-4s ", pe->p_name);
 156                 else
 157                         printf("%4u ", q[l].id.proto);
 158                 ina.s_addr = htonl(q[l].id.src_ip);
 159                 printf("%15s/%-5d ",
 160                     inet_ntoa(ina), q[l].id.src_port);
 161                 ina.s_addr = htonl(q[l].id.dst_ip);
 162                 printf("%15s/%-5d ",
 163                     inet_ntoa(ina), q[l].id.dst_port);
 164                 printf("%4llu %8llu %2u %4u %3u\n",
 165                     align_uint64(&q[l].tot_pkts),
 166                     align_uint64(&q[l].tot_bytes),
 167                     q[l].len, q[l].len_bytes, q[l].drops);
 168                 if (co.verbose)
 169                         printf("   S %20llu  F %20llu\n",
 170                             align_uint64(&q[l].S), align_uint64(&q[l].F));
 171         }
 172
 173         /* Print IPv6 flows */
 174         index_printed = 0;
 175         for (l = 0; l < fs->rq_elements; l++) {
 176                 if (!IS_IP6_FLOW_ID(&(q[l].id)))
 177                         continue;
 178
 179                 if (!index_printed) {
 180                         index_printed = 1;
 181                         if (indexes > 0)
 182                                 printf("\n");
 183                         indexes++;
 184                         printf("\n        mask: proto: 0x%02x, flow_id: 0x%08x,  ",
 185                             fs->flow_mask.proto, fs->flow_mask.flow_id6);
 186                         inet_ntop(AF_INET6, &(fs->flow_mask.src_ip6),
 187                             buff, sizeof(buff));
 188                         printf("%s/0x%04x -> ", buff, fs->flow_mask.src_port);
 189                         inet_ntop( AF_INET6, &(fs->flow_mask.dst_ip6),
 190                             buff, sizeof(buff) );
 191                         printf("%s/0x%04x\n", buff, fs->flow_mask.dst_port);
 192
 193                         printf("BKT ___Prot___ _flow-id_ "
 194                             "______________Source IPv6/port_______________ "
 195                             "_______________Dest. IPv6/port_______________ "
 196                             "Tot_pkt/bytes Pkt/Byte Drp\n");
 197                 }
 198                 printf("%3d ", q[l].hash_slot);
 199                 pe = getprotobynumber(q[l].id.proto);
 200                 if (pe != NULL)
 201                         printf("%9s ", pe->p_name);
 202                 else
 203                         printf("%9u ", q[l].id.proto);
 204                 printf("%7d  %39s/%-5d ", q[l].id.flow_id6,
 205                     inet_ntop(AF_INET6, &(q[l].id.src_ip6), buff, sizeof(buff)),
 206                     q[l].id.src_port);
 207                 printf(" %39s/%-5d ",
 208                     inet_ntop(AF_INET6, &(q[l].id.dst_ip6), buff, sizeof(buff)),
 209                     q[l].id.dst_port);
 210                 printf(" %4llu %8llu %2u %4u %3u\n",
 211                     align_uint64(&q[l].tot_pkts),
 212                     align_uint64(&q[l].tot_bytes),
 213                     q[l].len, q[l].len_bytes, q[l].drops);
 214                 if (co.verbose)
 215                         printf("   S %20llu  F %20llu\n",
 216                             align_uint64(&q[l].S),
 217                             align_uint64(&q[l].F));
 218         }
 219 }
 220
 221 static void
 222 print_flowset_parms(struct dn_flow_set *fs, char *prefix)
 223 {
 224         int l;
 225         char qs[30];
 226         char plr[30];
 227         char red[90];   /* Display RED parameters */
 228
 229         l = fs->qsize;
 230         if (fs->flags_fs & DN_QSIZE_IS_BYTES) {
 231                 if (l >= 8192)
 232                         sprintf(qs, "%d KB", l / 1024);
 233                 else
 234                         sprintf(qs, "%d B", l);
 235         } else
 236                 sprintf(qs, "%3d sl.", l);
 237         if (fs->plr)
 238                 sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff));
 239         else
 240                 plr[0] = '\0';
 241         if (fs->flags_fs & DN_IS_RED)   /* RED parameters */
 242                 sprintf(red,
 243                     "\n\t  %cRED w_q %f min_th %d max_th %d max_p %f",
 244                     (fs->flags_fs & DN_IS_GENTLE_RED) ? 'G' : ' ',
 245                     1.0 * fs->w_q / (double)(1 << SCALE_RED),
 246                     SCALE_VAL(fs->min_th),
 247                     SCALE_VAL(fs->max_th),
 248                     1.0 * fs->max_p / (double)(1 << SCALE_RED));
 249         else
 250                 sprintf(red, "droptail");
 251
 252         printf("%s %s%s %d queues (%d buckets) %s\n",
 253             prefix, qs, plr, fs->rq_elements, fs->rq_size, red);
 254 }
 255
 256 static void
 257 print_extra_delay_parms(struct dn_pipe *p, char *prefix)
 258 {
 259         double loss;
 260         if (p->samples_no <= 0)
 261                 return;
 262
 263         loss = p->loss_level;
 264         loss /= p->samples_no;
 265         printf("%s profile: name \"%s\" loss %f samples %d\n",
 266                 prefix, p->name, loss, p->samples_no);
 267 }
 268
 269 void
 270 ipfw_list_pipes(void *data, uint nbytes, int ac, char *av[])
 271 {
 272         int rulenum;
 273         void *next = data;
 274         struct dn_pipe *p = (struct dn_pipe *) data;
 275         struct dn_flow_set *fs;
 276         struct dn_flow_queue *q;
 277         int l;
 278
 279         if (ac > 0)
 280                 rulenum = strtoul(*av++, NULL, 10);
 281         else
 282                 rulenum = 0;
 283         for (; nbytes >= sizeof *p; p = (struct dn_pipe *)next) {
 284                 double b = p->bandwidth;
 285                 char buf[30];
 286                 char prefix[80];
 287
 288                 if (SLIST_NEXT(p, next) != (struct dn_pipe *)DN_IS_PIPE)
 289                         break;  /* done with pipes, now queues */
 290
 291                 /*
 292                  * compute length, as pipe have variable size
 293                  */
 294                 l = sizeof(*p) + p->fs.rq_elements * sizeof(*q);
 295                 next = (char *)p + l;
 296                 nbytes -= l;
 297
 298                 if ((rulenum != 0 && rulenum != p->pipe_nr) || co.do_pipe == 2)
 299                         continue;
 300
 301                 /*
 302                  * Print rate (or clocking interface)
 303                  */
 304                 if (p->if_name[0] != '\0')
 305                         sprintf(buf, "%s", p->if_name);
 306                 else if (b == 0)
 307                         sprintf(buf, "unlimited");
 308                 else if (b >= 1000000)
 309                         sprintf(buf, "%7.3f Mbit/s", b/1000000);
 310                 else if (b >= 1000)
 311                         sprintf(buf, "%7.3f Kbit/s", b/1000);
 312                 else
 313                         sprintf(buf, "%7.3f bit/s ", b);
 314
 315                 sprintf(prefix, "%05d: %s %4d ms ",
 316                     p->pipe_nr, buf, p->delay);
 317
 318                 print_extra_delay_parms(p, prefix);
 319
 320                 print_flowset_parms(&(p->fs), prefix);
 321
 322                 q = (struct dn_flow_queue *)(p+1);
 323                 list_queues(&(p->fs), q);
 324         }
 325         for (fs = next; nbytes >= sizeof *fs; fs = next) {
 326                 char prefix[80];
 327
 328                 if (SLIST_NEXT(fs, next) != (struct dn_flow_set *)DN_IS_QUEUE)
 329                         break;
 330                 l = sizeof(*fs) + fs->rq_elements * sizeof(*q);
 331                 next = (char *)fs + l;
 332                 nbytes -= l;
 333
 334                 if (rulenum != 0 && ((rulenum != fs->fs_nr && co.do_pipe == 2) ||
 335                     (rulenum != fs->parent_nr && co.do_pipe == 1))) {
 336                         continue;
 337                 }
 338
 339                 q = (struct dn_flow_queue *)(fs+1);
 340                 sprintf(prefix, "q%05d: weight %d pipe %d ",
 341                     fs->fs_nr, fs->weight, fs->parent_nr);
 342                 print_flowset_parms(fs, prefix);
 343                 list_queues(fs, q);
 344         }
 345 }
 346
 347 /*
 348  * Delete pipe or queue i
 349  */
 350 int
 351 ipfw_delete_pipe(int pipe_or_queue, int i)
 352 {
 353         struct dn_pipe p;
 354
 355         memset(&p, 0, sizeof p);
 356         if (pipe_or_queue == 1)
 357                 p.pipe_nr = i;          /* pipe */
 358         else
 359                 p.fs.fs_nr = i;         /* queue */
 360         i = do_cmd(IP_DUMMYNET_DEL, &p, sizeof p);
 361         if (i) {
 362                 i = 1;
 363                 warn("rule %u: setsockopt(IP_DUMMYNET_DEL)", i);
 364         }
 365         return i;
 366 }
 367
 368 /*
 369  * Code to parse delay profiles.
 370  *
 371  * Some link types introduce extra delays in the transmission
 372  * of a packet, e.g. because of MAC level framing, contention on
 373  * the use of the channel, MAC level retransmissions and so on.
 374  * From our point of view, the channel is effectively unavailable
 375  * for this extra time, which is constant or variable depending
 376  * on the link type. Additionally, packets may be dropped after this
 377  * time (e.g. on a wireless link after too many retransmissions).
 378  * We can model the additional delay with an empirical curve
 379  * that represents its distribution.
 380  *
 381  *      cumulative probability
 382  *      1.0 ^
 383  *          |
 384  *      L   +-- loss-level          x
 385  *          |                 ******
 386  *          |                *
 387  *          |           *****
 388  *          |          *
 389  *          |        **
 390  *          |       *
 391  *          +-------*------------------->
 392  *                      delay
 393  *
 394  * The empirical curve may have both vertical and horizontal lines.
 395  * Vertical lines represent constant delay for a range of
 396  * probabilities; horizontal lines correspond to a discontinuty
 397  * in the delay distribution: the pipe will use the largest delay
 398  * for a given probability.
 399  *
 400  * To pass the curve to dummynet, we must store the parameters
 401  * in a file as described below, and issue the command
 402  *
 403  *      ipfw pipe <n> config ... bw XXX profile <filename> ...
 404  *
 405  * The file format is the following, with whitespace acting as
 406  * a separator and '#' indicating the beginning a comment:
 407  *
 408  *      samples N
 409  *              the number of samples used in the internal
 410  *              representation (2..1024; default 100);
 411  *
 412  *      loss-level L
 413  *              The probability above which packets are lost.
 414  *               (0.0 <= L <= 1.0, default 1.0 i.e. no loss);
 415  *
 416  *      name identifier
 417  *              Optional a name (listed by "ipfw pipe show")
 418  *              to identify the distribution;
 419  *
 420  *      "delay prob" | "prob delay"
 421  *              One of these two lines is mandatory and defines
 422  *              the format of the following lines with data points.
 423  *
 424  *      XXX YYY
 425  *              2 or more lines representing points in the curve,
 426  *              with either delay or probability first, according
 427  *              to the chosen format.
 428  *              The unit for delay is milliseconds.
 429  *
 430  * Data points does not need to be ordered or equal to the number
 431  * specified in the "samples" line. ipfw will sort and interpolate
 432  * the curve as needed.
 433  *
 434  * Example of a profile file:
 435
 436         name    bla_bla_bla
 437         samples 100
 438         loss-level    0.86
 439         prob    delay
 440         0       200     # minimum overhead is 200ms
 441         0.5     200
 442         0.5     300
 443         0.8     1000
 444         0.9     1300
 445         1       1300
 446
 447  * Internally, we will convert the curve to a fixed number of
 448  * samples, and when it is time to transmit a packet we will
 449  * model the extra delay as extra bits in the packet.
 450  *
 451  */
 452
 453 /* XXX move to an array definition ? */
 454 #define ED_MAX_LINE_LEN 256+ED_MAX_NAME_LEN
 455 #define ED_TOK_SAMPLES  "samples"
 456 #define ED_TOK_LOSS     "loss-level"
 457 #define ED_TOK_NAME     "name"
 458 #define ED_TOK_DELAY    "delay"
 459 #define ED_TOK_PROB     "prob"
 460 #define ED_TOK_BW       "bw"
 461 #define ED_SEPARATORS   " \t\n"
 462 #define ED_MIN_SAMPLES_NO       2
 463
 464 /*
 465  * returns 1 if s is a non-negative number, with at least one '.'
 466  */
 467 static int
 468 is_valid_number(const char *s)
 469 {
 470         int i, dots_found = 0;
 471         int len = strlen(s);
 472
 473         for (i = 0; i<len; ++i)
 474                 if (!isdigit(s[i]) && (s[i] !='.' || ++dots_found > 1))
 475                         return 0;
 476         return 1;
 477 }
 478
 479 /*
 480  * Take as input a string describing a bandwidth value
 481  * and return the numeric bandwidth value.
 482  * set clocking interface or bandwidth value
 483  */
 484 static void
 485 read_bandwidth(char *arg, int *bandwidth, char *if_name, int namelen)
 486 {
 487         if (*bandwidth != -1)
 488                 warn("duplicate token, override bandwidth value!");
 489
 490         if (arg[0] >= 'a' && arg[0] <= 'z') {
 491                 if (namelen >= IFNAMSIZ)
 492                         warn("interface name truncated");
 493                 namelen--;
 494                 /* interface name */
 495                 strncpy(if_name, arg, namelen);
 496                 if_name[namelen] = '\0';
 497                 *bandwidth = 0;
 498         } else {        /* read bandwidth value */
 499                 int bw;
 500                 char *end = NULL;
 501
 502                 bw = strtoul(arg, &end, 0);
 503                 if (*end == 'K' || *end == 'k') {
 504                         end++;
 505                         bw *= 1000;
 506                 } else if (*end == 'M') {
 507                         end++;
 508                         bw *= 1000000;
 509                 }
 510                 if ((*end == 'B' &&
 511                     _substrcmp2(end, "Bi", "Bit/s") != 0) ||
 512                     _substrcmp2(end, "by", "bytes") == 0)
 513                         bw *= 8;
 514
 515                 if (bw < 0)
 516                         errx(EX_DATAERR, "bandwidth too large");
 517
 518                 *bandwidth = bw;
 519                 if_name[0] = '\0';
 520         }
 521 }
 522
 523 struct point {
 524         double prob;
 525         double delay;
 526 };
 527
 528 static int
 529 compare_points(const void *vp1, const void *vp2)
 530 {
 531         const struct point *p1 = vp1;
 532         const struct point *p2 = vp2;
 533         double res = 0;
 534
 535         res = p1->prob - p2->prob;
 536         if (res == 0)
 537                 res = p1->delay - p2->delay;
 538         if (res < 0)
 539                 return -1;
 540         else if (res > 0)
 541                 return 1;
 542         else
 543                 return 0;
 544 }
 545
 546 #define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno
 547
 548 static void
 549 load_extra_delays(const char *filename, struct dn_pipe *p)
 550 {
 551         char    line[ED_MAX_LINE_LEN];
 552         FILE    *f;
 553         int     lineno = 0;
 554         int     i;
 555
 556         int     samples = -1;
 557         double  loss = -1.0;
 558         char    profile_name[ED_MAX_NAME_LEN];
 559         int     delay_first = -1;
 560         int     do_points = 0;
 561         struct point    points[ED_MAX_SAMPLES_NO];
 562         int     points_no = 0;
 563
 564         profile_name[0] = '\0';
 565         f = fopen(filename, "r");
 566         if (f == NULL)
 567                 err(EX_UNAVAILABLE, "fopen: %s", filename);
 568
 569         while (fgets(line, ED_MAX_LINE_LEN, f)) {         /* read commands */
 570                 char *s, *cur = line, *name = NULL, *arg = NULL;
 571
 572                 ++lineno;
 573
 574                 /* parse the line */
 575                 while (cur) {
 576                         s = strsep(&cur, ED_SEPARATORS);
 577                         if (s == NULL || *s == '#')
 578                                 break;
 579                         if (*s == '\0')
 580                                 continue;
 581                         if (arg)
 582                                 errx(ED_EFMT("too many arguments"));
 583                         if (name == NULL)
 584                                 name = s;
 585                         else
 586                                 arg = s;
 587                 }
 588                 if (name == NULL)       /* empty line */
 589                         continue;
 590                 if (arg == NULL)
 591                         errx(ED_EFMT("missing arg for %s"), name);
 592
 593                 if (!strcasecmp(name, ED_TOK_SAMPLES)) {
 594                     if (samples > 0)
 595                         errx(ED_EFMT("duplicate ``samples'' line"));
 596                     if (atoi(arg) <=0)
 597                         errx(ED_EFMT("invalid number of samples"));
 598                     samples = atoi(arg);
 599                     if (samples>ED_MAX_SAMPLES_NO)
 600                             errx(ED_EFMT("too many samples, maximum is %d"),
 601                                 ED_MAX_SAMPLES_NO);
 602                     do_points = 0;
 603                 } else if (!strcasecmp(name, ED_TOK_BW)) {
 604                     read_bandwidth(arg, &p->bandwidth, p->if_name, sizeof(p->if_name));
 605                 } else if (!strcasecmp(name, ED_TOK_LOSS)) {
 606                     if (loss != -1.0)
 607                         errx(ED_EFMT("duplicated token: %s"), name);
 608                     if (!is_valid_number(arg))
 609                         errx(ED_EFMT("invalid %s"), arg);
 610                     loss = atof(arg);
 611                     if (loss > 1)
 612                         errx(ED_EFMT("%s greater than 1.0"), name);
 613                     do_points = 0;
 614                 } else if (!strcasecmp(name, ED_TOK_NAME)) {
 615                     if (profile_name[0] != '\0')
 616                         errx(ED_EFMT("duplicated token: %s"), name);
 617                     strncpy(profile_name, arg, sizeof(profile_name) - 1);
 618                     profile_name[sizeof(profile_name)-1] = '\0';
 619                     do_points = 0;
 620                 } else if (!strcasecmp(name, ED_TOK_DELAY)) {
 621                     if (do_points)
 622                         errx(ED_EFMT("duplicated token: %s"), name);
 623                     delay_first = 1;
 624                     do_points = 1;
 625                 } else if (!strcasecmp(name, ED_TOK_PROB)) {
 626                     if (do_points)
 627                         errx(ED_EFMT("duplicated token: %s"), name);
 628                     delay_first = 0;
 629                     do_points = 1;
 630                 } else if (do_points) {
 631                     if (!is_valid_number(name) || !is_valid_number(arg))
 632                         errx(ED_EFMT("invalid point found"));
 633                     if (delay_first) {
 634                         points[points_no].delay = atof(name);
 635                         points[points_no].prob = atof(arg);
 636                     } else {
 637                         points[points_no].delay = atof(arg);
 638                         points[points_no].prob = atof(name);
 639                     }
 640                     if (points[points_no].prob > 1.0)
 641                         errx(ED_EFMT("probability greater than 1.0"));
 642                     ++points_no;
 643                 } else {
 644                     errx(ED_EFMT("unrecognised command '%s'"), name);
 645                 }
 646         }
 647
 648         if (samples == -1) {
 649             warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES);
 650             samples = 100;
 651         }
 652
 653         if (loss == -1.0) {
 654             warnx("'%s' not found, assuming no loss", ED_TOK_LOSS);
 655             loss = 1;
 656         }
 657
 658         /* make sure that there are enough points. */
 659         if (points_no < ED_MIN_SAMPLES_NO)
 660             errx(ED_EFMT("too few samples, need at least %d"),
 661                 ED_MIN_SAMPLES_NO);
 662
 663         qsort(points, points_no, sizeof(struct point), compare_points);
 664
 665         /* interpolation */
 666         for (i = 0; i<points_no-1; ++i) {
 667             double y1 = points[i].prob * samples;
 668             double x1 = points[i].delay;
 669             double y2 = points[i+1].prob * samples;
 670             double x2 = points[i+1].delay;
 671
 672             int index = y1;
 673             int stop = y2;
 674
 675             if (x1 == x2) {
 676                 for (; index<stop; ++index)
 677                     p->samples[index] = x1;
 678             } else {
 679                 double m = (y2-y1)/(x2-x1);
 680                 double c = y1 - m*x1;
 681                 for (; index<stop ; ++index)
 682                     p->samples[index] = (index - c)/m;
 683             }
 684         }
 685         p->samples_no = samples;
 686         p->loss_level = loss * samples;
 687         strncpy(p->name, profile_name, sizeof(p->name));
 688 }
 689
 690 void
 691 ipfw_config_pipe(int ac, char **av)
 692 {
 693         int samples[ED_MAX_SAMPLES_NO];
 694         struct dn_pipe p;
 695         int i;
 696         char *end;
 697         void *par = NULL;
 698
 699         memset(&p, 0, sizeof p);
 700         p.bandwidth = -1;
 701
 702         av++; ac--;
 703         /* Pipe number */
 704         if (ac && isdigit(**av)) {
 705                 i = atoi(*av); av++; ac--;
 706                 if (co.do_pipe == 1)
 707                         p.pipe_nr = i;
 708                 else
 709                         p.fs.fs_nr = i;
 710         }
 711         while (ac > 0) {
 712                 double d;
 713                 int tok = match_token(dummynet_params, *av);
 714                 ac--; av++;
 715
 716                 switch(tok) {
 717                 case TOK_NOERROR:
 718                         p.fs.flags_fs |= DN_NOERROR;
 719                         break;
 720
 721                 case TOK_PLR:
 722                         NEED1("plr needs argument 0..1\n");
 723                         d = strtod(av[0], NULL);
 724                         if (d > 1)
 725                                 d = 1;
 726                         else if (d < 0)
 727                                 d = 0;
 728                         p.fs.plr = (int)(d*0x7fffffff);
 729                         ac--; av++;
 730                         break;
 731
 732                 case TOK_QUEUE:
 733                         NEED1("queue needs queue size\n");
 734                         end = NULL;
 735                         p.fs.qsize = strtoul(av[0], &end, 0);
 736                         if (*end == 'K' || *end == 'k') {
 737                                 p.fs.flags_fs |= DN_QSIZE_IS_BYTES;
 738                                 p.fs.qsize *= 1024;
 739                         } else if (*end == 'B' ||
 740                             _substrcmp2(end, "by", "bytes") == 0) {
 741                                 p.fs.flags_fs |= DN_QSIZE_IS_BYTES;
 742                         }
 743                         ac--; av++;
 744                         break;
 745
 746                 case TOK_BUCKETS:
 747                         NEED1("buckets needs argument\n");
 748                         p.fs.rq_size = strtoul(av[0], NULL, 0);
 749                         ac--; av++;
 750                         break;
 751
 752                 case TOK_MASK:
 753                         NEED1("mask needs mask specifier\n");
 754                         /*
 755                          * per-flow queue, mask is dst_ip, dst_port,
 756                          * src_ip, src_port, proto measured in bits
 757                          */
 758                         par = NULL;
 759
 760                         bzero(&p.fs.flow_mask, sizeof(p.fs.flow_mask));
 761                         end = NULL;
 762
 763                         while (ac >= 1) {
 764                             uint32_t *p32 = NULL;
 765                             uint16_t *p16 = NULL;
 766                             uint32_t *p20 = NULL;
 767                             struct in6_addr *pa6 = NULL;
 768                             uint32_t a;
 769
 770                             tok = match_token(dummynet_params, *av);
 771                             ac--; av++;
 772                             switch(tok) {
 773                             case TOK_ALL:
 774                                     /*
 775                                      * special case, all bits significant
 776                                      */
 777                                     p.fs.flow_mask.dst_ip = ~0;
 778                                     p.fs.flow_mask.src_ip = ~0;
 779                                     p.fs.flow_mask.dst_port = ~0;
 780                                     p.fs.flow_mask.src_port = ~0;
 781                                     p.fs.flow_mask.proto = ~0;
 782                                     n2mask(&(p.fs.flow_mask.dst_ip6), 128);
 783                                     n2mask(&(p.fs.flow_mask.src_ip6), 128);
 784                                     p.fs.flow_mask.flow_id6 = ~0;
 785                                     p.fs.flags_fs |= DN_HAVE_FLOW_MASK;
 786                                     goto end_mask;
 787
 788                             case TOK_DSTIP:
 789                                     p32 = &p.fs.flow_mask.dst_ip;
 790                                     break;
 791
 792                             case TOK_SRCIP:
 793                                     p32 = &p.fs.flow_mask.src_ip;
 794                                     break;
 795
 796                             case TOK_DSTIP6:
 797                                     pa6 = &(p.fs.flow_mask.dst_ip6);
 798                                     break;
 799
 800                             case TOK_SRCIP6:
 801                                     pa6 = &(p.fs.flow_mask.src_ip6);
 802                                     break;
 803
 804                             case TOK_FLOWID:
 805                                     p20 = &p.fs.flow_mask.flow_id6;
 806                                     break;
 807
 808                             case TOK_DSTPORT:
 809                                     p16 = &p.fs.flow_mask.dst_port;
 810                                     break;
 811
 812                             case TOK_SRCPORT:
 813                                     p16 = &p.fs.flow_mask.src_port;
 814                                     break;
 815
 816                             case TOK_PROTO:
 817                                     break;
 818
 819                             default:
 820                                     ac++; av--; /* backtrack */
 821                                     goto end_mask;
 822                             }
 823                             if (ac < 1)
 824                                     errx(EX_USAGE, "mask: value missing");
 825                             if (*av[0] == '/') {
 826                                     a = strtoul(av[0]+1, &end, 0);
 827                                     if (pa6 == NULL)
 828                                             a = (a == 32) ? ~0 : (1 << a) - 1;
 829                             } else
 830                                     a = strtoul(av[0], &end, 0);
 831                             if (p32 != NULL)
 832                                     *p32 = a;
 833                             else if (p16 != NULL) {
 834                                     if (a > 0xFFFF)
 835                                             errx(EX_DATAERR,
 836                                                 "port mask must be 16 bit");
 837                                     *p16 = (uint16_t)a;
 838                             } else if (p20 != NULL) {
 839                                     if (a > 0xfffff)
 840                                         errx(EX_DATAERR,
 841                                             "flow_id mask must be 20 bit");
 842                                     *p20 = (uint32_t)a;
 843                             } else if (pa6 != NULL) {
 844                                     if (a > 128)
 845                                         errx(EX_DATAERR,
 846                                             "in6addr invalid mask len");
 847                                     else
 848                                         n2mask(pa6, a);
 849                             } else {
 850                                     if (a > 0xFF)
 851                                             errx(EX_DATAERR,
 852                                                 "proto mask must be 8 bit");
 853                                     p.fs.flow_mask.proto = (uint8_t)a;
 854                             }
 855                             if (a != 0)
 856                                     p.fs.flags_fs |= DN_HAVE_FLOW_MASK;
 857                             ac--; av++;
 858                         } /* end while, config masks */
 859 end_mask:
 860                         break;
 861
 862                 case TOK_RED:
 863                 case TOK_GRED:
 864                         NEED1("red/gred needs w_q/min_th/max_th/max_p\n");
 865                         p.fs.flags_fs |= DN_IS_RED;
 866                         if (tok == TOK_GRED)
 867                                 p.fs.flags_fs |= DN_IS_GENTLE_RED;
 868                         /*
 869                          * the format for parameters is w_q/min_th/max_th/max_p
 870                          */
 871                         if ((end = strsep(&av[0], "/"))) {
 872                             double w_q = strtod(end, NULL);
 873                             if (w_q > 1 || w_q <= 0)
 874                                 errx(EX_DATAERR, "0 < w_q <= 1");
 875                             p.fs.w_q = (int) (w_q * (1 << SCALE_RED));
 876                         }
 877                         if ((end = strsep(&av[0], "/"))) {
 878                             p.fs.min_th = strtoul(end, &end, 0);
 879                             if (*end == 'K' || *end == 'k')
 880                                 p.fs.min_th *= 1024;
 881                         }
 882                         if ((end = strsep(&av[0], "/"))) {
 883                             p.fs.max_th = strtoul(end, &end, 0);
 884                             if (*end == 'K' || *end == 'k')
 885                                 p.fs.max_th *= 1024;
 886                         }
 887                         if ((end = strsep(&av[0], "/"))) {
 888                             double max_p = strtod(end, NULL);
 889                             if (max_p > 1 || max_p <= 0)
 890                                 errx(EX_DATAERR, "0 < max_p <= 1");
 891                             p.fs.max_p = (int)(max_p * (1 << SCALE_RED));
 892                         }
 893                         ac--; av++;
 894                         break;
 895
 896                 case TOK_DROPTAIL:
 897                         p.fs.flags_fs &= ~(DN_IS_RED|DN_IS_GENTLE_RED);
 898                         break;
 899
 900                 case TOK_BW:
 901                         NEED1("bw needs bandwidth or interface\n");
 902                         if (co.do_pipe != 1)
 903                             errx(EX_DATAERR, "bandwidth only valid for pipes");
 904                         read_bandwidth(av[0], &p.bandwidth, p.if_name, sizeof(p.if_name));
 905                         ac--; av++;
 906                         break;
 907
 908                 case TOK_DELAY:
 909                         if (co.do_pipe != 1)
 910                                 errx(EX_DATAERR, "delay only valid for pipes");
 911                         NEED1("delay needs argument 0..10000ms\n");
 912                         p.delay = strtoul(av[0], NULL, 0);
 913                         ac--; av++;
 914                         break;
 915
 916                 case TOK_WEIGHT:
 917                         if (co.do_pipe == 1)
 918                                 errx(EX_DATAERR,"weight only valid for queues");
 919                         NEED1("weight needs argument 0..100\n");
 920                         p.fs.weight = strtoul(av[0], &end, 0);
 921                         ac--; av++;
 922                         break;
 923
 924                 case TOK_PIPE:
 925                         if (co.do_pipe == 1)
 926                                 errx(EX_DATAERR,"pipe only valid for queues");
 927                         NEED1("pipe needs pipe_number\n");
 928                         p.fs.parent_nr = strtoul(av[0], &end, 0);
 929                         ac--; av++;
 930                         break;
 931
 932                 case TOK_PIPE_PROFILE:
 933                         if (co.do_pipe != 1)
 934                             errx(EX_DATAERR, "extra delay only valid for pipes");
 935                         NEED1("extra delay needs the file name\n");
 936                         p.samples = &samples[0];
 937                         load_extra_delays(av[0], &p);
 938                         --ac; ++av;
 939                         break;
 940
 941                 default:
 942                         errx(EX_DATAERR, "unrecognised option ``%s''", av[-1]);
 943                 }
 944         }
 945         if (co.do_pipe == 1) {
 946                 if (p.pipe_nr == 0)
 947                         errx(EX_DATAERR, "pipe_nr must be > 0");
 948                 if (p.delay > 10000)
 949                         errx(EX_DATAERR, "delay must be < 10000");
 950         } else { /* co.do_pipe == 2, queue */
 951                 if (p.fs.parent_nr == 0)
 952                         errx(EX_DATAERR, "pipe must be > 0");
 953                 if (p.fs.weight >100)
 954                         errx(EX_DATAERR, "weight must be <= 100");
 955         }
 956
 957         /* check for bandwidth value */
 958         if (p.bandwidth == -1) {
 959                 p.bandwidth = 0;
 960                 if (p.samples_no > 0)
 961                         errx(EX_DATAERR, "profile requires a bandwidth limit");
 962         }
 963
 964         if (p.fs.flags_fs & DN_QSIZE_IS_BYTES) {
 965                 size_t len;
 966                 long limit;
 967
 968                 len = sizeof(limit);
 969                 if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit",
 970                         &limit, &len, NULL, 0) == -1)
 971                         limit = 1024*1024;
 972                 if (p.fs.qsize > limit)
 973                         errx(EX_DATAERR, "queue size must be < %ldB", limit);
 974         } else {
 975                 size_t len;
 976                 long limit;
 977
 978                 len = sizeof(limit);
 979                 if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit",
 980                         &limit, &len, NULL, 0) == -1)
 981                         limit = 100;
 982                 if (p.fs.qsize > limit)
 983                         errx(EX_DATAERR, "2 <= queue size <= %ld", limit);
 984         }
 985         if (p.fs.flags_fs & DN_IS_RED) {
 986                 size_t len;
 987                 int lookup_depth, avg_pkt_size;
 988                 double s, idle, weight, w_q;
 989                 struct clockinfo ck;
 990                 int t;
 991
 992                 if (p.fs.min_th >= p.fs.max_th)
 993                     errx(EX_DATAERR, "min_th %d must be < than max_th %d",
 994                         p.fs.min_th, p.fs.max_th);
 995                 if (p.fs.max_th == 0)
 996                     errx(EX_DATAERR, "max_th must be > 0");
 997
 998                 len = sizeof(int);
 999                 if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth",
1000                         &lookup_depth, &len, NULL, 0) == -1)
1001                     errx(1, "sysctlbyname(\"%s\")",
1002                         "net.inet.ip.dummynet.red_lookup_depth");
1003                 if (lookup_depth == 0)
1004                     errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth"
1005                         " must be greater than zero");
1006
1007                 len = sizeof(int);
1008                 if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size",
1009                         &avg_pkt_size, &len, NULL, 0) == -1)
1010
1011                     errx(1, "sysctlbyname(\"%s\")",
1012                         "net.inet.ip.dummynet.red_avg_pkt_size");
1013                 if (avg_pkt_size == 0)
1014                         errx(EX_DATAERR,
1015                             "net.inet.ip.dummynet.red_avg_pkt_size must"
1016                             " be greater than zero");
1017
1018                 len = sizeof(struct clockinfo);
1019                 if (sysctlbyname("kern.clockrate", &ck, &len, NULL, 0) == -1)
1020                         errx(1, "sysctlbyname(\"%s\")", "kern.clockrate");
1021
1022                 /*
1023                  * Ticks needed for sending a medium-sized packet.
1024                  * Unfortunately, when we are configuring a WF2Q+ queue, we
1025                  * do not have bandwidth information, because that is stored
1026                  * in the parent pipe, and also we have multiple queues
1027                  * competing for it. So we set s=0, which is not very
1028                  * correct. But on the other hand, why do we want RED with
1029                  * WF2Q+ ?
1030                  */
1031                 if (p.bandwidth==0) /* this is a WF2Q+ queue */
1032                         s = 0;
1033                 else
1034                         s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth;
1035
1036                 /*
1037                  * max idle time (in ticks) before avg queue size becomes 0.
1038                  * NOTA:  (3/w_q) is approx the value x so that
1039                  * (1-w_q)^x < 10^-3.
1040                  */
1041                 w_q = ((double)p.fs.w_q) / (1 << SCALE_RED);
1042                 idle = s * 3. / w_q;
1043                 p.fs.lookup_step = (int)idle / lookup_depth;
1044                 if (!p.fs.lookup_step)
1045                         p.fs.lookup_step = 1;
1046                 weight = 1 - w_q;
1047                 for (t = p.fs.lookup_step; t > 1; --t)
1048                         weight *= 1 - w_q;
1049                 p.fs.lookup_weight = (int)(weight * (1 << SCALE_RED));
1050         }
1051         if (p.samples_no <= 0) {
1052         i = do_cmd(IP_DUMMYNET_CONFIGURE, &p, sizeof p);
1053         } else {
1054                 struct dn_pipe_max pm;
1055                 int len = sizeof(pm);
1056
1057                 memcpy(&pm.pipe, &p, sizeof(pm.pipe));
1058                 memcpy(&pm.samples, samples, sizeof(pm.samples));
1059
1060                 i = do_cmd(IP_DUMMYNET_CONFIGURE, &pm, len);
1061         }
1062
1063         if (i)
1064                 err(1, "setsockopt(%s)", "IP_DUMMYNET_CONFIGURE");
1065 }