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 <errno.h>
  36 #include <netdb.h>
  37 #include <stdio.h>
  38 #include <stdlib.h>
  39 #include <string.h>
  40 #include <sysexits.h>
  41
  42 #include <net/if.h>
  43 #include <netinet/in.h>
  44 #include <netinet/ip_fw.h>
  45 #include <netinet/ip_dummynet.h>
  46 #include <arpa/inet.h>  /* inet_ntoa */
  47
  48 static struct _s_x dummynet_params[] = {
  49         { "plr",                TOK_PLR },
  50         { "noerror",            TOK_NOERROR },
  51         { "buckets",            TOK_BUCKETS },
  52         { "dst-ip",             TOK_DSTIP },
  53         { "src-ip",             TOK_SRCIP },
  54         { "dst-port",           TOK_DSTPORT },
  55         { "src-port",           TOK_SRCPORT },
  56         { "proto",              TOK_PROTO },
  57         { "weight",             TOK_WEIGHT },
  58         { "all",                TOK_ALL },
  59         { "mask",               TOK_MASK },
  60         { "droptail",           TOK_DROPTAIL },
  61         { "red",                TOK_RED },
  62         { "gred",               TOK_GRED },
  63         { "bw",                 TOK_BW },
  64         { "bandwidth",          TOK_BW },
  65         { "delay",              TOK_DELAY },
  66         { "pipe",               TOK_PIPE },
  67         { "queue",              TOK_QUEUE },
  68         { "flow-id",            TOK_FLOWID},
  69         { "dst-ipv6",           TOK_DSTIP6},
  70         { "dst-ip6",            TOK_DSTIP6},
  71         { "src-ipv6",           TOK_SRCIP6},
  72         { "src-ip6",            TOK_SRCIP6},
  73         { "profile",            TOK_PIPE_PROFILE},
  74         { "burst",              TOK_BURST},
  75         { "dummynet-params",    TOK_NULL },
  76         { NULL, 0 }     /* terminator */
  77 };
  78
  79 static int
  80 sort_q(void *arg, const void *pa, const void *pb)
  81 {
  82         int rev = (co.do_sort < 0);
  83         int field = rev ? -co.do_sort : co.do_sort;
  84         long long res = 0;
  85         const struct dn_flow_queue *a = pa;
  86         const struct dn_flow_queue *b = pb;
  87
  88         switch (field) {
  89         case 1: /* pkts */
  90                 res = a->len - b->len;
  91                 break;
  92         case 2: /* bytes */
  93                 res = a->len_bytes - b->len_bytes;
  94                 break;
  95
  96         case 3: /* tot pkts */
  97                 res = a->tot_pkts - b->tot_pkts;
  98                 break;
  99
 100         case 4: /* tot bytes */
 101                 res = a->tot_bytes - b->tot_bytes;
 102                 break;
 103         }
 104         if (res < 0)
 105                 res = -1;
 106         if (res > 0)
 107                 res = 1;
 108         return (int)(rev ? res : -res);
 109 }
 110
 111 static void
 112 list_queues(struct dn_flow_set *fs, struct dn_flow_queue *q)
 113 {
 114         int l;
 115         int index_printed, indexes = 0;
 116         char buff[255];
 117         struct protoent *pe;
 118
 119         if (fs->rq_elements == 0)
 120                 return;
 121
 122         if (co.do_sort != 0)
 123                 qsort_r(q, fs->rq_elements, sizeof *q, NULL, sort_q);
 124
 125         /* Print IPv4 flows */
 126         index_printed = 0;
 127         for (l = 0; l < fs->rq_elements; l++) {
 128                 struct in_addr ina;
 129
 130                 /* XXX: Should check for IPv4 flows */
 131                 if (IS_IP6_FLOW_ID(&(q[l].id)))
 132                         continue;
 133
 134                 if (!index_printed) {
 135                         index_printed = 1;
 136                         if (indexes > 0)        /* currently a no-op */
 137                                 printf("\n");
 138                         indexes++;
 139                         printf("    "
 140                             "mask: 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n",
 141                             fs->flow_mask.proto,
 142                             fs->flow_mask.src_ip, fs->flow_mask.src_port,
 143                             fs->flow_mask.dst_ip, fs->flow_mask.dst_port);
 144
 145                         printf("BKT Prot ___Source IP/port____ "
 146                             "____Dest. IP/port____ "
 147                             "Tot_pkt/bytes Pkt/Byte Drp\n");
 148                 }
 149
 150                 printf("%3d ", q[l].hash_slot);
 151                 pe = getprotobynumber(q[l].id.proto);
 152                 if (pe)
 153                         printf("%-4s ", pe->p_name);
 154                 else
 155                         printf("%4u ", q[l].id.proto);
 156                 ina.s_addr = htonl(q[l].id.src_ip);
 157                 printf("%15s/%-5d ",
 158                     inet_ntoa(ina), q[l].id.src_port);
 159                 ina.s_addr = htonl(q[l].id.dst_ip);
 160                 printf("%15s/%-5d ",
 161                     inet_ntoa(ina), q[l].id.dst_port);
 162                 printf("%4llu %8llu %2u %4u %3u\n",
 163                     align_uint64(&q[l].tot_pkts),
 164                     align_uint64(&q[l].tot_bytes),
 165                     q[l].len, q[l].len_bytes, q[l].drops);
 166                 if (co.verbose)
 167                         printf("   S %20llu  F %20llu\n",
 168                             align_uint64(&q[l].S), align_uint64(&q[l].F));
 169         }
 170
 171         /* Print IPv6 flows */
 172         index_printed = 0;
 173         for (l = 0; l < fs->rq_elements; l++) {
 174                 if (!IS_IP6_FLOW_ID(&(q[l].id)))
 175                         continue;
 176
 177                 if (!index_printed) {
 178                         index_printed = 1;
 179                         if (indexes > 0)
 180                                 printf("\n");
 181                         indexes++;
 182                         printf("\n        mask: proto: 0x%02x, flow_id: 0x%08x,  ",
 183                             fs->flow_mask.proto, fs->flow_mask.flow_id6);
 184                         inet_ntop(AF_INET6, &(fs->flow_mask.src_ip6),
 185                             buff, sizeof(buff));
 186                         printf("%s/0x%04x -> ", buff, fs->flow_mask.src_port);
 187                         inet_ntop( AF_INET6, &(fs->flow_mask.dst_ip6),
 188                             buff, sizeof(buff) );
 189                         printf("%s/0x%04x\n", buff, fs->flow_mask.dst_port);
 190
 191                         printf("BKT ___Prot___ _flow-id_ "
 192                             "______________Source IPv6/port_______________ "
 193                             "_______________Dest. IPv6/port_______________ "
 194                             "Tot_pkt/bytes Pkt/Byte Drp\n");
 195                 }
 196                 printf("%3d ", q[l].hash_slot);
 197                 pe = getprotobynumber(q[l].id.proto);
 198                 if (pe != NULL)
 199                         printf("%9s ", pe->p_name);
 200                 else
 201                         printf("%9u ", q[l].id.proto);
 202                 printf("%7d  %39s/%-5d ", q[l].id.flow_id6,
 203                     inet_ntop(AF_INET6, &(q[l].id.src_ip6), buff, sizeof(buff)),
 204                     q[l].id.src_port);
 205                 printf(" %39s/%-5d ",
 206                     inet_ntop(AF_INET6, &(q[l].id.dst_ip6), buff, sizeof(buff)),
 207                     q[l].id.dst_port);
 208                 printf(" %4llu %8llu %2u %4u %3u\n",
 209                     align_uint64(&q[l].tot_pkts),
 210                     align_uint64(&q[l].tot_bytes),
 211                     q[l].len, q[l].len_bytes, q[l].drops);
 212                 if (co.verbose)
 213                         printf("   S %20llu  F %20llu\n",
 214                             align_uint64(&q[l].S),
 215                             align_uint64(&q[l].F));
 216         }
 217 }
 218
 219 static void
 220 print_flowset_parms(struct dn_flow_set *fs, char *prefix)
 221 {
 222         int l;
 223         char qs[30];
 224         char plr[30];
 225         char red[90];   /* Display RED parameters */
 226
 227         l = fs->qsize;
 228         if (fs->flags_fs & DN_QSIZE_IS_BYTES) {
 229                 if (l >= 8192)
 230                         sprintf(qs, "%d KB", l / 1024);
 231                 else
 232                         sprintf(qs, "%d B", l);
 233         } else
 234                 sprintf(qs, "%3d sl.", l);
 235         if (fs->plr)
 236                 sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff));
 237         else
 238                 plr[0] = '\0';
 239         if (fs->flags_fs & DN_IS_RED)   /* RED parameters */
 240                 sprintf(red,
 241                     "\n\t  %cRED w_q %f min_th %d max_th %d max_p %f",
 242                     (fs->flags_fs & DN_IS_GENTLE_RED) ? 'G' : ' ',
 243                     1.0 * fs->w_q / (double)(1 << SCALE_RED),
 244                     SCALE_VAL(fs->min_th),
 245                     SCALE_VAL(fs->max_th),
 246                     1.0 * fs->max_p / (double)(1 << SCALE_RED));
 247         else
 248                 sprintf(red, "droptail");
 249
 250         printf("%s %s%s %d queues (%d buckets) %s\n",
 251             prefix, qs, plr, fs->rq_elements, fs->rq_size, red);
 252 }
 253
 254 static void
 255 print_extra_delay_parms(struct dn_pipe *p)
 256 {
 257         double loss;
 258         if (p->samples_no <= 0)
 259                 return;
 260
 261         loss = p->loss_level;
 262         loss /= p->samples_no;
 263         printf("\t profile: name \"%s\" loss %f samples %d\n",
 264                 p->name, loss, p->samples_no);
 265 }
 266
 267 void
 268 ipfw_list_pipes(void *data, uint nbytes, int ac, char *av[])
 269 {
 270         int rulenum;
 271         void *next = data;
 272         struct dn_pipe *p = (struct dn_pipe *) data;
 273         struct dn_flow_set *fs;
 274         struct dn_flow_queue *q;
 275         int l;
 276
 277         if (ac > 0)
 278                 rulenum = strtoul(*av++, NULL, 10);
 279         else
 280                 rulenum = 0;
 281         for (; nbytes >= sizeof *p; p = (struct dn_pipe *)next) {
 282                 double b = p->bandwidth;
 283                 char buf[30];
 284                 char prefix[80];
 285                 char burst[5 + 7];
 286
 287                 if (SLIST_NEXT(p, next) != (struct dn_pipe *)DN_IS_PIPE)
 288                         break;  /* done with pipes, now queues */
 289
 290                 /*
 291                  * compute length, as pipe have variable size
 292                  */
 293                 l = sizeof(*p) + p->fs.rq_elements * sizeof(*q);
 294                 next = (char *)p + l;
 295                 nbytes -= l;
 296
 297                 if ((rulenum != 0 && rulenum != p->pipe_nr) || co.do_pipe == 2)
 298                         continue;
 299
 300                 /*
 301                  * Print rate (or clocking interface)
 302                  */
 303                 if (p->if_name[0] != '\0')
 304                         sprintf(buf, "%s", p->if_name);
 305                 else if (b == 0)
 306                         sprintf(buf, "unlimited");
 307                 else if (b >= 1000000)
 308                         sprintf(buf, "%7.3f Mbit/s", b/1000000);
 309                 else if (b >= 1000)
 310                         sprintf(buf, "%7.3f Kbit/s", b/1000);
 311                 else
 312                         sprintf(buf, "%7.3f bit/s ", b);
 313
 314                 sprintf(prefix, "%05d: %s %4d ms ",
 315                     p->pipe_nr, buf, p->delay);
 316
 317                 print_flowset_parms(&(p->fs), prefix);
 318
 319                 if (humanize_number(burst, sizeof(burst), p->burst,
 320                     "Byte", HN_AUTOSCALE, 0) < 0 || co.verbose)
 321                         printf("\t burst: %ju Byte\n", p->burst);
 322                 else
 323                         printf("\t burst: %s\n", burst);
 324
 325                 print_extra_delay_parms(p);
 326
 327                 q = (struct dn_flow_queue *)(p+1);
 328                 list_queues(&(p->fs), q);
 329         }
 330         for (fs = next; nbytes >= sizeof *fs; fs = next) {
 331                 char prefix[80];
 332
 333                 if (SLIST_NEXT(fs, next) != (struct dn_flow_set *)DN_IS_QUEUE)
 334                         break;
 335                 l = sizeof(*fs) + fs->rq_elements * sizeof(*q);
 336                 next = (char *)fs + l;
 337                 nbytes -= l;
 338
 339                 if (rulenum != 0 && ((rulenum != fs->fs_nr && co.do_pipe == 2) ||
 340                     (rulenum != fs->parent_nr && co.do_pipe == 1))) {
 341                         continue;
 342                 }
 343
 344                 q = (struct dn_flow_queue *)(fs+1);
 345                 sprintf(prefix, "q%05d: weight %d pipe %d ",
 346                     fs->fs_nr, fs->weight, fs->parent_nr);
 347                 print_flowset_parms(fs, prefix);
 348                 list_queues(fs, q);
 349         }
 350 }
 351
 352 /*
 353  * Delete pipe or queue i
 354  */
 355 int
 356 ipfw_delete_pipe(int pipe_or_queue, int i)
 357 {
 358         struct dn_pipe p;
 359
 360         memset(&p, 0, sizeof p);
 361         if (pipe_or_queue == 1)
 362                 p.pipe_nr = i;          /* pipe */
 363         else
 364                 p.fs.fs_nr = i;         /* queue */
 365         i = do_cmd(IP_DUMMYNET_DEL, &p, sizeof p);
 366         if (i) {
 367                 i = 1;
 368                 warn("rule %u: setsockopt(IP_DUMMYNET_DEL)", i);
 369         }
 370         return i;
 371 }
 372
 373 /*
 374  * Code to parse delay profiles.
 375  *
 376  * Some link types introduce extra delays in the transmission
 377  * of a packet, e.g. because of MAC level framing, contention on
 378  * the use of the channel, MAC level retransmissions and so on.
 379  * From our point of view, the channel is effectively unavailable
 380  * for this extra time, which is constant or variable depending
 381  * on the link type. Additionally, packets may be dropped after this
 382  * time (e.g. on a wireless link after too many retransmissions).
 383  * We can model the additional delay with an empirical curve
 384  * that represents its distribution.
 385  *
 386  *      cumulative probability
 387  *      1.0 ^
 388  *          |
 389  *      L   +-- loss-level          x
 390  *          |                 ******
 391  *          |                *
 392  *          |           *****
 393  *          |          *
 394  *          |        **
 395  *          |       *
 396  *          +-------*------------------->
 397  *                      delay
 398  *
 399  * The empirical curve may have both vertical and horizontal lines.
 400  * Vertical lines represent constant delay for a range of
 401  * probabilities; horizontal lines correspond to a discontinuty
 402  * in the delay distribution: the pipe will use the largest delay
 403  * for a given probability.
 404  *
 405  * To pass the curve to dummynet, we must store the parameters
 406  * in a file as described below, and issue the command
 407  *
 408  *      ipfw pipe <n> config ... bw XXX profile <filename> ...
 409  *
 410  * The file format is the following, with whitespace acting as
 411  * a separator and '#' indicating the beginning a comment:
 412  *
 413  *      samples N
 414  *              the number of samples used in the internal
 415  *              representation (2..1024; default 100);
 416  *
 417  *      loss-level L
 418  *              The probability above which packets are lost.
 419  *               (0.0 <= L <= 1.0, default 1.0 i.e. no loss);
 420  *
 421  *      name identifier
 422  *              Optional a name (listed by "ipfw pipe show")
 423  *              to identify the distribution;
 424  *
 425  *      "delay prob" | "prob delay"
 426  *              One of these two lines is mandatory and defines
 427  *              the format of the following lines with data points.
 428  *
 429  *      XXX YYY
 430  *              2 or more lines representing points in the curve,
 431  *              with either delay or probability first, according
 432  *              to the chosen format.
 433  *              The unit for delay is milliseconds.
 434  *
 435  * Data points does not need to be ordered or equal to the number
 436  * specified in the "samples" line. ipfw will sort and interpolate
 437  * the curve as needed.
 438  *
 439  * Example of a profile file:
 440
 441         name    bla_bla_bla
 442         samples 100
 443         loss-level    0.86
 444         prob    delay
 445         0       200     # minimum overhead is 200ms
 446         0.5     200
 447         0.5     300
 448         0.8     1000
 449         0.9     1300
 450         1       1300
 451
 452  * Internally, we will convert the curve to a fixed number of
 453  * samples, and when it is time to transmit a packet we will
 454  * model the extra delay as extra bits in the packet.
 455  *
 456  */
 457
 458 #define ED_MAX_LINE_LEN 256+ED_MAX_NAME_LEN
 459 #define ED_TOK_SAMPLES  "samples"
 460 #define ED_TOK_LOSS     "loss-level"
 461 #define ED_TOK_NAME     "name"
 462 #define ED_TOK_DELAY    "delay"
 463 #define ED_TOK_PROB     "prob"
 464 #define ED_TOK_BW       "bw"
 465 #define ED_SEPARATORS   " \t\n"
 466 #define ED_MIN_SAMPLES_NO       2
 467
 468 /*
 469  * returns 1 if s is a non-negative number, with at least one '.'
 470  */
 471 static int
 472 is_valid_number(const char *s)
 473 {
 474         int i, dots_found = 0;
 475         int len = strlen(s);
 476
 477         for (i = 0; i<len; ++i)
 478                 if (!isdigit(s[i]) && (s[i] !='.' || ++dots_found > 1))
 479                         return 0;
 480         return 1;
 481 }
 482
 483 /*
 484  * Take as input a string describing a bandwidth value
 485  * and return the numeric bandwidth value.
 486  * set clocking interface or bandwidth value
 487  */
 488 static void
 489 read_bandwidth(char *arg, int *bandwidth, char *if_name, int namelen)
 490 {
 491         if (*bandwidth != -1)
 492                 warn("duplicate token, override bandwidth value!");
 493
 494         if (arg[0] >= 'a' && arg[0] <= 'z') {
 495                 if (namelen >= IFNAMSIZ)
 496                         warn("interface name truncated");
 497                 namelen--;
 498                 /* interface name */
 499                 strncpy(if_name, arg, namelen);
 500                 if_name[namelen] = '\0';
 501                 *bandwidth = 0;
 502         } else {        /* read bandwidth value */
 503                 int bw;
 504                 char *end = NULL;
 505
 506                 bw = strtoul(arg, &end, 0);
 507                 if (*end == 'K' || *end == 'k') {
 508                         end++;
 509                         bw *= 1000;
 510                 } else if (*end == 'M') {
 511                         end++;
 512                         bw *= 1000000;
 513                 }
 514                 if ((*end == 'B' &&
 515                     _substrcmp2(end, "Bi", "Bit/s") != 0) ||
 516                     _substrcmp2(end, "by", "bytes") == 0)
 517                         bw *= 8;
 518
 519                 if (bw < 0)
 520                         errx(EX_DATAERR, "bandwidth too large");
 521
 522                 *bandwidth = bw;
 523                 if_name[0] = '\0';
 524         }
 525 }
 526
 527 struct point {
 528         double prob;
 529         double delay;
 530 };
 531
 532 static int
 533 compare_points(const void *vp1, const void *vp2)
 534 {
 535         const struct point *p1 = vp1;
 536         const struct point *p2 = vp2;
 537         double res = 0;
 538
 539         res = p1->prob - p2->prob;
 540         if (res == 0)
 541                 res = p1->delay - p2->delay;
 542         if (res < 0)
 543                 return -1;
 544         else if (res > 0)
 545                 return 1;
 546         else
 547                 return 0;
 548 }
 549
 550 #define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno
 551
 552 static void
 553 load_extra_delays(const char *filename, struct dn_pipe *p)
 554 {
 555         char    line[ED_MAX_LINE_LEN];
 556         FILE    *f;
 557         int     lineno = 0;
 558         int     i;
 559
 560         int     samples = -1;
 561         double  loss = -1.0;
 562         char    profile_name[ED_MAX_NAME_LEN];
 563         int     delay_first = -1;
 564         int     do_points = 0;
 565         struct point    points[ED_MAX_SAMPLES_NO];
 566         int     points_no = 0;
 567
 568         profile_name[0] = '\0';
 569         f = fopen(filename, "r");
 570         if (f == NULL)
 571                 err(EX_UNAVAILABLE, "fopen: %s", filename);
 572
 573         while (fgets(line, ED_MAX_LINE_LEN, f)) {         /* read commands */
 574                 char *s, *cur = line, *name = NULL, *arg = NULL;
 575
 576                 ++lineno;
 577
 578                 /* parse the line */
 579                 while (cur) {
 580                         s = strsep(&cur, ED_SEPARATORS);
 581                         if (s == NULL || *s == '#')
 582                                 break;
 583                         if (*s == '\0')
 584                                 continue;
 585                         if (arg)
 586                                 errx(ED_EFMT("too many arguments"));
 587                         if (name == NULL)
 588                                 name = s;
 589                         else
 590                                 arg = s;
 591                 }
 592                 if (name == NULL)       /* empty line */
 593                         continue;
 594                 if (arg == NULL)
 595                         errx(ED_EFMT("missing arg for %s"), name);
 596
 597                 if (!strcasecmp(name, ED_TOK_SAMPLES)) {
 598                     if (samples > 0)
 599                         errx(ED_EFMT("duplicate ``samples'' line"));
 600                     if (atoi(arg) <=0)
 601                         errx(ED_EFMT("invalid number of samples"));
 602                     samples = atoi(arg);
 603                     if (samples>ED_MAX_SAMPLES_NO)
 604                             errx(ED_EFMT("too many samples, maximum is %d"),
 605                                 ED_MAX_SAMPLES_NO);
 606                     do_points = 0;
 607                 } else if (!strcasecmp(name, ED_TOK_BW)) {
 608                     read_bandwidth(arg, &p->bandwidth, p->if_name, sizeof(p->if_name));
 609                 } else if (!strcasecmp(name, ED_TOK_LOSS)) {
 610                     if (loss != -1.0)
 611                         errx(ED_EFMT("duplicated token: %s"), name);
 612                     if (!is_valid_number(arg))
 613                         errx(ED_EFMT("invalid %s"), arg);
 614                     loss = atof(arg);
 615                     if (loss > 1)
 616                         errx(ED_EFMT("%s greater than 1.0"), name);
 617                     do_points = 0;
 618                 } else if (!strcasecmp(name, ED_TOK_NAME)) {
 619                     if (profile_name[0] != '\0')
 620                         errx(ED_EFMT("duplicated token: %s"), name);
 621                     strncpy(profile_name, arg, sizeof(profile_name) - 1);
 622                     profile_name[sizeof(profile_name)-1] = '\0';
 623                     do_points = 0;
 624                 } else if (!strcasecmp(name, ED_TOK_DELAY)) {
 625                     if (do_points)
 626                         errx(ED_EFMT("duplicated token: %s"), name);
 627                     delay_first = 1;
 628                     do_points = 1;
 629                 } else if (!strcasecmp(name, ED_TOK_PROB)) {
 630                     if (do_points)
 631                         errx(ED_EFMT("duplicated token: %s"), name);
 632                     delay_first = 0;
 633                     do_points = 1;
 634                 } else if (do_points) {
 635                     if (!is_valid_number(name) || !is_valid_number(arg))
 636                         errx(ED_EFMT("invalid point found"));
 637                     if (delay_first) {
 638                         points[points_no].delay = atof(name);
 639                         points[points_no].prob = atof(arg);
 640                     } else {
 641                         points[points_no].delay = atof(arg);
 642                         points[points_no].prob = atof(name);
 643                     }
 644                     if (points[points_no].prob > 1.0)
 645                         errx(ED_EFMT("probability greater than 1.0"));
 646                     ++points_no;
 647                 } else {
 648                     errx(ED_EFMT("unrecognised command '%s'"), name);
 649                 }
 650         }
 651
 652         fclose (f);
 653
 654         if (samples == -1) {
 655             warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES);
 656             samples = 100;
 657         }
 658
 659         if (loss == -1.0) {
 660             warnx("'%s' not found, assuming no loss", ED_TOK_LOSS);
 661             loss = 1;
 662         }
 663
 664         /* make sure that there are enough points. */
 665         if (points_no < ED_MIN_SAMPLES_NO)
 666             errx(ED_EFMT("too few samples, need at least %d"),
 667                 ED_MIN_SAMPLES_NO);
 668
 669         qsort(points, points_no, sizeof(struct point), compare_points);
 670
 671         /* interpolation */
 672         for (i = 0; i<points_no-1; ++i) {
 673             double y1 = points[i].prob * samples;
 674             double x1 = points[i].delay;
 675             double y2 = points[i+1].prob * samples;
 676             double x2 = points[i+1].delay;
 677
 678             int index = y1;
 679             int stop = y2;
 680
 681             if (x1 == x2) {
 682                 for (; index<stop; ++index)
 683                     p->samples[index] = x1;
 684             } else {
 685                 double m = (y2-y1)/(x2-x1);
 686                 double c = y1 - m*x1;
 687                 for (; index<stop ; ++index)
 688                     p->samples[index] = (index - c)/m;
 689             }
 690         }
 691         p->samples_no = samples;
 692         p->loss_level = loss * samples;
 693         strncpy(p->name, profile_name, sizeof(p->name));
 694 }
 695
 696 void
 697 ipfw_config_pipe(int ac, char **av)
 698 {
 699         int samples[ED_MAX_SAMPLES_NO];
 700         struct dn_pipe p;
 701         int i;
 702         char *end;
 703         void *par = NULL;
 704
 705         memset(&p, 0, sizeof p);
 706         p.bandwidth = -1;
 707
 708         av++; ac--;
 709         /* Pipe number */
 710         if (ac && isdigit(**av)) {
 711                 i = atoi(*av); av++; ac--;
 712                 if (co.do_pipe == 1)
 713                         p.pipe_nr = i;
 714                 else
 715                         p.fs.fs_nr = i;
 716         }
 717         while (ac > 0) {
 718                 double d;
 719                 int tok = match_token(dummynet_params, *av);
 720                 ac--; av++;
 721
 722                 switch(tok) {
 723                 case TOK_NOERROR:
 724                         p.fs.flags_fs |= DN_NOERROR;
 725                         break;
 726
 727                 case TOK_PLR:
 728                         NEED1("plr needs argument 0..1\n");
 729                         d = strtod(av[0], NULL);
 730                         if (d > 1)
 731                                 d = 1;
 732                         else if (d < 0)
 733                                 d = 0;
 734                         p.fs.plr = (int)(d*0x7fffffff);
 735                         ac--; av++;
 736                         break;
 737
 738                 case TOK_QUEUE:
 739                         NEED1("queue needs queue size\n");
 740                         end = NULL;
 741                         p.fs.qsize = strtoul(av[0], &end, 0);
 742                         if (*end == 'K' || *end == 'k') {
 743                                 p.fs.flags_fs |= DN_QSIZE_IS_BYTES;
 744                                 p.fs.qsize *= 1024;
 745                         } else if (*end == 'B' ||
 746                             _substrcmp2(end, "by", "bytes") == 0) {
 747                                 p.fs.flags_fs |= DN_QSIZE_IS_BYTES;
 748                         }
 749                         ac--; av++;
 750                         break;
 751
 752                 case TOK_BUCKETS:
 753                         NEED1("buckets needs argument\n");
 754                         p.fs.rq_size = strtoul(av[0], NULL, 0);
 755                         ac--; av++;
 756                         break;
 757
 758                 case TOK_MASK:
 759                         NEED1("mask needs mask specifier\n");
 760                         /*
 761                          * per-flow queue, mask is dst_ip, dst_port,
 762                          * src_ip, src_port, proto measured in bits
 763                          */
 764                         par = NULL;
 765
 766                         bzero(&p.fs.flow_mask, sizeof(p.fs.flow_mask));
 767                         end = NULL;
 768
 769                         while (ac >= 1) {
 770                             uint32_t *p32 = NULL;
 771                             uint16_t *p16 = NULL;
 772                             uint32_t *p20 = NULL;
 773                             struct in6_addr *pa6 = NULL;
 774                             uint32_t a;
 775
 776                             tok = match_token(dummynet_params, *av);
 777                             ac--; av++;
 778                             switch(tok) {
 779                             case TOK_ALL:
 780                                     /*
 781                                      * special case, all bits significant
 782                                      */
 783                                     p.fs.flow_mask.dst_ip = ~0;
 784                                     p.fs.flow_mask.src_ip = ~0;
 785                                     p.fs.flow_mask.dst_port = ~0;
 786                                     p.fs.flow_mask.src_port = ~0;
 787                                     p.fs.flow_mask.proto = ~0;
 788                                     n2mask(&(p.fs.flow_mask.dst_ip6), 128);
 789                                     n2mask(&(p.fs.flow_mask.src_ip6), 128);
 790                                     p.fs.flow_mask.flow_id6 = ~0;
 791                                     p.fs.flags_fs |= DN_HAVE_FLOW_MASK;
 792                                     goto end_mask;
 793
 794                             case TOK_DSTIP:
 795                                     p32 = &p.fs.flow_mask.dst_ip;
 796                                     break;
 797
 798                             case TOK_SRCIP:
 799                                     p32 = &p.fs.flow_mask.src_ip;
 800                                     break;
 801
 802                             case TOK_DSTIP6:
 803                                     pa6 = &(p.fs.flow_mask.dst_ip6);
 804                                     break;
 805
 806                             case TOK_SRCIP6:
 807                                     pa6 = &(p.fs.flow_mask.src_ip6);
 808                                     break;
 809
 810                             case TOK_FLOWID:
 811                                     p20 = &p.fs.flow_mask.flow_id6;
 812                                     break;
 813
 814                             case TOK_DSTPORT:
 815                                     p16 = &p.fs.flow_mask.dst_port;
 816                                     break;
 817
 818                             case TOK_SRCPORT:
 819                                     p16 = &p.fs.flow_mask.src_port;
 820                                     break;
 821
 822                             case TOK_PROTO:
 823                                     break;
 824
 825                             default:
 826                                     ac++; av--; /* backtrack */
 827                                     goto end_mask;
 828                             }
 829                             if (ac < 1)
 830                                     errx(EX_USAGE, "mask: value missing");
 831                             if (*av[0] == '/') {
 832                                     a = strtoul(av[0]+1, &end, 0);
 833                                     if (pa6 == NULL)
 834                                             a = (a == 32) ? ~0 : (1 << a) - 1;
 835                             } else
 836                                     a = strtoul(av[0], &end, 0);
 837                             if (p32 != NULL)
 838                                     *p32 = a;
 839                             else if (p16 != NULL) {
 840                                     if (a > 0xFFFF)
 841                                             errx(EX_DATAERR,
 842                                                 "port mask must be 16 bit");
 843                                     *p16 = (uint16_t)a;
 844                             } else if (p20 != NULL) {
 845                                     if (a > 0xfffff)
 846                                         errx(EX_DATAERR,
 847                                             "flow_id mask must be 20 bit");
 848                                     *p20 = (uint32_t)a;
 849                             } else if (pa6 != NULL) {
 850                                     if (a > 128)
 851                                         errx(EX_DATAERR,
 852                                             "in6addr invalid mask len");
 853                                     else
 854                                         n2mask(pa6, a);
 855                             } else {
 856                                     if (a > 0xFF)
 857                                             errx(EX_DATAERR,
 858                                                 "proto mask must be 8 bit");
 859                                     p.fs.flow_mask.proto = (uint8_t)a;
 860                             }
 861                             if (a != 0)
 862                                     p.fs.flags_fs |= DN_HAVE_FLOW_MASK;
 863                             ac--; av++;
 864                         } /* end while, config masks */
 865 end_mask:
 866                         break;
 867
 868                 case TOK_RED:
 869                 case TOK_GRED:
 870                         NEED1("red/gred needs w_q/min_th/max_th/max_p\n");
 871                         p.fs.flags_fs |= DN_IS_RED;
 872                         if (tok == TOK_GRED)
 873                                 p.fs.flags_fs |= DN_IS_GENTLE_RED;
 874                         /*
 875                          * the format for parameters is w_q/min_th/max_th/max_p
 876                          */
 877                         if ((end = strsep(&av[0], "/"))) {
 878                             double w_q = strtod(end, NULL);
 879                             if (w_q > 1 || w_q <= 0)
 880                                 errx(EX_DATAERR, "0 < w_q <= 1");
 881                             p.fs.w_q = (int) (w_q * (1 << SCALE_RED));
 882                         }
 883                         if ((end = strsep(&av[0], "/"))) {
 884                             p.fs.min_th = strtoul(end, &end, 0);
 885                             if (*end == 'K' || *end == 'k')
 886                                 p.fs.min_th *= 1024;
 887                         }
 888                         if ((end = strsep(&av[0], "/"))) {
 889                             p.fs.max_th = strtoul(end, &end, 0);
 890                             if (*end == 'K' || *end == 'k')
 891                                 p.fs.max_th *= 1024;
 892                         }
 893                         if ((end = strsep(&av[0], "/"))) {
 894                             double max_p = strtod(end, NULL);
 895                             if (max_p > 1 || max_p <= 0)
 896                                 errx(EX_DATAERR, "0 < max_p <= 1");
 897                             p.fs.max_p = (int)(max_p * (1 << SCALE_RED));
 898                         }
 899                         ac--; av++;
 900                         break;
 901
 902                 case TOK_DROPTAIL:
 903                         p.fs.flags_fs &= ~(DN_IS_RED|DN_IS_GENTLE_RED);
 904                         break;
 905
 906                 case TOK_BW:
 907                         NEED1("bw needs bandwidth or interface\n");
 908                         if (co.do_pipe != 1)
 909                             errx(EX_DATAERR, "bandwidth only valid for pipes");
 910                         read_bandwidth(av[0], &p.bandwidth, p.if_name, sizeof(p.if_name));
 911                         ac--; av++;
 912                         break;
 913
 914                 case TOK_DELAY:
 915                         if (co.do_pipe != 1)
 916                                 errx(EX_DATAERR, "delay only valid for pipes");
 917                         NEED1("delay needs argument 0..10000ms\n");
 918                         p.delay = strtoul(av[0], NULL, 0);
 919                         ac--; av++;
 920                         break;
 921
 922                 case TOK_WEIGHT:
 923                         if (co.do_pipe == 1)
 924                                 errx(EX_DATAERR,"weight only valid for queues");
 925                         NEED1("weight needs argument 0..100\n");
 926                         p.fs.weight = strtoul(av[0], &end, 0);
 927                         ac--; av++;
 928                         break;
 929
 930                 case TOK_PIPE:
 931                         if (co.do_pipe == 1)
 932                                 errx(EX_DATAERR,"pipe only valid for queues");
 933                         NEED1("pipe needs pipe_number\n");
 934                         p.fs.parent_nr = strtoul(av[0], &end, 0);
 935                         ac--; av++;
 936                         break;
 937
 938                 case TOK_PIPE_PROFILE:
 939                         if (co.do_pipe != 1)
 940                             errx(EX_DATAERR, "extra delay only valid for pipes");
 941                         NEED1("extra delay needs the file name\n");
 942                         p.samples = &samples[0];
 943                         load_extra_delays(av[0], &p);
 944                         --ac; ++av;
 945                         break;
 946
 947                 case TOK_BURST:
 948                         if (co.do_pipe != 1)
 949                                 errx(EX_DATAERR, "burst only valid for pipes");
 950                         NEED1("burst needs argument\n");
 951                         errno = 0;
 952                         if (expand_number(av[0], (int64_t *)&p.burst) < 0)
 953                                 if (errno != ERANGE)
 954                                         errx(EX_DATAERR,
 955                                                 "burst: invalid argument");
 956                         if (errno || p.burst > (1ULL << 48) - 1)
 957                                 errx(EX_DATAERR,
 958                                         "burst: out of range (0..2^48-1)");
 959                         ac--; av++;
 960                         break;
 961
 962                 default:
 963                         errx(EX_DATAERR, "unrecognised option ``%s''", av[-1]);
 964                 }
 965         }
 966         if (co.do_pipe == 1) {
 967                 if (p.pipe_nr == 0)
 968                         errx(EX_DATAERR, "pipe_nr must be > 0");
 969                 if (p.delay > 10000)
 970                         errx(EX_DATAERR, "delay must be < 10000");
 971         } else { /* co.do_pipe == 2, queue */
 972                 if (p.fs.parent_nr == 0)
 973                         errx(EX_DATAERR, "pipe must be > 0");
 974                 if (p.fs.weight >100)
 975                         errx(EX_DATAERR, "weight must be <= 100");
 976         }
 977
 978         /* check for bandwidth value */
 979         if (p.bandwidth == -1) {
 980                 p.bandwidth = 0;
 981                 if (p.samples_no > 0)
 982                         errx(EX_DATAERR, "profile requires a bandwidth limit");
 983         }
 984
 985         if (p.fs.flags_fs & DN_QSIZE_IS_BYTES) {
 986                 size_t len;
 987                 long limit;
 988
 989                 len = sizeof(limit);
 990                 if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit",
 991                         &limit, &len, NULL, 0) == -1)
 992                         limit = 1024*1024;
 993                 if (p.fs.qsize > limit)
 994                         errx(EX_DATAERR, "queue size must be < %ldB", limit);
 995         } else {
 996                 size_t len;
 997                 long limit;
 998
 999                 len = sizeof(limit);
1000                 if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit",
1001                         &limit, &len, NULL, 0) == -1)
1002                         limit = 100;
1003                 if (p.fs.qsize > limit)
1004                         errx(EX_DATAERR, "2 <= queue size <= %ld", limit);
1005         }
1006         if (p.fs.flags_fs & DN_IS_RED) {
1007                 size_t len;
1008                 int lookup_depth, avg_pkt_size;
1009                 double s, idle, weight, w_q;
1010                 struct clockinfo ck;
1011                 int t;
1012
1013                 if (p.fs.min_th >= p.fs.max_th)
1014                     errx(EX_DATAERR, "min_th %d must be < than max_th %d",
1015                         p.fs.min_th, p.fs.max_th);
1016                 if (p.fs.max_th == 0)
1017                     errx(EX_DATAERR, "max_th must be > 0");
1018
1019                 len = sizeof(int);
1020                 if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth",
1021                         &lookup_depth, &len, NULL, 0) == -1)
1022                     errx(1, "sysctlbyname(\"%s\")",
1023                         "net.inet.ip.dummynet.red_lookup_depth");
1024                 if (lookup_depth == 0)
1025                     errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth"
1026                         " must be greater than zero");
1027
1028                 len = sizeof(int);
1029                 if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size",
1030                         &avg_pkt_size, &len, NULL, 0) == -1)
1031
1032                     errx(1, "sysctlbyname(\"%s\")",
1033                         "net.inet.ip.dummynet.red_avg_pkt_size");
1034                 if (avg_pkt_size == 0)
1035                         errx(EX_DATAERR,
1036                             "net.inet.ip.dummynet.red_avg_pkt_size must"
1037                             " be greater than zero");
1038
1039                 len = sizeof(struct clockinfo);
1040                 if (sysctlbyname("kern.clockrate", &ck, &len, NULL, 0) == -1)
1041                         errx(1, "sysctlbyname(\"%s\")", "kern.clockrate");
1042
1043                 /*
1044                  * Ticks needed for sending a medium-sized packet.
1045                  * Unfortunately, when we are configuring a WF2Q+ queue, we
1046                  * do not have bandwidth information, because that is stored
1047                  * in the parent pipe, and also we have multiple queues
1048                  * competing for it. So we set s=0, which is not very
1049                  * correct. But on the other hand, why do we want RED with
1050                  * WF2Q+ ?
1051                  */
1052                 if (p.bandwidth==0) /* this is a WF2Q+ queue */
1053                         s = 0;
1054                 else
1055                         s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth;
1056
1057                 /*
1058                  * max idle time (in ticks) before avg queue size becomes 0.
1059                  * NOTA:  (3/w_q) is approx the value x so that
1060                  * (1-w_q)^x < 10^-3.
1061                  */
1062                 w_q = ((double)p.fs.w_q) / (1 << SCALE_RED);
1063                 idle = s * 3. / w_q;
1064                 p.fs.lookup_step = (int)idle / lookup_depth;
1065                 if (!p.fs.lookup_step)
1066                         p.fs.lookup_step = 1;
1067                 weight = 1 - w_q;
1068                 for (t = p.fs.lookup_step; t > 1; --t)
1069                         weight *= 1 - w_q;
1070                 p.fs.lookup_weight = (int)(weight * (1 << SCALE_RED));
1071         }
1072         if (p.samples_no <= 0) {
1073         i = do_cmd(IP_DUMMYNET_CONFIGURE, &p, sizeof p);
1074         } else {
1075                 struct dn_pipe_max pm;
1076                 int len = sizeof(pm);
1077
1078                 memcpy(&pm.pipe, &p, sizeof(pm.pipe));
1079                 memcpy(&pm.samples, samples, sizeof(pm.samples));
1080
1081                 i = do_cmd(IP_DUMMYNET_CONFIGURE, &pm, len);
1082         }
1083
1084         if (i)
1085                 err(1, "setsockopt(%s)", "IP_DUMMYNET_CONFIGURE");
1086 }