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