1 /*- 2 * Codel/FQ_Codel and PIE/FQ_PIE Code: 3 * Copyright (C) 2016 Centre for Advanced Internet Architectures, 4 * Swinburne University of Technology, Melbourne, Australia. 5 * Portions of this code were made possible in part by a gift from 6 * The Comcast Innovation Fund. 7 * Implemented by Rasool Al-Saadi <ralsaadi@swin.edu.au> 8 * 9 * Copyright (c) 2002-2003,2010 Luigi Rizzo 10 * 11 * Redistribution and use in source forms, with and without modification, 12 * are permitted provided that this entire comment appears intact. 13 * 14 * Redistribution in binary form may occur without any restrictions. 15 * Obviously, it would be nice if you gave credit where credit is due 16 * but requiring it would be too onerous. 17 * 18 * This software is provided ``AS IS'' without any warranties of any kind. 19 * 20 * dummynet support 21 */ 22 23 #define NEW_AQM 24 #include <sys/limits.h> 25 #include <sys/types.h> 26 #include <sys/socket.h> 27 /* XXX there are several sysctl leftover here */ 28 #include <sys/sysctl.h> 29 30 #include "ipfw2.h" 31 32 #ifdef NEW_AQM 33 #include <stdint.h> 34 #endif 35 36 #include <ctype.h> 37 #include <err.h> 38 #include <errno.h> 39 #include <libutil.h> 40 #include <netdb.h> 41 #include <stdio.h> 42 #include <stdlib.h> 43 #include <string.h> 44 #include <sysexits.h> 45 46 #include <net/if.h> 47 #include <netinet/in.h> 48 #include <netinet/ip_fw.h> 49 #include <netinet/ip_dummynet.h> 50 #include <arpa/inet.h> /* inet_ntoa */ 51 52 53 static struct _s_x dummynet_params[] = { 54 { "plr", TOK_PLR }, 55 { "noerror", TOK_NOERROR }, 56 { "buckets", TOK_BUCKETS }, 57 { "dst-ip", TOK_DSTIP }, 58 { "src-ip", TOK_SRCIP }, 59 { "dst-port", TOK_DSTPORT }, 60 { "src-port", TOK_SRCPORT }, 61 { "proto", TOK_PROTO }, 62 { "weight", TOK_WEIGHT }, 63 { "lmax", TOK_LMAX }, 64 { "maxlen", TOK_LMAX }, 65 { "all", TOK_ALL }, 66 { "mask", TOK_MASK }, /* alias for both */ 67 { "sched_mask", TOK_SCHED_MASK }, 68 { "flow_mask", TOK_FLOW_MASK }, 69 { "droptail", TOK_DROPTAIL }, 70 { "ecn", TOK_ECN }, 71 { "red", TOK_RED }, 72 { "gred", TOK_GRED }, 73 #ifdef NEW_AQM 74 { "codel", TOK_CODEL}, /* Codel AQM */ 75 { "fq_codel", TOK_FQ_CODEL}, /* FQ-Codel */ 76 { "pie", TOK_PIE}, /* PIE AQM */ 77 { "fq_pie", TOK_FQ_PIE}, /* FQ-PIE */ 78 #endif 79 { "bw", TOK_BW }, 80 { "bandwidth", TOK_BW }, 81 { "delay", TOK_DELAY }, 82 { "link", TOK_LINK }, 83 { "pipe", TOK_PIPE }, 84 { "queue", TOK_QUEUE }, 85 { "flowset", TOK_FLOWSET }, 86 { "sched", TOK_SCHED }, 87 { "pri", TOK_PRI }, 88 { "priority", TOK_PRI }, 89 { "type", TOK_TYPE }, 90 { "flow-id", TOK_FLOWID}, 91 { "dst-ipv6", TOK_DSTIP6}, 92 { "dst-ip6", TOK_DSTIP6}, 93 { "src-ipv6", TOK_SRCIP6}, 94 { "src-ip6", TOK_SRCIP6}, 95 { "profile", TOK_PROFILE}, 96 { "burst", TOK_BURST}, 97 { "dummynet-params", TOK_NULL }, 98 { NULL, 0 } /* terminator */ 99 }; 100 101 #ifdef NEW_AQM 102 /* AQM/extra sched parameters tokens*/ 103 static struct _s_x aqm_params[] = { 104 { "target", TOK_TARGET}, 105 { "interval", TOK_INTERVAL}, 106 { "limit", TOK_LIMIT}, 107 { "flows", TOK_FLOWS}, 108 { "quantum", TOK_QUANTUM}, 109 { "ecn", TOK_ECN}, 110 { "noecn", TOK_NO_ECN}, 111 { "tupdate", TOK_TUPDATE}, 112 { "max_burst", TOK_MAX_BURST}, 113 { "max_ecnth", TOK_MAX_ECNTH}, 114 { "alpha", TOK_ALPHA}, 115 { "beta", TOK_BETA}, 116 { "capdrop", TOK_CAPDROP}, 117 { "nocapdrop", TOK_NO_CAPDROP}, 118 { "onoff", TOK_ONOFF}, 119 { "dre", TOK_DRE}, 120 { "ts", TOK_TS}, 121 { "derand", TOK_DERAND}, 122 { "noderand", TOK_NO_DERAND}, 123 { NULL, 0 } /* terminator */ 124 }; 125 #endif 126 127 #define O_NEXT(p, len) ((void *)((char *)p + len)) 128 129 static void 130 oid_fill(struct dn_id *oid, int len, int type, uintptr_t id) 131 { 132 oid->len = len; 133 oid->type = type; 134 oid->subtype = 0; 135 oid->id = id; 136 } 137 138 /* make room in the buffer and move the pointer forward */ 139 static void * 140 o_next(struct dn_id **o, int len, int type) 141 { 142 struct dn_id *ret = *o; 143 oid_fill(ret, len, type, 0); 144 *o = O_NEXT(*o, len); 145 return ret; 146 } 147 148 #ifdef NEW_AQM 149 150 /* Codel flags */ 151 enum { 152 CODEL_ECN_ENABLED = 1 153 }; 154 155 /* PIE flags, from PIE kernel module */ 156 enum { 157 PIE_ECN_ENABLED = 1, 158 PIE_CAPDROP_ENABLED = 2, 159 PIE_ON_OFF_MODE_ENABLED = 4, 160 PIE_DEPRATEEST_ENABLED = 8, 161 PIE_DERAND_ENABLED = 16 162 }; 163 164 #define PIE_FIX_POINT_BITS 13 165 #define PIE_SCALE (1L<<PIE_FIX_POINT_BITS) 166 167 /* integer to time */ 168 static void 169 us_to_time(int t, char *strt) 170 { 171 if (t < 0) 172 strt[0]='\0'; 173 else if ( t==0 ) 174 sprintf(strt,"%d", t); 175 else if (t< 1000) 176 sprintf(strt,"%dus", t); 177 else if (t < 1000000) 178 sprintf(strt,"%gms", (float) t / 1000); 179 else 180 sprintf(strt,"%gfs", (float) t / 1000000); 181 } 182 183 /* 184 * returns -1 if s is not a valid time, otherwise, return time in us 185 */ 186 static long 187 time_to_us(const char *s) 188 { 189 int i, dots = 0; 190 int len = strlen(s); 191 char strt[16]="", stru[16]=""; 192 193 if (len>15) 194 return -1; 195 for (i = 0; i<len && (isdigit(s[i]) || s[i]=='.') ; i++) 196 if (s[i]=='.') { 197 if (dots) 198 return -1; 199 else 200 dots++; 201 } 202 203 if (!i) 204 return -1; 205 strncpy(strt, s, i); 206 if (i<len) 207 strcpy(stru, s+i); 208 else 209 strcpy(stru, "ms"); 210 211 if (!strcasecmp(stru, "us")) 212 return atol(strt); 213 if (!strcasecmp(stru, "ms")) 214 return (strtod(strt, NULL) * 1000); 215 if (!strcasecmp(stru, "s")) 216 return (strtod(strt, NULL)*1000000); 217 218 return -1; 219 } 220 221 222 /* Get AQM or scheduler extra parameters */ 223 static void 224 get_extra_parms(uint32_t nr, char *out, int subtype) 225 { 226 struct dn_extra_parms *ep; 227 int ret; 228 char strt1[15], strt2[15], strt3[15]; 229 u_int l; 230 231 /* prepare the request */ 232 l = sizeof(struct dn_extra_parms); 233 ep = safe_calloc(1, l); 234 memset(ep, 0, sizeof(*ep)); 235 *out = '\0'; 236 237 oid_fill(&ep->oid, l, DN_CMD_GET, DN_API_VERSION); 238 ep->oid.len = l; 239 ep->oid.subtype = subtype; 240 ep->nr = nr; 241 242 ret = do_cmd(-IP_DUMMYNET3, ep, (uintptr_t)&l); 243 if (ret) { 244 free(ep); 245 errx(EX_DATAERR, "Error getting extra parameters\n"); 246 } 247 248 switch (subtype) { 249 case DN_AQM_PARAMS: 250 if( !strcasecmp(ep->name, "codel")) { 251 us_to_time(ep->par[0], strt1); 252 us_to_time(ep->par[1], strt2); 253 l = sprintf(out, " AQM CoDel target %s interval %s", 254 strt1, strt2); 255 if (ep->par[2] & CODEL_ECN_ENABLED) 256 l = sprintf(out + l, " ECN"); 257 else 258 l += sprintf(out + l, " NoECN"); 259 } else if( !strcasecmp(ep->name, "pie")) { 260 us_to_time(ep->par[0], strt1); 261 us_to_time(ep->par[1], strt2); 262 us_to_time(ep->par[2], strt3); 263 l = sprintf(out, " AQM type PIE target %s tupdate %s alpha " 264 "%g beta %g max_burst %s max_ecnth %.3g", 265 strt1, 266 strt2, 267 ep->par[4] / (float) PIE_SCALE, 268 ep->par[5] / (float) PIE_SCALE, 269 strt3, 270 ep->par[3] / (float) PIE_SCALE 271 ); 272 273 if (ep->par[6] & PIE_ECN_ENABLED) 274 l += sprintf(out + l, " ECN"); 275 else 276 l += sprintf(out + l, " NoECN"); 277 if (ep->par[6] & PIE_CAPDROP_ENABLED) 278 l += sprintf(out + l, " CapDrop"); 279 else 280 l += sprintf(out + l, " NoCapDrop"); 281 if (ep->par[6] & PIE_ON_OFF_MODE_ENABLED) 282 l += sprintf(out + l, " OnOff"); 283 if (ep->par[6] & PIE_DEPRATEEST_ENABLED) 284 l += sprintf(out + l, " DRE"); 285 else 286 l += sprintf(out + l, " TS"); 287 if (ep->par[6] & PIE_DERAND_ENABLED) 288 l += sprintf(out + l, " Derand"); 289 else 290 l += sprintf(out + l, " NoDerand"); 291 } 292 break; 293 294 case DN_SCH_PARAMS: 295 if (!strcasecmp(ep->name,"FQ_CODEL")) { 296 us_to_time(ep->par[0], strt1); 297 us_to_time(ep->par[1], strt2); 298 l = sprintf(out," FQ_CODEL target %s interval %s" 299 " quantum %jd limit %jd flows %jd", 300 strt1, strt2, 301 (intmax_t) ep->par[3], 302 (intmax_t) ep->par[4], 303 (intmax_t) ep->par[5] 304 ); 305 if (ep->par[2] & CODEL_ECN_ENABLED) 306 l += sprintf(out + l, " ECN"); 307 else 308 l += sprintf(out + l, " NoECN"); 309 l += sprintf(out + l, "\n"); 310 } else if (!strcasecmp(ep->name,"FQ_PIE")) { 311 us_to_time(ep->par[0], strt1); 312 us_to_time(ep->par[1], strt2); 313 us_to_time(ep->par[2], strt3); 314 l = sprintf(out, " FQ_PIE target %s tupdate %s alpha " 315 "%g beta %g max_burst %s max_ecnth %.3g" 316 " quantum %jd limit %jd flows %jd", 317 strt1, 318 strt2, 319 ep->par[4] / (float) PIE_SCALE, 320 ep->par[5] / (float) PIE_SCALE, 321 strt3, 322 ep->par[3] / (float) PIE_SCALE, 323 (intmax_t) ep->par[7], 324 (intmax_t) ep->par[8], 325 (intmax_t) ep->par[9] 326 ); 327 328 if (ep->par[6] & PIE_ECN_ENABLED) 329 l += sprintf(out + l, " ECN"); 330 else 331 l += sprintf(out + l, " NoECN"); 332 if (ep->par[6] & PIE_CAPDROP_ENABLED) 333 l += sprintf(out + l, " CapDrop"); 334 else 335 l += sprintf(out + l, " NoCapDrop"); 336 if (ep->par[6] & PIE_ON_OFF_MODE_ENABLED) 337 l += sprintf(out + l, " OnOff"); 338 if (ep->par[6] & PIE_DEPRATEEST_ENABLED) 339 l += sprintf(out + l, " DRE"); 340 else 341 l += sprintf(out + l, " TS"); 342 if (ep->par[6] & PIE_DERAND_ENABLED) 343 l += sprintf(out + l, " Derand"); 344 else 345 l += sprintf(out + l, " NoDerand"); 346 l += sprintf(out + l, "\n"); 347 } 348 break; 349 } 350 351 free(ep); 352 } 353 #endif 354 355 356 #if 0 357 static int 358 sort_q(void *arg, const void *pa, const void *pb) 359 { 360 int rev = (co.do_sort < 0); 361 int field = rev ? -co.do_sort : co.do_sort; 362 long long res = 0; 363 const struct dn_flow_queue *a = pa; 364 const struct dn_flow_queue *b = pb; 365 366 switch (field) { 367 case 1: /* pkts */ 368 res = a->len - b->len; 369 break; 370 case 2: /* bytes */ 371 res = a->len_bytes - b->len_bytes; 372 break; 373 374 case 3: /* tot pkts */ 375 res = a->tot_pkts - b->tot_pkts; 376 break; 377 378 case 4: /* tot bytes */ 379 res = a->tot_bytes - b->tot_bytes; 380 break; 381 } 382 if (res < 0) 383 res = -1; 384 if (res > 0) 385 res = 1; 386 return (int)(rev ? res : -res); 387 } 388 #endif 389 390 /* print a mask and header for the subsequent list of flows */ 391 static void 392 print_mask(struct ipfw_flow_id *id) 393 { 394 if (!IS_IP6_FLOW_ID(id)) { 395 printf(" " 396 "mask: %s 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n", 397 id->extra ? "queue," : "", 398 id->proto, 399 id->src_ip, id->src_port, 400 id->dst_ip, id->dst_port); 401 } else { 402 char buf[255]; 403 printf("\n mask: %sproto: 0x%02x, flow_id: 0x%08x, ", 404 id->extra ? "queue," : "", 405 id->proto, id->flow_id6); 406 inet_ntop(AF_INET6, &(id->src_ip6), buf, sizeof(buf)); 407 printf("%s/0x%04x -> ", buf, id->src_port); 408 inet_ntop(AF_INET6, &(id->dst_ip6), buf, sizeof(buf)); 409 printf("%s/0x%04x\n", buf, id->dst_port); 410 } 411 } 412 413 static void 414 print_header(struct ipfw_flow_id *id) 415 { 416 if (!IS_IP6_FLOW_ID(id)) 417 printf("BKT Prot ___Source IP/port____ " 418 "____Dest. IP/port____ " 419 "Tot_pkt/bytes Pkt/Byte Drp\n"); 420 else 421 printf("BKT ___Prot___ _flow-id_ " 422 "______________Source IPv6/port_______________ " 423 "_______________Dest. IPv6/port_______________ " 424 "Tot_pkt/bytes Pkt/Byte Drp\n"); 425 } 426 427 static void 428 list_flow(struct buf_pr *bp, struct dn_flow *ni) 429 { 430 char buff[255]; 431 struct protoent *pe = NULL; 432 struct in_addr ina; 433 struct ipfw_flow_id *id = &ni->fid; 434 435 pe = getprotobynumber(id->proto); 436 /* XXX: Should check for IPv4 flows */ 437 bprintf(bp, "%3u%c", (ni->oid.id) & 0xff, 438 id->extra ? '*' : ' '); 439 if (!IS_IP6_FLOW_ID(id)) { 440 if (pe) 441 bprintf(bp, "%-4s ", pe->p_name); 442 else 443 bprintf(bp, "%4u ", id->proto); 444 ina.s_addr = htonl(id->src_ip); 445 bprintf(bp, "%15s/%-5d ", 446 inet_ntoa(ina), id->src_port); 447 ina.s_addr = htonl(id->dst_ip); 448 bprintf(bp, "%15s/%-5d ", 449 inet_ntoa(ina), id->dst_port); 450 } else { 451 /* Print IPv6 flows */ 452 if (pe != NULL) 453 bprintf(bp, "%9s ", pe->p_name); 454 else 455 bprintf(bp, "%9u ", id->proto); 456 bprintf(bp, "%7d %39s/%-5d ", id->flow_id6, 457 inet_ntop(AF_INET6, &(id->src_ip6), buff, sizeof(buff)), 458 id->src_port); 459 bprintf(bp, " %39s/%-5d ", 460 inet_ntop(AF_INET6, &(id->dst_ip6), buff, sizeof(buff)), 461 id->dst_port); 462 } 463 pr_u64(bp, &ni->tot_pkts, 4); 464 pr_u64(bp, &ni->tot_bytes, 8); 465 bprintf(bp, "%2u %4u %3u", 466 ni->length, ni->len_bytes, ni->drops); 467 } 468 469 static void 470 print_flowset_parms(struct dn_fs *fs, char *prefix) 471 { 472 int l; 473 char qs[30]; 474 char plr[30]; 475 char red[200]; /* Display RED parameters */ 476 477 l = fs->qsize; 478 if (fs->flags & DN_QSIZE_BYTES) { 479 if (l >= 8192) 480 sprintf(qs, "%d KB", l / 1024); 481 else 482 sprintf(qs, "%d B", l); 483 } else 484 sprintf(qs, "%3d sl.", l); 485 if (fs->plr) 486 sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff)); 487 else 488 plr[0] = '\0'; 489 490 if (fs->flags & DN_IS_RED) { /* RED parameters */ 491 sprintf(red, 492 "\n\t %cRED w_q %f min_th %d max_th %d max_p %f", 493 (fs->flags & DN_IS_GENTLE_RED) ? 'G' : ' ', 494 1.0 * fs->w_q / (double)(1 << SCALE_RED), 495 fs->min_th, 496 fs->max_th, 497 1.0 * fs->max_p / (double)(1 << SCALE_RED)); 498 if (fs->flags & DN_IS_ECN) 499 strlcat(red, " (ecn)", sizeof(red)); 500 #ifdef NEW_AQM 501 /* get AQM parameters */ 502 } else if (fs->flags & DN_IS_AQM) { 503 get_extra_parms(fs->fs_nr, red, DN_AQM_PARAMS); 504 #endif 505 } else 506 sprintf(red, "droptail"); 507 508 if (prefix[0]) { 509 printf("%s %s%s %d queues (%d buckets) %s\n", 510 prefix, qs, plr, fs->oid.id, fs->buckets, red); 511 prefix[0] = '\0'; 512 } else { 513 printf("q%05d %s%s %d flows (%d buckets) sched %d " 514 "weight %d lmax %d pri %d %s\n", 515 fs->fs_nr, qs, plr, fs->oid.id, fs->buckets, 516 fs->sched_nr, fs->par[0], fs->par[1], fs->par[2], red); 517 if (fs->flags & DN_HAVE_MASK) 518 print_mask(&fs->flow_mask); 519 } 520 } 521 522 static void 523 print_extra_delay_parms(struct dn_profile *p) 524 { 525 double loss; 526 if (p->samples_no <= 0) 527 return; 528 529 loss = p->loss_level; 530 loss /= p->samples_no; 531 printf("\t profile: name \"%s\" loss %f samples %d\n", 532 p->name, loss, p->samples_no); 533 } 534 535 static void 536 flush_buf(char *buf) 537 { 538 if (buf[0]) 539 printf("%s\n", buf); 540 buf[0] = '\0'; 541 } 542 543 /* 544 * generic list routine. We expect objects in a specific order, i.e. 545 * PIPES AND SCHEDULERS: 546 * link; scheduler; internal flowset if any; instances 547 * we can tell a pipe from the number. 548 * 549 * FLOWSETS: 550 * flowset; queues; 551 * link i (int queue); scheduler i; si(i) { flowsets() : queues } 552 */ 553 static void 554 list_pipes(struct dn_id *oid, struct dn_id *end) 555 { 556 char buf[160]; /* pending buffer */ 557 int toPrint = 1; /* print header */ 558 struct buf_pr bp; 559 560 buf[0] = '\0'; 561 bp_alloc(&bp, 4096); 562 for (; oid != end; oid = O_NEXT(oid, oid->len)) { 563 if (oid->len < sizeof(*oid)) 564 errx(1, "invalid oid len %d\n", oid->len); 565 566 switch (oid->type) { 567 default: 568 flush_buf(buf); 569 printf("unrecognized object %d size %d\n", oid->type, oid->len); 570 break; 571 case DN_TEXT: /* list of attached flowsets */ 572 { 573 int i, l; 574 struct { 575 struct dn_id id; 576 uint32_t p[0]; 577 } *d = (void *)oid; 578 l = (oid->len - sizeof(*oid))/sizeof(d->p[0]); 579 if (l == 0) 580 break; 581 printf(" Children flowsets: "); 582 for (i = 0; i < l; i++) 583 printf("%u ", d->p[i]); 584 printf("\n"); 585 break; 586 } 587 case DN_CMD_GET: 588 if (g_co.verbose) 589 printf("answer for cmd %d, len %d\n", oid->type, oid->id); 590 break; 591 case DN_SCH: { 592 struct dn_sch *s = (struct dn_sch *)oid; 593 flush_buf(buf); 594 printf(" sched %d type %s flags 0x%x %d buckets %d active\n", 595 s->sched_nr, 596 s->name, s->flags, s->buckets, s->oid.id); 597 #ifdef NEW_AQM 598 char parms[200]; 599 get_extra_parms(s->sched_nr, parms, DN_SCH_PARAMS); 600 printf("%s",parms); 601 #endif 602 if (s->flags & DN_HAVE_MASK) 603 print_mask(&s->sched_mask); 604 } 605 break; 606 607 case DN_FLOW: 608 if (toPrint != 0) { 609 print_header(&((struct dn_flow *)oid)->fid); 610 toPrint = 0; 611 } 612 list_flow(&bp, (struct dn_flow *)oid); 613 printf("%s\n", bp.buf); 614 bp_flush(&bp); 615 break; 616 617 case DN_LINK: { 618 struct dn_link *p = (struct dn_link *)oid; 619 double b = p->bandwidth; 620 char bwbuf[30]; 621 char burst[5 + 7]; 622 623 /* This starts a new object so flush buffer */ 624 flush_buf(buf); 625 /* data rate */ 626 if (b == 0) 627 sprintf(bwbuf, "unlimited "); 628 else if (b >= 1000000000) 629 sprintf(bwbuf, "%7.3f Gbit/s", b/1000000000); 630 else if (b >= 1000000) 631 sprintf(bwbuf, "%7.3f Mbit/s", b/1000000); 632 else if (b >= 1000) 633 sprintf(bwbuf, "%7.3f Kbit/s", b/1000); 634 else 635 sprintf(bwbuf, "%7.3f bit/s ", b); 636 637 if (humanize_number(burst, sizeof(burst), p->burst, 638 "", HN_AUTOSCALE, 0) < 0 || g_co.verbose) 639 sprintf(burst, "%d", (int)p->burst); 640 sprintf(buf, "%05d: %s %4d ms burst %s", 641 p->link_nr % DN_MAX_ID, bwbuf, p->delay, burst); 642 } 643 break; 644 645 case DN_FS: 646 print_flowset_parms((struct dn_fs *)oid, buf); 647 break; 648 case DN_PROFILE: 649 flush_buf(buf); 650 print_extra_delay_parms((struct dn_profile *)oid); 651 } 652 flush_buf(buf); // XXX does it really go here ? 653 } 654 655 bp_free(&bp); 656 } 657 658 /* 659 * Delete pipe, queue or scheduler i 660 */ 661 int 662 ipfw_delete_pipe(int do_pipe, int i) 663 { 664 struct { 665 struct dn_id oid; 666 uintptr_t a[1]; /* add more if we want a list */ 667 } cmd; 668 oid_fill((void *)&cmd, sizeof(cmd), DN_CMD_DELETE, DN_API_VERSION); 669 cmd.oid.subtype = (do_pipe == 1) ? DN_LINK : 670 ( (do_pipe == 2) ? DN_FS : DN_SCH); 671 cmd.a[0] = i; 672 i = do_cmd(IP_DUMMYNET3, &cmd, cmd.oid.len); 673 if (i) { 674 i = 1; 675 warn("rule %u: setsockopt(IP_DUMMYNET_DEL)", i); 676 } 677 return i; 678 } 679 680 /* 681 * Code to parse delay profiles. 682 * 683 * Some link types introduce extra delays in the transmission 684 * of a packet, e.g. because of MAC level framing, contention on 685 * the use of the channel, MAC level retransmissions and so on. 686 * From our point of view, the channel is effectively unavailable 687 * for this extra time, which is constant or variable depending 688 * on the link type. Additionally, packets may be dropped after this 689 * time (e.g. on a wireless link after too many retransmissions). 690 * We can model the additional delay with an empirical curve 691 * that represents its distribution. 692 * 693 * cumulative probability 694 * 1.0 ^ 695 * | 696 * L +-- loss-level x 697 * | ****** 698 * | * 699 * | ***** 700 * | * 701 * | ** 702 * | * 703 * +-------*-------------------> 704 * delay 705 * 706 * The empirical curve may have both vertical and horizontal lines. 707 * Vertical lines represent constant delay for a range of 708 * probabilities; horizontal lines correspond to a discontinuty 709 * in the delay distribution: the link will use the largest delay 710 * for a given probability. 711 * 712 * To pass the curve to dummynet, we must store the parameters 713 * in a file as described below, and issue the command 714 * 715 * ipfw pipe <n> config ... bw XXX profile <filename> ... 716 * 717 * The file format is the following, with whitespace acting as 718 * a separator and '#' indicating the beginning a comment: 719 * 720 * samples N 721 * the number of samples used in the internal 722 * representation (2..1024; default 100); 723 * 724 * loss-level L 725 * The probability above which packets are lost. 726 * (0.0 <= L <= 1.0, default 1.0 i.e. no loss); 727 * 728 * name identifier 729 * Optional a name (listed by "ipfw pipe show") 730 * to identify the distribution; 731 * 732 * "delay prob" | "prob delay" 733 * One of these two lines is mandatory and defines 734 * the format of the following lines with data points. 735 * 736 * XXX YYY 737 * 2 or more lines representing points in the curve, 738 * with either delay or probability first, according 739 * to the chosen format. 740 * The unit for delay is milliseconds. 741 * 742 * Data points does not need to be ordered or equal to the number 743 * specified in the "samples" line. ipfw will sort and interpolate 744 * the curve as needed. 745 * 746 * Example of a profile file: 747 748 name bla_bla_bla 749 samples 100 750 loss-level 0.86 751 prob delay 752 0 200 # minimum overhead is 200ms 753 0.5 200 754 0.5 300 755 0.8 1000 756 0.9 1300 757 1 1300 758 759 * Internally, we will convert the curve to a fixed number of 760 * samples, and when it is time to transmit a packet we will 761 * model the extra delay as extra bits in the packet. 762 * 763 */ 764 765 #define ED_MAX_LINE_LEN 256+ED_MAX_NAME_LEN 766 #define ED_TOK_SAMPLES "samples" 767 #define ED_TOK_LOSS "loss-level" 768 #define ED_TOK_NAME "name" 769 #define ED_TOK_DELAY "delay" 770 #define ED_TOK_PROB "prob" 771 #define ED_TOK_BW "bw" 772 #define ED_SEPARATORS " \t\n" 773 #define ED_MIN_SAMPLES_NO 2 774 775 /* 776 * returns 1 if s is a non-negative number, with at least one '.' 777 */ 778 static int 779 is_valid_number(const char *s) 780 { 781 int i, dots_found = 0; 782 int len = strlen(s); 783 784 for (i = 0; i<len; ++i) 785 if (!isdigit(s[i]) && (s[i] !='.' || ++dots_found > 1)) 786 return 0; 787 return 1; 788 } 789 790 /* 791 * Take as input a string describing a bandwidth value 792 * and return the numeric bandwidth value. 793 * set clocking interface or bandwidth value 794 */ 795 static void 796 read_bandwidth(char *arg, uint32_t *bandwidth, char *if_name, int namelen) 797 { 798 if (*bandwidth != (uint32_t)-1) 799 warnx("duplicate token, override bandwidth value!"); 800 801 if (arg[0] >= 'a' && arg[0] <= 'z') { 802 if (!if_name) { 803 errx(1, "no if support"); 804 } 805 if (namelen >= IFNAMSIZ) 806 warn("interface name truncated"); 807 namelen--; 808 /* interface name */ 809 strlcpy(if_name, arg, namelen); 810 *bandwidth = 0; 811 } else { /* read bandwidth value */ 812 uint64_t bw; 813 char *end = NULL; 814 815 bw = strtoul(arg, &end, 0); 816 if (*end == 'K' || *end == 'k') { 817 end++; 818 bw *= 1000; 819 } else if (*end == 'M' || *end == 'm') { 820 end++; 821 bw *= 1000000; 822 } else if (*end == 'G' || *end == 'g') { 823 end++; 824 bw *= 1000000000; 825 } 826 if ((*end == 'B' && 827 _substrcmp2(end, "Bi", "Bit/s") != 0) || 828 _substrcmp2(end, "by", "bytes") == 0) 829 bw *= 8; 830 831 if (bw > UINT_MAX) 832 errx(EX_DATAERR, "bandwidth too large"); 833 834 *bandwidth = (uint32_t)bw; 835 if (if_name) 836 if_name[0] = '\0'; 837 } 838 } 839 840 struct point { 841 double prob; 842 double delay; 843 }; 844 845 static int 846 compare_points(const void *vp1, const void *vp2) 847 { 848 const struct point *p1 = vp1; 849 const struct point *p2 = vp2; 850 double res = 0; 851 852 res = p1->prob - p2->prob; 853 if (res == 0) 854 res = p1->delay - p2->delay; 855 if (res < 0) 856 return -1; 857 else if (res > 0) 858 return 1; 859 else 860 return 0; 861 } 862 863 #define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno 864 865 static void 866 load_extra_delays(const char *filename, struct dn_profile *p, 867 struct dn_link *link) 868 { 869 char line[ED_MAX_LINE_LEN]; 870 FILE *f; 871 int lineno = 0; 872 int i; 873 874 int samples = -1; 875 double loss = -1.0; 876 char profile_name[ED_MAX_NAME_LEN]; 877 int delay_first = -1; 878 int do_points = 0; 879 struct point points[ED_MAX_SAMPLES_NO]; 880 int points_no = 0; 881 882 /* XXX link never NULL? */ 883 p->link_nr = link->link_nr; 884 885 profile_name[0] = '\0'; 886 f = fopen(filename, "r"); 887 if (f == NULL) 888 err(EX_UNAVAILABLE, "fopen: %s", filename); 889 890 while (fgets(line, ED_MAX_LINE_LEN, f)) { /* read commands */ 891 char *s, *cur = line, *name = NULL, *arg = NULL; 892 893 ++lineno; 894 895 /* parse the line */ 896 while (cur) { 897 s = strsep(&cur, ED_SEPARATORS); 898 if (s == NULL || *s == '#') 899 break; 900 if (*s == '\0') 901 continue; 902 if (arg) 903 errx(ED_EFMT("too many arguments")); 904 if (name == NULL) 905 name = s; 906 else 907 arg = s; 908 } 909 if (name == NULL) /* empty line */ 910 continue; 911 if (arg == NULL) 912 errx(ED_EFMT("missing arg for %s"), name); 913 914 if (!strcasecmp(name, ED_TOK_SAMPLES)) { 915 if (samples > 0) 916 errx(ED_EFMT("duplicate ``samples'' line")); 917 if (atoi(arg) <=0) 918 errx(ED_EFMT("invalid number of samples")); 919 samples = atoi(arg); 920 if (samples>ED_MAX_SAMPLES_NO) 921 errx(ED_EFMT("too many samples, maximum is %d"), 922 ED_MAX_SAMPLES_NO); 923 do_points = 0; 924 } else if (!strcasecmp(name, ED_TOK_BW)) { 925 char buf[IFNAMSIZ]; 926 read_bandwidth(arg, &link->bandwidth, buf, sizeof(buf)); 927 } else if (!strcasecmp(name, ED_TOK_LOSS)) { 928 if (loss != -1.0) 929 errx(ED_EFMT("duplicated token: %s"), name); 930 if (!is_valid_number(arg)) 931 errx(ED_EFMT("invalid %s"), arg); 932 loss = atof(arg); 933 if (loss > 1) 934 errx(ED_EFMT("%s greater than 1.0"), name); 935 do_points = 0; 936 } else if (!strcasecmp(name, ED_TOK_NAME)) { 937 if (profile_name[0] != '\0') 938 errx(ED_EFMT("duplicated token: %s"), name); 939 strlcpy(profile_name, arg, sizeof(profile_name)); 940 do_points = 0; 941 } else if (!strcasecmp(name, ED_TOK_DELAY)) { 942 if (do_points) 943 errx(ED_EFMT("duplicated token: %s"), name); 944 delay_first = 1; 945 do_points = 1; 946 } else if (!strcasecmp(name, ED_TOK_PROB)) { 947 if (do_points) 948 errx(ED_EFMT("duplicated token: %s"), name); 949 delay_first = 0; 950 do_points = 1; 951 } else if (do_points) { 952 if (!is_valid_number(name) || !is_valid_number(arg)) 953 errx(ED_EFMT("invalid point found")); 954 if (delay_first) { 955 points[points_no].delay = atof(name); 956 points[points_no].prob = atof(arg); 957 } else { 958 points[points_no].delay = atof(arg); 959 points[points_no].prob = atof(name); 960 } 961 if (points[points_no].prob > 1.0) 962 errx(ED_EFMT("probability greater than 1.0")); 963 ++points_no; 964 } else { 965 errx(ED_EFMT("unrecognised command '%s'"), name); 966 } 967 } 968 969 fclose (f); 970 971 if (samples == -1) { 972 warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES); 973 samples = 100; 974 } 975 976 if (loss == -1.0) { 977 warnx("'%s' not found, assuming no loss", ED_TOK_LOSS); 978 loss = 1; 979 } 980 981 /* make sure that there are enough points. */ 982 if (points_no < ED_MIN_SAMPLES_NO) 983 errx(ED_EFMT("too few samples, need at least %d"), 984 ED_MIN_SAMPLES_NO); 985 986 qsort(points, points_no, sizeof(struct point), compare_points); 987 988 /* interpolation */ 989 for (i = 0; i<points_no-1; ++i) { 990 double y1 = points[i].prob * samples; 991 double x1 = points[i].delay; 992 double y2 = points[i+1].prob * samples; 993 double x2 = points[i+1].delay; 994 995 int ix = y1; 996 int stop = y2; 997 998 if (x1 == x2) { 999 for (; ix<stop; ++ix) 1000 p->samples[ix] = x1; 1001 } else { 1002 double m = (y2-y1)/(x2-x1); 1003 double c = y1 - m*x1; 1004 for (; ix<stop ; ++ix) 1005 p->samples[ix] = (ix - c)/m; 1006 } 1007 } 1008 p->samples_no = samples; 1009 p->loss_level = loss * samples; 1010 strlcpy(p->name, profile_name, sizeof(p->name)); 1011 } 1012 1013 #ifdef NEW_AQM 1014 1015 /* Parse AQM/extra scheduler parameters */ 1016 static int 1017 process_extra_parms(int *ac, char **av, struct dn_extra_parms *ep, 1018 uint16_t type) 1019 { 1020 int i; 1021 1022 /* use kernel defaults */ 1023 for (i=0; i<DN_MAX_EXTRA_PARM; i++) 1024 ep->par[i] = -1; 1025 1026 switch(type) { 1027 case TOK_CODEL: 1028 case TOK_FQ_CODEL: 1029 /* Codel 1030 * 0- target, 1- interval, 2- flags, 1031 * FQ_CODEL 1032 * 3- quantum, 4- limit, 5- flows 1033 */ 1034 if (type==TOK_CODEL) 1035 ep->par[2] = 0; 1036 else 1037 ep->par[2] = CODEL_ECN_ENABLED; 1038 1039 while (*ac > 0) { 1040 int tok = match_token(aqm_params, *av); 1041 (*ac)--; av++; 1042 switch(tok) { 1043 case TOK_TARGET: 1044 if (*ac <= 0 || time_to_us(av[0]) < 0) 1045 errx(EX_DATAERR, "target needs time\n"); 1046 1047 ep->par[0] = time_to_us(av[0]); 1048 (*ac)--; av++; 1049 break; 1050 1051 case TOK_INTERVAL: 1052 if (*ac <= 0 || time_to_us(av[0]) < 0) 1053 errx(EX_DATAERR, "interval needs time\n"); 1054 1055 ep->par[1] = time_to_us(av[0]); 1056 (*ac)--; av++; 1057 break; 1058 1059 case TOK_ECN: 1060 ep->par[2] = CODEL_ECN_ENABLED; 1061 break; 1062 case TOK_NO_ECN: 1063 ep->par[2] &= ~CODEL_ECN_ENABLED; 1064 break; 1065 /* Config fq_codel parameters */ 1066 case TOK_QUANTUM: 1067 if (type != TOK_FQ_CODEL) 1068 errx(EX_DATAERR, "quantum is not for codel\n"); 1069 if (*ac <= 0 || !is_valid_number(av[0])) 1070 errx(EX_DATAERR, "quantum needs number\n"); 1071 1072 ep->par[3]= atoi(av[0]); 1073 (*ac)--; av++; 1074 break; 1075 1076 case TOK_LIMIT: 1077 if (type != TOK_FQ_CODEL) 1078 errx(EX_DATAERR, "limit is not for codel, use queue instead\n"); 1079 if (*ac <= 0 || !is_valid_number(av[0])) 1080 errx(EX_DATAERR, "limit needs number\n"); 1081 1082 ep->par[4] = atoi(av[0]); 1083 (*ac)--; av++; 1084 break; 1085 1086 case TOK_FLOWS: 1087 if (type != TOK_FQ_CODEL) 1088 errx(EX_DATAERR, "flows is not for codel\n"); 1089 if (*ac <= 0 || !is_valid_number(av[0])) 1090 errx(EX_DATAERR, "flows needs number\n"); 1091 1092 ep->par[5] = atoi(av[0]); 1093 (*ac)--; av++; 1094 break; 1095 1096 default: 1097 printf("%s is Invalid parameter\n", av[-1]); 1098 } 1099 } 1100 break; 1101 case TOK_PIE: 1102 case TOK_FQ_PIE: 1103 /* PIE 1104 * 0- target , 1- tupdate, 2- max_burst, 1105 * 3- max_ecnth, 4- alpha, 1106 * 5- beta, 6- flags 1107 * FQ_CODEL 1108 * 7- quantum, 8- limit, 9- flows 1109 */ 1110 1111 if ( type == TOK_PIE) 1112 ep->par[6] = PIE_CAPDROP_ENABLED | PIE_DEPRATEEST_ENABLED 1113 | PIE_DERAND_ENABLED; 1114 else 1115 /* for FQ-PIE, use TS mode */ 1116 ep->par[6] = PIE_CAPDROP_ENABLED | PIE_DERAND_ENABLED 1117 | PIE_ECN_ENABLED; 1118 1119 while (*ac > 0) { 1120 int tok = match_token(aqm_params, *av); 1121 (*ac)--; av++; 1122 switch(tok) { 1123 case TOK_TARGET: 1124 if (*ac <= 0 || time_to_us(av[0]) < 0) 1125 errx(EX_DATAERR, "target needs time\n"); 1126 1127 ep->par[0] = time_to_us(av[0]); 1128 (*ac)--; av++; 1129 break; 1130 1131 case TOK_TUPDATE: 1132 if (*ac <= 0 || time_to_us(av[0]) < 0) 1133 errx(EX_DATAERR, "tupdate needs time\n"); 1134 1135 ep->par[1] = time_to_us(av[0]); 1136 (*ac)--; av++; 1137 break; 1138 1139 case TOK_MAX_BURST: 1140 if (*ac <= 0 || time_to_us(av[0]) < 0) 1141 errx(EX_DATAERR, "max_burst needs time\n"); 1142 1143 ep->par[2] = time_to_us(av[0]); 1144 (*ac)--; av++; 1145 break; 1146 1147 case TOK_MAX_ECNTH: 1148 if (*ac <= 0 || !is_valid_number(av[0])) 1149 errx(EX_DATAERR, "max_ecnth needs number\n"); 1150 1151 ep->par[3] = atof(av[0]) * PIE_SCALE; 1152 (*ac)--; av++; 1153 break; 1154 1155 case TOK_ALPHA: 1156 if (*ac <= 0 || !is_valid_number(av[0])) 1157 errx(EX_DATAERR, "alpha needs number\n"); 1158 1159 ep->par[4] = atof(av[0]) * PIE_SCALE; 1160 (*ac)--; av++; 1161 break; 1162 1163 case TOK_BETA: 1164 if (*ac <= 0 || !is_valid_number(av[0])) 1165 errx(EX_DATAERR, "beta needs number\n"); 1166 1167 ep->par[5] = atof(av[0]) * PIE_SCALE; 1168 (*ac)--; av++; 1169 break; 1170 1171 case TOK_ECN: 1172 ep->par[6] |= PIE_ECN_ENABLED; 1173 break; 1174 case TOK_NO_ECN: 1175 ep->par[6] &= ~PIE_ECN_ENABLED; 1176 break; 1177 1178 case TOK_CAPDROP: 1179 ep->par[6] |= PIE_CAPDROP_ENABLED; 1180 break; 1181 case TOK_NO_CAPDROP: 1182 ep->par[6] &= ~PIE_CAPDROP_ENABLED; 1183 break; 1184 1185 case TOK_ONOFF: 1186 ep->par[6] |= PIE_ON_OFF_MODE_ENABLED; 1187 break; 1188 1189 case TOK_DRE: 1190 ep->par[6] |= PIE_DEPRATEEST_ENABLED; 1191 break; 1192 1193 case TOK_TS: 1194 ep->par[6] &= ~PIE_DEPRATEEST_ENABLED; 1195 break; 1196 1197 case TOK_DERAND: 1198 ep->par[6] |= PIE_DERAND_ENABLED; 1199 break; 1200 case TOK_NO_DERAND: 1201 ep->par[6] &= ~PIE_DERAND_ENABLED; 1202 break; 1203 1204 /* Config fq_pie parameters */ 1205 case TOK_QUANTUM: 1206 if (type != TOK_FQ_PIE) 1207 errx(EX_DATAERR, "quantum is not for pie\n"); 1208 if (*ac <= 0 || !is_valid_number(av[0])) 1209 errx(EX_DATAERR, "quantum needs number\n"); 1210 1211 ep->par[7]= atoi(av[0]); 1212 (*ac)--; av++; 1213 break; 1214 1215 case TOK_LIMIT: 1216 if (type != TOK_FQ_PIE) 1217 errx(EX_DATAERR, "limit is not for pie, use queue instead\n"); 1218 if (*ac <= 0 || !is_valid_number(av[0])) 1219 errx(EX_DATAERR, "limit needs number\n"); 1220 1221 ep->par[8] = atoi(av[0]); 1222 (*ac)--; av++; 1223 break; 1224 1225 case TOK_FLOWS: 1226 if (type != TOK_FQ_PIE) 1227 errx(EX_DATAERR, "flows is not for pie\n"); 1228 if (*ac <= 0 || !is_valid_number(av[0])) 1229 errx(EX_DATAERR, "flows needs number\n"); 1230 1231 ep->par[9] = atoi(av[0]); 1232 (*ac)--; av++; 1233 break; 1234 1235 1236 default: 1237 printf("%s is invalid parameter\n", av[-1]); 1238 } 1239 } 1240 break; 1241 } 1242 1243 return 0; 1244 } 1245 1246 #endif 1247 1248 1249 /* 1250 * configuration of pipes, schedulers, flowsets. 1251 * When we configure a new scheduler, an empty pipe is created, so: 1252 * 1253 * do_pipe = 1 -> "pipe N config ..." only for backward compatibility 1254 * sched N+Delta type fifo sched_mask ... 1255 * pipe N+Delta <parameters> 1256 * flowset N+Delta pipe N+Delta (no parameters) 1257 * sched N type wf2q+ sched_mask ... 1258 * pipe N <parameters> 1259 * 1260 * do_pipe = 2 -> flowset N config 1261 * flowset N parameters 1262 * 1263 * do_pipe = 3 -> sched N config 1264 * sched N parameters (default no pipe) 1265 * optional Pipe N config ... 1266 * pipe ==> 1267 */ 1268 void 1269 ipfw_config_pipe(int ac, char **av) 1270 { 1271 int i; 1272 u_int j; 1273 char *end; 1274 struct dn_id *buf, *base; 1275 struct dn_sch *sch = NULL; 1276 struct dn_link *p = NULL; 1277 struct dn_fs *fs = NULL; 1278 struct dn_profile *pf = NULL; 1279 struct ipfw_flow_id *mask = NULL; 1280 #ifdef NEW_AQM 1281 struct dn_extra_parms *aqm_extra = NULL; 1282 struct dn_extra_parms *sch_extra = NULL; 1283 int lmax_extra; 1284 #endif 1285 1286 int lmax; 1287 uint32_t _foo = 0, *flags = &_foo , *buckets = &_foo; 1288 1289 /* 1290 * allocate space for 1 header, 1291 * 1 scheduler, 1 link, 1 flowset, 1 profile 1292 */ 1293 lmax = sizeof(struct dn_id); /* command header */ 1294 lmax += sizeof(struct dn_sch) + sizeof(struct dn_link) + 1295 sizeof(struct dn_fs) + sizeof(struct dn_profile); 1296 1297 #ifdef NEW_AQM 1298 /* Extra Params */ 1299 lmax_extra = sizeof(struct dn_extra_parms); 1300 /* two lmax_extra because one for AQM params and another 1301 * sch params 1302 */ 1303 lmax += lmax_extra*2; 1304 #endif 1305 1306 av++; ac--; 1307 /* Pipe number */ 1308 if (ac && isdigit(**av)) { 1309 i = atoi(*av); av++; ac--; 1310 } else 1311 i = -1; 1312 if (i <= 0) 1313 errx(EX_USAGE, "need a pipe/flowset/sched number"); 1314 base = buf = safe_calloc(1, lmax); 1315 /* all commands start with a 'CONFIGURE' and a version */ 1316 o_next(&buf, sizeof(struct dn_id), DN_CMD_CONFIG); 1317 base->id = DN_API_VERSION; 1318 1319 switch (g_co.do_pipe) { 1320 case 1: /* "pipe N config ..." */ 1321 /* Allocate space for the WF2Q+ scheduler, its link 1322 * and the FIFO flowset. Set the number, but leave 1323 * the scheduler subtype and other parameters to 0 1324 * so the kernel will use appropriate defaults. 1325 * XXX todo: add a flag to record if a parameter 1326 * is actually configured. 1327 * If we do a 'pipe config' mask -> sched_mask. 1328 * The FIFO scheduler and link are derived from the 1329 * WF2Q+ one in the kernel. 1330 */ 1331 #ifdef NEW_AQM 1332 sch_extra = o_next(&buf, lmax_extra, DN_TEXT); 1333 sch_extra ->oid.subtype = 0; /* don't configure scheduler */ 1334 #endif 1335 sch = o_next(&buf, sizeof(*sch), DN_SCH); 1336 p = o_next(&buf, sizeof(*p), DN_LINK); 1337 #ifdef NEW_AQM 1338 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT); 1339 aqm_extra ->oid.subtype = 0; /* don't configure AQM */ 1340 #endif 1341 fs = o_next(&buf, sizeof(*fs), DN_FS); 1342 1343 sch->sched_nr = i; 1344 sch->oid.subtype = 0; /* defaults to WF2Q+ */ 1345 mask = &sch->sched_mask; 1346 flags = &sch->flags; 1347 buckets = &sch->buckets; 1348 *flags |= DN_PIPE_CMD; 1349 1350 p->link_nr = i; 1351 1352 /* This flowset is only for the FIFO scheduler */ 1353 fs->fs_nr = i + 2*DN_MAX_ID; 1354 fs->sched_nr = i + DN_MAX_ID; 1355 break; 1356 1357 case 2: /* "queue N config ... " */ 1358 #ifdef NEW_AQM 1359 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT); 1360 aqm_extra ->oid.subtype = 0; 1361 #endif 1362 fs = o_next(&buf, sizeof(*fs), DN_FS); 1363 fs->fs_nr = i; 1364 mask = &fs->flow_mask; 1365 flags = &fs->flags; 1366 buckets = &fs->buckets; 1367 break; 1368 1369 case 3: /* "sched N config ..." */ 1370 #ifdef NEW_AQM 1371 sch_extra = o_next(&buf, lmax_extra, DN_TEXT); 1372 sch_extra ->oid.subtype = 0; 1373 #endif 1374 sch = o_next(&buf, sizeof(*sch), DN_SCH); 1375 #ifdef NEW_AQM 1376 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT); 1377 aqm_extra ->oid.subtype = 0; 1378 #endif 1379 fs = o_next(&buf, sizeof(*fs), DN_FS); 1380 sch->sched_nr = i; 1381 mask = &sch->sched_mask; 1382 flags = &sch->flags; 1383 buckets = &sch->buckets; 1384 /* fs is used only with !MULTIQUEUE schedulers */ 1385 fs->fs_nr = i + DN_MAX_ID; 1386 fs->sched_nr = i; 1387 break; 1388 } 1389 /* set to -1 those fields for which we want to reuse existing 1390 * values from the kernel. 1391 * Also, *_nr and subtype = 0 mean reuse the value from the kernel. 1392 * XXX todo: support reuse of the mask. 1393 */ 1394 if (p) 1395 p->bandwidth = -1; 1396 for (j = 0; j < sizeof(fs->par)/sizeof(fs->par[0]); j++) 1397 fs->par[j] = -1; 1398 while (ac > 0) { 1399 double d; 1400 int tok = match_token(dummynet_params, *av); 1401 ac--; av++; 1402 1403 switch(tok) { 1404 case TOK_NOERROR: 1405 NEED(fs, "noerror is only for pipes"); 1406 fs->flags |= DN_NOERROR; 1407 break; 1408 1409 case TOK_PLR: 1410 NEED(fs, "plr is only for pipes"); 1411 NEED1("plr needs argument 0..1\n"); 1412 d = strtod(av[0], NULL); 1413 if (d > 1) 1414 d = 1; 1415 else if (d < 0) 1416 d = 0; 1417 fs->plr = (int)(d*0x7fffffff); 1418 ac--; av++; 1419 break; 1420 1421 case TOK_QUEUE: 1422 NEED(fs, "queue is only for pipes or flowsets"); 1423 NEED1("queue needs queue size\n"); 1424 end = NULL; 1425 fs->qsize = strtoul(av[0], &end, 0); 1426 if (*end == 'K' || *end == 'k') { 1427 fs->flags |= DN_QSIZE_BYTES; 1428 fs->qsize *= 1024; 1429 } else if (*end == 'B' || 1430 _substrcmp2(end, "by", "bytes") == 0) { 1431 fs->flags |= DN_QSIZE_BYTES; 1432 } 1433 ac--; av++; 1434 break; 1435 1436 case TOK_BUCKETS: 1437 NEED(fs, "buckets is only for pipes or flowsets"); 1438 NEED1("buckets needs argument\n"); 1439 *buckets = strtoul(av[0], NULL, 0); 1440 ac--; av++; 1441 break; 1442 1443 case TOK_FLOW_MASK: 1444 case TOK_SCHED_MASK: 1445 case TOK_MASK: 1446 NEED(mask, "tok_mask"); 1447 NEED1("mask needs mask specifier\n"); 1448 /* 1449 * per-flow queue, mask is dst_ip, dst_port, 1450 * src_ip, src_port, proto measured in bits 1451 */ 1452 1453 bzero(mask, sizeof(*mask)); 1454 end = NULL; 1455 1456 while (ac >= 1) { 1457 uint32_t *p32 = NULL; 1458 uint16_t *p16 = NULL; 1459 uint32_t *p20 = NULL; 1460 struct in6_addr *pa6 = NULL; 1461 uint32_t a; 1462 1463 tok = match_token(dummynet_params, *av); 1464 ac--; av++; 1465 switch(tok) { 1466 case TOK_ALL: 1467 /* 1468 * special case, all bits significant 1469 * except 'extra' (the queue number) 1470 */ 1471 mask->dst_ip = ~0; 1472 mask->src_ip = ~0; 1473 mask->dst_port = ~0; 1474 mask->src_port = ~0; 1475 mask->proto = ~0; 1476 n2mask(&mask->dst_ip6, 128); 1477 n2mask(&mask->src_ip6, 128); 1478 mask->flow_id6 = ~0; 1479 *flags |= DN_HAVE_MASK; 1480 goto end_mask; 1481 1482 case TOK_QUEUE: 1483 mask->extra = ~0; 1484 *flags |= DN_HAVE_MASK; 1485 goto end_mask; 1486 1487 case TOK_DSTIP: 1488 mask->addr_type = 4; 1489 p32 = &mask->dst_ip; 1490 break; 1491 1492 case TOK_SRCIP: 1493 mask->addr_type = 4; 1494 p32 = &mask->src_ip; 1495 break; 1496 1497 case TOK_DSTIP6: 1498 mask->addr_type = 6; 1499 pa6 = &mask->dst_ip6; 1500 break; 1501 1502 case TOK_SRCIP6: 1503 mask->addr_type = 6; 1504 pa6 = &mask->src_ip6; 1505 break; 1506 1507 case TOK_FLOWID: 1508 mask->addr_type = 6; 1509 p20 = &mask->flow_id6; 1510 break; 1511 1512 case TOK_DSTPORT: 1513 p16 = &mask->dst_port; 1514 break; 1515 1516 case TOK_SRCPORT: 1517 p16 = &mask->src_port; 1518 break; 1519 1520 case TOK_PROTO: 1521 break; 1522 1523 default: 1524 ac++; av--; /* backtrack */ 1525 goto end_mask; 1526 } 1527 if (ac < 1) 1528 errx(EX_USAGE, "mask: value missing"); 1529 if (*av[0] == '/') { 1530 a = strtoul(av[0]+1, &end, 0); 1531 if (pa6 == NULL) 1532 a = (a == 32) ? ~0 : (1 << a) - 1; 1533 } else 1534 a = strtoul(av[0], &end, 0); 1535 if (p32 != NULL) 1536 *p32 = a; 1537 else if (p16 != NULL) { 1538 if (a > 0xFFFF) 1539 errx(EX_DATAERR, 1540 "port mask must be 16 bit"); 1541 *p16 = (uint16_t)a; 1542 } else if (p20 != NULL) { 1543 if (a > 0xfffff) 1544 errx(EX_DATAERR, 1545 "flow_id mask must be 20 bit"); 1546 *p20 = (uint32_t)a; 1547 } else if (pa6 != NULL) { 1548 if (a > 128) 1549 errx(EX_DATAERR, 1550 "in6addr invalid mask len"); 1551 else 1552 n2mask(pa6, a); 1553 } else { 1554 if (a > 0xFF) 1555 errx(EX_DATAERR, 1556 "proto mask must be 8 bit"); 1557 mask->proto = (uint8_t)a; 1558 } 1559 if (a != 0) 1560 *flags |= DN_HAVE_MASK; 1561 ac--; av++; 1562 } /* end while, config masks */ 1563 end_mask: 1564 break; 1565 #ifdef NEW_AQM 1566 case TOK_CODEL: 1567 case TOK_PIE: 1568 NEED(fs, "codel/pie is only for flowsets"); 1569 1570 fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED); 1571 fs->flags |= DN_IS_AQM; 1572 1573 strlcpy(aqm_extra->name, av[-1], 1574 sizeof(aqm_extra->name)); 1575 aqm_extra->oid.subtype = DN_AQM_PARAMS; 1576 1577 process_extra_parms(&ac, av, aqm_extra, tok); 1578 break; 1579 1580 case TOK_FQ_CODEL: 1581 case TOK_FQ_PIE: 1582 if (!strcmp(av[-1],"type")) 1583 errx(EX_DATAERR, "use type before fq_codel/fq_pie"); 1584 1585 NEED(sch, "fq_codel/fq_pie is only for schd"); 1586 strlcpy(sch_extra->name, av[-1], 1587 sizeof(sch_extra->name)); 1588 sch_extra->oid.subtype = DN_SCH_PARAMS; 1589 process_extra_parms(&ac, av, sch_extra, tok); 1590 break; 1591 #endif 1592 case TOK_RED: 1593 case TOK_GRED: 1594 NEED1("red/gred needs w_q/min_th/max_th/max_p\n"); 1595 fs->flags |= DN_IS_RED; 1596 if (tok == TOK_GRED) 1597 fs->flags |= DN_IS_GENTLE_RED; 1598 /* 1599 * the format for parameters is w_q/min_th/max_th/max_p 1600 */ 1601 if ((end = strsep(&av[0], "/"))) { 1602 double w_q = strtod(end, NULL); 1603 if (w_q > 1 || w_q <= 0) 1604 errx(EX_DATAERR, "0 < w_q <= 1"); 1605 fs->w_q = (int) (w_q * (1 << SCALE_RED)); 1606 } 1607 if ((end = strsep(&av[0], "/"))) { 1608 fs->min_th = strtoul(end, &end, 0); 1609 if (*end == 'K' || *end == 'k') 1610 fs->min_th *= 1024; 1611 } 1612 if ((end = strsep(&av[0], "/"))) { 1613 fs->max_th = strtoul(end, &end, 0); 1614 if (*end == 'K' || *end == 'k') 1615 fs->max_th *= 1024; 1616 } 1617 if ((end = strsep(&av[0], "/"))) { 1618 double max_p = strtod(end, NULL); 1619 if (max_p > 1 || max_p < 0) 1620 errx(EX_DATAERR, "0 <= max_p <= 1"); 1621 fs->max_p = (int)(max_p * (1 << SCALE_RED)); 1622 } 1623 ac--; av++; 1624 break; 1625 1626 case TOK_ECN: 1627 fs->flags |= DN_IS_ECN; 1628 break; 1629 1630 case TOK_DROPTAIL: 1631 NEED(fs, "droptail is only for flowsets"); 1632 fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED); 1633 break; 1634 1635 case TOK_BW: 1636 NEED(p, "bw is only for links"); 1637 NEED1("bw needs bandwidth or interface\n"); 1638 read_bandwidth(av[0], &p->bandwidth, NULL, 0); 1639 ac--; av++; 1640 break; 1641 1642 case TOK_DELAY: 1643 NEED(p, "delay is only for links"); 1644 NEED1("delay needs argument 0..10000ms\n"); 1645 p->delay = strtoul(av[0], NULL, 0); 1646 ac--; av++; 1647 break; 1648 1649 case TOK_TYPE: { 1650 int l; 1651 NEED(sch, "type is only for schedulers"); 1652 NEED1("type needs a string"); 1653 l = strlen(av[0]); 1654 if (l == 0 || l > 15) 1655 errx(1, "type %s too long\n", av[0]); 1656 strlcpy(sch->name, av[0], sizeof(sch->name)); 1657 sch->oid.subtype = 0; /* use string */ 1658 #ifdef NEW_AQM 1659 /* if fq_codel is selected, consider all tokens after it 1660 * as parameters 1661 */ 1662 if (!strcasecmp(av[0],"fq_codel") || !strcasecmp(av[0],"fq_pie")){ 1663 strlcpy(sch_extra->name, av[0], 1664 sizeof(sch_extra->name)); 1665 sch_extra->oid.subtype = DN_SCH_PARAMS; 1666 process_extra_parms(&ac, av, sch_extra, tok); 1667 } else { 1668 ac--;av++; 1669 } 1670 #else 1671 ac--;av++; 1672 #endif 1673 break; 1674 } 1675 1676 case TOK_WEIGHT: 1677 NEED(fs, "weight is only for flowsets"); 1678 NEED1("weight needs argument\n"); 1679 fs->par[0] = strtol(av[0], &end, 0); 1680 ac--; av++; 1681 break; 1682 1683 case TOK_LMAX: 1684 NEED(fs, "lmax is only for flowsets"); 1685 NEED1("lmax needs argument\n"); 1686 fs->par[1] = strtol(av[0], &end, 0); 1687 ac--; av++; 1688 break; 1689 1690 case TOK_PRI: 1691 NEED(fs, "priority is only for flowsets"); 1692 NEED1("priority needs argument\n"); 1693 fs->par[2] = strtol(av[0], &end, 0); 1694 ac--; av++; 1695 break; 1696 1697 case TOK_SCHED: 1698 case TOK_PIPE: 1699 NEED(fs, "pipe/sched"); 1700 NEED1("pipe/link/sched needs number\n"); 1701 fs->sched_nr = strtoul(av[0], &end, 0); 1702 ac--; av++; 1703 break; 1704 1705 case TOK_PROFILE: 1706 NEED((!pf), "profile already set"); 1707 NEED(p, "profile"); 1708 { 1709 NEED1("extra delay needs the file name\n"); 1710 pf = o_next(&buf, sizeof(*pf), DN_PROFILE); 1711 load_extra_delays(av[0], pf, p); //XXX can't fail? 1712 --ac; ++av; 1713 } 1714 break; 1715 1716 case TOK_BURST: 1717 NEED(p, "burst"); 1718 NEED1("burst needs argument\n"); 1719 errno = 0; 1720 if (expand_number(av[0], &p->burst) < 0) 1721 if (errno != ERANGE) 1722 errx(EX_DATAERR, 1723 "burst: invalid argument"); 1724 if (errno || p->burst > (1ULL << 48) - 1) 1725 errx(EX_DATAERR, 1726 "burst: out of range (0..2^48-1)"); 1727 ac--; av++; 1728 break; 1729 1730 default: 1731 errx(EX_DATAERR, "unrecognised option ``%s''", av[-1]); 1732 } 1733 } 1734 1735 /* check validity of parameters */ 1736 if (p) { 1737 if (p->delay > 10000) 1738 errx(EX_DATAERR, "delay must be < 10000"); 1739 if (p->bandwidth == (uint32_t)-1) 1740 p->bandwidth = 0; 1741 } 1742 if (fs) { 1743 /* XXX accept a 0 scheduler to keep the default */ 1744 if (fs->flags & DN_QSIZE_BYTES) { 1745 size_t len; 1746 long limit; 1747 1748 len = sizeof(limit); 1749 if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit", 1750 &limit, &len, NULL, 0) == -1) 1751 limit = 1024*1024; 1752 if (fs->qsize > limit) 1753 errx(EX_DATAERR, "queue size must be < %ldB", limit); 1754 } else { 1755 size_t len; 1756 long limit; 1757 1758 len = sizeof(limit); 1759 if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit", 1760 &limit, &len, NULL, 0) == -1) 1761 limit = 100; 1762 if (fs->qsize > limit) 1763 errx(EX_DATAERR, "2 <= queue size <= %ld", limit); 1764 } 1765 1766 #ifdef NEW_AQM 1767 if ((fs->flags & DN_IS_ECN) && !((fs->flags & DN_IS_RED)|| 1768 (fs->flags & DN_IS_AQM))) 1769 errx(EX_USAGE, "ECN can be used with red/gred/" 1770 "codel/fq_codel only!"); 1771 #else 1772 if ((fs->flags & DN_IS_ECN) && !(fs->flags & DN_IS_RED)) 1773 errx(EX_USAGE, "enable red/gred for ECN"); 1774 1775 #endif 1776 1777 if (fs->flags & DN_IS_RED) { 1778 size_t len; 1779 int lookup_depth, avg_pkt_size; 1780 1781 if (!(fs->flags & DN_IS_ECN) && (fs->min_th >= fs->max_th)) 1782 errx(EX_DATAERR, "min_th %d must be < than max_th %d", 1783 fs->min_th, fs->max_th); 1784 else if ((fs->flags & DN_IS_ECN) && (fs->min_th > fs->max_th)) 1785 errx(EX_DATAERR, "min_th %d must be =< than max_th %d", 1786 fs->min_th, fs->max_th); 1787 1788 if (fs->max_th == 0) 1789 errx(EX_DATAERR, "max_th must be > 0"); 1790 1791 len = sizeof(int); 1792 if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth", 1793 &lookup_depth, &len, NULL, 0) == -1) 1794 lookup_depth = 256; 1795 if (lookup_depth == 0) 1796 errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth" 1797 " must be greater than zero"); 1798 1799 len = sizeof(int); 1800 if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size", 1801 &avg_pkt_size, &len, NULL, 0) == -1) 1802 avg_pkt_size = 512; 1803 1804 if (avg_pkt_size == 0) 1805 errx(EX_DATAERR, 1806 "net.inet.ip.dummynet.red_avg_pkt_size must" 1807 " be greater than zero"); 1808 1809 #if 0 /* the following computation is now done in the kernel */ 1810 /* 1811 * Ticks needed for sending a medium-sized packet. 1812 * Unfortunately, when we are configuring a WF2Q+ queue, we 1813 * do not have bandwidth information, because that is stored 1814 * in the parent pipe, and also we have multiple queues 1815 * competing for it. So we set s=0, which is not very 1816 * correct. But on the other hand, why do we want RED with 1817 * WF2Q+ ? 1818 */ 1819 if (p.bandwidth==0) /* this is a WF2Q+ queue */ 1820 s = 0; 1821 else 1822 s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth; 1823 /* 1824 * max idle time (in ticks) before avg queue size becomes 0. 1825 * NOTA: (3/w_q) is approx the value x so that 1826 * (1-w_q)^x < 10^-3. 1827 */ 1828 w_q = ((double)fs->w_q) / (1 << SCALE_RED); 1829 idle = s * 3. / w_q; 1830 fs->lookup_step = (int)idle / lookup_depth; 1831 if (!fs->lookup_step) 1832 fs->lookup_step = 1; 1833 weight = 1 - w_q; 1834 for (t = fs->lookup_step; t > 1; --t) 1835 weight *= 1 - w_q; 1836 fs->lookup_weight = (int)(weight * (1 << SCALE_RED)); 1837 #endif /* code moved in the kernel */ 1838 } 1839 } 1840 1841 i = do_cmd(IP_DUMMYNET3, base, (char *)buf - (char *)base); 1842 1843 if (i) 1844 err(1, "setsockopt(%s)", "IP_DUMMYNET_CONFIGURE"); 1845 } 1846 1847 void 1848 dummynet_flush(void) 1849 { 1850 struct dn_id oid; 1851 oid_fill(&oid, sizeof(oid), DN_CMD_FLUSH, DN_API_VERSION); 1852 do_cmd(IP_DUMMYNET3, &oid, oid.len); 1853 } 1854 1855 /* Parse input for 'ipfw [pipe|sched|queue] show [range list]' 1856 * Returns the number of ranges, and possibly stores them 1857 * in the array v of size len. 1858 */ 1859 static int 1860 parse_range(int ac, char *av[], uint32_t *v, int len) 1861 { 1862 int n = 0; 1863 char *endptr, *s; 1864 uint32_t base[2]; 1865 1866 if (v == NULL || len < 2) { 1867 v = base; 1868 len = 2; 1869 } 1870 1871 for (s = *av; s != NULL; av++, ac--) { 1872 v[0] = strtoul(s, &endptr, 10); 1873 v[1] = (*endptr != '-') ? v[0] : 1874 strtoul(endptr+1, &endptr, 10); 1875 if (*endptr == '\0') { /* prepare for next round */ 1876 s = (ac > 0) ? *(av+1) : NULL; 1877 } else { 1878 if (*endptr != ',') { 1879 warn("invalid number: %s", s); 1880 s = ++endptr; 1881 continue; 1882 } 1883 /* continue processing from here */ 1884 s = ++endptr; 1885 ac++; 1886 av--; 1887 } 1888 if (v[1] < v[0] || 1889 v[0] >= DN_MAX_ID-1 || 1890 v[1] >= DN_MAX_ID-1) { 1891 continue; /* invalid entry */ 1892 } 1893 n++; 1894 /* translate if 'pipe list' */ 1895 if (g_co.do_pipe == 1) { 1896 v[0] += DN_MAX_ID; 1897 v[1] += DN_MAX_ID; 1898 } 1899 v = (n*2 < len) ? v + 2 : base; 1900 } 1901 return n; 1902 } 1903 1904 /* main entry point for dummynet list functions. co.do_pipe indicates 1905 * which function we want to support. 1906 * av may contain filtering arguments, either individual entries 1907 * or ranges, or lists (space or commas are valid separators). 1908 * Format for a range can be n1-n2 or n3 n4 n5 ... 1909 * In a range n1 must be <= n2, otherwise the range is ignored. 1910 * A number 'n4' is translate in a range 'n4-n4' 1911 * All number must be > 0 and < DN_MAX_ID-1 1912 */ 1913 void 1914 dummynet_list(int ac, char *av[], int show_counters) 1915 { 1916 struct dn_id *oid, *x = NULL; 1917 int ret, i; 1918 int n; /* # of ranges */ 1919 u_int buflen, l; 1920 u_int max_size; /* largest obj passed up */ 1921 1922 (void)show_counters; // XXX unused, but we should use it. 1923 ac--; 1924 av++; /* skip 'list' | 'show' word */ 1925 1926 n = parse_range(ac, av, NULL, 0); /* Count # of ranges. */ 1927 1928 /* Allocate space to store ranges */ 1929 l = sizeof(*oid) + sizeof(uint32_t) * n * 2; 1930 oid = safe_calloc(1, l); 1931 oid_fill(oid, l, DN_CMD_GET, DN_API_VERSION); 1932 1933 if (n > 0) /* store ranges in idx */ 1934 parse_range(ac, av, (uint32_t *)(oid + 1), n*2); 1935 /* 1936 * Compute the size of the largest object returned. If the 1937 * response leaves at least this much spare space in the 1938 * buffer, then surely the response is complete; otherwise 1939 * there might be a risk of truncation and we will need to 1940 * retry with a larger buffer. 1941 * XXX don't bother with smaller structs. 1942 */ 1943 max_size = sizeof(struct dn_fs); 1944 if (max_size < sizeof(struct dn_sch)) 1945 max_size = sizeof(struct dn_sch); 1946 if (max_size < sizeof(struct dn_flow)) 1947 max_size = sizeof(struct dn_flow); 1948 1949 switch (g_co.do_pipe) { 1950 case 1: 1951 oid->subtype = DN_LINK; /* list pipe */ 1952 break; 1953 case 2: 1954 oid->subtype = DN_FS; /* list queue */ 1955 break; 1956 case 3: 1957 oid->subtype = DN_SCH; /* list sched */ 1958 break; 1959 } 1960 1961 /* 1962 * Ask the kernel an estimate of the required space (result 1963 * in oid.id), unless we are requesting a subset of objects, 1964 * in which case the kernel does not give an exact answer. 1965 * In any case, space might grow in the meantime due to the 1966 * creation of new queues, so we must be prepared to retry. 1967 */ 1968 if (n > 0) { 1969 buflen = 4*1024; 1970 } else { 1971 ret = do_cmd(-IP_DUMMYNET3, oid, (uintptr_t)&l); 1972 if (ret != 0 || oid->id <= sizeof(*oid)) 1973 goto done; 1974 buflen = oid->id + max_size; 1975 oid->len = sizeof(*oid); /* restore */ 1976 } 1977 /* Try a few times, until the buffer fits */ 1978 for (i = 0; i < 20; i++) { 1979 l = buflen; 1980 x = safe_realloc(x, l); 1981 bcopy(oid, x, oid->len); 1982 ret = do_cmd(-IP_DUMMYNET3, x, (uintptr_t)&l); 1983 if (ret != 0 || x->id <= sizeof(*oid)) 1984 goto done; /* no response */ 1985 if (l + max_size <= buflen) 1986 break; /* ok */ 1987 buflen *= 2; /* double for next attempt */ 1988 } 1989 list_pipes(x, O_NEXT(x, l)); 1990 done: 1991 if (x) 1992 free(x); 1993 free(oid); 1994 } 1995