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 * $FreeBSD: stable/10/sbin/ipfw/dummynet.c 320782 2017-07-07 15:35:42Z asomers $ 21 * 22 * dummynet support 23 */ 24 25#define NEW_AQM 26#include <sys/types.h> 27#include <sys/socket.h> 28/* XXX there are several sysctl leftover here */ 29#include <sys/sysctl.h> 30 31#include "ipfw2.h" 32 33#ifdef NEW_AQM 34#include <stdint.h> 35#endif 36 37#include <ctype.h> 38#include <err.h> 39#include <errno.h> 40#include <libutil.h> 41#include <netdb.h> 42#include <stdio.h> 43#include <stdlib.h> 44#include <string.h> 45#include <sysexits.h> 46 47#include <net/if.h> 48#include <netinet/in.h> 49#include <netinet/ip_fw.h> 50#include <netinet/ip_dummynet.h> 51#include <arpa/inet.h> /* inet_ntoa */ 52 53 54static struct _s_x dummynet_params[] = { 55 { "plr", TOK_PLR }, 56 { "noerror", TOK_NOERROR }, 57 { "buckets", TOK_BUCKETS }, 58 { "dst-ip", TOK_DSTIP }, 59 { "src-ip", TOK_SRCIP }, 60 { "dst-port", TOK_DSTPORT }, 61 { "src-port", TOK_SRCPORT }, 62 { "proto", TOK_PROTO }, 63 { "weight", TOK_WEIGHT }, 64 { "lmax", TOK_LMAX }, 65 { "maxlen", TOK_LMAX }, 66 { "all", TOK_ALL }, 67 { "mask", TOK_MASK }, /* alias for both */ 68 { "sched_mask", TOK_SCHED_MASK }, 69 { "flow_mask", TOK_FLOW_MASK }, 70 { "droptail", TOK_DROPTAIL }, 71 { "ecn", TOK_ECN }, 72 { "red", TOK_RED }, 73 { "gred", TOK_GRED }, 74#ifdef NEW_AQM 75 { "codel", TOK_CODEL}, /* Codel AQM */ 76 { "fq_codel", TOK_FQ_CODEL}, /* FQ-Codel */ 77 { "pie", TOK_PIE}, /* PIE AQM */ 78 { "fq_pie", TOK_FQ_PIE}, /* FQ-PIE */ 79#endif 80 { "bw", TOK_BW }, 81 { "bandwidth", TOK_BW }, 82 { "delay", TOK_DELAY }, 83 { "link", TOK_LINK }, 84 { "pipe", TOK_PIPE }, 85 { "queue", TOK_QUEUE }, 86 { "flowset", TOK_FLOWSET }, 87 { "sched", TOK_SCHED }, 88 { "pri", TOK_PRI }, 89 { "priority", TOK_PRI }, 90 { "type", TOK_TYPE }, 91 { "flow-id", TOK_FLOWID}, 92 { "dst-ipv6", TOK_DSTIP6}, 93 { "dst-ip6", TOK_DSTIP6}, 94 { "src-ipv6", TOK_SRCIP6}, 95 { "src-ip6", TOK_SRCIP6}, 96 { "profile", TOK_PROFILE}, 97 { "burst", TOK_BURST}, 98 { "dummynet-params", TOK_NULL }, 99 { NULL, 0 } /* terminator */ 100}; 101 102#ifdef NEW_AQM 103/* AQM/extra sched parameters tokens*/ 104static struct _s_x aqm_params[] = { 105 { "target", TOK_TARGET}, 106 { "interval", TOK_INTERVAL}, 107 { "limit", TOK_LIMIT}, 108 { "flows", TOK_FLOWS}, 109 { "quantum", TOK_QUANTUM}, 110 { "ecn", TOK_ECN}, 111 { "noecn", TOK_NO_ECN}, 112 { "tupdate", TOK_TUPDATE}, 113 { "max_burst", TOK_MAX_BURST}, 114 { "max_ecnth", TOK_MAX_ECNTH}, 115 { "alpha", TOK_ALPHA}, 116 { "beta", TOK_BETA}, 117 { "capdrop", TOK_CAPDROP}, 118 { "nocapdrop", TOK_NO_CAPDROP}, 119 { "onoff", TOK_ONOFF}, 120 { "dre", TOK_DRE}, 121 { "ts", TOK_TS}, 122 { "derand", TOK_DERAND}, 123 { "noderand", TOK_NO_DERAND}, 124 { NULL, 0 } /* terminator */ 125}; 126#endif 127 128#define O_NEXT(p, len) ((void *)((char *)p + len)) 129 130static void 131oid_fill(struct dn_id *oid, int len, int type, uintptr_t id) 132{ 133 oid->len = len; 134 oid->type = type; 135 oid->subtype = 0; 136 oid->id = id; 137} 138 139/* make room in the buffer and move the pointer forward */ 140static void * 141o_next(struct dn_id **o, int len, int type) 142{ 143 struct dn_id *ret = *o; 144 oid_fill(ret, len, type, 0); 145 *o = O_NEXT(*o, len); 146 return ret; 147} 148 149#ifdef NEW_AQM 150 151/* Codel flags */ 152enum { 153 CODEL_ECN_ENABLED = 1 154}; 155 156/* PIE flags, from PIE kernel module */ 157enum { 158 PIE_ECN_ENABLED = 1, 159 PIE_CAPDROP_ENABLED = 2, 160 PIE_ON_OFF_MODE_ENABLED = 4, 161 PIE_DEPRATEEST_ENABLED = 8, 162 PIE_DERAND_ENABLED = 16 163}; 164 165#define PIE_FIX_POINT_BITS 13 166#define PIE_SCALE (1L<<PIE_FIX_POINT_BITS) 167 168/* integer to time */ 169void 170us_to_time(int t,char *strt) 171{ 172 if (t < 0) 173 strt[0]='\0'; 174 else if ( t==0 ) 175 sprintf(strt,"%d", t); 176 else if (t< 1000) 177 sprintf(strt,"%dus", t); 178 else if (t < 1000000) 179 sprintf(strt,"%gms", (float) t / 1000); 180 else 181 sprintf(strt,"%gfs", (float) t / 1000000); 182} 183 184/* 185 * returns -1 if s is not a valid time, otherwise, return time in us 186 */ 187static long 188time_to_us(const char *s) 189{ 190 int i, dots = 0; 191 int len = strlen(s); 192 char strt[16]="", stru[16]=""; 193 194 if (len>15) 195 return -1; 196 for (i = 0; i<len && (isdigit(s[i]) || s[i]=='.') ; i++) 197 if (s[i]=='.') { 198 if (dots) 199 return -1; 200 else 201 dots++; 202 } 203 204 if (!i) 205 return -1; 206 strncpy(strt, s, i); 207 if (i<len) 208 strcpy(stru, s+i); 209 else 210 strcpy(stru, "ms"); 211 212 if (!strcasecmp(stru, "us")) 213 return atol(strt); 214 if (!strcasecmp(stru, "ms")) 215 return (strtod(strt, NULL) * 1000); 216 if (!strcasecmp(stru, "s")) 217 return (strtod(strt, NULL)*1000000); 218 219 return -1; 220} 221 222 223/* Get AQM or scheduler extra parameters */ 224void 225get_extra_parms(uint32_t nr, char *out, int subtype) 226{ 227 struct dn_extra_parms *ep; 228 int ret; 229 char strt1[15], strt2[15], strt3[15]; 230 u_int l; 231 232 /* prepare the request */ 233 l = sizeof(struct dn_extra_parms); 234 ep = safe_calloc(1, l); 235 memset(ep, 0, sizeof(*ep)); 236 *out = '\0'; 237 238 oid_fill(&ep->oid, l, DN_CMD_GET, DN_API_VERSION); 239 ep->oid.len = l; 240 ep->oid.subtype = subtype; 241 ep->nr = nr; 242 243 ret = do_cmd(-IP_DUMMYNET3, ep, (uintptr_t)&l); 244 if (ret) { 245 free(ep); 246 errx(EX_DATAERR, "Error getting extra parameters\n"); 247 } 248 249 switch (subtype) { 250 case DN_AQM_PARAMS: 251 if( !strcasecmp(ep->name, "codel")) { 252 us_to_time(ep->par[0], strt1); 253 us_to_time(ep->par[1], strt2); 254 l = sprintf(out, " AQM CoDel target %s interval %s", 255 strt1, strt2); 256 if (ep->par[2] & CODEL_ECN_ENABLED) 257 l = sprintf(out + l, " ECN"); 258 else 259 l += sprintf(out + l, " NoECN"); 260 } else if( !strcasecmp(ep->name, "pie")) { 261 us_to_time(ep->par[0], strt1); 262 us_to_time(ep->par[1], strt2); 263 us_to_time(ep->par[2], strt3); 264 l = sprintf(out, " AQM type PIE target %s tupdate %s alpha " 265 "%g beta %g max_burst %s max_ecnth %.3g", 266 strt1, 267 strt2, 268 ep->par[4] / (float) PIE_SCALE, 269 ep->par[5] / (float) PIE_SCALE, 270 strt3, 271 ep->par[3] / (float) PIE_SCALE 272 ); 273 274 if (ep->par[6] & PIE_ECN_ENABLED) 275 l += sprintf(out + l, " ECN"); 276 else 277 l += sprintf(out + l, " NoECN"); 278 if (ep->par[6] & PIE_CAPDROP_ENABLED) 279 l += sprintf(out + l, " CapDrop"); 280 else 281 l += sprintf(out + l, " NoCapDrop"); 282 if (ep->par[6] & PIE_ON_OFF_MODE_ENABLED) 283 l += sprintf(out + l, " OnOff"); 284 if (ep->par[6] & PIE_DEPRATEEST_ENABLED) 285 l += sprintf(out + l, " DRE"); 286 else 287 l += sprintf(out + l, " TS"); 288 if (ep->par[6] & PIE_DERAND_ENABLED) 289 l += sprintf(out + l, " Derand"); 290 else 291 l += sprintf(out + l, " NoDerand"); 292 } 293 break; 294 295 case DN_SCH_PARAMS: 296 if (!strcasecmp(ep->name,"FQ_CODEL")) { 297 us_to_time(ep->par[0], strt1); 298 us_to_time(ep->par[1], strt2); 299 l = sprintf(out," FQ_CODEL target %s interval %s" 300 " quantum %jd limit %jd flows %jd", 301 strt1, strt2, 302 (intmax_t) ep->par[3], 303 (intmax_t) ep->par[4], 304 (intmax_t) ep->par[5] 305 ); 306 if (ep->par[2] & CODEL_ECN_ENABLED) 307 l += sprintf(out + l, " ECN"); 308 else 309 l += sprintf(out + l, " NoECN"); 310 l += sprintf(out + l, "\n"); 311 } else if (!strcasecmp(ep->name,"FQ_PIE")) { 312 us_to_time(ep->par[0], strt1); 313 us_to_time(ep->par[1], strt2); 314 us_to_time(ep->par[2], strt3); 315 l = sprintf(out, " FQ_PIE target %s tupdate %s alpha " 316 "%g beta %g max_burst %s max_ecnth %.3g" 317 " quantum %jd limit %jd flows %jd", 318 strt1, 319 strt2, 320 ep->par[4] / (float) PIE_SCALE, 321 ep->par[5] / (float) PIE_SCALE, 322 strt3, 323 ep->par[3] / (float) PIE_SCALE, 324 (intmax_t) ep->par[7], 325 (intmax_t) ep->par[8], 326 (intmax_t) ep->par[9] 327 ); 328 329 if (ep->par[6] & PIE_ECN_ENABLED) 330 l += sprintf(out + l, " ECN"); 331 else 332 l += sprintf(out + l, " NoECN"); 333 if (ep->par[6] & PIE_CAPDROP_ENABLED) 334 l += sprintf(out + l, " CapDrop"); 335 else 336 l += sprintf(out + l, " NoCapDrop"); 337 if (ep->par[6] & PIE_ON_OFF_MODE_ENABLED) 338 l += sprintf(out + l, " OnOff"); 339 if (ep->par[6] & PIE_DEPRATEEST_ENABLED) 340 l += sprintf(out + l, " DRE"); 341 else 342 l += sprintf(out + l, " TS"); 343 if (ep->par[6] & PIE_DERAND_ENABLED) 344 l += sprintf(out + l, " Derand"); 345 else 346 l += sprintf(out + l, " NoDerand"); 347 l += sprintf(out + l, "\n"); 348 } 349 break; 350 } 351 352 free(ep); 353} 354#endif 355 356 357#if 0 358static int 359sort_q(void *arg, const void *pa, const void *pb) 360{ 361 int rev = (co.do_sort < 0); 362 int field = rev ? -co.do_sort : co.do_sort; 363 long long res = 0; 364 const struct dn_flow_queue *a = pa; 365 const struct dn_flow_queue *b = pb; 366 367 switch (field) { 368 case 1: /* pkts */ 369 res = a->len - b->len; 370 break; 371 case 2: /* bytes */ 372 res = a->len_bytes - b->len_bytes; 373 break; 374 375 case 3: /* tot pkts */ 376 res = a->tot_pkts - b->tot_pkts; 377 break; 378 379 case 4: /* tot bytes */ 380 res = a->tot_bytes - b->tot_bytes; 381 break; 382 } 383 if (res < 0) 384 res = -1; 385 if (res > 0) 386 res = 1; 387 return (int)(rev ? res : -res); 388} 389#endif 390 391/* print a mask and header for the subsequent list of flows */ 392static void 393print_mask(struct ipfw_flow_id *id) 394{ 395 if (!IS_IP6_FLOW_ID(id)) { 396 printf(" " 397 "mask: %s 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n", 398 id->extra ? "queue," : "", 399 id->proto, 400 id->src_ip, id->src_port, 401 id->dst_ip, id->dst_port); 402 } else { 403 char buf[255]; 404 printf("\n mask: %sproto: 0x%02x, flow_id: 0x%08x, ", 405 id->extra ? "queue," : "", 406 id->proto, id->flow_id6); 407 inet_ntop(AF_INET6, &(id->src_ip6), buf, sizeof(buf)); 408 printf("%s/0x%04x -> ", buf, id->src_port); 409 inet_ntop(AF_INET6, &(id->dst_ip6), buf, sizeof(buf)); 410 printf("%s/0x%04x\n", buf, id->dst_port); 411 } 412} 413 414static void 415print_header(struct ipfw_flow_id *id) 416{ 417 if (!IS_IP6_FLOW_ID(id)) 418 printf("BKT Prot ___Source IP/port____ " 419 "____Dest. IP/port____ " 420 "Tot_pkt/bytes Pkt/Byte Drp\n"); 421 else 422 printf("BKT ___Prot___ _flow-id_ " 423 "______________Source IPv6/port_______________ " 424 "_______________Dest. IPv6/port_______________ " 425 "Tot_pkt/bytes Pkt/Byte Drp\n"); 426} 427 428static void 429list_flow(struct dn_flow *ni, int *print) 430{ 431 char buff[255]; 432 struct protoent *pe = NULL; 433 struct in_addr ina; 434 struct ipfw_flow_id *id = &ni->fid; 435 436 if (*print) { 437 print_header(&ni->fid); 438 *print = 0; 439 } 440 pe = getprotobynumber(id->proto); 441 /* XXX: Should check for IPv4 flows */ 442 printf("%3u%c", (ni->oid.id) & 0xff, 443 id->extra ? '*' : ' '); 444 if (!IS_IP6_FLOW_ID(id)) { 445 if (pe) 446 printf("%-4s ", pe->p_name); 447 else 448 printf("%4u ", id->proto); 449 ina.s_addr = htonl(id->src_ip); 450 printf("%15s/%-5d ", 451 inet_ntoa(ina), id->src_port); 452 ina.s_addr = htonl(id->dst_ip); 453 printf("%15s/%-5d ", 454 inet_ntoa(ina), id->dst_port); 455 } else { 456 /* Print IPv6 flows */ 457 if (pe != NULL) 458 printf("%9s ", pe->p_name); 459 else 460 printf("%9u ", id->proto); 461 printf("%7d %39s/%-5d ", id->flow_id6, 462 inet_ntop(AF_INET6, &(id->src_ip6), buff, sizeof(buff)), 463 id->src_port); 464 printf(" %39s/%-5d ", 465 inet_ntop(AF_INET6, &(id->dst_ip6), buff, sizeof(buff)), 466 id->dst_port); 467 } 468 pr_u64(&ni->tot_pkts, 4); 469 pr_u64(&ni->tot_bytes, 8); 470 printf("%2u %4u %3u\n", 471 ni->length, ni->len_bytes, ni->drops); 472} 473 474static void 475print_flowset_parms(struct dn_fs *fs, char *prefix) 476{ 477 int l; 478 char qs[30]; 479 char plr[30]; 480 char red[200]; /* Display RED parameters */ 481 482 l = fs->qsize; 483 if (fs->flags & DN_QSIZE_BYTES) { 484 if (l >= 8192) 485 sprintf(qs, "%d KB", l / 1024); 486 else 487 sprintf(qs, "%d B", l); 488 } else 489 sprintf(qs, "%3d sl.", l); 490 if (fs->plr) 491 sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff)); 492 else 493 plr[0] = '\0'; 494 495 if (fs->flags & DN_IS_RED) { /* RED parameters */ 496 sprintf(red, 497 "\n\t %cRED w_q %f min_th %d max_th %d max_p %f", 498 (fs->flags & DN_IS_GENTLE_RED) ? 'G' : ' ', 499 1.0 * fs->w_q / (double)(1 << SCALE_RED), 500 fs->min_th, 501 fs->max_th, 502 1.0 * fs->max_p / (double)(1 << SCALE_RED)); 503 if (fs->flags & DN_IS_ECN) 504 strncat(red, " (ecn)", 6); 505#ifdef NEW_AQM 506 /* get AQM parameters */ 507 } else if (fs->flags & DN_IS_AQM) { 508 get_extra_parms(fs->fs_nr, red, DN_AQM_PARAMS); 509#endif 510 } else 511 sprintf(red, "droptail"); 512 513 if (prefix[0]) { 514 printf("%s %s%s %d queues (%d buckets) %s\n", 515 prefix, qs, plr, fs->oid.id, fs->buckets, red); 516 prefix[0] = '\0'; 517 } else { 518 printf("q%05d %s%s %d flows (%d buckets) sched %d " 519 "weight %d lmax %d pri %d %s\n", 520 fs->fs_nr, qs, plr, fs->oid.id, fs->buckets, 521 fs->sched_nr, fs->par[0], fs->par[1], fs->par[2], red); 522 if (fs->flags & DN_HAVE_MASK) 523 print_mask(&fs->flow_mask); 524 } 525} 526 527static void 528print_extra_delay_parms(struct dn_profile *p) 529{ 530 double loss; 531 if (p->samples_no <= 0) 532 return; 533 534 loss = p->loss_level; 535 loss /= p->samples_no; 536 printf("\t profile: name \"%s\" loss %f samples %d\n", 537 p->name, loss, p->samples_no); 538} 539 540static void 541flush_buf(char *buf) 542{ 543 if (buf[0]) 544 printf("%s\n", buf); 545 buf[0] = '\0'; 546} 547 548/* 549 * generic list routine. We expect objects in a specific order, i.e. 550 * PIPES AND SCHEDULERS: 551 * link; scheduler; internal flowset if any; instances 552 * we can tell a pipe from the number. 553 * 554 * FLOWSETS: 555 * flowset; queues; 556 * link i (int queue); scheduler i; si(i) { flowsets() : queues } 557 */ 558static void 559list_pipes(struct dn_id *oid, struct dn_id *end) 560{ 561 char buf[160]; /* pending buffer */ 562 int toPrint = 1; /* print header */ 563 564 buf[0] = '\0'; 565 for (; oid != end; oid = O_NEXT(oid, oid->len)) { 566 if (oid->len < sizeof(*oid)) 567 errx(1, "invalid oid len %d\n", oid->len); 568 569 switch (oid->type) { 570 default: 571 flush_buf(buf); 572 printf("unrecognized object %d size %d\n", oid->type, oid->len); 573 break; 574 case DN_TEXT: /* list of attached flowsets */ 575 { 576 int i, l; 577 struct { 578 struct dn_id id; 579 uint32_t p[0]; 580 } *d = (void *)oid; 581 l = (oid->len - sizeof(*oid))/sizeof(d->p[0]); 582 if (l == 0) 583 break; 584 printf(" Children flowsets: "); 585 for (i = 0; i < l; i++) 586 printf("%u ", d->p[i]); 587 printf("\n"); 588 break; 589 } 590 case DN_CMD_GET: 591 if (co.verbose) 592 printf("answer for cmd %d, len %d\n", oid->type, oid->id); 593 break; 594 case DN_SCH: { 595 struct dn_sch *s = (struct dn_sch *)oid; 596 flush_buf(buf); 597 printf(" sched %d type %s flags 0x%x %d buckets %d active\n", 598 s->sched_nr, 599 s->name, s->flags, s->buckets, s->oid.id); 600#ifdef NEW_AQM 601 char parms[200]; 602 get_extra_parms(s->sched_nr, parms, DN_SCH_PARAMS); 603 printf("%s",parms); 604#endif 605 if (s->flags & DN_HAVE_MASK) 606 print_mask(&s->sched_mask); 607 } 608 break; 609 610 case DN_FLOW: 611 list_flow((struct dn_flow *)oid, &toPrint); 612 break; 613 614 case DN_LINK: { 615 struct dn_link *p = (struct dn_link *)oid; 616 double b = p->bandwidth; 617 char bwbuf[30]; 618 char burst[5 + 7]; 619 620 /* This starts a new object so flush buffer */ 621 flush_buf(buf); 622 /* data rate */ 623 if (b == 0) 624 sprintf(bwbuf, "unlimited "); 625 else if (b >= 1000000) 626 sprintf(bwbuf, "%7.3f Mbit/s", b/1000000); 627 else if (b >= 1000) 628 sprintf(bwbuf, "%7.3f Kbit/s", b/1000); 629 else 630 sprintf(bwbuf, "%7.3f bit/s ", b); 631 632 if (humanize_number(burst, sizeof(burst), p->burst, 633 "", HN_AUTOSCALE, 0) < 0 || co.verbose) 634 sprintf(burst, "%d", (int)p->burst); 635 sprintf(buf, "%05d: %s %4d ms burst %s", 636 p->link_nr % DN_MAX_ID, bwbuf, p->delay, burst); 637 } 638 break; 639 640 case DN_FS: 641 print_flowset_parms((struct dn_fs *)oid, buf); 642 break; 643 case DN_PROFILE: 644 flush_buf(buf); 645 print_extra_delay_parms((struct dn_profile *)oid); 646 } 647 flush_buf(buf); // XXX does it really go here ? 648 } 649} 650 651/* 652 * Delete pipe, queue or scheduler i 653 */ 654int 655ipfw_delete_pipe(int do_pipe, int i) 656{ 657 struct { 658 struct dn_id oid; 659 uintptr_t a[1]; /* add more if we want a list */ 660 } cmd; 661 oid_fill((void *)&cmd, sizeof(cmd), DN_CMD_DELETE, DN_API_VERSION); 662 cmd.oid.subtype = (do_pipe == 1) ? DN_LINK : 663 ( (do_pipe == 2) ? DN_FS : DN_SCH); 664 cmd.a[0] = i; 665 i = do_cmd(IP_DUMMYNET3, &cmd, cmd.oid.len); 666 if (i) { 667 i = 1; 668 warn("rule %u: setsockopt(IP_DUMMYNET_DEL)", i); 669 } 670 return i; 671} 672 673/* 674 * Code to parse delay profiles. 675 * 676 * Some link types introduce extra delays in the transmission 677 * of a packet, e.g. because of MAC level framing, contention on 678 * the use of the channel, MAC level retransmissions and so on. 679 * From our point of view, the channel is effectively unavailable 680 * for this extra time, which is constant or variable depending 681 * on the link type. Additionally, packets may be dropped after this 682 * time (e.g. on a wireless link after too many retransmissions). 683 * We can model the additional delay with an empirical curve 684 * that represents its distribution. 685 * 686 * cumulative probability 687 * 1.0 ^ 688 * | 689 * L +-- loss-level x 690 * | ****** 691 * | * 692 * | ***** 693 * | * 694 * | ** 695 * | * 696 * +-------*-------------------> 697 * delay 698 * 699 * The empirical curve may have both vertical and horizontal lines. 700 * Vertical lines represent constant delay for a range of 701 * probabilities; horizontal lines correspond to a discontinuty 702 * in the delay distribution: the link will use the largest delay 703 * for a given probability. 704 * 705 * To pass the curve to dummynet, we must store the parameters 706 * in a file as described below, and issue the command 707 * 708 * ipfw pipe <n> config ... bw XXX profile <filename> ... 709 * 710 * The file format is the following, with whitespace acting as 711 * a separator and '#' indicating the beginning a comment: 712 * 713 * samples N 714 * the number of samples used in the internal 715 * representation (2..1024; default 100); 716 * 717 * loss-level L 718 * The probability above which packets are lost. 719 * (0.0 <= L <= 1.0, default 1.0 i.e. no loss); 720 * 721 * name identifier 722 * Optional a name (listed by "ipfw pipe show") 723 * to identify the distribution; 724 * 725 * "delay prob" | "prob delay" 726 * One of these two lines is mandatory and defines 727 * the format of the following lines with data points. 728 * 729 * XXX YYY 730 * 2 or more lines representing points in the curve, 731 * with either delay or probability first, according 732 * to the chosen format. 733 * The unit for delay is milliseconds. 734 * 735 * Data points does not need to be ordered or equal to the number 736 * specified in the "samples" line. ipfw will sort and interpolate 737 * the curve as needed. 738 * 739 * Example of a profile file: 740 741 name bla_bla_bla 742 samples 100 743 loss-level 0.86 744 prob delay 745 0 200 # minimum overhead is 200ms 746 0.5 200 747 0.5 300 748 0.8 1000 749 0.9 1300 750 1 1300 751 752 * Internally, we will convert the curve to a fixed number of 753 * samples, and when it is time to transmit a packet we will 754 * model the extra delay as extra bits in the packet. 755 * 756 */ 757 758#define ED_MAX_LINE_LEN 256+ED_MAX_NAME_LEN 759#define ED_TOK_SAMPLES "samples" 760#define ED_TOK_LOSS "loss-level" 761#define ED_TOK_NAME "name" 762#define ED_TOK_DELAY "delay" 763#define ED_TOK_PROB "prob" 764#define ED_TOK_BW "bw" 765#define ED_SEPARATORS " \t\n" 766#define ED_MIN_SAMPLES_NO 2 767 768/* 769 * returns 1 if s is a non-negative number, with at least one '.' 770 */ 771static int 772is_valid_number(const char *s) 773{ 774 int i, dots_found = 0; 775 int len = strlen(s); 776 777 for (i = 0; i<len; ++i) 778 if (!isdigit(s[i]) && (s[i] !='.' || ++dots_found > 1)) 779 return 0; 780 return 1; 781} 782 783/* 784 * Take as input a string describing a bandwidth value 785 * and return the numeric bandwidth value. 786 * set clocking interface or bandwidth value 787 */ 788static void 789read_bandwidth(char *arg, int *bandwidth, char *if_name, int namelen) 790{ 791 if (*bandwidth != -1) 792 warnx("duplicate token, override bandwidth value!"); 793 794 if (arg[0] >= 'a' && arg[0] <= 'z') { 795 if (!if_name) { 796 errx(1, "no if support"); 797 } 798 if (namelen >= IFNAMSIZ) 799 warn("interface name truncated"); 800 namelen--; 801 /* interface name */ 802 strlcpy(if_name, arg, namelen); 803 *bandwidth = 0; 804 } else { /* read bandwidth value */ 805 int bw; 806 char *end = NULL; 807 808 bw = strtoul(arg, &end, 0); 809 if (*end == 'K' || *end == 'k') { 810 end++; 811 bw *= 1000; 812 } else if (*end == 'M' || *end == 'm') { 813 end++; 814 bw *= 1000000; 815 } 816 if ((*end == 'B' && 817 _substrcmp2(end, "Bi", "Bit/s") != 0) || 818 _substrcmp2(end, "by", "bytes") == 0) 819 bw *= 8; 820 821 if (bw < 0) 822 errx(EX_DATAERR, "bandwidth too large"); 823 824 *bandwidth = bw; 825 if (if_name) 826 if_name[0] = '\0'; 827 } 828} 829 830struct point { 831 double prob; 832 double delay; 833}; 834 835static int 836compare_points(const void *vp1, const void *vp2) 837{ 838 const struct point *p1 = vp1; 839 const struct point *p2 = vp2; 840 double res = 0; 841 842 res = p1->prob - p2->prob; 843 if (res == 0) 844 res = p1->delay - p2->delay; 845 if (res < 0) 846 return -1; 847 else if (res > 0) 848 return 1; 849 else 850 return 0; 851} 852 853#define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno 854 855static void 856load_extra_delays(const char *filename, struct dn_profile *p, 857 struct dn_link *link) 858{ 859 char line[ED_MAX_LINE_LEN]; 860 FILE *f; 861 int lineno = 0; 862 int i; 863 864 int samples = -1; 865 double loss = -1.0; 866 char profile_name[ED_MAX_NAME_LEN]; 867 int delay_first = -1; 868 int do_points = 0; 869 struct point points[ED_MAX_SAMPLES_NO]; 870 int points_no = 0; 871 872 /* XXX link never NULL? */ 873 p->link_nr = link->link_nr; 874 875 profile_name[0] = '\0'; 876 f = fopen(filename, "r"); 877 if (f == NULL) 878 err(EX_UNAVAILABLE, "fopen: %s", filename); 879 880 while (fgets(line, ED_MAX_LINE_LEN, f)) { /* read commands */ 881 char *s, *cur = line, *name = NULL, *arg = NULL; 882 883 ++lineno; 884 885 /* parse the line */ 886 while (cur) { 887 s = strsep(&cur, ED_SEPARATORS); 888 if (s == NULL || *s == '#') 889 break; 890 if (*s == '\0') 891 continue; 892 if (arg) 893 errx(ED_EFMT("too many arguments")); 894 if (name == NULL) 895 name = s; 896 else 897 arg = s; 898 } 899 if (name == NULL) /* empty line */ 900 continue; 901 if (arg == NULL) 902 errx(ED_EFMT("missing arg for %s"), name); 903 904 if (!strcasecmp(name, ED_TOK_SAMPLES)) { 905 if (samples > 0) 906 errx(ED_EFMT("duplicate ``samples'' line")); 907 if (atoi(arg) <=0) 908 errx(ED_EFMT("invalid number of samples")); 909 samples = atoi(arg); 910 if (samples>ED_MAX_SAMPLES_NO) 911 errx(ED_EFMT("too many samples, maximum is %d"), 912 ED_MAX_SAMPLES_NO); 913 do_points = 0; 914 } else if (!strcasecmp(name, ED_TOK_BW)) { 915 char buf[IFNAMSIZ]; 916 read_bandwidth(arg, &link->bandwidth, buf, sizeof(buf)); 917 } else if (!strcasecmp(name, ED_TOK_LOSS)) { 918 if (loss != -1.0) 919 errx(ED_EFMT("duplicated token: %s"), name); 920 if (!is_valid_number(arg)) 921 errx(ED_EFMT("invalid %s"), arg); 922 loss = atof(arg); 923 if (loss > 1) 924 errx(ED_EFMT("%s greater than 1.0"), name); 925 do_points = 0; 926 } else if (!strcasecmp(name, ED_TOK_NAME)) { 927 if (profile_name[0] != '\0') 928 errx(ED_EFMT("duplicated token: %s"), name); 929 strlcpy(profile_name, arg, sizeof(profile_name)); 930 do_points = 0; 931 } else if (!strcasecmp(name, ED_TOK_DELAY)) { 932 if (do_points) 933 errx(ED_EFMT("duplicated token: %s"), name); 934 delay_first = 1; 935 do_points = 1; 936 } else if (!strcasecmp(name, ED_TOK_PROB)) { 937 if (do_points) 938 errx(ED_EFMT("duplicated token: %s"), name); 939 delay_first = 0; 940 do_points = 1; 941 } else if (do_points) { 942 if (!is_valid_number(name) || !is_valid_number(arg)) 943 errx(ED_EFMT("invalid point found")); 944 if (delay_first) { 945 points[points_no].delay = atof(name); 946 points[points_no].prob = atof(arg); 947 } else { 948 points[points_no].delay = atof(arg); 949 points[points_no].prob = atof(name); 950 } 951 if (points[points_no].prob > 1.0) 952 errx(ED_EFMT("probability greater than 1.0")); 953 ++points_no; 954 } else { 955 errx(ED_EFMT("unrecognised command '%s'"), name); 956 } 957 } 958 959 fclose (f); 960 961 if (samples == -1) { 962 warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES); 963 samples = 100; 964 } 965 966 if (loss == -1.0) { 967 warnx("'%s' not found, assuming no loss", ED_TOK_LOSS); 968 loss = 1; 969 } 970 971 /* make sure that there are enough points. */ 972 if (points_no < ED_MIN_SAMPLES_NO) 973 errx(ED_EFMT("too few samples, need at least %d"), 974 ED_MIN_SAMPLES_NO); 975 976 qsort(points, points_no, sizeof(struct point), compare_points); 977 978 /* interpolation */ 979 for (i = 0; i<points_no-1; ++i) { 980 double y1 = points[i].prob * samples; 981 double x1 = points[i].delay; 982 double y2 = points[i+1].prob * samples; 983 double x2 = points[i+1].delay; 984 985 int ix = y1; 986 int stop = y2; 987 988 if (x1 == x2) { 989 for (; ix<stop; ++ix) 990 p->samples[ix] = x1; 991 } else { 992 double m = (y2-y1)/(x2-x1); 993 double c = y1 - m*x1; 994 for (; ix<stop ; ++ix) 995 p->samples[ix] = (ix - c)/m; 996 } 997 } 998 p->samples_no = samples; 999 p->loss_level = loss * samples; 1000 strlcpy(p->name, profile_name, sizeof(p->name)); 1001} 1002 1003#ifdef NEW_AQM 1004 1005/* Parse AQM/extra scheduler parameters */ 1006static int 1007process_extra_parms(int *ac, char **av, struct dn_extra_parms *ep, 1008 uint16_t type) 1009{ 1010 int i; 1011 1012 /* use kernel defaults */ 1013 for (i=0; i<DN_MAX_EXTRA_PARM; i++) 1014 ep->par[i] = -1; 1015 1016 switch(type) { 1017 case TOK_CODEL: 1018 case TOK_FQ_CODEL: 1019 /* Codel 1020 * 0- target, 1- interval, 2- flags, 1021 * FQ_CODEL 1022 * 3- quantum, 4- limit, 5- flows 1023 */ 1024 if (type==TOK_CODEL) 1025 ep->par[2] = 0; 1026 else 1027 ep->par[2] = CODEL_ECN_ENABLED; 1028 1029 while (*ac > 0) { 1030 int tok = match_token(aqm_params, *av); 1031 (*ac)--; av++; 1032 switch(tok) { 1033 case TOK_TARGET: 1034 if (*ac <= 0 || time_to_us(av[0]) < 0) 1035 errx(EX_DATAERR, "target needs time\n"); 1036 1037 ep->par[0] = time_to_us(av[0]); 1038 (*ac)--; av++; 1039 break; 1040 1041 case TOK_INTERVAL: 1042 if (*ac <= 0 || time_to_us(av[0]) < 0) 1043 errx(EX_DATAERR, "interval needs time\n"); 1044 1045 ep->par[1] = time_to_us(av[0]); 1046 (*ac)--; av++; 1047 break; 1048 1049 case TOK_ECN: 1050 ep->par[2] = CODEL_ECN_ENABLED; 1051 break; 1052 case TOK_NO_ECN: 1053 ep->par[2] &= ~CODEL_ECN_ENABLED; 1054 break; 1055 /* Config fq_codel parameters */ 1056 case TOK_QUANTUM: 1057 if (type != TOK_FQ_CODEL) 1058 errx(EX_DATAERR, "quantum is not for codel\n"); 1059 if (*ac <= 0 || !is_valid_number(av[0])) 1060 errx(EX_DATAERR, "quantum needs number\n"); 1061 1062 ep->par[3]= atoi(av[0]); 1063 (*ac)--; av++; 1064 break; 1065 1066 case TOK_LIMIT: 1067 if (type != TOK_FQ_CODEL) 1068 errx(EX_DATAERR, "limit is not for codel, use queue instead\n"); 1069 if (*ac <= 0 || !is_valid_number(av[0])) 1070 errx(EX_DATAERR, "limit needs number\n"); 1071 1072 ep->par[4] = atoi(av[0]); 1073 (*ac)--; av++; 1074 break; 1075 1076 case TOK_FLOWS: 1077 if (type != TOK_FQ_CODEL) 1078 errx(EX_DATAERR, "flows is not for codel\n"); 1079 if (*ac <= 0 || !is_valid_number(av[0])) 1080 errx(EX_DATAERR, "flows needs number\n"); 1081 1082 ep->par[5] = atoi(av[0]); 1083 (*ac)--; av++; 1084 break; 1085 1086 default: 1087 printf("%s is Invalid parameter\n", av[-1]); 1088 } 1089 } 1090 break; 1091 case TOK_PIE: 1092 case TOK_FQ_PIE: 1093 /* PIE 1094 * 0- target , 1- tupdate, 2- max_burst, 1095 * 3- max_ecnth, 4- alpha, 1096 * 5- beta, 6- flags 1097 * FQ_CODEL 1098 * 7- quantum, 8- limit, 9- flows 1099 */ 1100 1101 if ( type == TOK_PIE) 1102 ep->par[6] = PIE_CAPDROP_ENABLED | PIE_DEPRATEEST_ENABLED 1103 | PIE_DERAND_ENABLED; 1104 else 1105 /* for FQ-PIE, use TS mode */ 1106 ep->par[6] = PIE_CAPDROP_ENABLED | PIE_DERAND_ENABLED 1107 | PIE_ECN_ENABLED; 1108 1109 while (*ac > 0) { 1110 int tok = match_token(aqm_params, *av); 1111 (*ac)--; av++; 1112 switch(tok) { 1113 case TOK_TARGET: 1114 if (*ac <= 0 || time_to_us(av[0]) < 0) 1115 errx(EX_DATAERR, "target needs time\n"); 1116 1117 ep->par[0] = time_to_us(av[0]); 1118 (*ac)--; av++; 1119 break; 1120 1121 case TOK_TUPDATE: 1122 if (*ac <= 0 || time_to_us(av[0]) < 0) 1123 errx(EX_DATAERR, "tupdate needs time\n"); 1124 1125 ep->par[1] = time_to_us(av[0]); 1126 (*ac)--; av++; 1127 break; 1128 1129 case TOK_MAX_BURST: 1130 if (*ac <= 0 || time_to_us(av[0]) < 0) 1131 errx(EX_DATAERR, "max_burst needs time\n"); 1132 1133 ep->par[2] = time_to_us(av[0]); 1134 (*ac)--; av++; 1135 break; 1136 1137 case TOK_MAX_ECNTH: 1138 if (*ac <= 0 || !is_valid_number(av[0])) 1139 errx(EX_DATAERR, "max_ecnth needs number\n"); 1140 1141 ep->par[3] = atof(av[0]) * PIE_SCALE; 1142 (*ac)--; av++; 1143 break; 1144 1145 case TOK_ALPHA: 1146 if (*ac <= 0 || !is_valid_number(av[0])) 1147 errx(EX_DATAERR, "alpha needs number\n"); 1148 1149 ep->par[4] = atof(av[0]) * PIE_SCALE; 1150 (*ac)--; av++; 1151 break; 1152 1153 case TOK_BETA: 1154 if (*ac <= 0 || !is_valid_number(av[0])) 1155 errx(EX_DATAERR, "beta needs number\n"); 1156 1157 ep->par[5] = atof(av[0]) * PIE_SCALE; 1158 (*ac)--; av++; 1159 break; 1160 1161 case TOK_ECN: 1162 ep->par[6] |= PIE_ECN_ENABLED; 1163 break; 1164 case TOK_NO_ECN: 1165 ep->par[6] &= ~PIE_ECN_ENABLED; 1166 break; 1167 1168 case TOK_CAPDROP: 1169 ep->par[6] |= PIE_CAPDROP_ENABLED; 1170 break; 1171 case TOK_NO_CAPDROP: 1172 ep->par[6] &= ~PIE_CAPDROP_ENABLED; 1173 break; 1174 1175 case TOK_ONOFF: 1176 ep->par[6] |= PIE_ON_OFF_MODE_ENABLED; 1177 break; 1178 1179 case TOK_DRE: 1180 ep->par[6] |= PIE_DEPRATEEST_ENABLED; 1181 break; 1182 1183 case TOK_TS: 1184 ep->par[6] &= ~PIE_DEPRATEEST_ENABLED; 1185 break; 1186 1187 case TOK_DERAND: 1188 ep->par[6] |= PIE_DERAND_ENABLED; 1189 break; 1190 case TOK_NO_DERAND: 1191 ep->par[6] &= ~PIE_DERAND_ENABLED; 1192 break; 1193 1194 /* Config fq_pie parameters */ 1195 case TOK_QUANTUM: 1196 if (type != TOK_FQ_PIE) 1197 errx(EX_DATAERR, "quantum is not for pie\n"); 1198 if (*ac <= 0 || !is_valid_number(av[0])) 1199 errx(EX_DATAERR, "quantum needs number\n"); 1200 1201 ep->par[7]= atoi(av[0]); 1202 (*ac)--; av++; 1203 break; 1204 1205 case TOK_LIMIT: 1206 if (type != TOK_FQ_PIE) 1207 errx(EX_DATAERR, "limit is not for pie, use queue instead\n"); 1208 if (*ac <= 0 || !is_valid_number(av[0])) 1209 errx(EX_DATAERR, "limit needs number\n"); 1210 1211 ep->par[8] = atoi(av[0]); 1212 (*ac)--; av++; 1213 break; 1214 1215 case TOK_FLOWS: 1216 if (type != TOK_FQ_PIE) 1217 errx(EX_DATAERR, "flows is not for pie\n"); 1218 if (*ac <= 0 || !is_valid_number(av[0])) 1219 errx(EX_DATAERR, "flows needs number\n"); 1220 1221 ep->par[9] = atoi(av[0]); 1222 (*ac)--; av++; 1223 break; 1224 1225 1226 default: 1227 printf("%s is invalid parameter\n", av[-1]); 1228 } 1229 } 1230 break; 1231 } 1232 1233 return 0; 1234} 1235 1236#endif 1237 1238 1239/* 1240 * configuration of pipes, schedulers, flowsets. 1241 * When we configure a new scheduler, an empty pipe is created, so: 1242 * 1243 * do_pipe = 1 -> "pipe N config ..." only for backward compatibility 1244 * sched N+Delta type fifo sched_mask ... 1245 * pipe N+Delta <parameters> 1246 * flowset N+Delta pipe N+Delta (no parameters) 1247 * sched N type wf2q+ sched_mask ... 1248 * pipe N <parameters> 1249 * 1250 * do_pipe = 2 -> flowset N config 1251 * flowset N parameters 1252 * 1253 * do_pipe = 3 -> sched N config 1254 * sched N parameters (default no pipe) 1255 * optional Pipe N config ... 1256 * pipe ==> 1257 */ 1258void 1259ipfw_config_pipe(int ac, char **av) 1260{ 1261 int i; 1262 u_int j; 1263 char *end; 1264 struct dn_id *buf, *base; 1265 struct dn_sch *sch = NULL; 1266 struct dn_link *p = NULL; 1267 struct dn_fs *fs = NULL; 1268 struct dn_profile *pf = NULL; 1269 struct ipfw_flow_id *mask = NULL; 1270#ifdef NEW_AQM 1271 struct dn_extra_parms *aqm_extra; 1272 struct dn_extra_parms *sch_extra; 1273 int lmax_extra; 1274#endif 1275 1276 int lmax; 1277 uint32_t _foo = 0, *flags = &_foo , *buckets = &_foo; 1278 1279 /* 1280 * allocate space for 1 header, 1281 * 1 scheduler, 1 link, 1 flowset, 1 profile 1282 */ 1283 lmax = sizeof(struct dn_id); /* command header */ 1284 lmax += sizeof(struct dn_sch) + sizeof(struct dn_link) + 1285 sizeof(struct dn_fs) + sizeof(struct dn_profile); 1286 1287#ifdef NEW_AQM 1288 /* Extra Params */ 1289 lmax_extra = sizeof(struct dn_extra_parms); 1290 /* two lmax_extra because one for AQM params and another 1291 * sch params 1292 */ 1293 lmax += lmax_extra*2; 1294#endif 1295 1296 av++; ac--; 1297 /* Pipe number */ 1298 if (ac && isdigit(**av)) { 1299 i = atoi(*av); av++; ac--; 1300 } else 1301 i = -1; 1302 if (i <= 0) 1303 errx(EX_USAGE, "need a pipe/flowset/sched number"); 1304 base = buf = safe_calloc(1, lmax); 1305 /* all commands start with a 'CONFIGURE' and a version */ 1306 o_next(&buf, sizeof(struct dn_id), DN_CMD_CONFIG); 1307 base->id = DN_API_VERSION; 1308 1309 switch (co.do_pipe) { 1310 case 1: /* "pipe N config ..." */ 1311 /* Allocate space for the WF2Q+ scheduler, its link 1312 * and the FIFO flowset. Set the number, but leave 1313 * the scheduler subtype and other parameters to 0 1314 * so the kernel will use appropriate defaults. 1315 * XXX todo: add a flag to record if a parameter 1316 * is actually configured. 1317 * If we do a 'pipe config' mask -> sched_mask. 1318 * The FIFO scheduler and link are derived from the 1319 * WF2Q+ one in the kernel. 1320 */ 1321#ifdef NEW_AQM 1322 sch_extra = o_next(&buf, lmax_extra, DN_TEXT); 1323 sch_extra ->oid.subtype = 0; /* don't configure scheduler */ 1324#endif 1325 sch = o_next(&buf, sizeof(*sch), DN_SCH); 1326 p = o_next(&buf, sizeof(*p), DN_LINK); 1327#ifdef NEW_AQM 1328 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT); 1329 aqm_extra ->oid.subtype = 0; /* don't configure AQM */ 1330#endif 1331 fs = o_next(&buf, sizeof(*fs), DN_FS); 1332 1333 sch->sched_nr = i; 1334 sch->oid.subtype = 0; /* defaults to WF2Q+ */ 1335 mask = &sch->sched_mask; 1336 flags = &sch->flags; 1337 buckets = &sch->buckets; 1338 *flags |= DN_PIPE_CMD; 1339 1340 p->link_nr = i; 1341 1342 /* This flowset is only for the FIFO scheduler */ 1343 fs->fs_nr = i + 2*DN_MAX_ID; 1344 fs->sched_nr = i + DN_MAX_ID; 1345 break; 1346 1347 case 2: /* "queue N config ... " */ 1348#ifdef NEW_AQM 1349 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT); 1350 aqm_extra ->oid.subtype = 0; 1351#endif 1352 fs = o_next(&buf, sizeof(*fs), DN_FS); 1353 fs->fs_nr = i; 1354 mask = &fs->flow_mask; 1355 flags = &fs->flags; 1356 buckets = &fs->buckets; 1357 break; 1358 1359 case 3: /* "sched N config ..." */ 1360#ifdef NEW_AQM 1361 sch_extra = o_next(&buf, lmax_extra, DN_TEXT); 1362 sch_extra ->oid.subtype = 0; 1363#endif 1364 sch = o_next(&buf, sizeof(*sch), DN_SCH); 1365#ifdef NEW_AQM 1366 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT); 1367 aqm_extra ->oid.subtype = 0; 1368#endif 1369 fs = o_next(&buf, sizeof(*fs), DN_FS); 1370 sch->sched_nr = i; 1371 mask = &sch->sched_mask; 1372 flags = &sch->flags; 1373 buckets = &sch->buckets; 1374 /* fs is used only with !MULTIQUEUE schedulers */ 1375 fs->fs_nr = i + DN_MAX_ID; 1376 fs->sched_nr = i; 1377 break; 1378 } 1379 /* set to -1 those fields for which we want to reuse existing 1380 * values from the kernel. 1381 * Also, *_nr and subtype = 0 mean reuse the value from the kernel. 1382 * XXX todo: support reuse of the mask. 1383 */ 1384 if (p) 1385 p->bandwidth = -1; 1386 for (j = 0; j < sizeof(fs->par)/sizeof(fs->par[0]); j++) 1387 fs->par[j] = -1; 1388 while (ac > 0) { 1389 double d; 1390 int tok = match_token(dummynet_params, *av); 1391 ac--; av++; 1392 1393 switch(tok) { 1394 case TOK_NOERROR: 1395 NEED(fs, "noerror is only for pipes"); 1396 fs->flags |= DN_NOERROR; 1397 break; 1398 1399 case TOK_PLR: 1400 NEED(fs, "plr is only for pipes"); 1401 NEED1("plr needs argument 0..1\n"); 1402 d = strtod(av[0], NULL); 1403 if (d > 1) 1404 d = 1; 1405 else if (d < 0) 1406 d = 0; 1407 fs->plr = (int)(d*0x7fffffff); 1408 ac--; av++; 1409 break; 1410 1411 case TOK_QUEUE: 1412 NEED(fs, "queue is only for pipes or flowsets"); 1413 NEED1("queue needs queue size\n"); 1414 end = NULL; 1415 fs->qsize = strtoul(av[0], &end, 0); 1416 if (*end == 'K' || *end == 'k') { 1417 fs->flags |= DN_QSIZE_BYTES; 1418 fs->qsize *= 1024; 1419 } else if (*end == 'B' || 1420 _substrcmp2(end, "by", "bytes") == 0) { 1421 fs->flags |= DN_QSIZE_BYTES; 1422 } 1423 ac--; av++; 1424 break; 1425 1426 case TOK_BUCKETS: 1427 NEED(fs, "buckets is only for pipes or flowsets"); 1428 NEED1("buckets needs argument\n"); 1429 *buckets = strtoul(av[0], NULL, 0); 1430 ac--; av++; 1431 break; 1432 1433 case TOK_FLOW_MASK: 1434 case TOK_SCHED_MASK: 1435 case TOK_MASK: 1436 NEED(mask, "tok_mask"); 1437 NEED1("mask needs mask specifier\n"); 1438 /* 1439 * per-flow queue, mask is dst_ip, dst_port, 1440 * src_ip, src_port, proto measured in bits 1441 */ 1442 1443 bzero(mask, sizeof(*mask)); 1444 end = NULL; 1445 1446 while (ac >= 1) { 1447 uint32_t *p32 = NULL; 1448 uint16_t *p16 = NULL; 1449 uint32_t *p20 = NULL; 1450 struct in6_addr *pa6 = NULL; 1451 uint32_t a; 1452 1453 tok = match_token(dummynet_params, *av); 1454 ac--; av++; 1455 switch(tok) { 1456 case TOK_ALL: 1457 /* 1458 * special case, all bits significant 1459 * except 'extra' (the queue number) 1460 */ 1461 mask->dst_ip = ~0; 1462 mask->src_ip = ~0; 1463 mask->dst_port = ~0; 1464 mask->src_port = ~0; 1465 mask->proto = ~0; 1466 n2mask(&mask->dst_ip6, 128); 1467 n2mask(&mask->src_ip6, 128); 1468 mask->flow_id6 = ~0; 1469 *flags |= DN_HAVE_MASK; 1470 goto end_mask; 1471 1472 case TOK_QUEUE: 1473 mask->extra = ~0; 1474 *flags |= DN_HAVE_MASK; 1475 goto end_mask; 1476 1477 case TOK_DSTIP: 1478 mask->addr_type = 4; 1479 p32 = &mask->dst_ip; 1480 break; 1481 1482 case TOK_SRCIP: 1483 mask->addr_type = 4; 1484 p32 = &mask->src_ip; 1485 break; 1486 1487 case TOK_DSTIP6: 1488 mask->addr_type = 6; 1489 pa6 = &mask->dst_ip6; 1490 break; 1491 1492 case TOK_SRCIP6: 1493 mask->addr_type = 6; 1494 pa6 = &mask->src_ip6; 1495 break; 1496 1497 case TOK_FLOWID: 1498 mask->addr_type = 6; 1499 p20 = &mask->flow_id6; 1500 break; 1501 1502 case TOK_DSTPORT: 1503 p16 = &mask->dst_port; 1504 break; 1505 1506 case TOK_SRCPORT: 1507 p16 = &mask->src_port; 1508 break; 1509 1510 case TOK_PROTO: 1511 break; 1512 1513 default: 1514 ac++; av--; /* backtrack */ 1515 goto end_mask; 1516 } 1517 if (ac < 1) 1518 errx(EX_USAGE, "mask: value missing"); 1519 if (*av[0] == '/') { 1520 a = strtoul(av[0]+1, &end, 0); 1521 if (pa6 == NULL) 1522 a = (a == 32) ? ~0 : (1 << a) - 1; 1523 } else 1524 a = strtoul(av[0], &end, 0); 1525 if (p32 != NULL) 1526 *p32 = a; 1527 else if (p16 != NULL) { 1528 if (a > 0xFFFF) 1529 errx(EX_DATAERR, 1530 "port mask must be 16 bit"); 1531 *p16 = (uint16_t)a; 1532 } else if (p20 != NULL) { 1533 if (a > 0xfffff) 1534 errx(EX_DATAERR, 1535 "flow_id mask must be 20 bit"); 1536 *p20 = (uint32_t)a; 1537 } else if (pa6 != NULL) { 1538 if (a > 128) 1539 errx(EX_DATAERR, 1540 "in6addr invalid mask len"); 1541 else 1542 n2mask(pa6, a); 1543 } else { 1544 if (a > 0xFF) 1545 errx(EX_DATAERR, 1546 "proto mask must be 8 bit"); 1547 mask->proto = (uint8_t)a; 1548 } 1549 if (a != 0) 1550 *flags |= DN_HAVE_MASK; 1551 ac--; av++; 1552 } /* end while, config masks */ 1553end_mask: 1554 break; 1555#ifdef NEW_AQM 1556 case TOK_CODEL: 1557 case TOK_PIE: 1558 NEED(fs, "codel/pie is only for flowsets"); 1559 1560 fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED); 1561 fs->flags |= DN_IS_AQM; 1562 1563 strlcpy(aqm_extra->name, av[-1], 1564 sizeof(aqm_extra->name)); 1565 aqm_extra->oid.subtype = DN_AQM_PARAMS; 1566 1567 process_extra_parms(&ac, av, aqm_extra, tok); 1568 break; 1569 1570 case TOK_FQ_CODEL: 1571 case TOK_FQ_PIE: 1572 if (!strcmp(av[-1],"type")) 1573 errx(EX_DATAERR, "use type before fq_codel/fq_pie"); 1574 1575 NEED(sch, "fq_codel/fq_pie is only for schd"); 1576 strlcpy(sch_extra->name, av[-1], 1577 sizeof(sch_extra->name)); 1578 sch_extra->oid.subtype = DN_SCH_PARAMS; 1579 process_extra_parms(&ac, av, sch_extra, tok); 1580 break; 1581#endif 1582 case TOK_RED: 1583 case TOK_GRED: 1584 NEED1("red/gred needs w_q/min_th/max_th/max_p\n"); 1585 fs->flags |= DN_IS_RED; 1586 if (tok == TOK_GRED) 1587 fs->flags |= DN_IS_GENTLE_RED; 1588 /* 1589 * the format for parameters is w_q/min_th/max_th/max_p 1590 */ 1591 if ((end = strsep(&av[0], "/"))) { 1592 double w_q = strtod(end, NULL); 1593 if (w_q > 1 || w_q <= 0) 1594 errx(EX_DATAERR, "0 < w_q <= 1"); 1595 fs->w_q = (int) (w_q * (1 << SCALE_RED)); 1596 } 1597 if ((end = strsep(&av[0], "/"))) { 1598 fs->min_th = strtoul(end, &end, 0); 1599 if (*end == 'K' || *end == 'k') 1600 fs->min_th *= 1024; 1601 } 1602 if ((end = strsep(&av[0], "/"))) { 1603 fs->max_th = strtoul(end, &end, 0); 1604 if (*end == 'K' || *end == 'k') 1605 fs->max_th *= 1024; 1606 } 1607 if ((end = strsep(&av[0], "/"))) { 1608 double max_p = strtod(end, NULL); 1609 if (max_p > 1 || max_p < 0) 1610 errx(EX_DATAERR, "0 <= max_p <= 1"); 1611 fs->max_p = (int)(max_p * (1 << SCALE_RED)); 1612 } 1613 ac--; av++; 1614 break; 1615 1616 case TOK_ECN: 1617 fs->flags |= DN_IS_ECN; 1618 break; 1619 1620 case TOK_DROPTAIL: 1621 NEED(fs, "droptail is only for flowsets"); 1622 fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED); 1623 break; 1624 1625 case TOK_BW: 1626 NEED(p, "bw is only for links"); 1627 NEED1("bw needs bandwidth or interface\n"); 1628 read_bandwidth(av[0], &p->bandwidth, NULL, 0); 1629 ac--; av++; 1630 break; 1631 1632 case TOK_DELAY: 1633 NEED(p, "delay is only for links"); 1634 NEED1("delay needs argument 0..10000ms\n"); 1635 p->delay = strtoul(av[0], NULL, 0); 1636 ac--; av++; 1637 break; 1638 1639 case TOK_TYPE: { 1640 int l; 1641 NEED(sch, "type is only for schedulers"); 1642 NEED1("type needs a string"); 1643 l = strlen(av[0]); 1644 if (l == 0 || l > 15) 1645 errx(1, "type %s too long\n", av[0]); 1646 strlcpy(sch->name, av[0], sizeof(sch->name)); 1647 sch->oid.subtype = 0; /* use string */ 1648#ifdef NEW_AQM 1649 /* if fq_codel is selected, consider all tokens after it 1650 * as parameters 1651 */ 1652 if (!strcasecmp(av[0],"fq_codel") || !strcasecmp(av[0],"fq_pie")){ 1653 strlcpy(sch_extra->name, av[0], 1654 sizeof(sch_extra->name)); 1655 sch_extra->oid.subtype = DN_SCH_PARAMS; 1656 process_extra_parms(&ac, av, sch_extra, tok); 1657 } else { 1658 ac--;av++; 1659 } 1660#else 1661 ac--;av++; 1662#endif 1663 break; 1664 } 1665 1666 case TOK_WEIGHT: 1667 NEED(fs, "weight is only for flowsets"); 1668 NEED1("weight needs argument\n"); 1669 fs->par[0] = strtol(av[0], &end, 0); 1670 ac--; av++; 1671 break; 1672 1673 case TOK_LMAX: 1674 NEED(fs, "lmax is only for flowsets"); 1675 NEED1("lmax needs argument\n"); 1676 fs->par[1] = strtol(av[0], &end, 0); 1677 ac--; av++; 1678 break; 1679 1680 case TOK_PRI: 1681 NEED(fs, "priority is only for flowsets"); 1682 NEED1("priority needs argument\n"); 1683 fs->par[2] = strtol(av[0], &end, 0); 1684 ac--; av++; 1685 break; 1686 1687 case TOK_SCHED: 1688 case TOK_PIPE: 1689 NEED(fs, "pipe/sched"); 1690 NEED1("pipe/link/sched needs number\n"); 1691 fs->sched_nr = strtoul(av[0], &end, 0); 1692 ac--; av++; 1693 break; 1694 1695 case TOK_PROFILE: 1696 NEED((!pf), "profile already set"); 1697 NEED(p, "profile"); 1698 { 1699 NEED1("extra delay needs the file name\n"); 1700 pf = o_next(&buf, sizeof(*pf), DN_PROFILE); 1701 load_extra_delays(av[0], pf, p); //XXX can't fail? 1702 --ac; ++av; 1703 } 1704 break; 1705 1706 case TOK_BURST: 1707 NEED(p, "burst"); 1708 NEED1("burst needs argument\n"); 1709 errno = 0; 1710 if (expand_number(av[0], &p->burst) < 0) 1711 if (errno != ERANGE) 1712 errx(EX_DATAERR, 1713 "burst: invalid argument"); 1714 if (errno || p->burst > (1ULL << 48) - 1) 1715 errx(EX_DATAERR, 1716 "burst: out of range (0..2^48-1)"); 1717 ac--; av++; 1718 break; 1719 1720 default: 1721 errx(EX_DATAERR, "unrecognised option ``%s''", av[-1]); 1722 } 1723 } 1724 1725 /* check validity of parameters */ 1726 if (p) { 1727 if (p->delay > 10000) 1728 errx(EX_DATAERR, "delay must be < 10000"); 1729 if (p->bandwidth == -1) 1730 p->bandwidth = 0; 1731 } 1732 if (fs) { 1733 /* XXX accept a 0 scheduler to keep the default */ 1734 if (fs->flags & DN_QSIZE_BYTES) { 1735 size_t len; 1736 long limit; 1737 1738 len = sizeof(limit); 1739 if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit", 1740 &limit, &len, NULL, 0) == -1) 1741 limit = 1024*1024; 1742 if (fs->qsize > limit) 1743 errx(EX_DATAERR, "queue size must be < %ldB", limit); 1744 } else { 1745 size_t len; 1746 long limit; 1747 1748 len = sizeof(limit); 1749 if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit", 1750 &limit, &len, NULL, 0) == -1) 1751 limit = 100; 1752 if (fs->qsize > limit) 1753 errx(EX_DATAERR, "2 <= queue size <= %ld", limit); 1754 } 1755 1756#ifdef NEW_AQM 1757 if ((fs->flags & DN_IS_ECN) && !((fs->flags & DN_IS_RED)|| 1758 (fs->flags & DN_IS_AQM))) 1759 errx(EX_USAGE, "ECN can be used with red/gred/" 1760 "codel/fq_codel only!"); 1761#else 1762 if ((fs->flags & DN_IS_ECN) && !(fs->flags & DN_IS_RED)) 1763 errx(EX_USAGE, "enable red/gred for ECN"); 1764 1765#endif 1766 1767 if (fs->flags & DN_IS_RED) { 1768 size_t len; 1769 int lookup_depth, avg_pkt_size; 1770 1771 if (!(fs->flags & DN_IS_ECN) && (fs->min_th >= fs->max_th)) 1772 errx(EX_DATAERR, "min_th %d must be < than max_th %d", 1773 fs->min_th, fs->max_th); 1774 else if ((fs->flags & DN_IS_ECN) && (fs->min_th > fs->max_th)) 1775 errx(EX_DATAERR, "min_th %d must be =< than max_th %d", 1776 fs->min_th, fs->max_th); 1777 1778 if (fs->max_th == 0) 1779 errx(EX_DATAERR, "max_th must be > 0"); 1780 1781 len = sizeof(int); 1782 if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth", 1783 &lookup_depth, &len, NULL, 0) == -1) 1784 lookup_depth = 256; 1785 if (lookup_depth == 0) 1786 errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth" 1787 " must be greater than zero"); 1788 1789 len = sizeof(int); 1790 if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size", 1791 &avg_pkt_size, &len, NULL, 0) == -1) 1792 avg_pkt_size = 512; 1793 1794 if (avg_pkt_size == 0) 1795 errx(EX_DATAERR, 1796 "net.inet.ip.dummynet.red_avg_pkt_size must" 1797 " be greater than zero"); 1798 1799#if 0 /* the following computation is now done in the kernel */ 1800 /* 1801 * Ticks needed for sending a medium-sized packet. 1802 * Unfortunately, when we are configuring a WF2Q+ queue, we 1803 * do not have bandwidth information, because that is stored 1804 * in the parent pipe, and also we have multiple queues 1805 * competing for it. So we set s=0, which is not very 1806 * correct. But on the other hand, why do we want RED with 1807 * WF2Q+ ? 1808 */ 1809 if (p.bandwidth==0) /* this is a WF2Q+ queue */ 1810 s = 0; 1811 else 1812 s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth; 1813 /* 1814 * max idle time (in ticks) before avg queue size becomes 0. 1815 * NOTA: (3/w_q) is approx the value x so that 1816 * (1-w_q)^x < 10^-3. 1817 */ 1818 w_q = ((double)fs->w_q) / (1 << SCALE_RED); 1819 idle = s * 3. / w_q; 1820 fs->lookup_step = (int)idle / lookup_depth; 1821 if (!fs->lookup_step) 1822 fs->lookup_step = 1; 1823 weight = 1 - w_q; 1824 for (t = fs->lookup_step; t > 1; --t) 1825 weight *= 1 - w_q; 1826 fs->lookup_weight = (int)(weight * (1 << SCALE_RED)); 1827#endif /* code moved in the kernel */ 1828 } 1829 } 1830 1831 i = do_cmd(IP_DUMMYNET3, base, (char *)buf - (char *)base); 1832 1833 if (i) 1834 err(1, "setsockopt(%s)", "IP_DUMMYNET_CONFIGURE"); 1835} 1836 1837void 1838dummynet_flush(void) 1839{ 1840 struct dn_id oid; 1841 oid_fill(&oid, sizeof(oid), DN_CMD_FLUSH, DN_API_VERSION); 1842 do_cmd(IP_DUMMYNET3, &oid, oid.len); 1843} 1844 1845/* Parse input for 'ipfw [pipe|sched|queue] show [range list]' 1846 * Returns the number of ranges, and possibly stores them 1847 * in the array v of size len. 1848 */ 1849static int 1850parse_range(int ac, char *av[], uint32_t *v, int len) 1851{ 1852 int n = 0; 1853 char *endptr, *s; 1854 uint32_t base[2]; 1855 1856 if (v == NULL || len < 2) { 1857 v = base; 1858 len = 2; 1859 } 1860 1861 for (s = *av; s != NULL; av++, ac--) { 1862 v[0] = strtoul(s, &endptr, 10); 1863 v[1] = (*endptr != '-') ? v[0] : 1864 strtoul(endptr+1, &endptr, 10); 1865 if (*endptr == '\0') { /* prepare for next round */ 1866 s = (ac > 0) ? *(av+1) : NULL; 1867 } else { 1868 if (*endptr != ',') { 1869 warn("invalid number: %s", s); 1870 s = ++endptr; 1871 continue; 1872 } 1873 /* continue processing from here */ 1874 s = ++endptr; 1875 ac++; 1876 av--; 1877 } 1878 if (v[1] < v[0] || 1879 v[1] >= DN_MAX_ID-1 || 1880 v[1] >= DN_MAX_ID-1) { 1881 continue; /* invalid entry */ 1882 } 1883 n++; 1884 /* translate if 'pipe list' */ 1885 if (co.do_pipe == 1) { 1886 v[0] += DN_MAX_ID; 1887 v[1] += DN_MAX_ID; 1888 } 1889 v = (n*2 < len) ? v + 2 : base; 1890 } 1891 return n; 1892} 1893 1894/* main entry point for dummynet list functions. co.do_pipe indicates 1895 * which function we want to support. 1896 * av may contain filtering arguments, either individual entries 1897 * or ranges, or lists (space or commas are valid separators). 1898 * Format for a range can be n1-n2 or n3 n4 n5 ... 1899 * In a range n1 must be <= n2, otherwise the range is ignored. 1900 * A number 'n4' is translate in a range 'n4-n4' 1901 * All number must be > 0 and < DN_MAX_ID-1 1902 */ 1903void 1904dummynet_list(int ac, char *av[], int show_counters) 1905{ 1906 struct dn_id *oid, *x = NULL; 1907 int ret, i; 1908 int n; /* # of ranges */ 1909 u_int buflen, l; 1910 u_int max_size; /* largest obj passed up */ 1911 1912 (void)show_counters; // XXX unused, but we should use it. 1913 ac--; 1914 av++; /* skip 'list' | 'show' word */ 1915 1916 n = parse_range(ac, av, NULL, 0); /* Count # of ranges. */ 1917 1918 /* Allocate space to store ranges */ 1919 l = sizeof(*oid) + sizeof(uint32_t) * n * 2; 1920 oid = safe_calloc(1, l); 1921 oid_fill(oid, l, DN_CMD_GET, DN_API_VERSION); 1922 1923 if (n > 0) /* store ranges in idx */ 1924 parse_range(ac, av, (uint32_t *)(oid + 1), n*2); 1925 /* 1926 * Compute the size of the largest object returned. If the 1927 * response leaves at least this much spare space in the 1928 * buffer, then surely the response is complete; otherwise 1929 * there might be a risk of truncation and we will need to 1930 * retry with a larger buffer. 1931 * XXX don't bother with smaller structs. 1932 */ 1933 max_size = sizeof(struct dn_fs); 1934 if (max_size < sizeof(struct dn_sch)) 1935 max_size = sizeof(struct dn_sch); 1936 if (max_size < sizeof(struct dn_flow)) 1937 max_size = sizeof(struct dn_flow); 1938 1939 switch (co.do_pipe) { 1940 case 1: 1941 oid->subtype = DN_LINK; /* list pipe */ 1942 break; 1943 case 2: 1944 oid->subtype = DN_FS; /* list queue */ 1945 break; 1946 case 3: 1947 oid->subtype = DN_SCH; /* list sched */ 1948 break; 1949 } 1950 1951 /* 1952 * Ask the kernel an estimate of the required space (result 1953 * in oid.id), unless we are requesting a subset of objects, 1954 * in which case the kernel does not give an exact answer. 1955 * In any case, space might grow in the meantime due to the 1956 * creation of new queues, so we must be prepared to retry. 1957 */ 1958 if (n > 0) { 1959 buflen = 4*1024; 1960 } else { 1961 ret = do_cmd(-IP_DUMMYNET3, oid, (uintptr_t)&l); 1962 if (ret != 0 || oid->id <= sizeof(*oid)) 1963 goto done; 1964 buflen = oid->id + max_size; 1965 oid->len = sizeof(*oid); /* restore */ 1966 } 1967 /* Try a few times, until the buffer fits */ 1968 for (i = 0; i < 20; i++) { 1969 l = buflen; 1970 x = safe_realloc(x, l); 1971 bcopy(oid, x, oid->len); 1972 ret = do_cmd(-IP_DUMMYNET3, x, (uintptr_t)&l); 1973 if (ret != 0 || x->id <= sizeof(*oid)) 1974 goto done; /* no response */ 1975 if (l + max_size <= buflen) 1976 break; /* ok */ 1977 buflen *= 2; /* double for next attempt */ 1978 } 1979 list_pipes(x, O_NEXT(x, l)); 1980done: 1981 if (x) 1982 free(x); 1983 free(oid); 1984} 1985