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