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/limits.h> 27 #include <sys/types.h> 28 #include <sys/socket.h> 29 /* XXX there are several sysctl leftover here */ 30 #include <sys/sysctl.h> 31 32 #include "ipfw2.h" 33 34 #ifdef NEW_AQM 35 #include <stdint.h> 36 #endif 37 38 #include <ctype.h> 39 #include <err.h> 40 #include <errno.h> 41 #include <libutil.h> 42 #include <netdb.h> 43 #include <stdio.h> 44 #include <stdlib.h> 45 #include <string.h> 46 #include <sysexits.h> 47 48 #include <net/if.h> 49 #include <netinet/in.h> 50 #include <netinet/ip_fw.h> 51 #include <netinet/ip_dummynet.h> 52 #include <arpa/inet.h> /* inet_ntoa */ 53 54 55 static struct _s_x dummynet_params[] = { 56 { "plr", TOK_PLR }, 57 { "noerror", TOK_NOERROR }, 58 { "buckets", TOK_BUCKETS }, 59 { "dst-ip", TOK_DSTIP }, 60 { "src-ip", TOK_SRCIP }, 61 { "dst-port", TOK_DSTPORT }, 62 { "src-port", TOK_SRCPORT }, 63 { "proto", TOK_PROTO }, 64 { "weight", TOK_WEIGHT }, 65 { "lmax", TOK_LMAX }, 66 { "maxlen", TOK_LMAX }, 67 { "all", TOK_ALL }, 68 { "mask", TOK_MASK }, /* alias for both */ 69 { "sched_mask", TOK_SCHED_MASK }, 70 { "flow_mask", TOK_FLOW_MASK }, 71 { "droptail", TOK_DROPTAIL }, 72 { "ecn", TOK_ECN }, 73 { "red", TOK_RED }, 74 { "gred", TOK_GRED }, 75 #ifdef NEW_AQM 76 { "codel", TOK_CODEL}, /* Codel AQM */ 77 { "fq_codel", TOK_FQ_CODEL}, /* FQ-Codel */ 78 { "pie", TOK_PIE}, /* PIE AQM */ 79 { "fq_pie", TOK_FQ_PIE}, /* FQ-PIE */ 80 #endif 81 { "bw", TOK_BW }, 82 { "bandwidth", TOK_BW }, 83 { "delay", TOK_DELAY }, 84 { "link", TOK_LINK }, 85 { "pipe", TOK_PIPE }, 86 { "queue", TOK_QUEUE }, 87 { "flowset", TOK_FLOWSET }, 88 { "sched", TOK_SCHED }, 89 { "pri", TOK_PRI }, 90 { "priority", TOK_PRI }, 91 { "type", TOK_TYPE }, 92 { "flow-id", TOK_FLOWID}, 93 { "dst-ipv6", TOK_DSTIP6}, 94 { "dst-ip6", TOK_DSTIP6}, 95 { "src-ipv6", TOK_SRCIP6}, 96 { "src-ip6", TOK_SRCIP6}, 97 { "profile", TOK_PROFILE}, 98 { "burst", TOK_BURST}, 99 { "dummynet-params", TOK_NULL }, 100 { NULL, 0 } /* terminator */ 101 }; 102 103 #ifdef NEW_AQM 104 /* AQM/extra sched parameters tokens*/ 105 static struct _s_x aqm_params[] = { 106 { "target", TOK_TARGET}, 107 { "interval", TOK_INTERVAL}, 108 { "limit", TOK_LIMIT}, 109 { "flows", TOK_FLOWS}, 110 { "quantum", TOK_QUANTUM}, 111 { "ecn", TOK_ECN}, 112 { "noecn", TOK_NO_ECN}, 113 { "tupdate", TOK_TUPDATE}, 114 { "max_burst", TOK_MAX_BURST}, 115 { "max_ecnth", TOK_MAX_ECNTH}, 116 { "alpha", TOK_ALPHA}, 117 { "beta", TOK_BETA}, 118 { "capdrop", TOK_CAPDROP}, 119 { "nocapdrop", TOK_NO_CAPDROP}, 120 { "onoff", TOK_ONOFF}, 121 { "dre", TOK_DRE}, 122 { "ts", TOK_TS}, 123 { "derand", TOK_DERAND}, 124 { "noderand", TOK_NO_DERAND}, 125 { NULL, 0 } /* terminator */ 126 }; 127 #endif 128 129 #define O_NEXT(p, len) ((void *)((char *)p + len)) 130 131 static void 132 oid_fill(struct dn_id *oid, int len, int type, uintptr_t id) 133 { 134 oid->len = len; 135 oid->type = type; 136 oid->subtype = 0; 137 oid->id = id; 138 } 139 140 /* make room in the buffer and move the pointer forward */ 141 static void * 142 o_next(struct dn_id **o, int len, int type) 143 { 144 struct dn_id *ret = *o; 145 oid_fill(ret, len, type, 0); 146 *o = O_NEXT(*o, len); 147 return ret; 148 } 149 150 #ifdef NEW_AQM 151 152 /* Codel flags */ 153 enum { 154 CODEL_ECN_ENABLED = 1 155 }; 156 157 /* PIE flags, from PIE kernel module */ 158 enum { 159 PIE_ECN_ENABLED = 1, 160 PIE_CAPDROP_ENABLED = 2, 161 PIE_ON_OFF_MODE_ENABLED = 4, 162 PIE_DEPRATEEST_ENABLED = 8, 163 PIE_DERAND_ENABLED = 16 164 }; 165 166 #define PIE_FIX_POINT_BITS 13 167 #define PIE_SCALE (1L<<PIE_FIX_POINT_BITS) 168 169 /* integer to time */ 170 static void 171 us_to_time(int t, char *strt) 172 { 173 if (t < 0) 174 strt[0]='\0'; 175 else if ( t==0 ) 176 sprintf(strt,"%d", t); 177 else if (t< 1000) 178 sprintf(strt,"%dus", t); 179 else if (t < 1000000) 180 sprintf(strt,"%gms", (float) t / 1000); 181 else 182 sprintf(strt,"%gfs", (float) t / 1000000); 183 } 184 185 /* 186 * returns -1 if s is not a valid time, otherwise, return time in us 187 */ 188 static long 189 time_to_us(const char *s) 190 { 191 int i, dots = 0; 192 int len = strlen(s); 193 char strt[16]="", stru[16]=""; 194 195 if (len>15) 196 return -1; 197 for (i = 0; i<len && (isdigit(s[i]) || s[i]=='.') ; i++) 198 if (s[i]=='.') { 199 if (dots) 200 return -1; 201 else 202 dots++; 203 } 204 205 if (!i) 206 return -1; 207 strncpy(strt, s, i); 208 if (i<len) 209 strcpy(stru, s+i); 210 else 211 strcpy(stru, "ms"); 212 213 if (!strcasecmp(stru, "us")) 214 return atol(strt); 215 if (!strcasecmp(stru, "ms")) 216 return (strtod(strt, NULL) * 1000); 217 if (!strcasecmp(stru, "s")) 218 return (strtod(strt, NULL)*1000000); 219 220 return -1; 221 } 222 223 224 /* Get AQM or scheduler extra parameters */ 225 static void 226 get_extra_parms(uint32_t nr, char *out, int subtype) 227 { 228 struct dn_extra_parms *ep; 229 int ret; 230 char strt1[15], strt2[15], strt3[15]; 231 u_int l; 232 233 /* prepare the request */ 234 l = sizeof(struct dn_extra_parms); 235 ep = safe_calloc(1, l); 236 memset(ep, 0, sizeof(*ep)); 237 *out = '\0'; 238 239 oid_fill(&ep->oid, l, DN_CMD_GET, DN_API_VERSION); 240 ep->oid.len = l; 241 ep->oid.subtype = subtype; 242 ep->nr = nr; 243 244 ret = do_cmd(-IP_DUMMYNET3, ep, (uintptr_t)&l); 245 if (ret) { 246 free(ep); 247 errx(EX_DATAERR, "Error getting extra parameters\n"); 248 } 249 250 switch (subtype) { 251 case DN_AQM_PARAMS: 252 if( !strcasecmp(ep->name, "codel")) { 253 us_to_time(ep->par[0], strt1); 254 us_to_time(ep->par[1], strt2); 255 l = sprintf(out, " AQM CoDel target %s interval %s", 256 strt1, strt2); 257 if (ep->par[2] & CODEL_ECN_ENABLED) 258 l = sprintf(out + l, " ECN"); 259 else 260 l += sprintf(out + l, " NoECN"); 261 } else if( !strcasecmp(ep->name, "pie")) { 262 us_to_time(ep->par[0], strt1); 263 us_to_time(ep->par[1], strt2); 264 us_to_time(ep->par[2], strt3); 265 l = sprintf(out, " AQM type PIE target %s tupdate %s alpha " 266 "%g beta %g max_burst %s max_ecnth %.3g", 267 strt1, 268 strt2, 269 ep->par[4] / (float) PIE_SCALE, 270 ep->par[5] / (float) PIE_SCALE, 271 strt3, 272 ep->par[3] / (float) PIE_SCALE 273 ); 274 275 if (ep->par[6] & PIE_ECN_ENABLED) 276 l += sprintf(out + l, " ECN"); 277 else 278 l += sprintf(out + l, " NoECN"); 279 if (ep->par[6] & PIE_CAPDROP_ENABLED) 280 l += sprintf(out + l, " CapDrop"); 281 else 282 l += sprintf(out + l, " NoCapDrop"); 283 if (ep->par[6] & PIE_ON_OFF_MODE_ENABLED) 284 l += sprintf(out + l, " OnOff"); 285 if (ep->par[6] & PIE_DEPRATEEST_ENABLED) 286 l += sprintf(out + l, " DRE"); 287 else 288 l += sprintf(out + l, " TS"); 289 if (ep->par[6] & PIE_DERAND_ENABLED) 290 l += sprintf(out + l, " Derand"); 291 else 292 l += sprintf(out + l, " NoDerand"); 293 } 294 break; 295 296 case DN_SCH_PARAMS: 297 if (!strcasecmp(ep->name,"FQ_CODEL")) { 298 us_to_time(ep->par[0], strt1); 299 us_to_time(ep->par[1], strt2); 300 l = sprintf(out," FQ_CODEL target %s interval %s" 301 " quantum %jd limit %jd flows %jd", 302 strt1, strt2, 303 (intmax_t) ep->par[3], 304 (intmax_t) ep->par[4], 305 (intmax_t) ep->par[5] 306 ); 307 if (ep->par[2] & CODEL_ECN_ENABLED) 308 l += sprintf(out + l, " ECN"); 309 else 310 l += sprintf(out + l, " NoECN"); 311 l += sprintf(out + l, "\n"); 312 } else if (!strcasecmp(ep->name,"FQ_PIE")) { 313 us_to_time(ep->par[0], strt1); 314 us_to_time(ep->par[1], strt2); 315 us_to_time(ep->par[2], strt3); 316 l = sprintf(out, " FQ_PIE target %s tupdate %s alpha " 317 "%g beta %g max_burst %s max_ecnth %.3g" 318 " quantum %jd limit %jd flows %jd", 319 strt1, 320 strt2, 321 ep->par[4] / (float) PIE_SCALE, 322 ep->par[5] / (float) PIE_SCALE, 323 strt3, 324 ep->par[3] / (float) PIE_SCALE, 325 (intmax_t) ep->par[7], 326 (intmax_t) ep->par[8], 327 (intmax_t) ep->par[9] 328 ); 329 330 if (ep->par[6] & PIE_ECN_ENABLED) 331 l += sprintf(out + l, " ECN"); 332 else 333 l += sprintf(out + l, " NoECN"); 334 if (ep->par[6] & PIE_CAPDROP_ENABLED) 335 l += sprintf(out + l, " CapDrop"); 336 else 337 l += sprintf(out + l, " NoCapDrop"); 338 if (ep->par[6] & PIE_ON_OFF_MODE_ENABLED) 339 l += sprintf(out + l, " OnOff"); 340 if (ep->par[6] & PIE_DEPRATEEST_ENABLED) 341 l += sprintf(out + l, " DRE"); 342 else 343 l += sprintf(out + l, " TS"); 344 if (ep->par[6] & PIE_DERAND_ENABLED) 345 l += sprintf(out + l, " Derand"); 346 else 347 l += sprintf(out + l, " NoDerand"); 348 l += sprintf(out + l, "\n"); 349 } 350 break; 351 } 352 353 free(ep); 354 } 355 #endif 356 357 358 #if 0 359 static int 360 sort_q(void *arg, const void *pa, const void *pb) 361 { 362 int rev = (co.do_sort < 0); 363 int field = rev ? -co.do_sort : co.do_sort; 364 long long res = 0; 365 const struct dn_flow_queue *a = pa; 366 const struct dn_flow_queue *b = pb; 367 368 switch (field) { 369 case 1: /* pkts */ 370 res = a->len - b->len; 371 break; 372 case 2: /* bytes */ 373 res = a->len_bytes - b->len_bytes; 374 break; 375 376 case 3: /* tot pkts */ 377 res = a->tot_pkts - b->tot_pkts; 378 break; 379 380 case 4: /* tot bytes */ 381 res = a->tot_bytes - b->tot_bytes; 382 break; 383 } 384 if (res < 0) 385 res = -1; 386 if (res > 0) 387 res = 1; 388 return (int)(rev ? res : -res); 389 } 390 #endif 391 392 /* print a mask and header for the subsequent list of flows */ 393 static void 394 print_mask(struct ipfw_flow_id *id) 395 { 396 if (!IS_IP6_FLOW_ID(id)) { 397 printf(" " 398 "mask: %s 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n", 399 id->extra ? "queue," : "", 400 id->proto, 401 id->src_ip, id->src_port, 402 id->dst_ip, id->dst_port); 403 } else { 404 char buf[255]; 405 printf("\n mask: %sproto: 0x%02x, flow_id: 0x%08x, ", 406 id->extra ? "queue," : "", 407 id->proto, id->flow_id6); 408 inet_ntop(AF_INET6, &(id->src_ip6), buf, sizeof(buf)); 409 printf("%s/0x%04x -> ", buf, id->src_port); 410 inet_ntop(AF_INET6, &(id->dst_ip6), buf, sizeof(buf)); 411 printf("%s/0x%04x\n", buf, id->dst_port); 412 } 413 } 414 415 static void 416 print_header(struct ipfw_flow_id *id) 417 { 418 if (!IS_IP6_FLOW_ID(id)) 419 printf("BKT Prot ___Source IP/port____ " 420 "____Dest. IP/port____ " 421 "Tot_pkt/bytes Pkt/Byte Drp\n"); 422 else 423 printf("BKT ___Prot___ _flow-id_ " 424 "______________Source IPv6/port_______________ " 425 "_______________Dest. IPv6/port_______________ " 426 "Tot_pkt/bytes Pkt/Byte Drp\n"); 427 } 428 429 static void 430 list_flow(struct buf_pr *bp, struct dn_flow *ni) 431 { 432 char buff[255]; 433 struct protoent *pe = NULL; 434 struct in_addr ina; 435 struct ipfw_flow_id *id = &ni->fid; 436 437 pe = getprotobynumber(id->proto); 438 /* XXX: Should check for IPv4 flows */ 439 bprintf(bp, "%3u%c", (ni->oid.id) & 0xff, 440 id->extra ? '*' : ' '); 441 if (!IS_IP6_FLOW_ID(id)) { 442 if (pe) 443 bprintf(bp, "%-4s ", pe->p_name); 444 else 445 bprintf(bp, "%4u ", id->proto); 446 ina.s_addr = htonl(id->src_ip); 447 bprintf(bp, "%15s/%-5d ", 448 inet_ntoa(ina), id->src_port); 449 ina.s_addr = htonl(id->dst_ip); 450 bprintf(bp, "%15s/%-5d ", 451 inet_ntoa(ina), id->dst_port); 452 } else { 453 /* Print IPv6 flows */ 454 if (pe != NULL) 455 bprintf(bp, "%9s ", pe->p_name); 456 else 457 bprintf(bp, "%9u ", id->proto); 458 bprintf(bp, "%7d %39s/%-5d ", id->flow_id6, 459 inet_ntop(AF_INET6, &(id->src_ip6), buff, sizeof(buff)), 460 id->src_port); 461 bprintf(bp, " %39s/%-5d ", 462 inet_ntop(AF_INET6, &(id->dst_ip6), buff, sizeof(buff)), 463 id->dst_port); 464 } 465 pr_u64(bp, &ni->tot_pkts, 4); 466 pr_u64(bp, &ni->tot_bytes, 8); 467 bprintf(bp, "%2u %4u %3u", 468 ni->length, ni->len_bytes, ni->drops); 469 } 470 471 static void 472 print_flowset_parms(struct dn_fs *fs, char *prefix) 473 { 474 int l; 475 char qs[30]; 476 char plr[30]; 477 char red[200]; /* Display RED parameters */ 478 479 l = fs->qsize; 480 if (fs->flags & DN_QSIZE_BYTES) { 481 if (l >= 8192) 482 sprintf(qs, "%d KB", l / 1024); 483 else 484 sprintf(qs, "%d B", l); 485 } else 486 sprintf(qs, "%3d sl.", l); 487 if (fs->plr) 488 sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff)); 489 else 490 plr[0] = '\0'; 491 492 if (fs->flags & DN_IS_RED) { /* RED parameters */ 493 sprintf(red, 494 "\n\t %cRED w_q %f min_th %d max_th %d max_p %f", 495 (fs->flags & DN_IS_GENTLE_RED) ? 'G' : ' ', 496 1.0 * fs->w_q / (double)(1 << SCALE_RED), 497 fs->min_th, 498 fs->max_th, 499 1.0 * fs->max_p / (double)(1 << SCALE_RED)); 500 if (fs->flags & DN_IS_ECN) 501 strlcat(red, " (ecn)", sizeof(red)); 502 #ifdef NEW_AQM 503 /* get AQM parameters */ 504 } else if (fs->flags & DN_IS_AQM) { 505 get_extra_parms(fs->fs_nr, red, DN_AQM_PARAMS); 506 #endif 507 } else 508 sprintf(red, "droptail"); 509 510 if (prefix[0]) { 511 printf("%s %s%s %d queues (%d buckets) %s\n", 512 prefix, qs, plr, fs->oid.id, fs->buckets, red); 513 prefix[0] = '\0'; 514 } else { 515 printf("q%05d %s%s %d flows (%d buckets) sched %d " 516 "weight %d lmax %d pri %d %s\n", 517 fs->fs_nr, qs, plr, fs->oid.id, fs->buckets, 518 fs->sched_nr, fs->par[0], fs->par[1], fs->par[2], red); 519 if (fs->flags & DN_HAVE_MASK) 520 print_mask(&fs->flow_mask); 521 } 522 } 523 524 static void 525 print_extra_delay_parms(struct dn_profile *p) 526 { 527 double loss; 528 if (p->samples_no <= 0) 529 return; 530 531 loss = p->loss_level; 532 loss /= p->samples_no; 533 printf("\t profile: name \"%s\" loss %f samples %d\n", 534 p->name, loss, p->samples_no); 535 } 536 537 static void 538 flush_buf(char *buf) 539 { 540 if (buf[0]) 541 printf("%s\n", buf); 542 buf[0] = '\0'; 543 } 544 545 /* 546 * generic list routine. We expect objects in a specific order, i.e. 547 * PIPES AND SCHEDULERS: 548 * link; scheduler; internal flowset if any; instances 549 * we can tell a pipe from the number. 550 * 551 * FLOWSETS: 552 * flowset; queues; 553 * link i (int queue); scheduler i; si(i) { flowsets() : queues } 554 */ 555 static void 556 list_pipes(struct dn_id *oid, struct dn_id *end) 557 { 558 char buf[160]; /* pending buffer */ 559 int toPrint = 1; /* print header */ 560 struct buf_pr bp; 561 562 buf[0] = '\0'; 563 bp_alloc(&bp, 4096); 564 for (; oid != end; oid = O_NEXT(oid, oid->len)) { 565 if (oid->len < sizeof(*oid)) 566 errx(1, "invalid oid len %d\n", oid->len); 567 568 switch (oid->type) { 569 default: 570 flush_buf(buf); 571 printf("unrecognized object %d size %d\n", oid->type, oid->len); 572 break; 573 case DN_TEXT: /* list of attached flowsets */ 574 { 575 int i, l; 576 struct { 577 struct dn_id id; 578 uint32_t p[0]; 579 } *d = (void *)oid; 580 l = (oid->len - sizeof(*oid))/sizeof(d->p[0]); 581 if (l == 0) 582 break; 583 printf(" Children flowsets: "); 584 for (i = 0; i < l; i++) 585 printf("%u ", d->p[i]); 586 printf("\n"); 587 break; 588 } 589 case DN_CMD_GET: 590 if (g_co.verbose) 591 printf("answer for cmd %d, len %d\n", oid->type, oid->id); 592 break; 593 case DN_SCH: { 594 struct dn_sch *s = (struct dn_sch *)oid; 595 flush_buf(buf); 596 printf(" sched %d type %s flags 0x%x %d buckets %d active\n", 597 s->sched_nr, 598 s->name, s->flags, s->buckets, s->oid.id); 599 #ifdef NEW_AQM 600 char parms[200]; 601 get_extra_parms(s->sched_nr, parms, DN_SCH_PARAMS); 602 printf("%s",parms); 603 #endif 604 if (s->flags & DN_HAVE_MASK) 605 print_mask(&s->sched_mask); 606 } 607 break; 608 609 case DN_FLOW: 610 if (toPrint != 0) { 611 print_header(&((struct dn_flow *)oid)->fid); 612 toPrint = 0; 613 } 614 list_flow(&bp, (struct dn_flow *)oid); 615 printf("%s\n", bp.buf); 616 bp_flush(&bp); 617 break; 618 619 case DN_LINK: { 620 struct dn_link *p = (struct dn_link *)oid; 621 double b = p->bandwidth; 622 char bwbuf[30]; 623 char burst[5 + 7]; 624 625 /* This starts a new object so flush buffer */ 626 flush_buf(buf); 627 /* data rate */ 628 if (b == 0) 629 sprintf(bwbuf, "unlimited "); 630 else if (b >= 1000000000) 631 sprintf(bwbuf, "%7.3f Gbit/s", b/1000000000); 632 else if (b >= 1000000) 633 sprintf(bwbuf, "%7.3f Mbit/s", b/1000000); 634 else if (b >= 1000) 635 sprintf(bwbuf, "%7.3f Kbit/s", b/1000); 636 else 637 sprintf(bwbuf, "%7.3f bit/s ", b); 638 639 if (humanize_number(burst, sizeof(burst), p->burst, 640 "", HN_AUTOSCALE, 0) < 0 || g_co.verbose) 641 sprintf(burst, "%d", (int)p->burst); 642 sprintf(buf, "%05d: %s %4d ms burst %s", 643 p->link_nr % DN_MAX_ID, bwbuf, p->delay, burst); 644 } 645 break; 646 647 case DN_FS: 648 print_flowset_parms((struct dn_fs *)oid, buf); 649 break; 650 case DN_PROFILE: 651 flush_buf(buf); 652 print_extra_delay_parms((struct dn_profile *)oid); 653 } 654 flush_buf(buf); // XXX does it really go here ? 655 } 656 657 bp_free(&bp); 658 } 659 660 /* 661 * Delete pipe, queue or scheduler i 662 */ 663 int 664 ipfw_delete_pipe(int do_pipe, int i) 665 { 666 struct { 667 struct dn_id oid; 668 uintptr_t a[1]; /* add more if we want a list */ 669 } cmd; 670 oid_fill((void *)&cmd, sizeof(cmd), DN_CMD_DELETE, DN_API_VERSION); 671 cmd.oid.subtype = (do_pipe == 1) ? DN_LINK : 672 ( (do_pipe == 2) ? DN_FS : DN_SCH); 673 cmd.a[0] = i; 674 i = do_cmd(IP_DUMMYNET3, &cmd, cmd.oid.len); 675 if (i) { 676 i = 1; 677 warn("rule %u: setsockopt(IP_DUMMYNET_DEL)", i); 678 } 679 return i; 680 } 681 682 /* 683 * Code to parse delay profiles. 684 * 685 * Some link types introduce extra delays in the transmission 686 * of a packet, e.g. because of MAC level framing, contention on 687 * the use of the channel, MAC level retransmissions and so on. 688 * From our point of view, the channel is effectively unavailable 689 * for this extra time, which is constant or variable depending 690 * on the link type. Additionally, packets may be dropped after this 691 * time (e.g. on a wireless link after too many retransmissions). 692 * We can model the additional delay with an empirical curve 693 * that represents its distribution. 694 * 695 * cumulative probability 696 * 1.0 ^ 697 * | 698 * L +-- loss-level x 699 * | ****** 700 * | * 701 * | ***** 702 * | * 703 * | ** 704 * | * 705 * +-------*-------------------> 706 * delay 707 * 708 * The empirical curve may have both vertical and horizontal lines. 709 * Vertical lines represent constant delay for a range of 710 * probabilities; horizontal lines correspond to a discontinuty 711 * in the delay distribution: the link will use the largest delay 712 * for a given probability. 713 * 714 * To pass the curve to dummynet, we must store the parameters 715 * in a file as described below, and issue the command 716 * 717 * ipfw pipe <n> config ... bw XXX profile <filename> ... 718 * 719 * The file format is the following, with whitespace acting as 720 * a separator and '#' indicating the beginning a comment: 721 * 722 * samples N 723 * the number of samples used in the internal 724 * representation (2..1024; default 100); 725 * 726 * loss-level L 727 * The probability above which packets are lost. 728 * (0.0 <= L <= 1.0, default 1.0 i.e. no loss); 729 * 730 * name identifier 731 * Optional a name (listed by "ipfw pipe show") 732 * to identify the distribution; 733 * 734 * "delay prob" | "prob delay" 735 * One of these two lines is mandatory and defines 736 * the format of the following lines with data points. 737 * 738 * XXX YYY 739 * 2 or more lines representing points in the curve, 740 * with either delay or probability first, according 741 * to the chosen format. 742 * The unit for delay is milliseconds. 743 * 744 * Data points does not need to be ordered or equal to the number 745 * specified in the "samples" line. ipfw will sort and interpolate 746 * the curve as needed. 747 * 748 * Example of a profile file: 749 750 name bla_bla_bla 751 samples 100 752 loss-level 0.86 753 prob delay 754 0 200 # minimum overhead is 200ms 755 0.5 200 756 0.5 300 757 0.8 1000 758 0.9 1300 759 1 1300 760 761 * Internally, we will convert the curve to a fixed number of 762 * samples, and when it is time to transmit a packet we will 763 * model the extra delay as extra bits in the packet. 764 * 765 */ 766 767 #define ED_MAX_LINE_LEN 256+ED_MAX_NAME_LEN 768 #define ED_TOK_SAMPLES "samples" 769 #define ED_TOK_LOSS "loss-level" 770 #define ED_TOK_NAME "name" 771 #define ED_TOK_DELAY "delay" 772 #define ED_TOK_PROB "prob" 773 #define ED_TOK_BW "bw" 774 #define ED_SEPARATORS " \t\n" 775 #define ED_MIN_SAMPLES_NO 2 776 777 /* 778 * returns 1 if s is a non-negative number, with at least one '.' 779 */ 780 static int 781 is_valid_number(const char *s) 782 { 783 int i, dots_found = 0; 784 int len = strlen(s); 785 786 for (i = 0; i<len; ++i) 787 if (!isdigit(s[i]) && (s[i] !='.' || ++dots_found > 1)) 788 return 0; 789 return 1; 790 } 791 792 /* 793 * Take as input a string describing a bandwidth value 794 * and return the numeric bandwidth value. 795 * set clocking interface or bandwidth value 796 */ 797 static void 798 read_bandwidth(char *arg, uint32_t *bandwidth, char *if_name, int namelen) 799 { 800 if (*bandwidth != (uint32_t)-1) 801 warnx("duplicate token, override bandwidth value!"); 802 803 if (arg[0] >= 'a' && arg[0] <= 'z') { 804 if (!if_name) { 805 errx(1, "no if support"); 806 } 807 if (namelen >= IFNAMSIZ) 808 warn("interface name truncated"); 809 namelen--; 810 /* interface name */ 811 strlcpy(if_name, arg, namelen); 812 *bandwidth = 0; 813 } else { /* read bandwidth value */ 814 uint64_t bw; 815 char *end = NULL; 816 817 bw = strtoul(arg, &end, 0); 818 if (*end == 'K' || *end == 'k') { 819 end++; 820 bw *= 1000; 821 } else if (*end == 'M' || *end == 'm') { 822 end++; 823 bw *= 1000000; 824 } else if (*end == 'G' || *end == 'g') { 825 end++; 826 bw *= 1000000000; 827 } 828 if ((*end == 'B' && 829 _substrcmp2(end, "Bi", "Bit/s") != 0) || 830 _substrcmp2(end, "by", "bytes") == 0) 831 bw *= 8; 832 833 if (bw > UINT_MAX) 834 errx(EX_DATAERR, "bandwidth too large"); 835 836 *bandwidth = (uint32_t)bw; 837 if (if_name) 838 if_name[0] = '\0'; 839 } 840 } 841 842 struct point { 843 double prob; 844 double delay; 845 }; 846 847 static int 848 compare_points(const void *vp1, const void *vp2) 849 { 850 const struct point *p1 = vp1; 851 const struct point *p2 = vp2; 852 double res = 0; 853 854 res = p1->prob - p2->prob; 855 if (res == 0) 856 res = p1->delay - p2->delay; 857 if (res < 0) 858 return -1; 859 else if (res > 0) 860 return 1; 861 else 862 return 0; 863 } 864 865 #define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno 866 867 static void 868 load_extra_delays(const char *filename, struct dn_profile *p, 869 struct dn_link *link) 870 { 871 char line[ED_MAX_LINE_LEN]; 872 FILE *f; 873 int lineno = 0; 874 int i; 875 876 int samples = -1; 877 double loss = -1.0; 878 char profile_name[ED_MAX_NAME_LEN]; 879 int delay_first = -1; 880 int do_points = 0; 881 struct point points[ED_MAX_SAMPLES_NO]; 882 int points_no = 0; 883 884 /* XXX link never NULL? */ 885 p->link_nr = link->link_nr; 886 887 profile_name[0] = '\0'; 888 f = fopen(filename, "r"); 889 if (f == NULL) 890 err(EX_UNAVAILABLE, "fopen: %s", filename); 891 892 while (fgets(line, ED_MAX_LINE_LEN, f)) { /* read commands */ 893 char *s, *cur = line, *name = NULL, *arg = NULL; 894 895 ++lineno; 896 897 /* parse the line */ 898 while (cur) { 899 s = strsep(&cur, ED_SEPARATORS); 900 if (s == NULL || *s == '#') 901 break; 902 if (*s == '\0') 903 continue; 904 if (arg) 905 errx(ED_EFMT("too many arguments")); 906 if (name == NULL) 907 name = s; 908 else 909 arg = s; 910 } 911 if (name == NULL) /* empty line */ 912 continue; 913 if (arg == NULL) 914 errx(ED_EFMT("missing arg for %s"), name); 915 916 if (!strcasecmp(name, ED_TOK_SAMPLES)) { 917 if (samples > 0) 918 errx(ED_EFMT("duplicate ``samples'' line")); 919 if (atoi(arg) <=0) 920 errx(ED_EFMT("invalid number of samples")); 921 samples = atoi(arg); 922 if (samples>ED_MAX_SAMPLES_NO) 923 errx(ED_EFMT("too many samples, maximum is %d"), 924 ED_MAX_SAMPLES_NO); 925 do_points = 0; 926 } else if (!strcasecmp(name, ED_TOK_BW)) { 927 char buf[IFNAMSIZ]; 928 read_bandwidth(arg, &link->bandwidth, buf, sizeof(buf)); 929 } else if (!strcasecmp(name, ED_TOK_LOSS)) { 930 if (loss != -1.0) 931 errx(ED_EFMT("duplicated token: %s"), name); 932 if (!is_valid_number(arg)) 933 errx(ED_EFMT("invalid %s"), arg); 934 loss = atof(arg); 935 if (loss > 1) 936 errx(ED_EFMT("%s greater than 1.0"), name); 937 do_points = 0; 938 } else if (!strcasecmp(name, ED_TOK_NAME)) { 939 if (profile_name[0] != '\0') 940 errx(ED_EFMT("duplicated token: %s"), name); 941 strlcpy(profile_name, arg, sizeof(profile_name)); 942 do_points = 0; 943 } else if (!strcasecmp(name, ED_TOK_DELAY)) { 944 if (do_points) 945 errx(ED_EFMT("duplicated token: %s"), name); 946 delay_first = 1; 947 do_points = 1; 948 } else if (!strcasecmp(name, ED_TOK_PROB)) { 949 if (do_points) 950 errx(ED_EFMT("duplicated token: %s"), name); 951 delay_first = 0; 952 do_points = 1; 953 } else if (do_points) { 954 if (!is_valid_number(name) || !is_valid_number(arg)) 955 errx(ED_EFMT("invalid point found")); 956 if (delay_first) { 957 points[points_no].delay = atof(name); 958 points[points_no].prob = atof(arg); 959 } else { 960 points[points_no].delay = atof(arg); 961 points[points_no].prob = atof(name); 962 } 963 if (points[points_no].prob > 1.0) 964 errx(ED_EFMT("probability greater than 1.0")); 965 ++points_no; 966 } else { 967 errx(ED_EFMT("unrecognised command '%s'"), name); 968 } 969 } 970 971 fclose (f); 972 973 if (samples == -1) { 974 warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES); 975 samples = 100; 976 } 977 978 if (loss == -1.0) { 979 warnx("'%s' not found, assuming no loss", ED_TOK_LOSS); 980 loss = 1; 981 } 982 983 /* make sure that there are enough points. */ 984 if (points_no < ED_MIN_SAMPLES_NO) 985 errx(ED_EFMT("too few samples, need at least %d"), 986 ED_MIN_SAMPLES_NO); 987 988 qsort(points, points_no, sizeof(struct point), compare_points); 989 990 /* interpolation */ 991 for (i = 0; i<points_no-1; ++i) { 992 double y1 = points[i].prob * samples; 993 double x1 = points[i].delay; 994 double y2 = points[i+1].prob * samples; 995 double x2 = points[i+1].delay; 996 997 int ix = y1; 998 int stop = y2; 999 1000 if (x1 == x2) { 1001 for (; ix<stop; ++ix) 1002 p->samples[ix] = x1; 1003 } else { 1004 double m = (y2-y1)/(x2-x1); 1005 double c = y1 - m*x1; 1006 for (; ix<stop ; ++ix) 1007 p->samples[ix] = (ix - c)/m; 1008 } 1009 } 1010 p->samples_no = samples; 1011 p->loss_level = loss * samples; 1012 strlcpy(p->name, profile_name, sizeof(p->name)); 1013 } 1014 1015 #ifdef NEW_AQM 1016 1017 /* Parse AQM/extra scheduler parameters */ 1018 static int 1019 process_extra_parms(int *ac, char **av, struct dn_extra_parms *ep, 1020 uint16_t type) 1021 { 1022 int i; 1023 1024 /* use kernel defaults */ 1025 for (i=0; i<DN_MAX_EXTRA_PARM; i++) 1026 ep->par[i] = -1; 1027 1028 switch(type) { 1029 case TOK_CODEL: 1030 case TOK_FQ_CODEL: 1031 /* Codel 1032 * 0- target, 1- interval, 2- flags, 1033 * FQ_CODEL 1034 * 3- quantum, 4- limit, 5- flows 1035 */ 1036 if (type==TOK_CODEL) 1037 ep->par[2] = 0; 1038 else 1039 ep->par[2] = CODEL_ECN_ENABLED; 1040 1041 while (*ac > 0) { 1042 int tok = match_token(aqm_params, *av); 1043 (*ac)--; av++; 1044 switch(tok) { 1045 case TOK_TARGET: 1046 if (*ac <= 0 || time_to_us(av[0]) < 0) 1047 errx(EX_DATAERR, "target needs time\n"); 1048 1049 ep->par[0] = time_to_us(av[0]); 1050 (*ac)--; av++; 1051 break; 1052 1053 case TOK_INTERVAL: 1054 if (*ac <= 0 || time_to_us(av[0]) < 0) 1055 errx(EX_DATAERR, "interval needs time\n"); 1056 1057 ep->par[1] = time_to_us(av[0]); 1058 (*ac)--; av++; 1059 break; 1060 1061 case TOK_ECN: 1062 ep->par[2] = CODEL_ECN_ENABLED; 1063 break; 1064 case TOK_NO_ECN: 1065 ep->par[2] &= ~CODEL_ECN_ENABLED; 1066 break; 1067 /* Config fq_codel parameters */ 1068 case TOK_QUANTUM: 1069 if (type != TOK_FQ_CODEL) 1070 errx(EX_DATAERR, "quantum is not for codel\n"); 1071 if (*ac <= 0 || !is_valid_number(av[0])) 1072 errx(EX_DATAERR, "quantum needs number\n"); 1073 1074 ep->par[3]= atoi(av[0]); 1075 (*ac)--; av++; 1076 break; 1077 1078 case TOK_LIMIT: 1079 if (type != TOK_FQ_CODEL) 1080 errx(EX_DATAERR, "limit is not for codel, use queue instead\n"); 1081 if (*ac <= 0 || !is_valid_number(av[0])) 1082 errx(EX_DATAERR, "limit needs number\n"); 1083 1084 ep->par[4] = atoi(av[0]); 1085 (*ac)--; av++; 1086 break; 1087 1088 case TOK_FLOWS: 1089 if (type != TOK_FQ_CODEL) 1090 errx(EX_DATAERR, "flows is not for codel\n"); 1091 if (*ac <= 0 || !is_valid_number(av[0])) 1092 errx(EX_DATAERR, "flows needs number\n"); 1093 1094 ep->par[5] = atoi(av[0]); 1095 (*ac)--; av++; 1096 break; 1097 1098 default: 1099 printf("%s is Invalid parameter\n", av[-1]); 1100 } 1101 } 1102 break; 1103 case TOK_PIE: 1104 case TOK_FQ_PIE: 1105 /* PIE 1106 * 0- target , 1- tupdate, 2- max_burst, 1107 * 3- max_ecnth, 4- alpha, 1108 * 5- beta, 6- flags 1109 * FQ_CODEL 1110 * 7- quantum, 8- limit, 9- flows 1111 */ 1112 1113 if ( type == TOK_PIE) 1114 ep->par[6] = PIE_CAPDROP_ENABLED | PIE_DEPRATEEST_ENABLED 1115 | PIE_DERAND_ENABLED; 1116 else 1117 /* for FQ-PIE, use TS mode */ 1118 ep->par[6] = PIE_CAPDROP_ENABLED | PIE_DERAND_ENABLED 1119 | PIE_ECN_ENABLED; 1120 1121 while (*ac > 0) { 1122 int tok = match_token(aqm_params, *av); 1123 (*ac)--; av++; 1124 switch(tok) { 1125 case TOK_TARGET: 1126 if (*ac <= 0 || time_to_us(av[0]) < 0) 1127 errx(EX_DATAERR, "target needs time\n"); 1128 1129 ep->par[0] = time_to_us(av[0]); 1130 (*ac)--; av++; 1131 break; 1132 1133 case TOK_TUPDATE: 1134 if (*ac <= 0 || time_to_us(av[0]) < 0) 1135 errx(EX_DATAERR, "tupdate needs time\n"); 1136 1137 ep->par[1] = time_to_us(av[0]); 1138 (*ac)--; av++; 1139 break; 1140 1141 case TOK_MAX_BURST: 1142 if (*ac <= 0 || time_to_us(av[0]) < 0) 1143 errx(EX_DATAERR, "max_burst needs time\n"); 1144 1145 ep->par[2] = time_to_us(av[0]); 1146 (*ac)--; av++; 1147 break; 1148 1149 case TOK_MAX_ECNTH: 1150 if (*ac <= 0 || !is_valid_number(av[0])) 1151 errx(EX_DATAERR, "max_ecnth needs number\n"); 1152 1153 ep->par[3] = atof(av[0]) * PIE_SCALE; 1154 (*ac)--; av++; 1155 break; 1156 1157 case TOK_ALPHA: 1158 if (*ac <= 0 || !is_valid_number(av[0])) 1159 errx(EX_DATAERR, "alpha needs number\n"); 1160 1161 ep->par[4] = atof(av[0]) * PIE_SCALE; 1162 (*ac)--; av++; 1163 break; 1164 1165 case TOK_BETA: 1166 if (*ac <= 0 || !is_valid_number(av[0])) 1167 errx(EX_DATAERR, "beta needs number\n"); 1168 1169 ep->par[5] = atof(av[0]) * PIE_SCALE; 1170 (*ac)--; av++; 1171 break; 1172 1173 case TOK_ECN: 1174 ep->par[6] |= PIE_ECN_ENABLED; 1175 break; 1176 case TOK_NO_ECN: 1177 ep->par[6] &= ~PIE_ECN_ENABLED; 1178 break; 1179 1180 case TOK_CAPDROP: 1181 ep->par[6] |= PIE_CAPDROP_ENABLED; 1182 break; 1183 case TOK_NO_CAPDROP: 1184 ep->par[6] &= ~PIE_CAPDROP_ENABLED; 1185 break; 1186 1187 case TOK_ONOFF: 1188 ep->par[6] |= PIE_ON_OFF_MODE_ENABLED; 1189 break; 1190 1191 case TOK_DRE: 1192 ep->par[6] |= PIE_DEPRATEEST_ENABLED; 1193 break; 1194 1195 case TOK_TS: 1196 ep->par[6] &= ~PIE_DEPRATEEST_ENABLED; 1197 break; 1198 1199 case TOK_DERAND: 1200 ep->par[6] |= PIE_DERAND_ENABLED; 1201 break; 1202 case TOK_NO_DERAND: 1203 ep->par[6] &= ~PIE_DERAND_ENABLED; 1204 break; 1205 1206 /* Config fq_pie parameters */ 1207 case TOK_QUANTUM: 1208 if (type != TOK_FQ_PIE) 1209 errx(EX_DATAERR, "quantum is not for pie\n"); 1210 if (*ac <= 0 || !is_valid_number(av[0])) 1211 errx(EX_DATAERR, "quantum needs number\n"); 1212 1213 ep->par[7]= atoi(av[0]); 1214 (*ac)--; av++; 1215 break; 1216 1217 case TOK_LIMIT: 1218 if (type != TOK_FQ_PIE) 1219 errx(EX_DATAERR, "limit is not for pie, use queue instead\n"); 1220 if (*ac <= 0 || !is_valid_number(av[0])) 1221 errx(EX_DATAERR, "limit needs number\n"); 1222 1223 ep->par[8] = atoi(av[0]); 1224 (*ac)--; av++; 1225 break; 1226 1227 case TOK_FLOWS: 1228 if (type != TOK_FQ_PIE) 1229 errx(EX_DATAERR, "flows is not for pie\n"); 1230 if (*ac <= 0 || !is_valid_number(av[0])) 1231 errx(EX_DATAERR, "flows needs number\n"); 1232 1233 ep->par[9] = atoi(av[0]); 1234 (*ac)--; av++; 1235 break; 1236 1237 1238 default: 1239 printf("%s is invalid parameter\n", av[-1]); 1240 } 1241 } 1242 break; 1243 } 1244 1245 return 0; 1246 } 1247 1248 #endif 1249 1250 1251 /* 1252 * configuration of pipes, schedulers, flowsets. 1253 * When we configure a new scheduler, an empty pipe is created, so: 1254 * 1255 * do_pipe = 1 -> "pipe N config ..." only for backward compatibility 1256 * sched N+Delta type fifo sched_mask ... 1257 * pipe N+Delta <parameters> 1258 * flowset N+Delta pipe N+Delta (no parameters) 1259 * sched N type wf2q+ sched_mask ... 1260 * pipe N <parameters> 1261 * 1262 * do_pipe = 2 -> flowset N config 1263 * flowset N parameters 1264 * 1265 * do_pipe = 3 -> sched N config 1266 * sched N parameters (default no pipe) 1267 * optional Pipe N config ... 1268 * pipe ==> 1269 */ 1270 void 1271 ipfw_config_pipe(int ac, char **av) 1272 { 1273 int i; 1274 u_int j; 1275 char *end; 1276 struct dn_id *buf, *base; 1277 struct dn_sch *sch = NULL; 1278 struct dn_link *p = NULL; 1279 struct dn_fs *fs = NULL; 1280 struct dn_profile *pf = NULL; 1281 struct ipfw_flow_id *mask = NULL; 1282 #ifdef NEW_AQM 1283 struct dn_extra_parms *aqm_extra = NULL; 1284 struct dn_extra_parms *sch_extra = NULL; 1285 int lmax_extra; 1286 #endif 1287 1288 int lmax; 1289 uint32_t _foo = 0, *flags = &_foo , *buckets = &_foo; 1290 1291 /* 1292 * allocate space for 1 header, 1293 * 1 scheduler, 1 link, 1 flowset, 1 profile 1294 */ 1295 lmax = sizeof(struct dn_id); /* command header */ 1296 lmax += sizeof(struct dn_sch) + sizeof(struct dn_link) + 1297 sizeof(struct dn_fs) + sizeof(struct dn_profile); 1298 1299 #ifdef NEW_AQM 1300 /* Extra Params */ 1301 lmax_extra = sizeof(struct dn_extra_parms); 1302 /* two lmax_extra because one for AQM params and another 1303 * sch params 1304 */ 1305 lmax += lmax_extra*2; 1306 #endif 1307 1308 av++; ac--; 1309 /* Pipe number */ 1310 if (ac && isdigit(**av)) { 1311 i = atoi(*av); av++; ac--; 1312 } else 1313 i = -1; 1314 if (i <= 0) 1315 errx(EX_USAGE, "need a pipe/flowset/sched number"); 1316 base = buf = safe_calloc(1, lmax); 1317 /* all commands start with a 'CONFIGURE' and a version */ 1318 o_next(&buf, sizeof(struct dn_id), DN_CMD_CONFIG); 1319 base->id = DN_API_VERSION; 1320 1321 switch (g_co.do_pipe) { 1322 case 1: /* "pipe N config ..." */ 1323 /* Allocate space for the WF2Q+ scheduler, its link 1324 * and the FIFO flowset. Set the number, but leave 1325 * the scheduler subtype and other parameters to 0 1326 * so the kernel will use appropriate defaults. 1327 * XXX todo: add a flag to record if a parameter 1328 * is actually configured. 1329 * If we do a 'pipe config' mask -> sched_mask. 1330 * The FIFO scheduler and link are derived from the 1331 * WF2Q+ one in the kernel. 1332 */ 1333 #ifdef NEW_AQM 1334 sch_extra = o_next(&buf, lmax_extra, DN_TEXT); 1335 sch_extra ->oid.subtype = 0; /* don't configure scheduler */ 1336 #endif 1337 sch = o_next(&buf, sizeof(*sch), DN_SCH); 1338 p = o_next(&buf, sizeof(*p), DN_LINK); 1339 #ifdef NEW_AQM 1340 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT); 1341 aqm_extra ->oid.subtype = 0; /* don't configure AQM */ 1342 #endif 1343 fs = o_next(&buf, sizeof(*fs), DN_FS); 1344 1345 sch->sched_nr = i; 1346 sch->oid.subtype = 0; /* defaults to WF2Q+ */ 1347 mask = &sch->sched_mask; 1348 flags = &sch->flags; 1349 buckets = &sch->buckets; 1350 *flags |= DN_PIPE_CMD; 1351 1352 p->link_nr = i; 1353 1354 /* This flowset is only for the FIFO scheduler */ 1355 fs->fs_nr = i + 2*DN_MAX_ID; 1356 fs->sched_nr = i + DN_MAX_ID; 1357 break; 1358 1359 case 2: /* "queue N config ... " */ 1360 #ifdef NEW_AQM 1361 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT); 1362 aqm_extra ->oid.subtype = 0; 1363 #endif 1364 fs = o_next(&buf, sizeof(*fs), DN_FS); 1365 fs->fs_nr = i; 1366 mask = &fs->flow_mask; 1367 flags = &fs->flags; 1368 buckets = &fs->buckets; 1369 break; 1370 1371 case 3: /* "sched N config ..." */ 1372 #ifdef NEW_AQM 1373 sch_extra = o_next(&buf, lmax_extra, DN_TEXT); 1374 sch_extra ->oid.subtype = 0; 1375 #endif 1376 sch = o_next(&buf, sizeof(*sch), DN_SCH); 1377 #ifdef NEW_AQM 1378 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT); 1379 aqm_extra ->oid.subtype = 0; 1380 #endif 1381 fs = o_next(&buf, sizeof(*fs), DN_FS); 1382 sch->sched_nr = i; 1383 mask = &sch->sched_mask; 1384 flags = &sch->flags; 1385 buckets = &sch->buckets; 1386 /* fs is used only with !MULTIQUEUE schedulers */ 1387 fs->fs_nr = i + DN_MAX_ID; 1388 fs->sched_nr = i; 1389 break; 1390 } 1391 /* set to -1 those fields for which we want to reuse existing 1392 * values from the kernel. 1393 * Also, *_nr and subtype = 0 mean reuse the value from the kernel. 1394 * XXX todo: support reuse of the mask. 1395 */ 1396 if (p) 1397 p->bandwidth = -1; 1398 for (j = 0; j < sizeof(fs->par)/sizeof(fs->par[0]); j++) 1399 fs->par[j] = -1; 1400 while (ac > 0) { 1401 double d; 1402 int tok = match_token(dummynet_params, *av); 1403 ac--; av++; 1404 1405 switch(tok) { 1406 case TOK_NOERROR: 1407 NEED(fs, "noerror is only for pipes"); 1408 fs->flags |= DN_NOERROR; 1409 break; 1410 1411 case TOK_PLR: 1412 NEED(fs, "plr is only for pipes"); 1413 NEED1("plr needs argument 0..1\n"); 1414 d = strtod(av[0], NULL); 1415 if (d > 1) 1416 d = 1; 1417 else if (d < 0) 1418 d = 0; 1419 fs->plr = (int)(d*0x7fffffff); 1420 ac--; av++; 1421 break; 1422 1423 case TOK_QUEUE: 1424 NEED(fs, "queue is only for pipes or flowsets"); 1425 NEED1("queue needs queue size\n"); 1426 end = NULL; 1427 fs->qsize = strtoul(av[0], &end, 0); 1428 if (*end == 'K' || *end == 'k') { 1429 fs->flags |= DN_QSIZE_BYTES; 1430 fs->qsize *= 1024; 1431 } else if (*end == 'B' || 1432 _substrcmp2(end, "by", "bytes") == 0) { 1433 fs->flags |= DN_QSIZE_BYTES; 1434 } 1435 ac--; av++; 1436 break; 1437 1438 case TOK_BUCKETS: 1439 NEED(fs, "buckets is only for pipes or flowsets"); 1440 NEED1("buckets needs argument\n"); 1441 *buckets = strtoul(av[0], NULL, 0); 1442 ac--; av++; 1443 break; 1444 1445 case TOK_FLOW_MASK: 1446 case TOK_SCHED_MASK: 1447 case TOK_MASK: 1448 NEED(mask, "tok_mask"); 1449 NEED1("mask needs mask specifier\n"); 1450 /* 1451 * per-flow queue, mask is dst_ip, dst_port, 1452 * src_ip, src_port, proto measured in bits 1453 */ 1454 1455 bzero(mask, sizeof(*mask)); 1456 end = NULL; 1457 1458 while (ac >= 1) { 1459 uint32_t *p32 = NULL; 1460 uint16_t *p16 = NULL; 1461 uint32_t *p20 = NULL; 1462 struct in6_addr *pa6 = NULL; 1463 uint32_t a; 1464 1465 tok = match_token(dummynet_params, *av); 1466 ac--; av++; 1467 switch(tok) { 1468 case TOK_ALL: 1469 /* 1470 * special case, all bits significant 1471 * except 'extra' (the queue number) 1472 */ 1473 mask->dst_ip = ~0; 1474 mask->src_ip = ~0; 1475 mask->dst_port = ~0; 1476 mask->src_port = ~0; 1477 mask->proto = ~0; 1478 n2mask(&mask->dst_ip6, 128); 1479 n2mask(&mask->src_ip6, 128); 1480 mask->flow_id6 = ~0; 1481 *flags |= DN_HAVE_MASK; 1482 goto end_mask; 1483 1484 case TOK_QUEUE: 1485 mask->extra = ~0; 1486 *flags |= DN_HAVE_MASK; 1487 goto end_mask; 1488 1489 case TOK_DSTIP: 1490 mask->addr_type = 4; 1491 p32 = &mask->dst_ip; 1492 break; 1493 1494 case TOK_SRCIP: 1495 mask->addr_type = 4; 1496 p32 = &mask->src_ip; 1497 break; 1498 1499 case TOK_DSTIP6: 1500 mask->addr_type = 6; 1501 pa6 = &mask->dst_ip6; 1502 break; 1503 1504 case TOK_SRCIP6: 1505 mask->addr_type = 6; 1506 pa6 = &mask->src_ip6; 1507 break; 1508 1509 case TOK_FLOWID: 1510 mask->addr_type = 6; 1511 p20 = &mask->flow_id6; 1512 break; 1513 1514 case TOK_DSTPORT: 1515 p16 = &mask->dst_port; 1516 break; 1517 1518 case TOK_SRCPORT: 1519 p16 = &mask->src_port; 1520 break; 1521 1522 case TOK_PROTO: 1523 break; 1524 1525 default: 1526 ac++; av--; /* backtrack */ 1527 goto end_mask; 1528 } 1529 if (ac < 1) 1530 errx(EX_USAGE, "mask: value missing"); 1531 if (*av[0] == '/') { 1532 a = strtoul(av[0]+1, &end, 0); 1533 if (pa6 == NULL) 1534 a = (a == 32) ? ~0 : (1 << a) - 1; 1535 } else 1536 a = strtoul(av[0], &end, 0); 1537 if (p32 != NULL) 1538 *p32 = a; 1539 else if (p16 != NULL) { 1540 if (a > 0xFFFF) 1541 errx(EX_DATAERR, 1542 "port mask must be 16 bit"); 1543 *p16 = (uint16_t)a; 1544 } else if (p20 != NULL) { 1545 if (a > 0xfffff) 1546 errx(EX_DATAERR, 1547 "flow_id mask must be 20 bit"); 1548 *p20 = (uint32_t)a; 1549 } else if (pa6 != NULL) { 1550 if (a > 128) 1551 errx(EX_DATAERR, 1552 "in6addr invalid mask len"); 1553 else 1554 n2mask(pa6, a); 1555 } else { 1556 if (a > 0xFF) 1557 errx(EX_DATAERR, 1558 "proto mask must be 8 bit"); 1559 mask->proto = (uint8_t)a; 1560 } 1561 if (a != 0) 1562 *flags |= DN_HAVE_MASK; 1563 ac--; av++; 1564 } /* end while, config masks */ 1565 end_mask: 1566 break; 1567 #ifdef NEW_AQM 1568 case TOK_CODEL: 1569 case TOK_PIE: 1570 NEED(fs, "codel/pie is only for flowsets"); 1571 1572 fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED); 1573 fs->flags |= DN_IS_AQM; 1574 1575 strlcpy(aqm_extra->name, av[-1], 1576 sizeof(aqm_extra->name)); 1577 aqm_extra->oid.subtype = DN_AQM_PARAMS; 1578 1579 process_extra_parms(&ac, av, aqm_extra, tok); 1580 break; 1581 1582 case TOK_FQ_CODEL: 1583 case TOK_FQ_PIE: 1584 if (!strcmp(av[-1],"type")) 1585 errx(EX_DATAERR, "use type before fq_codel/fq_pie"); 1586 1587 NEED(sch, "fq_codel/fq_pie is only for schd"); 1588 strlcpy(sch_extra->name, av[-1], 1589 sizeof(sch_extra->name)); 1590 sch_extra->oid.subtype = DN_SCH_PARAMS; 1591 process_extra_parms(&ac, av, sch_extra, tok); 1592 break; 1593 #endif 1594 case TOK_RED: 1595 case TOK_GRED: 1596 NEED1("red/gred needs w_q/min_th/max_th/max_p\n"); 1597 fs->flags |= DN_IS_RED; 1598 if (tok == TOK_GRED) 1599 fs->flags |= DN_IS_GENTLE_RED; 1600 /* 1601 * the format for parameters is w_q/min_th/max_th/max_p 1602 */ 1603 if ((end = strsep(&av[0], "/"))) { 1604 double w_q = strtod(end, NULL); 1605 if (w_q > 1 || w_q <= 0) 1606 errx(EX_DATAERR, "0 < w_q <= 1"); 1607 fs->w_q = (int) (w_q * (1 << SCALE_RED)); 1608 } 1609 if ((end = strsep(&av[0], "/"))) { 1610 fs->min_th = strtoul(end, &end, 0); 1611 if (*end == 'K' || *end == 'k') 1612 fs->min_th *= 1024; 1613 } 1614 if ((end = strsep(&av[0], "/"))) { 1615 fs->max_th = strtoul(end, &end, 0); 1616 if (*end == 'K' || *end == 'k') 1617 fs->max_th *= 1024; 1618 } 1619 if ((end = strsep(&av[0], "/"))) { 1620 double max_p = strtod(end, NULL); 1621 if (max_p > 1 || max_p < 0) 1622 errx(EX_DATAERR, "0 <= max_p <= 1"); 1623 fs->max_p = (int)(max_p * (1 << SCALE_RED)); 1624 } 1625 ac--; av++; 1626 break; 1627 1628 case TOK_ECN: 1629 fs->flags |= DN_IS_ECN; 1630 break; 1631 1632 case TOK_DROPTAIL: 1633 NEED(fs, "droptail is only for flowsets"); 1634 fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED); 1635 break; 1636 1637 case TOK_BW: 1638 NEED(p, "bw is only for links"); 1639 NEED1("bw needs bandwidth or interface\n"); 1640 read_bandwidth(av[0], &p->bandwidth, NULL, 0); 1641 ac--; av++; 1642 break; 1643 1644 case TOK_DELAY: 1645 NEED(p, "delay is only for links"); 1646 NEED1("delay needs argument 0..10000ms\n"); 1647 p->delay = strtoul(av[0], NULL, 0); 1648 ac--; av++; 1649 break; 1650 1651 case TOK_TYPE: { 1652 int l; 1653 NEED(sch, "type is only for schedulers"); 1654 NEED1("type needs a string"); 1655 l = strlen(av[0]); 1656 if (l == 0 || l > 15) 1657 errx(1, "type %s too long\n", av[0]); 1658 strlcpy(sch->name, av[0], sizeof(sch->name)); 1659 sch->oid.subtype = 0; /* use string */ 1660 #ifdef NEW_AQM 1661 /* if fq_codel is selected, consider all tokens after it 1662 * as parameters 1663 */ 1664 if (!strcasecmp(av[0],"fq_codel") || !strcasecmp(av[0],"fq_pie")){ 1665 strlcpy(sch_extra->name, av[0], 1666 sizeof(sch_extra->name)); 1667 sch_extra->oid.subtype = DN_SCH_PARAMS; 1668 process_extra_parms(&ac, av, sch_extra, tok); 1669 } else { 1670 ac--;av++; 1671 } 1672 #else 1673 ac--;av++; 1674 #endif 1675 break; 1676 } 1677 1678 case TOK_WEIGHT: 1679 NEED(fs, "weight is only for flowsets"); 1680 NEED1("weight needs argument\n"); 1681 fs->par[0] = strtol(av[0], &end, 0); 1682 ac--; av++; 1683 break; 1684 1685 case TOK_LMAX: 1686 NEED(fs, "lmax is only for flowsets"); 1687 NEED1("lmax needs argument\n"); 1688 fs->par[1] = strtol(av[0], &end, 0); 1689 ac--; av++; 1690 break; 1691 1692 case TOK_PRI: 1693 NEED(fs, "priority is only for flowsets"); 1694 NEED1("priority needs argument\n"); 1695 fs->par[2] = strtol(av[0], &end, 0); 1696 ac--; av++; 1697 break; 1698 1699 case TOK_SCHED: 1700 case TOK_PIPE: 1701 NEED(fs, "pipe/sched"); 1702 NEED1("pipe/link/sched needs number\n"); 1703 fs->sched_nr = strtoul(av[0], &end, 0); 1704 ac--; av++; 1705 break; 1706 1707 case TOK_PROFILE: 1708 NEED((!pf), "profile already set"); 1709 NEED(p, "profile"); 1710 { 1711 NEED1("extra delay needs the file name\n"); 1712 pf = o_next(&buf, sizeof(*pf), DN_PROFILE); 1713 load_extra_delays(av[0], pf, p); //XXX can't fail? 1714 --ac; ++av; 1715 } 1716 break; 1717 1718 case TOK_BURST: 1719 NEED(p, "burst"); 1720 NEED1("burst needs argument\n"); 1721 errno = 0; 1722 if (expand_number(av[0], &p->burst) < 0) 1723 if (errno != ERANGE) 1724 errx(EX_DATAERR, 1725 "burst: invalid argument"); 1726 if (errno || p->burst > (1ULL << 48) - 1) 1727 errx(EX_DATAERR, 1728 "burst: out of range (0..2^48-1)"); 1729 ac--; av++; 1730 break; 1731 1732 default: 1733 errx(EX_DATAERR, "unrecognised option ``%s''", av[-1]); 1734 } 1735 } 1736 1737 /* check validity of parameters */ 1738 if (p) { 1739 if (p->delay > 10000) 1740 errx(EX_DATAERR, "delay must be < 10000"); 1741 if (p->bandwidth == (uint32_t)-1) 1742 p->bandwidth = 0; 1743 } 1744 if (fs) { 1745 /* XXX accept a 0 scheduler to keep the default */ 1746 if (fs->flags & DN_QSIZE_BYTES) { 1747 size_t len; 1748 long limit; 1749 1750 len = sizeof(limit); 1751 if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit", 1752 &limit, &len, NULL, 0) == -1) 1753 limit = 1024*1024; 1754 if (fs->qsize > limit) 1755 errx(EX_DATAERR, "queue size must be < %ldB", limit); 1756 } else { 1757 size_t len; 1758 long limit; 1759 1760 len = sizeof(limit); 1761 if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit", 1762 &limit, &len, NULL, 0) == -1) 1763 limit = 100; 1764 if (fs->qsize > limit) 1765 errx(EX_DATAERR, "2 <= queue size <= %ld", limit); 1766 } 1767 1768 #ifdef NEW_AQM 1769 if ((fs->flags & DN_IS_ECN) && !((fs->flags & DN_IS_RED)|| 1770 (fs->flags & DN_IS_AQM))) 1771 errx(EX_USAGE, "ECN can be used with red/gred/" 1772 "codel/fq_codel only!"); 1773 #else 1774 if ((fs->flags & DN_IS_ECN) && !(fs->flags & DN_IS_RED)) 1775 errx(EX_USAGE, "enable red/gred for ECN"); 1776 1777 #endif 1778 1779 if (fs->flags & DN_IS_RED) { 1780 size_t len; 1781 int lookup_depth, avg_pkt_size; 1782 1783 if (!(fs->flags & DN_IS_ECN) && (fs->min_th >= fs->max_th)) 1784 errx(EX_DATAERR, "min_th %d must be < than max_th %d", 1785 fs->min_th, fs->max_th); 1786 else if ((fs->flags & DN_IS_ECN) && (fs->min_th > fs->max_th)) 1787 errx(EX_DATAERR, "min_th %d must be =< than max_th %d", 1788 fs->min_th, fs->max_th); 1789 1790 if (fs->max_th == 0) 1791 errx(EX_DATAERR, "max_th must be > 0"); 1792 1793 len = sizeof(int); 1794 if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth", 1795 &lookup_depth, &len, NULL, 0) == -1) 1796 lookup_depth = 256; 1797 if (lookup_depth == 0) 1798 errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth" 1799 " must be greater than zero"); 1800 1801 len = sizeof(int); 1802 if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size", 1803 &avg_pkt_size, &len, NULL, 0) == -1) 1804 avg_pkt_size = 512; 1805 1806 if (avg_pkt_size == 0) 1807 errx(EX_DATAERR, 1808 "net.inet.ip.dummynet.red_avg_pkt_size must" 1809 " be greater than zero"); 1810 1811 #if 0 /* the following computation is now done in the kernel */ 1812 /* 1813 * Ticks needed for sending a medium-sized packet. 1814 * Unfortunately, when we are configuring a WF2Q+ queue, we 1815 * do not have bandwidth information, because that is stored 1816 * in the parent pipe, and also we have multiple queues 1817 * competing for it. So we set s=0, which is not very 1818 * correct. But on the other hand, why do we want RED with 1819 * WF2Q+ ? 1820 */ 1821 if (p.bandwidth==0) /* this is a WF2Q+ queue */ 1822 s = 0; 1823 else 1824 s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth; 1825 /* 1826 * max idle time (in ticks) before avg queue size becomes 0. 1827 * NOTA: (3/w_q) is approx the value x so that 1828 * (1-w_q)^x < 10^-3. 1829 */ 1830 w_q = ((double)fs->w_q) / (1 << SCALE_RED); 1831 idle = s * 3. / w_q; 1832 fs->lookup_step = (int)idle / lookup_depth; 1833 if (!fs->lookup_step) 1834 fs->lookup_step = 1; 1835 weight = 1 - w_q; 1836 for (t = fs->lookup_step; t > 1; --t) 1837 weight *= 1 - w_q; 1838 fs->lookup_weight = (int)(weight * (1 << SCALE_RED)); 1839 #endif /* code moved in the kernel */ 1840 } 1841 } 1842 1843 i = do_cmd(IP_DUMMYNET3, base, (char *)buf - (char *)base); 1844 1845 if (i) 1846 err(1, "setsockopt(%s)", "IP_DUMMYNET_CONFIGURE"); 1847 } 1848 1849 void 1850 dummynet_flush(void) 1851 { 1852 struct dn_id oid; 1853 oid_fill(&oid, sizeof(oid), DN_CMD_FLUSH, DN_API_VERSION); 1854 do_cmd(IP_DUMMYNET3, &oid, oid.len); 1855 } 1856 1857 /* Parse input for 'ipfw [pipe|sched|queue] show [range list]' 1858 * Returns the number of ranges, and possibly stores them 1859 * in the array v of size len. 1860 */ 1861 static int 1862 parse_range(int ac, char *av[], uint32_t *v, int len) 1863 { 1864 int n = 0; 1865 char *endptr, *s; 1866 uint32_t base[2]; 1867 1868 if (v == NULL || len < 2) { 1869 v = base; 1870 len = 2; 1871 } 1872 1873 for (s = *av; s != NULL; av++, ac--) { 1874 v[0] = strtoul(s, &endptr, 10); 1875 v[1] = (*endptr != '-') ? v[0] : 1876 strtoul(endptr+1, &endptr, 10); 1877 if (*endptr == '\0') { /* prepare for next round */ 1878 s = (ac > 0) ? *(av+1) : NULL; 1879 } else { 1880 if (*endptr != ',') { 1881 warn("invalid number: %s", s); 1882 s = ++endptr; 1883 continue; 1884 } 1885 /* continue processing from here */ 1886 s = ++endptr; 1887 ac++; 1888 av--; 1889 } 1890 if (v[1] < v[0] || 1891 v[0] >= DN_MAX_ID-1 || 1892 v[1] >= DN_MAX_ID-1) { 1893 continue; /* invalid entry */ 1894 } 1895 n++; 1896 /* translate if 'pipe list' */ 1897 if (g_co.do_pipe == 1) { 1898 v[0] += DN_MAX_ID; 1899 v[1] += DN_MAX_ID; 1900 } 1901 v = (n*2 < len) ? v + 2 : base; 1902 } 1903 return n; 1904 } 1905 1906 /* main entry point for dummynet list functions. co.do_pipe indicates 1907 * which function we want to support. 1908 * av may contain filtering arguments, either individual entries 1909 * or ranges, or lists (space or commas are valid separators). 1910 * Format for a range can be n1-n2 or n3 n4 n5 ... 1911 * In a range n1 must be <= n2, otherwise the range is ignored. 1912 * A number 'n4' is translate in a range 'n4-n4' 1913 * All number must be > 0 and < DN_MAX_ID-1 1914 */ 1915 void 1916 dummynet_list(int ac, char *av[], int show_counters) 1917 { 1918 struct dn_id *oid, *x = NULL; 1919 int ret, i; 1920 int n; /* # of ranges */ 1921 u_int buflen, l; 1922 u_int max_size; /* largest obj passed up */ 1923 1924 (void)show_counters; // XXX unused, but we should use it. 1925 ac--; 1926 av++; /* skip 'list' | 'show' word */ 1927 1928 n = parse_range(ac, av, NULL, 0); /* Count # of ranges. */ 1929 1930 /* Allocate space to store ranges */ 1931 l = sizeof(*oid) + sizeof(uint32_t) * n * 2; 1932 oid = safe_calloc(1, l); 1933 oid_fill(oid, l, DN_CMD_GET, DN_API_VERSION); 1934 1935 if (n > 0) /* store ranges in idx */ 1936 parse_range(ac, av, (uint32_t *)(oid + 1), n*2); 1937 /* 1938 * Compute the size of the largest object returned. If the 1939 * response leaves at least this much spare space in the 1940 * buffer, then surely the response is complete; otherwise 1941 * there might be a risk of truncation and we will need to 1942 * retry with a larger buffer. 1943 * XXX don't bother with smaller structs. 1944 */ 1945 max_size = sizeof(struct dn_fs); 1946 if (max_size < sizeof(struct dn_sch)) 1947 max_size = sizeof(struct dn_sch); 1948 if (max_size < sizeof(struct dn_flow)) 1949 max_size = sizeof(struct dn_flow); 1950 1951 switch (g_co.do_pipe) { 1952 case 1: 1953 oid->subtype = DN_LINK; /* list pipe */ 1954 break; 1955 case 2: 1956 oid->subtype = DN_FS; /* list queue */ 1957 break; 1958 case 3: 1959 oid->subtype = DN_SCH; /* list sched */ 1960 break; 1961 } 1962 1963 /* 1964 * Ask the kernel an estimate of the required space (result 1965 * in oid.id), unless we are requesting a subset of objects, 1966 * in which case the kernel does not give an exact answer. 1967 * In any case, space might grow in the meantime due to the 1968 * creation of new queues, so we must be prepared to retry. 1969 */ 1970 if (n > 0) { 1971 buflen = 4*1024; 1972 } else { 1973 ret = do_cmd(-IP_DUMMYNET3, oid, (uintptr_t)&l); 1974 if (ret != 0 || oid->id <= sizeof(*oid)) 1975 goto done; 1976 buflen = oid->id + max_size; 1977 oid->len = sizeof(*oid); /* restore */ 1978 } 1979 /* Try a few times, until the buffer fits */ 1980 for (i = 0; i < 20; i++) { 1981 l = buflen; 1982 x = safe_realloc(x, l); 1983 bcopy(oid, x, oid->len); 1984 ret = do_cmd(-IP_DUMMYNET3, x, (uintptr_t)&l); 1985 if (ret != 0 || x->id <= sizeof(*oid)) 1986 goto done; /* no response */ 1987 if (l + max_size <= buflen) 1988 break; /* ok */ 1989 buflen *= 2; /* double for next attempt */ 1990 } 1991 list_pipes(x, O_NEXT(x, l)); 1992 done: 1993 if (x) 1994 free(x); 1995 free(oid); 1996 } 1997