1 /* $OpenBSD: pfctl_altq.c,v 1.83 2004/03/14 21:51:44 dhartmei Exp $ */ 2 /* $DragonFly: src/usr.sbin/pfctl/pfctl_altq.c,v 1.2 2005/02/11 22:31:45 joerg Exp $ */ 3 4 /* 5 * Copyright (c) 2002 6 * Sony Computer Science Laboratories Inc. 7 * Copyright (c) 2002, 2003 Henning Brauer <henning@openbsd.org> 8 * 9 * Permission to use, copy, modify, and distribute this software for any 10 * purpose with or without fee is hereby granted, provided that the above 11 * copyright notice and this permission notice appear in all copies. 12 * 13 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 14 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 15 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 16 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 17 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 18 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 19 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 20 */ 21 22 #include <sys/param.h> 23 #include <sys/ioctl.h> 24 #include <sys/socket.h> 25 #include <sys/sysctl.h> 26 27 #include <net/if.h> 28 #include <net/if_mib.h> 29 #include <netinet/in.h> 30 #include <net/pf/pfvar.h> 31 32 #include <err.h> 33 #include <errno.h> 34 #include <limits.h> 35 #include <math.h> 36 #include <stdio.h> 37 #include <stdlib.h> 38 #include <string.h> 39 #include <unistd.h> 40 41 #include <net/altq/altq.h> 42 #include <net/altq/altq_cbq.h> 43 #include <net/altq/altq_priq.h> 44 #include <net/altq/altq_hfsc.h> 45 46 #include "pfctl_parser.h" 47 #include "pfctl.h" 48 49 #define is_sc_null(sc) (((sc) == NULL) || ((sc)->m1 == 0 && (sc)->m2 == 0)) 50 51 TAILQ_HEAD(altqs, pf_altq) altqs = TAILQ_HEAD_INITIALIZER(altqs); 52 LIST_HEAD(gen_sc, segment) rtsc, lssc; 53 54 struct pf_altq *qname_to_pfaltq(const char *, const char *); 55 u_int32_t qname_to_qid(const char *); 56 57 static int eval_pfqueue_cbq(struct pfctl *, struct pf_altq *); 58 static int cbq_compute_idletime(struct pfctl *, struct pf_altq *); 59 static int check_commit_cbq(int, int, struct pf_altq *); 60 static int print_cbq_opts(const struct pf_altq *); 61 62 static int eval_pfqueue_priq(struct pfctl *, struct pf_altq *); 63 static int check_commit_priq(int, int, struct pf_altq *); 64 static int print_priq_opts(const struct pf_altq *); 65 66 static int eval_pfqueue_hfsc(struct pfctl *, struct pf_altq *); 67 static int check_commit_hfsc(int, int, struct pf_altq *); 68 static int print_hfsc_opts(const struct pf_altq *, 69 const struct node_queue_opt *); 70 71 static void gsc_add_sc(struct gen_sc *, struct service_curve *); 72 static int is_gsc_under_sc(struct gen_sc *, 73 struct service_curve *); 74 static void gsc_destroy(struct gen_sc *); 75 static struct segment *gsc_getentry(struct gen_sc *, double); 76 static int gsc_add_seg(struct gen_sc *, double, double, double, 77 double); 78 static double sc_x2y(struct service_curve *, double); 79 80 u_int32_t getifspeed(const char *); 81 u_long getifmtu(char *); 82 int eval_queue_opts(struct pf_altq *, struct node_queue_opt *, 83 u_int32_t); 84 u_int32_t eval_bwspec(struct node_queue_bw *, u_int32_t); 85 void print_hfsc_sc(const char *, u_int, u_int, u_int, 86 const struct node_hfsc_sc *); 87 88 void 89 pfaltq_store(struct pf_altq *a) 90 { 91 struct pf_altq *altq; 92 93 if ((altq = malloc(sizeof(*altq))) == NULL) 94 err(1, "malloc"); 95 memcpy(altq, a, sizeof(struct pf_altq)); 96 TAILQ_INSERT_TAIL(&altqs, altq, entries); 97 } 98 99 void 100 pfaltq_free(struct pf_altq *a) 101 { 102 struct pf_altq *altq; 103 104 TAILQ_FOREACH(altq, &altqs, entries) { 105 if (strncmp(a->ifname, altq->ifname, IFNAMSIZ) == 0 && 106 strncmp(a->qname, altq->qname, PF_QNAME_SIZE) == 0) { 107 TAILQ_REMOVE(&altqs, altq, entries); 108 free(altq); 109 return; 110 } 111 } 112 } 113 114 struct pf_altq * 115 pfaltq_lookup(const char *ifname) 116 { 117 struct pf_altq *altq; 118 119 TAILQ_FOREACH(altq, &altqs, entries) { 120 if (strncmp(ifname, altq->ifname, IFNAMSIZ) == 0 && 121 altq->qname[0] == 0) 122 return (altq); 123 } 124 return (NULL); 125 } 126 127 struct pf_altq * 128 qname_to_pfaltq(const char *qname, const char *ifname) 129 { 130 struct pf_altq *altq; 131 132 TAILQ_FOREACH(altq, &altqs, entries) { 133 if (strncmp(ifname, altq->ifname, IFNAMSIZ) == 0 && 134 strncmp(qname, altq->qname, PF_QNAME_SIZE) == 0) 135 return (altq); 136 } 137 return (NULL); 138 } 139 140 u_int32_t 141 qname_to_qid(const char *qname) 142 { 143 struct pf_altq *altq; 144 145 /* 146 * We guarantee that same named queues on different interfaces 147 * have the same qid, so we do NOT need to limit matching on 148 * one interface! 149 */ 150 151 TAILQ_FOREACH(altq, &altqs, entries) { 152 if (strncmp(qname, altq->qname, PF_QNAME_SIZE) == 0) 153 return (altq->qid); 154 } 155 return (0); 156 } 157 158 void 159 print_altq(const struct pf_altq *a, unsigned level, struct node_queue_bw *bw, 160 struct node_queue_opt *qopts) 161 { 162 if (a->qname[0] != 0) { 163 print_queue(a, level, bw, 0, qopts); 164 return; 165 } 166 167 printf("altq on %s ", a->ifname); 168 169 switch (a->scheduler) { 170 case ALTQT_CBQ: 171 if (!print_cbq_opts(a)) 172 printf("cbq "); 173 break; 174 case ALTQT_PRIQ: 175 if (!print_priq_opts(a)) 176 printf("priq "); 177 break; 178 case ALTQT_HFSC: 179 if (!print_hfsc_opts(a, qopts)) 180 printf("hfsc "); 181 break; 182 } 183 184 if (bw != NULL && bw->bw_percent > 0) { 185 if (bw->bw_percent < 100) 186 printf("bandwidth %u%% ", bw->bw_percent); 187 } else 188 printf("bandwidth %s ", rate2str((double)a->ifbandwidth)); 189 190 if (a->qlimit != DEFAULT_QLIMIT) 191 printf("qlimit %u ", a->qlimit); 192 printf("tbrsize %u ", a->tbrsize); 193 } 194 195 void 196 print_queue(const struct pf_altq *a, unsigned level, struct node_queue_bw *bw, 197 int print_interface, struct node_queue_opt *qopts) 198 { 199 unsigned i; 200 201 printf("queue "); 202 for (i = 0; i < level; ++i) 203 printf(" "); 204 printf("%s ", a->qname); 205 if (print_interface) 206 printf("on %s ", a->ifname); 207 if (a->scheduler == ALTQT_CBQ || a->scheduler == ALTQT_HFSC) { 208 if (bw != NULL && bw->bw_percent > 0) { 209 if (bw->bw_percent < 100) 210 printf("bandwidth %u%% ", bw->bw_percent); 211 } else 212 printf("bandwidth %s ", rate2str((double)a->bandwidth)); 213 } 214 if (a->priority != DEFAULT_PRIORITY) 215 printf("priority %u ", a->priority); 216 if (a->qlimit != DEFAULT_QLIMIT) 217 printf("qlimit %u ", a->qlimit); 218 switch (a->scheduler) { 219 case ALTQT_CBQ: 220 print_cbq_opts(a); 221 break; 222 case ALTQT_PRIQ: 223 print_priq_opts(a); 224 break; 225 case ALTQT_HFSC: 226 print_hfsc_opts(a, qopts); 227 break; 228 } 229 } 230 231 /* 232 * eval_pfaltq computes the discipline parameters. 233 */ 234 int 235 eval_pfaltq(struct pfctl *pf __unused, struct pf_altq *pa, 236 struct node_queue_bw *bw, struct node_queue_opt *opts) 237 { 238 u_int rate, size, errors = 0; 239 240 if (bw->bw_absolute > 0) 241 pa->ifbandwidth = bw->bw_absolute; 242 else 243 if ((rate = getifspeed(pa->ifname)) == 0) { 244 fprintf(stderr, "cannot determine interface bandwidth " 245 "for %s, specify an absolute bandwidth\n", 246 pa->ifname); 247 errors++; 248 } else if ((pa->ifbandwidth = eval_bwspec(bw, rate)) == 0) 249 pa->ifbandwidth = rate; 250 251 errors += eval_queue_opts(pa, opts, pa->ifbandwidth); 252 253 /* if tbrsize is not specified, use heuristics */ 254 if (pa->tbrsize == 0) { 255 rate = pa->ifbandwidth; 256 if (rate <= 1 * 1000 * 1000) 257 size = 1; 258 else if (rate <= 10 * 1000 * 1000) 259 size = 4; 260 else if (rate <= 200 * 1000 * 1000) 261 size = 8; 262 else 263 size = 24; 264 size = size * getifmtu(pa->ifname); 265 if (size > 0xffff) 266 size = 0xffff; 267 pa->tbrsize = size; 268 } 269 return (errors); 270 } 271 272 /* 273 * check_commit_altq does consistency check for each interface 274 */ 275 int 276 check_commit_altq(int dev, int opts) 277 { 278 struct pf_altq *altq; 279 int error = 0; 280 281 /* call the discipline check for each interface. */ 282 TAILQ_FOREACH(altq, &altqs, entries) { 283 if (altq->qname[0] == 0) { 284 switch (altq->scheduler) { 285 case ALTQT_CBQ: 286 error = check_commit_cbq(dev, opts, altq); 287 break; 288 case ALTQT_PRIQ: 289 error = check_commit_priq(dev, opts, altq); 290 break; 291 case ALTQT_HFSC: 292 error = check_commit_hfsc(dev, opts, altq); 293 break; 294 default: 295 break; 296 } 297 } 298 } 299 return (error); 300 } 301 302 /* 303 * eval_pfqueue computes the queue parameters. 304 */ 305 int 306 eval_pfqueue(struct pfctl *pf, struct pf_altq *pa, struct node_queue_bw *bw, 307 struct node_queue_opt *opts) 308 { 309 /* should be merged with expand_queue */ 310 struct pf_altq *if_pa, *parent; 311 int error = 0; 312 313 /* find the corresponding interface and copy fields used by queues */ 314 if ((if_pa = pfaltq_lookup(pa->ifname)) == NULL) { 315 fprintf(stderr, "altq not defined on %s\n", pa->ifname); 316 return (1); 317 } 318 pa->scheduler = if_pa->scheduler; 319 pa->ifbandwidth = if_pa->ifbandwidth; 320 321 if (qname_to_pfaltq(pa->qname, pa->ifname) != NULL) { 322 fprintf(stderr, "queue %s already exists on interface %s\n", 323 pa->qname, pa->ifname); 324 return (1); 325 } 326 pa->qid = qname_to_qid(pa->qname); 327 328 parent = NULL; 329 if (pa->parent[0] != 0) { 330 parent = qname_to_pfaltq(pa->parent, pa->ifname); 331 if (parent == NULL) { 332 fprintf(stderr, "parent %s not found for %s\n", 333 pa->parent, pa->qname); 334 return (1); 335 } 336 pa->parent_qid = parent->qid; 337 } 338 if (pa->qlimit == 0) 339 pa->qlimit = DEFAULT_QLIMIT; 340 341 if (pa->scheduler == ALTQT_CBQ || pa->scheduler == ALTQT_HFSC) { 342 if ((pa->bandwidth = eval_bwspec(bw, 343 parent == NULL ? 0 : parent->bandwidth)) == 0) { 344 fprintf(stderr, "bandwidth for %s invalid (%d / %d)\n", 345 pa->qname, bw->bw_absolute, bw->bw_percent); 346 return (1); 347 } 348 349 if (pa->bandwidth > pa->ifbandwidth) { 350 fprintf(stderr, "bandwidth for %s higher than " 351 "interface\n", pa->qname); 352 return (1); 353 } 354 if (parent != NULL && pa->bandwidth > parent->bandwidth) { 355 fprintf(stderr, "bandwidth for %s higher than parent\n", 356 pa->qname); 357 return (1); 358 } 359 } 360 361 if (eval_queue_opts(pa, opts, parent == NULL? 0 : parent->bandwidth)) 362 return (1); 363 364 switch (pa->scheduler) { 365 case ALTQT_CBQ: 366 error = eval_pfqueue_cbq(pf, pa); 367 break; 368 case ALTQT_PRIQ: 369 error = eval_pfqueue_priq(pf, pa); 370 break; 371 case ALTQT_HFSC: 372 error = eval_pfqueue_hfsc(pf, pa); 373 break; 374 default: 375 break; 376 } 377 return (error); 378 } 379 380 /* 381 * CBQ support functions 382 */ 383 #define RM_FILTER_GAIN 5 /* log2 of gain, e.g., 5 => 31/32 */ 384 #define RM_NS_PER_SEC (1000000000) 385 386 static int 387 eval_pfqueue_cbq(struct pfctl *pf, struct pf_altq *pa) 388 { 389 struct cbq_opts *opts; 390 u_int ifmtu; 391 392 if (pa->priority >= CBQ_MAXPRI) { 393 warnx("priority out of range: max %d", CBQ_MAXPRI - 1); 394 return (-1); 395 } 396 397 ifmtu = getifmtu(pa->ifname); 398 opts = &pa->pq_u.cbq_opts; 399 400 if (opts->pktsize == 0) { /* use default */ 401 opts->pktsize = ifmtu; 402 if (opts->pktsize > MCLBYTES) /* do what TCP does */ 403 opts->pktsize &= ~MCLBYTES; 404 } else if (opts->pktsize > ifmtu) 405 opts->pktsize = ifmtu; 406 if (opts->maxpktsize == 0) /* use default */ 407 opts->maxpktsize = ifmtu; 408 else if (opts->maxpktsize > ifmtu) 409 opts->pktsize = ifmtu; 410 411 if (opts->pktsize > opts->maxpktsize) 412 opts->pktsize = opts->maxpktsize; 413 414 if (pa->parent[0] == 0) 415 opts->flags |= (CBQCLF_ROOTCLASS | CBQCLF_WRR); 416 417 cbq_compute_idletime(pf, pa); 418 return (0); 419 } 420 421 /* 422 * compute ns_per_byte, maxidle, minidle, and offtime 423 */ 424 static int 425 cbq_compute_idletime(struct pfctl *pf, struct pf_altq *pa) 426 { 427 struct cbq_opts *opts; 428 double maxidle_s, maxidle, minidle; 429 double offtime, nsPerByte, ifnsPerByte, ptime, cptime; 430 double z, g, f, gton, gtom; 431 u_int minburst, maxburst; 432 433 opts = &pa->pq_u.cbq_opts; 434 ifnsPerByte = (1.0 / (double)pa->ifbandwidth) * RM_NS_PER_SEC * 8; 435 minburst = opts->minburst; 436 maxburst = opts->maxburst; 437 438 if (pa->bandwidth == 0) 439 f = 0.0001; /* small enough? */ 440 else 441 f = ((double) pa->bandwidth / (double) pa->ifbandwidth); 442 443 nsPerByte = ifnsPerByte / f; 444 ptime = (double)opts->pktsize * ifnsPerByte; 445 cptime = ptime * (1.0 - f) / f; 446 447 if (nsPerByte * (double)opts->maxpktsize > (double)INT_MAX) { 448 /* 449 * this causes integer overflow in kernel! 450 * (bandwidth < 6Kbps when max_pkt_size=1500) 451 */ 452 if (pa->bandwidth != 0 && (pf->opts & PF_OPT_QUIET) == 0) 453 warnx("queue bandwidth must be larger than %s", 454 rate2str(ifnsPerByte * (double)opts->maxpktsize / 455 (double)INT_MAX * (double)pa->ifbandwidth)); 456 fprintf(stderr, "cbq: queue %s is too slow!\n", 457 pa->qname); 458 nsPerByte = (double)(INT_MAX / opts->maxpktsize); 459 } 460 461 if (maxburst == 0) { /* use default */ 462 if (cptime > 10.0 * 1000000) 463 maxburst = 4; 464 else 465 maxburst = 16; 466 } 467 if (minburst == 0) /* use default */ 468 minburst = 2; 469 if (minburst > maxburst) 470 minburst = maxburst; 471 472 z = (double)(1 << RM_FILTER_GAIN); 473 g = (1.0 - 1.0 / z); 474 gton = pow(g, (double)maxburst); 475 gtom = pow(g, (double)(minburst-1)); 476 maxidle = ((1.0 / f - 1.0) * ((1.0 - gton) / gton)); 477 maxidle_s = (1.0 - g); 478 if (maxidle > maxidle_s) 479 maxidle = ptime * maxidle; 480 else 481 maxidle = ptime * maxidle_s; 482 if (minburst) 483 offtime = cptime * (1.0 + 1.0/(1.0 - g) * (1.0 - gtom) / gtom); 484 else 485 offtime = cptime; 486 minidle = -((double)opts->maxpktsize * (double)nsPerByte); 487 488 /* scale parameters */ 489 maxidle = ((maxidle * 8.0) / nsPerByte) * 490 pow(2.0, (double)RM_FILTER_GAIN); 491 offtime = (offtime * 8.0) / nsPerByte * 492 pow(2.0, (double)RM_FILTER_GAIN); 493 minidle = ((minidle * 8.0) / nsPerByte) * 494 pow(2.0, (double)RM_FILTER_GAIN); 495 496 maxidle = maxidle / 1000.0; 497 offtime = offtime / 1000.0; 498 minidle = minidle / 1000.0; 499 500 opts->minburst = minburst; 501 opts->maxburst = maxburst; 502 opts->ns_per_byte = (u_int)nsPerByte; 503 opts->maxidle = (u_int)fabs(maxidle); 504 opts->minidle = (int)minidle; 505 opts->offtime = (u_int)fabs(offtime); 506 507 return (0); 508 } 509 510 static int 511 check_commit_cbq(int dev __unused, int opts __unused, struct pf_altq *pa) 512 { 513 struct pf_altq *altq; 514 int root_class, default_class; 515 int error = 0; 516 517 /* 518 * check if cbq has one root queue and one default queue 519 * for this interface 520 */ 521 root_class = default_class = 0; 522 TAILQ_FOREACH(altq, &altqs, entries) { 523 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0) 524 continue; 525 if (altq->qname[0] == 0) /* this is for interface */ 526 continue; 527 if (altq->pq_u.cbq_opts.flags & CBQCLF_ROOTCLASS) 528 root_class++; 529 if (altq->pq_u.cbq_opts.flags & CBQCLF_DEFCLASS) 530 default_class++; 531 } 532 if (root_class != 1) { 533 warnx("should have one root queue on %s", pa->ifname); 534 error++; 535 } 536 if (default_class != 1) { 537 warnx("should have one default queue on %s", pa->ifname); 538 error++; 539 } 540 return (error); 541 } 542 543 static int 544 print_cbq_opts(const struct pf_altq *a) 545 { 546 const struct cbq_opts *opts; 547 548 opts = &a->pq_u.cbq_opts; 549 if (opts->flags) { 550 printf("cbq("); 551 if (opts->flags & CBQCLF_RED) 552 printf(" red"); 553 if (opts->flags & CBQCLF_ECN) 554 printf(" ecn"); 555 if (opts->flags & CBQCLF_RIO) 556 printf(" rio"); 557 if (opts->flags & CBQCLF_CLEARDSCP) 558 printf(" cleardscp"); 559 if (opts->flags & CBQCLF_BORROW) 560 printf(" borrow"); 561 if (opts->flags & CBQCLF_WRR) 562 printf(" wrr"); 563 if (opts->flags & CBQCLF_EFFICIENT) 564 printf(" efficient"); 565 if (opts->flags & CBQCLF_ROOTCLASS) 566 printf(" root"); 567 if (opts->flags & CBQCLF_DEFCLASS) 568 printf(" default"); 569 printf(" ) "); 570 571 return (1); 572 } else 573 return (0); 574 } 575 576 /* 577 * PRIQ support functions 578 */ 579 static int 580 eval_pfqueue_priq(struct pfctl *pf __unused, struct pf_altq *pa) 581 { 582 struct pf_altq *altq; 583 584 if (pa->priority >= PRIQ_MAXPRI) { 585 warnx("priority out of range: max %d", PRIQ_MAXPRI - 1); 586 return (-1); 587 } 588 /* the priority should be unique for the interface */ 589 TAILQ_FOREACH(altq, &altqs, entries) { 590 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) == 0 && 591 altq->qname[0] != 0 && altq->priority == pa->priority) { 592 warnx("%s and %s have the same priority", 593 altq->qname, pa->qname); 594 return (-1); 595 } 596 } 597 598 return (0); 599 } 600 601 static int 602 check_commit_priq(int dev __unused, int opts __unused, struct pf_altq *pa) 603 { 604 struct pf_altq *altq; 605 int default_class; 606 int error = 0; 607 608 /* 609 * check if priq has one default class for this interface 610 */ 611 default_class = 0; 612 TAILQ_FOREACH(altq, &altqs, entries) { 613 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0) 614 continue; 615 if (altq->qname[0] == 0) /* this is for interface */ 616 continue; 617 if (altq->pq_u.priq_opts.flags & PRCF_DEFAULTCLASS) 618 default_class++; 619 } 620 if (default_class != 1) { 621 warnx("should have one default queue on %s", pa->ifname); 622 error++; 623 } 624 return (error); 625 } 626 627 static int 628 print_priq_opts(const struct pf_altq *a) 629 { 630 const struct priq_opts *opts; 631 632 opts = &a->pq_u.priq_opts; 633 634 if (opts->flags) { 635 printf("priq("); 636 if (opts->flags & PRCF_RED) 637 printf(" red"); 638 if (opts->flags & PRCF_ECN) 639 printf(" ecn"); 640 if (opts->flags & PRCF_RIO) 641 printf(" rio"); 642 if (opts->flags & PRCF_CLEARDSCP) 643 printf(" cleardscp"); 644 if (opts->flags & PRCF_DEFAULTCLASS) 645 printf(" default"); 646 printf(" ) "); 647 648 return (1); 649 } else 650 return (0); 651 } 652 653 /* 654 * HFSC support functions 655 */ 656 static int 657 eval_pfqueue_hfsc(struct pfctl *pf __unused, struct pf_altq *pa) 658 { 659 struct pf_altq *altq, *parent; 660 struct hfsc_opts *opts; 661 struct service_curve sc; 662 663 opts = &pa->pq_u.hfsc_opts; 664 665 if (pa->parent[0] == 0) { 666 /* root queue */ 667 opts->lssc_m1 = pa->ifbandwidth; 668 opts->lssc_m2 = pa->ifbandwidth; 669 opts->lssc_d = 0; 670 return (0); 671 } 672 673 LIST_INIT(&rtsc); 674 LIST_INIT(&lssc); 675 676 /* if link_share is not specified, use bandwidth */ 677 if (opts->lssc_m2 == 0) 678 opts->lssc_m2 = pa->bandwidth; 679 680 if ((opts->rtsc_m1 > 0 && opts->rtsc_m2 == 0) || 681 (opts->lssc_m1 > 0 && opts->lssc_m2 == 0) || 682 (opts->ulsc_m1 > 0 && opts->ulsc_m2 == 0)) { 683 warnx("m2 is zero for %s", pa->qname); 684 return (-1); 685 } 686 687 if ((opts->rtsc_m1 < opts->rtsc_m2 && opts->rtsc_m1 != 0) || 688 (opts->rtsc_m1 < opts->rtsc_m2 && opts->rtsc_m1 != 0) || 689 (opts->rtsc_m1 < opts->rtsc_m2 && opts->rtsc_m1 != 0)) { 690 warnx("m1 must be zero for convex curve: %s", pa->qname); 691 return (-1); 692 } 693 694 /* 695 * admission control: 696 * for the real-time service curve, the sum of the service curves 697 * should not exceed 80% of the interface bandwidth. 20% is reserved 698 * not to over-commit the actual interface bandwidth. 699 * for the link-sharing service curve, the sum of the child service 700 * curve should not exceed the parent service curve. 701 * for the upper-limit service curve, the assigned bandwidth should 702 * be smaller than the interface bandwidth, and the upper-limit should 703 * be larger than the real-time service curve when both are defined. 704 */ 705 parent = qname_to_pfaltq(pa->parent, pa->ifname); 706 if (parent == NULL) 707 errx(1, "parent %s not found for %s", pa->parent, pa->qname); 708 709 TAILQ_FOREACH(altq, &altqs, entries) { 710 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0) 711 continue; 712 if (altq->qname[0] == 0) /* this is for interface */ 713 continue; 714 715 /* if the class has a real-time service curve, add it. */ 716 if (opts->rtsc_m2 != 0 && altq->pq_u.hfsc_opts.rtsc_m2 != 0) { 717 sc.m1 = altq->pq_u.hfsc_opts.rtsc_m1; 718 sc.d = altq->pq_u.hfsc_opts.rtsc_d; 719 sc.m2 = altq->pq_u.hfsc_opts.rtsc_m2; 720 gsc_add_sc(&rtsc, &sc); 721 } 722 723 if (strncmp(altq->parent, pa->parent, PF_QNAME_SIZE) != 0) 724 continue; 725 726 /* if the class has a link-sharing service curve, add it. */ 727 if (opts->lssc_m2 != 0 && altq->pq_u.hfsc_opts.lssc_m2 != 0) { 728 sc.m1 = altq->pq_u.hfsc_opts.lssc_m1; 729 sc.d = altq->pq_u.hfsc_opts.lssc_d; 730 sc.m2 = altq->pq_u.hfsc_opts.lssc_m2; 731 gsc_add_sc(&lssc, &sc); 732 } 733 } 734 735 /* check the real-time service curve. reserve 20% of interface bw */ 736 if (opts->rtsc_m2 != 0) { 737 sc.m1 = 0; 738 sc.d = 0; 739 sc.m2 = pa->ifbandwidth / 100 * 80; 740 if (!is_gsc_under_sc(&rtsc, &sc)) { 741 warnx("real-time sc exceeds the interface bandwidth"); 742 goto err_ret; 743 } 744 } 745 746 /* check the link-sharing service curve. */ 747 if (opts->lssc_m2 != 0) { 748 sc.m1 = parent->pq_u.hfsc_opts.lssc_m1; 749 sc.d = parent->pq_u.hfsc_opts.lssc_d; 750 sc.m2 = parent->pq_u.hfsc_opts.lssc_m2; 751 if (!is_gsc_under_sc(&lssc, &sc)) { 752 warnx("link-sharing sc exceeds parent's sc"); 753 goto err_ret; 754 } 755 } 756 757 /* check the upper-limit service curve. */ 758 if (opts->ulsc_m2 != 0) { 759 if (opts->ulsc_m1 > pa->ifbandwidth || 760 opts->ulsc_m2 > pa->ifbandwidth) { 761 warnx("upper-limit larger than interface bandwidth"); 762 goto err_ret; 763 } 764 if (opts->rtsc_m2 != 0 && opts->rtsc_m2 > opts->ulsc_m2) { 765 warnx("upper-limit sc smaller than real-time sc"); 766 goto err_ret; 767 } 768 } 769 770 gsc_destroy(&rtsc); 771 gsc_destroy(&lssc); 772 773 return (0); 774 775 err_ret: 776 gsc_destroy(&rtsc); 777 gsc_destroy(&lssc); 778 return (-1); 779 } 780 781 static int 782 check_commit_hfsc(int dev __unused, int opts __unused, struct pf_altq *pa) 783 { 784 struct pf_altq *altq, *def = NULL; 785 int default_class; 786 int error = 0; 787 788 /* check if hfsc has one default queue for this interface */ 789 default_class = 0; 790 TAILQ_FOREACH(altq, &altqs, entries) { 791 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0) 792 continue; 793 if (altq->qname[0] == 0) /* this is for interface */ 794 continue; 795 if (altq->parent[0] == 0) /* dummy root */ 796 continue; 797 if (altq->pq_u.hfsc_opts.flags & HFCF_DEFAULTCLASS) { 798 default_class++; 799 def = altq; 800 } 801 } 802 if (default_class != 1) { 803 warnx("should have one default queue on %s", pa->ifname); 804 return (1); 805 } 806 /* make sure the default queue is a leaf */ 807 TAILQ_FOREACH(altq, &altqs, entries) { 808 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0) 809 continue; 810 if (altq->qname[0] == 0) /* this is for interface */ 811 continue; 812 if (strncmp(altq->parent, def->qname, PF_QNAME_SIZE) == 0) { 813 warnx("default queue is not a leaf"); 814 error++; 815 } 816 } 817 return (error); 818 } 819 820 static int 821 print_hfsc_opts(const struct pf_altq *a, const struct node_queue_opt *qopts) 822 { 823 const struct hfsc_opts *opts; 824 const struct node_hfsc_sc *loc_rtsc, *loc_lssc, *ulsc; 825 826 opts = &a->pq_u.hfsc_opts; 827 if (qopts == NULL) 828 loc_rtsc = loc_lssc = ulsc = NULL; 829 else { 830 loc_rtsc = &qopts->data.hfsc_opts.realtime; 831 loc_lssc = &qopts->data.hfsc_opts.linkshare; 832 ulsc = &qopts->data.hfsc_opts.upperlimit; 833 } 834 835 if (opts->flags || opts->rtsc_m2 != 0 || opts->ulsc_m2 != 0 || 836 (opts->lssc_m2 != 0 && (opts->lssc_m2 != a->bandwidth || 837 opts->lssc_d != 0))) { 838 printf("hfsc("); 839 if (opts->flags & HFCF_RED) 840 printf(" red"); 841 if (opts->flags & HFCF_ECN) 842 printf(" ecn"); 843 if (opts->flags & HFCF_RIO) 844 printf(" rio"); 845 if (opts->flags & HFCF_CLEARDSCP) 846 printf(" cleardscp"); 847 if (opts->flags & HFCF_DEFAULTCLASS) 848 printf(" default"); 849 if (opts->rtsc_m2 != 0) 850 print_hfsc_sc("realtime", opts->rtsc_m1, opts->rtsc_d, 851 opts->rtsc_m2, loc_rtsc); 852 if (opts->lssc_m2 != 0 && (opts->lssc_m2 != a->bandwidth || 853 opts->lssc_d != 0)) 854 print_hfsc_sc("linkshare", opts->lssc_m1, opts->lssc_d, 855 opts->lssc_m2, loc_lssc); 856 if (opts->ulsc_m2 != 0) 857 print_hfsc_sc("upperlimit", opts->ulsc_m1, opts->ulsc_d, 858 opts->ulsc_m2, ulsc); 859 printf(" ) "); 860 861 return (1); 862 } else 863 return (0); 864 } 865 866 /* 867 * admission control using generalized service curve 868 */ 869 #define INFINITY HUGE_VAL /* positive infinity defined in <math.h> */ 870 871 /* add a new service curve to a generalized service curve */ 872 static void 873 gsc_add_sc(struct gen_sc *gsc, struct service_curve *sc) 874 { 875 if (is_sc_null(sc)) 876 return; 877 if (sc->d != 0) 878 gsc_add_seg(gsc, 0.0, 0.0, (double)sc->d, (double)sc->m1); 879 gsc_add_seg(gsc, (double)sc->d, 0.0, INFINITY, (double)sc->m2); 880 } 881 882 /* 883 * check whether all points of a generalized service curve have 884 * their y-coordinates no larger than a given two-piece linear 885 * service curve. 886 */ 887 static int 888 is_gsc_under_sc(struct gen_sc *gsc, struct service_curve *sc) 889 { 890 struct segment *s, *last, *end; 891 double y; 892 893 if (is_sc_null(sc)) { 894 if (LIST_EMPTY(gsc)) 895 return (1); 896 LIST_FOREACH(s, gsc, _next) { 897 if (s->m != 0) 898 return (0); 899 } 900 return (1); 901 } 902 /* 903 * gsc has a dummy entry at the end with x = INFINITY. 904 * loop through up to this dummy entry. 905 */ 906 end = gsc_getentry(gsc, INFINITY); 907 if (end == NULL) 908 return (1); 909 last = NULL; 910 for (s = LIST_FIRST(gsc); s != end; s = LIST_NEXT(s, _next)) { 911 if (s->y > sc_x2y(sc, s->x)) 912 return (0); 913 last = s; 914 } 915 /* last now holds the real last segment */ 916 if (last == NULL) 917 return (1); 918 if (last->m > sc->m2) 919 return (0); 920 if (last->x < sc->d && last->m > sc->m1) { 921 y = last->y + (sc->d - last->x) * last->m; 922 if (y > sc_x2y(sc, sc->d)) 923 return (0); 924 } 925 return (1); 926 } 927 928 static void 929 gsc_destroy(struct gen_sc *gsc) 930 { 931 struct segment *s; 932 933 while ((s = LIST_FIRST(gsc)) != NULL) { 934 LIST_REMOVE(s, _next); 935 free(s); 936 } 937 } 938 939 /* 940 * return a segment entry starting at x. 941 * if gsc has no entry starting at x, a new entry is created at x. 942 */ 943 static struct segment * 944 gsc_getentry(struct gen_sc *gsc, double x) 945 { 946 struct segment *new, *prev, *s; 947 948 prev = NULL; 949 LIST_FOREACH(s, gsc, _next) { 950 if (s->x == x) 951 return (s); /* matching entry found */ 952 else if (s->x < x) 953 prev = s; 954 else 955 break; 956 } 957 958 /* we have to create a new entry */ 959 if ((new = calloc(1, sizeof(struct segment))) == NULL) 960 return (NULL); 961 962 new->x = x; 963 if (x == INFINITY || s == NULL) 964 new->d = 0; 965 else if (s->x == INFINITY) 966 new->d = INFINITY; 967 else 968 new->d = s->x - x; 969 if (prev == NULL) { 970 /* insert the new entry at the head of the list */ 971 new->y = 0; 972 new->m = 0; 973 LIST_INSERT_HEAD(gsc, new, _next); 974 } else { 975 /* 976 * the start point intersects with the segment pointed by 977 * prev. divide prev into 2 segments 978 */ 979 if (x == INFINITY) { 980 prev->d = INFINITY; 981 if (prev->m == 0) 982 new->y = prev->y; 983 else 984 new->y = INFINITY; 985 } else { 986 prev->d = x - prev->x; 987 new->y = prev->d * prev->m + prev->y; 988 } 989 new->m = prev->m; 990 LIST_INSERT_AFTER(prev, new, _next); 991 } 992 return (new); 993 } 994 995 /* add a segment to a generalized service curve */ 996 static int 997 gsc_add_seg(struct gen_sc *gsc, double x, double y, double d, double m) 998 { 999 struct segment *start, *end, *s; 1000 double x2; 1001 1002 if (d == INFINITY) 1003 x2 = INFINITY; 1004 else 1005 x2 = x + d; 1006 start = gsc_getentry(gsc, x); 1007 end = gsc_getentry(gsc, x2); 1008 if (start == NULL || end == NULL) 1009 return (-1); 1010 1011 for (s = start; s != end; s = LIST_NEXT(s, _next)) { 1012 s->m += m; 1013 s->y += y + (s->x - x) * m; 1014 } 1015 1016 end = gsc_getentry(gsc, INFINITY); 1017 for (; s != end; s = LIST_NEXT(s, _next)) { 1018 s->y += m * d; 1019 } 1020 1021 return (0); 1022 } 1023 1024 /* get y-projection of a service curve */ 1025 static double 1026 sc_x2y(struct service_curve *sc, double x) 1027 { 1028 double y; 1029 1030 if (x <= (double)sc->d) 1031 /* y belongs to the 1st segment */ 1032 y = x * (double)sc->m1; 1033 else 1034 /* y belongs to the 2nd segment */ 1035 y = (double)sc->d * (double)sc->m1 1036 + (x - (double)sc->d) * (double)sc->m2; 1037 return (y); 1038 } 1039 1040 /* 1041 * misc utilities 1042 */ 1043 #define R2S_BUFS 8 1044 #define RATESTR_MAX 16 1045 1046 char * 1047 rate2str(double rate) 1048 { 1049 char *buf; 1050 static char r2sbuf[R2S_BUFS][RATESTR_MAX]; /* ring bufer */ 1051 static int idx = 0; 1052 int i; 1053 static const char unit[] = " KMG"; 1054 1055 buf = r2sbuf[idx++]; 1056 if (idx == R2S_BUFS) 1057 idx = 0; 1058 1059 for (i = 0; rate >= 1000 && i <= 3; i++) 1060 rate /= 1000; 1061 1062 if ((int)(rate * 100) % 100) 1063 snprintf(buf, RATESTR_MAX, "%.2f%cb", rate, unit[i]); 1064 else 1065 snprintf(buf, RATESTR_MAX, "%d%cb", (int)rate, unit[i]); 1066 1067 return (buf); 1068 } 1069 1070 u_int32_t 1071 getifspeed(const char *ifname) 1072 { 1073 size_t datalen; 1074 int idx; 1075 struct ifmibdata data; 1076 int name[] = { 1077 CTL_NET, 1078 PF_LINK, 1079 NETLINK_GENERIC, 1080 IFMIB_IFDATA, 1081 0, 1082 IFDATA_GENERAL 1083 }; 1084 1085 if ((idx = (int)if_nametoindex(ifname)) == 0) 1086 err(1, "getifspeed: if_nametoindex"); 1087 name[4] = idx; 1088 1089 datalen = sizeof(data); 1090 if (sysctl(name, 6, &data, &datalen, NULL, 0)) 1091 err(1, "getifspeed: sysctl"); 1092 1093 return(data.ifmd_data.ifi_baudrate); 1094 } 1095 1096 u_long 1097 getifmtu(char *ifname) 1098 { 1099 int s; 1100 struct ifreq ifr; 1101 1102 if ((s = socket(AF_INET, SOCK_DGRAM, 0)) < 0) 1103 err(1, "socket"); 1104 if (strlcpy(ifr.ifr_name, ifname, sizeof(ifr.ifr_name)) >= 1105 sizeof(ifr.ifr_name)) 1106 errx(1, "getifmtu: strlcpy"); 1107 if (ioctl(s, SIOCGIFMTU, (caddr_t)&ifr) == -1) 1108 err(1, "SIOCGIFMTU"); 1109 if (shutdown(s, SHUT_RDWR) == -1) 1110 err(1, "shutdown"); 1111 if (close(s)) 1112 err(1, "close"); 1113 if (ifr.ifr_mtu > 0) 1114 return (ifr.ifr_mtu); 1115 else { 1116 warnx("could not get mtu for %s, assuming 1500", ifname); 1117 return (1500); 1118 } 1119 } 1120 1121 int 1122 eval_queue_opts(struct pf_altq *pa, struct node_queue_opt *opts, 1123 u_int32_t ref_bw) 1124 { 1125 int errors = 0; 1126 1127 switch (pa->scheduler) { 1128 case ALTQT_CBQ: 1129 pa->pq_u.cbq_opts = opts->data.cbq_opts; 1130 break; 1131 case ALTQT_PRIQ: 1132 pa->pq_u.priq_opts = opts->data.priq_opts; 1133 break; 1134 case ALTQT_HFSC: 1135 pa->pq_u.hfsc_opts.flags = opts->data.hfsc_opts.flags; 1136 if (opts->data.hfsc_opts.linkshare.used) { 1137 pa->pq_u.hfsc_opts.lssc_m1 = 1138 eval_bwspec(&opts->data.hfsc_opts.linkshare.m1, 1139 ref_bw); 1140 pa->pq_u.hfsc_opts.lssc_m2 = 1141 eval_bwspec(&opts->data.hfsc_opts.linkshare.m2, 1142 ref_bw); 1143 pa->pq_u.hfsc_opts.lssc_d = 1144 opts->data.hfsc_opts.linkshare.d; 1145 } 1146 if (opts->data.hfsc_opts.realtime.used) { 1147 pa->pq_u.hfsc_opts.rtsc_m1 = 1148 eval_bwspec(&opts->data.hfsc_opts.realtime.m1, 1149 ref_bw); 1150 pa->pq_u.hfsc_opts.rtsc_m2 = 1151 eval_bwspec(&opts->data.hfsc_opts.realtime.m2, 1152 ref_bw); 1153 pa->pq_u.hfsc_opts.rtsc_d = 1154 opts->data.hfsc_opts.realtime.d; 1155 } 1156 if (opts->data.hfsc_opts.upperlimit.used) { 1157 pa->pq_u.hfsc_opts.ulsc_m1 = 1158 eval_bwspec(&opts->data.hfsc_opts.upperlimit.m1, 1159 ref_bw); 1160 pa->pq_u.hfsc_opts.ulsc_m2 = 1161 eval_bwspec(&opts->data.hfsc_opts.upperlimit.m2, 1162 ref_bw); 1163 pa->pq_u.hfsc_opts.ulsc_d = 1164 opts->data.hfsc_opts.upperlimit.d; 1165 } 1166 break; 1167 default: 1168 warnx("eval_queue_opts: unknown scheduler type %u", 1169 opts->qtype); 1170 errors++; 1171 break; 1172 } 1173 1174 return (errors); 1175 } 1176 1177 u_int32_t 1178 eval_bwspec(struct node_queue_bw *bw, u_int32_t ref_bw) 1179 { 1180 if (bw->bw_absolute > 0) 1181 return (bw->bw_absolute); 1182 1183 if (bw->bw_percent > 0) 1184 return (ref_bw / 100 * bw->bw_percent); 1185 1186 return (0); 1187 } 1188 1189 void 1190 print_hfsc_sc(const char *scname, u_int m1, u_int d, u_int m2, 1191 const struct node_hfsc_sc *sc) 1192 { 1193 printf(" %s", scname); 1194 1195 if (d != 0) { 1196 printf("("); 1197 if (sc != NULL && sc->m1.bw_percent > 0) 1198 printf("%u%%", sc->m1.bw_percent); 1199 else 1200 printf("%s", rate2str((double)m1)); 1201 printf(" %u", d); 1202 } 1203 1204 if (sc != NULL && sc->m2.bw_percent > 0) 1205 printf(" %u%%", sc->m2.bw_percent); 1206 else 1207 printf(" %s", rate2str((double)m2)); 1208 1209 if (d != 0) 1210 printf(")"); 1211 } 1212