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 1.1 2004/09/21 21:25:28 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/types.h> 23 #include <sys/ioctl.h> 24 #include <sys/socket.h> 25 26 #include <net/if.h> 27 #include <netinet/in.h> 28 #include <net/pf/pfvar.h> 29 30 #include <err.h> 31 #include <errno.h> 32 #include <limits.h> 33 #include <math.h> 34 #include <stdio.h> 35 #include <stdlib.h> 36 #include <string.h> 37 #include <unistd.h> 38 39 #include <altq/altq.h> 40 #include <altq/altq_cbq.h> 41 #include <altq/altq_priq.h> 42 #include <altq/altq_hfsc.h> 43 44 #include "pfctl_parser.h" 45 #include "pfctl.h" 46 47 #define is_sc_null(sc) (((sc) == NULL) || ((sc)->m1 == 0 && (sc)->m2 == 0)) 48 49 TAILQ_HEAD(altqs, pf_altq) altqs = TAILQ_HEAD_INITIALIZER(altqs); 50 LIST_HEAD(gen_sc, segment) rtsc, lssc; 51 52 struct pf_altq *qname_to_pfaltq(const char *, const char *); 53 u_int32_t qname_to_qid(const char *); 54 55 static int eval_pfqueue_cbq(struct pfctl *, struct pf_altq *); 56 static int cbq_compute_idletime(struct pfctl *, struct pf_altq *); 57 static int check_commit_cbq(int, int, struct pf_altq *); 58 static int print_cbq_opts(const struct pf_altq *); 59 60 static int eval_pfqueue_priq(struct pfctl *, struct pf_altq *); 61 static int check_commit_priq(int, int, struct pf_altq *); 62 static int print_priq_opts(const struct pf_altq *); 63 64 static int eval_pfqueue_hfsc(struct pfctl *, struct pf_altq *); 65 static int check_commit_hfsc(int, int, struct pf_altq *); 66 static int print_hfsc_opts(const struct pf_altq *, 67 const struct node_queue_opt *); 68 69 static void gsc_add_sc(struct gen_sc *, struct service_curve *); 70 static int is_gsc_under_sc(struct gen_sc *, 71 struct service_curve *); 72 static void gsc_destroy(struct gen_sc *); 73 static struct segment *gsc_getentry(struct gen_sc *, double); 74 static int gsc_add_seg(struct gen_sc *, double, double, double, 75 double); 76 static double sc_x2y(struct service_curve *, double); 77 78 u_int32_t getifspeed(char *); 79 u_long getifmtu(char *); 80 int eval_queue_opts(struct pf_altq *, struct node_queue_opt *, 81 u_int32_t); 82 u_int32_t eval_bwspec(struct node_queue_bw *, u_int32_t); 83 void print_hfsc_sc(const char *, u_int, u_int, u_int, 84 const struct node_hfsc_sc *); 85 86 void 87 pfaltq_store(struct pf_altq *a) 88 { 89 struct pf_altq *altq; 90 91 if ((altq = malloc(sizeof(*altq))) == NULL) 92 err(1, "malloc"); 93 memcpy(altq, a, sizeof(struct pf_altq)); 94 TAILQ_INSERT_TAIL(&altqs, altq, entries); 95 } 96 97 void 98 pfaltq_free(struct pf_altq *a) 99 { 100 struct pf_altq *altq; 101 102 TAILQ_FOREACH(altq, &altqs, entries) { 103 if (strncmp(a->ifname, altq->ifname, IFNAMSIZ) == 0 && 104 strncmp(a->qname, altq->qname, PF_QNAME_SIZE) == 0) { 105 TAILQ_REMOVE(&altqs, altq, entries); 106 free(altq); 107 return; 108 } 109 } 110 } 111 112 struct pf_altq * 113 pfaltq_lookup(const char *ifname) 114 { 115 struct pf_altq *altq; 116 117 TAILQ_FOREACH(altq, &altqs, entries) { 118 if (strncmp(ifname, altq->ifname, IFNAMSIZ) == 0 && 119 altq->qname[0] == 0) 120 return (altq); 121 } 122 return (NULL); 123 } 124 125 struct pf_altq * 126 qname_to_pfaltq(const char *qname, const char *ifname) 127 { 128 struct pf_altq *altq; 129 130 TAILQ_FOREACH(altq, &altqs, entries) { 131 if (strncmp(ifname, altq->ifname, IFNAMSIZ) == 0 && 132 strncmp(qname, altq->qname, PF_QNAME_SIZE) == 0) 133 return (altq); 134 } 135 return (NULL); 136 } 137 138 u_int32_t 139 qname_to_qid(const char *qname) 140 { 141 struct pf_altq *altq; 142 143 /* 144 * We guarantee that same named queues on different interfaces 145 * have the same qid, so we do NOT need to limit matching on 146 * one interface! 147 */ 148 149 TAILQ_FOREACH(altq, &altqs, entries) { 150 if (strncmp(qname, altq->qname, PF_QNAME_SIZE) == 0) 151 return (altq->qid); 152 } 153 return (0); 154 } 155 156 void 157 print_altq(const struct pf_altq *a, unsigned level, struct node_queue_bw *bw, 158 struct node_queue_opt *qopts) 159 { 160 if (a->qname[0] != 0) { 161 print_queue(a, level, bw, 0, qopts); 162 return; 163 } 164 165 printf("altq on %s ", a->ifname); 166 167 switch (a->scheduler) { 168 case ALTQT_CBQ: 169 if (!print_cbq_opts(a)) 170 printf("cbq "); 171 break; 172 case ALTQT_PRIQ: 173 if (!print_priq_opts(a)) 174 printf("priq "); 175 break; 176 case ALTQT_HFSC: 177 if (!print_hfsc_opts(a, qopts)) 178 printf("hfsc "); 179 break; 180 } 181 182 if (bw != NULL && bw->bw_percent > 0) { 183 if (bw->bw_percent < 100) 184 printf("bandwidth %u%% ", bw->bw_percent); 185 } else 186 printf("bandwidth %s ", rate2str((double)a->ifbandwidth)); 187 188 if (a->qlimit != DEFAULT_QLIMIT) 189 printf("qlimit %u ", a->qlimit); 190 printf("tbrsize %u ", a->tbrsize); 191 } 192 193 void 194 print_queue(const struct pf_altq *a, unsigned level, struct node_queue_bw *bw, 195 int print_interface, struct node_queue_opt *qopts) 196 { 197 unsigned i; 198 199 printf("queue "); 200 for (i = 0; i < level; ++i) 201 printf(" "); 202 printf("%s ", a->qname); 203 if (print_interface) 204 printf("on %s ", a->ifname); 205 if (a->scheduler == ALTQT_CBQ || a->scheduler == ALTQT_HFSC) { 206 if (bw != NULL && bw->bw_percent > 0) { 207 if (bw->bw_percent < 100) 208 printf("bandwidth %u%% ", bw->bw_percent); 209 } else 210 printf("bandwidth %s ", rate2str((double)a->bandwidth)); 211 } 212 if (a->priority != DEFAULT_PRIORITY) 213 printf("priority %u ", a->priority); 214 if (a->qlimit != DEFAULT_QLIMIT) 215 printf("qlimit %u ", a->qlimit); 216 switch (a->scheduler) { 217 case ALTQT_CBQ: 218 print_cbq_opts(a); 219 break; 220 case ALTQT_PRIQ: 221 print_priq_opts(a); 222 break; 223 case ALTQT_HFSC: 224 print_hfsc_opts(a, qopts); 225 break; 226 } 227 } 228 229 /* 230 * eval_pfaltq computes the discipline parameters. 231 */ 232 int 233 eval_pfaltq(struct pfctl *pf, struct pf_altq *pa, struct node_queue_bw *bw, 234 struct node_queue_opt *opts) 235 { 236 u_int rate, size, errors = 0; 237 238 if (bw->bw_absolute > 0) 239 pa->ifbandwidth = bw->bw_absolute; 240 else 241 if ((rate = getifspeed(pa->ifname)) == 0) { 242 fprintf(stderr, "cannot determine interface bandwidth " 243 "for %s, specify an absolute bandwidth\n", 244 pa->ifname); 245 errors++; 246 } else if ((pa->ifbandwidth = eval_bwspec(bw, rate)) == 0) 247 pa->ifbandwidth = rate; 248 249 errors += eval_queue_opts(pa, opts, pa->ifbandwidth); 250 251 /* if tbrsize is not specified, use heuristics */ 252 if (pa->tbrsize == 0) { 253 rate = pa->ifbandwidth; 254 if (rate <= 1 * 1000 * 1000) 255 size = 1; 256 else if (rate <= 10 * 1000 * 1000) 257 size = 4; 258 else if (rate <= 200 * 1000 * 1000) 259 size = 8; 260 else 261 size = 24; 262 size = size * getifmtu(pa->ifname); 263 if (size > 0xffff) 264 size = 0xffff; 265 pa->tbrsize = size; 266 } 267 return (errors); 268 } 269 270 /* 271 * check_commit_altq does consistency check for each interface 272 */ 273 int 274 check_commit_altq(int dev, int opts) 275 { 276 struct pf_altq *altq; 277 int error = 0; 278 279 /* call the discipline check for each interface. */ 280 TAILQ_FOREACH(altq, &altqs, entries) { 281 if (altq->qname[0] == 0) { 282 switch (altq->scheduler) { 283 case ALTQT_CBQ: 284 error = check_commit_cbq(dev, opts, altq); 285 break; 286 case ALTQT_PRIQ: 287 error = check_commit_priq(dev, opts, altq); 288 break; 289 case ALTQT_HFSC: 290 error = check_commit_hfsc(dev, opts, altq); 291 break; 292 default: 293 break; 294 } 295 } 296 } 297 return (error); 298 } 299 300 /* 301 * eval_pfqueue computes the queue parameters. 302 */ 303 int 304 eval_pfqueue(struct pfctl *pf, struct pf_altq *pa, struct node_queue_bw *bw, 305 struct node_queue_opt *opts) 306 { 307 /* should be merged with expand_queue */ 308 struct pf_altq *if_pa, *parent; 309 int error = 0; 310 311 /* find the corresponding interface and copy fields used by queues */ 312 if ((if_pa = pfaltq_lookup(pa->ifname)) == NULL) { 313 fprintf(stderr, "altq not defined on %s\n", pa->ifname); 314 return (1); 315 } 316 pa->scheduler = if_pa->scheduler; 317 pa->ifbandwidth = if_pa->ifbandwidth; 318 319 if (qname_to_pfaltq(pa->qname, pa->ifname) != NULL) { 320 fprintf(stderr, "queue %s already exists on interface %s\n", 321 pa->qname, pa->ifname); 322 return (1); 323 } 324 pa->qid = qname_to_qid(pa->qname); 325 326 parent = NULL; 327 if (pa->parent[0] != 0) { 328 parent = qname_to_pfaltq(pa->parent, pa->ifname); 329 if (parent == NULL) { 330 fprintf(stderr, "parent %s not found for %s\n", 331 pa->parent, pa->qname); 332 return (1); 333 } 334 pa->parent_qid = parent->qid; 335 } 336 if (pa->qlimit == 0) 337 pa->qlimit = DEFAULT_QLIMIT; 338 339 if (pa->scheduler == ALTQT_CBQ || pa->scheduler == ALTQT_HFSC) { 340 if ((pa->bandwidth = eval_bwspec(bw, 341 parent == NULL ? 0 : parent->bandwidth)) == 0) { 342 fprintf(stderr, "bandwidth for %s invalid (%d / %d)\n", 343 pa->qname, bw->bw_absolute, bw->bw_percent); 344 return (1); 345 } 346 347 if (pa->bandwidth > pa->ifbandwidth) { 348 fprintf(stderr, "bandwidth for %s higher than " 349 "interface\n", pa->qname); 350 return (1); 351 } 352 if (parent != NULL && pa->bandwidth > parent->bandwidth) { 353 fprintf(stderr, "bandwidth for %s higher than parent\n", 354 pa->qname); 355 return (1); 356 } 357 } 358 359 if (eval_queue_opts(pa, opts, parent == NULL? 0 : parent->bandwidth)) 360 return (1); 361 362 switch (pa->scheduler) { 363 case ALTQT_CBQ: 364 error = eval_pfqueue_cbq(pf, pa); 365 break; 366 case ALTQT_PRIQ: 367 error = eval_pfqueue_priq(pf, pa); 368 break; 369 case ALTQT_HFSC: 370 error = eval_pfqueue_hfsc(pf, pa); 371 break; 372 default: 373 break; 374 } 375 return (error); 376 } 377 378 /* 379 * CBQ support functions 380 */ 381 #define RM_FILTER_GAIN 5 /* log2 of gain, e.g., 5 => 31/32 */ 382 #define RM_NS_PER_SEC (1000000000) 383 384 static int 385 eval_pfqueue_cbq(struct pfctl *pf, struct pf_altq *pa) 386 { 387 struct cbq_opts *opts; 388 u_int ifmtu; 389 390 if (pa->priority >= CBQ_MAXPRI) { 391 warnx("priority out of range: max %d", CBQ_MAXPRI - 1); 392 return (-1); 393 } 394 395 ifmtu = getifmtu(pa->ifname); 396 opts = &pa->pq_u.cbq_opts; 397 398 if (opts->pktsize == 0) { /* use default */ 399 opts->pktsize = ifmtu; 400 if (opts->pktsize > MCLBYTES) /* do what TCP does */ 401 opts->pktsize &= ~MCLBYTES; 402 } else if (opts->pktsize > ifmtu) 403 opts->pktsize = ifmtu; 404 if (opts->maxpktsize == 0) /* use default */ 405 opts->maxpktsize = ifmtu; 406 else if (opts->maxpktsize > ifmtu) 407 opts->pktsize = ifmtu; 408 409 if (opts->pktsize > opts->maxpktsize) 410 opts->pktsize = opts->maxpktsize; 411 412 if (pa->parent[0] == 0) 413 opts->flags |= (CBQCLF_ROOTCLASS | CBQCLF_WRR); 414 415 cbq_compute_idletime(pf, pa); 416 return (0); 417 } 418 419 /* 420 * compute ns_per_byte, maxidle, minidle, and offtime 421 */ 422 static int 423 cbq_compute_idletime(struct pfctl *pf, struct pf_altq *pa) 424 { 425 struct cbq_opts *opts; 426 double maxidle_s, maxidle, minidle; 427 double offtime, nsPerByte, ifnsPerByte, ptime, cptime; 428 double z, g, f, gton, gtom; 429 u_int minburst, maxburst; 430 431 opts = &pa->pq_u.cbq_opts; 432 ifnsPerByte = (1.0 / (double)pa->ifbandwidth) * RM_NS_PER_SEC * 8; 433 minburst = opts->minburst; 434 maxburst = opts->maxburst; 435 436 if (pa->bandwidth == 0) 437 f = 0.0001; /* small enough? */ 438 else 439 f = ((double) pa->bandwidth / (double) pa->ifbandwidth); 440 441 nsPerByte = ifnsPerByte / f; 442 ptime = (double)opts->pktsize * ifnsPerByte; 443 cptime = ptime * (1.0 - f) / f; 444 445 if (nsPerByte * (double)opts->maxpktsize > (double)INT_MAX) { 446 /* 447 * this causes integer overflow in kernel! 448 * (bandwidth < 6Kbps when max_pkt_size=1500) 449 */ 450 if (pa->bandwidth != 0 && (pf->opts & PF_OPT_QUIET) == 0) 451 warnx("queue bandwidth must be larger than %s", 452 rate2str(ifnsPerByte * (double)opts->maxpktsize / 453 (double)INT_MAX * (double)pa->ifbandwidth)); 454 fprintf(stderr, "cbq: queue %s is too slow!\n", 455 pa->qname); 456 nsPerByte = (double)(INT_MAX / opts->maxpktsize); 457 } 458 459 if (maxburst == 0) { /* use default */ 460 if (cptime > 10.0 * 1000000) 461 maxburst = 4; 462 else 463 maxburst = 16; 464 } 465 if (minburst == 0) /* use default */ 466 minburst = 2; 467 if (minburst > maxburst) 468 minburst = maxburst; 469 470 z = (double)(1 << RM_FILTER_GAIN); 471 g = (1.0 - 1.0 / z); 472 gton = pow(g, (double)maxburst); 473 gtom = pow(g, (double)(minburst-1)); 474 maxidle = ((1.0 / f - 1.0) * ((1.0 - gton) / gton)); 475 maxidle_s = (1.0 - g); 476 if (maxidle > maxidle_s) 477 maxidle = ptime * maxidle; 478 else 479 maxidle = ptime * maxidle_s; 480 if (minburst) 481 offtime = cptime * (1.0 + 1.0/(1.0 - g) * (1.0 - gtom) / gtom); 482 else 483 offtime = cptime; 484 minidle = -((double)opts->maxpktsize * (double)nsPerByte); 485 486 /* scale parameters */ 487 maxidle = ((maxidle * 8.0) / nsPerByte) * 488 pow(2.0, (double)RM_FILTER_GAIN); 489 offtime = (offtime * 8.0) / nsPerByte * 490 pow(2.0, (double)RM_FILTER_GAIN); 491 minidle = ((minidle * 8.0) / nsPerByte) * 492 pow(2.0, (double)RM_FILTER_GAIN); 493 494 maxidle = maxidle / 1000.0; 495 offtime = offtime / 1000.0; 496 minidle = minidle / 1000.0; 497 498 opts->minburst = minburst; 499 opts->maxburst = maxburst; 500 opts->ns_per_byte = (u_int)nsPerByte; 501 opts->maxidle = (u_int)fabs(maxidle); 502 opts->minidle = (int)minidle; 503 opts->offtime = (u_int)fabs(offtime); 504 505 return (0); 506 } 507 508 static int 509 check_commit_cbq(int dev, int opts, struct pf_altq *pa) 510 { 511 struct pf_altq *altq; 512 int root_class, default_class; 513 int error = 0; 514 515 /* 516 * check if cbq has one root queue and one default queue 517 * for this interface 518 */ 519 root_class = default_class = 0; 520 TAILQ_FOREACH(altq, &altqs, entries) { 521 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0) 522 continue; 523 if (altq->qname[0] == 0) /* this is for interface */ 524 continue; 525 if (altq->pq_u.cbq_opts.flags & CBQCLF_ROOTCLASS) 526 root_class++; 527 if (altq->pq_u.cbq_opts.flags & CBQCLF_DEFCLASS) 528 default_class++; 529 } 530 if (root_class != 1) { 531 warnx("should have one root queue on %s", pa->ifname); 532 error++; 533 } 534 if (default_class != 1) { 535 warnx("should have one default queue on %s", pa->ifname); 536 error++; 537 } 538 return (error); 539 } 540 541 static int 542 print_cbq_opts(const struct pf_altq *a) 543 { 544 const struct cbq_opts *opts; 545 546 opts = &a->pq_u.cbq_opts; 547 if (opts->flags) { 548 printf("cbq("); 549 if (opts->flags & CBQCLF_RED) 550 printf(" red"); 551 if (opts->flags & CBQCLF_ECN) 552 printf(" ecn"); 553 if (opts->flags & CBQCLF_RIO) 554 printf(" rio"); 555 if (opts->flags & CBQCLF_CLEARDSCP) 556 printf(" cleardscp"); 557 if (opts->flags & CBQCLF_FLOWVALVE) 558 printf(" flowvalve"); 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, 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, int opts, 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, 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, int opts, 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 *rtsc, *lssc, *ulsc; 825 826 opts = &a->pq_u.hfsc_opts; 827 if (qopts == NULL) 828 rtsc = lssc = ulsc = NULL; 829 else { 830 rtsc = &qopts->data.hfsc_opts.realtime; 831 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, 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, 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(char *ifname) 1072 { 1073 int s; 1074 struct ifreq ifr; 1075 struct if_data ifrdat; 1076 1077 if ((s = socket(AF_INET, SOCK_DGRAM, 0)) < 0) 1078 err(1, "socket"); 1079 if (strlcpy(ifr.ifr_name, ifname, sizeof(ifr.ifr_name)) >= 1080 sizeof(ifr.ifr_name)) 1081 errx(1, "getifspeed: strlcpy"); 1082 ifr.ifr_data = (caddr_t)&ifrdat; 1083 if (ioctl(s, SIOCGIFDATA, (caddr_t)&ifr) == -1) 1084 err(1, "SIOCGIFDATA"); 1085 if (shutdown(s, SHUT_RDWR) == -1) 1086 err(1, "shutdown"); 1087 if (close(s)) 1088 err(1, "close"); 1089 return ((u_int32_t)ifrdat.ifi_baudrate); 1090 } 1091 1092 u_long 1093 getifmtu(char *ifname) 1094 { 1095 int s; 1096 struct ifreq ifr; 1097 1098 if ((s = socket(AF_INET, SOCK_DGRAM, 0)) < 0) 1099 err(1, "socket"); 1100 if (strlcpy(ifr.ifr_name, ifname, sizeof(ifr.ifr_name)) >= 1101 sizeof(ifr.ifr_name)) 1102 errx(1, "getifmtu: strlcpy"); 1103 if (ioctl(s, SIOCGIFMTU, (caddr_t)&ifr) == -1) 1104 err(1, "SIOCGIFMTU"); 1105 if (shutdown(s, SHUT_RDWR) == -1) 1106 err(1, "shutdown"); 1107 if (close(s)) 1108 err(1, "close"); 1109 if (ifr.ifr_mtu > 0) 1110 return (ifr.ifr_mtu); 1111 else { 1112 warnx("could not get mtu for %s, assuming 1500", ifname); 1113 return (1500); 1114 } 1115 } 1116 1117 int 1118 eval_queue_opts(struct pf_altq *pa, struct node_queue_opt *opts, 1119 u_int32_t ref_bw) 1120 { 1121 int errors = 0; 1122 1123 switch (pa->scheduler) { 1124 case ALTQT_CBQ: 1125 pa->pq_u.cbq_opts = opts->data.cbq_opts; 1126 break; 1127 case ALTQT_PRIQ: 1128 pa->pq_u.priq_opts = opts->data.priq_opts; 1129 break; 1130 case ALTQT_HFSC: 1131 pa->pq_u.hfsc_opts.flags = opts->data.hfsc_opts.flags; 1132 if (opts->data.hfsc_opts.linkshare.used) { 1133 pa->pq_u.hfsc_opts.lssc_m1 = 1134 eval_bwspec(&opts->data.hfsc_opts.linkshare.m1, 1135 ref_bw); 1136 pa->pq_u.hfsc_opts.lssc_m2 = 1137 eval_bwspec(&opts->data.hfsc_opts.linkshare.m2, 1138 ref_bw); 1139 pa->pq_u.hfsc_opts.lssc_d = 1140 opts->data.hfsc_opts.linkshare.d; 1141 } 1142 if (opts->data.hfsc_opts.realtime.used) { 1143 pa->pq_u.hfsc_opts.rtsc_m1 = 1144 eval_bwspec(&opts->data.hfsc_opts.realtime.m1, 1145 ref_bw); 1146 pa->pq_u.hfsc_opts.rtsc_m2 = 1147 eval_bwspec(&opts->data.hfsc_opts.realtime.m2, 1148 ref_bw); 1149 pa->pq_u.hfsc_opts.rtsc_d = 1150 opts->data.hfsc_opts.realtime.d; 1151 } 1152 if (opts->data.hfsc_opts.upperlimit.used) { 1153 pa->pq_u.hfsc_opts.ulsc_m1 = 1154 eval_bwspec(&opts->data.hfsc_opts.upperlimit.m1, 1155 ref_bw); 1156 pa->pq_u.hfsc_opts.ulsc_m2 = 1157 eval_bwspec(&opts->data.hfsc_opts.upperlimit.m2, 1158 ref_bw); 1159 pa->pq_u.hfsc_opts.ulsc_d = 1160 opts->data.hfsc_opts.upperlimit.d; 1161 } 1162 break; 1163 default: 1164 warnx("eval_queue_opts: unknown scheduler type %u", 1165 opts->qtype); 1166 errors++; 1167 break; 1168 } 1169 1170 return (errors); 1171 } 1172 1173 u_int32_t 1174 eval_bwspec(struct node_queue_bw *bw, u_int32_t ref_bw) 1175 { 1176 if (bw->bw_absolute > 0) 1177 return (bw->bw_absolute); 1178 1179 if (bw->bw_percent > 0) 1180 return (ref_bw / 100 * bw->bw_percent); 1181 1182 return (0); 1183 } 1184 1185 void 1186 print_hfsc_sc(const char *scname, u_int m1, u_int d, u_int m2, 1187 const struct node_hfsc_sc *sc) 1188 { 1189 printf(" %s", scname); 1190 1191 if (d != 0) { 1192 printf("("); 1193 if (sc != NULL && sc->m1.bw_percent > 0) 1194 printf("%u%%", sc->m1.bw_percent); 1195 else 1196 printf("%s", rate2str((double)m1)); 1197 printf(" %u", d); 1198 } 1199 1200 if (sc != NULL && sc->m2.bw_percent > 0) 1201 printf(" %u%%", sc->m2.bw_percent); 1202 else 1203 printf(" %s", rate2str((double)m2)); 1204 1205 if (d != 0) 1206 printf(")"); 1207 } 1208