altq_subr.c revision 1.14.10.1
1 /* $NetBSD: altq_subr.c,v 1.14.10.1 2006/10/22 06:04:30 yamt Exp $ */ 2 /* $KAME: altq_subr.c,v 1.24 2005/04/13 03:44:25 suz Exp $ */ 3 4 /* 5 * Copyright (C) 1997-2003 6 * Sony Computer Science Laboratories Inc. All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY SONY CSL AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL SONY CSL OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30 #include <sys/cdefs.h> 31 __KERNEL_RCSID(0, "$NetBSD: altq_subr.c,v 1.14.10.1 2006/10/22 06:04:30 yamt Exp $"); 32 33 #ifdef _KERNEL_OPT 34 #include "opt_altq.h" 35 #include "opt_inet.h" 36 #include "pf.h" 37 #endif 38 39 #include <sys/param.h> 40 #include <sys/malloc.h> 41 #include <sys/mbuf.h> 42 #include <sys/systm.h> 43 #include <sys/proc.h> 44 #include <sys/socket.h> 45 #include <sys/socketvar.h> 46 #include <sys/kernel.h> 47 #include <sys/errno.h> 48 #include <sys/syslog.h> 49 #include <sys/sysctl.h> 50 #include <sys/queue.h> 51 52 #include <net/if.h> 53 #include <net/if_dl.h> 54 #include <net/if_types.h> 55 56 #include <netinet/in.h> 57 #include <netinet/in_systm.h> 58 #include <netinet/ip.h> 59 #ifdef INET6 60 #include <netinet/ip6.h> 61 #endif 62 #include <netinet/tcp.h> 63 #include <netinet/udp.h> 64 65 #if NPF > 0 66 #include <net/pfvar.h> 67 #endif 68 #include <altq/altq.h> 69 #ifdef ALTQ3_COMPAT 70 #include <altq/altq_conf.h> 71 #endif 72 73 /* machine dependent clock related includes */ 74 #ifdef __FreeBSD__ 75 #include "opt_cpu.h" /* for FreeBSD-2.2.8 to get i586_ctr_freq */ 76 #include <machine/clock.h> 77 #endif 78 #if defined(__i386__) 79 #include <machine/cpufunc.h> /* for pentium tsc */ 80 #include <machine/specialreg.h> /* for CPUID_TSC */ 81 #ifdef __FreeBSD__ 82 #include <machine/md_var.h> /* for cpu_feature */ 83 #elif defined(__NetBSD__) || defined(__OpenBSD__) 84 #include <machine/cpu.h> /* for cpu_feature */ 85 #endif 86 #endif /* __i386__ */ 87 88 /* 89 * internal function prototypes 90 */ 91 static void tbr_timeout(void *); 92 int (*altq_input)(struct mbuf *, int) = NULL; 93 static int tbr_timer = 0; /* token bucket regulator timer */ 94 static struct callout tbr_callout = CALLOUT_INITIALIZER; 95 96 #ifdef ALTQ3_CLFIER_COMPAT 97 static int extract_ports4(struct mbuf *, struct ip *, struct flowinfo_in *); 98 #ifdef INET6 99 static int extract_ports6(struct mbuf *, struct ip6_hdr *, 100 struct flowinfo_in6 *); 101 #endif 102 static int apply_filter4(u_int32_t, struct flow_filter *, 103 struct flowinfo_in *); 104 static int apply_ppfilter4(u_int32_t, struct flow_filter *, 105 struct flowinfo_in *); 106 #ifdef INET6 107 static int apply_filter6(u_int32_t, struct flow_filter6 *, 108 struct flowinfo_in6 *); 109 #endif 110 static int apply_tosfilter4(u_int32_t, struct flow_filter *, 111 struct flowinfo_in *); 112 static u_long get_filt_handle(struct acc_classifier *, int); 113 static struct acc_filter *filth_to_filtp(struct acc_classifier *, u_long); 114 static u_int32_t filt2fibmask(struct flow_filter *); 115 116 static void ip4f_cache(struct ip *, struct flowinfo_in *); 117 static int ip4f_lookup(struct ip *, struct flowinfo_in *); 118 static int ip4f_init(void); 119 static struct ip4_frag *ip4f_alloc(void); 120 static void ip4f_free(struct ip4_frag *); 121 #endif /* ALTQ3_CLFIER_COMPAT */ 122 123 /* 124 * alternate queueing support routines 125 */ 126 127 /* look up the queue state by the interface name and the queueing type. */ 128 void * 129 altq_lookup(char *name, int type) 130 { 131 struct ifnet *ifp; 132 133 if ((ifp = ifunit(name)) != NULL) { 134 if (type != ALTQT_NONE && ifp->if_snd.altq_type == type) 135 return (ifp->if_snd.altq_disc); 136 } 137 138 return NULL; 139 } 140 141 int 142 altq_attach(struct ifaltq *ifq, int type, void *discipline, 143 int (*enqueue)(struct ifaltq *, struct mbuf *, struct altq_pktattr *), 144 struct mbuf *(*dequeue)(struct ifaltq *, int), 145 int (*request)(struct ifaltq *, int, void *), 146 void *clfier, void *(*classify)(void *, struct mbuf *, int)) 147 { 148 if (!ALTQ_IS_READY(ifq)) 149 return ENXIO; 150 151 #ifdef ALTQ3_COMPAT 152 /* 153 * pfaltq can override the existing discipline, but altq3 cannot. 154 * check these if clfier is not NULL (which implies altq3). 155 */ 156 if (clfier != NULL) { 157 if (ALTQ_IS_ENABLED(ifq)) 158 return EBUSY; 159 if (ALTQ_IS_ATTACHED(ifq)) 160 return EEXIST; 161 } 162 #endif 163 ifq->altq_type = type; 164 ifq->altq_disc = discipline; 165 ifq->altq_enqueue = enqueue; 166 ifq->altq_dequeue = dequeue; 167 ifq->altq_request = request; 168 ifq->altq_clfier = clfier; 169 ifq->altq_classify = classify; 170 ifq->altq_flags &= (ALTQF_CANTCHANGE|ALTQF_ENABLED); 171 #ifdef ALTQ3_COMPAT 172 #ifdef ALTQ_KLD 173 altq_module_incref(type); 174 #endif 175 #endif 176 return 0; 177 } 178 179 int 180 altq_detach(struct ifaltq *ifq) 181 { 182 if (!ALTQ_IS_READY(ifq)) 183 return ENXIO; 184 if (ALTQ_IS_ENABLED(ifq)) 185 return EBUSY; 186 if (!ALTQ_IS_ATTACHED(ifq)) 187 return (0); 188 #ifdef ALTQ3_COMPAT 189 #ifdef ALTQ_KLD 190 altq_module_declref(ifq->altq_type); 191 #endif 192 #endif 193 194 ifq->altq_type = ALTQT_NONE; 195 ifq->altq_disc = NULL; 196 ifq->altq_enqueue = NULL; 197 ifq->altq_dequeue = NULL; 198 ifq->altq_request = NULL; 199 ifq->altq_clfier = NULL; 200 ifq->altq_classify = NULL; 201 ifq->altq_flags &= ALTQF_CANTCHANGE; 202 return 0; 203 } 204 205 int 206 altq_enable(struct ifaltq *ifq) 207 { 208 int s; 209 210 if (!ALTQ_IS_READY(ifq)) 211 return ENXIO; 212 if (ALTQ_IS_ENABLED(ifq)) 213 return 0; 214 215 s = splnet(); 216 IFQ_PURGE(ifq); 217 ASSERT(ifq->ifq_len == 0); 218 ifq->altq_flags |= ALTQF_ENABLED; 219 if (ifq->altq_clfier != NULL) 220 ifq->altq_flags |= ALTQF_CLASSIFY; 221 splx(s); 222 223 return 0; 224 } 225 226 int 227 altq_disable(struct ifaltq *ifq) 228 { 229 int s; 230 231 if (!ALTQ_IS_ENABLED(ifq)) 232 return 0; 233 234 s = splnet(); 235 IFQ_PURGE(ifq); 236 ASSERT(ifq->ifq_len == 0); 237 ifq->altq_flags &= ~(ALTQF_ENABLED|ALTQF_CLASSIFY); 238 splx(s); 239 return 0; 240 } 241 242 #ifdef ALTQ_DEBUG 243 void 244 altq_assert(const char *file, int line, const char *failedexpr) 245 { 246 (void)printf("altq assertion \"%s\" failed: file \"%s\", line %d\n", 247 failedexpr, file, line); 248 panic("altq assertion"); 249 /* NOTREACHED */ 250 } 251 #endif 252 253 /* 254 * internal representation of token bucket parameters 255 * rate: byte_per_unittime << 32 256 * (((bits_per_sec) / 8) << 32) / machclk_freq 257 * depth: byte << 32 258 * 259 */ 260 #define TBR_SHIFT 32 261 #define TBR_SCALE(x) ((int64_t)(x) << TBR_SHIFT) 262 #define TBR_UNSCALE(x) ((x) >> TBR_SHIFT) 263 264 struct mbuf * 265 tbr_dequeue(struct ifaltq *ifq, int op) 266 { 267 struct tb_regulator *tbr; 268 struct mbuf *m; 269 int64_t interval; 270 u_int64_t now; 271 272 tbr = ifq->altq_tbr; 273 if (op == ALTDQ_REMOVE && tbr->tbr_lastop == ALTDQ_POLL) { 274 /* if this is a remove after poll, bypass tbr check */ 275 } else { 276 /* update token only when it is negative */ 277 if (tbr->tbr_token <= 0) { 278 now = read_machclk(); 279 interval = now - tbr->tbr_last; 280 if (interval >= tbr->tbr_filluptime) 281 tbr->tbr_token = tbr->tbr_depth; 282 else { 283 tbr->tbr_token += interval * tbr->tbr_rate; 284 if (tbr->tbr_token > tbr->tbr_depth) 285 tbr->tbr_token = tbr->tbr_depth; 286 } 287 tbr->tbr_last = now; 288 } 289 /* if token is still negative, don't allow dequeue */ 290 if (tbr->tbr_token <= 0) 291 return (NULL); 292 } 293 294 if (ALTQ_IS_ENABLED(ifq)) 295 m = (*ifq->altq_dequeue)(ifq, op); 296 else { 297 if (op == ALTDQ_POLL) 298 IF_POLL(ifq, m); 299 else 300 IF_DEQUEUE(ifq, m); 301 } 302 303 if (m != NULL && op == ALTDQ_REMOVE) 304 tbr->tbr_token -= TBR_SCALE(m_pktlen(m)); 305 tbr->tbr_lastop = op; 306 return (m); 307 } 308 309 /* 310 * set a token bucket regulator. 311 * if the specified rate is zero, the token bucket regulator is deleted. 312 */ 313 int 314 tbr_set(struct ifaltq *ifq, struct tb_profile *profile) 315 { 316 struct tb_regulator *tbr, *otbr; 317 318 if (machclk_freq == 0) 319 init_machclk(); 320 if (machclk_freq == 0) { 321 printf("tbr_set: no CPU clock available!\n"); 322 return (ENXIO); 323 } 324 325 if (profile->rate == 0) { 326 /* delete this tbr */ 327 if ((tbr = ifq->altq_tbr) == NULL) 328 return (ENOENT); 329 ifq->altq_tbr = NULL; 330 free(tbr, M_DEVBUF); 331 return (0); 332 } 333 334 tbr = malloc(sizeof(struct tb_regulator), M_DEVBUF, M_WAITOK|M_ZERO); 335 if (tbr == NULL) 336 return (ENOMEM); 337 338 tbr->tbr_rate = TBR_SCALE(profile->rate / 8) / machclk_freq; 339 tbr->tbr_depth = TBR_SCALE(profile->depth); 340 if (tbr->tbr_rate > 0) 341 tbr->tbr_filluptime = tbr->tbr_depth / tbr->tbr_rate; 342 else 343 tbr->tbr_filluptime = 0xffffffffffffffffLL; 344 tbr->tbr_token = tbr->tbr_depth; 345 tbr->tbr_last = read_machclk(); 346 tbr->tbr_lastop = ALTDQ_REMOVE; 347 348 otbr = ifq->altq_tbr; 349 ifq->altq_tbr = tbr; /* set the new tbr */ 350 351 if (otbr != NULL) 352 free(otbr, M_DEVBUF); 353 else { 354 if (tbr_timer == 0) { 355 CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0); 356 tbr_timer = 1; 357 } 358 } 359 return (0); 360 } 361 362 /* 363 * tbr_timeout goes through the interface list, and kicks the drivers 364 * if necessary. 365 */ 366 static void 367 tbr_timeout(void *arg __unused) 368 { 369 struct ifnet *ifp; 370 int active, s; 371 372 active = 0; 373 s = splnet(); 374 for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) { 375 if (!TBR_IS_ENABLED(&ifp->if_snd)) 376 continue; 377 active++; 378 if (!IFQ_IS_EMPTY(&ifp->if_snd) && ifp->if_start != NULL) 379 (*ifp->if_start)(ifp); 380 } 381 splx(s); 382 if (active > 0) 383 CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0); 384 else 385 tbr_timer = 0; /* don't need tbr_timer anymore */ 386 #if defined(__alpha__) && !defined(ALTQ_NOPCC) 387 { 388 /* 389 * XXX read out the machine dependent clock once a second 390 * to detect counter wrap-around. 391 */ 392 static u_int cnt; 393 394 if (++cnt >= hz) { 395 (void)read_machclk(); 396 cnt = 0; 397 } 398 } 399 #endif /* __alpha__ && !ALTQ_NOPCC */ 400 } 401 402 /* 403 * get token bucket regulator profile 404 */ 405 int 406 tbr_get(struct ifaltq *ifq, struct tb_profile *profile) 407 { 408 struct tb_regulator *tbr; 409 410 if ((tbr = ifq->altq_tbr) == NULL) { 411 profile->rate = 0; 412 profile->depth = 0; 413 } else { 414 profile->rate = 415 (u_int)TBR_UNSCALE(tbr->tbr_rate * 8 * machclk_freq); 416 profile->depth = (u_int)TBR_UNSCALE(tbr->tbr_depth); 417 } 418 return (0); 419 } 420 421 #if NPF > 0 422 /* 423 * attach a discipline to the interface. if one already exists, it is 424 * overridden. 425 */ 426 int 427 altq_pfattach(struct pf_altq *a) 428 { 429 int error = 0; 430 431 switch (a->scheduler) { 432 case ALTQT_NONE: 433 break; 434 #ifdef ALTQ_CBQ 435 case ALTQT_CBQ: 436 error = cbq_pfattach(a); 437 break; 438 #endif 439 #ifdef ALTQ_PRIQ 440 case ALTQT_PRIQ: 441 error = priq_pfattach(a); 442 break; 443 #endif 444 #ifdef ALTQ_HFSC 445 case ALTQT_HFSC: 446 error = hfsc_pfattach(a); 447 break; 448 #endif 449 default: 450 error = ENXIO; 451 } 452 453 return (error); 454 } 455 456 /* 457 * detach a discipline from the interface. 458 * it is possible that the discipline was already overridden by another 459 * discipline. 460 */ 461 int 462 altq_pfdetach(struct pf_altq *a) 463 { 464 struct ifnet *ifp; 465 int s, error = 0; 466 467 if ((ifp = ifunit(a->ifname)) == NULL) 468 return (EINVAL); 469 470 /* if this discipline is no longer referenced, just return */ 471 if (a->altq_disc == NULL || a->altq_disc != ifp->if_snd.altq_disc) 472 return (0); 473 474 s = splnet(); 475 if (ALTQ_IS_ENABLED(&ifp->if_snd)) 476 error = altq_disable(&ifp->if_snd); 477 if (error == 0) 478 error = altq_detach(&ifp->if_snd); 479 splx(s); 480 481 return (error); 482 } 483 484 /* 485 * add a discipline or a queue 486 */ 487 int 488 altq_add(struct pf_altq *a) 489 { 490 int error = 0; 491 492 if (a->qname[0] != 0) 493 return (altq_add_queue(a)); 494 495 if (machclk_freq == 0) 496 init_machclk(); 497 if (machclk_freq == 0) 498 panic("altq_add: no CPU clock"); 499 500 switch (a->scheduler) { 501 #ifdef ALTQ_CBQ 502 case ALTQT_CBQ: 503 error = cbq_add_altq(a); 504 break; 505 #endif 506 #ifdef ALTQ_PRIQ 507 case ALTQT_PRIQ: 508 error = priq_add_altq(a); 509 break; 510 #endif 511 #ifdef ALTQ_HFSC 512 case ALTQT_HFSC: 513 error = hfsc_add_altq(a); 514 break; 515 #endif 516 default: 517 error = ENXIO; 518 } 519 520 return (error); 521 } 522 523 /* 524 * remove a discipline or a queue 525 */ 526 int 527 altq_remove(struct pf_altq *a) 528 { 529 int error = 0; 530 531 if (a->qname[0] != 0) 532 return (altq_remove_queue(a)); 533 534 switch (a->scheduler) { 535 #ifdef ALTQ_CBQ 536 case ALTQT_CBQ: 537 error = cbq_remove_altq(a); 538 break; 539 #endif 540 #ifdef ALTQ_PRIQ 541 case ALTQT_PRIQ: 542 error = priq_remove_altq(a); 543 break; 544 #endif 545 #ifdef ALTQ_HFSC 546 case ALTQT_HFSC: 547 error = hfsc_remove_altq(a); 548 break; 549 #endif 550 default: 551 error = ENXIO; 552 } 553 554 return (error); 555 } 556 557 /* 558 * add a queue to the discipline 559 */ 560 int 561 altq_add_queue(struct pf_altq *a) 562 { 563 int error = 0; 564 565 switch (a->scheduler) { 566 #ifdef ALTQ_CBQ 567 case ALTQT_CBQ: 568 error = cbq_add_queue(a); 569 break; 570 #endif 571 #ifdef ALTQ_PRIQ 572 case ALTQT_PRIQ: 573 error = priq_add_queue(a); 574 break; 575 #endif 576 #ifdef ALTQ_HFSC 577 case ALTQT_HFSC: 578 error = hfsc_add_queue(a); 579 break; 580 #endif 581 default: 582 error = ENXIO; 583 } 584 585 return (error); 586 } 587 588 /* 589 * remove a queue from the discipline 590 */ 591 int 592 altq_remove_queue(struct pf_altq *a) 593 { 594 int error = 0; 595 596 switch (a->scheduler) { 597 #ifdef ALTQ_CBQ 598 case ALTQT_CBQ: 599 error = cbq_remove_queue(a); 600 break; 601 #endif 602 #ifdef ALTQ_PRIQ 603 case ALTQT_PRIQ: 604 error = priq_remove_queue(a); 605 break; 606 #endif 607 #ifdef ALTQ_HFSC 608 case ALTQT_HFSC: 609 error = hfsc_remove_queue(a); 610 break; 611 #endif 612 default: 613 error = ENXIO; 614 } 615 616 return (error); 617 } 618 619 /* 620 * get queue statistics 621 */ 622 int 623 altq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes) 624 { 625 int error = 0; 626 627 switch (a->scheduler) { 628 #ifdef ALTQ_CBQ 629 case ALTQT_CBQ: 630 error = cbq_getqstats(a, ubuf, nbytes); 631 break; 632 #endif 633 #ifdef ALTQ_PRIQ 634 case ALTQT_PRIQ: 635 error = priq_getqstats(a, ubuf, nbytes); 636 break; 637 #endif 638 #ifdef ALTQ_HFSC 639 case ALTQT_HFSC: 640 error = hfsc_getqstats(a, ubuf, nbytes); 641 break; 642 #endif 643 default: 644 error = ENXIO; 645 } 646 647 return (error); 648 } 649 #endif /* NPF > 0 */ 650 651 /* 652 * read and write diffserv field in IPv4 or IPv6 header 653 */ 654 u_int8_t 655 read_dsfield(struct mbuf *m, struct altq_pktattr *pktattr) 656 { 657 struct mbuf *m0; 658 u_int8_t ds_field = 0; 659 660 if (pktattr == NULL || 661 (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6)) 662 return ((u_int8_t)0); 663 664 /* verify that pattr_hdr is within the mbuf data */ 665 for (m0 = m; m0 != NULL; m0 = m0->m_next) 666 if ((pktattr->pattr_hdr >= m0->m_data) && 667 (pktattr->pattr_hdr < m0->m_data + m0->m_len)) 668 break; 669 if (m0 == NULL) { 670 /* ick, pattr_hdr is stale */ 671 pktattr->pattr_af = AF_UNSPEC; 672 #ifdef ALTQ_DEBUG 673 printf("read_dsfield: can't locate header!\n"); 674 #endif 675 return ((u_int8_t)0); 676 } 677 678 if (pktattr->pattr_af == AF_INET) { 679 struct ip *ip = (struct ip *)pktattr->pattr_hdr; 680 681 if (ip->ip_v != 4) 682 return ((u_int8_t)0); /* version mismatch! */ 683 ds_field = ip->ip_tos; 684 } 685 #ifdef INET6 686 else if (pktattr->pattr_af == AF_INET6) { 687 struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr; 688 u_int32_t flowlabel; 689 690 flowlabel = ntohl(ip6->ip6_flow); 691 if ((flowlabel >> 28) != 6) 692 return ((u_int8_t)0); /* version mismatch! */ 693 ds_field = (flowlabel >> 20) & 0xff; 694 } 695 #endif 696 return (ds_field); 697 } 698 699 void 700 write_dsfield(struct mbuf *m, struct altq_pktattr *pktattr, u_int8_t dsfield) 701 { 702 struct mbuf *m0; 703 704 if (pktattr == NULL || 705 (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6)) 706 return; 707 708 /* verify that pattr_hdr is within the mbuf data */ 709 for (m0 = m; m0 != NULL; m0 = m0->m_next) 710 if ((pktattr->pattr_hdr >= m0->m_data) && 711 (pktattr->pattr_hdr < m0->m_data + m0->m_len)) 712 break; 713 if (m0 == NULL) { 714 /* ick, pattr_hdr is stale */ 715 pktattr->pattr_af = AF_UNSPEC; 716 #ifdef ALTQ_DEBUG 717 printf("write_dsfield: can't locate header!\n"); 718 #endif 719 return; 720 } 721 722 if (pktattr->pattr_af == AF_INET) { 723 struct ip *ip = (struct ip *)pktattr->pattr_hdr; 724 u_int8_t old; 725 int32_t sum; 726 727 if (ip->ip_v != 4) 728 return; /* version mismatch! */ 729 old = ip->ip_tos; 730 dsfield |= old & 3; /* leave CU bits */ 731 if (old == dsfield) 732 return; 733 ip->ip_tos = dsfield; 734 /* 735 * update checksum (from RFC1624) 736 * HC' = ~(~HC + ~m + m') 737 */ 738 sum = ~ntohs(ip->ip_sum) & 0xffff; 739 sum += 0xff00 + (~old & 0xff) + dsfield; 740 sum = (sum >> 16) + (sum & 0xffff); 741 sum += (sum >> 16); /* add carry */ 742 743 ip->ip_sum = htons(~sum & 0xffff); 744 } 745 #ifdef INET6 746 else if (pktattr->pattr_af == AF_INET6) { 747 struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr; 748 u_int32_t flowlabel; 749 750 flowlabel = ntohl(ip6->ip6_flow); 751 if ((flowlabel >> 28) != 6) 752 return; /* version mismatch! */ 753 flowlabel = (flowlabel & 0xf03fffff) | (dsfield << 20); 754 ip6->ip6_flow = htonl(flowlabel); 755 } 756 #endif 757 return; 758 } 759 760 761 /* 762 * high resolution clock support taking advantage of a machine dependent 763 * high resolution time counter (e.g., timestamp counter of intel pentium). 764 * we assume 765 * - 64-bit-long monotonically-increasing counter 766 * - frequency range is 100M-4GHz (CPU speed) 767 */ 768 /* if pcc is not available or disabled, emulate 256MHz using microtime() */ 769 #define MACHCLK_SHIFT 8 770 771 int machclk_usepcc; 772 u_int32_t machclk_freq = 0; 773 u_int32_t machclk_per_tick = 0; 774 775 #ifdef __alpha__ 776 #ifdef __FreeBSD__ 777 extern u_int32_t cycles_per_sec; /* alpha cpu clock frequency */ 778 #elif defined(__NetBSD__) || defined(__OpenBSD__) 779 extern u_int64_t cycles_per_usec; /* alpha cpu clock frequency */ 780 #endif 781 #endif /* __alpha__ */ 782 783 void 784 init_machclk(void) 785 { 786 machclk_usepcc = 1; 787 788 #if (!defined(__i386__) && !defined(__alpha__)) || defined(ALTQ_NOPCC) 789 machclk_usepcc = 0; 790 #endif 791 #if defined(__FreeBSD__) && defined(SMP) 792 machclk_usepcc = 0; 793 #endif 794 #if defined(__NetBSD__) && defined(MULTIPROCESSOR) 795 machclk_usepcc = 0; 796 #endif 797 #ifdef __i386__ 798 /* check if TSC is available */ 799 if (machclk_usepcc == 1 && (cpu_feature & CPUID_TSC) == 0) 800 machclk_usepcc = 0; 801 #endif 802 803 if (machclk_usepcc == 0) { 804 /* emulate 256MHz using microtime() */ 805 machclk_freq = 1000000 << MACHCLK_SHIFT; 806 machclk_per_tick = machclk_freq / hz; 807 #ifdef ALTQ_DEBUG 808 printf("altq: emulate %uHz CPU clock\n", machclk_freq); 809 #endif 810 return; 811 } 812 813 /* 814 * if the clock frequency (of Pentium TSC or Alpha PCC) is 815 * accessible, just use it. 816 */ 817 #ifdef __i386__ 818 #ifdef __FreeBSD__ 819 #if (__FreeBSD_version > 300000) 820 machclk_freq = tsc_freq; 821 #else 822 machclk_freq = i586_ctr_freq; 823 #endif 824 #elif defined(__NetBSD__) 825 machclk_freq = (u_int32_t)curcpu()->ci_tsc_freq; 826 #elif defined(__OpenBSD__) && (defined(I586_CPU) || defined(I686_CPU)) 827 machclk_freq = pentium_mhz * 1000000; 828 #endif 829 #elif defined(__alpha__) 830 #ifdef __FreeBSD__ 831 machclk_freq = cycles_per_sec; 832 #elif defined(__NetBSD__) || defined(__OpenBSD__) 833 machclk_freq = (u_int32_t)(cycles_per_usec * 1000000); 834 #endif 835 #endif /* __alpha__ */ 836 837 /* 838 * if we don't know the clock frequency, measure it. 839 */ 840 if (machclk_freq == 0) { 841 static int wait; 842 struct timeval tv_start, tv_end; 843 u_int64_t start, end, diff; 844 int timo; 845 846 microtime(&tv_start); 847 start = read_machclk(); 848 timo = hz; /* 1 sec */ 849 (void)tsleep(&wait, PWAIT | PCATCH, "init_machclk", timo); 850 microtime(&tv_end); 851 end = read_machclk(); 852 diff = (u_int64_t)(tv_end.tv_sec - tv_start.tv_sec) * 1000000 853 + tv_end.tv_usec - tv_start.tv_usec; 854 if (diff != 0) 855 machclk_freq = (u_int)((end - start) * 1000000 / diff); 856 } 857 858 machclk_per_tick = machclk_freq / hz; 859 860 #ifdef ALTQ_DEBUG 861 printf("altq: CPU clock: %uHz\n", machclk_freq); 862 #endif 863 } 864 865 #if defined(__OpenBSD__) && defined(__i386__) 866 static inline u_int64_t 867 rdtsc(void) 868 { 869 u_int64_t rv; 870 __asm __volatile(".byte 0x0f, 0x31" : "=A" (rv)); 871 return (rv); 872 } 873 #endif /* __OpenBSD__ && __i386__ */ 874 875 u_int64_t 876 read_machclk(void) 877 { 878 u_int64_t val; 879 880 if (machclk_usepcc) { 881 #if defined(__i386__) 882 val = rdtsc(); 883 #elif defined(__alpha__) 884 static u_int32_t last_pcc, upper; 885 u_int32_t pcc; 886 887 /* 888 * for alpha, make a 64bit counter value out of the 32bit 889 * alpha processor cycle counter. 890 * read_machclk must be called within a half of its 891 * wrap-around cycle (about 5 sec for 400MHz cpu) to properly 892 * detect a counter wrap-around. 893 * tbr_timeout calls read_machclk once a second. 894 */ 895 pcc = (u_int32_t)alpha_rpcc(); 896 if (pcc <= last_pcc) 897 upper++; 898 last_pcc = pcc; 899 val = ((u_int64_t)upper << 32) + pcc; 900 #else 901 panic("read_machclk"); 902 #endif 903 } else { 904 struct timeval tv; 905 906 microtime(&tv); 907 val = (((u_int64_t)(tv.tv_sec - boottime.tv_sec) * 1000000 908 + tv.tv_usec) << MACHCLK_SHIFT); 909 } 910 return (val); 911 } 912 913 #ifdef ALTQ3_CLFIER_COMPAT 914 915 #ifndef IPPROTO_ESP 916 #define IPPROTO_ESP 50 /* encapsulating security payload */ 917 #endif 918 #ifndef IPPROTO_AH 919 #define IPPROTO_AH 51 /* authentication header */ 920 #endif 921 922 /* 923 * extract flow information from a given packet. 924 * filt_mask shows flowinfo fields required. 925 * we assume the ip header is in one mbuf, and addresses and ports are 926 * in network byte order. 927 */ 928 int 929 altq_extractflow(struct mbuf *m, int af, struct flowinfo *flow, 930 u_int32_t filt_bmask) 931 { 932 933 switch (af) { 934 case PF_INET: { 935 struct flowinfo_in *fin; 936 struct ip *ip; 937 938 ip = mtod(m, struct ip *); 939 940 if (ip->ip_v != 4) 941 break; 942 943 fin = (struct flowinfo_in *)flow; 944 fin->fi_len = sizeof(struct flowinfo_in); 945 fin->fi_family = AF_INET; 946 947 fin->fi_proto = ip->ip_p; 948 fin->fi_tos = ip->ip_tos; 949 950 fin->fi_src.s_addr = ip->ip_src.s_addr; 951 fin->fi_dst.s_addr = ip->ip_dst.s_addr; 952 953 if (filt_bmask & FIMB4_PORTS) 954 /* if port info is required, extract port numbers */ 955 extract_ports4(m, ip, fin); 956 else { 957 fin->fi_sport = 0; 958 fin->fi_dport = 0; 959 fin->fi_gpi = 0; 960 } 961 return (1); 962 } 963 964 #ifdef INET6 965 case PF_INET6: { 966 struct flowinfo_in6 *fin6; 967 struct ip6_hdr *ip6; 968 969 ip6 = mtod(m, struct ip6_hdr *); 970 /* should we check the ip version? */ 971 972 fin6 = (struct flowinfo_in6 *)flow; 973 fin6->fi6_len = sizeof(struct flowinfo_in6); 974 fin6->fi6_family = AF_INET6; 975 976 fin6->fi6_proto = ip6->ip6_nxt; 977 fin6->fi6_tclass = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 978 979 fin6->fi6_flowlabel = ip6->ip6_flow & htonl(0x000fffff); 980 fin6->fi6_src = ip6->ip6_src; 981 fin6->fi6_dst = ip6->ip6_dst; 982 983 if ((filt_bmask & FIMB6_PORTS) || 984 ((filt_bmask & FIMB6_PROTO) 985 && ip6->ip6_nxt > IPPROTO_IPV6)) 986 /* 987 * if port info is required, or proto is required 988 * but there are option headers, extract port 989 * and protocol numbers. 990 */ 991 extract_ports6(m, ip6, fin6); 992 else { 993 fin6->fi6_sport = 0; 994 fin6->fi6_dport = 0; 995 fin6->fi6_gpi = 0; 996 } 997 return (1); 998 } 999 #endif /* INET6 */ 1000 1001 default: 1002 break; 1003 } 1004 1005 /* failed */ 1006 flow->fi_len = sizeof(struct flowinfo); 1007 flow->fi_family = AF_UNSPEC; 1008 return (0); 1009 } 1010 1011 /* 1012 * helper routine to extract port numbers 1013 */ 1014 /* structure for ipsec and ipv6 option header template */ 1015 struct _opt6 { 1016 u_int8_t opt6_nxt; /* next header */ 1017 u_int8_t opt6_hlen; /* header extension length */ 1018 u_int16_t _pad; 1019 u_int32_t ah_spi; /* security parameter index 1020 for authentication header */ 1021 }; 1022 1023 /* 1024 * extract port numbers from a ipv4 packet. 1025 */ 1026 static int 1027 extract_ports4(struct mbuf *m, struct ip *ip, struct flowinfo_in *fin) 1028 { 1029 struct mbuf *m0; 1030 u_short ip_off; 1031 u_int8_t proto; 1032 int off; 1033 1034 fin->fi_sport = 0; 1035 fin->fi_dport = 0; 1036 fin->fi_gpi = 0; 1037 1038 ip_off = ntohs(ip->ip_off); 1039 /* if it is a fragment, try cached fragment info */ 1040 if (ip_off & IP_OFFMASK) { 1041 ip4f_lookup(ip, fin); 1042 return (1); 1043 } 1044 1045 /* locate the mbuf containing the protocol header */ 1046 for (m0 = m; m0 != NULL; m0 = m0->m_next) 1047 if (((caddr_t)ip >= m0->m_data) && 1048 ((caddr_t)ip < m0->m_data + m0->m_len)) 1049 break; 1050 if (m0 == NULL) { 1051 #ifdef ALTQ_DEBUG 1052 printf("extract_ports4: can't locate header! ip=%p\n", ip); 1053 #endif 1054 return (0); 1055 } 1056 off = ((caddr_t)ip - m0->m_data) + (ip->ip_hl << 2); 1057 proto = ip->ip_p; 1058 1059 #ifdef ALTQ_IPSEC 1060 again: 1061 #endif 1062 while (off >= m0->m_len) { 1063 off -= m0->m_len; 1064 m0 = m0->m_next; 1065 if (m0 == NULL) 1066 return (0); /* bogus ip_hl! */ 1067 } 1068 if (m0->m_len < off + 4) 1069 return (0); 1070 1071 switch (proto) { 1072 case IPPROTO_TCP: 1073 case IPPROTO_UDP: { 1074 struct udphdr *udp; 1075 1076 udp = (struct udphdr *)(mtod(m0, caddr_t) + off); 1077 fin->fi_sport = udp->uh_sport; 1078 fin->fi_dport = udp->uh_dport; 1079 fin->fi_proto = proto; 1080 } 1081 break; 1082 1083 #ifdef ALTQ_IPSEC 1084 case IPPROTO_ESP: 1085 if (fin->fi_gpi == 0){ 1086 u_int32_t *gpi; 1087 1088 gpi = (u_int32_t *)(mtod(m0, caddr_t) + off); 1089 fin->fi_gpi = *gpi; 1090 } 1091 fin->fi_proto = proto; 1092 break; 1093 1094 case IPPROTO_AH: { 1095 /* get next header and header length */ 1096 struct _opt6 *opt6; 1097 1098 opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off); 1099 proto = opt6->opt6_nxt; 1100 off += 8 + (opt6->opt6_hlen * 4); 1101 if (fin->fi_gpi == 0 && m0->m_len >= off + 8) 1102 fin->fi_gpi = opt6->ah_spi; 1103 } 1104 /* goto the next header */ 1105 goto again; 1106 #endif /* ALTQ_IPSEC */ 1107 1108 default: 1109 fin->fi_proto = proto; 1110 return (0); 1111 } 1112 1113 /* if this is a first fragment, cache it. */ 1114 if (ip_off & IP_MF) 1115 ip4f_cache(ip, fin); 1116 1117 return (1); 1118 } 1119 1120 #ifdef INET6 1121 static int 1122 extract_ports6(struct mbuf *m, struct ip6_hdr *ip6, struct flowinfo_in6 *fin6) 1123 { 1124 struct mbuf *m0; 1125 int off; 1126 u_int8_t proto; 1127 1128 fin6->fi6_gpi = 0; 1129 fin6->fi6_sport = 0; 1130 fin6->fi6_dport = 0; 1131 1132 /* locate the mbuf containing the protocol header */ 1133 for (m0 = m; m0 != NULL; m0 = m0->m_next) 1134 if (((caddr_t)ip6 >= m0->m_data) && 1135 ((caddr_t)ip6 < m0->m_data + m0->m_len)) 1136 break; 1137 if (m0 == NULL) { 1138 #ifdef ALTQ_DEBUG 1139 printf("extract_ports6: can't locate header! ip6=%p\n", ip6); 1140 #endif 1141 return (0); 1142 } 1143 off = ((caddr_t)ip6 - m0->m_data) + sizeof(struct ip6_hdr); 1144 1145 proto = ip6->ip6_nxt; 1146 do { 1147 while (off >= m0->m_len) { 1148 off -= m0->m_len; 1149 m0 = m0->m_next; 1150 if (m0 == NULL) 1151 return (0); 1152 } 1153 if (m0->m_len < off + 4) 1154 return (0); 1155 1156 switch (proto) { 1157 case IPPROTO_TCP: 1158 case IPPROTO_UDP: { 1159 struct udphdr *udp; 1160 1161 udp = (struct udphdr *)(mtod(m0, caddr_t) + off); 1162 fin6->fi6_sport = udp->uh_sport; 1163 fin6->fi6_dport = udp->uh_dport; 1164 fin6->fi6_proto = proto; 1165 } 1166 return (1); 1167 1168 case IPPROTO_ESP: 1169 if (fin6->fi6_gpi == 0) { 1170 u_int32_t *gpi; 1171 1172 gpi = (u_int32_t *)(mtod(m0, caddr_t) + off); 1173 fin6->fi6_gpi = *gpi; 1174 } 1175 fin6->fi6_proto = proto; 1176 return (1); 1177 1178 case IPPROTO_AH: { 1179 /* get next header and header length */ 1180 struct _opt6 *opt6; 1181 1182 opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off); 1183 if (fin6->fi6_gpi == 0 && m0->m_len >= off + 8) 1184 fin6->fi6_gpi = opt6->ah_spi; 1185 proto = opt6->opt6_nxt; 1186 off += 8 + (opt6->opt6_hlen * 4); 1187 /* goto the next header */ 1188 break; 1189 } 1190 1191 case IPPROTO_HOPOPTS: 1192 case IPPROTO_ROUTING: 1193 case IPPROTO_DSTOPTS: { 1194 /* get next header and header length */ 1195 struct _opt6 *opt6; 1196 1197 opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off); 1198 proto = opt6->opt6_nxt; 1199 off += (opt6->opt6_hlen + 1) * 8; 1200 /* goto the next header */ 1201 break; 1202 } 1203 1204 case IPPROTO_FRAGMENT: 1205 /* ipv6 fragmentations are not supported yet */ 1206 default: 1207 fin6->fi6_proto = proto; 1208 return (0); 1209 } 1210 } while (1); 1211 /*NOTREACHED*/ 1212 } 1213 #endif /* INET6 */ 1214 1215 /* 1216 * altq common classifier 1217 */ 1218 int 1219 acc_add_filter(struct acc_classifier *classifier, struct flow_filter *filter, 1220 void *class, u_long *phandle) 1221 { 1222 struct acc_filter *afp, *prev, *tmp; 1223 int i, s; 1224 1225 #ifdef INET6 1226 if (filter->ff_flow.fi_family != AF_INET && 1227 filter->ff_flow.fi_family != AF_INET6) 1228 return (EINVAL); 1229 #else 1230 if (filter->ff_flow.fi_family != AF_INET) 1231 return (EINVAL); 1232 #endif 1233 1234 afp = malloc(sizeof(struct acc_filter), M_DEVBUF, M_WAITOK|M_ZERO); 1235 if (afp == NULL) 1236 return (ENOMEM); 1237 1238 afp->f_filter = *filter; 1239 afp->f_class = class; 1240 1241 i = ACC_WILDCARD_INDEX; 1242 if (filter->ff_flow.fi_family == AF_INET) { 1243 struct flow_filter *filter4 = &afp->f_filter; 1244 1245 /* 1246 * if address is 0, it's a wildcard. if address mask 1247 * isn't set, use full mask. 1248 */ 1249 if (filter4->ff_flow.fi_dst.s_addr == 0) 1250 filter4->ff_mask.mask_dst.s_addr = 0; 1251 else if (filter4->ff_mask.mask_dst.s_addr == 0) 1252 filter4->ff_mask.mask_dst.s_addr = 0xffffffff; 1253 if (filter4->ff_flow.fi_src.s_addr == 0) 1254 filter4->ff_mask.mask_src.s_addr = 0; 1255 else if (filter4->ff_mask.mask_src.s_addr == 0) 1256 filter4->ff_mask.mask_src.s_addr = 0xffffffff; 1257 1258 /* clear extra bits in addresses */ 1259 filter4->ff_flow.fi_dst.s_addr &= 1260 filter4->ff_mask.mask_dst.s_addr; 1261 filter4->ff_flow.fi_src.s_addr &= 1262 filter4->ff_mask.mask_src.s_addr; 1263 1264 /* 1265 * if dst address is a wildcard, use hash-entry 1266 * ACC_WILDCARD_INDEX. 1267 */ 1268 if (filter4->ff_mask.mask_dst.s_addr != 0xffffffff) 1269 i = ACC_WILDCARD_INDEX; 1270 else 1271 i = ACC_GET_HASH_INDEX(filter4->ff_flow.fi_dst.s_addr); 1272 } 1273 #ifdef INET6 1274 else if (filter->ff_flow.fi_family == AF_INET6) { 1275 struct flow_filter6 *filter6 = 1276 (struct flow_filter6 *)&afp->f_filter; 1277 #ifndef IN6MASK0 /* taken from kame ipv6 */ 1278 #define IN6MASK0 {{{ 0, 0, 0, 0 }}} 1279 #define IN6MASK128 {{{ 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff }}} 1280 const struct in6_addr in6mask0 = IN6MASK0; 1281 const struct in6_addr in6mask128 = IN6MASK128; 1282 #endif 1283 1284 if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_dst)) 1285 filter6->ff_mask6.mask6_dst = in6mask0; 1286 else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_dst)) 1287 filter6->ff_mask6.mask6_dst = in6mask128; 1288 if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_src)) 1289 filter6->ff_mask6.mask6_src = in6mask0; 1290 else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_src)) 1291 filter6->ff_mask6.mask6_src = in6mask128; 1292 1293 /* clear extra bits in addresses */ 1294 for (i = 0; i < 16; i++) 1295 filter6->ff_flow6.fi6_dst.s6_addr[i] &= 1296 filter6->ff_mask6.mask6_dst.s6_addr[i]; 1297 for (i = 0; i < 16; i++) 1298 filter6->ff_flow6.fi6_src.s6_addr[i] &= 1299 filter6->ff_mask6.mask6_src.s6_addr[i]; 1300 1301 if (filter6->ff_flow6.fi6_flowlabel == 0) 1302 i = ACC_WILDCARD_INDEX; 1303 else 1304 i = ACC_GET_HASH_INDEX(filter6->ff_flow6.fi6_flowlabel); 1305 } 1306 #endif /* INET6 */ 1307 1308 afp->f_handle = get_filt_handle(classifier, i); 1309 1310 /* update filter bitmask */ 1311 afp->f_fbmask = filt2fibmask(filter); 1312 classifier->acc_fbmask |= afp->f_fbmask; 1313 1314 /* 1315 * add this filter to the filter list. 1316 * filters are ordered from the highest rule number. 1317 */ 1318 s = splnet(); 1319 prev = NULL; 1320 LIST_FOREACH(tmp, &classifier->acc_filters[i], f_chain) { 1321 if (tmp->f_filter.ff_ruleno > afp->f_filter.ff_ruleno) 1322 prev = tmp; 1323 else 1324 break; 1325 } 1326 if (prev == NULL) 1327 LIST_INSERT_HEAD(&classifier->acc_filters[i], afp, f_chain); 1328 else 1329 LIST_INSERT_AFTER(prev, afp, f_chain); 1330 splx(s); 1331 1332 *phandle = afp->f_handle; 1333 return (0); 1334 } 1335 1336 int 1337 acc_delete_filter(struct acc_classifier *classifier, u_long handle) 1338 { 1339 struct acc_filter *afp; 1340 int s; 1341 1342 if ((afp = filth_to_filtp(classifier, handle)) == NULL) 1343 return (EINVAL); 1344 1345 s = splnet(); 1346 LIST_REMOVE(afp, f_chain); 1347 splx(s); 1348 1349 free(afp, M_DEVBUF); 1350 1351 /* todo: update filt_bmask */ 1352 1353 return (0); 1354 } 1355 1356 /* 1357 * delete filters referencing to the specified class. 1358 * if the all flag is not 0, delete all the filters. 1359 */ 1360 int 1361 acc_discard_filters(struct acc_classifier *classifier, void *class, int all) 1362 { 1363 struct acc_filter *afp; 1364 int i, s; 1365 1366 s = splnet(); 1367 for (i = 0; i < ACC_FILTER_TABLESIZE; i++) { 1368 do { 1369 LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain) 1370 if (all || afp->f_class == class) { 1371 LIST_REMOVE(afp, f_chain); 1372 free(afp, M_DEVBUF); 1373 /* start again from the head */ 1374 break; 1375 } 1376 } while (afp != NULL); 1377 } 1378 splx(s); 1379 1380 if (all) 1381 classifier->acc_fbmask = 0; 1382 1383 return (0); 1384 } 1385 1386 void * 1387 acc_classify(void *clfier, struct mbuf *m, int af) 1388 { 1389 struct acc_classifier *classifier; 1390 struct flowinfo flow; 1391 struct acc_filter *afp; 1392 int i; 1393 1394 classifier = (struct acc_classifier *)clfier; 1395 altq_extractflow(m, af, &flow, classifier->acc_fbmask); 1396 1397 if (flow.fi_family == AF_INET) { 1398 struct flowinfo_in *fp = (struct flowinfo_in *)&flow; 1399 1400 if ((classifier->acc_fbmask & FIMB4_ALL) == FIMB4_TOS) { 1401 /* only tos is used */ 1402 LIST_FOREACH(afp, 1403 &classifier->acc_filters[ACC_WILDCARD_INDEX], 1404 f_chain) 1405 if (apply_tosfilter4(afp->f_fbmask, 1406 &afp->f_filter, fp)) 1407 /* filter matched */ 1408 return (afp->f_class); 1409 } else if ((classifier->acc_fbmask & 1410 (~(FIMB4_PROTO|FIMB4_SPORT|FIMB4_DPORT) & FIMB4_ALL)) 1411 == 0) { 1412 /* only proto and ports are used */ 1413 LIST_FOREACH(afp, 1414 &classifier->acc_filters[ACC_WILDCARD_INDEX], 1415 f_chain) 1416 if (apply_ppfilter4(afp->f_fbmask, 1417 &afp->f_filter, fp)) 1418 /* filter matched */ 1419 return (afp->f_class); 1420 } else { 1421 /* get the filter hash entry from its dest address */ 1422 i = ACC_GET_HASH_INDEX(fp->fi_dst.s_addr); 1423 do { 1424 /* 1425 * go through this loop twice. first for dst 1426 * hash, second for wildcards. 1427 */ 1428 LIST_FOREACH(afp, &classifier->acc_filters[i], 1429 f_chain) 1430 if (apply_filter4(afp->f_fbmask, 1431 &afp->f_filter, fp)) 1432 /* filter matched */ 1433 return (afp->f_class); 1434 1435 /* 1436 * check again for filters with a dst addr 1437 * wildcard. 1438 * (daddr == 0 || dmask != 0xffffffff). 1439 */ 1440 if (i != ACC_WILDCARD_INDEX) 1441 i = ACC_WILDCARD_INDEX; 1442 else 1443 break; 1444 } while (1); 1445 } 1446 } 1447 #ifdef INET6 1448 else if (flow.fi_family == AF_INET6) { 1449 struct flowinfo_in6 *fp6 = (struct flowinfo_in6 *)&flow; 1450 1451 /* get the filter hash entry from its flow ID */ 1452 if (fp6->fi6_flowlabel != 0) 1453 i = ACC_GET_HASH_INDEX(fp6->fi6_flowlabel); 1454 else 1455 /* flowlable can be zero */ 1456 i = ACC_WILDCARD_INDEX; 1457 1458 /* go through this loop twice. first for flow hash, second 1459 for wildcards. */ 1460 do { 1461 LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain) 1462 if (apply_filter6(afp->f_fbmask, 1463 (struct flow_filter6 *)&afp->f_filter, 1464 fp6)) 1465 /* filter matched */ 1466 return (afp->f_class); 1467 1468 /* 1469 * check again for filters with a wildcard. 1470 */ 1471 if (i != ACC_WILDCARD_INDEX) 1472 i = ACC_WILDCARD_INDEX; 1473 else 1474 break; 1475 } while (1); 1476 } 1477 #endif /* INET6 */ 1478 1479 /* no filter matched */ 1480 return (NULL); 1481 } 1482 1483 static int 1484 apply_filter4(u_int32_t fbmask, struct flow_filter *filt, 1485 struct flowinfo_in *pkt) 1486 { 1487 if (filt->ff_flow.fi_family != AF_INET) 1488 return (0); 1489 if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport) 1490 return (0); 1491 if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport) 1492 return (0); 1493 if ((fbmask & FIMB4_DADDR) && 1494 filt->ff_flow.fi_dst.s_addr != 1495 (pkt->fi_dst.s_addr & filt->ff_mask.mask_dst.s_addr)) 1496 return (0); 1497 if ((fbmask & FIMB4_SADDR) && 1498 filt->ff_flow.fi_src.s_addr != 1499 (pkt->fi_src.s_addr & filt->ff_mask.mask_src.s_addr)) 1500 return (0); 1501 if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto) 1502 return (0); 1503 if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos != 1504 (pkt->fi_tos & filt->ff_mask.mask_tos)) 1505 return (0); 1506 if ((fbmask & FIMB4_GPI) && filt->ff_flow.fi_gpi != (pkt->fi_gpi)) 1507 return (0); 1508 /* match */ 1509 return (1); 1510 } 1511 1512 /* 1513 * filter matching function optimized for a common case that checks 1514 * only protocol and port numbers 1515 */ 1516 static int 1517 apply_ppfilter4(u_int32_t fbmask, struct flow_filter *filt, 1518 struct flowinfo_in *pkt) 1519 { 1520 if (filt->ff_flow.fi_family != AF_INET) 1521 return (0); 1522 if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport) 1523 return (0); 1524 if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport) 1525 return (0); 1526 if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto) 1527 return (0); 1528 /* match */ 1529 return (1); 1530 } 1531 1532 /* 1533 * filter matching function only for tos field. 1534 */ 1535 static int 1536 apply_tosfilter4(u_int32_t fbmask, struct flow_filter *filt, 1537 struct flowinfo_in *pkt) 1538 { 1539 if (filt->ff_flow.fi_family != AF_INET) 1540 return (0); 1541 if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos != 1542 (pkt->fi_tos & filt->ff_mask.mask_tos)) 1543 return (0); 1544 /* match */ 1545 return (1); 1546 } 1547 1548 #ifdef INET6 1549 static int 1550 apply_filter6(u_int32_t fbmask, struct flow_filter6 *filt, 1551 struct flowinfo_in6 *pkt) 1552 { 1553 int i; 1554 1555 if (filt->ff_flow6.fi6_family != AF_INET6) 1556 return (0); 1557 if ((fbmask & FIMB6_FLABEL) && 1558 filt->ff_flow6.fi6_flowlabel != pkt->fi6_flowlabel) 1559 return (0); 1560 if ((fbmask & FIMB6_PROTO) && 1561 filt->ff_flow6.fi6_proto != pkt->fi6_proto) 1562 return (0); 1563 if ((fbmask & FIMB6_SPORT) && 1564 filt->ff_flow6.fi6_sport != pkt->fi6_sport) 1565 return (0); 1566 if ((fbmask & FIMB6_DPORT) && 1567 filt->ff_flow6.fi6_dport != pkt->fi6_dport) 1568 return (0); 1569 if (fbmask & FIMB6_SADDR) { 1570 for (i = 0; i < 4; i++) 1571 if (filt->ff_flow6.fi6_src.s6_addr32[i] != 1572 (pkt->fi6_src.s6_addr32[i] & 1573 filt->ff_mask6.mask6_src.s6_addr32[i])) 1574 return (0); 1575 } 1576 if (fbmask & FIMB6_DADDR) { 1577 for (i = 0; i < 4; i++) 1578 if (filt->ff_flow6.fi6_dst.s6_addr32[i] != 1579 (pkt->fi6_dst.s6_addr32[i] & 1580 filt->ff_mask6.mask6_dst.s6_addr32[i])) 1581 return (0); 1582 } 1583 if ((fbmask & FIMB6_TCLASS) && 1584 filt->ff_flow6.fi6_tclass != 1585 (pkt->fi6_tclass & filt->ff_mask6.mask6_tclass)) 1586 return (0); 1587 if ((fbmask & FIMB6_GPI) && 1588 filt->ff_flow6.fi6_gpi != pkt->fi6_gpi) 1589 return (0); 1590 /* match */ 1591 return (1); 1592 } 1593 #endif /* INET6 */ 1594 1595 /* 1596 * filter handle: 1597 * bit 20-28: index to the filter hash table 1598 * bit 0-19: unique id in the hash bucket. 1599 */ 1600 static u_long 1601 get_filt_handle(struct acc_classifier *classifier, int i) 1602 { 1603 static u_long handle_number = 1; 1604 u_long handle; 1605 struct acc_filter *afp; 1606 1607 while (1) { 1608 handle = handle_number++ & 0x000fffff; 1609 1610 if (LIST_EMPTY(&classifier->acc_filters[i])) 1611 break; 1612 1613 LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain) 1614 if ((afp->f_handle & 0x000fffff) == handle) 1615 break; 1616 if (afp == NULL) 1617 break; 1618 /* this handle is already used, try again */ 1619 } 1620 1621 return ((i << 20) | handle); 1622 } 1623 1624 /* convert filter handle to filter pointer */ 1625 static struct acc_filter * 1626 filth_to_filtp(struct acc_classifier *classifier, u_long handle) 1627 { 1628 struct acc_filter *afp; 1629 int i; 1630 1631 i = ACC_GET_HINDEX(handle); 1632 1633 LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain) 1634 if (afp->f_handle == handle) 1635 return (afp); 1636 1637 return (NULL); 1638 } 1639 1640 /* create flowinfo bitmask */ 1641 static u_int32_t 1642 filt2fibmask(struct flow_filter *filt) 1643 { 1644 u_int32_t mask = 0; 1645 #ifdef INET6 1646 struct flow_filter6 *filt6; 1647 #endif 1648 1649 switch (filt->ff_flow.fi_family) { 1650 case AF_INET: 1651 if (filt->ff_flow.fi_proto != 0) 1652 mask |= FIMB4_PROTO; 1653 if (filt->ff_flow.fi_tos != 0) 1654 mask |= FIMB4_TOS; 1655 if (filt->ff_flow.fi_dst.s_addr != 0) 1656 mask |= FIMB4_DADDR; 1657 if (filt->ff_flow.fi_src.s_addr != 0) 1658 mask |= FIMB4_SADDR; 1659 if (filt->ff_flow.fi_sport != 0) 1660 mask |= FIMB4_SPORT; 1661 if (filt->ff_flow.fi_dport != 0) 1662 mask |= FIMB4_DPORT; 1663 if (filt->ff_flow.fi_gpi != 0) 1664 mask |= FIMB4_GPI; 1665 break; 1666 #ifdef INET6 1667 case AF_INET6: 1668 filt6 = (struct flow_filter6 *)filt; 1669 1670 if (filt6->ff_flow6.fi6_proto != 0) 1671 mask |= FIMB6_PROTO; 1672 if (filt6->ff_flow6.fi6_tclass != 0) 1673 mask |= FIMB6_TCLASS; 1674 if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_dst)) 1675 mask |= FIMB6_DADDR; 1676 if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_src)) 1677 mask |= FIMB6_SADDR; 1678 if (filt6->ff_flow6.fi6_sport != 0) 1679 mask |= FIMB6_SPORT; 1680 if (filt6->ff_flow6.fi6_dport != 0) 1681 mask |= FIMB6_DPORT; 1682 if (filt6->ff_flow6.fi6_gpi != 0) 1683 mask |= FIMB6_GPI; 1684 if (filt6->ff_flow6.fi6_flowlabel != 0) 1685 mask |= FIMB6_FLABEL; 1686 break; 1687 #endif /* INET6 */ 1688 } 1689 return (mask); 1690 } 1691 1692 1693 /* 1694 * helper functions to handle IPv4 fragments. 1695 * currently only in-sequence fragments are handled. 1696 * - fragment info is cached in a LRU list. 1697 * - when a first fragment is found, cache its flow info. 1698 * - when a non-first fragment is found, lookup the cache. 1699 */ 1700 1701 struct ip4_frag { 1702 TAILQ_ENTRY(ip4_frag) ip4f_chain; 1703 char ip4f_valid; 1704 u_short ip4f_id; 1705 struct flowinfo_in ip4f_info; 1706 }; 1707 1708 static TAILQ_HEAD(ip4f_list, ip4_frag) ip4f_list; /* IPv4 fragment cache */ 1709 1710 #define IP4F_TABSIZE 16 /* IPv4 fragment cache size */ 1711 1712 1713 static void 1714 ip4f_cache(struct ip *ip, struct flowinfo_in *fin) 1715 { 1716 struct ip4_frag *fp; 1717 1718 if (TAILQ_EMPTY(&ip4f_list)) { 1719 /* first time call, allocate fragment cache entries. */ 1720 if (ip4f_init() < 0) 1721 /* allocation failed! */ 1722 return; 1723 } 1724 1725 fp = ip4f_alloc(); 1726 fp->ip4f_id = ip->ip_id; 1727 fp->ip4f_info.fi_proto = ip->ip_p; 1728 fp->ip4f_info.fi_src.s_addr = ip->ip_src.s_addr; 1729 fp->ip4f_info.fi_dst.s_addr = ip->ip_dst.s_addr; 1730 1731 /* save port numbers */ 1732 fp->ip4f_info.fi_sport = fin->fi_sport; 1733 fp->ip4f_info.fi_dport = fin->fi_dport; 1734 fp->ip4f_info.fi_gpi = fin->fi_gpi; 1735 } 1736 1737 static int 1738 ip4f_lookup(struct ip *ip, struct flowinfo_in *fin) 1739 { 1740 struct ip4_frag *fp; 1741 1742 for (fp = TAILQ_FIRST(&ip4f_list); fp != NULL && fp->ip4f_valid; 1743 fp = TAILQ_NEXT(fp, ip4f_chain)) 1744 if (ip->ip_id == fp->ip4f_id && 1745 ip->ip_src.s_addr == fp->ip4f_info.fi_src.s_addr && 1746 ip->ip_dst.s_addr == fp->ip4f_info.fi_dst.s_addr && 1747 ip->ip_p == fp->ip4f_info.fi_proto) { 1748 1749 /* found the matching entry */ 1750 fin->fi_sport = fp->ip4f_info.fi_sport; 1751 fin->fi_dport = fp->ip4f_info.fi_dport; 1752 fin->fi_gpi = fp->ip4f_info.fi_gpi; 1753 1754 if ((ntohs(ip->ip_off) & IP_MF) == 0) 1755 /* this is the last fragment, 1756 release the entry. */ 1757 ip4f_free(fp); 1758 1759 return (1); 1760 } 1761 1762 /* no matching entry found */ 1763 return (0); 1764 } 1765 1766 static int 1767 ip4f_init(void) 1768 { 1769 struct ip4_frag *fp; 1770 int i; 1771 1772 TAILQ_INIT(&ip4f_list); 1773 for (i=0; i<IP4F_TABSIZE; i++) { 1774 fp = malloc(sizeof(struct ip4_frag), M_DEVBUF, M_NOWAIT); 1775 if (fp == NULL) { 1776 printf("ip4f_init: can't alloc %dth entry!\n", i); 1777 if (i == 0) 1778 return (-1); 1779 return (0); 1780 } 1781 fp->ip4f_valid = 0; 1782 TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain); 1783 } 1784 return (0); 1785 } 1786 1787 static struct ip4_frag * 1788 ip4f_alloc(void) 1789 { 1790 struct ip4_frag *fp; 1791 1792 /* reclaim an entry at the tail, put it at the head */ 1793 fp = TAILQ_LAST(&ip4f_list, ip4f_list); 1794 TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain); 1795 fp->ip4f_valid = 1; 1796 TAILQ_INSERT_HEAD(&ip4f_list, fp, ip4f_chain); 1797 return (fp); 1798 } 1799 1800 static void 1801 ip4f_free(struct ip4_frag *fp) 1802 { 1803 TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain); 1804 fp->ip4f_valid = 0; 1805 TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain); 1806 } 1807 1808 #endif /* ALTQ3_CLFIER_COMPAT */ 1809
Indexes created Wed Oct 15 16:09:53 GMT 2025