1 /* $KAME: altq_subr.c,v 1.23 2004/04/20 16:10:06 itojun Exp $ */ 2 /* $DragonFly: src/sys/net/altq/altq_subr.c,v 1.12 2008/05/14 11:59:23 sephe 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 "opt_altq.h" 31 #include "opt_inet.h" 32 #include "opt_inet6.h" 33 34 #include <sys/param.h> 35 #include <sys/malloc.h> 36 #include <sys/mbuf.h> 37 #include <sys/systm.h> 38 #include <sys/proc.h> 39 #include <sys/socket.h> 40 #include <sys/socketvar.h> 41 #include <sys/kernel.h> 42 #include <sys/callout.h> 43 #include <sys/errno.h> 44 #include <sys/syslog.h> 45 #include <sys/sysctl.h> 46 #include <sys/queue.h> 47 #include <sys/thread2.h> 48 49 #include <net/if.h> 50 #include <net/if_dl.h> 51 #include <net/if_types.h> 52 #include <net/ifq_var.h> 53 #include <net/netmsg2.h> 54 #include <net/netisr2.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 #include <net/pf/pfvar.h> 66 #include <net/altq/altq.h> 67 68 /* machine dependent clock related includes */ 69 #include <machine/clock.h> /* for tsc_frequency */ 70 #include <machine/md_var.h> /* for cpu_feature */ 71 #include <machine/specialreg.h> /* for CPUID_TSC */ 72 73 /* 74 * internal function prototypes 75 */ 76 static void tbr_timeout(void *); 77 static void tbr_timeout_dispatch(netmsg_t); 78 static int altq_enable_locked(struct ifaltq *); 79 static int altq_disable_locked(struct ifaltq *); 80 static int altq_detach_locked(struct ifaltq *); 81 static int tbr_set_locked(struct ifaltq *, struct tb_profile *); 82 83 int (*altq_input)(struct mbuf *, int) = NULL; 84 static int tbr_timer = 0; /* token bucket regulator timer */ 85 static struct callout tbr_callout; 86 static struct netmsg_base tbr_timeout_netmsg; 87 88 int pfaltq_running; /* keep track of running state */ 89 90 MALLOC_DEFINE(M_ALTQ, "altq", "ALTQ structures"); 91 92 /* 93 * alternate queueing support routines 94 */ 95 96 /* look up the queue state by the interface name and the queueing type. */ 97 void * 98 altq_lookup(const char *name, int type) 99 { 100 struct ifnet *ifp; 101 102 if ((ifp = ifunit(name)) != NULL) { 103 if (type != ALTQT_NONE && ifp->if_snd.altq_type == type) 104 return (ifp->if_snd.altq_disc); 105 } 106 107 return (NULL); 108 } 109 110 int 111 altq_attach(struct ifaltq *ifq, int type, void *discipline, 112 altq_mapsubq_t mapsubq, 113 ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request, 114 void *clfier, 115 void *(*classify)(struct ifaltq *, struct mbuf *, struct altq_pktattr *)) 116 { 117 if (!ifq_is_ready(ifq)) 118 return ENXIO; 119 120 ifq->altq_type = type; 121 ifq->altq_disc = discipline; 122 ifq->altq_clfier = clfier; 123 ifq->altq_classify = classify; 124 ifq->altq_flags &= (ALTQF_CANTCHANGE|ALTQF_ENABLED); 125 ifq_set_methods(ifq, mapsubq, enqueue, dequeue, request); 126 return 0; 127 } 128 129 static int 130 altq_detach_locked(struct ifaltq *ifq) 131 { 132 if (!ifq_is_ready(ifq)) 133 return ENXIO; 134 if (ifq_is_enabled(ifq)) 135 return EBUSY; 136 if (!ifq_is_attached(ifq)) 137 return (0); 138 139 ifq_set_classic(ifq); 140 ifq->altq_type = ALTQT_NONE; 141 ifq->altq_disc = NULL; 142 ifq->altq_clfier = NULL; 143 ifq->altq_classify = NULL; 144 ifq->altq_flags &= ALTQF_CANTCHANGE; 145 return 0; 146 } 147 148 int 149 altq_detach(struct ifaltq *ifq) 150 { 151 int error; 152 153 ifq_lock_all(ifq); 154 error = altq_detach_locked(ifq); 155 ifq_unlock_all(ifq); 156 return error; 157 } 158 159 static int 160 altq_enable_locked(struct ifaltq *ifq) 161 { 162 if (!ifq_is_ready(ifq)) 163 return ENXIO; 164 if (ifq_is_enabled(ifq)) 165 return 0; 166 167 ifq_purge_all_locked(ifq); 168 169 ifq->altq_flags |= ALTQF_ENABLED; 170 if (ifq->altq_clfier != NULL) 171 ifq->altq_flags |= ALTQF_CLASSIFY; 172 return 0; 173 } 174 175 int 176 altq_enable(struct ifaltq *ifq) 177 { 178 int error; 179 180 ifq_lock_all(ifq); 181 error = altq_enable_locked(ifq); 182 ifq_unlock_all(ifq); 183 return error; 184 } 185 186 static int 187 altq_disable_locked(struct ifaltq *ifq) 188 { 189 if (!ifq_is_enabled(ifq)) 190 return 0; 191 192 ifq_purge_all_locked(ifq); 193 ifq->altq_flags &= ~(ALTQF_ENABLED|ALTQF_CLASSIFY); 194 return 0; 195 } 196 197 int 198 altq_disable(struct ifaltq *ifq) 199 { 200 int error; 201 202 ifq_lock_all(ifq); 203 error = altq_disable_locked(ifq); 204 ifq_unlock_all(ifq); 205 return error; 206 } 207 208 /* 209 * internal representation of token bucket parameters 210 * rate: byte_per_unittime << 32 211 * (((bits_per_sec) / 8) << 32) / machclk_freq 212 * depth: byte << 32 213 * 214 */ 215 #define TBR_SHIFT 32 216 #define TBR_SCALE(x) ((int64_t)(x) << TBR_SHIFT) 217 #define TBR_UNSCALE(x) ((x) >> TBR_SHIFT) 218 219 struct mbuf * 220 tbr_dequeue(struct ifaltq_subque *ifsq, int op) 221 { 222 struct ifaltq *ifq = ifsq->ifsq_altq; 223 struct tb_regulator *tbr; 224 struct mbuf *m; 225 int64_t interval; 226 uint64_t now; 227 228 if (ifsq_get_index(ifsq) != ALTQ_SUBQ_INDEX_DEFAULT) { 229 /* 230 * Race happened, the unrelated subqueue was 231 * picked during the packet scheduler transition. 232 */ 233 ifsq_classic_request(ifsq, ALTRQ_PURGE, NULL); 234 return NULL; 235 } 236 237 crit_enter(); 238 tbr = ifq->altq_tbr; 239 if (op == ALTDQ_REMOVE && tbr->tbr_lastop == ALTDQ_POLL) { 240 /* if this is a remove after poll, bypass tbr check */ 241 } else { 242 /* update token only when it is negative */ 243 if (tbr->tbr_token <= 0) { 244 now = read_machclk(); 245 interval = now - tbr->tbr_last; 246 if (interval >= tbr->tbr_filluptime) 247 tbr->tbr_token = tbr->tbr_depth; 248 else { 249 tbr->tbr_token += interval * tbr->tbr_rate; 250 if (tbr->tbr_token > tbr->tbr_depth) 251 tbr->tbr_token = tbr->tbr_depth; 252 } 253 tbr->tbr_last = now; 254 } 255 /* if token is still negative, don't allow dequeue */ 256 if (tbr->tbr_token <= 0) { 257 crit_exit(); 258 return (NULL); 259 } 260 } 261 262 if (ifq_is_enabled(ifq)) 263 m = (*ifsq->ifsq_dequeue)(ifsq, op); 264 else 265 m = ifsq_classic_dequeue(ifsq, op); 266 267 if (m != NULL && op == ALTDQ_REMOVE) 268 tbr->tbr_token -= TBR_SCALE(m_pktlen(m)); 269 tbr->tbr_lastop = op; 270 crit_exit(); 271 return (m); 272 } 273 274 /* 275 * set a token bucket regulator. 276 * if the specified rate is zero, the token bucket regulator is deleted. 277 */ 278 static int 279 tbr_set_locked(struct ifaltq *ifq, struct tb_profile *profile) 280 { 281 struct tb_regulator *tbr, *otbr; 282 283 if (machclk_freq == 0) 284 init_machclk(); 285 if (machclk_freq == 0) { 286 kprintf("%s: no cpu clock available!\n", __func__); 287 return (ENXIO); 288 } 289 290 if (profile->rate == 0) { 291 /* delete this tbr */ 292 if ((tbr = ifq->altq_tbr) == NULL) 293 return (ENOENT); 294 ifq->altq_tbr = NULL; 295 kfree(tbr, M_ALTQ); 296 return (0); 297 } 298 299 tbr = kmalloc(sizeof(*tbr), M_ALTQ, M_WAITOK | M_ZERO); 300 tbr->tbr_rate = TBR_SCALE(profile->rate / 8) / machclk_freq; 301 tbr->tbr_depth = TBR_SCALE(profile->depth); 302 if (tbr->tbr_rate > 0) 303 tbr->tbr_filluptime = tbr->tbr_depth / tbr->tbr_rate; 304 else 305 tbr->tbr_filluptime = 0xffffffffffffffffLL; 306 tbr->tbr_token = tbr->tbr_depth; 307 tbr->tbr_last = read_machclk(); 308 tbr->tbr_lastop = ALTDQ_REMOVE; 309 310 otbr = ifq->altq_tbr; 311 ifq->altq_tbr = tbr; /* set the new tbr */ 312 313 if (otbr != NULL) 314 kfree(otbr, M_ALTQ); 315 else if (tbr_timer == 0) { 316 callout_reset_bycpu(&tbr_callout, 1, tbr_timeout, NULL, 0); 317 tbr_timer = 1; 318 } 319 return (0); 320 } 321 322 int 323 tbr_set(struct ifaltq *ifq, struct tb_profile *profile) 324 { 325 int error; 326 327 ifq_lock_all(ifq); 328 error = tbr_set_locked(ifq, profile); 329 ifq_unlock_all(ifq); 330 return error; 331 } 332 333 static void 334 tbr_timeout(void *arg __unused) 335 { 336 struct lwkt_msg *lmsg = &tbr_timeout_netmsg.lmsg; 337 338 KASSERT(mycpuid == 0, ("not on cpu0")); 339 crit_enter(); 340 if (lmsg->ms_flags & MSGF_DONE) 341 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg); 342 crit_exit(); 343 } 344 345 /* 346 * tbr_timeout goes through the interface list, and kicks the drivers 347 * if necessary. 348 */ 349 static void 350 tbr_timeout_dispatch(netmsg_t nmsg) 351 { 352 const struct ifnet_array *arr; 353 int active, i; 354 355 KASSERT(&curthread->td_msgport == netisr_cpuport(0), 356 ("not in netisr0")); 357 358 crit_enter(); 359 lwkt_replymsg(&nmsg->lmsg, 0); /* reply ASAP */ 360 crit_exit(); 361 362 active = 0; 363 arr = ifnet_array_get(); 364 for (i = 0; i < arr->ifnet_count; ++i) { 365 struct ifnet *ifp = arr->ifnet_arr[i]; 366 struct ifaltq_subque *ifsq; 367 368 if (ifp->if_snd.altq_tbr == NULL) 369 continue; 370 371 ifsq = &ifp->if_snd.altq_subq[ALTQ_SUBQ_INDEX_DEFAULT]; 372 active++; 373 if (!ifsq_is_empty(ifsq) && ifp->if_start != NULL) { 374 ifsq_serialize_hw(ifsq); 375 (*ifp->if_start)(ifp, ifsq); 376 ifsq_deserialize_hw(ifsq); 377 } 378 } 379 if (active > 0) 380 callout_reset(&tbr_callout, 1, tbr_timeout, NULL); 381 else 382 tbr_timer = 0; /* don't need tbr_timer anymore */ 383 } 384 385 /* 386 * get token bucket regulator profile 387 */ 388 int 389 tbr_get(struct ifaltq *ifq, struct tb_profile *profile) 390 { 391 struct tb_regulator *tbr; 392 393 if ((tbr = ifq->altq_tbr) == NULL) { 394 profile->rate = 0; 395 profile->depth = 0; 396 } else { 397 profile->rate = 398 (u_int)TBR_UNSCALE(tbr->tbr_rate * 8 * machclk_freq); 399 profile->depth = (u_int)TBR_UNSCALE(tbr->tbr_depth); 400 } 401 return (0); 402 } 403 404 /* 405 * attach a discipline to the interface. if one already exists, it is 406 * overridden. 407 */ 408 int 409 altq_pfattach(struct pf_altq *a) 410 { 411 struct ifaltq *ifq; 412 struct ifnet *ifp; 413 int error; 414 415 if (a->scheduler == ALTQT_NONE) 416 return 0; 417 418 if (a->altq_disc == NULL) 419 return EINVAL; 420 421 ifnet_lock(); 422 423 ifp = ifunit(a->ifname); 424 if (ifp == NULL) { 425 ifnet_unlock(); 426 return EINVAL; 427 } 428 ifq = &ifp->if_snd; 429 430 ifq_lock_all(ifq); 431 432 switch (a->scheduler) { 433 #ifdef ALTQ_CBQ 434 case ALTQT_CBQ: 435 error = cbq_pfattach(a, ifq); 436 break; 437 #endif 438 #ifdef ALTQ_PRIQ 439 case ALTQT_PRIQ: 440 error = priq_pfattach(a, ifq); 441 break; 442 #endif 443 #ifdef ALTQ_HFSC 444 case ALTQT_HFSC: 445 error = hfsc_pfattach(a, ifq); 446 break; 447 #endif 448 #ifdef ALTQ_FAIRQ 449 case ALTQT_FAIRQ: 450 error = fairq_pfattach(a, ifq); 451 break; 452 #endif 453 default: 454 error = ENXIO; 455 goto back; 456 } 457 458 /* if the state is running, enable altq */ 459 if (error == 0 && pfaltq_running && ifq->altq_type != ALTQT_NONE && 460 !ifq_is_enabled(ifq)) 461 error = altq_enable_locked(ifq); 462 463 /* if altq is already enabled, reset set tokenbucket regulator */ 464 if (error == 0 && ifq_is_enabled(ifq)) { 465 struct tb_profile tb; 466 467 tb.rate = a->ifbandwidth; 468 tb.depth = a->tbrsize; 469 error = tbr_set_locked(ifq, &tb); 470 } 471 back: 472 ifq_unlock_all(ifq); 473 ifnet_unlock(); 474 return (error); 475 } 476 477 /* 478 * detach a discipline from the interface. 479 * it is possible that the discipline was already overridden by another 480 * discipline. 481 */ 482 int 483 altq_pfdetach(struct pf_altq *a) 484 { 485 struct ifnet *ifp; 486 struct ifaltq *ifq; 487 int error = 0; 488 489 ifnet_lock(); 490 491 ifp = ifunit(a->ifname); 492 if (ifp == NULL) { 493 ifnet_unlock(); 494 return (EINVAL); 495 } 496 ifq = &ifp->if_snd; 497 498 /* if this discipline is no longer referenced, just return */ 499 if (a->altq_disc == NULL) { 500 ifnet_unlock(); 501 return (0); 502 } 503 504 ifq_lock_all(ifq); 505 506 if (a->altq_disc != ifq->altq_disc) 507 goto back; 508 509 if (ifq_is_enabled(ifq)) 510 error = altq_disable_locked(ifq); 511 if (error == 0) 512 error = altq_detach_locked(ifq); 513 514 back: 515 ifq_unlock_all(ifq); 516 ifnet_unlock(); 517 return (error); 518 } 519 520 /* 521 * add a discipline or a queue 522 */ 523 int 524 altq_add(struct pf_altq *a) 525 { 526 int error = 0; 527 528 if (a->qname[0] != 0) 529 return (altq_add_queue(a)); 530 531 if (machclk_freq == 0) 532 init_machclk(); 533 if (machclk_freq == 0) 534 panic("altq_add: no cpu clock"); 535 536 switch (a->scheduler) { 537 #ifdef ALTQ_CBQ 538 case ALTQT_CBQ: 539 error = cbq_add_altq(a); 540 break; 541 #endif 542 #ifdef ALTQ_PRIQ 543 case ALTQT_PRIQ: 544 error = priq_add_altq(a); 545 break; 546 #endif 547 #ifdef ALTQ_HFSC 548 case ALTQT_HFSC: 549 error = hfsc_add_altq(a); 550 break; 551 #endif 552 #ifdef ALTQ_FAIRQ 553 case ALTQT_FAIRQ: 554 error = fairq_add_altq(a); 555 break; 556 #endif 557 default: 558 error = ENXIO; 559 } 560 561 return (error); 562 } 563 564 /* 565 * remove a discipline or a queue 566 */ 567 int 568 altq_remove(struct pf_altq *a) 569 { 570 int error = 0; 571 572 if (a->qname[0] != 0) 573 return (altq_remove_queue(a)); 574 575 switch (a->scheduler) { 576 #ifdef ALTQ_CBQ 577 case ALTQT_CBQ: 578 error = cbq_remove_altq(a); 579 break; 580 #endif 581 #ifdef ALTQ_PRIQ 582 case ALTQT_PRIQ: 583 error = priq_remove_altq(a); 584 break; 585 #endif 586 #ifdef ALTQ_HFSC 587 case ALTQT_HFSC: 588 error = hfsc_remove_altq(a); 589 break; 590 #endif 591 #ifdef ALTQ_FAIRQ 592 case ALTQT_FAIRQ: 593 error = fairq_remove_altq(a); 594 break; 595 #endif 596 default: 597 error = ENXIO; 598 } 599 600 return (error); 601 } 602 603 /* 604 * add a queue to the discipline 605 */ 606 int 607 altq_add_queue(struct pf_altq *a) 608 { 609 int error = 0; 610 611 switch (a->scheduler) { 612 #ifdef ALTQ_CBQ 613 case ALTQT_CBQ: 614 error = cbq_add_queue(a); 615 break; 616 #endif 617 #ifdef ALTQ_PRIQ 618 case ALTQT_PRIQ: 619 error = priq_add_queue(a); 620 break; 621 #endif 622 #ifdef ALTQ_HFSC 623 case ALTQT_HFSC: 624 error = hfsc_add_queue(a); 625 break; 626 #endif 627 #ifdef ALTQ_FAIRQ 628 case ALTQT_FAIRQ: 629 error = fairq_add_queue(a); 630 break; 631 #endif 632 default: 633 error = ENXIO; 634 } 635 636 return (error); 637 } 638 639 /* 640 * remove a queue from the discipline 641 */ 642 int 643 altq_remove_queue(struct pf_altq *a) 644 { 645 int error = 0; 646 647 switch (a->scheduler) { 648 #ifdef ALTQ_CBQ 649 case ALTQT_CBQ: 650 error = cbq_remove_queue(a); 651 break; 652 #endif 653 #ifdef ALTQ_PRIQ 654 case ALTQT_PRIQ: 655 error = priq_remove_queue(a); 656 break; 657 #endif 658 #ifdef ALTQ_HFSC 659 case ALTQT_HFSC: 660 error = hfsc_remove_queue(a); 661 break; 662 #endif 663 #ifdef ALTQ_FAIRQ 664 case ALTQT_FAIRQ: 665 error = fairq_remove_queue(a); 666 break; 667 #endif 668 default: 669 error = ENXIO; 670 } 671 672 return (error); 673 } 674 675 /* 676 * get queue statistics 677 */ 678 int 679 altq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes) 680 { 681 int error = 0; 682 683 switch (a->scheduler) { 684 #ifdef ALTQ_CBQ 685 case ALTQT_CBQ: 686 error = cbq_getqstats(a, ubuf, nbytes); 687 break; 688 #endif 689 #ifdef ALTQ_PRIQ 690 case ALTQT_PRIQ: 691 error = priq_getqstats(a, ubuf, nbytes); 692 break; 693 #endif 694 #ifdef ALTQ_HFSC 695 case ALTQT_HFSC: 696 error = hfsc_getqstats(a, ubuf, nbytes); 697 break; 698 #endif 699 #ifdef ALTQ_FAIRQ 700 case ALTQT_FAIRQ: 701 error = fairq_getqstats(a, ubuf, nbytes); 702 break; 703 #endif 704 default: 705 error = ENXIO; 706 } 707 708 return (error); 709 } 710 711 /* 712 * read and write diffserv field in IPv4 or IPv6 header 713 */ 714 uint8_t 715 read_dsfield(struct mbuf *m, struct altq_pktattr *pktattr) 716 { 717 struct mbuf *m0; 718 uint8_t ds_field = 0; 719 720 if (pktattr == NULL || 721 (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6)) 722 return ((uint8_t)0); 723 724 /* verify that pattr_hdr is within the mbuf data */ 725 for (m0 = m; m0 != NULL; m0 = m0->m_next) { 726 if ((pktattr->pattr_hdr >= m0->m_data) && 727 (pktattr->pattr_hdr < m0->m_data + m0->m_len)) 728 break; 729 } 730 if (m0 == NULL) { 731 /* ick, pattr_hdr is stale */ 732 pktattr->pattr_af = AF_UNSPEC; 733 #ifdef ALTQ_DEBUG 734 kprintf("read_dsfield: can't locate header!\n"); 735 #endif 736 return ((uint8_t)0); 737 } 738 739 if (pktattr->pattr_af == AF_INET) { 740 struct ip *ip = (struct ip *)pktattr->pattr_hdr; 741 742 if (ip->ip_v != 4) 743 return ((uint8_t)0); /* version mismatch! */ 744 ds_field = ip->ip_tos; 745 } 746 #ifdef INET6 747 else if (pktattr->pattr_af == AF_INET6) { 748 struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr; 749 uint32_t flowlabel; 750 751 flowlabel = ntohl(ip6->ip6_flow); 752 if ((flowlabel >> 28) != 6) 753 return ((uint8_t)0); /* version mismatch! */ 754 ds_field = (flowlabel >> 20) & 0xff; 755 } 756 #endif 757 return (ds_field); 758 } 759 760 void 761 write_dsfield(struct mbuf *m, struct altq_pktattr *pktattr, uint8_t dsfield) 762 { 763 struct mbuf *m0; 764 765 if (pktattr == NULL || 766 (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6)) 767 return; 768 769 /* verify that pattr_hdr is within the mbuf data */ 770 for (m0 = m; m0 != NULL; m0 = m0->m_next) { 771 if ((pktattr->pattr_hdr >= m0->m_data) && 772 (pktattr->pattr_hdr < m0->m_data + m0->m_len)) 773 break; 774 } 775 if (m0 == NULL) { 776 /* ick, pattr_hdr is stale */ 777 pktattr->pattr_af = AF_UNSPEC; 778 #ifdef ALTQ_DEBUG 779 kprintf("write_dsfield: can't locate header!\n"); 780 #endif 781 return; 782 } 783 784 if (pktattr->pattr_af == AF_INET) { 785 struct ip *ip = (struct ip *)pktattr->pattr_hdr; 786 uint8_t old; 787 int32_t sum; 788 789 if (ip->ip_v != 4) 790 return; /* version mismatch! */ 791 old = ip->ip_tos; 792 dsfield |= old & 3; /* leave CU bits */ 793 if (old == dsfield) 794 return; 795 ip->ip_tos = dsfield; 796 /* 797 * update checksum (from RFC1624) 798 * HC' = ~(~HC + ~m + m') 799 */ 800 sum = ~ntohs(ip->ip_sum) & 0xffff; 801 sum += 0xff00 + (~old & 0xff) + dsfield; 802 sum = (sum >> 16) + (sum & 0xffff); 803 sum += (sum >> 16); /* add carry */ 804 805 ip->ip_sum = htons(~sum & 0xffff); 806 } 807 #ifdef INET6 808 else if (pktattr->pattr_af == AF_INET6) { 809 struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr; 810 uint32_t flowlabel; 811 812 flowlabel = ntohl(ip6->ip6_flow); 813 if ((flowlabel >> 28) != 6) 814 return; /* version mismatch! */ 815 flowlabel = (flowlabel & 0xf03fffff) | (dsfield << 20); 816 ip6->ip6_flow = htonl(flowlabel); 817 } 818 #endif 819 } 820 821 /* 822 * high resolution clock support taking advantage of a machine dependent 823 * high resolution time counter (e.g., timestamp counter of intel pentium). 824 * we assume 825 * - 64-bit-long monotonically-increasing counter 826 * - frequency range is 100M-4GHz (CPU speed) 827 */ 828 /* if pcc is not available or disabled, emulate 256MHz using microtime() */ 829 #define MACHCLK_SHIFT 8 830 831 static int machclk_usepcc; 832 uint64_t machclk_freq = 0; 833 uint32_t machclk_per_tick = 0; 834 835 void 836 init_machclk(void) 837 { 838 callout_init_mp(&tbr_callout); 839 netmsg_init(&tbr_timeout_netmsg, NULL, &netisr_adone_rport, 840 MSGF_PRIORITY, tbr_timeout_dispatch); 841 842 #ifdef ALTQ_NOPCC 843 machclk_usepcc = 0; 844 #else 845 machclk_usepcc = 1; 846 #endif 847 848 #if defined(__i386__) || defined(__x86_64__) 849 if (!tsc_mpsync) 850 machclk_usepcc = 0; 851 #else 852 machclk_usepcc = 0; 853 #endif 854 855 if (!machclk_usepcc) { 856 /* emulate 256MHz using microtime() */ 857 machclk_freq = 1000000LLU << MACHCLK_SHIFT; 858 machclk_per_tick = machclk_freq / hz; 859 #ifdef ALTQ_DEBUG 860 kprintf("altq: emulate %juHz cpu clock\n", 861 (uintmax_t)machclk_freq); 862 #endif 863 return; 864 } 865 866 /* 867 * If the clock frequency (of Pentium TSC) is accessible, 868 * just use it. 869 */ 870 #ifdef _RDTSC_SUPPORTED_ 871 if (tsc_present) 872 machclk_freq = (uint64_t)tsc_frequency; 873 #endif 874 875 /* 876 * If we don't know the clock frequency, measure it. 877 */ 878 if (machclk_freq == 0) { 879 static int wait; 880 struct timeval tv_start, tv_end; 881 uint64_t start, end, diff; 882 int timo; 883 884 microtime(&tv_start); 885 start = read_machclk(); 886 timo = hz; /* 1 sec */ 887 tsleep(&wait, PCATCH, "init_machclk", timo); 888 microtime(&tv_end); 889 end = read_machclk(); 890 diff = (uint64_t)(tv_end.tv_sec - tv_start.tv_sec) * 1000000 891 + tv_end.tv_usec - tv_start.tv_usec; 892 if (diff != 0) 893 machclk_freq = (end - start) * 1000000 / diff; 894 } 895 896 machclk_per_tick = machclk_freq / hz; 897 898 #ifdef ALTQ_DEBUG 899 kprintf("altq: CPU clock: %juHz\n", (uintmax_t)machclk_freq); 900 #endif 901 } 902 903 uint64_t 904 read_machclk(void) 905 { 906 uint64_t val; 907 908 if (machclk_usepcc) { 909 #ifdef _RDTSC_SUPPORTED_ 910 val = rdtsc(); 911 #else 912 panic("read_machclk"); 913 #endif 914 } else { 915 struct timeval tv; 916 917 microtime(&tv); 918 val = (((uint64_t)(tv.tv_sec - boottime.tv_sec) * 1000000 919 + tv.tv_usec) << MACHCLK_SHIFT); 920 } 921 return (val); 922 } 923