altq_rio.c revision 1.12
1 /* $NetBSD: altq_rio.c,v 1.12 2006/07/21 16:48:46 ad Exp $ */ 2 /* $KAME: altq_rio.c,v 1.8 2000/12/14 08:12:46 thorpej Exp $ */ 3 4 /* 5 * Copyright (C) 1998-2000 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 * Copyright (c) 1990-1994 Regents of the University of California. 31 * All rights reserved. 32 * 33 * Redistribution and use in source and binary forms, with or without 34 * modification, are permitted provided that the following conditions 35 * are met: 36 * 1. Redistributions of source code must retain the above copyright 37 * notice, this list of conditions and the following disclaimer. 38 * 2. Redistributions in binary form must reproduce the above copyright 39 * notice, this list of conditions and the following disclaimer in the 40 * documentation and/or other materials provided with the distribution. 41 * 3. All advertising materials mentioning features or use of this software 42 * must display the following acknowledgement: 43 * This product includes software developed by the Computer Systems 44 * Engineering Group at Lawrence Berkeley Laboratory. 45 * 4. Neither the name of the University nor of the Laboratory may be used 46 * to endorse or promote products derived from this software without 47 * specific prior written permission. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 */ 61 62 #include <sys/cdefs.h> 63 __KERNEL_RCSID(0, "$NetBSD: altq_rio.c,v 1.12 2006/07/21 16:48:46 ad Exp $"); 64 65 #if defined(__FreeBSD__) || defined(__NetBSD__) 66 #include "opt_altq.h" 67 #if (__FreeBSD__ != 2) 68 #include "opt_inet.h" 69 #ifdef __FreeBSD__ 70 #include "opt_inet6.h" 71 #endif 72 #endif 73 #endif /* __FreeBSD__ || __NetBSD__ */ 74 #ifdef ALTQ_RIO /* rio is enabled by ALTQ_RIO option in opt_altq.h */ 75 76 #include <sys/param.h> 77 #include <sys/malloc.h> 78 #include <sys/mbuf.h> 79 #include <sys/socket.h> 80 #include <sys/sockio.h> 81 #include <sys/systm.h> 82 #include <sys/proc.h> 83 #include <sys/errno.h> 84 #include <sys/kernel.h> 85 #include <sys/kauth.h> 86 87 #include <net/if.h> 88 #include <net/if_types.h> 89 90 #include <netinet/in.h> 91 #include <netinet/in_systm.h> 92 #include <netinet/ip.h> 93 #ifdef INET6 94 #include <netinet/ip6.h> 95 #endif 96 97 #include <altq/altq.h> 98 #include <altq/altq_conf.h> 99 #include <altq/altq_cdnr.h> 100 #include <altq/altq_red.h> 101 #include <altq/altq_rio.h> 102 103 /* 104 * RIO: RED with IN/OUT bit 105 * described in 106 * "Explicit Allocation of Best Effort Packet Delivery Service" 107 * David D. Clark and Wenjia Fang, MIT Lab for Computer Science 108 * http://diffserv.lcs.mit.edu/Papers/exp-alloc-ddc-wf.{ps,pdf} 109 * 110 * this implementation is extended to support more than 2 drop precedence 111 * values as described in RFC2597 (Assured Forwarding PHB Group). 112 * 113 */ 114 /* 115 * AF DS (differentiated service) codepoints. 116 * (classes can be mapped to CBQ or H-FSC classes.) 117 * 118 * 0 1 2 3 4 5 6 7 119 * +---+---+---+---+---+---+---+---+ 120 * | CLASS |DropPre| 0 | CU | 121 * +---+---+---+---+---+---+---+---+ 122 * 123 * class 1: 001 124 * class 2: 010 125 * class 3: 011 126 * class 4: 100 127 * 128 * low drop prec: 01 129 * medium drop prec: 10 130 * high drop prec: 01 131 */ 132 133 /* normal red parameters */ 134 #define W_WEIGHT 512 /* inverse of weight of EWMA (511/512) */ 135 /* q_weight = 0.00195 */ 136 137 /* red parameters for a slow link */ 138 #define W_WEIGHT_1 128 /* inverse of weight of EWMA (127/128) */ 139 /* q_weight = 0.0078125 */ 140 141 /* red parameters for a very slow link (e.g., dialup) */ 142 #define W_WEIGHT_2 64 /* inverse of weight of EWMA (63/64) */ 143 /* q_weight = 0.015625 */ 144 145 /* fixed-point uses 12-bit decimal places */ 146 #define FP_SHIFT 12 /* fixed-point shift */ 147 148 /* red parameters for drop probability */ 149 #define INV_P_MAX 10 /* inverse of max drop probability */ 150 #define TH_MIN 5 /* min threshold */ 151 #define TH_MAX 15 /* max threshold */ 152 153 #define RIO_LIMIT 60 /* default max queue length */ 154 155 #define TV_DELTA(a, b, delta) { \ 156 register int xxs; \ 157 \ 158 delta = (a)->tv_usec - (b)->tv_usec; \ 159 if ((xxs = (a)->tv_sec - (b)->tv_sec) != 0) { \ 160 if (xxs < 0) { \ 161 printf("rm_class: bogus time values"); \ 162 delta = 60000000; \ 163 } else if (xxs > 4) { \ 164 if (xxs > 60) \ 165 delta = 60000000; \ 166 else \ 167 delta += xxs * 1000000; \ 168 } else while (xxs > 0) { \ 169 delta += 1000000; \ 170 xxs--; \ 171 } \ 172 } \ 173 } 174 175 /* rio_list keeps all rio_queue_t's allocated. */ 176 static rio_queue_t *rio_list = NULL; 177 /* default rio parameter values */ 178 static struct redparams default_rio_params[RIO_NDROPPREC] = { 179 /* th_min, th_max, inv_pmax */ 180 { TH_MAX * 2 + TH_MIN, TH_MAX * 3, INV_P_MAX }, /* low drop precedence */ 181 { TH_MAX + TH_MIN, TH_MAX * 2, INV_P_MAX }, /* medium drop precedence */ 182 { TH_MIN, TH_MAX, INV_P_MAX } /* high drop precedence */ 183 }; 184 185 /* internal function prototypes */ 186 static int rio_enqueue __P((struct ifaltq *, struct mbuf *, 187 struct altq_pktattr *)); 188 static struct mbuf *rio_dequeue __P((struct ifaltq *, int)); 189 static int rio_request __P((struct ifaltq *, int, void *)); 190 static int rio_detach __P((rio_queue_t *)); 191 static int dscp2index __P((u_int8_t)); 192 193 /* 194 * rio device interface 195 */ 196 altqdev_decl(rio); 197 198 int 199 rioopen(dev, flag, fmt, l) 200 dev_t dev; 201 int flag, fmt; 202 struct lwp *l; 203 { 204 /* everything will be done when the queueing scheme is attached. */ 205 return 0; 206 } 207 208 int 209 rioclose(dev, flag, fmt, l) 210 dev_t dev; 211 int flag, fmt; 212 struct lwp *l; 213 { 214 rio_queue_t *rqp; 215 int err, error = 0; 216 217 while ((rqp = rio_list) != NULL) { 218 /* destroy all */ 219 err = rio_detach(rqp); 220 if (err != 0 && error == 0) 221 error = err; 222 } 223 224 return error; 225 } 226 227 int 228 rioioctl(dev, cmd, addr, flag, l) 229 dev_t dev; 230 ioctlcmd_t cmd; 231 caddr_t addr; 232 int flag; 233 struct lwp *l; 234 { 235 rio_queue_t *rqp; 236 struct rio_interface *ifacep; 237 struct ifnet *ifp; 238 int error = 0; 239 240 /* check super-user privilege */ 241 switch (cmd) { 242 case RIO_GETSTATS: 243 break; 244 default: 245 #if (__FreeBSD_version > 400000) 246 if ((error = suser(p)) != 0) 247 return (error); 248 #else 249 if ((error = kauth_authorize_generic(l->l_cred, 250 KAUTH_GENERIC_ISSUSER, &l->l_acflag)) != 0) 251 return (error); 252 #endif 253 break; 254 } 255 256 switch (cmd) { 257 258 case RIO_ENABLE: 259 ifacep = (struct rio_interface *)addr; 260 if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) { 261 error = EBADF; 262 break; 263 } 264 error = altq_enable(rqp->rq_ifq); 265 break; 266 267 case RIO_DISABLE: 268 ifacep = (struct rio_interface *)addr; 269 if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) { 270 error = EBADF; 271 break; 272 } 273 error = altq_disable(rqp->rq_ifq); 274 break; 275 276 case RIO_IF_ATTACH: 277 ifp = ifunit(((struct rio_interface *)addr)->rio_ifname); 278 if (ifp == NULL) { 279 error = ENXIO; 280 break; 281 } 282 283 /* allocate and initialize rio_queue_t */ 284 rqp = malloc(sizeof(rio_queue_t), M_DEVBUF, M_WAITOK|M_ZERO); 285 if (rqp == NULL) { 286 error = ENOMEM; 287 break; 288 } 289 290 rqp->rq_q = malloc(sizeof(class_queue_t), M_DEVBUF, 291 M_WAITOK|M_ZERO); 292 if (rqp->rq_q == NULL) { 293 free(rqp, M_DEVBUF); 294 error = ENOMEM; 295 break; 296 } 297 298 rqp->rq_rio = rio_alloc(0, NULL, 0, 0); 299 if (rqp->rq_rio == NULL) { 300 free(rqp->rq_q, M_DEVBUF); 301 free(rqp, M_DEVBUF); 302 error = ENOMEM; 303 break; 304 } 305 306 rqp->rq_ifq = &ifp->if_snd; 307 qtail(rqp->rq_q) = NULL; 308 qlen(rqp->rq_q) = 0; 309 qlimit(rqp->rq_q) = RIO_LIMIT; 310 qtype(rqp->rq_q) = Q_RIO; 311 312 /* 313 * set RIO to this ifnet structure. 314 */ 315 error = altq_attach(rqp->rq_ifq, ALTQT_RIO, rqp, 316 rio_enqueue, rio_dequeue, rio_request, 317 NULL, NULL); 318 if (error) { 319 rio_destroy(rqp->rq_rio); 320 free(rqp->rq_q, M_DEVBUF); 321 free(rqp, M_DEVBUF); 322 break; 323 } 324 325 /* add this state to the rio list */ 326 rqp->rq_next = rio_list; 327 rio_list = rqp; 328 break; 329 330 case RIO_IF_DETACH: 331 ifacep = (struct rio_interface *)addr; 332 if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) { 333 error = EBADF; 334 break; 335 } 336 error = rio_detach(rqp); 337 break; 338 339 case RIO_GETSTATS: 340 do { 341 struct rio_stats *q_stats; 342 rio_t *rp; 343 int i; 344 345 q_stats = (struct rio_stats *)addr; 346 if ((rqp = altq_lookup(q_stats->iface.rio_ifname, 347 ALTQT_RIO)) == NULL) { 348 error = EBADF; 349 break; 350 } 351 352 rp = rqp->rq_rio; 353 354 q_stats->q_limit = qlimit(rqp->rq_q); 355 q_stats->weight = rp->rio_weight; 356 q_stats->flags = rp->rio_flags; 357 358 for (i = 0; i < RIO_NDROPPREC; i++) { 359 q_stats->q_len[i] = rp->rio_precstate[i].qlen; 360 (void)memcpy(&q_stats->q_stats[i], 361 &rp->q_stats[i], sizeof(struct redstats)); 362 q_stats->q_stats[i].q_avg = 363 rp->rio_precstate[i].avg >> rp->rio_wshift; 364 365 q_stats->q_params[i].inv_pmax 366 = rp->rio_precstate[i].inv_pmax; 367 q_stats->q_params[i].th_min 368 = rp->rio_precstate[i].th_min; 369 q_stats->q_params[i].th_max 370 = rp->rio_precstate[i].th_max; 371 } 372 } while (0); 373 break; 374 375 case RIO_CONFIG: 376 do { 377 struct rio_conf *fc; 378 rio_t *new; 379 int s, limit, i; 380 381 fc = (struct rio_conf *)addr; 382 if ((rqp = altq_lookup(fc->iface.rio_ifname, 383 ALTQT_RIO)) == NULL) { 384 error = EBADF; 385 break; 386 } 387 388 new = rio_alloc(fc->rio_weight, &fc->q_params[0], 389 fc->rio_flags, fc->rio_pkttime); 390 if (new == NULL) { 391 error = ENOMEM; 392 break; 393 } 394 395 s = splnet(); 396 _flushq(rqp->rq_q); 397 limit = fc->rio_limit; 398 if (limit < fc->q_params[RIO_NDROPPREC-1].th_max) 399 limit = fc->q_params[RIO_NDROPPREC-1].th_max; 400 qlimit(rqp->rq_q) = limit; 401 402 rio_destroy(rqp->rq_rio); 403 rqp->rq_rio = new; 404 405 splx(s); 406 407 /* write back new values */ 408 fc->rio_limit = limit; 409 for (i = 0; i < RIO_NDROPPREC; i++) { 410 fc->q_params[i].inv_pmax = 411 rqp->rq_rio->rio_precstate[i].inv_pmax; 412 fc->q_params[i].th_min = 413 rqp->rq_rio->rio_precstate[i].th_min; 414 fc->q_params[i].th_max = 415 rqp->rq_rio->rio_precstate[i].th_max; 416 } 417 } while (0); 418 break; 419 420 case RIO_SETDEFAULTS: 421 do { 422 struct redparams *rp; 423 int i; 424 425 rp = (struct redparams *)addr; 426 for (i = 0; i < RIO_NDROPPREC; i++) 427 default_rio_params[i] = rp[i]; 428 } while (0); 429 break; 430 431 default: 432 error = EINVAL; 433 break; 434 } 435 436 return error; 437 } 438 439 static int 440 rio_detach(rqp) 441 rio_queue_t *rqp; 442 { 443 rio_queue_t *tmp; 444 int error = 0; 445 446 if (ALTQ_IS_ENABLED(rqp->rq_ifq)) 447 altq_disable(rqp->rq_ifq); 448 449 if ((error = altq_detach(rqp->rq_ifq))) 450 return (error); 451 452 if (rio_list == rqp) 453 rio_list = rqp->rq_next; 454 else { 455 for (tmp = rio_list; tmp != NULL; tmp = tmp->rq_next) 456 if (tmp->rq_next == rqp) { 457 tmp->rq_next = rqp->rq_next; 458 break; 459 } 460 if (tmp == NULL) 461 printf("rio_detach: no state found in rio_list!\n"); 462 } 463 464 rio_destroy(rqp->rq_rio); 465 free(rqp->rq_q, M_DEVBUF); 466 free(rqp, M_DEVBUF); 467 return (error); 468 } 469 470 /* 471 * rio support routines 472 */ 473 static int 474 rio_request(ifq, req, arg) 475 struct ifaltq *ifq; 476 int req; 477 void *arg; 478 { 479 rio_queue_t *rqp = (rio_queue_t *)ifq->altq_disc; 480 481 switch (req) { 482 case ALTRQ_PURGE: 483 _flushq(rqp->rq_q); 484 if (ALTQ_IS_ENABLED(ifq)) 485 ifq->ifq_len = 0; 486 break; 487 } 488 return (0); 489 } 490 491 492 rio_t * 493 rio_alloc(weight, params, flags, pkttime) 494 int weight; 495 struct redparams *params; 496 int flags, pkttime; 497 { 498 rio_t *rp; 499 int w, i; 500 int npkts_per_sec; 501 502 rp = malloc(sizeof(rio_t), M_DEVBUF, M_WAITOK|M_ZERO); 503 if (rp == NULL) 504 return (NULL); 505 506 rp->rio_flags = flags; 507 if (pkttime == 0) 508 /* default packet time: 1000 bytes / 10Mbps * 8 * 1000000 */ 509 rp->rio_pkttime = 800; 510 else 511 rp->rio_pkttime = pkttime; 512 513 if (weight != 0) 514 rp->rio_weight = weight; 515 else { 516 /* use derfault */ 517 rp->rio_weight = W_WEIGHT; 518 519 /* when the link is very slow, adjust red parameters */ 520 npkts_per_sec = 1000000 / rp->rio_pkttime; 521 if (npkts_per_sec < 50) { 522 /* up to about 400Kbps */ 523 rp->rio_weight = W_WEIGHT_2; 524 } else if (npkts_per_sec < 300) { 525 /* up to about 2.4Mbps */ 526 rp->rio_weight = W_WEIGHT_1; 527 } 528 } 529 530 /* calculate wshift. weight must be power of 2 */ 531 w = rp->rio_weight; 532 for (i = 0; w > 1; i++) 533 w = w >> 1; 534 rp->rio_wshift = i; 535 w = 1 << rp->rio_wshift; 536 if (w != rp->rio_weight) { 537 printf("invalid weight value %d for red! use %d\n", 538 rp->rio_weight, w); 539 rp->rio_weight = w; 540 } 541 542 /* allocate weight table */ 543 rp->rio_wtab = wtab_alloc(rp->rio_weight); 544 545 for (i = 0; i < RIO_NDROPPREC; i++) { 546 struct dropprec_state *prec = &rp->rio_precstate[i]; 547 548 prec->avg = 0; 549 prec->idle = 1; 550 551 if (params == NULL || params[i].inv_pmax == 0) 552 prec->inv_pmax = default_rio_params[i].inv_pmax; 553 else 554 prec->inv_pmax = params[i].inv_pmax; 555 if (params == NULL || params[i].th_min == 0) 556 prec->th_min = default_rio_params[i].th_min; 557 else 558 prec->th_min = params[i].th_min; 559 if (params == NULL || params[i].th_max == 0) 560 prec->th_max = default_rio_params[i].th_max; 561 else 562 prec->th_max = params[i].th_max; 563 564 /* 565 * th_min_s and th_max_s are scaled versions of th_min 566 * and th_max to be compared with avg. 567 */ 568 prec->th_min_s = prec->th_min << (rp->rio_wshift + FP_SHIFT); 569 prec->th_max_s = prec->th_max << (rp->rio_wshift + FP_SHIFT); 570 571 /* 572 * precompute probability denominator 573 * probd = (2 * (TH_MAX-TH_MIN) / pmax) in fixed-point 574 */ 575 prec->probd = (2 * (prec->th_max - prec->th_min) 576 * prec->inv_pmax) << FP_SHIFT; 577 578 microtime(&prec->last); 579 } 580 581 return (rp); 582 } 583 584 void 585 rio_destroy(rp) 586 rio_t *rp; 587 { 588 wtab_destroy(rp->rio_wtab); 589 free(rp, M_DEVBUF); 590 } 591 592 void 593 rio_getstats(rp, sp) 594 rio_t *rp; 595 struct redstats *sp; 596 { 597 int i; 598 599 for (i = 0; i < RIO_NDROPPREC; i++) { 600 (void)memcpy(sp, &rp->q_stats[i], sizeof(struct redstats)); 601 sp->q_avg = rp->rio_precstate[i].avg >> rp->rio_wshift; 602 sp++; 603 } 604 } 605 606 /* 607 * enqueue routine: 608 * 609 * returns: 0 when successfully queued. 610 * ENOBUFS when drop occurs. 611 */ 612 static int 613 rio_enqueue(ifq, m, pktattr) 614 struct ifaltq *ifq; 615 struct mbuf *m; 616 struct altq_pktattr *pktattr; 617 { 618 rio_queue_t *rqp = (rio_queue_t *)ifq->altq_disc; 619 int error = 0; 620 621 if (rio_addq(rqp->rq_rio, rqp->rq_q, m, pktattr) == 0) 622 ifq->ifq_len++; 623 else 624 error = ENOBUFS; 625 return error; 626 } 627 628 #if (RIO_NDROPPREC == 3) 629 /* 630 * internally, a drop precedence value is converted to an index 631 * starting from 0. 632 */ 633 static int 634 dscp2index(u_int8_t dscp) 635 { 636 int dpindex = dscp & AF_DROPPRECMASK; 637 638 if (dpindex == 0) 639 return (0); 640 return ((dpindex >> 3) - 1); 641 } 642 #endif 643 644 #if 1 645 /* 646 * kludge: when a packet is dequeued, we need to know its drop precedence 647 * in order to keep the queue length of each drop precedence. 648 * use m_pkthdr.rcvif to pass this info. 649 */ 650 #define RIOM_SET_PRECINDEX(m, idx) \ 651 do { (m)->m_pkthdr.rcvif = (struct ifnet *)((long)(idx)); } while (0) 652 #define RIOM_GET_PRECINDEX(m) \ 653 ({ long idx; idx = (long)((m)->m_pkthdr.rcvif); \ 654 (m)->m_pkthdr.rcvif = NULL; idx; }) 655 #endif 656 657 int 658 rio_addq(rp, q, m, pktattr) 659 rio_t *rp; 660 class_queue_t *q; 661 struct mbuf *m; 662 struct altq_pktattr *pktattr; 663 { 664 int avg, droptype; 665 u_int8_t dsfield, odsfield; 666 int dpindex, i, n, t; 667 struct timeval now; 668 struct dropprec_state *prec; 669 670 dsfield = odsfield = read_dsfield(m, pktattr); 671 dpindex = dscp2index(dsfield); 672 673 /* 674 * update avg of the precedence states whose drop precedence 675 * is larger than or equal to the drop precedence of the packet 676 */ 677 now.tv_sec = 0; 678 for (i = dpindex; i < RIO_NDROPPREC; i++) { 679 prec = &rp->rio_precstate[i]; 680 avg = prec->avg; 681 if (prec->idle) { 682 prec->idle = 0; 683 if (now.tv_sec == 0) 684 microtime(&now); 685 t = (now.tv_sec - prec->last.tv_sec); 686 if (t > 60) 687 avg = 0; 688 else { 689 t = t * 1000000 + 690 (now.tv_usec - prec->last.tv_usec); 691 n = t / rp->rio_pkttime; 692 /* calculate (avg = (1 - Wq)^n * avg) */ 693 if (n > 0) 694 avg = (avg >> FP_SHIFT) * 695 pow_w(rp->rio_wtab, n); 696 } 697 } 698 699 /* run estimator. (avg is scaled by WEIGHT in fixed-point) */ 700 avg += (prec->qlen << FP_SHIFT) - (avg >> rp->rio_wshift); 701 prec->avg = avg; /* save the new value */ 702 /* 703 * count keeps a tally of arriving traffic that has not 704 * been dropped. 705 */ 706 prec->count++; 707 } 708 709 prec = &rp->rio_precstate[dpindex]; 710 avg = prec->avg; 711 712 /* see if we drop early */ 713 droptype = DTYPE_NODROP; 714 if (avg >= prec->th_min_s && prec->qlen > 1) { 715 if (avg >= prec->th_max_s) { 716 /* avg >= th_max: forced drop */ 717 droptype = DTYPE_FORCED; 718 } else if (prec->old == 0) { 719 /* first exceeds th_min */ 720 prec->count = 1; 721 prec->old = 1; 722 } else if (drop_early((avg - prec->th_min_s) >> rp->rio_wshift, 723 prec->probd, prec->count)) { 724 /* unforced drop by red */ 725 droptype = DTYPE_EARLY; 726 } 727 } else { 728 /* avg < th_min */ 729 prec->old = 0; 730 } 731 732 /* 733 * if the queue length hits the hard limit, it's a forced drop. 734 */ 735 if (droptype == DTYPE_NODROP && qlen(q) >= qlimit(q)) 736 droptype = DTYPE_FORCED; 737 738 if (droptype != DTYPE_NODROP) { 739 /* always drop incoming packet (as opposed to randomdrop) */ 740 for (i = dpindex; i < RIO_NDROPPREC; i++) 741 rp->rio_precstate[i].count = 0; 742 #ifdef RIO_STATS 743 if (droptype == DTYPE_EARLY) 744 rp->q_stats[dpindex].drop_unforced++; 745 else 746 rp->q_stats[dpindex].drop_forced++; 747 PKTCNTR_ADD(&rp->q_stats[dpindex].drop_cnt, m_pktlen(m)); 748 #endif 749 m_freem(m); 750 return (-1); 751 } 752 753 for (i = dpindex; i < RIO_NDROPPREC; i++) 754 rp->rio_precstate[i].qlen++; 755 756 /* save drop precedence index in mbuf hdr */ 757 RIOM_SET_PRECINDEX(m, dpindex); 758 759 if (rp->rio_flags & RIOF_CLEARDSCP) 760 dsfield &= ~DSCP_MASK; 761 762 if (dsfield != odsfield) 763 write_dsfield(m, pktattr, dsfield); 764 765 _addq(q, m); 766 767 #ifdef RIO_STATS 768 PKTCNTR_ADD(&rp->q_stats[dpindex].xmit_cnt, m_pktlen(m)); 769 #endif 770 return (0); 771 } 772 773 /* 774 * dequeue routine: 775 * must be called in splnet. 776 * 777 * returns: mbuf dequeued. 778 * NULL when no packet is available in the queue. 779 */ 780 781 static struct mbuf * 782 rio_dequeue(ifq, op) 783 struct ifaltq *ifq; 784 int op; 785 { 786 rio_queue_t *rqp = (rio_queue_t *)ifq->altq_disc; 787 struct mbuf *m = NULL; 788 789 if (op == ALTDQ_POLL) 790 return qhead(rqp->rq_q); 791 792 m = rio_getq(rqp->rq_rio, rqp->rq_q); 793 if (m != NULL) 794 ifq->ifq_len--; 795 return m; 796 } 797 798 struct mbuf * 799 rio_getq(rp, q) 800 rio_t *rp; 801 class_queue_t *q; 802 { 803 struct mbuf *m; 804 int dpindex, i; 805 806 if ((m = _getq(q)) == NULL) 807 return NULL; 808 809 dpindex = RIOM_GET_PRECINDEX(m); 810 for (i = dpindex; i < RIO_NDROPPREC; i++) { 811 if (--rp->rio_precstate[i].qlen == 0) { 812 if (rp->rio_precstate[i].idle == 0) { 813 rp->rio_precstate[i].idle = 1; 814 microtime(&rp->rio_precstate[i].last); 815 } 816 } 817 } 818 return (m); 819 } 820 821 #ifdef KLD_MODULE 822 823 static struct altqsw rio_sw = 824 {"rio", rioopen, rioclose, rioioctl}; 825 826 ALTQ_MODULE(altq_rio, ALTQT_RIO, &rio_sw); 827 828 #endif /* KLD_MODULE */ 829 830 #endif /* ALTQ_RIO */ 831
Indexes created Mon Sep 29 21:09:56 GMT 2025