altq_rio.c revision 1.14
1 /* $NetBSD: altq_rio.c,v 1.14 2006/10/12 19:59:08 peter Exp $ */ 2 /* $KAME: altq_rio.c,v 1.19 2005/04/13 03:44:25 suz Exp $ */ 3 4 /* 5 * Copyright (C) 1998-2003 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 * 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 1.14 2006/10/12 19:59:08 peter Exp $"); 64 65 #ifdef _KERNEL_OPT 66 #include "opt_altq.h" 67 #include "opt_inet.h" 68 #endif 69 70 #ifdef ALTQ_RIO /* rio is enabled by ALTQ_RIO option in opt_altq.h */ 71 72 #include <sys/param.h> 73 #include <sys/malloc.h> 74 #include <sys/mbuf.h> 75 #include <sys/socket.h> 76 #include <sys/systm.h> 77 #include <sys/errno.h> 78 #include <sys/kauth.h> 79 #if 1 /* ALTQ3_COMPAT */ 80 #include <sys/proc.h> 81 #include <sys/sockio.h> 82 #include <sys/kernel.h> 83 #endif 84 #include <sys/kauth.h> 85 86 #include <net/if.h> 87 88 #include <netinet/in.h> 89 #include <netinet/in_systm.h> 90 #include <netinet/ip.h> 91 #ifdef INET6 92 #include <netinet/ip6.h> 93 #endif 94 95 #include <net/pfvar.h> 96 #include <altq/altq.h> 97 #include <altq/altq_cdnr.h> 98 #include <altq/altq_red.h> 99 #include <altq/altq_rio.h> 100 #ifdef ALTQ3_COMPAT 101 #include <altq/altq_conf.h> 102 #endif 103 104 /* 105 * RIO: RED with IN/OUT bit 106 * described in 107 * "Explicit Allocation of Best Effort Packet Delivery Service" 108 * David D. Clark and Wenjia Fang, MIT Lab for Computer Science 109 * http://diffserv.lcs.mit.edu/Papers/exp-alloc-ddc-wf.{ps,pdf} 110 * 111 * this implementation is extended to support more than 2 drop precedence 112 * values as described in RFC2597 (Assured Forwarding PHB Group). 113 * 114 */ 115 /* 116 * AF DS (differentiated service) codepoints. 117 * (classes can be mapped to CBQ or H-FSC classes.) 118 * 119 * 0 1 2 3 4 5 6 7 120 * +---+---+---+---+---+---+---+---+ 121 * | CLASS |DropPre| 0 | CU | 122 * +---+---+---+---+---+---+---+---+ 123 * 124 * class 1: 001 125 * class 2: 010 126 * class 3: 011 127 * class 4: 100 128 * 129 * low drop prec: 01 130 * medium drop prec: 10 131 * high drop prec: 11 132 */ 133 134 /* normal red parameters */ 135 #define W_WEIGHT 512 /* inverse of weight of EWMA (511/512) */ 136 /* q_weight = 0.00195 */ 137 138 /* red parameters for a slow link */ 139 #define W_WEIGHT_1 128 /* inverse of weight of EWMA (127/128) */ 140 /* q_weight = 0.0078125 */ 141 142 /* red parameters for a very slow link (e.g., dialup) */ 143 #define W_WEIGHT_2 64 /* inverse of weight of EWMA (63/64) */ 144 /* q_weight = 0.015625 */ 145 146 /* fixed-point uses 12-bit decimal places */ 147 #define FP_SHIFT 12 /* fixed-point shift */ 148 149 /* red parameters for drop probability */ 150 #define INV_P_MAX 10 /* inverse of max drop probability */ 151 #define TH_MIN 5 /* min threshold */ 152 #define TH_MAX 15 /* max threshold */ 153 154 #define RIO_LIMIT 60 /* default max queue lenght */ 155 #define RIO_STATS /* collect statistics */ 156 157 #define TV_DELTA(a, b, delta) { \ 158 register int xxs; \ 159 \ 160 delta = (a)->tv_usec - (b)->tv_usec; \ 161 if ((xxs = (a)->tv_sec - (b)->tv_sec) != 0) { \ 162 if (xxs < 0) { \ 163 delta = 60000000; \ 164 } else if (xxs > 4) { \ 165 if (xxs > 60) \ 166 delta = 60000000; \ 167 else \ 168 delta += xxs * 1000000; \ 169 } else while (xxs > 0) { \ 170 delta += 1000000; \ 171 xxs--; \ 172 } \ 173 } \ 174 } 175 176 #ifdef ALTQ3_COMPAT 177 /* rio_list keeps all rio_queue_t's allocated. */ 178 static rio_queue_t *rio_list = NULL; 179 #endif 180 /* default rio parameter values */ 181 static struct redparams default_rio_params[RIO_NDROPPREC] = { 182 /* th_min, th_max, inv_pmax */ 183 { TH_MAX * 2 + TH_MIN, TH_MAX * 3, INV_P_MAX }, /* low drop precedence */ 184 { TH_MAX + TH_MIN, TH_MAX * 2, INV_P_MAX }, /* medium drop precedence */ 185 { TH_MIN, TH_MAX, INV_P_MAX } /* high drop precedence */ 186 }; 187 188 /* internal function prototypes */ 189 static int dscp2index(u_int8_t); 190 #ifdef ALTQ3_COMPAT 191 static int rio_enqueue(struct ifaltq *, struct mbuf *, struct altq_pktattr *); 192 static struct mbuf *rio_dequeue(struct ifaltq *, int); 193 static int rio_request(struct ifaltq *, int, void *); 194 static int rio_detach(rio_queue_t *); 195 196 /* 197 * rio device interface 198 */ 199 altqdev_decl(rio); 200 201 #endif /* ALTQ3_COMPAT */ 202 203 rio_t * 204 rio_alloc(int weight, struct redparams *params, int flags, int pkttime) 205 { 206 rio_t *rp; 207 int w, i; 208 int npkts_per_sec; 209 210 rp = malloc(sizeof(rio_t), M_DEVBUF, M_WAITOK|M_ZERO); 211 if (rp == NULL) 212 return (NULL); 213 214 rp->rio_flags = flags; 215 if (pkttime == 0) 216 /* default packet time: 1000 bytes / 10Mbps * 8 * 1000000 */ 217 rp->rio_pkttime = 800; 218 else 219 rp->rio_pkttime = pkttime; 220 221 if (weight != 0) 222 rp->rio_weight = weight; 223 else { 224 /* use default */ 225 rp->rio_weight = W_WEIGHT; 226 227 /* when the link is very slow, adjust red parameters */ 228 npkts_per_sec = 1000000 / rp->rio_pkttime; 229 if (npkts_per_sec < 50) { 230 /* up to about 400Kbps */ 231 rp->rio_weight = W_WEIGHT_2; 232 } else if (npkts_per_sec < 300) { 233 /* up to about 2.4Mbps */ 234 rp->rio_weight = W_WEIGHT_1; 235 } 236 } 237 238 /* calculate wshift. weight must be power of 2 */ 239 w = rp->rio_weight; 240 for (i = 0; w > 1; i++) 241 w = w >> 1; 242 rp->rio_wshift = i; 243 w = 1 << rp->rio_wshift; 244 if (w != rp->rio_weight) { 245 printf("invalid weight value %d for red! use %d\n", 246 rp->rio_weight, w); 247 rp->rio_weight = w; 248 } 249 250 /* allocate weight table */ 251 rp->rio_wtab = wtab_alloc(rp->rio_weight); 252 253 for (i = 0; i < RIO_NDROPPREC; i++) { 254 struct dropprec_state *prec = &rp->rio_precstate[i]; 255 256 prec->avg = 0; 257 prec->idle = 1; 258 259 if (params == NULL || params[i].inv_pmax == 0) 260 prec->inv_pmax = default_rio_params[i].inv_pmax; 261 else 262 prec->inv_pmax = params[i].inv_pmax; 263 if (params == NULL || params[i].th_min == 0) 264 prec->th_min = default_rio_params[i].th_min; 265 else 266 prec->th_min = params[i].th_min; 267 if (params == NULL || params[i].th_max == 0) 268 prec->th_max = default_rio_params[i].th_max; 269 else 270 prec->th_max = params[i].th_max; 271 272 /* 273 * th_min_s and th_max_s are scaled versions of th_min 274 * and th_max to be compared with avg. 275 */ 276 prec->th_min_s = prec->th_min << (rp->rio_wshift + FP_SHIFT); 277 prec->th_max_s = prec->th_max << (rp->rio_wshift + FP_SHIFT); 278 279 /* 280 * precompute probability denominator 281 * probd = (2 * (TH_MAX-TH_MIN) / pmax) in fixed-point 282 */ 283 prec->probd = (2 * (prec->th_max - prec->th_min) 284 * prec->inv_pmax) << FP_SHIFT; 285 286 microtime(&prec->last); 287 } 288 289 return (rp); 290 } 291 292 void 293 rio_destroy(rio_t *rp) 294 { 295 wtab_destroy(rp->rio_wtab); 296 free(rp, M_DEVBUF); 297 } 298 299 void 300 rio_getstats(rio_t *rp, struct redstats *sp) 301 { 302 int i; 303 304 for (i = 0; i < RIO_NDROPPREC; i++) { 305 bcopy(&rp->q_stats[i], sp, sizeof(struct redstats)); 306 sp->q_avg = rp->rio_precstate[i].avg >> rp->rio_wshift; 307 sp++; 308 } 309 } 310 311 #if (RIO_NDROPPREC == 3) 312 /* 313 * internally, a drop precedence value is converted to an index 314 * starting from 0. 315 */ 316 static int 317 dscp2index(u_int8_t dscp) 318 { 319 int dpindex = dscp & AF_DROPPRECMASK; 320 321 if (dpindex == 0) 322 return (0); 323 return ((dpindex >> 3) - 1); 324 } 325 #endif 326 327 #if 1 328 /* 329 * kludge: when a packet is dequeued, we need to know its drop precedence 330 * in order to keep the queue length of each drop precedence. 331 * use m_pkthdr.rcvif to pass this info. 332 */ 333 #define RIOM_SET_PRECINDEX(m, idx) \ 334 do { (m)->m_pkthdr.rcvif = (struct ifnet *)((long)(idx)); } while (0) 335 #define RIOM_GET_PRECINDEX(m) \ 336 ({ long idx; idx = (long)((m)->m_pkthdr.rcvif); \ 337 (m)->m_pkthdr.rcvif = NULL; idx; }) 338 #endif 339 340 int 341 rio_addq(rio_t *rp, class_queue_t *q, struct mbuf *m, 342 struct altq_pktattr *pktattr) 343 { 344 int avg, droptype; 345 u_int8_t dsfield, odsfield; 346 int dpindex, i, n, t; 347 struct timeval now; 348 struct dropprec_state *prec; 349 350 dsfield = odsfield = read_dsfield(m, pktattr); 351 dpindex = dscp2index(dsfield); 352 353 /* 354 * update avg of the precedence states whose drop precedence 355 * is larger than or equal to the drop precedence of the packet 356 */ 357 now.tv_sec = 0; 358 for (i = dpindex; i < RIO_NDROPPREC; i++) { 359 prec = &rp->rio_precstate[i]; 360 avg = prec->avg; 361 if (prec->idle) { 362 prec->idle = 0; 363 if (now.tv_sec == 0) 364 microtime(&now); 365 t = (now.tv_sec - prec->last.tv_sec); 366 if (t > 60) 367 avg = 0; 368 else { 369 t = t * 1000000 + 370 (now.tv_usec - prec->last.tv_usec); 371 n = t / rp->rio_pkttime; 372 /* calculate (avg = (1 - Wq)^n * avg) */ 373 if (n > 0) 374 avg = (avg >> FP_SHIFT) * 375 pow_w(rp->rio_wtab, n); 376 } 377 } 378 379 /* run estimator. (avg is scaled by WEIGHT in fixed-point) */ 380 avg += (prec->qlen << FP_SHIFT) - (avg >> rp->rio_wshift); 381 prec->avg = avg; /* save the new value */ 382 /* 383 * count keeps a tally of arriving traffic that has not 384 * been dropped. 385 */ 386 prec->count++; 387 } 388 389 prec = &rp->rio_precstate[dpindex]; 390 avg = prec->avg; 391 392 /* see if we drop early */ 393 droptype = DTYPE_NODROP; 394 if (avg >= prec->th_min_s && prec->qlen > 1) { 395 if (avg >= prec->th_max_s) { 396 /* avg >= th_max: forced drop */ 397 droptype = DTYPE_FORCED; 398 } else if (prec->old == 0) { 399 /* first exceeds th_min */ 400 prec->count = 1; 401 prec->old = 1; 402 } else if (drop_early((avg - prec->th_min_s) >> rp->rio_wshift, 403 prec->probd, prec->count)) { 404 /* unforced drop by red */ 405 droptype = DTYPE_EARLY; 406 } 407 } else { 408 /* avg < th_min */ 409 prec->old = 0; 410 } 411 412 /* 413 * if the queue length hits the hard limit, it's a forced drop. 414 */ 415 if (droptype == DTYPE_NODROP && qlen(q) >= qlimit(q)) 416 droptype = DTYPE_FORCED; 417 418 if (droptype != DTYPE_NODROP) { 419 /* always drop incoming packet (as opposed to randomdrop) */ 420 for (i = dpindex; i < RIO_NDROPPREC; i++) 421 rp->rio_precstate[i].count = 0; 422 #ifdef RIO_STATS 423 if (droptype == DTYPE_EARLY) 424 rp->q_stats[dpindex].drop_unforced++; 425 else 426 rp->q_stats[dpindex].drop_forced++; 427 PKTCNTR_ADD(&rp->q_stats[dpindex].drop_cnt, m_pktlen(m)); 428 #endif 429 m_freem(m); 430 return (-1); 431 } 432 433 for (i = dpindex; i < RIO_NDROPPREC; i++) 434 rp->rio_precstate[i].qlen++; 435 436 /* save drop precedence index in mbuf hdr */ 437 RIOM_SET_PRECINDEX(m, dpindex); 438 439 if (rp->rio_flags & RIOF_CLEARDSCP) 440 dsfield &= ~DSCP_MASK; 441 442 if (dsfield != odsfield) 443 write_dsfield(m, pktattr, dsfield); 444 445 _addq(q, m); 446 447 #ifdef RIO_STATS 448 PKTCNTR_ADD(&rp->q_stats[dpindex].xmit_cnt, m_pktlen(m)); 449 #endif 450 return (0); 451 } 452 453 struct mbuf * 454 rio_getq(rio_t *rp, class_queue_t *q) 455 { 456 struct mbuf *m; 457 int dpindex, i; 458 459 if ((m = _getq(q)) == NULL) 460 return NULL; 461 462 dpindex = RIOM_GET_PRECINDEX(m); 463 for (i = dpindex; i < RIO_NDROPPREC; i++) { 464 if (--rp->rio_precstate[i].qlen == 0) { 465 if (rp->rio_precstate[i].idle == 0) { 466 rp->rio_precstate[i].idle = 1; 467 microtime(&rp->rio_precstate[i].last); 468 } 469 } 470 } 471 return (m); 472 } 473 474 #ifdef ALTQ3_COMPAT 475 int 476 rioopen(dev_t dev __unused, int flag __unused, int fmt __unused, 477 struct lwp *l __unused) 478 { 479 /* everything will be done when the queueing scheme is attached. */ 480 return 0; 481 } 482 483 int 484 rioclose(dev_t dev __unused, int flag __unused, int fmt __unused, 485 struct lwp *l __unused) 486 { 487 rio_queue_t *rqp; 488 int err, error = 0; 489 490 while ((rqp = rio_list) != NULL) { 491 /* destroy all */ 492 err = rio_detach(rqp); 493 if (err != 0 && error == 0) 494 error = err; 495 } 496 497 return error; 498 } 499 500 int 501 rioioctl(dev_t dev __unused, ioctlcmd_t cmd, caddr_t addr, int flag __unused, 502 struct lwp *l) 503 { 504 rio_queue_t *rqp; 505 struct rio_interface *ifacep; 506 struct ifnet *ifp; 507 int error = 0; 508 509 /* check super-user privilege */ 510 switch (cmd) { 511 case RIO_GETSTATS: 512 break; 513 default: 514 #if (__FreeBSD_version > 400000) 515 if ((error = suser(p)) != 0) 516 return (error); 517 #else 518 if ((error = kauth_authorize_generic(l->l_cred, 519 KAUTH_GENERIC_ISSUSER, &l->l_acflag)) != 0) 520 return (error); 521 #endif 522 break; 523 } 524 525 switch (cmd) { 526 527 case RIO_ENABLE: 528 ifacep = (struct rio_interface *)addr; 529 if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) { 530 error = EBADF; 531 break; 532 } 533 error = altq_enable(rqp->rq_ifq); 534 break; 535 536 case RIO_DISABLE: 537 ifacep = (struct rio_interface *)addr; 538 if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) { 539 error = EBADF; 540 break; 541 } 542 error = altq_disable(rqp->rq_ifq); 543 break; 544 545 case RIO_IF_ATTACH: 546 ifp = ifunit(((struct rio_interface *)addr)->rio_ifname); 547 if (ifp == NULL) { 548 error = ENXIO; 549 break; 550 } 551 552 /* allocate and initialize rio_queue_t */ 553 rqp = malloc(sizeof(rio_queue_t), M_DEVBUF, M_WAITOK|M_ZERO); 554 if (rqp == NULL) { 555 error = ENOMEM; 556 break; 557 } 558 559 rqp->rq_q = malloc(sizeof(class_queue_t), M_DEVBUF, 560 M_WAITOK|M_ZERO); 561 if (rqp->rq_q == NULL) { 562 free(rqp, M_DEVBUF); 563 error = ENOMEM; 564 break; 565 } 566 567 rqp->rq_rio = rio_alloc(0, NULL, 0, 0); 568 if (rqp->rq_rio == NULL) { 569 free(rqp->rq_q, M_DEVBUF); 570 free(rqp, M_DEVBUF); 571 error = ENOMEM; 572 break; 573 } 574 575 rqp->rq_ifq = &ifp->if_snd; 576 qtail(rqp->rq_q) = NULL; 577 qlen(rqp->rq_q) = 0; 578 qlimit(rqp->rq_q) = RIO_LIMIT; 579 qtype(rqp->rq_q) = Q_RIO; 580 581 /* 582 * set RIO to this ifnet structure. 583 */ 584 error = altq_attach(rqp->rq_ifq, ALTQT_RIO, rqp, 585 rio_enqueue, rio_dequeue, rio_request, 586 NULL, NULL); 587 if (error) { 588 rio_destroy(rqp->rq_rio); 589 free(rqp->rq_q, M_DEVBUF); 590 free(rqp, M_DEVBUF); 591 break; 592 } 593 594 /* add this state to the rio list */ 595 rqp->rq_next = rio_list; 596 rio_list = rqp; 597 break; 598 599 case RIO_IF_DETACH: 600 ifacep = (struct rio_interface *)addr; 601 if ((rqp = altq_lookup(ifacep->rio_ifname, ALTQT_RIO)) == NULL) { 602 error = EBADF; 603 break; 604 } 605 error = rio_detach(rqp); 606 break; 607 608 case RIO_GETSTATS: 609 do { 610 struct rio_stats *q_stats; 611 rio_t *rp; 612 int i; 613 614 q_stats = (struct rio_stats *)addr; 615 if ((rqp = altq_lookup(q_stats->iface.rio_ifname, 616 ALTQT_RIO)) == NULL) { 617 error = EBADF; 618 break; 619 } 620 621 rp = rqp->rq_rio; 622 623 q_stats->q_limit = qlimit(rqp->rq_q); 624 q_stats->weight = rp->rio_weight; 625 q_stats->flags = rp->rio_flags; 626 627 for (i = 0; i < RIO_NDROPPREC; i++) { 628 q_stats->q_len[i] = rp->rio_precstate[i].qlen; 629 bcopy(&rp->q_stats[i], &q_stats->q_stats[i], 630 sizeof(struct redstats)); 631 q_stats->q_stats[i].q_avg = 632 rp->rio_precstate[i].avg >> rp->rio_wshift; 633 634 q_stats->q_params[i].inv_pmax 635 = rp->rio_precstate[i].inv_pmax; 636 q_stats->q_params[i].th_min 637 = rp->rio_precstate[i].th_min; 638 q_stats->q_params[i].th_max 639 = rp->rio_precstate[i].th_max; 640 } 641 } while (/*CONSTCOND*/ 0); 642 break; 643 644 case RIO_CONFIG: 645 do { 646 struct rio_conf *fc; 647 rio_t *new; 648 int s, limit, i; 649 650 fc = (struct rio_conf *)addr; 651 if ((rqp = altq_lookup(fc->iface.rio_ifname, 652 ALTQT_RIO)) == NULL) { 653 error = EBADF; 654 break; 655 } 656 657 new = rio_alloc(fc->rio_weight, &fc->q_params[0], 658 fc->rio_flags, fc->rio_pkttime); 659 if (new == NULL) { 660 error = ENOMEM; 661 break; 662 } 663 664 s = splnet(); 665 _flushq(rqp->rq_q); 666 limit = fc->rio_limit; 667 if (limit < fc->q_params[RIO_NDROPPREC-1].th_max) 668 limit = fc->q_params[RIO_NDROPPREC-1].th_max; 669 qlimit(rqp->rq_q) = limit; 670 671 rio_destroy(rqp->rq_rio); 672 rqp->rq_rio = new; 673 674 splx(s); 675 676 /* write back new values */ 677 fc->rio_limit = limit; 678 for (i = 0; i < RIO_NDROPPREC; i++) { 679 fc->q_params[i].inv_pmax = 680 rqp->rq_rio->rio_precstate[i].inv_pmax; 681 fc->q_params[i].th_min = 682 rqp->rq_rio->rio_precstate[i].th_min; 683 fc->q_params[i].th_max = 684 rqp->rq_rio->rio_precstate[i].th_max; 685 } 686 } while (/*CONSTCOND*/ 0); 687 break; 688 689 case RIO_SETDEFAULTS: 690 do { 691 struct redparams *rp; 692 int i; 693 694 rp = (struct redparams *)addr; 695 for (i = 0; i < RIO_NDROPPREC; i++) 696 default_rio_params[i] = rp[i]; 697 } while (/*CONSTCOND*/ 0); 698 break; 699 700 default: 701 error = EINVAL; 702 break; 703 } 704 705 return error; 706 } 707 708 static int 709 rio_detach(rio_queue_t *rqp) 710 { 711 rio_queue_t *tmp; 712 int error = 0; 713 714 if (ALTQ_IS_ENABLED(rqp->rq_ifq)) 715 altq_disable(rqp->rq_ifq); 716 717 if ((error = altq_detach(rqp->rq_ifq))) 718 return (error); 719 720 if (rio_list == rqp) 721 rio_list = rqp->rq_next; 722 else { 723 for (tmp = rio_list; tmp != NULL; tmp = tmp->rq_next) 724 if (tmp->rq_next == rqp) { 725 tmp->rq_next = rqp->rq_next; 726 break; 727 } 728 if (tmp == NULL) 729 printf("rio_detach: no state found in rio_list!\n"); 730 } 731 732 rio_destroy(rqp->rq_rio); 733 free(rqp->rq_q, M_DEVBUF); 734 free(rqp, M_DEVBUF); 735 return (error); 736 } 737 738 /* 739 * rio support routines 740 */ 741 static int 742 rio_request(struct ifaltq *ifq, int req, void *arg __unused) 743 { 744 rio_queue_t *rqp = (rio_queue_t *)ifq->altq_disc; 745 746 switch (req) { 747 case ALTRQ_PURGE: 748 _flushq(rqp->rq_q); 749 if (ALTQ_IS_ENABLED(ifq)) 750 ifq->ifq_len = 0; 751 break; 752 } 753 return (0); 754 } 755 756 /* 757 * enqueue routine: 758 * 759 * returns: 0 when successfully queued. 760 * ENOBUFS when drop occurs. 761 */ 762 static int 763 rio_enqueue(struct ifaltq *ifq, struct mbuf *m, struct altq_pktattr *pktattr) 764 { 765 rio_queue_t *rqp = (rio_queue_t *)ifq->altq_disc; 766 int error = 0; 767 768 if (rio_addq(rqp->rq_rio, rqp->rq_q, m, pktattr) == 0) 769 ifq->ifq_len++; 770 else 771 error = ENOBUFS; 772 return error; 773 } 774 775 /* 776 * dequeue routine: 777 * must be called in splnet. 778 * 779 * returns: mbuf dequeued. 780 * NULL when no packet is available in the queue. 781 */ 782 783 static struct mbuf * 784 rio_dequeue(struct ifaltq *ifq, 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 #ifdef KLD_MODULE 799 800 static struct altqsw rio_sw = 801 {"rio", rioopen, rioclose, rioioctl}; 802 803 ALTQ_MODULE(altq_rio, ALTQT_RIO, &rio_sw); 804 MODULE_VERSION(altq_rio, 1); 805 MODULE_DEPEND(altq_rio, altq_red, 1, 1, 1); 806 807 #endif /* KLD_MODULE */ 808 #endif /* ALTQ3_COMPAT */ 809 810 #endif /* ALTQ_RIO */ 811
Indexes created Sun Sep 28 12:09:53 GMT 2025