1 /* $KAME: altq_rio.c,v 1.17 2003/07/10 12:07:49 kjc Exp $ */ 2 /* $DragonFly: src/sys/net/altq/altq_rio.c,v 1.3 2006/12/22 23:44:55 swildner 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 "opt_altq.h" 63 #include "opt_inet.h" 64 #include "opt_inet6.h" 65 66 #ifdef ALTQ_RIO /* rio is enabled by ALTQ_RIO option in opt_altq.h */ 67 68 #include <sys/param.h> 69 #include <sys/malloc.h> 70 #include <sys/mbuf.h> 71 #include <sys/socket.h> 72 #include <sys/systm.h> 73 #include <sys/errno.h> 74 75 #include <net/if.h> 76 77 #include <netinet/in.h> 78 #include <netinet/in_systm.h> 79 #include <netinet/ip.h> 80 #ifdef INET6 81 #include <netinet/ip6.h> 82 #endif 83 84 #include <net/pf/pfvar.h> 85 #include <net/altq/altq.h> 86 #include <net/altq/altq_red.h> 87 #include <net/altq/altq_rio.h> 88 89 /* 90 * RIO: RED with IN/OUT bit 91 * described in 92 * "Explicit Allocation of Best Effort Packet Delivery Service" 93 * David D. Clark and Wenjia Fang, MIT Lab for Computer Science 94 * http://diffserv.lcs.mit.edu/Papers/exp-alloc-ddc-wf.{ps,pdf} 95 * 96 * this implementation is extended to support more than 2 drop precedence 97 * values as described in RFC2597 (Assured Forwarding PHB Group). 98 * 99 */ 100 /* 101 * AF DS (differentiated service) codepoints. 102 * (classes can be mapped to CBQ or H-FSC classes.) 103 * 104 * 0 1 2 3 4 5 6 7 105 * +---+---+---+---+---+---+---+---+ 106 * | CLASS |DropPre| 0 | CU | 107 * +---+---+---+---+---+---+---+---+ 108 * 109 * class 1: 001 110 * class 2: 010 111 * class 3: 011 112 * class 4: 100 113 * 114 * low drop prec: 01 115 * medium drop prec: 10 116 * high drop prec: 11 117 */ 118 119 /* normal red parameters */ 120 #define W_WEIGHT 512 /* inverse of weight of EWMA (511/512) */ 121 /* q_weight = 0.00195 */ 122 123 /* red parameters for a slow link */ 124 #define W_WEIGHT_1 128 /* inverse of weight of EWMA (127/128) */ 125 /* q_weight = 0.0078125 */ 126 127 /* red parameters for a very slow link (e.g., dialup) */ 128 #define W_WEIGHT_2 64 /* inverse of weight of EWMA (63/64) */ 129 /* q_weight = 0.015625 */ 130 131 /* fixed-point uses 12-bit decimal places */ 132 #define FP_SHIFT 12 /* fixed-point shift */ 133 134 /* red parameters for drop probability */ 135 #define INV_P_MAX 10 /* inverse of max drop probability */ 136 #define TH_MIN 5 /* min threshold */ 137 #define TH_MAX 15 /* max threshold */ 138 139 #define RIO_LIMIT 60 /* default max queue lenght */ 140 #define RIO_STATS /* collect statistics */ 141 142 /* default rio parameter values */ 143 static struct redparams default_rio_params[RIO_NDROPPREC] = { 144 /* th_min, th_max, inv_pmax */ 145 { TH_MAX * 2 + TH_MIN, TH_MAX * 3, INV_P_MAX }, /* low drop precedence */ 146 { TH_MAX + TH_MIN, TH_MAX * 2, INV_P_MAX }, /* medium drop precedence */ 147 { TH_MIN, TH_MAX, INV_P_MAX } /* high drop precedence */ 148 }; 149 150 /* internal function prototypes */ 151 static int dscp2index(uint8_t); 152 /* YYY Do we really need this? */ 153 #define AF_DROPPRECMASK 0x18 154 #define DSCP_MASK 0xfc 155 156 rio_t * 157 rio_alloc(int weight, struct redparams *params, int flags, int pkttime) 158 { 159 rio_t *rp; 160 int w, i; 161 int npkts_per_sec; 162 163 rp = kmalloc(sizeof(*rp), M_ALTQ, M_WAITOK | M_ZERO); 164 165 rp->rio_flags = flags; 166 if (pkttime == 0) 167 /* default packet time: 1000 bytes / 10Mbps * 8 * 1000000 */ 168 rp->rio_pkttime = 800; 169 else 170 rp->rio_pkttime = pkttime; 171 172 if (weight != 0) 173 rp->rio_weight = weight; 174 else { 175 /* use default */ 176 rp->rio_weight = W_WEIGHT; 177 178 /* when the link is very slow, adjust red parameters */ 179 npkts_per_sec = 1000000 / rp->rio_pkttime; 180 if (npkts_per_sec < 50) { 181 /* up to about 400Kbps */ 182 rp->rio_weight = W_WEIGHT_2; 183 } else if (npkts_per_sec < 300) { 184 /* up to about 2.4Mbps */ 185 rp->rio_weight = W_WEIGHT_1; 186 } 187 } 188 189 /* calculate wshift. weight must be power of 2 */ 190 w = rp->rio_weight; 191 for (i = 0; w > 1; i++) 192 w = w >> 1; 193 rp->rio_wshift = i; 194 w = 1 << rp->rio_wshift; 195 if (w != rp->rio_weight) { 196 kprintf("invalid weight value %d for red! use %d\n", 197 rp->rio_weight, w); 198 rp->rio_weight = w; 199 } 200 201 /* allocate weight table */ 202 rp->rio_wtab = wtab_alloc(rp->rio_weight); 203 204 for (i = 0; i < RIO_NDROPPREC; i++) { 205 struct dropprec_state *prec = &rp->rio_precstate[i]; 206 207 prec->avg = 0; 208 prec->idle = 1; 209 210 if (params == NULL || params[i].inv_pmax == 0) 211 prec->inv_pmax = default_rio_params[i].inv_pmax; 212 else 213 prec->inv_pmax = params[i].inv_pmax; 214 if (params == NULL || params[i].th_min == 0) 215 prec->th_min = default_rio_params[i].th_min; 216 else 217 prec->th_min = params[i].th_min; 218 if (params == NULL || params[i].th_max == 0) 219 prec->th_max = default_rio_params[i].th_max; 220 else 221 prec->th_max = params[i].th_max; 222 223 /* 224 * th_min_s and th_max_s are scaled versions of th_min 225 * and th_max to be compared with avg. 226 */ 227 prec->th_min_s = prec->th_min << (rp->rio_wshift + FP_SHIFT); 228 prec->th_max_s = prec->th_max << (rp->rio_wshift + FP_SHIFT); 229 230 /* 231 * precompute probability denominator 232 * probd = (2 * (TH_MAX-TH_MIN) / pmax) in fixed-point 233 */ 234 prec->probd = (2 * (prec->th_max - prec->th_min) 235 * prec->inv_pmax) << FP_SHIFT; 236 237 microtime(&prec->last); 238 } 239 240 return (rp); 241 } 242 243 void 244 rio_destroy(rio_t *rp) 245 { 246 wtab_destroy(rp->rio_wtab); 247 kfree(rp, M_ALTQ); 248 } 249 250 void 251 rio_getstats(rio_t *rp, struct redstats *sp) 252 { 253 int i; 254 255 for (i = 0; i < RIO_NDROPPREC; i++) { 256 bcopy(&rp->q_stats[i], sp, sizeof(struct redstats)); 257 sp->q_avg = rp->rio_precstate[i].avg >> rp->rio_wshift; 258 sp++; 259 } 260 } 261 262 #if (RIO_NDROPPREC == 3) 263 /* 264 * internally, a drop precedence value is converted to an index 265 * starting from 0. 266 */ 267 static int 268 dscp2index(uint8_t dscp) 269 { 270 int dpindex = dscp & AF_DROPPRECMASK; 271 272 if (dpindex == 0) 273 return (0); 274 return ((dpindex >> 3) - 1); 275 } 276 #endif 277 278 #if 1 279 /* 280 * kludge: when a packet is dequeued, we need to know its drop precedence 281 * in order to keep the queue length of each drop precedence. 282 * use m_pkthdr.rcvif to pass this info. 283 */ 284 #define RIOM_SET_PRECINDEX(m, idx) \ 285 do { (m)->m_pkthdr.rcvif = (struct ifnet *)((long)(idx)); } while (0) 286 #define RIOM_GET_PRECINDEX(m) \ 287 ({ long idx; idx = (long)((m)->m_pkthdr.rcvif); \ 288 (m)->m_pkthdr.rcvif = NULL; idx; }) 289 #endif 290 291 int 292 rio_addq(rio_t *rp, class_queue_t *q, struct mbuf *m, struct altq_pktattr *pktattr) 293 { 294 int avg, droptype; 295 uint8_t dsfield, odsfield; 296 int dpindex, i, n, t; 297 struct timeval now; 298 struct dropprec_state *prec; 299 300 dsfield = odsfield = read_dsfield(m, pktattr); 301 dpindex = dscp2index(dsfield); 302 303 /* 304 * update avg of the precedence states whose drop precedence 305 * is larger than or equal to the drop precedence of the packet 306 */ 307 now.tv_sec = 0; 308 for (i = dpindex; i < RIO_NDROPPREC; i++) { 309 prec = &rp->rio_precstate[i]; 310 avg = prec->avg; 311 if (prec->idle) { 312 prec->idle = 0; 313 if (now.tv_sec == 0) 314 microtime(&now); 315 t = (now.tv_sec - prec->last.tv_sec); 316 if (t > 60) 317 avg = 0; 318 else { 319 t = t * 1000000 + 320 (now.tv_usec - prec->last.tv_usec); 321 n = t / rp->rio_pkttime; 322 /* calculate (avg = (1 - Wq)^n * avg) */ 323 if (n > 0) 324 avg = (avg >> FP_SHIFT) * 325 pow_w(rp->rio_wtab, n); 326 } 327 } 328 329 /* run estimator. (avg is scaled by WEIGHT in fixed-point) */ 330 avg += (prec->qlen << FP_SHIFT) - (avg >> rp->rio_wshift); 331 prec->avg = avg; /* save the new value */ 332 /* 333 * count keeps a tally of arriving traffic that has not 334 * been dropped. 335 */ 336 prec->count++; 337 } 338 339 prec = &rp->rio_precstate[dpindex]; 340 avg = prec->avg; 341 342 /* see if we drop early */ 343 droptype = DTYPE_NODROP; 344 if (avg >= prec->th_min_s && prec->qlen > 1) { 345 if (avg >= prec->th_max_s) { 346 /* avg >= th_max: forced drop */ 347 droptype = DTYPE_FORCED; 348 } else if (prec->old == 0) { 349 /* first exceeds th_min */ 350 prec->count = 1; 351 prec->old = 1; 352 } else if (drop_early((avg - prec->th_min_s) >> rp->rio_wshift, 353 prec->probd, prec->count)) { 354 /* unforced drop by red */ 355 droptype = DTYPE_EARLY; 356 } 357 } else { 358 /* avg < th_min */ 359 prec->old = 0; 360 } 361 362 /* 363 * if the queue length hits the hard limit, it's a forced drop. 364 */ 365 if (droptype == DTYPE_NODROP && qlen(q) >= qlimit(q)) 366 droptype = DTYPE_FORCED; 367 368 if (droptype != DTYPE_NODROP) { 369 /* always drop incoming packet (as opposed to randomdrop) */ 370 for (i = dpindex; i < RIO_NDROPPREC; i++) 371 rp->rio_precstate[i].count = 0; 372 #ifdef RIO_STATS 373 if (droptype == DTYPE_EARLY) 374 rp->q_stats[dpindex].drop_unforced++; 375 else 376 rp->q_stats[dpindex].drop_forced++; 377 PKTCNTR_ADD(&rp->q_stats[dpindex].drop_cnt, m_pktlen(m)); 378 #endif 379 m_freem(m); 380 return (-1); 381 } 382 383 for (i = dpindex; i < RIO_NDROPPREC; i++) 384 rp->rio_precstate[i].qlen++; 385 386 /* save drop precedence index in mbuf hdr */ 387 RIOM_SET_PRECINDEX(m, dpindex); 388 389 if (rp->rio_flags & RIOF_CLEARDSCP) 390 dsfield &= ~DSCP_MASK; 391 392 if (dsfield != odsfield) 393 write_dsfield(m, pktattr, dsfield); 394 395 _addq(q, m); 396 397 #ifdef RIO_STATS 398 PKTCNTR_ADD(&rp->q_stats[dpindex].xmit_cnt, m_pktlen(m)); 399 #endif 400 return (0); 401 } 402 403 struct mbuf * 404 rio_getq(rio_t *rp, class_queue_t *q) 405 { 406 struct mbuf *m; 407 int dpindex, i; 408 409 if ((m = _getq(q)) == NULL) 410 return (NULL); 411 412 dpindex = RIOM_GET_PRECINDEX(m); 413 for (i = dpindex; i < RIO_NDROPPREC; i++) { 414 if (--rp->rio_precstate[i].qlen == 0) { 415 if (rp->rio_precstate[i].idle == 0) { 416 rp->rio_precstate[i].idle = 1; 417 microtime(&rp->rio_precstate[i].last); 418 } 419 } 420 } 421 return (m); 422 } 423 424 #endif /* ALTQ_RIO */ 425