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 #include <net/ifq_var.h>
77
78 #include <netinet/in.h>
79 #include <netinet/in_systm.h>
80 #include <netinet/ip.h>
81 #ifdef INET6
82 #include <netinet/ip6.h>
83 #endif
84
85 #include <net/pf/pfvar.h>
86 #include <net/altq/altq.h>
87 #include <net/altq/altq_red.h>
88 #include <net/altq/altq_rio.h>
89
90 /*
91 * RIO: RED with IN/OUT bit
92 * described in
93 * "Explicit Allocation of Best Effort Packet Delivery Service"
94 * David D. Clark and Wenjia Fang, MIT Lab for Computer Science
95 * http://diffserv.lcs.mit.edu/Papers/exp-alloc-ddc-wf.{ps,pdf}
96 *
97 * this implementation is extended to support more than 2 drop precedence
98 * values as described in RFC2597 (Assured Forwarding PHB Group).
99 *
100 */
101 /*
102 * AF DS (differentiated service) codepoints.
103 * (classes can be mapped to CBQ or H-FSC classes.)
104 *
105 * 0 1 2 3 4 5 6 7
106 * +---+---+---+---+---+---+---+---+
107 * | CLASS |DropPre| 0 | CU |
108 * +---+---+---+---+---+---+---+---+
109 *
110 * class 1: 001
111 * class 2: 010
112 * class 3: 011
113 * class 4: 100
114 *
115 * low drop prec: 01
116 * medium drop prec: 10
117 * high drop prec: 11
118 */
119
120 /* normal red parameters */
121 #define W_WEIGHT 512 /* inverse of weight of EWMA (511/512) */
122 /* q_weight = 0.00195 */
123
124 /* red parameters for a slow link */
125 #define W_WEIGHT_1 128 /* inverse of weight of EWMA (127/128) */
126 /* q_weight = 0.0078125 */
127
128 /* red parameters for a very slow link (e.g., dialup) */
129 #define W_WEIGHT_2 64 /* inverse of weight of EWMA (63/64) */
130 /* q_weight = 0.015625 */
131
132 /* fixed-point uses 12-bit decimal places */
133 #define FP_SHIFT 12 /* fixed-point shift */
134
135 /* red parameters for drop probability */
136 #define INV_P_MAX 10 /* inverse of max drop probability */
137 #define TH_MIN 5 /* min threshold */
138 #define TH_MAX 15 /* max threshold */
139
140 #define RIO_LIMIT 60 /* default max queue lenght */
141 #define RIO_STATS /* collect statistics */
142
143 /* default rio parameter values */
144 static struct redparams default_rio_params[RIO_NDROPPREC] = {
145 /* th_min, th_max, inv_pmax */
146 { TH_MAX * 2 + TH_MIN, TH_MAX * 3, INV_P_MAX }, /* low drop precedence */
147 { TH_MAX + TH_MIN, TH_MAX * 2, INV_P_MAX }, /* medium drop precedence */
148 { TH_MIN, TH_MAX, INV_P_MAX } /* high drop precedence */
149 };
150
151 /* internal function prototypes */
152 static int dscp2index(uint8_t);
153 /* YYY Do we really need this? */
154 #define AF_DROPPRECMASK 0x18
155 #define DSCP_MASK 0xfc
156
157 rio_t *
rio_alloc(int weight,struct redparams * params,int flags,int pkttime)158 rio_alloc(int weight, struct redparams *params, int flags, int pkttime)
159 {
160 rio_t *rp;
161 int w, i;
162 int npkts_per_sec;
163
164 rp = kmalloc(sizeof(*rp), M_ALTQ, M_WAITOK | M_ZERO);
165
166 rp->rio_flags = flags;
167 if (pkttime == 0)
168 /* default packet time: 1000 bytes / 10Mbps * 8 * 1000000 */
169 rp->rio_pkttime = 800;
170 else
171 rp->rio_pkttime = pkttime;
172
173 if (weight != 0)
174 rp->rio_weight = weight;
175 else {
176 /* use default */
177 rp->rio_weight = W_WEIGHT;
178
179 /* when the link is very slow, adjust red parameters */
180 npkts_per_sec = 1000000 / rp->rio_pkttime;
181 if (npkts_per_sec < 50) {
182 /* up to about 400Kbps */
183 rp->rio_weight = W_WEIGHT_2;
184 } else if (npkts_per_sec < 300) {
185 /* up to about 2.4Mbps */
186 rp->rio_weight = W_WEIGHT_1;
187 }
188 }
189
190 /* calculate wshift. weight must be power of 2 */
191 w = rp->rio_weight;
192 for (i = 0; w > 1; i++)
193 w = w >> 1;
194 rp->rio_wshift = i;
195 w = 1 << rp->rio_wshift;
196 if (w != rp->rio_weight) {
197 kprintf("invalid weight value %d for red! use %d\n",
198 rp->rio_weight, w);
199 rp->rio_weight = w;
200 }
201
202 /* allocate weight table */
203 rp->rio_wtab = wtab_alloc(rp->rio_weight);
204
205 for (i = 0; i < RIO_NDROPPREC; i++) {
206 struct dropprec_state *prec = &rp->rio_precstate[i];
207
208 prec->avg = 0;
209 prec->idle = 1;
210
211 if (params == NULL || params[i].inv_pmax == 0)
212 prec->inv_pmax = default_rio_params[i].inv_pmax;
213 else
214 prec->inv_pmax = params[i].inv_pmax;
215 if (params == NULL || params[i].th_min == 0)
216 prec->th_min = default_rio_params[i].th_min;
217 else
218 prec->th_min = params[i].th_min;
219 if (params == NULL || params[i].th_max == 0)
220 prec->th_max = default_rio_params[i].th_max;
221 else
222 prec->th_max = params[i].th_max;
223
224 /*
225 * th_min_s and th_max_s are scaled versions of th_min
226 * and th_max to be compared with avg.
227 */
228 prec->th_min_s = prec->th_min << (rp->rio_wshift + FP_SHIFT);
229 prec->th_max_s = prec->th_max << (rp->rio_wshift + FP_SHIFT);
230
231 /*
232 * precompute probability denominator
233 * probd = (2 * (TH_MAX-TH_MIN) / pmax) in fixed-point
234 */
235 prec->probd = (2 * (prec->th_max - prec->th_min)
236 * prec->inv_pmax) << FP_SHIFT;
237
238 microtime(&prec->last);
239 }
240
241 return (rp);
242 }
243
244 void
rio_destroy(rio_t * rp)245 rio_destroy(rio_t *rp)
246 {
247 wtab_destroy(rp->rio_wtab);
248 kfree(rp, M_ALTQ);
249 }
250
251 void
rio_getstats(rio_t * rp,struct redstats * sp)252 rio_getstats(rio_t *rp, struct redstats *sp)
253 {
254 int i;
255
256 for (i = 0; i < RIO_NDROPPREC; i++) {
257 bcopy(&rp->q_stats[i], sp, sizeof(struct redstats));
258 sp->q_avg = rp->rio_precstate[i].avg >> rp->rio_wshift;
259 sp++;
260 }
261 }
262
263 #if (RIO_NDROPPREC == 3)
264 /*
265 * internally, a drop precedence value is converted to an index
266 * starting from 0.
267 */
268 static int
dscp2index(uint8_t dscp)269 dscp2index(uint8_t dscp)
270 {
271 int dpindex = dscp & AF_DROPPRECMASK;
272
273 if (dpindex == 0)
274 return (0);
275 return ((dpindex >> 3) - 1);
276 }
277 #endif
278
279 #if 1
280 /*
281 * kludge: when a packet is dequeued, we need to know its drop precedence
282 * in order to keep the queue length of each drop precedence.
283 * use m_pkthdr.rcvif to pass this info.
284 */
285 #define RIOM_SET_PRECINDEX(m, idx) \
286 do { (m)->m_pkthdr.rcvif = (struct ifnet *)((long)(idx)); } while (0)
287 #define RIOM_GET_PRECINDEX(m) \
288 ({ long idx; idx = (long)((m)->m_pkthdr.rcvif); \
289 (m)->m_pkthdr.rcvif = NULL; idx; })
290 #endif
291
292 int
rio_addq(rio_t * rp,class_queue_t * q,struct mbuf * m,struct altq_pktattr * pktattr)293 rio_addq(rio_t *rp, class_queue_t *q, struct mbuf *m, struct altq_pktattr *pktattr)
294 {
295 int avg, droptype;
296 uint8_t dsfield, odsfield;
297 int dpindex, i, n, t;
298 struct timeval now;
299 struct dropprec_state *prec;
300
301 dsfield = odsfield = read_dsfield(m, pktattr);
302 dpindex = dscp2index(dsfield);
303
304 /*
305 * update avg of the precedence states whose drop precedence
306 * is larger than or equal to the drop precedence of the packet
307 */
308 now.tv_sec = 0;
309 for (i = dpindex; i < RIO_NDROPPREC; i++) {
310 prec = &rp->rio_precstate[i];
311 avg = prec->avg;
312 if (prec->idle) {
313 prec->idle = 0;
314 if (now.tv_sec == 0)
315 microtime(&now);
316 t = (now.tv_sec - prec->last.tv_sec);
317 if (t > 60)
318 avg = 0;
319 else {
320 t = t * 1000000 +
321 (now.tv_usec - prec->last.tv_usec);
322 n = t / rp->rio_pkttime;
323 /* calculate (avg = (1 - Wq)^n * avg) */
324 if (n > 0)
325 avg = (avg >> FP_SHIFT) *
326 pow_w(rp->rio_wtab, n);
327 }
328 }
329
330 /* run estimator. (avg is scaled by WEIGHT in fixed-point) */
331 avg += (prec->qlen << FP_SHIFT) - (avg >> rp->rio_wshift);
332 prec->avg = avg; /* save the new value */
333 /*
334 * count keeps a tally of arriving traffic that has not
335 * been dropped.
336 */
337 prec->count++;
338 }
339
340 prec = &rp->rio_precstate[dpindex];
341 avg = prec->avg;
342
343 /* see if we drop early */
344 droptype = DTYPE_NODROP;
345 if (avg >= prec->th_min_s && prec->qlen > 1) {
346 if (avg >= prec->th_max_s) {
347 /* avg >= th_max: forced drop */
348 droptype = DTYPE_FORCED;
349 } else if (prec->old == 0) {
350 /* first exceeds th_min */
351 prec->count = 1;
352 prec->old = 1;
353 } else if (drop_early((avg - prec->th_min_s) >> rp->rio_wshift,
354 prec->probd, prec->count)) {
355 /* unforced drop by red */
356 droptype = DTYPE_EARLY;
357 }
358 } else {
359 /* avg < th_min */
360 prec->old = 0;
361 }
362
363 /*
364 * if the queue length hits the hard limit, it's a forced drop.
365 */
366 if (droptype == DTYPE_NODROP && qlen(q) >= qlimit(q))
367 droptype = DTYPE_FORCED;
368
369 if (droptype != DTYPE_NODROP) {
370 /* always drop incoming packet (as opposed to randomdrop) */
371 for (i = dpindex; i < RIO_NDROPPREC; i++)
372 rp->rio_precstate[i].count = 0;
373 #ifdef RIO_STATS
374 if (droptype == DTYPE_EARLY)
375 rp->q_stats[dpindex].drop_unforced++;
376 else
377 rp->q_stats[dpindex].drop_forced++;
378 PKTCNTR_ADD(&rp->q_stats[dpindex].drop_cnt, m_pktlen(m));
379 #endif
380 m_freem(m);
381 return (-1);
382 }
383
384 for (i = dpindex; i < RIO_NDROPPREC; i++)
385 rp->rio_precstate[i].qlen++;
386
387 /* save drop precedence index in mbuf hdr */
388 RIOM_SET_PRECINDEX(m, dpindex);
389
390 if (rp->rio_flags & RIOF_CLEARDSCP)
391 dsfield &= ~DSCP_MASK;
392
393 if (dsfield != odsfield)
394 write_dsfield(m, pktattr, dsfield);
395
396 _addq(q, m);
397
398 #ifdef RIO_STATS
399 PKTCNTR_ADD(&rp->q_stats[dpindex].xmit_cnt, m_pktlen(m));
400 #endif
401 return (0);
402 }
403
404 struct mbuf *
rio_getq(rio_t * rp,class_queue_t * q)405 rio_getq(rio_t *rp, class_queue_t *q)
406 {
407 struct mbuf *m;
408 int dpindex, i;
409
410 if ((m = _getq(q)) == NULL)
411 return (NULL);
412
413 dpindex = RIOM_GET_PRECINDEX(m);
414 for (i = dpindex; i < RIO_NDROPPREC; i++) {
415 if (--rp->rio_precstate[i].qlen == 0) {
416 if (rp->rio_precstate[i].idle == 0) {
417 rp->rio_precstate[i].idle = 1;
418 microtime(&rp->rio_precstate[i].last);
419 }
420 }
421 }
422 return (m);
423 }
424
425 #endif /* ALTQ_RIO */
426