dummynet.c revision 187769
1/*
2 * Copyright (c) 2002-2003 Luigi Rizzo
3 * Copyright (c) 1996 Alex Nash, Paul Traina, Poul-Henning Kamp
4 * Copyright (c) 1994 Ugen J.S.Antsilevich
5 *
6 * Idea and grammar partially left from:
7 * Copyright (c) 1993 Daniel Boulet
8 *
9 * Redistribution and use in source forms, with and without modification,
10 * are permitted provided that this entire comment appears intact.
11 *
12 * Redistribution in binary form may occur without any restrictions.
13 * Obviously, it would be nice if you gave credit where credit is due
14 * but requiring it would be too onerous.
15 *
16 * This software is provided ``AS IS'' without any warranties of any kind.
17 *
18 * NEW command line interface for IP firewall facility
19 *
20 * $FreeBSD: head/sbin/ipfw/dummynet.c 187769 2009-01-27 11:06:59Z luigi $
21 *
22 * dummynet support
23 */
24
25#include <sys/types.h>
26#include <sys/socket.h>
27#include <sys/sockio.h>
28#include <sys/sysctl.h>
29
30#include "ipfw2.h"
31
32#include <ctype.h>
33#include <err.h>
34#include <netdb.h>
35#include <stdio.h>
36#include <stdlib.h>
37#include <string.h>
38#include <sysexits.h>
39
40#include <net/if.h>
41#include <netinet/in.h>
42#include <netinet/in_systm.h>
43#include <netinet/ip.h>
44#include <netinet/ip_fw.h>
45#include <netinet/ip_dummynet.h>
46#include <arpa/inet.h>
47
48static struct _s_x dummynet_params[] = {
49	{ "plr",		TOK_PLR },
50	{ "noerror",		TOK_NOERROR },
51	{ "buckets",		TOK_BUCKETS },
52	{ "dst-ip",		TOK_DSTIP },
53	{ "src-ip",		TOK_SRCIP },
54	{ "dst-port",		TOK_DSTPORT },
55	{ "src-port",		TOK_SRCPORT },
56	{ "proto",		TOK_PROTO },
57	{ "weight",		TOK_WEIGHT },
58	{ "all",		TOK_ALL },
59	{ "mask",		TOK_MASK },
60	{ "droptail",		TOK_DROPTAIL },
61	{ "red",		TOK_RED },
62	{ "gred",		TOK_GRED },
63	{ "bw",			TOK_BW },
64	{ "bandwidth",		TOK_BW },
65	{ "delay",		TOK_DELAY },
66	{ "pipe",		TOK_PIPE },
67	{ "queue",		TOK_QUEUE },
68	{ "flow-id",		TOK_FLOWID},
69	{ "dst-ipv6",		TOK_DSTIP6},
70	{ "dst-ip6",		TOK_DSTIP6},
71	{ "src-ipv6",		TOK_SRCIP6},
72	{ "src-ip6",		TOK_SRCIP6},
73	{ "dummynet-params",	TOK_NULL },
74	{ NULL, 0 }	/* terminator */
75};
76
77static int
78sort_q(const void *pa, const void *pb)
79{
80	int rev = (co.do_sort < 0);
81	int field = rev ? -co.do_sort : co.do_sort;
82	long long res = 0;
83	const struct dn_flow_queue *a = pa;
84	const struct dn_flow_queue *b = pb;
85
86	switch (field) {
87	case 1: /* pkts */
88		res = a->len - b->len;
89		break;
90	case 2: /* bytes */
91		res = a->len_bytes - b->len_bytes;
92		break;
93
94	case 3: /* tot pkts */
95		res = a->tot_pkts - b->tot_pkts;
96		break;
97
98	case 4: /* tot bytes */
99		res = a->tot_bytes - b->tot_bytes;
100		break;
101	}
102	if (res < 0)
103		res = -1;
104	if (res > 0)
105		res = 1;
106	return (int)(rev ? res : -res);
107}
108
109static void
110list_queues(struct dn_flow_set *fs, struct dn_flow_queue *q)
111{
112	int l;
113	int index_printed, indexes = 0;
114	char buff[255];
115	struct protoent *pe;
116
117	if (fs->rq_elements == 0)
118		return;
119
120	if (co.do_sort != 0)
121		heapsort(q, fs->rq_elements, sizeof *q, sort_q);
122
123	/* Print IPv4 flows */
124	index_printed = 0;
125	for (l = 0; l < fs->rq_elements; l++) {
126		struct in_addr ina;
127
128		/* XXX: Should check for IPv4 flows */
129		if (IS_IP6_FLOW_ID(&(q[l].id)))
130			continue;
131
132		if (!index_printed) {
133			index_printed = 1;
134			if (indexes > 0)	/* currently a no-op */
135				printf("\n");
136			indexes++;
137			printf("    "
138			    "mask: 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n",
139			    fs->flow_mask.proto,
140			    fs->flow_mask.src_ip, fs->flow_mask.src_port,
141			    fs->flow_mask.dst_ip, fs->flow_mask.dst_port);
142
143			printf("BKT Prot ___Source IP/port____ "
144			    "____Dest. IP/port____ "
145			    "Tot_pkt/bytes Pkt/Byte Drp\n");
146		}
147
148		printf("%3d ", q[l].hash_slot);
149		pe = getprotobynumber(q[l].id.proto);
150		if (pe)
151			printf("%-4s ", pe->p_name);
152		else
153			printf("%4u ", q[l].id.proto);
154		ina.s_addr = htonl(q[l].id.src_ip);
155		printf("%15s/%-5d ",
156		    inet_ntoa(ina), q[l].id.src_port);
157		ina.s_addr = htonl(q[l].id.dst_ip);
158		printf("%15s/%-5d ",
159		    inet_ntoa(ina), q[l].id.dst_port);
160		printf("%4qu %8qu %2u %4u %3u\n",
161		    q[l].tot_pkts, q[l].tot_bytes,
162		    q[l].len, q[l].len_bytes, q[l].drops);
163		if (co.verbose)
164			printf("   S %20qd  F %20qd\n",
165			    q[l].S, q[l].F);
166	}
167
168	/* Print IPv6 flows */
169	index_printed = 0;
170	for (l = 0; l < fs->rq_elements; l++) {
171		if (!IS_IP6_FLOW_ID(&(q[l].id)))
172			continue;
173
174		if (!index_printed) {
175			index_printed = 1;
176			if (indexes > 0)
177				printf("\n");
178			indexes++;
179			printf("\n        mask: proto: 0x%02x, flow_id: 0x%08x,  ",
180			    fs->flow_mask.proto, fs->flow_mask.flow_id6);
181			inet_ntop(AF_INET6, &(fs->flow_mask.src_ip6),
182			    buff, sizeof(buff));
183			printf("%s/0x%04x -> ", buff, fs->flow_mask.src_port);
184			inet_ntop( AF_INET6, &(fs->flow_mask.dst_ip6),
185			    buff, sizeof(buff) );
186			printf("%s/0x%04x\n", buff, fs->flow_mask.dst_port);
187
188			printf("BKT ___Prot___ _flow-id_ "
189			    "______________Source IPv6/port_______________ "
190			    "_______________Dest. IPv6/port_______________ "
191			    "Tot_pkt/bytes Pkt/Byte Drp\n");
192		}
193		printf("%3d ", q[l].hash_slot);
194		pe = getprotobynumber(q[l].id.proto);
195		if (pe != NULL)
196			printf("%9s ", pe->p_name);
197		else
198			printf("%9u ", q[l].id.proto);
199		printf("%7d  %39s/%-5d ", q[l].id.flow_id6,
200		    inet_ntop(AF_INET6, &(q[l].id.src_ip6), buff, sizeof(buff)),
201		    q[l].id.src_port);
202		printf(" %39s/%-5d ",
203		    inet_ntop(AF_INET6, &(q[l].id.dst_ip6), buff, sizeof(buff)),
204		    q[l].id.dst_port);
205		printf(" %4qu %8qu %2u %4u %3u\n",
206		    q[l].tot_pkts, q[l].tot_bytes,
207		    q[l].len, q[l].len_bytes, q[l].drops);
208		if (co.verbose)
209			printf("   S %20qd  F %20qd\n", q[l].S, q[l].F);
210	}
211}
212
213static void
214print_flowset_parms(struct dn_flow_set *fs, char *prefix)
215{
216	int l;
217	char qs[30];
218	char plr[30];
219	char red[90];	/* Display RED parameters */
220
221	l = fs->qsize;
222	if (fs->flags_fs & DN_QSIZE_IS_BYTES) {
223		if (l >= 8192)
224			sprintf(qs, "%d KB", l / 1024);
225		else
226			sprintf(qs, "%d B", l);
227	} else
228		sprintf(qs, "%3d sl.", l);
229	if (fs->plr)
230		sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff));
231	else
232		plr[0] = '\0';
233	if (fs->flags_fs & DN_IS_RED)	/* RED parameters */
234		sprintf(red,
235		    "\n\t  %cRED w_q %f min_th %d max_th %d max_p %f",
236		    (fs->flags_fs & DN_IS_GENTLE_RED) ? 'G' : ' ',
237		    1.0 * fs->w_q / (double)(1 << SCALE_RED),
238		    SCALE_VAL(fs->min_th),
239		    SCALE_VAL(fs->max_th),
240		    1.0 * fs->max_p / (double)(1 << SCALE_RED));
241	else
242		sprintf(red, "droptail");
243
244	printf("%s %s%s %d queues (%d buckets) %s\n",
245	    prefix, qs, plr, fs->rq_elements, fs->rq_size, red);
246}
247
248void
249ipfw_list_pipes(void *data, uint nbytes, int ac, char *av[])
250{
251	int rulenum;
252	void *next = data;
253	struct dn_pipe *p = (struct dn_pipe *) data;
254	struct dn_flow_set *fs;
255	struct dn_flow_queue *q;
256	int l;
257
258	if (ac > 0)
259		rulenum = strtoul(*av++, NULL, 10);
260	else
261		rulenum = 0;
262	for (; nbytes >= sizeof *p; p = (struct dn_pipe *)next) {
263		double b = p->bandwidth;
264		char buf[30];
265		char prefix[80];
266
267		if (SLIST_NEXT(p, next) != (struct dn_pipe *)DN_IS_PIPE)
268			break;	/* done with pipes, now queues */
269
270		/*
271		 * compute length, as pipe have variable size
272		 */
273		l = sizeof(*p) + p->fs.rq_elements * sizeof(*q);
274		next = (char *)p + l;
275		nbytes -= l;
276
277		if ((rulenum != 0 && rulenum != p->pipe_nr) || co.do_pipe == 2)
278			continue;
279
280		/*
281		 * Print rate (or clocking interface)
282		 */
283		if (p->if_name[0] != '\0')
284			sprintf(buf, "%s", p->if_name);
285		else if (b == 0)
286			sprintf(buf, "unlimited");
287		else if (b >= 1000000)
288			sprintf(buf, "%7.3f Mbit/s", b/1000000);
289		else if (b >= 1000)
290			sprintf(buf, "%7.3f Kbit/s", b/1000);
291		else
292			sprintf(buf, "%7.3f bit/s ", b);
293
294		sprintf(prefix, "%05d: %s %4d ms ",
295		    p->pipe_nr, buf, p->delay);
296		print_flowset_parms(&(p->fs), prefix);
297		if (co.verbose)
298			printf("   V %20qd\n", p->V >> MY_M);
299
300		q = (struct dn_flow_queue *)(p+1);
301		list_queues(&(p->fs), q);
302	}
303	for (fs = next; nbytes >= sizeof *fs; fs = next) {
304		char prefix[80];
305
306		if (SLIST_NEXT(fs, next) != (struct dn_flow_set *)DN_IS_QUEUE)
307			break;
308		l = sizeof(*fs) + fs->rq_elements * sizeof(*q);
309		next = (char *)fs + l;
310		nbytes -= l;
311
312		if (rulenum != 0 && ((rulenum != fs->fs_nr && co.do_pipe == 2) ||
313		    (rulenum != fs->parent_nr && co.do_pipe == 1))) {
314			continue;
315		}
316
317		q = (struct dn_flow_queue *)(fs+1);
318		sprintf(prefix, "q%05d: weight %d pipe %d ",
319		    fs->fs_nr, fs->weight, fs->parent_nr);
320		print_flowset_parms(fs, prefix);
321		list_queues(fs, q);
322	}
323}
324
325/*
326 * Delete pipe or queue i
327 */
328int
329ipfw_delete_pipe(int pipe_or_queue, int n)
330{
331	struct dn_pipe p;
332	int i;
333
334	memset(&p, 0, sizeof p);
335	if (pipe_or_queue == 1)
336		p.pipe_nr = i;		/* pipe */
337	else
338		p.fs.fs_nr = i;		/* queue */
339	i = do_cmd(IP_DUMMYNET_DEL, &p, sizeof p);
340	if (i) {
341		i = 1;
342		warn("rule %u: setsockopt(IP_DUMMYNET_DEL)", i);
343	}
344	return i;
345}
346
347void
348ipfw_config_pipe(int ac, char **av)
349{
350	struct dn_pipe p;
351	int i;
352	char *end;
353	void *par = NULL;
354
355	memset(&p, 0, sizeof p);
356
357	av++; ac--;
358	/* Pipe number */
359	if (ac && isdigit(**av)) {
360		i = atoi(*av); av++; ac--;
361		if (co.do_pipe == 1)
362			p.pipe_nr = i;
363		else
364			p.fs.fs_nr = i;
365	}
366	while (ac > 0) {
367		double d;
368		int tok = match_token(dummynet_params, *av);
369		ac--; av++;
370
371		switch(tok) {
372		case TOK_NOERROR:
373			p.fs.flags_fs |= DN_NOERROR;
374			break;
375
376		case TOK_PLR:
377			NEED1("plr needs argument 0..1\n");
378			d = strtod(av[0], NULL);
379			if (d > 1)
380				d = 1;
381			else if (d < 0)
382				d = 0;
383			p.fs.plr = (int)(d*0x7fffffff);
384			ac--; av++;
385			break;
386
387		case TOK_QUEUE:
388			NEED1("queue needs queue size\n");
389			end = NULL;
390			p.fs.qsize = strtoul(av[0], &end, 0);
391			if (*end == 'K' || *end == 'k') {
392				p.fs.flags_fs |= DN_QSIZE_IS_BYTES;
393				p.fs.qsize *= 1024;
394			} else if (*end == 'B' ||
395			    _substrcmp2(end, "by", "bytes") == 0) {
396				p.fs.flags_fs |= DN_QSIZE_IS_BYTES;
397			}
398			ac--; av++;
399			break;
400
401		case TOK_BUCKETS:
402			NEED1("buckets needs argument\n");
403			p.fs.rq_size = strtoul(av[0], NULL, 0);
404			ac--; av++;
405			break;
406
407		case TOK_MASK:
408			NEED1("mask needs mask specifier\n");
409			/*
410			 * per-flow queue, mask is dst_ip, dst_port,
411			 * src_ip, src_port, proto measured in bits
412			 */
413			par = NULL;
414
415			bzero(&p.fs.flow_mask, sizeof(p.fs.flow_mask));
416			end = NULL;
417
418			while (ac >= 1) {
419			    uint32_t *p32 = NULL;
420			    uint16_t *p16 = NULL;
421			    uint32_t *p20 = NULL;
422			    struct in6_addr *pa6 = NULL;
423			    uint32_t a;
424
425			    tok = match_token(dummynet_params, *av);
426			    ac--; av++;
427			    switch(tok) {
428			    case TOK_ALL:
429				    /*
430				     * special case, all bits significant
431				     */
432				    p.fs.flow_mask.dst_ip = ~0;
433				    p.fs.flow_mask.src_ip = ~0;
434				    p.fs.flow_mask.dst_port = ~0;
435				    p.fs.flow_mask.src_port = ~0;
436				    p.fs.flow_mask.proto = ~0;
437				    n2mask(&(p.fs.flow_mask.dst_ip6), 128);
438				    n2mask(&(p.fs.flow_mask.src_ip6), 128);
439				    p.fs.flow_mask.flow_id6 = ~0;
440				    p.fs.flags_fs |= DN_HAVE_FLOW_MASK;
441				    goto end_mask;
442
443			    case TOK_DSTIP:
444				    p32 = &p.fs.flow_mask.dst_ip;
445				    break;
446
447			    case TOK_SRCIP:
448				    p32 = &p.fs.flow_mask.src_ip;
449				    break;
450
451			    case TOK_DSTIP6:
452				    pa6 = &(p.fs.flow_mask.dst_ip6);
453				    break;
454
455			    case TOK_SRCIP6:
456				    pa6 = &(p.fs.flow_mask.src_ip6);
457				    break;
458
459			    case TOK_FLOWID:
460				    p20 = &p.fs.flow_mask.flow_id6;
461				    break;
462
463			    case TOK_DSTPORT:
464				    p16 = &p.fs.flow_mask.dst_port;
465				    break;
466
467			    case TOK_SRCPORT:
468				    p16 = &p.fs.flow_mask.src_port;
469				    break;
470
471			    case TOK_PROTO:
472				    break;
473
474			    default:
475				    ac++; av--; /* backtrack */
476				    goto end_mask;
477			    }
478			    if (ac < 1)
479				    errx(EX_USAGE, "mask: value missing");
480			    if (*av[0] == '/') {
481				    a = strtoul(av[0]+1, &end, 0);
482				    if (pa6 == NULL)
483					    a = (a == 32) ? ~0 : (1 << a) - 1;
484			    } else
485				    a = strtoul(av[0], &end, 0);
486			    if (p32 != NULL)
487				    *p32 = a;
488			    else if (p16 != NULL) {
489				    if (a > 0xFFFF)
490					    errx(EX_DATAERR,
491						"port mask must be 16 bit");
492				    *p16 = (uint16_t)a;
493			    } else if (p20 != NULL) {
494				    if (a > 0xfffff)
495					errx(EX_DATAERR,
496					    "flow_id mask must be 20 bit");
497				    *p20 = (uint32_t)a;
498			    } else if (pa6 != NULL) {
499				    if (a > 128)
500					errx(EX_DATAERR,
501					    "in6addr invalid mask len");
502				    else
503					n2mask(pa6, a);
504			    } else {
505				    if (a > 0xFF)
506					    errx(EX_DATAERR,
507						"proto mask must be 8 bit");
508				    p.fs.flow_mask.proto = (uint8_t)a;
509			    }
510			    if (a != 0)
511				    p.fs.flags_fs |= DN_HAVE_FLOW_MASK;
512			    ac--; av++;
513			} /* end while, config masks */
514end_mask:
515			break;
516
517		case TOK_RED:
518		case TOK_GRED:
519			NEED1("red/gred needs w_q/min_th/max_th/max_p\n");
520			p.fs.flags_fs |= DN_IS_RED;
521			if (tok == TOK_GRED)
522				p.fs.flags_fs |= DN_IS_GENTLE_RED;
523			/*
524			 * the format for parameters is w_q/min_th/max_th/max_p
525			 */
526			if ((end = strsep(&av[0], "/"))) {
527			    double w_q = strtod(end, NULL);
528			    if (w_q > 1 || w_q <= 0)
529				errx(EX_DATAERR, "0 < w_q <= 1");
530			    p.fs.w_q = (int) (w_q * (1 << SCALE_RED));
531			}
532			if ((end = strsep(&av[0], "/"))) {
533			    p.fs.min_th = strtoul(end, &end, 0);
534			    if (*end == 'K' || *end == 'k')
535				p.fs.min_th *= 1024;
536			}
537			if ((end = strsep(&av[0], "/"))) {
538			    p.fs.max_th = strtoul(end, &end, 0);
539			    if (*end == 'K' || *end == 'k')
540				p.fs.max_th *= 1024;
541			}
542			if ((end = strsep(&av[0], "/"))) {
543			    double max_p = strtod(end, NULL);
544			    if (max_p > 1 || max_p <= 0)
545				errx(EX_DATAERR, "0 < max_p <= 1");
546			    p.fs.max_p = (int)(max_p * (1 << SCALE_RED));
547			}
548			ac--; av++;
549			break;
550
551		case TOK_DROPTAIL:
552			p.fs.flags_fs &= ~(DN_IS_RED|DN_IS_GENTLE_RED);
553			break;
554
555		case TOK_BW:
556			NEED1("bw needs bandwidth or interface\n");
557			if (co.do_pipe != 1)
558			    errx(EX_DATAERR, "bandwidth only valid for pipes");
559			/*
560			 * set clocking interface or bandwidth value
561			 */
562			if (av[0][0] >= 'a' && av[0][0] <= 'z') {
563			    int l = sizeof(p.if_name)-1;
564			    /* interface name */
565			    strncpy(p.if_name, av[0], l);
566			    p.if_name[l] = '\0';
567			    p.bandwidth = 0;
568			} else {
569			    p.if_name[0] = '\0';
570			    p.bandwidth = strtoul(av[0], &end, 0);
571			    if (*end == 'K' || *end == 'k') {
572				end++;
573				p.bandwidth *= 1000;
574			    } else if (*end == 'M') {
575				end++;
576				p.bandwidth *= 1000000;
577			    }
578			    if ((*end == 'B' &&
579				  _substrcmp2(end, "Bi", "Bit/s") != 0) ||
580			        _substrcmp2(end, "by", "bytes") == 0)
581				p.bandwidth *= 8;
582			    if (p.bandwidth < 0)
583				errx(EX_DATAERR, "bandwidth too large");
584			}
585			ac--; av++;
586			break;
587
588		case TOK_DELAY:
589			if (co.do_pipe != 1)
590				errx(EX_DATAERR, "delay only valid for pipes");
591			NEED1("delay needs argument 0..10000ms\n");
592			p.delay = strtoul(av[0], NULL, 0);
593			ac--; av++;
594			break;
595
596		case TOK_WEIGHT:
597			if (co.do_pipe == 1)
598				errx(EX_DATAERR,"weight only valid for queues");
599			NEED1("weight needs argument 0..100\n");
600			p.fs.weight = strtoul(av[0], &end, 0);
601			ac--; av++;
602			break;
603
604		case TOK_PIPE:
605			if (co.do_pipe == 1)
606				errx(EX_DATAERR,"pipe only valid for queues");
607			NEED1("pipe needs pipe_number\n");
608			p.fs.parent_nr = strtoul(av[0], &end, 0);
609			ac--; av++;
610			break;
611
612		default:
613			errx(EX_DATAERR, "unrecognised option ``%s''", av[-1]);
614		}
615	}
616	if (co.do_pipe == 1) {
617		if (p.pipe_nr == 0)
618			errx(EX_DATAERR, "pipe_nr must be > 0");
619		if (p.delay > 10000)
620			errx(EX_DATAERR, "delay must be < 10000");
621	} else { /* co.do_pipe == 2, queue */
622		if (p.fs.parent_nr == 0)
623			errx(EX_DATAERR, "pipe must be > 0");
624		if (p.fs.weight >100)
625			errx(EX_DATAERR, "weight must be <= 100");
626	}
627	if (p.fs.flags_fs & DN_QSIZE_IS_BYTES) {
628		size_t len;
629		long limit;
630
631		len = sizeof(limit);
632		if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit",
633			&limit, &len, NULL, 0) == -1)
634			limit = 1024*1024;
635		if (p.fs.qsize > limit)
636			errx(EX_DATAERR, "queue size must be < %ldB", limit);
637	} else {
638		size_t len;
639		long limit;
640
641		len = sizeof(limit);
642		if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit",
643			&limit, &len, NULL, 0) == -1)
644			limit = 100;
645		if (p.fs.qsize > limit)
646			errx(EX_DATAERR, "2 <= queue size <= %ld", limit);
647	}
648	if (p.fs.flags_fs & DN_IS_RED) {
649		size_t len;
650		int lookup_depth, avg_pkt_size;
651		double s, idle, weight, w_q;
652		struct clockinfo ck;
653		int t;
654
655		if (p.fs.min_th >= p.fs.max_th)
656		    errx(EX_DATAERR, "min_th %d must be < than max_th %d",
657			p.fs.min_th, p.fs.max_th);
658		if (p.fs.max_th == 0)
659		    errx(EX_DATAERR, "max_th must be > 0");
660
661		len = sizeof(int);
662		if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth",
663			&lookup_depth, &len, NULL, 0) == -1)
664		    errx(1, "sysctlbyname(\"%s\")",
665			"net.inet.ip.dummynet.red_lookup_depth");
666		if (lookup_depth == 0)
667		    errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth"
668			" must be greater than zero");
669
670		len = sizeof(int);
671		if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size",
672			&avg_pkt_size, &len, NULL, 0) == -1)
673
674		    errx(1, "sysctlbyname(\"%s\")",
675			"net.inet.ip.dummynet.red_avg_pkt_size");
676		if (avg_pkt_size == 0)
677			errx(EX_DATAERR,
678			    "net.inet.ip.dummynet.red_avg_pkt_size must"
679			    " be greater than zero");
680
681		len = sizeof(struct clockinfo);
682		if (sysctlbyname("kern.clockrate", &ck, &len, NULL, 0) == -1)
683			errx(1, "sysctlbyname(\"%s\")", "kern.clockrate");
684
685		/*
686		 * Ticks needed for sending a medium-sized packet.
687		 * Unfortunately, when we are configuring a WF2Q+ queue, we
688		 * do not have bandwidth information, because that is stored
689		 * in the parent pipe, and also we have multiple queues
690		 * competing for it. So we set s=0, which is not very
691		 * correct. But on the other hand, why do we want RED with
692		 * WF2Q+ ?
693		 */
694		if (p.bandwidth==0) /* this is a WF2Q+ queue */
695			s = 0;
696		else
697			s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth;
698
699		/*
700		 * max idle time (in ticks) before avg queue size becomes 0.
701		 * NOTA:  (3/w_q) is approx the value x so that
702		 * (1-w_q)^x < 10^-3.
703		 */
704		w_q = ((double)p.fs.w_q) / (1 << SCALE_RED);
705		idle = s * 3. / w_q;
706		p.fs.lookup_step = (int)idle / lookup_depth;
707		if (!p.fs.lookup_step)
708			p.fs.lookup_step = 1;
709		weight = 1 - w_q;
710		for (t = p.fs.lookup_step; t > 1; --t)
711			weight *= 1 - w_q;
712		p.fs.lookup_weight = (int)(weight * (1 << SCALE_RED));
713	}
714	i = do_cmd(IP_DUMMYNET_CONFIGURE, &p, sizeof p);
715	if (i)
716		err(1, "setsockopt(%s)", "IP_DUMMYNET_CONFIGURE");
717}
718