xref: /freebsd/sbin/ipfw/dummynet.c (revision b1f9167f94059fd55c630891d359bcff987bd7eb)
1 /*
2  * Copyright (c) 2002-2003,2010 Luigi Rizzo
3  *
4  * Redistribution and use in source forms, with and without modification,
5  * are permitted provided that this entire comment appears intact.
6  *
7  * Redistribution in binary form may occur without any restrictions.
8  * Obviously, it would be nice if you gave credit where credit is due
9  * but requiring it would be too onerous.
10  *
11  * This software is provided ``AS IS'' without any warranties of any kind.
12  *
13  * $FreeBSD$
14  *
15  * dummynet support
16  */
17 
18 #include <sys/types.h>
19 #include <sys/socket.h>
20 /* XXX there are several sysctl leftover here */
21 #include <sys/sysctl.h>
22 
23 #include "ipfw2.h"
24 
25 #include <ctype.h>
26 #include <err.h>
27 #include <errno.h>
28 #include <libutil.h>
29 #include <netdb.h>
30 #include <stdio.h>
31 #include <stdlib.h>
32 #include <string.h>
33 #include <sysexits.h>
34 
35 #include <net/if.h>
36 #include <netinet/in.h>
37 #include <netinet/ip_fw.h>
38 #include <netinet/ip_dummynet.h>
39 #include <arpa/inet.h>	/* inet_ntoa */
40 
41 
42 static struct _s_x dummynet_params[] = {
43 	{ "plr",		TOK_PLR },
44 	{ "noerror",		TOK_NOERROR },
45 	{ "buckets",		TOK_BUCKETS },
46 	{ "dst-ip",		TOK_DSTIP },
47 	{ "src-ip",		TOK_SRCIP },
48 	{ "dst-port",		TOK_DSTPORT },
49 	{ "src-port",		TOK_SRCPORT },
50 	{ "proto",		TOK_PROTO },
51 	{ "weight",		TOK_WEIGHT },
52 	{ "lmax",		TOK_LMAX },
53 	{ "maxlen",		TOK_LMAX },
54 	{ "all",		TOK_ALL },
55 	{ "mask",		TOK_MASK }, /* alias for both */
56 	{ "sched_mask",		TOK_SCHED_MASK },
57 	{ "flow_mask",		TOK_FLOW_MASK },
58 	{ "droptail",		TOK_DROPTAIL },
59 	{ "ecn",		TOK_ECN },
60 	{ "red",		TOK_RED },
61 	{ "gred",		TOK_GRED },
62 	{ "bw",			TOK_BW },
63 	{ "bandwidth",		TOK_BW },
64 	{ "delay",		TOK_DELAY },
65 	{ "link",		TOK_LINK },
66 	{ "pipe",		TOK_PIPE },
67 	{ "queue",		TOK_QUEUE },
68 	{ "flowset",		TOK_FLOWSET },
69 	{ "sched",		TOK_SCHED },
70 	{ "pri",		TOK_PRI },
71 	{ "priority",		TOK_PRI },
72 	{ "type",		TOK_TYPE },
73 	{ "flow-id",		TOK_FLOWID},
74 	{ "dst-ipv6",		TOK_DSTIP6},
75 	{ "dst-ip6",		TOK_DSTIP6},
76 	{ "src-ipv6",		TOK_SRCIP6},
77 	{ "src-ip6",		TOK_SRCIP6},
78 	{ "profile",		TOK_PROFILE},
79 	{ "burst",		TOK_BURST},
80 	{ "dummynet-params",	TOK_NULL },
81 	{ NULL, 0 }	/* terminator */
82 };
83 
84 #define O_NEXT(p, len) ((void *)((char *)p + len))
85 
86 static void
87 oid_fill(struct dn_id *oid, int len, int type, uintptr_t id)
88 {
89 	oid->len = len;
90 	oid->type = type;
91 	oid->subtype = 0;
92 	oid->id = id;
93 }
94 
95 /* make room in the buffer and move the pointer forward */
96 static void *
97 o_next(struct dn_id **o, int len, int type)
98 {
99 	struct dn_id *ret = *o;
100 	oid_fill(ret, len, type, 0);
101 	*o = O_NEXT(*o, len);
102 	return ret;
103 }
104 
105 #if 0
106 static int
107 sort_q(void *arg, const void *pa, const void *pb)
108 {
109 	int rev = (co.do_sort < 0);
110 	int field = rev ? -co.do_sort : co.do_sort;
111 	long long res = 0;
112 	const struct dn_flow_queue *a = pa;
113 	const struct dn_flow_queue *b = pb;
114 
115 	switch (field) {
116 	case 1: /* pkts */
117 		res = a->len - b->len;
118 		break;
119 	case 2: /* bytes */
120 		res = a->len_bytes - b->len_bytes;
121 		break;
122 
123 	case 3: /* tot pkts */
124 		res = a->tot_pkts - b->tot_pkts;
125 		break;
126 
127 	case 4: /* tot bytes */
128 		res = a->tot_bytes - b->tot_bytes;
129 		break;
130 	}
131 	if (res < 0)
132 		res = -1;
133 	if (res > 0)
134 		res = 1;
135 	return (int)(rev ? res : -res);
136 }
137 #endif
138 
139 /* print a mask and header for the subsequent list of flows */
140 static void
141 print_mask(struct ipfw_flow_id *id)
142 {
143 	if (!IS_IP6_FLOW_ID(id)) {
144 		printf("    "
145 		    "mask: %s 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n",
146 		    id->extra ? "queue," : "",
147 		    id->proto,
148 		    id->src_ip, id->src_port,
149 		    id->dst_ip, id->dst_port);
150 	} else {
151 		char buf[255];
152 		printf("\n        mask: %sproto: 0x%02x, flow_id: 0x%08x,  ",
153 		    id->extra ? "queue," : "",
154 		    id->proto, id->flow_id6);
155 		inet_ntop(AF_INET6, &(id->src_ip6), buf, sizeof(buf));
156 		printf("%s/0x%04x -> ", buf, id->src_port);
157 		inet_ntop(AF_INET6, &(id->dst_ip6), buf, sizeof(buf));
158 		printf("%s/0x%04x\n", buf, id->dst_port);
159 	}
160 }
161 
162 static void
163 print_header(struct ipfw_flow_id *id)
164 {
165 	if (!IS_IP6_FLOW_ID(id))
166 		printf("BKT Prot ___Source IP/port____ "
167 		    "____Dest. IP/port____ "
168 		    "Tot_pkt/bytes Pkt/Byte Drp\n");
169 	else
170 		printf("BKT ___Prot___ _flow-id_ "
171 		    "______________Source IPv6/port_______________ "
172 		    "_______________Dest. IPv6/port_______________ "
173 		    "Tot_pkt/bytes Pkt/Byte Drp\n");
174 }
175 
176 static void
177 list_flow(struct dn_flow *ni, int *print)
178 {
179 	char buff[255];
180 	struct protoent *pe = NULL;
181 	struct in_addr ina;
182 	struct ipfw_flow_id *id = &ni->fid;
183 
184 	if (*print) {
185 		print_header(&ni->fid);
186 		*print = 0;
187 	}
188 	pe = getprotobynumber(id->proto);
189 		/* XXX: Should check for IPv4 flows */
190 	printf("%3u%c", (ni->oid.id) & 0xff,
191 		id->extra ? '*' : ' ');
192 	if (!IS_IP6_FLOW_ID(id)) {
193 		if (pe)
194 			printf("%-4s ", pe->p_name);
195 		else
196 			printf("%4u ", id->proto);
197 		ina.s_addr = htonl(id->src_ip);
198 		printf("%15s/%-5d ",
199 		    inet_ntoa(ina), id->src_port);
200 		ina.s_addr = htonl(id->dst_ip);
201 		printf("%15s/%-5d ",
202 		    inet_ntoa(ina), id->dst_port);
203 	} else {
204 		/* Print IPv6 flows */
205 		if (pe != NULL)
206 			printf("%9s ", pe->p_name);
207 		else
208 			printf("%9u ", id->proto);
209 		printf("%7d  %39s/%-5d ", id->flow_id6,
210 		    inet_ntop(AF_INET6, &(id->src_ip6), buff, sizeof(buff)),
211 		    id->src_port);
212 		printf(" %39s/%-5d ",
213 		    inet_ntop(AF_INET6, &(id->dst_ip6), buff, sizeof(buff)),
214 		    id->dst_port);
215 	}
216 	pr_u64(&ni->tot_pkts, 4);
217 	pr_u64(&ni->tot_bytes, 8);
218 	printf("%2u %4u %3u\n",
219 	    ni->length, ni->len_bytes, ni->drops);
220 }
221 
222 static void
223 print_flowset_parms(struct dn_fs *fs, char *prefix)
224 {
225 	int l;
226 	char qs[30];
227 	char plr[30];
228 	char red[90];	/* Display RED parameters */
229 
230 	l = fs->qsize;
231 	if (fs->flags & DN_QSIZE_BYTES) {
232 		if (l >= 8192)
233 			sprintf(qs, "%d KB", l / 1024);
234 		else
235 			sprintf(qs, "%d B", l);
236 	} else
237 		sprintf(qs, "%3d sl.", l);
238 	if (fs->plr)
239 		sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff));
240 	else
241 		plr[0] = '\0';
242 
243 	if (fs->flags & DN_IS_RED) {	/* RED parameters */
244 		sprintf(red,
245 		    "\n\t %cRED w_q %f min_th %d max_th %d max_p %f",
246 		    (fs->flags & DN_IS_GENTLE_RED) ? 'G' : ' ',
247 		    1.0 * fs->w_q / (double)(1 << SCALE_RED),
248 		    fs->min_th,
249 		    fs->max_th,
250 		    1.0 * fs->max_p / (double)(1 << SCALE_RED));
251 		if (fs->flags & DN_IS_ECN)
252 			strncat(red, " (ecn)", 6);
253 	} else
254 		sprintf(red, "droptail");
255 
256 	if (prefix[0]) {
257 	    printf("%s %s%s %d queues (%d buckets) %s\n",
258 		prefix, qs, plr, fs->oid.id, fs->buckets, red);
259 	    prefix[0] = '\0';
260 	} else {
261 	    printf("q%05d %s%s %d flows (%d buckets) sched %d "
262 			"weight %d lmax %d pri %d %s\n",
263 		fs->fs_nr, qs, plr, fs->oid.id, fs->buckets,
264 		fs->sched_nr, fs->par[0], fs->par[1], fs->par[2], red);
265 	    if (fs->flags & DN_HAVE_MASK)
266 		print_mask(&fs->flow_mask);
267 	}
268 }
269 
270 static void
271 print_extra_delay_parms(struct dn_profile *p)
272 {
273 	double loss;
274 	if (p->samples_no <= 0)
275 		return;
276 
277 	loss = p->loss_level;
278 	loss /= p->samples_no;
279 	printf("\t profile: name \"%s\" loss %f samples %d\n",
280 		p->name, loss, p->samples_no);
281 }
282 
283 static void
284 flush_buf(char *buf)
285 {
286 	if (buf[0])
287 		printf("%s\n", buf);
288 	buf[0] = '\0';
289 }
290 
291 /*
292  * generic list routine. We expect objects in a specific order, i.e.
293  * PIPES AND SCHEDULERS:
294  *	link; scheduler; internal flowset if any; instances
295  * we can tell a pipe from the number.
296  *
297  * FLOWSETS:
298  *	flowset; queues;
299  * link i (int queue); scheduler i; si(i) { flowsets() : queues }
300  */
301 static void
302 list_pipes(struct dn_id *oid, struct dn_id *end)
303 {
304     char buf[160];	/* pending buffer */
305     int toPrint = 1;	/* print header */
306 
307     buf[0] = '\0';
308     for (; oid != end; oid = O_NEXT(oid, oid->len)) {
309 	if (oid->len < sizeof(*oid))
310 		errx(1, "invalid oid len %d\n", oid->len);
311 
312 	switch (oid->type) {
313 	default:
314 	    flush_buf(buf);
315 	    printf("unrecognized object %d size %d\n", oid->type, oid->len);
316 	    break;
317 	case DN_TEXT: /* list of attached flowsets */
318 	    {
319 		int i, l;
320 		struct {
321 			struct dn_id id;
322 			uint32_t p[0];
323 		} *d = (void *)oid;
324 		l = (oid->len - sizeof(*oid))/sizeof(d->p[0]);
325 		if (l == 0)
326 		    break;
327 		printf("   Children flowsets: ");
328 		for (i = 0; i < l; i++)
329 			printf("%u ", d->p[i]);
330 		printf("\n");
331 		break;
332 	    }
333 	case DN_CMD_GET:
334 	    if (co.verbose)
335 		printf("answer for cmd %d, len %d\n", oid->type, oid->id);
336 	    break;
337 	case DN_SCH: {
338 	    struct dn_sch *s = (struct dn_sch *)oid;
339 	    flush_buf(buf);
340 	    printf(" sched %d type %s flags 0x%x %d buckets %d active\n",
341 			s->sched_nr,
342 			s->name, s->flags, s->buckets, s->oid.id);
343 	    if (s->flags & DN_HAVE_MASK)
344 		print_mask(&s->sched_mask);
345 	    }
346 	    break;
347 
348 	case DN_FLOW:
349 	    list_flow((struct dn_flow *)oid, &toPrint);
350 	    break;
351 
352 	case DN_LINK: {
353 	    struct dn_link *p = (struct dn_link *)oid;
354 	    double b = p->bandwidth;
355 	    char bwbuf[30];
356 	    char burst[5 + 7];
357 
358 	    /* This starts a new object so flush buffer */
359 	    flush_buf(buf);
360 	    /* data rate */
361 	    if (b == 0)
362 		sprintf(bwbuf, "unlimited     ");
363 	    else if (b >= 1000000)
364 		sprintf(bwbuf, "%7.3f Mbit/s", b/1000000);
365 	    else if (b >= 1000)
366 		sprintf(bwbuf, "%7.3f Kbit/s", b/1000);
367 	    else
368 		sprintf(bwbuf, "%7.3f bit/s ", b);
369 
370 	    if (humanize_number(burst, sizeof(burst), p->burst,
371 		    "", HN_AUTOSCALE, 0) < 0 || co.verbose)
372 		sprintf(burst, "%d", (int)p->burst);
373 	    sprintf(buf, "%05d: %s %4d ms burst %s",
374 		p->link_nr % DN_MAX_ID, bwbuf, p->delay, burst);
375 	    }
376 	    break;
377 
378 	case DN_FS:
379 	    print_flowset_parms((struct dn_fs *)oid, buf);
380 	    break;
381 	case DN_PROFILE:
382 	    flush_buf(buf);
383 	    print_extra_delay_parms((struct dn_profile *)oid);
384 	}
385 	flush_buf(buf); // XXX does it really go here ?
386     }
387 }
388 
389 /*
390  * Delete pipe, queue or scheduler i
391  */
392 int
393 ipfw_delete_pipe(int do_pipe, int i)
394 {
395 	struct {
396 		struct dn_id oid;
397 		uintptr_t a[1];	/* add more if we want a list */
398 	} cmd;
399 	oid_fill((void *)&cmd, sizeof(cmd), DN_CMD_DELETE, DN_API_VERSION);
400 	cmd.oid.subtype = (do_pipe == 1) ? DN_LINK :
401 		( (do_pipe == 2) ? DN_FS : DN_SCH);
402 	cmd.a[0] = i;
403 	i = do_cmd(IP_DUMMYNET3, &cmd, cmd.oid.len);
404 	if (i) {
405 		i = 1;
406 		warn("rule %u: setsockopt(IP_DUMMYNET_DEL)", i);
407 	}
408 	return i;
409 }
410 
411 /*
412  * Code to parse delay profiles.
413  *
414  * Some link types introduce extra delays in the transmission
415  * of a packet, e.g. because of MAC level framing, contention on
416  * the use of the channel, MAC level retransmissions and so on.
417  * From our point of view, the channel is effectively unavailable
418  * for this extra time, which is constant or variable depending
419  * on the link type. Additionally, packets may be dropped after this
420  * time (e.g. on a wireless link after too many retransmissions).
421  * We can model the additional delay with an empirical curve
422  * that represents its distribution.
423  *
424  *      cumulative probability
425  *      1.0 ^
426  *          |
427  *      L   +-- loss-level          x
428  *          |                 ******
429  *          |                *
430  *          |           *****
431  *          |          *
432  *          |        **
433  *          |       *
434  *          +-------*------------------->
435  *                      delay
436  *
437  * The empirical curve may have both vertical and horizontal lines.
438  * Vertical lines represent constant delay for a range of
439  * probabilities; horizontal lines correspond to a discontinuty
440  * in the delay distribution: the link will use the largest delay
441  * for a given probability.
442  *
443  * To pass the curve to dummynet, we must store the parameters
444  * in a file as described below, and issue the command
445  *
446  *      ipfw pipe <n> config ... bw XXX profile <filename> ...
447  *
448  * The file format is the following, with whitespace acting as
449  * a separator and '#' indicating the beginning a comment:
450  *
451  *	samples N
452  *		the number of samples used in the internal
453  *		representation (2..1024; default 100);
454  *
455  *	loss-level L
456  *		The probability above which packets are lost.
457  *	       (0.0 <= L <= 1.0, default 1.0 i.e. no loss);
458  *
459  *	name identifier
460  *		Optional a name (listed by "ipfw pipe show")
461  *		to identify the distribution;
462  *
463  *	"delay prob" | "prob delay"
464  *		One of these two lines is mandatory and defines
465  *		the format of the following lines with data points.
466  *
467  *	XXX YYY
468  *		2 or more lines representing points in the curve,
469  *		with either delay or probability first, according
470  *		to the chosen format.
471  *		The unit for delay is milliseconds.
472  *
473  * Data points does not need to be ordered or equal to the number
474  * specified in the "samples" line. ipfw will sort and interpolate
475  * the curve as needed.
476  *
477  * Example of a profile file:
478 
479 	name    bla_bla_bla
480 	samples 100
481 	loss-level    0.86
482 	prob    delay
483 	0       200	# minimum overhead is 200ms
484 	0.5     200
485 	0.5     300
486 	0.8     1000
487 	0.9     1300
488 	1       1300
489 
490  * Internally, we will convert the curve to a fixed number of
491  * samples, and when it is time to transmit a packet we will
492  * model the extra delay as extra bits in the packet.
493  *
494  */
495 
496 #define ED_MAX_LINE_LEN	256+ED_MAX_NAME_LEN
497 #define ED_TOK_SAMPLES	"samples"
498 #define ED_TOK_LOSS	"loss-level"
499 #define ED_TOK_NAME	"name"
500 #define ED_TOK_DELAY	"delay"
501 #define ED_TOK_PROB	"prob"
502 #define ED_TOK_BW	"bw"
503 #define ED_SEPARATORS	" \t\n"
504 #define ED_MIN_SAMPLES_NO	2
505 
506 /*
507  * returns 1 if s is a non-negative number, with at least one '.'
508  */
509 static int
510 is_valid_number(const char *s)
511 {
512 	int i, dots_found = 0;
513 	int len = strlen(s);
514 
515 	for (i = 0; i<len; ++i)
516 		if (!isdigit(s[i]) && (s[i] !='.' || ++dots_found > 1))
517 			return 0;
518 	return 1;
519 }
520 
521 /*
522  * Take as input a string describing a bandwidth value
523  * and return the numeric bandwidth value.
524  * set clocking interface or bandwidth value
525  */
526 static void
527 read_bandwidth(char *arg, int *bandwidth, char *if_name, int namelen)
528 {
529 	if (*bandwidth != -1)
530 		warnx("duplicate token, override bandwidth value!");
531 
532 	if (arg[0] >= 'a' && arg[0] <= 'z') {
533 		if (!if_name) {
534 			errx(1, "no if support");
535 		}
536 		if (namelen >= IFNAMSIZ)
537 			warn("interface name truncated");
538 		namelen--;
539 		/* interface name */
540 		strncpy(if_name, arg, namelen);
541 		if_name[namelen] = '\0';
542 		*bandwidth = 0;
543 	} else {	/* read bandwidth value */
544 		int bw;
545 		char *end = NULL;
546 
547 		bw = strtoul(arg, &end, 0);
548 		if (*end == 'K' || *end == 'k') {
549 			end++;
550 			bw *= 1000;
551 		} else if (*end == 'M' || *end == 'm') {
552 			end++;
553 			bw *= 1000000;
554 		}
555 		if ((*end == 'B' &&
556 			_substrcmp2(end, "Bi", "Bit/s") != 0) ||
557 		    _substrcmp2(end, "by", "bytes") == 0)
558 			bw *= 8;
559 
560 		if (bw < 0)
561 			errx(EX_DATAERR, "bandwidth too large");
562 
563 		*bandwidth = bw;
564 		if (if_name)
565 			if_name[0] = '\0';
566 	}
567 }
568 
569 struct point {
570 	double prob;
571 	double delay;
572 };
573 
574 static int
575 compare_points(const void *vp1, const void *vp2)
576 {
577 	const struct point *p1 = vp1;
578 	const struct point *p2 = vp2;
579 	double res = 0;
580 
581 	res = p1->prob - p2->prob;
582 	if (res == 0)
583 		res = p1->delay - p2->delay;
584 	if (res < 0)
585 		return -1;
586 	else if (res > 0)
587 		return 1;
588 	else
589 		return 0;
590 }
591 
592 #define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno
593 
594 static void
595 load_extra_delays(const char *filename, struct dn_profile *p,
596 	struct dn_link *link)
597 {
598 	char    line[ED_MAX_LINE_LEN];
599 	FILE    *f;
600 	int     lineno = 0;
601 	int     i;
602 
603 	int     samples = -1;
604 	double  loss = -1.0;
605 	char    profile_name[ED_MAX_NAME_LEN];
606 	int     delay_first = -1;
607 	int     do_points = 0;
608 	struct point    points[ED_MAX_SAMPLES_NO];
609 	int     points_no = 0;
610 
611 	/* XXX link never NULL? */
612 	p->link_nr = link->link_nr;
613 
614 	profile_name[0] = '\0';
615 	f = fopen(filename, "r");
616 	if (f == NULL)
617 		err(EX_UNAVAILABLE, "fopen: %s", filename);
618 
619 	while (fgets(line, ED_MAX_LINE_LEN, f)) {	 /* read commands */
620 		char *s, *cur = line, *name = NULL, *arg = NULL;
621 
622 		++lineno;
623 
624 		/* parse the line */
625 		while (cur) {
626 			s = strsep(&cur, ED_SEPARATORS);
627 			if (s == NULL || *s == '#')
628 				break;
629 			if (*s == '\0')
630 				continue;
631 			if (arg)
632 				errx(ED_EFMT("too many arguments"));
633 			if (name == NULL)
634 				name = s;
635 			else
636 				arg = s;
637 		}
638 		if (name == NULL)	/* empty line */
639 			continue;
640 		if (arg == NULL)
641 			errx(ED_EFMT("missing arg for %s"), name);
642 
643 		if (!strcasecmp(name, ED_TOK_SAMPLES)) {
644 		    if (samples > 0)
645 			errx(ED_EFMT("duplicate ``samples'' line"));
646 		    if (atoi(arg) <=0)
647 			errx(ED_EFMT("invalid number of samples"));
648 		    samples = atoi(arg);
649 		    if (samples>ED_MAX_SAMPLES_NO)
650 			    errx(ED_EFMT("too many samples, maximum is %d"),
651 				ED_MAX_SAMPLES_NO);
652 		    do_points = 0;
653 		} else if (!strcasecmp(name, ED_TOK_BW)) {
654 		    char buf[IFNAMSIZ];
655 		    read_bandwidth(arg, &link->bandwidth, buf, sizeof(buf));
656 		} else if (!strcasecmp(name, ED_TOK_LOSS)) {
657 		    if (loss != -1.0)
658 			errx(ED_EFMT("duplicated token: %s"), name);
659 		    if (!is_valid_number(arg))
660 			errx(ED_EFMT("invalid %s"), arg);
661 		    loss = atof(arg);
662 		    if (loss > 1)
663 			errx(ED_EFMT("%s greater than 1.0"), name);
664 		    do_points = 0;
665 		} else if (!strcasecmp(name, ED_TOK_NAME)) {
666 		    if (profile_name[0] != '\0')
667 			errx(ED_EFMT("duplicated token: %s"), name);
668 		    strncpy(profile_name, arg, sizeof(profile_name) - 1);
669 		    profile_name[sizeof(profile_name)-1] = '\0';
670 		    do_points = 0;
671 		} else if (!strcasecmp(name, ED_TOK_DELAY)) {
672 		    if (do_points)
673 			errx(ED_EFMT("duplicated token: %s"), name);
674 		    delay_first = 1;
675 		    do_points = 1;
676 		} else if (!strcasecmp(name, ED_TOK_PROB)) {
677 		    if (do_points)
678 			errx(ED_EFMT("duplicated token: %s"), name);
679 		    delay_first = 0;
680 		    do_points = 1;
681 		} else if (do_points) {
682 		    if (!is_valid_number(name) || !is_valid_number(arg))
683 			errx(ED_EFMT("invalid point found"));
684 		    if (delay_first) {
685 			points[points_no].delay = atof(name);
686 			points[points_no].prob = atof(arg);
687 		    } else {
688 			points[points_no].delay = atof(arg);
689 			points[points_no].prob = atof(name);
690 		    }
691 		    if (points[points_no].prob > 1.0)
692 			errx(ED_EFMT("probability greater than 1.0"));
693 		    ++points_no;
694 		} else {
695 		    errx(ED_EFMT("unrecognised command '%s'"), name);
696 		}
697 	}
698 
699 	fclose (f);
700 
701 	if (samples == -1) {
702 	    warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES);
703 	    samples = 100;
704 	}
705 
706 	if (loss == -1.0) {
707 	    warnx("'%s' not found, assuming no loss", ED_TOK_LOSS);
708 	    loss = 1;
709 	}
710 
711 	/* make sure that there are enough points. */
712 	if (points_no < ED_MIN_SAMPLES_NO)
713 	    errx(ED_EFMT("too few samples, need at least %d"),
714 		ED_MIN_SAMPLES_NO);
715 
716 	qsort(points, points_no, sizeof(struct point), compare_points);
717 
718 	/* interpolation */
719 	for (i = 0; i<points_no-1; ++i) {
720 	    double y1 = points[i].prob * samples;
721 	    double x1 = points[i].delay;
722 	    double y2 = points[i+1].prob * samples;
723 	    double x2 = points[i+1].delay;
724 
725 	    int ix = y1;
726 	    int stop = y2;
727 
728 	    if (x1 == x2) {
729 		for (; ix<stop; ++ix)
730 		    p->samples[ix] = x1;
731 	    } else {
732 		double m = (y2-y1)/(x2-x1);
733 		double c = y1 - m*x1;
734 		for (; ix<stop ; ++ix)
735 		    p->samples[ix] = (ix - c)/m;
736 	    }
737 	}
738 	p->samples_no = samples;
739 	p->loss_level = loss * samples;
740 	strncpy(p->name, profile_name, sizeof(p->name));
741 }
742 
743 /*
744  * configuration of pipes, schedulers, flowsets.
745  * When we configure a new scheduler, an empty pipe is created, so:
746  *
747  * do_pipe = 1 -> "pipe N config ..." only for backward compatibility
748  *	sched N+Delta type fifo sched_mask ...
749  *	pipe N+Delta <parameters>
750  *	flowset N+Delta pipe N+Delta (no parameters)
751  *	sched N type wf2q+ sched_mask ...
752  *	pipe N <parameters>
753  *
754  * do_pipe = 2 -> flowset N config
755  *	flowset N parameters
756  *
757  * do_pipe = 3 -> sched N config
758  *	sched N parameters (default no pipe)
759  *	optional Pipe N config ...
760  * pipe ==>
761  */
762 void
763 ipfw_config_pipe(int ac, char **av)
764 {
765 	int i;
766 	u_int j;
767 	char *end;
768 	struct dn_id *buf, *base;
769 	struct dn_sch *sch = NULL;
770 	struct dn_link *p = NULL;
771 	struct dn_fs *fs = NULL;
772 	struct dn_profile *pf = NULL;
773 	struct ipfw_flow_id *mask = NULL;
774 	int lmax;
775 	uint32_t _foo = 0, *flags = &_foo , *buckets = &_foo;
776 
777 	/*
778 	 * allocate space for 1 header,
779 	 * 1 scheduler, 1 link, 1 flowset, 1 profile
780 	 */
781 	lmax = sizeof(struct dn_id);	/* command header */
782 	lmax += sizeof(struct dn_sch) + sizeof(struct dn_link) +
783 		sizeof(struct dn_fs) + sizeof(struct dn_profile);
784 
785 	av++; ac--;
786 	/* Pipe number */
787 	if (ac && isdigit(**av)) {
788 		i = atoi(*av); av++; ac--;
789 	} else
790 		i = -1;
791 	if (i <= 0)
792 		errx(EX_USAGE, "need a pipe/flowset/sched number");
793 	base = buf = safe_calloc(1, lmax);
794 	/* all commands start with a 'CONFIGURE' and a version */
795 	o_next(&buf, sizeof(struct dn_id), DN_CMD_CONFIG);
796 	base->id = DN_API_VERSION;
797 
798 	switch (co.do_pipe) {
799 	case 1: /* "pipe N config ..." */
800 		/* Allocate space for the WF2Q+ scheduler, its link
801 		 * and the FIFO flowset. Set the number, but leave
802 		 * the scheduler subtype and other parameters to 0
803 		 * so the kernel will use appropriate defaults.
804 		 * XXX todo: add a flag to record if a parameter
805 		 * is actually configured.
806 		 * If we do a 'pipe config' mask -> sched_mask.
807 		 * The FIFO scheduler and link are derived from the
808 		 * WF2Q+ one in the kernel.
809 		 */
810 		sch = o_next(&buf, sizeof(*sch), DN_SCH);
811 		p = o_next(&buf, sizeof(*p), DN_LINK);
812 		fs = o_next(&buf, sizeof(*fs), DN_FS);
813 
814 		sch->sched_nr = i;
815 		sch->oid.subtype = 0;	/* defaults to WF2Q+ */
816 		mask = &sch->sched_mask;
817 		flags = &sch->flags;
818 		buckets = &sch->buckets;
819 		*flags |= DN_PIPE_CMD;
820 
821 		p->link_nr = i;
822 
823 		/* This flowset is only for the FIFO scheduler */
824 		fs->fs_nr = i + 2*DN_MAX_ID;
825 		fs->sched_nr = i + DN_MAX_ID;
826 		break;
827 
828 	case 2: /* "queue N config ... " */
829 		fs = o_next(&buf, sizeof(*fs), DN_FS);
830 		fs->fs_nr = i;
831 		mask = &fs->flow_mask;
832 		flags = &fs->flags;
833 		buckets = &fs->buckets;
834 		break;
835 
836 	case 3: /* "sched N config ..." */
837 		sch = o_next(&buf, sizeof(*sch), DN_SCH);
838 		fs = o_next(&buf, sizeof(*fs), DN_FS);
839 		sch->sched_nr = i;
840 		mask = &sch->sched_mask;
841 		flags = &sch->flags;
842 		buckets = &sch->buckets;
843 		/* fs is used only with !MULTIQUEUE schedulers */
844 		fs->fs_nr = i + DN_MAX_ID;
845 		fs->sched_nr = i;
846 		break;
847 	}
848 	/* set to -1 those fields for which we want to reuse existing
849 	 * values from the kernel.
850 	 * Also, *_nr and subtype = 0 mean reuse the value from the kernel.
851 	 * XXX todo: support reuse of the mask.
852 	 */
853 	if (p)
854 		p->bandwidth = -1;
855 	for (j = 0; j < sizeof(fs->par)/sizeof(fs->par[0]); j++)
856 		fs->par[j] = -1;
857 	while (ac > 0) {
858 		double d;
859 		int tok = match_token(dummynet_params, *av);
860 		ac--; av++;
861 
862 		switch(tok) {
863 		case TOK_NOERROR:
864 			NEED(fs, "noerror is only for pipes");
865 			fs->flags |= DN_NOERROR;
866 			break;
867 
868 		case TOK_PLR:
869 			NEED(fs, "plr is only for pipes");
870 			NEED1("plr needs argument 0..1\n");
871 			d = strtod(av[0], NULL);
872 			if (d > 1)
873 				d = 1;
874 			else if (d < 0)
875 				d = 0;
876 			fs->plr = (int)(d*0x7fffffff);
877 			ac--; av++;
878 			break;
879 
880 		case TOK_QUEUE:
881 			NEED(fs, "queue is only for pipes or flowsets");
882 			NEED1("queue needs queue size\n");
883 			end = NULL;
884 			fs->qsize = strtoul(av[0], &end, 0);
885 			if (*end == 'K' || *end == 'k') {
886 				fs->flags |= DN_QSIZE_BYTES;
887 				fs->qsize *= 1024;
888 			} else if (*end == 'B' ||
889 			    _substrcmp2(end, "by", "bytes") == 0) {
890 				fs->flags |= DN_QSIZE_BYTES;
891 			}
892 			ac--; av++;
893 			break;
894 
895 		case TOK_BUCKETS:
896 			NEED(fs, "buckets is only for pipes or flowsets");
897 			NEED1("buckets needs argument\n");
898 			*buckets = strtoul(av[0], NULL, 0);
899 			ac--; av++;
900 			break;
901 
902 		case TOK_FLOW_MASK:
903 		case TOK_SCHED_MASK:
904 		case TOK_MASK:
905 			NEED(mask, "tok_mask");
906 			NEED1("mask needs mask specifier\n");
907 			/*
908 			 * per-flow queue, mask is dst_ip, dst_port,
909 			 * src_ip, src_port, proto measured in bits
910 			 */
911 
912 			bzero(mask, sizeof(*mask));
913 			end = NULL;
914 
915 			while (ac >= 1) {
916 			    uint32_t *p32 = NULL;
917 			    uint16_t *p16 = NULL;
918 			    uint32_t *p20 = NULL;
919 			    struct in6_addr *pa6 = NULL;
920 			    uint32_t a;
921 
922 			    tok = match_token(dummynet_params, *av);
923 			    ac--; av++;
924 			    switch(tok) {
925 			    case TOK_ALL:
926 				    /*
927 				     * special case, all bits significant
928 				     * except 'extra' (the queue number)
929 				     */
930 				    mask->dst_ip = ~0;
931 				    mask->src_ip = ~0;
932 				    mask->dst_port = ~0;
933 				    mask->src_port = ~0;
934 				    mask->proto = ~0;
935 				    n2mask(&mask->dst_ip6, 128);
936 				    n2mask(&mask->src_ip6, 128);
937 				    mask->flow_id6 = ~0;
938 				    *flags |= DN_HAVE_MASK;
939 				    goto end_mask;
940 
941 			    case TOK_QUEUE:
942 				    mask->extra = ~0;
943 				    *flags |= DN_HAVE_MASK;
944 				    goto end_mask;
945 
946 			    case TOK_DSTIP:
947 				    mask->addr_type = 4;
948 				    p32 = &mask->dst_ip;
949 				    break;
950 
951 			    case TOK_SRCIP:
952 				    mask->addr_type = 4;
953 				    p32 = &mask->src_ip;
954 				    break;
955 
956 			    case TOK_DSTIP6:
957 				    mask->addr_type = 6;
958 				    pa6 = &mask->dst_ip6;
959 				    break;
960 
961 			    case TOK_SRCIP6:
962 				    mask->addr_type = 6;
963 				    pa6 = &mask->src_ip6;
964 				    break;
965 
966 			    case TOK_FLOWID:
967 				    mask->addr_type = 6;
968 				    p20 = &mask->flow_id6;
969 				    break;
970 
971 			    case TOK_DSTPORT:
972 				    p16 = &mask->dst_port;
973 				    break;
974 
975 			    case TOK_SRCPORT:
976 				    p16 = &mask->src_port;
977 				    break;
978 
979 			    case TOK_PROTO:
980 				    break;
981 
982 			    default:
983 				    ac++; av--; /* backtrack */
984 				    goto end_mask;
985 			    }
986 			    if (ac < 1)
987 				    errx(EX_USAGE, "mask: value missing");
988 			    if (*av[0] == '/') {
989 				    a = strtoul(av[0]+1, &end, 0);
990 				    if (pa6 == NULL)
991 					    a = (a == 32) ? ~0 : (1 << a) - 1;
992 			    } else
993 				    a = strtoul(av[0], &end, 0);
994 			    if (p32 != NULL)
995 				    *p32 = a;
996 			    else if (p16 != NULL) {
997 				    if (a > 0xFFFF)
998 					    errx(EX_DATAERR,
999 						"port mask must be 16 bit");
1000 				    *p16 = (uint16_t)a;
1001 			    } else if (p20 != NULL) {
1002 				    if (a > 0xfffff)
1003 					errx(EX_DATAERR,
1004 					    "flow_id mask must be 20 bit");
1005 				    *p20 = (uint32_t)a;
1006 			    } else if (pa6 != NULL) {
1007 				    if (a > 128)
1008 					errx(EX_DATAERR,
1009 					    "in6addr invalid mask len");
1010 				    else
1011 					n2mask(pa6, a);
1012 			    } else {
1013 				    if (a > 0xFF)
1014 					    errx(EX_DATAERR,
1015 						"proto mask must be 8 bit");
1016 				    mask->proto = (uint8_t)a;
1017 			    }
1018 			    if (a != 0)
1019 				    *flags |= DN_HAVE_MASK;
1020 			    ac--; av++;
1021 			} /* end while, config masks */
1022 end_mask:
1023 			break;
1024 
1025 		case TOK_RED:
1026 		case TOK_GRED:
1027 			NEED1("red/gred needs w_q/min_th/max_th/max_p\n");
1028 			fs->flags |= DN_IS_RED;
1029 			if (tok == TOK_GRED)
1030 				fs->flags |= DN_IS_GENTLE_RED;
1031 			/*
1032 			 * the format for parameters is w_q/min_th/max_th/max_p
1033 			 */
1034 			if ((end = strsep(&av[0], "/"))) {
1035 			    double w_q = strtod(end, NULL);
1036 			    if (w_q > 1 || w_q <= 0)
1037 				errx(EX_DATAERR, "0 < w_q <= 1");
1038 			    fs->w_q = (int) (w_q * (1 << SCALE_RED));
1039 			}
1040 			if ((end = strsep(&av[0], "/"))) {
1041 			    fs->min_th = strtoul(end, &end, 0);
1042 			    if (*end == 'K' || *end == 'k')
1043 				fs->min_th *= 1024;
1044 			}
1045 			if ((end = strsep(&av[0], "/"))) {
1046 			    fs->max_th = strtoul(end, &end, 0);
1047 			    if (*end == 'K' || *end == 'k')
1048 				fs->max_th *= 1024;
1049 			}
1050 			if ((end = strsep(&av[0], "/"))) {
1051 			    double max_p = strtod(end, NULL);
1052 			    if (max_p > 1 || max_p < 0)
1053 				errx(EX_DATAERR, "0 <= max_p <= 1");
1054 			    fs->max_p = (int)(max_p * (1 << SCALE_RED));
1055 			}
1056 			ac--; av++;
1057 			break;
1058 
1059 		case TOK_ECN:
1060 			fs->flags |= DN_IS_ECN;
1061 			break;
1062 
1063 		case TOK_DROPTAIL:
1064 			NEED(fs, "droptail is only for flowsets");
1065 			fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED);
1066 			break;
1067 
1068 		case TOK_BW:
1069 			NEED(p, "bw is only for links");
1070 			NEED1("bw needs bandwidth or interface\n");
1071 			read_bandwidth(av[0], &p->bandwidth, NULL, 0);
1072 			ac--; av++;
1073 			break;
1074 
1075 		case TOK_DELAY:
1076 			NEED(p, "delay is only for links");
1077 			NEED1("delay needs argument 0..10000ms\n");
1078 			p->delay = strtoul(av[0], NULL, 0);
1079 			ac--; av++;
1080 			break;
1081 
1082 		case TOK_TYPE: {
1083 			int l;
1084 			NEED(sch, "type is only for schedulers");
1085 			NEED1("type needs a string");
1086 			l = strlen(av[0]);
1087 			if (l == 0 || l > 15)
1088 				errx(1, "type %s too long\n", av[0]);
1089 			strcpy(sch->name, av[0]);
1090 			sch->oid.subtype = 0; /* use string */
1091 			ac--; av++;
1092 			break;
1093 		    }
1094 
1095 		case TOK_WEIGHT:
1096 			NEED(fs, "weight is only for flowsets");
1097 			NEED1("weight needs argument\n");
1098 			fs->par[0] = strtol(av[0], &end, 0);
1099 			ac--; av++;
1100 			break;
1101 
1102 		case TOK_LMAX:
1103 			NEED(fs, "lmax is only for flowsets");
1104 			NEED1("lmax needs argument\n");
1105 			fs->par[1] = strtol(av[0], &end, 0);
1106 			ac--; av++;
1107 			break;
1108 
1109 		case TOK_PRI:
1110 			NEED(fs, "priority is only for flowsets");
1111 			NEED1("priority needs argument\n");
1112 			fs->par[2] = strtol(av[0], &end, 0);
1113 			ac--; av++;
1114 			break;
1115 
1116 		case TOK_SCHED:
1117 		case TOK_PIPE:
1118 			NEED(fs, "pipe/sched");
1119 			NEED1("pipe/link/sched needs number\n");
1120 			fs->sched_nr = strtoul(av[0], &end, 0);
1121 			ac--; av++;
1122 			break;
1123 
1124 		case TOK_PROFILE:
1125 			NEED((!pf), "profile already set");
1126 			NEED(p, "profile");
1127 		    {
1128 			NEED1("extra delay needs the file name\n");
1129 			pf = o_next(&buf, sizeof(*pf), DN_PROFILE);
1130 			load_extra_delays(av[0], pf, p); //XXX can't fail?
1131 			--ac; ++av;
1132 		    }
1133 			break;
1134 
1135 		case TOK_BURST:
1136 			NEED(p, "burst");
1137 			NEED1("burst needs argument\n");
1138 			errno = 0;
1139 			if (expand_number(av[0], &p->burst) < 0)
1140 				if (errno != ERANGE)
1141 					errx(EX_DATAERR,
1142 					    "burst: invalid argument");
1143 			if (errno || p->burst > (1ULL << 48) - 1)
1144 				errx(EX_DATAERR,
1145 				    "burst: out of range (0..2^48-1)");
1146 			ac--; av++;
1147 			break;
1148 
1149 		default:
1150 			errx(EX_DATAERR, "unrecognised option ``%s''", av[-1]);
1151 		}
1152 	}
1153 
1154 	/* check validity of parameters */
1155 	if (p) {
1156 		if (p->delay > 10000)
1157 			errx(EX_DATAERR, "delay must be < 10000");
1158 		if (p->bandwidth == -1)
1159 			p->bandwidth = 0;
1160 	}
1161 	if (fs) {
1162 		/* XXX accept a 0 scheduler to keep the default */
1163 	    if (fs->flags & DN_QSIZE_BYTES) {
1164 		size_t len;
1165 		long limit;
1166 
1167 		len = sizeof(limit);
1168 		if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit",
1169 			&limit, &len, NULL, 0) == -1)
1170 			limit = 1024*1024;
1171 		if (fs->qsize > limit)
1172 			errx(EX_DATAERR, "queue size must be < %ldB", limit);
1173 	    } else {
1174 		size_t len;
1175 		long limit;
1176 
1177 		len = sizeof(limit);
1178 		if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit",
1179 			&limit, &len, NULL, 0) == -1)
1180 			limit = 100;
1181 		if (fs->qsize > limit)
1182 			errx(EX_DATAERR, "2 <= queue size <= %ld", limit);
1183 	    }
1184 
1185 	    if ((fs->flags & DN_IS_ECN) && !(fs->flags & DN_IS_RED))
1186 		errx(EX_USAGE, "enable red/gred for ECN");
1187 
1188 	    if (fs->flags & DN_IS_RED) {
1189 		size_t len;
1190 		int lookup_depth, avg_pkt_size;
1191 
1192 		if (!(fs->flags & DN_IS_ECN) && (fs->min_th >= fs->max_th))
1193 		    errx(EX_DATAERR, "min_th %d must be < than max_th %d",
1194 			fs->min_th, fs->max_th);
1195 		else if ((fs->flags & DN_IS_ECN) && (fs->min_th > fs->max_th))
1196 		    errx(EX_DATAERR, "min_th %d must be =< than max_th %d",
1197 			fs->min_th, fs->max_th);
1198 
1199 		if (fs->max_th == 0)
1200 		    errx(EX_DATAERR, "max_th must be > 0");
1201 
1202 		len = sizeof(int);
1203 		if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth",
1204 			&lookup_depth, &len, NULL, 0) == -1)
1205 			lookup_depth = 256;
1206 		if (lookup_depth == 0)
1207 		    errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth"
1208 			" must be greater than zero");
1209 
1210 		len = sizeof(int);
1211 		if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size",
1212 			&avg_pkt_size, &len, NULL, 0) == -1)
1213 			avg_pkt_size = 512;
1214 
1215 		if (avg_pkt_size == 0)
1216 			errx(EX_DATAERR,
1217 			    "net.inet.ip.dummynet.red_avg_pkt_size must"
1218 			    " be greater than zero");
1219 
1220 #if 0 /* the following computation is now done in the kernel */
1221 		/*
1222 		 * Ticks needed for sending a medium-sized packet.
1223 		 * Unfortunately, when we are configuring a WF2Q+ queue, we
1224 		 * do not have bandwidth information, because that is stored
1225 		 * in the parent pipe, and also we have multiple queues
1226 		 * competing for it. So we set s=0, which is not very
1227 		 * correct. But on the other hand, why do we want RED with
1228 		 * WF2Q+ ?
1229 		 */
1230 		if (p.bandwidth==0) /* this is a WF2Q+ queue */
1231 			s = 0;
1232 		else
1233 			s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth;
1234 		/*
1235 		 * max idle time (in ticks) before avg queue size becomes 0.
1236 		 * NOTA:  (3/w_q) is approx the value x so that
1237 		 * (1-w_q)^x < 10^-3.
1238 		 */
1239 		w_q = ((double)fs->w_q) / (1 << SCALE_RED);
1240 		idle = s * 3. / w_q;
1241 		fs->lookup_step = (int)idle / lookup_depth;
1242 		if (!fs->lookup_step)
1243 			fs->lookup_step = 1;
1244 		weight = 1 - w_q;
1245 		for (t = fs->lookup_step; t > 1; --t)
1246 			weight *= 1 - w_q;
1247 		fs->lookup_weight = (int)(weight * (1 << SCALE_RED));
1248 #endif /* code moved in the kernel */
1249 	    }
1250 	}
1251 
1252 	i = do_cmd(IP_DUMMYNET3, base, (char *)buf - (char *)base);
1253 
1254 	if (i)
1255 		err(1, "setsockopt(%s)", "IP_DUMMYNET_CONFIGURE");
1256 }
1257 
1258 void
1259 dummynet_flush(void)
1260 {
1261 	struct dn_id oid;
1262 	oid_fill(&oid, sizeof(oid), DN_CMD_FLUSH, DN_API_VERSION);
1263 	do_cmd(IP_DUMMYNET3, &oid, oid.len);
1264 }
1265 
1266 /* Parse input for 'ipfw [pipe|sched|queue] show [range list]'
1267  * Returns the number of ranges, and possibly stores them
1268  * in the array v of size len.
1269  */
1270 static int
1271 parse_range(int ac, char *av[], uint32_t *v, int len)
1272 {
1273 	int n = 0;
1274 	char *endptr, *s;
1275 	uint32_t base[2];
1276 
1277 	if (v == NULL || len < 2) {
1278 		v = base;
1279 		len = 2;
1280 	}
1281 
1282 	for (s = *av; s != NULL; av++, ac--) {
1283 		v[0] = strtoul(s, &endptr, 10);
1284 		v[1] = (*endptr != '-') ? v[0] :
1285 			 strtoul(endptr+1, &endptr, 10);
1286 		if (*endptr == '\0') { /* prepare for next round */
1287 			s = (ac > 0) ? *(av+1) : NULL;
1288 		} else {
1289 			if (*endptr != ',') {
1290 				warn("invalid number: %s", s);
1291 				s = ++endptr;
1292 				continue;
1293 			}
1294 			/* continue processing from here */
1295 			s = ++endptr;
1296 			ac++;
1297 			av--;
1298 		}
1299 		if (v[1] < v[0] ||
1300 			v[1] >= DN_MAX_ID-1 ||
1301 			v[1] >= DN_MAX_ID-1) {
1302 			continue; /* invalid entry */
1303 		}
1304 		n++;
1305 		/* translate if 'pipe list' */
1306 		if (co.do_pipe == 1) {
1307 			v[0] += DN_MAX_ID;
1308 			v[1] += DN_MAX_ID;
1309 		}
1310 		v = (n*2 < len) ? v + 2 : base;
1311 	}
1312 	return n;
1313 }
1314 
1315 /* main entry point for dummynet list functions. co.do_pipe indicates
1316  * which function we want to support.
1317  * av may contain filtering arguments, either individual entries
1318  * or ranges, or lists (space or commas are valid separators).
1319  * Format for a range can be n1-n2 or n3 n4 n5 ...
1320  * In a range n1 must be <= n2, otherwise the range is ignored.
1321  * A number 'n4' is translate in a range 'n4-n4'
1322  * All number must be > 0 and < DN_MAX_ID-1
1323  */
1324 void
1325 dummynet_list(int ac, char *av[], int show_counters)
1326 {
1327 	struct dn_id *oid, *x = NULL;
1328 	int ret, i;
1329 	int n; 		/* # of ranges */
1330 	u_int buflen, l;
1331 	u_int max_size;	/* largest obj passed up */
1332 
1333 	(void)show_counters;	// XXX unused, but we should use it.
1334 	ac--;
1335 	av++; 		/* skip 'list' | 'show' word */
1336 
1337 	n = parse_range(ac, av, NULL, 0);	/* Count # of ranges. */
1338 
1339 	/* Allocate space to store ranges */
1340 	l = sizeof(*oid) + sizeof(uint32_t) * n * 2;
1341 	oid = safe_calloc(1, l);
1342 	oid_fill(oid, l, DN_CMD_GET, DN_API_VERSION);
1343 
1344 	if (n > 0)	/* store ranges in idx */
1345 		parse_range(ac, av, (uint32_t *)(oid + 1), n*2);
1346 	/*
1347 	 * Compute the size of the largest object returned. If the
1348 	 * response leaves at least this much spare space in the
1349 	 * buffer, then surely the response is complete; otherwise
1350 	 * there might be a risk of truncation and we will need to
1351 	 * retry with a larger buffer.
1352 	 * XXX don't bother with smaller structs.
1353 	 */
1354 	max_size = sizeof(struct dn_fs);
1355 	if (max_size < sizeof(struct dn_sch))
1356 		max_size = sizeof(struct dn_sch);
1357 	if (max_size < sizeof(struct dn_flow))
1358 		max_size = sizeof(struct dn_flow);
1359 
1360 	switch (co.do_pipe) {
1361 	case 1:
1362 		oid->subtype = DN_LINK;	/* list pipe */
1363 		break;
1364 	case 2:
1365 		oid->subtype = DN_FS;	/* list queue */
1366 		break;
1367 	case 3:
1368 		oid->subtype = DN_SCH;	/* list sched */
1369 		break;
1370 	}
1371 
1372 	/*
1373 	 * Ask the kernel an estimate of the required space (result
1374 	 * in oid.id), unless we are requesting a subset of objects,
1375 	 * in which case the kernel does not give an exact answer.
1376 	 * In any case, space might grow in the meantime due to the
1377 	 * creation of new queues, so we must be prepared to retry.
1378 	 */
1379 	if (n > 0) {
1380 		buflen = 4*1024;
1381 	} else {
1382 		ret = do_cmd(-IP_DUMMYNET3, oid, (uintptr_t)&l);
1383 		if (ret != 0 || oid->id <= sizeof(*oid))
1384 			goto done;
1385 		buflen = oid->id + max_size;
1386 		oid->len = sizeof(*oid); /* restore */
1387 	}
1388 	/* Try a few times, until the buffer fits */
1389 	for (i = 0; i < 20; i++) {
1390 		l = buflen;
1391 		x = safe_realloc(x, l);
1392 		bcopy(oid, x, oid->len);
1393 		ret = do_cmd(-IP_DUMMYNET3, x, (uintptr_t)&l);
1394 		if (ret != 0 || x->id <= sizeof(*oid))
1395 			goto done; /* no response */
1396 		if (l + max_size <= buflen)
1397 			break; /* ok */
1398 		buflen *= 2;	 /* double for next attempt */
1399 	}
1400 	list_pipes(x, O_NEXT(x, l));
1401 done:
1402 	if (x)
1403 		free(x);
1404 	free(oid);
1405 }
1406