2 * Copyright (c) 2002-2003,2010 Luigi Rizzo
4 * Redistribution and use in source forms, with and without modification,
5 * are permitted provided that this entire comment appears intact.
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.
11 * This software is provided ``AS IS'' without any warranties of any kind.
13 * $FreeBSD: user/luigi/ipfw3-head/sbin/ipfw/dummynet.c 203321 2010-01-31 21:39:25Z luigi $
18 #include <sys/types.h>
19 #include <sys/socket.h>
20 /* XXX there are several sysctl leftover here */
21 #include <sys/sysctl.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 */
42 static struct _s_x dummynet_params[] = {
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 },
53 { "maxlen", TOK_LMAX },
55 { "mask", TOK_MASK }, /* alias for both */
56 { "sched_mask", TOK_SCHED_MASK },
57 { "flow_mask", TOK_FLOW_MASK },
58 { "droptail", TOK_DROPTAIL },
62 { "bandwidth", TOK_BW },
63 { "delay", TOK_DELAY },
66 { "queue", TOK_QUEUE },
67 { "flowset", TOK_FLOWSET },
68 { "sched", TOK_SCHED },
70 { "priority", TOK_PRI },
72 { "flow-id", TOK_FLOWID},
73 { "dst-ipv6", TOK_DSTIP6},
74 { "dst-ip6", TOK_DSTIP6},
75 { "src-ipv6", TOK_SRCIP6},
76 { "src-ip6", TOK_SRCIP6},
77 { "profile", TOK_PROFILE},
78 { "burst", TOK_BURST},
79 { "dummynet-params", TOK_NULL },
80 { NULL, 0 } /* terminator */
83 #define O_NEXT(p, len) ((void *)((char *)p + len))
86 oid_fill(struct dn_id *oid, int len, int type, uintptr_t id)
94 /* make room in the buffer and move the pointer forward */
96 o_next(struct dn_id **o, int len, int type)
98 struct dn_id *ret = *o;
99 oid_fill(ret, len, type, 0);
100 *o = O_NEXT(*o, len);
106 sort_q(void *arg, const void *pa, const void *pb)
108 int rev = (co.do_sort < 0);
109 int field = rev ? -co.do_sort : co.do_sort;
111 const struct dn_flow_queue *a = pa;
112 const struct dn_flow_queue *b = pb;
116 res = a->len - b->len;
119 res = a->len_bytes - b->len_bytes;
122 case 3: /* tot pkts */
123 res = a->tot_pkts - b->tot_pkts;
126 case 4: /* tot bytes */
127 res = a->tot_bytes - b->tot_bytes;
134 return (int)(rev ? res : -res);
138 /* print a mask and header for the subsequent list of flows */
140 print_mask(struct ipfw_flow_id *id)
142 if (!IS_IP6_FLOW_ID(id)) {
144 "mask: %s 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n",
145 id->extra ? "queue," : "",
147 id->src_ip, id->src_port,
148 id->dst_ip, id->dst_port);
150 printf("BKT Prot ___Source IP/port____ "
151 "____Dest. IP/port____ "
152 "Tot_pkt/bytes Pkt/Byte Drp\n");
155 printf("\n mask: %sproto: 0x%02x, flow_id: 0x%08x, ",
156 id->extra ? "queue," : "",
157 id->proto, id->flow_id6);
158 inet_ntop(AF_INET6, &(id->src_ip6), buf, sizeof(buf));
159 printf("%s/0x%04x -> ", buf, id->src_port);
160 inet_ntop(AF_INET6, &(id->dst_ip6), buf, sizeof(buf));
161 printf("%s/0x%04x\n", buf, id->dst_port);
163 printf("BKT ___Prot___ _flow-id_ "
164 "______________Source IPv6/port_______________ "
165 "_______________Dest. IPv6/port_______________ "
166 "Tot_pkt/bytes Pkt/Byte Drp\n");
171 list_flow(struct dn_flow *ni)
174 struct protoent *pe = NULL;
176 struct ipfw_flow_id *id = &ni->fid;
178 pe = getprotobynumber(id->proto);
179 /* XXX: Should check for IPv4 flows */
180 printf("%3u%c", (ni->oid.id) & 0xff,
181 id->extra ? '*' : ' ');
182 if (!IS_IP6_FLOW_ID(id)) {
184 printf("%-4s ", pe->p_name);
186 printf("%4u ", id->proto);
187 ina.s_addr = htonl(id->src_ip);
189 inet_ntoa(ina), id->src_port);
190 ina.s_addr = htonl(id->dst_ip);
192 inet_ntoa(ina), id->dst_port);
194 /* Print IPv6 flows */
196 printf("%9s ", pe->p_name);
198 printf("%9u ", id->proto);
199 printf("%7d %39s/%-5d ", id->flow_id6,
200 inet_ntop(AF_INET6, &(id->src_ip6), buff, sizeof(buff)),
202 printf(" %39s/%-5d ",
203 inet_ntop(AF_INET6, &(id->dst_ip6), buff, sizeof(buff)),
207 /* Tcc relies on msvcrt.dll for printf, and
208 * it does not support ANSI %llu syntax
211 printf("%4llu %8llu %2u %4u %3u\n",
212 align_uint64(&ni->tot_pkts),
213 align_uint64(&ni->tot_bytes),
214 ni->length, ni->len_bytes, ni->drops);
216 /* XXX This should be printed correctly, but for some
217 * weird reason, it is not. Making a printf for each
218 * value is a workaround, until we don't undestand what's wrong
220 /*printf("%4I64u %8I64u %2u %4u %3u\n",
221 align_uint64(&ni->tot_pkts),
222 align_uint64(&ni->tot_bytes),
223 ni->length, ni->len_bytes, ni->drops);*/
225 printf("%4I64u ",align_uint64(&ni->tot_pkts));
226 printf("%8I64u ",align_uint64(&ni->tot_bytes));
227 printf("%2u ",ni->length);
228 printf("%4u ",ni->len_bytes);
229 printf("%3u\n",ni->drops);
234 print_flowset_parms(struct dn_fs *fs, char *prefix)
239 char red[90]; /* Display RED parameters */
242 if (fs->flags & DN_QSIZE_BYTES) {
244 sprintf(qs, "%d KB", l / 1024);
246 sprintf(qs, "%d B", l);
248 sprintf(qs, "%3d sl.", l);
250 sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff));
254 if (fs->flags & DN_IS_RED) /* RED parameters */
256 "\n\t %cRED w_q %f min_th %d max_th %d max_p %f",
257 (fs->flags & DN_IS_GENTLE_RED) ? 'G' : ' ',
258 1.0 * fs->w_q / (double)(1 << SCALE_RED),
261 1.0 * fs->max_p / (double)(1 << SCALE_RED));
263 sprintf(red, "droptail");
266 printf("%s %s%s %d queues (%d buckets) %s\n",
267 prefix, qs, plr, fs->oid.id, fs->buckets, red);
270 printf("q%05d %s%s %d flows (%d buckets) sched %d "
271 "weight %d lmax %d pri %d %s\n",
272 fs->fs_nr, qs, plr, fs->oid.id, fs->buckets,
273 fs->sched_nr, fs->par[0], fs->par[1], fs->par[2], red);
274 if (fs->flags & DN_HAVE_MASK)
275 print_mask(&fs->flow_mask);
280 print_extra_delay_parms(struct dn_profile *p)
283 if (p->samples_no <= 0)
286 loss = p->loss_level;
287 loss /= p->samples_no;
288 printf("\t profile: name \"%s\" loss %f samples %d\n",
289 p->name, loss, p->samples_no);
301 * generic list routine. We expect objects in a specific order, i.e.
302 * PIPES AND SCHEDULERS:
303 * link; scheduler; internal flowset if any; instances
304 * we can tell a pipe from the number.
308 * link i (int queue); scheduler i; si(i) { flowsets() : queues }
311 list_pipes(struct dn_id *oid, struct dn_id *end)
313 char buf[160]; /* pending buffer */
316 for (; oid != end; oid = O_NEXT(oid, oid->len)) {
317 if (oid->len < sizeof(*oid))
318 errx(1, "invalid oid len %d\n", oid->len);
323 printf("unrecognized object %d size %d\n", oid->type, oid->len);
325 case DN_TEXT: /* list of attached flowsets */
332 l = (oid->len - sizeof(*oid))/sizeof(d->p[0]);
335 printf(" Children flowsets: ");
336 for (i = 0; i < l; i++)
337 printf("%u ", d->p[i]);
343 printf("answer for cmd %d, len %d\n", oid->type, oid->id);
346 struct dn_sch *s = (struct dn_sch *)oid;
348 printf(" sched %d type %s flags 0x%x %d buckets %d active\n",
350 s->name, s->flags, s->buckets, s->oid.id);
351 if (s->flags & DN_HAVE_MASK)
352 print_mask(&s->sched_mask);
357 list_flow((struct dn_flow *)oid);
361 struct dn_link *p = (struct dn_link *)oid;
362 double b = p->bandwidth;
366 /* This starts a new object so flush buffer */
370 sprintf(bwbuf, "unlimited ");
371 else if (b >= 1000000)
372 sprintf(bwbuf, "%7.3f Mbit/s", b/1000000);
374 sprintf(bwbuf, "%7.3f Kbit/s", b/1000);
376 sprintf(bwbuf, "%7.3f bit/s ", b);
378 if (humanize_number(burst, sizeof(burst), p->burst,
379 "", HN_AUTOSCALE, 0) < 0 || co.verbose)
380 sprintf(burst, "%d", (int)p->burst);
381 sprintf(buf, "%05d: %s %4d ms burst %s",
382 p->link_nr % DN_MAX_ID, bwbuf, p->delay, burst);
387 print_flowset_parms((struct dn_fs *)oid, buf);
391 print_extra_delay_parms((struct dn_profile *)oid);
393 flush_buf(buf); // XXX does it really go here ?
398 * Delete pipe, queue or scheduler i
401 ipfw_delete_pipe(int do_pipe, int i)
405 uintptr_t a[1]; /* add more if we want a list */
407 oid_fill((void *)&cmd, sizeof(cmd), DN_CMD_DELETE, DN_API_VERSION);
408 cmd.oid.subtype = (do_pipe == 1) ? DN_LINK :
409 ( (do_pipe == 2) ? DN_FS : DN_SCH);
411 i = do_cmd(IP_DUMMYNET3, &cmd, cmd.oid.len);
414 warn("rule %u: setsockopt(IP_DUMMYNET_DEL)", i);
420 * Code to parse delay profiles.
422 * Some link types introduce extra delays in the transmission
423 * of a packet, e.g. because of MAC level framing, contention on
424 * the use of the channel, MAC level retransmissions and so on.
425 * From our point of view, the channel is effectively unavailable
426 * for this extra time, which is constant or variable depending
427 * on the link type. Additionally, packets may be dropped after this
428 * time (e.g. on a wireless link after too many retransmissions).
429 * We can model the additional delay with an empirical curve
430 * that represents its distribution.
432 * cumulative probability
442 * +-------*------------------->
445 * The empirical curve may have both vertical and horizontal lines.
446 * Vertical lines represent constant delay for a range of
447 * probabilities; horizontal lines correspond to a discontinuty
448 * in the delay distribution: the link will use the largest delay
449 * for a given probability.
451 * To pass the curve to dummynet, we must store the parameters
452 * in a file as described below, and issue the command
454 * ipfw pipe <n> config ... bw XXX profile <filename> ...
456 * The file format is the following, with whitespace acting as
457 * a separator and '#' indicating the beginning a comment:
460 * the number of samples used in the internal
461 * representation (2..1024; default 100);
464 * The probability above which packets are lost.
465 * (0.0 <= L <= 1.0, default 1.0 i.e. no loss);
468 * Optional a name (listed by "ipfw pipe show")
469 * to identify the distribution;
471 * "delay prob" | "prob delay"
472 * One of these two lines is mandatory and defines
473 * the format of the following lines with data points.
476 * 2 or more lines representing points in the curve,
477 * with either delay or probability first, according
478 * to the chosen format.
479 * The unit for delay is milliseconds.
481 * Data points does not need to be ordered or equal to the number
482 * specified in the "samples" line. ipfw will sort and interpolate
483 * the curve as needed.
485 * Example of a profile file:
491 0 200 # minimum overhead is 200ms
498 * Internally, we will convert the curve to a fixed number of
499 * samples, and when it is time to transmit a packet we will
500 * model the extra delay as extra bits in the packet.
504 #define ED_MAX_LINE_LEN 256+ED_MAX_NAME_LEN
505 #define ED_TOK_SAMPLES "samples"
506 #define ED_TOK_LOSS "loss-level"
507 #define ED_TOK_NAME "name"
508 #define ED_TOK_DELAY "delay"
509 #define ED_TOK_PROB "prob"
510 #define ED_TOK_BW "bw"
511 #define ED_SEPARATORS " \t\n"
512 #define ED_MIN_SAMPLES_NO 2
515 * returns 1 if s is a non-negative number, with at least one '.'
518 is_valid_number(const char *s)
520 int i, dots_found = 0;
523 for (i = 0; i<len; ++i)
524 if (!isdigit(s[i]) && (s[i] !='.' || ++dots_found > 1))
530 * Take as input a string describing a bandwidth value
531 * and return the numeric bandwidth value.
532 * set clocking interface or bandwidth value
535 read_bandwidth(char *arg, int *bandwidth, char *if_name, int namelen)
537 if (*bandwidth != -1)
538 warnx("duplicate token, override bandwidth value!");
540 if (arg[0] >= 'a' && arg[0] <= 'z') {
542 errx(1, "no if support");
544 if (namelen >= IFNAMSIZ)
545 warn("interface name truncated");
548 strncpy(if_name, arg, namelen);
549 if_name[namelen] = '\0';
551 } else { /* read bandwidth value */
555 bw = strtoul(arg, &end, 0);
556 if (*end == 'K' || *end == 'k') {
559 } else if (*end == 'M' || *end == 'm') {
564 _substrcmp2(end, "Bi", "Bit/s") != 0) ||
565 _substrcmp2(end, "by", "bytes") == 0)
569 errx(EX_DATAERR, "bandwidth too large");
583 compare_points(const void *vp1, const void *vp2)
585 const struct point *p1 = vp1;
586 const struct point *p2 = vp2;
589 res = p1->prob - p2->prob;
591 res = p1->delay - p2->delay;
600 #define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno
603 load_extra_delays(const char *filename, struct dn_profile *p,
604 struct dn_link *link)
606 char line[ED_MAX_LINE_LEN];
613 char profile_name[ED_MAX_NAME_LEN];
614 int delay_first = -1;
616 struct point points[ED_MAX_SAMPLES_NO];
619 /* XXX link never NULL? */
620 p->link_nr = link->link_nr;
622 profile_name[0] = '\0';
623 f = fopen(filename, "r");
625 err(EX_UNAVAILABLE, "fopen: %s", filename);
627 while (fgets(line, ED_MAX_LINE_LEN, f)) { /* read commands */
628 char *s, *cur = line, *name = NULL, *arg = NULL;
634 s = strsep(&cur, ED_SEPARATORS);
635 if (s == NULL || *s == '#')
640 errx(ED_EFMT("too many arguments"));
646 if (name == NULL) /* empty line */
649 errx(ED_EFMT("missing arg for %s"), name);
651 if (!strcasecmp(name, ED_TOK_SAMPLES)) {
653 errx(ED_EFMT("duplicate ``samples'' line"));
655 errx(ED_EFMT("invalid number of samples"));
657 if (samples>ED_MAX_SAMPLES_NO)
658 errx(ED_EFMT("too many samples, maximum is %d"),
661 } else if (!strcasecmp(name, ED_TOK_BW)) {
663 read_bandwidth(arg, &link->bandwidth, buf, sizeof(buf));
664 } else if (!strcasecmp(name, ED_TOK_LOSS)) {
666 errx(ED_EFMT("duplicated token: %s"), name);
667 if (!is_valid_number(arg))
668 errx(ED_EFMT("invalid %s"), arg);
671 errx(ED_EFMT("%s greater than 1.0"), name);
673 } else if (!strcasecmp(name, ED_TOK_NAME)) {
674 if (profile_name[0] != '\0')
675 errx(ED_EFMT("duplicated token: %s"), name);
676 strncpy(profile_name, arg, sizeof(profile_name) - 1);
677 profile_name[sizeof(profile_name)-1] = '\0';
679 } else if (!strcasecmp(name, ED_TOK_DELAY)) {
681 errx(ED_EFMT("duplicated token: %s"), name);
684 } else if (!strcasecmp(name, ED_TOK_PROB)) {
686 errx(ED_EFMT("duplicated token: %s"), name);
689 } else if (do_points) {
690 if (!is_valid_number(name) || !is_valid_number(arg))
691 errx(ED_EFMT("invalid point found"));
693 points[points_no].delay = atof(name);
694 points[points_no].prob = atof(arg);
696 points[points_no].delay = atof(arg);
697 points[points_no].prob = atof(name);
699 if (points[points_no].prob > 1.0)
700 errx(ED_EFMT("probability greater than 1.0"));
703 errx(ED_EFMT("unrecognised command '%s'"), name);
710 warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES);
715 warnx("'%s' not found, assuming no loss", ED_TOK_LOSS);
719 /* make sure that there are enough points. */
720 if (points_no < ED_MIN_SAMPLES_NO)
721 errx(ED_EFMT("too few samples, need at least %d"),
724 qsort(points, points_no, sizeof(struct point), compare_points);
727 for (i = 0; i<points_no-1; ++i) {
728 double y1 = points[i].prob * samples;
729 double x1 = points[i].delay;
730 double y2 = points[i+1].prob * samples;
731 double x2 = points[i+1].delay;
737 for (; ix<stop; ++ix)
740 double m = (y2-y1)/(x2-x1);
741 double c = y1 - m*x1;
742 for (; ix<stop ; ++ix)
743 p->samples[ix] = (ix - c)/m;
746 p->samples_no = samples;
747 p->loss_level = loss * samples;
748 strncpy(p->name, profile_name, sizeof(p->name));
752 * configuration of pipes, schedulers, flowsets.
753 * When we configure a new scheduler, an empty pipe is created, so:
755 * do_pipe = 1 -> "pipe N config ..." only for backward compatibility
756 * sched N+Delta type fifo sched_mask ...
757 * pipe N+Delta <parameters>
758 * flowset N+Delta pipe N+Delta (no parameters)
759 * sched N type wf2q+ sched_mask ...
760 * pipe N <parameters>
762 * do_pipe = 2 -> flowset N config
763 * flowset N parameters
765 * do_pipe = 3 -> sched N config
766 * sched N parameters (default no pipe)
767 * optional Pipe N config ...
771 ipfw_config_pipe(int ac, char **av)
776 struct dn_id *buf, *base;
777 struct dn_sch *sch = NULL;
778 struct dn_link *p = NULL;
779 struct dn_fs *fs = NULL;
780 struct dn_profile *pf = NULL;
781 struct ipfw_flow_id *mask = NULL;
783 uint32_t _foo = 0, *flags = &_foo , *buckets = &_foo;
786 * allocate space for 1 header,
787 * 1 scheduler, 1 link, 1 flowset, 1 profile
789 lmax = sizeof(struct dn_id); /* command header */
790 lmax += sizeof(struct dn_sch) + sizeof(struct dn_link) +
791 sizeof(struct dn_fs) + sizeof(struct dn_profile);
795 if (ac && isdigit(**av)) {
796 i = atoi(*av); av++; ac--;
800 errx(EX_USAGE, "need a pipe/flowset/sched number");
801 base = buf = safe_calloc(1, lmax);
802 /* all commands start with a 'CONFIGURE' and a version */
803 o_next(&buf, sizeof(struct dn_id), DN_CMD_CONFIG);
804 base->id = DN_API_VERSION;
806 switch (co.do_pipe) {
807 case 1: /* "pipe N config ..." */
808 /* Allocate space for the WF2Q+ scheduler, its link
809 * and the FIFO flowset. Set the number, but leave
810 * the scheduler subtype and other parameters to 0
811 * so the kernel will use appropriate defaults.
812 * XXX todo: add a flag to record if a parameter
813 * is actually configured.
814 * If we do a 'pipe config' mask -> sched_mask.
815 * The FIFO scheduler and link are derived from the
816 * WF2Q+ one in the kernel.
818 sch = o_next(&buf, sizeof(*sch), DN_SCH);
819 p = o_next(&buf, sizeof(*p), DN_LINK);
820 fs = o_next(&buf, sizeof(*fs), DN_FS);
823 sch->oid.subtype = 0; /* defaults to WF2Q+ */
824 mask = &sch->sched_mask;
826 buckets = &sch->buckets;
827 *flags |= DN_PIPE_CMD;
831 /* This flowset is only for the FIFO scheduler */
832 fs->fs_nr = i + 2*DN_MAX_ID;
833 fs->sched_nr = i + DN_MAX_ID;
836 case 2: /* "queue N config ... " */
837 fs = o_next(&buf, sizeof(*fs), DN_FS);
839 mask = &fs->flow_mask;
841 buckets = &fs->buckets;
844 case 3: /* "sched N config ..." */
845 sch = o_next(&buf, sizeof(*sch), DN_SCH);
846 fs = o_next(&buf, sizeof(*fs), DN_FS);
848 mask = &sch->sched_mask;
850 buckets = &sch->buckets;
851 /* fs is used only with !MULTIQUEUE schedulers */
852 fs->fs_nr = i + DN_MAX_ID;
856 /* set to -1 those fields for which we want to reuse existing
857 * values from the kernel.
858 * Also, *_nr and subtype = 0 mean reuse the value from the kernel.
859 * XXX todo: support reuse of the mask.
863 for (j = 0; j < sizeof(fs->par)/sizeof(fs->par[0]); j++)
867 int tok = match_token(dummynet_params, *av);
872 NEED(fs, "noerror is only for pipes");
873 fs->flags |= DN_NOERROR;
877 NEED(fs, "plr is only for pipes");
878 NEED1("plr needs argument 0..1\n");
879 d = strtod(av[0], NULL);
884 fs->plr = (int)(d*0x7fffffff);
889 NEED(fs, "queue is only for pipes or flowsets");
890 NEED1("queue needs queue size\n");
892 fs->qsize = strtoul(av[0], &end, 0);
893 if (*end == 'K' || *end == 'k') {
894 fs->flags |= DN_QSIZE_BYTES;
896 } else if (*end == 'B' ||
897 _substrcmp2(end, "by", "bytes") == 0) {
898 fs->flags |= DN_QSIZE_BYTES;
904 NEED(fs, "buckets is only for pipes or flowsets");
905 NEED1("buckets needs argument\n");
906 *buckets = strtoul(av[0], NULL, 0);
913 NEED(mask, "tok_mask");
914 NEED1("mask needs mask specifier\n");
916 * per-flow queue, mask is dst_ip, dst_port,
917 * src_ip, src_port, proto measured in bits
921 bzero(mask, sizeof(*mask));
925 uint32_t *p32 = NULL;
926 uint16_t *p16 = NULL;
927 uint32_t *p20 = NULL;
928 struct in6_addr *pa6 = NULL;
931 tok = match_token(dummynet_params, *av);
936 * special case, all bits significant
937 * except 'extra' (the queue number)
944 n2mask(&mask->dst_ip6, 128);
945 n2mask(&mask->src_ip6, 128);
947 *flags |= DN_HAVE_MASK;
952 *flags |= DN_HAVE_MASK;
967 pa6 = &mask->dst_ip6;
972 pa6 = &mask->src_ip6;
977 p20 = &mask->flow_id6;
981 p16 = &mask->dst_port;
985 p16 = &mask->src_port;
992 ac++; av--; /* backtrack */
996 errx(EX_USAGE, "mask: value missing");
998 a = strtoul(av[0]+1, &end, 0);
1000 a = (a == 32) ? ~0 : (1 << a) - 1;
1002 a = strtoul(av[0], &end, 0);
1005 else if (p16 != NULL) {
1008 "port mask must be 16 bit");
1010 } else if (p20 != NULL) {
1013 "flow_id mask must be 20 bit");
1015 } else if (pa6 != NULL) {
1018 "in6addr invalid mask len");
1024 "proto mask must be 8 bit");
1025 mask->proto = (uint8_t)a;
1028 *flags |= DN_HAVE_MASK;
1030 } /* end while, config masks */
1036 NEED1("red/gred needs w_q/min_th/max_th/max_p\n");
1037 fs->flags |= DN_IS_RED;
1038 if (tok == TOK_GRED)
1039 fs->flags |= DN_IS_GENTLE_RED;
1041 * the format for parameters is w_q/min_th/max_th/max_p
1043 if ((end = strsep(&av[0], "/"))) {
1044 double w_q = strtod(end, NULL);
1045 if (w_q > 1 || w_q <= 0)
1046 errx(EX_DATAERR, "0 < w_q <= 1");
1047 fs->w_q = (int) (w_q * (1 << SCALE_RED));
1049 if ((end = strsep(&av[0], "/"))) {
1050 fs->min_th = strtoul(end, &end, 0);
1051 if (*end == 'K' || *end == 'k')
1054 if ((end = strsep(&av[0], "/"))) {
1055 fs->max_th = strtoul(end, &end, 0);
1056 if (*end == 'K' || *end == 'k')
1059 if ((end = strsep(&av[0], "/"))) {
1060 double max_p = strtod(end, NULL);
1061 if (max_p > 1 || max_p <= 0)
1062 errx(EX_DATAERR, "0 < max_p <= 1");
1063 fs->max_p = (int)(max_p * (1 << SCALE_RED));
1069 NEED(fs, "droptail is only for flowsets");
1070 fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED);
1074 NEED(p, "bw is only for links");
1075 NEED1("bw needs bandwidth or interface\n");
1076 read_bandwidth(av[0], &p->bandwidth, NULL, 0);
1081 NEED(p, "delay is only for links");
1082 NEED1("delay needs argument 0..10000ms\n");
1083 p->delay = strtoul(av[0], NULL, 0);
1089 NEED(sch, "type is only for schedulers");
1090 NEED1("type needs a string");
1092 if (l == 0 || l > 15)
1093 errx(1, "type %s too long\n", av[0]);
1094 strcpy(sch->name, av[0]);
1095 sch->oid.subtype = 0; /* use string */
1101 NEED(fs, "weight is only for flowsets");
1102 NEED1("weight needs argument\n");
1103 fs->par[0] = strtol(av[0], &end, 0);
1108 NEED(fs, "lmax is only for flowsets");
1109 NEED1("lmax needs argument\n");
1110 fs->par[1] = strtol(av[0], &end, 0);
1115 NEED(fs, "priority is only for flowsets");
1116 NEED1("priority needs argument\n");
1117 fs->par[2] = strtol(av[0], &end, 0);
1123 NEED(fs, "pipe/sched");
1124 NEED1("pipe/link/sched needs number\n");
1125 fs->sched_nr = strtoul(av[0], &end, 0);
1130 NEED((!pf), "profile already set");
1133 NEED1("extra delay needs the file name\n");
1134 pf = o_next(&buf, sizeof(*pf), DN_PROFILE);
1135 load_extra_delays(av[0], pf, p); //XXX can't fail?
1142 NEED1("burst needs argument\n");
1144 if (expand_number(av[0], (int64_t *)&p->burst) < 0)
1145 if (errno != ERANGE)
1147 "burst: invalid argument");
1148 if (errno || p->burst > (1ULL << 48) - 1)
1150 "burst: out of range (0..2^48-1)");
1155 errx(EX_DATAERR, "unrecognised option ``%s''", av[-1]);
1159 /* check validity of parameters */
1161 if (p->delay > 10000)
1162 errx(EX_DATAERR, "delay must be < 10000");
1163 if (p->bandwidth == -1)
1167 /* XXX accept a 0 scheduler to keep the default */
1168 if (fs->flags & DN_QSIZE_BYTES) {
1172 len = sizeof(limit);
1173 if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit",
1174 &limit, &len, NULL, 0) == -1)
1176 if (fs->qsize > limit)
1177 errx(EX_DATAERR, "queue size must be < %ldB", limit);
1182 len = sizeof(limit);
1183 if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit",
1184 &limit, &len, NULL, 0) == -1)
1186 if (fs->qsize > limit)
1187 errx(EX_DATAERR, "2 <= queue size <= %ld", limit);
1190 if (fs->flags & DN_IS_RED) {
1192 int lookup_depth, avg_pkt_size;
1195 if (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 if (fs->max_th == 0)
1199 errx(EX_DATAERR, "max_th must be > 0");
1202 if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth",
1203 &lookup_depth, &len, NULL, 0) == -1)
1205 if (lookup_depth == 0)
1206 errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth"
1207 " must be greater than zero");
1210 if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size",
1211 &avg_pkt_size, &len, NULL, 0) == -1)
1214 if (avg_pkt_size == 0)
1216 "net.inet.ip.dummynet.red_avg_pkt_size must"
1217 " be greater than zero");
1220 * Ticks needed for sending a medium-sized packet.
1221 * Unfortunately, when we are configuring a WF2Q+ queue, we
1222 * do not have bandwidth information, because that is stored
1223 * in the parent pipe, and also we have multiple queues
1224 * competing for it. So we set s=0, which is not very
1225 * correct. But on the other hand, why do we want RED with
1229 if (p.bandwidth==0) /* this is a WF2Q+ queue */
1232 s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth;
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.
1239 w_q = ((double)fs->w_q) / (1 << SCALE_RED);
1240 #if 0 // go in kernel
1241 idle = s * 3. / w_q;
1242 fs->lookup_step = (int)idle / lookup_depth;
1243 if (!fs->lookup_step)
1244 fs->lookup_step = 1;
1246 for (t = fs->lookup_step; t > 1; --t)
1248 fs->lookup_weight = (int)(weight * (1 << SCALE_RED));
1253 i = do_cmd(IP_DUMMYNET3, base, (char *)buf - (char *)base);
1256 err(1, "setsockopt(%s)", "IP_DUMMYNET_CONFIGURE");
1260 dummynet_flush(void)
1263 oid_fill(&oid, sizeof(oid), DN_CMD_FLUSH, DN_API_VERSION);
1264 do_cmd(IP_DUMMYNET3, &oid, oid.len);
1267 /* Parse input for 'ipfw [pipe|sched|queue] show [range list]'
1268 * Returns the number of ranges, and possibly stores them
1269 * in the array v of size len.
1272 parse_range(int ac, char *av[], uint32_t *v, int len)
1278 if (v == NULL || len < 2) {
1283 for (s = *av; s != NULL; av++, ac--) {
1284 v[0] = strtoul(s, &endptr, 10);
1285 v[1] = (*endptr != '-') ? v[0] :
1286 strtoul(endptr+1, &endptr, 10);
1287 if (*endptr == '\0') { /* prepare for next round */
1288 s = (ac > 0) ? *(av+1) : NULL;
1290 if (*endptr != ',') {
1291 warn("invalid number: %s", s);
1295 /* continue processing from here */
1301 v[1] < 0 || v[1] >= DN_MAX_ID-1 ||
1302 v[0] < 0 || v[1] >= DN_MAX_ID-1) {
1303 continue; /* invalid entry */
1306 /* translate if 'pipe list' */
1307 if (co.do_pipe == 1) {
1311 v = (n*2 < len) ? v + 2 : base;
1316 /* main entry point for dummynet list functions. co.do_pipe indicates
1317 * which function we want to support.
1318 * av may contain filtering arguments, either individual entries
1319 * or ranges, or lists (space or commas are valid separators).
1320 * Format for a range can be n1-n2 or n3 n4 n5 ...
1321 * In a range n1 must be <= n2, otherwise the range is ignored.
1322 * A number 'n4' is translate in a range 'n4-n4'
1323 * All number must be > 0 and < DN_MAX_ID-1
1326 dummynet_list(int ac, char *av[], int show_counters)
1328 struct dn_id *oid, *x = NULL;
1330 int n; /* # of ranges */
1332 int max_size; /* largest obj passed up */
1335 av++; /* skip 'list' | 'show' word */
1337 n = parse_range(ac, av, NULL, 0); /* Count # of ranges. */
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);
1344 if (n > 0) /* store ranges in idx */
1345 parse_range(ac, av, (uint32_t *)(oid + 1), n*2);
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.
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);
1360 switch (co.do_pipe) {
1362 oid->subtype = DN_LINK; /* list pipe */
1365 oid->subtype = DN_FS; /* list queue */
1368 oid->subtype = DN_SCH; /* list sched */
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.
1382 ret = do_cmd(-IP_DUMMYNET3, oid, (uintptr_t)&l);
1383 if (ret != 0 || oid->id <= sizeof(*oid))
1385 buflen = oid->id + max_size;
1386 oid->len = sizeof(*oid); /* restore */
1388 /* Try a few times, until the buffer fits */
1389 for (i = 0; i < 20; i++) {
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)
1398 buflen *= 2; /* double for next attempt */
1400 list_pipes(x, O_NEXT(x, l));