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.
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 },
63 { "bandwidth", TOK_BW },
64 { "delay", TOK_DELAY },
67 { "queue", TOK_QUEUE },
68 { "flowset", TOK_FLOWSET },
69 { "sched", TOK_SCHED },
71 { "priority", TOK_PRI },
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 */
84 #define O_NEXT(p, len) ((void *)((char *)p + len))
87 oid_fill(struct dn_id *oid, int len, int type, uintptr_t id)
95 /* make room in the buffer and move the pointer forward */
97 o_next(struct dn_id **o, int len, int type)
99 struct dn_id *ret = *o;
100 oid_fill(ret, len, type, 0);
101 *o = O_NEXT(*o, len);
107 sort_q(void *arg, const void *pa, const void *pb)
109 int rev = (co.do_sort < 0);
110 int field = rev ? -co.do_sort : co.do_sort;
112 const struct dn_flow_queue *a = pa;
113 const struct dn_flow_queue *b = pb;
117 res = a->len - b->len;
120 res = a->len_bytes - b->len_bytes;
123 case 3: /* tot pkts */
124 res = a->tot_pkts - b->tot_pkts;
127 case 4: /* tot bytes */
128 res = a->tot_bytes - b->tot_bytes;
135 return (int)(rev ? res : -res);
139 /* print a mask and header for the subsequent list of flows */
141 print_mask(struct ipfw_flow_id *id)
143 if (!IS_IP6_FLOW_ID(id)) {
145 "mask: %s 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n",
146 id->extra ? "queue," : "",
148 id->src_ip, id->src_port,
149 id->dst_ip, id->dst_port);
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);
163 print_header(struct ipfw_flow_id *id)
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");
170 printf("BKT ___Prot___ _flow-id_ "
171 "______________Source IPv6/port_______________ "
172 "_______________Dest. IPv6/port_______________ "
173 "Tot_pkt/bytes Pkt/Byte Drp\n");
177 list_flow(struct dn_flow *ni, int *print)
180 struct protoent *pe = NULL;
182 struct ipfw_flow_id *id = &ni->fid;
185 print_header(&ni->fid);
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)) {
194 printf("%-4s ", pe->p_name);
196 printf("%4u ", id->proto);
197 ina.s_addr = htonl(id->src_ip);
199 inet_ntoa(ina), id->src_port);
200 ina.s_addr = htonl(id->dst_ip);
202 inet_ntoa(ina), id->dst_port);
204 /* Print IPv6 flows */
206 printf("%9s ", pe->p_name);
208 printf("%9u ", id->proto);
209 printf("%7d %39s/%-5d ", id->flow_id6,
210 inet_ntop(AF_INET6, &(id->src_ip6), buff, sizeof(buff)),
212 printf(" %39s/%-5d ",
213 inet_ntop(AF_INET6, &(id->dst_ip6), buff, sizeof(buff)),
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);
223 print_flowset_parms(struct dn_fs *fs, char *prefix)
228 char red[90]; /* Display RED parameters */
231 if (fs->flags & DN_QSIZE_BYTES) {
233 sprintf(qs, "%d KB", l / 1024);
235 sprintf(qs, "%d B", l);
237 sprintf(qs, "%3d sl.", l);
239 sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff));
243 if (fs->flags & DN_IS_RED) { /* RED parameters */
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),
250 1.0 * fs->max_p / (double)(1 << SCALE_RED));
251 if (fs->flags & DN_IS_ECN)
252 strncat(red, " (ecn)", 6);
254 sprintf(red, "droptail");
257 printf("%s %s%s %d queues (%d buckets) %s\n",
258 prefix, qs, plr, fs->oid.id, fs->buckets, red);
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);
271 print_extra_delay_parms(struct dn_profile *p)
274 if (p->samples_no <= 0)
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);
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.
299 * link i (int queue); scheduler i; si(i) { flowsets() : queues }
302 list_pipes(struct dn_id *oid, struct dn_id *end)
304 char buf[160]; /* pending buffer */
305 int toPrint = 1; /* print header */
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);
315 printf("unrecognized object %d size %d\n", oid->type, oid->len);
317 case DN_TEXT: /* list of attached flowsets */
324 l = (oid->len - sizeof(*oid))/sizeof(d->p[0]);
327 printf(" Children flowsets: ");
328 for (i = 0; i < l; i++)
329 printf("%u ", d->p[i]);
335 printf("answer for cmd %d, len %d\n", oid->type, oid->id);
338 struct dn_sch *s = (struct dn_sch *)oid;
340 printf(" sched %d type %s flags 0x%x %d buckets %d active\n",
342 s->name, s->flags, s->buckets, s->oid.id);
343 if (s->flags & DN_HAVE_MASK)
344 print_mask(&s->sched_mask);
349 list_flow((struct dn_flow *)oid, &toPrint);
353 struct dn_link *p = (struct dn_link *)oid;
354 double b = p->bandwidth;
358 /* This starts a new object so flush buffer */
362 sprintf(bwbuf, "unlimited ");
363 else if (b >= 1000000)
364 sprintf(bwbuf, "%7.3f Mbit/s", b/1000000);
366 sprintf(bwbuf, "%7.3f Kbit/s", b/1000);
368 sprintf(bwbuf, "%7.3f bit/s ", b);
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);
379 print_flowset_parms((struct dn_fs *)oid, buf);
383 print_extra_delay_parms((struct dn_profile *)oid);
385 flush_buf(buf); // XXX does it really go here ?
390 * Delete pipe, queue or scheduler i
393 ipfw_delete_pipe(int do_pipe, int i)
397 uintptr_t a[1]; /* add more if we want a list */
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);
403 i = do_cmd(IP_DUMMYNET3, &cmd, cmd.oid.len);
406 warn("rule %u: setsockopt(IP_DUMMYNET_DEL)", i);
412 * Code to parse delay profiles.
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.
424 * cumulative probability
434 * +-------*------------------->
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.
443 * To pass the curve to dummynet, we must store the parameters
444 * in a file as described below, and issue the command
446 * ipfw pipe <n> config ... bw XXX profile <filename> ...
448 * The file format is the following, with whitespace acting as
449 * a separator and '#' indicating the beginning a comment:
452 * the number of samples used in the internal
453 * representation (2..1024; default 100);
456 * The probability above which packets are lost.
457 * (0.0 <= L <= 1.0, default 1.0 i.e. no loss);
460 * Optional a name (listed by "ipfw pipe show")
461 * to identify the distribution;
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.
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.
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.
477 * Example of a profile file:
483 0 200 # minimum overhead is 200ms
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.
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
507 * returns 1 if s is a non-negative number, with at least one '.'
510 is_valid_number(const char *s)
512 int i, dots_found = 0;
515 for (i = 0; i<len; ++i)
516 if (!isdigit(s[i]) && (s[i] !='.' || ++dots_found > 1))
522 * Take as input a string describing a bandwidth value
523 * and return the numeric bandwidth value.
524 * set clocking interface or bandwidth value
527 read_bandwidth(char *arg, int *bandwidth, char *if_name, int namelen)
529 if (*bandwidth != -1)
530 warnx("duplicate token, override bandwidth value!");
532 if (arg[0] >= 'a' && arg[0] <= 'z') {
534 errx(1, "no if support");
536 if (namelen >= IFNAMSIZ)
537 warn("interface name truncated");
540 strncpy(if_name, arg, namelen);
541 if_name[namelen] = '\0';
543 } else { /* read bandwidth value */
547 bw = strtoul(arg, &end, 0);
548 if (*end == 'K' || *end == 'k') {
551 } else if (*end == 'M' || *end == 'm') {
556 _substrcmp2(end, "Bi", "Bit/s") != 0) ||
557 _substrcmp2(end, "by", "bytes") == 0)
561 errx(EX_DATAERR, "bandwidth too large");
575 compare_points(const void *vp1, const void *vp2)
577 const struct point *p1 = vp1;
578 const struct point *p2 = vp2;
581 res = p1->prob - p2->prob;
583 res = p1->delay - p2->delay;
592 #define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno
595 load_extra_delays(const char *filename, struct dn_profile *p,
596 struct dn_link *link)
598 char line[ED_MAX_LINE_LEN];
605 char profile_name[ED_MAX_NAME_LEN];
606 int delay_first = -1;
608 struct point points[ED_MAX_SAMPLES_NO];
611 /* XXX link never NULL? */
612 p->link_nr = link->link_nr;
614 profile_name[0] = '\0';
615 f = fopen(filename, "r");
617 err(EX_UNAVAILABLE, "fopen: %s", filename);
619 while (fgets(line, ED_MAX_LINE_LEN, f)) { /* read commands */
620 char *s, *cur = line, *name = NULL, *arg = NULL;
626 s = strsep(&cur, ED_SEPARATORS);
627 if (s == NULL || *s == '#')
632 errx(ED_EFMT("too many arguments"));
638 if (name == NULL) /* empty line */
641 errx(ED_EFMT("missing arg for %s"), name);
643 if (!strcasecmp(name, ED_TOK_SAMPLES)) {
645 errx(ED_EFMT("duplicate ``samples'' line"));
647 errx(ED_EFMT("invalid number of samples"));
649 if (samples>ED_MAX_SAMPLES_NO)
650 errx(ED_EFMT("too many samples, maximum is %d"),
653 } else if (!strcasecmp(name, ED_TOK_BW)) {
655 read_bandwidth(arg, &link->bandwidth, buf, sizeof(buf));
656 } else if (!strcasecmp(name, ED_TOK_LOSS)) {
658 errx(ED_EFMT("duplicated token: %s"), name);
659 if (!is_valid_number(arg))
660 errx(ED_EFMT("invalid %s"), arg);
663 errx(ED_EFMT("%s greater than 1.0"), name);
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';
671 } else if (!strcasecmp(name, ED_TOK_DELAY)) {
673 errx(ED_EFMT("duplicated token: %s"), name);
676 } else if (!strcasecmp(name, ED_TOK_PROB)) {
678 errx(ED_EFMT("duplicated token: %s"), name);
681 } else if (do_points) {
682 if (!is_valid_number(name) || !is_valid_number(arg))
683 errx(ED_EFMT("invalid point found"));
685 points[points_no].delay = atof(name);
686 points[points_no].prob = atof(arg);
688 points[points_no].delay = atof(arg);
689 points[points_no].prob = atof(name);
691 if (points[points_no].prob > 1.0)
692 errx(ED_EFMT("probability greater than 1.0"));
695 errx(ED_EFMT("unrecognised command '%s'"), name);
702 warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES);
707 warnx("'%s' not found, assuming no loss", ED_TOK_LOSS);
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"),
716 qsort(points, points_no, sizeof(struct point), compare_points);
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;
729 for (; ix<stop; ++ix)
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;
738 p->samples_no = samples;
739 p->loss_level = loss * samples;
740 strncpy(p->name, profile_name, sizeof(p->name));
744 * configuration of pipes, schedulers, flowsets.
745 * When we configure a new scheduler, an empty pipe is created, so:
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>
754 * do_pipe = 2 -> flowset N config
755 * flowset N parameters
757 * do_pipe = 3 -> sched N config
758 * sched N parameters (default no pipe)
759 * optional Pipe N config ...
763 ipfw_config_pipe(int ac, char **av)
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;
775 uint32_t _foo = 0, *flags = &_foo , *buckets = &_foo;
778 * allocate space for 1 header,
779 * 1 scheduler, 1 link, 1 flowset, 1 profile
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);
787 if (ac && isdigit(**av)) {
788 i = atoi(*av); av++; ac--;
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;
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.
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);
815 sch->oid.subtype = 0; /* defaults to WF2Q+ */
816 mask = &sch->sched_mask;
818 buckets = &sch->buckets;
819 *flags |= DN_PIPE_CMD;
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;
828 case 2: /* "queue N config ... " */
829 fs = o_next(&buf, sizeof(*fs), DN_FS);
831 mask = &fs->flow_mask;
833 buckets = &fs->buckets;
836 case 3: /* "sched N config ..." */
837 sch = o_next(&buf, sizeof(*sch), DN_SCH);
838 fs = o_next(&buf, sizeof(*fs), DN_FS);
840 mask = &sch->sched_mask;
842 buckets = &sch->buckets;
843 /* fs is used only with !MULTIQUEUE schedulers */
844 fs->fs_nr = i + DN_MAX_ID;
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.
855 for (j = 0; j < sizeof(fs->par)/sizeof(fs->par[0]); j++)
859 int tok = match_token(dummynet_params, *av);
864 NEED(fs, "noerror is only for pipes");
865 fs->flags |= DN_NOERROR;
869 NEED(fs, "plr is only for pipes");
870 NEED1("plr needs argument 0..1\n");
871 d = strtod(av[0], NULL);
876 fs->plr = (int)(d*0x7fffffff);
881 NEED(fs, "queue is only for pipes or flowsets");
882 NEED1("queue needs queue size\n");
884 fs->qsize = strtoul(av[0], &end, 0);
885 if (*end == 'K' || *end == 'k') {
886 fs->flags |= DN_QSIZE_BYTES;
888 } else if (*end == 'B' ||
889 _substrcmp2(end, "by", "bytes") == 0) {
890 fs->flags |= DN_QSIZE_BYTES;
896 NEED(fs, "buckets is only for pipes or flowsets");
897 NEED1("buckets needs argument\n");
898 *buckets = strtoul(av[0], NULL, 0);
905 NEED(mask, "tok_mask");
906 NEED1("mask needs mask specifier\n");
908 * per-flow queue, mask is dst_ip, dst_port,
909 * src_ip, src_port, proto measured in bits
912 bzero(mask, sizeof(*mask));
916 uint32_t *p32 = NULL;
917 uint16_t *p16 = NULL;
918 uint32_t *p20 = NULL;
919 struct in6_addr *pa6 = NULL;
922 tok = match_token(dummynet_params, *av);
927 * special case, all bits significant
928 * except 'extra' (the queue number)
935 n2mask(&mask->dst_ip6, 128);
936 n2mask(&mask->src_ip6, 128);
938 *flags |= DN_HAVE_MASK;
943 *flags |= DN_HAVE_MASK;
958 pa6 = &mask->dst_ip6;
963 pa6 = &mask->src_ip6;
968 p20 = &mask->flow_id6;
972 p16 = &mask->dst_port;
976 p16 = &mask->src_port;
983 ac++; av--; /* backtrack */
987 errx(EX_USAGE, "mask: value missing");
989 a = strtoul(av[0]+1, &end, 0);
991 a = (a == 32) ? ~0 : (1 << a) - 1;
993 a = strtoul(av[0], &end, 0);
996 else if (p16 != NULL) {
999 "port mask must be 16 bit");
1001 } else if (p20 != NULL) {
1004 "flow_id mask must be 20 bit");
1006 } else if (pa6 != NULL) {
1009 "in6addr invalid mask len");
1015 "proto mask must be 8 bit");
1016 mask->proto = (uint8_t)a;
1019 *flags |= DN_HAVE_MASK;
1021 } /* end while, config masks */
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;
1032 * the format for parameters is w_q/min_th/max_th/max_p
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));
1040 if ((end = strsep(&av[0], "/"))) {
1041 fs->min_th = strtoul(end, &end, 0);
1042 if (*end == 'K' || *end == 'k')
1045 if ((end = strsep(&av[0], "/"))) {
1046 fs->max_th = strtoul(end, &end, 0);
1047 if (*end == 'K' || *end == 'k')
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));
1060 fs->flags |= DN_IS_ECN;
1064 NEED(fs, "droptail is only for flowsets");
1065 fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED);
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);
1076 NEED(p, "delay is only for links");
1077 NEED1("delay needs argument 0..10000ms\n");
1078 p->delay = strtoul(av[0], NULL, 0);
1084 NEED(sch, "type is only for schedulers");
1085 NEED1("type needs a string");
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 */
1096 NEED(fs, "weight is only for flowsets");
1097 NEED1("weight needs argument\n");
1098 fs->par[0] = strtol(av[0], &end, 0);
1103 NEED(fs, "lmax is only for flowsets");
1104 NEED1("lmax needs argument\n");
1105 fs->par[1] = strtol(av[0], &end, 0);
1110 NEED(fs, "priority is only for flowsets");
1111 NEED1("priority needs argument\n");
1112 fs->par[2] = strtol(av[0], &end, 0);
1118 NEED(fs, "pipe/sched");
1119 NEED1("pipe/link/sched needs number\n");
1120 fs->sched_nr = strtoul(av[0], &end, 0);
1125 NEED((!pf), "profile already set");
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?
1137 NEED1("burst needs argument\n");
1139 if (expand_number(av[0], &p->burst) < 0)
1140 if (errno != ERANGE)
1142 "burst: invalid argument");
1143 if (errno || p->burst > (1ULL << 48) - 1)
1145 "burst: out of range (0..2^48-1)");
1150 errx(EX_DATAERR, "unrecognised option ``%s''", av[-1]);
1154 /* check validity of parameters */
1156 if (p->delay > 10000)
1157 errx(EX_DATAERR, "delay must be < 10000");
1158 if (p->bandwidth == -1)
1162 /* XXX accept a 0 scheduler to keep the default */
1163 if (fs->flags & DN_QSIZE_BYTES) {
1167 len = sizeof(limit);
1168 if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit",
1169 &limit, &len, NULL, 0) == -1)
1171 if (fs->qsize > limit)
1172 errx(EX_DATAERR, "queue size must be < %ldB", limit);
1177 len = sizeof(limit);
1178 if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit",
1179 &limit, &len, NULL, 0) == -1)
1181 if (fs->qsize > limit)
1182 errx(EX_DATAERR, "2 <= queue size <= %ld", limit);
1185 if ((fs->flags & DN_IS_ECN) && !(fs->flags & DN_IS_RED))
1186 errx(EX_USAGE, "enable red/gred for ECN");
1188 if (fs->flags & DN_IS_RED) {
1190 int lookup_depth, avg_pkt_size;
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);
1199 if (fs->max_th == 0)
1200 errx(EX_DATAERR, "max_th must be > 0");
1203 if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth",
1204 &lookup_depth, &len, NULL, 0) == -1)
1206 if (lookup_depth == 0)
1207 errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth"
1208 " must be greater than zero");
1211 if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size",
1212 &avg_pkt_size, &len, NULL, 0) == -1)
1215 if (avg_pkt_size == 0)
1217 "net.inet.ip.dummynet.red_avg_pkt_size must"
1218 " be greater than zero");
1220 #if 0 /* the following computation is now done in the kernel */
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
1230 if (p.bandwidth==0) /* this is a WF2Q+ queue */
1233 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 idle = s * 3. / w_q;
1241 fs->lookup_step = (int)idle / lookup_depth;
1242 if (!fs->lookup_step)
1243 fs->lookup_step = 1;
1245 for (t = fs->lookup_step; t > 1; --t)
1247 fs->lookup_weight = (int)(weight * (1 << SCALE_RED));
1248 #endif /* code moved in the kernel */
1252 i = do_cmd(IP_DUMMYNET3, base, (char *)buf - (char *)base);
1255 err(1, "setsockopt(%s)", "IP_DUMMYNET_CONFIGURE");
1259 dummynet_flush(void)
1262 oid_fill(&oid, sizeof(oid), DN_CMD_FLUSH, DN_API_VERSION);
1263 do_cmd(IP_DUMMYNET3, &oid, oid.len);
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.
1271 parse_range(int ac, char *av[], uint32_t *v, int len)
1277 if (v == NULL || len < 2) {
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;
1289 if (*endptr != ',') {
1290 warn("invalid number: %s", s);
1294 /* continue processing from here */
1300 v[1] >= DN_MAX_ID-1 ||
1301 v[1] >= DN_MAX_ID-1) {
1302 continue; /* invalid entry */
1305 /* translate if 'pipe list' */
1306 if (co.do_pipe == 1) {
1310 v = (n*2 < len) ? v + 2 : base;
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
1325 dummynet_list(int ac, char *av[], int show_counters)
1327 struct dn_id *oid, *x = NULL;
1329 int n; /* # of ranges */
1331 u_int max_size; /* largest obj passed up */
1333 (void)show_counters; // XXX unused, but we should use it.
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));