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 buf_pr *bp, struct dn_flow *ni)
180 struct protoent *pe = NULL;
182 struct ipfw_flow_id *id = &ni->fid;
184 pe = getprotobynumber(id->proto);
185 /* XXX: Should check for IPv4 flows */
186 bprintf(bp, "%3u%c", (ni->oid.id) & 0xff,
187 id->extra ? '*' : ' ');
188 if (!IS_IP6_FLOW_ID(id)) {
190 bprintf(bp, "%-4s ", pe->p_name);
192 bprintf(bp, "%4u ", id->proto);
193 ina.s_addr = htonl(id->src_ip);
194 bprintf(bp, "%15s/%-5d ",
195 inet_ntoa(ina), id->src_port);
196 ina.s_addr = htonl(id->dst_ip);
197 bprintf(bp, "%15s/%-5d ",
198 inet_ntoa(ina), id->dst_port);
200 /* Print IPv6 flows */
202 bprintf(bp, "%9s ", pe->p_name);
204 bprintf(bp, "%9u ", id->proto);
205 bprintf(bp, "%7d %39s/%-5d ", id->flow_id6,
206 inet_ntop(AF_INET6, &(id->src_ip6), buff, sizeof(buff)),
208 bprintf(bp, " %39s/%-5d ",
209 inet_ntop(AF_INET6, &(id->dst_ip6), buff, sizeof(buff)),
212 pr_u64(bp, &ni->tot_pkts, 4);
213 pr_u64(bp, &ni->tot_bytes, 8);
214 bprintf(bp, "%2u %4u %3u",
215 ni->length, ni->len_bytes, ni->drops);
219 print_flowset_parms(struct dn_fs *fs, char *prefix)
224 char red[90]; /* Display RED parameters */
227 if (fs->flags & DN_QSIZE_BYTES) {
229 sprintf(qs, "%d KB", l / 1024);
231 sprintf(qs, "%d B", l);
233 sprintf(qs, "%3d sl.", l);
235 sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff));
239 if (fs->flags & DN_IS_RED) { /* RED parameters */
241 "\n\t %cRED w_q %f min_th %d max_th %d max_p %f",
242 (fs->flags & DN_IS_GENTLE_RED) ? 'G' : ' ',
243 1.0 * fs->w_q / (double)(1 << SCALE_RED),
246 1.0 * fs->max_p / (double)(1 << SCALE_RED));
247 if (fs->flags & DN_IS_ECN)
248 strncat(red, " (ecn)", 6);
250 sprintf(red, "droptail");
253 printf("%s %s%s %d queues (%d buckets) %s\n",
254 prefix, qs, plr, fs->oid.id, fs->buckets, red);
257 printf("q%05d %s%s %d flows (%d buckets) sched %d "
258 "weight %d lmax %d pri %d %s\n",
259 fs->fs_nr, qs, plr, fs->oid.id, fs->buckets,
260 fs->sched_nr, fs->par[0], fs->par[1], fs->par[2], red);
261 if (fs->flags & DN_HAVE_MASK)
262 print_mask(&fs->flow_mask);
267 print_extra_delay_parms(struct dn_profile *p)
270 if (p->samples_no <= 0)
273 loss = p->loss_level;
274 loss /= p->samples_no;
275 printf("\t profile: name \"%s\" loss %f samples %d\n",
276 p->name, loss, p->samples_no);
288 * generic list routine. We expect objects in a specific order, i.e.
289 * PIPES AND SCHEDULERS:
290 * link; scheduler; internal flowset if any; instances
291 * we can tell a pipe from the number.
295 * link i (int queue); scheduler i; si(i) { flowsets() : queues }
298 list_pipes(struct dn_id *oid, struct dn_id *end)
300 char buf[160]; /* pending buffer */
301 int toPrint = 1; /* print header */
306 for (; oid != end; oid = O_NEXT(oid, oid->len)) {
307 if (oid->len < sizeof(*oid))
308 errx(1, "invalid oid len %d\n", oid->len);
313 printf("unrecognized object %d size %d\n", oid->type, oid->len);
315 case DN_TEXT: /* list of attached flowsets */
322 l = (oid->len - sizeof(*oid))/sizeof(d->p[0]);
325 printf(" Children flowsets: ");
326 for (i = 0; i < l; i++)
327 printf("%u ", d->p[i]);
333 printf("answer for cmd %d, len %d\n", oid->type, oid->id);
336 struct dn_sch *s = (struct dn_sch *)oid;
338 printf(" sched %d type %s flags 0x%x %d buckets %d active\n",
340 s->name, s->flags, s->buckets, s->oid.id);
341 if (s->flags & DN_HAVE_MASK)
342 print_mask(&s->sched_mask);
348 print_header(&((struct dn_flow *)oid)->fid);
351 list_flow(&bp, (struct dn_flow *)oid);
352 printf("%s\n", bp.buf);
356 struct dn_link *p = (struct dn_link *)oid;
357 double b = p->bandwidth;
361 /* This starts a new object so flush buffer */
365 sprintf(bwbuf, "unlimited ");
366 else if (b >= 1000000)
367 sprintf(bwbuf, "%7.3f Mbit/s", b/1000000);
369 sprintf(bwbuf, "%7.3f Kbit/s", b/1000);
371 sprintf(bwbuf, "%7.3f bit/s ", b);
373 if (humanize_number(burst, sizeof(burst), p->burst,
374 "", HN_AUTOSCALE, 0) < 0 || co.verbose)
375 sprintf(burst, "%d", (int)p->burst);
376 sprintf(buf, "%05d: %s %4d ms burst %s",
377 p->link_nr % DN_MAX_ID, bwbuf, p->delay, burst);
382 print_flowset_parms((struct dn_fs *)oid, buf);
386 print_extra_delay_parms((struct dn_profile *)oid);
388 flush_buf(buf); // XXX does it really go here ?
395 * Delete pipe, queue or scheduler i
398 ipfw_delete_pipe(int do_pipe, int i)
402 uintptr_t a[1]; /* add more if we want a list */
404 oid_fill((void *)&cmd, sizeof(cmd), DN_CMD_DELETE, DN_API_VERSION);
405 cmd.oid.subtype = (do_pipe == 1) ? DN_LINK :
406 ( (do_pipe == 2) ? DN_FS : DN_SCH);
408 i = do_cmd(IP_DUMMYNET3, &cmd, cmd.oid.len);
411 warn("rule %u: setsockopt(IP_DUMMYNET_DEL)", i);
417 * Code to parse delay profiles.
419 * Some link types introduce extra delays in the transmission
420 * of a packet, e.g. because of MAC level framing, contention on
421 * the use of the channel, MAC level retransmissions and so on.
422 * From our point of view, the channel is effectively unavailable
423 * for this extra time, which is constant or variable depending
424 * on the link type. Additionally, packets may be dropped after this
425 * time (e.g. on a wireless link after too many retransmissions).
426 * We can model the additional delay with an empirical curve
427 * that represents its distribution.
429 * cumulative probability
439 * +-------*------------------->
442 * The empirical curve may have both vertical and horizontal lines.
443 * Vertical lines represent constant delay for a range of
444 * probabilities; horizontal lines correspond to a discontinuty
445 * in the delay distribution: the link will use the largest delay
446 * for a given probability.
448 * To pass the curve to dummynet, we must store the parameters
449 * in a file as described below, and issue the command
451 * ipfw pipe <n> config ... bw XXX profile <filename> ...
453 * The file format is the following, with whitespace acting as
454 * a separator and '#' indicating the beginning a comment:
457 * the number of samples used in the internal
458 * representation (2..1024; default 100);
461 * The probability above which packets are lost.
462 * (0.0 <= L <= 1.0, default 1.0 i.e. no loss);
465 * Optional a name (listed by "ipfw pipe show")
466 * to identify the distribution;
468 * "delay prob" | "prob delay"
469 * One of these two lines is mandatory and defines
470 * the format of the following lines with data points.
473 * 2 or more lines representing points in the curve,
474 * with either delay or probability first, according
475 * to the chosen format.
476 * The unit for delay is milliseconds.
478 * Data points does not need to be ordered or equal to the number
479 * specified in the "samples" line. ipfw will sort and interpolate
480 * the curve as needed.
482 * Example of a profile file:
488 0 200 # minimum overhead is 200ms
495 * Internally, we will convert the curve to a fixed number of
496 * samples, and when it is time to transmit a packet we will
497 * model the extra delay as extra bits in the packet.
501 #define ED_MAX_LINE_LEN 256+ED_MAX_NAME_LEN
502 #define ED_TOK_SAMPLES "samples"
503 #define ED_TOK_LOSS "loss-level"
504 #define ED_TOK_NAME "name"
505 #define ED_TOK_DELAY "delay"
506 #define ED_TOK_PROB "prob"
507 #define ED_TOK_BW "bw"
508 #define ED_SEPARATORS " \t\n"
509 #define ED_MIN_SAMPLES_NO 2
512 * returns 1 if s is a non-negative number, with at least one '.'
515 is_valid_number(const char *s)
517 int i, dots_found = 0;
520 for (i = 0; i<len; ++i)
521 if (!isdigit(s[i]) && (s[i] !='.' || ++dots_found > 1))
527 * Take as input a string describing a bandwidth value
528 * and return the numeric bandwidth value.
529 * set clocking interface or bandwidth value
532 read_bandwidth(char *arg, int *bandwidth, char *if_name, int namelen)
534 if (*bandwidth != -1)
535 warnx("duplicate token, override bandwidth value!");
537 if (arg[0] >= 'a' && arg[0] <= 'z') {
539 errx(1, "no if support");
541 if (namelen >= IFNAMSIZ)
542 warn("interface name truncated");
545 strncpy(if_name, arg, namelen);
546 if_name[namelen] = '\0';
548 } else { /* read bandwidth value */
552 bw = strtoul(arg, &end, 0);
553 if (*end == 'K' || *end == 'k') {
556 } else if (*end == 'M' || *end == 'm') {
561 _substrcmp2(end, "Bi", "Bit/s") != 0) ||
562 _substrcmp2(end, "by", "bytes") == 0)
566 errx(EX_DATAERR, "bandwidth too large");
580 compare_points(const void *vp1, const void *vp2)
582 const struct point *p1 = vp1;
583 const struct point *p2 = vp2;
586 res = p1->prob - p2->prob;
588 res = p1->delay - p2->delay;
597 #define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno
600 load_extra_delays(const char *filename, struct dn_profile *p,
601 struct dn_link *link)
603 char line[ED_MAX_LINE_LEN];
610 char profile_name[ED_MAX_NAME_LEN];
611 int delay_first = -1;
613 struct point points[ED_MAX_SAMPLES_NO];
616 /* XXX link never NULL? */
617 p->link_nr = link->link_nr;
619 profile_name[0] = '\0';
620 f = fopen(filename, "r");
622 err(EX_UNAVAILABLE, "fopen: %s", filename);
624 while (fgets(line, ED_MAX_LINE_LEN, f)) { /* read commands */
625 char *s, *cur = line, *name = NULL, *arg = NULL;
631 s = strsep(&cur, ED_SEPARATORS);
632 if (s == NULL || *s == '#')
637 errx(ED_EFMT("too many arguments"));
643 if (name == NULL) /* empty line */
646 errx(ED_EFMT("missing arg for %s"), name);
648 if (!strcasecmp(name, ED_TOK_SAMPLES)) {
650 errx(ED_EFMT("duplicate ``samples'' line"));
652 errx(ED_EFMT("invalid number of samples"));
654 if (samples>ED_MAX_SAMPLES_NO)
655 errx(ED_EFMT("too many samples, maximum is %d"),
658 } else if (!strcasecmp(name, ED_TOK_BW)) {
660 read_bandwidth(arg, &link->bandwidth, buf, sizeof(buf));
661 } else if (!strcasecmp(name, ED_TOK_LOSS)) {
663 errx(ED_EFMT("duplicated token: %s"), name);
664 if (!is_valid_number(arg))
665 errx(ED_EFMT("invalid %s"), arg);
668 errx(ED_EFMT("%s greater than 1.0"), name);
670 } else if (!strcasecmp(name, ED_TOK_NAME)) {
671 if (profile_name[0] != '\0')
672 errx(ED_EFMT("duplicated token: %s"), name);
673 strncpy(profile_name, arg, sizeof(profile_name) - 1);
674 profile_name[sizeof(profile_name)-1] = '\0';
676 } else if (!strcasecmp(name, ED_TOK_DELAY)) {
678 errx(ED_EFMT("duplicated token: %s"), name);
681 } else if (!strcasecmp(name, ED_TOK_PROB)) {
683 errx(ED_EFMT("duplicated token: %s"), name);
686 } else if (do_points) {
687 if (!is_valid_number(name) || !is_valid_number(arg))
688 errx(ED_EFMT("invalid point found"));
690 points[points_no].delay = atof(name);
691 points[points_no].prob = atof(arg);
693 points[points_no].delay = atof(arg);
694 points[points_no].prob = atof(name);
696 if (points[points_no].prob > 1.0)
697 errx(ED_EFMT("probability greater than 1.0"));
700 errx(ED_EFMT("unrecognised command '%s'"), name);
707 warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES);
712 warnx("'%s' not found, assuming no loss", ED_TOK_LOSS);
716 /* make sure that there are enough points. */
717 if (points_no < ED_MIN_SAMPLES_NO)
718 errx(ED_EFMT("too few samples, need at least %d"),
721 qsort(points, points_no, sizeof(struct point), compare_points);
724 for (i = 0; i<points_no-1; ++i) {
725 double y1 = points[i].prob * samples;
726 double x1 = points[i].delay;
727 double y2 = points[i+1].prob * samples;
728 double x2 = points[i+1].delay;
734 for (; ix<stop; ++ix)
737 double m = (y2-y1)/(x2-x1);
738 double c = y1 - m*x1;
739 for (; ix<stop ; ++ix)
740 p->samples[ix] = (ix - c)/m;
743 p->samples_no = samples;
744 p->loss_level = loss * samples;
745 strncpy(p->name, profile_name, sizeof(p->name));
749 * configuration of pipes, schedulers, flowsets.
750 * When we configure a new scheduler, an empty pipe is created, so:
752 * do_pipe = 1 -> "pipe N config ..." only for backward compatibility
753 * sched N+Delta type fifo sched_mask ...
754 * pipe N+Delta <parameters>
755 * flowset N+Delta pipe N+Delta (no parameters)
756 * sched N type wf2q+ sched_mask ...
757 * pipe N <parameters>
759 * do_pipe = 2 -> flowset N config
760 * flowset N parameters
762 * do_pipe = 3 -> sched N config
763 * sched N parameters (default no pipe)
764 * optional Pipe N config ...
768 ipfw_config_pipe(int ac, char **av)
773 struct dn_id *buf, *base;
774 struct dn_sch *sch = NULL;
775 struct dn_link *p = NULL;
776 struct dn_fs *fs = NULL;
777 struct dn_profile *pf = NULL;
778 struct ipfw_flow_id *mask = NULL;
780 uint32_t _foo = 0, *flags = &_foo , *buckets = &_foo;
783 * allocate space for 1 header,
784 * 1 scheduler, 1 link, 1 flowset, 1 profile
786 lmax = sizeof(struct dn_id); /* command header */
787 lmax += sizeof(struct dn_sch) + sizeof(struct dn_link) +
788 sizeof(struct dn_fs) + sizeof(struct dn_profile);
792 if (ac && isdigit(**av)) {
793 i = atoi(*av); av++; ac--;
797 errx(EX_USAGE, "need a pipe/flowset/sched number");
798 base = buf = safe_calloc(1, lmax);
799 /* all commands start with a 'CONFIGURE' and a version */
800 o_next(&buf, sizeof(struct dn_id), DN_CMD_CONFIG);
801 base->id = DN_API_VERSION;
803 switch (co.do_pipe) {
804 case 1: /* "pipe N config ..." */
805 /* Allocate space for the WF2Q+ scheduler, its link
806 * and the FIFO flowset. Set the number, but leave
807 * the scheduler subtype and other parameters to 0
808 * so the kernel will use appropriate defaults.
809 * XXX todo: add a flag to record if a parameter
810 * is actually configured.
811 * If we do a 'pipe config' mask -> sched_mask.
812 * The FIFO scheduler and link are derived from the
813 * WF2Q+ one in the kernel.
815 sch = o_next(&buf, sizeof(*sch), DN_SCH);
816 p = o_next(&buf, sizeof(*p), DN_LINK);
817 fs = o_next(&buf, sizeof(*fs), DN_FS);
820 sch->oid.subtype = 0; /* defaults to WF2Q+ */
821 mask = &sch->sched_mask;
823 buckets = &sch->buckets;
824 *flags |= DN_PIPE_CMD;
828 /* This flowset is only for the FIFO scheduler */
829 fs->fs_nr = i + 2*DN_MAX_ID;
830 fs->sched_nr = i + DN_MAX_ID;
833 case 2: /* "queue N config ... " */
834 fs = o_next(&buf, sizeof(*fs), DN_FS);
836 mask = &fs->flow_mask;
838 buckets = &fs->buckets;
841 case 3: /* "sched N config ..." */
842 sch = o_next(&buf, sizeof(*sch), DN_SCH);
843 fs = o_next(&buf, sizeof(*fs), DN_FS);
845 mask = &sch->sched_mask;
847 buckets = &sch->buckets;
848 /* fs is used only with !MULTIQUEUE schedulers */
849 fs->fs_nr = i + DN_MAX_ID;
853 /* set to -1 those fields for which we want to reuse existing
854 * values from the kernel.
855 * Also, *_nr and subtype = 0 mean reuse the value from the kernel.
856 * XXX todo: support reuse of the mask.
860 for (j = 0; j < sizeof(fs->par)/sizeof(fs->par[0]); j++)
864 int tok = match_token(dummynet_params, *av);
869 NEED(fs, "noerror is only for pipes");
870 fs->flags |= DN_NOERROR;
874 NEED(fs, "plr is only for pipes");
875 NEED1("plr needs argument 0..1\n");
876 d = strtod(av[0], NULL);
881 fs->plr = (int)(d*0x7fffffff);
886 NEED(fs, "queue is only for pipes or flowsets");
887 NEED1("queue needs queue size\n");
889 fs->qsize = strtoul(av[0], &end, 0);
890 if (*end == 'K' || *end == 'k') {
891 fs->flags |= DN_QSIZE_BYTES;
893 } else if (*end == 'B' ||
894 _substrcmp2(end, "by", "bytes") == 0) {
895 fs->flags |= DN_QSIZE_BYTES;
901 NEED(fs, "buckets is only for pipes or flowsets");
902 NEED1("buckets needs argument\n");
903 *buckets = strtoul(av[0], NULL, 0);
910 NEED(mask, "tok_mask");
911 NEED1("mask needs mask specifier\n");
913 * per-flow queue, mask is dst_ip, dst_port,
914 * src_ip, src_port, proto measured in bits
917 bzero(mask, sizeof(*mask));
921 uint32_t *p32 = NULL;
922 uint16_t *p16 = NULL;
923 uint32_t *p20 = NULL;
924 struct in6_addr *pa6 = NULL;
927 tok = match_token(dummynet_params, *av);
932 * special case, all bits significant
933 * except 'extra' (the queue number)
940 n2mask(&mask->dst_ip6, 128);
941 n2mask(&mask->src_ip6, 128);
943 *flags |= DN_HAVE_MASK;
948 *flags |= DN_HAVE_MASK;
963 pa6 = &mask->dst_ip6;
968 pa6 = &mask->src_ip6;
973 p20 = &mask->flow_id6;
977 p16 = &mask->dst_port;
981 p16 = &mask->src_port;
988 ac++; av--; /* backtrack */
992 errx(EX_USAGE, "mask: value missing");
994 a = strtoul(av[0]+1, &end, 0);
996 a = (a == 32) ? ~0 : (1 << a) - 1;
998 a = strtoul(av[0], &end, 0);
1001 else if (p16 != NULL) {
1004 "port mask must be 16 bit");
1006 } else if (p20 != NULL) {
1009 "flow_id mask must be 20 bit");
1011 } else if (pa6 != NULL) {
1014 "in6addr invalid mask len");
1020 "proto mask must be 8 bit");
1021 mask->proto = (uint8_t)a;
1024 *flags |= DN_HAVE_MASK;
1026 } /* end while, config masks */
1032 NEED1("red/gred needs w_q/min_th/max_th/max_p\n");
1033 fs->flags |= DN_IS_RED;
1034 if (tok == TOK_GRED)
1035 fs->flags |= DN_IS_GENTLE_RED;
1037 * the format for parameters is w_q/min_th/max_th/max_p
1039 if ((end = strsep(&av[0], "/"))) {
1040 double w_q = strtod(end, NULL);
1041 if (w_q > 1 || w_q <= 0)
1042 errx(EX_DATAERR, "0 < w_q <= 1");
1043 fs->w_q = (int) (w_q * (1 << SCALE_RED));
1045 if ((end = strsep(&av[0], "/"))) {
1046 fs->min_th = strtoul(end, &end, 0);
1047 if (*end == 'K' || *end == 'k')
1050 if ((end = strsep(&av[0], "/"))) {
1051 fs->max_th = strtoul(end, &end, 0);
1052 if (*end == 'K' || *end == 'k')
1055 if ((end = strsep(&av[0], "/"))) {
1056 double max_p = strtod(end, NULL);
1057 if (max_p > 1 || max_p < 0)
1058 errx(EX_DATAERR, "0 <= max_p <= 1");
1059 fs->max_p = (int)(max_p * (1 << SCALE_RED));
1065 fs->flags |= DN_IS_ECN;
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], &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_ECN) && !(fs->flags & DN_IS_RED))
1191 errx(EX_USAGE, "enable red/gred for ECN");
1193 if (fs->flags & DN_IS_RED) {
1195 int lookup_depth, avg_pkt_size;
1197 if (!(fs->flags & DN_IS_ECN) && (fs->min_th >= fs->max_th))
1198 errx(EX_DATAERR, "min_th %d must be < than max_th %d",
1199 fs->min_th, fs->max_th);
1200 else if ((fs->flags & DN_IS_ECN) && (fs->min_th > fs->max_th))
1201 errx(EX_DATAERR, "min_th %d must be =< than max_th %d",
1202 fs->min_th, fs->max_th);
1204 if (fs->max_th == 0)
1205 errx(EX_DATAERR, "max_th must be > 0");
1208 if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth",
1209 &lookup_depth, &len, NULL, 0) == -1)
1211 if (lookup_depth == 0)
1212 errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth"
1213 " must be greater than zero");
1216 if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size",
1217 &avg_pkt_size, &len, NULL, 0) == -1)
1220 if (avg_pkt_size == 0)
1222 "net.inet.ip.dummynet.red_avg_pkt_size must"
1223 " be greater than zero");
1225 #if 0 /* the following computation is now done in the kernel */
1227 * Ticks needed for sending a medium-sized packet.
1228 * Unfortunately, when we are configuring a WF2Q+ queue, we
1229 * do not have bandwidth information, because that is stored
1230 * in the parent pipe, and also we have multiple queues
1231 * competing for it. So we set s=0, which is not very
1232 * correct. But on the other hand, why do we want RED with
1235 if (p.bandwidth==0) /* this is a WF2Q+ queue */
1238 s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth;
1240 * max idle time (in ticks) before avg queue size becomes 0.
1241 * NOTA: (3/w_q) is approx the value x so that
1242 * (1-w_q)^x < 10^-3.
1244 w_q = ((double)fs->w_q) / (1 << SCALE_RED);
1245 idle = s * 3. / w_q;
1246 fs->lookup_step = (int)idle / lookup_depth;
1247 if (!fs->lookup_step)
1248 fs->lookup_step = 1;
1250 for (t = fs->lookup_step; t > 1; --t)
1252 fs->lookup_weight = (int)(weight * (1 << SCALE_RED));
1253 #endif /* code moved in the kernel */
1257 i = do_cmd(IP_DUMMYNET3, base, (char *)buf - (char *)base);
1260 err(1, "setsockopt(%s)", "IP_DUMMYNET_CONFIGURE");
1264 dummynet_flush(void)
1267 oid_fill(&oid, sizeof(oid), DN_CMD_FLUSH, DN_API_VERSION);
1268 do_cmd(IP_DUMMYNET3, &oid, oid.len);
1271 /* Parse input for 'ipfw [pipe|sched|queue] show [range list]'
1272 * Returns the number of ranges, and possibly stores them
1273 * in the array v of size len.
1276 parse_range(int ac, char *av[], uint32_t *v, int len)
1282 if (v == NULL || len < 2) {
1287 for (s = *av; s != NULL; av++, ac--) {
1288 v[0] = strtoul(s, &endptr, 10);
1289 v[1] = (*endptr != '-') ? v[0] :
1290 strtoul(endptr+1, &endptr, 10);
1291 if (*endptr == '\0') { /* prepare for next round */
1292 s = (ac > 0) ? *(av+1) : NULL;
1294 if (*endptr != ',') {
1295 warn("invalid number: %s", s);
1299 /* continue processing from here */
1305 v[1] >= DN_MAX_ID-1 ||
1306 v[1] >= DN_MAX_ID-1) {
1307 continue; /* invalid entry */
1310 /* translate if 'pipe list' */
1311 if (co.do_pipe == 1) {
1315 v = (n*2 < len) ? v + 2 : base;
1320 /* main entry point for dummynet list functions. co.do_pipe indicates
1321 * which function we want to support.
1322 * av may contain filtering arguments, either individual entries
1323 * or ranges, or lists (space or commas are valid separators).
1324 * Format for a range can be n1-n2 or n3 n4 n5 ...
1325 * In a range n1 must be <= n2, otherwise the range is ignored.
1326 * A number 'n4' is translate in a range 'n4-n4'
1327 * All number must be > 0 and < DN_MAX_ID-1
1330 dummynet_list(int ac, char *av[], int show_counters)
1332 struct dn_id *oid, *x = NULL;
1334 int n; /* # of ranges */
1336 u_int max_size; /* largest obj passed up */
1338 (void)show_counters; // XXX unused, but we should use it.
1340 av++; /* skip 'list' | 'show' word */
1342 n = parse_range(ac, av, NULL, 0); /* Count # of ranges. */
1344 /* Allocate space to store ranges */
1345 l = sizeof(*oid) + sizeof(uint32_t) * n * 2;
1346 oid = safe_calloc(1, l);
1347 oid_fill(oid, l, DN_CMD_GET, DN_API_VERSION);
1349 if (n > 0) /* store ranges in idx */
1350 parse_range(ac, av, (uint32_t *)(oid + 1), n*2);
1352 * Compute the size of the largest object returned. If the
1353 * response leaves at least this much spare space in the
1354 * buffer, then surely the response is complete; otherwise
1355 * there might be a risk of truncation and we will need to
1356 * retry with a larger buffer.
1357 * XXX don't bother with smaller structs.
1359 max_size = sizeof(struct dn_fs);
1360 if (max_size < sizeof(struct dn_sch))
1361 max_size = sizeof(struct dn_sch);
1362 if (max_size < sizeof(struct dn_flow))
1363 max_size = sizeof(struct dn_flow);
1365 switch (co.do_pipe) {
1367 oid->subtype = DN_LINK; /* list pipe */
1370 oid->subtype = DN_FS; /* list queue */
1373 oid->subtype = DN_SCH; /* list sched */
1378 * Ask the kernel an estimate of the required space (result
1379 * in oid.id), unless we are requesting a subset of objects,
1380 * in which case the kernel does not give an exact answer.
1381 * In any case, space might grow in the meantime due to the
1382 * creation of new queues, so we must be prepared to retry.
1387 ret = do_cmd(-IP_DUMMYNET3, oid, (uintptr_t)&l);
1388 if (ret != 0 || oid->id <= sizeof(*oid))
1390 buflen = oid->id + max_size;
1391 oid->len = sizeof(*oid); /* restore */
1393 /* Try a few times, until the buffer fits */
1394 for (i = 0; i < 20; i++) {
1396 x = safe_realloc(x, l);
1397 bcopy(oid, x, oid->len);
1398 ret = do_cmd(-IP_DUMMYNET3, x, (uintptr_t)&l);
1399 if (ret != 0 || x->id <= sizeof(*oid))
1400 goto done; /* no response */
1401 if (l + max_size <= buflen)
1403 buflen *= 2; /* double for next attempt */
1405 list_pipes(x, O_NEXT(x, l));