2 * Codel/FQ_Codel and PIE/FQ_PIE Code:
3 * Copyright (C) 2016 Centre for Advanced Internet Architectures,
4 * Swinburne University of Technology, Melbourne, Australia.
5 * Portions of this code were made possible in part by a gift from
6 * The Comcast Innovation Fund.
7 * Implemented by Rasool Al-Saadi <ralsaadi@swin.edu.au>
9 * Copyright (c) 2002-2003,2010 Luigi Rizzo
11 * Redistribution and use in source forms, with and without modification,
12 * are permitted provided that this entire comment appears intact.
14 * Redistribution in binary form may occur without any restrictions.
15 * Obviously, it would be nice if you gave credit where credit is due
16 * but requiring it would be too onerous.
18 * This software is provided ``AS IS'' without any warranties of any kind.
26 #include <sys/types.h>
27 #include <sys/socket.h>
28 /* XXX there are several sysctl leftover here */
29 #include <sys/sysctl.h>
48 #include <netinet/in.h>
49 #include <netinet/ip_fw.h>
50 #include <netinet/ip_dummynet.h>
51 #include <arpa/inet.h> /* inet_ntoa */
54 static struct _s_x dummynet_params[] = {
56 { "noerror", TOK_NOERROR },
57 { "buckets", TOK_BUCKETS },
58 { "dst-ip", TOK_DSTIP },
59 { "src-ip", TOK_SRCIP },
60 { "dst-port", TOK_DSTPORT },
61 { "src-port", TOK_SRCPORT },
62 { "proto", TOK_PROTO },
63 { "weight", TOK_WEIGHT },
65 { "maxlen", TOK_LMAX },
67 { "mask", TOK_MASK }, /* alias for both */
68 { "sched_mask", TOK_SCHED_MASK },
69 { "flow_mask", TOK_FLOW_MASK },
70 { "droptail", TOK_DROPTAIL },
75 { "codel", TOK_CODEL}, /* Codel AQM */
76 { "fq_codel", TOK_FQ_CODEL}, /* FQ-Codel */
77 { "pie", TOK_PIE}, /* PIE AQM */
78 { "fq_pie", TOK_FQ_PIE}, /* FQ-PIE */
81 { "bandwidth", TOK_BW },
82 { "delay", TOK_DELAY },
85 { "queue", TOK_QUEUE },
86 { "flowset", TOK_FLOWSET },
87 { "sched", TOK_SCHED },
89 { "priority", TOK_PRI },
91 { "flow-id", TOK_FLOWID},
92 { "dst-ipv6", TOK_DSTIP6},
93 { "dst-ip6", TOK_DSTIP6},
94 { "src-ipv6", TOK_SRCIP6},
95 { "src-ip6", TOK_SRCIP6},
96 { "profile", TOK_PROFILE},
97 { "burst", TOK_BURST},
98 { "dummynet-params", TOK_NULL },
99 { NULL, 0 } /* terminator */
103 /* AQM/extra sched parameters tokens*/
104 static struct _s_x aqm_params[] = {
105 { "target", TOK_TARGET},
106 { "interval", TOK_INTERVAL},
107 { "limit", TOK_LIMIT},
108 { "flows", TOK_FLOWS},
109 { "quantum", TOK_QUANTUM},
111 { "noecn", TOK_NO_ECN},
112 { "tupdate", TOK_TUPDATE},
113 { "max_burst", TOK_MAX_BURST},
114 { "max_ecnth", TOK_MAX_ECNTH},
115 { "alpha", TOK_ALPHA},
117 { "capdrop", TOK_CAPDROP},
118 { "nocapdrop", TOK_NO_CAPDROP},
119 { "onoff", TOK_ONOFF},
122 { "derand", TOK_DERAND},
123 { "noderand", TOK_NO_DERAND},
124 { NULL, 0 } /* terminator */
128 #define O_NEXT(p, len) ((void *)((char *)p + len))
131 oid_fill(struct dn_id *oid, int len, int type, uintptr_t id)
139 /* make room in the buffer and move the pointer forward */
141 o_next(struct dn_id **o, int len, int type)
143 struct dn_id *ret = *o;
144 oid_fill(ret, len, type, 0);
145 *o = O_NEXT(*o, len);
153 CODEL_ECN_ENABLED = 1
156 /* PIE flags, from PIE kernel module */
159 PIE_CAPDROP_ENABLED = 2,
160 PIE_ON_OFF_MODE_ENABLED = 4,
161 PIE_DEPRATEEST_ENABLED = 8,
162 PIE_DERAND_ENABLED = 16
165 #define PIE_FIX_POINT_BITS 13
166 #define PIE_SCALE (1L<<PIE_FIX_POINT_BITS)
168 /* integer to time */
170 us_to_time(int t,char *strt)
175 sprintf(strt,"%d", t);
177 sprintf(strt,"%dus", t);
178 else if (t < 1000000)
179 sprintf(strt,"%gms", (float) t / 1000);
181 sprintf(strt,"%gfs", (float) t / 1000000);
185 * returns -1 if s is not a valid time, otherwise, return time in us
188 time_to_us(const char *s)
192 char strt[16]="", stru[16]="";
196 for (i = 0; i<len && (isdigit(s[i]) || s[i]=='.') ; i++)
212 if (!strcasecmp(stru, "us"))
214 if (!strcasecmp(stru, "ms"))
215 return (strtod(strt, NULL) * 1000);
216 if (!strcasecmp(stru, "s"))
217 return (strtod(strt, NULL)*1000000);
223 /* Get AQM or scheduler extra parameters */
225 get_extra_parms(uint32_t nr, char *out, int subtype)
227 struct dn_extra_parms *ep;
229 char strt1[15], strt2[15], strt3[15];
232 /* prepare the request */
233 l = sizeof(struct dn_extra_parms);
234 ep = safe_calloc(1, l);
235 memset(ep, 0, sizeof(*ep));
238 oid_fill(&ep->oid, l, DN_CMD_GET, DN_API_VERSION);
240 ep->oid.subtype = subtype;
243 ret = do_cmd(-IP_DUMMYNET3, ep, (uintptr_t)&l);
246 errx(EX_DATAERR, "Error getting extra parameters\n");
251 if( !strcasecmp(ep->name, "codel")) {
252 us_to_time(ep->par[0], strt1);
253 us_to_time(ep->par[1], strt2);
254 l = sprintf(out, " AQM CoDel target %s interval %s",
256 if (ep->par[2] & CODEL_ECN_ENABLED)
257 l = sprintf(out + l, " ECN");
259 l += sprintf(out + l, " NoECN");
260 } else if( !strcasecmp(ep->name, "pie")) {
261 us_to_time(ep->par[0], strt1);
262 us_to_time(ep->par[1], strt2);
263 us_to_time(ep->par[2], strt3);
264 l = sprintf(out, " AQM type PIE target %s tupdate %s alpha "
265 "%g beta %g max_burst %s max_ecnth %.3g",
268 ep->par[4] / (float) PIE_SCALE,
269 ep->par[5] / (float) PIE_SCALE,
271 ep->par[3] / (float) PIE_SCALE
274 if (ep->par[6] & PIE_ECN_ENABLED)
275 l += sprintf(out + l, " ECN");
277 l += sprintf(out + l, " NoECN");
278 if (ep->par[6] & PIE_CAPDROP_ENABLED)
279 l += sprintf(out + l, " CapDrop");
281 l += sprintf(out + l, " NoCapDrop");
282 if (ep->par[6] & PIE_ON_OFF_MODE_ENABLED)
283 l += sprintf(out + l, " OnOff");
284 if (ep->par[6] & PIE_DEPRATEEST_ENABLED)
285 l += sprintf(out + l, " DRE");
287 l += sprintf(out + l, " TS");
288 if (ep->par[6] & PIE_DERAND_ENABLED)
289 l += sprintf(out + l, " Derand");
291 l += sprintf(out + l, " NoDerand");
296 if (!strcasecmp(ep->name,"FQ_CODEL")) {
297 us_to_time(ep->par[0], strt1);
298 us_to_time(ep->par[1], strt2);
299 l = sprintf(out," FQ_CODEL target %s interval %s"
300 " quantum %jd limit %jd flows %jd",
302 (intmax_t) ep->par[3],
303 (intmax_t) ep->par[4],
304 (intmax_t) ep->par[5]
306 if (ep->par[2] & CODEL_ECN_ENABLED)
307 l += sprintf(out + l, " ECN");
309 l += sprintf(out + l, " NoECN");
310 l += sprintf(out + l, "\n");
311 } else if (!strcasecmp(ep->name,"FQ_PIE")) {
312 us_to_time(ep->par[0], strt1);
313 us_to_time(ep->par[1], strt2);
314 us_to_time(ep->par[2], strt3);
315 l = sprintf(out, " FQ_PIE target %s tupdate %s alpha "
316 "%g beta %g max_burst %s max_ecnth %.3g"
317 " quantum %jd limit %jd flows %jd",
320 ep->par[4] / (float) PIE_SCALE,
321 ep->par[5] / (float) PIE_SCALE,
323 ep->par[3] / (float) PIE_SCALE,
324 (intmax_t) ep->par[7],
325 (intmax_t) ep->par[8],
326 (intmax_t) ep->par[9]
329 if (ep->par[6] & PIE_ECN_ENABLED)
330 l += sprintf(out + l, " ECN");
332 l += sprintf(out + l, " NoECN");
333 if (ep->par[6] & PIE_CAPDROP_ENABLED)
334 l += sprintf(out + l, " CapDrop");
336 l += sprintf(out + l, " NoCapDrop");
337 if (ep->par[6] & PIE_ON_OFF_MODE_ENABLED)
338 l += sprintf(out + l, " OnOff");
339 if (ep->par[6] & PIE_DEPRATEEST_ENABLED)
340 l += sprintf(out + l, " DRE");
342 l += sprintf(out + l, " TS");
343 if (ep->par[6] & PIE_DERAND_ENABLED)
344 l += sprintf(out + l, " Derand");
346 l += sprintf(out + l, " NoDerand");
347 l += sprintf(out + l, "\n");
359 sort_q(void *arg, const void *pa, const void *pb)
361 int rev = (co.do_sort < 0);
362 int field = rev ? -co.do_sort : co.do_sort;
364 const struct dn_flow_queue *a = pa;
365 const struct dn_flow_queue *b = pb;
369 res = a->len - b->len;
372 res = a->len_bytes - b->len_bytes;
375 case 3: /* tot pkts */
376 res = a->tot_pkts - b->tot_pkts;
379 case 4: /* tot bytes */
380 res = a->tot_bytes - b->tot_bytes;
387 return (int)(rev ? res : -res);
391 /* print a mask and header for the subsequent list of flows */
393 print_mask(struct ipfw_flow_id *id)
395 if (!IS_IP6_FLOW_ID(id)) {
397 "mask: %s 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n",
398 id->extra ? "queue," : "",
400 id->src_ip, id->src_port,
401 id->dst_ip, id->dst_port);
404 printf("\n mask: %sproto: 0x%02x, flow_id: 0x%08x, ",
405 id->extra ? "queue," : "",
406 id->proto, id->flow_id6);
407 inet_ntop(AF_INET6, &(id->src_ip6), buf, sizeof(buf));
408 printf("%s/0x%04x -> ", buf, id->src_port);
409 inet_ntop(AF_INET6, &(id->dst_ip6), buf, sizeof(buf));
410 printf("%s/0x%04x\n", buf, id->dst_port);
415 print_header(struct ipfw_flow_id *id)
417 if (!IS_IP6_FLOW_ID(id))
418 printf("BKT Prot ___Source IP/port____ "
419 "____Dest. IP/port____ "
420 "Tot_pkt/bytes Pkt/Byte Drp\n");
422 printf("BKT ___Prot___ _flow-id_ "
423 "______________Source IPv6/port_______________ "
424 "_______________Dest. IPv6/port_______________ "
425 "Tot_pkt/bytes Pkt/Byte Drp\n");
429 list_flow(struct buf_pr *bp, struct dn_flow *ni)
432 struct protoent *pe = NULL;
434 struct ipfw_flow_id *id = &ni->fid;
436 pe = getprotobynumber(id->proto);
437 /* XXX: Should check for IPv4 flows */
438 bprintf(bp, "%3u%c", (ni->oid.id) & 0xff,
439 id->extra ? '*' : ' ');
440 if (!IS_IP6_FLOW_ID(id)) {
442 bprintf(bp, "%-4s ", pe->p_name);
444 bprintf(bp, "%4u ", id->proto);
445 ina.s_addr = htonl(id->src_ip);
446 bprintf(bp, "%15s/%-5d ",
447 inet_ntoa(ina), id->src_port);
448 ina.s_addr = htonl(id->dst_ip);
449 bprintf(bp, "%15s/%-5d ",
450 inet_ntoa(ina), id->dst_port);
452 /* Print IPv6 flows */
454 bprintf(bp, "%9s ", pe->p_name);
456 bprintf(bp, "%9u ", id->proto);
457 bprintf(bp, "%7d %39s/%-5d ", id->flow_id6,
458 inet_ntop(AF_INET6, &(id->src_ip6), buff, sizeof(buff)),
460 bprintf(bp, " %39s/%-5d ",
461 inet_ntop(AF_INET6, &(id->dst_ip6), buff, sizeof(buff)),
464 pr_u64(bp, &ni->tot_pkts, 4);
465 pr_u64(bp, &ni->tot_bytes, 8);
466 bprintf(bp, "%2u %4u %3u",
467 ni->length, ni->len_bytes, ni->drops);
471 print_flowset_parms(struct dn_fs *fs, char *prefix)
476 char red[200]; /* Display RED parameters */
479 if (fs->flags & DN_QSIZE_BYTES) {
481 sprintf(qs, "%d KB", l / 1024);
483 sprintf(qs, "%d B", l);
485 sprintf(qs, "%3d sl.", l);
487 sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff));
491 if (fs->flags & DN_IS_RED) { /* RED parameters */
493 "\n\t %cRED w_q %f min_th %d max_th %d max_p %f",
494 (fs->flags & DN_IS_GENTLE_RED) ? 'G' : ' ',
495 1.0 * fs->w_q / (double)(1 << SCALE_RED),
498 1.0 * fs->max_p / (double)(1 << SCALE_RED));
499 if (fs->flags & DN_IS_ECN)
500 strncat(red, " (ecn)", 6);
502 /* get AQM parameters */
503 } else if (fs->flags & DN_IS_AQM) {
504 get_extra_parms(fs->fs_nr, red, DN_AQM_PARAMS);
507 sprintf(red, "droptail");
510 printf("%s %s%s %d queues (%d buckets) %s\n",
511 prefix, qs, plr, fs->oid.id, fs->buckets, red);
514 printf("q%05d %s%s %d flows (%d buckets) sched %d "
515 "weight %d lmax %d pri %d %s\n",
516 fs->fs_nr, qs, plr, fs->oid.id, fs->buckets,
517 fs->sched_nr, fs->par[0], fs->par[1], fs->par[2], red);
518 if (fs->flags & DN_HAVE_MASK)
519 print_mask(&fs->flow_mask);
524 print_extra_delay_parms(struct dn_profile *p)
527 if (p->samples_no <= 0)
530 loss = p->loss_level;
531 loss /= p->samples_no;
532 printf("\t profile: name \"%s\" loss %f samples %d\n",
533 p->name, loss, p->samples_no);
545 * generic list routine. We expect objects in a specific order, i.e.
546 * PIPES AND SCHEDULERS:
547 * link; scheduler; internal flowset if any; instances
548 * we can tell a pipe from the number.
552 * link i (int queue); scheduler i; si(i) { flowsets() : queues }
555 list_pipes(struct dn_id *oid, struct dn_id *end)
557 char buf[160]; /* pending buffer */
558 int toPrint = 1; /* print header */
563 for (; oid != end; oid = O_NEXT(oid, oid->len)) {
564 if (oid->len < sizeof(*oid))
565 errx(1, "invalid oid len %d\n", oid->len);
570 printf("unrecognized object %d size %d\n", oid->type, oid->len);
572 case DN_TEXT: /* list of attached flowsets */
579 l = (oid->len - sizeof(*oid))/sizeof(d->p[0]);
582 printf(" Children flowsets: ");
583 for (i = 0; i < l; i++)
584 printf("%u ", d->p[i]);
590 printf("answer for cmd %d, len %d\n", oid->type, oid->id);
593 struct dn_sch *s = (struct dn_sch *)oid;
595 printf(" sched %d type %s flags 0x%x %d buckets %d active\n",
597 s->name, s->flags, s->buckets, s->oid.id);
600 get_extra_parms(s->sched_nr, parms, DN_SCH_PARAMS);
603 if (s->flags & DN_HAVE_MASK)
604 print_mask(&s->sched_mask);
610 print_header(&((struct dn_flow *)oid)->fid);
613 list_flow(&bp, (struct dn_flow *)oid);
614 printf("%s\n", bp.buf);
619 struct dn_link *p = (struct dn_link *)oid;
620 double b = p->bandwidth;
624 /* This starts a new object so flush buffer */
628 sprintf(bwbuf, "unlimited ");
629 else if (b >= 1000000)
630 sprintf(bwbuf, "%7.3f Mbit/s", b/1000000);
632 sprintf(bwbuf, "%7.3f Kbit/s", b/1000);
634 sprintf(bwbuf, "%7.3f bit/s ", b);
636 if (humanize_number(burst, sizeof(burst), p->burst,
637 "", HN_AUTOSCALE, 0) < 0 || co.verbose)
638 sprintf(burst, "%d", (int)p->burst);
639 sprintf(buf, "%05d: %s %4d ms burst %s",
640 p->link_nr % DN_MAX_ID, bwbuf, p->delay, burst);
645 print_flowset_parms((struct dn_fs *)oid, buf);
649 print_extra_delay_parms((struct dn_profile *)oid);
651 flush_buf(buf); // XXX does it really go here ?
658 * Delete pipe, queue or scheduler i
661 ipfw_delete_pipe(int do_pipe, int i)
665 uintptr_t a[1]; /* add more if we want a list */
667 oid_fill((void *)&cmd, sizeof(cmd), DN_CMD_DELETE, DN_API_VERSION);
668 cmd.oid.subtype = (do_pipe == 1) ? DN_LINK :
669 ( (do_pipe == 2) ? DN_FS : DN_SCH);
671 i = do_cmd(IP_DUMMYNET3, &cmd, cmd.oid.len);
674 warn("rule %u: setsockopt(IP_DUMMYNET_DEL)", i);
680 * Code to parse delay profiles.
682 * Some link types introduce extra delays in the transmission
683 * of a packet, e.g. because of MAC level framing, contention on
684 * the use of the channel, MAC level retransmissions and so on.
685 * From our point of view, the channel is effectively unavailable
686 * for this extra time, which is constant or variable depending
687 * on the link type. Additionally, packets may be dropped after this
688 * time (e.g. on a wireless link after too many retransmissions).
689 * We can model the additional delay with an empirical curve
690 * that represents its distribution.
692 * cumulative probability
702 * +-------*------------------->
705 * The empirical curve may have both vertical and horizontal lines.
706 * Vertical lines represent constant delay for a range of
707 * probabilities; horizontal lines correspond to a discontinuty
708 * in the delay distribution: the link will use the largest delay
709 * for a given probability.
711 * To pass the curve to dummynet, we must store the parameters
712 * in a file as described below, and issue the command
714 * ipfw pipe <n> config ... bw XXX profile <filename> ...
716 * The file format is the following, with whitespace acting as
717 * a separator and '#' indicating the beginning a comment:
720 * the number of samples used in the internal
721 * representation (2..1024; default 100);
724 * The probability above which packets are lost.
725 * (0.0 <= L <= 1.0, default 1.0 i.e. no loss);
728 * Optional a name (listed by "ipfw pipe show")
729 * to identify the distribution;
731 * "delay prob" | "prob delay"
732 * One of these two lines is mandatory and defines
733 * the format of the following lines with data points.
736 * 2 or more lines representing points in the curve,
737 * with either delay or probability first, according
738 * to the chosen format.
739 * The unit for delay is milliseconds.
741 * Data points does not need to be ordered or equal to the number
742 * specified in the "samples" line. ipfw will sort and interpolate
743 * the curve as needed.
745 * Example of a profile file:
751 0 200 # minimum overhead is 200ms
758 * Internally, we will convert the curve to a fixed number of
759 * samples, and when it is time to transmit a packet we will
760 * model the extra delay as extra bits in the packet.
764 #define ED_MAX_LINE_LEN 256+ED_MAX_NAME_LEN
765 #define ED_TOK_SAMPLES "samples"
766 #define ED_TOK_LOSS "loss-level"
767 #define ED_TOK_NAME "name"
768 #define ED_TOK_DELAY "delay"
769 #define ED_TOK_PROB "prob"
770 #define ED_TOK_BW "bw"
771 #define ED_SEPARATORS " \t\n"
772 #define ED_MIN_SAMPLES_NO 2
775 * returns 1 if s is a non-negative number, with at least one '.'
778 is_valid_number(const char *s)
780 int i, dots_found = 0;
783 for (i = 0; i<len; ++i)
784 if (!isdigit(s[i]) && (s[i] !='.' || ++dots_found > 1))
790 * Take as input a string describing a bandwidth value
791 * and return the numeric bandwidth value.
792 * set clocking interface or bandwidth value
795 read_bandwidth(char *arg, int *bandwidth, char *if_name, int namelen)
797 if (*bandwidth != -1)
798 warnx("duplicate token, override bandwidth value!");
800 if (arg[0] >= 'a' && arg[0] <= 'z') {
802 errx(1, "no if support");
804 if (namelen >= IFNAMSIZ)
805 warn("interface name truncated");
808 strlcpy(if_name, arg, namelen);
810 } else { /* read bandwidth value */
814 bw = strtoul(arg, &end, 0);
815 if (*end == 'K' || *end == 'k') {
818 } else if (*end == 'M' || *end == 'm') {
823 _substrcmp2(end, "Bi", "Bit/s") != 0) ||
824 _substrcmp2(end, "by", "bytes") == 0)
828 errx(EX_DATAERR, "bandwidth too large");
842 compare_points(const void *vp1, const void *vp2)
844 const struct point *p1 = vp1;
845 const struct point *p2 = vp2;
848 res = p1->prob - p2->prob;
850 res = p1->delay - p2->delay;
859 #define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno
862 load_extra_delays(const char *filename, struct dn_profile *p,
863 struct dn_link *link)
865 char line[ED_MAX_LINE_LEN];
872 char profile_name[ED_MAX_NAME_LEN];
873 int delay_first = -1;
875 struct point points[ED_MAX_SAMPLES_NO];
878 /* XXX link never NULL? */
879 p->link_nr = link->link_nr;
881 profile_name[0] = '\0';
882 f = fopen(filename, "r");
884 err(EX_UNAVAILABLE, "fopen: %s", filename);
886 while (fgets(line, ED_MAX_LINE_LEN, f)) { /* read commands */
887 char *s, *cur = line, *name = NULL, *arg = NULL;
893 s = strsep(&cur, ED_SEPARATORS);
894 if (s == NULL || *s == '#')
899 errx(ED_EFMT("too many arguments"));
905 if (name == NULL) /* empty line */
908 errx(ED_EFMT("missing arg for %s"), name);
910 if (!strcasecmp(name, ED_TOK_SAMPLES)) {
912 errx(ED_EFMT("duplicate ``samples'' line"));
914 errx(ED_EFMT("invalid number of samples"));
916 if (samples>ED_MAX_SAMPLES_NO)
917 errx(ED_EFMT("too many samples, maximum is %d"),
920 } else if (!strcasecmp(name, ED_TOK_BW)) {
922 read_bandwidth(arg, &link->bandwidth, buf, sizeof(buf));
923 } else if (!strcasecmp(name, ED_TOK_LOSS)) {
925 errx(ED_EFMT("duplicated token: %s"), name);
926 if (!is_valid_number(arg))
927 errx(ED_EFMT("invalid %s"), arg);
930 errx(ED_EFMT("%s greater than 1.0"), name);
932 } else if (!strcasecmp(name, ED_TOK_NAME)) {
933 if (profile_name[0] != '\0')
934 errx(ED_EFMT("duplicated token: %s"), name);
935 strlcpy(profile_name, arg, sizeof(profile_name));
937 } else if (!strcasecmp(name, ED_TOK_DELAY)) {
939 errx(ED_EFMT("duplicated token: %s"), name);
942 } else if (!strcasecmp(name, ED_TOK_PROB)) {
944 errx(ED_EFMT("duplicated token: %s"), name);
947 } else if (do_points) {
948 if (!is_valid_number(name) || !is_valid_number(arg))
949 errx(ED_EFMT("invalid point found"));
951 points[points_no].delay = atof(name);
952 points[points_no].prob = atof(arg);
954 points[points_no].delay = atof(arg);
955 points[points_no].prob = atof(name);
957 if (points[points_no].prob > 1.0)
958 errx(ED_EFMT("probability greater than 1.0"));
961 errx(ED_EFMT("unrecognised command '%s'"), name);
968 warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES);
973 warnx("'%s' not found, assuming no loss", ED_TOK_LOSS);
977 /* make sure that there are enough points. */
978 if (points_no < ED_MIN_SAMPLES_NO)
979 errx(ED_EFMT("too few samples, need at least %d"),
982 qsort(points, points_no, sizeof(struct point), compare_points);
985 for (i = 0; i<points_no-1; ++i) {
986 double y1 = points[i].prob * samples;
987 double x1 = points[i].delay;
988 double y2 = points[i+1].prob * samples;
989 double x2 = points[i+1].delay;
995 for (; ix<stop; ++ix)
998 double m = (y2-y1)/(x2-x1);
999 double c = y1 - m*x1;
1000 for (; ix<stop ; ++ix)
1001 p->samples[ix] = (ix - c)/m;
1004 p->samples_no = samples;
1005 p->loss_level = loss * samples;
1006 strlcpy(p->name, profile_name, sizeof(p->name));
1011 /* Parse AQM/extra scheduler parameters */
1013 process_extra_parms(int *ac, char **av, struct dn_extra_parms *ep,
1018 /* use kernel defaults */
1019 for (i=0; i<DN_MAX_EXTRA_PARM; i++)
1026 * 0- target, 1- interval, 2- flags,
1028 * 3- quantum, 4- limit, 5- flows
1030 if (type==TOK_CODEL)
1033 ep->par[2] = CODEL_ECN_ENABLED;
1036 int tok = match_token(aqm_params, *av);
1040 if (*ac <= 0 || time_to_us(av[0]) < 0)
1041 errx(EX_DATAERR, "target needs time\n");
1043 ep->par[0] = time_to_us(av[0]);
1048 if (*ac <= 0 || time_to_us(av[0]) < 0)
1049 errx(EX_DATAERR, "interval needs time\n");
1051 ep->par[1] = time_to_us(av[0]);
1056 ep->par[2] = CODEL_ECN_ENABLED;
1059 ep->par[2] &= ~CODEL_ECN_ENABLED;
1061 /* Config fq_codel parameters */
1063 if (type != TOK_FQ_CODEL)
1064 errx(EX_DATAERR, "quantum is not for codel\n");
1065 if (*ac <= 0 || !is_valid_number(av[0]))
1066 errx(EX_DATAERR, "quantum needs number\n");
1068 ep->par[3]= atoi(av[0]);
1073 if (type != TOK_FQ_CODEL)
1074 errx(EX_DATAERR, "limit is not for codel, use queue instead\n");
1075 if (*ac <= 0 || !is_valid_number(av[0]))
1076 errx(EX_DATAERR, "limit needs number\n");
1078 ep->par[4] = atoi(av[0]);
1083 if (type != TOK_FQ_CODEL)
1084 errx(EX_DATAERR, "flows is not for codel\n");
1085 if (*ac <= 0 || !is_valid_number(av[0]))
1086 errx(EX_DATAERR, "flows needs number\n");
1088 ep->par[5] = atoi(av[0]);
1093 printf("%s is Invalid parameter\n", av[-1]);
1100 * 0- target , 1- tupdate, 2- max_burst,
1101 * 3- max_ecnth, 4- alpha,
1104 * 7- quantum, 8- limit, 9- flows
1107 if ( type == TOK_PIE)
1108 ep->par[6] = PIE_CAPDROP_ENABLED | PIE_DEPRATEEST_ENABLED
1109 | PIE_DERAND_ENABLED;
1111 /* for FQ-PIE, use TS mode */
1112 ep->par[6] = PIE_CAPDROP_ENABLED | PIE_DERAND_ENABLED
1116 int tok = match_token(aqm_params, *av);
1120 if (*ac <= 0 || time_to_us(av[0]) < 0)
1121 errx(EX_DATAERR, "target needs time\n");
1123 ep->par[0] = time_to_us(av[0]);
1128 if (*ac <= 0 || time_to_us(av[0]) < 0)
1129 errx(EX_DATAERR, "tupdate needs time\n");
1131 ep->par[1] = time_to_us(av[0]);
1136 if (*ac <= 0 || time_to_us(av[0]) < 0)
1137 errx(EX_DATAERR, "max_burst needs time\n");
1139 ep->par[2] = time_to_us(av[0]);
1144 if (*ac <= 0 || !is_valid_number(av[0]))
1145 errx(EX_DATAERR, "max_ecnth needs number\n");
1147 ep->par[3] = atof(av[0]) * PIE_SCALE;
1152 if (*ac <= 0 || !is_valid_number(av[0]))
1153 errx(EX_DATAERR, "alpha needs number\n");
1155 ep->par[4] = atof(av[0]) * PIE_SCALE;
1160 if (*ac <= 0 || !is_valid_number(av[0]))
1161 errx(EX_DATAERR, "beta needs number\n");
1163 ep->par[5] = atof(av[0]) * PIE_SCALE;
1168 ep->par[6] |= PIE_ECN_ENABLED;
1171 ep->par[6] &= ~PIE_ECN_ENABLED;
1175 ep->par[6] |= PIE_CAPDROP_ENABLED;
1177 case TOK_NO_CAPDROP:
1178 ep->par[6] &= ~PIE_CAPDROP_ENABLED;
1182 ep->par[6] |= PIE_ON_OFF_MODE_ENABLED;
1186 ep->par[6] |= PIE_DEPRATEEST_ENABLED;
1190 ep->par[6] &= ~PIE_DEPRATEEST_ENABLED;
1194 ep->par[6] |= PIE_DERAND_ENABLED;
1197 ep->par[6] &= ~PIE_DERAND_ENABLED;
1200 /* Config fq_pie parameters */
1202 if (type != TOK_FQ_PIE)
1203 errx(EX_DATAERR, "quantum is not for pie\n");
1204 if (*ac <= 0 || !is_valid_number(av[0]))
1205 errx(EX_DATAERR, "quantum needs number\n");
1207 ep->par[7]= atoi(av[0]);
1212 if (type != TOK_FQ_PIE)
1213 errx(EX_DATAERR, "limit is not for pie, use queue instead\n");
1214 if (*ac <= 0 || !is_valid_number(av[0]))
1215 errx(EX_DATAERR, "limit needs number\n");
1217 ep->par[8] = atoi(av[0]);
1222 if (type != TOK_FQ_PIE)
1223 errx(EX_DATAERR, "flows is not for pie\n");
1224 if (*ac <= 0 || !is_valid_number(av[0]))
1225 errx(EX_DATAERR, "flows needs number\n");
1227 ep->par[9] = atoi(av[0]);
1233 printf("%s is invalid parameter\n", av[-1]);
1246 * configuration of pipes, schedulers, flowsets.
1247 * When we configure a new scheduler, an empty pipe is created, so:
1249 * do_pipe = 1 -> "pipe N config ..." only for backward compatibility
1250 * sched N+Delta type fifo sched_mask ...
1251 * pipe N+Delta <parameters>
1252 * flowset N+Delta pipe N+Delta (no parameters)
1253 * sched N type wf2q+ sched_mask ...
1254 * pipe N <parameters>
1256 * do_pipe = 2 -> flowset N config
1257 * flowset N parameters
1259 * do_pipe = 3 -> sched N config
1260 * sched N parameters (default no pipe)
1261 * optional Pipe N config ...
1265 ipfw_config_pipe(int ac, char **av)
1270 struct dn_id *buf, *base;
1271 struct dn_sch *sch = NULL;
1272 struct dn_link *p = NULL;
1273 struct dn_fs *fs = NULL;
1274 struct dn_profile *pf = NULL;
1275 struct ipfw_flow_id *mask = NULL;
1277 struct dn_extra_parms *aqm_extra;
1278 struct dn_extra_parms *sch_extra;
1283 uint32_t _foo = 0, *flags = &_foo , *buckets = &_foo;
1286 * allocate space for 1 header,
1287 * 1 scheduler, 1 link, 1 flowset, 1 profile
1289 lmax = sizeof(struct dn_id); /* command header */
1290 lmax += sizeof(struct dn_sch) + sizeof(struct dn_link) +
1291 sizeof(struct dn_fs) + sizeof(struct dn_profile);
1295 lmax_extra = sizeof(struct dn_extra_parms);
1296 /* two lmax_extra because one for AQM params and another
1299 lmax += lmax_extra*2;
1304 if (ac && isdigit(**av)) {
1305 i = atoi(*av); av++; ac--;
1309 errx(EX_USAGE, "need a pipe/flowset/sched number");
1310 base = buf = safe_calloc(1, lmax);
1311 /* all commands start with a 'CONFIGURE' and a version */
1312 o_next(&buf, sizeof(struct dn_id), DN_CMD_CONFIG);
1313 base->id = DN_API_VERSION;
1315 switch (co.do_pipe) {
1316 case 1: /* "pipe N config ..." */
1317 /* Allocate space for the WF2Q+ scheduler, its link
1318 * and the FIFO flowset. Set the number, but leave
1319 * the scheduler subtype and other parameters to 0
1320 * so the kernel will use appropriate defaults.
1321 * XXX todo: add a flag to record if a parameter
1322 * is actually configured.
1323 * If we do a 'pipe config' mask -> sched_mask.
1324 * The FIFO scheduler and link are derived from the
1325 * WF2Q+ one in the kernel.
1328 sch_extra = o_next(&buf, lmax_extra, DN_TEXT);
1329 sch_extra ->oid.subtype = 0; /* don't configure scheduler */
1331 sch = o_next(&buf, sizeof(*sch), DN_SCH);
1332 p = o_next(&buf, sizeof(*p), DN_LINK);
1334 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT);
1335 aqm_extra ->oid.subtype = 0; /* don't configure AQM */
1337 fs = o_next(&buf, sizeof(*fs), DN_FS);
1340 sch->oid.subtype = 0; /* defaults to WF2Q+ */
1341 mask = &sch->sched_mask;
1342 flags = &sch->flags;
1343 buckets = &sch->buckets;
1344 *flags |= DN_PIPE_CMD;
1348 /* This flowset is only for the FIFO scheduler */
1349 fs->fs_nr = i + 2*DN_MAX_ID;
1350 fs->sched_nr = i + DN_MAX_ID;
1353 case 2: /* "queue N config ... " */
1355 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT);
1356 aqm_extra ->oid.subtype = 0;
1358 fs = o_next(&buf, sizeof(*fs), DN_FS);
1360 mask = &fs->flow_mask;
1362 buckets = &fs->buckets;
1365 case 3: /* "sched N config ..." */
1367 sch_extra = o_next(&buf, lmax_extra, DN_TEXT);
1368 sch_extra ->oid.subtype = 0;
1370 sch = o_next(&buf, sizeof(*sch), DN_SCH);
1372 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT);
1373 aqm_extra ->oid.subtype = 0;
1375 fs = o_next(&buf, sizeof(*fs), DN_FS);
1377 mask = &sch->sched_mask;
1378 flags = &sch->flags;
1379 buckets = &sch->buckets;
1380 /* fs is used only with !MULTIQUEUE schedulers */
1381 fs->fs_nr = i + DN_MAX_ID;
1385 /* set to -1 those fields for which we want to reuse existing
1386 * values from the kernel.
1387 * Also, *_nr and subtype = 0 mean reuse the value from the kernel.
1388 * XXX todo: support reuse of the mask.
1392 for (j = 0; j < sizeof(fs->par)/sizeof(fs->par[0]); j++)
1396 int tok = match_token(dummynet_params, *av);
1401 NEED(fs, "noerror is only for pipes");
1402 fs->flags |= DN_NOERROR;
1406 NEED(fs, "plr is only for pipes");
1407 NEED1("plr needs argument 0..1\n");
1408 d = strtod(av[0], NULL);
1413 fs->plr = (int)(d*0x7fffffff);
1418 NEED(fs, "queue is only for pipes or flowsets");
1419 NEED1("queue needs queue size\n");
1421 fs->qsize = strtoul(av[0], &end, 0);
1422 if (*end == 'K' || *end == 'k') {
1423 fs->flags |= DN_QSIZE_BYTES;
1425 } else if (*end == 'B' ||
1426 _substrcmp2(end, "by", "bytes") == 0) {
1427 fs->flags |= DN_QSIZE_BYTES;
1433 NEED(fs, "buckets is only for pipes or flowsets");
1434 NEED1("buckets needs argument\n");
1435 *buckets = strtoul(av[0], NULL, 0);
1440 case TOK_SCHED_MASK:
1442 NEED(mask, "tok_mask");
1443 NEED1("mask needs mask specifier\n");
1445 * per-flow queue, mask is dst_ip, dst_port,
1446 * src_ip, src_port, proto measured in bits
1449 bzero(mask, sizeof(*mask));
1453 uint32_t *p32 = NULL;
1454 uint16_t *p16 = NULL;
1455 uint32_t *p20 = NULL;
1456 struct in6_addr *pa6 = NULL;
1459 tok = match_token(dummynet_params, *av);
1464 * special case, all bits significant
1465 * except 'extra' (the queue number)
1469 mask->dst_port = ~0;
1470 mask->src_port = ~0;
1472 n2mask(&mask->dst_ip6, 128);
1473 n2mask(&mask->src_ip6, 128);
1474 mask->flow_id6 = ~0;
1475 *flags |= DN_HAVE_MASK;
1480 *flags |= DN_HAVE_MASK;
1484 mask->addr_type = 4;
1485 p32 = &mask->dst_ip;
1489 mask->addr_type = 4;
1490 p32 = &mask->src_ip;
1494 mask->addr_type = 6;
1495 pa6 = &mask->dst_ip6;
1499 mask->addr_type = 6;
1500 pa6 = &mask->src_ip6;
1504 mask->addr_type = 6;
1505 p20 = &mask->flow_id6;
1509 p16 = &mask->dst_port;
1513 p16 = &mask->src_port;
1520 ac++; av--; /* backtrack */
1524 errx(EX_USAGE, "mask: value missing");
1525 if (*av[0] == '/') {
1526 a = strtoul(av[0]+1, &end, 0);
1528 a = (a == 32) ? ~0 : (1 << a) - 1;
1530 a = strtoul(av[0], &end, 0);
1533 else if (p16 != NULL) {
1536 "port mask must be 16 bit");
1538 } else if (p20 != NULL) {
1541 "flow_id mask must be 20 bit");
1543 } else if (pa6 != NULL) {
1546 "in6addr invalid mask len");
1552 "proto mask must be 8 bit");
1553 mask->proto = (uint8_t)a;
1556 *flags |= DN_HAVE_MASK;
1558 } /* end while, config masks */
1564 NEED(fs, "codel/pie is only for flowsets");
1566 fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED);
1567 fs->flags |= DN_IS_AQM;
1569 strlcpy(aqm_extra->name, av[-1],
1570 sizeof(aqm_extra->name));
1571 aqm_extra->oid.subtype = DN_AQM_PARAMS;
1573 process_extra_parms(&ac, av, aqm_extra, tok);
1578 if (!strcmp(av[-1],"type"))
1579 errx(EX_DATAERR, "use type before fq_codel/fq_pie");
1581 NEED(sch, "fq_codel/fq_pie is only for schd");
1582 strlcpy(sch_extra->name, av[-1],
1583 sizeof(sch_extra->name));
1584 sch_extra->oid.subtype = DN_SCH_PARAMS;
1585 process_extra_parms(&ac, av, sch_extra, tok);
1590 NEED1("red/gred needs w_q/min_th/max_th/max_p\n");
1591 fs->flags |= DN_IS_RED;
1592 if (tok == TOK_GRED)
1593 fs->flags |= DN_IS_GENTLE_RED;
1595 * the format for parameters is w_q/min_th/max_th/max_p
1597 if ((end = strsep(&av[0], "/"))) {
1598 double w_q = strtod(end, NULL);
1599 if (w_q > 1 || w_q <= 0)
1600 errx(EX_DATAERR, "0 < w_q <= 1");
1601 fs->w_q = (int) (w_q * (1 << SCALE_RED));
1603 if ((end = strsep(&av[0], "/"))) {
1604 fs->min_th = strtoul(end, &end, 0);
1605 if (*end == 'K' || *end == 'k')
1608 if ((end = strsep(&av[0], "/"))) {
1609 fs->max_th = strtoul(end, &end, 0);
1610 if (*end == 'K' || *end == 'k')
1613 if ((end = strsep(&av[0], "/"))) {
1614 double max_p = strtod(end, NULL);
1615 if (max_p > 1 || max_p < 0)
1616 errx(EX_DATAERR, "0 <= max_p <= 1");
1617 fs->max_p = (int)(max_p * (1 << SCALE_RED));
1623 fs->flags |= DN_IS_ECN;
1627 NEED(fs, "droptail is only for flowsets");
1628 fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED);
1632 NEED(p, "bw is only for links");
1633 NEED1("bw needs bandwidth or interface\n");
1634 read_bandwidth(av[0], &p->bandwidth, NULL, 0);
1639 NEED(p, "delay is only for links");
1640 NEED1("delay needs argument 0..10000ms\n");
1641 p->delay = strtoul(av[0], NULL, 0);
1647 NEED(sch, "type is only for schedulers");
1648 NEED1("type needs a string");
1650 if (l == 0 || l > 15)
1651 errx(1, "type %s too long\n", av[0]);
1652 strlcpy(sch->name, av[0], sizeof(sch->name));
1653 sch->oid.subtype = 0; /* use string */
1655 /* if fq_codel is selected, consider all tokens after it
1658 if (!strcasecmp(av[0],"fq_codel") || !strcasecmp(av[0],"fq_pie")){
1659 strlcpy(sch_extra->name, av[0],
1660 sizeof(sch_extra->name));
1661 sch_extra->oid.subtype = DN_SCH_PARAMS;
1662 process_extra_parms(&ac, av, sch_extra, tok);
1673 NEED(fs, "weight is only for flowsets");
1674 NEED1("weight needs argument\n");
1675 fs->par[0] = strtol(av[0], &end, 0);
1680 NEED(fs, "lmax is only for flowsets");
1681 NEED1("lmax needs argument\n");
1682 fs->par[1] = strtol(av[0], &end, 0);
1687 NEED(fs, "priority is only for flowsets");
1688 NEED1("priority needs argument\n");
1689 fs->par[2] = strtol(av[0], &end, 0);
1695 NEED(fs, "pipe/sched");
1696 NEED1("pipe/link/sched needs number\n");
1697 fs->sched_nr = strtoul(av[0], &end, 0);
1702 NEED((!pf), "profile already set");
1705 NEED1("extra delay needs the file name\n");
1706 pf = o_next(&buf, sizeof(*pf), DN_PROFILE);
1707 load_extra_delays(av[0], pf, p); //XXX can't fail?
1714 NEED1("burst needs argument\n");
1716 if (expand_number(av[0], &p->burst) < 0)
1717 if (errno != ERANGE)
1719 "burst: invalid argument");
1720 if (errno || p->burst > (1ULL << 48) - 1)
1722 "burst: out of range (0..2^48-1)");
1727 errx(EX_DATAERR, "unrecognised option ``%s''", av[-1]);
1731 /* check validity of parameters */
1733 if (p->delay > 10000)
1734 errx(EX_DATAERR, "delay must be < 10000");
1735 if (p->bandwidth == -1)
1739 /* XXX accept a 0 scheduler to keep the default */
1740 if (fs->flags & DN_QSIZE_BYTES) {
1744 len = sizeof(limit);
1745 if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit",
1746 &limit, &len, NULL, 0) == -1)
1748 if (fs->qsize > limit)
1749 errx(EX_DATAERR, "queue size must be < %ldB", limit);
1754 len = sizeof(limit);
1755 if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit",
1756 &limit, &len, NULL, 0) == -1)
1758 if (fs->qsize > limit)
1759 errx(EX_DATAERR, "2 <= queue size <= %ld", limit);
1763 if ((fs->flags & DN_IS_ECN) && !((fs->flags & DN_IS_RED)||
1764 (fs->flags & DN_IS_AQM)))
1765 errx(EX_USAGE, "ECN can be used with red/gred/"
1766 "codel/fq_codel only!");
1768 if ((fs->flags & DN_IS_ECN) && !(fs->flags & DN_IS_RED))
1769 errx(EX_USAGE, "enable red/gred for ECN");
1773 if (fs->flags & DN_IS_RED) {
1775 int lookup_depth, avg_pkt_size;
1777 if (!(fs->flags & DN_IS_ECN) && (fs->min_th >= fs->max_th))
1778 errx(EX_DATAERR, "min_th %d must be < than max_th %d",
1779 fs->min_th, fs->max_th);
1780 else if ((fs->flags & DN_IS_ECN) && (fs->min_th > fs->max_th))
1781 errx(EX_DATAERR, "min_th %d must be =< than max_th %d",
1782 fs->min_th, fs->max_th);
1784 if (fs->max_th == 0)
1785 errx(EX_DATAERR, "max_th must be > 0");
1788 if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth",
1789 &lookup_depth, &len, NULL, 0) == -1)
1791 if (lookup_depth == 0)
1792 errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth"
1793 " must be greater than zero");
1796 if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size",
1797 &avg_pkt_size, &len, NULL, 0) == -1)
1800 if (avg_pkt_size == 0)
1802 "net.inet.ip.dummynet.red_avg_pkt_size must"
1803 " be greater than zero");
1805 #if 0 /* the following computation is now done in the kernel */
1807 * Ticks needed for sending a medium-sized packet.
1808 * Unfortunately, when we are configuring a WF2Q+ queue, we
1809 * do not have bandwidth information, because that is stored
1810 * in the parent pipe, and also we have multiple queues
1811 * competing for it. So we set s=0, which is not very
1812 * correct. But on the other hand, why do we want RED with
1815 if (p.bandwidth==0) /* this is a WF2Q+ queue */
1818 s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth;
1820 * max idle time (in ticks) before avg queue size becomes 0.
1821 * NOTA: (3/w_q) is approx the value x so that
1822 * (1-w_q)^x < 10^-3.
1824 w_q = ((double)fs->w_q) / (1 << SCALE_RED);
1825 idle = s * 3. / w_q;
1826 fs->lookup_step = (int)idle / lookup_depth;
1827 if (!fs->lookup_step)
1828 fs->lookup_step = 1;
1830 for (t = fs->lookup_step; t > 1; --t)
1832 fs->lookup_weight = (int)(weight * (1 << SCALE_RED));
1833 #endif /* code moved in the kernel */
1837 i = do_cmd(IP_DUMMYNET3, base, (char *)buf - (char *)base);
1840 err(1, "setsockopt(%s)", "IP_DUMMYNET_CONFIGURE");
1844 dummynet_flush(void)
1847 oid_fill(&oid, sizeof(oid), DN_CMD_FLUSH, DN_API_VERSION);
1848 do_cmd(IP_DUMMYNET3, &oid, oid.len);
1851 /* Parse input for 'ipfw [pipe|sched|queue] show [range list]'
1852 * Returns the number of ranges, and possibly stores them
1853 * in the array v of size len.
1856 parse_range(int ac, char *av[], uint32_t *v, int len)
1862 if (v == NULL || len < 2) {
1867 for (s = *av; s != NULL; av++, ac--) {
1868 v[0] = strtoul(s, &endptr, 10);
1869 v[1] = (*endptr != '-') ? v[0] :
1870 strtoul(endptr+1, &endptr, 10);
1871 if (*endptr == '\0') { /* prepare for next round */
1872 s = (ac > 0) ? *(av+1) : NULL;
1874 if (*endptr != ',') {
1875 warn("invalid number: %s", s);
1879 /* continue processing from here */
1885 v[0] >= DN_MAX_ID-1 ||
1886 v[1] >= DN_MAX_ID-1) {
1887 continue; /* invalid entry */
1890 /* translate if 'pipe list' */
1891 if (co.do_pipe == 1) {
1895 v = (n*2 < len) ? v + 2 : base;
1900 /* main entry point for dummynet list functions. co.do_pipe indicates
1901 * which function we want to support.
1902 * av may contain filtering arguments, either individual entries
1903 * or ranges, or lists (space or commas are valid separators).
1904 * Format for a range can be n1-n2 or n3 n4 n5 ...
1905 * In a range n1 must be <= n2, otherwise the range is ignored.
1906 * A number 'n4' is translate in a range 'n4-n4'
1907 * All number must be > 0 and < DN_MAX_ID-1
1910 dummynet_list(int ac, char *av[], int show_counters)
1912 struct dn_id *oid, *x = NULL;
1914 int n; /* # of ranges */
1916 u_int max_size; /* largest obj passed up */
1918 (void)show_counters; // XXX unused, but we should use it.
1920 av++; /* skip 'list' | 'show' word */
1922 n = parse_range(ac, av, NULL, 0); /* Count # of ranges. */
1924 /* Allocate space to store ranges */
1925 l = sizeof(*oid) + sizeof(uint32_t) * n * 2;
1926 oid = safe_calloc(1, l);
1927 oid_fill(oid, l, DN_CMD_GET, DN_API_VERSION);
1929 if (n > 0) /* store ranges in idx */
1930 parse_range(ac, av, (uint32_t *)(oid + 1), n*2);
1932 * Compute the size of the largest object returned. If the
1933 * response leaves at least this much spare space in the
1934 * buffer, then surely the response is complete; otherwise
1935 * there might be a risk of truncation and we will need to
1936 * retry with a larger buffer.
1937 * XXX don't bother with smaller structs.
1939 max_size = sizeof(struct dn_fs);
1940 if (max_size < sizeof(struct dn_sch))
1941 max_size = sizeof(struct dn_sch);
1942 if (max_size < sizeof(struct dn_flow))
1943 max_size = sizeof(struct dn_flow);
1945 switch (co.do_pipe) {
1947 oid->subtype = DN_LINK; /* list pipe */
1950 oid->subtype = DN_FS; /* list queue */
1953 oid->subtype = DN_SCH; /* list sched */
1958 * Ask the kernel an estimate of the required space (result
1959 * in oid.id), unless we are requesting a subset of objects,
1960 * in which case the kernel does not give an exact answer.
1961 * In any case, space might grow in the meantime due to the
1962 * creation of new queues, so we must be prepared to retry.
1967 ret = do_cmd(-IP_DUMMYNET3, oid, (uintptr_t)&l);
1968 if (ret != 0 || oid->id <= sizeof(*oid))
1970 buflen = oid->id + max_size;
1971 oid->len = sizeof(*oid); /* restore */
1973 /* Try a few times, until the buffer fits */
1974 for (i = 0; i < 20; i++) {
1976 x = safe_realloc(x, l);
1977 bcopy(oid, x, oid->len);
1978 ret = do_cmd(-IP_DUMMYNET3, x, (uintptr_t)&l);
1979 if (ret != 0 || x->id <= sizeof(*oid))
1980 goto done; /* no response */
1981 if (l + max_size <= buflen)
1983 buflen *= 2; /* double for next attempt */
1985 list_pipes(x, O_NEXT(x, l));