1 /*-
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>
8 *
9 * Copyright (c) 2002-2003,2010 Luigi Rizzo
10 *
11 * Redistribution and use in source forms, with and without modification,
12 * are permitted provided that this entire comment appears intact.
13 *
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.
17 *
18 * This software is provided ``AS IS'' without any warranties of any kind.
19 *
20 * dummynet support
21 */
22
23 #define NEW_AQM
24 #include <sys/limits.h>
25 #include <sys/param.h>
26 #include <sys/socket.h>
27 /* XXX there are several sysctl leftover here */
28 #include <sys/sysctl.h>
29
30 #include "ipfw2.h"
31
32 #ifdef NEW_AQM
33 #include <stdint.h>
34 #endif
35
36 #include <ctype.h>
37 #include <err.h>
38 #include <errno.h>
39 #include <libutil.h>
40 #include <netdb.h>
41 #include <stdio.h>
42 #include <stdlib.h>
43 #include <string.h>
44 #include <sysexits.h>
45
46 #include <net/if.h>
47 #include <netinet/in.h>
48 #include <netinet/ip_fw.h>
49 #include <netinet/ip_dummynet.h>
50 #include <arpa/inet.h> /* inet_ntoa */
51
52
53 static struct _s_x dummynet_params[] = {
54 { "plr", TOK_PLR },
55 { "noerror", TOK_NOERROR },
56 { "buckets", TOK_BUCKETS },
57 { "dst-ip", TOK_DSTIP },
58 { "src-ip", TOK_SRCIP },
59 { "dst-port", TOK_DSTPORT },
60 { "src-port", TOK_SRCPORT },
61 { "proto", TOK_PROTO },
62 { "weight", TOK_WEIGHT },
63 { "lmax", TOK_LMAX },
64 { "maxlen", TOK_LMAX },
65 { "all", TOK_ALL },
66 { "mask", TOK_MASK }, /* alias for both */
67 { "sched_mask", TOK_SCHED_MASK },
68 { "flow_mask", TOK_FLOW_MASK },
69 { "droptail", TOK_DROPTAIL },
70 { "ecn", TOK_ECN },
71 { "red", TOK_RED },
72 { "gred", TOK_GRED },
73 #ifdef NEW_AQM
74 { "codel", TOK_CODEL}, /* Codel AQM */
75 { "fq_codel", TOK_FQ_CODEL}, /* FQ-Codel */
76 { "pie", TOK_PIE}, /* PIE AQM */
77 { "fq_pie", TOK_FQ_PIE}, /* FQ-PIE */
78 #endif
79 { "bw", TOK_BW },
80 { "bandwidth", TOK_BW },
81 { "delay", TOK_DELAY },
82 { "link", TOK_LINK },
83 { "pipe", TOK_PIPE },
84 { "queue", TOK_QUEUE },
85 { "flowset", TOK_FLOWSET },
86 { "sched", TOK_SCHED },
87 { "pri", TOK_PRI },
88 { "priority", TOK_PRI },
89 { "type", TOK_TYPE },
90 { "flow-id", TOK_FLOWID},
91 { "dst-ipv6", TOK_DSTIP6},
92 { "dst-ip6", TOK_DSTIP6},
93 { "src-ipv6", TOK_SRCIP6},
94 { "src-ip6", TOK_SRCIP6},
95 { "profile", TOK_PROFILE},
96 { "burst", TOK_BURST},
97 { "dummynet-params", TOK_NULL },
98 { NULL, 0 } /* terminator */
99 };
100
101 #ifdef NEW_AQM
102 /* AQM/extra sched parameters tokens*/
103 static struct _s_x aqm_params[] = {
104 { "target", TOK_TARGET},
105 { "interval", TOK_INTERVAL},
106 { "limit", TOK_LIMIT},
107 { "flows", TOK_FLOWS},
108 { "quantum", TOK_QUANTUM},
109 { "ecn", TOK_ECN},
110 { "noecn", TOK_NO_ECN},
111 { "tupdate", TOK_TUPDATE},
112 { "max_burst", TOK_MAX_BURST},
113 { "max_ecnth", TOK_MAX_ECNTH},
114 { "alpha", TOK_ALPHA},
115 { "beta", TOK_BETA},
116 { "capdrop", TOK_CAPDROP},
117 { "nocapdrop", TOK_NO_CAPDROP},
118 { "onoff", TOK_ONOFF},
119 { "dre", TOK_DRE},
120 { "ts", TOK_TS},
121 { "derand", TOK_DERAND},
122 { "noderand", TOK_NO_DERAND},
123 { NULL, 0 } /* terminator */
124 };
125 #endif
126
127 #define O_NEXT(p, len) ((void *)((char *)p + len))
128
129 static void
oid_fill(struct dn_id * oid,int len,int type,uintptr_t id)130 oid_fill(struct dn_id *oid, int len, int type, uintptr_t id)
131 {
132 oid->len = len;
133 oid->type = type;
134 oid->subtype = 0;
135 oid->id = id;
136 }
137
138 /* make room in the buffer and move the pointer forward */
139 static void *
o_next(struct dn_id ** o,int len,int type)140 o_next(struct dn_id **o, int len, int type)
141 {
142 struct dn_id *ret = *o;
143 oid_fill(ret, len, type, 0);
144 *o = O_NEXT(*o, len);
145 return ret;
146 }
147
148 #ifdef NEW_AQM
149
150 /* Codel flags */
151 enum {
152 CODEL_ECN_ENABLED = 1
153 };
154
155 /* PIE flags, from PIE kernel module */
156 enum {
157 PIE_ECN_ENABLED = 1,
158 PIE_CAPDROP_ENABLED = 2,
159 PIE_ON_OFF_MODE_ENABLED = 4,
160 PIE_DEPRATEEST_ENABLED = 8,
161 PIE_DERAND_ENABLED = 16
162 };
163
164 #define PIE_FIX_POINT_BITS 13
165 #define PIE_SCALE (1L<<PIE_FIX_POINT_BITS)
166
167 /* integer to time */
168 static void
us_to_time(int t,char * strt)169 us_to_time(int t, char *strt)
170 {
171 if (t < 0)
172 strt[0]='\0';
173 else if ( t==0 )
174 sprintf(strt,"%d", t);
175 else if (t< 1000)
176 sprintf(strt,"%dus", t);
177 else if (t < 1000000)
178 sprintf(strt,"%gms", (float) t / 1000);
179 else
180 sprintf(strt,"%gfs", (float) t / 1000000);
181 }
182
183 /*
184 * returns -1 if s is not a valid time, otherwise, return time in us
185 */
186 static long
time_to_us(const char * s)187 time_to_us(const char *s)
188 {
189 int i, dots = 0;
190 int len = strlen(s);
191 char strt[16]="", stru[16]="";
192
193 if (len>15)
194 return -1;
195 for (i = 0; i<len && (isdigit(s[i]) || s[i]=='.') ; i++)
196 if (s[i]=='.') {
197 if (dots)
198 return -1;
199 else
200 dots++;
201 }
202
203 if (!i)
204 return -1;
205 strncpy(strt, s, i);
206 if (i<len)
207 strcpy(stru, s+i);
208 else
209 strcpy(stru, "ms");
210
211 if (!strcasecmp(stru, "us"))
212 return atol(strt);
213 if (!strcasecmp(stru, "ms"))
214 return (strtod(strt, NULL) * 1000);
215 if (!strcasecmp(stru, "s"))
216 return (strtod(strt, NULL)*1000000);
217
218 return -1;
219 }
220
221
222 /* Get AQM or scheduler extra parameters */
223 static void
get_extra_parms(uint32_t nr,char * out,int subtype)224 get_extra_parms(uint32_t nr, char *out, int subtype)
225 {
226 struct dn_extra_parms *ep;
227 int ret;
228 char strt1[15], strt2[15], strt3[15];
229 u_int l;
230
231 /* prepare the request */
232 l = sizeof(struct dn_extra_parms);
233 ep = safe_calloc(1, l);
234 memset(ep, 0, sizeof(*ep));
235 *out = '\0';
236
237 oid_fill(&ep->oid, l, DN_CMD_GET, DN_API_VERSION);
238 ep->oid.len = l;
239 ep->oid.subtype = subtype;
240 ep->nr = nr;
241
242 ret = do_cmd(-IP_DUMMYNET3, ep, (uintptr_t)&l);
243 if (ret) {
244 free(ep);
245 errx(EX_DATAERR, "Error getting extra parameters\n");
246 }
247
248 switch (subtype) {
249 case DN_AQM_PARAMS:
250 if( !strcasecmp(ep->name, "codel")) {
251 us_to_time(ep->par[0], strt1);
252 us_to_time(ep->par[1], strt2);
253 l = sprintf(out, " AQM CoDel target %s interval %s",
254 strt1, strt2);
255 if (ep->par[2] & CODEL_ECN_ENABLED)
256 l = sprintf(out + l, " ECN");
257 else
258 l += sprintf(out + l, " NoECN");
259 } else if( !strcasecmp(ep->name, "pie")) {
260 us_to_time(ep->par[0], strt1);
261 us_to_time(ep->par[1], strt2);
262 us_to_time(ep->par[2], strt3);
263 l = sprintf(out, " AQM type PIE target %s tupdate %s alpha "
264 "%g beta %g max_burst %s max_ecnth %.3g",
265 strt1,
266 strt2,
267 ep->par[4] / (float) PIE_SCALE,
268 ep->par[5] / (float) PIE_SCALE,
269 strt3,
270 ep->par[3] / (float) PIE_SCALE
271 );
272
273 if (ep->par[6] & PIE_ECN_ENABLED)
274 l += sprintf(out + l, " ECN");
275 else
276 l += sprintf(out + l, " NoECN");
277 if (ep->par[6] & PIE_CAPDROP_ENABLED)
278 l += sprintf(out + l, " CapDrop");
279 else
280 l += sprintf(out + l, " NoCapDrop");
281 if (ep->par[6] & PIE_ON_OFF_MODE_ENABLED)
282 l += sprintf(out + l, " OnOff");
283 if (ep->par[6] & PIE_DEPRATEEST_ENABLED)
284 l += sprintf(out + l, " DRE");
285 else
286 l += sprintf(out + l, " TS");
287 if (ep->par[6] & PIE_DERAND_ENABLED)
288 l += sprintf(out + l, " Derand");
289 else
290 l += sprintf(out + l, " NoDerand");
291 }
292 break;
293
294 case DN_SCH_PARAMS:
295 if (!strcasecmp(ep->name,"FQ_CODEL")) {
296 us_to_time(ep->par[0], strt1);
297 us_to_time(ep->par[1], strt2);
298 l = sprintf(out," FQ_CODEL target %s interval %s"
299 " quantum %jd limit %jd flows %jd",
300 strt1, strt2,
301 (intmax_t) ep->par[3],
302 (intmax_t) ep->par[4],
303 (intmax_t) ep->par[5]
304 );
305 if (ep->par[2] & CODEL_ECN_ENABLED)
306 l += sprintf(out + l, " ECN");
307 else
308 l += sprintf(out + l, " NoECN");
309 l += sprintf(out + l, "\n");
310 } else if (!strcasecmp(ep->name,"FQ_PIE")) {
311 us_to_time(ep->par[0], strt1);
312 us_to_time(ep->par[1], strt2);
313 us_to_time(ep->par[2], strt3);
314 l = sprintf(out, " FQ_PIE target %s tupdate %s alpha "
315 "%g beta %g max_burst %s max_ecnth %.3g"
316 " quantum %jd limit %jd flows %jd",
317 strt1,
318 strt2,
319 ep->par[4] / (float) PIE_SCALE,
320 ep->par[5] / (float) PIE_SCALE,
321 strt3,
322 ep->par[3] / (float) PIE_SCALE,
323 (intmax_t) ep->par[7],
324 (intmax_t) ep->par[8],
325 (intmax_t) ep->par[9]
326 );
327
328 if (ep->par[6] & PIE_ECN_ENABLED)
329 l += sprintf(out + l, " ECN");
330 else
331 l += sprintf(out + l, " NoECN");
332 if (ep->par[6] & PIE_CAPDROP_ENABLED)
333 l += sprintf(out + l, " CapDrop");
334 else
335 l += sprintf(out + l, " NoCapDrop");
336 if (ep->par[6] & PIE_ON_OFF_MODE_ENABLED)
337 l += sprintf(out + l, " OnOff");
338 if (ep->par[6] & PIE_DEPRATEEST_ENABLED)
339 l += sprintf(out + l, " DRE");
340 else
341 l += sprintf(out + l, " TS");
342 if (ep->par[6] & PIE_DERAND_ENABLED)
343 l += sprintf(out + l, " Derand");
344 else
345 l += sprintf(out + l, " NoDerand");
346 l += sprintf(out + l, "\n");
347 }
348 break;
349 }
350
351 free(ep);
352 }
353 #endif
354
355
356 #if 0
357 static int
358 sort_q(void *arg, const void *pa, const void *pb)
359 {
360 int rev = (co.do_sort < 0);
361 int field = rev ? -co.do_sort : co.do_sort;
362 long long res = 0;
363 const struct dn_flow_queue *a = pa;
364 const struct dn_flow_queue *b = pb;
365
366 switch (field) {
367 case 1: /* pkts */
368 res = a->len - b->len;
369 break;
370 case 2: /* bytes */
371 res = a->len_bytes - b->len_bytes;
372 break;
373
374 case 3: /* tot pkts */
375 res = a->tot_pkts - b->tot_pkts;
376 break;
377
378 case 4: /* tot bytes */
379 res = a->tot_bytes - b->tot_bytes;
380 break;
381 }
382 if (res < 0)
383 res = -1;
384 if (res > 0)
385 res = 1;
386 return (int)(rev ? res : -res);
387 }
388 #endif
389
390 /* print a mask and header for the subsequent list of flows */
391 static void
print_mask(struct ipfw_flow_id * id)392 print_mask(struct ipfw_flow_id *id)
393 {
394 if (!IS_IP6_FLOW_ID(id)) {
395 printf(" "
396 "mask: %s 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n",
397 id->extra ? "queue," : "",
398 id->proto,
399 id->src_ip, id->src_port,
400 id->dst_ip, id->dst_port);
401 } else {
402 char buf[255];
403 printf("\n mask: %sproto: 0x%02x, flow_id: 0x%08x, ",
404 id->extra ? "queue," : "",
405 id->proto, id->flow_id6);
406 inet_ntop(AF_INET6, &(id->src_ip6), buf, sizeof(buf));
407 printf("%s/0x%04x -> ", buf, id->src_port);
408 inet_ntop(AF_INET6, &(id->dst_ip6), buf, sizeof(buf));
409 printf("%s/0x%04x\n", buf, id->dst_port);
410 }
411 }
412
413 static void
print_header(struct ipfw_flow_id * id)414 print_header(struct ipfw_flow_id *id)
415 {
416 if (!IS_IP6_FLOW_ID(id))
417 printf("BKT Prot ___Source IP/port____ "
418 "____Dest. IP/port____ "
419 "Tot_pkt/bytes Pkt/Byte Drp\n");
420 else
421 printf("BKT ___Prot___ _flow-id_ "
422 "______________Source IPv6/port_______________ "
423 "_______________Dest. IPv6/port_______________ "
424 "Tot_pkt/bytes Pkt/Byte Drp\n");
425 }
426
427 static void
list_flow(struct buf_pr * bp,struct dn_flow * ni)428 list_flow(struct buf_pr *bp, struct dn_flow *ni)
429 {
430 char buff[255];
431 struct protoent *pe = NULL;
432 struct in_addr ina;
433 struct ipfw_flow_id *id = &ni->fid;
434
435 pe = getprotobynumber(id->proto);
436 /* XXX: Should check for IPv4 flows */
437 bprintf(bp, "%3u%c", (ni->oid.id) & 0xff,
438 id->extra ? '*' : ' ');
439 if (!IS_IP6_FLOW_ID(id)) {
440 if (pe)
441 bprintf(bp, "%-4s ", pe->p_name);
442 else
443 bprintf(bp, "%4u ", id->proto);
444 ina.s_addr = htonl(id->src_ip);
445 bprintf(bp, "%15s/%-5d ",
446 inet_ntoa(ina), id->src_port);
447 ina.s_addr = htonl(id->dst_ip);
448 bprintf(bp, "%15s/%-5d ",
449 inet_ntoa(ina), id->dst_port);
450 } else {
451 /* Print IPv6 flows */
452 if (pe != NULL)
453 bprintf(bp, "%9s ", pe->p_name);
454 else
455 bprintf(bp, "%9u ", id->proto);
456 bprintf(bp, "%7d %39s/%-5d ", id->flow_id6,
457 inet_ntop(AF_INET6, &(id->src_ip6), buff, sizeof(buff)),
458 id->src_port);
459 bprintf(bp, " %39s/%-5d ",
460 inet_ntop(AF_INET6, &(id->dst_ip6), buff, sizeof(buff)),
461 id->dst_port);
462 }
463 pr_u64(bp, &ni->tot_pkts, 4);
464 pr_u64(bp, &ni->tot_bytes, 8);
465 bprintf(bp, "%2u %4u %3u",
466 ni->length, ni->len_bytes, ni->drops);
467 }
468
469 static void
print_flowset_parms(struct dn_fs * fs,char * prefix)470 print_flowset_parms(struct dn_fs *fs, char *prefix)
471 {
472 int l;
473 char qs[30];
474 char plr[40];
475 char red[200]; /* Display RED parameters */
476
477 l = fs->qsize;
478 if (fs->flags & DN_QSIZE_BYTES) {
479 if (l >= 8192)
480 sprintf(qs, "%d KB", l / 1024);
481 else
482 sprintf(qs, "%d B", l);
483 } else
484 sprintf(qs, "%3d sl.", l);
485 if (fs->plr[0] || fs->plr[1]) {
486 if (fs->plr[1] == 0)
487 sprintf(plr, "plr %f",
488 1.0 * fs->plr[0] / (double)(0x7fffffff));
489 else
490 sprintf(plr, "plr %f,%f,%f,%f",
491 1.0 * fs->plr[0] / (double)(0x7fffffff),
492 1.0 * fs->plr[1] / (double)(0x7fffffff),
493 1.0 * fs->plr[2] / (double)(0x7fffffff),
494 1.0 * fs->plr[3] / (double)(0x7fffffff));
495 } else
496 plr[0] = '\0';
497
498 if (fs->flags & DN_IS_RED) { /* RED parameters */
499 sprintf(red,
500 "\n\t %cRED w_q %f min_th %d max_th %d max_p %f",
501 (fs->flags & DN_IS_GENTLE_RED) ? 'G' : ' ',
502 1.0 * fs->w_q / (double)(1 << SCALE_RED),
503 fs->min_th,
504 fs->max_th,
505 1.0 * fs->max_p / (double)(1 << SCALE_RED));
506 if (fs->flags & DN_IS_ECN)
507 strlcat(red, " (ecn)", sizeof(red));
508 #ifdef NEW_AQM
509 /* get AQM parameters */
510 } else if (fs->flags & DN_IS_AQM) {
511 get_extra_parms(fs->fs_nr, red, DN_AQM_PARAMS);
512 #endif
513 } else
514 sprintf(red, "droptail");
515
516 if (prefix[0]) {
517 printf("%s %s%s %d queues (%d buckets) %s\n",
518 prefix, qs, plr, fs->oid.id, fs->buckets, red);
519 prefix[0] = '\0';
520 } else {
521 printf("q%05d %s%s %d flows (%d buckets) sched %d "
522 "weight %d lmax %d pri %d %s\n",
523 fs->fs_nr, qs, plr, fs->oid.id, fs->buckets,
524 fs->sched_nr, fs->par[0], fs->par[1], fs->par[2], red);
525 if (fs->flags & DN_HAVE_MASK)
526 print_mask(&fs->flow_mask);
527 }
528 }
529
530 static void
print_extra_delay_parms(struct dn_profile * p)531 print_extra_delay_parms(struct dn_profile *p)
532 {
533 double loss;
534 if (p->samples_no <= 0)
535 return;
536
537 loss = p->loss_level;
538 loss /= p->samples_no;
539 printf("\t profile: name \"%s\" loss %f samples %d\n",
540 p->name, loss, p->samples_no);
541 }
542
543 static void
flush_buf(char * buf)544 flush_buf(char *buf)
545 {
546 if (buf[0])
547 printf("%s\n", buf);
548 buf[0] = '\0';
549 }
550
551 /*
552 * generic list routine. We expect objects in a specific order, i.e.
553 * PIPES AND SCHEDULERS:
554 * link; scheduler; internal flowset if any; instances
555 * we can tell a pipe from the number.
556 *
557 * FLOWSETS:
558 * flowset; queues;
559 * link i (int queue); scheduler i; si(i) { flowsets() : queues }
560 */
561 static void
list_pipes(struct dn_id * oid,struct dn_id * end)562 list_pipes(struct dn_id *oid, struct dn_id *end)
563 {
564 char buf[160]; /* pending buffer */
565 int toPrint = 1; /* print header */
566 struct buf_pr bp;
567
568 buf[0] = '\0';
569 bp_alloc(&bp, 4096);
570 for (; oid != end; oid = O_NEXT(oid, oid->len)) {
571 if (oid->len < sizeof(*oid))
572 errx(1, "invalid oid len %d\n", oid->len);
573
574 switch (oid->type) {
575 default:
576 flush_buf(buf);
577 printf("unrecognized object %d size %d\n", oid->type, oid->len);
578 break;
579 case DN_TEXT: /* list of attached flowsets */
580 {
581 int i, l;
582 struct {
583 struct dn_id id;
584 uint32_t p[0];
585 } *d = (void *)oid;
586 l = (oid->len - sizeof(*oid))/sizeof(d->p[0]);
587 if (l == 0)
588 break;
589 printf(" Children flowsets: ");
590 for (i = 0; i < l; i++)
591 printf("%u ", d->p[i]);
592 printf("\n");
593 break;
594 }
595 case DN_CMD_GET:
596 if (g_co.verbose)
597 printf("answer for cmd %d, len %d\n", oid->type, oid->id);
598 break;
599 case DN_SCH: {
600 struct dn_sch *s = (struct dn_sch *)oid;
601 flush_buf(buf);
602 printf(" sched %d type %s flags 0x%x %d buckets %d active\n",
603 s->sched_nr,
604 s->name, s->flags, s->buckets, s->oid.id);
605 #ifdef NEW_AQM
606 char parms[200];
607 get_extra_parms(s->sched_nr, parms, DN_SCH_PARAMS);
608 printf("%s",parms);
609 #endif
610 if (s->flags & DN_HAVE_MASK)
611 print_mask(&s->sched_mask);
612 }
613 break;
614
615 case DN_FLOW:
616 if (toPrint != 0) {
617 print_header(&((struct dn_flow *)oid)->fid);
618 toPrint = 0;
619 }
620 list_flow(&bp, (struct dn_flow *)oid);
621 printf("%s\n", bp.buf);
622 bp_flush(&bp);
623 break;
624
625 case DN_LINK: {
626 struct dn_link *p = (struct dn_link *)oid;
627 double b = p->bandwidth;
628 char bwbuf[30];
629 char burst[5 + 7];
630
631 /* This starts a new object so flush buffer */
632 flush_buf(buf);
633 /* data rate */
634 if (b == 0)
635 sprintf(bwbuf, "unlimited ");
636 else if (b >= 1000000000)
637 sprintf(bwbuf, "%7.3f Gbit/s", b/1000000000);
638 else if (b >= 1000000)
639 sprintf(bwbuf, "%7.3f Mbit/s", b/1000000);
640 else if (b >= 1000)
641 sprintf(bwbuf, "%7.3f Kbit/s", b/1000);
642 else
643 sprintf(bwbuf, "%7.3f bit/s ", b);
644
645 if (humanize_number(burst, sizeof(burst), p->burst,
646 "", HN_AUTOSCALE, 0) < 0 || g_co.verbose)
647 sprintf(burst, "%d", (int)p->burst);
648 sprintf(buf, "%05d: %s %4d ms burst %s",
649 p->link_nr % DN_MAX_ID, bwbuf, p->delay, burst);
650 }
651 break;
652
653 case DN_FS:
654 print_flowset_parms((struct dn_fs *)oid, buf);
655 break;
656 case DN_PROFILE:
657 flush_buf(buf);
658 print_extra_delay_parms((struct dn_profile *)oid);
659 }
660 flush_buf(buf); // XXX does it really go here ?
661 }
662
663 bp_free(&bp);
664 }
665
666 /*
667 * Delete pipe, queue or scheduler i
668 */
669 int
ipfw_delete_pipe(int do_pipe,int i)670 ipfw_delete_pipe(int do_pipe, int i)
671 {
672 struct {
673 struct dn_id oid;
674 uintptr_t a[1]; /* add more if we want a list */
675 } cmd;
676 oid_fill((void *)&cmd, sizeof(cmd), DN_CMD_DELETE, DN_API_VERSION);
677 cmd.oid.subtype = (do_pipe == 1) ? DN_LINK :
678 ( (do_pipe == 2) ? DN_FS : DN_SCH);
679 cmd.a[0] = i;
680 i = do_cmd(IP_DUMMYNET3, &cmd, cmd.oid.len);
681 if (i) {
682 i = 1;
683 warn("rule %u: setsockopt(IP_DUMMYNET3)", i);
684 }
685 return i;
686 }
687
688 /*
689 * Code to parse delay profiles.
690 *
691 * Some link types introduce extra delays in the transmission
692 * of a packet, e.g. because of MAC level framing, contention on
693 * the use of the channel, MAC level retransmissions and so on.
694 * From our point of view, the channel is effectively unavailable
695 * for this extra time, which is constant or variable depending
696 * on the link type. Additionally, packets may be dropped after this
697 * time (e.g. on a wireless link after too many retransmissions).
698 * We can model the additional delay with an empirical curve
699 * that represents its distribution.
700 *
701 * cumulative probability
702 * 1.0 ^
703 * |
704 * L +-- loss-level x
705 * | ******
706 * | *
707 * | *****
708 * | *
709 * | **
710 * | *
711 * +-------*------------------->
712 * delay
713 *
714 * The empirical curve may have both vertical and horizontal lines.
715 * Vertical lines represent constant delay for a range of
716 * probabilities; horizontal lines correspond to a discontinuty
717 * in the delay distribution: the link will use the largest delay
718 * for a given probability.
719 *
720 * To pass the curve to dummynet, we must store the parameters
721 * in a file as described below, and issue the command
722 *
723 * ipfw pipe <n> config ... bw XXX profile <filename> ...
724 *
725 * The file format is the following, with whitespace acting as
726 * a separator and '#' indicating the beginning a comment:
727 *
728 * samples N
729 * the number of samples used in the internal
730 * representation (2..1024; default 100);
731 *
732 * loss-level L
733 * The probability above which packets are lost.
734 * (0.0 <= L <= 1.0, default 1.0 i.e. no loss);
735 *
736 * name identifier
737 * Optional a name (listed by "ipfw pipe show")
738 * to identify the distribution;
739 *
740 * "delay prob" | "prob delay"
741 * One of these two lines is mandatory and defines
742 * the format of the following lines with data points.
743 *
744 * XXX YYY
745 * 2 or more lines representing points in the curve,
746 * with either delay or probability first, according
747 * to the chosen format.
748 * The unit for delay is milliseconds.
749 *
750 * Data points does not need to be ordered or equal to the number
751 * specified in the "samples" line. ipfw will sort and interpolate
752 * the curve as needed.
753 *
754 * Example of a profile file:
755
756 name bla_bla_bla
757 samples 100
758 loss-level 0.86
759 prob delay
760 0 200 # minimum overhead is 200ms
761 0.5 200
762 0.5 300
763 0.8 1000
764 0.9 1300
765 1 1300
766
767 * Internally, we will convert the curve to a fixed number of
768 * samples, and when it is time to transmit a packet we will
769 * model the extra delay as extra bits in the packet.
770 *
771 */
772
773 #define ED_MAX_LINE_LEN 256+ED_MAX_NAME_LEN
774 #define ED_TOK_SAMPLES "samples"
775 #define ED_TOK_LOSS "loss-level"
776 #define ED_TOK_NAME "name"
777 #define ED_TOK_DELAY "delay"
778 #define ED_TOK_PROB "prob"
779 #define ED_TOK_BW "bw"
780 #define ED_SEPARATORS " \t\n"
781 #define ED_MIN_SAMPLES_NO 2
782
783 /*
784 * returns 1 if s is a non-negative number, with at least one '.'
785 */
786 static int
is_valid_number(const char * s)787 is_valid_number(const char *s)
788 {
789 int i, dots_found = 0;
790 int len = strlen(s);
791
792 for (i = 0; i<len; ++i)
793 if (!isdigit(s[i]) && (s[i] !='.' || ++dots_found > 1))
794 return 0;
795 return 1;
796 }
797
798 /*
799 * Take as input a string describing a bandwidth value
800 * and return the numeric bandwidth value.
801 * set clocking interface or bandwidth value
802 */
803 static void
read_bandwidth(char * arg,uint32_t * bandwidth,char * if_name,int namelen)804 read_bandwidth(char *arg, uint32_t *bandwidth, char *if_name, int namelen)
805 {
806 if (*bandwidth != (uint32_t)-1)
807 warnx("duplicate token, override bandwidth value!");
808
809 if (arg[0] >= 'a' && arg[0] <= 'z') {
810 if (!if_name) {
811 errx(1, "no if support");
812 }
813 if (namelen >= IFNAMSIZ)
814 warn("interface name truncated");
815 namelen--;
816 /* interface name */
817 strlcpy(if_name, arg, namelen);
818 *bandwidth = 0;
819 } else { /* read bandwidth value */
820 uint64_t bw;
821 char *end = NULL;
822
823 bw = strtoul(arg, &end, 0);
824 if (*end == 'K' || *end == 'k') {
825 end++;
826 bw *= 1000;
827 } else if (*end == 'M' || *end == 'm') {
828 end++;
829 bw *= 1000000;
830 } else if (*end == 'G' || *end == 'g') {
831 end++;
832 bw *= 1000000000;
833 }
834 if ((*end == 'B' &&
835 _substrcmp2(end, "Bi", "Bit/s") != 0) ||
836 _substrcmp2(end, "by", "bytes") == 0)
837 bw *= 8;
838
839 if (bw > UINT_MAX)
840 errx(EX_DATAERR, "bandwidth too large");
841
842 *bandwidth = (uint32_t)bw;
843 if (if_name)
844 if_name[0] = '\0';
845 }
846 }
847
848 struct point {
849 double prob;
850 double delay;
851 };
852
853 static int
compare_points(const void * vp1,const void * vp2)854 compare_points(const void *vp1, const void *vp2)
855 {
856 const struct point *p1 = vp1;
857 const struct point *p2 = vp2;
858 double res = 0;
859
860 res = p1->prob - p2->prob;
861 if (res == 0)
862 res = p1->delay - p2->delay;
863 if (res < 0)
864 return -1;
865 else if (res > 0)
866 return 1;
867 else
868 return 0;
869 }
870
871 #define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno
872
873 static void
load_extra_delays(const char * filename,struct dn_profile * p,struct dn_link * link)874 load_extra_delays(const char *filename, struct dn_profile *p,
875 struct dn_link *link)
876 {
877 char line[ED_MAX_LINE_LEN];
878 FILE *f;
879 int lineno = 0;
880 int i;
881
882 int samples = -1;
883 double loss = -1.0;
884 char profile_name[ED_MAX_NAME_LEN];
885 int delay_first = -1;
886 int do_points = 0;
887 struct point points[ED_MAX_SAMPLES_NO];
888 int points_no = 0;
889
890 /* XXX link never NULL? */
891 p->link_nr = link->link_nr;
892
893 profile_name[0] = '\0';
894 f = fopen(filename, "r");
895 if (f == NULL)
896 err(EX_UNAVAILABLE, "fopen: %s", filename);
897
898 while (fgets(line, ED_MAX_LINE_LEN, f)) { /* read commands */
899 char *s, *cur = line, *name = NULL, *arg = NULL;
900
901 ++lineno;
902
903 /* parse the line */
904 while (cur) {
905 s = strsep(&cur, ED_SEPARATORS);
906 if (s == NULL || *s == '#')
907 break;
908 if (*s == '\0')
909 continue;
910 if (arg)
911 errx(ED_EFMT("too many arguments"));
912 if (name == NULL)
913 name = s;
914 else
915 arg = s;
916 }
917 if (name == NULL) /* empty line */
918 continue;
919 if (arg == NULL)
920 errx(ED_EFMT("missing arg for %s"), name);
921
922 if (!strcasecmp(name, ED_TOK_SAMPLES)) {
923 if (samples > 0)
924 errx(ED_EFMT("duplicate ``samples'' line"));
925 if (atoi(arg) <=0)
926 errx(ED_EFMT("invalid number of samples"));
927 samples = atoi(arg);
928 if (samples>ED_MAX_SAMPLES_NO)
929 errx(ED_EFMT("too many samples, maximum is %d"),
930 ED_MAX_SAMPLES_NO);
931 do_points = 0;
932 } else if (!strcasecmp(name, ED_TOK_BW)) {
933 char buf[IFNAMSIZ];
934 read_bandwidth(arg, &link->bandwidth, buf, sizeof(buf));
935 } else if (!strcasecmp(name, ED_TOK_LOSS)) {
936 if (loss != -1.0)
937 errx(ED_EFMT("duplicated token: %s"), name);
938 if (!is_valid_number(arg))
939 errx(ED_EFMT("invalid %s"), arg);
940 loss = atof(arg);
941 if (loss > 1)
942 errx(ED_EFMT("%s greater than 1.0"), name);
943 do_points = 0;
944 } else if (!strcasecmp(name, ED_TOK_NAME)) {
945 if (profile_name[0] != '\0')
946 errx(ED_EFMT("duplicated token: %s"), name);
947 strlcpy(profile_name, arg, sizeof(profile_name));
948 do_points = 0;
949 } else if (!strcasecmp(name, ED_TOK_DELAY)) {
950 if (do_points)
951 errx(ED_EFMT("duplicated token: %s"), name);
952 delay_first = 1;
953 do_points = 1;
954 } else if (!strcasecmp(name, ED_TOK_PROB)) {
955 if (do_points)
956 errx(ED_EFMT("duplicated token: %s"), name);
957 delay_first = 0;
958 do_points = 1;
959 } else if (do_points) {
960 if (!is_valid_number(name) || !is_valid_number(arg))
961 errx(ED_EFMT("invalid point found"));
962 if (points_no >= ED_MAX_SAMPLES_NO)
963 errx(ED_EFMT("too many samples, maximum is %d"),
964 ED_MAX_SAMPLES_NO);
965 if (delay_first) {
966 points[points_no].delay = atof(name);
967 points[points_no].prob = atof(arg);
968 } else {
969 points[points_no].delay = atof(arg);
970 points[points_no].prob = atof(name);
971 }
972 if (points[points_no].prob > 1.0)
973 errx(ED_EFMT("probability greater than 1.0"));
974 ++points_no;
975 } else {
976 errx(ED_EFMT("unrecognised command '%s'"), name);
977 }
978 }
979
980 fclose (f);
981
982 if (samples == -1) {
983 warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES);
984 samples = 100;
985 }
986
987 if (loss == -1.0) {
988 warnx("'%s' not found, assuming no loss", ED_TOK_LOSS);
989 loss = 1;
990 }
991
992 /* make sure that there are enough points. */
993 if (points_no < ED_MIN_SAMPLES_NO)
994 errx(ED_EFMT("too few samples, need at least %d"),
995 ED_MIN_SAMPLES_NO);
996
997 qsort(points, points_no, sizeof(struct point), compare_points);
998
999 /* interpolation */
1000 for (i = 0; i<points_no-1; ++i) {
1001 double y1 = points[i].prob * samples;
1002 double x1 = points[i].delay;
1003 double y2 = points[i+1].prob * samples;
1004 double x2 = points[i+1].delay;
1005
1006 int ix = y1;
1007 int stop = y2;
1008
1009 if (x1 == x2) {
1010 for (; ix<stop; ++ix)
1011 p->samples[ix] = x1;
1012 } else {
1013 double m = (y2-y1)/(x2-x1);
1014 double c = y1 - m*x1;
1015 for (; ix<stop ; ++ix)
1016 p->samples[ix] = (ix - c)/m;
1017 }
1018 }
1019 p->samples_no = samples;
1020 p->loss_level = loss * samples;
1021 strlcpy(p->name, profile_name, sizeof(p->name));
1022 }
1023
1024 #ifdef NEW_AQM
1025
1026 /* Parse AQM/extra scheduler parameters */
1027 static int
process_extra_parms(int * ac,char ** av,struct dn_extra_parms * ep,uint16_t type)1028 process_extra_parms(int *ac, char **av, struct dn_extra_parms *ep,
1029 uint16_t type)
1030 {
1031 int i;
1032
1033 /* use kernel defaults */
1034 for (i=0; i<DN_MAX_EXTRA_PARM; i++)
1035 ep->par[i] = -1;
1036
1037 switch(type) {
1038 case TOK_CODEL:
1039 case TOK_FQ_CODEL:
1040 /* Codel
1041 * 0- target, 1- interval, 2- flags,
1042 * FQ_CODEL
1043 * 3- quantum, 4- limit, 5- flows
1044 */
1045 if (type==TOK_CODEL)
1046 ep->par[2] = 0;
1047 else
1048 ep->par[2] = CODEL_ECN_ENABLED;
1049
1050 while (*ac > 0) {
1051 int tok = match_token(aqm_params, *av);
1052 (*ac)--; av++;
1053 switch(tok) {
1054 case TOK_TARGET:
1055 if (*ac <= 0 || time_to_us(av[0]) < 0)
1056 errx(EX_DATAERR, "target needs time\n");
1057
1058 ep->par[0] = time_to_us(av[0]);
1059 (*ac)--; av++;
1060 break;
1061
1062 case TOK_INTERVAL:
1063 if (*ac <= 0 || time_to_us(av[0]) < 0)
1064 errx(EX_DATAERR, "interval needs time\n");
1065
1066 ep->par[1] = time_to_us(av[0]);
1067 (*ac)--; av++;
1068 break;
1069
1070 case TOK_ECN:
1071 ep->par[2] = CODEL_ECN_ENABLED;
1072 break;
1073 case TOK_NO_ECN:
1074 ep->par[2] &= ~CODEL_ECN_ENABLED;
1075 break;
1076 /* Config fq_codel parameters */
1077 case TOK_QUANTUM:
1078 if (type != TOK_FQ_CODEL)
1079 errx(EX_DATAERR, "quantum is not for codel\n");
1080 if (*ac <= 0 || !is_valid_number(av[0]))
1081 errx(EX_DATAERR, "quantum needs number\n");
1082
1083 ep->par[3]= atoi(av[0]);
1084 (*ac)--; av++;
1085 break;
1086
1087 case TOK_LIMIT:
1088 if (type != TOK_FQ_CODEL)
1089 errx(EX_DATAERR, "limit is not for codel, use queue instead\n");
1090 if (*ac <= 0 || !is_valid_number(av[0]))
1091 errx(EX_DATAERR, "limit needs number\n");
1092
1093 ep->par[4] = atoi(av[0]);
1094 (*ac)--; av++;
1095 break;
1096
1097 case TOK_FLOWS:
1098 if (type != TOK_FQ_CODEL)
1099 errx(EX_DATAERR, "flows is not for codel\n");
1100 if (*ac <= 0 || !is_valid_number(av[0]))
1101 errx(EX_DATAERR, "flows needs number\n");
1102
1103 ep->par[5] = atoi(av[0]);
1104 (*ac)--; av++;
1105 break;
1106
1107 default:
1108 printf("%s is Invalid parameter\n", av[-1]);
1109 }
1110 }
1111 break;
1112 case TOK_PIE:
1113 case TOK_FQ_PIE:
1114 /* PIE
1115 * 0- target , 1- tupdate, 2- max_burst,
1116 * 3- max_ecnth, 4- alpha,
1117 * 5- beta, 6- flags
1118 * FQ_CODEL
1119 * 7- quantum, 8- limit, 9- flows
1120 */
1121
1122 if ( type == TOK_PIE)
1123 ep->par[6] = PIE_CAPDROP_ENABLED | PIE_DEPRATEEST_ENABLED
1124 | PIE_DERAND_ENABLED;
1125 else
1126 /* for FQ-PIE, use TS mode */
1127 ep->par[6] = PIE_CAPDROP_ENABLED | PIE_DERAND_ENABLED
1128 | PIE_ECN_ENABLED;
1129
1130 while (*ac > 0) {
1131 int tok = match_token(aqm_params, *av);
1132 (*ac)--; av++;
1133 switch(tok) {
1134 case TOK_TARGET:
1135 if (*ac <= 0 || time_to_us(av[0]) < 0)
1136 errx(EX_DATAERR, "target needs time\n");
1137
1138 ep->par[0] = time_to_us(av[0]);
1139 (*ac)--; av++;
1140 break;
1141
1142 case TOK_TUPDATE:
1143 if (*ac <= 0 || time_to_us(av[0]) < 0)
1144 errx(EX_DATAERR, "tupdate needs time\n");
1145
1146 ep->par[1] = time_to_us(av[0]);
1147 (*ac)--; av++;
1148 break;
1149
1150 case TOK_MAX_BURST:
1151 if (*ac <= 0 || time_to_us(av[0]) < 0)
1152 errx(EX_DATAERR, "max_burst needs time\n");
1153
1154 ep->par[2] = time_to_us(av[0]);
1155 (*ac)--; av++;
1156 break;
1157
1158 case TOK_MAX_ECNTH:
1159 if (*ac <= 0 || !is_valid_number(av[0]))
1160 errx(EX_DATAERR, "max_ecnth needs number\n");
1161
1162 ep->par[3] = atof(av[0]) * PIE_SCALE;
1163 (*ac)--; av++;
1164 break;
1165
1166 case TOK_ALPHA:
1167 if (*ac <= 0 || !is_valid_number(av[0]))
1168 errx(EX_DATAERR, "alpha needs number\n");
1169
1170 ep->par[4] = atof(av[0]) * PIE_SCALE;
1171 (*ac)--; av++;
1172 break;
1173
1174 case TOK_BETA:
1175 if (*ac <= 0 || !is_valid_number(av[0]))
1176 errx(EX_DATAERR, "beta needs number\n");
1177
1178 ep->par[5] = atof(av[0]) * PIE_SCALE;
1179 (*ac)--; av++;
1180 break;
1181
1182 case TOK_ECN:
1183 ep->par[6] |= PIE_ECN_ENABLED;
1184 break;
1185 case TOK_NO_ECN:
1186 ep->par[6] &= ~PIE_ECN_ENABLED;
1187 break;
1188
1189 case TOK_CAPDROP:
1190 ep->par[6] |= PIE_CAPDROP_ENABLED;
1191 break;
1192 case TOK_NO_CAPDROP:
1193 ep->par[6] &= ~PIE_CAPDROP_ENABLED;
1194 break;
1195
1196 case TOK_ONOFF:
1197 ep->par[6] |= PIE_ON_OFF_MODE_ENABLED;
1198 break;
1199
1200 case TOK_DRE:
1201 ep->par[6] |= PIE_DEPRATEEST_ENABLED;
1202 break;
1203
1204 case TOK_TS:
1205 ep->par[6] &= ~PIE_DEPRATEEST_ENABLED;
1206 break;
1207
1208 case TOK_DERAND:
1209 ep->par[6] |= PIE_DERAND_ENABLED;
1210 break;
1211 case TOK_NO_DERAND:
1212 ep->par[6] &= ~PIE_DERAND_ENABLED;
1213 break;
1214
1215 /* Config fq_pie parameters */
1216 case TOK_QUANTUM:
1217 if (type != TOK_FQ_PIE)
1218 errx(EX_DATAERR, "quantum is not for pie\n");
1219 if (*ac <= 0 || !is_valid_number(av[0]))
1220 errx(EX_DATAERR, "quantum needs number\n");
1221
1222 ep->par[7]= atoi(av[0]);
1223 (*ac)--; av++;
1224 break;
1225
1226 case TOK_LIMIT:
1227 if (type != TOK_FQ_PIE)
1228 errx(EX_DATAERR, "limit is not for pie, use queue instead\n");
1229 if (*ac <= 0 || !is_valid_number(av[0]))
1230 errx(EX_DATAERR, "limit needs number\n");
1231
1232 ep->par[8] = atoi(av[0]);
1233 (*ac)--; av++;
1234 break;
1235
1236 case TOK_FLOWS:
1237 if (type != TOK_FQ_PIE)
1238 errx(EX_DATAERR, "flows is not for pie\n");
1239 if (*ac <= 0 || !is_valid_number(av[0]))
1240 errx(EX_DATAERR, "flows needs number\n");
1241
1242 ep->par[9] = atoi(av[0]);
1243 (*ac)--; av++;
1244 break;
1245
1246
1247 default:
1248 printf("%s is invalid parameter\n", av[-1]);
1249 }
1250 }
1251 break;
1252 }
1253
1254 return 0;
1255 }
1256
1257 #endif
1258
1259
1260 /*
1261 * configuration of pipes, schedulers, flowsets.
1262 * When we configure a new scheduler, an empty pipe is created, so:
1263 *
1264 * do_pipe = 1 -> "pipe N config ..." only for backward compatibility
1265 * sched N+Delta type fifo sched_mask ...
1266 * pipe N+Delta <parameters>
1267 * flowset N+Delta pipe N+Delta (no parameters)
1268 * sched N type wf2q+ sched_mask ...
1269 * pipe N <parameters>
1270 *
1271 * do_pipe = 2 -> flowset N config
1272 * flowset N parameters
1273 *
1274 * do_pipe = 3 -> sched N config
1275 * sched N parameters (default no pipe)
1276 * optional Pipe N config ...
1277 * pipe ==>
1278 */
1279 void
ipfw_config_pipe(int ac,char ** av)1280 ipfw_config_pipe(int ac, char **av)
1281 {
1282 int i;
1283 u_int j;
1284 char *end;
1285 struct dn_id *buf, *base;
1286 struct dn_sch *sch = NULL;
1287 struct dn_link *p = NULL;
1288 struct dn_fs *fs = NULL;
1289 struct dn_profile *pf = NULL;
1290 struct ipfw_flow_id *mask = NULL;
1291 #ifdef NEW_AQM
1292 struct dn_extra_parms *aqm_extra = NULL;
1293 struct dn_extra_parms *sch_extra = NULL;
1294 int lmax_extra;
1295 #endif
1296
1297 int lmax;
1298 uint32_t _foo = 0, *flags = &_foo , *buckets = &_foo;
1299
1300 /*
1301 * allocate space for 1 header,
1302 * 1 scheduler, 1 link, 1 flowset, 1 profile
1303 */
1304 lmax = sizeof(struct dn_id); /* command header */
1305 lmax += sizeof(struct dn_sch) + sizeof(struct dn_link) +
1306 sizeof(struct dn_fs) + sizeof(struct dn_profile);
1307
1308 #ifdef NEW_AQM
1309 /* Extra Params */
1310 lmax_extra = sizeof(struct dn_extra_parms);
1311 /* two lmax_extra because one for AQM params and another
1312 * sch params
1313 */
1314 lmax += lmax_extra*2;
1315 #endif
1316
1317 av++; ac--;
1318 /* Pipe number */
1319 if (ac && isdigit(**av)) {
1320 i = atoi(*av); av++; ac--;
1321 } else
1322 i = -1;
1323 if (i <= 0)
1324 errx(EX_USAGE, "need a pipe/flowset/sched number");
1325 base = buf = safe_calloc(1, lmax);
1326 /* all commands start with a 'CONFIGURE' and a version */
1327 o_next(&buf, sizeof(struct dn_id), DN_CMD_CONFIG);
1328 base->id = DN_API_VERSION;
1329
1330 switch (g_co.do_pipe) {
1331 case 1: /* "pipe N config ..." */
1332 /* Allocate space for the WF2Q+ scheduler, its link
1333 * and the FIFO flowset. Set the number, but leave
1334 * the scheduler subtype and other parameters to 0
1335 * so the kernel will use appropriate defaults.
1336 * XXX todo: add a flag to record if a parameter
1337 * is actually configured.
1338 * If we do a 'pipe config' mask -> sched_mask.
1339 * The FIFO scheduler and link are derived from the
1340 * WF2Q+ one in the kernel.
1341 */
1342 #ifdef NEW_AQM
1343 sch_extra = o_next(&buf, lmax_extra, DN_TEXT);
1344 sch_extra ->oid.subtype = 0; /* don't configure scheduler */
1345 #endif
1346 sch = o_next(&buf, sizeof(*sch), DN_SCH);
1347 p = o_next(&buf, sizeof(*p), DN_LINK);
1348 #ifdef NEW_AQM
1349 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT);
1350 aqm_extra ->oid.subtype = 0; /* don't configure AQM */
1351 #endif
1352 fs = o_next(&buf, sizeof(*fs), DN_FS);
1353
1354 sch->sched_nr = i;
1355 sch->oid.subtype = 0; /* defaults to WF2Q+ */
1356 mask = &sch->sched_mask;
1357 flags = &sch->flags;
1358 buckets = &sch->buckets;
1359 *flags |= DN_PIPE_CMD;
1360
1361 p->link_nr = i;
1362
1363 /* This flowset is only for the FIFO scheduler */
1364 fs->fs_nr = i + 2*DN_MAX_ID;
1365 fs->sched_nr = i + DN_MAX_ID;
1366 break;
1367
1368 case 2: /* "queue N config ... " */
1369 #ifdef NEW_AQM
1370 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT);
1371 aqm_extra ->oid.subtype = 0;
1372 #endif
1373 fs = o_next(&buf, sizeof(*fs), DN_FS);
1374 fs->fs_nr = i;
1375 mask = &fs->flow_mask;
1376 flags = &fs->flags;
1377 buckets = &fs->buckets;
1378 break;
1379
1380 case 3: /* "sched N config ..." */
1381 #ifdef NEW_AQM
1382 sch_extra = o_next(&buf, lmax_extra, DN_TEXT);
1383 sch_extra ->oid.subtype = 0;
1384 #endif
1385 sch = o_next(&buf, sizeof(*sch), DN_SCH);
1386 #ifdef NEW_AQM
1387 aqm_extra = o_next(&buf, lmax_extra, DN_TEXT);
1388 aqm_extra ->oid.subtype = 0;
1389 #endif
1390 fs = o_next(&buf, sizeof(*fs), DN_FS);
1391 sch->sched_nr = i;
1392 mask = &sch->sched_mask;
1393 flags = &sch->flags;
1394 buckets = &sch->buckets;
1395 /* fs is used only with !MULTIQUEUE schedulers */
1396 fs->fs_nr = i + DN_MAX_ID;
1397 fs->sched_nr = i;
1398 break;
1399 }
1400 /* set to -1 those fields for which we want to reuse existing
1401 * values from the kernel.
1402 * Also, *_nr and subtype = 0 mean reuse the value from the kernel.
1403 * XXX todo: support reuse of the mask.
1404 */
1405 if (p)
1406 p->bandwidth = -1;
1407 for (j = 0; j < nitems(fs->par); j++)
1408 fs->par[j] = -1;
1409 while (ac > 0) {
1410 double d;
1411 int tok = match_token(dummynet_params, *av);
1412 ac--; av++;
1413
1414 switch(tok) {
1415 case TOK_NOERROR:
1416 NEED(fs, "noerror is only for pipes");
1417 fs->flags |= DN_NOERROR;
1418 break;
1419
1420 case TOK_PLR:
1421 NEED(fs, "plr is only for pipes");
1422 NEED1("plr needs one or four arguments 0..1\n");
1423 if ((end = strsep(&av[0], ","))) {
1424 d = strtod(end, NULL);
1425 d = (d < 0) ? 0 : (d <= 1) ? d : 1;
1426 fs->plr[0] = (int)(d*0x7fffffff);
1427 }
1428 if ((end = strsep(&av[0], ","))) {
1429 d = strtod(end, NULL);
1430 d = (d < 0) ? 0 : (d <= 1) ? d : 1;
1431 fs->plr[1] = (int)(d*0x7fffffff);
1432 }
1433 if ((end = strsep(&av[0], ","))) {
1434 d = strtod(end, NULL);
1435 d = (d < 0) ? 0 : (d <= 1) ? d : 1;
1436 fs->plr[2] = (int)(d*0x7fffffff);
1437 }
1438 if ((end = strsep(&av[0], ","))) {
1439 d = strtod(end, NULL);
1440 d = (d < 0) ? 0 : (d <= 1) ? d : 1;
1441 fs->plr[3] = (int)(d*0x7fffffff);
1442 }
1443 ac--; av++;
1444 break;
1445
1446 case TOK_QUEUE:
1447 NEED(fs, "queue is only for pipes or flowsets");
1448 NEED1("queue needs queue size\n");
1449 end = NULL;
1450 fs->qsize = strtoul(av[0], &end, 0);
1451 if (*end == 'K' || *end == 'k') {
1452 fs->flags |= DN_QSIZE_BYTES;
1453 fs->qsize *= 1024;
1454 } else if (*end == 'B' ||
1455 _substrcmp2(end, "by", "bytes") == 0) {
1456 fs->flags |= DN_QSIZE_BYTES;
1457 }
1458 ac--; av++;
1459 break;
1460
1461 case TOK_BUCKETS:
1462 NEED(fs, "buckets is only for pipes or flowsets");
1463 NEED1("buckets needs argument\n");
1464 *buckets = strtoul(av[0], NULL, 0);
1465 ac--; av++;
1466 break;
1467
1468 case TOK_FLOW_MASK:
1469 case TOK_SCHED_MASK:
1470 case TOK_MASK:
1471 NEED(mask, "tok_mask");
1472 NEED1("mask needs mask specifier\n");
1473 /*
1474 * per-flow queue, mask is dst_ip, dst_port,
1475 * src_ip, src_port, proto measured in bits
1476 */
1477
1478 bzero(mask, sizeof(*mask));
1479 end = NULL;
1480
1481 while (ac >= 1) {
1482 uint32_t *p32 = NULL;
1483 uint16_t *p16 = NULL;
1484 uint32_t *p20 = NULL;
1485 struct in6_addr *pa6 = NULL;
1486 uint32_t a;
1487
1488 tok = match_token(dummynet_params, *av);
1489 ac--; av++;
1490 switch(tok) {
1491 case TOK_ALL:
1492 /*
1493 * special case, all bits significant
1494 * except 'extra' (the queue number)
1495 */
1496 mask->dst_ip = ~0;
1497 mask->src_ip = ~0;
1498 mask->dst_port = ~0;
1499 mask->src_port = ~0;
1500 mask->proto = ~0;
1501 n2mask(&mask->dst_ip6, 128);
1502 n2mask(&mask->src_ip6, 128);
1503 mask->flow_id6 = ~0;
1504 *flags |= DN_HAVE_MASK;
1505 goto end_mask;
1506
1507 case TOK_QUEUE:
1508 mask->extra = ~0;
1509 *flags |= DN_HAVE_MASK;
1510 goto end_mask;
1511
1512 case TOK_DSTIP:
1513 mask->addr_type = 4;
1514 p32 = &mask->dst_ip;
1515 break;
1516
1517 case TOK_SRCIP:
1518 mask->addr_type = 4;
1519 p32 = &mask->src_ip;
1520 break;
1521
1522 case TOK_DSTIP6:
1523 mask->addr_type = 6;
1524 pa6 = &mask->dst_ip6;
1525 break;
1526
1527 case TOK_SRCIP6:
1528 mask->addr_type = 6;
1529 pa6 = &mask->src_ip6;
1530 break;
1531
1532 case TOK_FLOWID:
1533 mask->addr_type = 6;
1534 p20 = &mask->flow_id6;
1535 break;
1536
1537 case TOK_DSTPORT:
1538 p16 = &mask->dst_port;
1539 break;
1540
1541 case TOK_SRCPORT:
1542 p16 = &mask->src_port;
1543 break;
1544
1545 case TOK_PROTO:
1546 break;
1547
1548 default:
1549 ac++; av--; /* backtrack */
1550 goto end_mask;
1551 }
1552 if (ac < 1)
1553 errx(EX_USAGE, "mask: value missing");
1554 if (*av[0] == '/') {
1555 a = strtoul(av[0]+1, &end, 0);
1556 if (pa6 == NULL)
1557 a = (a == 32) ? ~0 : (1 << a) - 1;
1558 } else
1559 a = strtoul(av[0], &end, 0);
1560 if (p32 != NULL)
1561 *p32 = a;
1562 else if (p16 != NULL) {
1563 if (a > 0xFFFF)
1564 errx(EX_DATAERR,
1565 "port mask must be 16 bit");
1566 *p16 = (uint16_t)a;
1567 } else if (p20 != NULL) {
1568 if (a > 0xfffff)
1569 errx(EX_DATAERR,
1570 "flow_id mask must be 20 bit");
1571 *p20 = (uint32_t)a;
1572 } else if (pa6 != NULL) {
1573 if (a > 128)
1574 errx(EX_DATAERR,
1575 "in6addr invalid mask len");
1576 else
1577 n2mask(pa6, a);
1578 } else {
1579 if (a > 0xFF)
1580 errx(EX_DATAERR,
1581 "proto mask must be 8 bit");
1582 mask->proto = (uint8_t)a;
1583 }
1584 if (a != 0)
1585 *flags |= DN_HAVE_MASK;
1586 ac--; av++;
1587 } /* end while, config masks */
1588 end_mask:
1589 break;
1590 #ifdef NEW_AQM
1591 case TOK_CODEL:
1592 case TOK_PIE:
1593 NEED(fs, "codel/pie is only for flowsets");
1594
1595 fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED);
1596 fs->flags |= DN_IS_AQM;
1597
1598 strlcpy(aqm_extra->name, av[-1],
1599 sizeof(aqm_extra->name));
1600 aqm_extra->oid.subtype = DN_AQM_PARAMS;
1601
1602 process_extra_parms(&ac, av, aqm_extra, tok);
1603 break;
1604
1605 case TOK_FQ_CODEL:
1606 case TOK_FQ_PIE:
1607 if (!strcmp(av[-1],"type"))
1608 errx(EX_DATAERR, "use type before fq_codel/fq_pie");
1609
1610 NEED(sch, "fq_codel/fq_pie is only for schd");
1611 strlcpy(sch_extra->name, av[-1],
1612 sizeof(sch_extra->name));
1613 sch_extra->oid.subtype = DN_SCH_PARAMS;
1614 process_extra_parms(&ac, av, sch_extra, tok);
1615 break;
1616 #endif
1617 case TOK_RED:
1618 case TOK_GRED:
1619 NEED1("red/gred needs w_q/min_th/max_th/max_p\n");
1620 fs->flags |= DN_IS_RED;
1621 if (tok == TOK_GRED)
1622 fs->flags |= DN_IS_GENTLE_RED;
1623 /*
1624 * the format for parameters is w_q/min_th/max_th/max_p
1625 */
1626 if ((end = strsep(&av[0], "/"))) {
1627 double w_q = strtod(end, NULL);
1628 if (w_q > 1 || w_q <= 0)
1629 errx(EX_DATAERR, "0 < w_q <= 1");
1630 fs->w_q = (int) (w_q * (1 << SCALE_RED));
1631 }
1632 if ((end = strsep(&av[0], "/"))) {
1633 fs->min_th = strtoul(end, &end, 0);
1634 if (*end == 'K' || *end == 'k')
1635 fs->min_th *= 1024;
1636 }
1637 if ((end = strsep(&av[0], "/"))) {
1638 fs->max_th = strtoul(end, &end, 0);
1639 if (*end == 'K' || *end == 'k')
1640 fs->max_th *= 1024;
1641 }
1642 if ((end = strsep(&av[0], "/"))) {
1643 double max_p = strtod(end, NULL);
1644 if (max_p > 1 || max_p < 0)
1645 errx(EX_DATAERR, "0 <= max_p <= 1");
1646 fs->max_p = (int)(max_p * (1 << SCALE_RED));
1647 }
1648 ac--; av++;
1649 break;
1650
1651 case TOK_ECN:
1652 fs->flags |= DN_IS_ECN;
1653 break;
1654
1655 case TOK_DROPTAIL:
1656 NEED(fs, "droptail is only for flowsets");
1657 fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED);
1658 break;
1659
1660 case TOK_BW:
1661 NEED(p, "bw is only for links");
1662 NEED1("bw needs bandwidth or interface\n");
1663 read_bandwidth(av[0], &p->bandwidth, NULL, 0);
1664 ac--; av++;
1665 break;
1666
1667 case TOK_DELAY:
1668 NEED(p, "delay is only for links");
1669 NEED1("delay needs argument 0..10000ms\n");
1670 p->delay = strtoul(av[0], NULL, 0);
1671 ac--; av++;
1672 break;
1673
1674 case TOK_TYPE: {
1675 int l;
1676 NEED(sch, "type is only for schedulers");
1677 NEED1("type needs a string");
1678 l = strlen(av[0]);
1679 if (l == 0 || l > 15)
1680 errx(1, "type %s too long\n", av[0]);
1681 strlcpy(sch->name, av[0], sizeof(sch->name));
1682 sch->oid.subtype = 0; /* use string */
1683 #ifdef NEW_AQM
1684 /* if fq_codel is selected, consider all tokens after it
1685 * as parameters
1686 */
1687 if (!strcasecmp(av[0],"fq_codel") || !strcasecmp(av[0],"fq_pie")){
1688 strlcpy(sch_extra->name, av[0],
1689 sizeof(sch_extra->name));
1690 sch_extra->oid.subtype = DN_SCH_PARAMS;
1691 process_extra_parms(&ac, av, sch_extra, tok);
1692 } else {
1693 ac--;av++;
1694 }
1695 #else
1696 ac--;av++;
1697 #endif
1698 break;
1699 }
1700
1701 case TOK_WEIGHT:
1702 NEED(fs, "weight is only for flowsets");
1703 NEED1("weight needs argument\n");
1704 fs->par[0] = strtol(av[0], &end, 0);
1705 ac--; av++;
1706 break;
1707
1708 case TOK_LMAX:
1709 NEED(fs, "lmax is only for flowsets");
1710 NEED1("lmax needs argument\n");
1711 fs->par[1] = strtol(av[0], &end, 0);
1712 ac--; av++;
1713 break;
1714
1715 case TOK_PRI:
1716 NEED(fs, "priority is only for flowsets");
1717 NEED1("priority needs argument\n");
1718 fs->par[2] = strtol(av[0], &end, 0);
1719 ac--; av++;
1720 break;
1721
1722 case TOK_SCHED:
1723 case TOK_PIPE:
1724 NEED(fs, "pipe/sched");
1725 NEED1("pipe/link/sched needs number\n");
1726 fs->sched_nr = strtoul(av[0], &end, 0);
1727 ac--; av++;
1728 break;
1729
1730 case TOK_PROFILE:
1731 NEED((!pf), "profile already set");
1732 NEED(p, "profile");
1733 {
1734 NEED1("extra delay needs the file name\n");
1735 pf = o_next(&buf, sizeof(*pf), DN_PROFILE);
1736 load_extra_delays(av[0], pf, p); //XXX can't fail?
1737 --ac; ++av;
1738 }
1739 break;
1740
1741 case TOK_BURST:
1742 NEED(p, "burst");
1743 NEED1("burst needs argument\n");
1744 errno = 0;
1745 if (expand_number(av[0], &p->burst) < 0)
1746 if (errno != ERANGE)
1747 errx(EX_DATAERR,
1748 "burst: invalid argument");
1749 if (errno || p->burst > (1ULL << 48) - 1)
1750 errx(EX_DATAERR,
1751 "burst: out of range (0..2^48-1)");
1752 ac--; av++;
1753 break;
1754
1755 default:
1756 errx(EX_DATAERR, "unrecognised option ``%s''", av[-1]);
1757 }
1758 }
1759
1760 /* check validity of parameters */
1761 if (p) {
1762 if (p->delay > 10000)
1763 errx(EX_DATAERR, "delay must be < 10000");
1764 if (p->bandwidth == (uint32_t)-1)
1765 p->bandwidth = 0;
1766 }
1767 if (fs) {
1768 /* XXX accept a 0 scheduler to keep the default */
1769 if (fs->flags & DN_QSIZE_BYTES) {
1770 size_t len;
1771 long limit;
1772
1773 len = sizeof(limit);
1774 if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit",
1775 &limit, &len, NULL, 0) == -1)
1776 limit = 1024*1024;
1777 if (fs->qsize > limit)
1778 errx(EX_DATAERR, "queue size must be < %ldB", limit);
1779 } else {
1780 size_t len;
1781 long limit;
1782
1783 len = sizeof(limit);
1784 if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit",
1785 &limit, &len, NULL, 0) == -1)
1786 limit = 100;
1787 if (fs->qsize > limit)
1788 errx(EX_DATAERR, "2 <= queue size <= %ld", limit);
1789 }
1790
1791 #ifdef NEW_AQM
1792 if ((fs->flags & DN_IS_ECN) && !((fs->flags & DN_IS_RED)||
1793 (fs->flags & DN_IS_AQM)))
1794 errx(EX_USAGE, "ECN can be used with red/gred/"
1795 "codel/fq_codel only!");
1796 #else
1797 if ((fs->flags & DN_IS_ECN) && !(fs->flags & DN_IS_RED))
1798 errx(EX_USAGE, "enable red/gred for ECN");
1799
1800 #endif
1801
1802 if (fs->flags & DN_IS_RED) {
1803 size_t len;
1804 int lookup_depth, avg_pkt_size;
1805
1806 if (!(fs->flags & DN_IS_ECN) && (fs->min_th >= fs->max_th))
1807 errx(EX_DATAERR, "min_th %d must be < than max_th %d",
1808 fs->min_th, fs->max_th);
1809 else if ((fs->flags & DN_IS_ECN) && (fs->min_th > fs->max_th))
1810 errx(EX_DATAERR, "min_th %d must be =< than max_th %d",
1811 fs->min_th, fs->max_th);
1812
1813 if (fs->max_th == 0)
1814 errx(EX_DATAERR, "max_th must be > 0");
1815
1816 len = sizeof(int);
1817 if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth",
1818 &lookup_depth, &len, NULL, 0) == -1)
1819 lookup_depth = 256;
1820 if (lookup_depth == 0)
1821 errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth"
1822 " must be greater than zero");
1823
1824 len = sizeof(int);
1825 if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size",
1826 &avg_pkt_size, &len, NULL, 0) == -1)
1827 avg_pkt_size = 512;
1828
1829 if (avg_pkt_size == 0)
1830 errx(EX_DATAERR,
1831 "net.inet.ip.dummynet.red_avg_pkt_size must"
1832 " be greater than zero");
1833
1834 #if 0 /* the following computation is now done in the kernel */
1835 /*
1836 * Ticks needed for sending a medium-sized packet.
1837 * Unfortunately, when we are configuring a WF2Q+ queue, we
1838 * do not have bandwidth information, because that is stored
1839 * in the parent pipe, and also we have multiple queues
1840 * competing for it. So we set s=0, which is not very
1841 * correct. But on the other hand, why do we want RED with
1842 * WF2Q+ ?
1843 */
1844 if (p.bandwidth==0) /* this is a WF2Q+ queue */
1845 s = 0;
1846 else
1847 s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth;
1848 /*
1849 * max idle time (in ticks) before avg queue size becomes 0.
1850 * NOTA: (3/w_q) is approx the value x so that
1851 * (1-w_q)^x < 10^-3.
1852 */
1853 w_q = ((double)fs->w_q) / (1 << SCALE_RED);
1854 idle = s * 3. / w_q;
1855 fs->lookup_step = (int)idle / lookup_depth;
1856 if (!fs->lookup_step)
1857 fs->lookup_step = 1;
1858 weight = 1 - w_q;
1859 for (t = fs->lookup_step; t > 1; --t)
1860 weight *= 1 - w_q;
1861 fs->lookup_weight = (int)(weight * (1 << SCALE_RED));
1862 #endif /* code moved in the kernel */
1863 }
1864 }
1865
1866 i = do_cmd(IP_DUMMYNET3, base, (char *)buf - (char *)base);
1867
1868 if (i)
1869 err(1, "setsockopt(%s)", "IP_DUMMYNET3");
1870 }
1871
1872 void
dummynet_flush(void)1873 dummynet_flush(void)
1874 {
1875 struct dn_id oid;
1876 oid_fill(&oid, sizeof(oid), DN_CMD_FLUSH, DN_API_VERSION);
1877 do_cmd(IP_DUMMYNET3, &oid, oid.len);
1878 }
1879
1880 /* Parse input for 'ipfw [pipe|sched|queue] show [range list]'
1881 * Returns the number of ranges, and possibly stores them
1882 * in the array v of size len.
1883 */
1884 static int
parse_range(int ac,char * av[],uint32_t * v,int len)1885 parse_range(int ac, char *av[], uint32_t *v, int len)
1886 {
1887 int n = 0;
1888 char *endptr, *s;
1889 uint32_t base[2];
1890
1891 if (v == NULL || len < 2) {
1892 v = base;
1893 len = 2;
1894 }
1895
1896 for (s = *av; s != NULL; av++, ac--) {
1897 v[0] = strtoul(s, &endptr, 10);
1898 v[1] = (*endptr != '-') ? v[0] :
1899 strtoul(endptr+1, &endptr, 10);
1900 if (*endptr == '\0') { /* prepare for next round */
1901 s = (ac > 0) ? *(av+1) : NULL;
1902 } else {
1903 if (*endptr != ',') {
1904 warn("invalid number: %s", s);
1905 s = ++endptr;
1906 continue;
1907 }
1908 /* continue processing from here */
1909 s = ++endptr;
1910 ac++;
1911 av--;
1912 }
1913 if (v[1] < v[0] ||
1914 v[0] >= DN_MAX_ID-1 ||
1915 v[1] >= DN_MAX_ID-1) {
1916 continue; /* invalid entry */
1917 }
1918 n++;
1919 /* translate if 'pipe list' */
1920 if (g_co.do_pipe == 1) {
1921 v[0] += DN_MAX_ID;
1922 v[1] += DN_MAX_ID;
1923 }
1924 v = (n*2 < len) ? v + 2 : base;
1925 }
1926 return n;
1927 }
1928
1929 /* main entry point for dummynet list functions. co.do_pipe indicates
1930 * which function we want to support.
1931 * av may contain filtering arguments, either individual entries
1932 * or ranges, or lists (space or commas are valid separators).
1933 * Format for a range can be n1-n2 or n3 n4 n5 ...
1934 * In a range n1 must be <= n2, otherwise the range is ignored.
1935 * A number 'n4' is translate in a range 'n4-n4'
1936 * All number must be > 0 and < DN_MAX_ID-1
1937 */
1938 void
dummynet_list(int ac,char * av[],int show_counters)1939 dummynet_list(int ac, char *av[], int show_counters)
1940 {
1941 struct dn_id *oid, *x = NULL;
1942 int ret, i;
1943 int n; /* # of ranges */
1944 u_int buflen, l;
1945 u_int max_size; /* largest obj passed up */
1946
1947 (void)show_counters; // XXX unused, but we should use it.
1948 ac--;
1949 av++; /* skip 'list' | 'show' word */
1950
1951 n = parse_range(ac, av, NULL, 0); /* Count # of ranges. */
1952
1953 /* Allocate space to store ranges */
1954 l = sizeof(*oid) + sizeof(uint32_t) * n * 2;
1955 oid = safe_calloc(1, l);
1956 oid_fill(oid, l, DN_CMD_GET, DN_API_VERSION);
1957
1958 if (n > 0) /* store ranges in idx */
1959 parse_range(ac, av, (uint32_t *)(oid + 1), n*2);
1960 /*
1961 * Compute the size of the largest object returned. If the
1962 * response leaves at least this much spare space in the
1963 * buffer, then surely the response is complete; otherwise
1964 * there might be a risk of truncation and we will need to
1965 * retry with a larger buffer.
1966 * XXX don't bother with smaller structs.
1967 */
1968 max_size = sizeof(struct dn_fs);
1969 if (max_size < sizeof(struct dn_sch))
1970 max_size = sizeof(struct dn_sch);
1971 if (max_size < sizeof(struct dn_flow))
1972 max_size = sizeof(struct dn_flow);
1973
1974 switch (g_co.do_pipe) {
1975 case 1:
1976 oid->subtype = DN_LINK; /* list pipe */
1977 break;
1978 case 2:
1979 oid->subtype = DN_FS; /* list queue */
1980 break;
1981 case 3:
1982 oid->subtype = DN_SCH; /* list sched */
1983 break;
1984 }
1985
1986 /*
1987 * Ask the kernel an estimate of the required space (result
1988 * in oid.id), unless we are requesting a subset of objects,
1989 * in which case the kernel does not give an exact answer.
1990 * In any case, space might grow in the meantime due to the
1991 * creation of new queues, so we must be prepared to retry.
1992 */
1993 if (n > 0) {
1994 buflen = 4*1024;
1995 } else {
1996 ret = do_cmd(-IP_DUMMYNET3, oid, (uintptr_t)&l);
1997 if (ret != 0 || oid->id <= sizeof(*oid))
1998 goto done;
1999 buflen = oid->id + max_size;
2000 oid->len = sizeof(*oid); /* restore */
2001 }
2002 /* Try a few times, until the buffer fits */
2003 for (i = 0; i < 20; i++) {
2004 l = buflen;
2005 x = safe_realloc(x, l);
2006 bcopy(oid, x, oid->len);
2007 ret = do_cmd(-IP_DUMMYNET3, x, (uintptr_t)&l);
2008 if (ret != 0 || x->id <= sizeof(*oid))
2009 goto done; /* no response */
2010 if (l + max_size <= buflen)
2011 break; /* ok */
2012 buflen *= 2; /* double for next attempt */
2013 }
2014 list_pipes(x, O_NEXT(x, l));
2015 done:
2016 if (x)
2017 free(x);
2018 free(oid);
2019 }
2020