X-Git-Url: http://git.onelab.eu/?p=ipfw.git;a=blobdiff_plain;f=ipfw%2Fdummynet.c;h=231f52f1a07d88f9fb7af9cfe26b4100250cbaf2;hp=e2c5ff142f7bd4ce0ef981fc98be58523406ddfa;hb=HEAD;hpb=6c2e192c3237bd46db6ad4230fed71d28f362331 diff --git a/ipfw/dummynet.c b/ipfw/dummynet.c index e2c5ff1..231f52f 100644 --- a/ipfw/dummynet.c +++ b/ipfw/dummynet.c @@ -1,10 +1,5 @@ /* - * Copyright (c) 2002-2003 Luigi Rizzo - * Copyright (c) 1996 Alex Nash, Paul Traina, Poul-Henning Kamp - * Copyright (c) 1994 Ugen J.S.Antsilevich - * - * Idea and grammar partially left from: - * Copyright (c) 1993 Daniel Boulet + * Copyright (c) 2002-2003,2010 Luigi Rizzo * * Redistribution and use in source forms, with and without modification, * are permitted provided that this entire comment appears intact. @@ -15,16 +10,13 @@ * * This software is provided ``AS IS'' without any warranties of any kind. * - * NEW command line interface for IP firewall facility - * - * $FreeBSD: head/sbin/ipfw/dummynet.c 187769 2009-01-27 11:06:59Z luigi $ + * $FreeBSD: head/sbin/ipfw/dummynet.c 206843 2010-04-19 15:11:45Z luigi $ * * dummynet support */ #include #include -#include /* XXX there are several sysctl leftover here */ #include @@ -32,6 +24,8 @@ #include #include +#include +#include #include #include #include @@ -44,6 +38,7 @@ #include #include /* inet_ntoa */ + static struct _s_x dummynet_params[] = { { "plr", TOK_PLR }, { "noerror", TOK_NOERROR }, @@ -54,32 +49,73 @@ static struct _s_x dummynet_params[] = { { "src-port", TOK_SRCPORT }, { "proto", TOK_PROTO }, { "weight", TOK_WEIGHT }, + { "lmax", TOK_LMAX }, + { "maxlen", TOK_LMAX }, { "all", TOK_ALL }, - { "mask", TOK_MASK }, + { "mask", TOK_MASK }, /* alias for both */ + { "sched_mask", TOK_SCHED_MASK }, + { "flow_mask", TOK_FLOW_MASK }, { "droptail", TOK_DROPTAIL }, { "red", TOK_RED }, { "gred", TOK_GRED }, { "bw", TOK_BW }, { "bandwidth", TOK_BW }, { "delay", TOK_DELAY }, + { "link", TOK_LINK }, { "pipe", TOK_PIPE }, { "queue", TOK_QUEUE }, + { "flowset", TOK_FLOWSET }, + { "sched", TOK_SCHED }, + { "pri", TOK_PRI }, + { "priority", TOK_PRI }, + { "type", TOK_TYPE }, { "flow-id", TOK_FLOWID}, { "dst-ipv6", TOK_DSTIP6}, { "dst-ip6", TOK_DSTIP6}, { "src-ipv6", TOK_SRCIP6}, { "src-ip6", TOK_SRCIP6}, - { "profile", TOK_PIPE_PROFILE}, + { "profile", TOK_PROFILE}, + { "burst", TOK_BURST}, { "dummynet-params", TOK_NULL }, { NULL, 0 } /* terminator */ }; -/* - * XXX to be updated to the new version, - * without the global struct command_opts variable - */ +#define O_NEXT(p, len) ((void *)((char *)p + len)) + +static void +oid_fill(struct dn_id *oid, int len, int type, uintptr_t id) +{ + oid->len = len; + oid->type = type; + oid->subtype = 0; + oid->id = id; +} + +/* make room in the buffer and move the pointer forward */ +static void * +o_next(struct dn_id **o, int len, int type) +{ + struct dn_id *ret = *o; + oid_fill(ret, len, type, 0); + *o = O_NEXT(*o, len); + return ret; +} + +/* handle variable lenght structures moving back the pointer and fixing lenght */ +static void * +o_compact(struct dn_id **o, int len, int real_length, int type) +{ + struct dn_id *ret = *o; + + ret = O_NEXT(*o, -len); + oid_fill(ret, real_length, type, 0); + *o = O_NEXT(ret, real_length); + return ret; +} + +#if 0 static int -sort_q(void * to_be_done, const void *pa, const void *pb) +sort_q(void *arg, const void *pa, const void *pb) { int rev = (co.do_sort < 0); int field = rev ? -co.do_sort : co.do_sort; @@ -109,117 +145,93 @@ sort_q(void * to_be_done, const void *pa, const void *pb) res = 1; return (int)(rev ? res : -res); } +#endif +/* print a mask and header for the subsequent list of flows */ static void -list_queues(struct dn_flow_set *fs, struct dn_flow_queue *q) +print_mask(struct ipfw_flow_id *id) { - int l; - int index_printed, indexes = 0; - char buff[255]; - struct protoent *pe; - - if (fs->rq_elements == 0) - return; + if (!IS_IP6_FLOW_ID(id)) { + printf(" " + "mask: %s 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n", + id->extra ? "queue," : "", + id->proto, + id->src_ip, id->src_port, + id->dst_ip, id->dst_port); + } else { + char buf[255]; + printf("\n mask: %sproto: 0x%02x, flow_id: 0x%08x, ", + id->extra ? "queue," : "", + id->proto, id->flow_id6); + inet_ntop(AF_INET6, &(id->src_ip6), buf, sizeof(buf)); + printf("%s/0x%04x -> ", buf, id->src_port); + inet_ntop(AF_INET6, &(id->dst_ip6), buf, sizeof(buf)); + printf("%s/0x%04x\n", buf, id->dst_port); + } +} - if (co.do_sort != 0) - qsort_r(q, fs->rq_elements, sizeof *q, NULL, sort_q); +static void +print_header(struct ipfw_flow_id *id) +{ + if (!IS_IP6_FLOW_ID(id)) + printf("BKT Prot ___Source IP/port____ " + "____Dest. IP/port____ " + "Tot_pkt/bytes Pkt/Byte Drp\n"); + else + printf("BKT ___Prot___ _flow-id_ " + "______________Source IPv6/port_______________ " + "_______________Dest. IPv6/port_______________ " + "Tot_pkt/bytes Pkt/Byte Drp\n"); +} - /* Print IPv4 flows */ - index_printed = 0; - for (l = 0; l < fs->rq_elements; l++) { - struct in_addr ina; +static void +list_flow(struct dn_flow *ni, int *print) +{ + char buff[255]; + struct protoent *pe = NULL; + struct in_addr ina; + struct ipfw_flow_id *id = &ni->fid; + if (*print) { + print_header(&ni->fid); + *print = 0; + } + pe = getprotobynumber(id->proto); /* XXX: Should check for IPv4 flows */ - if (IS_IP6_FLOW_ID(&(q[l].id))) - continue; - - if (!index_printed) { - index_printed = 1; - if (indexes > 0) /* currently a no-op */ - printf("\n"); - indexes++; - printf(" " - "mask: 0x%02x 0x%08x/0x%04x -> 0x%08x/0x%04x\n", - fs->flow_mask.proto, - fs->flow_mask.src_ip, fs->flow_mask.src_port, - fs->flow_mask.dst_ip, fs->flow_mask.dst_port); - - printf("BKT Prot ___Source IP/port____ " - "____Dest. IP/port____ " - "Tot_pkt/bytes Pkt/Byte Drp\n"); - } - - printf("%3d ", q[l].hash_slot); - pe = getprotobynumber(q[l].id.proto); + printf("%3u%c", (ni->oid.id) & 0xff, + id->extra ? '*' : ' '); + if (!IS_IP6_FLOW_ID(id)) { if (pe) printf("%-4s ", pe->p_name); else - printf("%4u ", q[l].id.proto); - ina.s_addr = htonl(q[l].id.src_ip); + printf("%4u ", id->proto); + ina.s_addr = htonl(id->src_ip); printf("%15s/%-5d ", - inet_ntoa(ina), q[l].id.src_port); - ina.s_addr = htonl(q[l].id.dst_ip); + inet_ntoa(ina), id->src_port); + ina.s_addr = htonl(id->dst_ip); printf("%15s/%-5d ", - inet_ntoa(ina), q[l].id.dst_port); - printf("%4llu %8llu %2u %4u %3u\n", - align_uint64(&q[l].tot_pkts), - align_uint64(&q[l].tot_bytes), - q[l].len, q[l].len_bytes, q[l].drops); - if (co.verbose) - printf(" S %20llu F %20llu\n", - align_uint64(&q[l].S), align_uint64(&q[l].F)); - } - - /* Print IPv6 flows */ - index_printed = 0; - for (l = 0; l < fs->rq_elements; l++) { - if (!IS_IP6_FLOW_ID(&(q[l].id))) - continue; - - if (!index_printed) { - index_printed = 1; - if (indexes > 0) - printf("\n"); - indexes++; - printf("\n mask: proto: 0x%02x, flow_id: 0x%08x, ", - fs->flow_mask.proto, fs->flow_mask.flow_id6); - inet_ntop(AF_INET6, &(fs->flow_mask.src_ip6), - buff, sizeof(buff)); - printf("%s/0x%04x -> ", buff, fs->flow_mask.src_port); - inet_ntop( AF_INET6, &(fs->flow_mask.dst_ip6), - buff, sizeof(buff) ); - printf("%s/0x%04x\n", buff, fs->flow_mask.dst_port); - - printf("BKT ___Prot___ _flow-id_ " - "______________Source IPv6/port_______________ " - "_______________Dest. IPv6/port_______________ " - "Tot_pkt/bytes Pkt/Byte Drp\n"); - } - printf("%3d ", q[l].hash_slot); - pe = getprotobynumber(q[l].id.proto); + inet_ntoa(ina), id->dst_port); + } else { + /* Print IPv6 flows */ if (pe != NULL) printf("%9s ", pe->p_name); else - printf("%9u ", q[l].id.proto); - printf("%7d %39s/%-5d ", q[l].id.flow_id6, - inet_ntop(AF_INET6, &(q[l].id.src_ip6), buff, sizeof(buff)), - q[l].id.src_port); + printf("%9u ", id->proto); + printf("%7d %39s/%-5d ", id->flow_id6, + inet_ntop(AF_INET6, &(id->src_ip6), buff, sizeof(buff)), + id->src_port); printf(" %39s/%-5d ", - inet_ntop(AF_INET6, &(q[l].id.dst_ip6), buff, sizeof(buff)), - q[l].id.dst_port); - printf(" %4llu %8llu %2u %4u %3u\n", - align_uint64(&q[l].tot_pkts), - align_uint64(&q[l].tot_bytes), - q[l].len, q[l].len_bytes, q[l].drops); - if (co.verbose) - printf(" S %20llu F %20llu\n", - align_uint64(&q[l].S), - align_uint64(&q[l].F)); + inet_ntop(AF_INET6, &(id->dst_ip6), buff, sizeof(buff)), + id->dst_port); } + pr_u64(&ni->tot_pkts, 4); + pr_u64(&ni->tot_bytes, 8); + printf("%2u %4u %3u\n", + ni->length, ni->len_bytes, ni->drops); } static void -print_flowset_parms(struct dn_flow_set *fs, char *prefix) +print_flowset_parms(struct dn_fs *fs, char *prefix) { int l; char qs[30]; @@ -227,7 +239,7 @@ print_flowset_parms(struct dn_flow_set *fs, char *prefix) char red[90]; /* Display RED parameters */ l = fs->qsize; - if (fs->flags_fs & DN_QSIZE_IS_BYTES) { + if (fs->flags & DN_QSIZE_BYTES) { if (l >= 8192) sprintf(qs, "%d KB", l / 1024); else @@ -238,23 +250,34 @@ print_flowset_parms(struct dn_flow_set *fs, char *prefix) sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff)); else plr[0] = '\0'; - if (fs->flags_fs & DN_IS_RED) /* RED parameters */ + + if (fs->flags & DN_IS_RED) /* RED parameters */ sprintf(red, - "\n\t %cRED w_q %f min_th %d max_th %d max_p %f", - (fs->flags_fs & DN_IS_GENTLE_RED) ? 'G' : ' ', + "\n\t %cRED w_q %f min_th %d max_th %d max_p %f", + (fs->flags & DN_IS_GENTLE_RED) ? 'G' : ' ', 1.0 * fs->w_q / (double)(1 << SCALE_RED), - SCALE_VAL(fs->min_th), - SCALE_VAL(fs->max_th), + fs->min_th, + fs->max_th, 1.0 * fs->max_p / (double)(1 << SCALE_RED)); else sprintf(red, "droptail"); - printf("%s %s%s %d queues (%d buckets) %s\n", - prefix, qs, plr, fs->rq_elements, fs->rq_size, red); + if (prefix[0]) { + printf("%s %s%s %d queues (%d buckets) %s\n", + prefix, qs, plr, fs->oid.id, fs->buckets, red); + prefix[0] = '\0'; + } else { + printf("q%05d %s%s %d flows (%d buckets) sched %d " + "weight %d lmax %d pri %d %s\n", + fs->fs_nr, qs, plr, fs->oid.id, fs->buckets, + fs->sched_nr, fs->par[0], fs->par[1], fs->par[2], red); + if (fs->flags & DN_HAVE_MASK) + print_mask(&fs->flow_mask); + } } static void -print_extra_delay_parms(struct dn_pipe *p, char *prefix) +print_extra_delay_parms(struct dn_profile *p) { double loss; if (p->samples_no <= 0) @@ -262,102 +285,131 @@ print_extra_delay_parms(struct dn_pipe *p, char *prefix) loss = p->loss_level; loss /= p->samples_no; - printf("%s profile: name \"%s\" loss %f samples %d\n", - prefix, p->name, loss, p->samples_no); + printf("\t profile: name \"%s\" loss %f samples %d\n", + p->name, loss, p->samples_no); } -void -ipfw_list_pipes(void *data, uint nbytes, int ac, char *av[]) +static void +flush_buf(char *buf) { - int rulenum; - void *next = data; - struct dn_pipe *p = (struct dn_pipe *) data; - struct dn_flow_set *fs; - struct dn_flow_queue *q; - int l; - - if (ac > 0) - rulenum = strtoul(*av++, NULL, 10); - else - rulenum = 0; - for (; nbytes >= sizeof *p; p = (struct dn_pipe *)next) { - double b = p->bandwidth; - char buf[30]; - char prefix[80]; - - if (SLIST_NEXT(p, next) != (struct dn_pipe *)DN_IS_PIPE) - break; /* done with pipes, now queues */ - - /* - * compute length, as pipe have variable size - */ - l = sizeof(*p) + p->fs.rq_elements * sizeof(*q); - next = (char *)p + l; - nbytes -= l; - - if ((rulenum != 0 && rulenum != p->pipe_nr) || co.do_pipe == 2) - continue; - - /* - * Print rate (or clocking interface) - */ - if (p->if_name[0] != '\0') - sprintf(buf, "%s", p->if_name); - else if (b == 0) - sprintf(buf, "unlimited"); - else if (b >= 1000000) - sprintf(buf, "%7.3f Mbit/s", b/1000000); - else if (b >= 1000) - sprintf(buf, "%7.3f Kbit/s", b/1000); - else - sprintf(buf, "%7.3f bit/s ", b); - - sprintf(prefix, "%05d: %s %4d ms ", - p->pipe_nr, buf, p->delay); - - print_extra_delay_parms(p, prefix); - - print_flowset_parms(&(p->fs), prefix); - - q = (struct dn_flow_queue *)(p+1); - list_queues(&(p->fs), q); - } - for (fs = next; nbytes >= sizeof *fs; fs = next) { - char prefix[80]; - - if (SLIST_NEXT(fs, next) != (struct dn_flow_set *)DN_IS_QUEUE) - break; - l = sizeof(*fs) + fs->rq_elements * sizeof(*q); - next = (char *)fs + l; - nbytes -= l; - - if (rulenum != 0 && ((rulenum != fs->fs_nr && co.do_pipe == 2) || - (rulenum != fs->parent_nr && co.do_pipe == 1))) { - continue; - } - - q = (struct dn_flow_queue *)(fs+1); - sprintf(prefix, "q%05d: weight %d pipe %d ", - fs->fs_nr, fs->weight, fs->parent_nr); - print_flowset_parms(fs, prefix); - list_queues(fs, q); + if (buf[0]) + printf("%s\n", buf); + buf[0] = '\0'; +} + +/* + * generic list routine. We expect objects in a specific order, i.e. + * PIPES AND SCHEDULERS: + * link; scheduler; internal flowset if any; instances + * we can tell a pipe from the number. + * + * FLOWSETS: + * flowset; queues; + * link i (int queue); scheduler i; si(i) { flowsets() : queues } + */ +static void +list_pipes(struct dn_id *oid, struct dn_id *end) +{ + char buf[160]; /* pending buffer */ + int toPrint = 1; /* print header */ + + buf[0] = '\0'; + for (; oid != end; oid = O_NEXT(oid, oid->len)) { + if (oid->len < sizeof(*oid)) + errx(1, "invalid oid len %d\n", oid->len); + + switch (oid->type) { + default: + flush_buf(buf); + printf("unrecognized object %d size %d\n", oid->type, oid->len); + break; + case DN_TEXT: /* list of attached flowsets */ + { + int i, l; + struct { + struct dn_id id; + uint32_t p[0]; + } *d = (void *)oid; + l = (oid->len - sizeof(*oid))/sizeof(d->p[0]); + if (l == 0) + break; + printf(" Children flowsets: "); + for (i = 0; i < l; i++) + printf("%u ", d->p[i]); + printf("\n"); + break; + } + case DN_CMD_GET: + if (co.verbose) + printf("answer for cmd %d, len %d\n", oid->type, oid->id); + break; + case DN_SCH: { + struct dn_sch *s = (struct dn_sch *)oid; + flush_buf(buf); + printf(" sched %d type %s flags 0x%x %d buckets %d active\n", + s->sched_nr, + s->name, s->flags, s->buckets, s->oid.id); + if (s->flags & DN_HAVE_MASK) + print_mask(&s->sched_mask); + } + break; + + case DN_FLOW: + list_flow((struct dn_flow *)oid, &toPrint); + break; + + case DN_LINK: { + struct dn_link *p = (struct dn_link *)oid; + double b = p->bandwidth; + char bwbuf[30]; + char burst[5 + 7]; + + /* This starts a new object so flush buffer */ + flush_buf(buf); + /* data rate */ + if (b == 0) + sprintf(bwbuf, "unlimited "); + else if (b >= 1000000) + sprintf(bwbuf, "%7.3f Mbit/s", b/1000000); + else if (b >= 1000) + sprintf(bwbuf, "%7.3f Kbit/s", b/1000); + else + sprintf(bwbuf, "%7.3f bit/s ", b); + + if (humanize_number(burst, sizeof(burst), p->burst, + "", HN_AUTOSCALE, 0) < 0 || co.verbose) + sprintf(burst, "%d", (int)p->burst); + sprintf(buf, "%05d: %s %4d ms burst %s", + p->link_nr % DN_MAX_ID, bwbuf, p->delay, burst); + } + break; + + case DN_FS: + print_flowset_parms((struct dn_fs *)oid, buf); + break; + case DN_PROFILE: + flush_buf(buf); + print_extra_delay_parms((struct dn_profile *)oid); } + flush_buf(buf); // XXX does it really go here ? + } } /* - * Delete pipe or queue i + * Delete pipe, queue or scheduler i */ int -ipfw_delete_pipe(int pipe_or_queue, int i) +ipfw_delete_pipe(int do_pipe, int i) { - struct dn_pipe p; - - memset(&p, 0, sizeof p); - if (pipe_or_queue == 1) - p.pipe_nr = i; /* pipe */ - else - p.fs.fs_nr = i; /* queue */ - i = do_cmd(IP_DUMMYNET_DEL, &p, sizeof p); + struct { + struct dn_id oid; + uintptr_t a[1]; /* add more if we want a list */ + } cmd; + oid_fill((void *)&cmd, sizeof(cmd), DN_CMD_DELETE, DN_API_VERSION); + cmd.oid.subtype = (do_pipe == 1) ? DN_LINK : + ( (do_pipe == 2) ? DN_FS : DN_SCH); + cmd.a[0] = i; + i = do_cmd(IP_DUMMYNET3, &cmd, cmd.oid.len); if (i) { i = 1; warn("rule %u: setsockopt(IP_DUMMYNET_DEL)", i); @@ -394,7 +446,7 @@ ipfw_delete_pipe(int pipe_or_queue, int i) * The empirical curve may have both vertical and horizontal lines. * Vertical lines represent constant delay for a range of * probabilities; horizontal lines correspond to a discontinuty - * in the delay distribution: the pipe will use the largest delay + * in the delay distribution: the link will use the largest delay * for a given probability. * * To pass the curve to dummynet, we must store the parameters @@ -450,7 +502,6 @@ ipfw_delete_pipe(int pipe_or_queue, int i) * */ -/* XXX move to an array definition ? */ #define ED_MAX_LINE_LEN 256+ED_MAX_NAME_LEN #define ED_TOK_SAMPLES "samples" #define ED_TOK_LOSS "loss-level" @@ -485,9 +536,12 @@ static void read_bandwidth(char *arg, int *bandwidth, char *if_name, int namelen) { if (*bandwidth != -1) - warn("duplicate token, override bandwidth value!"); + warnx("duplicate token, override bandwidth value!"); if (arg[0] >= 'a' && arg[0] <= 'z') { + if (!if_name) { + errx(1, "no if support"); + } if (namelen >= IFNAMSIZ) warn("interface name truncated"); namelen--; @@ -503,12 +557,12 @@ read_bandwidth(char *arg, int *bandwidth, char *if_name, int namelen) if (*end == 'K' || *end == 'k') { end++; bw *= 1000; - } else if (*end == 'M') { + } else if (*end == 'M' || *end == 'm') { end++; bw *= 1000000; } if ((*end == 'B' && - _substrcmp2(end, "Bi", "Bit/s") != 0) || + _substrcmp2(end, "Bi", "Bit/s") != 0) || _substrcmp2(end, "by", "bytes") == 0) bw *= 8; @@ -516,7 +570,8 @@ read_bandwidth(char *arg, int *bandwidth, char *if_name, int namelen) errx(EX_DATAERR, "bandwidth too large"); *bandwidth = bw; - if_name[0] = '\0'; + if (if_name) + if_name[0] = '\0'; } } @@ -545,13 +600,77 @@ compare_points(const void *vp1, const void *vp2) #define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno +/* + * Interpolate a set of proability-value tuples. + * + * This function takes as input a tuple of values + * and samples the interpolated curve described from the tuples. + * + * The user defined points are stored in the ponts structure. + * The number of points is stored in points_no. + * The user defined sampling value is stored in samples_no. + * The resulting samples are in the "samples" pointer. + * + * We assume that The last point for the '1' value of the + * probability should be defined. (XXX add checks for this) + * + * The input data are points and points_no. + * The output data are s (the array of s_no samples) + * and s_no (the number of samples) + * + */ +static void +interpolate_samples(struct point *p, int points_no, + int *samples, int samples_no, const char *filename) +{ + double dy; /* delta on the y axis */ + double y; /* current value of y */ + double x; /* current value of x */ + double m; /* the y slope */ + int i; /* samples index */ + int curr; /* points current index */ + + /* make sure that there are enough points. */ + /* XXX Duplicated should be removed */ + if (points_no < 3) + errx(EX_DATAERR, "%s too few samples, need at least %d", + filename, 3); + + qsort(p, points_no, sizeof(struct point), compare_points); + + dy = 1.0/samples_no; + y = 0; + + for (i=0, curr = 0; i < samples_no; i++, y+=dy) { + /* This statment move the curr pointer to the next point + * skipping the points with the same x value. We are + * guaranteed to exit from the loop because the + * last possible value of y is stricly less than 1 + * and the last possible value of the y points is 1 */ + while ( y >= p[curr+1].prob ) curr++; + + /* compute the slope of the curve */ + m = (p[curr+1].delay - p[curr].delay) / (p[curr+1].prob - p[curr].prob); + /* compute the x value starting from the current point */ + x = p[curr].delay + (y - p[curr].prob) * m; + samples[i] = x; + } + + /* add the last sample */ + samples[i] = p[curr+1].delay; +} + +/* + * p is the link (old pipe) + * pf is the profile + */ static void -load_extra_delays(const char *filename, struct dn_pipe *p) +load_extra_delays(const char *filename, struct dn_profile *p, + struct dn_link *link) { char line[ED_MAX_LINE_LEN]; FILE *f; int lineno = 0; - int i; int samples = -1; double loss = -1.0; @@ -561,7 +680,11 @@ load_extra_delays(const char *filename, struct dn_pipe *p) struct point points[ED_MAX_SAMPLES_NO]; int points_no = 0; + /* XXX link never NULL? */ + p->link_nr = link->link_nr; + profile_name[0] = '\0'; + f = fopen(filename, "r"); if (f == NULL) err(EX_UNAVAILABLE, "fopen: %s", filename); @@ -585,10 +708,9 @@ load_extra_delays(const char *filename, struct dn_pipe *p) else arg = s; } - if (name == NULL) /* empty line */ + + if ((name == NULL) || (*name == '#')) /* empty line */ continue; - if (arg == NULL) - errx(ED_EFMT("missing arg for %s"), name); if (!strcasecmp(name, ED_TOK_SAMPLES)) { if (samples > 0) @@ -596,12 +718,14 @@ load_extra_delays(const char *filename, struct dn_pipe *p) if (atoi(arg) <=0) errx(ED_EFMT("invalid number of samples")); samples = atoi(arg); - if (samples>ED_MAX_SAMPLES_NO) + if (samples>=ED_MAX_SAMPLES_NO-1) errx(ED_EFMT("too many samples, maximum is %d"), - ED_MAX_SAMPLES_NO); + ED_MAX_SAMPLES_NO-1); do_points = 0; } else if (!strcasecmp(name, ED_TOK_BW)) { - read_bandwidth(arg, &p->bandwidth, p->if_name, sizeof(p->if_name)); + char buf[IFNAMSIZ]; + read_bandwidth(arg, &link->bandwidth, buf, sizeof(buf)); + p->bandwidth = link->bandwidth; } else if (!strcasecmp(name, ED_TOK_LOSS)) { if (loss != -1.0) errx(ED_EFMT("duplicated token: %s"), name); @@ -645,6 +769,8 @@ load_extra_delays(const char *filename, struct dn_pipe *p) } } + fclose (f); + if (samples == -1) { warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES); samples = 100; @@ -655,59 +781,129 @@ load_extra_delays(const char *filename, struct dn_pipe *p) loss = 1; } - /* make sure that there are enough points. */ - if (points_no < ED_MIN_SAMPLES_NO) - errx(ED_EFMT("too few samples, need at least %d"), - ED_MIN_SAMPLES_NO); - - qsort(points, points_no, sizeof(struct point), compare_points); + interpolate_samples(points, points_no, p->samples, samples, filename); - /* interpolation */ - for (i = 0; isamples[index] = x1; - } else { - double m = (y2-y1)/(x2-x1); - double c = y1 - m*x1; - for (; indexsamples[index] = (index - c)/m; - } - } - p->samples_no = samples; + p->samples_no = samples++; p->loss_level = loss * samples; strncpy(p->name, profile_name, sizeof(p->name)); } +/* + * configuration of pipes, schedulers, flowsets. + * When we configure a new scheduler, an empty pipe is created, so: + * + * do_pipe = 1 -> "pipe N config ..." only for backward compatibility + * sched N+Delta type fifo sched_mask ... + * pipe N+Delta + * flowset N+Delta pipe N+Delta (no parameters) + * sched N type wf2q+ sched_mask ... + * pipe N + * + * do_pipe = 2 -> flowset N config + * flowset N parameters + * + * do_pipe = 3 -> sched N config + * sched N parameters (default no pipe) + * optional Pipe N config ... + * pipe ==> + */ void ipfw_config_pipe(int ac, char **av) { - int samples[ED_MAX_SAMPLES_NO]; - struct dn_pipe p; - int i; + int i, j; char *end; void *par = NULL; - - memset(&p, 0, sizeof p); - p.bandwidth = -1; + struct dn_id *buf, *base; + struct dn_sch *sch = NULL; + struct dn_link *p = NULL; + struct dn_fs *fs = NULL; + struct dn_profile *pf = NULL; + struct ipfw_flow_id *mask = NULL; + int lmax; + uint32_t _foo = 0, *flags = &_foo , *buckets = &_foo; + size_t max_pf_size = sizeof(struct dn_profile) + ED_MAX_SAMPLES_NO * sizeof(int); + + /* + * allocate space for 1 header, + * 1 scheduler, 1 link, 1 flowset, 1 profile + */ + lmax = sizeof(struct dn_id); /* command header */ + lmax += sizeof(struct dn_sch) + sizeof(struct dn_link) + + sizeof(struct dn_fs); + lmax += max_pf_size; av++; ac--; /* Pipe number */ if (ac && isdigit(**av)) { i = atoi(*av); av++; ac--; - if (co.do_pipe == 1) - p.pipe_nr = i; - else - p.fs.fs_nr = i; + } else + i = -1; + if (i <= 0) + errx(EX_USAGE, "need a pipe/flowset/sched number"); + base = buf = safe_calloc(1, lmax); + /* all commands start with a 'CONFIGURE' and a version */ + o_next(&buf, sizeof(struct dn_id), DN_CMD_CONFIG); + base->id = DN_API_VERSION; + + switch (co.do_pipe) { + case 1: /* "pipe N config ..." */ + /* Allocate space for the WF2Q+ scheduler, its link + * and the FIFO flowset. Set the number, but leave + * the scheduler subtype and other parameters to 0 + * so the kernel will use appropriate defaults. + * XXX todo: add a flag to record if a parameter + * is actually configured. + * If we do a 'pipe config' mask -> sched_mask. + * The FIFO scheduler and link are derived from the + * WF2Q+ one in the kernel. + */ + sch = o_next(&buf, sizeof(*sch), DN_SCH); + p = o_next(&buf, sizeof(*p), DN_LINK); + fs = o_next(&buf, sizeof(*fs), DN_FS); + + sch->sched_nr = i; + sch->oid.subtype = 0; /* defaults to WF2Q+ */ + mask = &sch->sched_mask; + flags = &sch->flags; + buckets = &sch->buckets; + *flags |= DN_PIPE_CMD; + + p->link_nr = i; + + /* This flowset is only for the FIFO scheduler */ + fs->fs_nr = i + 2*DN_MAX_ID; + fs->sched_nr = i + DN_MAX_ID; + break; + + case 2: /* "queue N config ... " */ + fs = o_next(&buf, sizeof(*fs), DN_FS); + fs->fs_nr = i; + mask = &fs->flow_mask; + flags = &fs->flags; + buckets = &fs->buckets; + break; + + case 3: /* "sched N config ..." */ + sch = o_next(&buf, sizeof(*sch), DN_SCH); + fs = o_next(&buf, sizeof(*fs), DN_FS); + sch->sched_nr = i; + mask = &sch->sched_mask; + flags = &sch->flags; + buckets = &sch->buckets; + /* fs is used only with !MULTIQUEUE schedulers */ + fs->fs_nr = i + DN_MAX_ID; + fs->sched_nr = i; + break; } + /* set to -1 those fields for which we want to reuse existing + * values from the kernel. + * Also, *_nr and subtype = 0 mean reuse the value from the kernel. + * XXX todo: support reuse of the mask. + */ + if (p) + p->bandwidth = -1; + for (j = 0; j < sizeof(fs->par)/sizeof(fs->par[0]); j++) + fs->par[j] = -1; while (ac > 0) { double d; int tok = match_token(dummynet_params, *av); @@ -715,41 +911,48 @@ ipfw_config_pipe(int ac, char **av) switch(tok) { case TOK_NOERROR: - p.fs.flags_fs |= DN_NOERROR; + NEED(fs, "noerror is only for pipes"); + fs->flags |= DN_NOERROR; break; case TOK_PLR: + NEED(fs, "plr is only for pipes"); NEED1("plr needs argument 0..1\n"); d = strtod(av[0], NULL); if (d > 1) d = 1; else if (d < 0) d = 0; - p.fs.plr = (int)(d*0x7fffffff); + fs->plr = (int)(d*0x7fffffff); ac--; av++; break; case TOK_QUEUE: + NEED(fs, "queue is only for pipes or flowsets"); NEED1("queue needs queue size\n"); end = NULL; - p.fs.qsize = strtoul(av[0], &end, 0); + fs->qsize = strtoul(av[0], &end, 0); if (*end == 'K' || *end == 'k') { - p.fs.flags_fs |= DN_QSIZE_IS_BYTES; - p.fs.qsize *= 1024; + fs->flags |= DN_QSIZE_BYTES; + fs->qsize *= 1024; } else if (*end == 'B' || _substrcmp2(end, "by", "bytes") == 0) { - p.fs.flags_fs |= DN_QSIZE_IS_BYTES; + fs->flags |= DN_QSIZE_BYTES; } ac--; av++; break; case TOK_BUCKETS: + NEED(fs, "buckets is only for pipes or flowsets"); NEED1("buckets needs argument\n"); - p.fs.rq_size = strtoul(av[0], NULL, 0); + *buckets = strtoul(av[0], NULL, 0); ac--; av++; break; + case TOK_FLOW_MASK: + case TOK_SCHED_MASK: case TOK_MASK: + NEED(mask, "tok_mask"); NEED1("mask needs mask specifier\n"); /* * per-flow queue, mask is dst_ip, dst_port, @@ -757,7 +960,7 @@ ipfw_config_pipe(int ac, char **av) */ par = NULL; - bzero(&p.fs.flow_mask, sizeof(p.fs.flow_mask)); + bzero(mask, sizeof(*mask)); end = NULL; while (ac >= 1) { @@ -773,44 +976,55 @@ ipfw_config_pipe(int ac, char **av) case TOK_ALL: /* * special case, all bits significant + * except 'extra' (the queue number) */ - p.fs.flow_mask.dst_ip = ~0; - p.fs.flow_mask.src_ip = ~0; - p.fs.flow_mask.dst_port = ~0; - p.fs.flow_mask.src_port = ~0; - p.fs.flow_mask.proto = ~0; - n2mask(&(p.fs.flow_mask.dst_ip6), 128); - n2mask(&(p.fs.flow_mask.src_ip6), 128); - p.fs.flow_mask.flow_id6 = ~0; - p.fs.flags_fs |= DN_HAVE_FLOW_MASK; + mask->dst_ip = ~0; + mask->src_ip = ~0; + mask->dst_port = ~0; + mask->src_port = ~0; + mask->proto = ~0; + n2mask(&mask->dst_ip6, 128); + n2mask(&mask->src_ip6, 128); + mask->flow_id6 = ~0; + *flags |= DN_HAVE_MASK; + goto end_mask; + + case TOK_QUEUE: + mask->extra = ~0; + *flags |= DN_HAVE_MASK; goto end_mask; case TOK_DSTIP: - p32 = &p.fs.flow_mask.dst_ip; + mask->addr_type = 4; + p32 = &mask->dst_ip; break; case TOK_SRCIP: - p32 = &p.fs.flow_mask.src_ip; + mask->addr_type = 4; + p32 = &mask->src_ip; break; case TOK_DSTIP6: - pa6 = &(p.fs.flow_mask.dst_ip6); + mask->addr_type = 6; + pa6 = &mask->dst_ip6; break; case TOK_SRCIP6: - pa6 = &(p.fs.flow_mask.src_ip6); + mask->addr_type = 6; + pa6 = &mask->src_ip6; break; case TOK_FLOWID: - p20 = &p.fs.flow_mask.flow_id6; + mask->addr_type = 6; + p20 = &mask->flow_id6; break; case TOK_DSTPORT: - p16 = &p.fs.flow_mask.dst_port; + p16 = &mask->dst_port; break; case TOK_SRCPORT: - p16 = &p.fs.flow_mask.src_port; + p16 = &mask->src_port; break; case TOK_PROTO: @@ -850,10 +1064,10 @@ ipfw_config_pipe(int ac, char **av) if (a > 0xFF) errx(EX_DATAERR, "proto mask must be 8 bit"); - p.fs.flow_mask.proto = (uint8_t)a; + mask->proto = (uint8_t)a; } if (a != 0) - p.fs.flags_fs |= DN_HAVE_FLOW_MASK; + *flags |= DN_HAVE_MASK; ac--; av++; } /* end while, config masks */ end_mask: @@ -862,9 +1076,9 @@ end_mask: case TOK_RED: case TOK_GRED: NEED1("red/gred needs w_q/min_th/max_th/max_p\n"); - p.fs.flags_fs |= DN_IS_RED; + fs->flags |= DN_IS_RED; if (tok == TOK_GRED) - p.fs.flags_fs |= DN_IS_GENTLE_RED; + fs->flags |= DN_IS_GENTLE_RED; /* * the format for parameters is w_q/min_th/max_th/max_p */ @@ -872,96 +1086,137 @@ end_mask: double w_q = strtod(end, NULL); if (w_q > 1 || w_q <= 0) errx(EX_DATAERR, "0 < w_q <= 1"); - p.fs.w_q = (int) (w_q * (1 << SCALE_RED)); + fs->w_q = (int) (w_q * (1 << SCALE_RED)); } if ((end = strsep(&av[0], "/"))) { - p.fs.min_th = strtoul(end, &end, 0); + fs->min_th = strtoul(end, &end, 0); if (*end == 'K' || *end == 'k') - p.fs.min_th *= 1024; + fs->min_th *= 1024; } if ((end = strsep(&av[0], "/"))) { - p.fs.max_th = strtoul(end, &end, 0); + fs->max_th = strtoul(end, &end, 0); if (*end == 'K' || *end == 'k') - p.fs.max_th *= 1024; + fs->max_th *= 1024; } if ((end = strsep(&av[0], "/"))) { double max_p = strtod(end, NULL); if (max_p > 1 || max_p <= 0) errx(EX_DATAERR, "0 < max_p <= 1"); - p.fs.max_p = (int)(max_p * (1 << SCALE_RED)); + fs->max_p = (int)(max_p * (1 << SCALE_RED)); } ac--; av++; break; case TOK_DROPTAIL: - p.fs.flags_fs &= ~(DN_IS_RED|DN_IS_GENTLE_RED); + NEED(fs, "droptail is only for flowsets"); + fs->flags &= ~(DN_IS_RED|DN_IS_GENTLE_RED); break; case TOK_BW: + NEED(p, "bw is only for links"); NEED1("bw needs bandwidth or interface\n"); - if (co.do_pipe != 1) - errx(EX_DATAERR, "bandwidth only valid for pipes"); - read_bandwidth(av[0], &p.bandwidth, p.if_name, sizeof(p.if_name)); + read_bandwidth(av[0], &p->bandwidth, NULL, 0); ac--; av++; break; case TOK_DELAY: - if (co.do_pipe != 1) - errx(EX_DATAERR, "delay only valid for pipes"); + NEED(p, "delay is only for links"); NEED1("delay needs argument 0..10000ms\n"); - p.delay = strtoul(av[0], NULL, 0); + p->delay = strtoul(av[0], NULL, 0); ac--; av++; break; + case TOK_TYPE: { + int l; + NEED(sch, "type is only for schedulers"); + NEED1("type needs a string"); + l = strlen(av[0]); + if (l == 0 || l > 15) + errx(1, "type %s too long\n", av[0]); + strcpy(sch->name, av[0]); + sch->oid.subtype = 0; /* use string */ + ac--; av++; + break; + } + case TOK_WEIGHT: - if (co.do_pipe == 1) - errx(EX_DATAERR,"weight only valid for queues"); - NEED1("weight needs argument 0..100\n"); - p.fs.weight = strtoul(av[0], &end, 0); + NEED(fs, "weight is only for flowsets"); + NEED1("weight needs argument\n"); + fs->par[0] = strtol(av[0], &end, 0); + ac--; av++; + break; + + case TOK_LMAX: + NEED(fs, "lmax is only for flowsets"); + NEED1("lmax needs argument\n"); + fs->par[1] = strtol(av[0], &end, 0); + ac--; av++; + break; + + case TOK_PRI: + NEED(fs, "priority is only for flowsets"); + NEED1("priority needs argument\n"); + fs->par[2] = strtol(av[0], &end, 0); ac--; av++; break; + case TOK_SCHED: case TOK_PIPE: - if (co.do_pipe == 1) - errx(EX_DATAERR,"pipe only valid for queues"); - NEED1("pipe needs pipe_number\n"); - p.fs.parent_nr = strtoul(av[0], &end, 0); + NEED(fs, "pipe/sched"); + NEED1("pipe/link/sched needs number\n"); + fs->sched_nr = strtoul(av[0], &end, 0); ac--; av++; break; - case TOK_PIPE_PROFILE: - if (co.do_pipe != 1) - errx(EX_DATAERR, "extra delay only valid for pipes"); + case TOK_PROFILE: + { + size_t real_length; + + NEED((!pf), "profile already set"); + NEED(p, "profile"); NEED1("extra delay needs the file name\n"); - p.samples = &samples[0]; - load_extra_delays(av[0], &p); + + /* load the profile structure using the DN_API */ + pf = o_next(&buf, max_pf_size, DN_PROFILE); + load_extra_delays(av[0], pf, p); //XXX can't fail? + + /* compact the dn_id structure */ + real_length = sizeof(struct dn_profile) + + pf->samples_no * sizeof(int); + o_compact(&buf, max_pf_size, real_length, DN_PROFILE); --ac; ++av; + } + break; + + case TOK_BURST: + NEED(p, "burst"); + NEED1("burst needs argument\n"); + errno = 0; + if (expand_number(av[0], (int64_t *)&p->burst) < 0) + if (errno != ERANGE) + errx(EX_DATAERR, + "burst: invalid argument"); + if (errno || p->burst > (1ULL << 48) - 1) + errx(EX_DATAERR, + "burst: out of range (0..2^48-1)"); + ac--; av++; break; default: errx(EX_DATAERR, "unrecognised option ``%s''", av[-1]); } } - if (co.do_pipe == 1) { - if (p.pipe_nr == 0) - errx(EX_DATAERR, "pipe_nr must be > 0"); - if (p.delay > 10000) - errx(EX_DATAERR, "delay must be < 10000"); - } else { /* co.do_pipe == 2, queue */ - if (p.fs.parent_nr == 0) - errx(EX_DATAERR, "pipe must be > 0"); - if (p.fs.weight >100) - errx(EX_DATAERR, "weight must be <= 100"); - } - /* check for bandwidth value */ - if (p.bandwidth == -1) { - p.bandwidth = 0; - if (p.samples_no > 0) - errx(EX_DATAERR, "profile requires a bandwidth limit"); + /* check validity of parameters */ + if (p) { + if (p->delay > 10000) + errx(EX_DATAERR, "delay must be < 10000"); + if (p->bandwidth == -1) + p->bandwidth = 0; } - - if (p.fs.flags_fs & DN_QSIZE_IS_BYTES) { + if (fs) { + /* XXX accept a 0 scheduler to keep the default */ + if (fs->flags & DN_QSIZE_BYTES) { size_t len; long limit; @@ -969,9 +1224,9 @@ end_mask: if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit", &limit, &len, NULL, 0) == -1) limit = 1024*1024; - if (p.fs.qsize > limit) + if (fs->qsize > limit) errx(EX_DATAERR, "queue size must be < %ldB", limit); - } else { + } else { size_t len; long limit; @@ -979,27 +1234,25 @@ end_mask: if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit", &limit, &len, NULL, 0) == -1) limit = 100; - if (p.fs.qsize > limit) + if (fs->qsize > limit) errx(EX_DATAERR, "2 <= queue size <= %ld", limit); - } - if (p.fs.flags_fs & DN_IS_RED) { + } + + if (fs->flags & DN_IS_RED) { size_t len; int lookup_depth, avg_pkt_size; - double s, idle, weight, w_q; - struct clockinfo ck; - int t; + double w_q; - if (p.fs.min_th >= p.fs.max_th) + if (fs->min_th >= fs->max_th) errx(EX_DATAERR, "min_th %d must be < than max_th %d", - p.fs.min_th, p.fs.max_th); - if (p.fs.max_th == 0) + fs->min_th, fs->max_th); + if (fs->max_th == 0) errx(EX_DATAERR, "max_th must be > 0"); len = sizeof(int); if (sysctlbyname("net.inet.ip.dummynet.red_lookup_depth", &lookup_depth, &len, NULL, 0) == -1) - errx(1, "sysctlbyname(\"%s\")", - "net.inet.ip.dummynet.red_lookup_depth"); + lookup_depth = 256; if (lookup_depth == 0) errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth" " must be greater than zero"); @@ -1007,18 +1260,13 @@ end_mask: len = sizeof(int); if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size", &avg_pkt_size, &len, NULL, 0) == -1) + avg_pkt_size = 512; - errx(1, "sysctlbyname(\"%s\")", - "net.inet.ip.dummynet.red_avg_pkt_size"); if (avg_pkt_size == 0) errx(EX_DATAERR, "net.inet.ip.dummynet.red_avg_pkt_size must" " be greater than zero"); - len = sizeof(struct clockinfo); - if (sysctlbyname("kern.clockrate", &ck, &len, NULL, 0) == -1) - errx(1, "sysctlbyname(\"%s\")", "kern.clockrate"); - /* * Ticks needed for sending a medium-sized packet. * Unfortunately, when we are configuring a WF2Q+ queue, we @@ -1028,38 +1276,181 @@ end_mask: * correct. But on the other hand, why do we want RED with * WF2Q+ ? */ +#if 0 if (p.bandwidth==0) /* this is a WF2Q+ queue */ s = 0; else s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth; - +#endif /* * max idle time (in ticks) before avg queue size becomes 0. * NOTA: (3/w_q) is approx the value x so that * (1-w_q)^x < 10^-3. */ - w_q = ((double)p.fs.w_q) / (1 << SCALE_RED); + w_q = ((double)fs->w_q) / (1 << SCALE_RED); +#if 0 // go in kernel idle = s * 3. / w_q; - p.fs.lookup_step = (int)idle / lookup_depth; - if (!p.fs.lookup_step) - p.fs.lookup_step = 1; + fs->lookup_step = (int)idle / lookup_depth; + if (!fs->lookup_step) + fs->lookup_step = 1; weight = 1 - w_q; - for (t = p.fs.lookup_step; t > 1; --t) + for (t = fs->lookup_step; t > 1; --t) weight *= 1 - w_q; - p.fs.lookup_weight = (int)(weight * (1 << SCALE_RED)); + fs->lookup_weight = (int)(weight * (1 << SCALE_RED)); +#endif + } } - if (p.samples_no <= 0) { - i = do_cmd(IP_DUMMYNET_CONFIGURE, &p, sizeof p); - } else { - struct dn_pipe_max pm; - int len = sizeof(pm); - - memcpy(&pm.pipe, &p, sizeof(pm.pipe)); - memcpy(&pm.samples, samples, sizeof(pm.samples)); - i = do_cmd(IP_DUMMYNET_CONFIGURE, &pm, len); - } + i = do_cmd(IP_DUMMYNET3, base, (char *)buf - (char *)base); if (i) err(1, "setsockopt(%s)", "IP_DUMMYNET_CONFIGURE"); } + +void +dummynet_flush(void) +{ + struct dn_id oid; + oid_fill(&oid, sizeof(oid), DN_CMD_FLUSH, DN_API_VERSION); + do_cmd(IP_DUMMYNET3, &oid, oid.len); +} + +/* Parse input for 'ipfw [pipe|sched|queue] show [range list]' + * Returns the number of ranges, and possibly stores them + * in the array v of size len. + */ +static int +parse_range(int ac, char *av[], uint32_t *v, int len) +{ + int n = 0; + char *endptr, *s; + uint32_t base[2]; + + if (v == NULL || len < 2) { + v = base; + len = 2; + } + + for (s = *av; s != NULL; av++, ac--) { + v[0] = strtoul(s, &endptr, 10); + v[1] = (*endptr != '-') ? v[0] : + strtoul(endptr+1, &endptr, 10); + if (*endptr == '\0') { /* prepare for next round */ + s = (ac > 0) ? *(av+1) : NULL; + } else { + if (*endptr != ',') { + warn("invalid number: %s", s); + s = ++endptr; + continue; + } + /* continue processing from here */ + s = ++endptr; + ac++; + av--; + } + if (v[1] < v[0] || + v[1] < 0 || v[1] >= DN_MAX_ID-1 || + v[0] < 0 || v[1] >= DN_MAX_ID-1) { + continue; /* invalid entry */ + } + n++; + /* translate if 'pipe list' */ + if (co.do_pipe == 1) { + v[0] += DN_MAX_ID; + v[1] += DN_MAX_ID; + } + v = (n*2 < len) ? v + 2 : base; + } + return n; +} + +/* main entry point for dummynet list functions. co.do_pipe indicates + * which function we want to support. + * av may contain filtering arguments, either individual entries + * or ranges, or lists (space or commas are valid separators). + * Format for a range can be n1-n2 or n3 n4 n5 ... + * In a range n1 must be <= n2, otherwise the range is ignored. + * A number 'n4' is translate in a range 'n4-n4' + * All number must be > 0 and < DN_MAX_ID-1 + */ +void +dummynet_list(int ac, char *av[], int show_counters) +{ + struct dn_id *oid, *x = NULL; + int ret, i, l; + int n; /* # of ranges */ + int buflen; + int max_size; /* largest obj passed up */ + + ac--; + av++; /* skip 'list' | 'show' word */ + + n = parse_range(ac, av, NULL, 0); /* Count # of ranges. */ + + /* Allocate space to store ranges */ + l = sizeof(*oid) + sizeof(uint32_t) * n * 2; + oid = safe_calloc(1, l); + oid_fill(oid, l, DN_CMD_GET, DN_API_VERSION); + + if (n > 0) /* store ranges in idx */ + parse_range(ac, av, (uint32_t *)(oid + 1), n*2); + /* + * Compute the size of the largest object returned. If the + * response leaves at least this much spare space in the + * buffer, then surely the response is complete; otherwise + * there might be a risk of truncation and we will need to + * retry with a larger buffer. + * XXX don't bother with smaller structs. + */ + max_size = sizeof(struct dn_fs); + if (max_size < sizeof(struct dn_sch)) + max_size = sizeof(struct dn_sch); + if (max_size < sizeof(struct dn_flow)) + max_size = sizeof(struct dn_flow); + + switch (co.do_pipe) { + case 1: + oid->subtype = DN_LINK; /* list pipe */ + break; + case 2: + oid->subtype = DN_FS; /* list queue */ + break; + case 3: + oid->subtype = DN_SCH; /* list sched */ + break; + } + + /* + * Ask the kernel an estimate of the required space (result + * in oid.id), unless we are requesting a subset of objects, + * in which case the kernel does not give an exact answer. + * In any case, space might grow in the meantime due to the + * creation of new queues, so we must be prepared to retry. + */ + if (n > 0) { + buflen = 4*1024; + } else { + ret = do_cmd(-IP_DUMMYNET3, oid, (uintptr_t)&l); + if (ret != 0 || oid->id <= sizeof(*oid)) + goto done; + buflen = oid->id + max_size; + oid->len = sizeof(*oid); /* restore */ + } + /* Try a few times, until the buffer fits */ + for (i = 0; i < 20; i++) { + l = buflen; + x = safe_realloc(x, l); + bcopy(oid, x, oid->len); + ret = do_cmd(-IP_DUMMYNET3, x, (uintptr_t)&l); + if (ret != 0 || x->id <= sizeof(*oid)) + goto done; /* no response */ + if (l + max_size <= buflen) + break; /* ok */ + buflen *= 2; /* double for next attempt */ + } + list_pipes(x, O_NEXT(x, l)); +done: + if (x) + free(x); + free(oid); +}