Index: stable/10/sbin/ipfw/dummynet.c =================================================================== --- stable/10/sbin/ipfw/dummynet.c (revision 301230) +++ stable/10/sbin/ipfw/dummynet.c (revision 301231) @@ -1,1391 +1,1405 @@ /* * 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. * * Redistribution in binary form may occur without any restrictions. * Obviously, it would be nice if you gave credit where credit is due * but requiring it would be too onerous. * * This software is provided ``AS IS'' without any warranties of any kind. * * $FreeBSD$ * * dummynet support */ #include #include /* XXX there are several sysctl leftover here */ #include #include "ipfw2.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* inet_ntoa */ static struct _s_x dummynet_params[] = { { "plr", TOK_PLR }, { "noerror", TOK_NOERROR }, { "buckets", TOK_BUCKETS }, { "dst-ip", TOK_DSTIP }, { "src-ip", TOK_SRCIP }, { "dst-port", TOK_DSTPORT }, { "src-port", TOK_SRCPORT }, { "proto", TOK_PROTO }, { "weight", TOK_WEIGHT }, { "lmax", TOK_LMAX }, { "maxlen", TOK_LMAX }, { "all", TOK_ALL }, { "mask", TOK_MASK }, /* alias for both */ { "sched_mask", TOK_SCHED_MASK }, { "flow_mask", TOK_FLOW_MASK }, { "droptail", TOK_DROPTAIL }, + { "ecn", TOK_ECN }, { "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_PROFILE}, { "burst", TOK_BURST}, { "dummynet-params", TOK_NULL }, { NULL, 0 } /* terminator */ }; #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; } #if 0 static int 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; long long res = 0; const struct dn_flow_queue *a = pa; const struct dn_flow_queue *b = pb; switch (field) { case 1: /* pkts */ res = a->len - b->len; break; case 2: /* bytes */ res = a->len_bytes - b->len_bytes; break; case 3: /* tot pkts */ res = a->tot_pkts - b->tot_pkts; break; case 4: /* tot bytes */ res = a->tot_bytes - b->tot_bytes; break; } if (res < 0) res = -1; if (res > 0) res = 1; return (int)(rev ? res : -res); } #endif /* print a mask and header for the subsequent list of flows */ static void print_mask(struct ipfw_flow_id *id) { 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); } } 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"); } 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 */ 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 ", id->proto); ina.s_addr = htonl(id->src_ip); printf("%15s/%-5d ", inet_ntoa(ina), id->src_port); ina.s_addr = htonl(id->dst_ip); printf("%15s/%-5d ", inet_ntoa(ina), id->dst_port); } else { /* Print IPv6 flows */ if (pe != NULL) printf("%9s ", pe->p_name); else 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, &(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_fs *fs, char *prefix) { int l; char qs[30]; char plr[30]; char red[90]; /* Display RED parameters */ l = fs->qsize; if (fs->flags & DN_QSIZE_BYTES) { if (l >= 8192) sprintf(qs, "%d KB", l / 1024); else sprintf(qs, "%d B", l); } else sprintf(qs, "%3d sl.", l); if (fs->plr) sprintf(plr, "plr %f", 1.0 * fs->plr / (double)(0x7fffffff)); else plr[0] = '\0'; - if (fs->flags & 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 & DN_IS_GENTLE_RED) ? 'G' : ' ', 1.0 * fs->w_q / (double)(1 << SCALE_RED), fs->min_th, fs->max_th, 1.0 * fs->max_p / (double)(1 << SCALE_RED)); - else + if (fs->flags & DN_IS_ECN) + strncat(red, " (ecn)", 6); + } else sprintf(red, "droptail"); 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_profile *p) { double loss; if (p->samples_no <= 0) return; loss = p->loss_level; loss /= p->samples_no; printf("\t profile: name \"%s\" loss %f samples %d\n", p->name, loss, p->samples_no); } static void flush_buf(char *buf) { 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, queue or scheduler i */ int ipfw_delete_pipe(int do_pipe, int i) { 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); } return i; } /* * Code to parse delay profiles. * * Some link types introduce extra delays in the transmission * of a packet, e.g. because of MAC level framing, contention on * the use of the channel, MAC level retransmissions and so on. * From our point of view, the channel is effectively unavailable * for this extra time, which is constant or variable depending * on the link type. Additionally, packets may be dropped after this * time (e.g. on a wireless link after too many retransmissions). * We can model the additional delay with an empirical curve * that represents its distribution. * * cumulative probability * 1.0 ^ * | * L +-- loss-level x * | ****** * | * * | ***** * | * * | ** * | * * +-------*-------------------> * delay * * 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 link will use the largest delay * for a given probability. * * To pass the curve to dummynet, we must store the parameters * in a file as described below, and issue the command * * ipfw pipe config ... bw XXX profile ... * * The file format is the following, with whitespace acting as * a separator and '#' indicating the beginning a comment: * * samples N * the number of samples used in the internal * representation (2..1024; default 100); * * loss-level L * The probability above which packets are lost. * (0.0 <= L <= 1.0, default 1.0 i.e. no loss); * * name identifier * Optional a name (listed by "ipfw pipe show") * to identify the distribution; * * "delay prob" | "prob delay" * One of these two lines is mandatory and defines * the format of the following lines with data points. * * XXX YYY * 2 or more lines representing points in the curve, * with either delay or probability first, according * to the chosen format. * The unit for delay is milliseconds. * * Data points does not need to be ordered or equal to the number * specified in the "samples" line. ipfw will sort and interpolate * the curve as needed. * * Example of a profile file: name bla_bla_bla samples 100 loss-level 0.86 prob delay 0 200 # minimum overhead is 200ms 0.5 200 0.5 300 0.8 1000 0.9 1300 1 1300 * Internally, we will convert the curve to a fixed number of * samples, and when it is time to transmit a packet we will * model the extra delay as extra bits in the packet. * */ #define ED_MAX_LINE_LEN 256+ED_MAX_NAME_LEN #define ED_TOK_SAMPLES "samples" #define ED_TOK_LOSS "loss-level" #define ED_TOK_NAME "name" #define ED_TOK_DELAY "delay" #define ED_TOK_PROB "prob" #define ED_TOK_BW "bw" #define ED_SEPARATORS " \t\n" #define ED_MIN_SAMPLES_NO 2 /* * returns 1 if s is a non-negative number, with at least one '.' */ static int is_valid_number(const char *s) { int i, dots_found = 0; int len = strlen(s); for (i = 0; i 1)) return 0; return 1; } /* * Take as input a string describing a bandwidth value * and return the numeric bandwidth value. * set clocking interface or bandwidth value */ static void read_bandwidth(char *arg, int *bandwidth, char *if_name, int namelen) { if (*bandwidth != -1) 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--; /* interface name */ strncpy(if_name, arg, namelen); if_name[namelen] = '\0'; *bandwidth = 0; } else { /* read bandwidth value */ int bw; char *end = NULL; bw = strtoul(arg, &end, 0); if (*end == 'K' || *end == 'k') { end++; bw *= 1000; } else if (*end == 'M' || *end == 'm') { end++; bw *= 1000000; } if ((*end == 'B' && _substrcmp2(end, "Bi", "Bit/s") != 0) || _substrcmp2(end, "by", "bytes") == 0) bw *= 8; if (bw < 0) errx(EX_DATAERR, "bandwidth too large"); *bandwidth = bw; if (if_name) if_name[0] = '\0'; } } struct point { double prob; double delay; }; static int compare_points(const void *vp1, const void *vp2) { const struct point *p1 = vp1; const struct point *p2 = vp2; double res = 0; res = p1->prob - p2->prob; if (res == 0) res = p1->delay - p2->delay; if (res < 0) return -1; else if (res > 0) return 1; else return 0; } #define ED_EFMT(s) EX_DATAERR,"error in %s at line %d: "#s,filename,lineno static void 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; char profile_name[ED_MAX_NAME_LEN]; int delay_first = -1; int do_points = 0; 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); while (fgets(line, ED_MAX_LINE_LEN, f)) { /* read commands */ char *s, *cur = line, *name = NULL, *arg = NULL; ++lineno; /* parse the line */ while (cur) { s = strsep(&cur, ED_SEPARATORS); if (s == NULL || *s == '#') break; if (*s == '\0') continue; if (arg) errx(ED_EFMT("too many arguments")); if (name == NULL) name = s; else arg = s; } if (name == NULL) /* empty line */ continue; if (arg == NULL) errx(ED_EFMT("missing arg for %s"), name); if (!strcasecmp(name, ED_TOK_SAMPLES)) { if (samples > 0) errx(ED_EFMT("duplicate ``samples'' line")); if (atoi(arg) <=0) errx(ED_EFMT("invalid number of samples")); samples = atoi(arg); if (samples>ED_MAX_SAMPLES_NO) errx(ED_EFMT("too many samples, maximum is %d"), ED_MAX_SAMPLES_NO); do_points = 0; } else if (!strcasecmp(name, ED_TOK_BW)) { char buf[IFNAMSIZ]; read_bandwidth(arg, &link->bandwidth, buf, sizeof(buf)); } else if (!strcasecmp(name, ED_TOK_LOSS)) { if (loss != -1.0) errx(ED_EFMT("duplicated token: %s"), name); if (!is_valid_number(arg)) errx(ED_EFMT("invalid %s"), arg); loss = atof(arg); if (loss > 1) errx(ED_EFMT("%s greater than 1.0"), name); do_points = 0; } else if (!strcasecmp(name, ED_TOK_NAME)) { if (profile_name[0] != '\0') errx(ED_EFMT("duplicated token: %s"), name); strncpy(profile_name, arg, sizeof(profile_name) - 1); profile_name[sizeof(profile_name)-1] = '\0'; do_points = 0; } else if (!strcasecmp(name, ED_TOK_DELAY)) { if (do_points) errx(ED_EFMT("duplicated token: %s"), name); delay_first = 1; do_points = 1; } else if (!strcasecmp(name, ED_TOK_PROB)) { if (do_points) errx(ED_EFMT("duplicated token: %s"), name); delay_first = 0; do_points = 1; } else if (do_points) { if (!is_valid_number(name) || !is_valid_number(arg)) errx(ED_EFMT("invalid point found")); if (delay_first) { points[points_no].delay = atof(name); points[points_no].prob = atof(arg); } else { points[points_no].delay = atof(arg); points[points_no].prob = atof(name); } if (points[points_no].prob > 1.0) errx(ED_EFMT("probability greater than 1.0")); ++points_no; } else { errx(ED_EFMT("unrecognised command '%s'"), name); } } fclose (f); if (samples == -1) { warnx("'%s' not found, assuming 100", ED_TOK_SAMPLES); samples = 100; } if (loss == -1.0) { warnx("'%s' not found, assuming no loss", ED_TOK_LOSS); 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); /* interpolation */ for (i = 0; isamples[ix] = x1; } else { double m = (y2-y1)/(x2-x1); double c = y1 - m*x1; for (; ixsamples[ix] = (ix - c)/m; } } 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 i; u_int j; char *end; 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; /* * 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) + sizeof(struct dn_profile); av++; ac--; /* Pipe number */ if (ac && isdigit(**av)) { i = atoi(*av); av++; ac--; } 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); ac--; av++; switch(tok) { case TOK_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; 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; fs->qsize = strtoul(av[0], &end, 0); if (*end == 'K' || *end == 'k') { fs->flags |= DN_QSIZE_BYTES; fs->qsize *= 1024; } else if (*end == 'B' || _substrcmp2(end, "by", "bytes") == 0) { 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"); *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, * src_ip, src_port, proto measured in bits */ bzero(mask, sizeof(*mask)); end = NULL; while (ac >= 1) { uint32_t *p32 = NULL; uint16_t *p16 = NULL; uint32_t *p20 = NULL; struct in6_addr *pa6 = NULL; uint32_t a; tok = match_token(dummynet_params, *av); ac--; av++; switch(tok) { case TOK_ALL: /* * special case, all bits significant * except 'extra' (the queue number) */ 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: mask->addr_type = 4; p32 = &mask->dst_ip; break; case TOK_SRCIP: mask->addr_type = 4; p32 = &mask->src_ip; break; case TOK_DSTIP6: mask->addr_type = 6; pa6 = &mask->dst_ip6; break; case TOK_SRCIP6: mask->addr_type = 6; pa6 = &mask->src_ip6; break; case TOK_FLOWID: mask->addr_type = 6; p20 = &mask->flow_id6; break; case TOK_DSTPORT: p16 = &mask->dst_port; break; case TOK_SRCPORT: p16 = &mask->src_port; break; case TOK_PROTO: break; default: ac++; av--; /* backtrack */ goto end_mask; } if (ac < 1) errx(EX_USAGE, "mask: value missing"); if (*av[0] == '/') { a = strtoul(av[0]+1, &end, 0); if (pa6 == NULL) a = (a == 32) ? ~0 : (1 << a) - 1; } else a = strtoul(av[0], &end, 0); if (p32 != NULL) *p32 = a; else if (p16 != NULL) { if (a > 0xFFFF) errx(EX_DATAERR, "port mask must be 16 bit"); *p16 = (uint16_t)a; } else if (p20 != NULL) { if (a > 0xfffff) errx(EX_DATAERR, "flow_id mask must be 20 bit"); *p20 = (uint32_t)a; } else if (pa6 != NULL) { if (a > 128) errx(EX_DATAERR, "in6addr invalid mask len"); else n2mask(pa6, a); } else { if (a > 0xFF) errx(EX_DATAERR, "proto mask must be 8 bit"); mask->proto = (uint8_t)a; } if (a != 0) *flags |= DN_HAVE_MASK; ac--; av++; } /* end while, config masks */ end_mask: break; case TOK_RED: case TOK_GRED: NEED1("red/gred needs w_q/min_th/max_th/max_p\n"); fs->flags |= DN_IS_RED; if (tok == TOK_GRED) fs->flags |= DN_IS_GENTLE_RED; /* * the format for parameters is w_q/min_th/max_th/max_p */ if ((end = strsep(&av[0], "/"))) { double w_q = strtod(end, NULL); if (w_q > 1 || w_q <= 0) errx(EX_DATAERR, "0 < w_q <= 1"); fs->w_q = (int) (w_q * (1 << SCALE_RED)); } if ((end = strsep(&av[0], "/"))) { fs->min_th = strtoul(end, &end, 0); if (*end == 'K' || *end == 'k') fs->min_th *= 1024; } if ((end = strsep(&av[0], "/"))) { fs->max_th = strtoul(end, &end, 0); if (*end == 'K' || *end == 'k') 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"); + if (max_p > 1 || max_p < 0) + errx(EX_DATAERR, "0 <= max_p <= 1"); fs->max_p = (int)(max_p * (1 << SCALE_RED)); } ac--; av++; break; + case TOK_ECN: + fs->flags |= DN_IS_ECN; + break; + case TOK_DROPTAIL: 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"); read_bandwidth(av[0], &p->bandwidth, NULL, 0); ac--; av++; break; case TOK_DELAY: NEED(p, "delay is only for links"); NEED1("delay needs argument 0..10000ms\n"); 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: 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: NEED(fs, "pipe/sched"); NEED1("pipe/link/sched needs number\n"); fs->sched_nr = strtoul(av[0], &end, 0); ac--; av++; break; case TOK_PROFILE: NEED((!pf), "profile already set"); NEED(p, "profile"); { NEED1("extra delay needs the file name\n"); pf = o_next(&buf, sizeof(*pf), DN_PROFILE); load_extra_delays(av[0], pf, p); //XXX can't fail? --ac; ++av; } break; case TOK_BURST: NEED(p, "burst"); NEED1("burst needs argument\n"); errno = 0; if (expand_number(av[0], &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]); } } /* 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 (fs) { /* XXX accept a 0 scheduler to keep the default */ if (fs->flags & DN_QSIZE_BYTES) { size_t len; long limit; len = sizeof(limit); if (sysctlbyname("net.inet.ip.dummynet.pipe_byte_limit", &limit, &len, NULL, 0) == -1) limit = 1024*1024; if (fs->qsize > limit) errx(EX_DATAERR, "queue size must be < %ldB", limit); } else { size_t len; long limit; len = sizeof(limit); if (sysctlbyname("net.inet.ip.dummynet.pipe_slot_limit", &limit, &len, NULL, 0) == -1) limit = 100; if (fs->qsize > limit) errx(EX_DATAERR, "2 <= queue size <= %ld", limit); } + if ((fs->flags & DN_IS_ECN) && !(fs->flags & DN_IS_RED)) + errx(EX_USAGE, "enable red/gred for ECN"); + if (fs->flags & DN_IS_RED) { size_t len; int lookup_depth, avg_pkt_size; - if (fs->min_th >= fs->max_th) + if (!(fs->flags & DN_IS_ECN) && (fs->min_th >= fs->max_th)) errx(EX_DATAERR, "min_th %d must be < than max_th %d", fs->min_th, fs->max_th); + else if ((fs->flags & DN_IS_ECN) && (fs->min_th > fs->max_th)) + errx(EX_DATAERR, "min_th %d must be =< than max_th %d", + 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) lookup_depth = 256; if (lookup_depth == 0) errx(EX_DATAERR, "net.inet.ip.dummynet.red_lookup_depth" " must be greater than zero"); len = sizeof(int); if (sysctlbyname("net.inet.ip.dummynet.red_avg_pkt_size", &avg_pkt_size, &len, NULL, 0) == -1) avg_pkt_size = 512; if (avg_pkt_size == 0) errx(EX_DATAERR, "net.inet.ip.dummynet.red_avg_pkt_size must" " be greater than zero"); #if 0 /* the following computation is now done in the kernel */ /* * Ticks needed for sending a medium-sized packet. * Unfortunately, when we are configuring a WF2Q+ queue, we * do not have bandwidth information, because that is stored * in the parent pipe, and also we have multiple queues * competing for it. So we set s=0, which is not very * correct. But on the other hand, why do we want RED with * WF2Q+ ? */ if (p.bandwidth==0) /* this is a WF2Q+ queue */ s = 0; else s = (double)ck.hz * avg_pkt_size * 8 / p.bandwidth; /* * 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)fs->w_q) / (1 << SCALE_RED); idle = s * 3. / w_q; fs->lookup_step = (int)idle / lookup_depth; if (!fs->lookup_step) fs->lookup_step = 1; weight = 1 - w_q; for (t = fs->lookup_step; t > 1; --t) weight *= 1 - w_q; fs->lookup_weight = (int)(weight * (1 << SCALE_RED)); #endif /* code moved in the kernel */ } } 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] >= DN_MAX_ID-1 || 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; int n; /* # of ranges */ u_int buflen, l; u_int max_size; /* largest obj passed up */ (void)show_counters; // XXX unused, but we should use it. 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); } Index: stable/10/sbin/ipfw/ipfw.8 =================================================================== --- stable/10/sbin/ipfw/ipfw.8 (revision 301230) +++ stable/10/sbin/ipfw/ipfw.8 (revision 301231) @@ -1,3472 +1,3473 @@ .\" .\" $FreeBSD$ .\" -.Dd October 25, 2012 +.Dd May 31, 2014 .Dt IPFW 8 .Os .Sh NAME .Nm ipfw .Nd User interface for firewall, traffic shaper, packet scheduler, in-kernel NAT. .Sh SYNOPSIS .Ss FIREWALL CONFIGURATION .Nm .Op Fl cq .Cm add .Ar rule .Nm .Op Fl acdefnNStT .Op Cm set Ar N .Brq Cm list | show .Op Ar rule | first-last ... .Nm .Op Fl f | q .Op Cm set Ar N .Cm flush .Nm .Op Fl q .Op Cm set Ar N .Brq Cm delete | zero | resetlog .Op Ar number ... .Pp .Nm .Cm set Oo Cm disable Ar number ... Oc Op Cm enable Ar number ... .Nm .Cm set move .Op Cm rule .Ar number Cm to Ar number .Nm .Cm set swap Ar number number .Nm .Cm set show .Ss SYSCTL SHORTCUTS .Nm .Cm enable .Brq Cm firewall | altq | one_pass | debug | verbose | dyn_keepalive .Nm .Cm disable .Brq Cm firewall | altq | one_pass | debug | verbose | dyn_keepalive .Ss LOOKUP TABLES .Nm .Cm table Ar number Cm add Ar addr Ns Oo / Ns Ar masklen Oc Op Ar value .Nm .Cm table Ar number Cm delete Ar addr Ns Op / Ns Ar masklen .Nm .Cm table .Brq Ar number | all .Cm flush .Nm .Cm table .Brq Ar number | all .Cm list .Ss DUMMYNET CONFIGURATION (TRAFFIC SHAPER AND PACKET SCHEDULER) .Nm .Brq Cm pipe | queue | sched .Ar number .Cm config .Ar config-options .Nm .Op Fl s Op Ar field .Brq Cm pipe | queue | sched .Brq Cm delete | list | show .Op Ar number ... .Ss IN-KERNEL NAT .Nm .Op Fl q .Cm nat .Ar number .Cm config .Ar config-options .Pp .Nm .Op Fl cfnNqS .Oo .Fl p Ar preproc .Oo .Ar preproc-flags .Oc .Oc .Ar pathname .Sh DESCRIPTION The .Nm utility is the user interface for controlling the .Xr ipfw 4 firewall, the .Xr dummynet 4 traffic shaper/packet scheduler, and the in-kernel NAT services. .Pp A firewall configuration, or .Em ruleset , is made of a list of .Em rules numbered from 1 to 65535. Packets are passed to the firewall from a number of different places in the protocol stack (depending on the source and destination of the packet, it is possible for the firewall to be invoked multiple times on the same packet). The packet passed to the firewall is compared against each of the rules in the .Em ruleset , in rule-number order (multiple rules with the same number are permitted, in which case they are processed in order of insertion). When a match is found, the action corresponding to the matching rule is performed. .Pp Depending on the action and certain system settings, packets can be reinjected into the firewall at some rule after the matching one for further processing. .Pp A ruleset always includes a .Em default rule (numbered 65535) which cannot be modified or deleted, and matches all packets. The action associated with the .Em default rule can be either .Cm deny or .Cm allow depending on how the kernel is configured. .Pp If the ruleset includes one or more rules with the .Cm keep-state or .Cm limit option, the firewall will have a .Em stateful behaviour, i.e., upon a match it will create .Em dynamic rules , i.e., rules that match packets with the same 5-tuple (protocol, source and destination addresses and ports) as the packet which caused their creation. Dynamic rules, which have a limited lifetime, are checked at the first occurrence of a .Cm check-state , .Cm keep-state or .Cm limit rule, and are typically used to open the firewall on-demand to legitimate traffic only. See the .Sx STATEFUL FIREWALL and .Sx EXAMPLES Sections below for more information on the stateful behaviour of .Nm . .Pp All rules (including dynamic ones) have a few associated counters: a packet count, a byte count, a log count and a timestamp indicating the time of the last match. Counters can be displayed or reset with .Nm commands. .Pp Each rule belongs to one of 32 different .Em sets , and there are .Nm commands to atomically manipulate sets, such as enable, disable, swap sets, move all rules in a set to another one, delete all rules in a set. These can be useful to install temporary configurations, or to test them. See Section .Sx SETS OF RULES for more information on .Em sets . .Pp Rules can be added with the .Cm add command; deleted individually or in groups with the .Cm delete command, and globally (except those in set 31) with the .Cm flush command; displayed, optionally with the content of the counters, using the .Cm show and .Cm list commands. Finally, counters can be reset with the .Cm zero and .Cm resetlog commands. .Pp .Ss COMMAND OPTIONS The following general options are available when invoking .Nm : .Bl -tag -width indent .It Fl a Show counter values when listing rules. The .Cm show command implies this option. .It Fl b Only show the action and the comment, not the body of a rule. Implies .Fl c . .It Fl c When entering or showing rules, print them in compact form, i.e., omitting the "ip from any to any" string when this does not carry any additional information. .It Fl d When listing, show dynamic rules in addition to static ones. .It Fl e When listing and .Fl d is specified, also show expired dynamic rules. .It Fl f Do not ask for confirmation for commands that can cause problems if misused, i.e., .Cm flush . If there is no tty associated with the process, this is implied. .It Fl i When listing a table (see the .Sx LOOKUP TABLES section below for more information on lookup tables), format values as IP addresses. By default, values are shown as integers. .It Fl n Only check syntax of the command strings, without actually passing them to the kernel. .It Fl N Try to resolve addresses and service names in output. .It Fl q Be quiet when executing the .Cm add , .Cm nat , .Cm zero , .Cm resetlog or .Cm flush commands; (implies .Fl f ) . This is useful when updating rulesets by executing multiple .Nm commands in a script (e.g., .Ql sh\ /etc/rc.firewall ) , or by processing a file with many .Nm rules across a remote login session. It also stops a table add or delete from failing if the entry already exists or is not present. .Pp The reason why this option may be important is that for some of these actions, .Nm may print a message; if the action results in blocking the traffic to the remote client, the remote login session will be closed and the rest of the ruleset will not be processed. Access to the console would then be required to recover. .It Fl S When listing rules, show the .Em set each rule belongs to. If this flag is not specified, disabled rules will not be listed. .It Fl s Op Ar field When listing pipes, sort according to one of the four counters (total or current packets or bytes). .It Fl t When listing, show last match timestamp converted with ctime(). .It Fl T When listing, show last match timestamp as seconds from the epoch. This form can be more convenient for postprocessing by scripts. .El .Ss LIST OF RULES AND PREPROCESSING To ease configuration, rules can be put into a file which is processed using .Nm as shown in the last synopsis line. An absolute .Ar pathname must be used. The file will be read line by line and applied as arguments to the .Nm utility. .Pp Optionally, a preprocessor can be specified using .Fl p Ar preproc where .Ar pathname is to be piped through. Useful preprocessors include .Xr cpp 1 and .Xr m4 1 . If .Ar preproc does not start with a slash .Pq Ql / as its first character, the usual .Ev PATH name search is performed. Care should be taken with this in environments where not all file systems are mounted (yet) by the time .Nm is being run (e.g.\& when they are mounted over NFS). Once .Fl p has been specified, any additional arguments are passed on to the preprocessor for interpretation. This allows for flexible configuration files (like conditionalizing them on the local hostname) and the use of macros to centralize frequently required arguments like IP addresses. .Ss TRAFFIC SHAPER CONFIGURATION The .Nm .Cm pipe , queue and .Cm sched commands are used to configure the traffic shaper and packet scheduler. See the .Sx TRAFFIC SHAPER (DUMMYNET) CONFIGURATION Section below for details. .Pp If the world and the kernel get out of sync the .Nm ABI may break, preventing you from being able to add any rules. This can adversely effect the booting process. You can use .Nm .Cm disable .Cm firewall to temporarily disable the firewall to regain access to the network, allowing you to fix the problem. .Sh PACKET FLOW A packet is checked against the active ruleset in multiple places in the protocol stack, under control of several sysctl variables. These places and variables are shown below, and it is important to have this picture in mind in order to design a correct ruleset. .Bd -literal -offset indent ^ to upper layers V | | +----------->-----------+ ^ V [ip(6)_input] [ip(6)_output] net.inet(6).ip(6).fw.enable=1 | | ^ V [ether_demux] [ether_output_frame] net.link.ether.ipfw=1 | | +-->--[bdg_forward]-->--+ net.link.bridge.ipfw=1 ^ V | to devices | .Ed .Pp The number of times the same packet goes through the firewall can vary between 0 and 4 depending on packet source and destination, and system configuration. .Pp Note that as packets flow through the stack, headers can be stripped or added to it, and so they may or may not be available for inspection. E.g., incoming packets will include the MAC header when .Nm is invoked from .Cm ether_demux() , but the same packets will have the MAC header stripped off when .Nm is invoked from .Cm ip_input() or .Cm ip6_input() . .Pp Also note that each packet is always checked against the complete ruleset, irrespective of the place where the check occurs, or the source of the packet. If a rule contains some match patterns or actions which are not valid for the place of invocation (e.g.\& trying to match a MAC header within .Cm ip_input or .Cm ip6_input ), the match pattern will not match, but a .Cm not operator in front of such patterns .Em will cause the pattern to .Em always match on those packets. It is thus the responsibility of the programmer, if necessary, to write a suitable ruleset to differentiate among the possible places. .Cm skipto rules can be useful here, as an example: .Bd -literal -offset indent # packets from ether_demux or bdg_forward ipfw add 10 skipto 1000 all from any to any layer2 in # packets from ip_input ipfw add 10 skipto 2000 all from any to any not layer2 in # packets from ip_output ipfw add 10 skipto 3000 all from any to any not layer2 out # packets from ether_output_frame ipfw add 10 skipto 4000 all from any to any layer2 out .Ed .Pp (yes, at the moment there is no way to differentiate between ether_demux and bdg_forward). .Sh SYNTAX In general, each keyword or argument must be provided as a separate command line argument, with no leading or trailing spaces. Keywords are case-sensitive, whereas arguments may or may not be case-sensitive depending on their nature (e.g.\& uid's are, hostnames are not). .Pp Some arguments (e.g., port or address lists) are comma-separated lists of values. In this case, spaces after commas ',' are allowed to make the line more readable. You can also put the entire command (including flags) into a single argument. E.g., the following forms are equivalent: .Bd -literal -offset indent ipfw -q add deny src-ip 10.0.0.0/24,127.0.0.1/8 ipfw -q add deny src-ip 10.0.0.0/24, 127.0.0.1/8 ipfw "-q add deny src-ip 10.0.0.0/24, 127.0.0.1/8" .Ed .Sh RULE FORMAT The format of firewall rules is the following: .Bd -ragged -offset indent .Bk -words .Op Ar rule_number .Op Cm set Ar set_number .Op Cm prob Ar match_probability .Ar action .Op Cm log Op Cm logamount Ar number .Op Cm altq Ar queue .Oo .Bro Cm tag | untag .Brc Ar number .Oc .Ar body .Ek .Ed .Pp where the body of the rule specifies which information is used for filtering packets, among the following: .Pp .Bl -tag -width "Source and dest. addresses and ports" -offset XXX -compact .It Layer-2 header fields When available .It IPv4 and IPv6 Protocol TCP, UDP, ICMP, etc. .It Source and dest. addresses and ports .It Direction See Section .Sx PACKET FLOW .It Transmit and receive interface By name or address .It Misc. IP header fields Version, type of service, datagram length, identification, fragment flag (non-zero IP offset), Time To Live .It IP options .It IPv6 Extension headers Fragmentation, Hop-by-Hop options, Routing Headers, Source routing rthdr0, Mobile IPv6 rthdr2, IPSec options. .It IPv6 Flow-ID .It Misc. TCP header fields TCP flags (SYN, FIN, ACK, RST, etc.), sequence number, acknowledgment number, window .It TCP options .It ICMP types for ICMP packets .It ICMP6 types for ICMP6 packets .It User/group ID When the packet can be associated with a local socket. .It Divert status Whether a packet came from a divert socket (e.g., .Xr natd 8 ) . .It Fib annotation state Whether a packet has been tagged for using a specific FIB (routing table) in future forwarding decisions. .El .Pp Note that some of the above information, e.g.\& source MAC or IP addresses and TCP/UDP ports, can be easily spoofed, so filtering on those fields alone might not guarantee the desired results. .Bl -tag -width indent .It Ar rule_number Each rule is associated with a .Ar rule_number in the range 1..65535, with the latter reserved for the .Em default rule. Rules are checked sequentially by rule number. Multiple rules can have the same number, in which case they are checked (and listed) according to the order in which they have been added. If a rule is entered without specifying a number, the kernel will assign one in such a way that the rule becomes the last one before the .Em default rule. Automatic rule numbers are assigned by incrementing the last non-default rule number by the value of the sysctl variable .Ar net.inet.ip.fw.autoinc_step which defaults to 100. If this is not possible (e.g.\& because we would go beyond the maximum allowed rule number), the number of the last non-default value is used instead. .It Cm set Ar set_number Each rule is associated with a .Ar set_number in the range 0..31. Sets can be individually disabled and enabled, so this parameter is of fundamental importance for atomic ruleset manipulation. It can be also used to simplify deletion of groups of rules. If a rule is entered without specifying a set number, set 0 will be used. .br Set 31 is special in that it cannot be disabled, and rules in set 31 are not deleted by the .Nm ipfw flush command (but you can delete them with the .Nm ipfw delete set 31 command). Set 31 is also used for the .Em default rule. .It Cm prob Ar match_probability A match is only declared with the specified probability (floating point number between 0 and 1). This can be useful for a number of applications such as random packet drop or (in conjunction with .Nm dummynet ) to simulate the effect of multiple paths leading to out-of-order packet delivery. .Pp Note: this condition is checked before any other condition, including ones such as keep-state or check-state which might have side effects. .It Cm log Op Cm logamount Ar number Packets matching a rule with the .Cm log keyword will be made available for logging in two ways: if the sysctl variable .Va net.inet.ip.fw.verbose is set to 0 (default), one can use .Xr bpf 4 attached to the .Li ipfw0 pseudo interface. This pseudo interface can be created after a boot manually by using the following command: .Bd -literal -offset indent # ifconfig ipfw0 create .Ed .Pp Or, automatically at boot time by adding the following line to the .Xr rc.conf 5 file: .Bd -literal -offset indent firewall_logif="YES" .Ed .Pp There is no overhead if no .Xr bpf 4 is attached to the pseudo interface. .Pp If .Va net.inet.ip.fw.verbose is set to 1, packets will be logged to .Xr syslogd 8 with a .Dv LOG_SECURITY facility up to a maximum of .Cm logamount packets. If no .Cm logamount is specified, the limit is taken from the sysctl variable .Va net.inet.ip.fw.verbose_limit . In both cases, a value of 0 means unlimited logging. .Pp Once the limit is reached, logging can be re-enabled by clearing the logging counter or the packet counter for that entry, see the .Cm resetlog command. .Pp Note: logging is done after all other packet matching conditions have been successfully verified, and before performing the final action (accept, deny, etc.) on the packet. .It Cm tag Ar number When a packet matches a rule with the .Cm tag keyword, the numeric tag for the given .Ar number in the range 1..65534 will be attached to the packet. The tag acts as an internal marker (it is not sent out over the wire) that can be used to identify these packets later on. This can be used, for example, to provide trust between interfaces and to start doing policy-based filtering. A packet can have multiple tags at the same time. Tags are "sticky", meaning once a tag is applied to a packet by a matching rule it exists until explicit removal. Tags are kept with the packet everywhere within the kernel, but are lost when packet leaves the kernel, for example, on transmitting packet out to the network or sending packet to a .Xr divert 4 socket. .Pp To check for previously applied tags, use the .Cm tagged rule option. To delete previously applied tag, use the .Cm untag keyword. .Pp Note: since tags are kept with the packet everywhere in kernelspace, they can be set and unset anywhere in the kernel network subsystem (using the .Xr mbuf_tags 9 facility), not only by means of the .Xr ipfw 4 .Cm tag and .Cm untag keywords. For example, there can be a specialized .Xr netgraph 4 node doing traffic analyzing and tagging for later inspecting in firewall. .It Cm untag Ar number When a packet matches a rule with the .Cm untag keyword, the tag with the number .Ar number is searched among the tags attached to this packet and, if found, removed from it. Other tags bound to packet, if present, are left untouched. .It Cm altq Ar queue When a packet matches a rule with the .Cm altq keyword, the ALTQ identifier for the given .Ar queue (see .Xr altq 4 ) will be attached. Note that this ALTQ tag is only meaningful for packets going "out" of IPFW, and not being rejected or going to divert sockets. Note that if there is insufficient memory at the time the packet is processed, it will not be tagged, so it is wise to make your ALTQ "default" queue policy account for this. If multiple .Cm altq rules match a single packet, only the first one adds the ALTQ classification tag. In doing so, traffic may be shaped by using .Cm count Cm altq Ar queue rules for classification early in the ruleset, then later applying the filtering decision. For example, .Cm check-state and .Cm keep-state rules may come later and provide the actual filtering decisions in addition to the fallback ALTQ tag. .Pp You must run .Xr pfctl 8 to set up the queues before IPFW will be able to look them up by name, and if the ALTQ disciplines are rearranged, the rules in containing the queue identifiers in the kernel will likely have gone stale and need to be reloaded. Stale queue identifiers will probably result in misclassification. .Pp All system ALTQ processing can be turned on or off via .Nm .Cm enable Ar altq and .Nm .Cm disable Ar altq . The usage of .Va net.inet.ip.fw.one_pass is irrelevant to ALTQ traffic shaping, as the actual rule action is followed always after adding an ALTQ tag. .El .Ss RULE ACTIONS A rule can be associated with one of the following actions, which will be executed when the packet matches the body of the rule. .Bl -tag -width indent .It Cm allow | accept | pass | permit Allow packets that match rule. The search terminates. .It Cm check-state Checks the packet against the dynamic ruleset. If a match is found, execute the action associated with the rule which generated this dynamic rule, otherwise move to the next rule. .br .Cm Check-state rules do not have a body. If no .Cm check-state rule is found, the dynamic ruleset is checked at the first .Cm keep-state or .Cm limit rule. .It Cm count Update counters for all packets that match rule. The search continues with the next rule. .It Cm deny | drop Discard packets that match this rule. The search terminates. .It Cm divert Ar port Divert packets that match this rule to the .Xr divert 4 socket bound to port .Ar port . The search terminates. .It Cm fwd | forward Ar ipaddr | tablearg Ns Op , Ns Ar port Change the next-hop on matching packets to .Ar ipaddr , which can be an IP address or a host name. For IPv4, the next hop can also be supplied by the last table looked up for the packet by using the .Cm tablearg keyword instead of an explicit address. The search terminates if this rule matches. .Pp If .Ar ipaddr is a local address, then matching packets will be forwarded to .Ar port (or the port number in the packet if one is not specified in the rule) on the local machine. .br If .Ar ipaddr is not a local address, then the port number (if specified) is ignored, and the packet will be forwarded to the remote address, using the route as found in the local routing table for that IP. .br A .Ar fwd rule will not match layer-2 packets (those received on ether_input, ether_output, or bridged). .br The .Cm fwd action does not change the contents of the packet at all. In particular, the destination address remains unmodified, so packets forwarded to another system will usually be rejected by that system unless there is a matching rule on that system to capture them. For packets forwarded locally, the local address of the socket will be set to the original destination address of the packet. This makes the .Xr netstat 1 entry look rather weird but is intended for use with transparent proxy servers. .It Cm nat Ar nat_nr | tablearg Pass packet to a nat instance (for network address translation, address redirect, etc.): see the .Sx NETWORK ADDRESS TRANSLATION (NAT) Section for further information. .It Cm pipe Ar pipe_nr Pass packet to a .Nm dummynet .Dq pipe (for bandwidth limitation, delay, etc.). See the .Sx TRAFFIC SHAPER (DUMMYNET) CONFIGURATION Section for further information. The search terminates; however, on exit from the pipe and if the .Xr sysctl 8 variable .Va net.inet.ip.fw.one_pass is not set, the packet is passed again to the firewall code starting from the next rule. .It Cm queue Ar queue_nr Pass packet to a .Nm dummynet .Dq queue (for bandwidth limitation using WF2Q+). .It Cm reject (Deprecated). Synonym for .Cm unreach host . .It Cm reset Discard packets that match this rule, and if the packet is a TCP packet, try to send a TCP reset (RST) notice. The search terminates. .It Cm reset6 Discard packets that match this rule, and if the packet is a TCP packet, try to send a TCP reset (RST) notice. The search terminates. .It Cm skipto Ar number | tablearg Skip all subsequent rules numbered less than .Ar number . The search continues with the first rule numbered .Ar number or higher. It is possible to use the .Cm tablearg keyword with a skipto for a .Em computed skipto, but care should be used, as no destination caching is possible in this case so the rules are always walked to find it, starting from the .Cm skipto . .It Cm call Ar number | tablearg The current rule number is saved in the internal stack and ruleset processing continues with the first rule numbered .Ar number or higher. If later a rule with the .Cm return action is encountered, the processing returns to the first rule with number of this .Cm call rule plus one or higher (the same behaviour as with packets returning from .Xr divert 4 socket after a .Cm divert action). This could be used to make somewhat like an assembly language .Dq subroutine calls to rules with common checks for different interfaces, etc. .Pp Rule with any number could be called, not just forward jumps as with .Cm skipto . So, to prevent endless loops in case of mistakes, both .Cm call and .Cm return actions don't do any jumps and simply go to the next rule if memory cannot be allocated or stack overflowed/underflowed. .Pp Internally stack for rule numbers is implemented using .Xr mbuf_tags 9 facility and currently has size of 16 entries. As mbuf tags are lost when packet leaves the kernel, .Cm divert should not be used in subroutines to avoid endless loops and other undesired effects. .It Cm return Takes rule number saved to internal stack by the last .Cm call action and returns ruleset processing to the first rule with number greater than number of corresponding .Cm call rule. See description of the .Cm call action for more details. .Pp Note that .Cm return rules usually end a .Dq subroutine and thus are unconditional, but .Nm command-line utility currently requires every action except .Cm check-state to have body. While it is sometimes useful to return only on some packets, usually you want to print just .Dq return for readability. A workaround for this is to use new syntax and .Fl c switch: .Bd -literal -offset indent # Add a rule without actual body ipfw add 2999 return via any # List rules without "from any to any" part ipfw -c list .Ed .Pp This cosmetic annoyance may be fixed in future releases. .It Cm tee Ar port Send a copy of packets matching this rule to the .Xr divert 4 socket bound to port .Ar port . The search continues with the next rule. .It Cm unreach Ar code Discard packets that match this rule, and try to send an ICMP unreachable notice with code .Ar code , where .Ar code is a number from 0 to 255, or one of these aliases: .Cm net , host , protocol , port , .Cm needfrag , srcfail , net-unknown , host-unknown , .Cm isolated , net-prohib , host-prohib , tosnet , .Cm toshost , filter-prohib , host-precedence or .Cm precedence-cutoff . The search terminates. .It Cm unreach6 Ar code Discard packets that match this rule, and try to send an ICMPv6 unreachable notice with code .Ar code , where .Ar code is a number from 0, 1, 3 or 4, or one of these aliases: .Cm no-route, admin-prohib, address or .Cm port . The search terminates. .It Cm netgraph Ar cookie Divert packet into netgraph with given .Ar cookie . The search terminates. If packet is later returned from netgraph it is either accepted or continues with the next rule, depending on .Va net.inet.ip.fw.one_pass sysctl variable. .It Cm ngtee Ar cookie A copy of packet is diverted into netgraph, original packet continues with the next rule. See .Xr ng_ipfw 4 for more information on .Cm netgraph and .Cm ngtee actions. .It Cm setfib Ar fibnum | tablearg The packet is tagged so as to use the FIB (routing table) .Ar fibnum in any subsequent forwarding decisions. In the current implementation, this is limited to the values 0 through 15, see .Xr setfib 2 . Processing continues at the next rule. It is possible to use the .Cm tablearg keyword with setfib. If the tablearg value is not within the compiled range of fibs, the packet's fib is set to 0. .It Cm setdscp Ar DSCP | number | tablearg Set specified DiffServ codepoint for an IPv4/IPv6 packet. Processing continues at the next rule. Supported values are: .Pp .Cm CS0 .Pq Dv 000000 , .Cm CS1 .Pq Dv 001000 , .Cm CS2 .Pq Dv 010000 , .Cm CS3 .Pq Dv 011000 , .Cm CS4 .Pq Dv 100000 , .Cm CS5 .Pq Dv 101000 , .Cm CS6 .Pq Dv 110000 , .Cm CS7 .Pq Dv 111000 , .Cm AF11 .Pq Dv 001010 , .Cm AF12 .Pq Dv 001100 , .Cm AF13 .Pq Dv 001110 , .Cm AF21 .Pq Dv 010010 , .Cm AF22 .Pq Dv 010100 , .Cm AF23 .Pq Dv 010110 , .Cm AF31 .Pq Dv 011010 , .Cm AF32 .Pq Dv 011100 , .Cm AF33 .Pq Dv 011110 , .Cm AF41 .Pq Dv 100010 , .Cm AF42 .Pq Dv 100100 , .Cm AF43 .Pq Dv 100110 , .Cm EF .Pq Dv 101110 , .Cm BE .Pq Dv 000000 . Additionally, DSCP value can be specified by number (0..64). It is also possible to use the .Cm tablearg keyword with setdscp. If the tablearg value is not within the 0..64 range, lower 6 bits of supplied value are used. .It Cm reass Queue and reassemble IP fragments. If the packet is not fragmented, counters are updated and processing continues with the next rule. If the packet is the last logical fragment, the packet is reassembled and, if .Va net.inet.ip.fw.one_pass is set to 0, processing continues with the next rule. Otherwise, the packet is allowed to pass and the search terminates. If the packet is a fragment in the middle of a logical group of fragments, it is consumed and processing stops immediately. .Pp Fragment handling can be tuned via .Va net.inet.ip.maxfragpackets and .Va net.inet.ip.maxfragsperpacket which limit, respectively, the maximum number of processable fragments (default: 800) and the maximum number of fragments per packet (default: 16). .Pp NOTA BENE: since fragments do not contain port numbers, they should be avoided with the .Nm reass rule. Alternatively, direction-based (like .Nm in / .Nm out ) and source-based (like .Nm via ) match patterns can be used to select fragments. .Pp Usually a simple rule like: .Bd -literal -offset indent # reassemble incoming fragments ipfw add reass all from any to any in .Ed .Pp is all you need at the beginning of your ruleset. .El .Ss RULE BODY The body of a rule contains zero or more patterns (such as specific source and destination addresses or ports, protocol options, incoming or outgoing interfaces, etc.) that the packet must match in order to be recognised. In general, the patterns are connected by (implicit) .Cm and operators -- i.e., all must match in order for the rule to match. Individual patterns can be prefixed by the .Cm not operator to reverse the result of the match, as in .Pp .Dl "ipfw add 100 allow ip from not 1.2.3.4 to any" .Pp Additionally, sets of alternative match patterns .Pq Em or-blocks can be constructed by putting the patterns in lists enclosed between parentheses ( ) or braces { }, and using the .Cm or operator as follows: .Pp .Dl "ipfw add 100 allow ip from { x or not y or z } to any" .Pp Only one level of parentheses is allowed. Beware that most shells have special meanings for parentheses or braces, so it is advisable to put a backslash \\ in front of them to prevent such interpretations. .Pp The body of a rule must in general include a source and destination address specifier. The keyword .Ar any can be used in various places to specify that the content of a required field is irrelevant. .Pp The rule body has the following format: .Bd -ragged -offset indent .Op Ar proto Cm from Ar src Cm to Ar dst .Op Ar options .Ed .Pp The first part (proto from src to dst) is for backward compatibility with earlier versions of .Fx . In modern .Fx any match pattern (including MAC headers, IP protocols, addresses and ports) can be specified in the .Ar options section. .Pp Rule fields have the following meaning: .Bl -tag -width indent .It Ar proto : protocol | Cm { Ar protocol Cm or ... } .It Ar protocol : Oo Cm not Oc Ar protocol-name | protocol-number An IP protocol specified by number or name (for a complete list see .Pa /etc/protocols ) , or one of the following keywords: .Bl -tag -width indent .It Cm ip4 | ipv4 Matches IPv4 packets. .It Cm ip6 | ipv6 Matches IPv6 packets. .It Cm ip | all Matches any packet. .El .Pp The .Cm ipv6 in .Cm proto option will be treated as inner protocol. And, the .Cm ipv4 is not available in .Cm proto option. .Pp The .Cm { Ar protocol Cm or ... } format (an .Em or-block ) is provided for convenience only but its use is deprecated. .It Ar src No and Ar dst : Bro Cm addr | Cm { Ar addr Cm or ... } Brc Op Oo Cm not Oc Ar ports An address (or a list, see below) optionally followed by .Ar ports specifiers. .Pp The second format .Em ( or-block with multiple addresses) is provided for convenience only and its use is discouraged. .It Ar addr : Oo Cm not Oc Bro .Cm any | me | me6 | .Cm table Ns Pq Ar number Ns Op , Ns Ar value .Ar | addr-list | addr-set .Brc .Bl -tag -width indent .It Cm any matches any IP address. .It Cm me matches any IP address configured on an interface in the system. .It Cm me6 matches any IPv6 address configured on an interface in the system. The address list is evaluated at the time the packet is analysed. .It Cm table Ns Pq Ar number Ns Op , Ns Ar value Matches any IPv4 address for which an entry exists in the lookup table .Ar number . If an optional 32-bit unsigned .Ar value is also specified, an entry will match only if it has this value. See the .Sx LOOKUP TABLES section below for more information on lookup tables. .El .It Ar addr-list : ip-addr Ns Op Ns , Ns Ar addr-list .It Ar ip-addr : A host or subnet address specified in one of the following ways: .Bl -tag -width indent .It Ar numeric-ip | hostname Matches a single IPv4 address, specified as dotted-quad or a hostname. Hostnames are resolved at the time the rule is added to the firewall list. .It Ar addr Ns / Ns Ar masklen Matches all addresses with base .Ar addr (specified as an IP address, a network number, or a hostname) and mask width of .Cm masklen bits. As an example, 1.2.3.4/25 or 1.2.3.0/25 will match all IP numbers from 1.2.3.0 to 1.2.3.127 . .It Ar addr Ns : Ns Ar mask Matches all addresses with base .Ar addr (specified as an IP address, a network number, or a hostname) and the mask of .Ar mask , specified as a dotted quad. As an example, 1.2.3.4:255.0.255.0 or 1.0.3.0:255.0.255.0 will match 1.*.3.*. This form is advised only for non-contiguous masks. It is better to resort to the .Ar addr Ns / Ns Ar masklen format for contiguous masks, which is more compact and less error-prone. .El .It Ar addr-set : addr Ns Oo Ns / Ns Ar masklen Oc Ns Cm { Ns Ar list Ns Cm } .It Ar list : Bro Ar num | num-num Brc Ns Op Ns , Ns Ar list Matches all addresses with base address .Ar addr (specified as an IP address, a network number, or a hostname) and whose last byte is in the list between braces { } . Note that there must be no spaces between braces and numbers (spaces after commas are allowed). Elements of the list can be specified as single entries or ranges. The .Ar masklen field is used to limit the size of the set of addresses, and can have any value between 24 and 32. If not specified, it will be assumed as 24. .br This format is particularly useful to handle sparse address sets within a single rule. Because the matching occurs using a bitmask, it takes constant time and dramatically reduces the complexity of rulesets. .br As an example, an address specified as 1.2.3.4/24{128,35-55,89} or 1.2.3.0/24{128,35-55,89} will match the following IP addresses: .br 1.2.3.128, 1.2.3.35 to 1.2.3.55, 1.2.3.89 . .It Ar addr6-list : ip6-addr Ns Op Ns , Ns Ar addr6-list .It Ar ip6-addr : A host or subnet specified one of the following ways: .Bl -tag -width indent .It Ar numeric-ip | hostname Matches a single IPv6 address as allowed by .Xr inet_pton 3 or a hostname. Hostnames are resolved at the time the rule is added to the firewall list. .It Ar addr Ns / Ns Ar masklen Matches all IPv6 addresses with base .Ar addr (specified as allowed by .Xr inet_pton or a hostname) and mask width of .Cm masklen bits. .El .Pp No support for sets of IPv6 addresses is provided because IPv6 addresses are typically random past the initial prefix. .It Ar ports : Bro Ar port | port Ns \&- Ns Ar port Ns Brc Ns Op , Ns Ar ports For protocols which support port numbers (such as TCP and UDP), optional .Cm ports may be specified as one or more ports or port ranges, separated by commas but no spaces, and an optional .Cm not operator. The .Ql \&- notation specifies a range of ports (including boundaries). .Pp Service names (from .Pa /etc/services ) may be used instead of numeric port values. The length of the port list is limited to 30 ports or ranges, though one can specify larger ranges by using an .Em or-block in the .Cm options section of the rule. .Pp A backslash .Pq Ql \e can be used to escape the dash .Pq Ql - character in a service name (from a shell, the backslash must be typed twice to avoid the shell itself interpreting it as an escape character). .Pp .Dl "ipfw add count tcp from any ftp\e\e-data-ftp to any" .Pp Fragmented packets which have a non-zero offset (i.e., not the first fragment) will never match a rule which has one or more port specifications. See the .Cm frag option for details on matching fragmented packets. .El .Ss RULE OPTIONS (MATCH PATTERNS) Additional match patterns can be used within rules. Zero or more of these so-called .Em options can be present in a rule, optionally prefixed by the .Cm not operand, and possibly grouped into .Em or-blocks . .Pp The following match patterns can be used (listed in alphabetical order): .Bl -tag -width indent .It Cm // this is a comment. Inserts the specified text as a comment in the rule. Everything following // is considered as a comment and stored in the rule. You can have comment-only rules, which are listed as having a .Cm count action followed by the comment. .It Cm bridged Alias for .Cm layer2 . .It Cm diverted Matches only packets generated by a divert socket. .It Cm diverted-loopback Matches only packets coming from a divert socket back into the IP stack input for delivery. .It Cm diverted-output Matches only packets going from a divert socket back outward to the IP stack output for delivery. .It Cm dst-ip Ar ip-address Matches IPv4 packets whose destination IP is one of the address(es) specified as argument. .It Bro Cm dst-ip6 | dst-ipv6 Brc Ar ip6-address Matches IPv6 packets whose destination IP is one of the address(es) specified as argument. .It Cm dst-port Ar ports Matches IP packets whose destination port is one of the port(s) specified as argument. .It Cm established Matches TCP packets that have the RST or ACK bits set. .It Cm ext6hdr Ar header Matches IPv6 packets containing the extended header given by .Ar header . Supported headers are: .Pp Fragment, .Pq Cm frag , Hop-to-hop options .Pq Cm hopopt , any type of Routing Header .Pq Cm route , Source routing Routing Header Type 0 .Pq Cm rthdr0 , Mobile IPv6 Routing Header Type 2 .Pq Cm rthdr2 , Destination options .Pq Cm dstopt , IPSec authentication headers .Pq Cm ah , and IPsec encapsulated security payload headers .Pq Cm esp . .It Cm fib Ar fibnum Matches a packet that has been tagged to use the given FIB (routing table) number. .It Cm flow-id Ar labels Matches IPv6 packets containing any of the flow labels given in .Ar labels . .Ar labels is a comma separated list of numeric flow labels. .It Cm frag Matches packets that are fragments and not the first fragment of an IP datagram. Note that these packets will not have the next protocol header (e.g.\& TCP, UDP) so options that look into these headers cannot match. .It Cm gid Ar group Matches all TCP or UDP packets sent by or received for a .Ar group . A .Ar group may be specified by name or number. .It Cm jail Ar prisonID Matches all TCP or UDP packets sent by or received for the jail whos prison ID is .Ar prisonID . .It Cm icmptypes Ar types Matches ICMP packets whose ICMP type is in the list .Ar types . The list may be specified as any combination of individual types (numeric) separated by commas. .Em Ranges are not allowed . The supported ICMP types are: .Pp echo reply .Pq Cm 0 , destination unreachable .Pq Cm 3 , source quench .Pq Cm 4 , redirect .Pq Cm 5 , echo request .Pq Cm 8 , router advertisement .Pq Cm 9 , router solicitation .Pq Cm 10 , time-to-live exceeded .Pq Cm 11 , IP header bad .Pq Cm 12 , timestamp request .Pq Cm 13 , timestamp reply .Pq Cm 14 , information request .Pq Cm 15 , information reply .Pq Cm 16 , address mask request .Pq Cm 17 and address mask reply .Pq Cm 18 . .It Cm icmp6types Ar types Matches ICMP6 packets whose ICMP6 type is in the list of .Ar types . The list may be specified as any combination of individual types (numeric) separated by commas. .Em Ranges are not allowed . .It Cm in | out Matches incoming or outgoing packets, respectively. .Cm in and .Cm out are mutually exclusive (in fact, .Cm out is implemented as .Cm not in Ns No ). .It Cm ipid Ar id-list Matches IPv4 packets whose .Cm ip_id field has value included in .Ar id-list , which is either a single value or a list of values or ranges specified in the same way as .Ar ports . .It Cm iplen Ar len-list Matches IP packets whose total length, including header and data, is in the set .Ar len-list , which is either a single value or a list of values or ranges specified in the same way as .Ar ports . .It Cm ipoptions Ar spec Matches packets whose IPv4 header contains the comma separated list of options specified in .Ar spec . The supported IP options are: .Pp .Cm ssrr (strict source route), .Cm lsrr (loose source route), .Cm rr (record packet route) and .Cm ts (timestamp). The absence of a particular option may be denoted with a .Ql \&! . .It Cm ipprecedence Ar precedence Matches IPv4 packets whose precedence field is equal to .Ar precedence . .It Cm ipsec Matches packets that have IPSEC history associated with them (i.e., the packet comes encapsulated in IPSEC, the kernel has IPSEC support and IPSEC_FILTERTUNNEL option, and can correctly decapsulate it). .Pp Note that specifying .Cm ipsec is different from specifying .Cm proto Ar ipsec as the latter will only look at the specific IP protocol field, irrespective of IPSEC kernel support and the validity of the IPSEC data. .Pp Further note that this flag is silently ignored in kernels without IPSEC support. It does not affect rule processing when given and the rules are handled as if with no .Cm ipsec flag. .It Cm iptos Ar spec Matches IPv4 packets whose .Cm tos field contains the comma separated list of service types specified in .Ar spec . The supported IP types of service are: .Pp .Cm lowdelay .Pq Dv IPTOS_LOWDELAY , .Cm throughput .Pq Dv IPTOS_THROUGHPUT , .Cm reliability .Pq Dv IPTOS_RELIABILITY , .Cm mincost .Pq Dv IPTOS_MINCOST , .Cm congestion .Pq Dv IPTOS_ECN_CE . The absence of a particular type may be denoted with a .Ql \&! . .It Cm dscp spec Ns Op , Ns Ar spec Matches IPv4/IPv6 packets whose .Cm DS field value is contained in .Ar spec mask. Multiple values can be specified via the comma separated list. Value can be one of keywords used in .Cm setdscp action or exact number. .It Cm ipttl Ar ttl-list Matches IPv4 packets whose time to live is included in .Ar ttl-list , which is either a single value or a list of values or ranges specified in the same way as .Ar ports . .It Cm ipversion Ar ver Matches IP packets whose IP version field is .Ar ver . .It Cm keep-state Upon a match, the firewall will create a dynamic rule, whose default behaviour is to match bidirectional traffic between source and destination IP/port using the same protocol. The rule has a limited lifetime (controlled by a set of .Xr sysctl 8 variables), and the lifetime is refreshed every time a matching packet is found. .It Cm layer2 Matches only layer2 packets, i.e., those passed to .Nm from ether_demux() and ether_output_frame(). .It Cm limit Bro Cm src-addr | src-port | dst-addr | dst-port Brc Ar N The firewall will only allow .Ar N connections with the same set of parameters as specified in the rule. One or more of source and destination addresses and ports can be specified. Currently, only IPv4 flows are supported. .It Cm lookup Bro Cm dst-ip | dst-port | src-ip | src-port | uid | jail Brc Ar N Search an entry in lookup table .Ar N that matches the field specified as argument. If not found, the match fails. Otherwise, the match succeeds and .Cm tablearg is set to the value extracted from the table. .Pp This option can be useful to quickly dispatch traffic based on certain packet fields. See the .Sx LOOKUP TABLES section below for more information on lookup tables. .It Cm { MAC | mac } Ar dst-mac src-mac Match packets with a given .Ar dst-mac and .Ar src-mac addresses, specified as the .Cm any keyword (matching any MAC address), or six groups of hex digits separated by colons, and optionally followed by a mask indicating the significant bits. The mask may be specified using either of the following methods: .Bl -enum -width indent .It A slash .Pq / followed by the number of significant bits. For example, an address with 33 significant bits could be specified as: .Pp .Dl "MAC 10:20:30:40:50:60/33 any" .Pp .It An ampersand .Pq & followed by a bitmask specified as six groups of hex digits separated by colons. For example, an address in which the last 16 bits are significant could be specified as: .Pp .Dl "MAC 10:20:30:40:50:60&00:00:00:00:ff:ff any" .Pp Note that the ampersand character has a special meaning in many shells and should generally be escaped. .Pp .El Note that the order of MAC addresses (destination first, source second) is the same as on the wire, but the opposite of the one used for IP addresses. .It Cm mac-type Ar mac-type Matches packets whose Ethernet Type field corresponds to one of those specified as argument. .Ar mac-type is specified in the same way as .Cm port numbers (i.e., one or more comma-separated single values or ranges). You can use symbolic names for known values such as .Em vlan , ipv4, ipv6 . Values can be entered as decimal or hexadecimal (if prefixed by 0x), and they are always printed as hexadecimal (unless the .Cm -N option is used, in which case symbolic resolution will be attempted). .It Cm proto Ar protocol Matches packets with the corresponding IP protocol. .It Cm recv | xmit | via Brq Ar ifX | Ar if Ns Cm * | Ar table Ns Pq Ar number Ns Op , Ns Ar value | Ar ipno | Ar any Matches packets received, transmitted or going through, respectively, the interface specified by exact name .Po Ar ifX Pc , by device name .Po Ar if* Pc , by IP address, or through some interface. .Pp The .Cm via keyword causes the interface to always be checked. If .Cm recv or .Cm xmit is used instead of .Cm via , then only the receive or transmit interface (respectively) is checked. By specifying both, it is possible to match packets based on both receive and transmit interface, e.g.: .Pp .Dl "ipfw add deny ip from any to any out recv ed0 xmit ed1" .Pp The .Cm recv interface can be tested on either incoming or outgoing packets, while the .Cm xmit interface can only be tested on outgoing packets. So .Cm out is required (and .Cm in is invalid) whenever .Cm xmit is used. .Pp A packet might not have a receive or transmit interface: packets originating from the local host have no receive interface, while packets destined for the local host have no transmit interface. .It Cm setup Matches TCP packets that have the SYN bit set but no ACK bit. This is the short form of .Dq Li tcpflags\ syn,!ack . .It Cm sockarg Matches packets that are associated to a local socket and for which the SO_USER_COOKIE socket option has been set to a non-zero value. As a side effect, the value of the option is made available as .Cm tablearg value, which in turn can be used as .Cm skipto or .Cm pipe number. .It Cm src-ip Ar ip-address Matches IPv4 packets whose source IP is one of the address(es) specified as an argument. .It Cm src-ip6 Ar ip6-address Matches IPv6 packets whose source IP is one of the address(es) specified as an argument. .It Cm src-port Ar ports Matches IP packets whose source port is one of the port(s) specified as argument. .It Cm tagged Ar tag-list Matches packets whose tags are included in .Ar tag-list , which is either a single value or a list of values or ranges specified in the same way as .Ar ports . Tags can be applied to the packet using .Cm tag rule action parameter (see it's description for details on tags). .It Cm tcpack Ar ack TCP packets only. Match if the TCP header acknowledgment number field is set to .Ar ack . .It Cm tcpdatalen Ar tcpdatalen-list Matches TCP packets whose length of TCP data is .Ar tcpdatalen-list , which is either a single value or a list of values or ranges specified in the same way as .Ar ports . .It Cm tcpflags Ar spec TCP packets only. Match if the TCP header contains the comma separated list of flags specified in .Ar spec . The supported TCP flags are: .Pp .Cm fin , .Cm syn , .Cm rst , .Cm psh , .Cm ack and .Cm urg . The absence of a particular flag may be denoted with a .Ql \&! . A rule which contains a .Cm tcpflags specification can never match a fragmented packet which has a non-zero offset. See the .Cm frag option for details on matching fragmented packets. .It Cm tcpseq Ar seq TCP packets only. Match if the TCP header sequence number field is set to .Ar seq . .It Cm tcpwin Ar tcpwin-list Matches TCP packets whose header window field is set to .Ar tcpwin-list , which is either a single value or a list of values or ranges specified in the same way as .Ar ports . .It Cm tcpoptions Ar spec TCP packets only. Match if the TCP header contains the comma separated list of options specified in .Ar spec . The supported TCP options are: .Pp .Cm mss (maximum segment size), .Cm window (tcp window advertisement), .Cm sack (selective ack), .Cm ts (rfc1323 timestamp) and .Cm cc (rfc1644 t/tcp connection count). The absence of a particular option may be denoted with a .Ql \&! . .It Cm uid Ar user Match all TCP or UDP packets sent by or received for a .Ar user . A .Ar user may be matched by name or identification number. .It Cm verrevpath For incoming packets, a routing table lookup is done on the packet's source address. If the interface on which the packet entered the system matches the outgoing interface for the route, the packet matches. If the interfaces do not match up, the packet does not match. All outgoing packets or packets with no incoming interface match. .Pp The name and functionality of the option is intentionally similar to the Cisco IOS command: .Pp .Dl ip verify unicast reverse-path .Pp This option can be used to make anti-spoofing rules to reject all packets with source addresses not from this interface. See also the option .Cm antispoof . .It Cm versrcreach For incoming packets, a routing table lookup is done on the packet's source address. If a route to the source address exists, but not the default route or a blackhole/reject route, the packet matches. Otherwise, the packet does not match. All outgoing packets match. .Pp The name and functionality of the option is intentionally similar to the Cisco IOS command: .Pp .Dl ip verify unicast source reachable-via any .Pp This option can be used to make anti-spoofing rules to reject all packets whose source address is unreachable. .It Cm antispoof For incoming packets, the packet's source address is checked if it belongs to a directly connected network. If the network is directly connected, then the interface the packet came on in is compared to the interface the network is connected to. When incoming interface and directly connected interface are not the same, the packet does not match. Otherwise, the packet does match. All outgoing packets match. .Pp This option can be used to make anti-spoofing rules to reject all packets that pretend to be from a directly connected network but do not come in through that interface. This option is similar to but more restricted than .Cm verrevpath because it engages only on packets with source addresses of directly connected networks instead of all source addresses. .El .Sh LOOKUP TABLES Lookup tables are useful to handle large sparse sets of addresses or other search keys (e.g., ports, jail IDs, interface names). In the rest of this section we will use the term ``address''. There may be up to 65535 different lookup tables, numbered 0 to 65534. .Pp Each entry is represented by an .Ar addr Ns Op / Ns Ar masklen and will match all addresses with base .Ar addr (specified as an IPv4/IPv6 address, a hostname or an unsigned integer) and mask width of .Ar masklen bits. If .Ar masklen is not specified, it defaults to 32 for IPv4 and 128 for IPv6. When looking up an IP address in a table, the most specific entry will match. Associated with each entry is a 32-bit unsigned .Ar value , which can optionally be checked by a rule matching code. When adding an entry, if .Ar value is not specified, it defaults to 0. .Pp An entry can be added to a table .Pq Cm add , or removed from a table .Pq Cm delete . A table can be examined .Pq Cm list or flushed .Pq Cm flush . .Pp Internally, each table is stored in a Radix tree, the same way as the routing table (see .Xr route 4 ) . .Pp Lookup tables currently support only ports, jail IDs, IPv4/IPv6 addresses and interface names. Wildcards is not supported for interface names. .Pp The .Cm tablearg feature provides the ability to use a value, looked up in the table, as the argument for a rule action, action parameter or rule option. This can significantly reduce number of rules in some configurations. If two tables are used in a rule, the result of the second (destination) is used. The .Cm tablearg argument can be used with the following actions: .Cm nat, pipe , queue, divert, tee, netgraph, ngtee, fwd, skipto, setfib, action parameters: .Cm tag, untag, rule options: .Cm limit, tagged. .Pp When used with .Cm fwd it is possible to supply table entries with values that are in the form of IP addresses or hostnames. See the .Sx EXAMPLES Section for example usage of tables and the tablearg keyword. .Pp When used with the .Cm skipto action, the user should be aware that the code will walk the ruleset up to a rule equal to, or past, the given number, and should therefore try keep the ruleset compact between the skipto and the target rules. .Sh SETS OF RULES Each rule belongs to one of 32 different .Em sets , numbered 0 to 31. Set 31 is reserved for the default rule. .Pp By default, rules are put in set 0, unless you use the .Cm set N attribute when entering a new rule. Sets can be individually and atomically enabled or disabled, so this mechanism permits an easy way to store multiple configurations of the firewall and quickly (and atomically) switch between them. The command to enable/disable sets is .Bd -ragged -offset indent .Nm .Cm set Oo Cm disable Ar number ... Oc Op Cm enable Ar number ... .Ed .Pp where multiple .Cm enable or .Cm disable sections can be specified. Command execution is atomic on all the sets specified in the command. By default, all sets are enabled. .Pp When you disable a set, its rules behave as if they do not exist in the firewall configuration, with only one exception: .Bd -ragged -offset indent dynamic rules created from a rule before it had been disabled will still be active until they expire. In order to delete dynamic rules you have to explicitly delete the parent rule which generated them. .Ed .Pp The set number of rules can be changed with the command .Bd -ragged -offset indent .Nm .Cm set move .Brq Cm rule Ar rule-number | old-set .Cm to Ar new-set .Ed .Pp Also, you can atomically swap two rulesets with the command .Bd -ragged -offset indent .Nm .Cm set swap Ar first-set second-set .Ed .Pp See the .Sx EXAMPLES Section on some possible uses of sets of rules. .Sh STATEFUL FIREWALL Stateful operation is a way for the firewall to dynamically create rules for specific flows when packets that match a given pattern are detected. Support for stateful operation comes through the .Cm check-state , keep-state and .Cm limit options of .Nm rules . .Pp Dynamic rules are created when a packet matches a .Cm keep-state or .Cm limit rule, causing the creation of a .Em dynamic rule which will match all and only packets with a given .Em protocol between a .Em src-ip/src-port dst-ip/dst-port pair of addresses .Em ( src and .Em dst are used here only to denote the initial match addresses, but they are completely equivalent afterwards). Dynamic rules will be checked at the first .Cm check-state, keep-state or .Cm limit occurrence, and the action performed upon a match will be the same as in the parent rule. .Pp Note that no additional attributes other than protocol and IP addresses and ports are checked on dynamic rules. .Pp The typical use of dynamic rules is to keep a closed firewall configuration, but let the first TCP SYN packet from the inside network install a dynamic rule for the flow so that packets belonging to that session will be allowed through the firewall: .Pp .Dl "ipfw add check-state" .Dl "ipfw add allow tcp from my-subnet to any setup keep-state" .Dl "ipfw add deny tcp from any to any" .Pp A similar approach can be used for UDP, where an UDP packet coming from the inside will install a dynamic rule to let the response through the firewall: .Pp .Dl "ipfw add check-state" .Dl "ipfw add allow udp from my-subnet to any keep-state" .Dl "ipfw add deny udp from any to any" .Pp Dynamic rules expire after some time, which depends on the status of the flow and the setting of some .Cm sysctl variables. See Section .Sx SYSCTL VARIABLES for more details. For TCP sessions, dynamic rules can be instructed to periodically send keepalive packets to refresh the state of the rule when it is about to expire. .Pp See Section .Sx EXAMPLES for more examples on how to use dynamic rules. .Sh TRAFFIC SHAPER (DUMMYNET) CONFIGURATION .Nm is also the user interface for the .Nm dummynet traffic shaper, packet scheduler and network emulator, a subsystem that can artificially queue, delay or drop packets emulating the behaviour of certain network links or queueing systems. .Pp .Nm dummynet operates by first using the firewall to select packets using any match pattern that can be used in .Nm rules. Matching packets are then passed to either of two different objects, which implement the traffic regulation: .Bl -hang -offset XXXX .It Em pipe A .Em pipe emulates a .Em link with given bandwidth and propagation delay, driven by a FIFO scheduler and a single queue with programmable queue size and packet loss rate. Packets are appended to the queue as they come out from .Nm ipfw , and then transferred in FIFO order to the link at the desired rate. .It Em queue A .Em queue is an abstraction used to implement packet scheduling using one of several packet scheduling algorithms. Packets sent to a .Em queue are first grouped into flows according to a mask on the 5-tuple. Flows are then passed to the scheduler associated to the .Em queue , and each flow uses scheduling parameters (weight and others) as configured in the .Em queue itself. A scheduler in turn is connected to an emulated link, and arbitrates the link's bandwidth among backlogged flows according to weights and to the features of the scheduling algorithm in use. .El .Pp In practice, .Em pipes can be used to set hard limits to the bandwidth that a flow can use, whereas .Em queues can be used to determine how different flows share the available bandwidth. .Pp A graphical representation of the binding of queues, flows, schedulers and links is below. .Bd -literal -offset indent (flow_mask|sched_mask) sched_mask +---------+ weight Wx +-------------+ | |->-[flow]-->--| |-+ -->--| QUEUE x | ... | | | | |->-[flow]-->--| SCHEDuler N | | +---------+ | | | ... | +--[LINK N]-->-- +---------+ weight Wy | | +--[LINK N]-->-- | |->-[flow]-->--| | | -->--| QUEUE y | ... | | | | |->-[flow]-->--| | | +---------+ +-------------+ | +-------------+ .Ed It is important to understand the role of the SCHED_MASK and FLOW_MASK, which are configured through the commands .Dl "ipfw sched N config mask SCHED_MASK ..." and .Dl "ipfw queue X config mask FLOW_MASK ..." . .Pp The SCHED_MASK is used to assign flows to one or more scheduler instances, one for each value of the packet's 5-tuple after applying SCHED_MASK. As an example, using ``src-ip 0xffffff00'' creates one instance for each /24 destination subnet. .Pp The FLOW_MASK, together with the SCHED_MASK, is used to split packets into flows. As an example, using ``src-ip 0x000000ff'' together with the previous SCHED_MASK makes a flow for each individual source address. In turn, flows for each /24 subnet will be sent to the same scheduler instance. .Pp The above diagram holds even for the .Em pipe case, with the only restriction that a .Em pipe only supports a SCHED_MASK, and forces the use of a FIFO scheduler (these are for backward compatibility reasons; in fact, internally, a .Nm dummynet's pipe is implemented exactly as above). .Pp There are two modes of .Nm dummynet operation: .Dq normal and .Dq fast . The .Dq normal mode tries to emulate a real link: the .Nm dummynet scheduler ensures that the packet will not leave the pipe faster than it would on the real link with a given bandwidth. The .Dq fast mode allows certain packets to bypass the .Nm dummynet scheduler (if packet flow does not exceed pipe's bandwidth). This is the reason why the .Dq fast mode requires less CPU cycles per packet (on average) and packet latency can be significantly lower in comparison to a real link with the same bandwidth. The default mode is .Dq normal . The .Dq fast mode can be enabled by setting the .Va net.inet.ip.dummynet.io_fast .Xr sysctl 8 variable to a non-zero value. .Pp .Ss PIPE, QUEUE AND SCHEDULER CONFIGURATION The .Em pipe , .Em queue and .Em scheduler configuration commands are the following: .Bd -ragged -offset indent .Cm pipe Ar number Cm config Ar pipe-configuration .Pp .Cm queue Ar number Cm config Ar queue-configuration .Pp .Cm sched Ar number Cm config Ar sched-configuration .Ed .Pp The following parameters can be configured for a pipe: .Pp .Bl -tag -width indent -compact .It Cm bw Ar bandwidth | device Bandwidth, measured in .Sm off .Op Cm K | M .Brq Cm bit/s | Byte/s . .Sm on .Pp A value of 0 (default) means unlimited bandwidth. The unit must immediately follow the number, as in .Pp .Dl "ipfw pipe 1 config bw 300Kbit/s" .Pp If a device name is specified instead of a numeric value, as in .Pp .Dl "ipfw pipe 1 config bw tun0" .Pp then the transmit clock is supplied by the specified device. At the moment only the .Xr tun 4 device supports this functionality, for use in conjunction with .Xr ppp 8 . .Pp .It Cm delay Ar ms-delay Propagation delay, measured in milliseconds. The value is rounded to the next multiple of the clock tick (typically 10ms, but it is a good practice to run kernels with .Dq "options HZ=1000" to reduce the granularity to 1ms or less). The default value is 0, meaning no delay. .Pp .It Cm burst Ar size If the data to be sent exceeds the pipe's bandwidth limit (and the pipe was previously idle), up to .Ar size bytes of data are allowed to bypass the .Nm dummynet scheduler, and will be sent as fast as the physical link allows. Any additional data will be transmitted at the rate specified by the .Nm pipe bandwidth. The burst size depends on how long the pipe has been idle; the effective burst size is calculated as follows: MAX( .Ar size , .Nm bw * pipe_idle_time). .Pp .It Cm profile Ar filename A file specifying the additional overhead incurred in the transmission of a packet on the link. .Pp Some link types introduce extra delays in the transmission of a packet, e.g., because of MAC level framing, contention on the use of the channel, MAC level retransmissions and so on. From our point of view, the channel is effectively unavailable for this extra time, which is constant or variable depending on the link type. Additionally, packets may be dropped after this time (e.g., on a wireless link after too many retransmissions). We can model the additional delay with an empirical curve that represents its distribution. .Bd -literal -offset indent cumulative probability 1.0 ^ | L +-- loss-level x | ****** | * | ***** | * | ** | * +-------*-------------------> delay .Ed 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 discontinuity in the delay distribution: the pipe will use the largest delay for a given probability. .Pp The file format is the following, with whitespace acting as a separator and '#' indicating the beginning a comment: .Bl -tag -width indent .It Cm name Ar identifier optional name (listed by "ipfw pipe show") to identify the delay distribution; .It Cm bw Ar value the bandwidth used for the pipe. If not specified here, it must be present explicitly as a configuration parameter for the pipe; .It Cm loss-level Ar L the probability above which packets are lost. (0.0 <= L <= 1.0, default 1.0 i.e., no loss); .It Cm samples Ar N the number of samples used in the internal representation of the curve (2..1024; default 100); .It Cm "delay prob" | "prob delay" One of these two lines is mandatory and defines the format of the following lines with data points. .It Ar XXX Ar YYY 2 or more lines representing points in the curve, with either delay or probability first, according to the chosen format. The unit for delay is milliseconds. Data points do not need to be sorted. Also, the number of actual lines can be different from the value of the "samples" parameter: .Nm utility will sort and interpolate the curve as needed. .El .Pp Example of a profile file: .Bd -literal -offset indent name bla_bla_bla samples 100 loss-level 0.86 prob delay 0 200 # minimum overhead is 200ms 0.5 200 0.5 300 0.8 1000 0.9 1300 1 1300 #configuration file end .Ed .El .Pp The following parameters can be configured for a queue: .Pp .Bl -tag -width indent -compact .It Cm pipe Ar pipe_nr Connects a queue to the specified pipe. Multiple queues (with the same or different weights) can be connected to the same pipe, which specifies the aggregate rate for the set of queues. .Pp .It Cm weight Ar weight Specifies the weight to be used for flows matching this queue. The weight must be in the range 1..100, and defaults to 1. .El .Pp The following case-insensitive parameters can be configured for a scheduler: .Pp .Bl -tag -width indent -compact .It Cm type Ar {fifo | wf2q+ | rr | qfq} specifies the scheduling algorithm to use. .Bl -tag -width indent -compact .It Cm fifo is just a FIFO scheduler (which means that all packets are stored in the same queue as they arrive to the scheduler). FIFO has O(1) per-packet time complexity, with very low constants (estimate 60-80ns on a 2GHz desktop machine) but gives no service guarantees. .It Cm wf2q+ implements the WF2Q+ algorithm, which is a Weighted Fair Queueing algorithm which permits flows to share bandwidth according to their weights. Note that weights are not priorities; even a flow with a minuscule weight will never starve. WF2Q+ has O(log N) per-packet processing cost, where N is the number of flows, and is the default algorithm used by previous versions dummynet's queues. .It Cm rr implements the Deficit Round Robin algorithm, which has O(1) processing costs (roughly, 100-150ns per packet) and permits bandwidth allocation according to weights, but with poor service guarantees. .It Cm qfq implements the QFQ algorithm, which is a very fast variant of WF2Q+, with similar service guarantees and O(1) processing costs (roughly, 200-250ns per packet). .El .El .Pp In addition to the type, all parameters allowed for a pipe can also be specified for a scheduler. .Pp Finally, the following parameters can be configured for both pipes and queues: .Pp .Bl -tag -width XXXX -compact .It Cm buckets Ar hash-table-size Specifies the size of the hash table used for storing the various queues. Default value is 64 controlled by the .Xr sysctl 8 variable .Va net.inet.ip.dummynet.hash_size , allowed range is 16 to 65536. .Pp .It Cm mask Ar mask-specifier Packets sent to a given pipe or queue by an .Nm rule can be further classified into multiple flows, each of which is then sent to a different .Em dynamic pipe or queue. A flow identifier is constructed by masking the IP addresses, ports and protocol types as specified with the .Cm mask options in the configuration of the pipe or queue. For each different flow identifier, a new pipe or queue is created with the same parameters as the original object, and matching packets are sent to it. .Pp Thus, when .Em dynamic pipes are used, each flow will get the same bandwidth as defined by the pipe, whereas when .Em dynamic queues are used, each flow will share the parent's pipe bandwidth evenly with other flows generated by the same queue (note that other queues with different weights might be connected to the same pipe). .br Available mask specifiers are a combination of one or more of the following: .Pp .Cm dst-ip Ar mask , .Cm dst-ip6 Ar mask , .Cm src-ip Ar mask , .Cm src-ip6 Ar mask , .Cm dst-port Ar mask , .Cm src-port Ar mask , .Cm flow-id Ar mask , .Cm proto Ar mask or .Cm all , .Pp where the latter means all bits in all fields are significant. .Pp .It Cm noerror When a packet is dropped by a .Nm dummynet queue or pipe, the error is normally reported to the caller routine in the kernel, in the same way as it happens when a device queue fills up. Setting this option reports the packet as successfully delivered, which can be needed for some experimental setups where you want to simulate loss or congestion at a remote router. .Pp .It Cm plr Ar packet-loss-rate Packet loss rate. Argument .Ar packet-loss-rate is a floating-point number between 0 and 1, with 0 meaning no loss, 1 meaning 100% loss. The loss rate is internally represented on 31 bits. .Pp .It Cm queue Brq Ar slots | size Ns Cm Kbytes Queue size, in .Ar slots or .Cm KBytes . Default value is 50 slots, which is the typical queue size for Ethernet devices. Note that for slow speed links you should keep the queue size short or your traffic might be affected by a significant queueing delay. E.g., 50 max-sized ethernet packets (1500 bytes) mean 600Kbit or 20s of queue on a 30Kbit/s pipe. Even worse effects can result if you get packets from an interface with a much larger MTU, e.g.\& the loopback interface with its 16KB packets. The .Xr sysctl 8 variables .Em net.inet.ip.dummynet.pipe_byte_limit and .Em net.inet.ip.dummynet.pipe_slot_limit control the maximum lengths that can be specified. .Pp .It Cm red | gred Ar w_q Ns / Ns Ar min_th Ns / Ns Ar max_th Ns / Ns Ar max_p +[ecn] Make use of the RED (Random Early Detection) queue management algorithm. .Ar w_q and .Ar max_p are floating -point numbers between 0 and 1 (0 not included), while +point numbers between 0 and 1 (inclusive), while .Ar min_th and .Ar max_th are integer numbers specifying thresholds for queue management (thresholds are computed in bytes if the queue has been defined in bytes, in slots otherwise). -The +The two parameters can also be of the same value if needed. The .Nm dummynet -also supports the gentle RED variant (gred). -Three +also supports the gentle RED variant (gred) and ECN (Explicit Congestion +Notification) as optional. Three .Xr sysctl 8 variables can be used to control the RED behaviour: .Bl -tag -width indent .It Va net.inet.ip.dummynet.red_lookup_depth specifies the accuracy in computing the average queue when the link is idle (defaults to 256, must be greater than zero) .It Va net.inet.ip.dummynet.red_avg_pkt_size specifies the expected average packet size (defaults to 512, must be greater than zero) .It Va net.inet.ip.dummynet.red_max_pkt_size specifies the expected maximum packet size, only used when queue thresholds are in bytes (defaults to 1500, must be greater than zero). .El .El .Pp When used with IPv6 data, .Nm dummynet currently has several limitations. Information necessary to route link-local packets to an interface is not available after processing by .Nm dummynet so those packets are dropped in the output path. Care should be taken to ensure that link-local packets are not passed to .Nm dummynet . .Sh CHECKLIST Here are some important points to consider when designing your rules: .Bl -bullet .It Remember that you filter both packets going .Cm in and .Cm out . Most connections need packets going in both directions. .It Remember to test very carefully. It is a good idea to be near the console when doing this. If you cannot be near the console, use an auto-recovery script such as the one in .Pa /usr/share/examples/ipfw/change_rules.sh . .It Do not forget the loopback interface. .El .Sh FINE POINTS .Bl -bullet .It There are circumstances where fragmented datagrams are unconditionally dropped. TCP packets are dropped if they do not contain at least 20 bytes of TCP header, UDP packets are dropped if they do not contain a full 8 byte UDP header, and ICMP packets are dropped if they do not contain 4 bytes of ICMP header, enough to specify the ICMP type, code, and checksum. These packets are simply logged as .Dq pullup failed since there may not be enough good data in the packet to produce a meaningful log entry. .It Another type of packet is unconditionally dropped, a TCP packet with a fragment offset of one. This is a valid packet, but it only has one use, to try to circumvent firewalls. When logging is enabled, these packets are reported as being dropped by rule -1. .It If you are logged in over a network, loading the .Xr kld 4 version of .Nm is probably not as straightforward as you would think. The following command line is recommended: .Bd -literal -offset indent kldload ipfw && \e ipfw add 32000 allow ip from any to any .Ed .Pp Along the same lines, doing an .Bd -literal -offset indent ipfw flush .Ed .Pp in similar surroundings is also a bad idea. .It The .Nm filter list may not be modified if the system security level is set to 3 or higher (see .Xr init 8 for information on system security levels). .El .Sh PACKET DIVERSION A .Xr divert 4 socket bound to the specified port will receive all packets diverted to that port. If no socket is bound to the destination port, or if the divert module is not loaded, or if the kernel was not compiled with divert socket support, the packets are dropped. .Sh NETWORK ADDRESS TRANSLATION (NAT) .Nm support in-kernel NAT using the kernel version of .Xr libalias 3 . .Pp The nat configuration command is the following: .Bd -ragged -offset indent .Bk -words .Cm nat .Ar nat_number .Cm config .Ar nat-configuration .Ek .Ed .Pp The following parameters can be configured: .Bl -tag -width indent .It Cm ip Ar ip_address Define an ip address to use for aliasing. .It Cm if Ar nic Use ip address of NIC for aliasing, dynamically changing it if NIC's ip address changes. .It Cm log Enable logging on this nat instance. .It Cm deny_in Deny any incoming connection from outside world. .It Cm same_ports Try to leave the alias port numbers unchanged from the actual local port numbers. .It Cm unreg_only Traffic on the local network not originating from an unregistered address spaces will be ignored. .It Cm reset Reset table of the packet aliasing engine on address change. .It Cm reverse Reverse the way libalias handles aliasing. .It Cm proxy_only Obey transparent proxy rules only, packet aliasing is not performed. .It Cm skip_global Skip instance in case of global state lookup (see below). .El .Pp Some specials value can be supplied instead of .Va nat_number: .Bl -tag -width indent .It Cm global Looks up translation state in all configured nat instances. If an entry is found, packet is aliased according to that entry. If no entry was found in any of the instances, packet is passed unchanged, and no new entry will be created. See section .Sx MULTIPLE INSTANCES in .Xr natd 8 for more information. .It Cm tablearg Uses argument supplied in lookup table. See .Sx LOOKUP TABLES section below for more information on lookup tables. .El .Pp To let the packet continue after being (de)aliased, set the sysctl variable .Va net.inet.ip.fw.one_pass to 0. For more information about aliasing modes, refer to .Xr libalias 3 . See Section .Sx EXAMPLES for some examples about nat usage. .Ss REDIRECT AND LSNAT SUPPORT IN IPFW Redirect and LSNAT support follow closely the syntax used in .Xr natd 8 . See Section .Sx EXAMPLES for some examples on how to do redirect and lsnat. .Ss SCTP NAT SUPPORT SCTP nat can be configured in a similar manner to TCP through the .Nm command line tool. The main difference is that .Nm sctp nat does not do port translation. Since the local and global side ports will be the same, there is no need to specify both. Ports are redirected as follows: .Bd -ragged -offset indent .Bk -words .Cm nat .Ar nat_number .Cm config if .Ar nic .Cm redirect_port sctp .Ar ip_address [,addr_list] {[port | port-port] [,ports]} .Ek .Ed .Pp Most .Nm sctp nat configuration can be done in real-time through the .Xr sysctl 8 interface. All may be changed dynamically, though the hash_table size will only change for new .Nm nat instances. See .Sx SYSCTL VARIABLES for more info. .Sh LOADER TUNABLES Tunables can be set in .Xr loader 8 prompt, .Xr loader.conf 5 or .Xr kenv 1 before ipfw module gets loaded. .Bl -tag -width indent .It Va net.inet.ip.fw.default_to_accept: No 0 Defines ipfw last rule behavior. This value overrides .Cd "options IPFW_DEFAULT_TO_(ACCEPT|DENY)" from kernel configuration file. .It Va net.inet.ip.fw.tables_max: No 128 Defines number of tables available in ipfw. Number cannot exceed 65534. .El .Sh SYSCTL VARIABLES A set of .Xr sysctl 8 variables controls the behaviour of the firewall and associated modules .Pq Nm dummynet , bridge , sctp nat . These are shown below together with their default value (but always check with the .Xr sysctl 8 command what value is actually in use) and meaning: .Bl -tag -width indent .It Va net.inet.ip.alias.sctp.accept_global_ootb_addip: No 0 Defines how the .Nm nat responds to receipt of global OOTB ASCONF-AddIP: .Bl -tag -width indent .It Cm 0 No response (unless a partially matching association exists - ports and vtags match but global address does not) .It Cm 1 .Nm nat will accept and process all OOTB global AddIP messages. .El .Pp Option 1 should never be selected as this forms a security risk. An attacker can establish multiple fake associations by sending AddIP messages. .It Va net.inet.ip.alias.sctp.chunk_proc_limit: No 5 Defines the maximum number of chunks in an SCTP packet that will be parsed for a packet that matches an existing association. This value is enforced to be greater or equal than .Cm net.inet.ip.alias.sctp.initialising_chunk_proc_limit . A high value is a DoS risk yet setting too low a value may result in important control chunks in the packet not being located and parsed. .It Va net.inet.ip.alias.sctp.error_on_ootb: No 1 Defines when the .Nm nat responds to any Out-of-the-Blue (OOTB) packets with ErrorM packets. An OOTB packet is a packet that arrives with no existing association registered in the .Nm nat and is not an INIT or ASCONF-AddIP packet: .Bl -tag -width indent .It Cm 0 ErrorM is never sent in response to OOTB packets. .It Cm 1 ErrorM is only sent to OOTB packets received on the local side. .It Cm 2 ErrorM is sent to the local side and on the global side ONLY if there is a partial match (ports and vtags match but the source global IP does not). This value is only useful if the .Nm nat is tracking global IP addresses. .It Cm 3 ErrorM is sent in response to all OOTB packets on both the local and global side (DoS risk). .El .Pp At the moment the default is 0, since the ErrorM packet is not yet supported by most SCTP stacks. When it is supported, and if not tracking global addresses, we recommend setting this value to 1 to allow multi-homed local hosts to function with the .Nm nat . To track global addresses, we recommend setting this value to 2 to allow global hosts to be informed when they need to (re)send an ASCONF-AddIP. Value 3 should never be chosen (except for debugging) as the .Nm nat will respond to all OOTB global packets (a DoS risk). .It Va net.inet.ip.alias.sctp.hashtable_size: No 2003 Size of hash tables used for .Nm nat lookups (100 < prime_number > 1000001). This value sets the .Nm hash table size for any future created .Nm nat instance and therefore must be set prior to creating a .Nm nat instance. The table sizes may be changed to suit specific needs. If there will be few concurrent associations, and memory is scarce, you may make these smaller. If there will be many thousands (or millions) of concurrent associations, you should make these larger. A prime number is best for the table size. The sysctl update function will adjust your input value to the next highest prime number. .It Va net.inet.ip.alias.sctp.holddown_time: No 0 Hold association in table for this many seconds after receiving a SHUTDOWN-COMPLETE. This allows endpoints to correct shutdown gracefully if a shutdown_complete is lost and retransmissions are required. .It Va net.inet.ip.alias.sctp.init_timer: No 15 Timeout value while waiting for (INIT-ACK|AddIP-ACK). This value cannot be 0. .It Va net.inet.ip.alias.sctp.initialising_chunk_proc_limit: No 2 Defines the maximum number of chunks in an SCTP packet that will be parsed when no existing association exists that matches that packet. Ideally this packet will only be an INIT or ASCONF-AddIP packet. A higher value may become a DoS risk as malformed packets can consume processing resources. .It Va net.inet.ip.alias.sctp.param_proc_limit: No 25 Defines the maximum number of parameters within a chunk that will be parsed in a packet. As for other similar sysctl variables, larger values pose a DoS risk. .It Va net.inet.ip.alias.sctp.log_level: No 0 Level of detail in the system log messages (0 \- minimal, 1 \- event, 2 \- info, 3 \- detail, 4 \- debug, 5 \- max debug). May be a good option in high loss environments. .It Va net.inet.ip.alias.sctp.shutdown_time: No 15 Timeout value while waiting for SHUTDOWN-COMPLETE. This value cannot be 0. .It Va net.inet.ip.alias.sctp.track_global_addresses: No 0 Enables/disables global IP address tracking within the .Nm nat and places an upper limit on the number of addresses tracked for each association: .Bl -tag -width indent .It Cm 0 Global tracking is disabled .It Cm >1 Enables tracking, the maximum number of addresses tracked for each association is limited to this value .El .Pp This variable is fully dynamic, the new value will be adopted for all newly arriving associations, existing associations are treated as they were previously. Global tracking will decrease the number of collisions within the .Nm nat at a cost of increased processing load, memory usage, complexity, and possible .Nm nat state problems in complex networks with multiple .Nm nats . We recommend not tracking global IP addresses, this will still result in a fully functional .Nm nat . .It Va net.inet.ip.alias.sctp.up_timer: No 300 Timeout value to keep an association up with no traffic. This value cannot be 0. .It Va net.inet.ip.dummynet.expire : No 1 Lazily delete dynamic pipes/queue once they have no pending traffic. You can disable this by setting the variable to 0, in which case the pipes/queues will only be deleted when the threshold is reached. .It Va net.inet.ip.dummynet.hash_size : No 64 Default size of the hash table used for dynamic pipes/queues. This value is used when no .Cm buckets option is specified when configuring a pipe/queue. .It Va net.inet.ip.dummynet.io_fast : No 0 If set to a non-zero value, the .Dq fast mode of .Nm dummynet operation (see above) is enabled. .It Va net.inet.ip.dummynet.io_pkt Number of packets passed to .Nm dummynet . .It Va net.inet.ip.dummynet.io_pkt_drop Number of packets dropped by .Nm dummynet . .It Va net.inet.ip.dummynet.io_pkt_fast Number of packets bypassed by the .Nm dummynet scheduler. .It Va net.inet.ip.dummynet.max_chain_len : No 16 Target value for the maximum number of pipes/queues in a hash bucket. The product .Cm max_chain_len*hash_size is used to determine the threshold over which empty pipes/queues will be expired even when .Cm net.inet.ip.dummynet.expire=0 . .It Va net.inet.ip.dummynet.red_lookup_depth : No 256 .It Va net.inet.ip.dummynet.red_avg_pkt_size : No 512 .It Va net.inet.ip.dummynet.red_max_pkt_size : No 1500 Parameters used in the computations of the drop probability for the RED algorithm. .It Va net.inet.ip.dummynet.pipe_byte_limit : No 1048576 .It Va net.inet.ip.dummynet.pipe_slot_limit : No 100 The maximum queue size that can be specified in bytes or packets. These limits prevent accidental exhaustion of resources such as mbufs. If you raise these limits, you should make sure the system is configured so that sufficient resources are available. .It Va net.inet.ip.fw.autoinc_step : No 100 Delta between rule numbers when auto-generating them. The value must be in the range 1..1000. .It Va net.inet.ip.fw.curr_dyn_buckets : Va net.inet.ip.fw.dyn_buckets The current number of buckets in the hash table for dynamic rules (readonly). .It Va net.inet.ip.fw.debug : No 1 Controls debugging messages produced by .Nm . .It Va net.inet.ip.fw.default_rule : No 65535 The default rule number (read-only). By the design of .Nm , the default rule is the last one, so its number can also serve as the highest number allowed for a rule. .It Va net.inet.ip.fw.dyn_buckets : No 256 The number of buckets in the hash table for dynamic rules. Must be a power of 2, up to 65536. It only takes effect when all dynamic rules have expired, so you are advised to use a .Cm flush command to make sure that the hash table is resized. .It Va net.inet.ip.fw.dyn_count : No 3 Current number of dynamic rules (read-only). .It Va net.inet.ip.fw.dyn_keepalive : No 1 Enables generation of keepalive packets for .Cm keep-state rules on TCP sessions. A keepalive is generated to both sides of the connection every 5 seconds for the last 20 seconds of the lifetime of the rule. .It Va net.inet.ip.fw.dyn_max : No 8192 Maximum number of dynamic rules. When you hit this limit, no more dynamic rules can be installed until old ones expire. .It Va net.inet.ip.fw.dyn_ack_lifetime : No 300 .It Va net.inet.ip.fw.dyn_syn_lifetime : No 20 .It Va net.inet.ip.fw.dyn_fin_lifetime : No 1 .It Va net.inet.ip.fw.dyn_rst_lifetime : No 1 .It Va net.inet.ip.fw.dyn_udp_lifetime : No 5 .It Va net.inet.ip.fw.dyn_short_lifetime : No 30 These variables control the lifetime, in seconds, of dynamic rules. Upon the initial SYN exchange the lifetime is kept short, then increased after both SYN have been seen, then decreased again during the final FIN exchange or when a RST is received. Both .Em dyn_fin_lifetime and .Em dyn_rst_lifetime must be strictly lower than 5 seconds, the period of repetition of keepalives. The firewall enforces that. .It Va net.inet.ip.fw.enable : No 1 Enables the firewall. Setting this variable to 0 lets you run your machine without firewall even if compiled in. .It Va net.inet6.ip6.fw.enable : No 1 provides the same functionality as above for the IPv6 case. .It Va net.inet.ip.fw.one_pass : No 1 When set, the packet exiting from the .Nm dummynet pipe or from .Xr ng_ipfw 4 node is not passed though the firewall again. Otherwise, after an action, the packet is reinjected into the firewall at the next rule. .It Va net.inet.ip.fw.tables_max : No 128 Maximum number of tables. .It Va net.inet.ip.fw.verbose : No 1 Enables verbose messages. .It Va net.inet.ip.fw.verbose_limit : No 0 Limits the number of messages produced by a verbose firewall. .It Va net.inet6.ip6.fw.deny_unknown_exthdrs : No 1 If enabled packets with unknown IPv6 Extension Headers will be denied. .It Va net.link.ether.ipfw : No 0 Controls whether layer-2 packets are passed to .Nm . Default is no. .It Va net.link.bridge.ipfw : No 0 Controls whether bridged packets are passed to .Nm . Default is no. .El .Sh EXAMPLES There are far too many possible uses of .Nm so this Section will only give a small set of examples. .Pp .Ss BASIC PACKET FILTERING This command adds an entry which denies all tcp packets from .Em cracker.evil.org to the telnet port of .Em wolf.tambov.su from being forwarded by the host: .Pp .Dl "ipfw add deny tcp from cracker.evil.org to wolf.tambov.su telnet" .Pp This one disallows any connection from the entire cracker's network to my host: .Pp .Dl "ipfw add deny ip from 123.45.67.0/24 to my.host.org" .Pp A first and efficient way to limit access (not using dynamic rules) is the use of the following rules: .Pp .Dl "ipfw add allow tcp from any to any established" .Dl "ipfw add allow tcp from net1 portlist1 to net2 portlist2 setup" .Dl "ipfw add allow tcp from net3 portlist3 to net3 portlist3 setup" .Dl "..." .Dl "ipfw add deny tcp from any to any" .Pp The first rule will be a quick match for normal TCP packets, but it will not match the initial SYN packet, which will be matched by the .Cm setup rules only for selected source/destination pairs. All other SYN packets will be rejected by the final .Cm deny rule. .Pp If you administer one or more subnets, you can take advantage of the address sets and or-blocks and write extremely compact rulesets which selectively enable services to blocks of clients, as below: .Pp .Dl "goodguys=\*q{ 10.1.2.0/24{20,35,66,18} or 10.2.3.0/28{6,3,11} }\*q" .Dl "badguys=\*q10.1.2.0/24{8,38,60}\*q" .Dl "" .Dl "ipfw add allow ip from ${goodguys} to any" .Dl "ipfw add deny ip from ${badguys} to any" .Dl "... normal policies ..." .Pp The .Cm verrevpath option could be used to do automated anti-spoofing by adding the following to the top of a ruleset: .Pp .Dl "ipfw add deny ip from any to any not verrevpath in" .Pp This rule drops all incoming packets that appear to be coming to the system on the wrong interface. For example, a packet with a source address belonging to a host on a protected internal network would be dropped if it tried to enter the system from an external interface. .Pp The .Cm antispoof option could be used to do similar but more restricted anti-spoofing by adding the following to the top of a ruleset: .Pp .Dl "ipfw add deny ip from any to any not antispoof in" .Pp This rule drops all incoming packets that appear to be coming from another directly connected system but on the wrong interface. For example, a packet with a source address of .Li 192.168.0.0/24 , configured on .Li fxp0 , but coming in on .Li fxp1 would be dropped. .Pp The .Cm setdscp option could be used to (re)mark user traffic, by adding the following to the appropriate place in ruleset: .Pp .Dl "ipfw add setdscp be ip from any to any dscp af11,af21" .Ss DYNAMIC RULES In order to protect a site from flood attacks involving fake TCP packets, it is safer to use dynamic rules: .Pp .Dl "ipfw add check-state" .Dl "ipfw add deny tcp from any to any established" .Dl "ipfw add allow tcp from my-net to any setup keep-state" .Pp This will let the firewall install dynamic rules only for those connection which start with a regular SYN packet coming from the inside of our network. Dynamic rules are checked when encountering the first occurrence of a .Cm check-state , .Cm keep-state or .Cm limit rule. A .Cm check-state rule should usually be placed near the beginning of the ruleset to minimize the amount of work scanning the ruleset. Your mileage may vary. .Pp To limit the number of connections a user can open you can use the following type of rules: .Pp .Dl "ipfw add allow tcp from my-net/24 to any setup limit src-addr 10" .Dl "ipfw add allow tcp from any to me setup limit src-addr 4" .Pp The former (assuming it runs on a gateway) will allow each host on a /24 network to open at most 10 TCP connections. The latter can be placed on a server to make sure that a single client does not use more than 4 simultaneous connections. .Pp .Em BEWARE : stateful rules can be subject to denial-of-service attacks by a SYN-flood which opens a huge number of dynamic rules. The effects of such attacks can be partially limited by acting on a set of .Xr sysctl 8 variables which control the operation of the firewall. .Pp Here is a good usage of the .Cm list command to see accounting records and timestamp information: .Pp .Dl ipfw -at list .Pp or in short form without timestamps: .Pp .Dl ipfw -a list .Pp which is equivalent to: .Pp .Dl ipfw show .Pp Next rule diverts all incoming packets from 192.168.2.0/24 to divert port 5000: .Pp .Dl ipfw divert 5000 ip from 192.168.2.0/24 to any in .Ss TRAFFIC SHAPING The following rules show some of the applications of .Nm and .Nm dummynet for simulations and the like. .Pp This rule drops random incoming packets with a probability of 5%: .Pp .Dl "ipfw add prob 0.05 deny ip from any to any in" .Pp A similar effect can be achieved making use of .Nm dummynet pipes: .Pp .Dl "ipfw add pipe 10 ip from any to any" .Dl "ipfw pipe 10 config plr 0.05" .Pp We can use pipes to artificially limit bandwidth, e.g.\& on a machine acting as a router, if we want to limit traffic from local clients on 192.168.2.0/24 we do: .Pp .Dl "ipfw add pipe 1 ip from 192.168.2.0/24 to any out" .Dl "ipfw pipe 1 config bw 300Kbit/s queue 50KBytes" .Pp note that we use the .Cm out modifier so that the rule is not used twice. Remember in fact that .Nm rules are checked both on incoming and outgoing packets. .Pp Should we want to simulate a bidirectional link with bandwidth limitations, the correct way is the following: .Pp .Dl "ipfw add pipe 1 ip from any to any out" .Dl "ipfw add pipe 2 ip from any to any in" .Dl "ipfw pipe 1 config bw 64Kbit/s queue 10Kbytes" .Dl "ipfw pipe 2 config bw 64Kbit/s queue 10Kbytes" .Pp The above can be very useful, e.g.\& if you want to see how your fancy Web page will look for a residential user who is connected only through a slow link. You should not use only one pipe for both directions, unless you want to simulate a half-duplex medium (e.g.\& AppleTalk, Ethernet, IRDA). It is not necessary that both pipes have the same configuration, so we can also simulate asymmetric links. .Pp Should we want to verify network performance with the RED queue management algorithm: .Pp .Dl "ipfw add pipe 1 ip from any to any" .Dl "ipfw pipe 1 config bw 500Kbit/s queue 100 red 0.002/30/80/0.1" .Pp Another typical application of the traffic shaper is to introduce some delay in the communication. This can significantly affect applications which do a lot of Remote Procedure Calls, and where the round-trip-time of the connection often becomes a limiting factor much more than bandwidth: .Pp .Dl "ipfw add pipe 1 ip from any to any out" .Dl "ipfw add pipe 2 ip from any to any in" .Dl "ipfw pipe 1 config delay 250ms bw 1Mbit/s" .Dl "ipfw pipe 2 config delay 250ms bw 1Mbit/s" .Pp Per-flow queueing can be useful for a variety of purposes. A very simple one is counting traffic: .Pp .Dl "ipfw add pipe 1 tcp from any to any" .Dl "ipfw add pipe 1 udp from any to any" .Dl "ipfw add pipe 1 ip from any to any" .Dl "ipfw pipe 1 config mask all" .Pp The above set of rules will create queues (and collect statistics) for all traffic. Because the pipes have no limitations, the only effect is collecting statistics. Note that we need 3 rules, not just the last one, because when .Nm tries to match IP packets it will not consider ports, so we would not see connections on separate ports as different ones. .Pp A more sophisticated example is limiting the outbound traffic on a net with per-host limits, rather than per-network limits: .Pp .Dl "ipfw add pipe 1 ip from 192.168.2.0/24 to any out" .Dl "ipfw add pipe 2 ip from any to 192.168.2.0/24 in" .Dl "ipfw pipe 1 config mask src-ip 0x000000ff bw 200Kbit/s queue 20Kbytes" .Dl "ipfw pipe 2 config mask dst-ip 0x000000ff bw 200Kbit/s queue 20Kbytes" .Ss LOOKUP TABLES In the following example, we need to create several traffic bandwidth classes and we need different hosts/networks to fall into different classes. We create one pipe for each class and configure them accordingly. Then we create a single table and fill it with IP subnets and addresses. For each subnet/host we set the argument equal to the number of the pipe that it should use. Then we classify traffic using a single rule: .Pp .Dl "ipfw pipe 1 config bw 1000Kbyte/s" .Dl "ipfw pipe 4 config bw 4000Kbyte/s" .Dl "..." .Dl "ipfw table 1 add 192.168.2.0/24 1" .Dl "ipfw table 1 add 192.168.0.0/27 4" .Dl "ipfw table 1 add 192.168.0.2 1" .Dl "..." .Dl "ipfw add pipe tablearg ip from table(1) to any" .Pp Using the .Cm fwd action, the table entries may include hostnames and IP addresses. .Pp .Dl "ipfw table 1 add 192.168.2.0/24 10.23.2.1" .Dl "ipfw table 1 add 192.168.0.0/27 router1.dmz" .Dl "..." .Dl "ipfw add 100 fwd tablearg ip from any to table(1)" .Pp In the following example per-interface firewall is created: .Pp .Dl "ipfw table 10 add vlan20 12000" .Dl "ipfw table 10 add vlan30 13000" .Dl "ipfw table 20 add vlan20 22000" .Dl "ipfw table 20 add vlan30 23000" .Dl ".." .Dl "ipfw add 100 ipfw skipto tablearg ip from any to any recv 'table(10)' in" .Dl "ipfw add 200 ipfw skipto tablearg ip from any to any xmit 'table(10)' out" .Ss SETS OF RULES To add a set of rules atomically, e.g.\& set 18: .Pp .Dl "ipfw set disable 18" .Dl "ipfw add NN set 18 ... # repeat as needed" .Dl "ipfw set enable 18" .Pp To delete a set of rules atomically the command is simply: .Pp .Dl "ipfw delete set 18" .Pp To test a ruleset and disable it and regain control if something goes wrong: .Pp .Dl "ipfw set disable 18" .Dl "ipfw add NN set 18 ... # repeat as needed" .Dl "ipfw set enable 18; echo done; sleep 30 && ipfw set disable 18" .Pp Here if everything goes well, you press control-C before the "sleep" terminates, and your ruleset will be left active. Otherwise, e.g.\& if you cannot access your box, the ruleset will be disabled after the sleep terminates thus restoring the previous situation. .Pp To show rules of the specific set: .Pp .Dl "ipfw set 18 show" .Pp To show rules of the disabled set: .Pp .Dl "ipfw -S set 18 show" .Pp To clear a specific rule counters of the specific set: .Pp .Dl "ipfw set 18 zero NN" .Pp To delete a specific rule of the specific set: .Pp .Dl "ipfw set 18 delete NN" .Ss NAT, REDIRECT AND LSNAT First redirect all the traffic to nat instance 123: .Pp .Dl "ipfw add nat 123 all from any to any" .Pp Then to configure nat instance 123 to alias all the outgoing traffic with ip 192.168.0.123, blocking all incoming connections, trying to keep same ports on both sides, clearing aliasing table on address change and keeping a log of traffic/link statistics: .Pp .Dl "ipfw nat 123 config ip 192.168.0.123 log deny_in reset same_ports" .Pp Or to change address of instance 123, aliasing table will be cleared (see reset option): .Pp .Dl "ipfw nat 123 config ip 10.0.0.1" .Pp To see configuration of nat instance 123: .Pp .Dl "ipfw nat 123 show config" .Pp To show logs of all the instances in range 111-999: .Pp .Dl "ipfw nat 111-999 show" .Pp To see configurations of all instances: .Pp .Dl "ipfw nat show config" .Pp Or a redirect rule with mixed modes could looks like: .Pp .Dl "ipfw nat 123 config redirect_addr 10.0.0.1 10.0.0.66" .Dl " redirect_port tcp 192.168.0.1:80 500" .Dl " redirect_proto udp 192.168.1.43 192.168.1.1" .Dl " redirect_addr 192.168.0.10,192.168.0.11" .Dl " 10.0.0.100 # LSNAT" .Dl " redirect_port tcp 192.168.0.1:80,192.168.0.10:22" .Dl " 500 # LSNAT" .Pp or it could be split in: .Pp .Dl "ipfw nat 1 config redirect_addr 10.0.0.1 10.0.0.66" .Dl "ipfw nat 2 config redirect_port tcp 192.168.0.1:80 500" .Dl "ipfw nat 3 config redirect_proto udp 192.168.1.43 192.168.1.1" .Dl "ipfw nat 4 config redirect_addr 192.168.0.10,192.168.0.11,192.168.0.12" .Dl " 10.0.0.100" .Dl "ipfw nat 5 config redirect_port tcp" .Dl " 192.168.0.1:80,192.168.0.10:22,192.168.0.20:25 500" .Sh SEE ALSO .Xr cpp 1 , .Xr m4 1 , .Xr altq 4 , .Xr divert 4 , .Xr dummynet 4 , .Xr if_bridge 4 , .Xr ip 4 , .Xr ipfirewall 4 , .Xr ng_ipfw 4 , .Xr protocols 5 , .Xr services 5 , .Xr init 8 , .Xr kldload 8 , .Xr reboot 8 , .Xr sysctl 8 , .Xr syslogd 8 .Sh HISTORY The .Nm utility first appeared in .Fx 2.0 . .Nm dummynet was introduced in .Fx 2.2.8 . Stateful extensions were introduced in .Fx 4.0 . .Nm ipfw2 was introduced in Summer 2002. .Sh AUTHORS .An Ugen J. S. Antsilevich , .An Poul-Henning Kamp , .An Alex Nash , .An Archie Cobbs , .An Luigi Rizzo . .Pp .An -nosplit API based upon code written by .An Daniel Boulet for BSDI. .Pp Dummynet has been introduced by Luigi Rizzo in 1997-1998. .Pp Some early work (1999-2000) on the .Nm dummynet traffic shaper supported by Akamba Corp. .Pp The ipfw core (ipfw2) has been completely redesigned and reimplemented by Luigi Rizzo in summer 2002. Further actions and options have been added by various developer over the years. .Pp .An -nosplit In-kernel NAT support written by .An Paolo Pisati Aq piso@FreeBSD.org as part of a Summer of Code 2005 project. .Pp SCTP .Nm nat support has been developed by .An The Centre for Advanced Internet Architectures (CAIA) Aq http://www.caia.swin.edu.au . The primary developers and maintainers are David Hayes and Jason But. For further information visit: .Aq http://www.caia.swin.edu.au/urp/SONATA .Pp Delay profiles have been developed by Alessandro Cerri and Luigi Rizzo, supported by the European Commission within Projects Onelab and Onelab2. .Sh BUGS The syntax has grown over the years and sometimes it might be confusing. Unfortunately, backward compatibility prevents cleaning up mistakes made in the definition of the syntax. .Pp .Em !!! WARNING !!! .Pp Misconfiguring the firewall can put your computer in an unusable state, possibly shutting down network services and requiring console access to regain control of it. .Pp Incoming packet fragments diverted by .Cm divert are reassembled before delivery to the socket. The action used on those packet is the one from the rule which matches the first fragment of the packet. .Pp Packets diverted to userland, and then reinserted by a userland process may lose various packet attributes. The packet source interface name will be preserved if it is shorter than 8 bytes and the userland process saves and reuses the sockaddr_in (as does .Xr natd 8 ) ; otherwise, it may be lost. If a packet is reinserted in this manner, later rules may be incorrectly applied, making the order of .Cm divert rules in the rule sequence very important. .Pp Dummynet drops all packets with IPv6 link-local addresses. .Pp Rules using .Cm uid or .Cm gid may not behave as expected. In particular, incoming SYN packets may have no uid or gid associated with them since they do not yet belong to a TCP connection, and the uid/gid associated with a packet may not be as expected if the associated process calls .Xr setuid 2 or similar system calls. .Pp Rule syntax is subject to the command line environment and some patterns may need to be escaped with the backslash character or quoted appropriately. .Pp Due to the architecture of .Xr libalias 3 , ipfw nat is not compatible with the TCP segmentation offloading (TSO). Thus, to reliably nat your network traffic, please disable TSO on your NICs using .Xr ifconfig 8 . .Pp ICMP error messages are not implicitly matched by dynamic rules for the respective conversations. To avoid failures of network error detection and path MTU discovery, ICMP error messages may need to be allowed explicitly through static rules. .Pp Rules using .Cm call and .Cm return actions may lead to confusing behaviour if ruleset has mistakes, and/or interaction with other subsystems (netgraph, dummynet, etc.) is used. One possible case for this is packet leaving .Nm in subroutine on the input pass, while later on output encountering unpaired .Cm return first. As the call stack is kept intact after input pass, packet will suddenly return to the rule number used on input pass, not on output one. Order of processing should be checked carefully to avoid such mistakes. Index: stable/10/sbin/ipfw/ipfw2.h =================================================================== --- stable/10/sbin/ipfw/ipfw2.h (revision 301230) +++ stable/10/sbin/ipfw/ipfw2.h (revision 301231) @@ -1,298 +1,299 @@ /* * 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 * * Redistribution and use in source forms, with and without modification, * are permitted provided that this entire comment appears intact. * * Redistribution in binary form may occur without any restrictions. * Obviously, it would be nice if you gave credit where credit is due * but requiring it would be too onerous. * * This software is provided ``AS IS'' without any warranties of any kind. * * NEW command line interface for IP firewall facility * * $FreeBSD$ */ /* * Options that can be set on the command line. * When reading commands from a file, a subset of the options can also * be applied globally by specifying them before the file name. * After that, each line can contain its own option that changes * the global value. * XXX The context is not restored after each line. */ struct cmdline_opts { /* boolean options: */ int do_value_as_ip; /* show table value as IP */ int do_resolv; /* try to resolve all ip to names */ int do_time; /* Show time stamps */ int do_quiet; /* Be quiet in add and flush */ int do_pipe; /* this cmd refers to a pipe/queue/sched */ int do_nat; /* this cmd refers to a nat config */ int do_dynamic; /* display dynamic rules */ int do_expired; /* display expired dynamic rules */ int do_compact; /* show rules in compact mode */ int do_force; /* do not ask for confirmation */ int show_sets; /* display the set each rule belongs to */ int test_only; /* only check syntax */ int comment_only; /* only print action and comment */ int verbose; /* be verbose on some commands */ /* The options below can have multiple values. */ int do_sort; /* field to sort results (0 = no) */ /* valid fields are 1 and above */ int use_set; /* work with specified set number */ /* 0 means all sets, otherwise apply to set use_set - 1 */ }; extern struct cmdline_opts co; /* * _s_x is a structure that stores a string <-> token pairs, used in * various places in the parser. Entries are stored in arrays, * with an entry with s=NULL as terminator. * The search routines are match_token() and match_value(). * Often, an element with x=0 contains an error string. * */ struct _s_x { char const *s; int x; }; enum tokens { TOK_NULL=0, TOK_OR, TOK_NOT, TOK_STARTBRACE, TOK_ENDBRACE, TOK_ACCEPT, TOK_COUNT, TOK_PIPE, TOK_LINK, TOK_QUEUE, TOK_FLOWSET, TOK_SCHED, TOK_DIVERT, TOK_TEE, TOK_NETGRAPH, TOK_NGTEE, TOK_FORWARD, TOK_SKIPTO, TOK_DENY, TOK_REJECT, TOK_RESET, TOK_UNREACH, TOK_CHECKSTATE, TOK_NAT, TOK_REASS, TOK_CALL, TOK_RETURN, TOK_ALTQ, TOK_LOG, TOK_TAG, TOK_UNTAG, TOK_TAGGED, TOK_UID, TOK_GID, TOK_JAIL, TOK_IN, TOK_LIMIT, TOK_KEEPSTATE, TOK_LAYER2, TOK_OUT, TOK_DIVERTED, TOK_DIVERTEDLOOPBACK, TOK_DIVERTEDOUTPUT, TOK_XMIT, TOK_RECV, TOK_VIA, TOK_FRAG, TOK_IPOPTS, TOK_IPLEN, TOK_IPID, TOK_IPPRECEDENCE, TOK_DSCP, TOK_IPTOS, TOK_IPTTL, TOK_IPVER, TOK_ESTAB, TOK_SETUP, TOK_TCPDATALEN, TOK_TCPFLAGS, TOK_TCPOPTS, TOK_TCPSEQ, TOK_TCPACK, TOK_TCPWIN, TOK_ICMPTYPES, TOK_MAC, TOK_MACTYPE, TOK_VERREVPATH, TOK_VERSRCREACH, TOK_ANTISPOOF, TOK_IPSEC, TOK_COMMENT, TOK_PLR, TOK_NOERROR, TOK_BUCKETS, TOK_DSTIP, TOK_SRCIP, TOK_DSTPORT, TOK_SRCPORT, TOK_ALL, TOK_MASK, TOK_FLOW_MASK, TOK_SCHED_MASK, TOK_BW, TOK_DELAY, TOK_PROFILE, TOK_BURST, TOK_RED, TOK_GRED, + TOK_ECN, TOK_DROPTAIL, TOK_PROTO, /* dummynet tokens */ TOK_WEIGHT, TOK_LMAX, TOK_PRI, TOK_TYPE, TOK_SLOTSIZE, TOK_IP, TOK_IF, TOK_ALOG, TOK_DENY_INC, TOK_SAME_PORTS, TOK_UNREG_ONLY, TOK_SKIP_GLOBAL, TOK_RESET_ADDR, TOK_ALIAS_REV, TOK_PROXY_ONLY, TOK_REDIR_ADDR, TOK_REDIR_PORT, TOK_REDIR_PROTO, TOK_IPV6, TOK_FLOWID, TOK_ICMP6TYPES, TOK_EXT6HDR, TOK_DSTIP6, TOK_SRCIP6, TOK_IPV4, TOK_UNREACH6, TOK_RESET6, TOK_FIB, TOK_SETFIB, TOK_LOOKUP, TOK_SOCKARG, TOK_SETDSCP, }; /* * the following macro returns an error message if we run out of * arguments. */ #define NEED(_p, msg) {if (!_p) errx(EX_USAGE, msg);} #define NEED1(msg) {if (!(*av)) errx(EX_USAGE, msg);} int pr_u64(uint64_t *pd, int width); /* memory allocation support */ void *safe_calloc(size_t number, size_t size); void *safe_realloc(void *ptr, size_t size); /* string comparison functions used for historical compatibility */ int _substrcmp(const char *str1, const char* str2); int _substrcmp2(const char *str1, const char* str2, const char* str3); /* utility functions */ int match_token(struct _s_x *table, char *string); char const *match_value(struct _s_x *p, int value); int do_cmd(int optname, void *optval, uintptr_t optlen); uint32_t ipfw_get_tables_max(void); struct in6_addr; void n2mask(struct in6_addr *mask, int n); int contigmask(uint8_t *p, int len); /* * Forward declarations to avoid include way too many headers. * C does not allow duplicated typedefs, so we use the base struct * that the typedef points to. * Should the typedefs use a different type, the compiler will * still detect the change when compiling the body of the * functions involved, so we do not lose error checking. */ struct _ipfw_insn; struct _ipfw_insn_altq; struct _ipfw_insn_u32; struct _ipfw_insn_ip6; struct _ipfw_insn_icmp6; /* * The reserved set numer. This is a constant in ip_fw.h * but we store it in a variable so other files do not depend * in that header just for one constant. */ extern int resvd_set_number; /* first-level command handlers */ void ipfw_add(char *av[]); void ipfw_show_nat(int ac, char **av); void ipfw_config_pipe(int ac, char **av); void ipfw_config_nat(int ac, char **av); void ipfw_sets_handler(char *av[]); void ipfw_table_handler(int ac, char *av[]); void ipfw_sysctl_handler(char *av[], int which); void ipfw_delete(char *av[]); void ipfw_flush(int force); void ipfw_zero(int ac, char *av[], int optname); void ipfw_list(int ac, char *av[], int show_counters); #ifdef PF /* altq.c */ void altq_set_enabled(int enabled); u_int32_t altq_name_to_qid(const char *name); void print_altq_cmd(struct _ipfw_insn_altq *altqptr); #else #define NO_ALTQ #endif /* dummynet.c */ void dummynet_list(int ac, char *av[], int show_counters); void dummynet_flush(void); int ipfw_delete_pipe(int pipe_or_queue, int n); /* ipv6.c */ void print_unreach6_code(uint16_t code); void print_ip6(struct _ipfw_insn_ip6 *cmd, char const *s); void print_flow6id(struct _ipfw_insn_u32 *cmd); void print_icmp6types(struct _ipfw_insn_u32 *cmd); void print_ext6hdr(struct _ipfw_insn *cmd ); struct _ipfw_insn *add_srcip6(struct _ipfw_insn *cmd, char *av, int cblen); struct _ipfw_insn *add_dstip6(struct _ipfw_insn *cmd, char *av, int cblen); void fill_flow6(struct _ipfw_insn_u32 *cmd, char *av, int cblen); void fill_unreach6_code(u_short *codep, char *str); void fill_icmp6types(struct _ipfw_insn_icmp6 *cmd, char *av, int cblen); int fill_ext6hdr(struct _ipfw_insn *cmd, char *av); Index: stable/10/sys/netinet/ip_dummynet.h =================================================================== --- stable/10/sys/netinet/ip_dummynet.h (revision 301230) +++ stable/10/sys/netinet/ip_dummynet.h (revision 301231) @@ -1,263 +1,264 @@ /*- * Copyright (c) 1998-2010 Luigi Rizzo, Universita` di Pisa * Portions Copyright (c) 2000 Akamba Corp. * All rights reserved * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _IP_DUMMYNET_H #define _IP_DUMMYNET_H /* * Definition of the kernel-userland API for dummynet. * * Setsockopt() and getsockopt() pass a batch of objects, each * of them starting with a "struct dn_id" which should fully identify * the object and its relation with others in the sequence. * The first object in each request should have * type= DN_CMD_*, id = DN_API_VERSION. * For other objects, type and subtype specify the object, len indicates * the total length including the header, and 'id' identifies the specific * object. * * Most objects are numbered with an identifier in the range 1..65535. * DN_MAX_ID indicates the first value outside the range. */ #define DN_API_VERSION 12500000 #define DN_MAX_ID 0x10000 struct dn_id { uint16_t len; /* total obj len including this header */ uint8_t type; uint8_t subtype; uint32_t id; /* generic id */ }; /* * These values are in the type field of struct dn_id. * To preserve the ABI, never rearrange the list or delete * entries with the exception of DN_LAST */ enum { DN_NONE = 0, DN_LINK = 1, DN_FS, DN_SCH, DN_SCH_I, DN_QUEUE, DN_DELAY_LINE, DN_PROFILE, DN_FLOW, /* struct dn_flow */ DN_TEXT, /* opaque text is the object */ DN_CMD_CONFIG = 0x80, /* objects follow */ DN_CMD_DELETE, /* subtype + list of entries */ DN_CMD_GET, /* subtype + list of entries */ DN_CMD_FLUSH, /* for compatibility with FreeBSD 7.2/8 */ DN_COMPAT_PIPE, DN_COMPAT_QUEUE, DN_GET_COMPAT, /* special commands for emulation of sysctl variables */ DN_SYSCTL_GET, DN_SYSCTL_SET, DN_LAST, }; enum { /* subtype for schedulers, flowset and the like */ DN_SCHED_UNKNOWN = 0, DN_SCHED_FIFO = 1, DN_SCHED_WF2QP = 2, /* others are in individual modules */ }; enum { /* user flags */ DN_HAVE_MASK = 0x0001, /* fs or sched has a mask */ DN_NOERROR = 0x0002, /* do not report errors */ DN_QHT_HASH = 0x0004, /* qht is a hash table */ DN_QSIZE_BYTES = 0x0008, /* queue size is in bytes */ DN_HAS_PROFILE = 0x0010, /* a link has a profile */ DN_IS_RED = 0x0020, DN_IS_GENTLE_RED= 0x0040, + DN_IS_ECN = 0x0080, DN_PIPE_CMD = 0x1000, /* pipe config... */ }; /* * link template. */ struct dn_link { struct dn_id oid; /* * Userland sets bw and delay in bits/s and milliseconds. * The kernel converts this back and forth to bits/tick and ticks. * XXX what about burst ? */ int32_t link_nr; int bandwidth; /* bit/s or bits/tick. */ int delay; /* ms and ticks */ uint64_t burst; /* scaled. bits*Hz XXX */ }; /* * A flowset, which is a template for flows. Contains parameters * from the command line: id, target scheduler, queue sizes, plr, * flow masks, buckets for the flow hash, and possibly scheduler- * specific parameters (weight, quantum and so on). */ struct dn_fs { struct dn_id oid; uint32_t fs_nr; /* the flowset number */ uint32_t flags; /* userland flags */ int qsize; /* queue size in slots or bytes */ int32_t plr; /* PLR, pkt loss rate (2^31-1 means 100%) */ uint32_t buckets; /* buckets used for the queue hash table */ struct ipfw_flow_id flow_mask; uint32_t sched_nr; /* the scheduler we attach to */ /* generic scheduler parameters. Leave them at -1 if unset. * Now we use 0: weight, 1: lmax, 2: priority */ int par[4]; /* RED/GRED parameters. * weight and probabilities are in the range 0..1 represented * in fixed point arithmetic with SCALE_RED decimal bits. */ #define SCALE_RED 16 #define SCALE(x) ( (x) << SCALE_RED ) #define SCALE_VAL(x) ( (x) >> SCALE_RED ) #define SCALE_MUL(x,y) ( ( (x) * (y) ) >> SCALE_RED ) int w_q ; /* queue weight (scaled) */ int max_th ; /* maximum threshold for queue (scaled) */ int min_th ; /* minimum threshold for queue (scaled) */ int max_p ; /* maximum value for p_b (scaled) */ }; /* * dn_flow collects flow_id and stats for queues and scheduler * instances, and is used to pass these info to userland. * oid.type/oid.subtype describe the object, oid.id is number * of the parent object. */ struct dn_flow { struct dn_id oid; struct ipfw_flow_id fid; uint64_t tot_pkts; /* statistics counters */ uint64_t tot_bytes; uint32_t length; /* Queue length, in packets */ uint32_t len_bytes; /* Queue length, in bytes */ uint32_t drops; }; /* * Scheduler template, mostly indicating the name, number, * sched_mask and buckets. */ struct dn_sch { struct dn_id oid; uint32_t sched_nr; /* N, scheduler number */ uint32_t buckets; /* number of buckets for the instances */ uint32_t flags; /* have_mask, ... */ char name[16]; /* null terminated */ /* mask to select the appropriate scheduler instance */ struct ipfw_flow_id sched_mask; /* M */ }; /* A delay profile is attached to a link. * Note that a profile, as any other object, cannot be longer than 2^16 */ #define ED_MAX_SAMPLES_NO 1024 struct dn_profile { struct dn_id oid; /* fields to simulate a delay profile */ #define ED_MAX_NAME_LEN 32 char name[ED_MAX_NAME_LEN]; int link_nr; int loss_level; int bandwidth; // XXX use link bandwidth? int samples_no; /* actual len of samples[] */ int samples[ED_MAX_SAMPLES_NO]; /* may be shorter */ }; /* * Overall structure of dummynet In dummynet, packets are selected with the firewall rules, and passed to two different objects: PIPE or QUEUE (bad name). A QUEUE defines a classifier, which groups packets into flows according to a 'mask', puts them into independent queues (one per flow) with configurable size and queue management policy, and passes flows to a scheduler: (flow_mask|sched_mask) sched_mask +---------+ weight Wx +-------------+ | |->-[flow]-->--| |-+ -->--| QUEUE x | ... | | | | |->-[flow]-->--| SCHEDuler N | | +---------+ | | | ... | +--[LINK N]-->-- +---------+ weight Wy | | +--[LINK N]-->-- | |->-[flow]-->--| | | -->--| QUEUE y | ... | | | | |->-[flow]-->--| | | +---------+ +-------------+ | +-------------+ Many QUEUE objects can connect to the same scheduler, each QUEUE object can have its own set of parameters. In turn, the SCHEDuler 'forks' multiple instances according to a 'sched_mask', each instance manages its own set of queues and transmits on a private instance of a configurable LINK. A PIPE is a simplified version of the above, where there is no flow_mask, and each scheduler instance handles a single queue. The following data structures (visible from userland) describe the objects used by dummynet: + dn_link, contains the main configuration parameters related to delay and bandwidth; + dn_profile describes a delay profile; + dn_flow describes the flow status (flow id, statistics) + dn_sch describes a scheduler + dn_fs describes a flowset (msk, weight, queue parameters) * */ #endif /* _IP_DUMMYNET_H */ Index: stable/10/sys/netpfil/ipfw/ip_dn_io.c =================================================================== --- stable/10/sys/netpfil/ipfw/ip_dn_io.c (revision 301230) +++ stable/10/sys/netpfil/ipfw/ip_dn_io.c (revision 301231) @@ -1,854 +1,924 @@ /*- * Copyright (c) 2010 Luigi Rizzo, Riccardo Panicucci, Universita` di Pisa * All rights reserved * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * Dummynet portions related to packet handling. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* IFNAMSIZ, struct ifaddr, ifq head, lock.h mutex.h */ #include #include #include #include /* ip_len, ip_off */ #include /* ip_output(), IP_FORWARDING */ #include #include #include /* various ether_* routines */ #include /* for ip6_input, ip6_output prototypes */ #include #include #include #include #include /* * We keep a private variable for the simulation time, but we could * probably use an existing one ("softticks" in sys/kern/kern_timeout.c) * instead of dn_cfg.curr_time */ struct dn_parms dn_cfg; //VNET_DEFINE(struct dn_parms, _base_dn_cfg); static long tick_last; /* Last tick duration (usec). */ static long tick_delta; /* Last vs standard tick diff (usec). */ static long tick_delta_sum; /* Accumulated tick difference (usec).*/ static long tick_adjustment; /* Tick adjustments done. */ static long tick_lost; /* Lost(coalesced) ticks number. */ /* Adjusted vs non-adjusted curr_time difference (ticks). */ static long tick_diff; static unsigned long io_pkt; static unsigned long io_pkt_fast; static unsigned long io_pkt_drop; /* * We use a heap to store entities for which we have pending timer events. * The heap is checked at every tick and all entities with expired events * are extracted. */ MALLOC_DEFINE(M_DUMMYNET, "dummynet", "dummynet heap"); extern void (*bridge_dn_p)(struct mbuf *, struct ifnet *); #ifdef SYSCTL_NODE /* * Because of the way the SYSBEGIN/SYSEND macros work on other * platforms, there should not be functions between them. * So keep the handlers outside the block. */ static int sysctl_hash_size(SYSCTL_HANDLER_ARGS) { int error, value; value = dn_cfg.hash_size; error = sysctl_handle_int(oidp, &value, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (value < 16 || value > 65536) return (EINVAL); dn_cfg.hash_size = value; return (0); } static int sysctl_limits(SYSCTL_HANDLER_ARGS) { int error; long value; if (arg2 != 0) value = dn_cfg.slot_limit; else value = dn_cfg.byte_limit; error = sysctl_handle_long(oidp, &value, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (arg2 != 0) { if (value < 1) return (EINVAL); dn_cfg.slot_limit = value; } else { if (value < 1500) return (EINVAL); dn_cfg.byte_limit = value; } return (0); } SYSBEGIN(f4) SYSCTL_DECL(_net_inet); SYSCTL_DECL(_net_inet_ip); static SYSCTL_NODE(_net_inet_ip, OID_AUTO, dummynet, CTLFLAG_RW, 0, "Dummynet"); /* wrapper to pass dn_cfg fields to SYSCTL_* */ //#define DC(x) (&(VNET_NAME(_base_dn_cfg).x)) #define DC(x) (&(dn_cfg.x)) /* parameters */ SYSCTL_PROC(_net_inet_ip_dummynet, OID_AUTO, hash_size, CTLTYPE_INT | CTLFLAG_RW, 0, 0, sysctl_hash_size, "I", "Default hash table size"); SYSCTL_PROC(_net_inet_ip_dummynet, OID_AUTO, pipe_slot_limit, CTLTYPE_LONG | CTLFLAG_RW, 0, 1, sysctl_limits, "L", "Upper limit in slots for pipe queue."); SYSCTL_PROC(_net_inet_ip_dummynet, OID_AUTO, pipe_byte_limit, CTLTYPE_LONG | CTLFLAG_RW, 0, 0, sysctl_limits, "L", "Upper limit in bytes for pipe queue."); SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, io_fast, CTLFLAG_RW, DC(io_fast), 0, "Enable fast dummynet io."); SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, debug, CTLFLAG_RW, DC(debug), 0, "Dummynet debug level"); /* RED parameters */ SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_lookup_depth, CTLFLAG_RD, DC(red_lookup_depth), 0, "Depth of RED lookup table"); SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_avg_pkt_size, CTLFLAG_RD, DC(red_avg_pkt_size), 0, "RED Medium packet size"); SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_max_pkt_size, CTLFLAG_RD, DC(red_max_pkt_size), 0, "RED Max packet size"); /* time adjustment */ SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_delta, CTLFLAG_RD, &tick_delta, 0, "Last vs standard tick difference (usec)."); SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_delta_sum, CTLFLAG_RD, &tick_delta_sum, 0, "Accumulated tick difference (usec)."); SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_adjustment, CTLFLAG_RD, &tick_adjustment, 0, "Tick adjustments done."); SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_diff, CTLFLAG_RD, &tick_diff, 0, "Adjusted vs non-adjusted curr_time difference (ticks)."); SYSCTL_LONG(_net_inet_ip_dummynet, OID_AUTO, tick_lost, CTLFLAG_RD, &tick_lost, 0, "Number of ticks coalesced by dummynet taskqueue."); /* Drain parameters */ SYSCTL_UINT(_net_inet_ip_dummynet, OID_AUTO, expire, CTLFLAG_RW, DC(expire), 0, "Expire empty queues/pipes"); SYSCTL_UINT(_net_inet_ip_dummynet, OID_AUTO, expire_cycle, CTLFLAG_RD, DC(expire_cycle), 0, "Expire cycle for queues/pipes"); /* statistics */ SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, schk_count, CTLFLAG_RD, DC(schk_count), 0, "Number of schedulers"); SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, si_count, CTLFLAG_RD, DC(si_count), 0, "Number of scheduler instances"); SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, fsk_count, CTLFLAG_RD, DC(fsk_count), 0, "Number of flowsets"); SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, queue_count, CTLFLAG_RD, DC(queue_count), 0, "Number of queues"); SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt, CTLFLAG_RD, &io_pkt, 0, "Number of packets passed to dummynet."); SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt_fast, CTLFLAG_RD, &io_pkt_fast, 0, "Number of packets bypassed dummynet scheduler."); SYSCTL_ULONG(_net_inet_ip_dummynet, OID_AUTO, io_pkt_drop, CTLFLAG_RD, &io_pkt_drop, 0, "Number of packets dropped by dummynet."); #undef DC SYSEND #endif static void dummynet_send(struct mbuf *); /* * Packets processed by dummynet have an mbuf tag associated with * them that carries their dummynet state. * Outside dummynet, only the 'rule' field is relevant, and it must * be at the beginning of the structure. */ struct dn_pkt_tag { struct ipfw_rule_ref rule; /* matching rule */ /* second part, dummynet specific */ int dn_dir; /* action when packet comes out.*/ /* see ip_fw_private.h */ uint64_t output_time; /* when the pkt is due for delivery*/ struct ifnet *ifp; /* interface, for ip_output */ struct _ip6dn_args ip6opt; /* XXX ipv6 options */ }; /* * Return the mbuf tag holding the dummynet state (it should * be the first one on the list). */ static struct dn_pkt_tag * dn_tag_get(struct mbuf *m) { struct m_tag *mtag = m_tag_first(m); KASSERT(mtag != NULL && mtag->m_tag_cookie == MTAG_ABI_COMPAT && mtag->m_tag_id == PACKET_TAG_DUMMYNET, ("packet on dummynet queue w/o dummynet tag!")); return (struct dn_pkt_tag *)(mtag+1); } static inline void mq_append(struct mq *q, struct mbuf *m) { if (q->head == NULL) q->head = m; else q->tail->m_nextpkt = m; q->tail = m; m->m_nextpkt = NULL; } /* * Dispose a list of packet. Use a functions so if we need to do * more work, this is a central point to do it. */ void dn_free_pkts(struct mbuf *mnext) { struct mbuf *m; while ((m = mnext) != NULL) { mnext = m->m_nextpkt; FREE_PKT(m); } } static int red_drops (struct dn_queue *q, int len) { /* * RED algorithm * * RED calculates the average queue size (avg) using a low-pass filter * with an exponential weighted (w_q) moving average: * avg <- (1-w_q) * avg + w_q * q_size * where q_size is the queue length (measured in bytes or * packets). * * If q_size == 0, we compute the idle time for the link, and set * avg = (1 - w_q)^(idle/s) * where s is the time needed for transmitting a medium-sized packet. * * Now, if avg < min_th the packet is enqueued. * If avg > max_th the packet is dropped. Otherwise, the packet is * dropped with probability P function of avg. */ struct dn_fsk *fs = q->fs; int64_t p_b = 0; /* Queue in bytes or packets? */ uint32_t q_size = (fs->fs.flags & DN_QSIZE_BYTES) ? q->ni.len_bytes : q->ni.length; /* Average queue size estimation. */ if (q_size != 0) { /* Queue is not empty, avg <- avg + (q_size - avg) * w_q */ int diff = SCALE(q_size) - q->avg; int64_t v = SCALE_MUL((int64_t)diff, (int64_t)fs->w_q); q->avg += (int)v; } else { /* * Queue is empty, find for how long the queue has been * empty and use a lookup table for computing * (1 - * w_q)^(idle_time/s) where s is the time to send a * (small) packet. * XXX check wraps... */ if (q->avg) { u_int t = div64((dn_cfg.curr_time - q->q_time), fs->lookup_step); q->avg = (t < fs->lookup_depth) ? SCALE_MUL(q->avg, fs->w_q_lookup[t]) : 0; } } /* Should i drop? */ if (q->avg < fs->min_th) { q->count = -1; return (0); /* accept packet */ } if (q->avg >= fs->max_th) { /* average queue >= max threshold */ + if (fs->fs.flags & DN_IS_ECN) + return (1); if (fs->fs.flags & DN_IS_GENTLE_RED) { /* * According to Gentle-RED, if avg is greater than * max_th the packet is dropped with a probability * p_b = c_3 * avg - c_4 * where c_3 = (1 - max_p) / max_th * c_4 = 1 - 2 * max_p */ p_b = SCALE_MUL((int64_t)fs->c_3, (int64_t)q->avg) - fs->c_4; } else { q->count = -1; return (1); } } else if (q->avg > fs->min_th) { + if (fs->fs.flags & DN_IS_ECN) + return (1); /* * We compute p_b using the linear dropping function * p_b = c_1 * avg - c_2 * where c_1 = max_p / (max_th - min_th) * c_2 = max_p * min_th / (max_th - min_th) */ p_b = SCALE_MUL((int64_t)fs->c_1, (int64_t)q->avg) - fs->c_2; } if (fs->fs.flags & DN_QSIZE_BYTES) p_b = div64((p_b * len) , fs->max_pkt_size); if (++q->count == 0) q->random = random() & 0xffff; else { /* * q->count counts packets arrived since last drop, so a greater * value of q->count means a greater packet drop probability. */ if (SCALE_MUL(p_b, SCALE((int64_t)q->count)) > q->random) { q->count = 0; /* After a drop we calculate a new random value. */ q->random = random() & 0xffff; return (1); /* drop */ } } /* End of RED algorithm. */ return (0); /* accept */ } /* + * ECN/ECT Processing (partially adopted from altq) + */ +static int +ecn_mark(struct mbuf* m) +{ + struct ip *ip; + ip = mtod(m, struct ip *); + + switch (ip->ip_v) { + case IPVERSION: + { + u_int8_t otos; + int sum; + + if ((ip->ip_tos & IPTOS_ECN_MASK) == IPTOS_ECN_NOTECT) + return (0); /* not-ECT */ + if ((ip->ip_tos & IPTOS_ECN_MASK) == IPTOS_ECN_CE) + return (1); /* already marked */ + + /* + * ecn-capable but not marked, + * mark CE and update checksum + */ + otos = ip->ip_tos; + ip->ip_tos |= IPTOS_ECN_CE; + /* + * update checksum (from RFC1624) + * HC' = ~(~HC + ~m + m') + */ + sum = ~ntohs(ip->ip_sum) & 0xffff; + sum += (~otos & 0xffff) + ip->ip_tos; + sum = (sum >> 16) + (sum & 0xffff); + sum += (sum >> 16); /* add carry */ + ip->ip_sum = htons(~sum & 0xffff); + return (1); + } +#ifdef INET6 + case (IPV6_VERSION >> 4): + { + struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); + u_int32_t flowlabel; + + flowlabel = ntohl(ip6->ip6_flow); + if ((flowlabel >> 28) != 6) + return (0); /* version mismatch! */ + if ((flowlabel & (IPTOS_ECN_MASK << 20)) == + (IPTOS_ECN_NOTECT << 20)) + return (0); /* not-ECT */ + if ((flowlabel & (IPTOS_ECN_MASK << 20)) == + (IPTOS_ECN_CE << 20)) + return (1); /* already marked */ + /* + * ecn-capable but not marked, mark CE + */ + flowlabel |= (IPTOS_ECN_CE << 20); + ip6->ip6_flow = htonl(flowlabel); + return (1); + } +#endif + } + return (0); +} + +/* * Enqueue a packet in q, subject to space and queue management policy * (whose parameters are in q->fs). * Update stats for the queue and the scheduler. * Return 0 on success, 1 on drop. The packet is consumed anyways. */ int dn_enqueue(struct dn_queue *q, struct mbuf* m, int drop) { struct dn_fs *f; struct dn_flow *ni; /* stats for scheduler instance */ uint64_t len; if (q->fs == NULL || q->_si == NULL) { printf("%s fs %p si %p, dropping\n", __FUNCTION__, q->fs, q->_si); FREE_PKT(m); return 1; } f = &(q->fs->fs); ni = &q->_si->ni; len = m->m_pkthdr.len; /* Update statistics, then check reasons to drop pkt. */ q->ni.tot_bytes += len; q->ni.tot_pkts++; ni->tot_bytes += len; ni->tot_pkts++; if (drop) goto drop; if (f->plr && random() < f->plr) goto drop; - if (f->flags & DN_IS_RED && red_drops(q, m->m_pkthdr.len)) - goto drop; + if (f->flags & DN_IS_RED && red_drops(q, m->m_pkthdr.len)) { + if (!(f->flags & DN_IS_ECN) || !ecn_mark(m)) + goto drop; + } if (f->flags & DN_QSIZE_BYTES) { if (q->ni.len_bytes > f->qsize) goto drop; } else if (q->ni.length >= f->qsize) { goto drop; } mq_append(&q->mq, m); q->ni.length++; q->ni.len_bytes += len; ni->length++; ni->len_bytes += len; - return 0; + return (0); drop: io_pkt_drop++; q->ni.drops++; ni->drops++; FREE_PKT(m); - return 1; + return (1); } /* * Fetch packets from the delay line which are due now. If there are * leftover packets, reinsert the delay line in the heap. * Runs under scheduler lock. */ static void transmit_event(struct mq *q, struct delay_line *dline, uint64_t now) { struct mbuf *m; struct dn_pkt_tag *pkt = NULL; dline->oid.subtype = 0; /* not in heap */ while ((m = dline->mq.head) != NULL) { pkt = dn_tag_get(m); if (!DN_KEY_LEQ(pkt->output_time, now)) break; dline->mq.head = m->m_nextpkt; mq_append(q, m); } if (m != NULL) { dline->oid.subtype = 1; /* in heap */ heap_insert(&dn_cfg.evheap, pkt->output_time, dline); } } /* * Convert the additional MAC overheads/delays into an equivalent * number of bits for the given data rate. The samples are * in milliseconds so we need to divide by 1000. */ static uint64_t extra_bits(struct mbuf *m, struct dn_schk *s) { int index; uint64_t bits; struct dn_profile *pf = s->profile; if (!pf || pf->samples_no == 0) return 0; index = random() % pf->samples_no; bits = div64((uint64_t)pf->samples[index] * s->link.bandwidth, 1000); if (index >= pf->loss_level) { struct dn_pkt_tag *dt = dn_tag_get(m); if (dt) dt->dn_dir = DIR_DROP; } return bits; } /* * Send traffic from a scheduler instance due by 'now'. * Return a pointer to the head of the queue. */ static struct mbuf * serve_sched(struct mq *q, struct dn_sch_inst *si, uint64_t now) { struct mq def_q; struct dn_schk *s = si->sched; struct mbuf *m = NULL; int delay_line_idle = (si->dline.mq.head == NULL); int done, bw; if (q == NULL) { q = &def_q; q->head = NULL; } bw = s->link.bandwidth; si->kflags &= ~DN_ACTIVE; if (bw > 0) si->credit += (now - si->sched_time) * bw; else si->credit = 0; si->sched_time = now; done = 0; while (si->credit >= 0 && (m = s->fp->dequeue(si)) != NULL) { uint64_t len_scaled; done++; len_scaled = (bw == 0) ? 0 : hz * (m->m_pkthdr.len * 8 + extra_bits(m, s)); si->credit -= len_scaled; /* Move packet in the delay line */ dn_tag_get(m)->output_time = dn_cfg.curr_time + s->link.delay ; mq_append(&si->dline.mq, m); } /* * If credit >= 0 the instance is idle, mark time. * Otherwise put back in the heap, and adjust the output * time of the last inserted packet, m, which was too early. */ if (si->credit >= 0) { si->idle_time = now; } else { uint64_t t; KASSERT (bw > 0, ("bw=0 and credit<0 ?")); t = div64(bw - 1 - si->credit, bw); if (m) dn_tag_get(m)->output_time += t; si->kflags |= DN_ACTIVE; heap_insert(&dn_cfg.evheap, now + t, si); } if (delay_line_idle && done) transmit_event(q, &si->dline, now); return q->head; } /* * The timer handler for dummynet. Time is computed in ticks, but * but the code is tolerant to the actual rate at which this is called. * Once complete, the function reschedules itself for the next tick. */ void dummynet_task(void *context, int pending) { struct timeval t; struct mq q = { NULL, NULL }; /* queue to accumulate results */ CURVNET_SET((struct vnet *)context); DN_BH_WLOCK(); /* Update number of lost(coalesced) ticks. */ tick_lost += pending - 1; getmicrouptime(&t); /* Last tick duration (usec). */ tick_last = (t.tv_sec - dn_cfg.prev_t.tv_sec) * 1000000 + (t.tv_usec - dn_cfg.prev_t.tv_usec); /* Last tick vs standard tick difference (usec). */ tick_delta = (tick_last * hz - 1000000) / hz; /* Accumulated tick difference (usec). */ tick_delta_sum += tick_delta; dn_cfg.prev_t = t; /* * Adjust curr_time if the accumulated tick difference is * greater than the 'standard' tick. Since curr_time should * be monotonically increasing, we do positive adjustments * as required, and throttle curr_time in case of negative * adjustment. */ dn_cfg.curr_time++; if (tick_delta_sum - tick >= 0) { int diff = tick_delta_sum / tick; dn_cfg.curr_time += diff; tick_diff += diff; tick_delta_sum %= tick; tick_adjustment++; } else if (tick_delta_sum + tick <= 0) { dn_cfg.curr_time--; tick_diff--; tick_delta_sum += tick; tick_adjustment++; } /* serve pending events, accumulate in q */ for (;;) { struct dn_id *p; /* generic parameter to handler */ if (dn_cfg.evheap.elements == 0 || DN_KEY_LT(dn_cfg.curr_time, HEAP_TOP(&dn_cfg.evheap)->key)) break; p = HEAP_TOP(&dn_cfg.evheap)->object; heap_extract(&dn_cfg.evheap, NULL); if (p->type == DN_SCH_I) { serve_sched(&q, (struct dn_sch_inst *)p, dn_cfg.curr_time); } else { /* extracted a delay line */ transmit_event(&q, (struct delay_line *)p, dn_cfg.curr_time); } } if (dn_cfg.expire && ++dn_cfg.expire_cycle >= dn_cfg.expire) { dn_cfg.expire_cycle = 0; dn_drain_scheduler(); dn_drain_queue(); } dn_reschedule(); DN_BH_WUNLOCK(); if (q.head != NULL) dummynet_send(q.head); CURVNET_RESTORE(); } /* * forward a chain of packets to the proper destination. * This runs outside the dummynet lock. */ static void dummynet_send(struct mbuf *m) { struct mbuf *n; for (; m != NULL; m = n) { struct ifnet *ifp = NULL; /* gcc 3.4.6 complains */ struct m_tag *tag; int dst; n = m->m_nextpkt; m->m_nextpkt = NULL; tag = m_tag_first(m); if (tag == NULL) { /* should not happen */ dst = DIR_DROP; } else { struct dn_pkt_tag *pkt = dn_tag_get(m); /* extract the dummynet info, rename the tag * to carry reinject info. */ dst = pkt->dn_dir; ifp = pkt->ifp; tag->m_tag_cookie = MTAG_IPFW_RULE; tag->m_tag_id = 0; } switch (dst) { case DIR_OUT: ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL); break ; case DIR_IN : netisr_dispatch(NETISR_IP, m); break; #ifdef INET6 case DIR_IN | PROTO_IPV6: netisr_dispatch(NETISR_IPV6, m); break; case DIR_OUT | PROTO_IPV6: ip6_output(m, NULL, NULL, IPV6_FORWARDING, NULL, NULL, NULL); break; #endif case DIR_FWD | PROTO_IFB: /* DN_TO_IFB_FWD: */ if (bridge_dn_p != NULL) ((*bridge_dn_p)(m, ifp)); else printf("dummynet: if_bridge not loaded\n"); break; case DIR_IN | PROTO_LAYER2: /* DN_TO_ETH_DEMUX: */ /* * The Ethernet code assumes the Ethernet header is * contiguous in the first mbuf header. * Insure this is true. */ if (m->m_len < ETHER_HDR_LEN && (m = m_pullup(m, ETHER_HDR_LEN)) == NULL) { printf("dummynet/ether: pullup failed, " "dropping packet\n"); break; } ether_demux(m->m_pkthdr.rcvif, m); break; case DIR_OUT | PROTO_LAYER2: /* N_TO_ETH_OUT: */ ether_output_frame(ifp, m); break; case DIR_DROP: /* drop the packet after some time */ FREE_PKT(m); break; default: printf("dummynet: bad switch %d!\n", dst); FREE_PKT(m); break; } } } static inline int tag_mbuf(struct mbuf *m, int dir, struct ip_fw_args *fwa) { struct dn_pkt_tag *dt; struct m_tag *mtag; mtag = m_tag_get(PACKET_TAG_DUMMYNET, sizeof(*dt), M_NOWAIT | M_ZERO); if (mtag == NULL) return 1; /* Cannot allocate packet header. */ m_tag_prepend(m, mtag); /* Attach to mbuf chain. */ dt = (struct dn_pkt_tag *)(mtag + 1); dt->rule = fwa->rule; dt->rule.info &= IPFW_ONEPASS; /* only keep this info */ dt->dn_dir = dir; dt->ifp = fwa->oif; /* dt->output tame is updated as we move through */ dt->output_time = dn_cfg.curr_time; return 0; } /* * dummynet hook for packets. * We use the argument to locate the flowset fs and the sched_set sch * associated to it. The we apply flow_mask and sched_mask to * determine the queue and scheduler instances. * * dir where shall we send the packet after dummynet. * *m0 the mbuf with the packet * ifp the 'ifp' parameter from the caller. * NULL in ip_input, destination interface in ip_output, */ int dummynet_io(struct mbuf **m0, int dir, struct ip_fw_args *fwa) { struct mbuf *m = *m0; struct dn_fsk *fs = NULL; struct dn_sch_inst *si; struct dn_queue *q = NULL; /* default */ int fs_id = (fwa->rule.info & IPFW_INFO_MASK) + ((fwa->rule.info & IPFW_IS_PIPE) ? 2*DN_MAX_ID : 0); DN_BH_WLOCK(); io_pkt++; /* we could actually tag outside the lock, but who cares... */ if (tag_mbuf(m, dir, fwa)) goto dropit; if (dn_cfg.busy) { /* if the upper half is busy doing something expensive, * lets queue the packet and move forward */ mq_append(&dn_cfg.pending, m); m = *m0 = NULL; /* consumed */ goto done; /* already active, nothing to do */ } /* XXX locate_flowset could be optimised with a direct ref. */ fs = dn_ht_find(dn_cfg.fshash, fs_id, 0, NULL); if (fs == NULL) goto dropit; /* This queue/pipe does not exist! */ if (fs->sched == NULL) /* should not happen */ goto dropit; /* find scheduler instance, possibly applying sched_mask */ si = ipdn_si_find(fs->sched, &(fwa->f_id)); if (si == NULL) goto dropit; /* * If the scheduler supports multiple queues, find the right one * (otherwise it will be ignored by enqueue). */ if (fs->sched->fp->flags & DN_MULTIQUEUE) { q = ipdn_q_find(fs, si, &(fwa->f_id)); if (q == NULL) goto dropit; } if (fs->sched->fp->enqueue(si, q, m)) { /* packet was dropped by enqueue() */ m = *m0 = NULL; goto dropit; } if (si->kflags & DN_ACTIVE) { m = *m0 = NULL; /* consumed */ goto done; /* already active, nothing to do */ } /* compute the initial allowance */ if (si->idle_time < dn_cfg.curr_time) { /* Do this only on the first packet on an idle pipe */ struct dn_link *p = &fs->sched->link; si->sched_time = dn_cfg.curr_time; si->credit = dn_cfg.io_fast ? p->bandwidth : 0; if (p->burst) { uint64_t burst = (dn_cfg.curr_time - si->idle_time) * p->bandwidth; if (burst > p->burst) burst = p->burst; si->credit += burst; } } /* pass through scheduler and delay line */ m = serve_sched(NULL, si, dn_cfg.curr_time); /* optimization -- pass it back to ipfw for immediate send */ /* XXX Don't call dummynet_send() if scheduler return the packet * just enqueued. This avoid a lock order reversal. * */ if (/*dn_cfg.io_fast &&*/ m == *m0 && (dir & PROTO_LAYER2) == 0 ) { /* fast io, rename the tag * to carry reinject info. */ struct m_tag *tag = m_tag_first(m); tag->m_tag_cookie = MTAG_IPFW_RULE; tag->m_tag_id = 0; io_pkt_fast++; if (m->m_nextpkt != NULL) { printf("dummynet: fast io: pkt chain detected!\n"); m->m_nextpkt = NULL; } m = NULL; } else { *m0 = NULL; } done: DN_BH_WUNLOCK(); if (m) dummynet_send(m); return 0; dropit: io_pkt_drop++; DN_BH_WUNLOCK(); if (m) FREE_PKT(m); *m0 = NULL; return (fs && (fs->fs.flags & DN_NOERROR)) ? 0 : ENOBUFS; } Index: stable/10/sys/netpfil/ipfw/ip_dummynet.c =================================================================== --- stable/10/sys/netpfil/ipfw/ip_dummynet.c (revision 301230) +++ stable/10/sys/netpfil/ipfw/ip_dummynet.c (revision 301231) @@ -1,2317 +1,2320 @@ /*- * Copyright (c) 1998-2002,2010 Luigi Rizzo, Universita` di Pisa * Portions Copyright (c) 2000 Akamba Corp. * All rights reserved * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); /* * Configuration and internal object management for dummynet. */ #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* IFNAMSIZ, struct ifaddr, ifq head, lock.h mutex.h */ #include #include /* ip_output(), IP_FORWARDING */ #include #include #include #include #include #include /* which objects to copy */ #define DN_C_LINK 0x01 #define DN_C_SCH 0x02 #define DN_C_FLOW 0x04 #define DN_C_FS 0x08 #define DN_C_QUEUE 0x10 /* we use this argument in case of a schk_new */ struct schk_new_arg { struct dn_alg *fp; struct dn_sch *sch; }; /*---- callout hooks. ----*/ static struct callout dn_timeout; static int dn_gone; static struct task dn_task; static struct taskqueue *dn_tq = NULL; static void dummynet(void *arg) { (void)arg; /* UNUSED */ taskqueue_enqueue_fast(dn_tq, &dn_task); } void dn_reschedule(void) { if (dn_gone != 0) return; callout_reset_sbt(&dn_timeout, tick_sbt, 0, dummynet, NULL, C_HARDCLOCK | C_DIRECT_EXEC); } /*----- end of callout hooks -----*/ /* Return a scheduler descriptor given the type or name. */ static struct dn_alg * find_sched_type(int type, char *name) { struct dn_alg *d; SLIST_FOREACH(d, &dn_cfg.schedlist, next) { if (d->type == type || (name && !strcasecmp(d->name, name))) return d; } return NULL; /* not found */ } int ipdn_bound_var(int *v, int dflt, int lo, int hi, const char *msg) { int oldv = *v; const char *op = NULL; if (dflt < lo) dflt = lo; if (dflt > hi) dflt = hi; if (oldv < lo) { *v = dflt; op = "Bump"; } else if (oldv > hi) { *v = hi; op = "Clamp"; } else return *v; if (op && msg) printf("%s %s to %d (was %d)\n", op, msg, *v, oldv); return *v; } /*---- flow_id mask, hash and compare functions ---*/ /* * The flow_id includes the 5-tuple, the queue/pipe number * which we store in the extra area in host order, * and for ipv6 also the flow_id6. * XXX see if we want the tos byte (can store in 'flags') */ static struct ipfw_flow_id * flow_id_mask(struct ipfw_flow_id *mask, struct ipfw_flow_id *id) { int is_v6 = IS_IP6_FLOW_ID(id); id->dst_port &= mask->dst_port; id->src_port &= mask->src_port; id->proto &= mask->proto; id->extra &= mask->extra; if (is_v6) { APPLY_MASK(&id->dst_ip6, &mask->dst_ip6); APPLY_MASK(&id->src_ip6, &mask->src_ip6); id->flow_id6 &= mask->flow_id6; } else { id->dst_ip &= mask->dst_ip; id->src_ip &= mask->src_ip; } return id; } /* computes an OR of two masks, result in dst and also returned */ static struct ipfw_flow_id * flow_id_or(struct ipfw_flow_id *src, struct ipfw_flow_id *dst) { int is_v6 = IS_IP6_FLOW_ID(dst); dst->dst_port |= src->dst_port; dst->src_port |= src->src_port; dst->proto |= src->proto; dst->extra |= src->extra; if (is_v6) { #define OR_MASK(_d, _s) \ (_d)->__u6_addr.__u6_addr32[0] |= (_s)->__u6_addr.__u6_addr32[0]; \ (_d)->__u6_addr.__u6_addr32[1] |= (_s)->__u6_addr.__u6_addr32[1]; \ (_d)->__u6_addr.__u6_addr32[2] |= (_s)->__u6_addr.__u6_addr32[2]; \ (_d)->__u6_addr.__u6_addr32[3] |= (_s)->__u6_addr.__u6_addr32[3]; OR_MASK(&dst->dst_ip6, &src->dst_ip6); OR_MASK(&dst->src_ip6, &src->src_ip6); #undef OR_MASK dst->flow_id6 |= src->flow_id6; } else { dst->dst_ip |= src->dst_ip; dst->src_ip |= src->src_ip; } return dst; } static int nonzero_mask(struct ipfw_flow_id *m) { if (m->dst_port || m->src_port || m->proto || m->extra) return 1; if (IS_IP6_FLOW_ID(m)) { return m->dst_ip6.__u6_addr.__u6_addr32[0] || m->dst_ip6.__u6_addr.__u6_addr32[1] || m->dst_ip6.__u6_addr.__u6_addr32[2] || m->dst_ip6.__u6_addr.__u6_addr32[3] || m->src_ip6.__u6_addr.__u6_addr32[0] || m->src_ip6.__u6_addr.__u6_addr32[1] || m->src_ip6.__u6_addr.__u6_addr32[2] || m->src_ip6.__u6_addr.__u6_addr32[3] || m->flow_id6; } else { return m->dst_ip || m->src_ip; } } /* XXX we may want a better hash function */ static uint32_t flow_id_hash(struct ipfw_flow_id *id) { uint32_t i; if (IS_IP6_FLOW_ID(id)) { uint32_t *d = (uint32_t *)&id->dst_ip6; uint32_t *s = (uint32_t *)&id->src_ip6; i = (d[0] ) ^ (d[1]) ^ (d[2] ) ^ (d[3]) ^ (d[0] >> 15) ^ (d[1] >> 15) ^ (d[2] >> 15) ^ (d[3] >> 15) ^ (s[0] << 1) ^ (s[1] << 1) ^ (s[2] << 1) ^ (s[3] << 1) ^ (s[0] << 16) ^ (s[1] << 16) ^ (s[2] << 16) ^ (s[3] << 16) ^ (id->dst_port << 1) ^ (id->src_port) ^ (id->extra) ^ (id->proto ) ^ (id->flow_id6); } else { i = (id->dst_ip) ^ (id->dst_ip >> 15) ^ (id->src_ip << 1) ^ (id->src_ip >> 16) ^ (id->extra) ^ (id->dst_port << 1) ^ (id->src_port) ^ (id->proto); } return i; } /* Like bcmp, returns 0 if ids match, 1 otherwise. */ static int flow_id_cmp(struct ipfw_flow_id *id1, struct ipfw_flow_id *id2) { int is_v6 = IS_IP6_FLOW_ID(id1); if (!is_v6) { if (IS_IP6_FLOW_ID(id2)) return 1; /* different address families */ return (id1->dst_ip == id2->dst_ip && id1->src_ip == id2->src_ip && id1->dst_port == id2->dst_port && id1->src_port == id2->src_port && id1->proto == id2->proto && id1->extra == id2->extra) ? 0 : 1; } /* the ipv6 case */ return ( !bcmp(&id1->dst_ip6,&id2->dst_ip6, sizeof(id1->dst_ip6)) && !bcmp(&id1->src_ip6,&id2->src_ip6, sizeof(id1->src_ip6)) && id1->dst_port == id2->dst_port && id1->src_port == id2->src_port && id1->proto == id2->proto && id1->extra == id2->extra && id1->flow_id6 == id2->flow_id6) ? 0 : 1; } /*--------- end of flow-id mask, hash and compare ---------*/ /*--- support functions for the qht hashtable ---- * Entries are hashed by flow-id */ static uint32_t q_hash(uintptr_t key, int flags, void *arg) { /* compute the hash slot from the flow id */ struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ? &((struct dn_queue *)key)->ni.fid : (struct ipfw_flow_id *)key; return flow_id_hash(id); } static int q_match(void *obj, uintptr_t key, int flags, void *arg) { struct dn_queue *o = (struct dn_queue *)obj; struct ipfw_flow_id *id2; if (flags & DNHT_KEY_IS_OBJ) { /* compare pointers */ id2 = &((struct dn_queue *)key)->ni.fid; } else { id2 = (struct ipfw_flow_id *)key; } return (0 == flow_id_cmp(&o->ni.fid, id2)); } /* * create a new queue instance for the given 'key'. */ static void * q_new(uintptr_t key, int flags, void *arg) { struct dn_queue *q, *template = arg; struct dn_fsk *fs = template->fs; int size = sizeof(*q) + fs->sched->fp->q_datalen; q = malloc(size, M_DUMMYNET, M_NOWAIT | M_ZERO); if (q == NULL) { D("no memory for new queue"); return NULL; } set_oid(&q->ni.oid, DN_QUEUE, size); if (fs->fs.flags & DN_QHT_HASH) q->ni.fid = *(struct ipfw_flow_id *)key; q->fs = fs; q->_si = template->_si; q->_si->q_count++; if (fs->sched->fp->new_queue) fs->sched->fp->new_queue(q); dn_cfg.queue_count++; return q; } /* * Notify schedulers that a queue is going away. * If (flags & DN_DESTROY), also free the packets. * The version for callbacks is called q_delete_cb(). */ static void dn_delete_queue(struct dn_queue *q, int flags) { struct dn_fsk *fs = q->fs; // D("fs %p si %p\n", fs, q->_si); /* notify the parent scheduler that the queue is going away */ if (fs && fs->sched->fp->free_queue) fs->sched->fp->free_queue(q); q->_si->q_count--; q->_si = NULL; if (flags & DN_DESTROY) { if (q->mq.head) dn_free_pkts(q->mq.head); bzero(q, sizeof(*q)); // safety free(q, M_DUMMYNET); dn_cfg.queue_count--; } } static int q_delete_cb(void *q, void *arg) { int flags = (int)(uintptr_t)arg; dn_delete_queue(q, flags); return (flags & DN_DESTROY) ? DNHT_SCAN_DEL : 0; } /* * calls dn_delete_queue/q_delete_cb on all queues, * which notifies the parent scheduler and possibly drains packets. * flags & DN_DESTROY: drains queues and destroy qht; */ static void qht_delete(struct dn_fsk *fs, int flags) { ND("fs %d start flags %d qht %p", fs->fs.fs_nr, flags, fs->qht); if (!fs->qht) return; if (fs->fs.flags & DN_QHT_HASH) { dn_ht_scan(fs->qht, q_delete_cb, (void *)(uintptr_t)flags); if (flags & DN_DESTROY) { dn_ht_free(fs->qht, 0); fs->qht = NULL; } } else { dn_delete_queue((struct dn_queue *)(fs->qht), flags); if (flags & DN_DESTROY) fs->qht = NULL; } } /* * Find and possibly create the queue for a MULTIQUEUE scheduler. * We never call it for !MULTIQUEUE (the queue is in the sch_inst). */ struct dn_queue * ipdn_q_find(struct dn_fsk *fs, struct dn_sch_inst *si, struct ipfw_flow_id *id) { struct dn_queue template; template._si = si; template.fs = fs; if (fs->fs.flags & DN_QHT_HASH) { struct ipfw_flow_id masked_id; if (fs->qht == NULL) { fs->qht = dn_ht_init(NULL, fs->fs.buckets, offsetof(struct dn_queue, q_next), q_hash, q_match, q_new); if (fs->qht == NULL) return NULL; } masked_id = *id; flow_id_mask(&fs->fsk_mask, &masked_id); return dn_ht_find(fs->qht, (uintptr_t)&masked_id, DNHT_INSERT, &template); } else { if (fs->qht == NULL) fs->qht = q_new(0, 0, &template); return (struct dn_queue *)fs->qht; } } /*--- end of queue hash table ---*/ /*--- support functions for the sch_inst hashtable ---- * * These are hashed by flow-id */ static uint32_t si_hash(uintptr_t key, int flags, void *arg) { /* compute the hash slot from the flow id */ struct ipfw_flow_id *id = (flags & DNHT_KEY_IS_OBJ) ? &((struct dn_sch_inst *)key)->ni.fid : (struct ipfw_flow_id *)key; return flow_id_hash(id); } static int si_match(void *obj, uintptr_t key, int flags, void *arg) { struct dn_sch_inst *o = obj; struct ipfw_flow_id *id2; id2 = (flags & DNHT_KEY_IS_OBJ) ? &((struct dn_sch_inst *)key)->ni.fid : (struct ipfw_flow_id *)key; return flow_id_cmp(&o->ni.fid, id2) == 0; } /* * create a new instance for the given 'key' * Allocate memory for instance, delay line and scheduler private data. */ static void * si_new(uintptr_t key, int flags, void *arg) { struct dn_schk *s = arg; struct dn_sch_inst *si; int l = sizeof(*si) + s->fp->si_datalen; si = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO); if (si == NULL) goto error; /* Set length only for the part passed up to userland. */ set_oid(&si->ni.oid, DN_SCH_I, sizeof(struct dn_flow)); set_oid(&(si->dline.oid), DN_DELAY_LINE, sizeof(struct delay_line)); /* mark si and dline as outside the event queue */ si->ni.oid.id = si->dline.oid.id = -1; si->sched = s; si->dline.si = si; if (s->fp->new_sched && s->fp->new_sched(si)) { D("new_sched error"); goto error; } if (s->sch.flags & DN_HAVE_MASK) si->ni.fid = *(struct ipfw_flow_id *)key; dn_cfg.si_count++; return si; error: if (si) { bzero(si, sizeof(*si)); // safety free(si, M_DUMMYNET); } return NULL; } /* * Callback from siht to delete all scheduler instances. Remove * si and delay line from the system heap, destroy all queues. * We assume that all flowset have been notified and do not * point to us anymore. */ static int si_destroy(void *_si, void *arg) { struct dn_sch_inst *si = _si; struct dn_schk *s = si->sched; struct delay_line *dl = &si->dline; if (dl->oid.subtype) /* remove delay line from event heap */ heap_extract(&dn_cfg.evheap, dl); dn_free_pkts(dl->mq.head); /* drain delay line */ if (si->kflags & DN_ACTIVE) /* remove si from event heap */ heap_extract(&dn_cfg.evheap, si); if (s->fp->free_sched) s->fp->free_sched(si); bzero(si, sizeof(*si)); /* safety */ free(si, M_DUMMYNET); dn_cfg.si_count--; return DNHT_SCAN_DEL; } /* * Find the scheduler instance for this packet. If we need to apply * a mask, do on a local copy of the flow_id to preserve the original. * Assume siht is always initialized if we have a mask. */ struct dn_sch_inst * ipdn_si_find(struct dn_schk *s, struct ipfw_flow_id *id) { if (s->sch.flags & DN_HAVE_MASK) { struct ipfw_flow_id id_t = *id; flow_id_mask(&s->sch.sched_mask, &id_t); return dn_ht_find(s->siht, (uintptr_t)&id_t, DNHT_INSERT, s); } if (!s->siht) s->siht = si_new(0, 0, s); return (struct dn_sch_inst *)s->siht; } /* callback to flush credit for the scheduler instance */ static int si_reset_credit(void *_si, void *arg) { struct dn_sch_inst *si = _si; struct dn_link *p = &si->sched->link; si->credit = p->burst + (dn_cfg.io_fast ? p->bandwidth : 0); return 0; } static void schk_reset_credit(struct dn_schk *s) { if (s->sch.flags & DN_HAVE_MASK) dn_ht_scan(s->siht, si_reset_credit, NULL); else if (s->siht) si_reset_credit(s->siht, NULL); } /*---- end of sch_inst hashtable ---------------------*/ /*------------------------------------------------------- * flowset hash (fshash) support. Entries are hashed by fs_nr. * New allocations are put in the fsunlinked list, from which * they are removed when they point to a specific scheduler. */ static uint32_t fsk_hash(uintptr_t key, int flags, void *arg) { uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key : ((struct dn_fsk *)key)->fs.fs_nr; return ( (i>>8)^(i>>4)^i ); } static int fsk_match(void *obj, uintptr_t key, int flags, void *arg) { struct dn_fsk *fs = obj; int i = !(flags & DNHT_KEY_IS_OBJ) ? key : ((struct dn_fsk *)key)->fs.fs_nr; return (fs->fs.fs_nr == i); } static void * fsk_new(uintptr_t key, int flags, void *arg) { struct dn_fsk *fs; fs = malloc(sizeof(*fs), M_DUMMYNET, M_NOWAIT | M_ZERO); if (fs) { set_oid(&fs->fs.oid, DN_FS, sizeof(fs->fs)); dn_cfg.fsk_count++; fs->drain_bucket = 0; SLIST_INSERT_HEAD(&dn_cfg.fsu, fs, sch_chain); } return fs; } /* * detach flowset from its current scheduler. Flags as follows: * DN_DETACH removes from the fsk_list * DN_DESTROY deletes individual queues * DN_DELETE_FS destroys the flowset (otherwise goes in unlinked). */ static void fsk_detach(struct dn_fsk *fs, int flags) { if (flags & DN_DELETE_FS) flags |= DN_DESTROY; ND("fs %d from sched %d flags %s %s %s", fs->fs.fs_nr, fs->fs.sched_nr, (flags & DN_DELETE_FS) ? "DEL_FS":"", (flags & DN_DESTROY) ? "DEL":"", (flags & DN_DETACH) ? "DET":""); if (flags & DN_DETACH) { /* detach from the list */ struct dn_fsk_head *h; h = fs->sched ? &fs->sched->fsk_list : &dn_cfg.fsu; SLIST_REMOVE(h, fs, dn_fsk, sch_chain); } /* Free the RED parameters, they will be recomputed on * subsequent attach if needed. */ if (fs->w_q_lookup) free(fs->w_q_lookup, M_DUMMYNET); fs->w_q_lookup = NULL; qht_delete(fs, flags); if (fs->sched && fs->sched->fp->free_fsk) fs->sched->fp->free_fsk(fs); fs->sched = NULL; if (flags & DN_DELETE_FS) { bzero(fs, sizeof(*fs)); /* safety */ free(fs, M_DUMMYNET); dn_cfg.fsk_count--; } else { SLIST_INSERT_HEAD(&dn_cfg.fsu, fs, sch_chain); } } /* * Detach or destroy all flowsets in a list. * flags specifies what to do: * DN_DESTROY: flush all queues * DN_DELETE_FS: DN_DESTROY + destroy flowset * DN_DELETE_FS implies DN_DESTROY */ static void fsk_detach_list(struct dn_fsk_head *h, int flags) { struct dn_fsk *fs; int n = 0; /* only for stats */ ND("head %p flags %x", h, flags); while ((fs = SLIST_FIRST(h))) { SLIST_REMOVE_HEAD(h, sch_chain); n++; fsk_detach(fs, flags); } ND("done %d flowsets", n); } /* * called on 'queue X delete' -- removes the flowset from fshash, * deletes all queues for the flowset, and removes the flowset. */ static int delete_fs(int i, int locked) { struct dn_fsk *fs; int err = 0; if (!locked) DN_BH_WLOCK(); fs = dn_ht_find(dn_cfg.fshash, i, DNHT_REMOVE, NULL); ND("fs %d found %p", i, fs); if (fs) { fsk_detach(fs, DN_DETACH | DN_DELETE_FS); err = 0; } else err = EINVAL; if (!locked) DN_BH_WUNLOCK(); return err; } /*----- end of flowset hashtable support -------------*/ /*------------------------------------------------------------ * Scheduler hash. When searching by index we pass sched_nr, * otherwise we pass struct dn_sch * which is the first field in * struct dn_schk so we can cast between the two. We use this trick * because in the create phase (but it should be fixed). */ static uint32_t schk_hash(uintptr_t key, int flags, void *_arg) { uint32_t i = !(flags & DNHT_KEY_IS_OBJ) ? key : ((struct dn_schk *)key)->sch.sched_nr; return ( (i>>8)^(i>>4)^i ); } static int schk_match(void *obj, uintptr_t key, int flags, void *_arg) { struct dn_schk *s = (struct dn_schk *)obj; int i = !(flags & DNHT_KEY_IS_OBJ) ? key : ((struct dn_schk *)key)->sch.sched_nr; return (s->sch.sched_nr == i); } /* * Create the entry and intialize with the sched hash if needed. * Leave s->fp unset so we can tell whether a dn_ht_find() returns * a new object or a previously existing one. */ static void * schk_new(uintptr_t key, int flags, void *arg) { struct schk_new_arg *a = arg; struct dn_schk *s; int l = sizeof(*s) +a->fp->schk_datalen; s = malloc(l, M_DUMMYNET, M_NOWAIT | M_ZERO); if (s == NULL) return NULL; set_oid(&s->link.oid, DN_LINK, sizeof(s->link)); s->sch = *a->sch; // copy initial values s->link.link_nr = s->sch.sched_nr; SLIST_INIT(&s->fsk_list); /* initialize the hash table or create the single instance */ s->fp = a->fp; /* si_new needs this */ s->drain_bucket = 0; if (s->sch.flags & DN_HAVE_MASK) { s->siht = dn_ht_init(NULL, s->sch.buckets, offsetof(struct dn_sch_inst, si_next), si_hash, si_match, si_new); if (s->siht == NULL) { free(s, M_DUMMYNET); return NULL; } } s->fp = NULL; /* mark as a new scheduler */ dn_cfg.schk_count++; return s; } /* * Callback for sched delete. Notify all attached flowsets to * detach from the scheduler, destroy the internal flowset, and * all instances. The scheduler goes away too. * arg is 0 (only detach flowsets and destroy instances) * DN_DESTROY (detach & delete queues, delete schk) * or DN_DELETE_FS (delete queues and flowsets, delete schk) */ static int schk_delete_cb(void *obj, void *arg) { struct dn_schk *s = obj; #if 0 int a = (int)arg; ND("sched %d arg %s%s", s->sch.sched_nr, a&DN_DESTROY ? "DEL ":"", a&DN_DELETE_FS ? "DEL_FS":""); #endif fsk_detach_list(&s->fsk_list, arg ? DN_DESTROY : 0); /* no more flowset pointing to us now */ if (s->sch.flags & DN_HAVE_MASK) { dn_ht_scan(s->siht, si_destroy, NULL); dn_ht_free(s->siht, 0); } else if (s->siht) si_destroy(s->siht, NULL); if (s->profile) { free(s->profile, M_DUMMYNET); s->profile = NULL; } s->siht = NULL; if (s->fp->destroy) s->fp->destroy(s); bzero(s, sizeof(*s)); // safety free(obj, M_DUMMYNET); dn_cfg.schk_count--; return DNHT_SCAN_DEL; } /* * called on a 'sched X delete' command. Deletes a single scheduler. * This is done by removing from the schedhash, unlinking all * flowsets and deleting their traffic. */ static int delete_schk(int i) { struct dn_schk *s; s = dn_ht_find(dn_cfg.schedhash, i, DNHT_REMOVE, NULL); ND("%d %p", i, s); if (!s) return EINVAL; delete_fs(i + DN_MAX_ID, 1); /* first delete internal fs */ /* then detach flowsets, delete traffic */ schk_delete_cb(s, (void*)(uintptr_t)DN_DESTROY); return 0; } /*--- end of schk hashtable support ---*/ static int copy_obj(char **start, char *end, void *_o, const char *msg, int i) { struct dn_id *o = _o; int have = end - *start; if (have < o->len || o->len == 0 || o->type == 0) { D("(WARN) type %d %s %d have %d need %d", o->type, msg, i, have, o->len); return 1; } ND("type %d %s %d len %d", o->type, msg, i, o->len); bcopy(_o, *start, o->len); if (o->type == DN_LINK) { /* Adjust burst parameter for link */ struct dn_link *l = (struct dn_link *)*start; l->burst = div64(l->burst, 8 * hz); l->delay = l->delay * 1000 / hz; } else if (o->type == DN_SCH) { /* Set id->id to the number of instances */ struct dn_schk *s = _o; struct dn_id *id = (struct dn_id *)(*start); id->id = (s->sch.flags & DN_HAVE_MASK) ? dn_ht_entries(s->siht) : (s->siht ? 1 : 0); } *start += o->len; return 0; } /* Specific function to copy a queue. * Copies only the user-visible part of a queue (which is in * a struct dn_flow), and sets len accordingly. */ static int copy_obj_q(char **start, char *end, void *_o, const char *msg, int i) { struct dn_id *o = _o; int have = end - *start; int len = sizeof(struct dn_flow); /* see above comment */ if (have < len || o->len == 0 || o->type != DN_QUEUE) { D("ERROR type %d %s %d have %d need %d", o->type, msg, i, have, len); return 1; } ND("type %d %s %d len %d", o->type, msg, i, len); bcopy(_o, *start, len); ((struct dn_id*)(*start))->len = len; *start += len; return 0; } static int copy_q_cb(void *obj, void *arg) { struct dn_queue *q = obj; struct copy_args *a = arg; struct dn_flow *ni = (struct dn_flow *)(*a->start); if (copy_obj_q(a->start, a->end, &q->ni, "queue", -1)) return DNHT_SCAN_END; ni->oid.type = DN_FLOW; /* override the DN_QUEUE */ ni->oid.id = si_hash((uintptr_t)&ni->fid, 0, NULL); return 0; } static int copy_q(struct copy_args *a, struct dn_fsk *fs, int flags) { if (!fs->qht) return 0; if (fs->fs.flags & DN_QHT_HASH) dn_ht_scan(fs->qht, copy_q_cb, a); else copy_q_cb(fs->qht, a); return 0; } /* * This routine only copies the initial part of a profile ? XXX */ static int copy_profile(struct copy_args *a, struct dn_profile *p) { int have = a->end - *a->start; /* XXX here we check for max length */ int profile_len = sizeof(struct dn_profile) - ED_MAX_SAMPLES_NO*sizeof(int); if (p == NULL) return 0; if (have < profile_len) { D("error have %d need %d", have, profile_len); return 1; } bcopy(p, *a->start, profile_len); ((struct dn_id *)(*a->start))->len = profile_len; *a->start += profile_len; return 0; } static int copy_flowset(struct copy_args *a, struct dn_fsk *fs, int flags) { struct dn_fs *ufs = (struct dn_fs *)(*a->start); if (!fs) return 0; ND("flowset %d", fs->fs.fs_nr); if (copy_obj(a->start, a->end, &fs->fs, "flowset", fs->fs.fs_nr)) return DNHT_SCAN_END; ufs->oid.id = (fs->fs.flags & DN_QHT_HASH) ? dn_ht_entries(fs->qht) : (fs->qht ? 1 : 0); if (flags) { /* copy queues */ copy_q(a, fs, 0); } return 0; } static int copy_si_cb(void *obj, void *arg) { struct dn_sch_inst *si = obj; struct copy_args *a = arg; struct dn_flow *ni = (struct dn_flow *)(*a->start); if (copy_obj(a->start, a->end, &si->ni, "inst", si->sched->sch.sched_nr)) return DNHT_SCAN_END; ni->oid.type = DN_FLOW; /* override the DN_SCH_I */ ni->oid.id = si_hash((uintptr_t)si, DNHT_KEY_IS_OBJ, NULL); return 0; } static int copy_si(struct copy_args *a, struct dn_schk *s, int flags) { if (s->sch.flags & DN_HAVE_MASK) dn_ht_scan(s->siht, copy_si_cb, a); else if (s->siht) copy_si_cb(s->siht, a); return 0; } /* * compute a list of children of a scheduler and copy up */ static int copy_fsk_list(struct copy_args *a, struct dn_schk *s, int flags) { struct dn_fsk *fs; struct dn_id *o; uint32_t *p; int n = 0, space = sizeof(*o); SLIST_FOREACH(fs, &s->fsk_list, sch_chain) { if (fs->fs.fs_nr < DN_MAX_ID) n++; } space += n * sizeof(uint32_t); DX(3, "sched %d has %d flowsets", s->sch.sched_nr, n); if (a->end - *(a->start) < space) return DNHT_SCAN_END; o = (struct dn_id *)(*(a->start)); o->len = space; *a->start += o->len; o->type = DN_TEXT; p = (uint32_t *)(o+1); SLIST_FOREACH(fs, &s->fsk_list, sch_chain) if (fs->fs.fs_nr < DN_MAX_ID) *p++ = fs->fs.fs_nr; return 0; } static int copy_data_helper(void *_o, void *_arg) { struct copy_args *a = _arg; uint32_t *r = a->extra->r; /* start of first range */ uint32_t *lim; /* first invalid pointer */ int n; lim = (uint32_t *)((char *)(a->extra) + a->extra->o.len); if (a->type == DN_LINK || a->type == DN_SCH) { /* pipe|sched show, we receive a dn_schk */ struct dn_schk *s = _o; n = s->sch.sched_nr; if (a->type == DN_SCH && n >= DN_MAX_ID) return 0; /* not a scheduler */ if (a->type == DN_LINK && n <= DN_MAX_ID) return 0; /* not a pipe */ /* see if the object is within one of our ranges */ for (;r < lim; r += 2) { if (n < r[0] || n > r[1]) continue; /* Found a valid entry, copy and we are done */ if (a->flags & DN_C_LINK) { if (copy_obj(a->start, a->end, &s->link, "link", n)) return DNHT_SCAN_END; if (copy_profile(a, s->profile)) return DNHT_SCAN_END; if (copy_flowset(a, s->fs, 0)) return DNHT_SCAN_END; } if (a->flags & DN_C_SCH) { if (copy_obj(a->start, a->end, &s->sch, "sched", n)) return DNHT_SCAN_END; /* list all attached flowsets */ if (copy_fsk_list(a, s, 0)) return DNHT_SCAN_END; } if (a->flags & DN_C_FLOW) copy_si(a, s, 0); break; } } else if (a->type == DN_FS) { /* queue show, skip internal flowsets */ struct dn_fsk *fs = _o; n = fs->fs.fs_nr; if (n >= DN_MAX_ID) return 0; /* see if the object is within one of our ranges */ for (;r < lim; r += 2) { if (n < r[0] || n > r[1]) continue; if (copy_flowset(a, fs, 0)) return DNHT_SCAN_END; copy_q(a, fs, 0); break; /* we are done */ } } return 0; } static inline struct dn_schk * locate_scheduler(int i) { return dn_ht_find(dn_cfg.schedhash, i, 0, NULL); } /* * red parameters are in fixed point arithmetic. */ static int config_red(struct dn_fsk *fs) { int64_t s, idle, weight, w0; int t, i; fs->w_q = fs->fs.w_q; fs->max_p = fs->fs.max_p; ND("called"); /* Doing stuff that was in userland */ i = fs->sched->link.bandwidth; s = (i <= 0) ? 0 : hz * dn_cfg.red_avg_pkt_size * 8 * SCALE(1) / i; idle = div64((s * 3) , fs->w_q); /* s, fs->w_q scaled; idle not scaled */ fs->lookup_step = div64(idle , dn_cfg.red_lookup_depth); /* fs->lookup_step not scaled, */ if (!fs->lookup_step) fs->lookup_step = 1; w0 = weight = SCALE(1) - fs->w_q; //fs->w_q scaled for (t = fs->lookup_step; t > 1; --t) weight = SCALE_MUL(weight, w0); fs->lookup_weight = (int)(weight); // scaled /* Now doing stuff that was in kerneland */ fs->min_th = SCALE(fs->fs.min_th); fs->max_th = SCALE(fs->fs.max_th); - fs->c_1 = fs->max_p / (fs->fs.max_th - fs->fs.min_th); + if (fs->fs.max_th == fs->fs.min_th) + fs->c_1 = fs->max_p; + else + fs->c_1 = SCALE((int64_t)(fs->max_p)) / (fs->fs.max_th - fs->fs.min_th); fs->c_2 = SCALE_MUL(fs->c_1, SCALE(fs->fs.min_th)); if (fs->fs.flags & DN_IS_GENTLE_RED) { fs->c_3 = (SCALE(1) - fs->max_p) / fs->fs.max_th; fs->c_4 = SCALE(1) - 2 * fs->max_p; } /* If the lookup table already exist, free and create it again. */ if (fs->w_q_lookup) { free(fs->w_q_lookup, M_DUMMYNET); fs->w_q_lookup = NULL; } if (dn_cfg.red_lookup_depth == 0) { printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth" "must be > 0\n"); fs->fs.flags &= ~DN_IS_RED; fs->fs.flags &= ~DN_IS_GENTLE_RED; return (EINVAL); } fs->lookup_depth = dn_cfg.red_lookup_depth; fs->w_q_lookup = (u_int *)malloc(fs->lookup_depth * sizeof(int), M_DUMMYNET, M_NOWAIT); if (fs->w_q_lookup == NULL) { printf("dummynet: sorry, cannot allocate red lookup table\n"); fs->fs.flags &= ~DN_IS_RED; fs->fs.flags &= ~DN_IS_GENTLE_RED; return(ENOSPC); } /* Fill the lookup table with (1 - w_q)^x */ fs->w_q_lookup[0] = SCALE(1) - fs->w_q; for (i = 1; i < fs->lookup_depth; i++) fs->w_q_lookup[i] = SCALE_MUL(fs->w_q_lookup[i - 1], fs->lookup_weight); if (dn_cfg.red_avg_pkt_size < 1) dn_cfg.red_avg_pkt_size = 512; fs->avg_pkt_size = dn_cfg.red_avg_pkt_size; if (dn_cfg.red_max_pkt_size < 1) dn_cfg.red_max_pkt_size = 1500; fs->max_pkt_size = dn_cfg.red_max_pkt_size; ND("exit"); return 0; } /* Scan all flowset attached to this scheduler and update red */ static void update_red(struct dn_schk *s) { struct dn_fsk *fs; SLIST_FOREACH(fs, &s->fsk_list, sch_chain) { if (fs && (fs->fs.flags & DN_IS_RED)) config_red(fs); } } /* attach flowset to scheduler s, possibly requeue */ static void fsk_attach(struct dn_fsk *fs, struct dn_schk *s) { ND("remove fs %d from fsunlinked, link to sched %d", fs->fs.fs_nr, s->sch.sched_nr); SLIST_REMOVE(&dn_cfg.fsu, fs, dn_fsk, sch_chain); fs->sched = s; SLIST_INSERT_HEAD(&s->fsk_list, fs, sch_chain); if (s->fp->new_fsk) s->fp->new_fsk(fs); /* XXX compute fsk_mask */ fs->fsk_mask = fs->fs.flow_mask; if (fs->sched->sch.flags & DN_HAVE_MASK) flow_id_or(&fs->sched->sch.sched_mask, &fs->fsk_mask); if (fs->qht) { /* * we must drain qht according to the old * type, and reinsert according to the new one. * The requeue is complex -- in general we need to * reclassify every single packet. * For the time being, let's hope qht is never set * when we reach this point. */ D("XXX TODO requeue from fs %d to sch %d", fs->fs.fs_nr, s->sch.sched_nr); fs->qht = NULL; } /* set the new type for qht */ if (nonzero_mask(&fs->fsk_mask)) fs->fs.flags |= DN_QHT_HASH; else fs->fs.flags &= ~DN_QHT_HASH; /* XXX config_red() can fail... */ if (fs->fs.flags & DN_IS_RED) config_red(fs); } /* update all flowsets which may refer to this scheduler */ static void update_fs(struct dn_schk *s) { struct dn_fsk *fs, *tmp; SLIST_FOREACH_SAFE(fs, &dn_cfg.fsu, sch_chain, tmp) { if (s->sch.sched_nr != fs->fs.sched_nr) { D("fs %d for sch %d not %d still unlinked", fs->fs.fs_nr, fs->fs.sched_nr, s->sch.sched_nr); continue; } fsk_attach(fs, s); } } /* * Configuration -- to preserve backward compatibility we use * the following scheme (N is 65536) * NUMBER SCHED LINK FLOWSET * 1 .. N-1 (1)WFQ (2)WFQ (3)queue * N+1 .. 2N-1 (4)FIFO (5)FIFO (6)FIFO for sched 1..N-1 * 2N+1 .. 3N-1 -- -- (7)FIFO for sched N+1..2N-1 * * "pipe i config" configures #1, #2 and #3 * "sched i config" configures #1 and possibly #6 * "queue i config" configures #3 * #1 is configured with 'pipe i config' or 'sched i config' * #2 is configured with 'pipe i config', and created if not * existing with 'sched i config' * #3 is configured with 'queue i config' * #4 is automatically configured after #1, can only be FIFO * #5 is automatically configured after #2 * #6 is automatically created when #1 is !MULTIQUEUE, * and can be updated. * #7 is automatically configured after #2 */ /* * configure a link (and its FIFO instance) */ static int config_link(struct dn_link *p, struct dn_id *arg) { int i; if (p->oid.len != sizeof(*p)) { D("invalid pipe len %d", p->oid.len); return EINVAL; } i = p->link_nr; if (i <= 0 || i >= DN_MAX_ID) return EINVAL; /* * The config program passes parameters as follows: * bw = bits/second (0 means no limits), * delay = ms, must be translated into ticks. * qsize = slots/bytes * burst ??? */ p->delay = (p->delay * hz) / 1000; /* Scale burst size: bytes -> bits * hz */ p->burst *= 8 * hz; DN_BH_WLOCK(); /* do it twice, base link and FIFO link */ for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) { struct dn_schk *s = locate_scheduler(i); if (s == NULL) { DN_BH_WUNLOCK(); D("sched %d not found", i); return EINVAL; } /* remove profile if exists */ if (s->profile) { free(s->profile, M_DUMMYNET); s->profile = NULL; } /* copy all parameters */ s->link.oid = p->oid; s->link.link_nr = i; s->link.delay = p->delay; if (s->link.bandwidth != p->bandwidth) { /* XXX bandwidth changes, need to update red params */ s->link.bandwidth = p->bandwidth; update_red(s); } s->link.burst = p->burst; schk_reset_credit(s); } dn_cfg.id++; DN_BH_WUNLOCK(); return 0; } /* * configure a flowset. Can be called from inside with locked=1, */ static struct dn_fsk * config_fs(struct dn_fs *nfs, struct dn_id *arg, int locked) { int i; struct dn_fsk *fs; if (nfs->oid.len != sizeof(*nfs)) { D("invalid flowset len %d", nfs->oid.len); return NULL; } i = nfs->fs_nr; if (i <= 0 || i >= 3*DN_MAX_ID) return NULL; ND("flowset %d", i); /* XXX other sanity checks */ if (nfs->flags & DN_QSIZE_BYTES) { ipdn_bound_var(&nfs->qsize, 16384, 1500, dn_cfg.byte_limit, NULL); // "queue byte size"); } else { ipdn_bound_var(&nfs->qsize, 50, 1, dn_cfg.slot_limit, NULL); // "queue slot size"); } if (nfs->flags & DN_HAVE_MASK) { /* make sure we have some buckets */ ipdn_bound_var((int *)&nfs->buckets, dn_cfg.hash_size, 1, dn_cfg.max_hash_size, "flowset buckets"); } else { nfs->buckets = 1; /* we only need 1 */ } if (!locked) DN_BH_WLOCK(); do { /* exit with break when done */ struct dn_schk *s; int flags = nfs->sched_nr ? DNHT_INSERT : 0; int j; int oldc = dn_cfg.fsk_count; fs = dn_ht_find(dn_cfg.fshash, i, flags, NULL); if (fs == NULL) { D("missing sched for flowset %d", i); break; } /* grab some defaults from the existing one */ if (nfs->sched_nr == 0) /* reuse */ nfs->sched_nr = fs->fs.sched_nr; for (j = 0; j < sizeof(nfs->par)/sizeof(nfs->par[0]); j++) { if (nfs->par[j] == -1) /* reuse */ nfs->par[j] = fs->fs.par[j]; } if (bcmp(&fs->fs, nfs, sizeof(*nfs)) == 0) { ND("flowset %d unchanged", i); break; /* no change, nothing to do */ } if (oldc != dn_cfg.fsk_count) /* new item */ dn_cfg.id++; s = locate_scheduler(nfs->sched_nr); /* detach from old scheduler if needed, preserving * queues if we need to reattach. Then update the * configuration, and possibly attach to the new sched. */ DX(2, "fs %d changed sched %d@%p to %d@%p", fs->fs.fs_nr, fs->fs.sched_nr, fs->sched, nfs->sched_nr, s); if (fs->sched) { int flags = s ? DN_DETACH : (DN_DETACH | DN_DESTROY); flags |= DN_DESTROY; /* XXX temporary */ fsk_detach(fs, flags); } fs->fs = *nfs; /* copy configuration */ if (s != NULL) fsk_attach(fs, s); } while (0); if (!locked) DN_BH_WUNLOCK(); return fs; } /* * config/reconfig a scheduler and its FIFO variant. * For !MULTIQUEUE schedulers, also set up the flowset. * * On reconfigurations (detected because s->fp is set), * detach existing flowsets preserving traffic, preserve link, * and delete the old scheduler creating a new one. */ static int config_sched(struct dn_sch *_nsch, struct dn_id *arg) { struct dn_schk *s; struct schk_new_arg a; /* argument for schk_new */ int i; struct dn_link p; /* copy of oldlink */ struct dn_profile *pf = NULL; /* copy of old link profile */ /* Used to preserv mask parameter */ struct ipfw_flow_id new_mask; int new_buckets = 0; int new_flags = 0; int pipe_cmd; int err = ENOMEM; a.sch = _nsch; if (a.sch->oid.len != sizeof(*a.sch)) { D("bad sched len %d", a.sch->oid.len); return EINVAL; } i = a.sch->sched_nr; if (i <= 0 || i >= DN_MAX_ID) return EINVAL; /* make sure we have some buckets */ if (a.sch->flags & DN_HAVE_MASK) ipdn_bound_var((int *)&a.sch->buckets, dn_cfg.hash_size, 1, dn_cfg.max_hash_size, "sched buckets"); /* XXX other sanity checks */ bzero(&p, sizeof(p)); pipe_cmd = a.sch->flags & DN_PIPE_CMD; a.sch->flags &= ~DN_PIPE_CMD; //XXX do it even if is not set? if (pipe_cmd) { /* Copy mask parameter */ new_mask = a.sch->sched_mask; new_buckets = a.sch->buckets; new_flags = a.sch->flags; } DN_BH_WLOCK(); again: /* run twice, for wfq and fifo */ /* * lookup the type. If not supplied, use the previous one * or default to WF2Q+. Otherwise, return an error. */ dn_cfg.id++; a.fp = find_sched_type(a.sch->oid.subtype, a.sch->name); if (a.fp != NULL) { /* found. Lookup or create entry */ s = dn_ht_find(dn_cfg.schedhash, i, DNHT_INSERT, &a); } else if (a.sch->oid.subtype == 0 && !a.sch->name[0]) { /* No type. search existing s* or retry with WF2Q+ */ s = dn_ht_find(dn_cfg.schedhash, i, 0, &a); if (s != NULL) { a.fp = s->fp; /* Scheduler exists, skip to FIFO scheduler * if command was pipe config... */ if (pipe_cmd) goto next; } else { /* New scheduler, create a wf2q+ with no mask * if command was pipe config... */ if (pipe_cmd) { /* clear mask parameter */ bzero(&a.sch->sched_mask, sizeof(new_mask)); a.sch->buckets = 0; a.sch->flags &= ~DN_HAVE_MASK; } a.sch->oid.subtype = DN_SCHED_WF2QP; goto again; } } else { D("invalid scheduler type %d %s", a.sch->oid.subtype, a.sch->name); err = EINVAL; goto error; } /* normalize name and subtype */ a.sch->oid.subtype = a.fp->type; bzero(a.sch->name, sizeof(a.sch->name)); strlcpy(a.sch->name, a.fp->name, sizeof(a.sch->name)); if (s == NULL) { D("cannot allocate scheduler %d", i); goto error; } /* restore existing link if any */ if (p.link_nr) { s->link = p; if (!pf || pf->link_nr != p.link_nr) { /* no saved value */ s->profile = NULL; /* XXX maybe not needed */ } else { s->profile = malloc(sizeof(struct dn_profile), M_DUMMYNET, M_NOWAIT | M_ZERO); if (s->profile == NULL) { D("cannot allocate profile"); goto error; //XXX } bcopy(pf, s->profile, sizeof(*pf)); } } p.link_nr = 0; if (s->fp == NULL) { DX(2, "sched %d new type %s", i, a.fp->name); } else if (s->fp != a.fp || bcmp(a.sch, &s->sch, sizeof(*a.sch)) ) { /* already existing. */ DX(2, "sched %d type changed from %s to %s", i, s->fp->name, a.fp->name); DX(4, " type/sub %d/%d -> %d/%d", s->sch.oid.type, s->sch.oid.subtype, a.sch->oid.type, a.sch->oid.subtype); if (s->link.link_nr == 0) D("XXX WARNING link 0 for sched %d", i); p = s->link; /* preserve link */ if (s->profile) {/* preserve profile */ if (!pf) pf = malloc(sizeof(*pf), M_DUMMYNET, M_NOWAIT | M_ZERO); if (pf) /* XXX should issue a warning otherwise */ bcopy(s->profile, pf, sizeof(*pf)); } /* remove from the hash */ dn_ht_find(dn_cfg.schedhash, i, DNHT_REMOVE, NULL); /* Detach flowsets, preserve queues. */ // schk_delete_cb(s, NULL); // XXX temporarily, kill queues schk_delete_cb(s, (void *)DN_DESTROY); goto again; } else { DX(4, "sched %d unchanged type %s", i, a.fp->name); } /* complete initialization */ s->sch = *a.sch; s->fp = a.fp; s->cfg = arg; // XXX schk_reset_credit(s); /* create the internal flowset if needed, * trying to reuse existing ones if available */ if (!(s->fp->flags & DN_MULTIQUEUE) && !s->fs) { s->fs = dn_ht_find(dn_cfg.fshash, i, 0, NULL); if (!s->fs) { struct dn_fs fs; bzero(&fs, sizeof(fs)); set_oid(&fs.oid, DN_FS, sizeof(fs)); fs.fs_nr = i + DN_MAX_ID; fs.sched_nr = i; s->fs = config_fs(&fs, NULL, 1 /* locked */); } if (!s->fs) { schk_delete_cb(s, (void *)DN_DESTROY); D("error creating internal fs for %d", i); goto error; } } /* call init function after the flowset is created */ if (s->fp->config) s->fp->config(s); update_fs(s); next: if (i < DN_MAX_ID) { /* now configure the FIFO instance */ i += DN_MAX_ID; if (pipe_cmd) { /* Restore mask parameter for FIFO */ a.sch->sched_mask = new_mask; a.sch->buckets = new_buckets; a.sch->flags = new_flags; } else { /* sched config shouldn't modify the FIFO scheduler */ if (dn_ht_find(dn_cfg.schedhash, i, 0, &a) != NULL) { /* FIFO already exist, don't touch it */ err = 0; /* and this is not an error */ goto error; } } a.sch->sched_nr = i; a.sch->oid.subtype = DN_SCHED_FIFO; bzero(a.sch->name, sizeof(a.sch->name)); goto again; } err = 0; error: DN_BH_WUNLOCK(); if (pf) free(pf, M_DUMMYNET); return err; } /* * attach a profile to a link */ static int config_profile(struct dn_profile *pf, struct dn_id *arg) { struct dn_schk *s; int i, olen, err = 0; if (pf->oid.len < sizeof(*pf)) { D("short profile len %d", pf->oid.len); return EINVAL; } i = pf->link_nr; if (i <= 0 || i >= DN_MAX_ID) return EINVAL; /* XXX other sanity checks */ DN_BH_WLOCK(); for (; i < 2*DN_MAX_ID; i += DN_MAX_ID) { s = locate_scheduler(i); if (s == NULL) { err = EINVAL; break; } dn_cfg.id++; /* * If we had a profile and the new one does not fit, * or it is deleted, then we need to free memory. */ if (s->profile && (pf->samples_no == 0 || s->profile->oid.len < pf->oid.len)) { free(s->profile, M_DUMMYNET); s->profile = NULL; } if (pf->samples_no == 0) continue; /* * new profile, possibly allocate memory * and copy data. */ if (s->profile == NULL) s->profile = malloc(pf->oid.len, M_DUMMYNET, M_NOWAIT | M_ZERO); if (s->profile == NULL) { D("no memory for profile %d", i); err = ENOMEM; break; } /* preserve larger length XXX double check */ olen = s->profile->oid.len; if (olen < pf->oid.len) olen = pf->oid.len; bcopy(pf, s->profile, pf->oid.len); s->profile->oid.len = olen; } DN_BH_WUNLOCK(); return err; } /* * Delete all objects: */ static void dummynet_flush(void) { /* delete all schedulers and related links/queues/flowsets */ dn_ht_scan(dn_cfg.schedhash, schk_delete_cb, (void *)(uintptr_t)DN_DELETE_FS); /* delete all remaining (unlinked) flowsets */ DX(4, "still %d unlinked fs", dn_cfg.fsk_count); dn_ht_free(dn_cfg.fshash, DNHT_REMOVE); fsk_detach_list(&dn_cfg.fsu, DN_DELETE_FS); /* Reinitialize system heap... */ heap_init(&dn_cfg.evheap, 16, offsetof(struct dn_id, id)); } /* * Main handler for configuration. We are guaranteed to be called * with an oid which is at least a dn_id. * - the first object is the command (config, delete, flush, ...) * - config_link must be issued after the corresponding config_sched * - parameters (DN_TXT) for an object must preceed the object * processed on a config_sched. */ int do_config(void *p, int l) { struct dn_id *next, *o; int err = 0, err2 = 0; struct dn_id *arg = NULL; uintptr_t *a; o = p; if (o->id != DN_API_VERSION) { D("invalid api version got %d need %d", o->id, DN_API_VERSION); return EINVAL; } for (; l >= sizeof(*o); o = next) { struct dn_id *prev = arg; if (o->len < sizeof(*o) || l < o->len) { D("bad len o->len %d len %d", o->len, l); err = EINVAL; break; } l -= o->len; next = (struct dn_id *)((char *)o + o->len); err = 0; switch (o->type) { default: D("cmd %d not implemented", o->type); break; #ifdef EMULATE_SYSCTL /* sysctl emulation. * if we recognize the command, jump to the correct * handler and return */ case DN_SYSCTL_SET: err = kesysctl_emu_set(p, l); return err; #endif case DN_CMD_CONFIG: /* simply a header */ break; case DN_CMD_DELETE: /* the argument is in the first uintptr_t after o */ a = (uintptr_t *)(o+1); if (o->len < sizeof(*o) + sizeof(*a)) { err = EINVAL; break; } switch (o->subtype) { case DN_LINK: /* delete base and derived schedulers */ DN_BH_WLOCK(); err = delete_schk(*a); err2 = delete_schk(*a + DN_MAX_ID); DN_BH_WUNLOCK(); if (!err) err = err2; break; default: D("invalid delete type %d", o->subtype); err = EINVAL; break; case DN_FS: err = (*a <1 || *a >= DN_MAX_ID) ? EINVAL : delete_fs(*a, 0) ; break; } break; case DN_CMD_FLUSH: DN_BH_WLOCK(); dummynet_flush(); DN_BH_WUNLOCK(); break; case DN_TEXT: /* store argument the next block */ prev = NULL; arg = o; break; case DN_LINK: err = config_link((struct dn_link *)o, arg); break; case DN_PROFILE: err = config_profile((struct dn_profile *)o, arg); break; case DN_SCH: err = config_sched((struct dn_sch *)o, arg); break; case DN_FS: err = (NULL==config_fs((struct dn_fs *)o, arg, 0)); break; } if (prev) arg = NULL; if (err != 0) break; } return err; } static int compute_space(struct dn_id *cmd, struct copy_args *a) { int x = 0, need = 0; int profile_size = sizeof(struct dn_profile) - ED_MAX_SAMPLES_NO*sizeof(int); /* NOTE about compute space: * NP = dn_cfg.schk_count * NSI = dn_cfg.si_count * NF = dn_cfg.fsk_count * NQ = dn_cfg.queue_count * - ipfw pipe show * (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler * link, scheduler template, flowset * integrated in scheduler and header * for flowset list * (NSI)*(dn_flow) all scheduler instance (includes * the queue instance) * - ipfw sched show * (NP/2)*(dn_link + dn_sch + dn_id + dn_fs) only half scheduler * link, scheduler template, flowset * integrated in scheduler and header * for flowset list * (NSI * dn_flow) all scheduler instances * (NF * sizeof(uint_32)) space for flowset list linked to scheduler * (NQ * dn_queue) all queue [XXXfor now not listed] * - ipfw queue show * (NF * dn_fs) all flowset * (NQ * dn_queue) all queues */ switch (cmd->subtype) { default: return -1; /* XXX where do LINK and SCH differ ? */ /* 'ipfw sched show' could list all queues associated to * a scheduler. This feature for now is disabled */ case DN_LINK: /* pipe show */ x = DN_C_LINK | DN_C_SCH | DN_C_FLOW; need += dn_cfg.schk_count * (sizeof(struct dn_fs) + profile_size) / 2; need += dn_cfg.fsk_count * sizeof(uint32_t); break; case DN_SCH: /* sched show */ need += dn_cfg.schk_count * (sizeof(struct dn_fs) + profile_size) / 2; need += dn_cfg.fsk_count * sizeof(uint32_t); x = DN_C_SCH | DN_C_LINK | DN_C_FLOW; break; case DN_FS: /* queue show */ x = DN_C_FS | DN_C_QUEUE; break; case DN_GET_COMPAT: /* compatibility mode */ need = dn_compat_calc_size(); break; } a->flags = x; if (x & DN_C_SCH) { need += dn_cfg.schk_count * sizeof(struct dn_sch) / 2; /* NOT also, each fs might be attached to a sched */ need += dn_cfg.schk_count * sizeof(struct dn_id) / 2; } if (x & DN_C_FS) need += dn_cfg.fsk_count * sizeof(struct dn_fs); if (x & DN_C_LINK) { need += dn_cfg.schk_count * sizeof(struct dn_link) / 2; } /* * When exporting a queue to userland, only pass up the * struct dn_flow, which is the only visible part. */ if (x & DN_C_QUEUE) need += dn_cfg.queue_count * sizeof(struct dn_flow); if (x & DN_C_FLOW) need += dn_cfg.si_count * (sizeof(struct dn_flow)); return need; } /* * If compat != NULL dummynet_get is called in compatibility mode. * *compat will be the pointer to the buffer to pass to ipfw */ int dummynet_get(struct sockopt *sopt, void **compat) { int have, i, need, error; char *start = NULL, *buf; size_t sopt_valsize; struct dn_id *cmd; struct copy_args a; struct copy_range r; int l = sizeof(struct dn_id); bzero(&a, sizeof(a)); bzero(&r, sizeof(r)); /* save and restore original sopt_valsize around copyin */ sopt_valsize = sopt->sopt_valsize; cmd = &r.o; if (!compat) { /* copy at least an oid, and possibly a full object */ error = sooptcopyin(sopt, cmd, sizeof(r), sizeof(*cmd)); sopt->sopt_valsize = sopt_valsize; if (error) goto done; l = cmd->len; #ifdef EMULATE_SYSCTL /* sysctl emulation. */ if (cmd->type == DN_SYSCTL_GET) return kesysctl_emu_get(sopt); #endif if (l > sizeof(r)) { /* request larger than default, allocate buffer */ cmd = malloc(l, M_DUMMYNET, M_WAITOK); error = sooptcopyin(sopt, cmd, l, l); sopt->sopt_valsize = sopt_valsize; if (error) goto done; } } else { /* compatibility */ error = 0; cmd->type = DN_CMD_GET; cmd->len = sizeof(struct dn_id); cmd->subtype = DN_GET_COMPAT; // cmd->id = sopt_valsize; D("compatibility mode"); } a.extra = (struct copy_range *)cmd; if (cmd->len == sizeof(*cmd)) { /* no range, create a default */ uint32_t *rp = (uint32_t *)(cmd + 1); cmd->len += 2* sizeof(uint32_t); rp[0] = 1; rp[1] = DN_MAX_ID - 1; if (cmd->subtype == DN_LINK) { rp[0] += DN_MAX_ID; rp[1] += DN_MAX_ID; } } /* Count space (under lock) and allocate (outside lock). * Exit with lock held if we manage to get enough buffer. * Try a few times then give up. */ for (have = 0, i = 0; i < 10; i++) { DN_BH_WLOCK(); need = compute_space(cmd, &a); /* if there is a range, ignore value from compute_space() */ if (l > sizeof(*cmd)) need = sopt_valsize - sizeof(*cmd); if (need < 0) { DN_BH_WUNLOCK(); error = EINVAL; goto done; } need += sizeof(*cmd); cmd->id = need; if (have >= need) break; DN_BH_WUNLOCK(); if (start) free(start, M_DUMMYNET); start = NULL; if (need > sopt_valsize) break; have = need; start = malloc(have, M_DUMMYNET, M_WAITOK | M_ZERO); } if (start == NULL) { if (compat) { *compat = NULL; error = 1; // XXX } else { error = sooptcopyout(sopt, cmd, sizeof(*cmd)); } goto done; } ND("have %d:%d sched %d, %d:%d links %d, %d:%d flowsets %d, " "%d:%d si %d, %d:%d queues %d", dn_cfg.schk_count, sizeof(struct dn_sch), DN_SCH, dn_cfg.schk_count, sizeof(struct dn_link), DN_LINK, dn_cfg.fsk_count, sizeof(struct dn_fs), DN_FS, dn_cfg.si_count, sizeof(struct dn_flow), DN_SCH_I, dn_cfg.queue_count, sizeof(struct dn_queue), DN_QUEUE); sopt->sopt_valsize = sopt_valsize; a.type = cmd->subtype; if (compat == NULL) { bcopy(cmd, start, sizeof(*cmd)); ((struct dn_id*)(start))->len = sizeof(struct dn_id); buf = start + sizeof(*cmd); } else buf = start; a.start = &buf; a.end = start + have; /* start copying other objects */ if (compat) { a.type = DN_COMPAT_PIPE; dn_ht_scan(dn_cfg.schedhash, copy_data_helper_compat, &a); a.type = DN_COMPAT_QUEUE; dn_ht_scan(dn_cfg.fshash, copy_data_helper_compat, &a); } else if (a.type == DN_FS) { dn_ht_scan(dn_cfg.fshash, copy_data_helper, &a); } else { dn_ht_scan(dn_cfg.schedhash, copy_data_helper, &a); } DN_BH_WUNLOCK(); if (compat) { *compat = start; sopt->sopt_valsize = buf - start; /* free() is done by ip_dummynet_compat() */ start = NULL; //XXX hack } else { error = sooptcopyout(sopt, start, buf - start); } done: if (cmd && cmd != &r.o) free(cmd, M_DUMMYNET); if (start) free(start, M_DUMMYNET); return error; } /* Callback called on scheduler instance to delete it if idle */ static int drain_scheduler_cb(void *_si, void *arg) { struct dn_sch_inst *si = _si; if ((si->kflags & DN_ACTIVE) || si->dline.mq.head != NULL) return 0; if (si->sched->fp->flags & DN_MULTIQUEUE) { if (si->q_count == 0) return si_destroy(si, NULL); else return 0; } else { /* !DN_MULTIQUEUE */ if ((si+1)->ni.length == 0) return si_destroy(si, NULL); else return 0; } return 0; /* unreachable */ } /* Callback called on scheduler to check if it has instances */ static int drain_scheduler_sch_cb(void *_s, void *arg) { struct dn_schk *s = _s; if (s->sch.flags & DN_HAVE_MASK) { dn_ht_scan_bucket(s->siht, &s->drain_bucket, drain_scheduler_cb, NULL); s->drain_bucket++; } else { if (s->siht) { if (drain_scheduler_cb(s->siht, NULL) == DNHT_SCAN_DEL) s->siht = NULL; } } return 0; } /* Called every tick, try to delete a 'bucket' of scheduler */ void dn_drain_scheduler(void) { dn_ht_scan_bucket(dn_cfg.schedhash, &dn_cfg.drain_sch, drain_scheduler_sch_cb, NULL); dn_cfg.drain_sch++; } /* Callback called on queue to delete if it is idle */ static int drain_queue_cb(void *_q, void *arg) { struct dn_queue *q = _q; if (q->ni.length == 0) { dn_delete_queue(q, DN_DESTROY); return DNHT_SCAN_DEL; /* queue is deleted */ } return 0; /* queue isn't deleted */ } /* Callback called on flowset used to check if it has queues */ static int drain_queue_fs_cb(void *_fs, void *arg) { struct dn_fsk *fs = _fs; if (fs->fs.flags & DN_QHT_HASH) { /* Flowset has a hash table for queues */ dn_ht_scan_bucket(fs->qht, &fs->drain_bucket, drain_queue_cb, NULL); fs->drain_bucket++; } else { /* No hash table for this flowset, null the pointer * if the queue is deleted */ if (fs->qht) { if (drain_queue_cb(fs->qht, NULL) == DNHT_SCAN_DEL) fs->qht = NULL; } } return 0; } /* Called every tick, try to delete a 'bucket' of queue */ void dn_drain_queue(void) { /* scan a bucket of flowset */ dn_ht_scan_bucket(dn_cfg.fshash, &dn_cfg.drain_fs, drain_queue_fs_cb, NULL); dn_cfg.drain_fs++; } /* * Handler for the various dummynet socket options */ static int ip_dn_ctl(struct sockopt *sopt) { void *p = NULL; int error, l; error = priv_check(sopt->sopt_td, PRIV_NETINET_DUMMYNET); if (error) return (error); /* Disallow sets in really-really secure mode. */ if (sopt->sopt_dir == SOPT_SET) { error = securelevel_ge(sopt->sopt_td->td_ucred, 3); if (error) return (error); } switch (sopt->sopt_name) { default : D("dummynet: unknown option %d", sopt->sopt_name); error = EINVAL; break; case IP_DUMMYNET_FLUSH: case IP_DUMMYNET_CONFIGURE: case IP_DUMMYNET_DEL: /* remove a pipe or queue */ case IP_DUMMYNET_GET: D("dummynet: compat option %d", sopt->sopt_name); error = ip_dummynet_compat(sopt); break; case IP_DUMMYNET3 : if (sopt->sopt_dir == SOPT_GET) { error = dummynet_get(sopt, NULL); break; } l = sopt->sopt_valsize; if (l < sizeof(struct dn_id) || l > 12000) { D("argument len %d invalid", l); break; } p = malloc(l, M_TEMP, M_WAITOK); // XXX can it fail ? error = sooptcopyin(sopt, p, l, l); if (error) break ; error = do_config(p, l); break; } if (p != NULL) free(p, M_TEMP); return error ; } static void ip_dn_init(void) { if (dn_cfg.init_done) return; printf("DUMMYNET %p with IPv6 initialized (100409)\n", curvnet); dn_cfg.init_done = 1; /* Set defaults here. MSVC does not accept initializers, * and this is also useful for vimages */ /* queue limits */ dn_cfg.slot_limit = 100; /* Foot shooting limit for queues. */ dn_cfg.byte_limit = 1024 * 1024; dn_cfg.expire = 1; /* RED parameters */ dn_cfg.red_lookup_depth = 256; /* default lookup table depth */ dn_cfg.red_avg_pkt_size = 512; /* default medium packet size */ dn_cfg.red_max_pkt_size = 1500; /* default max packet size */ /* hash tables */ dn_cfg.max_hash_size = 65536; /* max in the hash tables */ dn_cfg.hash_size = 64; /* default hash size */ /* create hash tables for schedulers and flowsets. * In both we search by key and by pointer. */ dn_cfg.schedhash = dn_ht_init(NULL, dn_cfg.hash_size, offsetof(struct dn_schk, schk_next), schk_hash, schk_match, schk_new); dn_cfg.fshash = dn_ht_init(NULL, dn_cfg.hash_size, offsetof(struct dn_fsk, fsk_next), fsk_hash, fsk_match, fsk_new); /* bucket index to drain object */ dn_cfg.drain_fs = 0; dn_cfg.drain_sch = 0; heap_init(&dn_cfg.evheap, 16, offsetof(struct dn_id, id)); SLIST_INIT(&dn_cfg.fsu); SLIST_INIT(&dn_cfg.schedlist); DN_LOCK_INIT(); TASK_INIT(&dn_task, 0, dummynet_task, curvnet); dn_tq = taskqueue_create_fast("dummynet", M_WAITOK, taskqueue_thread_enqueue, &dn_tq); taskqueue_start_threads(&dn_tq, 1, PI_NET, "dummynet"); callout_init(&dn_timeout, CALLOUT_MPSAFE); dn_reschedule(); /* Initialize curr_time adjustment mechanics. */ getmicrouptime(&dn_cfg.prev_t); } static void ip_dn_destroy(int last) { DN_BH_WLOCK(); /* ensure no more callouts are started */ dn_gone = 1; /* check for last */ if (last) { ND("removing last instance\n"); ip_dn_ctl_ptr = NULL; ip_dn_io_ptr = NULL; } dummynet_flush(); DN_BH_WUNLOCK(); callout_drain(&dn_timeout); taskqueue_drain(dn_tq, &dn_task); taskqueue_free(dn_tq); dn_ht_free(dn_cfg.schedhash, 0); dn_ht_free(dn_cfg.fshash, 0); heap_free(&dn_cfg.evheap); DN_LOCK_DESTROY(); } static int dummynet_modevent(module_t mod, int type, void *data) { if (type == MOD_LOAD) { if (ip_dn_io_ptr) { printf("DUMMYNET already loaded\n"); return EEXIST ; } ip_dn_init(); ip_dn_ctl_ptr = ip_dn_ctl; ip_dn_io_ptr = dummynet_io; return 0; } else if (type == MOD_UNLOAD) { ip_dn_destroy(1 /* last */); return 0; } else return EOPNOTSUPP; } /* modevent helpers for the modules */ static int load_dn_sched(struct dn_alg *d) { struct dn_alg *s; if (d == NULL) return 1; /* error */ ip_dn_init(); /* just in case, we need the lock */ /* Check that mandatory funcs exists */ if (d->enqueue == NULL || d->dequeue == NULL) { D("missing enqueue or dequeue for %s", d->name); return 1; } /* Search if scheduler already exists */ DN_BH_WLOCK(); SLIST_FOREACH(s, &dn_cfg.schedlist, next) { if (strcmp(s->name, d->name) == 0) { D("%s already loaded", d->name); break; /* scheduler already exists */ } } if (s == NULL) SLIST_INSERT_HEAD(&dn_cfg.schedlist, d, next); DN_BH_WUNLOCK(); D("dn_sched %s %sloaded", d->name, s ? "not ":""); return s ? 1 : 0; } static int unload_dn_sched(struct dn_alg *s) { struct dn_alg *tmp, *r; int err = EINVAL; ND("called for %s", s->name); DN_BH_WLOCK(); SLIST_FOREACH_SAFE(r, &dn_cfg.schedlist, next, tmp) { if (strcmp(s->name, r->name) != 0) continue; ND("ref_count = %d", r->ref_count); err = (r->ref_count != 0) ? EBUSY : 0; if (err == 0) SLIST_REMOVE(&dn_cfg.schedlist, r, dn_alg, next); break; } DN_BH_WUNLOCK(); D("dn_sched %s %sunloaded", s->name, err ? "not ":""); return err; } int dn_sched_modevent(module_t mod, int cmd, void *arg) { struct dn_alg *sch = arg; if (cmd == MOD_LOAD) return load_dn_sched(sch); else if (cmd == MOD_UNLOAD) return unload_dn_sched(sch); else return EINVAL; } static moduledata_t dummynet_mod = { "dummynet", dummynet_modevent, NULL }; #define DN_SI_SUB SI_SUB_PROTO_IFATTACHDOMAIN #define DN_MODEV_ORD (SI_ORDER_ANY - 128) /* after ipfw */ DECLARE_MODULE(dummynet, dummynet_mod, DN_SI_SUB, DN_MODEV_ORD); MODULE_DEPEND(dummynet, ipfw, 2, 2, 2); MODULE_VERSION(dummynet, 3); /* * Starting up. Done in order after dummynet_modevent() has been called. * VNET_SYSINIT is also called for each existing vnet and each new vnet. */ //VNET_SYSINIT(vnet_dn_init, DN_SI_SUB, DN_MODEV_ORD+2, ip_dn_init, NULL); /* * Shutdown handlers up shop. These are done in REVERSE ORDER, but still * after dummynet_modevent() has been called. Not called on reboot. * VNET_SYSUNINIT is also called for each exiting vnet as it exits. * or when the module is unloaded. */ //VNET_SYSUNINIT(vnet_dn_uninit, DN_SI_SUB, DN_MODEV_ORD+2, ip_dn_destroy, NULL); /* end of file */ Index: stable/10 =================================================================== --- stable/10 (revision 301230) +++ stable/10 (revision 301231) Property changes on: stable/10 ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head:r266941,266955