net/ipv4/netfilter/ipt_REJECT.c

   1 /*
   2  * This is a module which is used for rejecting packets.
   3  * Added support for customized reject packets (Jozsef Kadlecsik).
   4  * Added support for ICMP type-3-code-13 (Maciej Soltysiak). [RFC 1812]
   5  */
   6
   7 /* (C) 1999-2001 Paul `Rusty' Russell
   8  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  */
  14
  15 #include <linux/config.h>
  16 #include <linux/module.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/ip.h>
  19 #include <linux/udp.h>
  20 #include <linux/icmp.h>
  21 #include <net/icmp.h>
  22 #include <net/ip.h>
  23 #include <net/tcp.h>
  24 #include <net/route.h>
  25 #include <linux/netfilter_ipv4/ip_tables.h>
  26 #include <linux/netfilter_ipv4/ipt_REJECT.h>
  27 #ifdef CONFIG_BRIDGE_NETFILTER
  28 #include <linux/netfilter_bridge.h>
  29 #endif
  30
  31 MODULE_LICENSE("GPL");
  32 MODULE_AUTHOR("Netfilter Core Team <coreteam@netfilter.org>");
  33 MODULE_DESCRIPTION("iptables REJECT target module");
  34
  35 #if 0
  36 #define DEBUGP printk
  37 #else
  38 #define DEBUGP(format, args...)
  39 #endif
  40
  41 /* If the original packet is part of a connection, but the connection
  42    is not confirmed, our manufactured reply will not be associated
  43    with it, so we need to do this manually. */
  44 static void connection_attach(struct sk_buff *new_skb, struct nf_ct_info *nfct)
  45 {
  46         void (*attach)(struct sk_buff *, struct nf_ct_info *);
  47
  48         /* Avoid module unload race with ip_ct_attach being NULLed out */
  49         if (nfct && (attach = ip_ct_attach) != NULL) {
  50                 mb(); /* Just to be sure: must be read before executing this */
  51                 attach(new_skb, nfct);
  52         }
  53 }
  54
  55 static inline struct rtable *route_reverse(struct sk_buff *skb, int hook)
  56 {
  57         struct iphdr *iph = skb->nh.iph;
  58         struct dst_entry *odst;
  59         struct flowi fl = {};
  60         struct rtable *rt;
  61
  62         /* We don't require ip forwarding to be enabled to be able to
  63          * send a RST reply for bridged traffic. */
  64         if (hook != NF_IP_FORWARD
  65 #ifdef CONFIG_BRIDGE_NETFILTER
  66             || (skb->nf_bridge && skb->nf_bridge->mask & BRNF_BRIDGED)
  67 #endif
  68            ) {
  69                 fl.nl_u.ip4_u.daddr = iph->saddr;
  70                 if (hook == NF_IP_LOCAL_IN)
  71                         fl.nl_u.ip4_u.saddr = iph->daddr;
  72                 fl.nl_u.ip4_u.tos = RT_TOS(iph->tos);
  73
  74                 if (ip_route_output_key(&rt, &fl) != 0)
  75                         return NULL;
  76         } else {
  77                 /* non-local src, find valid iif to satisfy
  78                  * rp-filter when calling ip_route_input. */
  79                 fl.nl_u.ip4_u.daddr = iph->daddr;
  80                 if (ip_route_output_key(&rt, &fl) != 0)
  81                         return NULL;
  82
  83                 odst = skb->dst;
  84                 if (ip_route_input(skb, iph->saddr, iph->daddr,
  85                                    RT_TOS(iph->tos), rt->u.dst.dev) != 0) {
  86                         dst_release(&rt->u.dst);
  87                         return NULL;
  88                 }
  89                 dst_release(&rt->u.dst);
  90                 rt = (struct rtable *)skb->dst;
  91                 skb->dst = odst;
  92         }
  93
  94         if (rt->u.dst.error) {
  95                 dst_release(&rt->u.dst);
  96                 rt = NULL;
  97         }
  98
  99         return rt;
 100 }
 101
 102 /* Send RST reply */
 103 static void send_reset(struct sk_buff *oldskb, int hook)
 104 {
 105         struct sk_buff *nskb;
 106         struct tcphdr otcph, *tcph;
 107         struct rtable *rt;
 108         u_int16_t tmp_port;
 109         u_int32_t tmp_addr;
 110         int needs_ack;
 111         int hh_len;
 112
 113         /* IP header checks: fragment. */
 114         if (oldskb->nh.iph->frag_off & htons(IP_OFFSET))
 115                 return;
 116
 117         if (skb_copy_bits(oldskb, oldskb->nh.iph->ihl*4,
 118                           &otcph, sizeof(otcph)) < 0)
 119                 return;
 120
 121         /* No RST for RST. */
 122         if (otcph.rst)
 123                 return;
 124
 125         /* FIXME: Check checksum --RR */
 126         if ((rt = route_reverse(oldskb, hook)) == NULL)
 127                 return;
 128
 129         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
 130
 131         /* We need a linear, writeable skb.  We also need to expand
 132            headroom in case hh_len of incoming interface < hh_len of
 133            outgoing interface */
 134         nskb = skb_copy_expand(oldskb, hh_len, skb_tailroom(oldskb),
 135                                GFP_ATOMIC);
 136         if (!nskb) {
 137                 dst_release(&rt->u.dst);
 138                 return;
 139         }
 140
 141         dst_release(nskb->dst);
 142         nskb->dst = &rt->u.dst;
 143
 144         /* This packet will not be the same as the other: clear nf fields */
 145         nf_reset(nskb);
 146         nskb->nfcache = 0;
 147         nskb->nfmark = 0;
 148 #ifdef CONFIG_BRIDGE_NETFILTER
 149         nf_bridge_put(nskb->nf_bridge);
 150         nskb->nf_bridge = NULL;
 151 #endif
 152
 153         tcph = (struct tcphdr *)((u_int32_t*)nskb->nh.iph + nskb->nh.iph->ihl);
 154
 155         /* Swap source and dest */
 156         tmp_addr = nskb->nh.iph->saddr;
 157         nskb->nh.iph->saddr = nskb->nh.iph->daddr;
 158         nskb->nh.iph->daddr = tmp_addr;
 159         tmp_port = tcph->source;
 160         tcph->source = tcph->dest;
 161         tcph->dest = tmp_port;
 162
 163         /* Truncate to length (no data) */
 164         tcph->doff = sizeof(struct tcphdr)/4;
 165         skb_trim(nskb, nskb->nh.iph->ihl*4 + sizeof(struct tcphdr));
 166         nskb->nh.iph->tot_len = htons(nskb->len);
 167
 168         if (tcph->ack) {
 169                 needs_ack = 0;
 170                 tcph->seq = otcph.ack_seq;
 171                 tcph->ack_seq = 0;
 172         } else {
 173                 needs_ack = 1;
 174                 tcph->ack_seq = htonl(ntohl(otcph.seq) + otcph.syn + otcph.fin
 175                                       + oldskb->len - oldskb->nh.iph->ihl*4
 176                                       - (otcph.doff<<2));
 177                 tcph->seq = 0;
 178         }
 179
 180         /* Reset flags */
 181         ((u_int8_t *)tcph)[13] = 0;
 182         tcph->rst = 1;
 183         tcph->ack = needs_ack;
 184
 185         tcph->window = 0;
 186         tcph->urg_ptr = 0;
 187
 188         /* Adjust TCP checksum */
 189         tcph->check = 0;
 190         tcph->check = tcp_v4_check(tcph, sizeof(struct tcphdr),
 191                                    nskb->nh.iph->saddr,
 192                                    nskb->nh.iph->daddr,
 193                                    csum_partial((char *)tcph,
 194                                                 sizeof(struct tcphdr), 0));
 195
 196         /* Adjust IP TTL, DF */
 197         nskb->nh.iph->ttl = MAXTTL;
 198         /* Set DF, id = 0 */
 199         nskb->nh.iph->frag_off = htons(IP_DF);
 200         nskb->nh.iph->id = 0;
 201
 202         /* Adjust IP checksum */
 203         nskb->nh.iph->check = 0;
 204         nskb->nh.iph->check = ip_fast_csum((unsigned char *)nskb->nh.iph,
 205                                            nskb->nh.iph->ihl);
 206
 207         /* "Never happens" */
 208         if (nskb->len > dst_pmtu(nskb->dst))
 209                 goto free_nskb;
 210
 211         connection_attach(nskb, oldskb->nfct);
 212
 213         NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, nskb, NULL, nskb->dst->dev,
 214                 ip_finish_output);
 215         return;
 216
 217  free_nskb:
 218         kfree_skb(nskb);
 219 }
 220
 221 static void send_unreach(struct sk_buff *skb_in, int code)
 222 {
 223         struct iphdr *iph;
 224         struct udphdr *udph;
 225         struct icmphdr *icmph;
 226         struct sk_buff *nskb;
 227         u32 saddr;
 228         u8 tos;
 229         int hh_len, length;
 230         struct rtable *rt = (struct rtable*)skb_in->dst;
 231         unsigned char *data;
 232
 233         if (!rt)
 234                 return;
 235
 236         /* FIXME: Use sysctl number. --RR */
 237         if (!xrlim_allow(&rt->u.dst, 1*HZ))
 238                 return;
 239
 240         iph = skb_in->nh.iph;
 241
 242         /* No replies to physical multicast/broadcast */
 243         if (skb_in->pkt_type!=PACKET_HOST)
 244                 return;
 245
 246         /* Now check at the protocol level */
 247         if (rt->rt_flags&(RTCF_BROADCAST|RTCF_MULTICAST))
 248                 return;
 249
 250         /* Only reply to fragment 0. */
 251         if (iph->frag_off&htons(IP_OFFSET))
 252                 return;
 253
 254         /* Ensure we have at least 8 bytes of proto header. */
 255         if (skb_in->len < skb_in->nh.iph->ihl*4 + 8)
 256                 return;
 257
 258         /* if UDP checksum is set, verify it's correct */
 259         if (iph->protocol == IPPROTO_UDP
 260             && skb_in->tail-(u8*)iph >= sizeof(struct udphdr)) {
 261                 int datalen = skb_in->len - (iph->ihl<<2);
 262                 udph = (struct udphdr *)((char *)iph + (iph->ihl<<2));
 263                 if (udph->check
 264                     && csum_tcpudp_magic(iph->saddr, iph->daddr,
 265                                          datalen, IPPROTO_UDP,
 266                                          csum_partial((char *)udph, datalen,
 267                                                       0)) != 0)
 268                         return;
 269         }
 270
 271         /* If we send an ICMP error to an ICMP error a mess would result.. */
 272         if (iph->protocol == IPPROTO_ICMP
 273             && skb_in->tail-(u8*)iph >= sizeof(struct icmphdr)) {
 274                 icmph = (struct icmphdr *)((char *)iph + (iph->ihl<<2));
 275
 276                 if (skb_copy_bits(skb_in, skb_in->nh.iph->ihl*4,
 277                                   icmph, sizeof(*icmph)) < 0)
 278                         return;
 279
 280                 /* Between echo-reply (0) and timestamp (13),
 281                    everything except echo-request (8) is an error.
 282                    Also, anything greater than NR_ICMP_TYPES is
 283                    unknown, and hence should be treated as an error... */
 284                 if ((icmph->type < ICMP_TIMESTAMP
 285                      && icmph->type != ICMP_ECHOREPLY
 286                      && icmph->type != ICMP_ECHO)
 287                     || icmph->type > NR_ICMP_TYPES)
 288                         return;
 289         }
 290
 291         saddr = iph->daddr;
 292         if (!(rt->rt_flags & RTCF_LOCAL))
 293                 saddr = 0;
 294
 295         tos = (iph->tos & IPTOS_TOS_MASK) | IPTOS_PREC_INTERNETCONTROL;
 296
 297         {
 298                 struct flowi fl = { .nl_u = { .ip4_u =
 299                                               { .daddr = skb_in->nh.iph->saddr,
 300                                                 .saddr = saddr,
 301                                                 .tos = RT_TOS(tos) } } };
 302                 if (ip_route_output_key(&rt, &fl))
 303                         return;
 304         }
 305         /* RFC says return as much as we can without exceeding 576 bytes. */
 306         length = skb_in->len + sizeof(struct iphdr) + sizeof(struct icmphdr);
 307
 308         if (length > dst_pmtu(&rt->u.dst))
 309                 length = dst_pmtu(&rt->u.dst);
 310         if (length > 576)
 311                 length = 576;
 312
 313         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
 314
 315         nskb = alloc_skb(hh_len + length, GFP_ATOMIC);
 316         if (!nskb) {
 317                 ip_rt_put(rt);
 318                 return;
 319         }
 320
 321         nskb->priority = 0;
 322         nskb->dst = &rt->u.dst;
 323         skb_reserve(nskb, hh_len);
 324
 325         /* Set up IP header */
 326         iph = nskb->nh.iph
 327                 = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
 328         iph->version=4;
 329         iph->ihl=5;
 330         iph->tos=tos;
 331         iph->tot_len = htons(length);
 332
 333         /* PMTU discovery never applies to ICMP packets. */
 334         iph->frag_off = 0;
 335
 336         iph->ttl = MAXTTL;
 337         ip_select_ident(iph, &rt->u.dst, NULL);
 338         iph->protocol=IPPROTO_ICMP;
 339         iph->saddr=rt->rt_src;
 340         iph->daddr=rt->rt_dst;
 341         iph->check=0;
 342         iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
 343
 344         /* Set up ICMP header. */
 345         icmph = nskb->h.icmph
 346                 = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
 347         icmph->type = ICMP_DEST_UNREACH;
 348         icmph->code = code;
 349         icmph->un.gateway = 0;
 350         icmph->checksum = 0;
 351
 352         /* Copy as much of original packet as will fit */
 353         data = skb_put(nskb,
 354                        length - sizeof(struct iphdr) - sizeof(struct icmphdr));
 355
 356         skb_copy_bits(skb_in, 0, data,
 357                       length - sizeof(struct iphdr) - sizeof(struct icmphdr));
 358
 359         icmph->checksum = ip_compute_csum((unsigned char *)icmph,
 360                                           length - sizeof(struct iphdr));
 361
 362         connection_attach(nskb, skb_in->nfct);
 363
 364         NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, nskb, NULL, nskb->dst->dev,
 365                 ip_finish_output);
 366 }
 367
 368 static unsigned int reject(struct sk_buff **pskb,
 369                            const struct net_device *in,
 370                            const struct net_device *out,
 371                            unsigned int hooknum,
 372                            const void *targinfo,
 373                            void *userinfo)
 374 {
 375         const struct ipt_reject_info *reject = targinfo;
 376
 377         /* Our naive response construction doesn't deal with IP
 378            options, and probably shouldn't try. */
 379         if ((*pskb)->nh.iph->ihl<<2 != sizeof(struct iphdr))
 380                 return NF_DROP;
 381
 382         /* WARNING: This code causes reentry within iptables.
 383            This means that the iptables jump stack is now crap.  We
 384            must return an absolute verdict. --RR */
 385         switch (reject->with) {
 386         case IPT_ICMP_NET_UNREACHABLE:
 387                 send_unreach(*pskb, ICMP_NET_UNREACH);
 388                 break;
 389         case IPT_ICMP_HOST_UNREACHABLE:
 390                 send_unreach(*pskb, ICMP_HOST_UNREACH);
 391                 break;
 392         case IPT_ICMP_PROT_UNREACHABLE:
 393                 send_unreach(*pskb, ICMP_PROT_UNREACH);
 394                 break;
 395         case IPT_ICMP_PORT_UNREACHABLE:
 396                 send_unreach(*pskb, ICMP_PORT_UNREACH);
 397                 break;
 398         case IPT_ICMP_NET_PROHIBITED:
 399                 send_unreach(*pskb, ICMP_NET_ANO);
 400                 break;
 401         case IPT_ICMP_HOST_PROHIBITED:
 402                 send_unreach(*pskb, ICMP_HOST_ANO);
 403                 break;
 404         case IPT_ICMP_ADMIN_PROHIBITED:
 405                 send_unreach(*pskb, ICMP_PKT_FILTERED);
 406                 break;
 407         case IPT_TCP_RESET:
 408                 send_reset(*pskb, hooknum);
 409         case IPT_ICMP_ECHOREPLY:
 410                 /* Doesn't happen. */
 411                 break;
 412         }
 413
 414         return NF_DROP;
 415 }
 416
 417 static int check(const char *tablename,
 418                  const struct ipt_entry *e,
 419                  void *targinfo,
 420                  unsigned int targinfosize,
 421                  unsigned int hook_mask)
 422 {
 423         const struct ipt_reject_info *rejinfo = targinfo;
 424
 425         if (targinfosize != IPT_ALIGN(sizeof(struct ipt_reject_info))) {
 426                 DEBUGP("REJECT: targinfosize %u != 0\n", targinfosize);
 427                 return 0;
 428         }
 429
 430         /* Only allow these for packet filtering. */
 431         if (strcmp(tablename, "filter") != 0) {
 432                 DEBUGP("REJECT: bad table `%s'.\n", tablename);
 433                 return 0;
 434         }
 435         if ((hook_mask & ~((1 << NF_IP_LOCAL_IN)
 436                            | (1 << NF_IP_FORWARD)
 437                            | (1 << NF_IP_LOCAL_OUT))) != 0) {
 438                 DEBUGP("REJECT: bad hook mask %X\n", hook_mask);
 439                 return 0;
 440         }
 441
 442         if (rejinfo->with == IPT_ICMP_ECHOREPLY) {
 443                 printk("REJECT: ECHOREPLY no longer supported.\n");
 444                 return 0;
 445         } else if (rejinfo->with == IPT_TCP_RESET) {
 446                 /* Must specify that it's a TCP packet */
 447                 if (e->ip.proto != IPPROTO_TCP
 448                     || (e->ip.invflags & IPT_INV_PROTO)) {
 449                         DEBUGP("REJECT: TCP_RESET invalid for non-tcp\n");
 450                         return 0;
 451                 }
 452         }
 453
 454         return 1;
 455 }
 456
 457 static struct ipt_target ipt_reject_reg = {
 458         .name           = "REJECT",
 459         .target         = reject,
 460         .checkentry     = check,
 461         .me             = THIS_MODULE,
 462 };
 463
 464 static int __init init(void)
 465 {
 466         return ipt_register_target(&ipt_reject_reg);
 467 }
 468
 469 static void __exit fini(void)
 470 {
 471         ipt_unregister_target(&ipt_reject_reg);
 472 }
 473
 474 module_init(init);
 475 module_exit(fini);