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