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_conntrack_put(nskb->nfct);
 146         nskb->nfct = NULL;
 147         nskb->nfcache = 0;
 148 #ifdef CONFIG_NETFILTER_DEBUG
 149         nskb->nf_debug = 0;
 150 #endif
 151         nskb->nfmark = 0;
 152 #ifdef CONFIG_BRIDGE_NETFILTER
 153         nf_bridge_put(nskb->nf_bridge);
 154         nskb->nf_bridge = NULL;
 155 #endif
 156
 157         tcph = (struct tcphdr *)((u_int32_t*)nskb->nh.iph + nskb->nh.iph->ihl);
 158
 159         /* Swap source and dest */
 160         tmp_addr = nskb->nh.iph->saddr;
 161         nskb->nh.iph->saddr = nskb->nh.iph->daddr;
 162         nskb->nh.iph->daddr = tmp_addr;
 163         tmp_port = tcph->source;
 164         tcph->source = tcph->dest;
 165         tcph->dest = tmp_port;
 166
 167         /* Truncate to length (no data) */
 168         tcph->doff = sizeof(struct tcphdr)/4;
 169         skb_trim(nskb, nskb->nh.iph->ihl*4 + sizeof(struct tcphdr));
 170         nskb->nh.iph->tot_len = htons(nskb->len);
 171
 172         if (tcph->ack) {
 173                 needs_ack = 0;
 174                 tcph->seq = otcph.ack_seq;
 175                 tcph->ack_seq = 0;
 176         } else {
 177                 needs_ack = 1;
 178                 tcph->ack_seq = htonl(ntohl(otcph.seq) + otcph.syn + otcph.fin
 179                                       + oldskb->len - oldskb->nh.iph->ihl*4
 180                                       - (otcph.doff<<2));
 181                 tcph->seq = 0;
 182         }
 183
 184         /* Reset flags */
 185         ((u_int8_t *)tcph)[13] = 0;
 186         tcph->rst = 1;
 187         tcph->ack = needs_ack;
 188
 189         tcph->window = 0;
 190         tcph->urg_ptr = 0;
 191
 192         /* Adjust TCP checksum */
 193         tcph->check = 0;
 194         tcph->check = tcp_v4_check(tcph, sizeof(struct tcphdr),
 195                                    nskb->nh.iph->saddr,
 196                                    nskb->nh.iph->daddr,
 197                                    csum_partial((char *)tcph,
 198                                                 sizeof(struct tcphdr), 0));
 199
 200         /* Adjust IP TTL, DF */
 201         nskb->nh.iph->ttl = MAXTTL;
 202         /* Set DF, id = 0 */
 203         nskb->nh.iph->frag_off = htons(IP_DF);
 204         nskb->nh.iph->id = 0;
 205
 206         /* Adjust IP checksum */
 207         nskb->nh.iph->check = 0;
 208         nskb->nh.iph->check = ip_fast_csum((unsigned char *)nskb->nh.iph,
 209                                            nskb->nh.iph->ihl);
 210
 211         /* "Never happens" */
 212         if (nskb->len > dst_pmtu(nskb->dst))
 213                 goto free_nskb;
 214
 215         connection_attach(nskb, oldskb->nfct);
 216
 217         NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, nskb, NULL, nskb->dst->dev,
 218                 ip_finish_output);
 219         return;
 220
 221  free_nskb:
 222         kfree_skb(nskb);
 223 }
 224
 225 static void send_unreach(struct sk_buff *skb_in, int code)
 226 {
 227         struct iphdr *iph;
 228         struct udphdr *udph;
 229         struct icmphdr *icmph;
 230         struct sk_buff *nskb;
 231         u32 saddr;
 232         u8 tos;
 233         int hh_len, length;
 234         struct rtable *rt = (struct rtable*)skb_in->dst;
 235         unsigned char *data;
 236
 237         if (!rt)
 238                 return;
 239
 240         /* FIXME: Use sysctl number. --RR */
 241         if (!xrlim_allow(&rt->u.dst, 1*HZ))
 242                 return;
 243
 244         iph = skb_in->nh.iph;
 245
 246         /* No replies to physical multicast/broadcast */
 247         if (skb_in->pkt_type!=PACKET_HOST)
 248                 return;
 249
 250         /* Now check at the protocol level */
 251         if (rt->rt_flags&(RTCF_BROADCAST|RTCF_MULTICAST))
 252                 return;
 253
 254         /* Only reply to fragment 0. */
 255         if (iph->frag_off&htons(IP_OFFSET))
 256                 return;
 257
 258         /* Ensure we have at least 8 bytes of proto header. */
 259         if (skb_in->len < skb_in->nh.iph->ihl*4 + 8)
 260                 return;
 261
 262         /* if UDP checksum is set, verify it's correct */
 263         if (iph->protocol == IPPROTO_UDP
 264             && skb_in->tail-(u8*)iph >= sizeof(struct udphdr)) {
 265                 int datalen = skb_in->len - (iph->ihl<<2);
 266                 udph = (struct udphdr *)((char *)iph + (iph->ihl<<2));
 267                 if (udph->check
 268                     && csum_tcpudp_magic(iph->saddr, iph->daddr,
 269                                          datalen, IPPROTO_UDP,
 270                                          csum_partial((char *)udph, datalen,
 271                                                       0)) != 0)
 272                         return;
 273         }
 274
 275         /* If we send an ICMP error to an ICMP error a mess would result.. */
 276         if (iph->protocol == IPPROTO_ICMP
 277             && skb_in->tail-(u8*)iph >= sizeof(struct icmphdr)) {
 278                 icmph = (struct icmphdr *)((char *)iph + (iph->ihl<<2));
 279
 280                 if (skb_copy_bits(skb_in, skb_in->nh.iph->ihl*4,
 281                                   icmph, sizeof(*icmph)) < 0)
 282                         return;
 283
 284                 /* Between echo-reply (0) and timestamp (13),
 285                    everything except echo-request (8) is an error.
 286                    Also, anything greater than NR_ICMP_TYPES is
 287                    unknown, and hence should be treated as an error... */
 288                 if ((icmph->type < ICMP_TIMESTAMP
 289                      && icmph->type != ICMP_ECHOREPLY
 290                      && icmph->type != ICMP_ECHO)
 291                     || icmph->type > NR_ICMP_TYPES)
 292                         return;
 293         }
 294
 295         saddr = iph->daddr;
 296         if (!(rt->rt_flags & RTCF_LOCAL))
 297                 saddr = 0;
 298
 299         tos = (iph->tos & IPTOS_TOS_MASK) | IPTOS_PREC_INTERNETCONTROL;
 300
 301         {
 302                 struct flowi fl = { .nl_u = { .ip4_u =
 303                                               { .daddr = skb_in->nh.iph->saddr,
 304                                                 .saddr = saddr,
 305                                                 .tos = RT_TOS(tos) } } };
 306                 if (ip_route_output_key(&rt, &fl))
 307                         return;
 308         }
 309         /* RFC says return as much as we can without exceeding 576 bytes. */
 310         length = skb_in->len + sizeof(struct iphdr) + sizeof(struct icmphdr);
 311
 312         if (length > dst_pmtu(&rt->u.dst))
 313                 length = dst_pmtu(&rt->u.dst);
 314         if (length > 576)
 315                 length = 576;
 316
 317         hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
 318
 319         nskb = alloc_skb(hh_len + length, GFP_ATOMIC);
 320         if (!nskb) {
 321                 ip_rt_put(rt);
 322                 return;
 323         }
 324
 325         nskb->priority = 0;
 326         nskb->dst = &rt->u.dst;
 327         skb_reserve(nskb, hh_len);
 328
 329         /* Set up IP header */
 330         iph = nskb->nh.iph
 331                 = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
 332         iph->version=4;
 333         iph->ihl=5;
 334         iph->tos=tos;
 335         iph->tot_len = htons(length);
 336
 337         /* PMTU discovery never applies to ICMP packets. */
 338         iph->frag_off = 0;
 339
 340         iph->ttl = MAXTTL;
 341         ip_select_ident(iph, &rt->u.dst, NULL);
 342         iph->protocol=IPPROTO_ICMP;
 343         iph->saddr=rt->rt_src;
 344         iph->daddr=rt->rt_dst;
 345         iph->check=0;
 346         iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
 347
 348         /* Set up ICMP header. */
 349         icmph = nskb->h.icmph
 350                 = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
 351         icmph->type = ICMP_DEST_UNREACH;
 352         icmph->code = code;
 353         icmph->un.gateway = 0;
 354         icmph->checksum = 0;
 355
 356         /* Copy as much of original packet as will fit */
 357         data = skb_put(nskb,
 358                        length - sizeof(struct iphdr) - sizeof(struct icmphdr));
 359
 360         skb_copy_bits(skb_in, 0, data,
 361                       length - sizeof(struct iphdr) - sizeof(struct icmphdr));
 362
 363         icmph->checksum = ip_compute_csum((unsigned char *)icmph,
 364                                           length - sizeof(struct iphdr));
 365
 366         connection_attach(nskb, skb_in->nfct);
 367
 368         NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, nskb, NULL, nskb->dst->dev,
 369                 ip_finish_output);
 370 }
 371
 372 static unsigned int reject(struct sk_buff **pskb,
 373                            const struct net_device *in,
 374                            const struct net_device *out,
 375                            unsigned int hooknum,
 376                            const void *targinfo,
 377                            void *userinfo)
 378 {
 379         const struct ipt_reject_info *reject = targinfo;
 380
 381         /* Our naive response construction doesn't deal with IP
 382            options, and probably shouldn't try. */
 383         if ((*pskb)->nh.iph->ihl<<2 != sizeof(struct iphdr))
 384                 return NF_DROP;
 385
 386         /* WARNING: This code causes reentry within iptables.
 387            This means that the iptables jump stack is now crap.  We
 388            must return an absolute verdict. --RR */
 389         switch (reject->with) {
 390         case IPT_ICMP_NET_UNREACHABLE:
 391                 send_unreach(*pskb, ICMP_NET_UNREACH);
 392                 break;
 393         case IPT_ICMP_HOST_UNREACHABLE:
 394                 send_unreach(*pskb, ICMP_HOST_UNREACH);
 395                 break;
 396         case IPT_ICMP_PROT_UNREACHABLE:
 397                 send_unreach(*pskb, ICMP_PROT_UNREACH);
 398                 break;
 399         case IPT_ICMP_PORT_UNREACHABLE:
 400                 send_unreach(*pskb, ICMP_PORT_UNREACH);
 401                 break;
 402         case IPT_ICMP_NET_PROHIBITED:
 403                 send_unreach(*pskb, ICMP_NET_ANO);
 404                 break;
 405         case IPT_ICMP_HOST_PROHIBITED:
 406                 send_unreach(*pskb, ICMP_HOST_ANO);
 407                 break;
 408         case IPT_ICMP_ADMIN_PROHIBITED:
 409                 send_unreach(*pskb, ICMP_PKT_FILTERED);
 410                 break;
 411         case IPT_TCP_RESET:
 412                 send_reset(*pskb, hooknum);
 413         case IPT_ICMP_ECHOREPLY:
 414                 /* Doesn't happen. */
 415                 break;
 416         }
 417
 418         return NF_DROP;
 419 }
 420
 421 static int check(const char *tablename,
 422                  const struct ipt_entry *e,
 423                  void *targinfo,
 424                  unsigned int targinfosize,
 425                  unsigned int hook_mask)
 426 {
 427         const struct ipt_reject_info *rejinfo = targinfo;
 428
 429         if (targinfosize != IPT_ALIGN(sizeof(struct ipt_reject_info))) {
 430                 DEBUGP("REJECT: targinfosize %u != 0\n", targinfosize);
 431                 return 0;
 432         }
 433
 434         /* Only allow these for packet filtering. */
 435         if (strcmp(tablename, "filter") != 0) {
 436                 DEBUGP("REJECT: bad table `%s'.\n", tablename);
 437                 return 0;
 438         }
 439         if ((hook_mask & ~((1 << NF_IP_LOCAL_IN)
 440                            | (1 << NF_IP_FORWARD)
 441                            | (1 << NF_IP_LOCAL_OUT))) != 0) {
 442                 DEBUGP("REJECT: bad hook mask %X\n", hook_mask);
 443                 return 0;
 444         }
 445
 446         if (rejinfo->with == IPT_ICMP_ECHOREPLY) {
 447                 printk("REJECT: ECHOREPLY no longer supported.\n");
 448                 return 0;
 449         } else if (rejinfo->with == IPT_TCP_RESET) {
 450                 /* Must specify that it's a TCP packet */
 451                 if (e->ip.proto != IPPROTO_TCP
 452                     || (e->ip.invflags & IPT_INV_PROTO)) {
 453                         DEBUGP("REJECT: TCP_RESET invalid for non-tcp\n");
 454                         return 0;
 455                 }
 456         }
 457
 458         return 1;
 459 }
 460
 461 static struct ipt_target ipt_reject_reg = {
 462         .name           = "REJECT",
 463         .target         = reject,
 464         .checkentry     = check,
 465         .me             = THIS_MODULE,
 466 };
 467
 468 static int __init init(void)
 469 {
 470         return ipt_register_target(&ipt_reject_reg);
 471 }
 472
 473 static void __exit fini(void)
 474 {
 475         ipt_unregister_target(&ipt_reject_reg);
 476 }
 477
 478 module_init(init);
 479 module_exit(fini);