net/ipv4/netfilter/ip_fw_compat_masq.c

   1 /* Masquerading compatibility layer.
   2
   3    Note that there are no restrictions on other programs binding to
   4    ports 61000:65095 (in 2.0 and 2.2 they get EADDRINUSE).  Just DON'T
   5    DO IT.
   6  */
   7
   8 /* (C) 1999-2001 Paul `Rusty' Russell
   9  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
  10  *
  11  * This program is free software; you can redistribute it and/or modify
  12  * it under the terms of the GNU General Public License version 2 as
  13  * published by the Free Software Foundation.
  14  */
  15
  16 #include <linux/skbuff.h>
  17 #include <linux/in.h>
  18 #include <linux/ip.h>
  19 #include <linux/icmp.h>
  20 #include <linux/udp.h>
  21 #include <linux/netfilter_ipv4.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/inetdevice.h>
  24 #include <linux/proc_fs.h>
  25 #include <linux/module.h>
  26 #include <net/route.h>
  27 #include <net/ip.h>
  28
  29 #define ASSERT_READ_LOCK(x) MUST_BE_READ_LOCKED(&ip_conntrack_lock)
  30 #define ASSERT_WRITE_LOCK(x) MUST_BE_WRITE_LOCKED(&ip_conntrack_lock)
  31
  32 #include <linux/netfilter_ipv4/ip_conntrack.h>
  33 #include <linux/netfilter_ipv4/ip_conntrack_core.h>
  34 #include <linux/netfilter_ipv4/ip_nat.h>
  35 #include <linux/netfilter_ipv4/ip_nat_core.h>
  36 #include <linux/netfilter_ipv4/listhelp.h>
  37 #include "ip_fw_compat.h"
  38
  39 #if 0
  40 #define DEBUGP printk
  41 #else
  42 #define DEBUGP(format, args...)
  43 #endif
  44
  45 unsigned int
  46 do_masquerade(struct sk_buff **pskb, const struct net_device *dev)
  47 {
  48         struct ip_nat_info *info;
  49         enum ip_conntrack_info ctinfo;
  50         struct ip_conntrack *ct;
  51         unsigned int ret;
  52
  53         /* Sorry, only ICMP, TCP and UDP. */
  54         if ((*pskb)->nh.iph->protocol != IPPROTO_ICMP
  55             && (*pskb)->nh.iph->protocol != IPPROTO_TCP
  56             && (*pskb)->nh.iph->protocol != IPPROTO_UDP)
  57                 return NF_DROP;
  58
  59         /* Feed it to connection tracking; in fact we're in NF_IP_FORWARD,
  60            but connection tracking doesn't expect that */
  61         ret = ip_conntrack_in(NF_IP_POST_ROUTING, pskb, dev, NULL, NULL);
  62         if (ret != NF_ACCEPT) {
  63                 DEBUGP("ip_conntrack_in returned %u.\n", ret);
  64                 return ret;
  65         }
  66
  67         ct = ip_conntrack_get(*pskb, &ctinfo);
  68
  69         if (!ct) {
  70                 DEBUGP("ip_conntrack_in set to invalid conntrack.\n");
  71                 return NF_DROP;
  72         }
  73
  74         info = &ct->nat.info;
  75
  76         WRITE_LOCK(&ip_nat_lock);
  77         /* Setup the masquerade, if not already */
  78         if (!info->initialized) {
  79                 u_int32_t newsrc;
  80                 struct flowi fl = { .nl_u = { .ip4_u = { .daddr = (*pskb)->nh.iph->daddr } } };
  81                 struct rtable *rt;
  82                 struct ip_nat_multi_range range;
  83
  84                 /* Pass 0 instead of saddr, since it's going to be changed
  85                    anyway. */
  86                 if (ip_route_output_key(&rt, &fl) != 0) {
  87                         DEBUGP("ipnat_rule_masquerade: Can't reroute.\n");
  88                         return NF_DROP;
  89                 }
  90                 newsrc = inet_select_addr(rt->u.dst.dev, rt->rt_gateway,
  91                                           RT_SCOPE_UNIVERSE);
  92                 ip_rt_put(rt);
  93                 range = ((struct ip_nat_multi_range)
  94                          { 1,
  95                            {{IP_NAT_RANGE_MAP_IPS|IP_NAT_RANGE_PROTO_SPECIFIED,
  96                              newsrc, newsrc,
  97                              { htons(61000) }, { htons(65095) } } } });
  98
  99                 ret = ip_nat_setup_info(ct, &range, NF_IP_POST_ROUTING);
 100                 if (ret != NF_ACCEPT) {
 101                         WRITE_UNLOCK(&ip_nat_lock);
 102                         return ret;
 103                 }
 104         } else
 105                 DEBUGP("Masquerading already done on this conn.\n");
 106         WRITE_UNLOCK(&ip_nat_lock);
 107
 108         return do_bindings(ct, ctinfo, info, NF_IP_POST_ROUTING, pskb);
 109 }
 110
 111 void
 112 check_for_masq_error(struct sk_buff **pskb)
 113 {
 114         enum ip_conntrack_info ctinfo;
 115         struct ip_conntrack *ct;
 116
 117         ct = ip_conntrack_get(*pskb, &ctinfo);
 118         /* Wouldn't be here if not tracked already => masq'ed ICMP
 119            ping or error related to masq'd connection */
 120         IP_NF_ASSERT(ct);
 121         if (ctinfo == IP_CT_RELATED) {
 122                 icmp_reply_translation(pskb, ct, NF_IP_PRE_ROUTING,
 123                                        CTINFO2DIR(ctinfo));
 124                 icmp_reply_translation(pskb, ct, NF_IP_POST_ROUTING,
 125                                        CTINFO2DIR(ctinfo));
 126         }
 127 }
 128
 129 unsigned int
 130 check_for_demasq(struct sk_buff **pskb)
 131 {
 132         struct ip_conntrack_tuple tuple;
 133         struct ip_conntrack_protocol *protocol;
 134         struct ip_conntrack_tuple_hash *h;
 135         enum ip_conntrack_info ctinfo;
 136         struct ip_conntrack *ct;
 137         int ret;
 138
 139         protocol = ip_ct_find_proto((*pskb)->nh.iph->protocol);
 140
 141         /* We don't feed packets to conntrack system unless we know
 142            they're part of an connection already established by an
 143            explicit masq command. */
 144         switch ((*pskb)->nh.iph->protocol) {
 145         case IPPROTO_ICMP:
 146                 /* ICMP errors. */
 147                 ct = icmp_error_track(*pskb, &ctinfo, NF_IP_PRE_ROUTING);
 148                 if (ct) {
 149                         /* We only do SNAT in the compatibility layer.
 150                            So we can manipulate ICMP errors from
 151                            server here (== DNAT).  Do SNAT icmp manips
 152                            in POST_ROUTING handling. */
 153                         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY) {
 154                                 icmp_reply_translation(pskb, ct,
 155                                                        NF_IP_PRE_ROUTING,
 156                                                        CTINFO2DIR(ctinfo));
 157                                 icmp_reply_translation(pskb, ct,
 158                                                        NF_IP_POST_ROUTING,
 159                                                        CTINFO2DIR(ctinfo));
 160                         }
 161                         return NF_ACCEPT;
 162                 }
 163                 /* Fall thru... */
 164         case IPPROTO_TCP:
 165         case IPPROTO_UDP:
 166                 IP_NF_ASSERT(((*pskb)->nh.iph->frag_off & htons(IP_OFFSET)) == 0);
 167
 168                 if (!get_tuple((*pskb)->nh.iph, *pskb, (*pskb)->nh.iph->ihl*4, &tuple, protocol)) {
 169                         if (net_ratelimit())
 170                                 printk("ip_fw_compat_masq: Can't get tuple\n");
 171                         return NF_ACCEPT;
 172                 }
 173                 break;
 174
 175         default:
 176                 /* Not ours... */
 177                 return NF_ACCEPT;
 178         }
 179         h = ip_conntrack_find_get(&tuple, NULL);
 180
 181         /* MUST be found, and MUST be reply. */
 182         if (h && DIRECTION(h) == 1) {
 183                 ret = ip_conntrack_in(NF_IP_PRE_ROUTING, pskb,
 184                                       NULL, NULL, NULL);
 185
 186                 /* Put back the reference gained from find_get */
 187                 nf_conntrack_put(&h->ctrack->infos[0]);
 188                 if (ret == NF_ACCEPT) {
 189                         struct ip_conntrack *ct;
 190                         ct = ip_conntrack_get(*pskb, &ctinfo);
 191
 192                         if (ct) {
 193                                 struct ip_nat_info *info = &ct->nat.info;
 194
 195                                 do_bindings(ct, ctinfo, info,
 196                                             NF_IP_PRE_ROUTING,
 197                                             pskb);
 198                         } else
 199                                 if (net_ratelimit())
 200                                         printk("ip_fw_compat_masq: conntrack"
 201                                                " didn't like\n");
 202                 }
 203         } else {
 204                 if (h)
 205                         /* Put back the reference gained from find_get */
 206                         nf_conntrack_put(&h->ctrack->infos[0]);
 207                 ret = NF_ACCEPT;
 208         }
 209
 210         return ret;
 211 }
 212
 213 int ip_fw_masq_timeouts(void *user, int len)
 214 {
 215         printk("Sorry: masquerading timeouts set 5DAYS/2MINS/60SECS\n");
 216         return 0;
 217 }
 218
 219 static const char *masq_proto_name(u_int16_t protonum)
 220 {
 221         switch (protonum) {
 222         case IPPROTO_TCP: return "TCP";
 223         case IPPROTO_UDP: return "UDP";
 224         case IPPROTO_ICMP: return "ICMP";
 225         default: return "MORE-CAFFEINE-FOR-RUSTY";
 226         }
 227 }
 228
 229 static unsigned int
 230 print_masq(char *buffer, const struct ip_conntrack *conntrack)
 231 {
 232         char temp[129];
 233
 234         /* This is for backwards compatibility, but ick!.
 235            We should never export jiffies to userspace.
 236         */
 237         sprintf(temp,"%s %08X:%04X %08X:%04X %04X %08X %6d %6d %7lu",
 238                 masq_proto_name(conntrack->tuplehash[0].tuple.dst.protonum),
 239                 ntohl(conntrack->tuplehash[0].tuple.src.ip),
 240                 ntohs(conntrack->tuplehash[0].tuple.src.u.all),
 241                 ntohl(conntrack->tuplehash[0].tuple.dst.ip),
 242                 ntohs(conntrack->tuplehash[0].tuple.dst.u.all),
 243                 ntohs(conntrack->tuplehash[1].tuple.dst.u.all),
 244                 /* Sorry, no init_seq, delta or previous_delta (yet). */
 245                 0, 0, 0,
 246                 conntrack->timeout.expires - jiffies);
 247
 248         return sprintf(buffer, "%-127s\n", temp);
 249 }
 250
 251 /* Returns true when finished. */
 252 static int
 253 masq_iterate(const struct ip_conntrack_tuple_hash *hash,
 254              char *buffer, off_t offset, off_t *upto,
 255              unsigned int *len, unsigned int maxlen)
 256 {
 257         unsigned int newlen;
 258
 259         IP_NF_ASSERT(hash->ctrack);
 260
 261         /* Only count originals */
 262         if (DIRECTION(hash))
 263                 return 0;
 264
 265         if ((*upto)++ < offset)
 266                 return 0;
 267
 268         newlen = print_masq(buffer + *len, hash->ctrack);
 269         if (*len + newlen > maxlen)
 270                 return 1;
 271         else *len += newlen;
 272
 273         return 0;
 274 }
 275
 276 /* Everything in the hash is masqueraded. */
 277 static int
 278 masq_procinfo(char *buffer, char **start, off_t offset, int length)
 279 {
 280         unsigned int i;
 281         int len = 0;
 282         off_t upto = 1;
 283
 284         /* Header: first record */
 285         if (offset == 0) {
 286                 char temp[128];
 287
 288                 sprintf(temp,
 289                         "Prc FromIP   FPrt ToIP     TPrt Masq Init-seq  Delta PDelta Expires (free=0,0,0)");
 290                 len = sprintf(buffer, "%-127s\n", temp);
 291                 offset = 1;
 292         }
 293
 294         READ_LOCK(&ip_conntrack_lock);
 295         /* Traverse hash; print originals then reply. */
 296         for (i = 0; i < ip_conntrack_htable_size; i++) {
 297                 if (LIST_FIND(&ip_conntrack_hash[i], masq_iterate,
 298                               struct ip_conntrack_tuple_hash *,
 299                               buffer, offset, &upto, &len, length))
 300                         break;
 301         }
 302         READ_UNLOCK(&ip_conntrack_lock);
 303
 304         /* `start' hack - see fs/proc/generic.c line ~165 */
 305         *start = (char *)((unsigned int)upto - offset);
 306         return len;
 307 }
 308
 309 int __init masq_init(void)
 310 {
 311         int ret;
 312         struct proc_dir_entry *proc;
 313
 314         ret = ip_conntrack_init();
 315         if (ret == 0) {
 316                 ret = ip_nat_init();
 317                 if (ret == 0) {
 318                         proc = proc_net_create("ip_masquerade",
 319                                                0, masq_procinfo);
 320                         if (proc)
 321                                 proc->owner = THIS_MODULE;
 322                         else {
 323                                 ip_nat_cleanup();
 324                                 ip_conntrack_cleanup();
 325                                 ret = -ENOMEM;
 326                         }
 327                 } else
 328                         ip_conntrack_cleanup();
 329         }
 330
 331         return ret;
 332 }
 333
 334 void masq_cleanup(void)
 335 {
 336         ip_nat_cleanup();
 337         ip_conntrack_cleanup();
 338         proc_net_remove("ip_masquerade");
 339 }