net/core/netfilter.c

   1 /* netfilter.c: look after the filters for various protocols.
   2  * Heavily influenced by the old firewall.c by David Bonn and Alan Cox.
   3  *
   4  * Thanks to Rob `CmdrTaco' Malda for not influencing this code in any
   5  * way.
   6  *
   7  * Rusty Russell (C)2000 -- This code is GPL.
   8  *
   9  * February 2000: Modified by James Morris to have 1 queue per protocol.
  10  * 15-Mar-2000:   Added NF_REPEAT --RR.
  11  * 08-May-2003:   Internal logging interface added by Jozsef Kadlecsik.
  12  */
  13 #include <linux/config.h>
  14 #include <linux/kernel.h>
  15 #include <linux/netfilter.h>
  16 #include <net/protocol.h>
  17 #include <linux/init.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/wait.h>
  20 #include <linux/module.h>
  21 #include <linux/interrupt.h>
  22 #include <linux/if.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/inetdevice.h>
  25 #include <linux/tcp.h>
  26 #include <linux/udp.h>
  27 #include <linux/icmp.h>
  28 #include <net/sock.h>
  29 #include <net/route.h>
  30 #include <linux/ip.h>
  31
  32 /* In this code, we can be waiting indefinitely for userspace to
  33  * service a packet if a hook returns NF_QUEUE.  We could keep a count
  34  * of skbuffs queued for userspace, and not deregister a hook unless
  35  * this is zero, but that sucks.  Now, we simply check when the
  36  * packets come back: if the hook is gone, the packet is discarded. */
  37 #ifdef CONFIG_NETFILTER_DEBUG
  38 #define NFDEBUG(format, args...)  printk(format , ## args)
  39 #else
  40 #define NFDEBUG(format, args...)
  41 #endif
  42
  43 /* Sockopts only registered and called from user context, so
  44    net locking would be overkill.  Also, [gs]etsockopt calls may
  45    sleep. */
  46 static DECLARE_MUTEX(nf_sockopt_mutex);
  47
  48 struct list_head nf_hooks[NPROTO][NF_MAX_HOOKS];
  49 static LIST_HEAD(nf_sockopts);
  50 static DEFINE_SPINLOCK(nf_hook_lock);
  51
  52 /*
  53  * A queue handler may be registered for each protocol.  Each is protected by
  54  * long term mutex.  The handler must provide an an outfn() to accept packets
  55  * for queueing and must reinject all packets it receives, no matter what.
  56  */
  57 static struct nf_queue_handler_t {
  58         nf_queue_outfn_t outfn;
  59         void *data;
  60 } queue_handler[NPROTO];
  61 static DEFINE_RWLOCK(queue_handler_lock);
  62
  63 int nf_register_hook(struct nf_hook_ops *reg)
  64 {
  65         struct list_head *i;
  66
  67         spin_lock_bh(&nf_hook_lock);
  68         list_for_each(i, &nf_hooks[reg->pf][reg->hooknum]) {
  69                 if (reg->priority < ((struct nf_hook_ops *)i)->priority)
  70                         break;
  71         }
  72         list_add_rcu(&reg->list, i->prev);
  73         spin_unlock_bh(&nf_hook_lock);
  74
  75         synchronize_net();
  76         return 0;
  77 }
  78
  79 void nf_unregister_hook(struct nf_hook_ops *reg)
  80 {
  81         spin_lock_bh(&nf_hook_lock);
  82         list_del_rcu(&reg->list);
  83         spin_unlock_bh(&nf_hook_lock);
  84
  85         synchronize_net();
  86 }
  87
  88 /* Do exclusive ranges overlap? */
  89 static inline int overlap(int min1, int max1, int min2, int max2)
  90 {
  91         return max1 > min2 && min1 < max2;
  92 }
  93
  94 /* Functions to register sockopt ranges (exclusive). */
  95 int nf_register_sockopt(struct nf_sockopt_ops *reg)
  96 {
  97         struct list_head *i;
  98         int ret = 0;
  99
 100         if (down_interruptible(&nf_sockopt_mutex) != 0)
 101                 return -EINTR;
 102
 103         list_for_each(i, &nf_sockopts) {
 104                 struct nf_sockopt_ops *ops = (struct nf_sockopt_ops *)i;
 105                 if (ops->pf == reg->pf
 106                     && (overlap(ops->set_optmin, ops->set_optmax,
 107                                 reg->set_optmin, reg->set_optmax)
 108                         || overlap(ops->get_optmin, ops->get_optmax,
 109                                    reg->get_optmin, reg->get_optmax))) {
 110                         NFDEBUG("nf_sock overlap: %u-%u/%u-%u v %u-%u/%u-%u\n",
 111                                 ops->set_optmin, ops->set_optmax,
 112                                 ops->get_optmin, ops->get_optmax,
 113                                 reg->set_optmin, reg->set_optmax,
 114                                 reg->get_optmin, reg->get_optmax);
 115                         ret = -EBUSY;
 116                         goto out;
 117                 }
 118         }
 119
 120         list_add(&reg->list, &nf_sockopts);
 121 out:
 122         up(&nf_sockopt_mutex);
 123         return ret;
 124 }
 125
 126 void nf_unregister_sockopt(struct nf_sockopt_ops *reg)
 127 {
 128         /* No point being interruptible: we're probably in cleanup_module() */
 129  restart:
 130         down(&nf_sockopt_mutex);
 131         if (reg->use != 0) {
 132                 /* To be woken by nf_sockopt call... */
 133                 /* FIXME: Stuart Young's name appears gratuitously. */
 134                 set_current_state(TASK_UNINTERRUPTIBLE);
 135                 reg->cleanup_task = current;
 136                 up(&nf_sockopt_mutex);
 137                 schedule();
 138                 goto restart;
 139         }
 140         list_del(&reg->list);
 141         up(&nf_sockopt_mutex);
 142 }
 143
 144 #ifdef CONFIG_NETFILTER_DEBUG
 145 #include <net/ip.h>
 146 #include <net/tcp.h>
 147 #include <linux/netfilter_ipv4.h>
 148
 149 static void debug_print_hooks_ip(unsigned int nf_debug)
 150 {
 151         if (nf_debug & (1 << NF_IP_PRE_ROUTING)) {
 152                 printk("PRE_ROUTING ");
 153                 nf_debug ^= (1 << NF_IP_PRE_ROUTING);
 154         }
 155         if (nf_debug & (1 << NF_IP_LOCAL_IN)) {
 156                 printk("LOCAL_IN ");
 157                 nf_debug ^= (1 << NF_IP_LOCAL_IN);
 158         }
 159         if (nf_debug & (1 << NF_IP_FORWARD)) {
 160                 printk("FORWARD ");
 161                 nf_debug ^= (1 << NF_IP_FORWARD);
 162         }
 163         if (nf_debug & (1 << NF_IP_LOCAL_OUT)) {
 164                 printk("LOCAL_OUT ");
 165                 nf_debug ^= (1 << NF_IP_LOCAL_OUT);
 166         }
 167         if (nf_debug & (1 << NF_IP_POST_ROUTING)) {
 168                 printk("POST_ROUTING ");
 169                 nf_debug ^= (1 << NF_IP_POST_ROUTING);
 170         }
 171         if (nf_debug)
 172                 printk("Crap bits: 0x%04X", nf_debug);
 173         printk("\n");
 174 }
 175
 176 static void nf_dump_skb(int pf, struct sk_buff *skb)
 177 {
 178         printk("skb: pf=%i %s dev=%s len=%u\n",
 179                pf,
 180                skb->sk ? "(owned)" : "(unowned)",
 181                skb->dev ? skb->dev->name : "(no dev)",
 182                skb->len);
 183         switch (pf) {
 184         case PF_INET: {
 185                 const struct iphdr *ip = skb->nh.iph;
 186                 __u32 *opt = (__u32 *) (ip + 1);
 187                 int opti;
 188                 __u16 src_port = 0, dst_port = 0;
 189
 190                 if (ip->protocol == IPPROTO_TCP
 191                     || ip->protocol == IPPROTO_UDP) {
 192                         struct tcphdr *tcp=(struct tcphdr *)((__u32 *)ip+ip->ihl);
 193                         src_port = ntohs(tcp->source);
 194                         dst_port = ntohs(tcp->dest);
 195                 }
 196
 197                 printk("PROTO=%d %u.%u.%u.%u:%hu %u.%u.%u.%u:%hu"
 198                        " L=%hu S=0x%2.2hX I=%hu F=0x%4.4hX T=%hu",
 199                        ip->protocol, NIPQUAD(ip->saddr),
 200                        src_port, NIPQUAD(ip->daddr),
 201                        dst_port,
 202                        ntohs(ip->tot_len), ip->tos, ntohs(ip->id),
 203                        ntohs(ip->frag_off), ip->ttl);
 204
 205                 for (opti = 0; opti < (ip->ihl - sizeof(struct iphdr) / 4); opti++)
 206                         printk(" O=0x%8.8X", *opt++);
 207                 printk("\n");
 208         }
 209         }
 210 }
 211
 212 void nf_debug_ip_local_deliver(struct sk_buff *skb)
 213 {
 214         /* If it's a loopback packet, it must have come through
 215          * NF_IP_LOCAL_OUT, NF_IP_RAW_INPUT, NF_IP_PRE_ROUTING and
 216          * NF_IP_LOCAL_IN.  Otherwise, must have gone through
 217          * NF_IP_RAW_INPUT and NF_IP_PRE_ROUTING.  */
 218         if (!skb->dev) {
 219                 printk("ip_local_deliver: skb->dev is NULL.\n");
 220         } else {
 221                 if (skb->nf_debug != ((1<<NF_IP_PRE_ROUTING)
 222                                       | (1<<NF_IP_LOCAL_IN))) {
 223                         printk("ip_local_deliver: bad skb: ");
 224                         debug_print_hooks_ip(skb->nf_debug);
 225                         nf_dump_skb(PF_INET, skb);
 226                 }
 227         }
 228 }
 229
 230 void nf_debug_ip_loopback_xmit(struct sk_buff *newskb)
 231 {
 232         if (newskb->nf_debug != ((1 << NF_IP_LOCAL_OUT)
 233                                  | (1 << NF_IP_POST_ROUTING))) {
 234                 printk("ip_dev_loopback_xmit: bad owned skb = %p: ",
 235                        newskb);
 236                 debug_print_hooks_ip(newskb->nf_debug);
 237                 nf_dump_skb(PF_INET, newskb);
 238         }
 239 }
 240
 241 void nf_debug_ip_finish_output2(struct sk_buff *skb)
 242 {
 243         /* If it's owned, it must have gone through the
 244          * NF_IP_LOCAL_OUT and NF_IP_POST_ROUTING.
 245          * Otherwise, must have gone through
 246          * NF_IP_PRE_ROUTING, NF_IP_FORWARD and NF_IP_POST_ROUTING.
 247          */
 248         if (skb->sk) {
 249                 if (skb->nf_debug != ((1 << NF_IP_LOCAL_OUT)
 250                                       | (1 << NF_IP_POST_ROUTING))) {
 251                         printk("ip_finish_output: bad owned skb = %p: ", skb);
 252                         debug_print_hooks_ip(skb->nf_debug);
 253                         nf_dump_skb(PF_INET, skb);
 254                 }
 255         } else {
 256                 if (skb->nf_debug != ((1 << NF_IP_PRE_ROUTING)
 257                                       | (1 << NF_IP_FORWARD)
 258                                       | (1 << NF_IP_POST_ROUTING))) {
 259                         /* Fragments, entunnelled packets, TCP RSTs
 260                            generated by ipt_REJECT will have no
 261                            owners, but still may be local */
 262                         if (skb->nf_debug != ((1 << NF_IP_LOCAL_OUT)
 263                                               | (1 << NF_IP_POST_ROUTING))){
 264                                 printk("ip_finish_output:"
 265                                        " bad unowned skb = %p: ",skb);
 266                                 debug_print_hooks_ip(skb->nf_debug);
 267                                 nf_dump_skb(PF_INET, skb);
 268                         }
 269                 }
 270         }
 271 }
 272 #endif /*CONFIG_NETFILTER_DEBUG*/
 273
 274 /* Call get/setsockopt() */
 275 static int nf_sockopt(struct sock *sk, int pf, int val,
 276                       char __user *opt, int *len, int get)
 277 {
 278         struct list_head *i;
 279         struct nf_sockopt_ops *ops;
 280         int ret;
 281
 282         if (down_interruptible(&nf_sockopt_mutex) != 0)
 283                 return -EINTR;
 284
 285         list_for_each(i, &nf_sockopts) {
 286                 ops = (struct nf_sockopt_ops *)i;
 287                 if (ops->pf == pf) {
 288                         if (get) {
 289                                 if (val >= ops->get_optmin
 290                                     && val < ops->get_optmax) {
 291                                         ops->use++;
 292                                         up(&nf_sockopt_mutex);
 293                                         ret = ops->get(sk, val, opt, len);
 294                                         goto out;
 295                                 }
 296                         } else {
 297                                 if (val >= ops->set_optmin
 298                                     && val < ops->set_optmax) {
 299                                         ops->use++;
 300                                         up(&nf_sockopt_mutex);
 301                                         ret = ops->set(sk, val, opt, *len);
 302                                         goto out;
 303                                 }
 304                         }
 305                 }
 306         }
 307         up(&nf_sockopt_mutex);
 308         return -ENOPROTOOPT;
 309
 310  out:
 311         down(&nf_sockopt_mutex);
 312         ops->use--;
 313         if (ops->cleanup_task)
 314                 wake_up_process(ops->cleanup_task);
 315         up(&nf_sockopt_mutex);
 316         return ret;
 317 }
 318
 319 int nf_setsockopt(struct sock *sk, int pf, int val, char __user *opt,
 320                   int len)
 321 {
 322         return nf_sockopt(sk, pf, val, opt, &len, 0);
 323 }
 324
 325 int nf_getsockopt(struct sock *sk, int pf, int val, char __user *opt, int *len)
 326 {
 327         return nf_sockopt(sk, pf, val, opt, len, 1);
 328 }
 329
 330 static unsigned int nf_iterate(struct list_head *head,
 331                                struct sk_buff **skb,
 332                                int hook,
 333                                const struct net_device *indev,
 334                                const struct net_device *outdev,
 335                                struct list_head **i,
 336                                int (*okfn)(struct sk_buff *),
 337                                int hook_thresh)
 338 {
 339         unsigned int verdict;
 340
 341         /*
 342          * The caller must not block between calls to this
 343          * function because of risk of continuing from deleted element.
 344          */
 345         list_for_each_continue_rcu(*i, head) {
 346                 struct nf_hook_ops *elem = (struct nf_hook_ops *)*i;
 347
 348                 if (hook_thresh > elem->priority)
 349                         continue;
 350
 351                 /* Optimization: we don't need to hold module
 352                    reference here, since function can't sleep. --RR */
 353                 verdict = elem->hook(hook, skb, indev, outdev, okfn);
 354                 if (verdict != NF_ACCEPT) {
 355 #ifdef CONFIG_NETFILTER_DEBUG
 356                         if (unlikely(verdict > NF_MAX_VERDICT)) {
 357                                 NFDEBUG("Evil return from %p(%u).\n",
 358                                         elem->hook, hook);
 359                                 continue;
 360                         }
 361 #endif
 362                         if (verdict != NF_REPEAT)
 363                                 return verdict;
 364                         *i = (*i)->prev;
 365                 }
 366         }
 367         return NF_ACCEPT;
 368 }
 369
 370 int nf_register_queue_handler(int pf, nf_queue_outfn_t outfn, void *data)
 371 {
 372         int ret;
 373
 374         write_lock_bh(&queue_handler_lock);
 375         if (queue_handler[pf].outfn)
 376                 ret = -EBUSY;
 377         else {
 378                 queue_handler[pf].outfn = outfn;
 379                 queue_handler[pf].data = data;
 380                 ret = 0;
 381         }
 382         write_unlock_bh(&queue_handler_lock);
 383
 384         return ret;
 385 }
 386
 387 /* The caller must flush their queue before this */
 388 int nf_unregister_queue_handler(int pf)
 389 {
 390         write_lock_bh(&queue_handler_lock);
 391         queue_handler[pf].outfn = NULL;
 392         queue_handler[pf].data = NULL;
 393         write_unlock_bh(&queue_handler_lock);
 394
 395         return 0;
 396 }
 397
 398 /*
 399  * Any packet that leaves via this function must come back
 400  * through nf_reinject().
 401  */
 402 static int nf_queue(struct sk_buff *skb,
 403                     struct list_head *elem,
 404                     int pf, unsigned int hook,
 405                     struct net_device *indev,
 406                     struct net_device *outdev,
 407                     int (*okfn)(struct sk_buff *))
 408 {
 409         int status;
 410         struct nf_info *info;
 411 #ifdef CONFIG_BRIDGE_NETFILTER
 412         struct net_device *physindev = NULL;
 413         struct net_device *physoutdev = NULL;
 414 #endif
 415
 416         /* QUEUE == DROP if noone is waiting, to be safe. */
 417         read_lock(&queue_handler_lock);
 418         if (!queue_handler[pf].outfn) {
 419                 read_unlock(&queue_handler_lock);
 420                 kfree_skb(skb);
 421                 return 1;
 422         }
 423
 424         info = kmalloc(sizeof(*info), GFP_ATOMIC);
 425         if (!info) {
 426                 if (net_ratelimit())
 427                         printk(KERN_ERR "OOM queueing packet %p\n",
 428                                skb);
 429                 read_unlock(&queue_handler_lock);
 430                 kfree_skb(skb);
 431                 return 1;
 432         }
 433
 434         *info = (struct nf_info) {
 435                 (struct nf_hook_ops *)elem, pf, hook, indev, outdev, okfn };
 436
 437         /* If it's going away, ignore hook. */
 438         if (!try_module_get(info->elem->owner)) {
 439                 read_unlock(&queue_handler_lock);
 440                 kfree(info);
 441                 return 0;
 442         }
 443
 444         /* Bump dev refs so they don't vanish while packet is out */
 445         if (indev) dev_hold(indev);
 446         if (outdev) dev_hold(outdev);
 447
 448 #ifdef CONFIG_BRIDGE_NETFILTER
 449         if (skb->nf_bridge) {
 450                 physindev = skb->nf_bridge->physindev;
 451                 if (physindev) dev_hold(physindev);
 452                 physoutdev = skb->nf_bridge->physoutdev;
 453                 if (physoutdev) dev_hold(physoutdev);
 454         }
 455 #endif
 456
 457         status = queue_handler[pf].outfn(skb, info, queue_handler[pf].data);
 458         read_unlock(&queue_handler_lock);
 459
 460         if (status < 0) {
 461                 /* James M doesn't say fuck enough. */
 462                 if (indev) dev_put(indev);
 463                 if (outdev) dev_put(outdev);
 464 #ifdef CONFIG_BRIDGE_NETFILTER
 465                 if (physindev) dev_put(physindev);
 466                 if (physoutdev) dev_put(physoutdev);
 467 #endif
 468                 module_put(info->elem->owner);
 469                 kfree(info);
 470                 kfree_skb(skb);
 471                 return 1;
 472         }
 473         return 1;
 474 }
 475
 476 /* Returns 1 if okfn() needs to be executed by the caller,
 477  * -EPERM for NF_DROP, 0 otherwise. */
 478 int nf_hook_slow(int pf, unsigned int hook, struct sk_buff **pskb,
 479                  struct net_device *indev,
 480                  struct net_device *outdev,
 481                  int (*okfn)(struct sk_buff *),
 482                  int hook_thresh)
 483 {
 484         struct list_head *elem;
 485         unsigned int verdict;
 486         int ret = 0;
 487
 488         /* We may already have this, but read-locks nest anyway */
 489         rcu_read_lock();
 490
 491 #ifdef CONFIG_NETFILTER_DEBUG
 492         if (unlikely((*pskb)->nf_debug & (1 << hook))) {
 493                 printk("nf_hook: hook %i already set.\n", hook);
 494                 nf_dump_skb(pf, *pskb);
 495         }
 496         (*pskb)->nf_debug |= (1 << hook);
 497 #endif
 498
 499         elem = &nf_hooks[pf][hook];
 500 next_hook:
 501         verdict = nf_iterate(&nf_hooks[pf][hook], pskb, hook, indev,
 502                              outdev, &elem, okfn, hook_thresh);
 503         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 504                 ret = 1;
 505                 goto unlock;
 506         } else if (verdict == NF_DROP) {
 507                 kfree_skb(*pskb);
 508                 ret = -EPERM;
 509         } else if (verdict == NF_QUEUE) {
 510                 NFDEBUG("nf_hook: Verdict = QUEUE.\n");
 511                 if (!nf_queue(*pskb, elem, pf, hook, indev, outdev, okfn))
 512                         goto next_hook;
 513         }
 514 unlock:
 515         rcu_read_unlock();
 516         return ret;
 517 }
 518
 519 void nf_reinject(struct sk_buff *skb, struct nf_info *info,
 520                  unsigned int verdict)
 521 {
 522         struct list_head *elem = &info->elem->list;
 523         struct list_head *i;
 524
 525         rcu_read_lock();
 526
 527         /* Release those devices we held, or Alexey will kill me. */
 528         if (info->indev) dev_put(info->indev);
 529         if (info->outdev) dev_put(info->outdev);
 530 #ifdef CONFIG_BRIDGE_NETFILTER
 531         if (skb->nf_bridge) {
 532                 if (skb->nf_bridge->physindev)
 533                         dev_put(skb->nf_bridge->physindev);
 534                 if (skb->nf_bridge->physoutdev)
 535                         dev_put(skb->nf_bridge->physoutdev);
 536         }
 537 #endif
 538
 539         /* Drop reference to owner of hook which queued us. */
 540         module_put(info->elem->owner);
 541
 542         list_for_each_rcu(i, &nf_hooks[info->pf][info->hook]) {
 543                 if (i == elem)
 544                         break;
 545         }
 546
 547         if (elem == &nf_hooks[info->pf][info->hook]) {
 548                 /* The module which sent it to userspace is gone. */
 549                 NFDEBUG("%s: module disappeared, dropping packet.\n",
 550                         __FUNCTION__);
 551                 verdict = NF_DROP;
 552         }
 553
 554         /* Continue traversal iff userspace said ok... */
 555         if (verdict == NF_REPEAT) {
 556                 elem = elem->prev;
 557                 verdict = NF_ACCEPT;
 558         }
 559
 560         if (verdict == NF_ACCEPT) {
 561         next_hook:
 562                 verdict = nf_iterate(&nf_hooks[info->pf][info->hook],
 563                                      &skb, info->hook,
 564                                      info->indev, info->outdev, &elem,
 565                                      info->okfn, INT_MIN);
 566         }
 567
 568         switch (verdict) {
 569         case NF_ACCEPT:
 570                 info->okfn(skb);
 571                 break;
 572
 573         case NF_QUEUE:
 574                 if (!nf_queue(skb, elem, info->pf, info->hook,
 575                               info->indev, info->outdev, info->okfn))
 576                         goto next_hook;
 577                 break;
 578         }
 579         rcu_read_unlock();
 580
 581         if (verdict == NF_DROP)
 582                 kfree_skb(skb);
 583
 584         kfree(info);
 585         return;
 586 }
 587
 588 #ifdef CONFIG_INET
 589 /* route_me_harder function, used by iptable_nat, iptable_mangle + ip_queue */
 590 int ip_route_me_harder(struct sk_buff **pskb)
 591 {
 592         struct iphdr *iph = (*pskb)->nh.iph;
 593         struct rtable *rt;
 594         struct flowi fl = {};
 595         struct dst_entry *odst;
 596         unsigned int hh_len;
 597
 598         /* some non-standard hacks like ipt_REJECT.c:send_reset() can cause
 599          * packets with foreign saddr to appear on the NF_IP_LOCAL_OUT hook.
 600          */
 601         if (inet_addr_type(iph->saddr) == RTN_LOCAL) {
 602                 fl.nl_u.ip4_u.daddr = iph->daddr;
 603                 fl.nl_u.ip4_u.saddr = iph->saddr;
 604                 fl.nl_u.ip4_u.tos = RT_TOS(iph->tos);
 605                 fl.oif = (*pskb)->sk ? (*pskb)->sk->sk_bound_dev_if : 0;
 606 #ifdef CONFIG_IP_ROUTE_FWMARK
 607                 fl.nl_u.ip4_u.fwmark = (*pskb)->nfmark;
 608 #endif
 609                 fl.proto = iph->protocol;
 610                 if (ip_route_output_key(&rt, &fl) != 0)
 611                         return -1;
 612
 613                 /* Drop old route. */
 614                 dst_release((*pskb)->dst);
 615                 (*pskb)->dst = &rt->u.dst;
 616         } else {
 617                 /* non-local src, find valid iif to satisfy
 618                  * rp-filter when calling ip_route_input. */
 619                 fl.nl_u.ip4_u.daddr = iph->saddr;
 620                 if (ip_route_output_key(&rt, &fl) != 0)
 621                         return -1;
 622
 623                 odst = (*pskb)->dst;
 624                 if (ip_route_input(*pskb, iph->daddr, iph->saddr,
 625                                    RT_TOS(iph->tos), rt->u.dst.dev) != 0) {
 626                         dst_release(&rt->u.dst);
 627                         return -1;
 628                 }
 629                 dst_release(&rt->u.dst);
 630                 dst_release(odst);
 631         }
 632
 633         if ((*pskb)->dst->error)
 634                 return -1;
 635
 636         /* Change in oif may mean change in hh_len. */
 637         hh_len = (*pskb)->dst->dev->hard_header_len;
 638         if (skb_headroom(*pskb) < hh_len) {
 639                 struct sk_buff *nskb;
 640
 641                 nskb = skb_realloc_headroom(*pskb, hh_len);
 642                 if (!nskb)
 643                         return -1;
 644                 if ((*pskb)->sk)
 645                         skb_set_owner_w(nskb, (*pskb)->sk);
 646                 kfree_skb(*pskb);
 647                 *pskb = nskb;
 648         }
 649
 650         return 0;
 651 }
 652 EXPORT_SYMBOL(ip_route_me_harder);
 653
 654 int skb_ip_make_writable(struct sk_buff **pskb, unsigned int writable_len)
 655 {
 656         struct sk_buff *nskb;
 657
 658         if (writable_len > (*pskb)->len)
 659                 return 0;
 660
 661         /* Not exclusive use of packet?  Must copy. */
 662         if (skb_shared(*pskb) || skb_cloned(*pskb))
 663                 goto copy_skb;
 664
 665         return pskb_may_pull(*pskb, writable_len);
 666
 667 copy_skb:
 668         nskb = skb_copy(*pskb, GFP_ATOMIC);
 669         if (!nskb)
 670                 return 0;
 671         BUG_ON(skb_is_nonlinear(nskb));
 672
 673         /* Rest of kernel will get very unhappy if we pass it a
 674            suddenly-orphaned skbuff */
 675         if ((*pskb)->sk)
 676                 skb_set_owner_w(nskb, (*pskb)->sk);
 677         kfree_skb(*pskb);
 678         *pskb = nskb;
 679         return 1;
 680 }
 681 EXPORT_SYMBOL(skb_ip_make_writable);
 682 #endif /*CONFIG_INET*/
 683
 684 /* Internal logging interface, which relies on the real
 685    LOG target modules */
 686
 687 #define NF_LOG_PREFIXLEN                128
 688
 689 static nf_logfn *nf_logging[NPROTO]; /* = NULL */
 690 static int reported = 0;
 691 static DEFINE_SPINLOCK(nf_log_lock);
 692
 693 int nf_log_register(int pf, nf_logfn *logfn)
 694 {
 695         int ret = -EBUSY;
 696
 697         /* Any setup of logging members must be done before
 698          * substituting pointer. */
 699         spin_lock(&nf_log_lock);
 700         if (!nf_logging[pf]) {
 701                 rcu_assign_pointer(nf_logging[pf], logfn);
 702                 ret = 0;
 703         }
 704         spin_unlock(&nf_log_lock);
 705         return ret;
 706 }
 707
 708 void nf_log_unregister(int pf, nf_logfn *logfn)
 709 {
 710         spin_lock(&nf_log_lock);
 711         if (nf_logging[pf] == logfn)
 712                 nf_logging[pf] = NULL;
 713         spin_unlock(&nf_log_lock);
 714
 715         /* Give time to concurrent readers. */
 716         synchronize_net();
 717 }
 718
 719 void nf_log_packet(int pf,
 720                    unsigned int hooknum,
 721                    const struct sk_buff *skb,
 722                    const struct net_device *in,
 723                    const struct net_device *out,
 724                    const char *fmt, ...)
 725 {
 726         va_list args;
 727         char prefix[NF_LOG_PREFIXLEN];
 728         nf_logfn *logfn;
 729
 730         rcu_read_lock();
 731         logfn = rcu_dereference(nf_logging[pf]);
 732         if (logfn) {
 733                 va_start(args, fmt);
 734                 vsnprintf(prefix, sizeof(prefix), fmt, args);
 735                 va_end(args);
 736                 /* We must read logging before nf_logfn[pf] */
 737                 logfn(hooknum, skb, in, out, prefix);
 738         } else if (!reported) {
 739                 printk(KERN_WARNING "nf_log_packet: can\'t log yet, "
 740                        "no backend logging module loaded in!\n");
 741                 reported++;
 742         }
 743         rcu_read_unlock();
 744 }
 745 EXPORT_SYMBOL(nf_log_register);
 746 EXPORT_SYMBOL(nf_log_unregister);
 747 EXPORT_SYMBOL(nf_log_packet);
 748
 749 /* This does not belong here, but locally generated errors need it if connection
 750    tracking in use: without this, connection may not be in hash table, and hence
 751    manufactured ICMP or RST packets will not be associated with it. */
 752 void (*ip_ct_attach)(struct sk_buff *, struct sk_buff *);
 753
 754 void nf_ct_attach(struct sk_buff *new, struct sk_buff *skb)
 755 {
 756         void (*attach)(struct sk_buff *, struct sk_buff *);
 757
 758         if (skb->nfct && (attach = ip_ct_attach) != NULL) {
 759                 mb(); /* Just to be sure: must be read before executing this */
 760                 attach(new, skb);
 761         }
 762 }
 763
 764 void __init netfilter_init(void)
 765 {
 766         int i, h;
 767
 768         for (i = 0; i < NPROTO; i++) {
 769                 for (h = 0; h < NF_MAX_HOOKS; h++)
 770                         INIT_LIST_HEAD(&nf_hooks[i][h]);
 771         }
 772 }
 773
 774 EXPORT_SYMBOL(ip_ct_attach);
 775 EXPORT_SYMBOL(nf_ct_attach);
 776 EXPORT_SYMBOL(nf_getsockopt);
 777 EXPORT_SYMBOL(nf_hook_slow);
 778 EXPORT_SYMBOL(nf_hooks);
 779 EXPORT_SYMBOL(nf_register_hook);
 780 EXPORT_SYMBOL(nf_register_queue_handler);
 781 EXPORT_SYMBOL(nf_register_sockopt);
 782 EXPORT_SYMBOL(nf_reinject);
 783 EXPORT_SYMBOL(nf_setsockopt);
 784 EXPORT_SYMBOL(nf_unregister_hook);
 785 EXPORT_SYMBOL(nf_unregister_queue_handler);
 786 EXPORT_SYMBOL(nf_unregister_sockopt);