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 spinlock_t nf_hook_lock = SPIN_LOCK_UNLOCKED;
  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 rwlock_t queue_handler_lock = RW_LOCK_UNLOCKED;
  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 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         }
 221         else if (strcmp(skb->dev->name, "lo") == 0) {
 222                 if (skb->nf_debug != ((1 << NF_IP_LOCAL_OUT)
 223                                       | (1 << NF_IP_POST_ROUTING)
 224                                       | (1 << NF_IP_PRE_ROUTING)
 225                                       | (1 << NF_IP_LOCAL_IN))) {
 226                         printk("ip_local_deliver: bad loopback skb: ");
 227                         debug_print_hooks_ip(skb->nf_debug);
 228                         nf_dump_skb(PF_INET, skb);
 229                 }
 230         }
 231         else {
 232                 if (skb->nf_debug != ((1<<NF_IP_PRE_ROUTING)
 233                                       | (1<<NF_IP_LOCAL_IN))) {
 234                         printk("ip_local_deliver: bad non-lo skb: ");
 235                         debug_print_hooks_ip(skb->nf_debug);
 236                         nf_dump_skb(PF_INET, skb);
 237                 }
 238         }
 239 }
 240
 241 void nf_debug_ip_loopback_xmit(struct sk_buff *newskb)
 242 {
 243         if (newskb->nf_debug != ((1 << NF_IP_LOCAL_OUT)
 244                                  | (1 << NF_IP_POST_ROUTING))) {
 245                 printk("ip_dev_loopback_xmit: bad owned skb = %p: ",
 246                        newskb);
 247                 debug_print_hooks_ip(newskb->nf_debug);
 248                 nf_dump_skb(PF_INET, newskb);
 249         }
 250         /* Clear to avoid confusing input check */
 251         newskb->nf_debug = 0;
 252 }
 253
 254 void nf_debug_ip_finish_output2(struct sk_buff *skb)
 255 {
 256         /* If it's owned, it must have gone through the
 257          * NF_IP_LOCAL_OUT and NF_IP_POST_ROUTING.
 258          * Otherwise, must have gone through
 259          * NF_IP_PRE_ROUTING, NF_IP_FORWARD and NF_IP_POST_ROUTING.
 260          */
 261         if (skb->sk) {
 262                 if (skb->nf_debug != ((1 << NF_IP_LOCAL_OUT)
 263                                       | (1 << NF_IP_POST_ROUTING))) {
 264                         printk("ip_finish_output: bad owned skb = %p: ", skb);
 265                         debug_print_hooks_ip(skb->nf_debug);
 266                         nf_dump_skb(PF_INET, skb);
 267                 }
 268         } else {
 269                 if (skb->nf_debug != ((1 << NF_IP_PRE_ROUTING)
 270                                       | (1 << NF_IP_FORWARD)
 271                                       | (1 << NF_IP_POST_ROUTING))) {
 272                         /* Fragments, entunnelled packets, TCP RSTs
 273                            generated by ipt_REJECT will have no
 274                            owners, but still may be local */
 275                         if (skb->nf_debug != ((1 << NF_IP_LOCAL_OUT)
 276                                               | (1 << NF_IP_POST_ROUTING))){
 277                                 printk("ip_finish_output:"
 278                                        " bad unowned skb = %p: ",skb);
 279                                 debug_print_hooks_ip(skb->nf_debug);
 280                                 nf_dump_skb(PF_INET, skb);
 281                         }
 282                 }
 283         }
 284 }
 285 #endif /*CONFIG_NETFILTER_DEBUG*/
 286
 287 /* Call get/setsockopt() */
 288 static int nf_sockopt(struct sock *sk, int pf, int val,
 289                       char __user *opt, int *len, int get)
 290 {
 291         struct list_head *i;
 292         struct nf_sockopt_ops *ops;
 293         int ret;
 294
 295         if (down_interruptible(&nf_sockopt_mutex) != 0)
 296                 return -EINTR;
 297
 298         list_for_each(i, &nf_sockopts) {
 299                 ops = (struct nf_sockopt_ops *)i;
 300                 if (ops->pf == pf) {
 301                         if (get) {
 302                                 if (val >= ops->get_optmin
 303                                     && val < ops->get_optmax) {
 304                                         ops->use++;
 305                                         up(&nf_sockopt_mutex);
 306                                         ret = ops->get(sk, val, opt, len);
 307                                         goto out;
 308                                 }
 309                         } else {
 310                                 if (val >= ops->set_optmin
 311                                     && val < ops->set_optmax) {
 312                                         ops->use++;
 313                                         up(&nf_sockopt_mutex);
 314                                         ret = ops->set(sk, val, opt, *len);
 315                                         goto out;
 316                                 }
 317                         }
 318                 }
 319         }
 320         up(&nf_sockopt_mutex);
 321         return -ENOPROTOOPT;
 322
 323  out:
 324         down(&nf_sockopt_mutex);
 325         ops->use--;
 326         if (ops->cleanup_task)
 327                 wake_up_process(ops->cleanup_task);
 328         up(&nf_sockopt_mutex);
 329         return ret;
 330 }
 331
 332 int nf_setsockopt(struct sock *sk, int pf, int val, char __user *opt,
 333                   int len)
 334 {
 335         return nf_sockopt(sk, pf, val, opt, &len, 0);
 336 }
 337
 338 int nf_getsockopt(struct sock *sk, int pf, int val, char __user *opt, int *len)
 339 {
 340         return nf_sockopt(sk, pf, val, opt, len, 1);
 341 }
 342
 343 static unsigned int nf_iterate(struct list_head *head,
 344                                struct sk_buff **skb,
 345                                int hook,
 346                                const struct net_device *indev,
 347                                const struct net_device *outdev,
 348                                struct list_head **i,
 349                                int (*okfn)(struct sk_buff *),
 350                                int hook_thresh)
 351 {
 352         /*
 353          * The caller must not block between calls to this
 354          * function because of risk of continuing from deleted element.
 355          */
 356         list_for_each_continue_rcu(*i, head) {
 357                 struct nf_hook_ops *elem = (struct nf_hook_ops *)*i;
 358
 359                 if (hook_thresh > elem->priority)
 360                         continue;
 361
 362                 /* Optimization: we don't need to hold module
 363                    reference here, since function can't sleep. --RR */
 364                 switch (elem->hook(hook, skb, indev, outdev, okfn)) {
 365                 case NF_QUEUE:
 366                         return NF_QUEUE;
 367
 368                 case NF_STOLEN:
 369                         return NF_STOLEN;
 370
 371                 case NF_DROP:
 372                         return NF_DROP;
 373
 374                 case NF_REPEAT:
 375                         *i = (*i)->prev;
 376                         break;
 377
 378 #ifdef CONFIG_NETFILTER_DEBUG
 379                 case NF_ACCEPT:
 380                         break;
 381
 382                 default:
 383                         NFDEBUG("Evil return from %p(%u).\n",
 384                                 elem->hook, hook);
 385 #endif
 386                 }
 387         }
 388         return NF_ACCEPT;
 389 }
 390
 391 int nf_register_queue_handler(int pf, nf_queue_outfn_t outfn, void *data)
 392 {
 393         int ret;
 394
 395         write_lock_bh(&queue_handler_lock);
 396         if (queue_handler[pf].outfn)
 397                 ret = -EBUSY;
 398         else {
 399                 queue_handler[pf].outfn = outfn;
 400                 queue_handler[pf].data = data;
 401                 ret = 0;
 402         }
 403         write_unlock_bh(&queue_handler_lock);
 404
 405         return ret;
 406 }
 407
 408 /* The caller must flush their queue before this */
 409 int nf_unregister_queue_handler(int pf)
 410 {
 411         write_lock_bh(&queue_handler_lock);
 412         queue_handler[pf].outfn = NULL;
 413         queue_handler[pf].data = NULL;
 414         write_unlock_bh(&queue_handler_lock);
 415
 416         return 0;
 417 }
 418
 419 /*
 420  * Any packet that leaves via this function must come back
 421  * through nf_reinject().
 422  */
 423 static int nf_queue(struct sk_buff *skb,
 424                     struct list_head *elem,
 425                     int pf, unsigned int hook,
 426                     struct net_device *indev,
 427                     struct net_device *outdev,
 428                     int (*okfn)(struct sk_buff *))
 429 {
 430         int status;
 431         struct nf_info *info;
 432 #ifdef CONFIG_BRIDGE_NETFILTER
 433         struct net_device *physindev = NULL;
 434         struct net_device *physoutdev = NULL;
 435 #endif
 436
 437         /* QUEUE == DROP if noone is waiting, to be safe. */
 438         read_lock(&queue_handler_lock);
 439         if (!queue_handler[pf].outfn) {
 440                 read_unlock(&queue_handler_lock);
 441                 kfree_skb(skb);
 442                 return 1;
 443         }
 444
 445         info = kmalloc(sizeof(*info), GFP_ATOMIC);
 446         if (!info) {
 447                 if (net_ratelimit())
 448                         printk(KERN_ERR "OOM queueing packet %p\n",
 449                                skb);
 450                 read_unlock(&queue_handler_lock);
 451                 kfree_skb(skb);
 452                 return 1;
 453         }
 454
 455         *info = (struct nf_info) {
 456                 (struct nf_hook_ops *)elem, pf, hook, indev, outdev, okfn };
 457
 458         /* If it's going away, ignore hook. */
 459         if (!try_module_get(info->elem->owner)) {
 460                 read_unlock(&queue_handler_lock);
 461                 kfree(info);
 462                 return 0;
 463         }
 464
 465         /* Bump dev refs so they don't vanish while packet is out */
 466         if (indev) dev_hold(indev);
 467         if (outdev) dev_hold(outdev);
 468
 469 #ifdef CONFIG_BRIDGE_NETFILTER
 470         if (skb->nf_bridge) {
 471                 physindev = skb->nf_bridge->physindev;
 472                 if (physindev) dev_hold(physindev);
 473                 physoutdev = skb->nf_bridge->physoutdev;
 474                 if (physoutdev) dev_hold(physoutdev);
 475         }
 476 #endif
 477
 478         status = queue_handler[pf].outfn(skb, info, queue_handler[pf].data);
 479         read_unlock(&queue_handler_lock);
 480
 481         if (status < 0) {
 482                 /* James M doesn't say fuck enough. */
 483                 if (indev) dev_put(indev);
 484                 if (outdev) dev_put(outdev);
 485 #ifdef CONFIG_BRIDGE_NETFILTER
 486                 if (physindev) dev_put(physindev);
 487                 if (physoutdev) dev_put(physoutdev);
 488 #endif
 489                 module_put(info->elem->owner);
 490                 kfree(info);
 491                 kfree_skb(skb);
 492                 return 1;
 493         }
 494         return 1;
 495 }
 496
 497 int nf_hook_slow(int pf, unsigned int hook, struct sk_buff *skb,
 498                  struct net_device *indev,
 499                  struct net_device *outdev,
 500                  int (*okfn)(struct sk_buff *),
 501                  int hook_thresh)
 502 {
 503         struct list_head *elem;
 504         unsigned int verdict;
 505         int ret = 0;
 506
 507         /* We may already have this, but read-locks nest anyway */
 508         rcu_read_lock();
 509
 510 #ifdef CONFIG_NETFILTER_DEBUG
 511         if (skb->nf_debug & (1 << hook)) {
 512                 printk("nf_hook: hook %i already set.\n", hook);
 513                 nf_dump_skb(pf, skb);
 514         }
 515         skb->nf_debug |= (1 << hook);
 516 #endif
 517
 518         elem = &nf_hooks[pf][hook];
 519  next_hook:
 520         verdict = nf_iterate(&nf_hooks[pf][hook], &skb, hook, indev,
 521                              outdev, &elem, okfn, hook_thresh);
 522         if (verdict == NF_QUEUE) {
 523                 NFDEBUG("nf_hook: Verdict = QUEUE.\n");
 524                 if (!nf_queue(skb, elem, pf, hook, indev, outdev, okfn))
 525                         goto next_hook;
 526         }
 527
 528         switch (verdict) {
 529         case NF_ACCEPT:
 530                 ret = okfn(skb);
 531                 break;
 532
 533         case NF_DROP:
 534                 kfree_skb(skb);
 535                 ret = -EPERM;
 536                 break;
 537         }
 538
 539         rcu_read_unlock();
 540         return ret;
 541 }
 542
 543 void nf_reinject(struct sk_buff *skb, struct nf_info *info,
 544                  unsigned int verdict)
 545 {
 546         struct list_head *elem = &info->elem->list;
 547         struct list_head *i;
 548
 549         rcu_read_lock();
 550
 551         /* Release those devices we held, or Alexey will kill me. */
 552         if (info->indev) dev_put(info->indev);
 553         if (info->outdev) dev_put(info->outdev);
 554 #ifdef CONFIG_BRIDGE_NETFILTER
 555         if (skb->nf_bridge) {
 556                 if (skb->nf_bridge->physindev)
 557                         dev_put(skb->nf_bridge->physindev);
 558                 if (skb->nf_bridge->physoutdev)
 559                         dev_put(skb->nf_bridge->physoutdev);
 560         }
 561 #endif
 562
 563         /* Drop reference to owner of hook which queued us. */
 564         module_put(info->elem->owner);
 565
 566         list_for_each_rcu(i, &nf_hooks[info->pf][info->hook]) {
 567                 if (i == elem)
 568                         break;
 569         }
 570
 571         if (elem == &nf_hooks[info->pf][info->hook]) {
 572                 /* The module which sent it to userspace is gone. */
 573                 NFDEBUG("%s: module disappeared, dropping packet.\n",
 574                         __FUNCTION__);
 575                 verdict = NF_DROP;
 576         }
 577
 578         /* Continue traversal iff userspace said ok... */
 579         if (verdict == NF_REPEAT) {
 580                 elem = elem->prev;
 581                 verdict = NF_ACCEPT;
 582         }
 583
 584         if (verdict == NF_ACCEPT) {
 585         next_hook:
 586                 verdict = nf_iterate(&nf_hooks[info->pf][info->hook],
 587                                      &skb, info->hook,
 588                                      info->indev, info->outdev, &elem,
 589                                      info->okfn, INT_MIN);
 590         }
 591
 592         switch (verdict) {
 593         case NF_ACCEPT:
 594                 info->okfn(skb);
 595                 break;
 596
 597         case NF_QUEUE:
 598                 if (!nf_queue(skb, elem, info->pf, info->hook,
 599                               info->indev, info->outdev, info->okfn))
 600                         goto next_hook;
 601                 break;
 602         }
 603         rcu_read_unlock();
 604
 605         if (verdict == NF_DROP)
 606                 kfree_skb(skb);
 607
 608         kfree(info);
 609         return;
 610 }
 611
 612 #ifdef CONFIG_INET
 613 /* route_me_harder function, used by iptable_nat, iptable_mangle + ip_queue */
 614 int ip_route_me_harder(struct sk_buff **pskb)
 615 {
 616         struct iphdr *iph = (*pskb)->nh.iph;
 617         struct rtable *rt;
 618         struct flowi fl = {};
 619         struct dst_entry *odst;
 620         unsigned int hh_len;
 621
 622         /* some non-standard hacks like ipt_REJECT.c:send_reset() can cause
 623          * packets with foreign saddr to appear on the NF_IP_LOCAL_OUT hook.
 624          */
 625         if (inet_addr_type(iph->saddr) == RTN_LOCAL) {
 626                 fl.nl_u.ip4_u.daddr = iph->daddr;
 627                 fl.nl_u.ip4_u.saddr = iph->saddr;
 628                 fl.nl_u.ip4_u.tos = RT_TOS(iph->tos);
 629                 fl.oif = (*pskb)->sk ? (*pskb)->sk->sk_bound_dev_if : 0;
 630 #ifdef CONFIG_IP_ROUTE_FWMARK
 631                 fl.nl_u.ip4_u.fwmark = (*pskb)->nfmark;
 632 #endif
 633                 fl.proto = iph->protocol;
 634                 if (ip_route_output_key(&rt, &fl) != 0)
 635                         return -1;
 636
 637                 /* Drop old route. */
 638                 dst_release((*pskb)->dst);
 639                 (*pskb)->dst = &rt->u.dst;
 640         } else {
 641                 /* non-local src, find valid iif to satisfy
 642                  * rp-filter when calling ip_route_input. */
 643                 fl.nl_u.ip4_u.daddr = iph->saddr;
 644                 if (ip_route_output_key(&rt, &fl) != 0)
 645                         return -1;
 646
 647                 odst = (*pskb)->dst;
 648                 if (ip_route_input(*pskb, iph->daddr, iph->saddr,
 649                                    RT_TOS(iph->tos), rt->u.dst.dev) != 0) {
 650                         dst_release(&rt->u.dst);
 651                         return -1;
 652                 }
 653                 dst_release(&rt->u.dst);
 654                 dst_release(odst);
 655         }
 656
 657         if ((*pskb)->dst->error)
 658                 return -1;
 659
 660         /* Change in oif may mean change in hh_len. */
 661         hh_len = (*pskb)->dst->dev->hard_header_len;
 662         if (skb_headroom(*pskb) < hh_len) {
 663                 struct sk_buff *nskb;
 664
 665                 nskb = skb_realloc_headroom(*pskb, hh_len);
 666                 if (!nskb)
 667                         return -1;
 668                 if ((*pskb)->sk)
 669                         skb_set_owner_w(nskb, (*pskb)->sk);
 670                 kfree_skb(*pskb);
 671                 *pskb = nskb;
 672         }
 673
 674         return 0;
 675 }
 676
 677 int skb_ip_make_writable(struct sk_buff **pskb, unsigned int writable_len)
 678 {
 679         struct sk_buff *nskb;
 680         unsigned int iplen;
 681
 682         if (writable_len > (*pskb)->len)
 683                 return 0;
 684
 685         /* Not exclusive use of packet?  Must copy. */
 686         if (skb_shared(*pskb) || skb_cloned(*pskb))
 687                 goto copy_skb;
 688
 689         /* Alexey says IP hdr is always modifiable and linear, so ok. */
 690         if (writable_len <= (*pskb)->nh.iph->ihl*4)
 691                 return 1;
 692
 693         iplen = writable_len - (*pskb)->nh.iph->ihl*4;
 694
 695         /* DaveM says protocol headers are also modifiable. */
 696         switch ((*pskb)->nh.iph->protocol) {
 697         case IPPROTO_TCP: {
 698                 struct tcphdr hdr;
 699                 if (skb_copy_bits(*pskb, (*pskb)->nh.iph->ihl*4,
 700                                   &hdr, sizeof(hdr)) != 0)
 701                         goto copy_skb;
 702                 if (writable_len <= (*pskb)->nh.iph->ihl*4 + hdr.doff*4)
 703                         goto pull_skb;
 704                 goto copy_skb;
 705         }
 706         case IPPROTO_UDP:
 707                 if (writable_len<=(*pskb)->nh.iph->ihl*4+sizeof(struct udphdr))
 708                         goto pull_skb;
 709                 goto copy_skb;
 710         case IPPROTO_ICMP:
 711                 if (writable_len
 712                     <= (*pskb)->nh.iph->ihl*4 + sizeof(struct icmphdr))
 713                         goto pull_skb;
 714                 goto copy_skb;
 715         /* Insert other cases here as desired */
 716         }
 717
 718 copy_skb:
 719         nskb = skb_copy(*pskb, GFP_ATOMIC);
 720         if (!nskb)
 721                 return 0;
 722         BUG_ON(skb_is_nonlinear(nskb));
 723
 724         /* Rest of kernel will get very unhappy if we pass it a
 725            suddenly-orphaned skbuff */
 726         if ((*pskb)->sk)
 727                 skb_set_owner_w(nskb, (*pskb)->sk);
 728         kfree_skb(*pskb);
 729         *pskb = nskb;
 730         return 1;
 731
 732 pull_skb:
 733         return pskb_may_pull(*pskb, writable_len);
 734 }
 735 EXPORT_SYMBOL(skb_ip_make_writable);
 736 #endif /*CONFIG_INET*/
 737
 738 /* Internal logging interface, which relies on the real
 739    LOG target modules */
 740
 741 #define NF_LOG_PREFIXLEN                128
 742
 743 static nf_logfn *nf_logging[NPROTO]; /* = NULL */
 744 static int reported = 0;
 745 static spinlock_t nf_log_lock = SPIN_LOCK_UNLOCKED;
 746
 747 int nf_log_register(int pf, nf_logfn *logfn)
 748 {
 749         int ret = -EBUSY;
 750
 751         /* Any setup of logging members must be done before
 752          * substituting pointer. */
 753         smp_wmb();
 754         spin_lock(&nf_log_lock);
 755         if (!nf_logging[pf]) {
 756                 nf_logging[pf] = logfn;
 757                 ret = 0;
 758         }
 759         spin_unlock(&nf_log_lock);
 760         return ret;
 761 }
 762
 763 void nf_log_unregister(int pf, nf_logfn *logfn)
 764 {
 765         spin_lock(&nf_log_lock);
 766         if (nf_logging[pf] == logfn)
 767                 nf_logging[pf] = NULL;
 768         spin_unlock(&nf_log_lock);
 769
 770         /* Give time to concurrent readers. */
 771         synchronize_net();
 772 }
 773
 774 void nf_log_packet(int pf,
 775                    unsigned int hooknum,
 776                    const struct sk_buff *skb,
 777                    const struct net_device *in,
 778                    const struct net_device *out,
 779                    const char *fmt, ...)
 780 {
 781         va_list args;
 782         char prefix[NF_LOG_PREFIXLEN];
 783         nf_logfn *logfn;
 784
 785         rcu_read_lock();
 786         logfn = nf_logging[pf];
 787         if (logfn) {
 788                 va_start(args, fmt);
 789                 vsnprintf(prefix, sizeof(prefix), fmt, args);
 790                 va_end(args);
 791                 /* We must read logging before nf_logfn[pf] */
 792                 smp_read_barrier_depends();
 793                 logfn(hooknum, skb, in, out, prefix);
 794         } else if (!reported) {
 795                 printk(KERN_WARNING "nf_log_packet: can\'t log yet, "
 796                        "no backend logging module loaded in!\n");
 797                 reported++;
 798         }
 799         rcu_read_unlock();
 800 }
 801 EXPORT_SYMBOL(nf_log_register);
 802 EXPORT_SYMBOL(nf_log_unregister);
 803 EXPORT_SYMBOL(nf_log_packet);
 804
 805 /* This does not belong here, but ipt_REJECT needs it if connection
 806    tracking in use: without this, connection may not be in hash table,
 807    and hence manufactured ICMP or RST packets will not be associated
 808    with it. */
 809 void (*ip_ct_attach)(struct sk_buff *, struct nf_ct_info *);
 810
 811 void __init netfilter_init(void)
 812 {
 813         int i, h;
 814
 815         for (i = 0; i < NPROTO; i++) {
 816                 for (h = 0; h < NF_MAX_HOOKS; h++)
 817                         INIT_LIST_HEAD(&nf_hooks[i][h]);
 818         }
 819 }
 820
 821 EXPORT_SYMBOL(ip_ct_attach);
 822 EXPORT_SYMBOL(ip_route_me_harder);
 823 EXPORT_SYMBOL(nf_getsockopt);
 824 EXPORT_SYMBOL(nf_hook_slow);
 825 EXPORT_SYMBOL(nf_hooks);
 826 EXPORT_SYMBOL(nf_register_hook);
 827 EXPORT_SYMBOL(nf_register_queue_handler);
 828 EXPORT_SYMBOL(nf_register_sockopt);
 829 EXPORT_SYMBOL(nf_reinject);
 830 EXPORT_SYMBOL(nf_setsockopt);
 831 EXPORT_SYMBOL(nf_unregister_hook);
 832 EXPORT_SYMBOL(nf_unregister_queue_handler);
 833 EXPORT_SYMBOL(nf_unregister_sockopt);