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(" MARK=%lu (0x%lu)",
 208                        (long unsigned int)skb->nfmark,
 209                        (long unsigned int)skb->nfmark);
 210                 printk("\n");
 211         }
 212         }
 213 }
 214
 215 void nf_debug_ip_local_deliver(struct sk_buff *skb)
 216 {
 217         /* If it's a loopback packet, it must have come through
 218          * NF_IP_LOCAL_OUT, NF_IP_RAW_INPUT, NF_IP_PRE_ROUTING and
 219          * NF_IP_LOCAL_IN.  Otherwise, must have gone through
 220          * NF_IP_RAW_INPUT and NF_IP_PRE_ROUTING.  */
 221         if (!skb->dev) {
 222                 printk("ip_local_deliver: skb->dev is NULL.\n");
 223         }
 224         else if (strcmp(skb->dev->name, "lo") == 0) {
 225                 if (skb->nf_debug != ((1 << NF_IP_LOCAL_OUT)
 226                                       | (1 << NF_IP_POST_ROUTING)
 227                                       | (1 << NF_IP_PRE_ROUTING)
 228                                       | (1 << NF_IP_LOCAL_IN))) {
 229                         printk("ip_local_deliver: bad loopback skb: ");
 230                         debug_print_hooks_ip(skb->nf_debug);
 231                         nf_dump_skb(PF_INET, skb);
 232                 }
 233         }
 234         else {
 235                 if (skb->nf_debug != ((1<<NF_IP_PRE_ROUTING)
 236                                       | (1<<NF_IP_LOCAL_IN))) {
 237                         printk("ip_local_deliver: bad non-lo skb: ");
 238                         debug_print_hooks_ip(skb->nf_debug);
 239                         nf_dump_skb(PF_INET, skb);
 240                 }
 241         }
 242 }
 243
 244 void nf_debug_ip_loopback_xmit(struct sk_buff *newskb)
 245 {
 246         if (newskb->nf_debug != ((1 << NF_IP_LOCAL_OUT)
 247                                  | (1 << NF_IP_POST_ROUTING))) {
 248                 printk("ip_dev_loopback_xmit: bad owned skb = %p: ",
 249                        newskb);
 250                 debug_print_hooks_ip(newskb->nf_debug);
 251                 nf_dump_skb(PF_INET, newskb);
 252         }
 253         /* Clear to avoid confusing input check */
 254         newskb->nf_debug = 0;
 255 }
 256
 257 void nf_debug_ip_finish_output2(struct sk_buff *skb)
 258 {
 259         /* If it's owned, it must have gone through the
 260          * NF_IP_LOCAL_OUT and NF_IP_POST_ROUTING.
 261          * Otherwise, must have gone through
 262          * NF_IP_PRE_ROUTING, NF_IP_FORWARD and NF_IP_POST_ROUTING.
 263          */
 264         if (skb->sk) {
 265                 if (skb->nf_debug != ((1 << NF_IP_LOCAL_OUT)
 266                                       | (1 << NF_IP_POST_ROUTING))) {
 267                         printk("ip_finish_output: bad owned skb = %p: ", skb);
 268                         debug_print_hooks_ip(skb->nf_debug);
 269                         nf_dump_skb(PF_INET, skb);
 270                 }
 271         } else {
 272                 if (skb->nf_debug != ((1 << NF_IP_PRE_ROUTING)
 273                                       | (1 << NF_IP_FORWARD)
 274                                       | (1 << NF_IP_POST_ROUTING))) {
 275                         /* Fragments, entunnelled packets, TCP RSTs
 276                            generated by ipt_REJECT will have no
 277                            owners, but still may be local */
 278                         if (skb->nf_debug != ((1 << NF_IP_LOCAL_OUT)
 279                                               | (1 << NF_IP_POST_ROUTING))){
 280                                 printk("ip_finish_output:"
 281                                        " bad unowned skb = %p: ",skb);
 282                                 debug_print_hooks_ip(skb->nf_debug);
 283                                 nf_dump_skb(PF_INET, skb);
 284                         }
 285                 }
 286         }
 287 }
 288 #endif /*CONFIG_NETFILTER_DEBUG*/
 289
 290 /* Call get/setsockopt() */
 291 static int nf_sockopt(struct sock *sk, int pf, int val,
 292                       char __user *opt, int *len, int get)
 293 {
 294         struct list_head *i;
 295         struct nf_sockopt_ops *ops;
 296         int ret;
 297
 298         if (down_interruptible(&nf_sockopt_mutex) != 0)
 299                 return -EINTR;
 300
 301         list_for_each(i, &nf_sockopts) {
 302                 ops = (struct nf_sockopt_ops *)i;
 303                 if (ops->pf == pf) {
 304                         if (get) {
 305                                 if (val >= ops->get_optmin
 306                                     && val < ops->get_optmax) {
 307                                         ops->use++;
 308                                         up(&nf_sockopt_mutex);
 309                                         ret = ops->get(sk, val, opt, len);
 310                                         goto out;
 311                                 }
 312                         } else {
 313                                 if (val >= ops->set_optmin
 314                                     && val < ops->set_optmax) {
 315                                         ops->use++;
 316                                         up(&nf_sockopt_mutex);
 317                                         ret = ops->set(sk, val, opt, *len);
 318                                         goto out;
 319                                 }
 320                         }
 321                 }
 322         }
 323         up(&nf_sockopt_mutex);
 324         return -ENOPROTOOPT;
 325
 326  out:
 327         down(&nf_sockopt_mutex);
 328         ops->use--;
 329         if (ops->cleanup_task)
 330                 wake_up_process(ops->cleanup_task);
 331         up(&nf_sockopt_mutex);
 332         return ret;
 333 }
 334
 335 int nf_setsockopt(struct sock *sk, int pf, int val, char __user *opt,
 336                   int len)
 337 {
 338         return nf_sockopt(sk, pf, val, opt, &len, 0);
 339 }
 340
 341 int nf_getsockopt(struct sock *sk, int pf, int val, char __user *opt, int *len)
 342 {
 343         return nf_sockopt(sk, pf, val, opt, len, 1);
 344 }
 345
 346 static unsigned int nf_iterate(struct list_head *head,
 347                                struct sk_buff **skb,
 348                                int hook,
 349                                const struct net_device *indev,
 350                                const struct net_device *outdev,
 351                                struct list_head **i,
 352                                int (*okfn)(struct sk_buff *),
 353                                int hook_thresh)
 354 {
 355         /*
 356          * The caller must not block between calls to this
 357          * function because of risk of continuing from deleted element.
 358          */
 359         list_for_each_continue_rcu(*i, head) {
 360                 struct nf_hook_ops *elem = (struct nf_hook_ops *)*i;
 361
 362                 if (hook_thresh > elem->priority)
 363                         continue;
 364
 365                 /* Optimization: we don't need to hold module
 366                    reference here, since function can't sleep. --RR */
 367                 switch (elem->hook(hook, skb, indev, outdev, okfn)) {
 368                 case NF_QUEUE:
 369                         return NF_QUEUE;
 370
 371                 case NF_STOLEN:
 372                         return NF_STOLEN;
 373
 374                 case NF_DROP:
 375                         return NF_DROP;
 376
 377                 case NF_REPEAT:
 378                         *i = (*i)->prev;
 379                         break;
 380
 381 #ifdef CONFIG_NETFILTER_DEBUG
 382                 case NF_ACCEPT:
 383                         break;
 384
 385                 default:
 386                         NFDEBUG("Evil return from %p(%u).\n",
 387                                 elem->hook, hook);
 388 #endif
 389                 }
 390         }
 391         return NF_ACCEPT;
 392 }
 393
 394 int nf_register_queue_handler(int pf, nf_queue_outfn_t outfn, void *data)
 395 {
 396         int ret;
 397
 398         write_lock_bh(&queue_handler_lock);
 399         if (queue_handler[pf].outfn)
 400                 ret = -EBUSY;
 401         else {
 402                 queue_handler[pf].outfn = outfn;
 403                 queue_handler[pf].data = data;
 404                 ret = 0;
 405         }
 406         write_unlock_bh(&queue_handler_lock);
 407
 408         return ret;
 409 }
 410
 411 /* The caller must flush their queue before this */
 412 int nf_unregister_queue_handler(int pf)
 413 {
 414         write_lock_bh(&queue_handler_lock);
 415         queue_handler[pf].outfn = NULL;
 416         queue_handler[pf].data = NULL;
 417         write_unlock_bh(&queue_handler_lock);
 418
 419         return 0;
 420 }
 421
 422 /*
 423  * Any packet that leaves via this function must come back
 424  * through nf_reinject().
 425  */
 426 static int nf_queue(struct sk_buff *skb,
 427                     struct list_head *elem,
 428                     int pf, unsigned int hook,
 429                     struct net_device *indev,
 430                     struct net_device *outdev,
 431                     int (*okfn)(struct sk_buff *))
 432 {
 433         int status;
 434         struct nf_info *info;
 435 #ifdef CONFIG_BRIDGE_NETFILTER
 436         struct net_device *physindev = NULL;
 437         struct net_device *physoutdev = NULL;
 438 #endif
 439
 440         /* QUEUE == DROP if noone is waiting, to be safe. */
 441         read_lock(&queue_handler_lock);
 442         if (!queue_handler[pf].outfn) {
 443                 read_unlock(&queue_handler_lock);
 444                 kfree_skb(skb);
 445                 return 1;
 446         }
 447
 448         info = kmalloc(sizeof(*info), GFP_ATOMIC);
 449         if (!info) {
 450                 if (net_ratelimit())
 451                         printk(KERN_ERR "OOM queueing packet %p\n",
 452                                skb);
 453                 read_unlock(&queue_handler_lock);
 454                 kfree_skb(skb);
 455                 return 1;
 456         }
 457
 458         *info = (struct nf_info) {
 459                 (struct nf_hook_ops *)elem, pf, hook, indev, outdev, okfn };
 460
 461         /* If it's going away, ignore hook. */
 462         if (!try_module_get(info->elem->owner)) {
 463                 read_unlock(&queue_handler_lock);
 464                 kfree(info);
 465                 return 0;
 466         }
 467
 468         /* Bump dev refs so they don't vanish while packet is out */
 469         if (indev) dev_hold(indev);
 470         if (outdev) dev_hold(outdev);
 471
 472 #ifdef CONFIG_BRIDGE_NETFILTER
 473         if (skb->nf_bridge) {
 474                 physindev = skb->nf_bridge->physindev;
 475                 if (physindev) dev_hold(physindev);
 476                 physoutdev = skb->nf_bridge->physoutdev;
 477                 if (physoutdev) dev_hold(physoutdev);
 478         }
 479 #endif
 480
 481         status = queue_handler[pf].outfn(skb, info, queue_handler[pf].data);
 482         read_unlock(&queue_handler_lock);
 483
 484         if (status < 0) {
 485                 /* James M doesn't say fuck enough. */
 486                 if (indev) dev_put(indev);
 487                 if (outdev) dev_put(outdev);
 488 #ifdef CONFIG_BRIDGE_NETFILTER
 489                 if (physindev) dev_put(physindev);
 490                 if (physoutdev) dev_put(physoutdev);
 491 #endif
 492                 module_put(info->elem->owner);
 493                 kfree(info);
 494                 kfree_skb(skb);
 495                 return 1;
 496         }
 497         return 1;
 498 }
 499
 500 int nf_hook_slow(int pf, unsigned int hook, struct sk_buff *skb,
 501                  struct net_device *indev,
 502                  struct net_device *outdev,
 503                  int (*okfn)(struct sk_buff *),
 504                  int hook_thresh)
 505 {
 506         struct list_head *elem;
 507         unsigned int verdict;
 508         int ret = 0;
 509
 510         /* We may already have this, but read-locks nest anyway */
 511         rcu_read_lock();
 512
 513 #ifdef CONFIG_NETFILTER_DEBUG
 514         if (skb->nf_debug & (1 << hook)) {
 515                 printk("nf_hook: hook %i already set.\n", hook);
 516                 nf_dump_skb(pf, skb);
 517         }
 518         skb->nf_debug |= (1 << hook);
 519 #endif
 520
 521         elem = &nf_hooks[pf][hook];
 522  next_hook:
 523         verdict = nf_iterate(&nf_hooks[pf][hook], &skb, hook, indev,
 524                              outdev, &elem, okfn, hook_thresh);
 525         if (verdict == NF_QUEUE) {
 526                 NFDEBUG("nf_hook: Verdict = QUEUE.\n");
 527                 if (!nf_queue(skb, elem, pf, hook, indev, outdev, okfn))
 528                         goto next_hook;
 529         }
 530
 531         switch (verdict) {
 532         case NF_ACCEPT:
 533                 ret = okfn(skb);
 534                 break;
 535
 536         case NF_DROP:
 537                 kfree_skb(skb);
 538                 ret = -EPERM;
 539                 break;
 540         }
 541
 542         rcu_read_unlock();
 543         return ret;
 544 }
 545
 546 void nf_reinject(struct sk_buff *skb, struct nf_info *info,
 547                  unsigned int verdict)
 548 {
 549         struct list_head *elem = &info->elem->list;
 550         struct list_head *i;
 551
 552         rcu_read_lock();
 553
 554         /* Release those devices we held, or Alexey will kill me. */
 555         if (info->indev) dev_put(info->indev);
 556         if (info->outdev) dev_put(info->outdev);
 557 #ifdef CONFIG_BRIDGE_NETFILTER
 558         if (skb->nf_bridge) {
 559                 if (skb->nf_bridge->physindev)
 560                         dev_put(skb->nf_bridge->physindev);
 561                 if (skb->nf_bridge->physoutdev)
 562                         dev_put(skb->nf_bridge->physoutdev);
 563         }
 564 #endif
 565
 566         /* Drop reference to owner of hook which queued us. */
 567         module_put(info->elem->owner);
 568
 569         list_for_each_rcu(i, &nf_hooks[info->pf][info->hook]) {
 570                 if (i == elem)
 571                         break;
 572         }
 573
 574         if (elem == &nf_hooks[info->pf][info->hook]) {
 575                 /* The module which sent it to userspace is gone. */
 576                 NFDEBUG("%s: module disappeared, dropping packet.\n",
 577                         __FUNCTION__);
 578                 verdict = NF_DROP;
 579         }
 580
 581         /* Continue traversal iff userspace said ok... */
 582         if (verdict == NF_REPEAT) {
 583                 elem = elem->prev;
 584                 verdict = NF_ACCEPT;
 585         }
 586
 587         if (verdict == NF_ACCEPT) {
 588         next_hook:
 589                 verdict = nf_iterate(&nf_hooks[info->pf][info->hook],
 590                                      &skb, info->hook,
 591                                      info->indev, info->outdev, &elem,
 592                                      info->okfn, INT_MIN);
 593         }
 594
 595         switch (verdict) {
 596         case NF_ACCEPT:
 597                 info->okfn(skb);
 598                 break;
 599
 600         case NF_QUEUE:
 601                 if (!nf_queue(skb, elem, info->pf, info->hook,
 602                               info->indev, info->outdev, info->okfn))
 603                         goto next_hook;
 604                 break;
 605         }
 606         rcu_read_unlock();
 607
 608         if (verdict == NF_DROP)
 609                 kfree_skb(skb);
 610
 611         kfree(info);
 612         return;
 613 }
 614
 615 #ifdef CONFIG_INET
 616 /* route_me_harder function, used by iptable_nat, iptable_mangle + ip_queue */
 617 int ip_route_me_harder(struct sk_buff **pskb)
 618 {
 619         struct iphdr *iph = (*pskb)->nh.iph;
 620         struct rtable *rt;
 621         struct flowi fl = {};
 622         struct dst_entry *odst;
 623         unsigned int hh_len;
 624
 625         /* some non-standard hacks like ipt_REJECT.c:send_reset() can cause
 626          * packets with foreign saddr to appear on the NF_IP_LOCAL_OUT hook.
 627          */
 628         if (inet_addr_type(iph->saddr) == RTN_LOCAL) {
 629                 fl.nl_u.ip4_u.daddr = iph->daddr;
 630                 fl.nl_u.ip4_u.saddr = iph->saddr;
 631                 fl.nl_u.ip4_u.tos = RT_TOS(iph->tos);
 632                 fl.oif = (*pskb)->sk ? (*pskb)->sk->sk_bound_dev_if : 0;
 633 #ifdef CONFIG_IP_ROUTE_FWMARK
 634                 fl.nl_u.ip4_u.fwmark = (*pskb)->nfmark;
 635 #endif
 636                 fl.proto = iph->protocol;
 637                 if (ip_route_output_key(&rt, &fl) != 0)
 638                         return -1;
 639
 640                 /* Drop old route. */
 641                 dst_release((*pskb)->dst);
 642                 (*pskb)->dst = &rt->u.dst;
 643         } else {
 644                 /* non-local src, find valid iif to satisfy
 645                  * rp-filter when calling ip_route_input. */
 646                 fl.nl_u.ip4_u.daddr = iph->saddr;
 647                 if (ip_route_output_key(&rt, &fl) != 0)
 648                         return -1;
 649
 650                 odst = (*pskb)->dst;
 651                 if (ip_route_input(*pskb, iph->daddr, iph->saddr,
 652                                    RT_TOS(iph->tos), rt->u.dst.dev) != 0) {
 653                         dst_release(&rt->u.dst);
 654                         return -1;
 655                 }
 656                 dst_release(&rt->u.dst);
 657                 dst_release(odst);
 658         }
 659
 660         if ((*pskb)->dst->error)
 661                 return -1;
 662
 663         /* Change in oif may mean change in hh_len. */
 664         hh_len = (*pskb)->dst->dev->hard_header_len;
 665         if (skb_headroom(*pskb) < hh_len) {
 666                 struct sk_buff *nskb;
 667
 668                 nskb = skb_realloc_headroom(*pskb, hh_len);
 669                 if (!nskb)
 670                         return -1;
 671                 if ((*pskb)->sk)
 672                         skb_set_owner_w(nskb, (*pskb)->sk);
 673                 kfree_skb(*pskb);
 674                 *pskb = nskb;
 675         }
 676
 677         return 0;
 678 }
 679
 680 int skb_ip_make_writable(struct sk_buff **pskb, unsigned int writable_len)
 681 {
 682         struct sk_buff *nskb;
 683         unsigned int iplen;
 684
 685         if (writable_len > (*pskb)->len)
 686                 return 0;
 687
 688         /* Not exclusive use of packet?  Must copy. */
 689         if (skb_shared(*pskb) || skb_cloned(*pskb))
 690                 goto copy_skb;
 691
 692         /* Alexey says IP hdr is always modifiable and linear, so ok. */
 693         if (writable_len <= (*pskb)->nh.iph->ihl*4)
 694                 return 1;
 695
 696         iplen = writable_len - (*pskb)->nh.iph->ihl*4;
 697
 698         /* DaveM says protocol headers are also modifiable. */
 699         switch ((*pskb)->nh.iph->protocol) {
 700         case IPPROTO_TCP: {
 701                 struct tcphdr hdr;
 702                 if (skb_copy_bits(*pskb, (*pskb)->nh.iph->ihl*4,
 703                                   &hdr, sizeof(hdr)) != 0)
 704                         goto copy_skb;
 705                 if (writable_len <= (*pskb)->nh.iph->ihl*4 + hdr.doff*4)
 706                         goto pull_skb;
 707                 goto copy_skb;
 708         }
 709         case IPPROTO_UDP:
 710                 if (writable_len<=(*pskb)->nh.iph->ihl*4+sizeof(struct udphdr))
 711                         goto pull_skb;
 712                 goto copy_skb;
 713         case IPPROTO_ICMP:
 714                 if (writable_len
 715                     <= (*pskb)->nh.iph->ihl*4 + sizeof(struct icmphdr))
 716                         goto pull_skb;
 717                 goto copy_skb;
 718         /* Insert other cases here as desired */
 719         }
 720
 721 copy_skb:
 722         nskb = skb_copy(*pskb, GFP_ATOMIC);
 723         if (!nskb)
 724                 return 0;
 725         BUG_ON(skb_is_nonlinear(nskb));
 726
 727         /* Rest of kernel will get very unhappy if we pass it a
 728            suddenly-orphaned skbuff */
 729         if ((*pskb)->sk)
 730                 skb_set_owner_w(nskb, (*pskb)->sk);
 731         kfree_skb(*pskb);
 732         *pskb = nskb;
 733         return 1;
 734
 735 pull_skb:
 736         return pskb_may_pull(*pskb, writable_len);
 737 }
 738 EXPORT_SYMBOL(skb_ip_make_writable);
 739 #endif /*CONFIG_INET*/
 740
 741 /* Internal logging interface, which relies on the real
 742    LOG target modules */
 743
 744 #define NF_LOG_PREFIXLEN                128
 745
 746 static nf_logfn *nf_logging[NPROTO]; /* = NULL */
 747 static int reported = 0;
 748 static spinlock_t nf_log_lock = SPIN_LOCK_UNLOCKED;
 749
 750 int nf_log_register(int pf, nf_logfn *logfn)
 751 {
 752         int ret = -EBUSY;
 753
 754         /* Any setup of logging members must be done before
 755          * substituting pointer. */
 756         smp_wmb();
 757         spin_lock(&nf_log_lock);
 758         if (!nf_logging[pf]) {
 759                 nf_logging[pf] = logfn;
 760                 ret = 0;
 761         }
 762         spin_unlock(&nf_log_lock);
 763         return ret;
 764 }
 765
 766 void nf_log_unregister(int pf, nf_logfn *logfn)
 767 {
 768         spin_lock(&nf_log_lock);
 769         if (nf_logging[pf] == logfn)
 770                 nf_logging[pf] = NULL;
 771         spin_unlock(&nf_log_lock);
 772
 773         /* Give time to concurrent readers. */
 774         synchronize_net();
 775 }
 776
 777 void nf_log_packet(int pf,
 778                    unsigned int hooknum,
 779                    const struct sk_buff *skb,
 780                    const struct net_device *in,
 781                    const struct net_device *out,
 782                    const char *fmt, ...)
 783 {
 784         va_list args;
 785         char prefix[NF_LOG_PREFIXLEN];
 786         nf_logfn *logfn;
 787
 788         rcu_read_lock();
 789         logfn = nf_logging[pf];
 790         if (logfn) {
 791                 va_start(args, fmt);
 792                 vsnprintf(prefix, sizeof(prefix), fmt, args);
 793                 va_end(args);
 794                 /* We must read logging before nf_logfn[pf] */
 795                 smp_read_barrier_depends();
 796                 logfn(hooknum, skb, in, out, prefix);
 797         } else if (!reported) {
 798                 printk(KERN_WARNING "nf_log_packet: can\'t log yet, "
 799                        "no backend logging module loaded in!\n");
 800                 reported++;
 801         }
 802         rcu_read_unlock();
 803 }
 804 EXPORT_SYMBOL(nf_log_register);
 805 EXPORT_SYMBOL(nf_log_unregister);
 806 EXPORT_SYMBOL(nf_log_packet);
 807
 808 /* This does not belong here, but ipt_REJECT needs it if connection
 809    tracking in use: without this, connection may not be in hash table,
 810    and hence manufactured ICMP or RST packets will not be associated
 811    with it. */
 812 void (*ip_ct_attach)(struct sk_buff *, struct nf_ct_info *);
 813
 814 void __init netfilter_init(void)
 815 {
 816         int i, h;
 817
 818         for (i = 0; i < NPROTO; i++) {
 819                 for (h = 0; h < NF_MAX_HOOKS; h++)
 820                         INIT_LIST_HEAD(&nf_hooks[i][h]);
 821         }
 822 }
 823
 824 EXPORT_SYMBOL(ip_ct_attach);
 825 EXPORT_SYMBOL(ip_route_me_harder);
 826 EXPORT_SYMBOL(nf_getsockopt);
 827 EXPORT_SYMBOL(nf_hook_slow);
 828 EXPORT_SYMBOL(nf_hooks);
 829 EXPORT_SYMBOL(nf_register_hook);
 830 EXPORT_SYMBOL(nf_register_queue_handler);
 831 EXPORT_SYMBOL(nf_register_sockopt);
 832 EXPORT_SYMBOL(nf_reinject);
 833 EXPORT_SYMBOL(nf_setsockopt);
 834 EXPORT_SYMBOL(nf_unregister_hook);
 835 EXPORT_SYMBOL(nf_unregister_queue_handler);
 836 EXPORT_SYMBOL(nf_unregister_sockopt);
 837 #ifdef CONFIG_NETFILTER_DEBUG
 838 EXPORT_SYMBOL(nf_dump_skb);
 839 #endif