net/netfilter/nf_queue.c

   1 /*
   2  * Rusty Russell (C)2000 -- This code is GPL.
   3  * Patrick McHardy (c) 2006-2012
   4  */
   5
   6 #include <linux/kernel.h>
   7 #include <linux/slab.h>
   8 #include <linux/init.h>
   9 #include <linux/module.h>
  10 #include <linux/proc_fs.h>
  11 #include <linux/skbuff.h>
  12 #include <linux/netfilter.h>
  13 #include <linux/netfilter_ipv4.h>
  14 #include <linux/netfilter_ipv6.h>
  15 #include <linux/netfilter_bridge.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/rcupdate.h>
  18 #include <net/protocol.h>
  19 #include <net/netfilter/nf_queue.h>
  20 #include <net/dst.h>
  21
  22 #include "nf_internals.h"
  23
  24 static const struct nf_queue_handler __rcu *nf_queue_handler;
  25
  26 /*
  27  * Hook for nfnetlink_queue to register its queue handler.
  28  * We do this so that most of the NFQUEUE code can be modular.
  29  *
  30  * Once the queue is registered it must reinject all packets it
  31  * receives, no matter what.
  32  */
  33
  34 void nf_register_queue_handler(const struct nf_queue_handler *qh)
  35 {
  36         /* should never happen, we only have one queueing backend in kernel */
  37         WARN_ON(rcu_access_pointer(nf_queue_handler));
  38         rcu_assign_pointer(nf_queue_handler, qh);
  39 }
  40 EXPORT_SYMBOL(nf_register_queue_handler);
  41
  42 /* The caller must flush their queue before this */
  43 void nf_unregister_queue_handler(void)
  44 {
  45         RCU_INIT_POINTER(nf_queue_handler, NULL);
  46 }
  47 EXPORT_SYMBOL(nf_unregister_queue_handler);
  48
  49 static void nf_queue_sock_put(struct sock *sk)
  50 {
  51 #ifdef CONFIG_INET
  52         sock_gen_put(sk);
  53 #else
  54         sock_put(sk);
  55 #endif
  56 }
  57
  58 static void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
  59 {
  60         struct nf_hook_state *state = &entry->state;
  61
  62         /* Release those devices we held, or Alexey will kill me. */
  63         dev_put(state->in);
  64         dev_put(state->out);
  65         if (state->sk)
  66                 nf_queue_sock_put(state->sk);
  67
  68 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  69         dev_put(entry->physin);
  70         dev_put(entry->physout);
  71 #endif
  72 }
  73
  74 void nf_queue_entry_free(struct nf_queue_entry *entry)
  75 {
  76         nf_queue_entry_release_refs(entry);
  77         kfree(entry);
  78 }
  79 EXPORT_SYMBOL_GPL(nf_queue_entry_free);
  80
  81 static void __nf_queue_entry_init_physdevs(struct nf_queue_entry *entry)
  82 {
  83 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  84         const struct sk_buff *skb = entry->skb;
  85         struct nf_bridge_info *nf_bridge;
  86
  87         nf_bridge = nf_bridge_info_get(skb);
  88         if (nf_bridge) {
  89                 entry->physin = nf_bridge_get_physindev(skb);
  90                 entry->physout = nf_bridge_get_physoutdev(skb);
  91         } else {
  92                 entry->physin = NULL;
  93                 entry->physout = NULL;
  94         }
  95 #endif
  96 }
  97
  98 /* Bump dev refs so they don't vanish while packet is out */
  99 bool nf_queue_entry_get_refs(struct nf_queue_entry *entry)
 100 {
 101         struct nf_hook_state *state = &entry->state;
 102
 103         if (state->sk && !refcount_inc_not_zero(&state->sk->sk_refcnt))
 104                 return false;
 105
 106         dev_hold(state->in);
 107         dev_hold(state->out);
 108
 109 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 110         dev_hold(entry->physin);
 111         dev_hold(entry->physout);
 112 #endif
 113         return true;
 114 }
 115 EXPORT_SYMBOL_GPL(nf_queue_entry_get_refs);
 116
 117 void nf_queue_nf_hook_drop(struct net *net)
 118 {
 119         const struct nf_queue_handler *qh;
 120
 121         rcu_read_lock();
 122         qh = rcu_dereference(nf_queue_handler);
 123         if (qh)
 124                 qh->nf_hook_drop(net);
 125         rcu_read_unlock();
 126 }
 127 EXPORT_SYMBOL_GPL(nf_queue_nf_hook_drop);
 128
 129 static void nf_ip_saveroute(const struct sk_buff *skb,
 130                             struct nf_queue_entry *entry)
 131 {
 132         struct ip_rt_info *rt_info = nf_queue_entry_reroute(entry);
 133
 134         if (entry->state.hook == NF_INET_LOCAL_OUT) {
 135                 const struct iphdr *iph = ip_hdr(skb);
 136
 137                 rt_info->tos = iph->tos;
 138                 rt_info->daddr = iph->daddr;
 139                 rt_info->saddr = iph->saddr;
 140                 rt_info->mark = skb->mark;
 141         }
 142 }
 143
 144 static void nf_ip6_saveroute(const struct sk_buff *skb,
 145                              struct nf_queue_entry *entry)
 146 {
 147         struct ip6_rt_info *rt_info = nf_queue_entry_reroute(entry);
 148
 149         if (entry->state.hook == NF_INET_LOCAL_OUT) {
 150                 const struct ipv6hdr *iph = ipv6_hdr(skb);
 151
 152                 rt_info->daddr = iph->daddr;
 153                 rt_info->saddr = iph->saddr;
 154                 rt_info->mark = skb->mark;
 155         }
 156 }
 157
 158 static int __nf_queue(struct sk_buff *skb, const struct nf_hook_state *state,
 159                       unsigned int index, unsigned int queuenum)
 160 {
 161         struct nf_queue_entry *entry = NULL;
 162         const struct nf_queue_handler *qh;
 163         unsigned int route_key_size;
 164         int status;
 165
 166         /* QUEUE == DROP if no one is waiting, to be safe. */
 167         qh = rcu_dereference(nf_queue_handler);
 168         if (!qh)
 169                 return -ESRCH;
 170
 171         switch (state->pf) {
 172         case AF_INET:
 173                 route_key_size = sizeof(struct ip_rt_info);
 174                 break;
 175         case AF_INET6:
 176                 route_key_size = sizeof(struct ip6_rt_info);
 177                 break;
 178         default:
 179                 route_key_size = 0;
 180                 break;
 181         }
 182
 183         if (skb_sk_is_prefetched(skb)) {
 184                 struct sock *sk = skb->sk;
 185
 186                 if (!sk_is_refcounted(sk)) {
 187                         if (!refcount_inc_not_zero(&sk->sk_refcnt))
 188                                 return -ENOTCONN;
 189
 190                         /* drop refcount on skb_orphan */
 191                         skb->destructor = sock_edemux;
 192                 }
 193         }
 194
 195         entry = kmalloc(sizeof(*entry) + route_key_size, GFP_ATOMIC);
 196         if (!entry)
 197                 return -ENOMEM;
 198
 199         if (skb_dst(skb) && !skb_dst_force(skb)) {
 200                 kfree(entry);
 201                 return -ENETDOWN;
 202         }
 203
 204         *entry = (struct nf_queue_entry) {
 205                 .skb    = skb,
 206                 .state  = *state,
 207                 .hook_index = index,
 208                 .size   = sizeof(*entry) + route_key_size,
 209         };
 210
 211         __nf_queue_entry_init_physdevs(entry);
 212
 213         if (!nf_queue_entry_get_refs(entry)) {
 214                 kfree(entry);
 215                 return -ENOTCONN;
 216         }
 217
 218         switch (entry->state.pf) {
 219         case AF_INET:
 220                 nf_ip_saveroute(skb, entry);
 221                 break;
 222         case AF_INET6:
 223                 nf_ip6_saveroute(skb, entry);
 224                 break;
 225         }
 226
 227         status = qh->outfn(entry, queuenum);
 228         if (status < 0) {
 229                 nf_queue_entry_free(entry);
 230                 return status;
 231         }
 232
 233         return 0;
 234 }
 235
 236 /* Packets leaving via this function must come back through nf_reinject(). */
 237 int nf_queue(struct sk_buff *skb, struct nf_hook_state *state,
 238              unsigned int index, unsigned int verdict)
 239 {
 240         int ret;
 241
 242         ret = __nf_queue(skb, state, index, verdict >> NF_VERDICT_QBITS);
 243         if (ret < 0) {
 244                 if (ret == -ESRCH &&
 245                     (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 246                         return 1;
 247                 kfree_skb(skb);
 248         }
 249
 250         return 0;
 251 }
 252 EXPORT_SYMBOL_GPL(nf_queue);
 253
 254 static unsigned int nf_iterate(struct sk_buff *skb,
 255                                struct nf_hook_state *state,
 256                                const struct nf_hook_entries *hooks,
 257                                unsigned int *index)
 258 {
 259         const struct nf_hook_entry *hook;
 260         unsigned int verdict, i = *index;
 261
 262         while (i < hooks->num_hook_entries) {
 263                 hook = &hooks->hooks[i];
 264 repeat:
 265                 verdict = nf_hook_entry_hookfn(hook, skb, state);
 266                 if (verdict != NF_ACCEPT) {
 267                         *index = i;
 268                         if (verdict != NF_REPEAT)
 269                                 return verdict;
 270                         goto repeat;
 271                 }
 272                 i++;
 273         }
 274
 275         *index = i;
 276         return NF_ACCEPT;
 277 }
 278
 279 static struct nf_hook_entries *nf_hook_entries_head(const struct net *net, u8 pf, u8 hooknum)
 280 {
 281         switch (pf) {
 282 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
 283         case NFPROTO_BRIDGE:
 284                 return rcu_dereference(net->nf.hooks_bridge[hooknum]);
 285 #endif
 286         case NFPROTO_IPV4:
 287                 return rcu_dereference(net->nf.hooks_ipv4[hooknum]);
 288         case NFPROTO_IPV6:
 289                 return rcu_dereference(net->nf.hooks_ipv6[hooknum]);
 290         default:
 291                 WARN_ON_ONCE(1);
 292                 return NULL;
 293         }
 294
 295         return NULL;
 296 }
 297
 298 /* Caller must hold rcu read-side lock */
 299 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 300 {
 301         const struct nf_hook_entry *hook_entry;
 302         const struct nf_hook_entries *hooks;
 303         struct sk_buff *skb = entry->skb;
 304         const struct net *net;
 305         unsigned int i;
 306         int err;
 307         u8 pf;
 308
 309         net = entry->state.net;
 310         pf = entry->state.pf;
 311
 312         hooks = nf_hook_entries_head(net, pf, entry->state.hook);
 313
 314         i = entry->hook_index;
 315         if (WARN_ON_ONCE(!hooks || i >= hooks->num_hook_entries)) {
 316                 kfree_skb(skb);
 317                 nf_queue_entry_free(entry);
 318                 return;
 319         }
 320
 321         hook_entry = &hooks->hooks[i];
 322
 323         /* Continue traversal iff userspace said ok... */
 324         if (verdict == NF_REPEAT)
 325                 verdict = nf_hook_entry_hookfn(hook_entry, skb, &entry->state);
 326
 327         if (verdict == NF_ACCEPT) {
 328                 if (nf_reroute(skb, entry) < 0)
 329                         verdict = NF_DROP;
 330         }
 331
 332         if (verdict == NF_ACCEPT) {
 333 next_hook:
 334                 ++i;
 335                 verdict = nf_iterate(skb, &entry->state, hooks, &i);
 336         }
 337
 338         switch (verdict & NF_VERDICT_MASK) {
 339         case NF_ACCEPT:
 340         case NF_STOP:
 341                 local_bh_disable();
 342                 entry->state.okfn(entry->state.net, entry->state.sk, skb);
 343                 local_bh_enable();
 344                 break;
 345         case NF_QUEUE:
 346                 err = nf_queue(skb, &entry->state, i, verdict);
 347                 if (err == 1)
 348                         goto next_hook;
 349                 break;
 350         case NF_STOLEN:
 351                 break;
 352         default:
 353                 kfree_skb(skb);
 354         }
 355
 356         nf_queue_entry_free(entry);
 357 }
 358 EXPORT_SYMBOL(nf_reinject);