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
  86         if (nf_bridge_info_exists(skb)) {
  87                 entry->physin = nf_bridge_get_physindev(skb);
  88                 entry->physout = nf_bridge_get_physoutdev(skb);
  89         } else {
  90                 entry->physin = NULL;
  91                 entry->physout = NULL;
  92         }
  93 #endif
  94 }
  95
  96 /* Bump dev refs so they don't vanish while packet is out */
  97 bool nf_queue_entry_get_refs(struct nf_queue_entry *entry)
  98 {
  99         struct nf_hook_state *state = &entry->state;
 100
 101         if (state->sk && !refcount_inc_not_zero(&state->sk->sk_refcnt))
 102                 return false;
 103
 104         dev_hold(state->in);
 105         dev_hold(state->out);
 106
 107 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 108         dev_hold(entry->physin);
 109         dev_hold(entry->physout);
 110 #endif
 111         return true;
 112 }
 113 EXPORT_SYMBOL_GPL(nf_queue_entry_get_refs);
 114
 115 void nf_queue_nf_hook_drop(struct net *net)
 116 {
 117         const struct nf_queue_handler *qh;
 118
 119         rcu_read_lock();
 120         qh = rcu_dereference(nf_queue_handler);
 121         if (qh)
 122                 qh->nf_hook_drop(net);
 123         rcu_read_unlock();
 124 }
 125 EXPORT_SYMBOL_GPL(nf_queue_nf_hook_drop);
 126
 127 static void nf_ip_saveroute(const struct sk_buff *skb,
 128                             struct nf_queue_entry *entry)
 129 {
 130         struct ip_rt_info *rt_info = nf_queue_entry_reroute(entry);
 131
 132         if (entry->state.hook == NF_INET_LOCAL_OUT) {
 133                 const struct iphdr *iph = ip_hdr(skb);
 134
 135                 rt_info->tos = iph->tos;
 136                 rt_info->daddr = iph->daddr;
 137                 rt_info->saddr = iph->saddr;
 138                 rt_info->mark = skb->mark;
 139         }
 140 }
 141
 142 static void nf_ip6_saveroute(const struct sk_buff *skb,
 143                              struct nf_queue_entry *entry)
 144 {
 145         struct ip6_rt_info *rt_info = nf_queue_entry_reroute(entry);
 146
 147         if (entry->state.hook == NF_INET_LOCAL_OUT) {
 148                 const struct ipv6hdr *iph = ipv6_hdr(skb);
 149
 150                 rt_info->daddr = iph->daddr;
 151                 rt_info->saddr = iph->saddr;
 152                 rt_info->mark = skb->mark;
 153         }
 154 }
 155
 156 static int __nf_queue(struct sk_buff *skb, const struct nf_hook_state *state,
 157                       unsigned int index, unsigned int queuenum)
 158 {
 159         struct nf_queue_entry *entry = NULL;
 160         const struct nf_queue_handler *qh;
 161         unsigned int route_key_size;
 162         int status;
 163
 164         /* QUEUE == DROP if no one is waiting, to be safe. */
 165         qh = rcu_dereference(nf_queue_handler);
 166         if (!qh)
 167                 return -ESRCH;
 168
 169         switch (state->pf) {
 170         case AF_INET:
 171                 route_key_size = sizeof(struct ip_rt_info);
 172                 break;
 173         case AF_INET6:
 174                 route_key_size = sizeof(struct ip6_rt_info);
 175                 break;
 176         default:
 177                 route_key_size = 0;
 178                 break;
 179         }
 180
 181         if (skb_sk_is_prefetched(skb)) {
 182                 struct sock *sk = skb->sk;
 183
 184                 if (!sk_is_refcounted(sk)) {
 185                         if (!refcount_inc_not_zero(&sk->sk_refcnt))
 186                                 return -ENOTCONN;
 187
 188                         /* drop refcount on skb_orphan */
 189                         skb->destructor = sock_edemux;
 190                 }
 191         }
 192
 193         entry = kmalloc(sizeof(*entry) + route_key_size, GFP_ATOMIC);
 194         if (!entry)
 195                 return -ENOMEM;
 196
 197         if (skb_dst(skb) && !skb_dst_force(skb)) {
 198                 kfree(entry);
 199                 return -ENETDOWN;
 200         }
 201
 202         *entry = (struct nf_queue_entry) {
 203                 .skb    = skb,
 204                 .state  = *state,
 205                 .hook_index = index,
 206                 .size   = sizeof(*entry) + route_key_size,
 207         };
 208
 209         __nf_queue_entry_init_physdevs(entry);
 210
 211         if (!nf_queue_entry_get_refs(entry)) {
 212                 kfree(entry);
 213                 return -ENOTCONN;
 214         }
 215
 216         switch (entry->state.pf) {
 217         case AF_INET:
 218                 nf_ip_saveroute(skb, entry);
 219                 break;
 220         case AF_INET6:
 221                 nf_ip6_saveroute(skb, entry);
 222                 break;
 223         }
 224
 225         status = qh->outfn(entry, queuenum);
 226         if (status < 0) {
 227                 nf_queue_entry_free(entry);
 228                 return status;
 229         }
 230
 231         return 0;
 232 }
 233
 234 /* Packets leaving via this function must come back through nf_reinject(). */
 235 int nf_queue(struct sk_buff *skb, struct nf_hook_state *state,
 236              unsigned int index, unsigned int verdict)
 237 {
 238         int ret;
 239
 240         ret = __nf_queue(skb, state, index, verdict >> NF_VERDICT_QBITS);
 241         if (ret < 0) {
 242                 if (ret == -ESRCH &&
 243                     (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 244                         return 1;
 245                 kfree_skb(skb);
 246         }
 247
 248         return 0;
 249 }
 250 EXPORT_SYMBOL_GPL(nf_queue);
 251
 252 static unsigned int nf_iterate(struct sk_buff *skb,
 253                                struct nf_hook_state *state,
 254                                const struct nf_hook_entries *hooks,
 255                                unsigned int *index)
 256 {
 257         const struct nf_hook_entry *hook;
 258         unsigned int verdict, i = *index;
 259
 260         while (i < hooks->num_hook_entries) {
 261                 hook = &hooks->hooks[i];
 262 repeat:
 263                 verdict = nf_hook_entry_hookfn(hook, skb, state);
 264                 if (verdict != NF_ACCEPT) {
 265                         *index = i;
 266                         if (verdict != NF_REPEAT)
 267                                 return verdict;
 268                         goto repeat;
 269                 }
 270                 i++;
 271         }
 272
 273         *index = i;
 274         return NF_ACCEPT;
 275 }
 276
 277 static struct nf_hook_entries *nf_hook_entries_head(const struct net *net, u8 pf, u8 hooknum)
 278 {
 279         switch (pf) {
 280 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
 281         case NFPROTO_BRIDGE:
 282                 return rcu_dereference(net->nf.hooks_bridge[hooknum]);
 283 #endif
 284         case NFPROTO_IPV4:
 285                 return rcu_dereference(net->nf.hooks_ipv4[hooknum]);
 286         case NFPROTO_IPV6:
 287                 return rcu_dereference(net->nf.hooks_ipv6[hooknum]);
 288         default:
 289                 WARN_ON_ONCE(1);
 290                 return NULL;
 291         }
 292
 293         return NULL;
 294 }
 295
 296 /* Caller must hold rcu read-side lock */
 297 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 298 {
 299         const struct nf_hook_entry *hook_entry;
 300         const struct nf_hook_entries *hooks;
 301         struct sk_buff *skb = entry->skb;
 302         const struct net *net;
 303         unsigned int i;
 304         int err;
 305         u8 pf;
 306
 307         net = entry->state.net;
 308         pf = entry->state.pf;
 309
 310         hooks = nf_hook_entries_head(net, pf, entry->state.hook);
 311
 312         i = entry->hook_index;
 313         if (WARN_ON_ONCE(!hooks || i >= hooks->num_hook_entries)) {
 314                 kfree_skb(skb);
 315                 nf_queue_entry_free(entry);
 316                 return;
 317         }
 318
 319         hook_entry = &hooks->hooks[i];
 320
 321         /* Continue traversal iff userspace said ok... */
 322         if (verdict == NF_REPEAT)
 323                 verdict = nf_hook_entry_hookfn(hook_entry, skb, &entry->state);
 324
 325         if (verdict == NF_ACCEPT) {
 326                 if (nf_reroute(skb, entry) < 0)
 327                         verdict = NF_DROP;
 328         }
 329
 330         if (verdict == NF_ACCEPT) {
 331 next_hook:
 332                 ++i;
 333                 verdict = nf_iterate(skb, &entry->state, hooks, &i);
 334         }
 335
 336         switch (verdict & NF_VERDICT_MASK) {
 337         case NF_ACCEPT:
 338         case NF_STOP:
 339                 local_bh_disable();
 340                 entry->state.okfn(entry->state.net, entry->state.sk, skb);
 341                 local_bh_enable();
 342                 break;
 343         case NF_QUEUE:
 344                 err = nf_queue(skb, &entry->state, i, verdict);
 345                 if (err == 1)
 346                         goto next_hook;
 347                 break;
 348         case NF_STOLEN:
 349                 break;
 350         default:
 351                 kfree_skb(skb);
 352         }
 353
 354         nf_queue_entry_free(entry);
 355 }
 356 EXPORT_SYMBOL(nf_reinject);