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 /*
  25  * Hook for nfnetlink_queue to register its queue handler.
  26  * We do this so that most of the NFQUEUE code can be modular.
  27  *
  28  * Once the queue is registered it must reinject all packets it
  29  * receives, no matter what.
  30  */
  31
  32 /* return EBUSY when somebody else is registered, return EEXIST if the
  33  * same handler is registered, return 0 in case of success. */
  34 void nf_register_queue_handler(struct net *net, 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(net->nf.queue_handler));
  38         rcu_assign_pointer(net->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(struct net *net)
  44 {
  45         RCU_INIT_POINTER(net->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 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         if (state->in)
  64                 dev_put(state->in);
  65         if (state->out)
  66                 dev_put(state->out);
  67         if (state->sk)
  68                 nf_queue_sock_put(state->sk);
  69 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  70         if (entry->skb->nf_bridge) {
  71                 struct net_device *physdev;
  72
  73                 physdev = nf_bridge_get_physindev(entry->skb);
  74                 if (physdev)
  75                         dev_put(physdev);
  76                 physdev = nf_bridge_get_physoutdev(entry->skb);
  77                 if (physdev)
  78                         dev_put(physdev);
  79         }
  80 #endif
  81 }
  82 EXPORT_SYMBOL_GPL(nf_queue_entry_release_refs);
  83
  84 /* Bump dev refs so they don't vanish while packet is out */
  85 bool nf_queue_entry_get_refs(struct nf_queue_entry *entry)
  86 {
  87         struct nf_hook_state *state = &entry->state;
  88
  89         if (state->sk && !refcount_inc_not_zero(&state->sk->sk_refcnt))
  90                 return false;
  91
  92         if (state->in)
  93                 dev_hold(state->in);
  94         if (state->out)
  95                 dev_hold(state->out);
  96         if (state->sk)
  97                 sock_hold(state->sk);
  98 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  99         if (entry->skb->nf_bridge) {
 100                 struct net_device *physdev;
 101
 102                 physdev = nf_bridge_get_physindev(entry->skb);
 103                 if (physdev)
 104                         dev_hold(physdev);
 105                 physdev = nf_bridge_get_physoutdev(entry->skb);
 106                 if (physdev)
 107                         dev_hold(physdev);
 108         }
 109 #endif
 110         return true;
 111 }
 112 EXPORT_SYMBOL_GPL(nf_queue_entry_get_refs);
 113
 114 void nf_queue_nf_hook_drop(struct net *net)
 115 {
 116         const struct nf_queue_handler *qh;
 117
 118         rcu_read_lock();
 119         qh = rcu_dereference(net->nf.queue_handler);
 120         if (qh)
 121                 qh->nf_hook_drop(net);
 122         rcu_read_unlock();
 123 }
 124 EXPORT_SYMBOL_GPL(nf_queue_nf_hook_drop);
 125
 126 static void nf_ip_saveroute(const struct sk_buff *skb,
 127                             struct nf_queue_entry *entry)
 128 {
 129         struct ip_rt_info *rt_info = nf_queue_entry_reroute(entry);
 130
 131         if (entry->state.hook == NF_INET_LOCAL_OUT) {
 132                 const struct iphdr *iph = ip_hdr(skb);
 133
 134                 rt_info->tos = iph->tos;
 135                 rt_info->daddr = iph->daddr;
 136                 rt_info->saddr = iph->saddr;
 137                 rt_info->mark = skb->mark;
 138         }
 139 }
 140
 141 static void nf_ip6_saveroute(const struct sk_buff *skb,
 142                              struct nf_queue_entry *entry)
 143 {
 144         struct ip6_rt_info *rt_info = nf_queue_entry_reroute(entry);
 145
 146         if (entry->state.hook == NF_INET_LOCAL_OUT) {
 147                 const struct ipv6hdr *iph = ipv6_hdr(skb);
 148
 149                 rt_info->daddr = iph->daddr;
 150                 rt_info->saddr = iph->saddr;
 151                 rt_info->mark = skb->mark;
 152         }
 153 }
 154
 155 static int __nf_queue(struct sk_buff *skb, const struct nf_hook_state *state,
 156                       const struct nf_hook_entries *entries,
 157                       unsigned int index, unsigned int queuenum)
 158 {
 159         int status = -ENOENT;
 160         struct nf_queue_entry *entry = NULL;
 161         const struct nf_queue_handler *qh;
 162         struct net *net = state->net;
 163         unsigned int route_key_size;
 164
 165         /* QUEUE == DROP if no one is waiting, to be safe. */
 166         qh = rcu_dereference(net->nf.queue_handler);
 167         if (!qh) {
 168                 status = -ESRCH;
 169                 goto err;
 170         }
 171
 172         switch (state->pf) {
 173         case AF_INET:
 174                 route_key_size = sizeof(struct ip_rt_info);
 175                 break;
 176         case AF_INET6:
 177                 route_key_size = sizeof(struct ip6_rt_info);
 178                 break;
 179         default:
 180                 route_key_size = 0;
 181                 break;
 182         }
 183
 184         entry = kmalloc(sizeof(*entry) + route_key_size, GFP_ATOMIC);
 185         if (!entry) {
 186                 status = -ENOMEM;
 187                 goto err;
 188         }
 189
 190         if (skb_dst(skb) && !skb_dst_force(skb)) {
 191                 status = -ENETDOWN;
 192                 goto err;
 193         }
 194
 195         *entry = (struct nf_queue_entry) {
 196                 .skb    = skb,
 197                 .state  = *state,
 198                 .hook_index = index,
 199                 .size   = sizeof(*entry) + route_key_size,
 200         };
 201
 202         if (!nf_queue_entry_get_refs(entry)) {
 203                 kfree(entry);
 204                 return -ENOTCONN;
 205         }
 206
 207         switch (entry->state.pf) {
 208         case AF_INET:
 209                 nf_ip_saveroute(skb, entry);
 210                 break;
 211         case AF_INET6:
 212                 nf_ip6_saveroute(skb, entry);
 213                 break;
 214         }
 215
 216         status = qh->outfn(entry, queuenum);
 217
 218         if (status < 0) {
 219                 nf_queue_entry_release_refs(entry);
 220                 goto err;
 221         }
 222
 223         return 0;
 224
 225 err:
 226         kfree(entry);
 227         return status;
 228 }
 229
 230 /* Packets leaving via this function must come back through nf_reinject(). */
 231 int nf_queue(struct sk_buff *skb, struct nf_hook_state *state,
 232              const struct nf_hook_entries *entries, unsigned int index,
 233              unsigned int verdict)
 234 {
 235         int ret;
 236
 237         ret = __nf_queue(skb, state, entries, index, verdict >> NF_VERDICT_QBITS);
 238         if (ret < 0) {
 239                 if (ret == -ESRCH &&
 240                     (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 241                         return 1;
 242                 kfree_skb(skb);
 243         }
 244
 245         return 0;
 246 }
 247
 248 static unsigned int nf_iterate(struct sk_buff *skb,
 249                                struct nf_hook_state *state,
 250                                const struct nf_hook_entries *hooks,
 251                                unsigned int *index)
 252 {
 253         const struct nf_hook_entry *hook;
 254         unsigned int verdict, i = *index;
 255
 256         while (i < hooks->num_hook_entries) {
 257                 hook = &hooks->hooks[i];
 258 repeat:
 259                 verdict = nf_hook_entry_hookfn(hook, skb, state);
 260                 if (verdict != NF_ACCEPT) {
 261                         *index = i;
 262                         if (verdict != NF_REPEAT)
 263                                 return verdict;
 264                         goto repeat;
 265                 }
 266                 i++;
 267         }
 268
 269         *index = i;
 270         return NF_ACCEPT;
 271 }
 272
 273 static struct nf_hook_entries *nf_hook_entries_head(const struct net *net, u8 pf, u8 hooknum)
 274 {
 275         switch (pf) {
 276 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
 277         case NFPROTO_BRIDGE:
 278                 return rcu_dereference(net->nf.hooks_bridge[hooknum]);
 279 #endif
 280         case NFPROTO_IPV4:
 281                 return rcu_dereference(net->nf.hooks_ipv4[hooknum]);
 282         case NFPROTO_IPV6:
 283                 return rcu_dereference(net->nf.hooks_ipv6[hooknum]);
 284         default:
 285                 WARN_ON_ONCE(1);
 286                 return NULL;
 287         }
 288
 289         return NULL;
 290 }
 291
 292 /* Caller must hold rcu read-side lock */
 293 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 294 {
 295         const struct nf_hook_entry *hook_entry;
 296         const struct nf_hook_entries *hooks;
 297         struct sk_buff *skb = entry->skb;
 298         const struct net *net;
 299         unsigned int i;
 300         int err;
 301         u8 pf;
 302
 303         net = entry->state.net;
 304         pf = entry->state.pf;
 305
 306         hooks = nf_hook_entries_head(net, pf, entry->state.hook);
 307
 308         nf_queue_entry_release_refs(entry);
 309
 310         i = entry->hook_index;
 311         if (WARN_ON_ONCE(!hooks || i >= hooks->num_hook_entries)) {
 312                 kfree_skb(skb);
 313                 kfree(entry);
 314                 return;
 315         }
 316
 317         hook_entry = &hooks->hooks[i];
 318
 319         /* Continue traversal iff userspace said ok... */
 320         if (verdict == NF_REPEAT)
 321                 verdict = nf_hook_entry_hookfn(hook_entry, skb, &entry->state);
 322
 323         if (verdict == NF_ACCEPT) {
 324                 if (nf_reroute(skb, entry) < 0)
 325                         verdict = NF_DROP;
 326         }
 327
 328         if (verdict == NF_ACCEPT) {
 329 next_hook:
 330                 ++i;
 331                 verdict = nf_iterate(skb, &entry->state, hooks, &i);
 332         }
 333
 334         switch (verdict & NF_VERDICT_MASK) {
 335         case NF_ACCEPT:
 336         case NF_STOP:
 337                 local_bh_disable();
 338                 entry->state.okfn(entry->state.net, entry->state.sk, skb);
 339                 local_bh_enable();
 340                 break;
 341         case NF_QUEUE:
 342                 err = nf_queue(skb, &entry->state, hooks, i, verdict);
 343                 if (err == 1)
 344                         goto next_hook;
 345                 break;
 346         case NF_STOLEN:
 347                 break;
 348         default:
 349                 kfree_skb(skb);
 350         }
 351
 352         kfree(entry);
 353 }
 354 EXPORT_SYMBOL(nf_reinject);