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_entry_release_br_nf_refs(struct sk_buff *skb)
  50 {
  51 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  52         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
  53
  54         if (nf_bridge) {
  55                 struct net_device *physdev;
  56
  57                 physdev = nf_bridge_get_physindev(skb);
  58                 if (physdev)
  59                         dev_put(physdev);
  60                 physdev = nf_bridge_get_physoutdev(skb);
  61                 if (physdev)
  62                         dev_put(physdev);
  63         }
  64 #endif
  65 }
  66
  67 void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
  68 {
  69         struct nf_hook_state *state = &entry->state;
  70
  71         /* Release those devices we held, or Alexey will kill me. */
  72         if (state->in)
  73                 dev_put(state->in);
  74         if (state->out)
  75                 dev_put(state->out);
  76         if (state->sk)
  77                 sock_put(state->sk);
  78
  79         nf_queue_entry_release_br_nf_refs(entry->skb);
  80 }
  81 EXPORT_SYMBOL_GPL(nf_queue_entry_release_refs);
  82
  83 static void nf_queue_entry_get_br_nf_refs(struct sk_buff *skb)
  84 {
  85 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  86         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
  87
  88         if (nf_bridge) {
  89                 struct net_device *physdev;
  90
  91                 physdev = nf_bridge_get_physindev(skb);
  92                 if (physdev)
  93                         dev_hold(physdev);
  94                 physdev = nf_bridge_get_physoutdev(skb);
  95                 if (physdev)
  96                         dev_hold(physdev);
  97         }
  98 #endif
  99 }
 100
 101 /* Bump dev refs so they don't vanish while packet is out */
 102 void nf_queue_entry_get_refs(struct nf_queue_entry *entry)
 103 {
 104         struct nf_hook_state *state = &entry->state;
 105
 106         if (state->in)
 107                 dev_hold(state->in);
 108         if (state->out)
 109                 dev_hold(state->out);
 110         if (state->sk)
 111                 sock_hold(state->sk);
 112
 113         nf_queue_entry_get_br_nf_refs(entry->skb);
 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(net->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                       const struct nf_hook_entries *entries,
 160                       unsigned int index, unsigned int queuenum)
 161 {
 162         int status = -ENOENT;
 163         struct nf_queue_entry *entry = NULL;
 164         const struct nf_queue_handler *qh;
 165         struct net *net = state->net;
 166         unsigned int route_key_size;
 167
 168         /* QUEUE == DROP if no one is waiting, to be safe. */
 169         qh = rcu_dereference(net->nf.queue_handler);
 170         if (!qh) {
 171                 status = -ESRCH;
 172                 goto err;
 173         }
 174
 175         switch (state->pf) {
 176         case AF_INET:
 177                 route_key_size = sizeof(struct ip_rt_info);
 178                 break;
 179         case AF_INET6:
 180                 route_key_size = sizeof(struct ip6_rt_info);
 181                 break;
 182         default:
 183                 route_key_size = 0;
 184                 break;
 185         }
 186
 187         entry = kmalloc(sizeof(*entry) + route_key_size, GFP_ATOMIC);
 188         if (!entry) {
 189                 status = -ENOMEM;
 190                 goto err;
 191         }
 192
 193         *entry = (struct nf_queue_entry) {
 194                 .skb    = skb,
 195                 .state  = *state,
 196                 .hook_index = index,
 197                 .size   = sizeof(*entry) + route_key_size,
 198         };
 199
 200         nf_queue_entry_get_refs(entry);
 201         skb_dst_force(skb);
 202
 203         switch (entry->state.pf) {
 204         case AF_INET:
 205                 nf_ip_saveroute(skb, entry);
 206                 break;
 207         case AF_INET6:
 208                 nf_ip6_saveroute(skb, entry);
 209                 break;
 210         }
 211
 212         status = qh->outfn(entry, queuenum);
 213
 214         if (status < 0) {
 215                 nf_queue_entry_release_refs(entry);
 216                 goto err;
 217         }
 218
 219         return 0;
 220
 221 err:
 222         kfree(entry);
 223         return status;
 224 }
 225
 226 /* Packets leaving via this function must come back through nf_reinject(). */
 227 int nf_queue(struct sk_buff *skb, struct nf_hook_state *state,
 228              const struct nf_hook_entries *entries, unsigned int index,
 229              unsigned int verdict)
 230 {
 231         int ret;
 232
 233         ret = __nf_queue(skb, state, entries, index, verdict >> NF_VERDICT_QBITS);
 234         if (ret < 0) {
 235                 if (ret == -ESRCH &&
 236                     (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 237                         return 1;
 238                 kfree_skb(skb);
 239         }
 240
 241         return 0;
 242 }
 243 EXPORT_SYMBOL_GPL(nf_queue);
 244
 245 static unsigned int nf_iterate(struct sk_buff *skb,
 246                                struct nf_hook_state *state,
 247                                const struct nf_hook_entries *hooks,
 248                                unsigned int *index)
 249 {
 250         const struct nf_hook_entry *hook;
 251         unsigned int verdict, i = *index;
 252
 253         while (i < hooks->num_hook_entries) {
 254                 hook = &hooks->hooks[i];
 255 repeat:
 256                 verdict = nf_hook_entry_hookfn(hook, skb, state);
 257                 if (verdict != NF_ACCEPT) {
 258                         *index = i;
 259                         if (verdict != NF_REPEAT)
 260                                 return verdict;
 261                         goto repeat;
 262                 }
 263                 i++;
 264         }
 265
 266         *index = i;
 267         return NF_ACCEPT;
 268 }
 269
 270 static struct nf_hook_entries *nf_hook_entries_head(const struct net *net, u8 pf, u8 hooknum)
 271 {
 272         switch (pf) {
 273 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
 274         case NFPROTO_BRIDGE:
 275                 return rcu_dereference(net->nf.hooks_bridge[hooknum]);
 276 #endif
 277         case NFPROTO_IPV4:
 278                 return rcu_dereference(net->nf.hooks_ipv4[hooknum]);
 279         case NFPROTO_IPV6:
 280                 return rcu_dereference(net->nf.hooks_ipv6[hooknum]);
 281         default:
 282                 WARN_ON_ONCE(1);
 283                 return NULL;
 284         }
 285
 286         return NULL;
 287 }
 288
 289 /* Caller must hold rcu read-side lock */
 290 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 291 {
 292         const struct nf_hook_entry *hook_entry;
 293         const struct nf_hook_entries *hooks;
 294         struct sk_buff *skb = entry->skb;
 295         const struct net *net;
 296         unsigned int i;
 297         int err;
 298         u8 pf;
 299
 300         net = entry->state.net;
 301         pf = entry->state.pf;
 302
 303         hooks = nf_hook_entries_head(net, pf, entry->state.hook);
 304
 305         nf_queue_entry_release_refs(entry);
 306
 307         i = entry->hook_index;
 308         if (WARN_ON_ONCE(!hooks || i >= hooks->num_hook_entries)) {
 309                 kfree_skb(skb);
 310                 kfree(entry);
 311                 return;
 312         }
 313
 314         hook_entry = &hooks->hooks[i];
 315
 316         /* Continue traversal iff userspace said ok... */
 317         if (verdict == NF_REPEAT)
 318                 verdict = nf_hook_entry_hookfn(hook_entry, skb, &entry->state);
 319
 320         if (verdict == NF_ACCEPT) {
 321                 if (nf_reroute(skb, entry) < 0)
 322                         verdict = NF_DROP;
 323         }
 324
 325         if (verdict == NF_ACCEPT) {
 326 next_hook:
 327                 ++i;
 328                 verdict = nf_iterate(skb, &entry->state, hooks, &i);
 329         }
 330
 331         switch (verdict & NF_VERDICT_MASK) {
 332         case NF_ACCEPT:
 333         case NF_STOP:
 334                 local_bh_disable();
 335                 entry->state.okfn(entry->state.net, entry->state.sk, skb);
 336                 local_bh_enable();
 337                 break;
 338         case NF_QUEUE:
 339                 err = nf_queue(skb, &entry->state, hooks, i, verdict);
 340                 if (err == 1)
 341                         goto next_hook;
 342                 break;
 343         case NF_STOLEN:
 344                 break;
 345         default:
 346                 kfree_skb(skb);
 347         }
 348
 349         kfree(entry);
 350 }
 351 EXPORT_SYMBOL(nf_reinject);