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 void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
  50 {
  51         struct nf_hook_state *state = &entry->state;
  52
  53         /* Release those devices we held, or Alexey will kill me. */
  54         if (state->in)
  55                 dev_put(state->in);
  56         if (state->out)
  57                 dev_put(state->out);
  58         if (state->sk)
  59                 sock_put(state->sk);
  60 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  61         if (entry->skb->nf_bridge) {
  62                 struct net_device *physdev;
  63
  64                 physdev = nf_bridge_get_physindev(entry->skb);
  65                 if (physdev)
  66                         dev_put(physdev);
  67                 physdev = nf_bridge_get_physoutdev(entry->skb);
  68                 if (physdev)
  69                         dev_put(physdev);
  70         }
  71 #endif
  72 }
  73 EXPORT_SYMBOL_GPL(nf_queue_entry_release_refs);
  74
  75 /* Bump dev refs so they don't vanish while packet is out */
  76 void nf_queue_entry_get_refs(struct nf_queue_entry *entry)
  77 {
  78         struct nf_hook_state *state = &entry->state;
  79
  80         if (state->in)
  81                 dev_hold(state->in);
  82         if (state->out)
  83                 dev_hold(state->out);
  84         if (state->sk)
  85                 sock_hold(state->sk);
  86 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  87         if (entry->skb->nf_bridge) {
  88                 struct net_device *physdev;
  89
  90                 physdev = nf_bridge_get_physindev(entry->skb);
  91                 if (physdev)
  92                         dev_hold(physdev);
  93                 physdev = nf_bridge_get_physoutdev(entry->skb);
  94                 if (physdev)
  95                         dev_hold(physdev);
  96         }
  97 #endif
  98 }
  99 EXPORT_SYMBOL_GPL(nf_queue_entry_get_refs);
 100
 101 void nf_queue_nf_hook_drop(struct net *net)
 102 {
 103         const struct nf_queue_handler *qh;
 104
 105         rcu_read_lock();
 106         qh = rcu_dereference(net->nf.queue_handler);
 107         if (qh)
 108                 qh->nf_hook_drop(net);
 109         rcu_read_unlock();
 110 }
 111 EXPORT_SYMBOL_GPL(nf_queue_nf_hook_drop);
 112
 113 static void nf_ip_saveroute(const struct sk_buff *skb,
 114                             struct nf_queue_entry *entry)
 115 {
 116         struct ip_rt_info *rt_info = nf_queue_entry_reroute(entry);
 117
 118         if (entry->state.hook == NF_INET_LOCAL_OUT) {
 119                 const struct iphdr *iph = ip_hdr(skb);
 120
 121                 rt_info->tos = iph->tos;
 122                 rt_info->daddr = iph->daddr;
 123                 rt_info->saddr = iph->saddr;
 124                 rt_info->mark = skb->mark;
 125         }
 126 }
 127
 128 static void nf_ip6_saveroute(const struct sk_buff *skb,
 129                              struct nf_queue_entry *entry)
 130 {
 131         struct ip6_rt_info *rt_info = nf_queue_entry_reroute(entry);
 132
 133         if (entry->state.hook == NF_INET_LOCAL_OUT) {
 134                 const struct ipv6hdr *iph = ipv6_hdr(skb);
 135
 136                 rt_info->daddr = iph->daddr;
 137                 rt_info->saddr = iph->saddr;
 138                 rt_info->mark = skb->mark;
 139         }
 140 }
 141
 142 static int __nf_queue(struct sk_buff *skb, const struct nf_hook_state *state,
 143                       const struct nf_hook_entries *entries,
 144                       unsigned int index, unsigned int queuenum)
 145 {
 146         int status = -ENOENT;
 147         struct nf_queue_entry *entry = NULL;
 148         const struct nf_queue_handler *qh;
 149         struct net *net = state->net;
 150         unsigned int route_key_size;
 151
 152         /* QUEUE == DROP if no one is waiting, to be safe. */
 153         qh = rcu_dereference(net->nf.queue_handler);
 154         if (!qh) {
 155                 status = -ESRCH;
 156                 goto err;
 157         }
 158
 159         switch (state->pf) {
 160         case AF_INET:
 161                 route_key_size = sizeof(struct ip_rt_info);
 162                 break;
 163         case AF_INET6:
 164                 route_key_size = sizeof(struct ip6_rt_info);
 165                 break;
 166         default:
 167                 route_key_size = 0;
 168                 break;
 169         }
 170
 171         entry = kmalloc(sizeof(*entry) + route_key_size, GFP_ATOMIC);
 172         if (!entry) {
 173                 status = -ENOMEM;
 174                 goto err;
 175         }
 176
 177         if (skb_dst(skb) && !skb_dst_force(skb)) {
 178                 status = -ENETDOWN;
 179                 goto err;
 180         }
 181
 182         *entry = (struct nf_queue_entry) {
 183                 .skb    = skb,
 184                 .state  = *state,
 185                 .hook_index = index,
 186                 .size   = sizeof(*entry) + route_key_size,
 187         };
 188
 189         nf_queue_entry_get_refs(entry);
 190
 191         switch (entry->state.pf) {
 192         case AF_INET:
 193                 nf_ip_saveroute(skb, entry);
 194                 break;
 195         case AF_INET6:
 196                 nf_ip6_saveroute(skb, entry);
 197                 break;
 198         }
 199
 200         status = qh->outfn(entry, queuenum);
 201
 202         if (status < 0) {
 203                 nf_queue_entry_release_refs(entry);
 204                 goto err;
 205         }
 206
 207         return 0;
 208
 209 err:
 210         kfree(entry);
 211         return status;
 212 }
 213
 214 /* Packets leaving via this function must come back through nf_reinject(). */
 215 int nf_queue(struct sk_buff *skb, struct nf_hook_state *state,
 216              const struct nf_hook_entries *entries, unsigned int index,
 217              unsigned int verdict)
 218 {
 219         int ret;
 220
 221         ret = __nf_queue(skb, state, entries, index, verdict >> NF_VERDICT_QBITS);
 222         if (ret < 0) {
 223                 if (ret == -ESRCH &&
 224                     (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 225                         return 1;
 226                 kfree_skb(skb);
 227         }
 228
 229         return 0;
 230 }
 231
 232 static unsigned int nf_iterate(struct sk_buff *skb,
 233                                struct nf_hook_state *state,
 234                                const struct nf_hook_entries *hooks,
 235                                unsigned int *index)
 236 {
 237         const struct nf_hook_entry *hook;
 238         unsigned int verdict, i = *index;
 239
 240         while (i < hooks->num_hook_entries) {
 241                 hook = &hooks->hooks[i];
 242 repeat:
 243                 verdict = nf_hook_entry_hookfn(hook, skb, state);
 244                 if (verdict != NF_ACCEPT) {
 245                         *index = i;
 246                         if (verdict != NF_REPEAT)
 247                                 return verdict;
 248                         goto repeat;
 249                 }
 250                 i++;
 251         }
 252
 253         *index = i;
 254         return NF_ACCEPT;
 255 }
 256
 257 static struct nf_hook_entries *nf_hook_entries_head(const struct net *net, u8 pf, u8 hooknum)
 258 {
 259         switch (pf) {
 260 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
 261         case NFPROTO_BRIDGE:
 262                 return rcu_dereference(net->nf.hooks_bridge[hooknum]);
 263 #endif
 264         case NFPROTO_IPV4:
 265                 return rcu_dereference(net->nf.hooks_ipv4[hooknum]);
 266         case NFPROTO_IPV6:
 267                 return rcu_dereference(net->nf.hooks_ipv6[hooknum]);
 268         default:
 269                 WARN_ON_ONCE(1);
 270                 return NULL;
 271         }
 272
 273         return NULL;
 274 }
 275
 276 /* Caller must hold rcu read-side lock */
 277 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 278 {
 279         const struct nf_hook_entry *hook_entry;
 280         const struct nf_hook_entries *hooks;
 281         struct sk_buff *skb = entry->skb;
 282         const struct net *net;
 283         unsigned int i;
 284         int err;
 285         u8 pf;
 286
 287         net = entry->state.net;
 288         pf = entry->state.pf;
 289
 290         hooks = nf_hook_entries_head(net, pf, entry->state.hook);
 291
 292         nf_queue_entry_release_refs(entry);
 293
 294         i = entry->hook_index;
 295         if (WARN_ON_ONCE(!hooks || i >= hooks->num_hook_entries)) {
 296                 kfree_skb(skb);
 297                 kfree(entry);
 298                 return;
 299         }
 300
 301         hook_entry = &hooks->hooks[i];
 302
 303         /* Continue traversal iff userspace said ok... */
 304         if (verdict == NF_REPEAT)
 305                 verdict = nf_hook_entry_hookfn(hook_entry, skb, &entry->state);
 306
 307         if (verdict == NF_ACCEPT) {
 308                 if (nf_reroute(skb, entry) < 0)
 309                         verdict = NF_DROP;
 310         }
 311
 312         if (verdict == NF_ACCEPT) {
 313 next_hook:
 314                 ++i;
 315                 verdict = nf_iterate(skb, &entry->state, hooks, &i);
 316         }
 317
 318         switch (verdict & NF_VERDICT_MASK) {
 319         case NF_ACCEPT:
 320         case NF_STOP:
 321                 local_bh_disable();
 322                 entry->state.okfn(entry->state.net, entry->state.sk, skb);
 323                 local_bh_enable();
 324                 break;
 325         case NF_QUEUE:
 326                 err = nf_queue(skb, &entry->state, hooks, i, verdict);
 327                 if (err == 1)
 328                         goto next_hook;
 329                 break;
 330         case NF_STOLEN:
 331                 break;
 332         default:
 333                 kfree_skb(skb);
 334         }
 335
 336         kfree(entry);
 337 }
 338 EXPORT_SYMBOL(nf_reinject);