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