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_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
  61 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  62         if (entry->physin)
  63                 dev_put(entry->physin);
  64         if (entry->physout)
  65                 dev_put(entry->physout);
  66 #endif
  67 }
  68
  69 void nf_queue_entry_free(struct nf_queue_entry *entry)
  70 {
  71         nf_queue_entry_release_refs(entry);
  72         kfree(entry);
  73 }
  74 EXPORT_SYMBOL_GPL(nf_queue_entry_free);
  75
  76 static void __nf_queue_entry_init_physdevs(struct nf_queue_entry *entry)
  77 {
  78 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
  79         const struct sk_buff *skb = entry->skb;
  80         struct nf_bridge_info *nf_bridge;
  81
  82         nf_bridge = nf_bridge_info_get(skb);
  83         if (nf_bridge) {
  84                 entry->physin = nf_bridge_get_physindev(skb);
  85                 entry->physout = nf_bridge_get_physoutdev(skb);
  86         } else {
  87                 entry->physin = NULL;
  88                 entry->physout = NULL;
  89         }
  90 #endif
  91 }
  92
  93 /* Bump dev refs so they don't vanish while packet is out */
  94 void nf_queue_entry_get_refs(struct nf_queue_entry *entry)
  95 {
  96         struct nf_hook_state *state = &entry->state;
  97
  98         if (state->in)
  99                 dev_hold(state->in);
 100         if (state->out)
 101                 dev_hold(state->out);
 102         if (state->sk)
 103                 sock_hold(state->sk);
 104
 105 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 106         if (entry->physin)
 107                 dev_hold(entry->physin);
 108         if (entry->physout)
 109                 dev_hold(entry->physout);
 110 #endif
 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                       unsigned int index, unsigned int queuenum)
 157 {
 158         struct nf_queue_entry *entry = NULL;
 159         const struct nf_queue_handler *qh;
 160         struct net *net = state->net;
 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(net->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         entry = kmalloc(sizeof(*entry) + route_key_size, GFP_ATOMIC);
 182         if (!entry)
 183                 return -ENOMEM;
 184
 185         if (skb_dst(skb) && !skb_dst_force(skb)) {
 186                 kfree(entry);
 187                 return -ENETDOWN;
 188         }
 189
 190         *entry = (struct nf_queue_entry) {
 191                 .skb    = skb,
 192                 .state  = *state,
 193                 .hook_index = index,
 194                 .size   = sizeof(*entry) + route_key_size,
 195         };
 196
 197         __nf_queue_entry_init_physdevs(entry);
 198
 199         nf_queue_entry_get_refs(entry);
 200
 201         switch (entry->state.pf) {
 202         case AF_INET:
 203                 nf_ip_saveroute(skb, entry);
 204                 break;
 205         case AF_INET6:
 206                 nf_ip6_saveroute(skb, entry);
 207                 break;
 208         }
 209
 210         status = qh->outfn(entry, queuenum);
 211         if (status < 0) {
 212                 nf_queue_entry_free(entry);
 213                 return status;
 214         }
 215
 216         return 0;
 217 }
 218
 219 /* Packets leaving via this function must come back through nf_reinject(). */
 220 int nf_queue(struct sk_buff *skb, struct nf_hook_state *state,
 221              unsigned int index, unsigned int verdict)
 222 {
 223         int ret;
 224
 225         ret = __nf_queue(skb, state, index, verdict >> NF_VERDICT_QBITS);
 226         if (ret < 0) {
 227                 if (ret == -ESRCH &&
 228                     (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 229                         return 1;
 230                 kfree_skb(skb);
 231         }
 232
 233         return 0;
 234 }
 235 EXPORT_SYMBOL_GPL(nf_queue);
 236
 237 static unsigned int nf_iterate(struct sk_buff *skb,
 238                                struct nf_hook_state *state,
 239                                const struct nf_hook_entries *hooks,
 240                                unsigned int *index)
 241 {
 242         const struct nf_hook_entry *hook;
 243         unsigned int verdict, i = *index;
 244
 245         while (i < hooks->num_hook_entries) {
 246                 hook = &hooks->hooks[i];
 247 repeat:
 248                 verdict = nf_hook_entry_hookfn(hook, skb, state);
 249                 if (verdict != NF_ACCEPT) {
 250                         *index = i;
 251                         if (verdict != NF_REPEAT)
 252                                 return verdict;
 253                         goto repeat;
 254                 }
 255                 i++;
 256         }
 257
 258         *index = i;
 259         return NF_ACCEPT;
 260 }
 261
 262 static struct nf_hook_entries *nf_hook_entries_head(const struct net *net, u8 pf, u8 hooknum)
 263 {
 264         switch (pf) {
 265 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
 266         case NFPROTO_BRIDGE:
 267                 return rcu_dereference(net->nf.hooks_bridge[hooknum]);
 268 #endif
 269         case NFPROTO_IPV4:
 270                 return rcu_dereference(net->nf.hooks_ipv4[hooknum]);
 271         case NFPROTO_IPV6:
 272                 return rcu_dereference(net->nf.hooks_ipv6[hooknum]);
 273         default:
 274                 WARN_ON_ONCE(1);
 275                 return NULL;
 276         }
 277
 278         return NULL;
 279 }
 280
 281 /* Caller must hold rcu read-side lock */
 282 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 283 {
 284         const struct nf_hook_entry *hook_entry;
 285         const struct nf_hook_entries *hooks;
 286         struct sk_buff *skb = entry->skb;
 287         const struct net *net;
 288         unsigned int i;
 289         int err;
 290         u8 pf;
 291
 292         net = entry->state.net;
 293         pf = entry->state.pf;
 294
 295         hooks = nf_hook_entries_head(net, pf, entry->state.hook);
 296
 297         i = entry->hook_index;
 298         if (WARN_ON_ONCE(!hooks || i >= hooks->num_hook_entries)) {
 299                 kfree_skb(skb);
 300                 nf_queue_entry_free(entry);
 301                 return;
 302         }
 303
 304         hook_entry = &hooks->hooks[i];
 305
 306         /* Continue traversal iff userspace said ok... */
 307         if (verdict == NF_REPEAT)
 308                 verdict = nf_hook_entry_hookfn(hook_entry, skb, &entry->state);
 309
 310         if (verdict == NF_ACCEPT) {
 311                 if (nf_reroute(skb, entry) < 0)
 312                         verdict = NF_DROP;
 313         }
 314
 315         if (verdict == NF_ACCEPT) {
 316 next_hook:
 317                 ++i;
 318                 verdict = nf_iterate(skb, &entry->state, hooks, &i);
 319         }
 320
 321         switch (verdict & NF_VERDICT_MASK) {
 322         case NF_ACCEPT:
 323         case NF_STOP:
 324                 local_bh_disable();
 325                 entry->state.okfn(entry->state.net, entry->state.sk, skb);
 326                 local_bh_enable();
 327                 break;
 328         case NF_QUEUE:
 329                 err = nf_queue(skb, &entry->state, i, verdict);
 330                 if (err == 1)
 331                         goto next_hook;
 332                 break;
 333         case NF_STOLEN:
 334                 break;
 335         default:
 336                 kfree_skb(skb);
 337         }
 338
 339         nf_queue_entry_free(entry);
 340 }
 341 EXPORT_SYMBOL(nf_reinject);