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                       unsigned int index, unsigned int queuenum)
 160 {
 161         int status = -ENOENT;
 162         struct nf_queue_entry *entry = NULL;
 163         const struct nf_queue_handler *qh;
 164         struct net *net = state->net;
 165         unsigned int route_key_size;
 166
 167         /* QUEUE == DROP if no one is waiting, to be safe. */
 168         qh = rcu_dereference(net->nf.queue_handler);
 169         if (!qh) {
 170                 status = -ESRCH;
 171                 goto err;
 172         }
 173
 174         switch (state->pf) {
 175         case AF_INET:
 176                 route_key_size = sizeof(struct ip_rt_info);
 177                 break;
 178         case AF_INET6:
 179                 route_key_size = sizeof(struct ip6_rt_info);
 180                 break;
 181         default:
 182                 route_key_size = 0;
 183                 break;
 184         }
 185
 186         entry = kmalloc(sizeof(*entry) + route_key_size, GFP_ATOMIC);
 187         if (!entry) {
 188                 status = -ENOMEM;
 189                 goto err;
 190         }
 191
 192         if (!skb_dst_force(skb) && state->hook != NF_INET_PRE_ROUTING) {
 193                 status = -ENETDOWN;
 194                 goto err;
 195         }
 196
 197         *entry = (struct nf_queue_entry) {
 198                 .skb    = skb,
 199                 .state  = *state,
 200                 .hook_index = index,
 201                 .size   = sizeof(*entry) + route_key_size,
 202         };
 203
 204         nf_queue_entry_get_refs(entry);
 205
 206         switch (entry->state.pf) {
 207         case AF_INET:
 208                 nf_ip_saveroute(skb, entry);
 209                 break;
 210         case AF_INET6:
 211                 nf_ip6_saveroute(skb, entry);
 212                 break;
 213         }
 214
 215         status = qh->outfn(entry, queuenum);
 216
 217         if (status < 0) {
 218                 nf_queue_entry_release_refs(entry);
 219                 goto err;
 220         }
 221
 222         return 0;
 223
 224 err:
 225         kfree(entry);
 226         return status;
 227 }
 228
 229 /* Packets leaving via this function must come back through nf_reinject(). */
 230 int nf_queue(struct sk_buff *skb, struct nf_hook_state *state,
 231              unsigned int index, unsigned int verdict)
 232 {
 233         int ret;
 234
 235         ret = __nf_queue(skb, state, index, verdict >> NF_VERDICT_QBITS);
 236         if (ret < 0) {
 237                 if (ret == -ESRCH &&
 238                     (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 239                         return 1;
 240                 kfree_skb(skb);
 241         }
 242
 243         return 0;
 244 }
 245 EXPORT_SYMBOL_GPL(nf_queue);
 246
 247 static unsigned int nf_iterate(struct sk_buff *skb,
 248                                struct nf_hook_state *state,
 249                                const struct nf_hook_entries *hooks,
 250                                unsigned int *index)
 251 {
 252         const struct nf_hook_entry *hook;
 253         unsigned int verdict, i = *index;
 254
 255         while (i < hooks->num_hook_entries) {
 256                 hook = &hooks->hooks[i];
 257 repeat:
 258                 verdict = nf_hook_entry_hookfn(hook, skb, state);
 259                 if (verdict != NF_ACCEPT) {
 260                         *index = i;
 261                         if (verdict != NF_REPEAT)
 262                                 return verdict;
 263                         goto repeat;
 264                 }
 265                 i++;
 266         }
 267
 268         *index = i;
 269         return NF_ACCEPT;
 270 }
 271
 272 static struct nf_hook_entries *nf_hook_entries_head(const struct net *net, u8 pf, u8 hooknum)
 273 {
 274         switch (pf) {
 275 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
 276         case NFPROTO_BRIDGE:
 277                 return rcu_dereference(net->nf.hooks_bridge[hooknum]);
 278 #endif
 279         case NFPROTO_IPV4:
 280                 return rcu_dereference(net->nf.hooks_ipv4[hooknum]);
 281         case NFPROTO_IPV6:
 282                 return rcu_dereference(net->nf.hooks_ipv6[hooknum]);
 283         default:
 284                 WARN_ON_ONCE(1);
 285                 return NULL;
 286         }
 287
 288         return NULL;
 289 }
 290
 291 /* Caller must hold rcu read-side lock */
 292 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 293 {
 294         const struct nf_hook_entry *hook_entry;
 295         const struct nf_hook_entries *hooks;
 296         struct sk_buff *skb = entry->skb;
 297         const struct net *net;
 298         unsigned int i;
 299         int err;
 300         u8 pf;
 301
 302         net = entry->state.net;
 303         pf = entry->state.pf;
 304
 305         hooks = nf_hook_entries_head(net, pf, entry->state.hook);
 306
 307         nf_queue_entry_release_refs(entry);
 308
 309         i = entry->hook_index;
 310         if (WARN_ON_ONCE(!hooks || i >= hooks->num_hook_entries)) {
 311                 kfree_skb(skb);
 312                 kfree(entry);
 313                 return;
 314         }
 315
 316         hook_entry = &hooks->hooks[i];
 317
 318         /* Continue traversal iff userspace said ok... */
 319         if (verdict == NF_REPEAT)
 320                 verdict = nf_hook_entry_hookfn(hook_entry, skb, &entry->state);
 321
 322         if (verdict == NF_ACCEPT) {
 323                 if (nf_reroute(skb, entry) < 0)
 324                         verdict = NF_DROP;
 325         }
 326
 327         if (verdict == NF_ACCEPT) {
 328 next_hook:
 329                 ++i;
 330                 verdict = nf_iterate(skb, &entry->state, hooks, &i);
 331         }
 332
 333         switch (verdict & NF_VERDICT_MASK) {
 334         case NF_ACCEPT:
 335         case NF_STOP:
 336                 local_bh_disable();
 337                 entry->state.okfn(entry->state.net, entry->state.sk, skb);
 338                 local_bh_enable();
 339                 break;
 340         case NF_QUEUE:
 341                 err = nf_queue(skb, &entry->state, i, verdict);
 342                 if (err == 1)
 343                         goto next_hook;
 344                 break;
 345         case NF_STOLEN:
 346                 break;
 347         default:
 348                 kfree_skb(skb);
 349         }
 350
 351         kfree(entry);
 352 }
 353 EXPORT_SYMBOL(nf_reinject);