net/netfilter/nf_queue.c

   1 #include <linux/kernel.h>
   2 #include <linux/slab.h>
   3 #include <linux/init.h>
   4 #include <linux/module.h>
   5 #include <linux/proc_fs.h>
   6 #include <linux/skbuff.h>
   7 #include <linux/netfilter.h>
   8 #include <linux/seq_file.h>
   9 #include <linux/rcupdate.h>
  10 #include <net/protocol.h>
  11 #include <net/netfilter/nf_queue.h>
  12 #include <net/dst.h>
  13
  14 #include "nf_internals.h"
  15
  16 /*
  17  * A queue handler may be registered for each protocol.  Each is protected by
  18  * long term mutex.  The handler must provide an an outfn() to accept packets
  19  * for queueing and must reinject all packets it receives, no matter what.
  20  */
  21 static const struct nf_queue_handler __rcu *queue_handler[NFPROTO_NUMPROTO] __read_mostly;
  22
  23 static DEFINE_MUTEX(queue_handler_mutex);
  24
  25 /* return EBUSY when somebody else is registered, return EEXIST if the
  26  * same handler is registered, return 0 in case of success. */
  27 int nf_register_queue_handler(u_int8_t pf, const struct nf_queue_handler *qh)
  28 {
  29         int ret;
  30         const struct nf_queue_handler *old;
  31
  32         if (pf >= ARRAY_SIZE(queue_handler))
  33                 return -EINVAL;
  34
  35         mutex_lock(&queue_handler_mutex);
  36         old = rcu_dereference_protected(queue_handler[pf],
  37                                         lockdep_is_held(&queue_handler_mutex));
  38         if (old == qh)
  39                 ret = -EEXIST;
  40         else if (old)
  41                 ret = -EBUSY;
  42         else {
  43                 rcu_assign_pointer(queue_handler[pf], qh);
  44                 ret = 0;
  45         }
  46         mutex_unlock(&queue_handler_mutex);
  47
  48         return ret;
  49 }
  50 EXPORT_SYMBOL(nf_register_queue_handler);
  51
  52 /* The caller must flush their queue before this */
  53 int nf_unregister_queue_handler(u_int8_t pf, const struct nf_queue_handler *qh)
  54 {
  55         const struct nf_queue_handler *old;
  56
  57         if (pf >= ARRAY_SIZE(queue_handler))
  58                 return -EINVAL;
  59
  60         mutex_lock(&queue_handler_mutex);
  61         old = rcu_dereference_protected(queue_handler[pf],
  62                                         lockdep_is_held(&queue_handler_mutex));
  63         if (old && old != qh) {
  64                 mutex_unlock(&queue_handler_mutex);
  65                 return -EINVAL;
  66         }
  67
  68         rcu_assign_pointer(queue_handler[pf], NULL);
  69         mutex_unlock(&queue_handler_mutex);
  70
  71         synchronize_rcu();
  72
  73         return 0;
  74 }
  75 EXPORT_SYMBOL(nf_unregister_queue_handler);
  76
  77 void nf_unregister_queue_handlers(const struct nf_queue_handler *qh)
  78 {
  79         u_int8_t pf;
  80
  81         mutex_lock(&queue_handler_mutex);
  82         for (pf = 0; pf < ARRAY_SIZE(queue_handler); pf++)  {
  83                 if (rcu_dereference_protected(
  84                                 queue_handler[pf],
  85                                 lockdep_is_held(&queue_handler_mutex)
  86                                 ) == qh)
  87                         rcu_assign_pointer(queue_handler[pf], NULL);
  88         }
  89         mutex_unlock(&queue_handler_mutex);
  90
  91         synchronize_rcu();
  92 }
  93 EXPORT_SYMBOL_GPL(nf_unregister_queue_handlers);
  94
  95 static void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
  96 {
  97         /* Release those devices we held, or Alexey will kill me. */
  98         if (entry->indev)
  99                 dev_put(entry->indev);
 100         if (entry->outdev)
 101                 dev_put(entry->outdev);
 102 #ifdef CONFIG_BRIDGE_NETFILTER
 103         if (entry->skb->nf_bridge) {
 104                 struct nf_bridge_info *nf_bridge = entry->skb->nf_bridge;
 105
 106                 if (nf_bridge->physindev)
 107                         dev_put(nf_bridge->physindev);
 108                 if (nf_bridge->physoutdev)
 109                         dev_put(nf_bridge->physoutdev);
 110         }
 111 #endif
 112         /* Drop reference to owner of hook which queued us. */
 113         module_put(entry->elem->owner);
 114 }
 115
 116 /*
 117  * Any packet that leaves via this function must come back
 118  * through nf_reinject().
 119  */
 120 static int __nf_queue(struct sk_buff *skb,
 121                       struct list_head *elem,
 122                       u_int8_t pf, unsigned int hook,
 123                       struct net_device *indev,
 124                       struct net_device *outdev,
 125                       int (*okfn)(struct sk_buff *),
 126                       unsigned int queuenum)
 127 {
 128         int status = -ENOENT;
 129         struct nf_queue_entry *entry = NULL;
 130 #ifdef CONFIG_BRIDGE_NETFILTER
 131         struct net_device *physindev;
 132         struct net_device *physoutdev;
 133 #endif
 134         const struct nf_afinfo *afinfo;
 135         const struct nf_queue_handler *qh;
 136
 137         /* QUEUE == DROP if noone is waiting, to be safe. */
 138         rcu_read_lock();
 139
 140         qh = rcu_dereference(queue_handler[pf]);
 141         if (!qh)
 142                 goto err_unlock;
 143
 144         afinfo = nf_get_afinfo(pf);
 145         if (!afinfo)
 146                 goto err_unlock;
 147
 148         entry = kmalloc(sizeof(*entry) + afinfo->route_key_size, GFP_ATOMIC);
 149         if (!entry) {
 150                 status = -ENOMEM;
 151                 goto err_unlock;
 152         }
 153
 154         *entry = (struct nf_queue_entry) {
 155                 .skb    = skb,
 156                 .elem   = list_entry(elem, struct nf_hook_ops, list),
 157                 .pf     = pf,
 158                 .hook   = hook,
 159                 .indev  = indev,
 160                 .outdev = outdev,
 161                 .okfn   = okfn,
 162         };
 163
 164         /* If it's going away, ignore hook. */
 165         if (!try_module_get(entry->elem->owner)) {
 166                 status = -ECANCELED;
 167                 goto err_unlock;
 168         }
 169         /* Bump dev refs so they don't vanish while packet is out */
 170         if (indev)
 171                 dev_hold(indev);
 172         if (outdev)
 173                 dev_hold(outdev);
 174 #ifdef CONFIG_BRIDGE_NETFILTER
 175         if (skb->nf_bridge) {
 176                 physindev = skb->nf_bridge->physindev;
 177                 if (physindev)
 178                         dev_hold(physindev);
 179                 physoutdev = skb->nf_bridge->physoutdev;
 180                 if (physoutdev)
 181                         dev_hold(physoutdev);
 182         }
 183 #endif
 184         skb_dst_force(skb);
 185         afinfo->saveroute(skb, entry);
 186         status = qh->outfn(entry, queuenum);
 187
 188         rcu_read_unlock();
 189
 190         if (status < 0) {
 191                 nf_queue_entry_release_refs(entry);
 192                 goto err;
 193         }
 194
 195         return 0;
 196
 197 err_unlock:
 198         rcu_read_unlock();
 199 err:
 200         kfree(entry);
 201         return status;
 202 }
 203
 204 int nf_queue(struct sk_buff *skb,
 205              struct list_head *elem,
 206              u_int8_t pf, unsigned int hook,
 207              struct net_device *indev,
 208              struct net_device *outdev,
 209              int (*okfn)(struct sk_buff *),
 210              unsigned int queuenum)
 211 {
 212         struct sk_buff *segs;
 213         int err;
 214         unsigned int queued;
 215
 216         if (!skb_is_gso(skb))
 217                 return __nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
 218                                   queuenum);
 219
 220         switch (pf) {
 221         case NFPROTO_IPV4:
 222                 skb->protocol = htons(ETH_P_IP);
 223                 break;
 224         case NFPROTO_IPV6:
 225                 skb->protocol = htons(ETH_P_IPV6);
 226                 break;
 227         }
 228
 229         segs = skb_gso_segment(skb, 0);
 230         /* Does not use PTR_ERR to limit the number of error codes that can be
 231          * returned by nf_queue.  For instance, callers rely on -ECANCELED to mean
 232          * 'ignore this hook'.
 233          */
 234         if (IS_ERR(segs))
 235                 return -EINVAL;
 236
 237         queued = 0;
 238         err = 0;
 239         do {
 240                 struct sk_buff *nskb = segs->next;
 241
 242                 segs->next = NULL;
 243                 if (err == 0)
 244                         err = __nf_queue(segs, elem, pf, hook, indev,
 245                                            outdev, okfn, queuenum);
 246                 if (err == 0)
 247                         queued++;
 248                 else
 249                         kfree_skb(segs);
 250                 segs = nskb;
 251         } while (segs);
 252
 253         /* also free orig skb if only some segments were queued */
 254         if (unlikely(err && queued))
 255                 err = 0;
 256         if (err == 0)
 257                 kfree_skb(skb);
 258         return err;
 259 }
 260
 261 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 262 {
 263         struct sk_buff *skb = entry->skb;
 264         struct list_head *elem = &entry->elem->list;
 265         const struct nf_afinfo *afinfo;
 266         int err;
 267
 268         rcu_read_lock();
 269
 270         nf_queue_entry_release_refs(entry);
 271
 272         /* Continue traversal iff userspace said ok... */
 273         if (verdict == NF_REPEAT) {
 274                 elem = elem->prev;
 275                 verdict = NF_ACCEPT;
 276         }
 277
 278         if (verdict == NF_ACCEPT) {
 279                 afinfo = nf_get_afinfo(entry->pf);
 280                 if (!afinfo || afinfo->reroute(skb, entry) < 0)
 281                         verdict = NF_DROP;
 282         }
 283
 284         if (verdict == NF_ACCEPT) {
 285         next_hook:
 286                 verdict = nf_iterate(&nf_hooks[entry->pf][entry->hook],
 287                                      skb, entry->hook,
 288                                      entry->indev, entry->outdev, &elem,
 289                                      entry->okfn, INT_MIN);
 290         }
 291
 292         switch (verdict & NF_VERDICT_MASK) {
 293         case NF_ACCEPT:
 294         case NF_STOP:
 295                 local_bh_disable();
 296                 entry->okfn(skb);
 297                 local_bh_enable();
 298                 break;
 299         case NF_QUEUE:
 300                 err = __nf_queue(skb, elem, entry->pf, entry->hook,
 301                                  entry->indev, entry->outdev, entry->okfn,
 302                                  verdict >> NF_VERDICT_BITS);
 303                 if (err < 0) {
 304                         if (err == -ECANCELED)
 305                                 goto next_hook;
 306                         kfree_skb(skb);
 307                 }
 308                 break;
 309         case NF_STOLEN:
 310         default:
 311                 kfree_skb(skb);
 312         }
 313         rcu_read_unlock();
 314         kfree(entry);
 315 }
 316 EXPORT_SYMBOL(nf_reinject);
 317
 318 #ifdef CONFIG_PROC_FS
 319 static void *seq_start(struct seq_file *seq, loff_t *pos)
 320 {
 321         if (*pos >= ARRAY_SIZE(queue_handler))
 322                 return NULL;
 323
 324         return pos;
 325 }
 326
 327 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 328 {
 329         (*pos)++;
 330
 331         if (*pos >= ARRAY_SIZE(queue_handler))
 332                 return NULL;
 333
 334         return pos;
 335 }
 336
 337 static void seq_stop(struct seq_file *s, void *v)
 338 {
 339
 340 }
 341
 342 static int seq_show(struct seq_file *s, void *v)
 343 {
 344         int ret;
 345         loff_t *pos = v;
 346         const struct nf_queue_handler *qh;
 347
 348         rcu_read_lock();
 349         qh = rcu_dereference(queue_handler[*pos]);
 350         if (!qh)
 351                 ret = seq_printf(s, "%2lld NONE\n", *pos);
 352         else
 353                 ret = seq_printf(s, "%2lld %s\n", *pos, qh->name);
 354         rcu_read_unlock();
 355
 356         return ret;
 357 }
 358
 359 static const struct seq_operations nfqueue_seq_ops = {
 360         .start  = seq_start,
 361         .next   = seq_next,
 362         .stop   = seq_stop,
 363         .show   = seq_show,
 364 };
 365
 366 static int nfqueue_open(struct inode *inode, struct file *file)
 367 {
 368         return seq_open(file, &nfqueue_seq_ops);
 369 }
 370
 371 static const struct file_operations nfqueue_file_ops = {
 372         .owner   = THIS_MODULE,
 373         .open    = nfqueue_open,
 374         .read    = seq_read,
 375         .llseek  = seq_lseek,
 376         .release = seq_release,
 377 };
 378 #endif /* PROC_FS */
 379
 380
 381 int __init netfilter_queue_init(void)
 382 {
 383 #ifdef CONFIG_PROC_FS
 384         if (!proc_create("nf_queue", S_IRUGO,
 385                          proc_net_netfilter, &nfqueue_file_ops))
 386                 return -1;
 387 #endif
 388         return 0;
 389 }
 390