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/seq_file.h>
  14 #include <linux/rcupdate.h>
  15 #include <net/protocol.h>
  16 #include <net/netfilter/nf_queue.h>
  17 #include <net/dst.h>
  18
  19 #include "nf_internals.h"
  20
  21 /*
  22  * Hook for nfnetlink_queue to register its queue handler.
  23  * We do this so that most of the NFQUEUE code can be modular.
  24  *
  25  * Once the queue is registered it must reinject all packets it
  26  * receives, no matter what.
  27  */
  28 static const struct nf_queue_handler __rcu *queue_handler __read_mostly;
  29
  30 /* return EBUSY when somebody else is registered, return EEXIST if the
  31  * same handler is registered, return 0 in case of success. */
  32 void nf_register_queue_handler(const struct nf_queue_handler *qh)
  33 {
  34         /* should never happen, we only have one queueing backend in kernel */
  35         WARN_ON(rcu_access_pointer(queue_handler));
  36         rcu_assign_pointer(queue_handler, qh);
  37 }
  38 EXPORT_SYMBOL(nf_register_queue_handler);
  39
  40 /* The caller must flush their queue before this */
  41 void nf_unregister_queue_handler(void)
  42 {
  43         RCU_INIT_POINTER(queue_handler, NULL);
  44         synchronize_rcu();
  45 }
  46 EXPORT_SYMBOL(nf_unregister_queue_handler);
  47
  48 static void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
  49 {
  50         /* Release those devices we held, or Alexey will kill me. */
  51         if (entry->indev)
  52                 dev_put(entry->indev);
  53         if (entry->outdev)
  54                 dev_put(entry->outdev);
  55 #ifdef CONFIG_BRIDGE_NETFILTER
  56         if (entry->skb->nf_bridge) {
  57                 struct nf_bridge_info *nf_bridge = entry->skb->nf_bridge;
  58
  59                 if (nf_bridge->physindev)
  60                         dev_put(nf_bridge->physindev);
  61                 if (nf_bridge->physoutdev)
  62                         dev_put(nf_bridge->physoutdev);
  63         }
  64 #endif
  65         /* Drop reference to owner of hook which queued us. */
  66         module_put(entry->elem->owner);
  67 }
  68
  69 /* Bump dev refs so they don't vanish while packet is out */
  70 static bool nf_queue_entry_get_refs(struct nf_queue_entry *entry)
  71 {
  72         if (!try_module_get(entry->elem->owner))
  73                 return false;
  74
  75         if (entry->indev)
  76                 dev_hold(entry->indev);
  77         if (entry->outdev)
  78                 dev_hold(entry->outdev);
  79 #ifdef CONFIG_BRIDGE_NETFILTER
  80         if (entry->skb->nf_bridge) {
  81                 struct nf_bridge_info *nf_bridge = entry->skb->nf_bridge;
  82                 struct net_device *physdev;
  83
  84                 physdev = nf_bridge->physindev;
  85                 if (physdev)
  86                         dev_hold(physdev);
  87                 physdev = nf_bridge->physoutdev;
  88                 if (physdev)
  89                         dev_hold(physdev);
  90         }
  91 #endif
  92
  93         return true;
  94 }
  95
  96 /*
  97  * Any packet that leaves via this function must come back
  98  * through nf_reinject().
  99  */
 100 static int __nf_queue(struct sk_buff *skb,
 101                       struct nf_hook_ops *elem,
 102                       u_int8_t pf, unsigned int hook,
 103                       struct net_device *indev,
 104                       struct net_device *outdev,
 105                       int (*okfn)(struct sk_buff *),
 106                       unsigned int queuenum)
 107 {
 108         int status = -ENOENT;
 109         struct nf_queue_entry *entry = NULL;
 110         const struct nf_afinfo *afinfo;
 111         const struct nf_queue_handler *qh;
 112
 113         /* QUEUE == DROP if no one is waiting, to be safe. */
 114         rcu_read_lock();
 115
 116         qh = rcu_dereference(queue_handler);
 117         if (!qh) {
 118                 status = -ESRCH;
 119                 goto err_unlock;
 120         }
 121
 122         afinfo = nf_get_afinfo(pf);
 123         if (!afinfo)
 124                 goto err_unlock;
 125
 126         entry = kmalloc(sizeof(*entry) + afinfo->route_key_size, GFP_ATOMIC);
 127         if (!entry) {
 128                 status = -ENOMEM;
 129                 goto err_unlock;
 130         }
 131
 132         *entry = (struct nf_queue_entry) {
 133                 .skb    = skb,
 134                 .elem   = elem,
 135                 .pf     = pf,
 136                 .hook   = hook,
 137                 .indev  = indev,
 138                 .outdev = outdev,
 139                 .okfn   = okfn,
 140         };
 141
 142         if (!nf_queue_entry_get_refs(entry)) {
 143                 status = -ECANCELED;
 144                 goto err_unlock;
 145         }
 146         skb_dst_force(skb);
 147         afinfo->saveroute(skb, entry);
 148         status = qh->outfn(entry, queuenum);
 149
 150         rcu_read_unlock();
 151
 152         if (status < 0) {
 153                 nf_queue_entry_release_refs(entry);
 154                 goto err;
 155         }
 156
 157         return 0;
 158
 159 err_unlock:
 160         rcu_read_unlock();
 161 err:
 162         kfree(entry);
 163         return status;
 164 }
 165
 166 #ifdef CONFIG_BRIDGE_NETFILTER
 167 /* When called from bridge netfilter, skb->data must point to MAC header
 168  * before calling skb_gso_segment(). Else, original MAC header is lost
 169  * and segmented skbs will be sent to wrong destination.
 170  */
 171 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
 172 {
 173         if (skb->nf_bridge)
 174                 __skb_push(skb, skb->network_header - skb->mac_header);
 175 }
 176
 177 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
 178 {
 179         if (skb->nf_bridge)
 180                 __skb_pull(skb, skb->network_header - skb->mac_header);
 181 }
 182 #else
 183 #define nf_bridge_adjust_skb_data(s) do {} while (0)
 184 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
 185 #endif
 186
 187 int nf_queue(struct sk_buff *skb,
 188              struct nf_hook_ops *elem,
 189              u_int8_t pf, unsigned int hook,
 190              struct net_device *indev,
 191              struct net_device *outdev,
 192              int (*okfn)(struct sk_buff *),
 193              unsigned int queuenum)
 194 {
 195         struct sk_buff *segs;
 196         int err = -EINVAL;
 197         unsigned int queued;
 198
 199         if (!skb_is_gso(skb))
 200                 return __nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
 201                                   queuenum);
 202
 203         switch (pf) {
 204         case NFPROTO_IPV4:
 205                 skb->protocol = htons(ETH_P_IP);
 206                 break;
 207         case NFPROTO_IPV6:
 208                 skb->protocol = htons(ETH_P_IPV6);
 209                 break;
 210         }
 211
 212         nf_bridge_adjust_skb_data(skb);
 213         segs = skb_gso_segment(skb, 0);
 214         /* Does not use PTR_ERR to limit the number of error codes that can be
 215          * returned by nf_queue.  For instance, callers rely on -ECANCELED to mean
 216          * 'ignore this hook'.
 217          */
 218         if (IS_ERR(segs))
 219                 goto out_err;
 220         queued = 0;
 221         err = 0;
 222         do {
 223                 struct sk_buff *nskb = segs->next;
 224
 225                 segs->next = NULL;
 226                 if (err == 0) {
 227                         nf_bridge_adjust_segmented_data(segs);
 228                         err = __nf_queue(segs, elem, pf, hook, indev,
 229                                            outdev, okfn, queuenum);
 230                 }
 231                 if (err == 0)
 232                         queued++;
 233                 else
 234                         kfree_skb(segs);
 235                 segs = nskb;
 236         } while (segs);
 237
 238         if (queued) {
 239                 kfree_skb(skb);
 240                 return 0;
 241         }
 242   out_err:
 243         nf_bridge_adjust_segmented_data(skb);
 244         return err;
 245 }
 246
 247 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 248 {
 249         struct sk_buff *skb = entry->skb;
 250         struct nf_hook_ops *elem = entry->elem;
 251         const struct nf_afinfo *afinfo;
 252         int err;
 253
 254         rcu_read_lock();
 255
 256         nf_queue_entry_release_refs(entry);
 257
 258         /* Continue traversal iff userspace said ok... */
 259         if (verdict == NF_REPEAT) {
 260                 elem = list_entry(elem->list.prev, struct nf_hook_ops, list);
 261                 verdict = NF_ACCEPT;
 262         }
 263
 264         if (verdict == NF_ACCEPT) {
 265                 afinfo = nf_get_afinfo(entry->pf);
 266                 if (!afinfo || afinfo->reroute(skb, entry) < 0)
 267                         verdict = NF_DROP;
 268         }
 269
 270         if (verdict == NF_ACCEPT) {
 271         next_hook:
 272                 verdict = nf_iterate(&nf_hooks[entry->pf][entry->hook],
 273                                      skb, entry->hook,
 274                                      entry->indev, entry->outdev, &elem,
 275                                      entry->okfn, INT_MIN);
 276         }
 277
 278         switch (verdict & NF_VERDICT_MASK) {
 279         case NF_ACCEPT:
 280         case NF_STOP:
 281                 local_bh_disable();
 282                 entry->okfn(skb);
 283                 local_bh_enable();
 284                 break;
 285         case NF_QUEUE:
 286                 err = __nf_queue(skb, elem, entry->pf, entry->hook,
 287                                  entry->indev, entry->outdev, entry->okfn,
 288                                  verdict >> NF_VERDICT_QBITS);
 289                 if (err < 0) {
 290                         if (err == -ECANCELED)
 291                                 goto next_hook;
 292                         if (err == -ESRCH &&
 293                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 294                                 goto next_hook;
 295                         kfree_skb(skb);
 296                 }
 297                 break;
 298         case NF_STOLEN:
 299                 break;
 300         default:
 301                 kfree_skb(skb);
 302         }
 303         rcu_read_unlock();
 304         kfree(entry);
 305 }
 306 EXPORT_SYMBOL(nf_reinject);