Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
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_bridge.h>
14#include <linux/seq_file.h>
15#include <linux/rcupdate.h>
16#include <net/protocol.h>
17#include <net/netfilter/nf_queue.h>
18#include <net/dst.h>
19
20#include "nf_internals.h"
21
22/*
23 * Hook for nfnetlink_queue to register its queue handler.
24 * We do this so that most of the NFQUEUE code can be modular.
25 *
26 * Once the queue is registered it must reinject all packets it
27 * receives, no matter what.
28 */
29static const struct nf_queue_handler __rcu *queue_handler __read_mostly;
30
31/* return EBUSY when somebody else is registered, return EEXIST if the
32 * same handler is registered, return 0 in case of success. */
33void nf_register_queue_handler(const struct nf_queue_handler *qh)
34{
35 /* should never happen, we only have one queueing backend in kernel */
36 WARN_ON(rcu_access_pointer(queue_handler));
37 rcu_assign_pointer(queue_handler, qh);
38}
39EXPORT_SYMBOL(nf_register_queue_handler);
40
41/* The caller must flush their queue before this */
42void nf_unregister_queue_handler(void)
43{
44 RCU_INIT_POINTER(queue_handler, NULL);
45 synchronize_rcu();
46}
47EXPORT_SYMBOL(nf_unregister_queue_handler);
48
49void 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#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
61 if (entry->skb->nf_bridge) {
62 struct net_device *physdev;
63
64 physdev = nf_bridge_get_physindev(entry->skb);
65 if (physdev)
66 dev_put(physdev);
67 physdev = nf_bridge_get_physoutdev(entry->skb);
68 if (physdev)
69 dev_put(physdev);
70 }
71#endif
72 /* Drop reference to owner of hook which queued us. */
73 module_put(entry->elem->owner);
74}
75EXPORT_SYMBOL_GPL(nf_queue_entry_release_refs);
76
77/* Bump dev refs so they don't vanish while packet is out */
78bool nf_queue_entry_get_refs(struct nf_queue_entry *entry)
79{
80 struct nf_hook_state *state = &entry->state;
81
82 if (!try_module_get(entry->elem->owner))
83 return false;
84
85 if (state->in)
86 dev_hold(state->in);
87 if (state->out)
88 dev_hold(state->out);
89 if (state->sk)
90 sock_hold(state->sk);
91#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
92 if (entry->skb->nf_bridge) {
93 struct net_device *physdev;
94
95 physdev = nf_bridge_get_physindev(entry->skb);
96 if (physdev)
97 dev_hold(physdev);
98 physdev = nf_bridge_get_physoutdev(entry->skb);
99 if (physdev)
100 dev_hold(physdev);
101 }
102#endif
103
104 return true;
105}
106EXPORT_SYMBOL_GPL(nf_queue_entry_get_refs);
107
108void nf_queue_nf_hook_drop(struct net *net, struct nf_hook_ops *ops)
109{
110 const struct nf_queue_handler *qh;
111
112 rcu_read_lock();
113 qh = rcu_dereference(queue_handler);
114 if (qh)
115 qh->nf_hook_drop(net, ops);
116 rcu_read_unlock();
117}
118
119/*
120 * Any packet that leaves via this function must come back
121 * through nf_reinject().
122 */
123int nf_queue(struct sk_buff *skb,
124 struct nf_hook_ops *elem,
125 struct nf_hook_state *state,
126 unsigned int queuenum)
127{
128 int status = -ENOENT;
129 struct nf_queue_entry *entry = NULL;
130 const struct nf_afinfo *afinfo;
131 const struct nf_queue_handler *qh;
132
133 /* QUEUE == DROP if no one is waiting, to be safe. */
134 rcu_read_lock();
135
136 qh = rcu_dereference(queue_handler);
137 if (!qh) {
138 status = -ESRCH;
139 goto err_unlock;
140 }
141
142 afinfo = nf_get_afinfo(state->pf);
143 if (!afinfo)
144 goto err_unlock;
145
146 entry = kmalloc(sizeof(*entry) + afinfo->route_key_size, GFP_ATOMIC);
147 if (!entry) {
148 status = -ENOMEM;
149 goto err_unlock;
150 }
151
152 *entry = (struct nf_queue_entry) {
153 .skb = skb,
154 .elem = elem,
155 .state = *state,
156 .size = sizeof(*entry) + afinfo->route_key_size,
157 };
158
159 if (!nf_queue_entry_get_refs(entry)) {
160 status = -ECANCELED;
161 goto err_unlock;
162 }
163 skb_dst_force(skb);
164 afinfo->saveroute(skb, entry);
165 status = qh->outfn(entry, queuenum);
166
167 rcu_read_unlock();
168
169 if (status < 0) {
170 nf_queue_entry_release_refs(entry);
171 goto err;
172 }
173
174 return 0;
175
176err_unlock:
177 rcu_read_unlock();
178err:
179 kfree(entry);
180 return status;
181}
182
183void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
184{
185 struct sk_buff *skb = entry->skb;
186 struct nf_hook_ops *elem = entry->elem;
187 const struct nf_afinfo *afinfo;
188 int err;
189
190 rcu_read_lock();
191
192 nf_queue_entry_release_refs(entry);
193
194 /* Continue traversal iff userspace said ok... */
195 if (verdict == NF_REPEAT) {
196 elem = list_entry(elem->list.prev, struct nf_hook_ops, list);
197 verdict = NF_ACCEPT;
198 }
199
200 if (verdict == NF_ACCEPT) {
201 afinfo = nf_get_afinfo(entry->state.pf);
202 if (!afinfo || afinfo->reroute(skb, entry) < 0)
203 verdict = NF_DROP;
204 }
205
206 entry->state.thresh = INT_MIN;
207
208 if (verdict == NF_ACCEPT) {
209 next_hook:
210 verdict = nf_iterate(entry->state.hook_list,
211 skb, &entry->state, &elem);
212 }
213
214 switch (verdict & NF_VERDICT_MASK) {
215 case NF_ACCEPT:
216 case NF_STOP:
217 local_bh_disable();
218 entry->state.okfn(entry->state.sk, skb);
219 local_bh_enable();
220 break;
221 case NF_QUEUE:
222 err = nf_queue(skb, elem, &entry->state,
223 verdict >> NF_VERDICT_QBITS);
224 if (err < 0) {
225 if (err == -ECANCELED)
226 goto next_hook;
227 if (err == -ESRCH &&
228 (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
229 goto next_hook;
230 kfree_skb(skb);
231 }
232 break;
233 case NF_STOLEN:
234 break;
235 default:
236 kfree_skb(skb);
237 }
238 rcu_read_unlock();
239 kfree(entry);
240}
241EXPORT_SYMBOL(nf_reinject);