xref: /linux/include/linux/netfilter.h (revision c6fbb759) 
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __LINUX_NETFILTER_H
3 #define __LINUX_NETFILTER_H
4 
5 #include <linux/init.h>
6 #include <linux/skbuff.h>
7 #include <linux/net.h>
8 #include <linux/if.h>
9 #include <linux/in.h>
10 #include <linux/in6.h>
11 #include <linux/wait.h>
12 #include <linux/list.h>
13 #include <linux/static_key.h>
14 #include <linux/netfilter_defs.h>
15 #include <linux/netdevice.h>
16 #include <linux/sockptr.h>
17 #include <net/net_namespace.h>
18 
19 static inline int NF_DROP_GETERR(int verdict)
20 {
21 	return -(verdict >> NF_VERDICT_QBITS);
22 }
23 
24 static inline int nf_inet_addr_cmp(const union nf_inet_addr *a1,
25 				   const union nf_inet_addr *a2)
26 {
27 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
28 	const unsigned long *ul1 = (const unsigned long *)a1;
29 	const unsigned long *ul2 = (const unsigned long *)a2;
30 
31 	return ((ul1[0] ^ ul2[0]) | (ul1[1] ^ ul2[1])) == 0UL;
32 #else
33 	return a1->all[0] == a2->all[0] &&
34 	       a1->all[1] == a2->all[1] &&
35 	       a1->all[2] == a2->all[2] &&
36 	       a1->all[3] == a2->all[3];
37 #endif
38 }
39 
40 static inline void nf_inet_addr_mask(const union nf_inet_addr *a1,
41 				     union nf_inet_addr *result,
42 				     const union nf_inet_addr *mask)
43 {
44 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
45 	const unsigned long *ua = (const unsigned long *)a1;
46 	unsigned long *ur = (unsigned long *)result;
47 	const unsigned long *um = (const unsigned long *)mask;
48 
49 	ur[0] = ua[0] & um[0];
50 	ur[1] = ua[1] & um[1];
51 #else
52 	result->all[0] = a1->all[0] & mask->all[0];
53 	result->all[1] = a1->all[1] & mask->all[1];
54 	result->all[2] = a1->all[2] & mask->all[2];
55 	result->all[3] = a1->all[3] & mask->all[3];
56 #endif
57 }
58 
59 int netfilter_init(void);
60 
61 struct sk_buff;
62 
63 struct nf_hook_ops;
64 
65 struct sock;
66 
67 struct nf_hook_state {
68 	u8 hook;
69 	u8 pf;
70 	struct net_device *in;
71 	struct net_device *out;
72 	struct sock *sk;
73 	struct net *net;
74 	int (*okfn)(struct net *, struct sock *, struct sk_buff *);
75 };
76 
77 typedef unsigned int nf_hookfn(void *priv,
78 			       struct sk_buff *skb,
79 			       const struct nf_hook_state *state);
80 enum nf_hook_ops_type {
81 	NF_HOOK_OP_UNDEFINED,
82 	NF_HOOK_OP_NF_TABLES,
83 };
84 
85 struct nf_hook_ops {
86 	/* User fills in from here down. */
87 	nf_hookfn		*hook;
88 	struct net_device	*dev;
89 	void			*priv;
90 	u8			pf;
91 	enum nf_hook_ops_type	hook_ops_type:8;
92 	unsigned int		hooknum;
93 	/* Hooks are ordered in ascending priority. */
94 	int			priority;
95 };
96 
97 struct nf_hook_entry {
98 	nf_hookfn			*hook;
99 	void				*priv;
100 };
101 
102 struct nf_hook_entries_rcu_head {
103 	struct rcu_head head;
104 	void	*allocation;
105 };
106 
107 struct nf_hook_entries {
108 	u16				num_hook_entries;
109 	/* padding */
110 	struct nf_hook_entry		hooks[];
111 
112 	/* trailer: pointers to original orig_ops of each hook,
113 	 * followed by rcu_head and scratch space used for freeing
114 	 * the structure via call_rcu.
115 	 *
116 	 *   This is not part of struct nf_hook_entry since its only
117 	 *   needed in slow path (hook register/unregister):
118 	 * const struct nf_hook_ops     *orig_ops[]
119 	 *
120 	 *   For the same reason, we store this at end -- its
121 	 *   only needed when a hook is deleted, not during
122 	 *   packet path processing:
123 	 * struct nf_hook_entries_rcu_head     head
124 	 */
125 };
126 
127 #ifdef CONFIG_NETFILTER
128 static inline struct nf_hook_ops **nf_hook_entries_get_hook_ops(const struct nf_hook_entries *e)
129 {
130 	unsigned int n = e->num_hook_entries;
131 	const void *hook_end;
132 
133 	hook_end = &e->hooks[n]; /* this is *past* ->hooks[]! */
134 
135 	return (struct nf_hook_ops **)hook_end;
136 }
137 
138 static inline int
139 nf_hook_entry_hookfn(const struct nf_hook_entry *entry, struct sk_buff *skb,
140 		     struct nf_hook_state *state)
141 {
142 	return entry->hook(entry->priv, skb, state);
143 }
144 
145 static inline void nf_hook_state_init(struct nf_hook_state *p,
146 				      unsigned int hook,
147 				      u_int8_t pf,
148 				      struct net_device *indev,
149 				      struct net_device *outdev,
150 				      struct sock *sk,
151 				      struct net *net,
152 				      int (*okfn)(struct net *, struct sock *, struct sk_buff *))
153 {
154 	p->hook = hook;
155 	p->pf = pf;
156 	p->in = indev;
157 	p->out = outdev;
158 	p->sk = sk;
159 	p->net = net;
160 	p->okfn = okfn;
161 }
162 
163 
164 
165 struct nf_sockopt_ops {
166 	struct list_head list;
167 
168 	u_int8_t pf;
169 
170 	/* Non-inclusive ranges: use 0/0/NULL to never get called. */
171 	int set_optmin;
172 	int set_optmax;
173 	int (*set)(struct sock *sk, int optval, sockptr_t arg,
174 		   unsigned int len);
175 	int get_optmin;
176 	int get_optmax;
177 	int (*get)(struct sock *sk, int optval, void __user *user, int *len);
178 	/* Use the module struct to lock set/get code in place */
179 	struct module *owner;
180 };
181 
182 /* Function to register/unregister hook points. */
183 int nf_register_net_hook(struct net *net, const struct nf_hook_ops *ops);
184 void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *ops);
185 int nf_register_net_hooks(struct net *net, const struct nf_hook_ops *reg,
186 			  unsigned int n);
187 void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg,
188 			     unsigned int n);
189 
190 /* Functions to register get/setsockopt ranges (non-inclusive).  You
191    need to check permissions yourself! */
192 int nf_register_sockopt(struct nf_sockopt_ops *reg);
193 void nf_unregister_sockopt(struct nf_sockopt_ops *reg);
194 
195 #ifdef CONFIG_JUMP_LABEL
196 extern struct static_key nf_hooks_needed[NFPROTO_NUMPROTO][NF_MAX_HOOKS];
197 #endif
198 
199 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state,
200 		 const struct nf_hook_entries *e, unsigned int i);
201 
202 void nf_hook_slow_list(struct list_head *head, struct nf_hook_state *state,
203 		       const struct nf_hook_entries *e);
204 /**
205  *	nf_hook - call a netfilter hook
206  *
207  *	Returns 1 if the hook has allowed the packet to pass.  The function
208  *	okfn must be invoked by the caller in this case.  Any other return
209  *	value indicates the packet has been consumed by the hook.
210  */
211 static inline int nf_hook(u_int8_t pf, unsigned int hook, struct net *net,
212 			  struct sock *sk, struct sk_buff *skb,
213 			  struct net_device *indev, struct net_device *outdev,
214 			  int (*okfn)(struct net *, struct sock *, struct sk_buff *))
215 {
216 	struct nf_hook_entries *hook_head = NULL;
217 	int ret = 1;
218 
219 #ifdef CONFIG_JUMP_LABEL
220 	if (__builtin_constant_p(pf) &&
221 	    __builtin_constant_p(hook) &&
222 	    !static_key_false(&nf_hooks_needed[pf][hook]))
223 		return 1;
224 #endif
225 
226 	rcu_read_lock();
227 	switch (pf) {
228 	case NFPROTO_IPV4:
229 		hook_head = rcu_dereference(net->nf.hooks_ipv4[hook]);
230 		break;
231 	case NFPROTO_IPV6:
232 		hook_head = rcu_dereference(net->nf.hooks_ipv6[hook]);
233 		break;
234 	case NFPROTO_ARP:
235 #ifdef CONFIG_NETFILTER_FAMILY_ARP
236 		if (WARN_ON_ONCE(hook >= ARRAY_SIZE(net->nf.hooks_arp)))
237 			break;
238 		hook_head = rcu_dereference(net->nf.hooks_arp[hook]);
239 #endif
240 		break;
241 	case NFPROTO_BRIDGE:
242 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
243 		hook_head = rcu_dereference(net->nf.hooks_bridge[hook]);
244 #endif
245 		break;
246 	default:
247 		WARN_ON_ONCE(1);
248 		break;
249 	}
250 
251 	if (hook_head) {
252 		struct nf_hook_state state;
253 
254 		nf_hook_state_init(&state, hook, pf, indev, outdev,
255 				   sk, net, okfn);
256 
257 		ret = nf_hook_slow(skb, &state, hook_head, 0);
258 	}
259 	rcu_read_unlock();
260 
261 	return ret;
262 }
263 
264 /* Activate hook; either okfn or kfree_skb called, unless a hook
265    returns NF_STOLEN (in which case, it's up to the hook to deal with
266    the consequences).
267 
268    Returns -ERRNO if packet dropped.  Zero means queued, stolen or
269    accepted.
270 */
271 
272 /* RR:
273    > I don't want nf_hook to return anything because people might forget
274    > about async and trust the return value to mean "packet was ok".
275 
276    AK:
277    Just document it clearly, then you can expect some sense from kernel
278    coders :)
279 */
280 
281 static inline int
282 NF_HOOK_COND(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
283 	     struct sk_buff *skb, struct net_device *in, struct net_device *out,
284 	     int (*okfn)(struct net *, struct sock *, struct sk_buff *),
285 	     bool cond)
286 {
287 	int ret;
288 
289 	if (!cond ||
290 	    ((ret = nf_hook(pf, hook, net, sk, skb, in, out, okfn)) == 1))
291 		ret = okfn(net, sk, skb);
292 	return ret;
293 }
294 
295 static inline int
296 NF_HOOK(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk, struct sk_buff *skb,
297 	struct net_device *in, struct net_device *out,
298 	int (*okfn)(struct net *, struct sock *, struct sk_buff *))
299 {
300 	int ret = nf_hook(pf, hook, net, sk, skb, in, out, okfn);
301 	if (ret == 1)
302 		ret = okfn(net, sk, skb);
303 	return ret;
304 }
305 
306 static inline void
307 NF_HOOK_LIST(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
308 	     struct list_head *head, struct net_device *in, struct net_device *out,
309 	     int (*okfn)(struct net *, struct sock *, struct sk_buff *))
310 {
311 	struct nf_hook_entries *hook_head = NULL;
312 
313 #ifdef CONFIG_JUMP_LABEL
314 	if (__builtin_constant_p(pf) &&
315 	    __builtin_constant_p(hook) &&
316 	    !static_key_false(&nf_hooks_needed[pf][hook]))
317 		return;
318 #endif
319 
320 	rcu_read_lock();
321 	switch (pf) {
322 	case NFPROTO_IPV4:
323 		hook_head = rcu_dereference(net->nf.hooks_ipv4[hook]);
324 		break;
325 	case NFPROTO_IPV6:
326 		hook_head = rcu_dereference(net->nf.hooks_ipv6[hook]);
327 		break;
328 	default:
329 		WARN_ON_ONCE(1);
330 		break;
331 	}
332 
333 	if (hook_head) {
334 		struct nf_hook_state state;
335 
336 		nf_hook_state_init(&state, hook, pf, in, out, sk, net, okfn);
337 
338 		nf_hook_slow_list(head, &state, hook_head);
339 	}
340 	rcu_read_unlock();
341 }
342 
343 /* Call setsockopt() */
344 int nf_setsockopt(struct sock *sk, u_int8_t pf, int optval, sockptr_t opt,
345 		  unsigned int len);
346 int nf_getsockopt(struct sock *sk, u_int8_t pf, int optval, char __user *opt,
347 		  int *len);
348 
349 struct flowi;
350 struct nf_queue_entry;
351 
352 __sum16 nf_checksum(struct sk_buff *skb, unsigned int hook,
353 		    unsigned int dataoff, u_int8_t protocol,
354 		    unsigned short family);
355 
356 __sum16 nf_checksum_partial(struct sk_buff *skb, unsigned int hook,
357 			    unsigned int dataoff, unsigned int len,
358 			    u_int8_t protocol, unsigned short family);
359 int nf_route(struct net *net, struct dst_entry **dst, struct flowi *fl,
360 	     bool strict, unsigned short family);
361 int nf_reroute(struct sk_buff *skb, struct nf_queue_entry *entry);
362 
363 #include <net/flow.h>
364 
365 struct nf_conn;
366 enum nf_nat_manip_type;
367 struct nlattr;
368 enum ip_conntrack_dir;
369 
370 struct nf_nat_hook {
371 	int (*parse_nat_setup)(struct nf_conn *ct, enum nf_nat_manip_type manip,
372 			       const struct nlattr *attr);
373 	void (*decode_session)(struct sk_buff *skb, struct flowi *fl);
374 	unsigned int (*manip_pkt)(struct sk_buff *skb, struct nf_conn *ct,
375 				  enum nf_nat_manip_type mtype,
376 				  enum ip_conntrack_dir dir);
377 	void (*remove_nat_bysrc)(struct nf_conn *ct);
378 };
379 
380 extern const struct nf_nat_hook __rcu *nf_nat_hook;
381 
382 static inline void
383 nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl, u_int8_t family)
384 {
385 #if IS_ENABLED(CONFIG_NF_NAT)
386 	const struct nf_nat_hook *nat_hook;
387 
388 	rcu_read_lock();
389 	nat_hook = rcu_dereference(nf_nat_hook);
390 	if (nat_hook && nat_hook->decode_session)
391 		nat_hook->decode_session(skb, fl);
392 	rcu_read_unlock();
393 #endif
394 }
395 
396 #else /* !CONFIG_NETFILTER */
397 static inline int
398 NF_HOOK_COND(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
399 	     struct sk_buff *skb, struct net_device *in, struct net_device *out,
400 	     int (*okfn)(struct net *, struct sock *, struct sk_buff *),
401 	     bool cond)
402 {
403 	return okfn(net, sk, skb);
404 }
405 
406 static inline int
407 NF_HOOK(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
408 	struct sk_buff *skb, struct net_device *in, struct net_device *out,
409 	int (*okfn)(struct net *, struct sock *, struct sk_buff *))
410 {
411 	return okfn(net, sk, skb);
412 }
413 
414 static inline void
415 NF_HOOK_LIST(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
416 	     struct list_head *head, struct net_device *in, struct net_device *out,
417 	     int (*okfn)(struct net *, struct sock *, struct sk_buff *))
418 {
419 	/* nothing to do */
420 }
421 
422 static inline int nf_hook(u_int8_t pf, unsigned int hook, struct net *net,
423 			  struct sock *sk, struct sk_buff *skb,
424 			  struct net_device *indev, struct net_device *outdev,
425 			  int (*okfn)(struct net *, struct sock *, struct sk_buff *))
426 {
427 	return 1;
428 }
429 struct flowi;
430 static inline void
431 nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl, u_int8_t family)
432 {
433 }
434 #endif /*CONFIG_NETFILTER*/
435 
436 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
437 #include <linux/netfilter/nf_conntrack_zones_common.h>
438 
439 void nf_ct_attach(struct sk_buff *, const struct sk_buff *);
440 struct nf_conntrack_tuple;
441 bool nf_ct_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
442 			 const struct sk_buff *skb);
443 #else
444 static inline void nf_ct_attach(struct sk_buff *new, struct sk_buff *skb) {}
445 struct nf_conntrack_tuple;
446 static inline bool nf_ct_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
447 				       const struct sk_buff *skb)
448 {
449 	return false;
450 }
451 #endif
452 
453 struct nf_conn;
454 enum ip_conntrack_info;
455 
456 struct nf_ct_hook {
457 	int (*update)(struct net *net, struct sk_buff *skb);
458 	void (*destroy)(struct nf_conntrack *);
459 	bool (*get_tuple_skb)(struct nf_conntrack_tuple *,
460 			      const struct sk_buff *);
461 	void (*attach)(struct sk_buff *nskb, const struct sk_buff *skb);
462 };
463 extern const struct nf_ct_hook __rcu *nf_ct_hook;
464 
465 struct nlattr;
466 
467 struct nfnl_ct_hook {
468 	size_t (*build_size)(const struct nf_conn *ct);
469 	int (*build)(struct sk_buff *skb, struct nf_conn *ct,
470 		     enum ip_conntrack_info ctinfo,
471 		     u_int16_t ct_attr, u_int16_t ct_info_attr);
472 	int (*parse)(const struct nlattr *attr, struct nf_conn *ct);
473 	int (*attach_expect)(const struct nlattr *attr, struct nf_conn *ct,
474 			     u32 portid, u32 report);
475 	void (*seq_adjust)(struct sk_buff *skb, struct nf_conn *ct,
476 			   enum ip_conntrack_info ctinfo, s32 off);
477 };
478 extern const struct nfnl_ct_hook __rcu *nfnl_ct_hook;
479 
480 /**
481  * nf_skb_duplicated - TEE target has sent a packet
482  *
483  * When a xtables target sends a packet, the OUTPUT and POSTROUTING
484  * hooks are traversed again, i.e. nft and xtables are invoked recursively.
485  *
486  * This is used by xtables TEE target to prevent the duplicated skb from
487  * being duplicated again.
488  */
489 DECLARE_PER_CPU(bool, nf_skb_duplicated);
490 
491 #endif /*__LINUX_NETFILTER_H*/
492