xref: /linux/include/linux/netfilter.h (revision 0be3ff0c) 
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 #if IS_ENABLED(CONFIG_DECNET)
247 	case NFPROTO_DECNET:
248 		hook_head = rcu_dereference(net->nf.hooks_decnet[hook]);
249 		break;
250 #endif
251 	default:
252 		WARN_ON_ONCE(1);
253 		break;
254 	}
255 
256 	if (hook_head) {
257 		struct nf_hook_state state;
258 
259 		nf_hook_state_init(&state, hook, pf, indev, outdev,
260 				   sk, net, okfn);
261 
262 		ret = nf_hook_slow(skb, &state, hook_head, 0);
263 	}
264 	rcu_read_unlock();
265 
266 	return ret;
267 }
268 
269 /* Activate hook; either okfn or kfree_skb called, unless a hook
270    returns NF_STOLEN (in which case, it's up to the hook to deal with
271    the consequences).
272 
273    Returns -ERRNO if packet dropped.  Zero means queued, stolen or
274    accepted.
275 */
276 
277 /* RR:
278    > I don't want nf_hook to return anything because people might forget
279    > about async and trust the return value to mean "packet was ok".
280 
281    AK:
282    Just document it clearly, then you can expect some sense from kernel
283    coders :)
284 */
285 
286 static inline int
287 NF_HOOK_COND(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
288 	     struct sk_buff *skb, struct net_device *in, struct net_device *out,
289 	     int (*okfn)(struct net *, struct sock *, struct sk_buff *),
290 	     bool cond)
291 {
292 	int ret;
293 
294 	if (!cond ||
295 	    ((ret = nf_hook(pf, hook, net, sk, skb, in, out, okfn)) == 1))
296 		ret = okfn(net, sk, skb);
297 	return ret;
298 }
299 
300 static inline int
301 NF_HOOK(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk, struct sk_buff *skb,
302 	struct net_device *in, struct net_device *out,
303 	int (*okfn)(struct net *, struct sock *, struct sk_buff *))
304 {
305 	int ret = nf_hook(pf, hook, net, sk, skb, in, out, okfn);
306 	if (ret == 1)
307 		ret = okfn(net, sk, skb);
308 	return ret;
309 }
310 
311 static inline void
312 NF_HOOK_LIST(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
313 	     struct list_head *head, struct net_device *in, struct net_device *out,
314 	     int (*okfn)(struct net *, struct sock *, struct sk_buff *))
315 {
316 	struct nf_hook_entries *hook_head = NULL;
317 
318 #ifdef CONFIG_JUMP_LABEL
319 	if (__builtin_constant_p(pf) &&
320 	    __builtin_constant_p(hook) &&
321 	    !static_key_false(&nf_hooks_needed[pf][hook]))
322 		return;
323 #endif
324 
325 	rcu_read_lock();
326 	switch (pf) {
327 	case NFPROTO_IPV4:
328 		hook_head = rcu_dereference(net->nf.hooks_ipv4[hook]);
329 		break;
330 	case NFPROTO_IPV6:
331 		hook_head = rcu_dereference(net->nf.hooks_ipv6[hook]);
332 		break;
333 	default:
334 		WARN_ON_ONCE(1);
335 		break;
336 	}
337 
338 	if (hook_head) {
339 		struct nf_hook_state state;
340 
341 		nf_hook_state_init(&state, hook, pf, in, out, sk, net, okfn);
342 
343 		nf_hook_slow_list(head, &state, hook_head);
344 	}
345 	rcu_read_unlock();
346 }
347 
348 /* Call setsockopt() */
349 int nf_setsockopt(struct sock *sk, u_int8_t pf, int optval, sockptr_t opt,
350 		  unsigned int len);
351 int nf_getsockopt(struct sock *sk, u_int8_t pf, int optval, char __user *opt,
352 		  int *len);
353 
354 struct flowi;
355 struct nf_queue_entry;
356 
357 __sum16 nf_checksum(struct sk_buff *skb, unsigned int hook,
358 		    unsigned int dataoff, u_int8_t protocol,
359 		    unsigned short family);
360 
361 __sum16 nf_checksum_partial(struct sk_buff *skb, unsigned int hook,
362 			    unsigned int dataoff, unsigned int len,
363 			    u_int8_t protocol, unsigned short family);
364 int nf_route(struct net *net, struct dst_entry **dst, struct flowi *fl,
365 	     bool strict, unsigned short family);
366 int nf_reroute(struct sk_buff *skb, struct nf_queue_entry *entry);
367 
368 #include <net/flow.h>
369 
370 struct nf_conn;
371 enum nf_nat_manip_type;
372 struct nlattr;
373 enum ip_conntrack_dir;
374 
375 struct nf_nat_hook {
376 	int (*parse_nat_setup)(struct nf_conn *ct, enum nf_nat_manip_type manip,
377 			       const struct nlattr *attr);
378 	void (*decode_session)(struct sk_buff *skb, struct flowi *fl);
379 	unsigned int (*manip_pkt)(struct sk_buff *skb, struct nf_conn *ct,
380 				  enum nf_nat_manip_type mtype,
381 				  enum ip_conntrack_dir dir);
382 	void (*remove_nat_bysrc)(struct nf_conn *ct);
383 };
384 
385 extern const struct nf_nat_hook __rcu *nf_nat_hook;
386 
387 static inline void
388 nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl, u_int8_t family)
389 {
390 #if IS_ENABLED(CONFIG_NF_NAT)
391 	const struct nf_nat_hook *nat_hook;
392 
393 	rcu_read_lock();
394 	nat_hook = rcu_dereference(nf_nat_hook);
395 	if (nat_hook && nat_hook->decode_session)
396 		nat_hook->decode_session(skb, fl);
397 	rcu_read_unlock();
398 #endif
399 }
400 
401 #else /* !CONFIG_NETFILTER */
402 static inline int
403 NF_HOOK_COND(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
404 	     struct sk_buff *skb, struct net_device *in, struct net_device *out,
405 	     int (*okfn)(struct net *, struct sock *, struct sk_buff *),
406 	     bool cond)
407 {
408 	return okfn(net, sk, skb);
409 }
410 
411 static inline int
412 NF_HOOK(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
413 	struct sk_buff *skb, struct net_device *in, struct net_device *out,
414 	int (*okfn)(struct net *, struct sock *, struct sk_buff *))
415 {
416 	return okfn(net, sk, skb);
417 }
418 
419 static inline void
420 NF_HOOK_LIST(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
421 	     struct list_head *head, struct net_device *in, struct net_device *out,
422 	     int (*okfn)(struct net *, struct sock *, struct sk_buff *))
423 {
424 	/* nothing to do */
425 }
426 
427 static inline int nf_hook(u_int8_t pf, unsigned int hook, struct net *net,
428 			  struct sock *sk, struct sk_buff *skb,
429 			  struct net_device *indev, struct net_device *outdev,
430 			  int (*okfn)(struct net *, struct sock *, struct sk_buff *))
431 {
432 	return 1;
433 }
434 struct flowi;
435 static inline void
436 nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl, u_int8_t family)
437 {
438 }
439 #endif /*CONFIG_NETFILTER*/
440 
441 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
442 #include <linux/netfilter/nf_conntrack_zones_common.h>
443 
444 void nf_ct_attach(struct sk_buff *, const struct sk_buff *);
445 struct nf_conntrack_tuple;
446 bool nf_ct_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
447 			 const struct sk_buff *skb);
448 #else
449 static inline void nf_ct_attach(struct sk_buff *new, struct sk_buff *skb) {}
450 struct nf_conntrack_tuple;
451 static inline bool nf_ct_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
452 				       const struct sk_buff *skb)
453 {
454 	return false;
455 }
456 #endif
457 
458 struct nf_conn;
459 enum ip_conntrack_info;
460 
461 struct nf_ct_hook {
462 	int (*update)(struct net *net, struct sk_buff *skb);
463 	void (*destroy)(struct nf_conntrack *);
464 	bool (*get_tuple_skb)(struct nf_conntrack_tuple *,
465 			      const struct sk_buff *);
466 	void (*attach)(struct sk_buff *nskb, const struct sk_buff *skb);
467 };
468 extern const struct nf_ct_hook __rcu *nf_ct_hook;
469 
470 struct nlattr;
471 
472 struct nfnl_ct_hook {
473 	size_t (*build_size)(const struct nf_conn *ct);
474 	int (*build)(struct sk_buff *skb, struct nf_conn *ct,
475 		     enum ip_conntrack_info ctinfo,
476 		     u_int16_t ct_attr, u_int16_t ct_info_attr);
477 	int (*parse)(const struct nlattr *attr, struct nf_conn *ct);
478 	int (*attach_expect)(const struct nlattr *attr, struct nf_conn *ct,
479 			     u32 portid, u32 report);
480 	void (*seq_adjust)(struct sk_buff *skb, struct nf_conn *ct,
481 			   enum ip_conntrack_info ctinfo, s32 off);
482 };
483 extern const struct nfnl_ct_hook __rcu *nfnl_ct_hook;
484 
485 /**
486  * nf_skb_duplicated - TEE target has sent a packet
487  *
488  * When a xtables target sends a packet, the OUTPUT and POSTROUTING
489  * hooks are traversed again, i.e. nft and xtables are invoked recursively.
490  *
491  * This is used by xtables TEE target to prevent the duplicated skb from
492  * being duplicated again.
493  */
494 DECLARE_PER_CPU(bool, nf_skb_duplicated);
495 
496 #endif /*__LINUX_NETFILTER_H*/
497