xref: /linux-6.15/include/linux/netfilter.h (revision 77d4df41) 
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 <net/net_namespace.h>
17 
18 static inline int NF_DROP_GETERR(int verdict)
19 {
20 	return -(verdict >> NF_VERDICT_QBITS);
21 }
22 
23 static inline int nf_inet_addr_cmp(const union nf_inet_addr *a1,
24 				   const union nf_inet_addr *a2)
25 {
26 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
27 	const unsigned long *ul1 = (const unsigned long *)a1;
28 	const unsigned long *ul2 = (const unsigned long *)a2;
29 
30 	return ((ul1[0] ^ ul2[0]) | (ul1[1] ^ ul2[1])) == 0UL;
31 #else
32 	return a1->all[0] == a2->all[0] &&
33 	       a1->all[1] == a2->all[1] &&
34 	       a1->all[2] == a2->all[2] &&
35 	       a1->all[3] == a2->all[3];
36 #endif
37 }
38 
39 static inline void nf_inet_addr_mask(const union nf_inet_addr *a1,
40 				     union nf_inet_addr *result,
41 				     const union nf_inet_addr *mask)
42 {
43 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
44 	const unsigned long *ua = (const unsigned long *)a1;
45 	unsigned long *ur = (unsigned long *)result;
46 	const unsigned long *um = (const unsigned long *)mask;
47 
48 	ur[0] = ua[0] & um[0];
49 	ur[1] = ua[1] & um[1];
50 #else
51 	result->all[0] = a1->all[0] & mask->all[0];
52 	result->all[1] = a1->all[1] & mask->all[1];
53 	result->all[2] = a1->all[2] & mask->all[2];
54 	result->all[3] = a1->all[3] & mask->all[3];
55 #endif
56 }
57 
58 int netfilter_init(void);
59 
60 struct sk_buff;
61 
62 struct nf_hook_ops;
63 
64 struct sock;
65 
66 struct nf_hook_state {
67 	unsigned int hook;
68 	u_int8_t pf;
69 	struct net_device *in;
70 	struct net_device *out;
71 	struct sock *sk;
72 	struct net *net;
73 	int (*okfn)(struct net *, struct sock *, struct sk_buff *);
74 };
75 
76 typedef unsigned int nf_hookfn(void *priv,
77 			       struct sk_buff *skb,
78 			       const struct nf_hook_state *state);
79 struct nf_hook_ops {
80 	/* User fills in from here down. */
81 	nf_hookfn		*hook;
82 	struct net_device	*dev;
83 	void			*priv;
84 	u_int8_t		pf;
85 	unsigned int		hooknum;
86 	/* Hooks are ordered in ascending priority. */
87 	int			priority;
88 };
89 
90 struct nf_hook_entry {
91 	nf_hookfn			*hook;
92 	void				*priv;
93 };
94 
95 struct nf_hook_entries_rcu_head {
96 	struct rcu_head head;
97 	void	*allocation;
98 };
99 
100 struct nf_hook_entries {
101 	u16				num_hook_entries;
102 	/* padding */
103 	struct nf_hook_entry		hooks[];
104 
105 	/* trailer: pointers to original orig_ops of each hook,
106 	 * followed by rcu_head and scratch space used for freeing
107 	 * the structure via call_rcu.
108 	 *
109 	 *   This is not part of struct nf_hook_entry since its only
110 	 *   needed in slow path (hook register/unregister):
111 	 * const struct nf_hook_ops     *orig_ops[]
112 	 *
113 	 *   For the same reason, we store this at end -- its
114 	 *   only needed when a hook is deleted, not during
115 	 *   packet path processing:
116 	 * struct nf_hook_entries_rcu_head     head
117 	 */
118 };
119 
120 #ifdef CONFIG_NETFILTER
121 static inline struct nf_hook_ops **nf_hook_entries_get_hook_ops(const struct nf_hook_entries *e)
122 {
123 	unsigned int n = e->num_hook_entries;
124 	const void *hook_end;
125 
126 	hook_end = &e->hooks[n]; /* this is *past* ->hooks[]! */
127 
128 	return (struct nf_hook_ops **)hook_end;
129 }
130 
131 static inline int
132 nf_hook_entry_hookfn(const struct nf_hook_entry *entry, struct sk_buff *skb,
133 		     struct nf_hook_state *state)
134 {
135 	return entry->hook(entry->priv, skb, state);
136 }
137 
138 static inline void nf_hook_state_init(struct nf_hook_state *p,
139 				      unsigned int hook,
140 				      u_int8_t pf,
141 				      struct net_device *indev,
142 				      struct net_device *outdev,
143 				      struct sock *sk,
144 				      struct net *net,
145 				      int (*okfn)(struct net *, struct sock *, struct sk_buff *))
146 {
147 	p->hook = hook;
148 	p->pf = pf;
149 	p->in = indev;
150 	p->out = outdev;
151 	p->sk = sk;
152 	p->net = net;
153 	p->okfn = okfn;
154 }
155 
156 
157 
158 struct nf_sockopt_ops {
159 	struct list_head list;
160 
161 	u_int8_t pf;
162 
163 	/* Non-inclusive ranges: use 0/0/NULL to never get called. */
164 	int set_optmin;
165 	int set_optmax;
166 	int (*set)(struct sock *sk, int optval, void __user *user, unsigned int len);
167 	int get_optmin;
168 	int get_optmax;
169 	int (*get)(struct sock *sk, int optval, void __user *user, int *len);
170 	/* Use the module struct to lock set/get code in place */
171 	struct module *owner;
172 };
173 
174 /* Function to register/unregister hook points. */
175 int nf_register_net_hook(struct net *net, const struct nf_hook_ops *ops);
176 void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *ops);
177 int nf_register_net_hooks(struct net *net, const struct nf_hook_ops *reg,
178 			  unsigned int n);
179 void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg,
180 			     unsigned int n);
181 
182 /* Functions to register get/setsockopt ranges (non-inclusive).  You
183    need to check permissions yourself! */
184 int nf_register_sockopt(struct nf_sockopt_ops *reg);
185 void nf_unregister_sockopt(struct nf_sockopt_ops *reg);
186 
187 #ifdef CONFIG_JUMP_LABEL
188 extern struct static_key nf_hooks_needed[NFPROTO_NUMPROTO][NF_MAX_HOOKS];
189 #endif
190 
191 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state,
192 		 const struct nf_hook_entries *e, unsigned int i);
193 
194 void nf_hook_slow_list(struct list_head *head, struct nf_hook_state *state,
195 		       const struct nf_hook_entries *e);
196 /**
197  *	nf_hook - call a netfilter hook
198  *
199  *	Returns 1 if the hook has allowed the packet to pass.  The function
200  *	okfn must be invoked by the caller in this case.  Any other return
201  *	value indicates the packet has been consumed by the hook.
202  */
203 static inline int nf_hook(u_int8_t pf, unsigned int hook, struct net *net,
204 			  struct sock *sk, struct sk_buff *skb,
205 			  struct net_device *indev, struct net_device *outdev,
206 			  int (*okfn)(struct net *, struct sock *, struct sk_buff *))
207 {
208 	struct nf_hook_entries *hook_head = NULL;
209 	int ret = 1;
210 
211 #ifdef CONFIG_JUMP_LABEL
212 	if (__builtin_constant_p(pf) &&
213 	    __builtin_constant_p(hook) &&
214 	    !static_key_false(&nf_hooks_needed[pf][hook]))
215 		return 1;
216 #endif
217 
218 	rcu_read_lock();
219 	switch (pf) {
220 	case NFPROTO_IPV4:
221 		hook_head = rcu_dereference(net->nf.hooks_ipv4[hook]);
222 		break;
223 	case NFPROTO_IPV6:
224 		hook_head = rcu_dereference(net->nf.hooks_ipv6[hook]);
225 		break;
226 	case NFPROTO_ARP:
227 #ifdef CONFIG_NETFILTER_FAMILY_ARP
228 		if (WARN_ON_ONCE(hook >= ARRAY_SIZE(net->nf.hooks_arp)))
229 			break;
230 		hook_head = rcu_dereference(net->nf.hooks_arp[hook]);
231 #endif
232 		break;
233 	case NFPROTO_BRIDGE:
234 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
235 		hook_head = rcu_dereference(net->nf.hooks_bridge[hook]);
236 #endif
237 		break;
238 #if IS_ENABLED(CONFIG_DECNET)
239 	case NFPROTO_DECNET:
240 		hook_head = rcu_dereference(net->nf.hooks_decnet[hook]);
241 		break;
242 #endif
243 	default:
244 		WARN_ON_ONCE(1);
245 		break;
246 	}
247 
248 	if (hook_head) {
249 		struct nf_hook_state state;
250 
251 		nf_hook_state_init(&state, hook, pf, indev, outdev,
252 				   sk, net, okfn);
253 
254 		ret = nf_hook_slow(skb, &state, hook_head, 0);
255 	}
256 	rcu_read_unlock();
257 
258 	return ret;
259 }
260 
261 /* Activate hook; either okfn or kfree_skb called, unless a hook
262    returns NF_STOLEN (in which case, it's up to the hook to deal with
263    the consequences).
264 
265    Returns -ERRNO if packet dropped.  Zero means queued, stolen or
266    accepted.
267 */
268 
269 /* RR:
270    > I don't want nf_hook to return anything because people might forget
271    > about async and trust the return value to mean "packet was ok".
272 
273    AK:
274    Just document it clearly, then you can expect some sense from kernel
275    coders :)
276 */
277 
278 static inline int
279 NF_HOOK_COND(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
280 	     struct sk_buff *skb, struct net_device *in, struct net_device *out,
281 	     int (*okfn)(struct net *, struct sock *, struct sk_buff *),
282 	     bool cond)
283 {
284 	int ret;
285 
286 	if (!cond ||
287 	    ((ret = nf_hook(pf, hook, net, sk, skb, in, out, okfn)) == 1))
288 		ret = okfn(net, sk, skb);
289 	return ret;
290 }
291 
292 static inline int
293 NF_HOOK(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk, struct sk_buff *skb,
294 	struct net_device *in, struct net_device *out,
295 	int (*okfn)(struct net *, struct sock *, struct sk_buff *))
296 {
297 	int ret = nf_hook(pf, hook, net, sk, skb, in, out, okfn);
298 	if (ret == 1)
299 		ret = okfn(net, sk, skb);
300 	return ret;
301 }
302 
303 static inline void
304 NF_HOOK_LIST(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
305 	     struct list_head *head, struct net_device *in, struct net_device *out,
306 	     int (*okfn)(struct net *, struct sock *, struct sk_buff *))
307 {
308 	struct nf_hook_entries *hook_head = NULL;
309 
310 #ifdef CONFIG_JUMP_LABEL
311 	if (__builtin_constant_p(pf) &&
312 	    __builtin_constant_p(hook) &&
313 	    !static_key_false(&nf_hooks_needed[pf][hook]))
314 		return;
315 #endif
316 
317 	rcu_read_lock();
318 	switch (pf) {
319 	case NFPROTO_IPV4:
320 		hook_head = rcu_dereference(net->nf.hooks_ipv4[hook]);
321 		break;
322 	case NFPROTO_IPV6:
323 		hook_head = rcu_dereference(net->nf.hooks_ipv6[hook]);
324 		break;
325 	default:
326 		WARN_ON_ONCE(1);
327 		break;
328 	}
329 
330 	if (hook_head) {
331 		struct nf_hook_state state;
332 
333 		nf_hook_state_init(&state, hook, pf, in, out, sk, net, okfn);
334 
335 		nf_hook_slow_list(head, &state, hook_head);
336 	}
337 	rcu_read_unlock();
338 }
339 
340 /* Call setsockopt() */
341 int nf_setsockopt(struct sock *sk, u_int8_t pf, int optval, char __user *opt,
342 		  unsigned int len);
343 int nf_getsockopt(struct sock *sk, u_int8_t pf, int optval, char __user *opt,
344 		  int *len);
345 
346 struct flowi;
347 struct nf_queue_entry;
348 
349 __sum16 nf_checksum(struct sk_buff *skb, unsigned int hook,
350 		    unsigned int dataoff, u_int8_t protocol,
351 		    unsigned short family);
352 
353 __sum16 nf_checksum_partial(struct sk_buff *skb, unsigned int hook,
354 			    unsigned int dataoff, unsigned int len,
355 			    u_int8_t protocol, unsigned short family);
356 int nf_route(struct net *net, struct dst_entry **dst, struct flowi *fl,
357 	     bool strict, unsigned short family);
358 int nf_reroute(struct sk_buff *skb, struct nf_queue_entry *entry);
359 
360 #include <net/flow.h>
361 
362 struct nf_conn;
363 enum nf_nat_manip_type;
364 struct nlattr;
365 enum ip_conntrack_dir;
366 
367 struct nf_nat_hook {
368 	int (*parse_nat_setup)(struct nf_conn *ct, enum nf_nat_manip_type manip,
369 			       const struct nlattr *attr);
370 	void (*decode_session)(struct sk_buff *skb, struct flowi *fl);
371 	unsigned int (*manip_pkt)(struct sk_buff *skb, struct nf_conn *ct,
372 				  enum nf_nat_manip_type mtype,
373 				  enum ip_conntrack_dir dir);
374 };
375 
376 extern struct nf_nat_hook __rcu *nf_nat_hook;
377 
378 static inline void
379 nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl, u_int8_t family)
380 {
381 #if IS_ENABLED(CONFIG_NF_NAT)
382 	struct nf_nat_hook *nat_hook;
383 
384 	rcu_read_lock();
385 	nat_hook = rcu_dereference(nf_nat_hook);
386 	if (nat_hook && nat_hook->decode_session)
387 		nat_hook->decode_session(skb, fl);
388 	rcu_read_unlock();
389 #endif
390 }
391 
392 #else /* !CONFIG_NETFILTER */
393 static inline int
394 NF_HOOK_COND(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
395 	     struct sk_buff *skb, struct net_device *in, struct net_device *out,
396 	     int (*okfn)(struct net *, struct sock *, struct sk_buff *),
397 	     bool cond)
398 {
399 	return okfn(net, sk, skb);
400 }
401 
402 static inline int
403 NF_HOOK(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 {
407 	return okfn(net, sk, skb);
408 }
409 
410 static inline void
411 NF_HOOK_LIST(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
412 	     struct list_head *head, struct net_device *in, struct net_device *out,
413 	     int (*okfn)(struct net *, struct sock *, struct sk_buff *))
414 {
415 	/* nothing to do */
416 }
417 
418 static inline int nf_hook(u_int8_t pf, unsigned int hook, struct net *net,
419 			  struct sock *sk, struct sk_buff *skb,
420 			  struct net_device *indev, struct net_device *outdev,
421 			  int (*okfn)(struct net *, struct sock *, struct sk_buff *))
422 {
423 	return 1;
424 }
425 struct flowi;
426 static inline void
427 nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl, u_int8_t family)
428 {
429 }
430 #endif /*CONFIG_NETFILTER*/
431 
432 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
433 #include <linux/netfilter/nf_conntrack_zones_common.h>
434 
435 extern void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *) __rcu;
436 void nf_ct_attach(struct sk_buff *, const struct sk_buff *);
437 struct nf_conntrack_tuple;
438 bool nf_ct_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
439 			 const struct sk_buff *skb);
440 #else
441 static inline void nf_ct_attach(struct sk_buff *new, struct sk_buff *skb) {}
442 struct nf_conntrack_tuple;
443 static inline bool nf_ct_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
444 				       const struct sk_buff *skb)
445 {
446 	return false;
447 }
448 #endif
449 
450 struct nf_conn;
451 enum ip_conntrack_info;
452 
453 struct nf_ct_hook {
454 	int (*update)(struct net *net, struct sk_buff *skb);
455 	void (*destroy)(struct nf_conntrack *);
456 	bool (*get_tuple_skb)(struct nf_conntrack_tuple *,
457 			      const struct sk_buff *);
458 };
459 extern struct nf_ct_hook __rcu *nf_ct_hook;
460 
461 struct nlattr;
462 
463 struct nfnl_ct_hook {
464 	struct nf_conn *(*get_ct)(const struct sk_buff *skb,
465 				  enum ip_conntrack_info *ctinfo);
466 	size_t (*build_size)(const struct nf_conn *ct);
467 	int (*build)(struct sk_buff *skb, struct nf_conn *ct,
468 		     enum ip_conntrack_info ctinfo,
469 		     u_int16_t ct_attr, u_int16_t ct_info_attr);
470 	int (*parse)(const struct nlattr *attr, struct nf_conn *ct);
471 	int (*attach_expect)(const struct nlattr *attr, struct nf_conn *ct,
472 			     u32 portid, u32 report);
473 	void (*seq_adjust)(struct sk_buff *skb, struct nf_conn *ct,
474 			   enum ip_conntrack_info ctinfo, s32 off);
475 };
476 extern struct nfnl_ct_hook __rcu *nfnl_ct_hook;
477 
478 /**
479  * nf_skb_duplicated - TEE target has sent a packet
480  *
481  * When a xtables target sends a packet, the OUTPUT and POSTROUTING
482  * hooks are traversed again, i.e. nft and xtables are invoked recursively.
483  *
484  * This is used by xtables TEE target to prevent the duplicated skb from
485  * being duplicated again.
486  */
487 DECLARE_PER_CPU(bool, nf_skb_duplicated);
488 
489 #endif /*__LINUX_NETFILTER_H*/
490