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/module.h>
15#include <linux/netfilter_defs.h>
16#include <linux/netdevice.h>
17#include <linux/sockptr.h>
18#include <net/net_namespace.h>
19
20static inline int NF_DROP_GETERR(int verdict)
21{
22 return -(verdict >> NF_VERDICT_QBITS);
23}
24
25static __always_inline int
26NF_DROP_REASON(struct sk_buff *skb, enum skb_drop_reason reason, u32 err)
27{
28 BUILD_BUG_ON(err > 0xffff);
29
30 kfree_skb_reason(skb, reason);
31
32 return ((err << 16) | NF_STOLEN);
33}
34
35static inline int nf_inet_addr_cmp(const union nf_inet_addr *a1,
36 const union nf_inet_addr *a2)
37{
38#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
39 const unsigned long *ul1 = (const unsigned long *)a1;
40 const unsigned long *ul2 = (const unsigned long *)a2;
41
42 return ((ul1[0] ^ ul2[0]) | (ul1[1] ^ ul2[1])) == 0UL;
43#else
44 return a1->all[0] == a2->all[0] &&
45 a1->all[1] == a2->all[1] &&
46 a1->all[2] == a2->all[2] &&
47 a1->all[3] == a2->all[3];
48#endif
49}
50
51static inline void nf_inet_addr_mask(const union nf_inet_addr *a1,
52 union nf_inet_addr *result,
53 const union nf_inet_addr *mask)
54{
55#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
56 const unsigned long *ua = (const unsigned long *)a1;
57 unsigned long *ur = (unsigned long *)result;
58 const unsigned long *um = (const unsigned long *)mask;
59
60 ur[0] = ua[0] & um[0];
61 ur[1] = ua[1] & um[1];
62#else
63 result->all[0] = a1->all[0] & mask->all[0];
64 result->all[1] = a1->all[1] & mask->all[1];
65 result->all[2] = a1->all[2] & mask->all[2];
66 result->all[3] = a1->all[3] & mask->all[3];
67#endif
68}
69
70int netfilter_init(void);
71
72struct sk_buff;
73
74struct nf_hook_ops;
75
76struct sock;
77
78struct nf_hook_state {
79 u8 hook;
80 u8 pf;
81 struct net_device *in;
82 struct net_device *out;
83 struct sock *sk;
84 struct net *net;
85 int (*okfn)(struct net *, struct sock *, struct sk_buff *);
86};
87
88typedef unsigned int nf_hookfn(void *priv,
89 struct sk_buff *skb,
90 const struct nf_hook_state *state);
91enum nf_hook_ops_type {
92 NF_HOOK_OP_UNDEFINED,
93 NF_HOOK_OP_NF_TABLES,
94 NF_HOOK_OP_BPF,
95 NF_HOOK_OP_NFT_FT,
96};
97
98struct nf_hook_ops {
99 struct list_head list;
100 struct rcu_head rcu;
101
102 /* User fills in from here down. */
103 nf_hookfn *hook;
104 struct net_device *dev;
105 void *priv;
106 u8 pf;
107 enum nf_hook_ops_type hook_ops_type:8;
108 unsigned int hooknum;
109 /* Hooks are ordered in ascending priority. */
110 int priority;
111};
112
113struct nf_hook_entry {
114 nf_hookfn *hook;
115 void *priv;
116};
117
118struct nf_hook_entries_rcu_head {
119 struct rcu_head head;
120 void *allocation;
121};
122
123struct nf_hook_entries {
124 u16 num_hook_entries;
125 /* padding */
126 struct nf_hook_entry hooks[];
127
128 /* trailer: pointers to original orig_ops of each hook,
129 * followed by rcu_head and scratch space used for freeing
130 * the structure via call_rcu.
131 *
132 * This is not part of struct nf_hook_entry since its only
133 * needed in slow path (hook register/unregister):
134 * const struct nf_hook_ops *orig_ops[]
135 *
136 * For the same reason, we store this at end -- its
137 * only needed when a hook is deleted, not during
138 * packet path processing:
139 * struct nf_hook_entries_rcu_head head
140 */
141};
142
143#ifdef CONFIG_NETFILTER
144static inline struct nf_hook_ops **nf_hook_entries_get_hook_ops(const struct nf_hook_entries *e)
145{
146 unsigned int n = e->num_hook_entries;
147 const void *hook_end;
148
149 hook_end = &e->hooks[n]; /* this is *past* ->hooks[]! */
150
151 return (struct nf_hook_ops **)hook_end;
152}
153
154static inline int
155nf_hook_entry_hookfn(const struct nf_hook_entry *entry, struct sk_buff *skb,
156 struct nf_hook_state *state)
157{
158 return entry->hook(entry->priv, skb, state);
159}
160
161static inline void nf_hook_state_init(struct nf_hook_state *p,
162 unsigned int hook,
163 u_int8_t pf,
164 struct net_device *indev,
165 struct net_device *outdev,
166 struct sock *sk,
167 struct net *net,
168 int (*okfn)(struct net *, struct sock *, struct sk_buff *))
169{
170 p->hook = hook;
171 p->pf = pf;
172 p->in = indev;
173 p->out = outdev;
174 p->sk = sk;
175 p->net = net;
176 p->okfn = okfn;
177}
178
179
180
181struct nf_sockopt_ops {
182 struct list_head list;
183
184 u_int8_t pf;
185
186 /* Non-inclusive ranges: use 0/0/NULL to never get called. */
187 int set_optmin;
188 int set_optmax;
189 int (*set)(struct sock *sk, int optval, sockptr_t arg,
190 unsigned int len);
191 int get_optmin;
192 int get_optmax;
193 int (*get)(struct sock *sk, int optval, void __user *user, int *len);
194 /* Use the module struct to lock set/get code in place */
195 struct module *owner;
196};
197
198/* Function to register/unregister hook points. */
199int nf_register_net_hook(struct net *net, const struct nf_hook_ops *ops);
200void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *ops);
201int nf_register_net_hooks(struct net *net, const struct nf_hook_ops *reg,
202 unsigned int n);
203void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg,
204 unsigned int n);
205
206/* Functions to register get/setsockopt ranges (non-inclusive). You
207 need to check permissions yourself! */
208int nf_register_sockopt(struct nf_sockopt_ops *reg);
209void nf_unregister_sockopt(struct nf_sockopt_ops *reg);
210
211#ifdef CONFIG_JUMP_LABEL
212extern struct static_key nf_hooks_needed[NFPROTO_NUMPROTO][NF_MAX_HOOKS];
213#endif
214
215int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state,
216 const struct nf_hook_entries *e, unsigned int i);
217
218void nf_hook_slow_list(struct list_head *head, struct nf_hook_state *state,
219 const struct nf_hook_entries *e);
220/**
221 * nf_hook - call a netfilter hook
222 *
223 * Returns 1 if the hook has allowed the packet to pass. The function
224 * okfn must be invoked by the caller in this case. Any other return
225 * value indicates the packet has been consumed by the hook.
226 */
227static inline int nf_hook(u_int8_t pf, unsigned int hook, struct net *net,
228 struct sock *sk, struct sk_buff *skb,
229 struct net_device *indev, struct net_device *outdev,
230 int (*okfn)(struct net *, struct sock *, struct sk_buff *))
231{
232 struct nf_hook_entries *hook_head = NULL;
233 int ret = 1;
234
235#ifdef CONFIG_JUMP_LABEL
236 if (__builtin_constant_p(pf) &&
237 __builtin_constant_p(hook) &&
238 !static_key_false(key: &nf_hooks_needed[pf][hook]))
239 return 1;
240#endif
241
242 rcu_read_lock();
243 switch (pf) {
244 case NFPROTO_IPV4:
245 hook_head = rcu_dereference(net->nf.hooks_ipv4[hook]);
246 break;
247 case NFPROTO_IPV6:
248 hook_head = rcu_dereference(net->nf.hooks_ipv6[hook]);
249 break;
250 case NFPROTO_ARP:
251#ifdef CONFIG_NETFILTER_FAMILY_ARP
252 if (WARN_ON_ONCE(hook >= ARRAY_SIZE(net->nf.hooks_arp)))
253 break;
254 hook_head = rcu_dereference(net->nf.hooks_arp[hook]);
255#endif
256 break;
257 case NFPROTO_BRIDGE:
258#ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
259 hook_head = rcu_dereference(net->nf.hooks_bridge[hook]);
260#endif
261 break;
262 default:
263 WARN_ON_ONCE(1);
264 break;
265 }
266
267 if (hook_head) {
268 struct nf_hook_state state;
269
270 nf_hook_state_init(p: &state, hook, pf, indev, outdev,
271 sk, net, okfn);
272
273 ret = nf_hook_slow(skb, state: &state, e: hook_head, i: 0);
274 }
275 rcu_read_unlock();
276
277 return ret;
278}
279
280/* Activate hook; either okfn or kfree_skb called, unless a hook
281 returns NF_STOLEN (in which case, it's up to the hook to deal with
282 the consequences).
283
284 Returns -ERRNO if packet dropped. Zero means queued, stolen or
285 accepted.
286*/
287
288/* RR:
289 > I don't want nf_hook to return anything because people might forget
290 > about async and trust the return value to mean "packet was ok".
291
292 AK:
293 Just document it clearly, then you can expect some sense from kernel
294 coders :)
295*/
296
297static inline int
298NF_HOOK_COND(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
299 struct sk_buff *skb, struct net_device *in, struct net_device *out,
300 int (*okfn)(struct net *, struct sock *, struct sk_buff *),
301 bool cond)
302{
303 int ret;
304
305 if (!cond ||
306 ((ret = nf_hook(pf, hook, net, sk, skb, indev: in, outdev: out, okfn)) == 1))
307 ret = okfn(net, sk, skb);
308 return ret;
309}
310
311static inline int
312NF_HOOK(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk, struct sk_buff *skb,
313 struct net_device *in, struct net_device *out,
314 int (*okfn)(struct net *, struct sock *, struct sk_buff *))
315{
316 int ret = nf_hook(pf, hook, net, sk, skb, indev: in, outdev: out, okfn);
317 if (ret == 1)
318 ret = okfn(net, sk, skb);
319 return ret;
320}
321
322static inline void
323NF_HOOK_LIST(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
324 struct list_head *head, struct net_device *in, struct net_device *out,
325 int (*okfn)(struct net *, struct sock *, struct sk_buff *))
326{
327 struct nf_hook_entries *hook_head = NULL;
328
329#ifdef CONFIG_JUMP_LABEL
330 if (__builtin_constant_p(pf) &&
331 __builtin_constant_p(hook) &&
332 !static_key_false(key: &nf_hooks_needed[pf][hook]))
333 return;
334#endif
335
336 rcu_read_lock();
337 switch (pf) {
338 case NFPROTO_IPV4:
339 hook_head = rcu_dereference(net->nf.hooks_ipv4[hook]);
340 break;
341 case NFPROTO_IPV6:
342 hook_head = rcu_dereference(net->nf.hooks_ipv6[hook]);
343 break;
344 default:
345 WARN_ON_ONCE(1);
346 break;
347 }
348
349 if (hook_head) {
350 struct nf_hook_state state;
351
352 nf_hook_state_init(p: &state, hook, pf, indev: in, outdev: out, sk, net, okfn);
353
354 nf_hook_slow_list(head, state: &state, e: hook_head);
355 }
356 rcu_read_unlock();
357}
358
359/* Call setsockopt() */
360int nf_setsockopt(struct sock *sk, u_int8_t pf, int optval, sockptr_t opt,
361 unsigned int len);
362int nf_getsockopt(struct sock *sk, u_int8_t pf, int optval, char __user *opt,
363 int *len);
364
365struct flowi;
366struct nf_queue_entry;
367
368__sum16 nf_checksum(struct sk_buff *skb, unsigned int hook,
369 unsigned int dataoff, u_int8_t protocol,
370 unsigned short family);
371
372__sum16 nf_checksum_partial(struct sk_buff *skb, unsigned int hook,
373 unsigned int dataoff, unsigned int len,
374 u_int8_t protocol, unsigned short family);
375int nf_route(struct net *net, struct dst_entry **dst, struct flowi *fl,
376 bool strict, unsigned short family);
377
378#include <net/flow.h>
379
380struct nf_conn;
381enum nf_nat_manip_type;
382struct nlattr;
383
384struct nf_nat_hook {
385 int (*parse_nat_setup)(struct nf_conn *ct, enum nf_nat_manip_type manip,
386 const struct nlattr *attr);
387 void (*decode_session)(struct sk_buff *skb, struct flowi *fl);
388 void (*remove_nat_bysrc)(struct nf_conn *ct);
389};
390
391extern const struct nf_nat_hook __rcu *nf_nat_hook;
392
393static inline void
394nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl, u_int8_t family)
395{
396#if IS_ENABLED(CONFIG_NF_NAT)
397 const struct nf_nat_hook *nat_hook;
398
399 rcu_read_lock();
400 nat_hook = rcu_dereference(nf_nat_hook);
401 if (nat_hook && nat_hook->decode_session)
402 nat_hook->decode_session(skb, fl);
403 rcu_read_unlock();
404#endif
405}
406
407#else /* !CONFIG_NETFILTER */
408static inline int
409NF_HOOK_COND(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
410 struct sk_buff *skb, struct net_device *in, struct net_device *out,
411 int (*okfn)(struct net *, struct sock *, struct sk_buff *),
412 bool cond)
413{
414 return okfn(net, sk, skb);
415}
416
417static inline int
418NF_HOOK(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
419 struct sk_buff *skb, struct net_device *in, struct net_device *out,
420 int (*okfn)(struct net *, struct sock *, struct sk_buff *))
421{
422 return okfn(net, sk, skb);
423}
424
425static inline void
426NF_HOOK_LIST(uint8_t pf, unsigned int hook, struct net *net, struct sock *sk,
427 struct list_head *head, struct net_device *in, struct net_device *out,
428 int (*okfn)(struct net *, struct sock *, struct sk_buff *))
429{
430 /* nothing to do */
431}
432
433static inline int nf_hook(u_int8_t pf, unsigned int hook, struct net *net,
434 struct sock *sk, struct sk_buff *skb,
435 struct net_device *indev, struct net_device *outdev,
436 int (*okfn)(struct net *, struct sock *, struct sk_buff *))
437{
438 return 1;
439}
440struct flowi;
441static inline void
442nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl, u_int8_t family)
443{
444}
445#endif /*CONFIG_NETFILTER*/
446
447#if IS_ENABLED(CONFIG_NF_CONNTRACK)
448#include <linux/netfilter/nf_conntrack_zones_common.h>
449
450void nf_ct_attach(struct sk_buff *, const struct sk_buff *);
451void nf_ct_set_closing(struct nf_conntrack *nfct);
452struct nf_conntrack_tuple;
453bool nf_ct_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
454 const struct sk_buff *skb);
455#else
456static inline void nf_ct_attach(struct sk_buff *new, struct sk_buff *skb) {}
457static inline void nf_ct_set_closing(struct nf_conntrack *nfct) {}
458struct nf_conntrack_tuple;
459static inline bool nf_ct_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
460 const struct sk_buff *skb)
461{
462 return false;
463}
464#endif
465
466struct nf_conn;
467enum ip_conntrack_info;
468
469struct nf_ct_hook {
470 int (*update)(struct net *net, struct sk_buff *skb);
471 void (*destroy)(struct nf_conntrack *);
472 bool (*get_tuple_skb)(struct nf_conntrack_tuple *,
473 const struct sk_buff *);
474 void (*attach)(struct sk_buff *nskb, const struct sk_buff *skb);
475 void (*set_closing)(struct nf_conntrack *nfct);
476 int (*confirm)(struct sk_buff *skb);
477 u32 (*get_id)(const struct nf_conntrack *nfct);
478};
479extern const struct nf_ct_hook __rcu *nf_ct_hook;
480
481struct nlattr;
482
483struct nfnl_ct_hook {
484 size_t (*build_size)(const struct nf_conn *ct);
485 int (*build)(struct sk_buff *skb, struct nf_conn *ct,
486 enum ip_conntrack_info ctinfo,
487 u_int16_t ct_attr, u_int16_t ct_info_attr);
488 int (*parse)(const struct nlattr *attr, struct nf_conn *ct);
489 int (*attach_expect)(const struct nlattr *attr, struct nf_conn *ct,
490 u32 portid, u32 report);
491 void (*seq_adjust)(struct sk_buff *skb, struct nf_conn *ct,
492 enum ip_conntrack_info ctinfo, s32 off);
493};
494extern const struct nfnl_ct_hook __rcu *nfnl_ct_hook;
495
496struct nf_defrag_hook {
497 struct module *owner;
498 int (*enable)(struct net *net);
499 void (*disable)(struct net *net);
500};
501
502extern const struct nf_defrag_hook __rcu *nf_defrag_v4_hook;
503extern const struct nf_defrag_hook __rcu *nf_defrag_v6_hook;
504
505/*
506 * Contains bitmask of ctnetlink event subscribers, if any.
507 * Can't be pernet due to NETLINK_LISTEN_ALL_NSID setsockopt flag.
508 */
509extern u8 nf_ctnetlink_has_listener;
510#endif /*__LINUX_NETFILTER_H*/
511