include/linux/netfilter.h

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