1/* SPDX-License-Identifier: GPL-2.0 */ 2/* 3 * Operations on the network namespace 4 */ 5#ifndef __NET_NET_NAMESPACE_H 6#define __NET_NET_NAMESPACE_H 7 8#include <linux/atomic.h> 9#include <linux/refcount.h> 10#include <linux/workqueue.h> 11#include <linux/list.h> 12#include <linux/sysctl.h> 13#include <linux/uidgid.h> 14 15#include <net/flow.h> 16#include <net/netns/core.h> 17#include <net/netns/mib.h> 18#include <net/netns/unix.h> 19#include <net/netns/packet.h> 20#include <net/netns/ipv4.h> 21#include <net/netns/ipv6.h> 22#include <net/netns/nexthop.h> 23#include <net/netns/ieee802154_6lowpan.h> 24#include <net/netns/sctp.h> 25#include <net/netns/dccp.h> 26#include <net/netns/netfilter.h> 27#include <net/netns/x_tables.h> 28#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 29#include <net/netns/conntrack.h> 30#endif 31#include <net/netns/nftables.h> 32#include <net/netns/xfrm.h> 33#include <net/netns/mpls.h> 34#include <net/netns/can.h> 35#include <net/netns/xdp.h> 36#include <net/netns/bpf.h> 37#include <linux/ns_common.h> 38#include <linux/idr.h> 39#include <linux/skbuff.h> 40#include <linux/notifier.h> 41#ifdef CONFIG_NEWIP 42#include <net/netns/nip.h> 43#endif 44 45struct user_namespace; 46struct proc_dir_entry; 47struct net_device; 48struct sock; 49struct ctl_table_header; 50struct net_generic; 51struct uevent_sock; 52struct netns_ipvs; 53struct bpf_prog; 54 55 56#define NETDEV_HASHBITS 8 57#define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS) 58 59struct net { 60 /* First cache line can be often dirtied. 61 * Do not place here read-mostly fields. 62 */ 63 refcount_t passive; /* To decide when the network 64 * namespace should be freed. 65 */ 66 refcount_t count; /* To decided when the network 67 * namespace should be shut down. 68 */ 69 spinlock_t rules_mod_lock; 70 71 unsigned int dev_unreg_count; 72 73 unsigned int dev_base_seq; /* protected by rtnl_mutex */ 74 int ifindex; 75 76 spinlock_t nsid_lock; 77 atomic_t fnhe_genid; 78 79 struct list_head list; /* list of network namespaces */ 80 struct list_head exit_list; /* To linked to call pernet exit 81 * methods on dead net ( 82 * pernet_ops_rwsem read locked), 83 * or to unregister pernet ops 84 * (pernet_ops_rwsem write locked). 85 */ 86 struct llist_node cleanup_list; /* namespaces on death row */ 87 88#ifdef CONFIG_KEYS 89 struct key_tag *key_domain; /* Key domain of operation tag */ 90#endif 91 struct user_namespace *user_ns; /* Owning user namespace */ 92 struct ucounts *ucounts; 93 struct idr netns_ids; 94 95 struct ns_common ns; 96 97 struct list_head dev_base_head; 98 struct proc_dir_entry *proc_net; 99 struct proc_dir_entry *proc_net_stat; 100 101#ifdef CONFIG_SYSCTL 102 struct ctl_table_set sysctls; 103#endif 104 105 struct sock *rtnl; /* rtnetlink socket */ 106 struct sock *genl_sock; 107 108 struct uevent_sock *uevent_sock; /* uevent socket */ 109 110 struct hlist_head *dev_name_head; 111 struct hlist_head *dev_index_head; 112 struct raw_notifier_head netdev_chain; 113 114 /* Note that @hash_mix can be read millions times per second, 115 * it is critical that it is on a read_mostly cache line. 116 */ 117 u32 hash_mix; 118 119 struct net_device *loopback_dev; /* The loopback */ 120 121 /* core fib_rules */ 122 struct list_head rules_ops; 123 124 struct netns_core core; 125 struct netns_mib mib; 126 struct netns_packet packet; 127 struct netns_unix unx; 128 struct netns_nexthop nexthop; 129 struct netns_ipv4 ipv4; 130#if IS_ENABLED(CONFIG_IPV6) 131 struct netns_ipv6 ipv6; 132#endif 133#if IS_ENABLED(CONFIG_NEWIP) 134 struct netns_newip newip; /* NIP */ 135#endif 136#if IS_ENABLED(CONFIG_IEEE802154_6LOWPAN) 137 struct netns_ieee802154_lowpan ieee802154_lowpan; 138#endif 139#if defined(CONFIG_IP_SCTP) || defined(CONFIG_IP_SCTP_MODULE) 140 struct netns_sctp sctp; 141#endif 142#if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE) 143 struct netns_dccp dccp; 144#endif 145#ifdef CONFIG_NETFILTER 146 struct netns_nf nf; 147 struct netns_xt xt; 148#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 149 struct netns_ct ct; 150#endif 151#if defined(CONFIG_NF_TABLES) || defined(CONFIG_NF_TABLES_MODULE) 152 struct netns_nftables nft; 153#endif 154#if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) 155 struct netns_nf_frag nf_frag; 156 struct ctl_table_header *nf_frag_frags_hdr; 157#endif 158 struct sock *nfnl; 159 struct sock *nfnl_stash; 160#if IS_ENABLED(CONFIG_NETFILTER_NETLINK_ACCT) 161 struct list_head nfnl_acct_list; 162#endif 163#if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT) 164 struct list_head nfct_timeout_list; 165#endif 166#endif 167#ifdef CONFIG_WEXT_CORE 168 struct sk_buff_head wext_nlevents; 169#endif 170 struct net_generic __rcu *gen; 171 172 /* Used to store attached BPF programs */ 173 struct netns_bpf bpf; 174 175 /* Note : following structs are cache line aligned */ 176#ifdef CONFIG_XFRM 177 struct netns_xfrm xfrm; 178#endif 179 180 atomic64_t net_cookie; /* written once */ 181 182#if IS_ENABLED(CONFIG_IP_VS) 183 struct netns_ipvs *ipvs; 184#endif 185#if IS_ENABLED(CONFIG_MPLS) 186 struct netns_mpls mpls; 187#endif 188#if IS_ENABLED(CONFIG_CAN) 189 struct netns_can can; 190#endif 191#ifdef CONFIG_XDP_SOCKETS 192 struct netns_xdp xdp; 193#endif 194#if IS_ENABLED(CONFIG_CRYPTO_USER) 195 struct sock *crypto_nlsk; 196#endif 197 struct sock *diag_nlsk; 198} __randomize_layout; 199 200#include <linux/seq_file_net.h> 201 202/* Init's network namespace */ 203extern struct net init_net; 204 205#ifdef CONFIG_NET_NS 206struct net *copy_net_ns(unsigned long flags, struct user_namespace *user_ns, 207 struct net *old_net); 208 209void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid); 210 211void net_ns_barrier(void); 212 213struct ns_common *get_net_ns(struct ns_common *ns); 214#else /* CONFIG_NET_NS */ 215#include <linux/sched.h> 216#include <linux/nsproxy.h> 217static inline struct net *copy_net_ns(unsigned long flags, 218 struct user_namespace *user_ns, struct net *old_net) 219{ 220 if (flags & CLONE_NEWNET) 221 return ERR_PTR(-EINVAL); 222 return old_net; 223} 224 225static inline void net_ns_get_ownership(const struct net *net, 226 kuid_t *uid, kgid_t *gid) 227{ 228 *uid = GLOBAL_ROOT_UID; 229 *gid = GLOBAL_ROOT_GID; 230} 231 232static inline void net_ns_barrier(void) {} 233 234static inline struct ns_common *get_net_ns(struct ns_common *ns) 235{ 236 return ERR_PTR(-EINVAL); 237} 238#endif /* CONFIG_NET_NS */ 239 240 241extern struct list_head net_namespace_list; 242 243struct net *get_net_ns_by_pid(pid_t pid); 244struct net *get_net_ns_by_fd(int fd); 245 246u64 __net_gen_cookie(struct net *net); 247 248#ifdef CONFIG_SYSCTL 249void ipx_register_sysctl(void); 250void ipx_unregister_sysctl(void); 251#else 252#define ipx_register_sysctl() 253#define ipx_unregister_sysctl() 254#endif 255 256#ifdef CONFIG_NET_NS 257void __put_net(struct net *net); 258 259static inline struct net *get_net(struct net *net) 260{ 261 refcount_inc(&net->count); 262 return net; 263} 264 265static inline struct net *maybe_get_net(struct net *net) 266{ 267 /* Used when we know struct net exists but we 268 * aren't guaranteed a previous reference count 269 * exists. If the reference count is zero this 270 * function fails and returns NULL. 271 */ 272 if (!refcount_inc_not_zero(&net->count)) 273 net = NULL; 274 return net; 275} 276 277static inline void put_net(struct net *net) 278{ 279 if (refcount_dec_and_test(&net->count)) 280 __put_net(net); 281} 282 283static inline 284int net_eq(const struct net *net1, const struct net *net2) 285{ 286 return net1 == net2; 287} 288 289static inline int check_net(const struct net *net) 290{ 291 return refcount_read(&net->count) != 0; 292} 293 294void net_drop_ns(void *); 295 296#else 297 298static inline struct net *get_net(struct net *net) 299{ 300 return net; 301} 302 303static inline void put_net(struct net *net) 304{ 305} 306 307static inline struct net *maybe_get_net(struct net *net) 308{ 309 return net; 310} 311 312static inline 313int net_eq(const struct net *net1, const struct net *net2) 314{ 315 return 1; 316} 317 318static inline int check_net(const struct net *net) 319{ 320 return 1; 321} 322 323#define net_drop_ns NULL 324#endif 325 326 327typedef struct { 328#ifdef CONFIG_NET_NS 329 struct net *net; 330#endif 331} possible_net_t; 332 333static inline void write_pnet(possible_net_t *pnet, struct net *net) 334{ 335#ifdef CONFIG_NET_NS 336 pnet->net = net; 337#endif 338} 339 340static inline struct net *read_pnet(const possible_net_t *pnet) 341{ 342#ifdef CONFIG_NET_NS 343 return pnet->net; 344#else 345 return &init_net; 346#endif 347} 348 349/* Protected by net_rwsem */ 350#define for_each_net(VAR) \ 351 list_for_each_entry(VAR, &net_namespace_list, list) 352#define for_each_net_continue_reverse(VAR) \ 353 list_for_each_entry_continue_reverse(VAR, &net_namespace_list, list) 354#define for_each_net_rcu(VAR) \ 355 list_for_each_entry_rcu(VAR, &net_namespace_list, list) 356 357#ifdef CONFIG_NET_NS 358#define __net_init 359#define __net_exit 360#define __net_initdata 361#define __net_initconst 362#else 363#define __net_init __init 364#define __net_exit __ref 365#define __net_initdata __initdata 366#define __net_initconst __initconst 367#endif 368 369int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp); 370int peernet2id(const struct net *net, struct net *peer); 371bool peernet_has_id(const struct net *net, struct net *peer); 372struct net *get_net_ns_by_id(const struct net *net, int id); 373 374struct pernet_operations { 375 struct list_head list; 376 /* 377 * Below methods are called without any exclusive locks. 378 * More than one net may be constructed and destructed 379 * in parallel on several cpus. Every pernet_operations 380 * have to keep in mind all other pernet_operations and 381 * to introduce a locking, if they share common resources. 382 * 383 * The only time they are called with exclusive lock is 384 * from register_pernet_subsys(), unregister_pernet_subsys() 385 * register_pernet_device() and unregister_pernet_device(). 386 * 387 * Exit methods using blocking RCU primitives, such as 388 * synchronize_rcu(), should be implemented via exit_batch. 389 * Then, destruction of a group of net requires single 390 * synchronize_rcu() related to these pernet_operations, 391 * instead of separate synchronize_rcu() for every net. 392 * Please, avoid synchronize_rcu() at all, where it's possible. 393 * 394 * Note that a combination of pre_exit() and exit() can 395 * be used, since a synchronize_rcu() is guaranteed between 396 * the calls. 397 */ 398 int (*init)(struct net *net); 399 void (*pre_exit)(struct net *net); 400 void (*exit)(struct net *net); 401 void (*exit_batch)(struct list_head *net_exit_list); 402 unsigned int *id; 403 size_t size; 404}; 405 406/* 407 * Use these carefully. If you implement a network device and it 408 * needs per network namespace operations use device pernet operations, 409 * otherwise use pernet subsys operations. 410 * 411 * Network interfaces need to be removed from a dying netns _before_ 412 * subsys notifiers can be called, as most of the network code cleanup 413 * (which is done from subsys notifiers) runs with the assumption that 414 * dev_remove_pack has been called so no new packets will arrive during 415 * and after the cleanup functions have been called. dev_remove_pack 416 * is not per namespace so instead the guarantee of no more packets 417 * arriving in a network namespace is provided by ensuring that all 418 * network devices and all sockets have left the network namespace 419 * before the cleanup methods are called. 420 * 421 * For the longest time the ipv4 icmp code was registered as a pernet 422 * device which caused kernel oops, and panics during network 423 * namespace cleanup. So please don't get this wrong. 424 */ 425int register_pernet_subsys(struct pernet_operations *); 426void unregister_pernet_subsys(struct pernet_operations *); 427int register_pernet_device(struct pernet_operations *); 428void unregister_pernet_device(struct pernet_operations *); 429 430struct ctl_table; 431struct ctl_table_header; 432 433#ifdef CONFIG_SYSCTL 434int net_sysctl_init(void); 435struct ctl_table_header *register_net_sysctl(struct net *net, const char *path, 436 struct ctl_table *table); 437void unregister_net_sysctl_table(struct ctl_table_header *header); 438#else 439static inline int net_sysctl_init(void) { return 0; } 440static inline struct ctl_table_header *register_net_sysctl(struct net *net, 441 const char *path, struct ctl_table *table) 442{ 443 return NULL; 444} 445static inline void unregister_net_sysctl_table(struct ctl_table_header *header) 446{ 447} 448#endif 449 450static inline int rt_genid_ipv4(const struct net *net) 451{ 452 return atomic_read(&net->ipv4.rt_genid); 453} 454 455#if IS_ENABLED(CONFIG_IPV6) 456static inline int rt_genid_ipv6(const struct net *net) 457{ 458 return atomic_read(&net->ipv6.fib6_sernum); 459} 460#endif 461 462static inline void rt_genid_bump_ipv4(struct net *net) 463{ 464 atomic_inc(&net->ipv4.rt_genid); 465} 466 467extern void (*__fib6_flush_trees)(struct net *net); 468static inline void rt_genid_bump_ipv6(struct net *net) 469{ 470 if (__fib6_flush_trees) 471 __fib6_flush_trees(net); 472} 473 474#if IS_ENABLED(CONFIG_IEEE802154_6LOWPAN) 475static inline struct netns_ieee802154_lowpan * 476net_ieee802154_lowpan(struct net *net) 477{ 478 return &net->ieee802154_lowpan; 479} 480#endif 481 482/* For callers who don't really care about whether it's IPv4 or IPv6 */ 483static inline void rt_genid_bump_all(struct net *net) 484{ 485 rt_genid_bump_ipv4(net); 486 rt_genid_bump_ipv6(net); 487} 488 489static inline int fnhe_genid(const struct net *net) 490{ 491 return atomic_read(&net->fnhe_genid); 492} 493 494static inline void fnhe_genid_bump(struct net *net) 495{ 496 atomic_inc(&net->fnhe_genid); 497} 498 499#endif /* __NET_NET_NAMESPACE_H */ 500