xref: /linux/include/net/net_namespace.h (revision aaa44952bbd1d4db14a4d676bf9595bb5db7e7b0)
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/netfilter.h>
26 #include <net/netns/x_tables.h>
27 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
28 #include <net/netns/conntrack.h>
29 #endif
30 #include <net/netns/nftables.h>
31 #include <net/netns/xfrm.h>
32 #include <net/netns/mpls.h>
33 #include <net/netns/can.h>
34 #include <net/netns/xdp.h>
35 #include <net/netns/bpf.h>
36 #include <linux/ns_common.h>
37 #include <linux/idr.h>
38 #include <linux/skbuff.h>
39 #include <linux/notifier.h>
40 
41 struct user_namespace;
42 struct proc_dir_entry;
43 struct net_device;
44 struct sock;
45 struct ctl_table_header;
46 struct net_generic;
47 struct uevent_sock;
48 struct netns_ipvs;
49 struct bpf_prog;
50 
51 
52 #define NETDEV_HASHBITS    8
53 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
54 
55 struct net {
56 	/* First cache line can be often dirtied.
57 	 * Do not place here read-mostly fields.
58 	 */
59 	refcount_t		passive;	/* To decide when the network
60 						 * namespace should be freed.
61 						 */
62 	spinlock_t		rules_mod_lock;
63 
64 	unsigned int		dev_unreg_count;
65 
66 	unsigned int		dev_base_seq;	/* protected by rtnl_mutex */
67 	int			ifindex;
68 
69 	spinlock_t		nsid_lock;
70 	atomic_t		fnhe_genid;
71 
72 	struct list_head	list;		/* list of network namespaces */
73 	struct list_head	exit_list;	/* To linked to call pernet exit
74 						 * methods on dead net (
75 						 * pernet_ops_rwsem read locked),
76 						 * or to unregister pernet ops
77 						 * (pernet_ops_rwsem write locked).
78 						 */
79 	struct llist_node	cleanup_list;	/* namespaces on death row */
80 
81 #ifdef CONFIG_KEYS
82 	struct key_tag		*key_domain;	/* Key domain of operation tag */
83 #endif
84 	struct user_namespace   *user_ns;	/* Owning user namespace */
85 	struct ucounts		*ucounts;
86 	struct idr		netns_ids;
87 
88 	struct ns_common	ns;
89 
90 	struct list_head 	dev_base_head;
91 	struct proc_dir_entry 	*proc_net;
92 	struct proc_dir_entry 	*proc_net_stat;
93 
94 #ifdef CONFIG_SYSCTL
95 	struct ctl_table_set	sysctls;
96 #endif
97 
98 	struct sock 		*rtnl;			/* rtnetlink socket */
99 	struct sock		*genl_sock;
100 
101 	struct uevent_sock	*uevent_sock;		/* uevent socket */
102 
103 	struct hlist_head 	*dev_name_head;
104 	struct hlist_head	*dev_index_head;
105 	struct raw_notifier_head	netdev_chain;
106 
107 	/* Note that @hash_mix can be read millions times per second,
108 	 * it is critical that it is on a read_mostly cache line.
109 	 */
110 	u32			hash_mix;
111 
112 	struct net_device       *loopback_dev;          /* The loopback */
113 
114 	/* core fib_rules */
115 	struct list_head	rules_ops;
116 
117 	struct netns_core	core;
118 	struct netns_mib	mib;
119 	struct netns_packet	packet;
120 	struct netns_unix	unx;
121 	struct netns_nexthop	nexthop;
122 	struct netns_ipv4	ipv4;
123 #if IS_ENABLED(CONFIG_IPV6)
124 	struct netns_ipv6	ipv6;
125 #endif
126 #if IS_ENABLED(CONFIG_IEEE802154_6LOWPAN)
127 	struct netns_ieee802154_lowpan	ieee802154_lowpan;
128 #endif
129 #if defined(CONFIG_IP_SCTP) || defined(CONFIG_IP_SCTP_MODULE)
130 	struct netns_sctp	sctp;
131 #endif
132 #ifdef CONFIG_NETFILTER
133 	struct netns_nf		nf;
134 	struct netns_xt		xt;
135 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
136 	struct netns_ct		ct;
137 #endif
138 #if defined(CONFIG_NF_TABLES) || defined(CONFIG_NF_TABLES_MODULE)
139 	struct netns_nftables	nft;
140 #endif
141 #endif
142 #ifdef CONFIG_WEXT_CORE
143 	struct sk_buff_head	wext_nlevents;
144 #endif
145 	struct net_generic __rcu	*gen;
146 
147 	/* Used to store attached BPF programs */
148 	struct netns_bpf	bpf;
149 
150 	/* Note : following structs are cache line aligned */
151 #ifdef CONFIG_XFRM
152 	struct netns_xfrm	xfrm;
153 #endif
154 
155 	u64			net_cookie; /* written once */
156 
157 #if IS_ENABLED(CONFIG_IP_VS)
158 	struct netns_ipvs	*ipvs;
159 #endif
160 #if IS_ENABLED(CONFIG_MPLS)
161 	struct netns_mpls	mpls;
162 #endif
163 #if IS_ENABLED(CONFIG_CAN)
164 	struct netns_can	can;
165 #endif
166 #ifdef CONFIG_XDP_SOCKETS
167 	struct netns_xdp	xdp;
168 #endif
169 #if IS_ENABLED(CONFIG_CRYPTO_USER)
170 	struct sock		*crypto_nlsk;
171 #endif
172 	struct sock		*diag_nlsk;
173 } __randomize_layout;
174 
175 #include <linux/seq_file_net.h>
176 
177 /* Init's network namespace */
178 extern struct net init_net;
179 
180 #ifdef CONFIG_NET_NS
181 struct net *copy_net_ns(unsigned long flags, struct user_namespace *user_ns,
182 			struct net *old_net);
183 
184 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid);
185 
186 void net_ns_barrier(void);
187 #else /* CONFIG_NET_NS */
188 #include <linux/sched.h>
189 #include <linux/nsproxy.h>
190 static inline struct net *copy_net_ns(unsigned long flags,
191 	struct user_namespace *user_ns, struct net *old_net)
192 {
193 	if (flags & CLONE_NEWNET)
194 		return ERR_PTR(-EINVAL);
195 	return old_net;
196 }
197 
198 static inline void net_ns_get_ownership(const struct net *net,
199 					kuid_t *uid, kgid_t *gid)
200 {
201 	*uid = GLOBAL_ROOT_UID;
202 	*gid = GLOBAL_ROOT_GID;
203 }
204 
205 static inline void net_ns_barrier(void) {}
206 #endif /* CONFIG_NET_NS */
207 
208 
209 extern struct list_head net_namespace_list;
210 
211 struct net *get_net_ns_by_pid(pid_t pid);
212 struct net *get_net_ns_by_fd(int fd);
213 
214 #ifdef CONFIG_SYSCTL
215 void ipx_register_sysctl(void);
216 void ipx_unregister_sysctl(void);
217 #else
218 #define ipx_register_sysctl()
219 #define ipx_unregister_sysctl()
220 #endif
221 
222 #ifdef CONFIG_NET_NS
223 void __put_net(struct net *net);
224 
225 static inline struct net *get_net(struct net *net)
226 {
227 	refcount_inc(&net->ns.count);
228 	return net;
229 }
230 
231 static inline struct net *maybe_get_net(struct net *net)
232 {
233 	/* Used when we know struct net exists but we
234 	 * aren't guaranteed a previous reference count
235 	 * exists.  If the reference count is zero this
236 	 * function fails and returns NULL.
237 	 */
238 	if (!refcount_inc_not_zero(&net->ns.count))
239 		net = NULL;
240 	return net;
241 }
242 
243 static inline void put_net(struct net *net)
244 {
245 	if (refcount_dec_and_test(&net->ns.count))
246 		__put_net(net);
247 }
248 
249 static inline
250 int net_eq(const struct net *net1, const struct net *net2)
251 {
252 	return net1 == net2;
253 }
254 
255 static inline int check_net(const struct net *net)
256 {
257 	return refcount_read(&net->ns.count) != 0;
258 }
259 
260 void net_drop_ns(void *);
261 
262 #else
263 
264 static inline struct net *get_net(struct net *net)
265 {
266 	return net;
267 }
268 
269 static inline void put_net(struct net *net)
270 {
271 }
272 
273 static inline struct net *maybe_get_net(struct net *net)
274 {
275 	return net;
276 }
277 
278 static inline
279 int net_eq(const struct net *net1, const struct net *net2)
280 {
281 	return 1;
282 }
283 
284 static inline int check_net(const struct net *net)
285 {
286 	return 1;
287 }
288 
289 #define net_drop_ns NULL
290 #endif
291 
292 
293 typedef struct {
294 #ifdef CONFIG_NET_NS
295 	struct net *net;
296 #endif
297 } possible_net_t;
298 
299 static inline void write_pnet(possible_net_t *pnet, struct net *net)
300 {
301 #ifdef CONFIG_NET_NS
302 	pnet->net = net;
303 #endif
304 }
305 
306 static inline struct net *read_pnet(const possible_net_t *pnet)
307 {
308 #ifdef CONFIG_NET_NS
309 	return pnet->net;
310 #else
311 	return &init_net;
312 #endif
313 }
314 
315 /* Protected by net_rwsem */
316 #define for_each_net(VAR)				\
317 	list_for_each_entry(VAR, &net_namespace_list, list)
318 #define for_each_net_continue_reverse(VAR)		\
319 	list_for_each_entry_continue_reverse(VAR, &net_namespace_list, list)
320 #define for_each_net_rcu(VAR)				\
321 	list_for_each_entry_rcu(VAR, &net_namespace_list, list)
322 
323 #ifdef CONFIG_NET_NS
324 #define __net_init
325 #define __net_exit
326 #define __net_initdata
327 #define __net_initconst
328 #else
329 #define __net_init	__init
330 #define __net_exit	__ref
331 #define __net_initdata	__initdata
332 #define __net_initconst	__initconst
333 #endif
334 
335 int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp);
336 int peernet2id(const struct net *net, struct net *peer);
337 bool peernet_has_id(const struct net *net, struct net *peer);
338 struct net *get_net_ns_by_id(const struct net *net, int id);
339 
340 struct pernet_operations {
341 	struct list_head list;
342 	/*
343 	 * Below methods are called without any exclusive locks.
344 	 * More than one net may be constructed and destructed
345 	 * in parallel on several cpus. Every pernet_operations
346 	 * have to keep in mind all other pernet_operations and
347 	 * to introduce a locking, if they share common resources.
348 	 *
349 	 * The only time they are called with exclusive lock is
350 	 * from register_pernet_subsys(), unregister_pernet_subsys()
351 	 * register_pernet_device() and unregister_pernet_device().
352 	 *
353 	 * Exit methods using blocking RCU primitives, such as
354 	 * synchronize_rcu(), should be implemented via exit_batch.
355 	 * Then, destruction of a group of net requires single
356 	 * synchronize_rcu() related to these pernet_operations,
357 	 * instead of separate synchronize_rcu() for every net.
358 	 * Please, avoid synchronize_rcu() at all, where it's possible.
359 	 *
360 	 * Note that a combination of pre_exit() and exit() can
361 	 * be used, since a synchronize_rcu() is guaranteed between
362 	 * the calls.
363 	 */
364 	int (*init)(struct net *net);
365 	void (*pre_exit)(struct net *net);
366 	void (*exit)(struct net *net);
367 	void (*exit_batch)(struct list_head *net_exit_list);
368 	unsigned int *id;
369 	size_t size;
370 };
371 
372 /*
373  * Use these carefully.  If you implement a network device and it
374  * needs per network namespace operations use device pernet operations,
375  * otherwise use pernet subsys operations.
376  *
377  * Network interfaces need to be removed from a dying netns _before_
378  * subsys notifiers can be called, as most of the network code cleanup
379  * (which is done from subsys notifiers) runs with the assumption that
380  * dev_remove_pack has been called so no new packets will arrive during
381  * and after the cleanup functions have been called.  dev_remove_pack
382  * is not per namespace so instead the guarantee of no more packets
383  * arriving in a network namespace is provided by ensuring that all
384  * network devices and all sockets have left the network namespace
385  * before the cleanup methods are called.
386  *
387  * For the longest time the ipv4 icmp code was registered as a pernet
388  * device which caused kernel oops, and panics during network
389  * namespace cleanup.   So please don't get this wrong.
390  */
391 int register_pernet_subsys(struct pernet_operations *);
392 void unregister_pernet_subsys(struct pernet_operations *);
393 int register_pernet_device(struct pernet_operations *);
394 void unregister_pernet_device(struct pernet_operations *);
395 
396 struct ctl_table;
397 
398 #ifdef CONFIG_SYSCTL
399 int net_sysctl_init(void);
400 struct ctl_table_header *register_net_sysctl(struct net *net, const char *path,
401 					     struct ctl_table *table);
402 void unregister_net_sysctl_table(struct ctl_table_header *header);
403 #else
404 static inline int net_sysctl_init(void) { return 0; }
405 static inline struct ctl_table_header *register_net_sysctl(struct net *net,
406 	const char *path, struct ctl_table *table)
407 {
408 	return NULL;
409 }
410 static inline void unregister_net_sysctl_table(struct ctl_table_header *header)
411 {
412 }
413 #endif
414 
415 static inline int rt_genid_ipv4(const struct net *net)
416 {
417 	return atomic_read(&net->ipv4.rt_genid);
418 }
419 
420 #if IS_ENABLED(CONFIG_IPV6)
421 static inline int rt_genid_ipv6(const struct net *net)
422 {
423 	return atomic_read(&net->ipv6.fib6_sernum);
424 }
425 #endif
426 
427 static inline void rt_genid_bump_ipv4(struct net *net)
428 {
429 	atomic_inc(&net->ipv4.rt_genid);
430 }
431 
432 extern void (*__fib6_flush_trees)(struct net *net);
433 static inline void rt_genid_bump_ipv6(struct net *net)
434 {
435 	if (__fib6_flush_trees)
436 		__fib6_flush_trees(net);
437 }
438 
439 #if IS_ENABLED(CONFIG_IEEE802154_6LOWPAN)
440 static inline struct netns_ieee802154_lowpan *
441 net_ieee802154_lowpan(struct net *net)
442 {
443 	return &net->ieee802154_lowpan;
444 }
445 #endif
446 
447 /* For callers who don't really care about whether it's IPv4 or IPv6 */
448 static inline void rt_genid_bump_all(struct net *net)
449 {
450 	rt_genid_bump_ipv4(net);
451 	rt_genid_bump_ipv6(net);
452 }
453 
454 static inline int fnhe_genid(const struct net *net)
455 {
456 	return atomic_read(&net->fnhe_genid);
457 }
458 
459 static inline void fnhe_genid_bump(struct net *net)
460 {
461 	atomic_inc(&net->fnhe_genid);
462 }
463 
464 #endif /* __NET_NET_NAMESPACE_H */
465