xref: /linux/include/net/net_namespace.h (revision 18f90d372cf35b387663f1567de701e5393f6eb5)
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/ieee802154_6lowpan.h>
23 #include <net/netns/sctp.h>
24 #include <net/netns/dccp.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 <linux/ns_common.h>
36 #include <linux/idr.h>
37 #include <linux/skbuff.h>
38 
39 struct user_namespace;
40 struct proc_dir_entry;
41 struct net_device;
42 struct sock;
43 struct ctl_table_header;
44 struct net_generic;
45 struct uevent_sock;
46 struct netns_ipvs;
47 struct bpf_prog;
48 
49 
50 #define NETDEV_HASHBITS    8
51 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
52 
53 struct net {
54 	refcount_t		passive;	/* To decided when the network
55 						 * namespace should be freed.
56 						 */
57 	refcount_t		count;		/* To decided when the network
58 						 *  namespace should be shut down.
59 						 */
60 	spinlock_t		rules_mod_lock;
61 
62 	u32			hash_mix;
63 	atomic64_t		cookie_gen;
64 
65 	struct list_head	list;		/* list of network namespaces */
66 	struct list_head	exit_list;	/* To linked to call pernet exit
67 						 * methods on dead net (
68 						 * pernet_ops_rwsem read locked),
69 						 * or to unregister pernet ops
70 						 * (pernet_ops_rwsem write locked).
71 						 */
72 	struct llist_node	cleanup_list;	/* namespaces on death row */
73 
74 	struct user_namespace   *user_ns;	/* Owning user namespace */
75 	struct ucounts		*ucounts;
76 	spinlock_t		nsid_lock;
77 	struct idr		netns_ids;
78 
79 	struct ns_common	ns;
80 
81 	struct proc_dir_entry 	*proc_net;
82 	struct proc_dir_entry 	*proc_net_stat;
83 
84 #ifdef CONFIG_SYSCTL
85 	struct ctl_table_set	sysctls;
86 #endif
87 
88 	struct sock 		*rtnl;			/* rtnetlink socket */
89 	struct sock		*genl_sock;
90 
91 	struct uevent_sock	*uevent_sock;		/* uevent socket */
92 
93 	struct list_head 	dev_base_head;
94 	struct hlist_head 	*dev_name_head;
95 	struct hlist_head	*dev_index_head;
96 	unsigned int		dev_base_seq;	/* protected by rtnl_mutex */
97 	int			ifindex;
98 	unsigned int		dev_unreg_count;
99 
100 	/* core fib_rules */
101 	struct list_head	rules_ops;
102 
103 	struct list_head	fib_notifier_ops;  /* Populated by
104 						    * register_pernet_subsys()
105 						    */
106 	struct net_device       *loopback_dev;          /* The loopback */
107 	struct netns_core	core;
108 	struct netns_mib	mib;
109 	struct netns_packet	packet;
110 	struct netns_unix	unx;
111 	struct netns_ipv4	ipv4;
112 #if IS_ENABLED(CONFIG_IPV6)
113 	struct netns_ipv6	ipv6;
114 #endif
115 #if IS_ENABLED(CONFIG_IEEE802154_6LOWPAN)
116 	struct netns_ieee802154_lowpan	ieee802154_lowpan;
117 #endif
118 #if defined(CONFIG_IP_SCTP) || defined(CONFIG_IP_SCTP_MODULE)
119 	struct netns_sctp	sctp;
120 #endif
121 #if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE)
122 	struct netns_dccp	dccp;
123 #endif
124 #ifdef CONFIG_NETFILTER
125 	struct netns_nf		nf;
126 	struct netns_xt		xt;
127 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
128 	struct netns_ct		ct;
129 #endif
130 #if defined(CONFIG_NF_TABLES) || defined(CONFIG_NF_TABLES_MODULE)
131 	struct netns_nftables	nft;
132 #endif
133 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
134 	struct netns_nf_frag	nf_frag;
135 	struct ctl_table_header *nf_frag_frags_hdr;
136 #endif
137 	struct sock		*nfnl;
138 	struct sock		*nfnl_stash;
139 #if IS_ENABLED(CONFIG_NETFILTER_NETLINK_ACCT)
140 	struct list_head        nfnl_acct_list;
141 #endif
142 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
143 	struct list_head	nfct_timeout_list;
144 #endif
145 #endif
146 #ifdef CONFIG_WEXT_CORE
147 	struct sk_buff_head	wext_nlevents;
148 #endif
149 	struct net_generic __rcu	*gen;
150 
151 	struct bpf_prog __rcu	*flow_dissector_prog;
152 
153 	/* Note : following structs are cache line aligned */
154 #ifdef CONFIG_XFRM
155 	struct netns_xfrm	xfrm;
156 #endif
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 	struct sock		*diag_nlsk;
170 	atomic_t		fnhe_genid;
171 } __randomize_layout;
172 
173 #include <linux/seq_file_net.h>
174 
175 /* Init's network namespace */
176 extern struct net init_net;
177 
178 #ifdef CONFIG_NET_NS
179 struct net *copy_net_ns(unsigned long flags, struct user_namespace *user_ns,
180 			struct net *old_net);
181 
182 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid);
183 
184 void net_ns_barrier(void);
185 #else /* CONFIG_NET_NS */
186 #include <linux/sched.h>
187 #include <linux/nsproxy.h>
188 static inline struct net *copy_net_ns(unsigned long flags,
189 	struct user_namespace *user_ns, struct net *old_net)
190 {
191 	if (flags & CLONE_NEWNET)
192 		return ERR_PTR(-EINVAL);
193 	return old_net;
194 }
195 
196 static inline void net_ns_get_ownership(const struct net *net,
197 					kuid_t *uid, kgid_t *gid)
198 {
199 	*uid = GLOBAL_ROOT_UID;
200 	*gid = GLOBAL_ROOT_GID;
201 }
202 
203 static inline void net_ns_barrier(void) {}
204 #endif /* CONFIG_NET_NS */
205 
206 
207 extern struct list_head net_namespace_list;
208 
209 struct net *get_net_ns_by_pid(pid_t pid);
210 struct net *get_net_ns_by_fd(int fd);
211 
212 #ifdef CONFIG_SYSCTL
213 void ipx_register_sysctl(void);
214 void ipx_unregister_sysctl(void);
215 #else
216 #define ipx_register_sysctl()
217 #define ipx_unregister_sysctl()
218 #endif
219 
220 #ifdef CONFIG_NET_NS
221 void __put_net(struct net *net);
222 
223 static inline struct net *get_net(struct net *net)
224 {
225 	refcount_inc(&net->count);
226 	return net;
227 }
228 
229 static inline struct net *maybe_get_net(struct net *net)
230 {
231 	/* Used when we know struct net exists but we
232 	 * aren't guaranteed a previous reference count
233 	 * exists.  If the reference count is zero this
234 	 * function fails and returns NULL.
235 	 */
236 	if (!refcount_inc_not_zero(&net->count))
237 		net = NULL;
238 	return net;
239 }
240 
241 static inline void put_net(struct net *net)
242 {
243 	if (refcount_dec_and_test(&net->count))
244 		__put_net(net);
245 }
246 
247 static inline
248 int net_eq(const struct net *net1, const struct net *net2)
249 {
250 	return net1 == net2;
251 }
252 
253 static inline int check_net(const struct net *net)
254 {
255 	return refcount_read(&net->count) != 0;
256 }
257 
258 void net_drop_ns(void *);
259 
260 #else
261 
262 static inline struct net *get_net(struct net *net)
263 {
264 	return net;
265 }
266 
267 static inline void put_net(struct net *net)
268 {
269 }
270 
271 static inline struct net *maybe_get_net(struct net *net)
272 {
273 	return net;
274 }
275 
276 static inline
277 int net_eq(const struct net *net1, const struct net *net2)
278 {
279 	return 1;
280 }
281 
282 static inline int check_net(const struct net *net)
283 {
284 	return 1;
285 }
286 
287 #define net_drop_ns NULL
288 #endif
289 
290 
291 typedef struct {
292 #ifdef CONFIG_NET_NS
293 	struct net *net;
294 #endif
295 } possible_net_t;
296 
297 static inline void write_pnet(possible_net_t *pnet, struct net *net)
298 {
299 #ifdef CONFIG_NET_NS
300 	pnet->net = net;
301 #endif
302 }
303 
304 static inline struct net *read_pnet(const possible_net_t *pnet)
305 {
306 #ifdef CONFIG_NET_NS
307 	return pnet->net;
308 #else
309 	return &init_net;
310 #endif
311 }
312 
313 /* Protected by net_rwsem */
314 #define for_each_net(VAR)				\
315 	list_for_each_entry(VAR, &net_namespace_list, list)
316 
317 #define for_each_net_rcu(VAR)				\
318 	list_for_each_entry_rcu(VAR, &net_namespace_list, list)
319 
320 #ifdef CONFIG_NET_NS
321 #define __net_init
322 #define __net_exit
323 #define __net_initdata
324 #define __net_initconst
325 #else
326 #define __net_init	__init
327 #define __net_exit	__ref
328 #define __net_initdata	__initdata
329 #define __net_initconst	__initconst
330 #endif
331 
332 int peernet2id_alloc(struct net *net, struct net *peer);
333 int peernet2id(struct net *net, struct net *peer);
334 bool peernet_has_id(struct net *net, struct net *peer);
335 struct net *get_net_ns_by_id(struct net *net, int id);
336 
337 struct pernet_operations {
338 	struct list_head list;
339 	/*
340 	 * Below methods are called without any exclusive locks.
341 	 * More than one net may be constructed and destructed
342 	 * in parallel on several cpus. Every pernet_operations
343 	 * have to keep in mind all other pernet_operations and
344 	 * to introduce a locking, if they share common resources.
345 	 *
346 	 * The only time they are called with exclusive lock is
347 	 * from register_pernet_subsys(), unregister_pernet_subsys()
348 	 * register_pernet_device() and unregister_pernet_device().
349 	 *
350 	 * Exit methods using blocking RCU primitives, such as
351 	 * synchronize_rcu(), should be implemented via exit_batch.
352 	 * Then, destruction of a group of net requires single
353 	 * synchronize_rcu() related to these pernet_operations,
354 	 * instead of separate synchronize_rcu() for every net.
355 	 * Please, avoid synchronize_rcu() at all, where it's possible.
356 	 */
357 	int (*init)(struct net *net);
358 	void (*exit)(struct net *net);
359 	void (*exit_batch)(struct list_head *net_exit_list);
360 	unsigned int *id;
361 	size_t size;
362 };
363 
364 /*
365  * Use these carefully.  If you implement a network device and it
366  * needs per network namespace operations use device pernet operations,
367  * otherwise use pernet subsys operations.
368  *
369  * Network interfaces need to be removed from a dying netns _before_
370  * subsys notifiers can be called, as most of the network code cleanup
371  * (which is done from subsys notifiers) runs with the assumption that
372  * dev_remove_pack has been called so no new packets will arrive during
373  * and after the cleanup functions have been called.  dev_remove_pack
374  * is not per namespace so instead the guarantee of no more packets
375  * arriving in a network namespace is provided by ensuring that all
376  * network devices and all sockets have left the network namespace
377  * before the cleanup methods are called.
378  *
379  * For the longest time the ipv4 icmp code was registered as a pernet
380  * device which caused kernel oops, and panics during network
381  * namespace cleanup.   So please don't get this wrong.
382  */
383 int register_pernet_subsys(struct pernet_operations *);
384 void unregister_pernet_subsys(struct pernet_operations *);
385 int register_pernet_device(struct pernet_operations *);
386 void unregister_pernet_device(struct pernet_operations *);
387 
388 struct ctl_table;
389 struct ctl_table_header;
390 
391 #ifdef CONFIG_SYSCTL
392 int net_sysctl_init(void);
393 struct ctl_table_header *register_net_sysctl(struct net *net, const char *path,
394 					     struct ctl_table *table);
395 void unregister_net_sysctl_table(struct ctl_table_header *header);
396 #else
397 static inline int net_sysctl_init(void) { return 0; }
398 static inline struct ctl_table_header *register_net_sysctl(struct net *net,
399 	const char *path, struct ctl_table *table)
400 {
401 	return NULL;
402 }
403 static inline void unregister_net_sysctl_table(struct ctl_table_header *header)
404 {
405 }
406 #endif
407 
408 static inline int rt_genid_ipv4(struct net *net)
409 {
410 	return atomic_read(&net->ipv4.rt_genid);
411 }
412 
413 static inline void rt_genid_bump_ipv4(struct net *net)
414 {
415 	atomic_inc(&net->ipv4.rt_genid);
416 }
417 
418 extern void (*__fib6_flush_trees)(struct net *net);
419 static inline void rt_genid_bump_ipv6(struct net *net)
420 {
421 	if (__fib6_flush_trees)
422 		__fib6_flush_trees(net);
423 }
424 
425 #if IS_ENABLED(CONFIG_IEEE802154_6LOWPAN)
426 static inline struct netns_ieee802154_lowpan *
427 net_ieee802154_lowpan(struct net *net)
428 {
429 	return &net->ieee802154_lowpan;
430 }
431 #endif
432 
433 /* For callers who don't really care about whether it's IPv4 or IPv6 */
434 static inline void rt_genid_bump_all(struct net *net)
435 {
436 	rt_genid_bump_ipv4(net);
437 	rt_genid_bump_ipv6(net);
438 }
439 
440 static inline int fnhe_genid(struct net *net)
441 {
442 	return atomic_read(&net->fnhe_genid);
443 }
444 
445 static inline void fnhe_genid_bump(struct net *net)
446 {
447 	atomic_inc(&net->fnhe_genid);
448 }
449 
450 #endif /* __NET_NET_NAMESPACE_H */
451