xref: /linux/include/net/net_namespace.h (revision a58130ddc896e5a15e4de2bf50a1d89247118c23)
1 /*
2  * Operations on the network namespace
3  */
4 #ifndef __NET_NET_NAMESPACE_H
5 #define __NET_NET_NAMESPACE_H
6 
7 #include <linux/atomic.h>
8 #include <linux/workqueue.h>
9 #include <linux/list.h>
10 #include <linux/sysctl.h>
11 
12 #include <net/netns/core.h>
13 #include <net/netns/mib.h>
14 #include <net/netns/unix.h>
15 #include <net/netns/packet.h>
16 #include <net/netns/ipv4.h>
17 #include <net/netns/ipv6.h>
18 #include <net/netns/sctp.h>
19 #include <net/netns/dccp.h>
20 #include <net/netns/x_tables.h>
21 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
22 #include <net/netns/conntrack.h>
23 #endif
24 #include <net/netns/xfrm.h>
25 
26 struct user_namespace;
27 struct proc_dir_entry;
28 struct net_device;
29 struct sock;
30 struct ctl_table_header;
31 struct net_generic;
32 struct sock;
33 struct netns_ipvs;
34 
35 
36 #define NETDEV_HASHBITS    8
37 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
38 
39 struct net {
40 	atomic_t		passive;	/* To decided when the network
41 						 * namespace should be freed.
42 						 */
43 	atomic_t		count;		/* To decided when the network
44 						 *  namespace should be shut down.
45 						 */
46 #ifdef NETNS_REFCNT_DEBUG
47 	atomic_t		use_count;	/* To track references we
48 						 * destroy on demand
49 						 */
50 #endif
51 	spinlock_t		rules_mod_lock;
52 
53 	struct list_head	list;		/* list of network namespaces */
54 	struct list_head	cleanup_list;	/* namespaces on death row */
55 	struct list_head	exit_list;	/* Use only net_mutex */
56 
57 	struct user_namespace   *user_ns;	/* Owning user namespace */
58 
59 	struct proc_dir_entry 	*proc_net;
60 	struct proc_dir_entry 	*proc_net_stat;
61 
62 #ifdef CONFIG_SYSCTL
63 	struct ctl_table_set	sysctls;
64 #endif
65 
66 	struct sock 		*rtnl;			/* rtnetlink socket */
67 	struct sock		*genl_sock;
68 
69 	struct list_head 	dev_base_head;
70 	struct hlist_head 	*dev_name_head;
71 	struct hlist_head	*dev_index_head;
72 	unsigned int		dev_base_seq;	/* protected by rtnl_mutex */
73 	int			ifindex;
74 
75 	/* core fib_rules */
76 	struct list_head	rules_ops;
77 
78 
79 	struct net_device       *loopback_dev;          /* The loopback */
80 	struct netns_core	core;
81 	struct netns_mib	mib;
82 	struct netns_packet	packet;
83 	struct netns_unix	unx;
84 	struct netns_ipv4	ipv4;
85 #if IS_ENABLED(CONFIG_IPV6)
86 	struct netns_ipv6	ipv6;
87 #endif
88 #if defined(CONFIG_IP_SCTP) || defined(CONFIG_IP_SCTP_MODULE)
89 	struct netns_sctp	sctp;
90 #endif
91 #if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE)
92 	struct netns_dccp	dccp;
93 #endif
94 #ifdef CONFIG_NETFILTER
95 	struct netns_xt		xt;
96 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
97 	struct netns_ct		ct;
98 #endif
99 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
100 	struct netns_nf_frag	nf_frag;
101 #endif
102 	struct sock		*nfnl;
103 	struct sock		*nfnl_stash;
104 #endif
105 #ifdef CONFIG_WEXT_CORE
106 	struct sk_buff_head	wext_nlevents;
107 #endif
108 	struct net_generic __rcu	*gen;
109 
110 	/* Note : following structs are cache line aligned */
111 #ifdef CONFIG_XFRM
112 	struct netns_xfrm	xfrm;
113 #endif
114 	struct netns_ipvs	*ipvs;
115 	struct sock		*diag_nlsk;
116 	atomic_t		rt_genid;
117 };
118 
119 /*
120  * ifindex generation is per-net namespace, and loopback is
121  * always the 1st device in ns (see net_dev_init), thus any
122  * loopback device should get ifindex 1
123  */
124 
125 #define LOOPBACK_IFINDEX	1
126 
127 #include <linux/seq_file_net.h>
128 
129 /* Init's network namespace */
130 extern struct net init_net;
131 
132 #ifdef CONFIG_NET_NS
133 extern struct net *copy_net_ns(unsigned long flags,
134 	struct user_namespace *user_ns, struct net *old_net);
135 
136 #else /* CONFIG_NET_NS */
137 #include <linux/sched.h>
138 #include <linux/nsproxy.h>
139 static inline struct net *copy_net_ns(unsigned long flags,
140 	struct user_namespace *user_ns, struct net *old_net)
141 {
142 	if (flags & CLONE_NEWNET)
143 		return ERR_PTR(-EINVAL);
144 	return old_net;
145 }
146 #endif /* CONFIG_NET_NS */
147 
148 
149 extern struct list_head net_namespace_list;
150 
151 extern struct net *get_net_ns_by_pid(pid_t pid);
152 extern struct net *get_net_ns_by_fd(int pid);
153 
154 #ifdef CONFIG_NET_NS
155 extern void __put_net(struct net *net);
156 
157 static inline struct net *get_net(struct net *net)
158 {
159 	atomic_inc(&net->count);
160 	return net;
161 }
162 
163 static inline struct net *maybe_get_net(struct net *net)
164 {
165 	/* Used when we know struct net exists but we
166 	 * aren't guaranteed a previous reference count
167 	 * exists.  If the reference count is zero this
168 	 * function fails and returns NULL.
169 	 */
170 	if (!atomic_inc_not_zero(&net->count))
171 		net = NULL;
172 	return net;
173 }
174 
175 static inline void put_net(struct net *net)
176 {
177 	if (atomic_dec_and_test(&net->count))
178 		__put_net(net);
179 }
180 
181 static inline
182 int net_eq(const struct net *net1, const struct net *net2)
183 {
184 	return net1 == net2;
185 }
186 
187 extern void net_drop_ns(void *);
188 
189 #else
190 
191 static inline struct net *get_net(struct net *net)
192 {
193 	return net;
194 }
195 
196 static inline void put_net(struct net *net)
197 {
198 }
199 
200 static inline struct net *maybe_get_net(struct net *net)
201 {
202 	return net;
203 }
204 
205 static inline
206 int net_eq(const struct net *net1, const struct net *net2)
207 {
208 	return 1;
209 }
210 
211 #define net_drop_ns NULL
212 #endif
213 
214 
215 #ifdef NETNS_REFCNT_DEBUG
216 static inline struct net *hold_net(struct net *net)
217 {
218 	if (net)
219 		atomic_inc(&net->use_count);
220 	return net;
221 }
222 
223 static inline void release_net(struct net *net)
224 {
225 	if (net)
226 		atomic_dec(&net->use_count);
227 }
228 #else
229 static inline struct net *hold_net(struct net *net)
230 {
231 	return net;
232 }
233 
234 static inline void release_net(struct net *net)
235 {
236 }
237 #endif
238 
239 #ifdef CONFIG_NET_NS
240 
241 static inline void write_pnet(struct net **pnet, struct net *net)
242 {
243 	*pnet = net;
244 }
245 
246 static inline struct net *read_pnet(struct net * const *pnet)
247 {
248 	return *pnet;
249 }
250 
251 #else
252 
253 #define write_pnet(pnet, net)	do { (void)(net);} while (0)
254 #define read_pnet(pnet)		(&init_net)
255 
256 #endif
257 
258 #define for_each_net(VAR)				\
259 	list_for_each_entry(VAR, &net_namespace_list, list)
260 
261 #define for_each_net_rcu(VAR)				\
262 	list_for_each_entry_rcu(VAR, &net_namespace_list, list)
263 
264 #ifdef CONFIG_NET_NS
265 #define __net_init
266 #define __net_exit
267 #define __net_initdata
268 #define __net_initconst
269 #else
270 #define __net_init	__init
271 #define __net_exit	__exit_refok
272 #define __net_initdata	__initdata
273 #define __net_initconst	__initconst
274 #endif
275 
276 struct pernet_operations {
277 	struct list_head list;
278 	int (*init)(struct net *net);
279 	void (*exit)(struct net *net);
280 	void (*exit_batch)(struct list_head *net_exit_list);
281 	int *id;
282 	size_t size;
283 };
284 
285 /*
286  * Use these carefully.  If you implement a network device and it
287  * needs per network namespace operations use device pernet operations,
288  * otherwise use pernet subsys operations.
289  *
290  * Network interfaces need to be removed from a dying netns _before_
291  * subsys notifiers can be called, as most of the network code cleanup
292  * (which is done from subsys notifiers) runs with the assumption that
293  * dev_remove_pack has been called so no new packets will arrive during
294  * and after the cleanup functions have been called.  dev_remove_pack
295  * is not per namespace so instead the guarantee of no more packets
296  * arriving in a network namespace is provided by ensuring that all
297  * network devices and all sockets have left the network namespace
298  * before the cleanup methods are called.
299  *
300  * For the longest time the ipv4 icmp code was registered as a pernet
301  * device which caused kernel oops, and panics during network
302  * namespace cleanup.   So please don't get this wrong.
303  */
304 extern int register_pernet_subsys(struct pernet_operations *);
305 extern void unregister_pernet_subsys(struct pernet_operations *);
306 extern int register_pernet_device(struct pernet_operations *);
307 extern void unregister_pernet_device(struct pernet_operations *);
308 
309 struct ctl_table;
310 struct ctl_table_header;
311 
312 #ifdef CONFIG_SYSCTL
313 extern int net_sysctl_init(void);
314 extern struct ctl_table_header *register_net_sysctl(struct net *net,
315 	const char *path, struct ctl_table *table);
316 extern void unregister_net_sysctl_table(struct ctl_table_header *header);
317 #else
318 static inline int net_sysctl_init(void) { return 0; }
319 static inline struct ctl_table_header *register_net_sysctl(struct net *net,
320 	const char *path, struct ctl_table *table)
321 {
322 	return NULL;
323 }
324 static inline void unregister_net_sysctl_table(struct ctl_table_header *header)
325 {
326 }
327 #endif
328 
329 static inline int rt_genid(struct net *net)
330 {
331 	return atomic_read(&net->rt_genid);
332 }
333 
334 static inline void rt_genid_bump(struct net *net)
335 {
336 	atomic_inc(&net->rt_genid);
337 }
338 
339 #endif /* __NET_NET_NAMESPACE_H */
340