blob: 84e37b1ca9e17824ef33ac4fa7708d684f520edc [file] [log] [blame]
Eric W. Biederman5f256be2007-09-12 11:50:50 +02001/*
2 * Operations on the network namespace
3 */
4#ifndef __NET_NET_NAMESPACE_H
5#define __NET_NET_NAMESPACE_H
6
Arun Sharma600634972011-07-26 16:09:06 -07007#include <linux/atomic.h>
Eric W. Biederman5f256be2007-09-12 11:50:50 +02008#include <linux/workqueue.h>
9#include <linux/list.h>
David S. Millerbee95252011-05-26 16:40:37 -040010#include <linux/sysctl.h>
Eric W. Biederman5f256be2007-09-12 11:50:50 +020011
Pavel Emelyanov8efa6e92008-03-31 19:41:14 -070012#include <net/netns/core.h>
Pavel Emelyanov852566f2008-07-18 04:01:24 -070013#include <net/netns/mib.h>
Denis V. Luneva0a53c82007-12-11 04:19:17 -080014#include <net/netns/unix.h>
Denis V. Lunev2aaef4e2007-12-11 04:19:54 -080015#include <net/netns/packet.h>
Pavel Emelyanov8afd3512007-12-16 13:29:36 -080016#include <net/netns/ipv4.h>
Daniel Lezcanob0f159d2008-01-10 02:49:06 -080017#include <net/netns/ipv6.h>
Eric W. Biederman4db67e82012-08-06 08:42:04 +000018#include <net/netns/sctp.h>
Pavel Emelyanov67019cc2008-04-13 22:28:42 -070019#include <net/netns/dccp.h>
Gao fengf3c1a442013-03-24 23:50:39 +000020#include <net/netns/netfilter.h>
Alexey Dobriyan8d870052008-01-31 04:02:13 -080021#include <net/netns/x_tables.h>
Alexey Dobriyandfdb8d72008-10-08 11:35:02 +020022#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
23#include <net/netns/conntrack.h>
24#endif
Alexey Dobriyand62ddc22008-11-25 17:14:31 -080025#include <net/netns/xfrm.h>
Denis V. Luneva0a53c82007-12-11 04:19:17 -080026
Eric W. Biederman038e7332012-06-14 02:31:10 -070027struct user_namespace;
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020028struct proc_dir_entry;
Eric W. Biederman2774c7a2007-09-26 22:10:56 -070029struct net_device;
Denis V. Lunev97c53ca2007-11-19 22:26:51 -080030struct sock;
Pavel Emelyanov1597fbc2007-12-01 23:51:01 +110031struct ctl_table_header;
Pavel Emelyanovdec827d2008-04-15 00:36:08 -070032struct net_generic;
Johannes Berg134e6372009-07-10 09:51:34 +000033struct sock;
Julian Anastasov2553d062011-03-04 12:18:07 +020034struct netns_ipvs;
Pavel Emelyanov1597fbc2007-12-01 23:51:01 +110035
Eric Dumazet7c28bd02009-10-24 06:13:17 -070036
37#define NETDEV_HASHBITS 8
38#define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
39
Eric W. Biederman5f256be2007-09-12 11:50:50 +020040struct net {
Al Viroa685e082011-06-08 21:13:01 -040041 atomic_t passive; /* To decided when the network
42 * namespace should be freed.
43 */
Eric W. Biederman5f256be2007-09-12 11:50:50 +020044 atomic_t count; /* To decided when the network
Al Viroa685e082011-06-08 21:13:01 -040045 * namespace should be shut down.
Eric W. Biederman5f256be2007-09-12 11:50:50 +020046 */
Denis V. Lunev5d1e4462008-04-16 01:58:04 -070047#ifdef NETNS_REFCNT_DEBUG
Eric W. Biederman5f256be2007-09-12 11:50:50 +020048 atomic_t use_count; /* To track references we
49 * destroy on demand
50 */
Denis V. Lunev5d1e4462008-04-16 01:58:04 -070051#endif
Eric Dumazet8e602ce2010-10-14 05:56:18 +000052 spinlock_t rules_mod_lock;
53
Eric W. Biederman5f256be2007-09-12 11:50:50 +020054 struct list_head list; /* list of network namespaces */
Eric W. Biederman2b035b32009-11-29 22:25:27 +000055 struct list_head cleanup_list; /* namespaces on death row */
Eric W. Biederman72ad9372009-12-03 02:29:03 +000056 struct list_head exit_list; /* Use only net_mutex */
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020057
Eric W. Biederman038e7332012-06-14 02:31:10 -070058 struct user_namespace *user_ns; /* Owning user namespace */
59
Eric W. Biederman98f842e2011-06-15 10:21:48 -070060 unsigned int proc_inum;
61
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020062 struct proc_dir_entry *proc_net;
63 struct proc_dir_entry *proc_net_stat;
Eric W. Biederman881d9662007-09-17 11:56:21 -070064
Al Viro73455092008-07-14 21:22:20 -040065#ifdef CONFIG_SYSCTL
66 struct ctl_table_set sysctls;
67#endif
Eric W. Biederman95bdfcc2007-11-30 23:55:42 +110068
Eric Dumazet8e602ce2010-10-14 05:56:18 +000069 struct sock *rtnl; /* rtnetlink socket */
70 struct sock *genl_sock;
Eric W. Biederman2774c7a2007-09-26 22:10:56 -070071
Eric W. Biederman881d9662007-09-17 11:56:21 -070072 struct list_head dev_base_head;
73 struct hlist_head *dev_name_head;
74 struct hlist_head *dev_index_head;
Thomas Graf4e985ad2011-06-21 03:11:20 +000075 unsigned int dev_base_seq; /* protected by rtnl_mutex */
Pavel Emelyanovaa79e662012-08-08 21:53:19 +000076 int ifindex;
Denis V. Lunev97c53ca2007-11-19 22:26:51 -080077
Denis V. Lunev5fd30ee2008-01-10 03:20:28 -080078 /* core fib_rules */
79 struct list_head rules_ops;
Denis V. Lunev5fd30ee2008-01-10 03:20:28 -080080
Denis V. Lunevd12d01d2007-11-19 22:28:35 -080081
Eric Dumazet8e602ce2010-10-14 05:56:18 +000082 struct net_device *loopback_dev; /* The loopback */
Pavel Emelyanov8efa6e92008-03-31 19:41:14 -070083 struct netns_core core;
Pavel Emelyanov852566f2008-07-18 04:01:24 -070084 struct netns_mib mib;
Denis V. Lunev2aaef4e2007-12-11 04:19:54 -080085 struct netns_packet packet;
Denis V. Luneva0a53c82007-12-11 04:19:17 -080086 struct netns_unix unx;
Pavel Emelyanov8afd3512007-12-16 13:29:36 -080087 struct netns_ipv4 ipv4;
Eric Dumazetdfd56b82011-12-10 09:48:31 +000088#if IS_ENABLED(CONFIG_IPV6)
Daniel Lezcanob0f159d2008-01-10 02:49:06 -080089 struct netns_ipv6 ipv6;
90#endif
Eric W. Biederman4db67e82012-08-06 08:42:04 +000091#if defined(CONFIG_IP_SCTP) || defined(CONFIG_IP_SCTP_MODULE)
92 struct netns_sctp sctp;
93#endif
Pavel Emelyanov67019cc2008-04-13 22:28:42 -070094#if defined(CONFIG_IP_DCCP) || defined(CONFIG_IP_DCCP_MODULE)
95 struct netns_dccp dccp;
96#endif
Alexey Dobriyan8d870052008-01-31 04:02:13 -080097#ifdef CONFIG_NETFILTER
Gao fengf3c1a442013-03-24 23:50:39 +000098 struct netns_nf nf;
Alexey Dobriyan8d870052008-01-31 04:02:13 -080099 struct netns_xt xt;
Alexey Dobriyandfdb8d72008-10-08 11:35:02 +0200100#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
101 struct netns_ct ct;
102#endif
Amerigo Wangc038a762012-09-18 16:50:08 +0000103#if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
104 struct netns_nf_frag nf_frag;
105#endif
Alexey Dobriyancd8c20b2010-01-13 16:02:14 +0100106 struct sock *nfnl;
107 struct sock *nfnl_stash;
Alexey Dobriyan8d870052008-01-31 04:02:13 -0800108#endif
Johannes Berg3d23e342009-09-29 23:27:28 +0200109#ifdef CONFIG_WEXT_CORE
Johannes Bergb333b3d222009-06-24 01:34:48 +0000110 struct sk_buff_head wext_nlevents;
111#endif
Eric Dumazet1c877332010-10-25 03:20:11 +0000112 struct net_generic __rcu *gen;
Eric Dumazet8e602ce2010-10-14 05:56:18 +0000113
114 /* Note : following structs are cache line aligned */
115#ifdef CONFIG_XFRM
116 struct netns_xfrm xfrm;
117#endif
JunweiZhang8b4d14d2013-06-26 16:40:06 +0800118#if IS_ENABLED(CONFIG_IP_VS)
Hans Schillstrom61b1ab42011-01-03 14:44:42 +0100119 struct netns_ipvs *ipvs;
JunweiZhang8b4d14d2013-06-26 16:40:06 +0800120#endif
Andrey Vagin51d7ccc2012-07-16 04:28:49 +0000121 struct sock *diag_nlsk;
Nicolas Dichtelb42664f2012-09-10 22:09:44 +0000122 atomic_t rt_genid;
Timo Teräs5aad1de2013-05-27 20:46:33 +0000123 atomic_t fnhe_genid;
Eric W. Biederman5f256be2007-09-12 11:50:50 +0200124};
125
Pavel Emelyanov1fb94892012-08-08 21:53:36 +0000126/*
127 * ifindex generation is per-net namespace, and loopback is
128 * always the 1st device in ns (see net_dev_init), thus any
129 * loopback device should get ifindex 1
130 */
131
132#define LOOPBACK_IFINDEX 1
Denis V. Lunev225c0a02008-04-02 00:09:29 -0700133
Denis V. Lunevc0f39322008-04-02 00:10:28 -0700134#include <linux/seq_file_net.h>
135
Daniel Lezcano4fabcd72007-09-13 09:16:29 +0200136/* Init's network namespace */
Eric W. Biederman5f256be2007-09-12 11:50:50 +0200137extern struct net init_net;
Denis V. Luneva4aa8342008-04-03 13:04:33 -0700138
Eric W. Biedermand727abc2012-06-14 02:16:42 -0700139#ifdef CONFIG_NET_NS
Eric W. Biederman038e7332012-06-14 02:31:10 -0700140extern struct net *copy_net_ns(unsigned long flags,
141 struct user_namespace *user_ns, struct net *old_net);
Denis V. Lunev225c0a02008-04-02 00:09:29 -0700142
Eric W. Biedermand727abc2012-06-14 02:16:42 -0700143#else /* CONFIG_NET_NS */
144#include <linux/sched.h>
145#include <linux/nsproxy.h>
Eric W. Biederman038e7332012-06-14 02:31:10 -0700146static inline struct net *copy_net_ns(unsigned long flags,
147 struct user_namespace *user_ns, struct net *old_net)
Eric W. Biederman9dd776b2007-09-26 22:04:26 -0700148{
Eric W. Biedermand727abc2012-06-14 02:16:42 -0700149 if (flags & CLONE_NEWNET)
150 return ERR_PTR(-EINVAL);
151 return old_net;
Eric W. Biederman9dd776b2007-09-26 22:04:26 -0700152}
Eric W. Biedermand727abc2012-06-14 02:16:42 -0700153#endif /* CONFIG_NET_NS */
Denis V. Lunev225c0a02008-04-02 00:09:29 -0700154
155
156extern struct list_head net_namespace_list;
Eric W. Biederman9dd776b2007-09-26 22:04:26 -0700157
Johannes Berg30ffee82009-07-10 09:51:35 +0000158extern struct net *get_net_ns_by_pid(pid_t pid);
Eric W. Biedermanf0630522011-05-04 17:51:50 -0700159extern struct net *get_net_ns_by_fd(int pid);
Johannes Berg30ffee82009-07-10 09:51:35 +0000160
Pavel Emelyanovd4655792007-11-01 00:43:49 -0700161#ifdef CONFIG_NET_NS
Eric W. Biederman5f256be2007-09-12 11:50:50 +0200162extern void __put_net(struct net *net);
163
164static inline struct net *get_net(struct net *net)
165{
166 atomic_inc(&net->count);
167 return net;
168}
169
Eric W. Biederman077130c2007-09-13 09:18:57 +0200170static inline struct net *maybe_get_net(struct net *net)
171{
172 /* Used when we know struct net exists but we
173 * aren't guaranteed a previous reference count
174 * exists. If the reference count is zero this
175 * function fails and returns NULL.
176 */
177 if (!atomic_inc_not_zero(&net->count))
178 net = NULL;
179 return net;
180}
181
Eric W. Biederman5f256be2007-09-12 11:50:50 +0200182static inline void put_net(struct net *net)
183{
184 if (atomic_dec_and_test(&net->count))
185 __put_net(net);
186}
187
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +0900188static inline
189int net_eq(const struct net *net1, const struct net *net2)
190{
191 return net1 == net2;
192}
Al Viroa685e082011-06-08 21:13:01 -0400193
194extern void net_drop_ns(void *);
195
Pavel Emelyanovd4655792007-11-01 00:43:49 -0700196#else
Eric W. Biedermanb9f75f42008-06-20 22:16:51 -0700197
Pavel Emelyanovd4655792007-11-01 00:43:49 -0700198static inline struct net *get_net(struct net *net)
199{
200 return net;
201}
202
203static inline void put_net(struct net *net)
204{
205}
206
Pavel Emelyanovd4655792007-11-01 00:43:49 -0700207static inline struct net *maybe_get_net(struct net *net)
208{
209 return net;
210}
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +0900211
212static inline
213int net_eq(const struct net *net1, const struct net *net2)
214{
215 return 1;
216}
Al Viroa685e082011-06-08 21:13:01 -0400217
218#define net_drop_ns NULL
Pavel Emelyanovd4655792007-11-01 00:43:49 -0700219#endif
Eric W. Biederman5f256be2007-09-12 11:50:50 +0200220
Denis V. Lunev5d1e4462008-04-16 01:58:04 -0700221
222#ifdef NETNS_REFCNT_DEBUG
223static inline struct net *hold_net(struct net *net)
224{
225 if (net)
226 atomic_inc(&net->use_count);
227 return net;
228}
229
230static inline void release_net(struct net *net)
231{
232 if (net)
233 atomic_dec(&net->use_count);
234}
235#else
236static inline struct net *hold_net(struct net *net)
237{
238 return net;
239}
240
241static inline void release_net(struct net *net)
242{
243}
244#endif
245
Eric Dumazet8f424b52008-11-12 00:53:30 -0800246#ifdef CONFIG_NET_NS
247
248static inline void write_pnet(struct net **pnet, struct net *net)
249{
250 *pnet = net;
251}
252
253static inline struct net *read_pnet(struct net * const *pnet)
254{
255 return *pnet;
256}
257
258#else
259
260#define write_pnet(pnet, net) do { (void)(net);} while (0)
261#define read_pnet(pnet) (&init_net)
262
263#endif
Denis V. Lunev5d1e4462008-04-16 01:58:04 -0700264
Eric W. Biederman5f256be2007-09-12 11:50:50 +0200265#define for_each_net(VAR) \
266 list_for_each_entry(VAR, &net_namespace_list, list)
267
Johannes Berg11a28d32009-07-10 09:51:33 +0000268#define for_each_net_rcu(VAR) \
269 list_for_each_entry_rcu(VAR, &net_namespace_list, list)
270
Pavel Emelyanov46650792007-10-08 20:38:39 -0700271#ifdef CONFIG_NET_NS
272#define __net_init
273#define __net_exit
Denis V. Lunev022cbae2007-11-13 03:23:50 -0800274#define __net_initdata
Andi Kleen04a6f822012-10-04 17:12:11 -0700275#define __net_initconst
Pavel Emelyanov46650792007-10-08 20:38:39 -0700276#else
277#define __net_init __init
278#define __net_exit __exit_refok
Denis V. Lunev022cbae2007-11-13 03:23:50 -0800279#define __net_initdata __initdata
Andi Kleen04a6f822012-10-04 17:12:11 -0700280#define __net_initconst __initconst
Pavel Emelyanov46650792007-10-08 20:38:39 -0700281#endif
Eric W. Biederman5f256be2007-09-12 11:50:50 +0200282
283struct pernet_operations {
284 struct list_head list;
285 int (*init)(struct net *net);
286 void (*exit)(struct net *net);
Eric W. Biederman72ad9372009-12-03 02:29:03 +0000287 void (*exit_batch)(struct list_head *net_exit_list);
Eric W. Biedermanf875bae2009-11-29 22:25:28 +0000288 int *id;
289 size_t size;
Eric W. Biederman5f256be2007-09-12 11:50:50 +0200290};
291
Eric W. Biederman17edde52009-02-22 00:11:09 -0800292/*
293 * Use these carefully. If you implement a network device and it
294 * needs per network namespace operations use device pernet operations,
295 * otherwise use pernet subsys operations.
296 *
Johannes Berg4edf5472009-07-15 06:16:34 +0000297 * Network interfaces need to be removed from a dying netns _before_
298 * subsys notifiers can be called, as most of the network code cleanup
299 * (which is done from subsys notifiers) runs with the assumption that
300 * dev_remove_pack has been called so no new packets will arrive during
301 * and after the cleanup functions have been called. dev_remove_pack
302 * is not per namespace so instead the guarantee of no more packets
303 * arriving in a network namespace is provided by ensuring that all
304 * network devices and all sockets have left the network namespace
305 * before the cleanup methods are called.
Eric W. Biederman17edde52009-02-22 00:11:09 -0800306 *
307 * For the longest time the ipv4 icmp code was registered as a pernet
308 * device which caused kernel oops, and panics during network
309 * namespace cleanup. So please don't get this wrong.
310 */
Eric W. Biederman5f256be2007-09-12 11:50:50 +0200311extern int register_pernet_subsys(struct pernet_operations *);
312extern void unregister_pernet_subsys(struct pernet_operations *);
313extern int register_pernet_device(struct pernet_operations *);
314extern void unregister_pernet_device(struct pernet_operations *);
Eric W. Biedermanf875bae2009-11-29 22:25:28 +0000315
Eric W. Biederman95bdfcc2007-11-30 23:55:42 +1100316struct ctl_table;
317struct ctl_table_header;
Pavel Emelyanovd62c6122008-05-19 13:45:33 -0700318
Eric W. Biederman2ca794e2012-04-19 13:20:32 +0000319#ifdef CONFIG_SYSCTL
320extern int net_sysctl_init(void);
Eric W. Biedermanab41a2c2012-04-19 13:18:47 +0000321extern struct ctl_table_header *register_net_sysctl(struct net *net,
322 const char *path, struct ctl_table *table);
Eric W. Biederman95bdfcc2007-11-30 23:55:42 +1100323extern void unregister_net_sysctl_table(struct ctl_table_header *header);
Eric W. Biederman48c74952012-04-23 12:13:02 +0000324#else
325static inline int net_sysctl_init(void) { return 0; }
326static inline struct ctl_table_header *register_net_sysctl(struct net *net,
327 const char *path, struct ctl_table *table)
328{
329 return NULL;
330}
331static inline void unregister_net_sysctl_table(struct ctl_table_header *header)
332{
333}
334#endif
335
Nicolas Dichtelb42664f2012-09-10 22:09:44 +0000336static inline int rt_genid(struct net *net)
337{
338 return atomic_read(&net->rt_genid);
339}
340
341static inline void rt_genid_bump(struct net *net)
342{
343 atomic_inc(&net->rt_genid);
344}
Eric W. Biederman95bdfcc2007-11-30 23:55:42 +1100345
Timo Teräs5aad1de2013-05-27 20:46:33 +0000346static inline int fnhe_genid(struct net *net)
347{
348 return atomic_read(&net->fnhe_genid);
349}
350
351static inline void fnhe_genid_bump(struct net *net)
352{
353 atomic_inc(&net->fnhe_genid);
354}
355
Eric W. Biederman5f256be2007-09-12 11:50:50 +0200356#endif /* __NET_NET_NAMESPACE_H */