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#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
27#include <net/netns/conntrack.h>
28#endif
29#if IS_ENABLED(CONFIG_NF_FLOW_TABLE)
30#include <net/netns/flow_table.h>
31#endif
32#include <net/netns/nftables.h>
33#include <net/netns/xfrm.h>
34#include <net/netns/mpls.h>
35#include <net/netns/can.h>
36#include <net/netns/xdp.h>
37#include <net/netns/smc.h>
38#include <net/netns/bpf.h>
39#include <net/netns/mctp.h>
40#include <net/net_trackers.h>
41#include <linux/ns_common.h>
42#include <linux/idr.h>
43#include <linux/skbuff.h>
44#include <linux/notifier.h>
45#include <linux/xarray.h>
46
47struct user_namespace;
48struct proc_dir_entry;
49struct net_device;
50struct sock;
51struct ctl_table_header;
52struct net_generic;
53struct uevent_sock;
54struct netns_ipvs;
55struct bpf_prog;
56
57
58#define NETDEV_HASHBITS    8
59#define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
60
61struct net {
62	/* First cache line can be often dirtied.
63	 * Do not place here read-mostly fields.
64	 */
65	refcount_t		passive;	/* To decide when the network
66						 * namespace should be freed.
67						 */
68	spinlock_t		rules_mod_lock;
69
70	unsigned int		dev_base_seq;	/* protected by rtnl_mutex */
71	u32			ifindex;
72
73	spinlock_t		nsid_lock;
74	atomic_t		fnhe_genid;
75
76	struct list_head	list;		/* list of network namespaces */
77	struct list_head	exit_list;	/* To linked to call pernet exit
78						 * methods on dead net (
79						 * pernet_ops_rwsem read locked),
80						 * or to unregister pernet ops
81						 * (pernet_ops_rwsem write locked).
82						 */
83	struct llist_node	cleanup_list;	/* namespaces on death row */
84
85#ifdef CONFIG_KEYS
86	struct key_tag		*key_domain;	/* Key domain of operation tag */
87#endif
88	struct user_namespace   *user_ns;	/* Owning user namespace */
89	struct ucounts		*ucounts;
90	struct idr		netns_ids;
91
92	struct ns_common	ns;
93	struct ref_tracker_dir  refcnt_tracker;
94	struct ref_tracker_dir  notrefcnt_tracker; /* tracker for objects not
95						    * refcounted against netns
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 xarray		dev_by_index;
113	struct raw_notifier_head	netdev_chain;
114
115	/* Note that @hash_mix can be read millions times per second,
116	 * it is critical that it is on a read_mostly cache line.
117	 */
118	u32			hash_mix;
119
120	struct net_device       *loopback_dev;          /* The loopback */
121
122	/* core fib_rules */
123	struct list_head	rules_ops;
124
125	struct netns_core	core;
126	struct netns_mib	mib;
127	struct netns_packet	packet;
128#if IS_ENABLED(CONFIG_UNIX)
129	struct netns_unix	unx;
130#endif
131	struct netns_nexthop	nexthop;
132	struct netns_ipv4	ipv4;
133#if IS_ENABLED(CONFIG_IPV6)
134	struct netns_ipv6	ipv6;
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#ifdef CONFIG_NETFILTER
143	struct netns_nf		nf;
144#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
145	struct netns_ct		ct;
146#endif
147#if defined(CONFIG_NF_TABLES) || defined(CONFIG_NF_TABLES_MODULE)
148	struct netns_nftables	nft;
149#endif
150#if IS_ENABLED(CONFIG_NF_FLOW_TABLE)
151	struct netns_ft ft;
152#endif
153#endif
154#ifdef CONFIG_WEXT_CORE
155	struct sk_buff_head	wext_nlevents;
156#endif
157	struct net_generic __rcu	*gen;
158
159	/* Used to store attached BPF programs */
160	struct netns_bpf	bpf;
161
162	/* Note : following structs are cache line aligned */
163#ifdef CONFIG_XFRM
164	struct netns_xfrm	xfrm;
165#endif
166
167	u64			net_cookie; /* written once */
168
169#if IS_ENABLED(CONFIG_IP_VS)
170	struct netns_ipvs	*ipvs;
171#endif
172#if IS_ENABLED(CONFIG_MPLS)
173	struct netns_mpls	mpls;
174#endif
175#if IS_ENABLED(CONFIG_CAN)
176	struct netns_can	can;
177#endif
178#ifdef CONFIG_XDP_SOCKETS
179	struct netns_xdp	xdp;
180#endif
181#if IS_ENABLED(CONFIG_MCTP)
182	struct netns_mctp	mctp;
183#endif
184#if IS_ENABLED(CONFIG_CRYPTO_USER)
185	struct sock		*crypto_nlsk;
186#endif
187	struct sock		*diag_nlsk;
188#if IS_ENABLED(CONFIG_SMC)
189	struct netns_smc	smc;
190#endif
191} __randomize_layout;
192
193#include <linux/seq_file_net.h>
194
195/* Init's network namespace */
196extern struct net init_net;
197
198#ifdef CONFIG_NET_NS
199struct net *copy_net_ns(unsigned long flags, struct user_namespace *user_ns,
200			struct net *old_net);
201
202void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid);
203
204void net_ns_barrier(void);
205
206struct ns_common *get_net_ns(struct ns_common *ns);
207struct net *get_net_ns_by_fd(int fd);
208#else /* CONFIG_NET_NS */
209#include <linux/sched.h>
210#include <linux/nsproxy.h>
211static inline struct net *copy_net_ns(unsigned long flags,
212	struct user_namespace *user_ns, struct net *old_net)
213{
214	if (flags & CLONE_NEWNET)
215		return ERR_PTR(-EINVAL);
216	return old_net;
217}
218
219static inline void net_ns_get_ownership(const struct net *net,
220					kuid_t *uid, kgid_t *gid)
221{
222	*uid = GLOBAL_ROOT_UID;
223	*gid = GLOBAL_ROOT_GID;
224}
225
226static inline void net_ns_barrier(void) {}
227
228static inline struct ns_common *get_net_ns(struct ns_common *ns)
229{
230	return ERR_PTR(-EINVAL);
231}
232
233static inline struct net *get_net_ns_by_fd(int fd)
234{
235	return ERR_PTR(-EINVAL);
236}
237#endif /* CONFIG_NET_NS */
238
239
240extern struct list_head net_namespace_list;
241
242struct net *get_net_ns_by_pid(pid_t pid);
243
244#ifdef CONFIG_SYSCTL
245void ipx_register_sysctl(void);
246void ipx_unregister_sysctl(void);
247#else
248#define ipx_register_sysctl()
249#define ipx_unregister_sysctl()
250#endif
251
252#ifdef CONFIG_NET_NS
253void __put_net(struct net *net);
254
255/* Try using get_net_track() instead */
256static inline struct net *get_net(struct net *net)
257{
258	refcount_inc(&net->ns.count);
259	return net;
260}
261
262static inline struct net *maybe_get_net(struct net *net)
263{
264	/* Used when we know struct net exists but we
265	 * aren't guaranteed a previous reference count
266	 * exists.  If the reference count is zero this
267	 * function fails and returns NULL.
268	 */
269	if (!refcount_inc_not_zero(&net->ns.count))
270		net = NULL;
271	return net;
272}
273
274/* Try using put_net_track() instead */
275static inline void put_net(struct net *net)
276{
277	if (refcount_dec_and_test(&net->ns.count))
278		__put_net(net);
279}
280
281static inline
282int net_eq(const struct net *net1, const struct net *net2)
283{
284	return net1 == net2;
285}
286
287static inline int check_net(const struct net *net)
288{
289	return refcount_read(&net->ns.count) != 0;
290}
291
292void net_drop_ns(void *);
293
294#else
295
296static inline struct net *get_net(struct net *net)
297{
298	return net;
299}
300
301static inline void put_net(struct net *net)
302{
303}
304
305static inline struct net *maybe_get_net(struct net *net)
306{
307	return net;
308}
309
310static inline
311int net_eq(const struct net *net1, const struct net *net2)
312{
313	return 1;
314}
315
316static inline int check_net(const struct net *net)
317{
318	return 1;
319}
320
321#define net_drop_ns NULL
322#endif
323
324
325static inline void __netns_tracker_alloc(struct net *net,
326					 netns_tracker *tracker,
327					 bool refcounted,
328					 gfp_t gfp)
329{
330#ifdef CONFIG_NET_NS_REFCNT_TRACKER
331	ref_tracker_alloc(refcounted ? &net->refcnt_tracker :
332				       &net->notrefcnt_tracker,
333			  tracker, gfp);
334#endif
335}
336
337static inline void netns_tracker_alloc(struct net *net, netns_tracker *tracker,
338				       gfp_t gfp)
339{
340	__netns_tracker_alloc(net, tracker, true, gfp);
341}
342
343static inline void __netns_tracker_free(struct net *net,
344					netns_tracker *tracker,
345					bool refcounted)
346{
347#ifdef CONFIG_NET_NS_REFCNT_TRACKER
348       ref_tracker_free(refcounted ? &net->refcnt_tracker :
349				     &net->notrefcnt_tracker, tracker);
350#endif
351}
352
353static inline struct net *get_net_track(struct net *net,
354					netns_tracker *tracker, gfp_t gfp)
355{
356	get_net(net);
357	netns_tracker_alloc(net, tracker, gfp);
358	return net;
359}
360
361static inline void put_net_track(struct net *net, netns_tracker *tracker)
362{
363	__netns_tracker_free(net, tracker, true);
364	put_net(net);
365}
366
367typedef struct {
368#ifdef CONFIG_NET_NS
369	struct net __rcu *net;
370#endif
371} possible_net_t;
372
373static inline void write_pnet(possible_net_t *pnet, struct net *net)
374{
375#ifdef CONFIG_NET_NS
376	rcu_assign_pointer(pnet->net, net);
377#endif
378}
379
380static inline struct net *read_pnet(const possible_net_t *pnet)
381{
382#ifdef CONFIG_NET_NS
383	return rcu_dereference_protected(pnet->net, true);
384#else
385	return &init_net;
386#endif
387}
388
389static inline struct net *read_pnet_rcu(possible_net_t *pnet)
390{
391#ifdef CONFIG_NET_NS
392	return rcu_dereference(pnet->net);
393#else
394	return &init_net;
395#endif
396}
397
398/* Protected by net_rwsem */
399#define for_each_net(VAR)				\
400	list_for_each_entry(VAR, &net_namespace_list, list)
401#define for_each_net_continue_reverse(VAR)		\
402	list_for_each_entry_continue_reverse(VAR, &net_namespace_list, list)
403#define for_each_net_rcu(VAR)				\
404	list_for_each_entry_rcu(VAR, &net_namespace_list, list)
405
406#ifdef CONFIG_NET_NS
407#define __net_init
408#define __net_exit
409#define __net_initdata
410#define __net_initconst
411#else
412#define __net_init	__init
413#define __net_exit	__ref
414#define __net_initdata	__initdata
415#define __net_initconst	__initconst
416#endif
417
418int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp);
419int peernet2id(const struct net *net, struct net *peer);
420bool peernet_has_id(const struct net *net, struct net *peer);
421struct net *get_net_ns_by_id(const struct net *net, int id);
422
423struct pernet_operations {
424	struct list_head list;
425	/*
426	 * Below methods are called without any exclusive locks.
427	 * More than one net may be constructed and destructed
428	 * in parallel on several cpus. Every pernet_operations
429	 * have to keep in mind all other pernet_operations and
430	 * to introduce a locking, if they share common resources.
431	 *
432	 * The only time they are called with exclusive lock is
433	 * from register_pernet_subsys(), unregister_pernet_subsys()
434	 * register_pernet_device() and unregister_pernet_device().
435	 *
436	 * Exit methods using blocking RCU primitives, such as
437	 * synchronize_rcu(), should be implemented via exit_batch.
438	 * Then, destruction of a group of net requires single
439	 * synchronize_rcu() related to these pernet_operations,
440	 * instead of separate synchronize_rcu() for every net.
441	 * Please, avoid synchronize_rcu() at all, where it's possible.
442	 *
443	 * Note that a combination of pre_exit() and exit() can
444	 * be used, since a synchronize_rcu() is guaranteed between
445	 * the calls.
446	 */
447	int (*init)(struct net *net);
448	void (*pre_exit)(struct net *net);
449	void (*exit)(struct net *net);
450	void (*exit_batch)(struct list_head *net_exit_list);
451	/* Following method is called with RTNL held. */
452	void (*exit_batch_rtnl)(struct list_head *net_exit_list,
453				struct list_head *dev_kill_list);
454	unsigned int *id;
455	size_t size;
456};
457
458/*
459 * Use these carefully.  If you implement a network device and it
460 * needs per network namespace operations use device pernet operations,
461 * otherwise use pernet subsys operations.
462 *
463 * Network interfaces need to be removed from a dying netns _before_
464 * subsys notifiers can be called, as most of the network code cleanup
465 * (which is done from subsys notifiers) runs with the assumption that
466 * dev_remove_pack has been called so no new packets will arrive during
467 * and after the cleanup functions have been called.  dev_remove_pack
468 * is not per namespace so instead the guarantee of no more packets
469 * arriving in a network namespace is provided by ensuring that all
470 * network devices and all sockets have left the network namespace
471 * before the cleanup methods are called.
472 *
473 * For the longest time the ipv4 icmp code was registered as a pernet
474 * device which caused kernel oops, and panics during network
475 * namespace cleanup.   So please don't get this wrong.
476 */
477int register_pernet_subsys(struct pernet_operations *);
478void unregister_pernet_subsys(struct pernet_operations *);
479int register_pernet_device(struct pernet_operations *);
480void unregister_pernet_device(struct pernet_operations *);
481
482struct ctl_table;
483
484#define register_net_sysctl(net, path, table)	\
485	register_net_sysctl_sz(net, path, table, ARRAY_SIZE(table))
486#ifdef CONFIG_SYSCTL
487int net_sysctl_init(void);
488struct ctl_table_header *register_net_sysctl_sz(struct net *net, const char *path,
489					     struct ctl_table *table, size_t table_size);
490void unregister_net_sysctl_table(struct ctl_table_header *header);
491#else
492static inline int net_sysctl_init(void) { return 0; }
493static inline struct ctl_table_header *register_net_sysctl_sz(struct net *net,
494	const char *path, struct ctl_table *table, size_t table_size)
495{
496	return NULL;
497}
498static inline void unregister_net_sysctl_table(struct ctl_table_header *header)
499{
500}
501#endif
502
503static inline int rt_genid_ipv4(const struct net *net)
504{
505	return atomic_read(&net->ipv4.rt_genid);
506}
507
508#if IS_ENABLED(CONFIG_IPV6)
509static inline int rt_genid_ipv6(const struct net *net)
510{
511	return atomic_read(&net->ipv6.fib6_sernum);
512}
513#endif
514
515static inline void rt_genid_bump_ipv4(struct net *net)
516{
517	atomic_inc(&net->ipv4.rt_genid);
518}
519
520extern void (*__fib6_flush_trees)(struct net *net);
521static inline void rt_genid_bump_ipv6(struct net *net)
522{
523	if (__fib6_flush_trees)
524		__fib6_flush_trees(net);
525}
526
527#if IS_ENABLED(CONFIG_IEEE802154_6LOWPAN)
528static inline struct netns_ieee802154_lowpan *
529net_ieee802154_lowpan(struct net *net)
530{
531	return &net->ieee802154_lowpan;
532}
533#endif
534
535/* For callers who don't really care about whether it's IPv4 or IPv6 */
536static inline void rt_genid_bump_all(struct net *net)
537{
538	rt_genid_bump_ipv4(net);
539	rt_genid_bump_ipv6(net);
540}
541
542static inline int fnhe_genid(const struct net *net)
543{
544	return atomic_read(&net->fnhe_genid);
545}
546
547static inline void fnhe_genid_bump(struct net *net)
548{
549	atomic_inc(&net->fnhe_genid);
550}
551
552#ifdef CONFIG_NET
553void net_ns_init(void);
554#else
555static inline void net_ns_init(void) {}
556#endif
557
558#endif /* __NET_NET_NAMESPACE_H */
559