1/*
2 *  Copyright (C) 2006 IBM Corporation
3 *
4 *  Author: Serge Hallyn <serue@us.ibm.com>
5 *
6 *  This program is free software; you can redistribute it and/or
7 *  modify it under the terms of the GNU General Public License as
8 *  published by the Free Software Foundation, version 2 of the
9 *  License.
10 *
11 *  Jun 2006 - namespaces support
12 *             OpenVZ, SWsoft Inc.
13 *             Pavel Emelianov <xemul@openvz.org>
14 */
15
16#include <linux/slab.h>
17#include <linux/export.h>
18#include <linux/nsproxy.h>
19#include <linux/init_task.h>
20#include <linux/mnt_namespace.h>
21#include <linux/utsname.h>
22#include <linux/pid_namespace.h>
23#include <net/net_namespace.h>
24#include <linux/ipc_namespace.h>
25#include <linux/proc_ns.h>
26#include <linux/file.h>
27#include <linux/syscalls.h>
28
29static struct kmem_cache *nsproxy_cachep;
30
31struct nsproxy init_nsproxy = {
32	.count			= ATOMIC_INIT(1),
33	.uts_ns			= &init_uts_ns,
34#if defined(CONFIG_POSIX_MQUEUE) || defined(CONFIG_SYSVIPC)
35	.ipc_ns			= &init_ipc_ns,
36#endif
37	.mnt_ns			= NULL,
38	.pid_ns_for_children	= &init_pid_ns,
39#ifdef CONFIG_NET
40	.net_ns			= &init_net,
41#endif
42};
43
44static inline struct nsproxy *create_nsproxy(void)
45{
46	struct nsproxy *nsproxy;
47
48	nsproxy = kmem_cache_alloc(nsproxy_cachep, GFP_KERNEL);
49	if (nsproxy)
50		atomic_set(&nsproxy->count, 1);
51	return nsproxy;
52}
53
54/*
55 * Create new nsproxy and all of its the associated namespaces.
56 * Return the newly created nsproxy.  Do not attach this to the task,
57 * leave it to the caller to do proper locking and attach it to task.
58 */
59static struct nsproxy *create_new_namespaces(unsigned long flags,
60	struct task_struct *tsk, struct user_namespace *user_ns,
61	struct fs_struct *new_fs)
62{
63	struct nsproxy *new_nsp;
64	int err;
65
66	new_nsp = create_nsproxy();
67	if (!new_nsp)
68		return ERR_PTR(-ENOMEM);
69
70	new_nsp->mnt_ns = copy_mnt_ns(flags, tsk->nsproxy->mnt_ns, user_ns, new_fs);
71	if (IS_ERR(new_nsp->mnt_ns)) {
72		err = PTR_ERR(new_nsp->mnt_ns);
73		goto out_ns;
74	}
75
76	new_nsp->uts_ns = copy_utsname(flags, user_ns, tsk->nsproxy->uts_ns);
77	if (IS_ERR(new_nsp->uts_ns)) {
78		err = PTR_ERR(new_nsp->uts_ns);
79		goto out_uts;
80	}
81
82	new_nsp->ipc_ns = copy_ipcs(flags, user_ns, tsk->nsproxy->ipc_ns);
83	if (IS_ERR(new_nsp->ipc_ns)) {
84		err = PTR_ERR(new_nsp->ipc_ns);
85		goto out_ipc;
86	}
87
88	new_nsp->pid_ns_for_children =
89		copy_pid_ns(flags, user_ns, tsk->nsproxy->pid_ns_for_children);
90	if (IS_ERR(new_nsp->pid_ns_for_children)) {
91		err = PTR_ERR(new_nsp->pid_ns_for_children);
92		goto out_pid;
93	}
94
95	new_nsp->net_ns = copy_net_ns(flags, user_ns, tsk->nsproxy->net_ns);
96	if (IS_ERR(new_nsp->net_ns)) {
97		err = PTR_ERR(new_nsp->net_ns);
98		goto out_net;
99	}
100
101	return new_nsp;
102
103out_net:
104	if (new_nsp->pid_ns_for_children)
105		put_pid_ns(new_nsp->pid_ns_for_children);
106out_pid:
107	if (new_nsp->ipc_ns)
108		put_ipc_ns(new_nsp->ipc_ns);
109out_ipc:
110	if (new_nsp->uts_ns)
111		put_uts_ns(new_nsp->uts_ns);
112out_uts:
113	if (new_nsp->mnt_ns)
114		put_mnt_ns(new_nsp->mnt_ns);
115out_ns:
116	kmem_cache_free(nsproxy_cachep, new_nsp);
117	return ERR_PTR(err);
118}
119
120/*
121 * called from clone.  This now handles copy for nsproxy and all
122 * namespaces therein.
123 */
124int copy_namespaces(unsigned long flags, struct task_struct *tsk)
125{
126	struct nsproxy *old_ns = tsk->nsproxy;
127	struct user_namespace *user_ns = task_cred_xxx(tsk, user_ns);
128	struct nsproxy *new_ns;
129
130	if (likely(!(flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
131			      CLONE_NEWPID | CLONE_NEWNET)))) {
132		get_nsproxy(old_ns);
133		return 0;
134	}
135
136	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
137		return -EPERM;
138
139	/*
140	 * CLONE_NEWIPC must detach from the undolist: after switching
141	 * to a new ipc namespace, the semaphore arrays from the old
142	 * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
143	 * means share undolist with parent, so we must forbid using
144	 * it along with CLONE_NEWIPC.
145	 */
146	if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
147		(CLONE_NEWIPC | CLONE_SYSVSEM))
148		return -EINVAL;
149
150	new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs);
151	if (IS_ERR(new_ns))
152		return  PTR_ERR(new_ns);
153
154	tsk->nsproxy = new_ns;
155	return 0;
156}
157
158void free_nsproxy(struct nsproxy *ns)
159{
160	if (ns->mnt_ns)
161		put_mnt_ns(ns->mnt_ns);
162	if (ns->uts_ns)
163		put_uts_ns(ns->uts_ns);
164	if (ns->ipc_ns)
165		put_ipc_ns(ns->ipc_ns);
166	if (ns->pid_ns_for_children)
167		put_pid_ns(ns->pid_ns_for_children);
168	put_net(ns->net_ns);
169	kmem_cache_free(nsproxy_cachep, ns);
170}
171
172/*
173 * Called from unshare. Unshare all the namespaces part of nsproxy.
174 * On success, returns the new nsproxy.
175 */
176int unshare_nsproxy_namespaces(unsigned long unshare_flags,
177	struct nsproxy **new_nsp, struct cred *new_cred, struct fs_struct *new_fs)
178{
179	struct user_namespace *user_ns;
180	int err = 0;
181
182	if (!(unshare_flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
183			       CLONE_NEWNET | CLONE_NEWPID)))
184		return 0;
185
186	user_ns = new_cred ? new_cred->user_ns : current_user_ns();
187	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
188		return -EPERM;
189
190	*new_nsp = create_new_namespaces(unshare_flags, current, user_ns,
191					 new_fs ? new_fs : current->fs);
192	if (IS_ERR(*new_nsp)) {
193		err = PTR_ERR(*new_nsp);
194		goto out;
195	}
196
197out:
198	return err;
199}
200
201void switch_task_namespaces(struct task_struct *p, struct nsproxy *new)
202{
203	struct nsproxy *ns;
204
205	might_sleep();
206
207	task_lock(p);
208	ns = p->nsproxy;
209	p->nsproxy = new;
210	task_unlock(p);
211
212	if (ns && atomic_dec_and_test(&ns->count))
213		free_nsproxy(ns);
214}
215
216void exit_task_namespaces(struct task_struct *p)
217{
218	switch_task_namespaces(p, NULL);
219}
220
221SYSCALL_DEFINE2(setns, int, fd, int, nstype)
222{
223	const struct proc_ns_operations *ops;
224	struct task_struct *tsk = current;
225	struct nsproxy *new_nsproxy;
226	struct proc_ns *ei;
227	struct file *file;
228	int err;
229
230	file = proc_ns_fget(fd);
231	if (IS_ERR(file))
232		return PTR_ERR(file);
233
234	err = -EINVAL;
235	ei = get_proc_ns(file_inode(file));
236	ops = ei->ns_ops;
237	if (nstype && (ops->type != nstype))
238		goto out;
239
240	new_nsproxy = create_new_namespaces(0, tsk, current_user_ns(), tsk->fs);
241	if (IS_ERR(new_nsproxy)) {
242		err = PTR_ERR(new_nsproxy);
243		goto out;
244	}
245
246	err = ops->install(new_nsproxy, ei->ns);
247	if (err) {
248		free_nsproxy(new_nsproxy);
249		goto out;
250	}
251	switch_task_namespaces(tsk, new_nsproxy);
252out:
253	fput(file);
254	return err;
255}
256
257int __init nsproxy_cache_init(void)
258{
259	nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC);
260	return 0;
261}
262