xref: /linux-6.15/kernel/nsproxy.c (revision dbef0c1c)
1 /*
2  *  Copyright (C) 2006 IBM Corporation
3  *
4  *  Author: Serge Hallyn <[email protected]>
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 <[email protected]>
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 
29 static struct kmem_cache *nsproxy_cachep;
30 
31 struct 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	= &init_pid_ns,
39 #ifdef CONFIG_NET
40 	.net_ns	= &init_net,
41 #endif
42 };
43 
44 static 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  */
59 static 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 = copy_pid_ns(flags, user_ns, tsk->nsproxy->pid_ns);
89 	if (IS_ERR(new_nsp->pid_ns)) {
90 		err = PTR_ERR(new_nsp->pid_ns);
91 		goto out_pid;
92 	}
93 
94 	new_nsp->net_ns = copy_net_ns(flags, user_ns, tsk->nsproxy->net_ns);
95 	if (IS_ERR(new_nsp->net_ns)) {
96 		err = PTR_ERR(new_nsp->net_ns);
97 		goto out_net;
98 	}
99 
100 	return new_nsp;
101 
102 out_net:
103 	if (new_nsp->pid_ns)
104 		put_pid_ns(new_nsp->pid_ns);
105 out_pid:
106 	if (new_nsp->ipc_ns)
107 		put_ipc_ns(new_nsp->ipc_ns);
108 out_ipc:
109 	if (new_nsp->uts_ns)
110 		put_uts_ns(new_nsp->uts_ns);
111 out_uts:
112 	if (new_nsp->mnt_ns)
113 		put_mnt_ns(new_nsp->mnt_ns);
114 out_ns:
115 	kmem_cache_free(nsproxy_cachep, new_nsp);
116 	return ERR_PTR(err);
117 }
118 
119 /*
120  * called from clone.  This now handles copy for nsproxy and all
121  * namespaces therein.
122  */
123 int copy_namespaces(unsigned long flags, struct task_struct *tsk)
124 {
125 	struct nsproxy *old_ns = tsk->nsproxy;
126 	struct user_namespace *user_ns = task_cred_xxx(tsk, user_ns);
127 	struct nsproxy *new_ns;
128 
129 	if (likely(!(flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
130 			      CLONE_NEWPID | CLONE_NEWNET)))) {
131 		get_nsproxy(old_ns);
132 		return 0;
133 	}
134 
135 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
136 		return -EPERM;
137 
138 	/*
139 	 * CLONE_NEWIPC must detach from the undolist: after switching
140 	 * to a new ipc namespace, the semaphore arrays from the old
141 	 * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
142 	 * means share undolist with parent, so we must forbid using
143 	 * it along with CLONE_NEWIPC.
144 	 */
145 	if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
146 		(CLONE_NEWIPC | CLONE_SYSVSEM))
147 		return -EINVAL;
148 
149 	new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs);
150 	if (IS_ERR(new_ns))
151 		return  PTR_ERR(new_ns);
152 
153 	tsk->nsproxy = new_ns;
154 	return 0;
155 }
156 
157 void free_nsproxy(struct nsproxy *ns)
158 {
159 	if (ns->mnt_ns)
160 		put_mnt_ns(ns->mnt_ns);
161 	if (ns->uts_ns)
162 		put_uts_ns(ns->uts_ns);
163 	if (ns->ipc_ns)
164 		put_ipc_ns(ns->ipc_ns);
165 	if (ns->pid_ns)
166 		put_pid_ns(ns->pid_ns);
167 	put_net(ns->net_ns);
168 	kmem_cache_free(nsproxy_cachep, ns);
169 }
170 
171 /*
172  * Called from unshare. Unshare all the namespaces part of nsproxy.
173  * On success, returns the new nsproxy.
174  */
175 int unshare_nsproxy_namespaces(unsigned long unshare_flags,
176 	struct nsproxy **new_nsp, struct cred *new_cred, struct fs_struct *new_fs)
177 {
178 	struct user_namespace *user_ns;
179 	int err = 0;
180 
181 	if (!(unshare_flags & (CLONE_NEWNS | CLONE_NEWUTS | CLONE_NEWIPC |
182 			       CLONE_NEWNET | CLONE_NEWPID)))
183 		return 0;
184 
185 	user_ns = new_cred ? new_cred->user_ns : current_user_ns();
186 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
187 		return -EPERM;
188 
189 	*new_nsp = create_new_namespaces(unshare_flags, current, user_ns,
190 					 new_fs ? new_fs : current->fs);
191 	if (IS_ERR(*new_nsp)) {
192 		err = PTR_ERR(*new_nsp);
193 		goto out;
194 	}
195 
196 out:
197 	return err;
198 }
199 
200 void switch_task_namespaces(struct task_struct *p, struct nsproxy *new)
201 {
202 	struct nsproxy *ns;
203 
204 	might_sleep();
205 
206 	ns = p->nsproxy;
207 
208 	rcu_assign_pointer(p->nsproxy, new);
209 
210 	if (ns && atomic_dec_and_test(&ns->count)) {
211 		/*
212 		 * wait for others to get what they want from this nsproxy.
213 		 *
214 		 * cannot release this nsproxy via the call_rcu() since
215 		 * put_mnt_ns() will want to sleep
216 		 */
217 		synchronize_rcu();
218 		free_nsproxy(ns);
219 	}
220 }
221 
222 void exit_task_namespaces(struct task_struct *p)
223 {
224 	switch_task_namespaces(p, NULL);
225 }
226 
227 SYSCALL_DEFINE2(setns, int, fd, int, nstype)
228 {
229 	const struct proc_ns_operations *ops;
230 	struct task_struct *tsk = current;
231 	struct nsproxy *new_nsproxy;
232 	struct proc_ns *ei;
233 	struct file *file;
234 	int err;
235 
236 	file = proc_ns_fget(fd);
237 	if (IS_ERR(file))
238 		return PTR_ERR(file);
239 
240 	err = -EINVAL;
241 	ei = get_proc_ns(file_inode(file));
242 	ops = ei->ns_ops;
243 	if (nstype && (ops->type != nstype))
244 		goto out;
245 
246 	new_nsproxy = create_new_namespaces(0, tsk, current_user_ns(), tsk->fs);
247 	if (IS_ERR(new_nsproxy)) {
248 		err = PTR_ERR(new_nsproxy);
249 		goto out;
250 	}
251 
252 	err = ops->install(new_nsproxy, ei->ns);
253 	if (err) {
254 		free_nsproxy(new_nsproxy);
255 		goto out;
256 	}
257 	switch_task_namespaces(tsk, new_nsproxy);
258 out:
259 	fput(file);
260 	return err;
261 }
262 
263 int __init nsproxy_cache_init(void)
264 {
265 	nsproxy_cachep = KMEM_CACHE(nsproxy, SLAB_PANIC);
266 	return 0;
267 }
268