1SETNS(2) Linux Programmer's Manual SETNS(2)
2
3
4
6 setns - reassociate thread with a namespace
7
9 #define _GNU_SOURCE /* See feature_test_macros(7) */
10 #include <sched.h>
11
12 int setns(int fd, int nstype);
13
15 Given a file descriptor referring to a namespace, reassociate the call‐
16 ing thread with that namespace.
17
18 The fd argument is a file descriptor referring to one of the namespace
19 entries in a /proc/[pid]/ns/ directory; see namespaces(7) for further
20 information on /proc/[pid]/ns/. The calling thread will be reassoci‐
21 ated with the corresponding namespace, subject to any constraints
22 imposed by the nstype argument.
23
24 The nstype argument specifies which type of namespace the calling
25 thread may be reassociated with. This argument can have one of the
26 following values:
27
28 0 Allow any type of namespace to be joined.
29
30 CLONE_NEWCGROUP (since Linux 4.6)
31 fd must refer to a cgroup namespace.
32
33 CLONE_NEWIPC (since Linux 3.0)
34 fd must refer to an IPC namespace.
35
36 CLONE_NEWNET (since Linux 3.0)
37 fd must refer to a network namespace.
38
39 CLONE_NEWNS (since Linux 3.8)
40 fd must refer to a mount namespace.
41
42 CLONE_NEWPID (since Linux 3.8)
43 fd must refer to a descendant PID namespace.
44
45 CLONE_NEWUSER (since Linux 3.8)
46 fd must refer to a user namespace.
47
48 CLONE_NEWUTS (since Linux 3.0)
49 fd must refer to a UTS namespace.
50
51 Specifying nstype as 0 suffices if the caller knows (or does not care)
52 what type of namespace is referred to by fd. Specifying a nonzero
53 value for nstype is useful if the caller does not know what type of
54 namespace is referred to by fd and wants to ensure that the namespace
55 is of a particular type. (The caller might not know the type of the
56 namespace referred to by fd if the file descriptor was opened by
57 another process and, for example, passed to the caller via a UNIX
58 domain socket.)
59
60 If fd refers to a PID namespaces, the semantics are somewhat different
61 from other namespace types: reassociating the calling thread with a PID
62 namespace changes only the PID namespace that subsequently created
63 child processes of the caller will be placed in; it does not change the
64 PID namespace of the caller itself. Reassociating with a PID namespace
65 is allowed only if the PID namespace specified by fd is a descendant
66 (child, grandchild, etc.) of the PID namespace of the caller. For
67 further details on PID namespaces, see pid_namespaces(7).
68
69 A process reassociating itself with a user namespace must have the
70 CAP_SYS_ADMIN capability in the target user namespace. Upon success‐
71 fully joining a user namespace, a process is granted all capabilities
72 in that namespace, regardless of its user and group IDs. A multi‐
73 threaded process may not change user namespace with setns(). It is not
74 permitted to use setns() to reenter the caller's current user names‐
75 pace. This prevents a caller that has dropped capabilities from
76 regaining those capabilities via a call to setns(). For security rea‐
77 sons, a process can't join a new user namespace if it is sharing
78 filesystem-related attributes (the attributes whose sharing is con‐
79 trolled by the clone(2) CLONE_FS flag) with another process. For fur‐
80 ther details on user namespaces, see user_namespaces(7).
81
82 A process may not be reassociated with a new mount namespace if it is
83 multithreaded. Changing the mount namespace requires that the caller
84 possess both CAP_SYS_CHROOT and CAP_SYS_ADMIN capabilities in its own
85 user namespace and CAP_SYS_ADMIN in the target mount namespace. See
86 user_namespaces(7) for details on the interaction of user namespaces
87 and mount namespaces.
88
89 Using setns() to change the caller's cgroup namespace does not change
90 the caller's cgroup memberships.
91
93 On success, setns() returns 0. On failure, -1 is returned and errno is
94 set to indicate the error.
95
97 EBADF fd is not a valid file descriptor.
98
99 EINVAL fd refers to a namespace whose type does not match that speci‐
100 fied in nstype.
101
102 EINVAL There is problem with reassociating the thread with the speci‐
103 fied namespace.
104
105 EINVAL The caller tried to join an ancestor (parent, grandparent, and
106 so on) PID namespace.
107
108 EINVAL The caller attempted to join the user namespace in which it is
109 already a member.
110
111 EINVAL The caller shares filesystem (CLONE_FS) state (in particular,
112 the root directory) with other processes and tried to join a new
113 user namespace.
114
115 EINVAL The caller is multithreaded and tried to join a new user names‐
116 pace.
117
118 ENOMEM Cannot allocate sufficient memory to change the specified names‐
119 pace.
120
121 EPERM The calling thread did not have the required capability for this
122 operation.
123
125 The setns() system call first appeared in Linux in kernel 3.0; library
126 support was added to glibc in version 2.14.
127
129 The setns() system call is Linux-specific.
130
132 Not all of the attributes that can be shared when a new thread is cre‐
133 ated using clone(2) can be changed using setns().
134
136 The program below takes two or more arguments. The first argument
137 specifies the pathname of a namespace file in an existing
138 /proc/[pid]/ns/ directory. The remaining arguments specify a command
139 and its arguments. The program opens the namespace file, joins that
140 namespace using setns(), and executes the specified command inside that
141 namespace.
142
143 The following shell session demonstrates the use of this program (com‐
144 piled as a binary named ns_exec) in conjunction with the CLONE_NEWUTS
145 example program in the clone(2) man page (complied as a binary named
146 newuts).
147
148 We begin by executing the example program in clone(2) in the back‐
149 ground. That program creates a child in a separate UTS namespace. The
150 child changes the hostname in its namespace, and then both processes
151 display the hostnames in their UTS namespaces, so that we can see that
152 they are different.
153
154 $ su # Need privilege for namespace operations
155 Password:
156 # ./newuts bizarro &
157 [1] 3549
158 clone() returned 3550
159 uts.nodename in child: bizarro
160 uts.nodename in parent: antero
161 # uname -n # Verify hostname in the shell
162 antero
163
164 We then run the program shown below, using it to execute a shell.
165 Inside that shell, we verify that the hostname is the one set by the
166 child created by the first program:
167
168 # ./ns_exec /proc/3550/ns/uts /bin/bash
169 # uname -n # Executed in shell started by ns_exec
170 bizarro
171
172 Program source
173 #define _GNU_SOURCE
174 #include <fcntl.h>
175 #include <sched.h>
176 #include <unistd.h>
177 #include <stdlib.h>
178 #include <stdio.h>
179
180 #define errExit(msg) do { perror(msg); exit(EXIT_FAILURE); \
181 } while (0)
182
183 int
184 main(int argc, char *argv[])
185 {
186 int fd;
187
188 if (argc < 3) {
189 fprintf(stderr, "%s /proc/PID/ns/FILE cmd args...\n", argv[0]);
190 exit(EXIT_FAILURE);
191 }
192
193 fd = open(argv[1], O_RDONLY); /* Get file descriptor for namespace */
194 if (fd == -1)
195 errExit("open");
196
197 if (setns(fd, 0) == -1) /* Join that namespace */
198 errExit("setns");
199
200 execvp(argv[2], &argv[2]); /* Execute a command in namespace */
201 errExit("execvp");
202 }
203
205 nsenter(1), clone(2), fork(2), unshare(2), vfork(2), namespaces(7),
206 unix(7)
207
209 This page is part of release 4.15 of the Linux man-pages project. A
210 description of the project, information about reporting bugs, and the
211 latest version of this page, can be found at
212 https://www.kernel.org/doc/man-pages/.
213
214
215
216Linux 2017-09-15 SETNS(2)