1 .\" Copyright (C) 2006, Janak Desai <janak@us.ibm.com>
2 .\" and Copyright (C) 2006, 2012 Michael Kerrisk <mtk.manpages@gmail.com>
4 .\" %%%LICENSE_START(GPL_NOVERSION_ONELINE)
5 .\" Licensed under the GPL
8 .\" Patch Justification:
9 .\" unshare system call is needed to implement, using PAM,
10 .\" per-security_context and/or per-user namespace to provide
11 .\" polyinstantiated directories. Using unshare and bind mounts, a
12 .\" PAM module can create private namespace with appropriate
13 .\" directories(based on user's security context) bind mounted on
14 .\" public directories such as /tmp, thus providing an instance of
15 .\" /tmp that is based on user's security context. Without the
16 .\" unshare system call, namespace separation can only be achieved
17 .\" by clone, which would require porting and maintaining all commands
18 .\" such as login, and su, that establish a user session.
20 .TH UNSHARE 2 2017-05-03 "Linux" "Linux Programmer's Manual"
22 unshare \- disassociate parts of the process execution context
25 .B #define _GNU_SOURCE
28 .BI "int unshare(int " flags );
32 allows a process (or thread) to disassociate parts of its execution
33 context that are currently being shared with other processes (or threads).
34 Part of the execution context, such as the mount namespace, is shared
35 implicitly when a new process is created using
39 while other parts, such as virtual memory, may be
40 shared by explicit request when creating a process or thread using
45 is to allow a process to control its
46 shared execution context without creating a new process.
50 argument is a bit mask that specifies which parts of
51 the execution context should be unshared.
52 This argument is specified by ORing together zero or more
53 of the following constants:
56 Reverse the effect of the
60 Unshare the file descriptor table, so that the calling process
61 no longer shares its file descriptors with any other process.
64 Reverse the effect of the
68 Unshare filesystem attributes, so that the calling process
69 no longer shares its root directory
75 attributes with any other process.
77 .BR CLONE_NEWCGROUP " (since Linux 4.6)"
78 This flag has the same effect as the
82 Unshare the cgroup namespace.
89 .BR CLONE_NEWIPC " (since Linux 2.6.19)"
90 This flag has the same effect as the
94 Unshare the IPC namespace,
95 so that the calling process has a private copy of the
96 IPC namespace which is not shared with any other process.
97 Specifying this flag automatically implies
106 .BR CLONE_NEWNET " (since Linux 2.6.24)"
107 This flag has the same effect as the
111 Unshare the network namespace,
112 so that the calling process is moved into a
113 new network namespace which is not shared
114 with any previously existing process.
122 .\" These flag name are inconsistent:
123 .\" CLONE_NEWNS does the same thing in clone(), but CLONE_VM,
124 .\" CLONE_FS, and CLONE_FILES reverse the action of the clone()
125 .\" flags of the same name.
126 This flag has the same effect as the
130 Unshare the mount namespace,
131 so that the calling process has a private copy of
132 its namespace which is not shared with any other process.
133 Specifying this flag automatically implies
141 For further information, see
142 .BR mount_namespaces (7).
144 .BR CLONE_NEWPID " (since Linux 3.8)"
145 This flag has the same effect as the
149 Unshare the PID namespace,
150 so that the calling process has a new PID namespace for its children
151 which is not shared with any previously existing process.
152 The calling process is
154 moved into the new namespace.
155 The first child created by the calling process will have
156 the process ID 1 and will assume the role of
158 in the new namespace.
160 automatically implies
168 For further information, see
169 .BR pid_namespaces (7).
171 .BR CLONE_NEWUSER " (since Linux 3.8)"
172 This flag has the same effect as the
176 Unshare the user namespace,
177 so that the calling process is moved into a new user namespace
178 which is not shared with any previously existing process.
179 As with the child process created by
183 flag, the caller obtains a full set of capabilities in the new namespace.
186 requires that the calling process is not threaded; specifying
188 automatically implies
191 .\" commit e66eded8309ebf679d3d3c1f5820d1f2ca332c71
192 .\" https://lwn.net/Articles/543273/
194 also automatically implies
197 requires that the user ID and group ID
198 of the calling process are mapped to user IDs and group IDs in the
199 user namespace of the calling process at the time of the call.
201 For further information on user namespaces, see
202 .BR user_namespaces (7).
204 .BR CLONE_NEWUTS " (since Linux 2.6.19)"
205 This flag has the same effect as the
209 Unshare the UTS IPC namespace,
210 so that the calling process has a private copy of the
211 UTS namespace which is not shared with any other process.
218 .BR CLONE_SYSVSEM " (since Linux 2.6.26)
219 .\" commit 9edff4ab1f8d82675277a04e359d0ed8bf14a7b7
220 This flag reverses the effect of the
224 Unshare System\ V semaphore adjustment
227 so that the calling process has a new empty
229 list that is not shared with any other process.
230 If this is the last process that has a reference to the process's current
232 list, then the adjustments in that list are applied
233 to the corresponding semaphores, as described in
235 .\" CLONE_NEWNS If CLONE_SIGHAND is set and signals are also being shared
236 .\" (i.e., current->signal->count > 1), force CLONE_THREAD.
245 if the caller is single threaded (i.e., it is not sharing
246 its address space with another process or thread).
247 In this case, these flags have no effect.
248 (Note also that specifying
250 automatically implies
254 automatically implies
256 .\" As at 3.9, the following forced implications also apply,
257 .\" although the relevant flags are not yet implemented.
258 .\" If CLONE_THREAD is set force CLONE_VM.
259 .\" If CLONE_VM is set, force CLONE_SIGHAND.
261 If the process is multithreaded, then
262 the use of these flags results in an error.
263 .\" See kernel/fork.c::check_unshare_flags()
267 is specified as zero, then
270 no changes are made to the calling process's execution context.
272 On success, zero returned.
273 On failure, \-1 is returned and
275 is set to indicate the error.
279 An invalid bit was specified in
289 and the caller is multithreaded.
292 Cannot allocate sufficient memory to copy parts of caller's
293 context that need to be unshared.
295 .BR ENOSPC " (since Linux 3.7)"
296 .\" commit f2302505775fd13ba93f034206f1e2a587017929
298 was specified in flags,
299 but the limit on the nesting depth of PID namespaces
300 would have been exceeded; see
301 .BR pid_namespaces (7).
303 .BR ENOSPC " (since Linux 4.9; beforehand " EUSERS )
307 and the call would cause the limit on the number of
308 nested user namespaces to be exceeded.
310 .BR user_namespaces (7).
312 From Linux 3.11 to Linux 4.8, the error diagnosed in this case was
315 .BR ENOSPC " (since Linux 4.9)"
318 specified the creation of a new user namespace,
319 but doing so would have caused the limit defined by the corresponding file in
322 For further details, see
326 The calling process did not have the required privileges for this operation.
332 but either the effective user ID or the effective group ID of the caller
333 does not have a mapping in the parent namespace (see
334 .BR user_namespaces (7)).
336 .BR EPERM " (since Linux 3.9)"
337 .\" commit 3151527ee007b73a0ebd296010f1c0454a919c7d
341 and the caller is in a chroot environment
342 .\" FIXME What is the rationale for this restriction?
343 (i.e., the caller's root directory does not match the root directory
344 of the mount namespace in which it resides).
346 .BR EUSERS " (since Linux 3.11)"
350 and the limit on the number of nested user namespaces would be exceeded.
351 See the discussion of the
357 system call was added to Linux in kernel 2.6.16.
361 system call is Linux-specific.
363 Not all of the process attributes that can be shared when
364 a new process is created using
366 can be unshared using
368 In particular, as at kernel 3.8,
369 .\" FIXME all of the following needs to be reviewed for the current kernel
371 does not implement flags that reverse the effects of
373 .\" However, we can do unshare(CLONE_SIGHAND) if CLONE_SIGHAND
374 .\" was not specified when doing clone(); i.e., unsharing
375 .\" signal handlers is permitted if we are not actually
376 .\" sharing signal handlers. mtk
380 .\" However, we can do unshare(CLONE_VM) if CLONE_VM
381 .\" was not specified when doing clone(); i.e., unsharing
382 .\" virtual memory is permitted if we are not actually
383 .\" sharing virtual memory. mtk
384 Such functionality may be added in the future, if required.
388 .\"The current implementation of unshare does not allow unsharing of
389 .\"signals and signal handlers. Signals are complex to begin with and
390 .\"to unshare signals and/or signal handlers of a currently running
391 .\"process is even more complex. If in the future there is a specific
392 .\"need to allow unsharing of signals and/or signal handlers, it can
393 .\"be incrementally added to unshare without affecting legacy
394 .\"applications using unshare.
397 The program below provides a simple implementation of the
399 command, which unshares one or more namespaces and executes the
400 command supplied in its command-line arguments.
401 Here's an example of the use of this program,
402 running a shell in a new mount namespace,
403 and verifying that the original shell and the
404 new shell are in separate mount namespaces:
408 $ \fBreadlink /proc/$$/ns/mnt\fP
410 $ \fBsudo ./unshare -m /bin/bash\fP
411 [sudo] password for cecilia:
412 # \fBreadlink /proc/$$/ns/mnt\fP
417 The differing output of the two
419 commands shows that the two shells are in different mount namespaces.
425 A simple implementation of the unshare(1) command: unshare
426 namespaces and execute a command.
434 /* A simple error\-handling function: print an error message based
435 on the value in \(aqerrno\(aq and terminate the calling process */
437 #define errExit(msg) do { perror(msg); exit(EXIT_FAILURE); \\
443 fprintf(stderr, "Usage: %s [options] program [arg...]\\n", pname);
444 fprintf(stderr, "Options can be:\\n");
445 fprintf(stderr, " \-i unshare IPC namespace\\n");
446 fprintf(stderr, " \-m unshare mount namespace\\n");
447 fprintf(stderr, " \-n unshare network namespace\\n");
448 fprintf(stderr, " \-p unshare PID namespace\\n");
449 fprintf(stderr, " \-u unshare UTS namespace\\n");
450 fprintf(stderr, " \-U unshare user namespace\\n");
455 main(int argc, char *argv[])
461 while ((opt = getopt(argc, argv, "imnpuU")) != \-1) {
463 case \(aqi\(aq: flags |= CLONE_NEWIPC; break;
464 case \(aqm\(aq: flags |= CLONE_NEWNS; break;
465 case \(aqn\(aq: flags |= CLONE_NEWNET; break;
466 case \(aqp\(aq: flags |= CLONE_NEWPID; break;
467 case \(aqu\(aq: flags |= CLONE_NEWUTS; break;
468 case \(aqU\(aq: flags |= CLONE_NEWUSER; break;
469 default: usage(argv[0]);
476 if (unshare(flags) == \-1)
479 execvp(argv[optind], &argv[optind]);
492 .I Documentation/unshare.txt
493 in the Linux kernel source tree