2unshare system call:
4This document describes the new system call, unshare. The document
5provides an overview of the feature, why it is needed, how it can
6be used, its interface specification, design, implementation and
7how it can be tested.
9Change Log:
11version 0.1 Initial document, Janak Desai (, Jan 11, 2006
15    1) Overview
16    2) Benefits
17    3) Cost
18    4) Requirements
19    5) Functional Specification
20    6) High Level Design
21    7) Low Level Design
22    8) Test Specification
23    9) Future Work
251) Overview
27Most legacy operating system kernels support an abstraction of threads
28as multiple execution contexts within a process. These kernels provide
29special resources and mechanisms to maintain these "threads". The Linux
30kernel, in a clever and simple manner, does not make distinction
31between processes and "threads". The kernel allows processes to share
32resources and thus they can achieve legacy "threads" behavior without
33requiring additional data structures and mechanisms in the kernel. The
34power of implementing threads in this manner comes not only from
35its simplicity but also from allowing application programmers to work
36outside the confinement of all-or-nothing shared resources of legacy
37threads. On Linux, at the time of thread creation using the clone system
38call, applications can selectively choose which resources to share
39between threads.
41unshare system call adds a primitive to the Linux thread model that
42allows threads to selectively 'unshare' any resources that were being
43shared at the time of their creation. unshare was conceptualized by
44Al Viro in the August of 2000, on the Linux-Kernel mailing list, as part
45of the discussion on POSIX threads on Linux. unshare augments the
46usefulness of Linux threads for applications that would like to control
47shared resources without creating a new process. unshare is a natural
48addition to the set of available primitives on Linux that implement
49the concept of process/thread as a virtual machine.
512) Benefits
53unshare would be useful to large application frameworks such as PAM
54where creating a new process to control sharing/unsharing of process
55resources is not possible. Since namespaces are shared by default
56when creating a new process using fork or clone, unshare can benefit
57even non-threaded applications if they have a need to disassociate
58from default shared namespace. The following lists two use-cases
59where unshare can be used.
612.1 Per-security context namespaces
63unshare can be used to implement polyinstantiated directories using
64the kernel's per-process namespace mechanism. Polyinstantiated directories,
65such as per-user and/or per-security context instance of /tmp, /var/tmp or
66per-security context instance of a user's home directory, isolate user
67processes when working with these directories. Using unshare, a PAM
68module can easily setup a private namespace for a user at login.
69Polyinstantiated directories are required for Common Criteria certification
70with Labeled System Protection Profile, however, with the availability
71of shared-tree feature in the Linux kernel, even regular Linux systems
72can benefit from setting up private namespaces at login and
73polyinstantiating /tmp, /var/tmp and other directories deemed
74appropriate by system administrators.
762.2 unsharing of virtual memory and/or open files
78Consider a client/server application where the server is processing
79client requests by creating processes that share resources such as
80virtual memory and open files. Without unshare, the server has to
81decide what needs to be shared at the time of creating the process
82which services the request. unshare allows the server an ability to
83disassociate parts of the context during the servicing of the
84request. For large and complex middleware application frameworks, this
85ability to unshare after the process was created can be very
883) Cost
90In order to not duplicate code and to handle the fact that unshare
91works on an active task (as opposed to clone/fork working on a newly
92allocated inactive task) unshare had to make minor reorganizational
93changes to copy_* functions utilized by clone/fork system call.
94There is a cost associated with altering existing, well tested and
95stable code to implement a new feature that may not get exercised
96extensively in the beginning. However, with proper design and code
97review of the changes and creation of an unshare test for the LTP
98the benefits of this new feature can exceed its cost.
1004) Requirements
102unshare reverses sharing that was done using clone(2) system call,
103so unshare should have a similar interface as clone(2). That is,
104since flags in clone(int flags, void *stack) specifies what should
105be shared, similar flags in unshare(int flags) should specify
106what should be unshared. Unfortunately, this may appear to invert
107the meaning of the flags from the way they are used in clone(2).
108However, there was no easy solution that was less confusing and that
109allowed incremental context unsharing in future without an ABI change.
111unshare interface should accommodate possible future addition of
112new context flags without requiring a rebuild of old applications.
113If and when new context flags are added, unshare design should allow
114incremental unsharing of those resources on an as needed basis.
1165) Functional Specification
119    unshare - disassociate parts of the process execution context
122    #include <sched.h>
124    int unshare(int flags);
127    unshare allows a process to disassociate parts of its execution
128    context that are currently being shared with other processes. Part
129    of execution context, such as the namespace, is shared by default
130    when a new process is created using fork(2), while other parts,
131    such as the virtual memory, open file descriptors, etc, may be
132    shared by explicit request to share them when creating a process
133    using clone(2).
135    The main use of unshare is to allow a process to control its
136    shared execution context without creating a new process.
138    The flags argument specifies one or bitwise-or'ed of several of
139    the following constants.
141    CLONE_FS
142        If CLONE_FS is set, file system information of the caller
143        is disassociated from the shared file system information.
146        If CLONE_FILES is set, the file descriptor table of the
147        caller is disassociated from the shared file descriptor
148        table.
151        If CLONE_NEWNS is set, the namespace of the caller is
152        disassociated from the shared namespace.
154    CLONE_VM
155        If CLONE_VM is set, the virtual memory of the caller is
156        disassociated from the shared virtual memory.
159    On success, zero returned. On failure, -1 is returned and errno is
162    EPERM CLONE_NEWNS was specified by a non-root process (process
163        without CAP_SYS_ADMIN).
165    ENOMEM Cannot allocate sufficient memory to copy parts of caller's
166        context that need to be unshared.
168    EINVAL Invalid flag was specified as an argument.
171    The unshare() call is Linux-specific and should not be used
172    in programs intended to be portable.
175    clone(2), fork(2)
1776) High Level Design
179Depending on the flags argument, the unshare system call allocates
180appropriate process context structures, populates it with values from
181the current shared version, associates newly duplicated structures
182with the current task structure and releases corresponding shared
183versions. Helper functions of clone (copy_*) could not be used
184directly by unshare because of the following two reasons.
185  1) clone operates on a newly allocated not-yet-active task
186     structure, where as unshare operates on the current active
187     task. Therefore unshare has to take appropriate task_lock()
188     before associating newly duplicated context structures
189  2) unshare has to allocate and duplicate all context structures
190     that are being unshared, before associating them with the
191     current task and releasing older shared structures. Failure
192     do so will create race conditions and/or oops when trying
193     to backout due to an error. Consider the case of unsharing
194     both virtual memory and namespace. After successfully unsharing
195     vm, if the system call encounters an error while allocating
196     new namespace structure, the error return code will have to
197     reverse the unsharing of vm. As part of the reversal the
198     system call will have to go back to older, shared, vm
199     structure, which may not exist anymore.
201Therefore code from copy_* functions that allocated and duplicated
202current context structure was moved into new dup_* functions. Now,
203copy_* functions call dup_* functions to allocate and duplicate
204appropriate context structures and then associate them with the
205task structure that is being constructed. unshare system call on
206the other hand performs the following:
207  1) Check flags to force missing, but implied, flags
208  2) For each context structure, call the corresponding unshare
209     helper function to allocate and duplicate a new context
210     structure, if the appropriate bit is set in the flags argument.
211  3) If there is no error in allocation and duplication and there
212     are new context structures then lock the current task structure,
213     associate new context structures with the current task structure,
214     and release the lock on the current task structure.
215  4) Appropriately release older, shared, context structures.
2177) Low Level Design
219Implementation of unshare can be grouped in the following 4 different
221  a) Reorganization of existing copy_* functions
222  b) unshare system call service function
223  c) unshare helper functions for each different process context
224  d) Registration of system call number for different architectures
226  7.1) Reorganization of copy_* functions
227       Each copy function such as copy_mm, copy_namespace, copy_files,
228       etc, had roughly two components. The first component allocated
229       and duplicated the appropriate structure and the second component
230       linked it to the task structure passed in as an argument to the copy
231       function. The first component was split into its own function.
232       These dup_* functions allocated and duplicated the appropriate
233       context structure. The reorganized copy_* functions invoked
234       their corresponding dup_* functions and then linked the newly
235       duplicated structures to the task structure with which the
236       copy function was called.
238  7.2) unshare system call service function
239       * Check flags
240     Force implied flags. If CLONE_THREAD is set force CLONE_VM.
241     If CLONE_VM is set, force CLONE_SIGHAND. If CLONE_SIGHAND is
242     set and signals are also being shared, force CLONE_THREAD. If
243     CLONE_NEWNS is set, force CLONE_FS.
244       * For each context flag, invoke the corresponding unshare_*
245     helper routine with flags passed into the system call and a
246     reference to pointer pointing the new unshared structure
247       * If any new structures are created by unshare_* helper
248     functions, take the task_lock() on the current task,
249     modify appropriate context pointers, and release the
250         task lock.
251       * For all newly unshared structures, release the corresponding
252         older, shared, structures.
254  7.3) unshare_* helper functions
255       For unshare_* helpers corresponding to CLONE_SYSVSEM, CLONE_SIGHAND,
256       and CLONE_THREAD, return -EINVAL since they are not implemented yet.
257       For others, check the flag value to see if the unsharing is
258       required for that structure. If it is, invoke the corresponding
259       dup_* function to allocate and duplicate the structure and return
260       a pointer to it.
262  7.4) Appropriately modify architecture specific code to register the
263       new system call.
2658) Test Specification
267The test for unshare should test the following:
268  1) Valid flags: Test to check that clone flags for signal and
269    signal handlers, for which unsharing is not implemented
270    yet, return -EINVAL.
271  2) Missing/implied flags: Test to make sure that if unsharing
272    namespace without specifying unsharing of filesystem, correctly
273    unshares both namespace and filesystem information.
274  3) For each of the four (namespace, filesystem, files and vm)
275    supported unsharing, verify that the system call correctly
276    unshares the appropriate structure. Verify that unsharing
277    them individually as well as in combination with each
278    other works as expected.
279  4) Concurrent execution: Use shared memory segments and futex on
280    an address in the shm segment to synchronize execution of
281    about 10 threads. Have a couple of threads execute execve,
282    a couple _exit and the rest unshare with different combination
283    of flags. Verify that unsharing is performed as expected and
284    that there are no oops or hangs.
2869) Future Work
288The current implementation of unshare does not allow unsharing of
289signals and signal handlers. Signals are complex to begin with and
290to unshare signals and/or signal handlers of a currently running
291process is even more complex. If in the future there is a specific
292need to allow unsharing of signals and/or signal handlers, it can
293be incrementally added to unshare without affecting legacy
294applications using unshare.

Archive Download this file