1CLONE(2) Linux Programmer's Manual CLONE(2)
2
6 clone - create a child process
7
9 #include <sched.h>
10 int clone(int (*fn)(void *), void *child_stack,
12 int flags, void *arg, ...
13 /* pid_t *pid, struct user_desc *tls, pid_t *ctid */ );
14 _syscall2(int, clone, int, flags, void *, child_stack)
16 _syscall5(int, clone, int, flags, void *, child_stack,
18 int *, parent_tidptr, struct user_desc *, newtls,
19 int *, child_tidptr)
20 /* Using syscall(2) may be preferable; see intro(2) */
21
23 clone() creates a new process, in a manner similar to fork(2). clone()
24 is a library function layered on top of the underlying clone() system
25 call, hereinafter referred to as sys_clone. A description of sys_clone
26 is given towards the end of this page.
27 Unlike fork(2), these calls allow the child process to share parts of
29 its execution context with the calling process, such as the memory
30 space, the table of file descriptors, and the table of signal handlers.
31 (Note that on this manual page, "calling process" normally corresponds
32 to "parent process". But see the description of CLONE_PARENT below.)
33 The main use of clone() is to implement threads: multiple threads of
35 control in a program that run concurrently in a shared memory space.
36 When the child process is created with clone(), it executes the func‐
38 tion application fn(arg). (This differs from fork(2), where execution
39 continues in the child from the point of the fork(2) call.) The fn
40 argument is a pointer to a function that is called by the child process
41 at the beginning of its execution. The arg argument is passed to the
42 fn function.
43 When the fn(arg) function application returns, the child process termi‐
45 nates. The integer returned by fn is the exit code for the child
46 process. The child process may also terminate explicitly by calling
47 exit(2) or after receiving a fatal signal.
48 The child_stack argument specifies the location of the stack used by
50 the child process. Since the child and calling process may share mem‐
51 ory, it is not possible for the child process to execute in the same
52 stack as the calling process. The calling process must therefore set
53 up memory space for the child stack and pass a pointer to this space to
54 clone(). Stacks grow downwards on all processors that run Linux
55 (except the HP PA processors), so child_stack usually points to the
56 topmost address of the memory space set up for the child stack.
57 The low byte of flags contains the number of the termination signal
59 sent to the parent when the child dies. If this signal is specified as
60 anything other than SIGCHLD, then the parent process must specify the
61 __WALL or __WCLONE options when waiting for the child with wait(2). If
62 no signal is specified, then the parent process is not signaled when
63 the child terminates.
64 flags may also be bitwise-or'ed with zero or more of the following con‐
66 stants, in order to specify what is shared between the calling process
67 and the child process:
68 CLONE_PARENT (since Linux 2.3.12)
70 If CLONE_PARENT is set, then the parent of the new child (as
71 returned by getppid(2)) will be the same as that of the calling
72 process.
73 If CLONE_PARENT is not set, then (as with fork(2)) the child's
75 parent is the calling process.
76 Note that it is the parent process, as returned by getppid(2),
78 which is signaled when the child terminates, so that if
79 CLONE_PARENT is set, then the parent of the calling process,
80 rather than the calling process itself, will be signaled.
81 CLONE_FS
83 If CLONE_FS is set, the caller and the child processes share the
84 same file system information. This includes the root of the
85 file system, the current working directory, and the umask. Any
86 call to chroot(2), chdir(2), or umask(2) performed by the call‐
87 ing process or the child process also affects the other process.
88 If CLONE_FS is not set, the child process works on a copy of the
90 file system information of the calling process at the time of
91 the clone() call. Calls to chroot(2), chdir(2), umask(2) per‐
92 formed later by one of the processes do not affect the other
93 process.
94 CLONE_FILES
96 If CLONE_FILES is set, the calling process and the child pro‐
97 cesses share the same file descriptor table. Any file descrip‐
98 tor created by the calling process or by the child process is
99 also valid in the other process. Similarly, if one of the pro‐
100 cesses closes a file descriptor, or changes its associated flags
101 (using the fcntl(2) F_SETFD operation), the other process is
102 also affected.
103 If CLONE_FILES is not set, the child process inherits a copy of
105 all file descriptors opened in the calling process at the time
106 of clone(). (The duplicated file descriptors in the child refer
107 to the same open file descriptions (see open(2)) as the corre‐
108 sponding file descriptors in the calling process.) Subsequent
109 operations that open or close file descriptors, or change file
110 descriptor flags, performed by either the calling process or the
111 child process do not affect the other process.
112 CLONE_NEWNS (since Linux 2.4.19)
114 Start the child in a new namespace.
115 Every process lives in a namespace. The namespace of a process
117 is the data (the set of mounts) describing the file hierarchy as
118 seen by that process. After a fork(2) or clone(2) where the
119 CLONE_NEWNS flag is not set, the child lives in the same names‐
120 pace as the parent. The system calls mount(2) and umount(2)
121 change the namespace of the calling process, and hence affect
122 all processes that live in the same namespace, but do not affect
123 processes in a different namespace.
124 After a clone(2) where the CLONE_NEWNS flag is set, the cloned
126 child is started in a new namespace, initialized with a copy of
127 the namespace of the parent.
128 Only a privileged process (one having the CAP_SYS_ADMIN capabil‐
130 ity) may specify the CLONE_NEWNS flag. It is not permitted to
131 specify both CLONE_NEWNS and CLONE_FS in the same clone() call.
132 CLONE_SIGHAND
134 If CLONE_SIGHAND is set, the calling process and the child pro‐
135 cesses share the same table of signal handlers. If the calling
136 process or child process calls sigaction(2) to change the behav‐
137 ior associated with a signal, the behavior is changed in the
138 other process as well. However, the calling process and child
139 processes still have distinct signal masks and sets of pending
140 signals. So, one of them may block or unblock some signals
141 using sigprocmask(2) without affecting the other process.
142 If CLONE_SIGHAND is not set, the child process inherits a copy
144 of the signal handlers of the calling process at the time
145 clone() is called. Calls to sigaction(2) performed later by one
146 of the processes have no effect on the other process.
147 Since Linux 2.6.0-test6, flags must also include CLONE_VM if
149 CLONE_SIGHAND is specified
150 CLONE_PTRACE
152 If CLONE_PTRACE is specified, and the calling process is being
153 traced, then trace the child also (see ptrace(2)).
154 CLONE_UNTRACED (since Linux 2.5.46)
156 If CLONE_UNTRACED is specified, then a tracing process cannot
157 force CLONE_PTRACE on this child process.
158 CLONE_STOPPED (since Linux 2.6.0-test2)
160 If CLONE_STOPPED is set, then the child is initially stopped (as
161 though it was sent a SIGSTOP signal), and must be resumed by
162 sending it a SIGCONT signal.
163 CLONE_VFORK
165 If CLONE_VFORK is set, the execution of the calling process is
166 suspended until the child releases its virtual memory resources
167 via a call to execve(2) or _exit(2) (as with vfork(2)).
168 If CLONE_VFORK is not set then both the calling process and the
170 child are schedulable after the call, and an application should
171 not rely on execution occurring in any particular order.
172 CLONE_VM
174 If CLONE_VM is set, the calling process and the child processes
175 run in the same memory space. In particular, memory writes per‐
176 formed by the calling process or by the child process are also
177 visible in the other process. Moreover, any memory mapping or
178 unmapping performed with mmap(2) or munmap(2) by the child or
179 calling process also affects the other process.
180 If CLONE_VM is not set, the child process runs in a separate
182 copy of the memory space of the calling process at the time of
183 clone(). Memory writes or file mappings/unmappings performed by
184 one of the processes do not affect the other, as with fork(2).
185 CLONE_PID (obsolete)
187 If CLONE_PID is set, the child process is created with the same
188 process ID as the calling process. This is good for hacking the
189 system, but otherwise of not much use. Since 2.3.21 this flag
190 can be specified only by the system boot process (PID 0). It
191 disappeared in Linux 2.5.16.
192 CLONE_THREAD (since Linux 2.4.0-test8)
194 If CLONE_THREAD is set, the child is placed in the same thread
195 group as the calling process. To make the remainder of the dis‐
196 cussion of CLONE_THREAD more readable, the term "thread" is used
197 to refer to the processes within a thread group.
198 Thread groups were a feature added in Linux 2.4 to support the
200 POSIX threads notion of a set of threads that share a single
201 PID. Internally, this shared PID is the so-called thread group
202 identifier (TGID) for the thread group. Since Linux 2.4, calls
203 to getpid(2) return the TGID of the caller.
204 The threads within a group can be distinguished by their (sys‐
206 tem-wide) unique thread IDs (TID). A new thread's TID is avail‐
207 able as the function result returned to the caller of clone(),
208 and a thread can obtain its own TID using gettid(2).
209 When a call is made to clone() without specifying CLONE_THREAD,
211 then the resulting thread is placed in a new thread group whose
212 TGID is the same as the thread's TID. This thread is the leader
213 of the new thread group.
214 A new thread created with CLONE_THREAD has the same parent
216 process as the caller of clone() (i.e., like CLONE_PARENT), so
217 that calls to getppid(2) return the same value for all of the
218 threads in a thread group. When a CLONE_THREAD thread termi‐
219 nates, the thread that created it using clone() is not sent a
220 SIGCHLD (or other termination) signal; nor can the status of
221 such a thread be obtained using wait(2). (The thread is said to
222 be detached.)
223 After all of the threads in a thread group terminate the parent
225 process of the thread group is sent a SIGCHLD (or other termina‐
226 tion) signal.
227 If any of the threads in a thread group performs an execve(2),
229 then all threads other than the thread group leader are termi‐
230 nated, and the new program is executed in the thread group
231 leader.
232 If one of the threads in a thread group creates a child using
234 fork(2), then any thread in the group can wait(2) for that
235 child.
236 Since Linux 2.5.35, flags must also include CLONE_SIGHAND if
238 CLONE_THREAD is specified.
239 Signals may be sent to a thread group as a whole (i.e., a TGID)
241 using kill(2), or to a specific thread (i.e., TID) using
242 tgkill(2).
243 Signal dispositions and actions are process-wide: if an unhan‐
245 dled signal is delivered to a thread, then it will affect (ter‐
246 minate, stop, continue, be ignored in) all members of the thread
247 group.
248 Each thread has its own signal mask, as set by sigprocmask(2),
250 but signals can be pending either: for the whole process (i.e.,
251 deliverable to any member of the thread group), when sent with
252 kill(2); or for an individual thread, when sent with tgkill(2).
253 A call to sigpending(2) returns a signal set that is the union
254 of the signals pending for the whole process and the signals
255 that are pending for the calling thread.
256 If kill(2) is used to send a signal to a thread group, and the
258 thread group has installed a handler for the signal, then the
259 handler will be invoked in exactly one, arbitrarily selected
260 member of the thread group that has not blocked the signal. If
261 multiple threads in a group are waiting to accept the same sig‐
262 nal using sigwaitinfo(2), the kernel will arbitrarily select one
263 of these threads to receive a signal sent using kill(2).
264 CLONE_SYSVSEM (since Linux 2.5.10)
266 If CLONE_SYSVSEM is set, then the child and the calling process
267 share a single list of System V semaphore undo values (see
268 semop(2)). If this flag is not set, then the child has a sepa‐
269 rate undo list, which is initially empty.
270 CLONE_SETTLS (since Linux 2.5.32)
272 The newtls parameter is the new TLS (Thread Local Storage)
273 descriptor. (See set_thread_area(2).)
274 CLONE_PARENT_SETTID (since Linux 2.5.49)
276 Store child thread ID at location parent_tidptr in parent and
277 child memory. (In Linux 2.5.32-2.5.48 there was a flag
278 CLONE_SETTID that did this.)
279 CLONE_CHILD_SETTID (since Linux 2.5.49)
281 Store child thread ID at location child_tidptr in child memory.
282 CLONE_CHILD_CLEARTID (since Linux 2.5.49)
284 Erase child thread ID at location child_tidptr in child memory
285 when the child exits, and do a wakeup on the futex at that
286 address. The address involved may be changed by the
287 set_tid_address(2) system call. This is used by threading
288 libraries.
289 sys_clone
292 The sys_clone system call corresponds more closely to fork(2) in that
293 execution in the child continues from the point of the call. Thus,
294 sys_clone only requires the flags and child_stack arguments, which have
295 the same meaning as for clone(). (Note that the order of these argu‐
296 ments differs from clone().)
297 Another difference for sys_clone is that the child_stack argument may
299 be zero, in which case copy-on-write semantics ensure that the child
300 gets separate copies of stack pages when either process modifies the
301 stack. In this case, for correct operation, the CLONE_VM option should
302 not be specified.
303 Since Linux 2.5.49 the system call has five parameters. The two new
305 parameters are parent_tidptr which points to the location (in parent
306 and child memory) where the child thread ID will be written in case
307 CLONE_PARENT_SETTID was specified, and child_tidptr which points to the
308 location (in child memory) where the child thread ID will be written in
309 case CLONE_CHILD_SETTID was specified.
310
312 On success, the thread ID of the child process is returned in the call‐
313 er's thread of execution. On failure, a -1 will be returned in the
314 caller's context, no child process will be created, and errno will be
315 set appropriately.
316
318 EAGAIN Too many processes are already running.
319 EINVAL CLONE_SIGHAND was specified, but CLONE_VM was not. (Since Linux
321 2.6.0-test6.)
322 EINVAL CLONE_THREAD was specified, but CLONE_SIGHAND was not. (Since
324 Linux 2.5.35.)
325 EINVAL Both CLONE_FS and CLONE_NEWNS were specified in flags.
327 EINVAL Returned by clone() when a zero value is specified for
329 child_stack.
330 ENOMEM Cannot allocate sufficient memory to allocate a task structure
332 for the child, or to copy those parts of the caller's context
333 that need to be copied.
334 EPERM CLONE_NEWNS was specified by a non-root process (process without
336 CAP_SYS_ADMIN).
337 EPERM CLONE_PID was specified by a process other than process 0.
339
341 There is no entry for clone() in libc5. glibc2 provides clone() as
342 described in this manual page.
343
345 In the kernel 2.4.x series, CLONE_THREAD generally does not make the
346 parent of the new thread the same as the parent of the calling process.
347 However, for kernel versions 2.4.7 to 2.4.18 the CLONE_THREAD flag
348 implied the CLONE_PARENT flag (as in kernel 2.6).
349 For a while there was CLONE_DETACHED (introduced in 2.5.32): parent
351 wants no child-exit signal. In 2.6.2 the need to give this together
352 with CLONE_THREAD disappeared. This flag is still defined, but has no
353 effect.
354 On x86, clone() should not be called through vsyscall, but directly
356 through int $0x80.
357 On IA-64, a different system call is used:
359 int clone2(int (*fn)(void *),
361 void *child_stack_base, size_t stack_size,
362 int flags, void *arg, ...
363 /* pid_t *pid, struct user_desc *tls, pid_t *ctid */ );
364 The clone2() system call operates in the same way as clone(), except
366 that child_stack_base points to the lowest address of the child's stack
367 area, and stack_size specifies the size of the stack pointed to by
368 child_stack_base.
369
371 The clone() and sys_clone calls are Linux specific and should not be
372 used in programs intended to be portable.
373
375 Versions of the GNU C library that include the NPTL threading library
376 contain a wrapper function for getpid() that performs caching of PIDs.
377 In programs linked against such libraries, calls to getpid() may return
378 the same value, even when the threads were not created using
379 CLONE_THREAD (and thus are not in the same thread group). To get the
380 truth, it may be necessary to use code such as the following
381 #include <syscall.h>
383 pid_t mypid;
385 mypid = syscall(SYS_getpid);
387
389 fork(2), futex(2), getpid(2), gettid(2), set_thread_area(2),
390 set_tid_address(2), tkill(2), unshare(2), wait(2), capabilities(7),
391 pthreads(7)
392Linux 2.6 2005-05-17 CLONE(2)