1 .\" Copyright (c) 1992 Drew Eckhardt <drew@cs.colorado.edu>, March 28, 1992
2 .\" and Copyright (c) Michael Kerrisk, 2001, 2002, 2005, 2013, 2019
4 .\" %%%LICENSE_START(GPL_NOVERSION_ONELINE)
5 .\" May be distributed under the GNU General Public License.
8 .\" Modified by Michael Haardt <michael@moria.de>
9 .\" Modified 24 Jul 1993 by Rik Faith <faith@cs.unc.edu>
10 .\" Modified 21 Aug 1994 by Michael Chastain <mec@shell.portal.com>:
11 .\" New man page (copied from 'fork.2').
12 .\" Modified 10 June 1995 by Andries Brouwer <aeb@cwi.nl>
13 .\" Modified 25 April 1998 by Xavier Leroy <Xavier.Leroy@inria.fr>
14 .\" Modified 26 Jun 2001 by Michael Kerrisk
15 .\" Mostly upgraded to 2.4.x
16 .\" Added prototype for sys_clone() plus description
17 .\" Added CLONE_THREAD with a brief description of thread groups
18 .\" Added CLONE_PARENT and revised entire page remove ambiguity
19 .\" between "calling process" and "parent process"
20 .\" Added CLONE_PTRACE and CLONE_VFORK
21 .\" Added EPERM and EINVAL error codes
22 .\" Renamed "__clone" to "clone" (which is the prototype in <sched.h>)
23 .\" various other minor tidy ups and clarifications.
24 .\" Modified 26 Jun 2001 by Michael Kerrisk <mtk.manpages@gmail.com>
25 .\" Updated notes for 2.4.7+ behavior of CLONE_THREAD
26 .\" Modified 15 Oct 2002 by Michael Kerrisk <mtk.manpages@gmail.com>
27 .\" Added description for CLONE_NEWNS, which was added in 2.4.19
28 .\" Slightly rephrased, aeb.
29 .\" Modified 1 Feb 2003 - added CLONE_SIGHAND restriction, aeb.
30 .\" Modified 1 Jan 2004 - various updates, aeb
31 .\" Modified 2004-09-10 - added CLONE_PARENT_SETTID etc. - aeb.
32 .\" 2005-04-12, mtk, noted the PID caching behavior of NPTL's getpid()
33 .\" wrapper under BUGS.
34 .\" 2005-05-10, mtk, added CLONE_SYSVSEM, CLONE_UNTRACED, CLONE_STOPPED.
35 .\" 2005-05-17, mtk, Substantially enhanced discussion of CLONE_THREAD.
36 .\" 2008-11-18, mtk, order CLONE_* flags alphabetically
37 .\" 2008-11-18, mtk, document CLONE_NEWPID
38 .\" 2008-11-19, mtk, document CLONE_NEWUTS
39 .\" 2008-11-19, mtk, document CLONE_NEWIPC
40 .\" 2008-11-19, Jens Axboe, mtk, document CLONE_IO
42 .TH CLONE 2 2019-11-19 "Linux" "Linux Programmer's Manual"
44 clone, __clone2, clone3 \- create a child process
47 /* Prototype for the glibc wrapper function */
49 .B #define _GNU_SOURCE
52 .BI "int clone(int (*" "fn" ")(void *), void *" stack \
53 ", int " flags ", void *" "arg" ", ... "
54 .BI " /* pid_t *" parent_tid ", void *" tls \
55 ", pid_t *" child_tid " */ );"
57 /* For the prototype of the raw clone() system call, see NOTES */
59 .BI "long clone3(struct clone_args *" cl_args ", size_t " size );
63 There is not yet a glibc wrapper for
68 create a new ("child") process, in a manner similar to
73 these system calls provide more precise control over what pieces of execution
74 context are shared between the calling process and the child process.
75 For example, using these system calls, the caller can control whether
76 or not the two processes share the virtual address space,
77 the table of file descriptors, and the table of signal handlers.
78 These system calls also allow the new child process to be placed
82 Note that in this manual
83 page, "calling process" normally corresponds to "parent process".
84 But see the descriptions of
90 This page describes the following interfaces:
94 wrapper function and the underlying system call on which it is based.
95 The main text describes the wrapper function;
96 the differences for the raw system call
97 are described toward the end of this page.
103 In the remainder of this page, the terminology "the clone call" is used
104 when noting details that apply to all of these interfaces,
106 .SS The clone() wrapper function
108 When the child process is created with the
111 it commences execution by calling the function pointed to by the argument
115 where execution continues in the child from the point
121 argument is passed as the argument of the function
126 function returns, the child process terminates.
127 The integer returned by
129 is the exit status for the child process.
130 The child process may also terminate explicitly by calling
132 or after receiving a fatal signal.
136 argument specifies the location of the stack used by the child process.
137 Since the child and calling process may share memory,
138 it is not possible for the child process to execute in the
139 same stack as the calling process.
140 The calling process must therefore
141 set up memory space for the child stack and pass a pointer to this
144 Stacks grow downward on all processors that run Linux
145 (except the HP PA processors), so
147 usually points to the topmost address of the memory space set up for
151 does not provide a means whereby the caller can inform the kernel of the
152 size of the stack area.
154 The remaining arguments to
162 system call provides a superset of the functionality of the older
165 It also provides a number of API improvements, including:
166 space for additional flags bits;
167 cleaner separation in the use of various arguments;
168 and the ability to specify the size of the child's stack area.
173 returns in both the parent and the child.
174 It returns 0 in the child process and returns the PID of the child
181 is a structure of the following form:
186 u64 flags; /* Flags bit mask */
187 u64 pidfd; /* Where to store PID file descriptor
189 u64 child_tid; /* Where to store child TID,
190 in child's memory (\fIpid_t *\fP) */
191 u64 parent_tid; /* Where to store child TID,
192 in parent's memory (\fIint *\fP) */
193 u64 exit_signal; /* Signal to deliver to parent on
195 u64 stack; /* Pointer to lowest byte of stack */
196 u64 stack_size; /* Size of stack */
197 u64 tls; /* Location of new TLS */
198 u64 set_tid; /* Pointer to a \fIpid_t\fP array */
199 u64 set_tid_size; /* Number of elements in \fIset_tid\fP */
206 argument that is supplied to
208 should be initialized to the size of this structure.
209 (The existence of the
211 argument permits future extensions to the
215 The stack for the child process is specified via
217 which points to the lowest byte of the stack area,
219 .IR cl_args.stack_size ,
220 which specifies the size of the stack in bytes.
221 In the case where the
223 flag (see below) is specified, a stack must be explicitly allocated
225 Otherwise, these two fields can be specified as NULL and 0,
226 which causes the child to use the same stack area as the parent
227 (in the child's own virtual address space).
229 The remaining fields in the
231 argument are discussed below.
233 .SS Equivalence between clone() and clone3() arguments
237 interface, where arguments are passed individually, in the newer
239 interface the arguments are packaged into the
241 structure shown above.
242 This structure allows for a superset of the information passed via the
246 The following table shows the equivalence between the arguments of
248 and the fields in the
257 clone() clone3() Notes
259 flags & ~0xff flags For most flags; details below
260 parent_tid pidfd See CLONE_PIDFD
261 child_tid child_tid See CLONE_CHILD_SETTID
262 parent_tid parent_tid See CLONE_PARENT_SETTID
263 flags & 0xff exit_signal
266 tls tls See CLONE_SETTLS
267 \fP---\fP set_tid See below for details
268 \fP---\fP set_tid_size
272 .SS The child termination signal
274 When the child process terminates, a signal may be sent to the parent.
275 The termination signal is specified in the low byte of
279 .I cl_args.exit_signal
281 If this signal is specified as anything other than
283 then the parent process must specify the
287 options when waiting for the child with
289 If no signal (i.e., zero) is specified, then the parent process is not signaled
290 when the child terminates.
292 .SS The set_tid array
294 By default, the kernel chooses the next sequential PID for the new
295 process in each of the PID namespaces where it is present.
296 When creating a process with
300 array (available since Linux 5.5)
301 can be used to select specific PIDs for the process in some
302 or all of the PID namespaces where it is present.
303 If the PID of the newly created process should be set only for the current
304 PID namespace or in the newly created PID namespace (if
308 then the first element in the
310 array has to be the desired PID and
314 If the PID of the newly created process should have a certain value in
315 multiple PID namespaces, then the
317 array can have multiple entries.
318 The first entry defines the PID in the most
319 deeply nested PID namespace and each of the following entries contains
321 corresponding ancestor PID namespace.
322 The number of PID namespaces in which a PID
323 should be set is defined by
325 which cannot be larger than the number of currently nested PID namespaces.
327 To create a process with the following PIDs in a PID namespace hierarchy:
332 PID NS level Requested PID Notes
333 0 31496 Outermost PID namespace
335 2 7 Innermost PID namespace
350 If only the PIDs in the two innermost PID namespaces
351 need to be specified, set the array to:
361 The PID in the PID namespaces outside the two innermost PID namespaces
362 will be selected the same way as any other PID is selected.
368 in all owning user namespaces of the target PID namespaces.
370 Callers may only choose a PID greater than 1 in a given PID namespace
373 process (i.e., a process with PID 1) already exists in that namespace.
375 entry for this PID namespace must be 1.
383 allow a flags bit mask that modifies their behavior
384 and allows the caller to specify what is shared between the calling process
385 and the child process.
396 mask in the remainder of this page.
400 mask is specified as a bitwise-OR of zero or more of
401 the constants listed below.
402 Except as noted below, these flags are available
403 (and have the same effect) in both
408 .BR CLONE_CHILD_CLEARTID " (since Linux 2.5.49)"
409 Clear (zero) the child thread ID at the location pointed to by
415 in child memory when the child exits, and do a wakeup on the futex
417 The address involved may be changed by the
418 .BR set_tid_address (2)
420 This is used by threading libraries.
422 .BR CLONE_CHILD_SETTID " (since Linux 2.5.49)"
423 Store the child thread ID at the location pointed to by
429 in the child's memory.
430 The store operation completes before the clone call
431 returns control to user space in the child process.
432 (Note that the store operation may not have completed before the clone call
433 returns in the parent process, which will be relevant if the
435 flag is also employed.)
437 .BR CLONE_CLEAR_SIGHAND " (since Linux 5.5)"
438 .\" commit b612e5df4587c934bd056bf05f4a1deca4de4f75
439 By default, signal dispositions in the child thread are the same as
441 If this flag is specified,
442 then all signals that are handled in the parent
443 are reset to their default dispositions
447 Specifying this flag together with
449 is nonsensical and disallowed.
451 .BR CLONE_DETACHED " (historical)"
452 For a while (during the Linux 2.5 development series)
453 .\" added in 2.5.32; removed in 2.6.0-test4
457 which caused the parent not to receive a signal when the child terminated.
458 Ultimately, the effect of this flag was subsumed under the
460 flag and by the time Linux 2.6.0 was released, this flag had no effect.
461 Starting in Linux 2.6.2, the need to give this flag together with
465 This flag is still defined, but it is usually ignored when calling
467 However, see the description of
471 .BR CLONE_FILES " (since Linux 2.0)"
474 is set, the calling process and the child process share the same file
476 Any file descriptor created by the calling process or by the child
477 process is also valid in the other process.
478 Similarly, if one of the processes closes a file descriptor,
479 or changes its associated flags (using the
482 operation), the other process is also affected.
483 If a process sharing a file descriptor table calls
485 its file descriptor table is duplicated (unshared).
489 is not set, the child process inherits a copy of all file descriptors
490 opened in the calling process at the time of the clone call.
491 Subsequent operations that open or close file descriptors,
492 or change file descriptor flags,
493 performed by either the calling
494 process or the child process do not affect the other process.
496 that the duplicated file descriptors in the child refer to the same
497 open file descriptions as the corresponding file descriptors
498 in the calling process,
499 and thus share file offsets and file status flags (see
502 .BR CLONE_FS " (since Linux 2.0)"
505 is set, the caller and the child process share the same filesystem
507 This includes the root of the filesystem, the current
508 working directory, and the umask.
514 performed by the calling process or the child process also affects the
519 is not set, the child process works on a copy of the filesystem
520 information of the calling process at the time of the clone call.
526 performed later by one of the processes do not affect the other process.
528 .BR CLONE_IO " (since Linux 2.6.25)"
531 is set, then the new process shares an I/O context with
533 If this flag is not set, then (as with
535 the new process has its own I/O context.
537 .\" The following based on text from Jens Axboe
538 The I/O context is the I/O scope of the disk scheduler (i.e.,
539 what the I/O scheduler uses to model scheduling of a process's I/O).
540 If processes share the same I/O context,
541 they are treated as one by the I/O scheduler.
542 As a consequence, they get to share disk time.
543 For some I/O schedulers,
544 .\" the anticipatory and CFQ scheduler
545 if two processes share an I/O context,
546 they will be allowed to interleave their disk access.
547 If several threads are doing I/O on behalf of the same process
549 for instance), they should employ
551 to get better I/O performance.
554 If the kernel is not configured with the
556 option, this flag is a no-op.
558 .BR CLONE_NEWCGROUP " (since Linux 4.6)"
559 Create the process in a new cgroup namespace.
560 If this flag is not set, then (as with
562 the process is created in the same cgroup namespaces as the calling process.
564 For further information on cgroup namespaces, see
565 .BR cgroup_namespaces (7).
567 Only a privileged process
568 .RB ( CAP_SYS_ADMIN )
570 .BR CLONE_NEWCGROUP .
573 .BR CLONE_NEWIPC " (since Linux 2.6.19)"
576 is set, then create the process in a new IPC namespace.
577 If this flag is not set, then (as with
579 the process is created in the same IPC namespace as
582 For further information on IPC namespaces, see
583 .BR ipc_namespaces (7).
585 Only a privileged process
586 .RB ( CAP_SYS_ADMIN )
589 This flag can't be specified in conjunction with
592 .BR CLONE_NEWNET " (since Linux 2.6.24)"
593 (The implementation of this flag was completed only
594 by about kernel version 2.6.29.)
598 is set, then create the process in a new network namespace.
599 If this flag is not set, then (as with
601 the process is created in the same network namespace as
604 For further information on network namespaces, see
605 .BR network_namespaces (7).
607 Only a privileged process
608 .RB ( CAP_SYS_ADMIN )
612 .BR CLONE_NEWNS " (since Linux 2.4.19)"
615 is set, the cloned child is started in a new mount namespace,
616 initialized with a copy of the namespace of the parent.
619 is not set, the child lives in the same mount
620 namespace as the parent.
622 For further information on mount namespaces, see
625 .BR mount_namespaces (7).
627 Only a privileged process
628 .RB ( CAP_SYS_ADMIN )
631 It is not permitted to specify both
635 .\" See https://lwn.net/Articles/543273/
636 in the same clone call.
638 .BR CLONE_NEWPID " (since Linux 2.6.24)"
639 .\" This explanation draws a lot of details from
640 .\" http://lwn.net/Articles/259217/
641 .\" Authors: Pavel Emelyanov <xemul@openvz.org>
642 .\" and Kir Kolyshkin <kir@openvz.org>
644 .\" The primary kernel commit is 30e49c263e36341b60b735cbef5ca37912549264
645 .\" Author: Pavel Emelyanov <xemul@openvz.org>
648 is set, then create the process in a new PID namespace.
649 If this flag is not set, then (as with
651 the process is created in the same PID namespace as
654 For further information on PID namespaces, see
657 .BR pid_namespaces (7).
659 Only a privileged process
660 .RB ( CAP_SYS_ADMIN )
663 This flag can't be specified in conjunction with
669 (This flag first became meaningful for
674 semantics were merged in Linux 3.5,
675 and the final pieces to make the user namespaces completely usable were
676 merged in Linux 3.8.)
680 is set, then create the process in a new user namespace.
681 If this flag is not set, then (as with
683 the process is created in the same user namespace as the calling process.
685 For further information on user namespaces, see
688 .BR user_namespaces (7).
690 Before Linux 3.8, use of
692 required that the caller have three capabilities:
697 .\" Before Linux 2.6.29, it appears that only CAP_SYS_ADMIN was needed
698 Starting with Linux 3.8,
699 no privileges are needed to create a user namespace.
701 This flag can't be specified in conjunction with
705 For security reasons,
706 .\" commit e66eded8309ebf679d3d3c1f5820d1f2ca332c71
707 .\" https://lwn.net/Articles/543273/
708 .\" The fix actually went into 3.9 and into 3.8.3. However, user namespaces
709 .\" were, for practical purposes, unusable in earlier 3.8.x because of the
710 .\" various filesystems that didn't support userns.
712 cannot be specified in conjunction with
715 .BR CLONE_NEWUTS " (since Linux 2.6.19)"
718 is set, then create the process in a new UTS namespace,
719 whose identifiers are initialized by duplicating the identifiers
720 from the UTS namespace of the calling process.
721 If this flag is not set, then (as with
723 the process is created in the same UTS namespace as
726 For further information on UTS namespaces, see
727 .BR uts_namespaces (7).
729 Only a privileged process
730 .RB ( CAP_SYS_ADMIN )
734 .BR CLONE_PARENT " (since Linux 2.3.12)"
737 is set, then the parent of the new child (as returned by
739 will be the same as that of the calling process.
743 is not set, then (as with
745 the child's parent is the calling process.
747 Note that it is the parent process, as returned by
749 which is signaled when the child terminates, so that
752 is set, then the parent of the calling process, rather than the
753 calling process itself, will be signaled.
757 flag can't be used in clone calls by the
758 global init process (PID 1 in the initial PID namespace)
759 and init processes in other PID namespaces.
760 This restriction prevents the creation of multi-rooted process trees
761 as well as the creation of unreapable zombies in the initial PID namespace.
763 .BR CLONE_PARENT_SETTID " (since Linux 2.5.49)"
764 Store the child thread ID at the location pointed to by
768 .I cl_args.parent_tid
770 in the parent's memory.
771 (In Linux 2.5.32-2.5.48 there was a flag
774 The store operation completes before the clone call
775 returns control to user space.
777 .BR CLONE_PID " (Linux 2.0 to 2.5.15)"
780 is set, the child process is created with the same process ID as
782 This is good for hacking the system, but otherwise
784 From Linux 2.3.21 onward, this flag could be
785 specified only by the system boot process (PID 0).
786 The flag disappeared completely from the kernel sources in Linux 2.5.16.
787 Subsequently, the kernel silently ignored this bit if it was specified in the
790 Much later, the same bit was recycled for use as the
794 .BR CLONE_PIDFD " (since Linux 5.2)"
795 .\" commit b3e5838252665ee4cfa76b82bdf1198dca81e5be
796 If this flag is specified,
797 a PID file descriptor referring to the child process is allocated
798 and placed at a specified location in the parent's memory.
799 The close-on-exec flag is set on this new file descriptor.
800 PID file descriptors can be used for the purposes described in
806 the PID file descriptor is placed at the location pointed to by
811 the PID file descriptor is placed at the location pointed to by
815 argument is used to return the PID file descriptor,
818 .B CLONE_PARENT_SETTID
823 It is currently not possible to use this flag together with
825 This means that the process identified by the PID file descriptor
826 will always be a thread group leader.
830 flag is specified alongside
834 an error is returned.
835 An error also results if
837 is specified when calling
839 This error behavior ensures that the bit corresponding to
841 can be reused for further PID file descriptor features in the future.
843 .BR CLONE_PTRACE " (since Linux 2.2)"
846 is specified, and the calling process is being traced,
847 then trace the child also (see
850 .BR CLONE_SETTLS " (since Linux 2.5.32)"
851 The TLS (Thread Local Storage) descriptor is set to
854 The interpretation of
856 and the resulting effect is architecture dependent.
860 .IR "struct user_desc\ *"
862 .BR set_thread_area (2)).
863 On x86-64 it is the new value to be set for the %fs base register
868 On architectures with a dedicated TLS register, it is the new value
871 Use of this flag requires detailed knowledge and generally it
872 should not be used except in libraries implementing threading.
874 .BR CLONE_SIGHAND " (since Linux 2.0)"
877 is set, the calling process and the child process share the same table of
879 If the calling process or child process calls
881 to change the behavior associated with a signal, the behavior is
882 changed in the other process as well.
883 However, the calling process and child
884 processes still have distinct signal masks and sets of pending
886 So, one of them may block or unblock signals using
888 without affecting the other process.
892 is not set, the child process inherits a copy of the signal handlers
893 of the calling process at the time of the clone call.
896 performed later by one of the processes have no effect on the other
900 .\" Precisely: Linux 2.6.0-test6
903 mask must also include
909 .BR CLONE_STOPPED " (since Linux 2.6.0)"
910 .\" Precisely: Linux 2.6.0-test2
913 is set, then the child is initially stopped (as though it was sent a
915 signal), and must be resumed by sending it a
921 from Linux 2.6.25 onward,
924 altogether in Linux 2.6.38.
925 Since then, the kernel silently ignores it without error.
926 .\" glibc 2.8 removed this defn from bits/sched.h
927 Starting with Linux 4.6, the same bit was reused for the
931 .BR CLONE_SYSVSEM " (since Linux 2.5.10)"
934 is set, then the child and the calling process share
935 a single list of System V semaphore adjustment
939 In this case, the shared list accumulates
941 values across all processes sharing the list,
942 and semaphore adjustments are performed only when the last process
943 that is sharing the list terminates (or ceases sharing the list using
945 If this flag is not set, then the child has a separate
947 list that is initially empty.
949 .BR CLONE_THREAD " (since Linux 2.4.0)"
950 .\" Precisely: Linux 2.6.0-test8
953 is set, the child is placed in the same thread group as the calling process.
954 To make the remainder of the discussion of
956 more readable, the term "thread" is used to refer to the
957 processes within a thread group.
959 Thread groups were a feature added in Linux 2.4 to support the
960 POSIX threads notion of a set of threads that share a single PID.
961 Internally, this shared PID is the so-called
962 thread group identifier (TGID) for the thread group.
963 Since Linux 2.4, calls to
965 return the TGID of the caller.
967 The threads within a group can be distinguished by their (system-wide)
968 unique thread IDs (TID).
969 A new thread's TID is available as the function result
970 returned to the caller,
971 and a thread can obtain
975 When a clone call is made without specifying
977 then the resulting thread is placed in a new thread group
978 whose TGID is the same as the thread's TID.
981 of the new thread group.
983 A new thread created with
985 has the same parent process as the process that made the clone call
990 return the same value for all of the threads in a thread group.
993 thread terminates, the thread that created it is not sent a
995 (or other termination) signal;
996 nor can the status of such a thread be obtained
999 (The thread is said to be
1002 After all of the threads in a thread group terminate
1003 the parent process of the thread group is sent a
1005 (or other termination) signal.
1007 If any of the threads in a thread group performs an
1009 then all threads other than the thread group leader are terminated,
1010 and the new program is executed in the thread group leader.
1012 If one of the threads in a thread group creates a child using
1014 then any thread in the group can
1018 Since Linux 2.5.35, the
1020 mask must also include
1025 (and note that, since Linux 2.6.0,
1026 .\" Precisely: Linux 2.6.0-test6
1032 Signal dispositions and actions are process-wide:
1033 if an unhandled signal is delivered to a thread, then
1034 it will affect (terminate, stop, continue, be ignored in)
1035 all members of the thread group.
1037 Each thread has its own signal mask, as set by
1038 .BR sigprocmask (2).
1040 A signal may be process-directed or thread-directed.
1041 A process-directed signal is targeted at a thread group (i.e., a TGID),
1042 and is delivered to an arbitrarily selected thread from among those
1043 that are not blocking the signal.
1044 A signal may be process-directed because it was generated by the kernel
1045 for reasons other than a hardware exception, or because it was sent using
1049 A thread-directed signal is targeted at (i.e., delivered to)
1051 A signal may be thread directed because it was sent using
1054 .BR pthread_sigqueue (3),
1055 or because the thread executed a machine language instruction that triggered
1056 a hardware exception
1057 (e.g., invalid memory access triggering
1059 or a floating-point exception triggering
1064 returns a signal set that is the union of the pending process-directed
1065 signals and the signals that are pending for the calling thread.
1067 If a process-directed signal is delivered to a thread group,
1068 and the thread group has installed a handler for the signal, then
1069 the handler will be invoked in exactly one, arbitrarily selected
1070 member of the thread group that has not blocked the signal.
1071 If multiple threads in a group are waiting to accept the same signal using
1072 .BR sigwaitinfo (2),
1073 the kernel will arbitrarily select one of these threads
1074 to receive the signal.
1076 .BR CLONE_UNTRACED " (since Linux 2.5.46)"
1079 is specified, then a tracing process cannot force
1081 on this child process.
1083 .BR CLONE_VFORK " (since Linux 2.2)"
1086 is set, the execution of the calling process is suspended
1087 until the child releases its virtual memory
1088 resources via a call to
1097 is not set, then both the calling process and the child are schedulable
1098 after the call, and an application should not rely on execution occurring
1099 in any particular order.
1101 .BR CLONE_VM " (since Linux 2.0)"
1104 is set, the calling process and the child process run in the same memory
1106 In particular, memory writes performed by the calling process
1107 or by the child process are also visible in the other process.
1108 Moreover, any memory mapping or unmapping performed with
1112 by the child or calling process also affects the other process.
1116 is not set, the child process runs in a separate copy of the memory
1117 space of the calling process at the time of the clone call.
1118 Memory writes or file mappings/unmappings performed by one of the
1119 processes do not affect the other, as with
1123 One use of these systems calls
1124 is to implement threads: multiple flows of control in a program that
1125 run concurrently in a shared address space.
1127 Glibc does not provide a wrapper for
1134 wrapper function makes some changes
1135 in the memory pointed to by
1137 (changes required to set the stack up correctly for the child)
1144 is used to recursively create children,
1145 do not use the buffer employed for the parent's stack
1146 as the stack of the child.
1148 .SS C library/kernel differences
1151 system call corresponds more closely to
1153 in that execution in the child continues from the point of the
1161 wrapper function are omitted.
1163 In contrast to the glibc wrapper, the raw
1165 system call accepts NULL as a
1172 In this case, the child uses a duplicate of the parent's stack.
1173 (Copy-on-write semantics ensure that the child gets separate copies
1174 of stack pages when either process modifies the stack.)
1175 In this case, for correct operation, the
1177 option should not be specified.
1180 the parent's memory because of the use of the
1183 then no copy-on-write duplication occurs and chaos is likely to result.)
1185 The order of the arguments also differs in the raw system call,
1186 and there are variations in the arguments across architectures,
1187 as detailed in the following paragraphs.
1189 The raw system call interface on x86-64 and some other architectures
1190 (including sh, tile, and alpha) is:
1194 .BI "long clone(unsigned long " flags ", void *" stack ,
1195 .BI " int *" parent_tid ", int *" child_tid ,
1196 .BI " unsigned long " tls );
1200 On x86-32, and several other common architectures
1201 (including score, ARM, ARM 64, PA-RISC, arc, Power PC, xtensa,
1203 .\" CONFIG_CLONE_BACKWARDS
1204 the order of the last two arguments is reversed:
1208 .BI "long clone(unsigned long " flags ", void *" stack ,
1209 .BI " int *" parent_tid ", unsigned long " tls ,
1210 .BI " int *" child_tid );
1214 On the cris and s390 architectures,
1215 .\" CONFIG_CLONE_BACKWARDS2
1216 the order of the first two arguments is reversed:
1220 .BI "long clone(void *" stack ", unsigned long " flags ,
1221 .BI " int *" parent_tid ", int *" child_tid ,
1222 .BI " unsigned long " tls );
1226 On the microblaze architecture,
1227 .\" CONFIG_CLONE_BACKWARDS3
1228 an additional argument is supplied:
1232 .BI "long clone(unsigned long " flags ", void *" stack ,
1233 .BI " int " stack_size , "\fR /* Size of stack */"
1234 .BI " int *" parent_tid ", int *" child_tid ,
1235 .BI " unsigned long " tls );
1239 .SS blackfin, m68k, and sparc
1240 .\" Mike Frysinger noted in a 2013 mail:
1241 .\" these arches don't define __ARCH_WANT_SYS_CLONE:
1242 .\" blackfin ia64 m68k sparc
1243 The argument-passing conventions on
1244 blackfin, m68k, and sparc are different from the descriptions above.
1245 For details, see the kernel (and glibc) source.
1247 On ia64, a different interface is used:
1251 .BI "int __clone2(int (*" "fn" ")(void *), "
1252 .BI " void *" stack_base ", size_t " stack_size ,
1253 .BI " int " flags ", void *" "arg" ", ... "
1254 .BI " /* pid_t *" parent_tid ", struct user_desc *" tls ,
1255 .BI " pid_t *" child_tid " */ );"
1259 The prototype shown above is for the glibc wrapper function;
1260 for the system call itself,
1261 the prototype can be described as follows (it is identical to the
1263 prototype on microblaze):
1267 .BI "long clone2(unsigned long " flags ", void *" stack_base ,
1268 .BI " int " stack_size , "\fR /* Size of stack */"
1269 .BI " int *" parent_tid ", int *" child_tid ,
1270 .BI " unsigned long " tls );
1275 operates in the same way as
1279 points to the lowest address of the child's stack area,
1282 specifies the size of the stack pointed to by
1284 .SS Linux 2.4 and earlier
1285 In Linux 2.4 and earlier,
1287 does not take arguments
1293 .\" gettid(2) returns current->pid;
1294 .\" getpid(2) returns current->tgid;
1295 On success, the thread ID of the child process is returned
1296 in the caller's thread of execution.
1297 On failure, \-1 is returned
1298 in the caller's context, no child process will be created, and
1300 will be set appropriately.
1304 Too many processes are already running; see
1307 .BR EEXIST " (" clone3 "() only)"
1308 One (or more) of the PIDs specified in
1310 already exists in the corresponding PID namespace.
1316 .B CLONE_CLEAR_SIGHAND
1317 were specified in the
1323 was specified in the
1328 (Since Linux 2.6.0.)
1329 .\" Precisely: Linux 2.6.0-test6
1333 was specified in the
1338 (Since Linux 2.5.35.)
1341 .\" Precisely one of
1342 .\" .B CLONE_DETACHED
1346 .\" (Since Linux 2.6.0-test6.)
1350 was specified in the
1352 mask, but the current process previously called
1358 to reassociate itself with a PID namespace.
1361 .\" commit e66eded8309ebf679d3d3c1f5820d1f2ca332c71
1366 were specified in the
1370 .BR EINVAL " (since Linux 3.9)"
1375 were specified in the
1384 were specified in the
1393 and one (or both) of
1397 were specified in the
1401 .BR EINVAL " (since Linux 2.6.32)"
1402 .\" commit 123be07b0b399670a7cc3d82fef0cb4f93ef885c
1404 was specified, and the caller is an init process.
1407 Returned by the glibc
1409 wrapper function when
1413 is specified as NULL.
1417 was specified in the
1420 but the kernel was not configured with the
1428 was specified in the
1431 but the kernel was not configured with the
1437 was specified in the
1440 but the kernel was not configured with the
1446 was specified in the
1449 but the kernel was not configured with the
1455 was specified in the
1458 but the kernel was not configured with the
1464 is not aligned to a suitable boundary for this architecture.
1465 For example, on aarch64,
1467 must be a multiple of 16.
1469 .BR EINVAL " (" clone3 "() only)"
1471 was specified in the
1475 .BR EINVAL " (" clone "() only)"
1477 was specified together with
1485 was specified together with
1491 .BR "EINVAL " "(" clone "() only)"
1493 was specified together with
1494 .B CLONE_PARENT_SETTID
1499 .BR EINVAL " (" clone3 "() only)"
1501 is greater than the number of nested PID namespaces.
1503 .BR EINVAL " (" clone3 "() only)"
1504 One of the PIDs specified in
1508 .BR EINVAL " (AArch64 only, Linux 4.6 and earlier)"
1510 was not aligned to a 126-bit boundary.
1513 Cannot allocate sufficient memory to allocate a task structure for the
1514 child, or to copy those parts of the caller's context that need to be
1517 .BR ENOSPC " (since Linux 3.7)"
1518 .\" commit f2302505775fd13ba93f034206f1e2a587017929
1520 was specified in the
1523 but the limit on the nesting depth of PID namespaces
1524 would have been exceeded; see
1525 .BR pid_namespaces (7).
1527 .BR ENOSPC " (since Linux 4.9; beforehand " EUSERS )
1529 was specified in the
1531 mask, and the call would cause the limit on the number of
1532 nested user namespaces to be exceeded.
1534 .BR user_namespaces (7).
1536 From Linux 3.11 to Linux 4.8, the error diagnosed in this case was
1539 .BR ENOSPC " (since Linux 4.9)"
1540 One of the values in the
1542 mask specified the creation of a new user namespace,
1543 but doing so would have caused the limit defined by the corresponding file in
1546 For further details, see
1550 .BR CLONE_NEWCGROUP ,
1557 was specified by an unprivileged process (process without \fBCAP_SYS_ADMIN\fP).
1561 was specified by a process other than process 0.
1562 (This error occurs only on Linux 2.5.15 and earlier.)
1566 was specified in the
1569 but either the effective user ID or the effective group ID of the caller
1570 does not have a mapping in the parent namespace (see
1571 .BR user_namespaces (7)).
1573 .BR EPERM " (since Linux 3.9)"
1574 .\" commit 3151527ee007b73a0ebd296010f1c0454a919c7d
1576 was specified in the
1578 mask and the caller is in a chroot environment
1579 .\" FIXME What is the rationale for this restriction?
1580 (i.e., the caller's root directory does not match the root directory
1581 of the mount namespace in which it resides).
1583 .BR EPERM " (" clone3 "() only)"
1585 was greater than zero, and the caller lacks the
1587 capability in one or more of the user namespaces that own the
1588 corresponding PID namespaces.
1590 .BR ERESTARTNOINTR " (since Linux 2.6.17)"
1591 .\" commit 4a2c7a7837da1b91468e50426066d988050e4d56
1592 System call was interrupted by a signal and will be restarted.
1593 (This can be seen only during a trace.)
1595 .BR EUSERS " (Linux 3.11 to Linux 4.8)"
1597 was specified in the
1600 and the limit on the number of nested user namespaces would be exceeded.
1601 See the discussion of the
1607 system call first appeared in Linux 5.3.
1608 .\" There is no entry for
1613 .\" as described in this manual page.
1616 are Linux-specific and should not be used in programs
1617 intended to be portable.
1621 system call can be used to test whether two processes share various
1622 resources such as a file descriptor table,
1623 System V semaphore undo operations, or a virtual address space.
1626 Handlers registered using
1627 .BR pthread_atfork (3)
1628 are not executed during a clone call.
1630 In the Linux 2.4.x series,
1632 generally does not make the parent of the new thread the same
1633 as the parent of the calling process.
1634 However, for kernel versions 2.4.7 to 2.4.18 the
1638 flag (as in Linux 2.6.0 and later).
1642 should not be called through vsyscall, but directly through
1645 GNU C library versions 2.3.4 up to and including 2.24
1646 contained a wrapper function for
1648 that performed caching of PIDs.
1649 This caching relied on support in the glibc wrapper for
1651 but limitations in the implementation
1652 meant that the cache was not up to date in some circumstances.
1654 if a signal was delivered to the child immediately after the
1656 call, then a call to
1658 in a handler for the signal could return the PID
1659 of the calling process ("the parent"),
1660 if the clone wrapper had not yet had a chance to update the PID
1662 (This discussion ignores the case where the child was created using
1667 return the same value in the child and in the process that called
1669 since the caller and the child are in the same thread group.
1670 The stale-cache problem also does not occur if the
1674 To get the truth, it was sometimes necessary to use code such as the following:
1678 #include <syscall.h>
1682 mypid = syscall(SYS_getpid);
1685 .\" See also the following bug reports
1686 .\" https://bugzilla.redhat.com/show_bug.cgi?id=417521
1687 .\" http://sourceware.org/bugzilla/show_bug.cgi?id=6910
1689 Because of the stale-cache problem, as well as other problems noted in
1691 the PID caching feature was removed in glibc 2.25.
1693 The following program demonstrates the use of
1695 to create a child process that executes in a separate UTS namespace.
1696 The child changes the hostname in its UTS namespace.
1697 Both parent and child then display the system hostname,
1698 making it possible to see that the hostname
1699 differs in the UTS namespaces of the parent and child.
1700 For an example of the use of this program, see
1703 Within the sample program, we allocate the memory that is to
1704 be used for the child's stack using
1708 for the following reasons:
1711 allocates a block of memory that starts on a page
1712 boundary and is a multiple of the page size.
1713 This is useful if we want to establish a guard page (a page with protection
1715 at the end of the stack using
1720 flag to request a mapping that is suitable for a stack.
1721 For the moment, this flag is a no-op on Linux,
1722 but it exists and has effect on some other systems,
1723 so we should include it for portability.
1727 #include <sys/wait.h>
1728 #include <sys/utsname.h>
1734 #include <sys/mman.h>
1736 #define errExit(msg) do { perror(msg); exit(EXIT_FAILURE); \e
1739 static int /* Start function for cloned child */
1740 childFunc(void *arg)
1744 /* Change hostname in UTS namespace of child */
1746 if (sethostname(arg, strlen(arg)) == \-1)
1747 errExit("sethostname");
1749 /* Retrieve and display hostname */
1751 if (uname(&uts) == \-1)
1753 printf("uts.nodename in child: %s\en", uts.nodename);
1755 /* Keep the namespace open for a while, by sleeping.
1756 This allows some experimentation\-\-for example, another
1757 process might join the namespace. */
1761 return 0; /* Child terminates now */
1764 #define STACK_SIZE (1024 * 1024) /* Stack size for cloned child */
1767 main(int argc, char *argv[])
1769 char *stack; /* Start of stack buffer */
1770 char *stackTop; /* End of stack buffer */
1775 fprintf(stderr, "Usage: %s <child\-hostname>\en", argv[0]);
1779 /* Allocate memory to be used for the stack of the child */
1781 stack = mmap(NULL, STACK_SIZE, PROT_READ | PROT_WRITE,
1782 MAP_PRIVATE | MAP_ANONYMOUS | MAP_STACK, \-1, 0);
1783 if (stack == MAP_FAILED)
1786 stackTop = stack + STACK_SIZE; /* Assume stack grows downward */
1788 /* Create child that has its own UTS namespace;
1789 child commences execution in childFunc() */
1791 pid = clone(childFunc, stackTop, CLONE_NEWUTS | SIGCHLD, argv[1]);
1794 printf("clone() returned %ld\en", (long) pid);
1796 /* Parent falls through to here */
1798 sleep(1); /* Give child time to change its hostname */
1800 /* Display hostname in parent\(aqs UTS namespace. This will be
1801 different from hostname in child\(aqs UTS namespace. */
1803 if (uname(&uts) == \-1)
1805 printf("uts.nodename in parent: %s\en", uts.nodename);
1807 if (waitpid(pid, NULL, 0) == \-1) /* Wait for child */
1809 printf("child has terminated\en");
1822 .BR set_thread_area (2),
1823 .BR set_tid_address (2),
1828 .BR capabilities (7),