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
200 u64 set_tid_size; /* Number of elements in \fIset_tid\fP
202 u64 cgroup; /* Target cgroup file descriptor for the
203 child process (since Linux 5.7) */
210 argument that is supplied to
212 should be initialized to the size of this structure.
213 (The existence of the
215 argument permits future extensions to the
219 The stack for the child process is specified via
221 which points to the lowest byte of the stack area,
223 .IR cl_args.stack_size ,
224 which specifies the size of the stack in bytes.
225 In the case where the
227 flag (see below) is specified, a stack must be explicitly allocated
229 Otherwise, these two fields can be specified as NULL and 0,
230 which causes the child to use the same stack area as the parent
231 (in the child's own virtual address space).
233 The remaining fields in the
235 argument are discussed below.
237 .SS Equivalence between clone() and clone3() arguments
241 interface, where arguments are passed individually, in the newer
243 interface the arguments are packaged into the
245 structure shown above.
246 This structure allows for a superset of the information passed via the
250 The following table shows the equivalence between the arguments of
252 and the fields in the
261 clone() clone3() Notes
263 flags & ~0xff flags For most flags; details below
264 parent_tid pidfd See CLONE_PIDFD
265 child_tid child_tid See CLONE_CHILD_SETTID
266 parent_tid parent_tid See CLONE_PARENT_SETTID
267 flags & 0xff exit_signal
270 tls tls See CLONE_SETTLS
271 \fP---\fP set_tid See below for details
272 \fP---\fP set_tid_size
276 .SS The child termination signal
278 When the child process terminates, a signal may be sent to the parent.
279 The termination signal is specified in the low byte of
283 .I cl_args.exit_signal
285 If this signal is specified as anything other than
287 then the parent process must specify the
291 options when waiting for the child with
293 If no signal (i.e., zero) is specified, then the parent process is not signaled
294 when the child terminates.
296 .SS The set_tid array
298 By default, the kernel chooses the next sequential PID for the new
299 process in each of the PID namespaces where it is present.
300 When creating a process with
304 array (available since Linux 5.5)
305 can be used to select specific PIDs for the process in some
306 or all of the PID namespaces where it is present.
307 If the PID of the newly created process should be set only for the current
308 PID namespace or in the newly created PID namespace (if
312 then the first element in the
314 array has to be the desired PID and
318 If the PID of the newly created process should have a certain value in
319 multiple PID namespaces, then the
321 array can have multiple entries.
322 The first entry defines the PID in the most
323 deeply nested PID namespace and each of the following entries contains
325 corresponding ancestor PID namespace.
326 The number of PID namespaces in which a PID
327 should be set is defined by
329 which cannot be larger than the number of currently nested PID namespaces.
331 To create a process with the following PIDs in a PID namespace hierarchy:
336 PID NS level Requested PID Notes
337 0 31496 Outermost PID namespace
339 2 7 Innermost PID namespace
354 If only the PIDs in the two innermost PID namespaces
355 need to be specified, set the array to:
365 The PID in the PID namespaces outside the two innermost PID namespaces
366 will be selected the same way as any other PID is selected.
372 in all owning user namespaces of the target PID namespaces.
374 Callers may only choose a PID greater than 1 in a given PID namespace
377 process (i.e., a process with PID 1) already exists in that namespace.
379 entry for this PID namespace must be 1.
387 allow a flags bit mask that modifies their behavior
388 and allows the caller to specify what is shared between the calling process
389 and the child process.
400 mask in the remainder of this page.
404 mask is specified as a bitwise-OR of zero or more of
405 the constants listed below.
406 Except as noted below, these flags are available
407 (and have the same effect) in both
412 .BR CLONE_CHILD_CLEARTID " (since Linux 2.5.49)"
413 Clear (zero) the child thread ID at the location pointed to by
419 in child memory when the child exits, and do a wakeup on the futex
421 The address involved may be changed by the
422 .BR set_tid_address (2)
424 This is used by threading libraries.
426 .BR CLONE_CHILD_SETTID " (since Linux 2.5.49)"
427 Store the child thread ID at the location pointed to by
433 in the child's memory.
434 The store operation completes before the clone call
435 returns control to user space in the child process.
436 (Note that the store operation may not have completed before the clone call
437 returns in the parent process, which will be relevant if the
439 flag is also employed.)
441 .BR CLONE_CLEAR_SIGHAND " (since Linux 5.5)"
442 .\" commit b612e5df4587c934bd056bf05f4a1deca4de4f75
443 By default, signal dispositions in the child thread are the same as
445 If this flag is specified,
446 then all signals that are handled in the parent
447 are reset to their default dispositions
451 Specifying this flag together with
453 is nonsensical and disallowed.
455 .BR CLONE_INTO_CGROUP " (since Linux 5.7)"
456 .\" commit ef2c41cf38a7559bbf91af42d5b6a4429db8fc68
457 By default, a child process is placed in the same version 2
458 cgroup as its parent.
459 If this flag is specified, the child process will be created in a
460 different version 2 cgroup than its parent.
462 .BR CLONE_INTO_CGROUP
463 has effect only for version 2 cgroups.)
465 In order to place the child process in a different version 2 cgroup,
467 .BR CLONE_INTO_CGROUP
470 and passes a directory file descriptor (see the
477 The caller may also pass an
481 file descriptor for the target cgroup.
482 Note, that all usual version 2 cgroup migration restrictions (see
486 Spawning a process into a cgroup different from the parent's cgroup
487 makes it possible for a service manager to directly spawn new
488 services into dedicated cgroups.
489 This eliminates the accounting
490 jitter that would be caused by the new process living in the
491 parent's cgroup for a short time before being
492 moved into the target cgroup.
494 .BR CLONE_INTO_CGROUP
495 flag also allows the creation of
496 frozen child processes by spawning them into a frozen cgroup.
499 for a description of the freezer controller.)
500 For threaded applications (or even thread implementations which
501 make use of cgroups to limit individual threads), it is possible to
502 establish a fixed cgroup layout before spawning each thread
503 directly into its target cgroup.
505 .BR CLONE_DETACHED " (historical)"
506 For a while (during the Linux 2.5 development series)
507 .\" added in 2.5.32; removed in 2.6.0-test4
511 which caused the parent not to receive a signal when the child terminated.
512 Ultimately, the effect of this flag was subsumed under the
514 flag and by the time Linux 2.6.0 was released, this flag had no effect.
515 Starting in Linux 2.6.2, the need to give this flag together with
519 This flag is still defined, but it is usually ignored when calling
521 However, see the description of
525 .BR CLONE_FILES " (since Linux 2.0)"
528 is set, the calling process and the child process share the same file
530 Any file descriptor created by the calling process or by the child
531 process is also valid in the other process.
532 Similarly, if one of the processes closes a file descriptor,
533 or changes its associated flags (using the
536 operation), the other process is also affected.
537 If a process sharing a file descriptor table calls
539 its file descriptor table is duplicated (unshared).
543 is not set, the child process inherits a copy of all file descriptors
544 opened in the calling process at the time of the clone call.
545 Subsequent operations that open or close file descriptors,
546 or change file descriptor flags,
547 performed by either the calling
548 process or the child process do not affect the other process.
550 that the duplicated file descriptors in the child refer to the same
551 open file descriptions as the corresponding file descriptors
552 in the calling process,
553 and thus share file offsets and file status flags (see
556 .BR CLONE_FS " (since Linux 2.0)"
559 is set, the caller and the child process share the same filesystem
561 This includes the root of the filesystem, the current
562 working directory, and the umask.
568 performed by the calling process or the child process also affects the
573 is not set, the child process works on a copy of the filesystem
574 information of the calling process at the time of the clone call.
580 performed later by one of the processes do not affect the other process.
582 .BR CLONE_IO " (since Linux 2.6.25)"
585 is set, then the new process shares an I/O context with
587 If this flag is not set, then (as with
589 the new process has its own I/O context.
591 .\" The following based on text from Jens Axboe
592 The I/O context is the I/O scope of the disk scheduler (i.e.,
593 what the I/O scheduler uses to model scheduling of a process's I/O).
594 If processes share the same I/O context,
595 they are treated as one by the I/O scheduler.
596 As a consequence, they get to share disk time.
597 For some I/O schedulers,
598 .\" the anticipatory and CFQ scheduler
599 if two processes share an I/O context,
600 they will be allowed to interleave their disk access.
601 If several threads are doing I/O on behalf of the same process
603 for instance), they should employ
605 to get better I/O performance.
608 If the kernel is not configured with the
610 option, this flag is a no-op.
612 .BR CLONE_NEWCGROUP " (since Linux 4.6)"
613 Create the process in a new cgroup namespace.
614 If this flag is not set, then (as with
616 the process is created in the same cgroup namespaces as the calling process.
618 For further information on cgroup namespaces, see
619 .BR cgroup_namespaces (7).
621 Only a privileged process
622 .RB ( CAP_SYS_ADMIN )
624 .BR CLONE_NEWCGROUP .
627 .BR CLONE_NEWIPC " (since Linux 2.6.19)"
630 is set, then create the process in a new IPC namespace.
631 If this flag is not set, then (as with
633 the process is created in the same IPC namespace as
636 For further information on IPC namespaces, see
637 .BR ipc_namespaces (7).
639 Only a privileged process
640 .RB ( CAP_SYS_ADMIN )
643 This flag can't be specified in conjunction with
646 .BR CLONE_NEWNET " (since Linux 2.6.24)"
647 (The implementation of this flag was completed only
648 by about kernel version 2.6.29.)
652 is set, then create the process in a new network namespace.
653 If this flag is not set, then (as with
655 the process is created in the same network namespace as
658 For further information on network namespaces, see
659 .BR network_namespaces (7).
661 Only a privileged process
662 .RB ( CAP_SYS_ADMIN )
666 .BR CLONE_NEWNS " (since Linux 2.4.19)"
669 is set, the cloned child is started in a new mount namespace,
670 initialized with a copy of the namespace of the parent.
673 is not set, the child lives in the same mount
674 namespace as the parent.
676 For further information on mount namespaces, see
679 .BR mount_namespaces (7).
681 Only a privileged process
682 .RB ( CAP_SYS_ADMIN )
685 It is not permitted to specify both
689 .\" See https://lwn.net/Articles/543273/
690 in the same clone call.
692 .BR CLONE_NEWPID " (since Linux 2.6.24)"
693 .\" This explanation draws a lot of details from
694 .\" http://lwn.net/Articles/259217/
695 .\" Authors: Pavel Emelyanov <xemul@openvz.org>
696 .\" and Kir Kolyshkin <kir@openvz.org>
698 .\" The primary kernel commit is 30e49c263e36341b60b735cbef5ca37912549264
699 .\" Author: Pavel Emelyanov <xemul@openvz.org>
702 is set, then create the process in a new PID namespace.
703 If this flag is not set, then (as with
705 the process is created in the same PID namespace as
708 For further information on PID namespaces, see
711 .BR pid_namespaces (7).
713 Only a privileged process
714 .RB ( CAP_SYS_ADMIN )
717 This flag can't be specified in conjunction with
723 (This flag first became meaningful for
728 semantics were merged in Linux 3.5,
729 and the final pieces to make the user namespaces completely usable were
730 merged in Linux 3.8.)
734 is set, then create the process in a new user namespace.
735 If this flag is not set, then (as with
737 the process is created in the same user namespace as the calling process.
739 For further information on user namespaces, see
742 .BR user_namespaces (7).
744 Before Linux 3.8, use of
746 required that the caller have three capabilities:
751 .\" Before Linux 2.6.29, it appears that only CAP_SYS_ADMIN was needed
752 Starting with Linux 3.8,
753 no privileges are needed to create a user namespace.
755 This flag can't be specified in conjunction with
759 For security reasons,
760 .\" commit e66eded8309ebf679d3d3c1f5820d1f2ca332c71
761 .\" https://lwn.net/Articles/543273/
762 .\" The fix actually went into 3.9 and into 3.8.3. However, user namespaces
763 .\" were, for practical purposes, unusable in earlier 3.8.x because of the
764 .\" various filesystems that didn't support userns.
766 cannot be specified in conjunction with
769 .BR CLONE_NEWUTS " (since Linux 2.6.19)"
772 is set, then create the process in a new UTS namespace,
773 whose identifiers are initialized by duplicating the identifiers
774 from the UTS namespace of the calling process.
775 If this flag is not set, then (as with
777 the process is created in the same UTS namespace as
780 For further information on UTS namespaces, see
781 .BR uts_namespaces (7).
783 Only a privileged process
784 .RB ( CAP_SYS_ADMIN )
788 .BR CLONE_PARENT " (since Linux 2.3.12)"
791 is set, then the parent of the new child (as returned by
793 will be the same as that of the calling process.
797 is not set, then (as with
799 the child's parent is the calling process.
801 Note that it is the parent process, as returned by
803 which is signaled when the child terminates, so that
806 is set, then the parent of the calling process, rather than the
807 calling process itself, will be signaled.
811 flag can't be used in clone calls by the
812 global init process (PID 1 in the initial PID namespace)
813 and init processes in other PID namespaces.
814 This restriction prevents the creation of multi-rooted process trees
815 as well as the creation of unreapable zombies in the initial PID namespace.
817 .BR CLONE_PARENT_SETTID " (since Linux 2.5.49)"
818 Store the child thread ID at the location pointed to by
822 .I cl_args.parent_tid
824 in the parent's memory.
825 (In Linux 2.5.32-2.5.48 there was a flag
828 The store operation completes before the clone call
829 returns control to user space.
831 .BR CLONE_PID " (Linux 2.0 to 2.5.15)"
834 is set, the child process is created with the same process ID as
836 This is good for hacking the system, but otherwise
838 From Linux 2.3.21 onward, this flag could be
839 specified only by the system boot process (PID 0).
840 The flag disappeared completely from the kernel sources in Linux 2.5.16.
841 Subsequently, the kernel silently ignored this bit if it was specified in the
844 Much later, the same bit was recycled for use as the
848 .BR CLONE_PIDFD " (since Linux 5.2)"
849 .\" commit b3e5838252665ee4cfa76b82bdf1198dca81e5be
850 If this flag is specified,
851 a PID file descriptor referring to the child process is allocated
852 and placed at a specified location in the parent's memory.
853 The close-on-exec flag is set on this new file descriptor.
854 PID file descriptors can be used for the purposes described in
860 the PID file descriptor is placed at the location pointed to by
865 the PID file descriptor is placed at the location pointed to by
869 argument is used to return the PID file descriptor,
872 .B CLONE_PARENT_SETTID
877 It is currently not possible to use this flag together with
879 This means that the process identified by the PID file descriptor
880 will always be a thread group leader.
884 flag is specified alongside
888 an error is returned.
889 An error also results if
891 is specified when calling
893 This error behavior ensures that the bit corresponding to
895 can be reused for further PID file descriptor features in the future.
897 .BR CLONE_PTRACE " (since Linux 2.2)"
900 is specified, and the calling process is being traced,
901 then trace the child also (see
904 .BR CLONE_SETTLS " (since Linux 2.5.32)"
905 The TLS (Thread Local Storage) descriptor is set to
908 The interpretation of
910 and the resulting effect is architecture dependent.
914 .IR "struct user_desc\ *"
916 .BR set_thread_area (2)).
917 On x86-64 it is the new value to be set for the %fs base register
922 On architectures with a dedicated TLS register, it is the new value
925 Use of this flag requires detailed knowledge and generally it
926 should not be used except in libraries implementing threading.
928 .BR CLONE_SIGHAND " (since Linux 2.0)"
931 is set, the calling process and the child process share the same table of
933 If the calling process or child process calls
935 to change the behavior associated with a signal, the behavior is
936 changed in the other process as well.
937 However, the calling process and child
938 processes still have distinct signal masks and sets of pending
940 So, one of them may block or unblock signals using
942 without affecting the other process.
946 is not set, the child process inherits a copy of the signal handlers
947 of the calling process at the time of the clone call.
950 performed later by one of the processes have no effect on the other
954 .\" Precisely: Linux 2.6.0-test6
957 mask must also include
963 .BR CLONE_STOPPED " (since Linux 2.6.0)"
964 .\" Precisely: Linux 2.6.0-test2
967 is set, then the child is initially stopped (as though it was sent a
969 signal), and must be resumed by sending it a
975 from Linux 2.6.25 onward,
978 altogether in Linux 2.6.38.
979 Since then, the kernel silently ignores it without error.
980 .\" glibc 2.8 removed this defn from bits/sched.h
981 Starting with Linux 4.6, the same bit was reused for the
985 .BR CLONE_SYSVSEM " (since Linux 2.5.10)"
988 is set, then the child and the calling process share
989 a single list of System V semaphore adjustment
993 In this case, the shared list accumulates
995 values across all processes sharing the list,
996 and semaphore adjustments are performed only when the last process
997 that is sharing the list terminates (or ceases sharing the list using
999 If this flag is not set, then the child has a separate
1001 list that is initially empty.
1003 .BR CLONE_THREAD " (since Linux 2.4.0)"
1004 .\" Precisely: Linux 2.6.0-test8
1007 is set, the child is placed in the same thread group as the calling process.
1008 To make the remainder of the discussion of
1010 more readable, the term "thread" is used to refer to the
1011 processes within a thread group.
1013 Thread groups were a feature added in Linux 2.4 to support the
1014 POSIX threads notion of a set of threads that share a single PID.
1015 Internally, this shared PID is the so-called
1016 thread group identifier (TGID) for the thread group.
1017 Since Linux 2.4, calls to
1019 return the TGID of the caller.
1021 The threads within a group can be distinguished by their (system-wide)
1022 unique thread IDs (TID).
1023 A new thread's TID is available as the function result
1024 returned to the caller,
1025 and a thread can obtain
1029 When a clone call is made without specifying
1031 then the resulting thread is placed in a new thread group
1032 whose TGID is the same as the thread's TID.
1035 of the new thread group.
1037 A new thread created with
1039 has the same parent process as the process that made the clone call
1044 return the same value for all of the threads in a thread group.
1047 thread terminates, the thread that created it is not sent a
1049 (or other termination) signal;
1050 nor can the status of such a thread be obtained
1053 (The thread is said to be
1056 After all of the threads in a thread group terminate
1057 the parent process of the thread group is sent a
1059 (or other termination) signal.
1061 If any of the threads in a thread group performs an
1063 then all threads other than the thread group leader are terminated,
1064 and the new program is executed in the thread group leader.
1066 If one of the threads in a thread group creates a child using
1068 then any thread in the group can
1072 Since Linux 2.5.35, the
1074 mask must also include
1079 (and note that, since Linux 2.6.0,
1080 .\" Precisely: Linux 2.6.0-test6
1086 Signal dispositions and actions are process-wide:
1087 if an unhandled signal is delivered to a thread, then
1088 it will affect (terminate, stop, continue, be ignored in)
1089 all members of the thread group.
1091 Each thread has its own signal mask, as set by
1092 .BR sigprocmask (2).
1094 A signal may be process-directed or thread-directed.
1095 A process-directed signal is targeted at a thread group (i.e., a TGID),
1096 and is delivered to an arbitrarily selected thread from among those
1097 that are not blocking the signal.
1098 A signal may be process-directed because it was generated by the kernel
1099 for reasons other than a hardware exception, or because it was sent using
1103 A thread-directed signal is targeted at (i.e., delivered to)
1105 A signal may be thread directed because it was sent using
1108 .BR pthread_sigqueue (3),
1109 or because the thread executed a machine language instruction that triggered
1110 a hardware exception
1111 (e.g., invalid memory access triggering
1113 or a floating-point exception triggering
1118 returns a signal set that is the union of the pending process-directed
1119 signals and the signals that are pending for the calling thread.
1121 If a process-directed signal is delivered to a thread group,
1122 and the thread group has installed a handler for the signal, then
1123 the handler will be invoked in exactly one, arbitrarily selected
1124 member of the thread group that has not blocked the signal.
1125 If multiple threads in a group are waiting to accept the same signal using
1126 .BR sigwaitinfo (2),
1127 the kernel will arbitrarily select one of these threads
1128 to receive the signal.
1130 .BR CLONE_UNTRACED " (since Linux 2.5.46)"
1133 is specified, then a tracing process cannot force
1135 on this child process.
1137 .BR CLONE_VFORK " (since Linux 2.2)"
1140 is set, the execution of the calling process is suspended
1141 until the child releases its virtual memory
1142 resources via a call to
1151 is not set, then both the calling process and the child are schedulable
1152 after the call, and an application should not rely on execution occurring
1153 in any particular order.
1155 .BR CLONE_VM " (since Linux 2.0)"
1158 is set, the calling process and the child process run in the same memory
1160 In particular, memory writes performed by the calling process
1161 or by the child process are also visible in the other process.
1162 Moreover, any memory mapping or unmapping performed with
1166 by the child or calling process also affects the other process.
1170 is not set, the child process runs in a separate copy of the memory
1171 space of the calling process at the time of the clone call.
1172 Memory writes or file mappings/unmappings performed by one of the
1173 processes do not affect the other, as with
1177 One use of these systems calls
1178 is to implement threads: multiple flows of control in a program that
1179 run concurrently in a shared address space.
1181 Glibc does not provide a wrapper for
1188 wrapper function makes some changes
1189 in the memory pointed to by
1191 (changes required to set the stack up correctly for the child)
1198 is used to recursively create children,
1199 do not use the buffer employed for the parent's stack
1200 as the stack of the child.
1202 .SS C library/kernel differences
1205 system call corresponds more closely to
1207 in that execution in the child continues from the point of the
1215 wrapper function are omitted.
1217 In contrast to the glibc wrapper, the raw
1219 system call accepts NULL as a
1226 In this case, the child uses a duplicate of the parent's stack.
1227 (Copy-on-write semantics ensure that the child gets separate copies
1228 of stack pages when either process modifies the stack.)
1229 In this case, for correct operation, the
1231 option should not be specified.
1234 the parent's memory because of the use of the
1237 then no copy-on-write duplication occurs and chaos is likely to result.)
1239 The order of the arguments also differs in the raw system call,
1240 and there are variations in the arguments across architectures,
1241 as detailed in the following paragraphs.
1243 The raw system call interface on x86-64 and some other architectures
1244 (including sh, tile, and alpha) is:
1248 .BI "long clone(unsigned long " flags ", void *" stack ,
1249 .BI " int *" parent_tid ", int *" child_tid ,
1250 .BI " unsigned long " tls );
1254 On x86-32, and several other common architectures
1255 (including score, ARM, ARM 64, PA-RISC, arc, Power PC, xtensa,
1257 .\" CONFIG_CLONE_BACKWARDS
1258 the order of the last two arguments is reversed:
1262 .BI "long clone(unsigned long " flags ", void *" stack ,
1263 .BI " int *" parent_tid ", unsigned long " tls ,
1264 .BI " int *" child_tid );
1268 On the cris and s390 architectures,
1269 .\" CONFIG_CLONE_BACKWARDS2
1270 the order of the first two arguments is reversed:
1274 .BI "long clone(void *" stack ", unsigned long " flags ,
1275 .BI " int *" parent_tid ", int *" child_tid ,
1276 .BI " unsigned long " tls );
1280 On the microblaze architecture,
1281 .\" CONFIG_CLONE_BACKWARDS3
1282 an additional argument is supplied:
1286 .BI "long clone(unsigned long " flags ", void *" stack ,
1287 .BI " int " stack_size , "\fR /* Size of stack */"
1288 .BI " int *" parent_tid ", int *" child_tid ,
1289 .BI " unsigned long " tls );
1293 .SS blackfin, m68k, and sparc
1294 .\" Mike Frysinger noted in a 2013 mail:
1295 .\" these arches don't define __ARCH_WANT_SYS_CLONE:
1296 .\" blackfin ia64 m68k sparc
1297 The argument-passing conventions on
1298 blackfin, m68k, and sparc are different from the descriptions above.
1299 For details, see the kernel (and glibc) source.
1301 On ia64, a different interface is used:
1305 .BI "int __clone2(int (*" "fn" ")(void *), "
1306 .BI " void *" stack_base ", size_t " stack_size ,
1307 .BI " int " flags ", void *" "arg" ", ... "
1308 .BI " /* pid_t *" parent_tid ", struct user_desc *" tls ,
1309 .BI " pid_t *" child_tid " */ );"
1313 The prototype shown above is for the glibc wrapper function;
1314 for the system call itself,
1315 the prototype can be described as follows (it is identical to the
1317 prototype on microblaze):
1321 .BI "long clone2(unsigned long " flags ", void *" stack_base ,
1322 .BI " int " stack_size , "\fR /* Size of stack */"
1323 .BI " int *" parent_tid ", int *" child_tid ,
1324 .BI " unsigned long " tls );
1329 operates in the same way as
1333 points to the lowest address of the child's stack area,
1336 specifies the size of the stack pointed to by
1338 .SS Linux 2.4 and earlier
1339 In Linux 2.4 and earlier,
1341 does not take arguments
1347 .\" gettid(2) returns current->pid;
1348 .\" getpid(2) returns current->tgid;
1349 On success, the thread ID of the child process is returned
1350 in the caller's thread of execution.
1351 On failure, \-1 is returned
1352 in the caller's context, no child process will be created, and
1354 will be set appropriately.
1358 Too many processes are already running; see
1361 .BR EEXIST " (" clone3 "() only)"
1362 One (or more) of the PIDs specified in
1364 already exists in the corresponding PID namespace.
1370 .B CLONE_CLEAR_SIGHAND
1371 were specified in the
1377 was specified in the
1382 (Since Linux 2.6.0.)
1383 .\" Precisely: Linux 2.6.0-test6
1387 was specified in the
1392 (Since Linux 2.5.35.)
1395 .\" Precisely one of
1396 .\" .B CLONE_DETACHED
1400 .\" (Since Linux 2.6.0-test6.)
1404 was specified in the
1406 mask, but the current process previously called
1412 to reassociate itself with a PID namespace.
1415 .\" commit e66eded8309ebf679d3d3c1f5820d1f2ca332c71
1420 were specified in the
1424 .BR EINVAL " (since Linux 3.9)"
1429 were specified in the
1438 were specified in the
1447 and one (or both) of
1451 were specified in the
1455 .BR EINVAL " (since Linux 2.6.32)"
1456 .\" commit 123be07b0b399670a7cc3d82fef0cb4f93ef885c
1458 was specified, and the caller is an init process.
1461 Returned by the glibc
1463 wrapper function when
1467 is specified as NULL.
1471 was specified in the
1474 but the kernel was not configured with the
1482 was specified in the
1485 but the kernel was not configured with the
1491 was specified in the
1494 but the kernel was not configured with the
1500 was specified in the
1503 but the kernel was not configured with the
1509 was specified in the
1512 but the kernel was not configured with the
1518 is not aligned to a suitable boundary for this architecture.
1519 For example, on aarch64,
1521 must be a multiple of 16.
1523 .BR EINVAL " (" clone3 "() only)"
1525 was specified in the
1529 .BR EINVAL " (" clone "() only)"
1531 was specified together with
1539 was specified together with
1545 .BR "EINVAL " "(" clone "() only)"
1547 was specified together with
1548 .B CLONE_PARENT_SETTID
1553 .BR EINVAL " (" clone3 "() only)"
1555 is greater than the number of nested PID namespaces.
1557 .BR EINVAL " (" clone3 "() only)"
1558 One of the PIDs specified in
1562 .BR EINVAL " (AArch64 only, Linux 4.6 and earlier)"
1564 was not aligned to a 126-bit boundary.
1567 Cannot allocate sufficient memory to allocate a task structure for the
1568 child, or to copy those parts of the caller's context that need to be
1571 .BR ENOSPC " (since Linux 3.7)"
1572 .\" commit f2302505775fd13ba93f034206f1e2a587017929
1574 was specified in the
1577 but the limit on the nesting depth of PID namespaces
1578 would have been exceeded; see
1579 .BR pid_namespaces (7).
1581 .BR ENOSPC " (since Linux 4.9; beforehand " EUSERS )
1583 was specified in the
1585 mask, and the call would cause the limit on the number of
1586 nested user namespaces to be exceeded.
1588 .BR user_namespaces (7).
1590 From Linux 3.11 to Linux 4.8, the error diagnosed in this case was
1593 .BR ENOSPC " (since Linux 4.9)"
1594 One of the values in the
1596 mask specified the creation of a new user namespace,
1597 but doing so would have caused the limit defined by the corresponding file in
1600 For further details, see
1604 .BR CLONE_NEWCGROUP ,
1611 was specified by an unprivileged process (process without \fBCAP_SYS_ADMIN\fP).
1615 was specified by a process other than process 0.
1616 (This error occurs only on Linux 2.5.15 and earlier.)
1620 was specified in the
1623 but either the effective user ID or the effective group ID of the caller
1624 does not have a mapping in the parent namespace (see
1625 .BR user_namespaces (7)).
1627 .BR EPERM " (since Linux 3.9)"
1628 .\" commit 3151527ee007b73a0ebd296010f1c0454a919c7d
1630 was specified in the
1632 mask and the caller is in a chroot environment
1633 .\" FIXME What is the rationale for this restriction?
1634 (i.e., the caller's root directory does not match the root directory
1635 of the mount namespace in which it resides).
1637 .BR EPERM " (" clone3 "() only)"
1639 was greater than zero, and the caller lacks the
1641 capability in one or more of the user namespaces that own the
1642 corresponding PID namespaces.
1644 .BR ERESTARTNOINTR " (since Linux 2.6.17)"
1645 .\" commit 4a2c7a7837da1b91468e50426066d988050e4d56
1646 System call was interrupted by a signal and will be restarted.
1647 (This can be seen only during a trace.)
1649 .BR EUSERS " (Linux 3.11 to Linux 4.8)"
1651 was specified in the
1654 and the limit on the number of nested user namespaces would be exceeded.
1655 See the discussion of the
1661 system call first appeared in Linux 5.3.
1662 .\" There is no entry for
1667 .\" as described in this manual page.
1670 are Linux-specific and should not be used in programs
1671 intended to be portable.
1675 system call can be used to test whether two processes share various
1676 resources such as a file descriptor table,
1677 System V semaphore undo operations, or a virtual address space.
1680 Handlers registered using
1681 .BR pthread_atfork (3)
1682 are not executed during a clone call.
1684 In the Linux 2.4.x series,
1686 generally does not make the parent of the new thread the same
1687 as the parent of the calling process.
1688 However, for kernel versions 2.4.7 to 2.4.18 the
1692 flag (as in Linux 2.6.0 and later).
1696 should not be called through vsyscall, but directly through
1699 GNU C library versions 2.3.4 up to and including 2.24
1700 contained a wrapper function for
1702 that performed caching of PIDs.
1703 This caching relied on support in the glibc wrapper for
1705 but limitations in the implementation
1706 meant that the cache was not up to date in some circumstances.
1708 if a signal was delivered to the child immediately after the
1710 call, then a call to
1712 in a handler for the signal could return the PID
1713 of the calling process ("the parent"),
1714 if the clone wrapper had not yet had a chance to update the PID
1716 (This discussion ignores the case where the child was created using
1721 return the same value in the child and in the process that called
1723 since the caller and the child are in the same thread group.
1724 The stale-cache problem also does not occur if the
1728 To get the truth, it was sometimes necessary to use code such as the following:
1732 #include <syscall.h>
1736 mypid = syscall(SYS_getpid);
1739 .\" See also the following bug reports
1740 .\" https://bugzilla.redhat.com/show_bug.cgi?id=417521
1741 .\" http://sourceware.org/bugzilla/show_bug.cgi?id=6910
1743 Because of the stale-cache problem, as well as other problems noted in
1745 the PID caching feature was removed in glibc 2.25.
1747 The following program demonstrates the use of
1749 to create a child process that executes in a separate UTS namespace.
1750 The child changes the hostname in its UTS namespace.
1751 Both parent and child then display the system hostname,
1752 making it possible to see that the hostname
1753 differs in the UTS namespaces of the parent and child.
1754 For an example of the use of this program, see
1757 Within the sample program, we allocate the memory that is to
1758 be used for the child's stack using
1762 for the following reasons:
1765 allocates a block of memory that starts on a page
1766 boundary and is a multiple of the page size.
1767 This is useful if we want to establish a guard page (a page with protection
1769 at the end of the stack using
1774 flag to request a mapping that is suitable for a stack.
1775 For the moment, this flag is a no-op on Linux,
1776 but it exists and has effect on some other systems,
1777 so we should include it for portability.
1781 #include <sys/wait.h>
1782 #include <sys/utsname.h>
1788 #include <sys/mman.h>
1790 #define errExit(msg) do { perror(msg); exit(EXIT_FAILURE); \e
1793 static int /* Start function for cloned child */
1794 childFunc(void *arg)
1798 /* Change hostname in UTS namespace of child */
1800 if (sethostname(arg, strlen(arg)) == \-1)
1801 errExit("sethostname");
1803 /* Retrieve and display hostname */
1805 if (uname(&uts) == \-1)
1807 printf("uts.nodename in child: %s\en", uts.nodename);
1809 /* Keep the namespace open for a while, by sleeping.
1810 This allows some experimentation\-\-for example, another
1811 process might join the namespace. */
1815 return 0; /* Child terminates now */
1818 #define STACK_SIZE (1024 * 1024) /* Stack size for cloned child */
1821 main(int argc, char *argv[])
1823 char *stack; /* Start of stack buffer */
1824 char *stackTop; /* End of stack buffer */
1829 fprintf(stderr, "Usage: %s <child\-hostname>\en", argv[0]);
1833 /* Allocate memory to be used for the stack of the child */
1835 stack = mmap(NULL, STACK_SIZE, PROT_READ | PROT_WRITE,
1836 MAP_PRIVATE | MAP_ANONYMOUS | MAP_STACK, \-1, 0);
1837 if (stack == MAP_FAILED)
1840 stackTop = stack + STACK_SIZE; /* Assume stack grows downward */
1842 /* Create child that has its own UTS namespace;
1843 child commences execution in childFunc() */
1845 pid = clone(childFunc, stackTop, CLONE_NEWUTS | SIGCHLD, argv[1]);
1848 printf("clone() returned %ld\en", (long) pid);
1850 /* Parent falls through to here */
1852 sleep(1); /* Give child time to change its hostname */
1854 /* Display hostname in parent\(aqs UTS namespace. This will be
1855 different from hostname in child\(aqs UTS namespace. */
1857 if (uname(&uts) == \-1)
1859 printf("uts.nodename in parent: %s\en", uts.nodename);
1861 if (waitpid(pid, NULL, 0) == \-1) /* Wait for child */
1863 printf("child has terminated\en");
1876 .BR set_thread_area (2),
1877 .BR set_tid_address (2),
1882 .BR capabilities (7),