man2/clone.2

   1 .\" Copyright (c) 1992 Drew Eckhardt <drew@cs.colorado.edu>, March 28, 1992
   2 .\" and Copyright (c) Michael Kerrisk, 2001, 2002, 2005, 2013
   3 .\"
   4 .\" %%%LICENSE_START(GPL_NOVERSION_ONELINE)
   5 .\" May be distributed under the GNU General Public License.
   6 .\" %%%LICENSE_END
   7 .\"
   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
  41 .\"
  42 .TH CLONE 2 2017-09-15 "Linux" "Linux Programmer's Manual"
  43 .SH NAME
  44 clone, __clone2 \- create a child process
  45 .SH SYNOPSIS
  46 .nf
  47 /* Prototype for the glibc wrapper function */
  48 .PP
  49 .B #define _GNU_SOURCE
  50 .B #include <sched.h>
  51 .PP
  52 .BI "int clone(int (*" "fn" ")(void *), void *" child_stack ,
  53 .BI "          int " flags ", void *" "arg" ", ... "
  54 .BI "          /* pid_t *" ptid ", void *" newtls \
  55 ", pid_t *" ctid " */ );"
  56 .PP
  57 /* For the prototype of the raw system call, see NOTES */
  58 .fi
  59 .SH DESCRIPTION
  60 .BR clone ()
  61 creates a new process, in a manner similar to
  62 .BR fork (2).
  63 .PP
  64 This page describes both the glibc
  65 .BR clone ()
  66 wrapper function and the underlying system call on which it is based.
  67 The main text describes the wrapper function;
  68 the differences for the raw system call
  69 are described toward the end of this page.
  70 .PP
  71 Unlike
  72 .BR fork (2),
  73 .BR clone ()
  74 allows the child process to share parts of its execution context with
  75 the calling process, such as the virtual address space, the table of file
  76 descriptors, and the table of signal handlers.
  77 (Note that on this manual
  78 page, "calling process" normally corresponds to "parent process".
  79 But see the description of
  80 .B CLONE_PARENT
  81 below.)
  82 .PP
  83 One use of
  84 .BR clone ()
  85 is to implement threads: multiple flows of control in a program that
  86 run concurrently in a shared address space.
  87 .PP
  88 When the child process is created with
  89 .BR clone (),
  90 it commences execution by calling the function pointed to by the argument
  91 .IR fn .
  92 (This differs from
  93 .BR fork (2),
  94 where execution continues in the child from the point
  95 of the
  96 .BR fork (2)
  97 call.)
  98 The
  99 .I arg
 100 argument is passed as the argument of the function
 101 .IR fn .
 102 .PP
 103 When the
 104 .IR fn ( arg )
 105 function returns, the child process terminates.
 106 The integer returned by
 107 .I fn
 108 is the exit status for the child process.
 109 The child process may also terminate explicitly by calling
 110 .BR exit (2)
 111 or after receiving a fatal signal.
 112 .PP
 113 The
 114 .I child_stack
 115 argument specifies the location of the stack used by the child process.
 116 Since the child and calling process may share memory,
 117 it is not possible for the child process to execute in the
 118 same stack as the calling process.
 119 The calling process must therefore
 120 set up memory space for the child stack and pass a pointer to this
 121 space to
 122 .BR clone ().
 123 Stacks grow downward on all processors that run Linux
 124 (except the HP PA processors), so
 125 .I child_stack
 126 usually points to the topmost address of the memory space set up for
 127 the child stack.
 128 .PP
 129 The low byte of
 130 .I flags
 131 contains the number of the
 132 .I "termination signal"
 133 sent to the parent when the child dies.
 134 If this signal is specified as anything other than
 135 .BR SIGCHLD ,
 136 then the parent process must specify the
 137 .B __WALL
 138 or
 139 .B __WCLONE
 140 options when waiting for the child with
 141 .BR wait (2).
 142 If no signal is specified, then the parent process is not signaled
 143 when the child terminates.
 144 .PP
 145 .I flags
 146 may also be bitwise-ORed with zero or more of the following constants,
 147 in order to specify what is shared between the calling process
 148 and the child process:
 149 .TP
 150 .BR CLONE_CHILD_CLEARTID " (since Linux 2.5.49)"
 151 Clear (zero) the child thread ID at the location
 152 .I ctid
 153 in child memory when the child exits, and do a wakeup on the futex
 154 at that address.
 155 The address involved may be changed by the
 156 .BR set_tid_address (2)
 157 system call.
 158 This is used by threading libraries.
 159 .TP
 160 .BR CLONE_CHILD_SETTID " (since Linux 2.5.49)"
 161 Store the child thread ID at the location
 162 .I ctid
 163 in the child's memory.
 164 The store operation completes before
 165 .BR clone ()
 166 returns control to user space.
 167 .TP
 168 .BR CLONE_FILES " (since Linux 2.0)"
 169 If
 170 .B CLONE_FILES
 171 is set, the calling process and the child process share the same file
 172 descriptor table.
 173 Any file descriptor created by the calling process or by the child
 174 process is also valid in the other process.
 175 Similarly, if one of the processes closes a file descriptor,
 176 or changes its associated flags (using the
 177 .BR fcntl (2)
 178 .B F_SETFD
 179 operation), the other process is also affected.
 180 If a process sharing a file descriptor table calls
 181 .BR execve (2),
 182 its file descriptor table is duplicated (unshared).
 183 .IP
 184 If
 185 .B CLONE_FILES
 186 is not set, the child process inherits a copy of all file descriptors
 187 opened in the calling process at the time of
 188 .BR clone ().
 189 Subsequent operations that open or close file descriptors,
 190 or change file descriptor flags,
 191 performed by either the calling
 192 process or the child process do not affect the other process.
 193 Note, however,
 194 that the duplicated file descriptors in the child refer to the same open file
 195 descriptions as the corresponding file descriptors in the calling process,
 196 and thus share file offsets and file status flags (see
 197 .BR open (2)).
 198 .TP
 199 .BR CLONE_FS " (since Linux 2.0)"
 200 If
 201 .B CLONE_FS
 202 is set, the caller and the child process share the same filesystem
 203 information.
 204 This includes the root of the filesystem, the current
 205 working directory, and the umask.
 206 Any call to
 207 .BR chroot (2),
 208 .BR chdir (2),
 209 or
 210 .BR umask (2)
 211 performed by the calling process or the child process also affects the
 212 other process.
 213 .IP
 214 If
 215 .B CLONE_FS
 216 is not set, the child process works on a copy of the filesystem
 217 information of the calling process at the time of the
 218 .BR clone ()
 219 call.
 220 Calls to
 221 .BR chroot (2),
 222 .BR chdir (2),
 223 or
 224 .BR umask (2)
 225 performed later by one of the processes do not affect the other process.
 226 .TP
 227 .BR CLONE_IO " (since Linux 2.6.25)"
 228 If
 229 .B CLONE_IO
 230 is set, then the new process shares an I/O context with
 231 the calling process.
 232 If this flag is not set, then (as with
 233 .BR fork (2))
 234 the new process has its own I/O context.
 235 .IP
 236 .\" The following based on text from Jens Axboe
 237 The I/O context is the I/O scope of the disk scheduler (i.e.,
 238 what the I/O scheduler uses to model scheduling of a process's I/O).
 239 If processes share the same I/O context,
 240 they are treated as one by the I/O scheduler.
 241 As a consequence, they get to share disk time.
 242 For some I/O schedulers,
 243 .\" the anticipatory and CFQ scheduler
 244 if two processes share an I/O context,
 245 they will be allowed to interleave their disk access.
 246 If several threads are doing I/O on behalf of the same process
 247 .RB ( aio_read (3),
 248 for instance), they should employ
 249 .BR CLONE_IO
 250 to get better I/O performance.
 251 .\" with CFQ and AS.
 252 .IP
 253 If the kernel is not configured with the
 254 .B CONFIG_BLOCK
 255 option, this flag is a no-op.
 256 .TP
 257 .BR CLONE_NEWCGROUP " (since Linux 4.6)"
 258 Create the process in a new cgroup namespace.
 259 If this flag is not set, then (as with
 260 .BR fork (2))
 261 the process is created in the same cgroup namespaces as the calling process.
 262 This flag is intended for the implementation of containers.
 263 .IP
 264 For further information on cgroup namespaces, see
 265 .BR cgroup_namespaces (7).
 266 .IP
 267 Only a privileged process
 268 .RB ( CAP_SYS_ADMIN )
 269 can employ
 270 .BR CLONE_NEWCGROUP .
 271 .\"
 272 .TP
 273 .BR CLONE_NEWIPC " (since Linux 2.6.19)"
 274 If
 275 .B CLONE_NEWIPC
 276 is set, then create the process in a new IPC namespace.
 277 If this flag is not set, then (as with
 278 .BR fork (2)),
 279 the process is created in the same IPC namespace as
 280 the calling process.
 281 This flag is intended for the implementation of containers.
 282 .IP
 283 An IPC namespace provides an isolated view of System\ V IPC objects (see
 284 .BR svipc (7))
 285 and (since Linux 2.6.30)
 286 .\" commit 7eafd7c74c3f2e67c27621b987b28397110d643f
 287 .\" https://lwn.net/Articles/312232/
 288 POSIX message queues
 289 (see
 290 .BR mq_overview (7)).
 291 The common characteristic of these IPC mechanisms is that IPC
 292 objects are identified by mechanisms other than filesystem
 293 pathnames.
 294 .IP
 295 Objects created in an IPC namespace are visible to all other processes
 296 that are members of that namespace,
 297 but are not visible to processes in other IPC namespaces.
 298 .IP
 299 When an IPC namespace is destroyed
 300 (i.e., when the last process that is a member of the namespace terminates),
 301 all IPC objects in the namespace are automatically destroyed.
 302 .IP
 303 Only a privileged process
 304 .RB ( CAP_SYS_ADMIN )
 305 can employ
 306 .BR CLONE_NEWIPC .
 307 This flag can't be specified in conjunction with
 308 .BR CLONE_SYSVSEM .
 309 .IP
 310 For further information on IPC namespaces, see
 311 .BR namespaces (7).
 312 .TP
 313 .BR CLONE_NEWNET " (since Linux 2.6.24)"
 314 (The implementation of this flag was completed only
 315 by about kernel version 2.6.29.)
 316 .IP
 317 If
 318 .B CLONE_NEWNET
 319 is set, then create the process in a new network namespace.
 320 If this flag is not set, then (as with
 321 .BR fork (2))
 322 the process is created in the same network namespace as
 323 the calling process.
 324 This flag is intended for the implementation of containers.
 325 .IP
 326 A network namespace provides an isolated view of the networking stack
 327 (network device interfaces, IPv4 and IPv6 protocol stacks,
 328 IP routing tables, firewall rules, the
 329 .I /proc/net
 330 and
 331 .I /sys/class/net
 332 directory trees, sockets, etc.).
 333 A physical network device can live in exactly one
 334 network namespace.
 335 A virtual network
 336 .RB ( veth (4))
 337 device pair provides a pipe-like abstraction
 338 that can be used to create tunnels between network namespaces,
 339 and can be used to create a bridge to a physical network device
 340 in another namespace.
 341 .IP
 342 When a network namespace is freed
 343 (i.e., when the last process in the namespace terminates),
 344 its physical network devices are moved back to the
 345 initial network namespace (not to the parent of the process).
 346 For further information on network namespaces, see
 347 .BR namespaces (7).
 348 .IP
 349 Only a privileged process
 350 .RB ( CAP_SYS_ADMIN )
 351 can employ
 352 .BR CLONE_NEWNET .
 353 .TP
 354 .BR CLONE_NEWNS " (since Linux 2.4.19)"
 355 If
 356 .B CLONE_NEWNS
 357 is set, the cloned child is started in a new mount namespace,
 358 initialized with a copy of the namespace of the parent.
 359 If
 360 .B CLONE_NEWNS
 361 is not set, the child lives in the same mount
 362 namespace as the parent.
 363 .IP
 364 Only a privileged process
 365 .RB ( CAP_SYS_ADMIN )
 366 can employ
 367 .BR CLONE_NEWNS .
 368 It is not permitted to specify both
 369 .B CLONE_NEWNS
 370 and
 371 .B CLONE_FS
 372 .\" See https://lwn.net/Articles/543273/
 373 in the same
 374 .BR clone ()
 375 call.
 376 .IP
 377 For further information on mount namespaces, see
 378 .BR namespaces (7)
 379 and
 380 .BR mount_namespaces (7).
 381 .TP
 382 .BR CLONE_NEWPID " (since Linux 2.6.24)"
 383 .\" This explanation draws a lot of details from
 384 .\" http://lwn.net/Articles/259217/
 385 .\" Authors: Pavel Emelyanov <xemul@openvz.org>
 386 .\" and Kir Kolyshkin <kir@openvz.org>
 387 .\"
 388 .\" The primary kernel commit is 30e49c263e36341b60b735cbef5ca37912549264
 389 .\" Author: Pavel Emelyanov <xemul@openvz.org>
 390 If
 391 .B CLONE_NEWPID
 392 is set, then create the process in a new PID namespace.
 393 If this flag is not set, then (as with
 394 .BR fork (2))
 395 the process is created in the same PID namespace as
 396 the calling process.
 397 This flag is intended for the implementation of containers.
 398 .IP
 399 For further information on PID namespaces, see
 400 .BR namespaces (7)
 401 and
 402 .BR pid_namespaces (7).
 403 .IP
 404 Only a privileged process
 405 .RB ( CAP_SYS_ADMIN )
 406 can employ
 407 .BR CLONE_NEWPID .
 408 This flag can't be specified in conjunction with
 409 .BR CLONE_THREAD
 410 or
 411 .BR CLONE_PARENT .
 412 .TP
 413 .BR CLONE_NEWUSER
 414 (This flag first became meaningful for
 415 .BR clone ()
 416 in Linux 2.6.23,
 417 the current
 418 .BR clone ()
 419 semantics were merged in Linux 3.5,
 420 and the final pieces to make the user namespaces completely usable were
 421 merged in Linux 3.8.)
 422 .IP
 423 If
 424 .B CLONE_NEWUSER
 425 is set, then create the process in a new user namespace.
 426 If this flag is not set, then (as with
 427 .BR fork (2))
 428 the process is created in the same user namespace as the calling process.
 429 .IP
 430 Before Linux 3.8, use of
 431 .BR CLONE_NEWUSER
 432 required that the caller have three capabilities:
 433 .BR CAP_SYS_ADMIN ,
 434 .BR CAP_SETUID ,
 435 and
 436 .BR CAP_SETGID .
 437 .\" Before Linux 2.6.29, it appears that only CAP_SYS_ADMIN was needed
 438 Starting with Linux 3.8,
 439 no privileges are needed to create a user namespace.
 440 .IP
 441 This flag can't be specified in conjunction with
 442 .BR CLONE_THREAD
 443 or
 444 .BR CLONE_PARENT .
 445 For security reasons,
 446 .\" commit e66eded8309ebf679d3d3c1f5820d1f2ca332c71
 447 .\" https://lwn.net/Articles/543273/
 448 .\" The fix actually went into 3.9 and into 3.8.3. However, user namespaces
 449 .\" were, for practical purposes, unusable in earlier 3.8.x because of the
 450 .\" various filesystems that didn't support userns.
 451 .BR CLONE_NEWUSER
 452 cannot be specified in conjunction with
 453 .BR CLONE_FS .
 454 .IP
 455 For further information on user namespaces, see
 456 .BR namespaces (7)
 457 and
 458 .BR user_namespaces (7).
 459 .TP
 460 .BR CLONE_NEWUTS " (since Linux 2.6.19)"
 461 If
 462 .B CLONE_NEWUTS
 463 is set, then create the process in a new UTS namespace,
 464 whose identifiers are initialized by duplicating the identifiers
 465 from the UTS namespace of the calling process.
 466 If this flag is not set, then (as with
 467 .BR fork (2))
 468 the process is created in the same UTS namespace as
 469 the calling process.
 470 This flag is intended for the implementation of containers.
 471 .IP
 472 A UTS namespace is the set of identifiers returned by
 473 .BR uname (2);
 474 among these, the domain name and the hostname can be modified by
 475 .BR setdomainname (2)
 476 and
 477 .BR sethostname (2),
 478 respectively.
 479 Changes made to the identifiers in a UTS namespace
 480 are visible to all other processes in the same namespace,
 481 but are not visible to processes in other UTS namespaces.
 482 .IP
 483 Only a privileged process
 484 .RB ( CAP_SYS_ADMIN )
 485 can employ
 486 .BR CLONE_NEWUTS .
 487 .IP
 488 For further information on UTS namespaces, see
 489 .BR namespaces (7).
 490 .TP
 491 .BR CLONE_PARENT " (since Linux 2.3.12)"
 492 If
 493 .B CLONE_PARENT
 494 is set, then the parent of the new child (as returned by
 495 .BR getppid (2))
 496 will be the same as that of the calling process.
 497 .IP
 498 If
 499 .B CLONE_PARENT
 500 is not set, then (as with
 501 .BR fork (2))
 502 the child's parent is the calling process.
 503 .IP
 504 Note that it is the parent process, as returned by
 505 .BR getppid (2),
 506 which is signaled when the child terminates, so that
 507 if
 508 .B CLONE_PARENT
 509 is set, then the parent of the calling process, rather than the
 510 calling process itself, will be signaled.
 511 .TP
 512 .BR CLONE_PARENT_SETTID " (since Linux 2.5.49)"
 513 Store the child thread ID at the location
 514 .I ptid
 515 in the parent's memory.
 516 (In Linux 2.5.32-2.5.48 there was a flag
 517 .B CLONE_SETTID
 518 that did this.)
 519 The store operation completes before
 520 .BR clone ()
 521 returns control to user space.
 522 .TP
 523 .BR CLONE_PID " (Linux 2.0 to 2.5.15)"
 524 If
 525 .B CLONE_PID
 526 is set, the child process is created with the same process ID as
 527 the calling process.
 528 This is good for hacking the system, but otherwise
 529 of not much use.
 530 From Linux 2.3.21 onward, this flag could be
 531 specified only by the system boot process (PID 0).
 532 The flag disappeared completely from the kernel sources in Linux 2.5.16.
 533 Since then, the kernel silently ignores this bit if it is specified in
 534 .IR flags .
 535 .TP
 536 .BR CLONE_PTRACE " (since Linux 2.2)"
 537 If
 538 .B CLONE_PTRACE
 539 is specified, and the calling process is being traced,
 540 then trace the child also (see
 541 .BR ptrace (2)).
 542 .TP
 543 .BR CLONE_SETTLS " (since Linux 2.5.32)"
 544 The TLS (Thread Local Storage) descriptor is set to
 545 .IR newtls .
 546 .IP
 547 The interpretation of
 548 .I newtls
 549 and the resulting effect is architecture dependent.
 550 On x86,
 551 .I newtls
 552 is interpreted as a
 553 .IR "struct user_desc\ *"
 554 (see
 555 .BR set_thread_area (2)).
 556 On x86-64 it is the new value to be set for the %fs base register
 557 (see the
 558 .B ARCH_SET_FS
 559 argument to
 560 .BR arch_prctl (2)).
 561 On architectures with a dedicated TLS register, it is the new value
 562 of that register.
 563 .TP
 564 .BR CLONE_SIGHAND " (since Linux 2.0)"
 565 If
 566 .B CLONE_SIGHAND
 567 is set, the calling process and the child process share the same table of
 568 signal handlers.
 569 If the calling process or child process calls
 570 .BR sigaction (2)
 571 to change the behavior associated with a signal, the behavior is
 572 changed in the other process as well.
 573 However, the calling process and child
 574 processes still have distinct signal masks and sets of pending
 575 signals.
 576 So, one of them may block or unblock signals using
 577 .BR sigprocmask (2)
 578 without affecting the other process.
 579 .IP
 580 If
 581 .B CLONE_SIGHAND
 582 is not set, the child process inherits a copy of the signal handlers
 583 of the calling process at the time
 584 .BR clone ()
 585 is called.
 586 Calls to
 587 .BR sigaction (2)
 588 performed later by one of the processes have no effect on the other
 589 process.
 590 .IP
 591 Since Linux 2.6.0-test6,
 592 .I flags
 593 must also include
 594 .B CLONE_VM
 595 if
 596 .B CLONE_SIGHAND
 597 is specified
 598 .TP
 599 .BR CLONE_STOPPED " (since Linux 2.6.0-test2)"
 600 If
 601 .B CLONE_STOPPED
 602 is set, then the child is initially stopped (as though it was sent a
 603 .B SIGSTOP
 604 signal), and must be resumed by sending it a
 605 .B SIGCONT
 606 signal.
 607 .IP
 608 This flag was
 609 .I deprecated
 610 from Linux 2.6.25 onward,
 611 and was
 612 .I removed
 613 altogether in Linux 2.6.38.
 614 Since then, the kernel silently ignores it without error.
 615 .\" glibc 2.8 removed this defn from bits/sched.h
 616 Starting with Linux 4.6, the same bit was reused for the
 617 .BR CLONE_NEWCGROUP
 618 flag.
 619 .TP
 620 .BR CLONE_SYSVSEM " (since Linux 2.5.10)"
 621 If
 622 .B CLONE_SYSVSEM
 623 is set, then the child and the calling process share
 624 a single list of System V semaphore adjustment
 625 .RI ( semadj )
 626 values (see
 627 .BR semop (2)).
 628 In this case, the shared list accumulates
 629 .I semadj
 630 values across all processes sharing the list,
 631 and semaphore adjustments are performed only when the last process
 632 that is sharing the list terminates (or ceases sharing the list using
 633 .BR unshare (2)).
 634 If this flag is not set, then the child has a separate
 635 .I semadj
 636 list that is initially empty.
 637 .TP
 638 .BR CLONE_THREAD " (since Linux 2.4.0-test8)"
 639 If
 640 .B CLONE_THREAD
 641 is set, the child is placed in the same thread group as the calling process.
 642 To make the remainder of the discussion of
 643 .B CLONE_THREAD
 644 more readable, the term "thread" is used to refer to the
 645 processes within a thread group.
 646 .IP
 647 Thread groups were a feature added in Linux 2.4 to support the
 648 POSIX threads notion of a set of threads that share a single PID.
 649 Internally, this shared PID is the so-called
 650 thread group identifier (TGID) for the thread group.
 651 Since Linux 2.4, calls to
 652 .BR getpid (2)
 653 return the TGID of the caller.
 654 .IP
 655 The threads within a group can be distinguished by their (system-wide)
 656 unique thread IDs (TID).
 657 A new thread's TID is available as the function result
 658 returned to the caller of
 659 .BR clone (),
 660 and a thread can obtain
 661 its own TID using
 662 .BR gettid (2).
 663 .IP
 664 When a call is made to
 665 .BR clone ()
 666 without specifying
 667 .BR CLONE_THREAD ,
 668 then the resulting thread is placed in a new thread group
 669 whose TGID is the same as the thread's TID.
 670 This thread is the
 671 .I leader
 672 of the new thread group.
 673 .IP
 674 A new thread created with
 675 .B CLONE_THREAD
 676 has the same parent process as the caller of
 677 .BR clone ()
 678 (i.e., like
 679 .BR CLONE_PARENT ),
 680 so that calls to
 681 .BR getppid (2)
 682 return the same value for all of the threads in a thread group.
 683 When a
 684 .B CLONE_THREAD
 685 thread terminates, the thread that created it using
 686 .BR clone ()
 687 is not sent a
 688 .B SIGCHLD
 689 (or other termination) signal;
 690 nor can the status of such a thread be obtained
 691 using
 692 .BR wait (2).
 693 (The thread is said to be
 694 .IR detached .)
 695 .IP
 696 After all of the threads in a thread group terminate
 697 the parent process of the thread group is sent a
 698 .B SIGCHLD
 699 (or other termination) signal.
 700 .IP
 701 If any of the threads in a thread group performs an
 702 .BR execve (2),
 703 then all threads other than the thread group leader are terminated,
 704 and the new program is executed in the thread group leader.
 705 .IP
 706 If one of the threads in a thread group creates a child using
 707 .BR fork (2),
 708 then any thread in the group can
 709 .BR wait (2)
 710 for that child.
 711 .IP
 712 Since Linux 2.5.35,
 713 .I flags
 714 must also include
 715 .B CLONE_SIGHAND
 716 if
 717 .B CLONE_THREAD
 718 is specified
 719 (and note that, since Linux 2.6.0-test6,
 720 .BR CLONE_SIGHAND
 721 also requires
 722 .BR CLONE_VM
 723 to be included).
 724 .IP
 725 Signals may be sent to a thread group as a whole (i.e., a TGID) using
 726 .BR kill (2),
 727 or to a specific thread (i.e., TID) using
 728 .BR tgkill (2).
 729 .IP
 730 Signal dispositions and actions are process-wide:
 731 if an unhandled signal is delivered to a thread, then
 732 it will affect (terminate, stop, continue, be ignored in)
 733 all members of the thread group.
 734 .IP
 735 Each thread has its own signal mask, as set by
 736 .BR sigprocmask (2),
 737 but signals can be pending either: for the whole process
 738 (i.e., deliverable to any member of the thread group),
 739 when sent with
 740 .BR kill (2);
 741 or for an individual thread, when sent with
 742 .BR tgkill (2).
 743 A call to
 744 .BR sigpending (2)
 745 returns a signal set that is the union of the signals pending for the
 746 whole process and the signals that are pending for the calling thread.
 747 .IP
 748 If a process-directed signal is delivered to a thread group,
 749 and the thread group has installed a handler for the signal, then
 750 the handler will be invoked in exactly one, arbitrarily selected
 751 member of the thread group that has not blocked the signal.
 752 If multiple threads in a group are waiting to accept the same signal using
 753 .BR sigwaitinfo (2),
 754 the kernel will arbitrarily select one of these threads
 755 to receive the signal.
 756 .TP
 757 .BR CLONE_UNTRACED " (since Linux 2.5.46)"
 758 If
 759 .B CLONE_UNTRACED
 760 is specified, then a tracing process cannot force
 761 .B CLONE_PTRACE
 762 on this child process.
 763 .TP
 764 .BR CLONE_VFORK " (since Linux 2.2)"
 765 If
 766 .B CLONE_VFORK
 767 is set, the execution of the calling process is suspended
 768 until the child releases its virtual memory
 769 resources via a call to
 770 .BR execve (2)
 771 or
 772 .BR _exit (2)
 773 (as with
 774 .BR vfork (2)).
 775 .IP
 776 If
 777 .B CLONE_VFORK
 778 is not set, then both the calling process and the child are schedulable
 779 after the call, and an application should not rely on execution occurring
 780 in any particular order.
 781 .TP
 782 .BR CLONE_VM " (since Linux 2.0)"
 783 If
 784 .B CLONE_VM
 785 is set, the calling process and the child process run in the same memory
 786 space.
 787 In particular, memory writes performed by the calling process
 788 or by the child process are also visible in the other process.
 789 Moreover, any memory mapping or unmapping performed with
 790 .BR mmap (2)
 791 or
 792 .BR munmap (2)
 793 by the child or calling process also affects the other process.
 794 .IP
 795 If
 796 .B CLONE_VM
 797 is not set, the child process runs in a separate copy of the memory
 798 space of the calling process at the time of
 799 .BR clone ().
 800 Memory writes or file mappings/unmappings performed by one of the
 801 processes do not affect the other, as with
 802 .BR fork (2).
 803 .SH NOTES
 804 Note that the glibc
 805 .BR clone ()
 806 wrapper function makes some changes
 807 in the memory pointed to by
 808 .I child_stack
 809 (changes required to set the stack up correctly for the child)
 810 .I before
 811 invoking the
 812 .BR clone ()
 813 system call.
 814 So, in cases where
 815 .BR clone ()
 816 is used to recursively create children,
 817 do not use the buffer employed for the parent's stack
 818 as the stack of the child.
 819 .\"
 820 .SS C library/kernel differences
 821 The raw
 822 .BR clone ()
 823 system call corresponds more closely to
 824 .BR fork (2)
 825 in that execution in the child continues from the point of the
 826 call.
 827 As such, the
 828 .I fn
 829 and
 830 .I arg
 831 arguments of the
 832 .BR clone ()
 833 wrapper function are omitted.
 834 .PP
 835 Another difference for the raw
 836 .BR clone ()
 837 system call is that the
 838 .I child_stack
 839 argument may be NULL,
 840 in which case the child uses a duplicate of the parent's stack.
 841 (Copy-on-write semantics ensure that the child gets separate copies
 842 of stack pages when either process modifies the stack.)
 843 In this case, for correct operation, the
 844 .B CLONE_VM
 845 option should not be specified.
 846 (If the child
 847 .I shares
 848 the parent's memory because of the use of the
 849 .BR CLONE_VM
 850 flag,
 851 then no copy-on-write duplication occurs and chaos is likely to result.)
 852 .PP
 853 The order of the arguments also differs in the raw system call,
 854 and there are variations in the arguments across architectures,
 855 as detailed in the following paragraphs.
 856 .PP
 857 The raw system call interface on x86-64 and some other architectures
 858 (including sh, tile, and alpha) is:
 859 .PP
 860 .in +4
 861 .EX
 862 .BI "long clone(unsigned long " flags ", void *" child_stack ,
 863 .BI "           int *" ptid ", int *" ctid ,
 864 .BI "           unsigned long " newtls );
 865 .EE
 866 .in
 867 .PP
 868 On x86-32, and several other common architectures
 869 (including score, ARM, ARM 64, PA-RISC, arc, Power PC, xtensa,
 870 and MIPS),
 871 .\" CONFIG_CLONE_BACKWARDS
 872 the order of the last two arguments is reversed:
 873 .PP
 874 .in +4
 875 .EX
 876 .BI "long clone(unsigned long " flags ", void *" child_stack ,
 877 .BI "          int *" ptid ", unsigned long " newtls ,
 878 .BI "          int *" ctid );
 879 .EE
 880 .in
 881 .PP
 882 On the cris and s390 architectures,
 883 .\" CONFIG_CLONE_BACKWARDS2
 884 the order of the first two arguments is reversed:
 885 .PP
 886 .in +4
 887 .EX
 888 .BI "long clone(void *" child_stack ", unsigned long " flags ,
 889 .BI "           int *" ptid ", int *" ctid ,
 890 .BI "           unsigned long " newtls );
 891 .EE
 892 .in
 893 .PP
 894 On the microblaze architecture,
 895 .\" CONFIG_CLONE_BACKWARDS3
 896 an additional argument is supplied:
 897 .PP
 898 .in +4
 899 .EX
 900 .BI "long clone(unsigned long " flags ", void *" child_stack ,
 901 .BI "           int " stack_size , "\fR         /* Size of stack */"
 902 .BI "           int *" ptid ", int *" ctid ,
 903 .BI "           unsigned long " newtls );
 904 .EE
 905 .in
 906 .\"
 907 .SS blackfin, m68k, and sparc
 908 .\" Mike Frysinger noted in a 2013 mail:
 909 .\"     these arches don't define __ARCH_WANT_SYS_CLONE:
 910 .\"     blackfin ia64 m68k sparc
 911 The argument-passing conventions on
 912 blackfin, m68k, and sparc are different from the descriptions above.
 913 For details, see the kernel (and glibc) source.
 914 .SS ia64
 915 On ia64, a different interface is used:
 916 .PP
 917 .nf
 918 .BI "int __clone2(int (*" "fn" ")(void *), "
 919 .BI "             void *" child_stack_base ", size_t " stack_size ,
 920 .BI "             int " flags ", void *" "arg" ", ... "
 921 .BI "          /* pid_t *" ptid ", struct user_desc *" tls \
 922 ", pid_t *" ctid " */ );"
 923 .fi
 924 .PP
 925 The prototype shown above is for the glibc wrapper function;
 926 the raw system call interface has no
 927 .I fn
 928 or
 929 .I arg
 930 argument, and changes the order of the arguments so that
 931 .I flags
 932 is the first argument, and
 933 .I tls
 934 is the last argument.
 935 .PP
 936 .BR __clone2 ()
 937 operates in the same way as
 938 .BR clone (),
 939 except that
 940 .I child_stack_base
 941 points to the lowest address of the child's stack area,
 942 and
 943 .I stack_size
 944 specifies the size of the stack pointed to by
 945 .IR child_stack_base .
 946 .SS Linux 2.4 and earlier
 947 In Linux 2.4 and earlier,
 948 .BR clone ()
 949 does not take arguments
 950 .IR ptid ,
 951 .IR tls ,
 952 and
 953 .IR ctid .
 954 .SH RETURN VALUE
 955 .\" gettid(2) returns current->pid;
 956 .\" getpid(2) returns current->tgid;
 957 On success, the thread ID of the child process is returned
 958 in the caller's thread of execution.
 959 On failure, \-1 is returned
 960 in the caller's context, no child process will be created, and
 961 .I errno
 962 will be set appropriately.
 963 .SH ERRORS
 964 .TP
 965 .B EAGAIN
 966 Too many processes are already running; see
 967 .BR fork (2).
 968 .TP
 969 .B EINVAL
 970 .B CLONE_SIGHAND
 971 was specified, but
 972 .B CLONE_VM
 973 was not.
 974 (Since Linux 2.6.0-test6.)
 975 .TP
 976 .B EINVAL
 977 .B CLONE_THREAD
 978 was specified, but
 979 .B CLONE_SIGHAND
 980 was not.
 981 (Since Linux 2.5.35.)
 982 .\" .TP
 983 .\" .B EINVAL
 984 .\" Precisely one of
 985 .\" .B CLONE_DETACHED
 986 .\" and
 987 .\" .B CLONE_THREAD
 988 .\" was specified.
 989 .\" (Since Linux 2.6.0-test6.)
 990 .TP
 991 .B EINVAL
 992 .\" commit e66eded8309ebf679d3d3c1f5820d1f2ca332c71
 993 Both
 994 .B CLONE_FS
 995 and
 996 .B CLONE_NEWNS
 997 were specified in
 998 .IR flags .
 999 .TP
1000 .BR EINVAL " (since Linux 3.9)"
1001 Both
1002 .B CLONE_NEWUSER
1003 and
1004 .B CLONE_FS
1005 were specified in
1006 .IR flags .
1007 .TP
1008 .B EINVAL
1009 Both
1010 .B CLONE_NEWIPC
1011 and
1012 .B CLONE_SYSVSEM
1013 were specified in
1014 .IR flags .
1015 .TP
1016 .B EINVAL
1017 One (or both) of
1018 .BR CLONE_NEWPID
1019 or
1020 .BR CLONE_NEWUSER
1021 and one (or both) of
1022 .BR CLONE_THREAD
1023 or
1024 .BR CLONE_PARENT
1025 were specified in
1026 .IR flags .
1027 .TP
1028 .B EINVAL
1029 Returned by the glibc
1030 .BR clone ()
1031 wrapper function when
1032 .IR fn
1033 or
1034 .IR child_stack
1035 is specified as NULL.
1036 .TP
1037 .B EINVAL
1038 .BR CLONE_NEWIPC
1039 was specified in
1040 .IR flags ,
1041 but the kernel was not configured with the
1042 .B CONFIG_SYSVIPC
1043 and
1044 .BR CONFIG_IPC_NS
1045 options.
1046 .TP
1047 .B EINVAL
1048 .BR CLONE_NEWNET
1049 was specified in
1050 .IR flags ,
1051 but the kernel was not configured with the
1052 .B CONFIG_NET_NS
1053 option.
1054 .TP
1055 .B EINVAL
1056 .BR CLONE_NEWPID
1057 was specified in
1058 .IR flags ,
1059 but the kernel was not configured with the
1060 .B CONFIG_PID_NS
1061 option.
1062 .TP
1063 .B EINVAL
1064 .BR CLONE_NEWUSER
1065 was specified in
1066 .IR flags ,
1067 but the kernel was not configured with the
1068 .B CONFIG_USER_NS
1069 option.
1070 .TP
1071 .B EINVAL
1072 .BR CLONE_NEWUTS
1073 was specified in
1074 .IR flags ,
1075 but the kernel was not configured with the
1076 .B CONFIG_UTS_NS
1077 option.
1078 .TP
1079 .B EINVAL
1080 .I child_stack
1081 is not aligned to a suitable boundary for this architecture.
1082 For example, on aarch64,
1083 .I child_stack
1084 must be a multiple of 16.
1085 .TP
1086 .B ENOMEM
1087 Cannot allocate sufficient memory to allocate a task structure for the
1088 child, or to copy those parts of the caller's context that need to be
1089 copied.
1090 .TP
1091 .BR ENOSPC " (since Linux 3.7)"
1092 .\" commit f2302505775fd13ba93f034206f1e2a587017929
1093 .B CLONE_NEWPID
1094 was specified in flags,
1095 but the limit on the nesting depth of PID namespaces
1096 would have been exceeded; see
1097 .BR pid_namespaces (7).
1098 .TP
1099 .BR ENOSPC " (since Linux 4.9; beforehand " EUSERS )
1100 .B CLONE_NEWUSER
1101 was specified in
1102 .IR flags ,
1103 and the call would cause the limit on the number of
1104 nested user namespaces to be exceeded.
1105 See
1106 .BR user_namespaces (7).
1107 .IP
1108 From Linux 3.11 to Linux 4.8, the error diagnosed in this case was
1109 .BR EUSERS .
1110 .TP
1111 .BR ENOSPC " (since Linux 4.9)"
1112 One of the values in
1113 .I flags
1114 specified the creation of a new user namespace,
1115 but doing so would have caused the limit defined by the corresponding file in
1116 .IR /proc/sys/user
1117 to be exceeded.
1118 For further details, see
1119 .BR namespaces (7).
1120 .TP
1121 .B EPERM
1122 .BR CLONE_NEWCGROUP ,
1123 .BR CLONE_NEWIPC ,
1124 .BR CLONE_NEWNET ,
1125 .BR CLONE_NEWNS ,
1126 .BR CLONE_NEWPID ,
1127 or
1128 .BR CLONE_NEWUTS
1129 was specified by an unprivileged process (process without \fBCAP_SYS_ADMIN\fP).
1130 .TP
1131 .B EPERM
1132 .B CLONE_PID
1133 was specified by a process other than process 0.
1134 (This error occurs only on Linux 2.5.15 and earlier.)
1135 .TP
1136 .B EPERM
1137 .BR CLONE_NEWUSER
1138 was specified in
1139 .IR flags ,
1140 but either the effective user ID or the effective group ID of the caller
1141 does not have a mapping in the parent namespace (see
1142 .BR user_namespaces (7)).
1143 .TP
1144 .BR EPERM " (since Linux 3.9)"
1145 .\" commit 3151527ee007b73a0ebd296010f1c0454a919c7d
1146 .B CLONE_NEWUSER
1147 was specified in
1148 .I flags
1149 and the caller is in a chroot environment
1150 .\" FIXME What is the rationale for this restriction?
1151 (i.e., the caller's root directory does not match the root directory
1152 of the mount namespace in which it resides).
1153 .TP
1154 .BR ERESTARTNOINTR " (since Linux 2.6.17)"
1155 .\" commit 4a2c7a7837da1b91468e50426066d988050e4d56
1156 System call was interrupted by a signal and will be restarted.
1157 (This can be seen only during a trace.)
1158 .TP
1159 .BR EUSERS " (Linux 3.11 to Linux 4.8)"
1160 .B CLONE_NEWUSER
1161 was specified in
1162 .IR flags ,
1163 and the limit on the number of nested user namespaces would be exceeded.
1164 See the discussion of the
1165 .BR ENOSPC
1166 error above.
1167 .\" .SH VERSIONS
1168 .\" There is no entry for
1169 .\" .BR clone ()
1170 .\" in libc5.
1171 .\" glibc2 provides
1172 .\" .BR clone ()
1173 .\" as described in this manual page.
1174 .SH CONFORMING TO
1175 .BR clone ()
1176 is Linux-specific and should not be used in programs
1177 intended to be portable.
1178 .SH NOTES
1179 The
1180 .BR kcmp (2)
1181 system call can be used to test whether two processes share various
1182 resources such as a file descriptor table,
1183 System V semaphore undo operations, or a virtual address space.
1184 .PP
1185 .PP
1186 Handlers registered using
1187 .BR pthread_atfork (3)
1188 are not executed during a call to
1189 .BR clone ().
1190 .PP
1191 In the Linux 2.4.x series,
1192 .B CLONE_THREAD
1193 generally does not make the parent of the new thread the same
1194 as the parent of the calling process.
1195 However, for kernel versions 2.4.7 to 2.4.18 the
1196 .B CLONE_THREAD
1197 flag implied the
1198 .B CLONE_PARENT
1199 flag (as in Linux 2.6.0 and later).
1200 .PP
1201 For a while there was
1202 .B CLONE_DETACHED
1203 (introduced in 2.5.32):
1204 parent wants no child-exit signal.
1205 In Linux 2.6.2, the need to give this flag together with
1206 .B CLONE_THREAD
1207 disappeared.
1208 This flag is still defined, but has no effect.
1209 .PP
1210 On i386,
1211 .BR clone ()
1212 should not be called through vsyscall, but directly through
1213 .IR "int $0x80" .
1214 .SH BUGS
1215 GNU C library versions 2.3.4 up to and including 2.24
1216 contained a wrapper function for
1217 .BR getpid (2)
1218 that performed caching of PIDs.
1219 This caching relied on support in the glibc wrapper for
1220 .BR clone (),
1221 but limitations in the implementation
1222 meant that the cache was not up to date in some circumstances.
1223 In particular,
1224 if a signal was delivered to the child immediately after the
1225 .BR clone ()
1226 call, then a call to
1227 .BR getpid (2)
1228 in a handler for the signal could return the PID
1229 of the calling process ("the parent"),
1230 if the clone wrapper had not yet had a chance to update the PID
1231 cache in the child.
1232 (This discussion ignores the case where the child was created using
1233 .BR CLONE_THREAD ,
1234 when
1235 .BR getpid (2)
1236 .I should
1237 return the same value in the child and in the process that called
1238 .BR clone (),
1239 since the caller and the child are in the same thread group.
1240 The stale-cache problem also does not occur if the
1241 .I flags
1242 argument includes
1243 .BR CLONE_VM .)
1244 To get the truth, it was sometimes necessary to use code such as the following:
1245 .PP
1246 .in +4n
1247 .EX
1248 #include <syscall.h>
1249
1250 pid_t mypid;
1251
1252 mypid = syscall(SYS_getpid);
1253 .EE
1254 .in
1255 .\" See also the following bug reports
1256 .\" https://bugzilla.redhat.com/show_bug.cgi?id=417521
1257 .\" http://sourceware.org/bugzilla/show_bug.cgi?id=6910
1258 .PP
1259 Because of the stale-cache problem, as well as other problems noted in
1260 .BR getpid (2),
1261 the PID caching feature was removed in glibc 2.25.
1262 .SH EXAMPLE
1263 The following program demonstrates the use of
1264 .BR clone ()
1265 to create a child process that executes in a separate UTS namespace.
1266 The child changes the hostname in its UTS namespace.
1267 Both parent and child then display the system hostname,
1268 making it possible to see that the hostname
1269 differs in the UTS namespaces of the parent and child.
1270 For an example of the use of this program, see
1271 .BR setns (2).
1272 .SS Program source
1273 .EX
1274 #define _GNU_SOURCE
1275 #include <sys/wait.h>
1276 #include <sys/utsname.h>
1277 #include <sched.h>
1278 #include <string.h>
1279 #include <stdio.h>
1280 #include <stdlib.h>
1281 #include <unistd.h>
1282
1283 #define errExit(msg)    do { perror(msg); exit(EXIT_FAILURE); \\
1284                         } while (0)
1285
1286 static int              /* Start function for cloned child */
1287 childFunc(void *arg)
1288 {
1289     struct utsname uts;
1290
1291     /* Change hostname in UTS namespace of child */
1292
1293     if (sethostname(arg, strlen(arg)) == \-1)
1294         errExit("sethostname");
1295
1296     /* Retrieve and display hostname */
1297
1298     if (uname(&uts) == \-1)
1299         errExit("uname");
1300     printf("uts.nodename in child:  %s\\n", uts.nodename);
1301
1302     /* Keep the namespace open for a while, by sleeping.
1303        This allows some experimentation\-\-for example, another
1304        process might join the namespace. */
1305
1306     sleep(200);
1307
1308     return 0;           /* Child terminates now */
1309 }
1310
1311 #define STACK_SIZE (1024 * 1024)    /* Stack size for cloned child */
1312
1313 int
1314 main(int argc, char *argv[])
1315 {
1316     char *stack;                    /* Start of stack buffer */
1317     char *stackTop;                 /* End of stack buffer */
1318     pid_t pid;
1319     struct utsname uts;
1320
1321     if (argc < 2) {
1322         fprintf(stderr, "Usage: %s <child\-hostname>\\n", argv[0]);
1323         exit(EXIT_SUCCESS);
1324     }
1325
1326     /* Allocate stack for child */
1327
1328     stack = malloc(STACK_SIZE);
1329     if (stack == NULL)
1330         errExit("malloc");
1331     stackTop = stack + STACK_SIZE;  /* Assume stack grows downward */
1332
1333     /* Create child that has its own UTS namespace;
1334        child commences execution in childFunc() */
1335
1336     pid = clone(childFunc, stackTop, CLONE_NEWUTS | SIGCHLD, argv[1]);
1337     if (pid == \-1)
1338         errExit("clone");
1339     printf("clone() returned %ld\\n", (long) pid);
1340
1341     /* Parent falls through to here */
1342
1343     sleep(1);           /* Give child time to change its hostname */
1344
1345     /* Display hostname in parent\(aqs UTS namespace. This will be
1346        different from hostname in child\(aqs UTS namespace. */
1347
1348     if (uname(&uts) == \-1)
1349         errExit("uname");
1350     printf("uts.nodename in parent: %s\\n", uts.nodename);
1351
1352     if (waitpid(pid, NULL, 0) == \-1)    /* Wait for child */
1353         errExit("waitpid");
1354     printf("child has terminated\\n");
1355
1356     exit(EXIT_SUCCESS);
1357 }
1358 .EE
1359 .SH SEE ALSO
1360 .BR fork (2),
1361 .BR futex (2),
1362 .BR getpid (2),
1363 .BR gettid (2),
1364 .BR kcmp (2),
1365 .BR set_thread_area (2),
1366 .BR set_tid_address (2),
1367 .BR setns (2),
1368 .BR tkill (2),
1369 .BR unshare (2),
1370 .BR wait (2),
1371 .BR capabilities (7),
1372 .BR namespaces (7),
1373 .BR pthreads (7)