Linux::Clone - an interface to the linux clone, unshare, setns, pivot_root and kcmp syscalls


 use Linux::Clone;


This module exposes the linux clone(2), unshare(2) and some related syscalls to Perl.

$retval = unshare $flags

The following CLONE_ flag values (without CLONE_ prefix) are supported for unshare, if found, in this release. See the documentation for unshare(2) for more info on what they do:

   Linux::Clone::NEWNS   (in unshare, implies FS)
   Linux::Clone::VM      (in unshare, implies SIGHAND)
   Linux::Clone::THREAD  (in unshare, implies VM, SIGHAND)
   Linux::Clone::NEWUSER (in unshare, implies CLONE_THREAD)

Example: unshare the network namespace and prove that by calling ifconfig, showing only the unconfigured lo interface.

   Linux::Clone::unshare Linux::Clone::NEWNET
      and "unshare: $!";
   system "ifconfig";

Example: unshare the network namespace, initialise the loopback interface, create a veth interface pair, put one interface into the parent processes namespace (use ifconfig -a from another shell), configure the other interface with -> and start a shell.

   use Linux::Clone;

   # unshare our network namespace
   Linux::Clone::unshare Linux::Clone::NEWNET
     and "unshare: $!";


   my $ppid = getppid;

   system "
      # create veth pair
      ip link add name veth_master type veth peer name veth_slave

      # move veth_master to our parent process' namespace
      ip link set veth_master netns $ppid

      # configure the local interface
      ip link set veth_slave up
      ip addr add dev veth_slave
      ip route add dev veth_slave

   print <<EOF;
   say hi to your new network namespace, use exit to return.

   try this from another shell to get networking up:

   ip link set veth_master up
   ip addr add dev veth_master
   ip route add dev veth_master

   system "bash";

Example: unshare the filesystem namespace and make a confusing bind mount only visible to the current process.

   use Linux::Clone;

   Linux::Clone::unshare Linux::Clone::NEWNS
      and die "unshare: $!";

   # now bind-mount /lib over /etc and ls -l /etc - looks scary
   system "mount -n --bind /lib /etc";
   system "ls -l /etc";
$retval = Linux::Clone::clone $coderef, $stacksize, $flags[, $ptid, $tls, $ctid]

Clones a new process as specified via $flags and calls $coderef without any arguments (a closure might help you if you need to pass arguments without global variables). The return value from coderef is returned to the system.

The $stacksize specifies how large a stack to allocate for the child. If it is 0, then a default stack size (currently 4MB) will be allocated. There is currently no way to free this area again in the child.

$ptid, if specified, will receive the thread id, $tls, if specified, must contain a struct user_desc and $ctid is currently totally unsupported and must not be specified.

Since this call basically bypasses both perl and your libc (for example, $$ might reflect the parent or child pid in the child), you need to be very careful when using this call, which means you should probably have a very good understanding of perl memory management and how fork and clone work.

The following flags are supported for clone, in addition to all flags supported by unshare, above, and a signal number. When in doubt, refer to the clone(2) manual page.

   Linux::Clone::SETTLS         (not yet implemented)
   Linux::Clone::PARENT_SETTID  (not yet implemented)
   Linux::Clone::CHILD_SETTID   (not yet implemented)
   Linux::Clone::CHILD_CLEARTID (not yet implemented)
   Linux::Clone::PIDFD          (not yet implemented)
   Linux::Clone::CSIGNAL exit signal mask

Note that for practical reasons you basically must not use Linux::Clone::VM or Linux::Clone::VFORK, as perl is unlikely to cope with that.

This is the glibc clone call, it cannot be used to emulate fork.

Example: do a fork-like clone, sharing nothing, slightly confusing perl and your libc, and exit immediately.

   my $pid = Linux::Clone::clone sub { warn "in child"; 77 }, 0, POSIX::SIGCHLD;
Linux::Clone::setns $fh_or_fd[, $nstype]

Calls setns(2) on the file descriptor (or file handle) $fh_or_fd. If $nstype is missing, then 0 is used.

The argument $nstype can be 0, Linux::Clone::NEWIPC, Linux::Clone::NEWNET, Linux::Clone::NEWUTS, Linux::Clone::NEWCGROUP, Linux::Clone::NEWNS, Linux::Clone::NEWPID or Linux::Clone::NEWUSER.

Linux::Clone::pivot_root $new_root, $old_root

Calls pivot_root(2) - refer to its manpage for details.

Linux::Clone::kcmp $pid1, $pid2, $type[, $idx1, $idx2]

Calls kcmp(2) - refer to its manpage for details on operations.

The following $type constants are available if the kcmp syscall number was available during compilation:

Linux::Clone::KCMP_FILE, Linux::Clone::KCMP_VM, Linux::Clone::KCMP_FILES, Linux::Clone::KCMP_FS, Linux::Clone::KCMP_SIGHAND, Linux::Clone::KCMP_IO, Linux::Clone::KCMP_SYSVSEM and Linux::Clone::KCMP_EPOLL_TFD.


Configures a working loopback interface (basically, does the equivalent of "ifconfig lo up" which automatically adds ipv4/ipv6 addresses and routes), which can be useful to get a network namespace going.

Dies on error and returns nothing.

ioctl symbols

The following ioctl symbols are also provided by this module (see ioctl_ns(8)).



IO::AIO has some related functions, such as pidfd_send_signal, and some unrelated functions that might be useful.

namspaces(7), cgroup_namespaces(7), pid_namespaces(7), user_namespaces(7), time_namespaces(7), ip-netns(8), switch_root(8), ioctl_ns(2), lsns(8)Q


 Marc Lehmann <>