Scroll to navigation

LIBUSERBINDMOUNT(3) Linux Programmer's Manual LIBUSERBINDMOUNT(3)

NAME

userbindmount, userbindmount_unshare, userbindmount_set_cap_sysadm, userbindmount_string, userbindmount_data, userbindmount_fd - bind-mount in user-namespaces

SYNOPSIS

#include <userbindmount.h>
int userbindmount(const char *source, const char *target);
int userbindmount_unshare(void);
int userbindmount_set_cap_sysadm(void);
int userbindmount_string(const char *string, const char *target, mode_t mode);
int userbindmount_data(const void *data, size_t count, const char *target, mode_t mode);
int userbindmount_fd(int fd, const char *target, mode_t mode);

These functions are provided by libuserbindmount. Link with -luserbindmount.

DESCRIPTION

Libuserbindmount is a library providing support for bind mount in user userspaces.

userbindmount bind mounts source on destination. If it is not permitted in the current namespace it creates (unshare) a new user-namespace.

userbindmount_unshare creates a new user-namespace where bind mount is allowed.

userbindmount_set_cap_sysadm add the CAP_SYS_ADMIN ambient capability to the current namespace so that the ability of bind mount files and directories can be exported to new programs (ambient capabilities survive to execve(2)).

userbindmount_string bind-mounts on destination a temporary file whose (text) contents is provided by string. The temporary file is automatically deleted when the namespace is closed or the file/directory unmounted.

userbindmount_data bind-mounts on destination a temporary file whose (binary) contents is provided by data and has the size of count bytes. The temporary file is automatically deleted when the namespace is closed or the file/directory unmounted.

userbindmount_fd bind-mounts on destination a temporary file whose contents is read from the by the file descriptor fd (up to the end of file). The temporary file is automatically deleted when the namespace is closed or the file/directory unmounted.

RETURN VALUE

All the functions provided by libuserbindmount return 0 in case of success. -1 is returned elseways and errno is set appropriately.

NOTES

Libuserbindmount fails if user namespaces have not been configured in the running kernel and enabled for users. In Debian the sysctl knob kernel.unprivileged_userns_clone must be set to 1.

EXAMPLES

The following excerpts of C code shows the use of libuserbindmount: the inclusion of the header file for this library is required:

#include <usrbindmount.h>

Bind-mount /tmp/resolv.conf on /etc/resolv.conf:

userbindmount("/tmp/resolv.conf", "/etc/resolv.conf");

Bind-mount a string on /etc/resolv.conf:

userbindmount_string("nameserver 9.9.9.90, "/etc/resolv.conf", 0600);

Bind-mount a binary data on /proc/self/cmdline:

static char fakeargv[] = {'c','m','d',0,
	'a','r','g','1',0,
	'a','r','g','2',0,
	0};
userbindmount_data(fakeargv, sizeof(fakeargv), "/proc/self/cmdline", 0600);

Bind-mount the data read from a file descriptor on /etc/resolv.conf:

userbindmount_fd(STDIN_FILENO, "/etc/resolv.conf", 0600);

Bind-mount several files or directories:

userbindmount("/tmp/resolv.conf", "/etc/resolv.conf");
userbindmount("/tmp/passwd", "/etc/passwd");
userbindmount("/tmp/hosts", "/etc/hosts");
Only the first userbindmount creates a new namespace if needed.

The following program creates a namespace and runs a program in it. In the new namespace bind-mount is allowed.

#include <stdio.h>
#include <unistd.h>
#include <userbindmount.h>
int main(int argc, char *argv[]) {
	userbindmount_unshare();
	userbindmount_set_cap_sysadm();
	execvp(argv[1], argv+1);
}

It can be compiled and tested in the following way:

$ gcc -o unshare_sysadm unshare_sysadm.c -luserbindmount
$ unshare_sysadm bash
$ cat /etc/resolv.conf 
nameserver 127.0.0.1
$ echo "nameserver 9.9.9.9" > /tmp/resolv.conf
$ busybox mount --bind /tmp/resolv.conf /etc/resolv.conf 
$ cat /etc/resolv.conf
nameserver 9.9.9.9
$ exit
$
please note that in the example the mount command by busybox has been used instead of the standard mount by util-linux. In fact the standard mount command has not been updated to support the capabilities, and forbids the access to the mount system call if the effective user is not root, denying in this way a legal operation.

SEE ALSO

mount(2), mount(8), user_namespaces(7)

BUGS

Bug reports should be addressed to <info@virtualsquare.org>

AUTHORS

Renzo Davoli <renzo@cs.unibo.it>

2017-08-22 VirtualSquare