CLONE_NEWNS and bind mounts to make "chroot" jail

CLONE_NEWNS and bind mounts to make "chroot" jail

[Leibowitz, Michael]

I have been trying to use the combination of CLONE_NEWNS and bind mounts
to create a better (than) chroot jail.  I wish to have the ability to
bind (ro will be possible in the future, I understand) certain
directories into the jail (perhaps /bin, /lib, /usr), but not have
parallel directories in the jail (no /etc, ..., /home). 

I have heard that this should be possible, but have yet to get a working
solution.

I have tried something analogous to:
chdir("/jail");
unshare(CLONE_NEWNS);
/* mount(8) syntax given for simplicity, but mount(2) used below */
mount --bind / /jail/old_root
mount --bind /jail/old_root/bin /jail/bin
mount --bind /jail/old_root/usr /jail/usr
mount --bind /jail/old_root/lib /jail/lib
mount --bind /jail /               # does nothing?
umount2("/old_root", MNT_DETACH);  # never happens.
exec("bin/sh");

When bin/sh runs, I can still see old_root from /jail and the bind of
/jail over / seems to have not done anything.  

Is it possible to create such a jail with bind mounts?  Is there a
recommended method for doing so?  Thank you for your time.

--
Michael Leibowitz
Software Engineer, UMG
Intel Corporation
michael.leibowitz at intel.com

Re: CLONE_NEWNS and bind mounts to make "chroot" jail

[serge]

Quoting Leibowitz, Michael (michael.leibowitz@intel.com):
> I have been trying to use the combination of CLONE_NEWNS and bind mounts
> to create a better (than) chroot jail.  I wish to have the ability to
> bind (ro will be possible in the future, I understand) certain
> directories into the jail (perhaps /bin, /lib, /usr), but not have
> parallel directories in the jail (no /etc, ..., /home). 
> 
> I have heard that this should be possible, but have yet to get a working
> solution.
> 
> I have tried something analogous to:

Try a few more things.  Since you had entered /jail, you can view '/' by
looking at .. .  But if you look at /, you dereference your
task->fsroot.  You never changed that, so it points to the original
mount.  If however you 'ls ..', you should see your 'jail' directory.
However it won't have the /bin and /lib mounted because you didn't
	mount --rbind /jail /
What you really want to do is
	mount --bind /jail /jail
to make sure it's a mountpoint, then set up the new /jail using bind
mounts like you're doing (and likely some rbinds in some places), then
use pivot_root() to change your root.  Then umount2("/old_root",
MNT_DETACH).

-serge

> chdir("/jail");
> unshare(CLONE_NEWNS);
> /* mount(8) syntax given for simplicity, but mount(2) used below */
> mount --bind / /jail/old_root
> mount --bind /jail/old_root/bin /jail/bin
> mount --bind /jail/old_root/usr /jail/usr
> mount --bind /jail/old_root/lib /jail/lib
> mount --bind /jail /               # does nothing?
> umount2("/old_root", MNT_DETACH);  # never happens.
> exec("bin/sh");
> 
> When bin/sh runs, I can still see old_root from /jail and the bind of
> /jail over / seems to have not done anything.  
> 
> Is it possible to create such a jail with bind mounts?  Is there a
> recommended method for doing so?  Thank you for your time.
> 
> --
> Michael Leibowitz
> Software Engineer, UMG
> Intel Corporation
> michael.leibowitz at intel.com
> 
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

RE: CLONE_NEWNS and bind mounts to make "chroot" jail

[Leibowitz, Michael]

If I understand correctly, the following should accomplish what I'm
looking for.  However, pivot_root gives me EBUSY.  I played around with
moving the mount --bind /jail /jail to before the unshared, as well as
making old_root a bind mount to itself.  However, pivot_root always
seems to fail.  Is there something obvious that I'm doing wrong?  The
following is my test code (error checking has been removed for clarity,
except for pivot_root).  

  char *newargv[]= { "sh", NULL };

  chdir("/jail");
  unshare(CLONE_NEWNS));
  mount("/jail", "/jail", NULL, MS_BIND, NULL));
  mount("/bin", "bin", NULL, MS_BIND, NULL));
  mount("/usr", "usr", NULL, MS_BIND, NULL));
  mount("/lib", "lib", NULL, MS_BIND, NULL));
  if (pivot_root(".", "old_root")) perror("pivot_root . old_root");
  exec("./bash-static"); /* copied to /jail prior to running */

Thanks. 

>Serge replies:
 [snip...snip]
>Try a few more things.  Since you had entered /jail, you can view '/'
by
>looking at .. .  But if you look at /, you dereference your
>task->fsroot.  You never changed that, so it points to the original
>mount.  If however you 'ls ..', you should see your 'jail' directory.
>However it won't have the /bin and /lib mounted because you didn't
>	mount --rbind /jail /
>What you really want to do is
>	mount --bind /jail /jail
>to make sure it's a mountpoint, then set up the new /jail using bind
>mounts like you're doing (and likely some rbinds in some places), then
>use pivot_root() to change your root.  Then umount2("/old_root",
>MNT_DETACH).
>
>-serge

--
Michael Leibowitz
Software Engineer, UMG
Intel Corporation
michael.leibowitz at intel.com

Re: CLONE_NEWNS and bind mounts to make "chroot" jail

[serge]

Quoting Leibowitz, Michael (michael.leibowitz@intel.com):
> If I understand correctly, the following should accomplish what I'm
> looking for.  However, pivot_root gives me EBUSY.  I played around with
> moving the mount --bind /jail /jail to before the unshared, as well as
> making old_root a bind mount to itself.  However, pivot_root always
> seems to fail.  Is there something obvious that I'm doing wrong?  The

Yes, you
	cd /jail
	mount --bind /jail /jail
	pivot_root . old_root

but . is now mounted over.

-serge

> following is my test code (error checking has been removed for clarity,
> except for pivot_root).  
> 
>   char *newargv[]= { "sh", NULL };
> 
>   chdir("/jail");
>   unshare(CLONE_NEWNS));
>   mount("/jail", "/jail", NULL, MS_BIND, NULL));
>   mount("/bin", "bin", NULL, MS_BIND, NULL));
>   mount("/usr", "usr", NULL, MS_BIND, NULL));
>   mount("/lib", "lib", NULL, MS_BIND, NULL));
>   if (pivot_root(".", "old_root")) perror("pivot_root . old_root");
>   exec("./bash-static"); /* copied to /jail prior to running */
> 
> Thanks. 
> 
> >Serge replies:
>  [snip...snip]
> >Try a few more things.  Since you had entered /jail, you can view '/'
> by
> >looking at .. .  But if you look at /, you dereference your
> >task->fsroot.  You never changed that, so it points to the original
> >mount.  If however you 'ls ..', you should see your 'jail' directory.
> >However it won't have the /bin and /lib mounted because you didn't
> >	mount --rbind /jail /
> >What you really want to do is
> >	mount --bind /jail /jail
> >to make sure it's a mountpoint, then set up the new /jail using bind
> >mounts like you're doing (and likely some rbinds in some places), then
> >use pivot_root() to change your root.  Then umount2("/old_root",
> >MNT_DETACH).
> >
> >-serge
> 
> --
> Michael Leibowitz
> Software Engineer, UMG
> Intel Corporation
> michael.leibowitz at intel.com
> 
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Re: CLONE_NEWNS and bind mounts to make "chroot" jail

[Leibowitz, Michael]

[-- Attachment #1: Type: text/plain, Size: 1208 bytes --]

I'm not 100% sure if this is what you meant, but I did get the following
to work: 

chdir("/jail"); 
unshare(CLONE_NEWNS);
mount("/jail", "/jail", NULL, MS_BIND, NULL);
pivot_root("/jail", "/jail/old_root");
chdir("/");
mount("/old_root/bin", "bin", NULL, MS_BIND, NULL);
mount("/old_root/usr", "usr", NULL, MS_BIND, NULL);
mount("/old_root/lib", "lib", NULL, MS_BIND, NULL);
umount2("/old_root", MNT_DETACH);
exec("/busybox");

Thanks for the help.  

On Tue, 2008-03-04 at 15:45 -0600, serge@hallyn.com wrote:
> Quoting Leibowitz, Michael (michael.leibowitz@intel.com):
> Yes, you
> 	cd /jail
> 	mount --bind /jail /jail
> 	pivot_root . old_root
> 
> but . is now mounted over.
> >   char *newargv[]= { "sh", NULL };
> > 
> >   chdir("/jail");
> >   unshare(CLONE_NEWNS));
> >   mount("/jail", "/jail", NULL, MS_BIND, NULL));
> >   mount("/bin", "bin", NULL, MS_BIND, NULL));
> >   mount("/usr", "usr", NULL, MS_BIND, NULL));
> >   mount("/lib", "lib", NULL, MS_BIND, NULL));
> >   if (pivot_root(".", "old_root")) perror("pivot_root . old_root");
> >   exec("./bash-static"); /* copied to /jail prior to running */

-- 
Michael Leibowitz <michael.leibowitz@intel.com>

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

RE: CLONE_NEWNS and bind mounts to make "chroot" jail

[Bodo Eggert]

Leibowitz, Michael <michael.leibowitz@intel.com> wrote:

> If I understand correctly, the following should accomplish what I'm
> looking for.  However, pivot_root gives me EBUSY.  I played around with
> moving the mount --bind /jail /jail to before the unshared, as well as
> making old_root a bind mount to itself.  However, pivot_root always
> seems to fail.  Is there something obvious that I'm doing wrong?  The
> following is my test code (error checking has been removed for clarity,
> except for pivot_root).
> 
>   char *newargv[]= { "sh", NULL };
> 
>   chdir("/jail");
>   unshare(CLONE_NEWNS));
>   mount("/jail", "/jail", NULL, MS_BIND, NULL));
>   mount("/bin", "bin", NULL, MS_BIND, NULL));
>   mount("/usr", "usr", NULL, MS_BIND, NULL));
>   mount("/lib", "lib", NULL, MS_BIND, NULL));
>   if (pivot_root(".", "old_root")) perror("pivot_root . old_root");
>   exec("./bash-static"); /* copied to /jail prior to running */

This works for me:

#include <sys/mount.h>
#include <unistd.h>

#define _GNU_SOURCE
#include <sched.h>

#define MNT_DETACH      2       /* Detach from tree only */

int main()
{
  unshare(CLONE_NEWNS);
  mount("jail", "jail", NULL, MS_BIND, NULL);
  mount("/bin", "jail/bin", NULL, MS_BIND, NULL);
  mount("/usr", "jail/usr", NULL, MS_BIND, NULL);
  mount("/lib", "jail/lib", NULL, MS_BIND, NULL);
  /* abuse bin as the temporary old root directory */
  if (pivot_root("jail", "jail/bin"))
    perror("pivot_root");
  chdir("/");
  umount2("bin", MNT_DETACH);
  execl("./sash", NULL); /* copied to /jail prior to running */
}