Re: [Monotone-devel] Please review quickly [Fwd: [bug #19137] permissions on ~/.monotone/keys/ are too permissive]

[Nathaniel Smith]

On Sat, Apr 14, 2007 at 01:27:33PM -0400, Stephen Leake wrote:
> This is my first time compiling monotone. Looking thru other code, it
> seems the correct fix might be:
[...]
> +#ifndef WIN32
> +  // umask not in MinGW. We assume MinGW boxes have single users, so
> +  // this doesn't matter.
>    mode_t mask = umask(S_IRWXG|S_IRWXO);
> +#endif /* WIN32 */
> +
>    L(FL("writing key '%s' to file '%s' in dir '%s'") % ident % file % 
> key_dir);
>    write_data(file, dat, key_dir);
> +
> +#ifndef WIN32
>    umask(mask);
> +#endif /* WIN32 */
[...]

No... it's entirely possible you found other code like this, alas, but
that code is wrong and misleading :-).

The correct fix would be adding something like

struct enable_restrictive_umask
{
  enable_restrictive_umask();
  ~enable_restrictive_umask();
};

And then implementing those two methods twice, once somewhere in
win32/ and once somewhere in unix/, with either empty bodies or bodies
that call umask.

...Except that the unix version needs a field in there to temporarily
stash the old umask.  After a little more looking around, it might be
far smarter to use fchmod(2) (inside an appropriate platform.hh
wrapper, of course) than to mess around with umask?

-- Nathaniel

-- 
Eternity is very long, especially towards the end.
  -- Woody Allen