Re: SMBFS mount's file cannot be made executable

[Takashi Yano <takashi.yano@nifty.ne.jp>]

On Sun, 8 Dec 2024 08:13:38 +0900
Takashi Yano wrote:
> On Tue, 19 Nov 2024 21:54:44 +0100
> Corinna Vinschen wrote:
> > On Nov 19 17:58, Takashi Yano via Cygwin wrote:
> > > On Mon, 18 Nov 2024 17:26:12 +0100
> > > Corinna Vinschen wrote:
> > > > We can safely assume that the current user is already authorized on the
> > > > SMB server.  So... shouldn't AuthzInitializeResourceManager be
> > > > sufficient and the code from class authz_ctx already does what we want?
> > > > We may just have to use in in place of calling NtCheckAccess(),
> > > > maybe with a tweak or two...
> > > 
> > > I already tried AuthzInitializeResourceManager(), but the result
> > > was the same with current implementation...
> > 
> > So you tried to call authz_get_user_attribute()?
> 
> Yes. But resulted in the same.
> 
> > > BTW, I come up with another implementation. This make the things
> > > much simpler. What do you think of the patch attached?
> > 
> > > [...]
> > >  int
> > >  check_file_access (path_conv &pc, int flags, bool effective)
> > >  {
> > > @@ -711,10 +618,14 @@ check_file_access (path_conv &pc, int flags, bool effective)
> > >      desired |= FILE_EXECUTE;
> > >    if (!get_file_sd (pc.handle (), pc, sd, false))
> > >      {
> > > -      /* Tweak Samba security descriptor as necessary. */
> > > -      if (pc.fs_is_samba ())
> > > -	convert_samba_sd (sd);
> > > -      ret = check_access (sd, file_mapping, desired, flags, effective);
> > > +      HANDLE h = CreateFileW (pc.get_nt_native_path ()->Buffer, desired,
> > > +			      0, NULL, OPEN_EXISTING,
> > > +			      FILE_FLAG_BACKUP_SEMANTICS, NULL);
> > > +      if (h != INVALID_HANDLE_VALUE)
> > > +	{
> > > +	  CloseHandle (h);
> > > +	  ret = 0;
> > > +	}
> > >      }
> > >    debug_printf ("flags %y, ret %d", flags, ret);
> > >    return ret;
> > 
> > No, we can't do that, it's too simple.
> > 
> > Just kidding.
> > 
> > This is so simple, I'm puzzled we never tried that before.  Or, if we
> > did, it's a loooong time ago...
> > 
> > If we really do this, we don't even need to call get_file_sd().  And it
> > should use NtOpenFile and reopen semantics i.e.  pc.init_reopen_attr().
> > Also, the sharing flags should allow all access.  And the `effective'
> > argument needs to be taken into account.
> 
> I have a question. What pc.init_reopen_attr() is for? I tested with
> pc.get_object_attr() instead, it works. What handle should I pass
> to pc.init_reopen_attr()?
> 
> Anyway, I revised the patch as attached. What do you think?

diff --git a/winsup/cygwin/sec/base.cc b/winsup/cygwin/sec/base.cc
index fcc5e1ff7..68e7fd9e1 100644
--- a/winsup/cygwin/sec/base.cc
+++ b/winsup/cygwin/sec/base.cc
@@ -643,6 +643,8 @@ check_file_access (path_conv &pc, int flags, bool effective)
       NtClose (h);
       ret = 0;
     }
+  else
+    __seterrno_from_nt_status (status);
   if (!effective && cygheap->user.issetuid ())
     {
       /* Recover impersonation token */

-- 
Takashi Yano <takashi.yano@nifty.ne.jp>