Re: [fuse-devel] umount(8) returning before destroy completes

[Archie C. <ar...@de...>]

On Tue, May 7, 2013 at 1:25 PM, Archie Cobbs <ar...@de...> wrote:

> On Tue, May 7, 2013 at 10:00 AM, Miklos Szeredi <mi...@sz...> wrote:
>
>> On Tue, May 7, 2013 at 4:25 PM, Archie Cobbs <ar...@de...> wrote:
>> > Presumably, at some point this option has to be communicated to the
>> kernel.
>> > So that system call can return an error if the user is not root, right?
>>
>> Yes.  The problem is that fusermount is suid-root and calls the
>> mount(2) syscall with superuser privileges whether it was invoked by
>> root or not.  So the kernel cannot check for the usual
>> capable(CAP_FOO_BAR) when checking whether the user has enough
>> privileges for -osync_umount.
>>
>> But fusermount does pass "user_id=UID" option to the kernel.  And the
>> kernel can check for user_id=0.  Which is not very nice but looks like
>> the only option.
>>
>
> Got it. If time permits I will try to look into this idea further.
>

I've attached a partial patch. Not being at all familiar with the code,
it's likely incomplete and/or wrong, and I didn't even attempt to address
the kernel side. Note the fuse_kernel.h file was not up-to-date wrt. Linux
(minor version 21 instead of 22) so I included the version 22 update as
well.

Miklos please let me know your thoughts.

Thanks,
-Archie

-- 
Archie L. Cobbs