Re: inotify cookie regression/info leak in latest mainline

[Vegard Nossum <vegard.nossum@oracle.com>]

On 02/17/2014 01:59 PM, Jan Kara wrote:
>    Hello,
>
> On Sat 15-02-14 22:39:38, Vegard Nossum wrote:
>> It would seem that
>>
>> commit 7053aee26a3548ebaba046ae2e52396ccf56ac6c
>> Author: Jan Kara <jack@suse.cz>
>> Date:   Tue Jan 21 15:48:14 2014 -0800
>>
>>      fsnotify: do not share events between notification groups
>>
>> introduced a bug where the cookie field of struct inotify_event
>> never gets initialised. In particular, it used to be initialised
>> when send_to_group() called fsnotify_create_event(), but that no
>> longer happens, and the 'cookie' parameter of send_to_group() never
>> gets used.
>>
>> The problem manifests itself in copy_event_to_user() where the
>> cookie field is copied to userspace without being initialised.
>>
>> I tested this with a simple userspace program, I seem to get mostly
>> 0xffff8800 in the cookie field for non-move events (which should
>> always have 0 here).
>    That's a really embarassing bug. I've extented LTP inotify tests to
> verify the cookie value is sane (so far the tests completely ignored the
> value which is why I didn't notice the breakage).
>
> Attached patch fixes the problem for me. I'll send it to Linus tomorrow.
> Thanks for spotting the problem!

That seems to fix it for me too, thanks for the quick fix!


Vegard