Seems like a bug with mkfifo -m

[Ken Brown]

On 5/28/2020 10:52 AM, Corinna Vinschen wrote:
> On May 28 09:16, Ken Brown via Cygwin wrote:
>> On 5/28/2020 4:12 AM, Дмитрий Есарев via Cygwin wrote:
>>> Hi, all
>>>
>>> When i ran cygwin 2.x, i used mkfifo -m 0600 file to create a named pipe with no user and group permissions.
>>>
>>> in the latest cygwin the above command creates device with 0644 permissions. And i cant drop it to 0600:
>>>
>>> cygcheck.exe -V
>>> cygcheck (cygwin) 3.1.4
>>>
>>> $ umask 0077
>>> $ touch somefile; ls -l somefile
>>> -rw------- 1 admin absent 0 may 26 18:15 somefile
>>>
>>> $ mkfifo -m 0600 somefifo; ls -l somefifo
>>> prw-r--r-- 1 admin absent 0 may 26 18:16 somefifo
>>>
>>> $ chmod 600 somefifo; ls -l somefifo
>>> prw-r--r-- 1 admin absent 0 may 26 18:16 somefifo
>>>
>>>
>>>
>>> In old-good cygwin 2.x the command works as expected:
>>>
>>> $ cygcheck.exe -V
>>> cygcheck (cygwin) 2.9.0
>>>
>>> $ umask
>>> 0022
>>>
>>> $ mkfifo -m 0600 somefifo; ls -l somefifo
>>> prw------- 1 builduser Domain Users 0 May 26 18:21 somefifo
>>
>> Thanks for the report.  The problem isn't with mkfifo, it's with the
>> permission information reported by ls.  I did a bisection of the Cygwin
>> development repo and found that the regression was introduced by the
>> following commit:
>>
>> commit f36262d56ac78f04de147746ce4a85c6155e4a23
>> Author: Corinna Vinschen <corinna@vinschen.de>
>> Date:   Wed Jan 29 15:14:05 2020 +0100
>>
>>      Cygwin: stat: fix st_mode of fifos
>>
>> I'll take a look if Corinna doesn't get to it first.
> 
> Not sure what I was thinking at the time.  I recall having observed
> something funny, but the patch was apparently wrong.  Just revert
> it at your discretion, Ken.

I remember we had an IRC discussion about it, but I can't remember what the 
issue was.  I'll look a little more closely before reverting.

Ken