Weird issue with file permissions

Lavrentiev, Anton (NIH/NLM/NCBI) [C]
Sun Jul 3 01:51:07 GMT 2022

> That's not what I'm seeing when I run your test program on Linux:
> $ ./sun
> fstat mode = 140666
> stat mode = 140777

True, but it creates the socket file with exactly how umask(0) told it to,
and stat() shows that.  So yeah, I should retract that it works on Linux with
fchmod() -- on Linux the fchmod() call won't be at all necessary.  And I just
checked BSD and MacOS, too.  [Truly, it's an old code that used to work everywhere
but failed on Cygwin -- that's how I noticed, so I assumed it was because of
fchmod() -- but it actually because of umask(0).]

On Cygwin, however, I have to resort to good old chmod().

> Which is not necessarily related to the permissions on the file. Windows
> socket is an in-memory object, the file is used merely for naming purposes.

Sockets are in-memory objects everywhere.  The UNIX socket file is a just connection
"point" (much like devices and other special files) that has to have proper permissions
for an accessor to be able to connect / read / etc.  The permissions are checked first,
then everything else goes.  So if a socket file in the filesystem isn't "readable"
for your permission category, you won't be able to connect regardless of what the
in-memory things are, IIRC.

> "Works", all right. But HOW does it works? Aren't the permissions seen on the
> socket file merely a coincidence/convenience?

No, they are NOT.  They are actually granting / denying the access.

Anyways, I think that I know how to fix this.  As to whether or not Cygwin must be
brought in line with Linux -- I can't tell, because I don't have files with ACLs
on Linux, so I can't see how umask(0), when ignored, would screw permissions there...

Anton Lavrentiev
Contractor NIH/NLM/NCBI

More information about the Cygwin mailing list