On Aug 27 12:50, Corinna Vinschen wrote:
> On Aug 27 12:41, Corinna Vinschen wrote:
> > On Aug 27 11:09, Corinna Vinschen wrote:
> > > On Aug 26 20:32, L A Walsh wrote:
> > > > On 8/23/2018 1:11 AM, Corinna Vinschen wrote:
> > > > ...
> > > > > No, that's a wrong assumption.  Think about it.  The ACL given to
> > > > > acl_to_text is the binary form, so it doesn't contain user or group
> > > > > names, only uids and gids.  The usernames are only generated in the
> > > > > output.
> > > > ---
> > > > 	Rats.  Of course, you're right.  	Then I nominate the problem being that it
> > > > can't convert from domain "Unknown"-user + "Unknown"-group to something it
> > > > can store in tar.
> > > 
> > > The problem with unknown SIDs is that there's no bijective
> > > transformation between SID <-> uid/gid.  You get the uid/gid -1 and
> > > then... what?  How do you restore the information?  There's no SID for
> > > uid/gid -1.
> > > 
> > > > As far as duplication, I have /etc/passwd+/etc/group files that mirror my
> > > > accounts on the linux-based PDC (samba 3.x).
> > > 
> > > What for?  This should work automatically and you would get rid of those
> > > dreaded backslashes in the account names.  Using passwd/group files also
> > > have a higher probability of account overlap with weird results.
> > > 
> > > Passwd and group files should only be used if you have very specific
> > > problems to solve (like offline usage or see below), otherwise just use
> > > the values you get from the account DBs.
> > > 
> > > > In this case, that user+group appear to correspond
> > > > to non-existent users. (S-1-5-21-oldsystem-ID-1001 + -1005).
> > > > The domain/system part appears to be from some previous
> > > > value for the machine's "sid"?  Not sure how to deliberately
> > > > reproduce that, but maybe you have a tool to create an
> > > > invalid acl entry for a user like: Unknown+User:*:4294967295:4294967295:S-1-5-21-3457732827-2369206082-2151550420-1001
> > > > in /etc/passwd.
> > > > and something similar in /etc/group?
> > 
> > Actually, I just did that.  I added a user and a group to the files with
> > weird SIDs, then I switched /etc/nsswitch.conf to "db" only.  With
> > different ACLs (created by Cygwin, created by native Windows) there are
> > different results.  The problem is that uid/gid -1 can be created as a
> > file ACL entry *and* at the same time have the meaning of "don't look
> > for the uid/gid" when checking the ACL for validity.  To make matters
> > worse, if you have multiple ACEs of unknown users, the resulting ACL is
> > *always* invalid.
> > 
> > Bottom line is, there are at least two bugs here in Cygwin.  I'm looking
> > into a fix.
> 
> The only sane way to handle unknown SIDs in file ACLs is to ignore them
> entirely.  The result will be that you never see them in getfacl, nor
> will they be stored by tar or rsync.  They are just not there from the
> Cygwin perspective.

I created a patch, uploaded developer snapshots to
https://cygwin.com/snapshots/ and released a new Cygwin test
release 2.11.0-0.4 with this change.  Please giver any of
them a try.


Thanks,
Corinna

-- 
Corinna Vinschen                  Please, send mails regarding Cygwin to
Cygwin Maintainer                 cygwin AT cygwin DOT com
Red Hat