O3X 1.5.2 and xdarwin ACLs?

[roemer]

First of all, thank you very much for the great development work and for releasing a new version of OpenZFS on OSX!

I am currently on O3X version 1.4.5 with xattr-darwinacls (forum build).
I noted that the Changelog for 1.5.2 mentions that default ACLs are now hidden.
That sounds great, but if I remember correctly there were other issues with the Finder too when it comes to ACLs and O3X. The xattr-darwinacls version works perfect in this regard though. Will there be another forum build of 1.5.2 with xattr ACLs, or are the ACL issues now resolved in 1.5.2?

Searching a bit around, I found on Github Issue #433 (acltype property) which mentions a potentially new acltype property 'apple_xsec_compat' to support Apple's extended-attribute-based compatibility ACL implementation. Is this already implemented, or just an enhancement request at this moment?

Thanks

[lundman]

https://github.com/openzfsonosx/zfs/issues/275

It is just a difference between ZFS and XNU, one always includes the 6 trivials, and the other doesn't. So now we simply hide them when sending ACLs out, and merge them back in when we receive ACLs back. So in theory, making it transparent to the user.

[ilovezfs]

@roemer I'm willing to consider putting out another xattr darwins acls release, but would prefer to get some feedback first (especially from you) about whether the changes made for 1.5* are sufficient to meet your needs without the compat "hack."

[roemer]

Fair enough. I only haven't yet found the time to do the update and especially the corresponding tests.
I certainly will let you know once I have done it.

[roemer]

@ilovezfs: Yes, v1.5.2 looks very good indeed. Granted, I did not have much time to spend with it yet, but I tried a few file syncs using ChronoSync and various test files and directories, and so far no problem. ACLs are copied and displayed as expected, which means that the default ACLs from ZFS are kept hidden as expected.
That's great.

I also note that installation on 10.11.4 worked perfectly without problems.
I encounter just one small issue with one test, in that the order of extended attributes was different than on the source. The content was the same, just 'xattr -l' listed them in a different order. Not sure whether this is a problem with ChronoSync, or whether this is actually not specified - or whether it is a speciality of O3X.
But a test case with "test `xattr -l $1` = `xattr -l $2`" did fail due to this. Semantically, it should not make a difference, I just noted it.

[ilovezfs]

Thanks for testing it out.

Does Apple's or brew's rsync preserve the xattr order?

[roemer]

ACLs are copied successfully with Apple's rsync, as well as the cp and ditto commands.

By the way: What happens with ACLs which previously were created using the xattr-darwinacls built?
I have one directory with existing ACLs and they are displayed correctly with 1.5.2.
How do I tell whether those ACLs are 'real' ACLs or whether they come out of some special extended attributes?

[ilovezfs]

Are we talking about ACL order or xattr order?

You could use `sudo zdb -dddd foo/bar/wonder` to look for which files have the hidden com.apple.system.Security xattr.

Or you could use the old xattr darwin acls version to remove them.