getfacl: implement --skip-base (-s)

[kevans]

Summary

This skips files that only have a trivial ACL, which is useful for
callers that want to examine files with more unique ACLs.

While we're here, don't issue the file name/ownership header if we're
just going to error out on the file; this adds extra noise for little
gain.

Sponsored by: Klara, Inc.

[markj]

bin/getfacl/getfacl.c

136

Why clear errno here?

151

Perhaps warn or exit if acl_is_trivial_np() fails?

[kevans]

bin/getfacl/getfacl.c

136

I hadn't done any due diligence here, this is just moved from old line 147... that said, I'm not sure I see anything that should really be affected if we just didn't.

[olce]

bin/getfacl/getfacl.c

151

I concur.

That said, acl_is_trivial_np() can fail only on an unknown ACL type or on out of memory (only for NFSv4 ACLs). Assuming there is no bug, only the second could occur in practice, and in this case it may be best to warn but also continue to output ACLs.

Exiting with a message, e.g., indicating one to run getfacl again without -s, would work interactively. On the other hand, scripts gathering state at a given time would have to explicitly cope with this condition (e.g., getfacl -s <paths> || getfact <paths>).

So I have a slight bias in favor of a warning and continuing to print ACLs even if not trivial.

Thanks.

[kevans]

• Stop resetting errno
• Warn if acl_is_trivial_np failed (likely out of memory?), but proceed with both this ACL (err on the side of caution since we don't know if it's trivial) and following ACLs

This also plugs one leak of acl in this new code, but there's notably at least
one more in a later failure path.

[olce]

Thanks!