On Mon, Oct 12, 2015 at 6:05 AM, Dave Chinner [off-list ref] wrote:

On Mon, Oct 12, 2015 at 03:51:15AM +0200, Andreas Grünbacher wrote:

quoted

2015-10-12 2:10 GMT+02:00 Dave Chinner [off-list ref]:

quoted

Also, I really dislike the API where passing a NULL acl means to
"set this acl" actually means "remove the existing ACL". Why no
->remove_acl method called from the generic code?

It's not uncommon, it saves inode operations and wiring-up code.

I know it's common. All it does is put extra branches in the
filesystem code to do this, because remove is a different operation
to set. The API sucks, and we're not limited on inode operations,
and the operator overloading makes the filesystem code unnecessarily
complex as it has to detect when to branch out ot remove or not...

I've tried it out. The filesystem code could be simplified (see the
richacl-wip [*] branch until the next posting). Adding a
remove_richacl inode operation on top of that really doesn't help.

Thanks,
Andreas

[*] git://git.kernel.org/pub/scm/linux/kernel/git/agruen/linux-richacl.git
richacl-wip