On 3/23/2022 7:00 AM, Miklos Szeredi wrote:

On Wed, 23 Mar 2022 at 14:51, Casey Schaufler [off-list ref] wrote:

quoted

You also need a way to get a list off what attributes are available
and/or a way to get all available attributes. Applications and especially
libraries shouldn't have to guess what information is relevant. If the
attributes change depending on the filesystem and/or LSM involved, and
they do, how can a general purpose library function know what data to
ask for?

Oh, yes.  Even the current prototype does that:

# ~/getvalues / ""
[] = "mnt" "mntns" "xattr" "data" (len=21)
# ~/getvalues / "mnt"
[mnt] = "id" "parentid" "root" "mountpoint" "options" "shared"
"master" "propagate_from" "unbindable" (len=76)
# ~/getvalues / "mntns"
[mntns] = "21" "22" "24" "25" "23" "26" "27" "28" "29" "30" "31" "32" (len=36)
  ~/getvalues / "mntns:21"
[mntns:21] = "id" "parentid" "root" "mountpoint" "options" "shared"
"master" "propagate_from" "unbindable" (len=76)

That requires multiple calls and hierarchy tracking by the caller.
Not to mention that in this case the caller needs to understand
how mount namespaces are being used. I don't see that you've made
anything cleaner. You have discarded the type checking provided
by the "classic" APIs. Elsewhere in this thread the claims of
improved performance have been questioned, but I can't say boo
about that. Is this interface targeted for languages other than C
for which the paradigm might provide (more?) value?

I didn't implement enumeration for "data" and "xattr" but that is
certainly possible and not even difficult to do.

Thanks,
Miklos