Re: [RFC PATCH] getting misc stats/attributes via xattr API

[Dave Chinner <david@fromorbit.com>]

On Thu, May 05, 2022 at 04:38:12PM -0700, tytso wrote:
> On Tue, May 03, 2022 at 02:23:23PM +0200, Miklos Szeredi wrote:
> > 
> > : - root
> > bar - an attribute
> > foo: - a folder (can contain attributes and/or folders)
> > 
> > The contents of a folder is represented by a null separated list of names.
> > 
> > Examples:
> > 
> > $ getfattr -etext -n ":" .
> > # file: .
> > :="mnt:\000mntns:"
> 
> In your example, does it matter what "." is?  It looks like in some
> cases, it makes no difference at all, and in other cases, like this,
> '.' *does* matter:
> 
> > $ getfattr -etext -n ":mnt:info" .
> > # file: .
> > :mnt:info="21 1 254:0 / / rw,relatime - ext4 /dev/root rw\012"
> 
> Is that right?
> 
> > $ getfattr -etext -n ":mntns:" .
> > # file: .
> > :mntns:="21:\00022:\00024:\00025:\00023:\00026:\00027:\00028:\00029:\00030:\00031:"
> 
> What is this returning?  All possible mount name spaces?  Or all of
> the mount spaces where '.' happens to exist?
> 
> Also, using the null character means that we can't really use shell
> scripts calling getfattr.

Yeah, it should be returning an attr per namespace, not an attr
whose value contains all the valid namespaces.

i.e. if the next level of the heirachy is 21, 22, 24, .... we should
be seeing a listing of multiple attributes with naming like:

:mntns:21:
:mntns:22:
:mntns:24:
....

rather than an attribute whose value contains the names of the
attrbiutes in the next layer of the heirarchy. Then we can just
pull the namespace we want and feed it directly to:

$ getfattr -n ":mntns:21:"

and we get a list of all the attributes available for that
namespace...

> I understand that the problem is that in
> some cases, you might want to return a pathname, and NULL is the only
> character which is guaranteed not to show up in a pathname.  However,
> it makes parsing the returned value in a shell script exciting.

We shouldn't be returning the names of children in an attribute
value. We have a syscall API for doing this that - listxattr() will
iterate attribute names just like a directory does with readdir()
via listxattr(). IOWs, we should not need to encode the next layer
of child attribute names into the value of the parent attribute - if
we do a listxattr on a parent that has children, return the list of
child names as individual attribute names....

(Yes, I know listxattr null separates the attribute names - it's a
godawful kernel API - but that's not the programmatic interface
we should expose at the shell script level.)

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com