Re: [PATCH 0/2] Support for posix ACLs in fuse

["Jean-Pierre André" <jean-pierre.andre@wanadoo.fr>]

Hi,

Miklos Szeredi wrote:
> [Adding Andreas Gruenbacher to Cc]
>
> On Mon, Aug 29, 2016 at 3:46 PM, Seth Forshee
> <seth.forshee@canonical.com> wrote:
>> Hi Miklos,
>>
>> Here's an updated set of patches for supporting posix ACLs in fuse. I
>> think I've incorporated all the feedback from the last RFC series, and
>> so I've dropped the RFC this time.
>
> Pushed, with minor changes, to
>
>    git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse.git#for-next
>
> Please verify that I didn't break it.
>
>> I also pushed to github the changes I made to libfuse for testing this.
>> They're a little rough and probably not 100% complete, but it is
>> sufficient for exercising the functionality of these patches with
>> fusexmp.
>>
>>   https://github.com/sforshee/libfuse/tree/posix-acl
>
> As for the libfuse part:
>
> 1) Please don't mess with fusexmp.c.  The added code is really an
> anti-example.  Posix acls will will work fine in such pass-through
> filesystems without doing anything.  The added complexity just makes
> it brittle and racy without actually doing anything positive.
>
> 2) You define some constants and structures (POSIX_ACL_*) in
> fuse_common.h that don't seem to belong there.  There's <sys/acl.h>
> that contains some parts of that, but I'm not sure how much we want to
> tie libfuse to libacl...  It's a difficult thing.  Generally I'd try
> to keep the interface as narrow as possible.  Perhaps it's enough to
> have a a function to return the equivalent mode from the xattr?
>
> 3) How will richacl's fit into this?
>
> 4) We really need a better example to check the efficiency of the new
> interface, but that's hard because we need a "real" filesystem for
> that and those are rare.  Ntfs-3g is one such, and it would be
> interesting to "port" it to using the new API.
>
> Jean-Pierre, how difficult would that be?

The code for Posix ACLs support within the kernel is
already present in ntfs-3g (just have to define the
macro KERNELACLS in order to disable the checks
within ntfs-3g and rely on the kernel ones).

This however relies on assumptions I made a few years
back, and might be invalid now. I will at least have
to set some flag in the init callback.

The ACLs are supposed to be set and retrieved through
the extended attributes system.posix_acl_access and
system.posix_acl_default. Recent posts about it in
this list mention "xattr handlers", do they mean
getxattr() and setxattr() on these extended attributes ?

Also the sync with the file mode is done natively,
as the ACLs and modes are translated to a single
object (an NTFS ACL). There is no need for fuse to
duplicate the mode to a ACL setting or conversely.

Now the main problem for ntfs-3g is to migrate to
libfuse3, still keeping the compatibility with non-Linux
implementations (MacOSX, OpenIndiana, and others).
I have not done anything about it yet.

Does the Posix ACL in-kernel support require libfuse3 ?

Finally, I am still using an old kernel, so I will
have to setup a test environment...

Jean-Pierre

>
> Thanks,
> Miklos
>