Re: [PATCH v15 00/22] Richacls (Core and Ext4)

[Andreas Gruenbacher]

On Tue, Nov 10, 2015 at 6:07 PM, J. Bruce Fields <bfields@xxxxxxxxxxxx> wrote:
> On Tue, Nov 10, 2015 at 10:43:46AM -0600, Steve French wrote:
>> On Tue, Nov 10, 2015 at 6:39 AM, Andreas Gruenbacher
>> <agruenba@xxxxxxxxxx> wrote:
>> > On Tue, Nov 10, 2015 at 12:29 PM, Christoph Hellwig <hch@xxxxxxxxxxxxx> wrote:
>> >> On Mon, Nov 09, 2015 at 12:08:41PM +0100, Andreas Gruenbacher wrote:
>> >>> Here is another update to the richacl patch queue.  This posting contains
>> >>> the patches ready to be merged; the patches later in the queue still need
>> >>> some more review.
>> <snip>
>> >> and still abuses xattrs instead of a proper syscall interface.
>> >> That's far from being ready to merge.
>> >
>> > The xattr syscall interface is what's used for very similar kinds of
>> > things today; using it for richacls as well sure does not count as
>> > abuse. Things could be improved in the xattr interface and in its
>> > implementation, but we need more substantial reasons than that for
>> > reimplementing the wheel once again.
>>
>> I don't have strong disagreement with using pseudo-xattrs to
>> store/retrieve ACLs (we already do this) but retrieving/setting an ACL
>> all at once can be awkward  when ACLs are quite large e.g. when it
>> encodes to over 1MB
>
> At least in the NFS case, that's also a limitation of the protocol.

I couldn't find a limit in the NFSv4 specification, but the client and
server implementations both define arbitrary ACL size limits. In
addition, the xattr syscalls allow attributes to be up to 64k long.

> If
> we really wanted to support massive ACLs then we'd need both syscall and
> NFS interfaces to allow incrementally reading and writing ACLs, and I
> don't even know what those would look like.
>
> So this is a fine limitation as far as I'm concerned.

The bigger problem would be incrementally setting ACLs. To prevent
processes from racing with each other, we would need a locking
mechanism. In addition, the memory overhead would be prohibitive and
access decisions would become extremely slow; we would have to come up
with mechanisms to avoid those problems.

Thanks,
Andreas
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/