As you may have heard, Linus has accepted the revised extended
attributes interfaces which several projects have been working
on during the past few months, including ours, into 2.5.3.
>From an XFS development point of view, this is great news and
will result in a significant decrease in the overlap which the
XFS patches have with the rest of the kernel, and also means
that we will no longer need to use unreserved system calls.
Inclusion of XFS in certain distributions has been held up due
to this, and once the transition to these interfaces is complete,
I understand these will be able to proceed. We will also be able
to converge on a single set of user tools for EAs and POSIX ACLs
that other filesystems will be using as well (which is good news
for Samba and any other upstream libacl.so users).
>From an end-user point of view, we will need to make a transition
from using our current system call interfaces for EAs and ACLs to
the new interfaces. This affects both the kernel and a number of
the userspace tools and libraries. We've considered attempting to
keep compatibility with the existing interfaces too, but ultimately
have discarded that option and will instead make a clean break from
one interface to the other (this applies to both the 2.4.x kernels
and the 2.5.x kernels). This means that with a future kernel merge
we will also release new userspace packages (I expect to move all
our packages up to version 2.0.0) and I also expect responsibility
for the ACL package we use to be taken over by Andreas Gruenbacher
- the author/maintainer of the [http://acl.bestbits.at/] ext2/ext3
ACL implementation - at some point in the not-too-distant-future.
XFS 1.1 is scheduled for this quarter, I plan on having this all
integrated for that release for those who track the official SGI
releases rather than CVS or patches.
Further details will follow later, this is just a heads-up that a
significant interface change is in the pipeline.