ocfs2 has a system file called "slot_map". A "slot" is a collection of
files local to particular mounted node, including the journal and
allocators that node is using. The slot map converts the slot number to
a node number, so when a node dies, ocfs2 knows which slot to recover.
The old ocfs2 slot map is a very limited. It has a physical maximum of
254 entries - specifically, it must fit within one disk block. It only
allows node numbers up to 254, and cannot be extended past INT16_MAX
(32767). This is a problem in the world of userspace cluster stacks,
where the node numbers are often sparse and can be up to UINT32_MAX.
It also has the structural problem that empty slots are signified by a
magic number. That number happens to be -1 (0xFFFF). It makes for code
that isn't as obvious as one would like.
Thus, we introduce a new slot map format, referred to hence as the
"extended slot map". The extended slot map is allocated as regular file
space, and so is bound by i_size. The new format adds a "valid" field,
distinct from the node number. Finally, it has room for extension
should it be needed.
The kernel code is available on the 'new-slot-map' branch of my git
repository.
View:
http://oss.oracle.com/git/?p=jlbec/linux-2.6.git;a=shortlog;h=new-slot-map
Pull:
git pull git://oss.oracle.com/git/jlbec/linux-2.6.git new-slot-map
The tools code is also available via git, in the 'new-slot-map' branch
as well.
View:
http://oss.oracle.com/git/?p=ocfs2-tools.git;a=shortlog;h=new-slot-map
Pull:
git pull git://oss.oracle.com/git/ocfs2-tools.git new-slot-map
--
