Re: [PATCH]: reiser4 [8/8] reiser4 code

[Nikita Danilov]

Alexander Viro writes:
 > 
 > 
 > On Thu, 31 Oct 2002, Nikita Danilov wrote:
 > 
 > >  > And you want that to be reviewed until tonight?
 > >  > 
 > > 
 > > No. But changes to the core are not very complicated. If Linus "reviews"
 > > and accepts them life of reiser4 would be much simpler.
 > 
 > Changes to the core consist (AFAICS) of exporting a bunch of functions
 > with no explanation of the way they are used - with some of them it's
 > really straightforward (and can go in at any point), with some one
 > would expect really detailed explanation and code review (your comments
 > regarding fsync_super() export trigger all sorts of alarms for me).

Let's start from fsync_super() then.

Reiser4 has data journalling.

When ->writepage() is called on dirtied page, page joins transaction.
During umount all out-standing transaction have to be committed.  But if
file were mmapped, then, at the moment of the call to ->kill_super()
pages can be dirtied without ->writepage() ever called on them.

generic_shutdown_super() calls fsync_super(sb) (which will call
->writepage() on each dirty page) and then invalidate_inodes().

Reiser4 has commit out-standing transactions -between- these two points:
after ->writepage() has been called on all dirty pages, but before
inodes were destroyed. Thus, we cannot use
kill_block_super()/generic_shutdown_super().

 > 
 > PS: Cc'ing a posting on a public list to a subscribers-only one is
 > generally not a nice thing to do...  Cc: trimmed.

Arghh... Sorry, it will be fixed... soon.

 > 

Nikita.