Re: Reiser4 for 3.6 a.k.a. removal of ->write_super()

Re: Reiser4 for 3.6 a.k.a. removal of ->write_super()

[Edward Shishkin]

On 09/19/2012 08:28 PM, Ivan Shapovalov wrote:
> Hi, Edward!
>

Hello.

> The 3.6 is will be released soon, and among trivial vfs changes there is one
> pretty non-trivial: removal of sync_supers thread and ->write_super() of
> super_operations.
> Reiserfs (v3) compensates for it by introducing its own workqueue; does
> reiser4 need something like that (or write_super() may be just removed
> completely)?
> If yes, then I'll attempt to make analogous changes to reiser4...
>

Long time ago (before 2.6.32, or so) ->write_super() was used by vfs
for sync(). However, starting from .32 (or so) it isn't, so I have
written a reiser4 implementation of ->sync_fs() for this purpose.

It seems that nowadays reiser4_sync_super is completely defuncted
(please, check) and is not needed any more. Reiser4 has its own
periodic writeout means (ktxnmgrd). ReiserFS (v3) didn't have this one.

So, I think, remove...

Thanks,
Edward.

Re: Reiser4 for 3.6 a.k.a. removal of ->write_super()

[Ivan Shapovalov]

On Wednesday 19 September 2012 22:55:48 Edward Shishkin wrote:
> On 09/19/2012 08:28 PM, Ivan Shapovalov wrote:
> > Hi, Edward!
> 
> Hello.
> 
> > The 3.6 is will be released soon, and among trivial vfs changes there is
> > one pretty non-trivial: removal of sync_supers thread and ->write_super()
> > of super_operations.
> > Reiserfs (v3) compensates for it by introducing its own workqueue; does
> > reiser4 need something like that (or write_super() may be just removed
> > completely)?
> > If yes, then I'll attempt to make analogous changes to reiser4...
> 
> Long time ago (before 2.6.32, or so) ->write_super() was used by vfs
> for sync(). However, starting from .32 (or so) it isn't, so I have
> written a reiser4 implementation of ->sync_fs() for this purpose.
> 
> It seems that nowadays reiser4_sync_super is completely defuncted
> (please, check) and is not needed any more. Reiser4 has its own
> periodic writeout means (ktxnmgrd). ReiserFS (v3) didn't have this one.
> 
> So, I think, remove...
Right, I've just checked the code - it never sets superblock's ->s_dirt to 1,
so the code appears to be completely unused.

Actually, the fs doesn't appear to be damaged after 6 days of usage without
any compensations (I'm running 3.6-rc5 since Sep 14), so I guess it's true. :)

Thanks,
Ivan.
P. S.: I'll send the patch rebased on top of linux 3.6 once it's released...