Re: [PATCH] 2.5.x write_super is not for syncing

[Andrew Morton <akpm@digeo.com>]

Bryan Henderson wrote:
> 
> ...
> But of course a ->write_super call is the only evidence I have that the
> user did a sync(), which is something I need to know.

You and the rest of the world.

>  Unfortunately, a
> ->write_super call can also simply mean kupdated is doing its periodic
> thing.  So the code I had to write to implement sync(), while not looking
> like ext2 in a bunch of other ways, horribly abuses the interface, works
> differently on different versions of Linux, and in some circumstances isn't
> even 100% correct.
> 
> A driver for the same filesystem type on AIX has no problem because AIX has
> an equivalent to ->sync_fs and nothing like ->write_super (AIX VFS is more
> general -- doesn't pretend to know about things like super blocks).
> 
> So this change looks like it makes Linux as hospitable as AIX to STFS, but
> here's where it appears to fall short:  It still uses the same VFS call to
> implement the sync() system call as to implement kupdated's purpose.
> Could there be a separate function or an argument, or do I have to continue
> looking at the process name to distinguish them?

Oh darn.  Yes, both kupdate and sync call sync_supers().  But
kupdate should not call ->sync_fs.

I'll fix that up.  The intent is that ->sync_fs() should only be
called for sync, unmount, etc.