Re: [fuse-devel] FUSE and handling of O_APPEND for network filesystems

[Roman V. S. <rv...@su...>]

On Tue, 2009-01-06 at 15:26 +0100, Miklos Szeredi wrote:
> > Would this be possible? Are there any better suggestions for
> > handling O_APPEND?
> 
> The best way to handle this is to use the O_APPEND flag in open().

That's not a problem -- O_APPEND is being set in the open() and
I get it as part of FUSE_OPEN message. The fact that O_APPEND
is set also makes generic_write_checks() adjust the pos argument
based on ->i_size:
   http://lxr.linux.no/linux+v2.6.28/mm/filemap.c#L1961
but it doesn't make it revalidate ->i_size!

And that's why I would like to tell FUSE: "please don't forget
to revalidate ->i_size before you call generic_write_checks()
for any file opened as O_APPEND". Exactly like nfs_file_write() 
does it:
   http://lxr.linux.no/linux+v2.6.28/fs/nfs/file.c#L524

Would something like this be possible? If not right away, may
be in the future versions of FUSE?

Thanks,
Roman.