[Qemu-devel] Re: [RFC][PATCH 2/2] qcow2: Use bdrv_(p)write_sync for metadata writes

[Christoph Hellwig <hch@lst.de>]

On Fri, Jun 18, 2010 at 09:54:24AM +0200, Kevin Wolf wrote:
> qcow2 files are expected to grow, so this probably doesn't help us. At
> best we could try to detect if the file grows and decide if to use
> sync_file_range or fdatasync.

No.  sync_file_range is always unsafe for use with a real filesystem.
It never even calls into the filesystem and never flushes the disk write
cache.  For our use case it's exactly a no-op.

> Of course, it would be better to have a flag or something to include the
> necessary metadata, but I have no idea how it's implemented and if this
> is easily possible.

We could add a flag to it.  But rather than adding more crap to this
bastard that should never have been added I'd rather add a real
fsync_range system call.

Note that for our use case it doesn't matter anyway.  For an image file
accessed with cache=none there will be no data in the pagecache, and
the writeback of the inode and associated metadata (btree blocks or
indirect blocks) will be same wether we give it a range or not, and same
for the disk cache flush.