On Tue, Jul 19, 2011 at 10:08:21AM -0600, Eric Blake wrote:
On 07/19/2011 10:06 AM, Daniel P. Berrange wrote:
The motivation for using O_DIRECT is that allowing pollution of the host cache causes stability problems for the host as a whole. As such IMHO, apps would likely want an error back if O_DIRECT cannot be supported,
NB, even some Linux filesystems can't do O_DIRECT, so this isn't an obscure mingw32 issue.
Conversely, open() on Linux silently ignores unknown flags - so if you are using a really old kernel but newer glibc headers, then O_DIRECT is non-zero and open() succeeds, but you _don't_ get direct I/O.
If O_DIRECT is 0, then it is pretty easy to diagnose that the request is unsupported. But if O_DIRECT is non-zero, then how do I tell whether the open(O_DIRECT) really meant that I have direct I/O, or whether it was a nice hint but still ignored and I'm still polluting the file system cache?
Hmm, I could have sworn we've seen QEMU itself fail to start when requesting O_DIRECT on say, tmpfs. Perhaps open() isn't failing, but rather read/write fail ? Daniel -- |: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :| |: http://libvirt.org -o- http://virt-manager.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: http://entangle-photo.org -o- http://live.gnome.org/gtk-vnc :|