On Fri, Sep 06, 2013 at 04:32:43PM -0400, Jeff Cody wrote: > On Thu, Aug 22, 2013 at 11:29:03AM +0200, Stefan Hajnoczi wrote: > > Print a warning when opening a file O_DIRECT fails with EINVAL. This > > saves users a lot of time trying to figure out the EINVAL error, which > > is typical when attempting to open a file O_DIRECT on Linux tmpfs. > > > > Reported-by: Deepak C Shetty <deepa...@linux.vnet.ibm.com> > > Signed-off-by: Stefan Hajnoczi <stefa...@redhat.com> > > --- > > util/osdep.c | 7 +++++++ > > 1 file changed, 7 insertions(+) > > > > diff --git a/util/osdep.c b/util/osdep.c > > index 685c8ae..62072b4 100644 > > --- a/util/osdep.c > > +++ b/util/osdep.c > > @@ -207,6 +207,13 @@ int qemu_open(const char *name, int flags, ...) > > } > > #endif > > > > +#ifdef O_DIRECT > > + if (ret == -1 && errno == EINVAL && (flags & O_DIRECT)) { > > + error_report("file system may not support O_DIRECT"); > > + errno = EINVAL; /* in case it was clobbered */ > > + } > > +#endif /* O_DIRECT */ > > + > > return ret; > > } > > > > -- > > 1.8.3.1 > > > > > > What about putting something similar in error_setg_file_open(), in > util/error.c? There are other occasions when O_DIRECT causes a file > open to fail (e.g. live snapshots with 'cache=none' to a tmpfs > filesystem). That would give additional info then for QMP commands. > Of course, it would then be necessary to also pass the open flags (or > other equivalent info) to error_setg_file_open().
It would be nice to cover all file open cases. But I think error_setg_file_open() is a bit beyond the scope here. error_setg_file_open() is not used in cases where O_DIRECT is relevant. I checked all callers and they don't seem to allow O_DIRECT. Some of them even use qemu_fopen(). Stefan