On Wed, Aug 21, 2013 at 9:08 AM, Markus Armbruster <arm...@redhat.com>wrote:
> Stefan Hajnoczi <stefa...@redhat.com> writes:
>
> > Print a warning when opening a file O_DIRECT on tmpfs fails. This saves
>
> Only when it fails with EINVAL, actually. Suggest "on tmpfs fails with
> EINVAL."
>
> > users a lot of time trying to figure out the EINVAL error.
> >
> > Daniel P. Berrange <berra...@redhat.com> suggested opening the file
> > without O_DIRECT as a portable way to check whether the file system
> > supports O_DIRECT. That gets messy when flags contains O_CREAT since
> > we'd create a file but return an error - or a race condition if we try
> > to unlink the file. It's simpler to check the file system type.
> >
> > Reported-by: Deepak C Shetty <deepa...@linux.vnet.ibm.com>
> > Signed-off-by: Stefan Hajnoczi <stefa...@redhat.com>
> > ---
> > util/osdep.c | 19 +++++++++++++++++++
> > 1 file changed, 19 insertions(+)
> >
> > diff --git a/util/osdep.c b/util/osdep.c
> > index 685c8ae..446a1dc 100644
> > --- a/util/osdep.c
> > +++ b/util/osdep.c
> > @@ -30,6 +30,11 @@
> > #include <unistd.h>
> > #include <fcntl.h>
> >
> > +#ifdef __linux__
> > +#include <sys/vfs.h>
> > +#include <linux/magic.h>
> > +#endif
> > +
> > /* Needed early for CONFIG_BSD etc. */
> > #include "config-host.h"
> >
> > @@ -207,6 +212,20 @@ int qemu_open(const char *name, int flags, ...)
> > }
> > #endif
> >
> > +#ifdef __linux__
> > + /* It is not possible to open files O_DIRECT on tmpfs. Provide a
> hint that
> > + * this may be the case (of course it could change in future kernel
> > + * versions).
> > + */
> > + if (ret == -1 && errno == EINVAL && (flags & O_DIRECT)) {
> > + struct statfs st;
> > + if (statfs(name, &st) == 0 && st.f_type == TMPFS_MAGIC) {
> > + error_report("tmpfs file systems may not support O_DIRECT");
> > + }
> > + errno = EINVAL; /* in case it was clobbered */
> > + }
> > +#endif /* __linux__ */
> > +
> > return ret;
> > }
>
> In theory, the warning could be misleading, because we can get EINVAL
> for reasons not related to flags & O_DIRECT. In practice, the warning
> probably does much more good than harm.
>
>
Actually EINVAL is only for O_DIRECT with open(). There are other
filesystems that will return this other than tmpfs as well, but we were
discussing /var/run so the assumption was tmpfs [1]. The code that performs
the check to see if EINVAL should be returned checks to see if the
direct_IO address space op is defined [2] (or get_xip_mem but that's
another story).
[1] https://www.redhat.com/archives/libvir-list/2013-August/msg00768.html
[2] http://lxr.linux.no/#linux+v3.10.9/fs/open.c#L651
--
Doug Goldstein
