On Wed, May 30, 2012 at 08:48:22 -0600, Eric Blake wrote:
On 05/30/2012 08:34 AM, Jiri Denemark wrote:
> ---
> src/util/virfile.c | 10 ++++++++++
> 1 file changed, 10 insertions(+)
>
> diff --git a/src/util/virfile.c b/src/util/virfile.c
> index db3d737..bc7f2c9 100644
> --- a/src/util/virfile.c
> +++ b/src/util/virfile.c
> @@ -35,6 +35,7 @@
> #include "configmake.h"
> #include "memory.h"
> #include "virterror_internal.h"
> +#include "logging.h"
>
> #define VIR_FROM_THIS VIR_FROM_NONE
> #define virFileError(code, ...) \
> @@ -51,6 +52,15 @@ int virFileClose(int *fdptr, bool preserve_errno)
> if (preserve_errno)
> saved_errno = errno;
> rc = close(*fdptr);
> + if (rc < 0) {
> + if (errno != EBADF) {
> + char ebuf[1024] ATTRIBUTE_UNUSED;
> + VIR_DEBUG("Failed to close fd %d: %d",
> + *fdptr, virStrerror(errno, ebuf, sizeof(ebuf)));
I think we _also_ need to log EBADF errors, those are usually the sign
of developer bugs, as you are probably guilty of a double-close race.
In fact, I think EBADF should be logged with higher severity than
VIR_DEBUG, as we want to see that right away.
I explicitly didn't want to log them since you would see a lot of errors when
after fork we are closing all available file descriptors except those we
explicitly want to leave open... I'll add EBADF logging and handle those
mass-closes separately.
Jirka