On Wed, Jul 16, 2014 at 08:30:02PM +0200, Martin Kletzander wrote:
This eliminates the need for active waiting.
Resolves:
https://bugzilla.redhat.com/show_bug.cgi?id=927369
Signed-off-by: Martin Kletzander <mkletzan(a)redhat.com>
---
src/rpc/virnetsocket.c | 58 +++++++++++++++++++++++++++++++++-----------------
1 file changed, 39 insertions(+), 19 deletions(-)
diff --git a/src/rpc/virnetsocket.c b/src/rpc/virnetsocket.c
index a94b2bc..c00209c 100644
--- a/src/rpc/virnetsocket.c
+++ b/src/rpc/virnetsocket.c
[...]
@@ -569,28 +572,45 @@ int virNetSocketNewConnectUNIX(const char *path,
[...]
+ /*
+ * We cannot do the umask() trick here because that's not
+ * thread-safe. fchmod(), however, is not guaranteed to work on
+ * some BSD favours, but *should* work on Linux before the socket
+ * is bound. POSIX says the behaviour of fchmod() called on
+ * socket is unspecified, though.
+ */
+ if (fchmod(passfd, 0700) < 0) {
+ virReportSystemError(errno, "%s",
+ _("Failed to change permissions on socket"));
+ goto error;
}
I've finally found a way out of this. We can fork() and in the child
do only umask() and bind(). It shouldn't be a problem that fstat()
returns different mode for the socket than stat(), it should work
everywhere and thanks to the fact that we do this pretty rarely and
copy-on-write pages there shouldn't be significant impact.
Is this acceptable?
Martin