Fixes devnix's V lookup

The issue was that the QID for chans that were genned from a Qnixdir had
a bad nixid.  We were passing either 0 (for ctl) or 1 (for image) (which
is what you get when you take (s - Qctl).  This happened to work okay
for ctl, since we usually have a nix0.  But image thought its struct v
was for nix1.

I think we only need to do this for "state machine genned" entries,
meaning the entries under Qnixdir.  The "directly genned" entries should
come in to gen with a properly filled out QID (which came from a
previous gen).

kern/drivers/dev/nix.c

diff --git a/kern/drivers/dev/nix.c b/kern/drivers/dev/nix.c
index c1cb598..7ed4386 100644
--- a/kern/drivers/dev/nix.c
+++ b/kern/drivers/dev/nix.c
@@ -130,6 +130,11 @@
 	return ((index << INDEX_SHIFT) | type);
 }
 
+static inline int QID2ID(struct qid q)
+{
+	return q.path >> INDEX_SHIFT;
+}
+
 /* TODO: (MGMT) not called yet.  -- we have to unlink the nix */
 static void nix_release(struct kref *kref)
 {
@@ -205,11 +210,11 @@
 		s += Qctl;	/* first time through, start on Qctl */
 		switch (s) {
 		case Qctl:
-			mkqid(&q, QID(s-Qctl, Qctl), 0, QTFILE);
+			mkqid(&q, QID(QID2ID(c->qid), Qctl), 0, QTFILE);
 			devdir(c, q, "ctl", 0, eve, 0666, dp);
 			return 1;
 		case Qimage:
-			mkqid(&q, QID(s-Qctl, Qimage), 0, QTFILE);
+			mkqid(&q, QID(QID2ID(c->qid), Qimage), 0, QTFILE);
 			devdir(c, q, "image", 0, eve, 0666, dp);
 			return 1;
 		}