Implemented asynchronous DNS lookups in nmbd.

I realised this afternoon just how easy it is to add this, so I
thought I'd implement it while the idea was fresh.

nmbd forks at startup and uses a pipe to talk to its child. The child
does the DNS lookups and the file descriptor of the child is added to
the main select loop.

While I was doing this I discovered a bug in nmbd that explains why
the dns proxy option has been so expensive. The DNS cache entries in
the WINS list were never being checked, which means we always did a
DNS lookup even if we have done it before and it is in cache. I'm sure
this used to work (I tested the DNS cache when I added it) so someone
broke it :-(

Anyway, the async DNS gets rid of the problem completely. I'll commit
just the fix to the DNS cache bug to the 1.9.17 tree.

You can disable async DNS by adding -DSYNC_DNS to the compile flags.
(This used to be commit 178e27de0791c1ff3268cb456ed5c5efc9ac2a01)

1 files changed, 20 insertions, 12 deletions

diff --git a/source3/nameservreply.c b/source3/nameservreply.c
index c901059f9b..6585a02261 100644
--- a/source3/nameservreply.c
+++ b/source3/nameservreply.c
@@ -569,23 +569,31 @@ void reply_name_query(struct packet_struct *p)
     /* look up the name in the cache */
     n = find_name_search(&d, question, FIND_LOCAL, p->ip);
 
+    /* check for a previous DNS lookup */
+    if (!n && (n = find_name_search(&d, question, FIND_WINS, p->ip))) {
+	    if (n->source != DNS && n->source != DNSFAIL) {
+		    n = NULL;
+	    } else {
+		    DEBUG(5,("Found DNS cache entry %s\n", namestr(&n->name)));
+	    }
+    }
+
     /* it is a name that already failed DNS lookup or it's expired */
     if (n && (n->source == DNSFAIL ||
-              (n->death_time && n->death_time < p->timestamp)))
-    {
-      success = False;
+              (n->death_time && n->death_time < p->timestamp))) {
+	    DEBUG(5,("expired name %s\n", namestr(&n->name)));
+	    success = False;
     }
+
    
     /* do we want to do dns lookups? */
-    /* XXXX this DELAYS nmbd while it does a search.  lp_dns_proxy()
-       can be switched off, to ensure that the blocking doesn't occur.
-       a better solution would be to fork, but this will require a
-       mechanism to carry on processing after the query is resolved
-       (similar to the netbios queue).
-     */
-    if (success && !n && (lp_dns_proxy() || !bcast))
-    {
-      n = dns_name_search(question, p->timestamp);
+    if (success && !n && (lp_dns_proxy() || !bcast)) {
+	    BOOL dns_type = (name_type == 0x20 || name_type == 0);
+	    if (dns_type && wins_subnet) {
+		    /* add it to the dns name query queue */
+		    if (queue_dns_query(p, question, &n))
+			    return;
+	    }
     }
   }