[Pkg-dns-devel] Bug#879633: knot secondary fails to transfer zone from primary

[Jens Meißner]

Source: knot
Version: 2.4.0-3+deb9u1
Severity: normal

Hi,

we are using knot as secondary name server with powerdns as primary.

With the version of knot shipped with stretch the zone transfer always fails
with the following error message:

Oct 23 18:42:02 ns1 knotd[7587]: info: [example.org.] refresh, outgoing,
192.0.2.53 at 53: remote serial 2017102301, zone is outdated
Oct 23 18:42:02 ns1 knotd[7587]: info: [example.org.] IXFR, incoming,
192.0.2.53 at 53: starting
Oct 23 18:42:02 ns1 knotd[7587]: warning: [example.org.] IXFR, incoming,
192.0.2.53 at 53: failed (malformed data)
Oct 23 18:42:02 ns1 knotd[7587]: warning: [example.org.] refresh, outgoing,
192.0.2.53 at 53: fallback to AXFR
Oct 23 18:42:02 ns1 knotd[7587]: warning: [example.org.] refresh, remote
'master' not usable
Oct 23 18:42:02 ns1 knotd[7587]: error: [example.org.] refresh, failed (no
usable master)

After several minutes the transfer is retried and fails again with the same
error.

The bug is fixed upstream already. The corresponding commit is:
https://gitlab.labs.nic.cz/knot/knot-
dns/commit/b4ff623a1fbe410e1ab2eaa3413f38f613190b8a
More information about this bug can be found on the knot mailing list:
https://lists.nic.cz/pipermail/knot-dns-users/2017-January/001043.html

An adapted patch for the knot version in stretch is attached to this message.

Regards,
Jens
-------------- next part --------------
diff -rNu a/src/knot/events/handlers/refresh.c b/src/knot/events/handlers/refresh.c
--- a/src/knot/events/handlers/refresh.c	2017-01-18 16:35:39.000000000 +0100
+++ b/src/knot/events/handlers/refresh.c	2017-10-23 19:15:30.125334866 +0200
@@ -748,6 +748,7 @@
 		REFRESH_LOG(LOG_WARNING, data->zone->name, data->remote,
 		            "fallback to AXFR");
 		ixfr_cleanup(data);
+		layer->flags |= KNOT_RQ_LAYER_CLOSE;
 		data->is_ixfr = false;
 		return KNOT_STATE_RESET;
 	}
diff -rNu a/src/knot/query/layer.h b/src/knot/query/layer.h
--- a/src/knot/query/layer.h	2017-01-18 16:35:39.000000000 +0100
+++ b/src/knot/query/layer.h	2017-10-23 19:15:30.125334866 +0200
@@ -48,6 +48,7 @@
 	void *data;                   //!< Module specific.
 	const struct knot_layer_api *api;
 	tsig_ctx_t *tsig;             //!< TODO: remove
+	unsigned flags;               //!< Custom flags.
 };
 
 /*! \brief Packet processing module API. */
diff -rNu a/src/knot/query/requestor.c b/src/knot/query/requestor.c
--- a/src/knot/query/requestor.c	2017-01-18 16:35:39.000000000 +0100
+++ b/src/knot/query/requestor.c	2017-10-23 19:15:30.125334866 +0200
@@ -197,6 +197,14 @@
 	knot_layer_reset(&req->layer);
 	tsig_reset(&last->tsig);
 
+	if (req->layer.flags & KNOT_RQ_LAYER_CLOSE) {
+		req->layer.flags &= ~KNOT_RQ_LAYER_CLOSE;
+		if (last->fd >= 0) {
+			close(last->fd);
+			last->fd = -1;
+		}
+	}
+
 	if (req->layer.state == KNOT_STATE_RESET) {
 		return KNOT_LAYER_ERROR;
 	}
diff -rNu a/src/knot/query/requestor.h b/src/knot/query/requestor.h
--- a/src/knot/query/requestor.h	2017-01-18 16:35:39.000000000 +0100
+++ b/src/knot/query/requestor.h	2017-10-23 19:15:30.125334866 +0200
@@ -31,6 +31,10 @@
 	KNOT_RQ_UDP = 1 << 0  /* Use UDP for requests. */
 };
 
+enum {
+	KNOT_RQ_LAYER_CLOSE = 1 << 0
+};
+
 /*! \brief Requestor structure.
  *
  *  Requestor holds a FIFO of pending queries.
@@ -48,7 +52,6 @@
 	knot_pkt_t *query;
 	knot_pkt_t *resp;
 	tsig_ctx_t tsig;
-	knot_layer_t layer;
 
 	knot_sign_context_t sign; /* TODO: Remove. Used in updates only, should
 	                             be part of the zone update context. */