#! /bin/sh /usr/share/dpatch/dpatch-run
## bts770688_snmp_memleak.dpatch by Pierre-Yves Ritschard <pyr@spootnik.org>
## and Marc Fournier <marc.fournier@camptocamp.com>
##
## DP: Let snmp_synch_response deal with PDU freeing
## DP:
## DP: When reading from tables, upon errors the PDUs sent are already
## DP: freed by snmp_synch_response since they are right after
## DP: snmp_send is called.
## DP:
## DP: This commit syncs collectd's approach with other occurences of
## DP: snmp_synch_response calls.
## DP:
## DP: There might be a few corner cases where we leak PDUs, but it
## DP: is unclear how to check for those since we would need to
## DP: have an indication that snmp_send was never called, which
## DP: as far as I can tell is not possible.
## DP:
## DP: The potential for failure in snmp_send is rather low and will
## DP: be easily spotted though, since when crafting invalid PDUs
## DP: snmp send will constantly fail and since valid configurations
## DP: can never leak memory.
## DP:
## DP: Upstream bug reports:
## DP: https://github.com/collectd/collectd/issues/610
## DP: https://github.com/collectd/collectd/issues/804

@DPATCH@

diff --git a/src/snmp.c b/src/snmp.c
index ad81c89..7d340d1 100644
--- a/src/snmp.c
+++ b/src/snmp.c
@@ -1316,6 +1316,8 @@ static int csnmp_read_table (host_definition_t *host, data_definition_t *data)
         snmp_free_pdu (res);
       res = NULL;
 
+      /* snmp_synch_response already freed our PDU */
+      req = NULL;
       sfree (errstr);
       csnmp_host_close_session (host);
 
@@ -1437,6 +1439,10 @@ static int csnmp_read_table (host_definition_t *host, data_definition_t *data)
     snmp_free_pdu (res);
   res = NULL;
 
+  if (req != NULL)
+    snmp_free_pdu (req);
+  req = NULL;
+
   if (status == 0)
     csnmp_dispatch_table (host, data, instance_list_head, value_list_head);