[master] Honor prefix length hint for returning clients

Merges in rt45780.

RELNOTES

@@ -183,11 +183,14 @@ dhcp-users@lists.isc.org.
   Correct parsing of DUID from config file, previously the LL type
   was put in the wrong place in the DUID string.
   [ISC-Bugs #20962]
+
   Add code to parse "do-forward-updates" as well as "do-forward-update"
   Thanks to Jiri Popelka at Red Hat.
   [ISC-Bugs #31328]
+
   Remove log_priority as it isn't currently used.
   [ISC-Bugs #33397]
+
   Increase the size of the buffer used for reading interface information.
   [ISC-Bugs #34858]
 
@@ -1126,6 +1129,14 @@ dhcp-users@lists.isc.org.
   Thanks to Tim Dean for reporting this issue.
   [ISC-bugs #31573]
 
+- Corrected an issue where the server would return a client's previously
+  released prefix lease even when the client provides a prefix length
+  hint that does not match the prior lease.  Now the server will only
+  return the previous lease if it exactly matches the hint.  If not
+  it will attempt to allocate a new prefix based on the hint and the
+  prefix-length-mode.
+  [ISC-bugs #45780]
+
 			Changes since 4.2.0 (new features)
 
 - If a client renews before 'dhcp-cache-threshold' percent of its lease

server/dhcpv6.c

@@ -4664,6 +4664,19 @@ find_client_prefix(struct reply_state *reply) {
 			best_prefix = prefix_compare(reply, prefix,
 						     best_prefix);
 		}
+
+		/*
+		 * If we have prefix length hint and we're not igoring them,
+		 * then toss the best match if it doesn't match the hint,
+		 * unless this is in response to a rebind.  In the latter
+		 * case we're supposed to return it with zero lifetimes.
+		 * (See rt45780) */
+		if (best_prefix && (reply->preflen > 0)
+		    && (prefix_length_mode != PLM_IGNORE)
+		    && (reply->preflen != best_prefix->plen)
+		    && (reply->packet->dhcpv6_msg_type != DHCPV6_REBIND)) {
+			best_prefix = NULL;
+		}
 	}
 
 	/* Try to pick a new prefix if we didn't find one, or if we found an